Nature Chemistry Volume 02 No 06 Pp425-510

nature chemistry | VOL 2 | JUNE 2010 | www.nature.com/naturechemistry 425

editorial

The relationship between science and politics has been likened to a marriage1, with the inference being that, to develop, the partners must not become alike but must respect their differences — and that the odd quarrel along the way is no big deal. Recently, however, science has taken the role of the meek, misunderstood spouse that has little influence over their all-powerful partner. Science must become stronger in this relationship; at present it does not have the respect it deserves from most politicians, and so its champions must become louder within the political arena if we are to address the grand challenges of the coming century.

Two recent incidents in the UK suggest that scientists hold little political power, with the real crux of the matter being a lack of science-literate politicians. Although some prominent politicians have science backgrounds (Margaret Thatcher and Angela Merkel were chemists) out of the 650 (pre-2010 election) UK members of parliament (MPs), 27 held science degrees and 584 admitted to having no political interest in science and technology — and taking into account upcoming retirements, it’s about to get worse2. This alarming finding calls into question whether the people responsible for making important policy decisions, either based on scientific research or about its funding, fully understand its importance or crucially the scientific method at its core.

The scientific method relies on the search for and critical consideration of evidence on which to base explanations and decisions, but it is often trumped by political or economic considerations, or more worryingly, is just not understood. Although UK government policy may be more informed by evidence than in the past, the way in which hard scientific evidence is handled and debated often seems to result in two steps forward and one step back. Take for example the way in which a frightening number of UK MPs have reacted to a recent government report on homeopathy — a practice currently funded publicly by the National Health Service.

In a progressive move, a science and technology committee, made up of the UK’s more science-minded MPs, was tasked to look at the supporting evidence for homeopathy to help re-evaluate government policy. The

thorough, evidence-led report, found that “homeopathy is not efficacious (that is, it does not work beyond the placebo effect) and that explanations for why homeopathy would work are scientifically implausible”. The committee therefore recommended the withdrawal of funding. On publication of the report, MP David Tredinnick — also an advocate of unscientific practices such as ‘medical astrology’ and ‘remote energetic healing’ — spearheaded a movement to reject the findings. Even though the rigours of the scientific method are there to behold in this report — which dismisses outright the value of homeopathy — 70 MPs supported Tredinnick’s campaign.

Science has fought back in the form of science writer, Michael Brooks. He took the view that if our politicians don’t listen to rational, reasoned arguments he would change things by trying to become one of them, standing against Tredinnick as a candidate in the recent 2010 election3. Although ultimately unsuccessful, this unusual move has gone some way to raising the profile of science.

The case of homeopathy is just one extreme example of how politicians can harm the development of science by undermining and questioning its credibility. Further troubles undoubtedly lie ahead because so many of the challenges faced by society today rely on knowledge afforded through scientific research, in topics far more complicated and with greater ramifications than homeopathy: climate change, energy provision and genetic modification to name but a few. To deal with them ably, governments must become more science-literate. This doesn’t require our politicians to be scientists (although this would help), rather they must have an appreciation of science. Foremost, however, they must learn how scientific research actually works to reach the best possible explanation for a given set of hypotheses.

There are obviously no expectations that politicians must be experts in cutting-edge science — any government must also have professional science advisers who present evidence around which policy can be moulded. Again, however, the UK government has shown how, even on the most scientific of subjects where expert testimony is of paramount importance, their

policies can be chosen without full regard for the evidence with which they are provided, as in the case of Professor David Nutt, the government’s former chief drug adviser.

In the first of a series of spats over Nutt contradicting the government’s hard line on illegal drugs, it reprimanded him for his efforts to show how the harm drugs can cause compares with other potentially harmful activities — for example, he highlighted that horse riding was statistically riskier than using the drug ecstasy. Then, after ignoring his advice to not reclassify cannabis from class C to B, it later sacked him for publicly presenting evidence suggesting that LSD, ecstasy and cannabis were in fact less harmful than alcohol and tobacco — again contradicting the government’s stance. Although the issues involved in the classification of illegal drugs are not clear cut, and governments must balance many different factors, ignoring expert advice and attempting to silence its communication to the public — rather than explaining their decisions in the face of it — undermines science and its worth.

Winston Churchill once said that “courage is what it takes to stand up and speak; courage is also what it takes to sit down and listen”. Science needs courage. It needs courage from scientists like David Nutt and Michael Brooks to stand up and speak out assertively to communicate its importance to politicians and to not back down when the evidence is clear. Scientists must question policies that ignore the evidence and must try to educate those who make them. In return, we need courage from governments and politicians. Governments must try to teach their members more about the basics and value of science so that policy decisions can be informed by a greater understanding of the technical issues. Politicians should seek advice from scientists and be open to learning more about what may seem like difficult topics. They must be willing to trust the scientific method and, in doing so, make tough decisions that not only consider the political ramifications, but take into account the underlying evidence. ❐

References1. Price, D. K. The Scientific Estate (Harvard Univ. Press, 1965).2. http://go.nature.com/6UsL5q3. http://www.scienceparty.org.uk/

Although politics has been defined as the ‘science of government’, there is little science in government. Recent events in UK politics have highlighted the lack of scientifically literate elected representatives — a situation that must change for the good of society.

Bringing science to the party

nchem_.671_JUN10.indd 425 7/5/10 11:31:40

© 20 Macmillan Publishers Limited. All rights reserved10

http://go.nature.com/6UsL5q

http://www.scienceparty.org.uk/

426 nature chemistry | VOL 2 | JUNE 2010 | www.nature.com/naturechemistry

research highlightsWater-OXiDatiOn cataLysts

stable and ableScience 328, 342–345 (2010)

Developing a cheap and effective water-oxidation catalyst (WOC) is one of the key breakthroughs needed to produce clean energy from artificial photosynthesis. Although heterogeneous WOCs have some advantages over homogeneous ones, the latter are easier to study and this increased understanding makes them easier to improve. Many homogeneous WOCs, however, have organic ligands that undergo oxidative degradation, or include expensive and rare metals, such as ruthenium.

Now, Craig Hill and colleagues from Emory University in Atlanta have developed a homogeneous WOC with a cobalt oxide core, stabilized by polytungstate ligands. The catalyst is made from cheap and readily available starting materials in a fairly simple one-pot synthesis. Although Hill and co-workers made a range of similar cobalt-based polyoxometalate compounds, only one showed catalytic activity. This had a turnover frequency — the number of molecules of water oxidized by a molecule of the catalyst — of 5 s−1, much higher than a similar heterogeneous cobalt phosphate catalyst.

To show that their catalyst was stable and did not break down into aqueous cobalt ions — which can themselves act as WOCs — Hill and colleagues performed a range of experiments. As well as UV–visible and NMR spectroscopy showing no changes over a month, adding a ligand known to inhibit the catalytic activity of aqueous cobalt ions had no effect on the catalyst. Furthermore, computational studies showed that the highest occupied molecular orbitals of the catalyst were located on the cobalt core, with no contribution from the polytungstate, showing that the ligands are effectively inert.

eXhaust cataLysts

Perovskites prove potentScience 327, 1624–1627 (2010)

Diesel engines are more fuel efficient than petrol ones but, because they run at higher air-to-fuel ratios, they give off more harmful NOx (NO and NO2). Removing NOx in this oxygen-rich environment is challenging, but made

easier by increasing the proportion of NO2 compared with NO. Current exhaust systems require the use of expensive platinum-based catalysts to do this, and these also have poor stability at the high operating temperatures.

Now, Wei Li and colleagues from General Motors Global Research and Development in Michigan have developed perovskite oxide catalysts that do the job as well or better than the commercially available ones. The oxides were LaCoO3 and LaMnO3 with some of the lanthanum replaced with a small amount of strontium. This doping not only almost doubled the surface area of the solids, but also promoted the catalysis itself in LaCoO3, by increasing the number of weakly bonded oxygen atoms.

A useful exhaust catalyst also must be able to oxidize carbon monoxide and unburnt hydrocarbons. Although the perovskite catalysts on their own were less effective at catalysing these reactions, combining them with palladium particles made them as effective as commercial platinum catalysts. The palladium additive also helped prevent the new catalysts from becoming poisoned by sulfur, a common problem in exhaust catalysts. The manganese-based catalyst was structurally stable enough to withstand the high temperatures as well as the reducing environments required to regenerate them.

DenDrimers

sticky situationAngew. Chem. Int. Ed. 49, 3030–3033 (2010)

Muscles contract because of the concerted molecular-level interactions between numerous actin and myosin proteins that work together to shorten muscle fibre. Myosin acts like an active ratchet, which binds to a filament of actin, pulls it in one direction, lets go, and then realigns itself before binding again. ATP drives the process and, in particular, it binds to myosin, which reduces the affinity of myosin for binding with actin, causing the two proteins to detach.

Now Takuzo Aida, Kazushi Kinbara and colleagues at the Universities of Tokyo and Tohoku have shown that this process can be hindered or completely stopped by ‘gluing’ the proteins together with a dendrimer that binds to both actin and myosin. The dendrimer is made up of ether branches and is terminated with nine guanidinium ions, which can bind to the oxyanions prevalent on protein surfaces.

To show that it could bind to both proteins and arrest the motion of actin, Aida, Kinbara and colleagues attached numerous myosin proteins to a surface and observed the behaviour of fluorescently tagged actin molecules when added to the mix. When no dendrimer was present the actin filaments

The synthesis of complex molecules relies on methods for the formation of carbon–carbon bonds. Methods that allow more than one carbon–carbon bond to be formed in a single step improve the efficiency of the process. The reaction of organometallic reagents with carbonyl compounds is one of the most popular methods of carbon–carbon bond formation. Adding two different groups to one carbonyl group, however, usually requires a cycle of reductive addition, followed by oxidation and then a second addition.

Now, Pei-Qiang Huang and co-workers from Xiamen University in China have developed a sequential addition of two different organometallic reagents to an amide or a lactam to produce a tertiary alkyl amine in a single pot. The carbonyl oxygen of the amide or lactam is first activated by formation of a triflate, which is a good leaving group. The iminium triflate thus formed is reacted with a Grignard reagent. The loss of the triflate group then results in a new iminium species that can react with a further organometallic reagent — and, importantly, this can be a different one from the first — to form the product.

A wide variety of different organometallic reagents can be employed for the second addition, including further Grignard reagents, alkynyl lithium reagents and even lithium enolates. Huang and co-workers also demonstrate that for a reaction with a lactam that already bears another substituent, the second addition can be made in a diastereoselective fashion.

reDuctiVe aLKyLatiOn

twice as nice Angew. Chem. Int. Ed. 49, 3037–3040 (2010).

© 2

010

AA

AS

nchem_.672-673_JUN10.indd 426 7/5/10 12:00:18



research highlights

moved with a velocity of approximately 4.6 μm s−1 but when the dendrimer was added, their motion stopped completely. A lower-generation dendrimer with only three guanidinium-terminated branches does not act like a molecular glue, so it seems that the number of sites on the dendrimer that can interact with the proteins is key to its ability to bind two proteins simultaneously.

PermeaBLe memBranes

Peptides make the differenceAngew. Chem. Int. Ed. 49, 3034–3036 (2010)

The materials that come to mind when ‘porous solids’ are mentioned are usually zeolites, silicates and carbon-based frameworks. Dipeptides, however, have recently been shown to form hydrogen-bonded microporous crystals, and have attracted interest for their gas sorption properties. Now, a team of Portuguese researchers led by Luís Gales at the Institute of Molecular and Cell Biology in Porto have observed that dipeptide single-crystals can act as permeable membranes able to distinguish between argon, nitrogen and oxygen — a process of interest for air separation, but difficult to carry out because of the similarity in size between the species.

The permeability of three dipeptides — l-leucyl-l-serine (LS), l-valyl-l-isoleucine (VI), and l-alanyl-l-alanine (AA), which crystallize into different lattices with different porosities — was found to be size-dependent. The crystals that displayed narrower pores combined lower gas absorption abilities and guest diffusivities, resulting in poorer permeabilities. Thus, LS, with nanochannels that are larger than argon, nitrogen and oxygen, is permeable to all three. VI, whose channel size is close to that of the gas molecules, was only permeable to oxygen and nitrogen. And although AA’s porosity consists of discrete pockets rather than channels, it was found to be permeable to oxygen.

The AA dipeptide retained its crystallinity after oxygen permeation, suggesting that its guest-induced flexibility is reversible. This dynamic response was also guest-dependent: AA was more permeable to oxygen than to helium — even though helium is a smaller species. This excellent selectivity suggests that dipeptide crystals hold promise for a variety of separation processes.

GLucOse cOnVersiOn

a solid combinationProc. Natl Acad. Sci. USA 107, 6164–6168 (2010)

The conversion of glucose into fructose is widespread in the food industry for the production of the sweetener high-fructose corn syrup, and has also emerged in the field of renewable energy as a path to degrading biomass into fuel or other valuable chemicals. Enzymatic catalysts are typically used for this isomerization, but their lifetime is limited, they require pre-purified substrates, and only work under specific conditions (neutral pH and around 333 K). To remedy these limitations, research is now focusing on chemical catalysts. Basic catalysts have shown promising activity, but are not practical as glucose and fructose easily decompose in alkaline solutions.

Now, Mark Davis and colleagues from the California Institute of Technology have prepared an efficient inorganic, heterogeneous catalyst by incorporating tin centres into a microporous zeolite with large pores (Beta). Much lower or no conversion was observed when tin was inserted into an ordered mesoporous silica or a medium-pore zeolite, respectively, showing that the catalyst’s activity greatly depends on the size of the pores. The tin–Beta heterogeneous catalyst was efficient even under high glucose concentrations (up to 45%), reusable over three cycles, and remained active after subsequently undergoing a typical zeolite regeneration process (involving calcination at 813 K).

Its activity in acidic aqueous solutions also makes the tin-containing zeolite promising for one-pot processes, as demonstrated with the degradation of starch into fructose, coupling hydrolysis and isomerization steps. The isomerization mechanism remains to be elucidated, but the researchers suggest that it occurs through the formation of a five-membered ring involving the tin centre, followed by an intramolecular hydrogen transfer.

The definitive versions of these Research Highlights first appeared on the Nature Chemistry website, along with other articles that will not appear in print. If citing these articles, please refer to the web version.

smart cookiesWord goes chemistry, science goes cool and cookies go nuclear.

A chemistry ‘add-in’ for Word (http://go.nature.com/KynNV9), created by Microsoft in collaboration with the Unilever Centre for Molecular Science Informatics at Cambridge University, aims to make it easier to get your chemistry into the widely used word processor. Lauren Wolf on Newscripts (http://go.nature.com/xkBm6h) pointed out that you need either Word 2007 or 2010, so as she “operates in the Stone Age (or is it Bronze Age?—I’m not sure) and have only Word 2003”, she hasn’t had a chance to try it yet. Wolf also picks up on the fact that many academics use Macs, for which there isn’t a version available. Worry not, Mac users: Peter Murray-Rust, one of the collaborators, hopes that Chem4Word’s open-source nature will mean it should “be available to the Mac Word and Open Office communities”. We must confess that our office uses Stone Age versions of Word too, so we haven’t been able to test it out either.

Is science cool? This question was asked by KJHaxton on Endless Possibilities (http://go.nature.com/HBgcyt) after reading a newspaper article suggesting that it is (http://go.nature.com/NXGfxO). The article quotes a Large Hadron Collider physicist, astronomers, numerous science writers and a comedian, but Haxton isn’t entirely convinced. She welcomes the increase in science on TV and that the press are reporting more scientific experiments, because “We need publicity for the science of the future, the science that the viewer or their children or grandchildren will help fashion.” But Haxton concludes that the only measure is “the number of young people who view it as a viable career choice”.

And finally, following on from last month’s atomic-emission-spectra scarves, we have atomic-orbital cookies. Windell Oskay at Evil Mad Scientist Laboratories (http://www.evilmadscientist.com/article.php/atomiccookies) cut some melamine extrusion plates for a cookie press in the shape of the s, p, d and f orbitals we all know and love. One standard cookie recipe later and, hey presto, we’ve got atomic orbitals in a new and tasty form.

© 2

010

Wil

ey

nchem_.672-673_JUN10.indd 427 7/5/10 12:00:19


http://go.nature.com/KynNV9

http://go.nature.com/KynNV9

http://go.nature.com/xkBm6h

http://go.nature.com/HBgcyt

http://go.nature.com/NXGfxO

http://www.evilmadscientist.com/article.php/atomiccookies

http://www.evilmadscientist.com/article.php/atomiccookies


news & views

Virtually every biological process, from intracellular transport to muscle contraction, is driven by

a molecular machine. These biological machines carry out pretty much the same functions as the macroscopic ones, but they take very different forms. Whereas macroscopic machines typically operate under the constraints of friction, inertia and gravity, their biological counterparts operate in the nanoworld against the Brownian storms created by the random movements of molecules in solution1.

The apparent ease with which biological machines operate fascinates and inspires scientists, while at the same time challenging us to learn how to construct compounds that can perform mechanical functions2,3. A variety of ingenious systems have been prepared, including molecular muscles, motors and shuttles4–6 — but it remains notoriously difficult to control these movements and translate them from the nanoscale into a macroscopic motion. In particular, despite its ubiquity in biological systems, mimicking spring-like function at the nanoscale has proven highly challenging. Writing in Nature Chemistry, Yoshio Furusho and co-workers now describe7 a molecular double helix that behaves in a similar manner to a macroscopic spring — that is, undergoes contraction and extension while winding and unwinding in a unidirectional sense.

Ion binding and release processes are key mechanisms in biological machines. The functioning of some muscle tissue, for example, relies on calcium ions binding. The interaction between the muscle components myosin and actin is governed by the calcium-sensing complex troponin C, which activates or inhibits muscle contraction through calcium binding and release. Taking a leaf out of nature’s book, Furusho and co-workers have used ion-binding events to trigger the spring-like motion of a helicate — a motif that is also omnipresent in nature, the best known being the DNA double helix.

The researchers had previously constructed a double-stranded helicate8

consisting of two hexaphenol strands bridged by two boron atoms through the formation of two spiroborate (–BO4) moieties, and accommodating a sodium cation in its central position. The sodium ion was coordinated to eight oxygen atoms — the two central hydroxyl groups of each hexaphenol strand and the two closest oxygen atoms of each spiroborate moiety. Furusho and colleagues noticed, however, that the four central hydroxyl groups weren’t necessary to hold the complex together and replaced them with hydrogen atoms, thus preparing a new double helicate in which the central sodium cation is coordinated only to the spiroborate moieties (shown in Fig. 1).

The inclusion and removal of the central sodium ion triggers the contraction and extension of the helix. As a sodium ion binds to the spiroborate moieties, it shields the electrostatic repulsion within the helix’s core, causing its contraction along its long axis. When cryptands are added to the solution, they bind to the sodium ions, removing them from the double-helicate complex. This unmasks the negative charges between the spiroborate moieties,

causing them to repel each other and resulting in the partial unwinding of the helix. A detailed analysis of the structure by X-ray crystallography and nuclear magnetic resonance reveals that the helix approximately doubled its length, from 6 to 13 Å (Fig. 1).

In macroscopic and biological springs, the contraction and expansion of springs is typically accompanied with unidirectional twisting, but this has rarely been observed in synthetic molecular systems. Typically, on contraction or extension, synthetic helicates adopt a non-helical conformation, which leads to a racemization of the helicate and a twisting in both the right- and left-handed directions. The double helicate described here, however, retained its inherent chirality; circular dichroism studies show that extension by unwinding of the double-stranded helix proceeds clockwise, and contraction by winding proceeds anticlockwise.

The spring-like motion can be repeated many times simply by adding sodium ions or cryptands (which equates to removing sodium ions) to the solution. The rate of the extension process is slower than that

MOLECULAR MACHINES

Springing into actionControlling the movements of molecular systems through external stimuli is crucial for the construction of nanoscale mechanical machines. A spring-like compound has now been prepared — a double helicate that retains its handedness under ion-triggered extension and contraction.

Ben L. Feringa

N

O

O

N

O

OONa

B B

B B

Figure 1 | Spring-like molecular motion. Inclusion and removal (through trapping by a cryptand) of a sodium ion triggers the contraction and extension of a double helicate. These events are accompanied by a unidirectional twisting.

nchem_N&V_JUN10.indd 429 10/5/10 12:03:47



news & views

of the contraction one — a feature that is attributed to the differences between the sodium binding processes involved (sodium–helicate and sodium–cryptand).

This elegant system shows how a clever design taking advantage of the unique features of helices can lead to a spring-like mechanical motion at the nanoscale. The next step could be an autonomous spring-like motion, in which the metal ion would bind successively to different helical strands. Making the leap to translating the molecular systems operation to macroscopic movement9, reminiscent of the actin–myosin system that achieves muscle contraction, will be a challenge. Any such system will necessitate binding an

ensemble of molecular springs to a surface and achieving concerted action. One could also foresee associating the ion binding and release events to a catalytic function, or controlling it by light irradiation. This would set the stage for the construction of a truly molecular mechanical device.

The molecular spring presented by Furusho and co-workers7 combines the beauty of molecular helicity with a useful function and is a significant step on the long and winding road towards molecular nanotechnological devices. ❐

Ben L. Feringa is at the Centre for Systems Chemistry, Stratingh Institute for Chemistry and Zernike Insitute for Advanced Materials,

University of Groningen, Nijenborgh 4, 9747 AG, Groningen, The Netherlands. e-mail: [email protected]

References1. Astumian, R. D. Science 276, 917–922 (1997).2. Browne, W. R. & Feringa, B. L. Nature Nanotech. 1, 25–35 (2006).3. Euan, R., Kay, E. R., Leigh, D. A. & Zerbetto, F.

Angew. Chem. Int. Ed. 46, 72–191 (2006).4. Huang, J. et al. Appl. Phys. Lett. 85, 5391–5393 (2003).5. Collin, J-P., Dietrich-Buchecker, C., Gavina, P.,

Jimenez-Molero, M. C. & Sauvage, J-P. Acc. Chem. Res. 34, 477–487 (2001).

6. Kinbara, K. & Aida, T. Chem. Rev. 105, 1377–1400 (2005).7. Miwa, K., Furusho, Y. & Yashima, E. Nature Chem.

2, 444–449 (2010).8. Katagiri, H., Miyagawa, T., Furusho, Y. & Yashima, E.

Angew. Chem. Int. Ed. 45, 1741–1744 (2006).9. Percec, V., Rudick, J. G., Peterca, M. & Heiney, P. A.

J. Am. Chem. Soc. 130, 7503–7508 (2008).

For a reaction to have superfast kinetics, compared with expectations from textbook equations, it helps if a higher

power intervenes and microscopically arranges the reacting molecules to be close to each other — closer than in a random distribution. Such ‘non-classical’ kinetics do occur for some heterogeneous chemical reactions and may play a large role in biology. Writing in Nature Chemistry, Olivier Bénichou and colleagues use theory to investigate such “geometrically controlled reactions”1. They study reactions in a geometrically confined (topologically tortuous) reaction space in which the reactants have a spatially ordered distribution (Fig. 1). Such situations are of much interest at present because they may be typical of highly significant subcellular biochemical reactions of potential biomedical importance, such as gene transcription. Similar situations are also encountered in condensed-state physical and chemical reactions, such as exciton and electron–hole recombination or trapping, which have relevance to photonics and solar-energy science.

Bénichou and colleagues1 use theory that goes beyond what has been generally termed non-classical or fractal-like kinetics, but still use a random-walk-based approach (that is, diffusion-limited

reaction kinetics) to obtain analytical expressions that allow straightforward computations. To understand such theory we must first introduce the classical concepts of chemical reaction kinetics, where an elementary bimolecular reaction, at time t, is described by R(t) = kA(t)B(t). Here, A(t) is the instantaneous concentration (or activity) of reactant A, and B(t) is that of reactant B. R(t) is the instantaneous reaction rate and k is a constant that doesn’t change with time and depends on the transport coefficients of the reactant molecules, as well as on the so-called reaction cross-section, or reaction probability at collision. What this classical formula does not seem to depend on is the size and shape of the reaction vessel, or the locations of the molecules. What is assumed implicitly in this expression is that the chemistry is taking place within a large reaction vessel with a homogeneous and random distribution of molecules at all times, that is, with perfect stirring. Furthermore, it is implicit that the so-called exploration volume, V(t), which is the volume that a reactant visits in a given period of time, increases (at least) linearly with time. As soon as any of these idealized conditions is relaxed, the equation above has to be modified. For instance, without perfect stirring,

k becomes a time-dependent quantity, that is, k(t), which has a monotonically descending dependence on time2.

To understand this better, let’s describe the distinction between diffusion in one-dimensional and three-dimensional (3D) exploration spaces with the following analogies: (1) ‘the drunk in an alley always returns to the bar’; (2) ‘the drunk space pilot hardly ever manages to return to the bar planet’. Implicit in the above analogies is that the drunk performs a diffusive (random) walk along the alley, and likewise, the pilot randomly changes the direction of flight, that is, performs a random walk in 3D space.

Mathematics teaches us that, in one dimension, V(t) increases (asymptotically) as t 1/2 (even if the alley is infinitely long),

whereas in three dimensions it increases linearly with t. Two equivalent ways of looking at that3 are: in one dimension (even if infinitely long, and in infinite time) the probability of the drunk (random walker) returning to the bar (origin) is unity, and thus his ‘escape probability’ is zero; however, the probability of the drunk space pilot returning to the bar planet is far from unity, for the drunk pilot manoeuvres in three dimensions, and thus his escape probability is finite. The above considerations distinguish

REACtION kINEtICS

Catalysis without a catalystCan two identical reactors with the same concentrations, under identical physical conditions, have reaction rates that differ by a factor of a thousand? A study now shows that, although not true in uncrowded environments, a reactant’s starting point makes a large difference to reaction kinetics in identically crowded systems, such as cellular nuclei.

Raoul kopelman

nchem_N&V_JUN10.indd 430 10/5/10 12:03:47


mailto:[email protected]


news & views

between diffusion in ‘compact’ exploration spaces with zero escape probability and a sublinear V(t), and in non-compact exploration spaces with finite escape probability and a linear V(t). As most fractal topologies (even if embedded in three dimensions) are compact, the resulting non-classical reaction kinetics, with sublinear exploration space, V(t), and time-dependent reaction coefficient, k(t), have been termed “fractal-like reaction kinetics”2. Although the above analogy considers one reactant exploring a volume, when we consider a distribution it turns out that fractal-like reaction kinetics are also accompanied by non-random reactant distributions in space, even if the initial distribution was random. Bénichou and co-workers look at how such a non-random distribution of reactants influences the kinetics of a reaction in confined, compact reaction spaces.

Historically, the early demonstrations of non-classical reaction kinetics were those of photophysical reactions in confined and tortuous domains. Such systems include exciton trapping in disordered molecular crystals4,5, or biological photosynthetic units6, and photochemical reactions in confined domains with co-localization of

photoexcited molecules, that is, a non-random initial distribution owing to laser speckles (a laser optics phenomena that causes a non-random distribution of photons at the target)7.

Specifically, the situation that Bénichou et al.1 study is that of a spatially confined reaction space with labyrinthine topology, and a non-random (quasi-intelligently designed) reactant distribution. Their contribution is to give a unified theoretical treatment, with universal results, using the first passage time (FPT) approach — the FPT is the time it takes a diffusing molecule to reach a target with the implicit assumption that reaching the target means reacting with it. The approach taken by Bénichou and colleagues allows them to derive the full distribution of the FPT, rather than just the mean, and this is crucial for quantifying the reaction kinetics. Although the mean gives satisfactory results for non-compact exploration, it does not for the compact case.

The main questions posed are: how does the FPT distribution depend on the volume of the confining domain and the initial position of the diffusing molecule, and are these factors important? If so,

could they potentially be used to control the kinetics?

The general derivation uses a random walker in a fractal medium with a fractal dimension, and the dynamics characterized by the dimension of the walk3. The results differ significantly depending on the relative size of these dimensions. When the fractal dimension is larger — resulting in non-compact exploration and an exploration volume that grows linearly in time — it is the mean of the FPT that characterizes the kinetics, as in the simple case of regular 3D diffusion. The result is that the initial position of the reactants has little effect on the kinetics (except in recombination reactions, which are not characteristic of biological situations and are not addressed). For the opposite case of compact exploration however — with exploration volume growing sublinearly in time — the full distribution of the FPT is required and not just the mean; the result is that the kinetics depend strongly on the original distance between the initial starting position and the target. It is the real-estate principle of ‘location, location, location’ that is key.

Bénichou and colleagues use the term geometrically controlled kinetics to describe such situations and it is typical of certain important reactions inside biological cells and nuclei, as is shown schematically in Fig. 1. To illustrate the power of such geometrically localized reactions, they discuss the specific situations of transcription kinetics in cases of gene co-localization, showing that a 100 nm co-localization in the nucleus may speed up the transcription kinetics by three orders of magnitude. These results demonstrate the importance of correctly treating such chemical reactions, rather than using the classical reaction-kinetics approach. One also wonders if evolution has in fact used the idea of geometrically controlling the reactants to speed things up. ❐

Raoul Kopelman is in the Department of Chemistry, University of Michigan, 930 N. University, Ann Arbor, Michigan 48109-1055, USA. e-mail: [email protected]

References1. Bénichou, O., Chevalier, C., Klafter, J., Meyer, B. & Voituriez, R.

Nature Chem. 2, 472–477 (2010).2. Kopelman, R. Science 241, 1620–1626 (1988).3. Ben-Avraham, D. & Havlin, S. Diffusion and Reactions in Fractals

and Disordered Systems (Cambridge Univ. Press, 2000).4. Kopelman, R. in Radiationless Processes in Molecules and

Condensed Phases Vol. 15 (ed. Fong, F. K.) 297–346 (Topics in Applied Physics, Springer-Verlag, 1976).

5. Parson, R. P. & Kopelman, R. Chem. Phys. Lett. 87, 528–532 (1982).

6. Kopelman, R. J. Phys. Chem. 80, 2191–2195 (1976).7. Monson, E. & Kopelman, R. Phys. Rev. Lett. 85, 666–669 (2000).

S1

S2

T

Figure 1 | A random race through crowded space: here reactant 1 starts at site S1 and reactant 2 at S2. Which one will get to the target T first? Both travel equally fast, and randomly change direction equally often. As may be intuitively expected here, the closer reactant is much more likely to win the race because of the shorter random path. Had there been no obstacles, however, both reactants would be equally likely to win the race. The obstacles lead to apparent catalysis in the kinetics of reactant 2, and this demonstrates the concept of geometry controlled kinetics. To apply this picture to biology, consider the elliptical area to be a cell nucleus, the reactants to be transcription factors, and the target to be a gene. Has nature ordered all transcription factors to originate in close lying locations such as site 2, and not in farther ones like site 1, so as to speed up this basic reaction of life? Image reproduced from ref. 1.

nchem_N&V_JUN10.indd 431 10/5/10 12:03:48




news & views

Many enzymes rely on iron porphyrin ‘haem’ centres (Fig. 1a) to carry out important tasks in biology,

in particular using their ability to bind small molecules such as O2 and water. For example, the metalloproteins haemoglobin and myoglobin, found in red blood cells and muscle tissue, respectively, use haem centres to reversibly bind O2 through their Fe(ii) ions and act as oxygen carriers. In contrast, in cytochrome b5 proteins, their role is to transport electrons through changes in the oxidation state of iron. Both of these abilities — oxygen binding and redox changes — are used by the cytochrome P450 family to catalyse oxygen-transfer reactions, ranging from the enantioselective epoxidation of olefins to the conversion of alkanes to alcohols (depending on the enzyme).

What if synthetic iron porphyrins could be coaxed to catalyse valuable chemical transformations in much the same fashion as cytochromes, but without the complex protein machinery? Over the years this idea has motivated some terrific biomimetic chemistry1,2 and, at the same time, revealed some practical difficulties. Most notably, iron porphyrins were found to easily form oxo-bridged dimers, which renders them catalytically inactive. Furthermore, there are challenges in positioning the nitrogen- or sulfur-based ligands — needed for catalyst activation — at one of the two available axial iron coordination sites while avoiding coordination at the second.

Collman and co-workers showed several years ago that both problems, in principle, could be overcome through sophisticated functionalization of the porphyrins (for example by using organic ‘picket fences’ to prevent dimerization, or by overarching straps to control ligation)2. Unfortunately, these derivatives are difficult to synthesize, making them impractical for routine oxidative catalysis, either in academic labs or in industry.

Writing in Science, McKeown and co-workers3 have now proposed an interesting alternative solution: build crystalline nanoporous arrays. Crystallinity ensures that the haem-like active sites are precisely positioned. Regular spacing (nanoporosity)

ensures that they are suitably isolated from each other, and at the same time provides channels to transport reactants and products to and from the sites. Furthermore, with practical applications in mind, McKeown and colleagues chose to use phthalocyanines as building blocks. These macrocycles are closely related to porphyrins (Fig. 1a,b) but are much easier to synthesize, and metallated

phthalocyanines behave catalytically much like their metallated porphyrin cousins.

At first, phthalocyanines seem an odd choice of components to build nanoporous arrays. They are large planar molecules that easily stack, forming poorly soluble aggregates. In previous studies, however, McKeown and co-workers had addressed this problem and arranged zinc phthalocyanines

CRYStAL ENGINEERING

towards artificial enzymesDespite knowing that the active centres of many metalloprotein enzymes are iron porphyrin ‘haem’ complexes, chemists find them difficult to imitate. Now, the assembly of haem-like centres into a crystalline, stable, nanoporous array shows promise for biomimetic catalysis.

Joseph t. Hupp

a b

c

Bipyridine linkers

N

NN

N

N

NN N

O O

O

O

O O

O

O

iPr

iPr

iPr

iPr

iPr iPr

iPriPr

iPr

iPriPr

iPr

iPriPr

iPr iPr

FeN

N

N

N

Fe

NN

Figure 1 | Schematic representation of the nanoporous crystalline haem-like array. a, The structure of iron porphyrin. b, Structure of a non-aggregating iron phthalocyanine. c, 3D nanoporous crystalline arrays of iron phthalocyanine molecules connected by bipyridine linkers.

nchem_N&V_JUN10.indd 432 10/5/10 12:03:49



news & views

into regular arrays first for phthalocyanines4 and subsequently for azaphthalocyanines5. This was achieved by introducing peripheral phenyl ethers — eight of them per molecule — as side groups that lie out of plane and prevent stacking. The arrays consist of stacked cubes, each filled with solvent molecules and delimited by phthalocyanines acting as faces. The corners of the cubes contain small openings — about 4 Å in diameter — forming small solvent-filled cavities that interconnect with the remarkably large solvent-filled voids defined by each cube — 8 nm3 each.

Why do the arrays form? They are examples of clathrates — molecular crystals that trap a second type of molecule — in this case, solvent. In these zinc phthalocyanine arrays, the clathrates were held together solely by van der Waals forces. This cube geometry, discovered serendipitously, obviously represented a minimum energy structure. McKeown and colleagues had shown that the solvent (mainly methanol) present initially could be exchanged for a variety of other solvents (including water) without loss of crystallinity. Nevertheless, the presence of solvent was essential, and removing it — as opposed to exchanging it — caused a loss of porosity, a loss of crystallinity, and therefore a loss of information about the nanoscale structure.

Now, McKeown et al. have replaced Zn(ii) by Fe(ii) and neatly resolved the array stability problem by introducing molecular tie bars (linear ligands such as 4,4′-bipyridine). The tie bars span the cavities between neighbouring cubes (separated by roughly 11 Å) and coordinate to the iron centres of two adjacent faces, keeping them linked together (Fig. 1c). With these tie bars in place, the assemblies retain their structure on removal of solvent. This is directly characterized by single-crystal X-ray crystallographic structural measurements, and indirectly by the N2 adsorption measurements of the array’s microporosity and internal surface area.

McKeown and co-workers point out that the tie bars could, in principle, also play the role of activating ligands in catalysis applications. The X-ray crystallographic measurements (carried out under a nitrogen atmosphere) show that the second iron axial coordination site is occupied not by a second tie bar, but instead by N2. It is therefore reasonable to assume that exposure to air would lead to coordination of O2.

With the synthesis of these remarkable haem-like arrays demonstrated, the next step will clearly be to assess their catalytic competency. It will be interesting to see to what extent the arrays can mimic the activity

of various types of cytochrome P450. Can the catalytic activity and reaction selectivity be modulated by changing the identity of the tie bars? Will the catalytic chemistry be limited to comparatively small substrates, given the small portals between cubes? Or will the portals prove to be sufficiently flexible to allow large molecules to reach reactive metal sites? Will the site-isolation inherent to the array structure lead to unusually high turnover numbers and exceptional catalyst longevity? And, will the solid-state nature of the catalytic arrays allow for recovery and reuse? Finally, will other kinds of catalytic chemistry be accessible based on available ruthenium and cobalt analogues3 of the iron-containing nanoporous structures? ❐

Joseph T. Hupp is in the Department of Chemistry, Northwestern University, Evanston, Illinois 60208, USA, and in the Materials Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA. e-mail: [email protected]

References1. Collman, J. P., Boulatov, R., Sunderland, C. J. & Fu, L. Chem. Rev.

104, 561–588 (2004).2. Groves, J. T. Proc. Natl Acad. Sci. USA 100, 3569–3574 (2003).3. Bezzu, C. G., Helliwell, M., Warren, J. E., Allan, D. R. &

McKeown, N. B. Science 327, 1627–1630 (2010).4. McKeown, N. B. et al. Angew. Chem. Int. Ed. 44, 7546–7549 (2005).5. Makhseed, S. et al. Chem. Eur. J. 14, 4810–4815 (2008).

Although laboratory applications of processes centred on the principle of Darwinian evolution are now

commonplace for nucleic acid and peptide biopolymers, it is only recently that researchers have brought molecular evolution to bear on organic compounds. One example of this is dynamic combinatorial chemistry (or dynamic covalent chemistry, both abbreviated DCC), a conceptual framework whereby collections of molecules (dynamic combinatorial libraries, or DCLs) are generated under equilibrating conditions and allowed to undergo ‘evolution’ as a function of some thermodynamic selection pressure1. In many experiments, the selection pressure is binding affinity for a small-molecule ‘guest’ or a biopolymer ‘host’ of biomedical relevance.

The exchange reactions used to provide mixture equilibration and ‘evolution’ in DCC are in some ways the antithesis of modern synthetic chemistry. Ideally, they are completely reversible under certain conditions, but irreversible under others, and chemoselective (the reaction only occurs between specific functional groups) but otherwise not influenced by compound structure. If the goal of a DCC experiment is the selection of compounds that bind to a biopolymer target such as an oligonucleotide or protein, an even more stringent set of constraints must be considered, because the exchange reaction has to operate under conditions compatible with the biological target. As one might imagine, the list of transformations satisfying all of these criteria

is limited. This represents a problem for the field, as constraints on reaction diversity concomitantly limit the structural diversity accessible to DCLs.

In arguably the earliest example of a biologically targeted DCL, Venton and colleagues showed that a biological catalyst (the enzyme thermolysin) could be employed to equilibrate peptide libraries targeting antibody binding2. Much like chemical catalysis has transformed modern synthetic chemistry, we are now beginning to learn that non-biological catalysts can also extend the range of chemistry accessible to DCC, and make it applicable to problems in biomolecule-targeted library selection. Now, writing in Nature Chemistry, Greaney, Campopiano and co-workers show3 that

DYNAMIC COVALENt CHEMIStRY

Catalysing dynamic librariesThe composition of dynamic small-molecule libraries can be biased by the addition of a target compound — such as a protein — that binds selectively to one of the components in the mixture. The chemistry of the library must, however, be compatible with the target and it has now been shown that aniline-catalysed exchange of acylhydrazones fits the bill.

Benjamin L. Miller

nchem_N&V_JUN10.indd 433 10/5/10 12:03:49




news & views

a catalysed version of acylhydrazone formation enables formation of a DCL under biopolymer-friendly conditions, thereby providing a new strategy for the generation and screening of protein-targeted DCLs.

Introduced in the context of DCC by the Sanders group4 and studied by them extensively, acylhydrazone formation has many of the characteristics of an ideal exchange reaction. The aldehyde and hydrazide components are easily obtainable, the reaction is readily reversible and chemoselective, and the exchange process can be halted by simply changing the pH of the solution. Unfortunately, the pH range at which acylhydrazone exchange occurs rapidly is inhospitable to most protein targets. Pointing the way towards a solution to this problem, Dawson and colleagues reported in 2006 that aniline could be employed as a convenient catalyst for the reaction, with 10 mM aniline providing as much as a 70-fold rate enhancement on the reaction of a 1 mM peptide substrate in a pH 5.7 buffer5. This work paved the way for further studies by the Dawson group on the application of the reaction to the preparation of exchangeable chemical linkers for protein enrichment6, as well as to the library studies described here.

Greaney and colleagues began their efforts by generating a small library of acylhydrazones from one aldehyde, chosen based on its structural resemblance to a known substrate of glutathione-S-transferase (GST), and 10 hydrazides. This library required five days to reach equilibrium in the absence of a catalyst. In contrast, incorporating aniline into the mixture accelerated the system dramatically, providing a fully equilibrated library in six hours in a pH 6.2 buffer. Control experiments (varying starting conditions) verified that the system

had reached a ‘true’ equilibrium. With these data in hand, it was demonstrated that aniline-catalysed library equilibration in the presence of two different GST enzymes (hGST P1-1, a human isoform of interest as a potential drug target in reducing drug resistance to chemotherapy, and SjGST, an isoform from the helminth worm) resulted in amplification of isoform-selective binders (Fig. 1).

After halting equilibration by raising the pH to 8.0, HPLC analysis of the mixture showed that hGST P1-1 and SjGST had selected distinctly different library members. In contrast, the control protein — bovine serum albumin — did not alter the composition of the library in comparison with that obtained for the protein-free system. Moreover, incorporation of a glutathione moiety into the library enhanced the solubility of its members, and led to the selection of compounds with significant binding ability. Interestingly, a catalytically inactive SjGST mutant selected the same library member as its active counterpart, confirming that the catalytic activity of the enzyme was not critical to the selection process. Subsequent binding studies confirmed that the selected compounds were indeed the most potent members of the library.

Although the libraries examined in this study are of modest size, and thus could have been screened using ‘standard’ parallel techniques, the dual strengths of DCC are that it (1) enables researchers to rapidly generate and screen large numbers of chemical entities with a minimum of resources (or effort), and (2) as one moves from simple dimeric compounds to oligomers and macrocycles one can begin to identify ‘surprising’ structures that would otherwise be difficult to access in the laboratory. We can anticipate that catalysed

acylhydrazone exchange will now find utility in both these areas. It is worth noting that even with a relatively small library it was possible, nonetheless, to identify compounds with significant selectivity for specific GST isoforms.

The success of Greaney and co-workers in applying Dawson’s simple organic catalyst to acylhydrazone equilibration for non-peptide library evolution targeting GST will hopefully provide encouragement for researchers engaged in the search for catalysts for other reactions, thus expanding the chemical repertoire of DCC. Indeed, efforts are underway to develop catalysts suitable for amide formation7, although these are not yet in a form suitable for libraries targeting biomolecules. Jeremy Knowles famously used the title of a paper to state that enzyme catalysis was “not different, just better”8; we can look forward to an expanded range of chemical catalysts enabling DCC to fulfil its promise of being not just different but also better. ❐

Benjamin L. Miller is at the University of Rochester Medical Center, Rochester, New York 14642, USA. e-mail: [email protected]

References1. Miller, B. L. (ed.) Dynamic Combinatorial Chemistry (Wiley, 2010).2. Swann, P. G. et al. Biopolymers 40, 617–625 (1996).3. Bhat, V. T. et al. Nature Chem. 2, 490–497 (2010).4. Furlan, R. L. E., Ng, Y. F., Otto, S. & Sanders, J. K. M.

J. Am. Chem. Soc. 123, 8876–8877 (2001).5. Dirksen, A., Dirksen, S., Hackeng, T. M. & Dawson, P. E.

J. Am. Chem. Soc. 128, 15602–15603 (2006).6. Dirksen, A., Yegneswaran, S. & Dawson, P. E.

Angew. Chem. Int. Ed. 49, 2023–2027 (2010).7. Stephenson, N. A., Zhu, J., Gellman, S. H. & Stahl, S. S.

J. Am. Chem. Soc. 131, 10003–10008 (2009).8. Knowles, J. R. Nature 350, 121–124 (1991).

Published online: 16 May 2010

O

NO2

ClNH2

SjGST

hGSTP1-1

N

NO2

Cl

HN

O

S

Isoform-selective amplification

N

NO2

Cl

HN

O

NO2

Cl

N

O2N

Cl

NH

R2

N

NO2Cl

NH

R3

N

NO2

Cl

HN

Rn

R2

HN

H2N

R3

HN

H2N

Rn

HN

H2N

Catalyst

Dynamic combinatorial library

BSANo amplification of any library member

N

HN

R1

Figure 1 | An aniline-catalysed acylhydrazone dynamic cominatorial library and the influence of protein targets on its composition. Two different isoforms of GST lead to amplification of different members of the dynamic library, whereas bovine serum albumin has no effect.

nchem_N&V_JUN10.indd 434 10/5/10 12:03:49




news & views

Metal–organic frameworks (MOFs) and porous coordination polymers are promising materials for a variety

of applications, including sorption, selection, catalysis, sensing or microelectronics, and the efforts of many research groups worldwide have produced numerous extended systems of ever growing structural complexity1. A mathematical (topological) description of the structure of these complex systems is necessary to achieve a correct classification and thus help to clarify the fundamental relationship between structure and properties. Indeed, new topological configurations are continuously discovered, including mechanically linked arrays derived from the simplest two-ring link, the Hopf link (Fig. 1). Hopf links are well known to the chemistry community as they appear in catenanes — compounds that consist of molecular rings held together by mechanical bonds2.

A recent review3 suggested the extension of catenation to create infinite objects in one, two or even three dimensions by the formation of several Hopf links. At the time, however, no real examples had actually been observed. It is only recently that a one-dimensional (1D) [n]-catenane (Fig. 1a) has been observed in the realm of coordination polymers4, and no examples extending into two or three dimensions have appeared until now. Writing in this issue of Nature Chemistry, Can-Zhong Lu and co-workers5 describe the exciting discovery of a coordination material in which the catenation of adamantane-like molecular cages — considered to be a zero-dimensional (0D) building block — extends in three directions to form a three-dimensional (3D) polycatenated architecture. They also show that two such identical extended structures interpenetrate one another in the final structure (Fig. 1b). Indeed, whereas interpenetration is frequently found in coordination networks, it is exceptional to observe structures based on polycatenated molecular basic motifs3,6.

The unique polycatenated and interpenetrated array described by Lu and co-workers is obtained thanks to the templating effect of the polyoxometalate counter-anions employed in the synthesis. The very existence of the structure, however, suggests that the use of mechanically interlinked cages as

building blocks — instead of the traditional use of metal ions as nodes coordinatively bound to polydentate organic linkers — represents a new synthetic strategy to obtain interesting MOFs.

A few words on the topology of the entanglements are necessary here. The Hopf link is the basic unit defining an inextricable entanglement — the only way to separate the links is by breaking one of the rings.

Complexity arises when one considers the dimensionality of the starting building blocks versus that of the resulting final architecture. Polycatenation is defined by an increase in the dimensionality of the final architecture over the dimensionality of the building blocks. On the contrary, in interpenetration there is no change in dimensionality. This distinction is therefore truly topological rather than just semantic7 (Fig. 2).

tOPOLOGICAL CRYStAL CHEMIStRY

Polycatenation weaves a 3D webMechanical linking of small cage structures leads to a type of metal–organic framework with an architecture topologically distinct from those constructed so far.

Davide M. Proserpio

Cd(imidazolate)2INTERPENETRATION

Hopf link[2]-catenane

0D + 0D 1D [n]-catenane

0D + 0D 3D pcuPOLYCATENATION

0D

3D + 3D 3D pcu twofoldINTERPENETRATION

3D + 3D 3D dia twofold

+

a

b

c

Figure 1 | Interpenetrated and polycatenated arrays. a, The Hopf link (left) is the basic unit of inextricable entanglement. Multiple Hopf links (right) result in an [n]-catenane and an increase in dimensionality. b, Polycatenation of 0D cages results in a 3D octahedral array of cages. In the final structure, two of these octahedral arrays (shown in blue and red) are interpenetrated. c, Cd(imidazolate)2 is an example of the type of MOF more commonly observed in which there is only interpenetration, and precisely twofold dia. This MOF is formed between cadmium ions acting as nodes and the organic imidazolate ligands acting as bidentate linkers.

nchem_N&V_JUN10.indd 435 10/5/10 12:03:54



news & views

By way of example, [n]-catenanes are the simplest example of polycatenation (0D + 0D → 1D), whereas the structure of Cd(imidazolate)2 is an example of a twofold interpenetrated6 diamondoid (called dia) network (3D + 3D → 3D) (see Figs 1 and 2). A further criterion for polycatenation is that, unlike interpenetration, each distinct building block is never interlaced with all the others in the array. So, in structures such as Cd(imidazolate)2, we can define a degree of interpenetration, because there is always a finite number of interpenetrated components. This is impossible for polycatenated arrays where there are an infinite number of entangled components, as shown in Fig. 2 for hexagonal layers3.

Starting from simple rings, it is perfectly possible to imagine structures more complex than the [n]-catenane chains — in which each

ring is linked to just two adjacent rings — and obtain something two-dimensional (2D) akin to medieval chain-mail2,3. If the finite building block is a 3D cage it is possible to extend the catenation in three directions, as in the compound reported by Lu and co-workers.

Here, each cage links to six other equivalent cages, giving rise to a type of octahedral six-coordination that is the basic node of the primitive cubic net (called pcu according to the modern nomenclature for nets)8 (Fig. 1). It is important to keep in mind that — as is usual in solid-state chemistry — such descriptions depend on the types of interaction that are being considered. If we take into account only the strongest interactions, such as covalent bonds, we describe the polycatenated and interpenetrated array as above. In general, topological descriptions may consider

supramolecular interactions — in particular hydrogen bonding — as important building factors of solid-state architectures. Here, if we also consider secondary weak interactions, then the independent motifs become connected and a different entanglement arises.

How can such complex topologies be detected? In the past, the crystallographer needed great experience in 3D visualization and the help of ball-and-stick models to explain these complicated arrangements. Thankfully, modern tailored software allows us to compute, detect and classify entanglements in periodic structures9,10. Such computer methods give us much better design capabilities for new extended architectures.

The results of Lu and co-workers demonstrate that, provided it is theoretically possible, almost any — even bizarre — entanglement can be realized in nature. First, however, it is important to thoroughly explore the relationships between the intricate sub-architectures to identify possible pathways for their synthesis. ❐

Davide M. Proserpio is in the Department of Structural Chemistry DCSSI, Università degli Studi di Milano, Via Venezian 21, 20133 Milano, Italy. e-mail: [email protected]

References1. Long, J. R. & Yaghi, O. M. Chem. Soc. Rev. 38, 1213–1214 (2009).2. Fang, L. et al. Chem. Soc. Rev. 39, 17–29 (2010).3. Carlucci, L., Ciani, G. & Proserpio, D. M. Coord. Chem. Rev.

246, 247–289 (2003).4. Jin, C. M., Lu, H., Wu, L. Y. & Huang, J. Chem. Commun.

5039–5041 (2006).5. Kuang, X. et al. Nature Chem. 2, 461–465 (2010).6. Blatov, V. A., Carlucci, L., Ciani, G. & Proserpio, D. M.

CrystEngComm 6, 377–395 (2004).7. Francl, M. Nature Chem. 1, 334–335 (2009).8. O’Keeffe, M., Peskov, M. A., Ramsden, S. J. & Yaghi, O. M.

Acc. Chem. Res. 30, 1782–1789 (2008).9. Blatov, V. A. IUCr CompComm Newsletter 7, 4–38 (2006).10. www.topos.ssu.samara.ru/starting.html

2D + 2D 2D

3D

3D

Polycatenated inclinedPolycatenated parallel

Interpenetrated threefold

Figure 2 | Topologically distinct entanglements of hexagonal layers. Two different modes of polycatenation are shown, which both result in an increase in dimensionality, versus interpenetration in which the dimensionality remains the same.

‘tug of War’ is a game that tests the strength of two teams pulling on a rope in opposite directions. If it is

being played on a molecular scale, the game is known as polymer mechanochemistry, and the interest is not so much in the muscle power of the contestants as in the strength of the rope and its fate under stress. Writing in the Journal of the American Chemical

Society, Bielawski and co-workers1 have brought this increasingly popular game to a higher level of sophistication by showing that pulling on molecular ropes may be used to promote racemization of chiral molecules.

In the past, chemists have focused almost exclusively on the use of light or heat to bring about chemical reactions. Promoting reactions with force is an alternative that is

much less popular, although the principle is well established, particularly when the transformation is simply a matter of breaking bonds in the main chain of a polymer — mastication of rubber to reduce its molecular weight is an example of a bond breaking reaction that is widely used in industry.

Recently there has been a flurry of reports that show how mechanical forces applied

MECHANOCHEMIStRY

Forcing a molecule’s handUltrasound can be used to control molecular processes as delicate as rotation around a single carbon–carbon bond.

S. karthikeyan and Rint P. Sijbesma

nchem_N&V_JUN10.indd 436 10/5/10 12:03:56



http://www.topos.ssu.samara.ru/starting.html


news & views

to polymer chains can be used for far more subtle manipulation of chemical bonds. Polymer mechanochemistry has become a burgeoning field that uses mechanical forces to bias reaction pathways2, to change the colour of materials3, to trigger multiple reactions in a single polymer chain4, and to selectively break weak coordinate bonds to activate dormant catalysts5. Recent theoretical work has helped to create a much better understanding of mechanical force as a unique stimulus for chemical reactions3,6. Chemists have begun to work in this area with renewed effort because of its potential applications in self-healing polymers, molecular strain gauges and controlled drug release, and because they recognize the possibility that it will provide greater understanding of the mechanochemical transduction mechanisms in biological systems.

Bielawski and co-workers have now elegantly shown that mechanical force can be used to surmount the high energy barrier associated with rotation about the C–C single bond in binaphthyl derivatives and thus convert one mirror image of this molecule to the other (Fig. 1). These mirror-image molecules are called atropisomers — stereoisomers resulting from the hindered rotation around a C–C single bond. The barrier for C–C bond rotation in binaphthyl derivatives is high (~30 kcal mol–1), which makes their racemization at ambient temperatures extremely slow. In fact, even at 195 °C, the half-life (t1/2) of the parent binaphthol molecule is as long as 4.5 hours (ref. 7).

How did the team, led by Texas-based chemist Chris Bielawski, manage to isomerize these stable chiral molecules? One of the most efficient ways to exert pulling forces on a molecule is to use the intense flow fields around collapsing cavitation bubbles in sonicated solutions. For effective transfer of the mechanical force to the reactive unit, it is essential to functionalize it with polymer chains. Therefore, Bielawski and co-workers appended poly(methyl acrylate) chains with total molecular weights between 10 and 100 kDa to binaphthyl derivatives. When the polymeric (S)-binaphthyl derivative (with Mn = 98.7 kDa) was sonicated in acetonitrile solution, circular dichroism spectroscopy showed that more than 95% of the derivative had racemized after 24 hours. As a measure of the relative efficiency of the mechanical force, heating of the same (S)-binaphthyl-functionalized polymer was also investigated, but heating at 250 °C for 72 hours gave no change in the intensity of the circular dichroism signal.

The mechanical action of ultrasonication is always accompanied by thermal effects, and the heating effect of collapsing cavitation

bubbles can be strong. Therefore, control experiments that establish the contribution of heating are important. To this end, ultrasonication experiments were performed on binaphthyl derivatives without the attached polymer chains. In these experiments, no racemization was observed. Further convincing evidence for the mechanochemical origin of the racemization comes from the molecular-weight dependence of the racemization rate observed for binaphthyl derivatives attached to polymer chains with Mn varying from 10 to 100 kDa. The ultrasound-induced isomerization shows a limiting molecular weight (between 25 and 50 kDa), below which no change was observed.

The present work raises interesting questions concerning the pathway by which the stereoisomers are interconverted. The two stereoisomers are stable because the steric bulk of the substituents adjacent to the biaryl bond prevents free rotation. Isomerization of the binaphthyl may therefore proceed by passage of 2,8′- and 2′,8 substituents (the so-called anti route) or 2,2′- and 8,8′ substituents (the so-called syn route)7,8 (Fig. 1). Detailed theoretical studies on thermal racemization pathways of 1,1′-binaphthyl analogues suggest that the most favourable pathway proceeds through the centrosymmetric anti transition state, but this is favoured over the syn route by only 4 kcal mol–1. Does the mechanically facilitated reaction follow the preferred thermal reaction pathway, as the authors propose? Investigating this question may reveal unexpected complexities. It is, for instance, imaginable that the favoured pathway has several transition states and includes mechanically as well as thermally surmounted barriers. Given the fact that

applied mechanical forces change the potential energy surface in a direction-dependent manner, the question can be answered conclusively only with detailed calculations using methods specifically developed to study mechanochemical reactions2,6.

The importance of the work lies in showing the path to selectivity to others who consider entering the fascinating field of mechanochemistry. Although polymer scission using ultrasound has been used for many decades, the use of mechanical forces to perform useful transformations and applications is still in its infancy. In the near future, efforts to use mechanochemistry productively will undoubtedly increase, and selectivity will be the focus of attention. Bielawski and co-workers have shown that the seemingly untamed force of ultrasound can be used to control a process as simple and fundamental as rotation about a C–C single bond. ❐

S. Karthikeyan and Rint P. Sijbesma are in the Laboratory of Macromolecular and Organic Chemistry, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands. e-mail: [email protected]; [email protected]

References1. Wiggins, K. M. et al. J. Am. Chem. Soc. 132, 3256–3257 (2010).2. Hickenboth, C. R. et al. Nature 446, 423–427 (2007).3. Davis, D. A. et al. Nature 459, 68–72 (2009).4. Lenhardt, J. M., Black, A. L. & Craig, S. L. J. Am. Chem. Soc.

131, 10818–10819 (2009).5. Piermattei, A., Karthikeyan, S. & Sijbesma, R. P. Nature Chem.

1, 133–137 (2009).6. Ribas-Arino, J., Shiga, M. & Marx, D. Angew. Chem. Int. Ed.

48, 4190–4193 (2009).7. Meca, L., Reha, D. & Havlas, Z. J. Org. Chem. 68, 5677–5680 (2003).8. Kranz, M., Clark, T. & von Rague Schleyer, P. J. Org. Chem.

58, 3317–3325 (1993).

O

O

O

O(S)-Binaphthyl polymer

Syn isomer

Anti isomer

(R)-Binaphthyl polymer

1

23

456

78 8'

7'

6'5' 4'

3'

2'

1'

Planar transitionstructures

O

O

O

O

Force

Force

Figure 1 | Racemization of binaphthyl-based polymers with ultrasound. Starting with the (S)-configured binaphthyl polymer, ultrasonic irradiation leads to rapid racemization, which may occur through one of the planar transition structures shown.

nchem_N&V_JUN10.indd 437 10/5/10 12:03:56





news & views

Proteins are typically preserved in the form of dehydrated powders to avoid their degradation or the growth of microbes that can occur in solution. Such powders are typically obtained by freeze-drying, but some proteins can be damaged during the process. Now, using an organic solvent (decanol) as drying agent, David Needham and co-workers at Duke University in the USA have successfully dehydrated a protein (lysozyme) to form beads of controllable size through a simple glassification procedure (pictured; Biophys. J. 98, 1075–1084; 2010)

In aqueous solutions, biological molecules are closely surrounded by hydration water molecules that are more difficult to remove than those of the bulk solution, and which keep the molecules apart. When small droplets of an aqueous lysozyme solution were

added into a decanol solution, all the bulk and hydration water molecules dissolved into the organic solvent within minutes. This process was too fast for the protein to crystallize and instead it arranged into microbeads with a glassy, amorphous

structure. On re-hydration, the lysozyme recovered most of its activity.

Using a packing model, the researchers determined the level of protein hydration, and therefore the separation distance between the lysozyme molecules, from the water activity measured in decanol (its ‘concentration’ in the non-ideal mixture). This means that by adjusting the water activity in the drying solvent, they were able to control the final protein concentration, and thus the size of the resulting glassy beads. This drying method was also faster and cheaper than the freeze-drying process, showing great promise for biological applications.

ANNE PICHON

The original version of this story first appeared on the Research Highlights section of the Nature Chemistry website.

Protein under glassPRESERVAtION PROCESSES

© 2

010

The

Bio

phys

ica

l so

cie

Ty

nchem_N&V_JUN10.indd 438 10/5/10 12:03:56



perspectivePublished online: 16 may 2010 | doi: 10.1038/nchem.654

robust dynamicshexiang deng1, mark a. olson2, J. Fraser stoddart2* and omar m. yaghi1*

Although metal–organic frameworks are extensive in number and have found widespread applications, there remains a need to add complexity to their structures in a controlled manner. It is inevitable that frameworks capable of dynamics will be required. However, as in other extended structures, when they are flexible, they fail. We propose that mechanically interlocked molecules be inserted covalently into the rigid framework backbone such that they are mounted as integrated components, capable of dynamics, without compromising the fidelity of the entire system. We have coined the term ‘robust dynamics’ to describe constructs where the repeated dynamics of one entity does not affect the integrity of any others linked to it. The implication of this concept for dynamic molecules, whose performance has the disadvantages of random motion, is to bring them to a standstill in three-dimensional extended structures and thus significantly enhance their order, and ultimately their coherence and performance.

Stitching molecular building blocks into extended frameworks using strong bonds — reticular chemistry — is one of the most widely investigated areas in chemistry today1. A library of organic

and inorganic building blocks has been used to build a large number of structures, named metal–organic frameworks2 (MOFs). Generally, the MOF construct is based on the principle of linking metal-oxide joints with organic struts as illustrated in Fig. 1. This process has been repeated over and over again in various different chemical contexts to afford extensive classes of porous MOFs with a diversity and mul-tiplicity previously unknown in the realm of artificial materials. The rigidity and directionality of the joints and struts ensure the MOFs’ architectural stability and therefore permanent porosity: both are vitally important for their applications in catalysis, gas storage and separation3. It is these very same features, however, that rob MOFs of their dynamics and give rise to the question: how can we preserve the important characteristics and properties of MOFs, while accessing the dynamics that could provide the key to enhancing their functions?

An obvious strategy is to make flexible frameworks from pli-able struts4. Another strategy is to use multi-interpenetrating

frameworks, wherein one framework shifts with respect to the oth-ers, thereby closing or opening the pores5. Both of these strategies are severely limited because frameworks that flex back-and-forth fail when subjected to further repeated dynamics, and interpen-etrating frameworks are highly dependent on the uncontrollable behaviour of guests that fill their pores. In fact, this problem is not unique to MOFs — it is understood that structural failure is to be expected when the backbones of polymers and other extended structures are subjected to repeated dynamics. How then do we overcome the challenge of introducing dynamics into MOFs and, for that matter, other extended chemical structures, while retain-ing their robustness?

To answer this question, we turn our attention to another field of endeavour that has been progressing at an equally rapid pace of late — that of artificial molecular switches and machines6. In one of their most highly studied manifestations, they are composed (Fig. 2a–d) of circular, and sometimes also linear, components that are linked together mechanically7. Given the use of templation8 in their synthesis, they are capable of elaborate and repeated dynamics,

1Center for Reticular Chemistry, Department of Chemistry and Biochemistry, University of California, Los Angeles, California 90095, USA, 2Center for the Chemistry of Integrated Systems, Department of Chemistry, Northwestern University, Evanston, Illinois 60208, USA. e-mail: [email protected]; [email protected]

≡

90˚

90˚

Framework backbone

≡

Six-way connection

+

Triphenylene unit

Organic strut

‘O-C-O’ Claws

Zn4O

Metal-oxide joint

IRMOF-16

90˚

Figure 1 | A rare view into the construction of metal–organic frameworks (MOFs). Herein the phenylene rings are bonded together by pivot joints to make the struts that link six-way tetra-zinc oxide clusters, likened to ball-joints, to form a MOF-5-type structure, previously named IRMOF-16. The carboxyl units act as ‘claws’ to keep the zinc centres in invariant positions and disallow any major structural perturbations. These features, when combined with the perpendicular orientations of all the claws, provide a glimpse into the reason for the well-known architectural stability of this MOF construct. The phenylene rings (red and black), O–C–O claws (grey and blue), and zinc oxide joint (Zn4O, pink).

nchem_.654_JUN10.indd 439 7/5/10 12:30:47


http://www.nature.com/doifinder/10.1038/nchem.654




perspective NATure cHeMIsTry doi: 10.1038/nchem.654

yet so far, for the most part, in an incoherent manner in solution or in condensed phases9. The problem with their present design is that they lack the rigid backbone that could provide a platform for their strategic and precise placement in two- (2D) or three-dimensional (3D) space so that they can express their dynamics — namely, the coherent switching between their mounted components.

During a switching process, only weak non-covalent bonds get broken and reformed again in a wholly reversible and highly con-trollable fashion7,9. We therefore propose that coupling the dynam-ics of molecular switches and machines with the rigid structures

of MOFs will yield materials that are intrinsically robust and rigid, yet dynamic — a property we term ‘robust dynamics’. The idea is that molecular switches and machines will be incorporated sym-metrically into the struts of MOFs to graft dynamics onto their frameworks. In such materials, the repeated relative motions of the mechanically interlocked components will not affect the robust-ness of the framework backbone because their relative movements do not subject their constituent covalent bonds to undue stress and hence breakage: only non-covalent bonds are being ruptured and remade. In this way, the material’s fidelity and longevity will be overwhelmingly enhanced. It is our opinion that a system capable of robust dynamics must inevitably encompass a rigid framework into which flexible units (Fig. 2a–d) are inserted. The result of this thought process is illustrated in Fig. 2e–h, using the well-known MOF-5-type structure10 and a bistable [2]catenane6,7. Here, we detail some of the underlying principles and thinking that now need to go into the blending of these two types of struc-tural architecture, while emphasising the vast potential inherent in this union.

The process of designing a MOF structure is not unlike how we design and construct macroscopic objects such as bridges and sky-scrapers; we link together girders and junctions of various shapes according to a blueprint using fasteners and rivets. The way the construction is done on the ångström or nanometre scale is to employ the chemical architect’s blueprint, which is a net — that is, a 3D array of points joined together by links, ideally related to each other by symmetry11. There is virtually a limitless number of net topologies of widely varying connectivities: we select the simplest and most symmetrical as feasible targets for synthesis. One such net is the primitive cubic topology. It is composed of two-connectors (struts or girders) that link (riveting) the verti-ces (junctions) leading to six-way connectivity. In our MOF-5-type structure, the intersecting points are joints with octahedral geometry. To make a MOF based on this topology, we take the phenylene ring of 1,4-benzenedicarboxylate as a two-way connec-tor, and the zinc cluster as the six-coordinated unit and together they impose (Fig. 1) a primitive cubic structure on the MOF. Using related links, the same strategy can be applied to produce12 yet more MOFs based on the same net (isoreticular) with predeter-mined pore sizes and shapes.

Ideally, the joints are rigid entities with well-defined geometries, which impart directionality and control over the resulting struc-ture. By predetermining the geometry of the joint, one dictates the connectivity of the underlying net. The key to making rigid joints is to choose clusters that have an intrinsically 3D structure that is entirely composed of common-sized rings. In the case of these MOFs, the Zn4O(CO2)6 joints may be viewed as being made up of six 6-membered rings of Zn2–O–CO2 composition that are sharing edges and are perpendicular to adjacent rings while also positioned (Fig. 1) opposite to other such rings. Indeed, when one considers this arrangement and the fact that the rings act as O–C–O ‘claws’, which hold the zinc atoms in position, then it becomes apparent why this construction does not shear. Once reaction conditions are identified for forming a joint, a small number of high-symmetry nets can be targeted, employing struts of the appropriate shapes and symmetries. We note that the use of joints made up of one metal atom, where no common rings exist, leads to flexing motions of large amplitude that inevitably destroy the framework, and the variable connectivity of such metal atoms preclude making frameworks by design. By contrast, metal-clus-ter joints ensure an invariant connectivity and therefore allow the design of frameworks, whereas the presence of multiple six-mem-bered rings within such joints imparts that crucial but slight flex-ing at those joints ultimately, leading to robust structures of high fidelity. With all these considerations in mind, the stage is now set for introducing bistable catenanes and rotaxanes as two-way

a

b

c

d

e

f

g

h

Struts MOFs

Figure 2 | Illustrative examples of how elaborate units can be mounted onto the organic struts to introduce complexity and dynamics into MOFs. The struts (a to d) are linked by tetra-zinc oxide centres to produce MOFs (e to h) in which the metrics of the struts and their functionality can be varied to give highly ordered 3D systems with controlled complexity and of vast openness such as to allow access to, and dynamics at, the mounted units. The polyether oxygen (pink), carbon (black) and phenylene (red) form the crown ether receptor/template (a and e), which forms pseudorotaxanes with Paraquat dication (b and f), degenerate catenanes with a cyclophane (blue) containing two Paraquat units linked by phenylene rings interlocking the crown ether (c and g), and a cyclophane (blue) containing a dimethyl diazapyrenium unit (purple) in addition to a Paraquat unit linked by phenylene rings interlocking the crown ether (d and h).

nchem_.654_JUN10.indd 440 7/5/10 12:30:58




perspectiveNATure cHeMIsTry doi: 10.1038/nchem.654

connectors into exactly the same type of net that we just described for MOF-5-type structures.

For mechanically interlocked molecules (MIMs), known13,14 as catenanes (Lat. Catena = chain) and rotaxanes (Lat. Rota = wheel, axis = axle) to behave as molecular switches and machines they must incorporate two characteristic features6,7,15–17. One is that their components — two mechanically interlocked rings (Fig. 2) in the case of a bistable [2]catenane, and a ring encircling a dumbbell in the case of a bistable [2]rotaxane — must communicate with each other by means of intramolecular forces that can be modu-lated subsequently6,7,13–17 with chemicals (for example, pH change), electricity or light, that is, redox change. The other feature is that, as a result of some judiciously chosen constitutional dissymmetry, as well as the required orthogonality to stimuli associated with the two different recognitions sites — the more interactive of which can be switched OFF and ON reversibly with complete fidelity in the presence of the weaker site. By means of this modulation, bistable MIMs can be raised from a ground to a metastable state, such that there are two translationally isomeric13 forms expressing their bistability9.

The first feature is associated intimately with the efficient synthesis of bistable MIMs by protocols that rely on templation8 — that is, the use of molecular recognition processes involving a steadily increasing number of intermolecular forces, which become intramolecular on the formation of a mechanical bond7, a factor that guides the assembly of the components of the bistable MIMs. It is the very fact that the non-covalent bonding, which is introduced incrementally in a step-wise manner into these bist-able MIMs each time a covalent bond is formed, ‘lives on’ inside the molecules afterwards, which endows them with their unique properties. The situation is a true chicken-and-egg one: without the progressive build-up of non-covalent bonds during templa-tion8, the outcome of a synthesis will be no better than statistical in nature — that is, the yields will be miniscule and the product will contain little or no information. One attribute of their construc-tion feeds off the other to the extent that, if reversibility is intro-duced into the covalent bond-forming steps, then proofreading and error checking will often lead18 to all but quantitative yields.

In pursuit of systems expressing functions, there has been a drive to self-assemble them (for example, as thiols on gold) on surfaces19, or to place them at interfaces by self-organization — for example, by Langmuir–Blodgett transfer of monolayers20 of amphiphilic MIMs — to create nanoelectromechanical systems21 or molecular elec-tronic devices22, respectively. The incoherence that characterizes the operation of artificial molecular switches and machines only finds a practical expression when they are constituted as self-assembled monolayers on surfaces19 or in molecular switch tunnel junctions at interfaces20,22,23 one molecule thick between two electrodes. Not only do these self-assembly processes and self-organizational pro-cedures all come with a downside in the shape of disorder, which introduces blemishes into devices, the artificial machinery is also often impeded in its function and usually becomes exhausted after tens, or, at the most, hundreds, of cycles. In other words, the dis-tribution of orientation of the active molecules within these envi-ronments leads to their drastically reduced performance. So the inevitable question arises — how can we remove these blemishes and impediments and, at the same time, improve on the perform-ance of molecular switches and machines?

The act of introducing MIMs into MOFs can be likened to the building of helicopters, jet planes and rockets. These particular fly-ing machines are all constructed around robust fuselages to which engines are attached to the top, or on the sides, or at the back, or on the bottom. By appealing to a combination of robustness and motive power, we have all but overcome the hurdle of flight, which fascinated, yet evaded human beings for centuries. By the same token, we can envisage incorporating the artificial switches

and machines, as part of the two-way connectors held together by cluster joints during the template-directed synthesis8 of the ena-bled struts. We foresee the possibility of being able to locate arrays of molecular switches and machines symmetrically and efficiently inside the 3D structures of MOFs, while retaining their inherent robustness. By virtue of the highly ordered MOF structure, its ult-rahigh porosity, and facile accessibility to all its internal sites, the dynamic components of the MIMs, mounted within the extended structure provided by the MOF, will, in principle, be completely addressable and, under the right set of circumstances, behave in a coherent and reproducible manner. In essence, the random motion that plagues untethered MIMs in solution and condensed phases is curbed (Fig. 3) in MOFs.

Recently, the feasibility of mounting MIMs within MOFs has become evident from the successful introduction of a strut (Fig. 2a) into a MOF (Fig. 2e) and the formation of its [2]pseudorotaxane24-27 MOF (Fig. 2f). Although MOFs incorporating the degenerate28 and the non-degenerate29 [2]catenane (Fig. 2c,d and g,h) have not been synthesized as yet, a 2D MOF containing the strut in Fig. 2c, as illustrated in Fig. 3a, has already been reported30. In these extended structures (Fig. 2e,f), long organic struts (~2 nm) incorporating 34- and 36-membered polyether rings27 were used to build new MOFs (MOF-1001 and MOF-1002) whose structures are based on that of the MOF-5-type. The polyether chains are known to be highly dynamic in their free state, where they are folded in on them-selves in the absence of guests, but are capable of readily unfolding to accommodate guests in their interior. Accordingly, this dynamic behaviour is also present in the new MOFs, in which the polyether units are found to be disordered in the crystals.

There is, however, yet another reason for the macrocycles to be disordered in MOF-1001 and MOF-1002: the presence of planar chirality. As the method of their synthesis did not permit any con-trol over the handedness of the macrocycles, enantiomeric forms of them are presumably distributed in three dimensions through-out the crystals. In the future, however, it is going to be possible

a

b

Figure 3 | The 2D and 3D merger of MIMs and MOFs to bring about robust dynamics. A representation of a, 2D and b, 3D MOFs incorporating units (catenated light blue and red in a for degenerate rings of blue positions and blue/purple for the non-degenerate bistable rings in b) that typically have random motions as discrete molecules, but such motions are brought to a standstill when the units are mounted covalently to the rigid framework backbone (pink spheres and grey struts). This merger between MIMs and MOFs brings order to the appended units and endows them with the ability to carry out well-defined repeated dynamics without compromising the integrity of the entire system — a concept termed robust dynamics.

nchem_.654_JUN10.indd 441 7/5/10 12:31:00




perspective NATure cHeMIsTry doi: 10.1038/nchem.654

to employ asymmetric catalysis31 to produce homochiral MOFs with receptors that incorporate planar chirality. Such dramatically new materials, in which the chiral receptor sites for the stereose-lective docking of enantiomers are arranged precisely in space and are easily accessible to racemic analytes in a mobile phase, could revolutionize the production of chiral stationary phases for high-performance liquid chromatography for the efficient separation of enantiomeric compounds.

Remarkably, when MOF-1001 (Fig. 2e) is exposed to Paraquat dications in acetonitrile solutions, its polyether loops unfold and bind to the Paraquat guests in a stereoelectronically specific man-ner. The process can be repeated many times with the full preser-vation of the MOF backbone structure and without leaving any imprint on it. We attribute this framework fidelity to the fact that the only segments of the MOF structure that are flexible are the polyether loops, and none of their dynamics require any alteration to the metrics of the framework. It is the ideal construct because the rigidity of the framework allows the permanent openness of the structure so that guests may move in and out without obstruc-tion and the large interstices provide sufficient space for the poly-ether loops to fold and unfold, and to do so independently of the framework. Therefore, the key to achieving robust dynamics in extended systems is only present in the segment of the structure where dynamics is desirable — leaving unperturbed the remain-der of the structure. The stereoelectronically selective manner in which Paraquat dication is bound to the polyether loops in MOF-1001 introduces (Fig. 4) molecular recognition into porous crystals that so far have operated on either a shape/size selective or compacting capability.

Further independence of the flexible units in MOFs is poten-tially achieved by employing mechanically interlocked compo-nents. In principle, the synthesis of such MOFs was shown30 to be feasible by the successful incorporation of degenerate, donor–acceptor [2]catenanes28 into the 2D structure of MOF-1011. The layered nature of this MOF and the strong layer–layer interac-tions in the crystal preclude any dynamics involving the mechani-cally interlocked rings. Indeed, the construction of a 3D MOF structure, such as that of the MOF-5-type, would be necessary to realize the full potential of the interlocking rings’ dynamics (Figs 2h and 3b). Nevertheless, it is encouraging to observe that these mechanically interlocked components, which are proto-typical molecular machines6, can be mounted successfully inside extended MOF systems.

We believe that once the non-degenerate MOF-5-like structure is made, it will be just a matter of time before ultradense, 3D arrays

of molecular memory based on switchable [2]catenanes make their way into state-of-the-art device settings23. Ultradense memory, however, is the tip of the iceberg, provided it can be addressed. In the fullness of time, coupling switching with the ever-increasing capabilities of carrying out recognition processes — for exam-ple, microcontact printing on the surfaces and extending into the highly sophisticated interiors of these new switchable MOFs — will become commonplace. In principle, MOFs with molecular machinery mounted appropriately within their extended struc-tures would not be unlike airplanes carrying their strategically mounted propeller-driven or jet engines on robust wings. One can easily conceptualize a chemical world where chameleon-like MOF crystals can be induced under the influence of chemicals (pH change), electricity (redox change) and light to travel in solu-tion from one environment to another. In essence, the concept of robust dynamics is not only a necessary requirement for the lon-gevity of dynamic extended structures, but it is also a strategy for adding yet another layer of complexity to the present functioning capabilities of MOFs.

The ability to build integrated systems that are capable of a complex set of functions is reminiscent of operations in the bio-logical world. In a sense, robust dynamics is an example of how the concept of repeating dynamics, which is so prevalent in biol-ogy, can be transferred from the biological world to an artificial arena occupied by MOFs and MIMs and filled with integrated systems, without actually mimicking biology. Robust dynamics, therefore, is about concept transfer from the life sciences into the chemistry of materials. We can define32 concept transfer as “adapt-ing and applying the recognition processes, employed by living systems in achieving their forms and fulfilling their functions, to the construction of chemical systems with well-defined forms and prescribed uses”. In the wake of this definition, we propose that concept transfer, rather than trying to mimic biology, is a more via-ble approach33 for making the next leap in the design and synthesis of useful materials. Moreover, the ordered, extended structures of MOFs and the flexibility with which they can be synthesized and functionalized with MIMs render the marriage between the two chemistries ideally suited for uncovering, testing and developing other concept transfer strategies.

references1. Long, J. R. & Yaghi O. M. The pervasive chemistry of metal-organic

frameworks. Chem. Soc. Rev. 38, 1213–1214 (2009).2. Yaghi, O. M. et al. Reticular synthesis and the design of new materials. Nature

423, 705–714 (2003).

Sorting domain Coverage domain Active domain

Figure 4 | The sorting, coverage and active domains of MOFs. The three porous domains that the new MOFs (Figs 2 and 3) combine when they recognize and bind incoming substrates (guests). These domains are principally characterized by use of the pore opening to sort guests by shape and size selection (sorting domain), the internal adsorption sites to compact guests (coverage domain), and the crown ether receptors designed to bind guests in a stereoelectronically selective manner (active domain).

nchem_.654_JUN10.indd 442 7/5/10 12:31:01




perspectiveNATure cHeMIsTry doi: 10.1038/nchem.654

3. Horike, S., Shimomura, S. & Kitagawa, S. Soft porous crystals. Nature Chem. 1, 695–704 (2009).4. Serre, C. et al. Role of solvent-host interactions that lead to very large swelling of

hybrid frameworks. Science 315, 1828–1831 (2007). 5. Maji, T. K., Matsuda, R. & Kitagawa, S. A flexible interpenetrating

coordination framework with a bimodal porous functionality. Nature Mater. 6, 142–148 (2007).

6. Balzani, V., Credi, A. & Venturi, M. Molecular Devices and Machines – Concepts and Perspectives for the Nanoworld (Wiley-VCH, 2008).

7. Stoddart, J. F. The chemistry of the mechanical bond. Chem. Soc. Rev. 38, 1802–1820 (2009).

8. Diederich, F. & Stang, P. J. (eds) Templated Organic Synthesis (Wiley-VCH, 1999).

9. Choi, J. W. et al. Ground-state equilibrium thermodynamics and switching kinetics of bistable [2]rotaxane switches in solution, polymer gels, and molecular electronic devices. Chem. Eur. J. 12, 261–279 (2006).

10. Li, H., Eddaoudi, M., O’Keeffe, O. & Yaghi, O. M. Design and synthesis of an exceptionally stable and highly porous metal-organic framework. Nature 402, 276–279 (1999).

11. Ockwig, N., Friedrichs, O. D., O’Keeffe, M. & Yaghi, O. M. Reticular chemistry: occurrence and taxonomy of nets, and grammar for the design of frameworks. Acc. Chem. Res. 38, 176–182 (2005).

12. Eddaoudi, M. et al. Systematic design of pore size and functionality in metal-organic frameworks and application in methane storage. Science 295, 469–472 (2002).

13. Schill, G. Catenanes, Rotaxanes and Knots (Academic Press, 1971).14. Sauvage, J-P. & Dietrich-Buchecker, C. (eds) Molecular Catenanes, Rotaxanes and

Knots: A Journey Through the World of Molecular Topology (Wiley-VCH, 1999).15. Bissell, R. A., Cordova, E., Kaifer, A. E. & Stoddart, J. F. A chemically and

electrochemically switchable molecular shuttle. Nature 369, 133–137 (2004).16. Livoreil, A., Dietrich-Buchecker, C. O. & Sauvage, J-P.

Electrochemically triggered swinging of a [2]catenane. J. Am. Chem. Soc. 116, 9399–9400 (1994).

17. Kay, E. R., Leigh, D. A. & Zerbetto, F. Synthetic molecular motors and mechanical machines. Angew. Chem. Int. Ed. 46, 72–191 (2007).

18. Chichak, K. S. et al. Molecular Borromean rings. Science 304, 1308–1312 (2004).19. Klajn, R. et al. Metal nanoparticles functionalized with molecular and

supramolecular switches. J. Am. Chem. Soc. 131, 4233–4235 (2009).

20. Collier, C. P. et al. A [2]catenane-based solid-state electronically reconfigurable switch. Science 289, 1172–1175 (2000).21. Juluri, B. K. et al. A mechanical actuator driven electrochemically by artificial

molecular muscles. ACS Nano 3, 291–300 (2009). 22. Luo, Y. et al. Two-dimensional molecular electronic circuits. ChemPhysChem

3, 519–525 (2002). 23. Green, J. E. et al. A 160-kilobit molecular electronic memory patterned at 1011

bits per square centimetre. Nature 445, 414–417 (2007).24. Li, Q. et al. Docking in metal-organic frameworks. Science 325, 855–859 (2009).25. Kim, K. Entering the recognition domain. Nature Chem. 1, 603–604 (2009). 26. Alavi, S. Selective guest docking in metal-organic framework materials.

ChemPhysChem 11, 55–57 (2010). 27. Zhao, Y-L. et al. Rigid strut-containing crown ethers and [2]catenanes

for incorporation into metal-organic frameworks. Chem. Eur. J. 15, 13356–13380 (2009).

28. Anelli, P-L. et al. Molecular meccano 1. [2]Rotaxane and a [2]catenane made to order. J. Am. Chem. Soc. 114, 193–218 (1992).

29. Balzani, V. et al. Constructing molecular machinery. A chemically switchable [2]catenane. J. Am. Chem. Soc. 122, 3542–3543 (2000).

30. Li, Q. et al. A metal-organic framework replete with ordered donor-acceptor catenanes. Chem. Commun. 46, 380–382 (2010).

31. Kandra, K., Koike, T., Endo, K. & Shibata, T. The first asymmetric Sonogashira coupling for enantioselective generation of planar chirality in paracyclophanes. Chem. Commun. 1870–1872 (2009).

32. Glink, P. T. & Stoddart, J. F. Concept transfer from the life sciences into materials science. Pure Appl. Chem. 70, 419–424 (1998).

33. Stoddart, J. F. Thither supramolecular chemistry. Nature Chem. 1, 14–15 (2009).

acknowledgementsWe acknowledge the Department of Energy (BES-Separation Program), the Department of Defense (Defense Reduction Threat Agency), the Air Force Office of Scientific Research under their Multidisciplinary University Research Initiative (FA9550-07-1-0534), the Microelectronics Advanced Research Corporation and its Focus Center Research Program, the Center on Functional Engineered Nano-Architectonics, and the NSF-MRSEC Program through the Northwestern University Materials Research Science and Engineering Center for their continued support of this research.

nchem_.654_JUN10.indd 443 7/5/10 12:31:01





© 2010 Macmillan Publishers Limited. All rights reserved.

Ion-triggered spring-like motion of a doublehelicate accompanied by anisotropic twistingKazuhiro Miwa, Yoshio Furusho* and Eiji Yashima*

Molecules that extend and contract under external stimuli are used to build molecular machines with nanomechanicalfunctions. But although common in biological systems, such extension and contraction motions with helical molecules haverarely been accompanied by unidirectional twisting in synthetic systems. Here we show that sodium ions can trigger thereversible anisotropic twisting of an enantiomeric double-stranded helicate, without racemization. An optically activehelicate consisting of two tetraphenol strands bridged by two spiroborate groups sandwiches a sodium ion. On removal ofthe central sodium—through addition of a cryptand [2.2.1] in solution—the double helicate extends. Crystallographic andnuclear magnetic resonance studies reveal that the extended helicate is over twice as long as the initial molecule, and istwisted in the right-handed direction. Circular dichroism analysis suggests that the twisting doesn’t affect the helicate’shandedness. This anisotropic extension–contraction process is reversibly triggered by the successive addition and removalof sodium ions in solution.

The design and construction of molecular actuators that canundergo a reversible extension–contraction motion at themolecular level triggered by chemical, electrochemical or

photochemical stimulation has recently been the subject of con-siderable interest in the context of molecular machines1–16. Themuscle-like extension–contraction molecular motion has beenachieved by a number of molecular devices, such as macrocycles5,rotaxanes6–8, oligomeric and polymeric helical systems9–16, andcoordination solids10. In particular, helical molecules are of signifi-cant interest as their inherently chiral structures suggest a seductiveanisotropic twisting (unidirectional spring-like motion) during theextension–contraction process. Several synthetic helical moleculesand polymers exhibit such extension–contraction motions triggeredby ion binding5,9 and changes in pH12, solvent13 or temperature11,16;these helical systems have rarely undergone a unidirectional twistingmotion, even though it is common in biological systems17–19.

Recently, Percec et al. reported one precedent using dendronizedhelical poly(phenylacetylene) molecules that exhibit a thermallyinduced anisotropic twisting16. In these biological and syntheticpolymer systems, however, the resulting anisotropic twisting is con-trolled entirely by the homochirality of the biopolymers and by theintroduction of stereocentres into the monomer units, respectively;therefore, one of the diastereomeric helices is favourably formed,leading to a unidirectional twisting motion. In contrast, thesynthetic helical molecules readily racemize or take a non-helicalconformation or transition state during the extension–contractionprocess, which results in a bidirectional twisting motion. Helicalmolecules that undergo a spring-like motion accompanied by aniso-tropic twisting could generate a torque on microscopic objects androtate them unidirectionally, when embedded in liquid crystals20.Ultimately, this could lead to a sophisticated unimolecularmachine that can perform intelligent work.

Over the course of our research on synthetic helical moleculesbased on oligo- and poly(m-phenylene) structures21–23, we unex-pectedly found that the reaction of a hexaphenol, H6L1 (1), withan equimolar amount of NaBH4 afforded a unique double-strandedhelicate (DH1BNaB

2.Naþ, Fig. 1), which consists of the two

hexaphenol strands bridged by two spiroborate groups accommo-dating a Naþ ion in the centre coordinated by eight oxygenatoms24. The structure of DH1BNaB

2 predicts that the two hydroxylgroups on the two central benzene rings of each strand are notnecessary for the formation of the helicate because they do not par-ticipate in the spiroborate bridges. With this in mind, we havedesigned and synthesized a new double-stranded helicate(DH2BNaB

2) that contains a tetraphenol lacking the two centralhydroxyl groups on the hexa(m-phenylene) backbone (H4L2, 2).Here we report an enantiomeric double helical molecule ofDH2BNaB

2 obtained by optical resolution of the racemichelices that undergoes Naþ ion-triggered, reversible extension–contraction motion coupled with a twisting motion in onedirection (right-handed).

Results and discussionThe ligand H4L2 (2) was allowed to react with an equimolar amountof NaBH4 in 1,2-dichloroethane-ethanol at 80 8C for 20 h, affordingthe boron complex [B2Na(L2)2]2.Naþ (DH2BNaB

2.Naþ) in 22%yield (Fig. 1). Countercation exchange with N-benzyl-N,N,N-trimethyl-ammonium bromide (BMAmmþ.Br2) gave DH2BNaB

2.BMAmmþ,of which single crystals suitable for an X-ray study were grownfrom an acetonitrile solution. The X-ray crystallographic analysisunambiguously revealed that DH2BNaB

2.BMAmmþ adopts adouble helical structure with a pseudo-D2-symmetry, which isvirtually isomorphic to that of the hexaphenol-based helicateDH1BNaB

2 (Fig. 2 and Supplementary Fig. S1).The two tetraphenol strands are intertwined with each other

through two spiroborate bridges, and a Naþ ion is embraced inthe centre of the complex coordinated by four oxygen atoms ofthe spiroborate moieties. The two terminal benzene rings of eachtetraphenol strand are twisted by approximately 2808.Interestingly, the two negatively charged spiroborate moieties arevery close to each other, with a B–B distance of 6.0 Å, as they areelectrostatically attracted by the positively charged central Naþ

ion. Electrospray ionization (ESI) mass measurements in the nega-tive mode support the monoanionic nature of the boron helicate,

Department of Molecular Design and Engineering, Graduate School of Engineering, Nagoya University, Chikusa-ku, Nagoya 464-8603, Japan.

*e-mail: [email protected]; [email protected]

ARTICLESPUBLISHED ONLINE: 2 MAY 2010 | DOI: 10.1038/NCHEM.649

NATURE CHEMISTRY | VOL 2 | JUNE 2010 | www.nature.com/naturechemistry444

http://www.nature.com/compfinder/10.1038/nchem.649_comp1






www.nature.com/naturechemistry


tBu

O

tBu

O

tBu

O

tBu

O

tBu

O

tBu

O

tBu

O

tBu

ONaB B

–

Na+

X X

XX

Y Y

YY

NC12H25

Ph

OH

+Br–

N+

Br–C4H9

C4H9C4H9

C4H9

1 H6L1: X = OH, Y = tBu2 H4L2: X = Y = H

NaBH4

Cl(CH2)2Cl-EtOH

80 °C, 20 h

DH1BNaB–·Na+: [B2Na(H2L1)2]–·Na+

DH2BNaB–·Na+: [B2Na(L2)2]–·Na+

DH2BNaB–·Na+

(+)-DH2BNaB–·(–)-DMEph+

(–)-DH2BNaB–·(–)-DMEph+ (–)-DH2BNaB

–·TBAmm+

(+)-DH2BNaB–·TBAmm+

TBAmm+·Br–

TBAmm+·Br–

93% d.e.

29% d.e.

Precipitates

Filtrate

(–)-DMEph+·Br–

a

b

((–)-DMEph+·Br–)

(TBAmm+·Br–)

OH

OH

X

Y

X

Y

OH

tBu

tBu

tBu

tBu

OH

93% e.e.

29% e.e.

Figure 1 | Synthesis and optical resolution. a, Synthesis of the boron helicates DH1BNaB2.Naþ and DH2BNaB

2.Naþ. b, Optical resolution of DH2BNaB2.Naþ by

a diastereomeric salt formation using (–)-DMEphþ.Br2. The Naþ cation located outside the helicate moiety was exchanged with (–)-DMEphþ in an

acetonitrile solution to form a pair of diastereomeric salts, (þ)-DH2BNaB2.(–)-DMEphþ and (–)-DH2BNaB

2.(–)-DMEphþ. The former precipitated from the

solution and the diastereomeric excess was determined to be 93%, whereas the latter was collected by evaporating the filtrate as a white solid with 29% d.e.

Both diastereomeric salts were converted to a pair of enantiomeric salts through countercation exchange with achiral TBAmmþ.Br2 in acetonitrile

without racemization.

Na

B

B

a

b

(±)-DH2BNaB–.BMAmm+

B

Na

B

B

Ring current

B

~280°

Ringcurrent

(±)-DH1BNaB–.H+

Figure 2 | Capped-stick representations of the crystal structures of the trinuclear boron helicates. a, Side (left) and top (right) views of the crystal

structure of (+)-DH2BNaB2.BMAmmþ, which reveal the double-stranded helical structure of the helicate bridged by the spiroborates formed from the

terminal biphenol units and the boron atoms. The t-Bu groups are located in close proximity to the benzene rings of the other strand, and their 1H NMR

signals are therefore shifted upfield owing to an aromatic ring current effect. The two terminal benzene rings of each strand are twisted by �2808. b, Side

(left) and top (right) views of the crystal structure of DH1BNaB2.Hþ. Similar upfield shifts of the terminal t-Bu groups can be seen for each strand as the

double-stranded helical structure of DH1BNaB2.Hþ is virtually isomorphic to DH2BNaB

2. All the hydrogen atoms, solvent molecules, and countercations are

omitted for clarity.

NATURE CHEMISTRY DOI: 10.1038/NCHEM.649 ARTICLES

NATURE CHEMISTRY | VOL 2 | JUNE 2010 | www.nature.com/naturechemistry 445




showing a strong signal due to the monovalent anion (DH2BNaB2) at

m/z¼ 1529.94 along with a minor signal due to the divalent anion(DH2BB

22) at m/z¼ 753.46 (Supplementary Fig. S2). The 1Hnuclear magnetic resonance (NMR) spectrum of the complex inCD3CN also revealed the pseudo-D2-symmetric structure as deter-mined by X-ray analysis in the solid state. The two t-Bu signalsshifted upfield owing to the ring current effect of the benzenerings of the other strand, which is in good agreement with thecrystal structure (Fig. 2). Furthermore, the 1H two-dimensional(2D) nuclear Overhauser effect spectroscopy (NOESY) experimentsshowed strong negative cross peaks between the phenol rings A andC (Supplementary Figs S3–S7), which are attributed to the inter-strand nuclear Overhauser effects (distance approximately 2.5 Å).This indicates that the complex retains the double-stranded helicaltrinuclear structure in solution.

The optical resolution of the helicate ((+)-DH2BNaB2.Naþ) was

then carried out by diastereomeric salt formation through cationexchange with (–)-N-dodecyl-N-methylephedrinium bromide((–)-DMEphþ.Br2), which had been successfully used forthe optical resolution of (+)-DH1BNaB

2.Naþ (ref. 24; Fig. 1). Onaddition of a tenfold excess of (–)-DMEphþ.Br2 to a solution of(+)-DH2BNaB

2.Naþ in acetonitrile, (þ)-DH2BNaB2.(–)-DMEphþ

was obtained as a white crystalline solid (the prefixes (þ) and (–)

for the helicate moieties denote the signs of the Cotton effect—that is, the characteristic change in optical rotatory dispersionand/or circular dichroism (CD) in the vicinity of an absorptionband of a substance—at 313 nm). Its diastereomeric excess(d.e.) was determined to be 93% on the basis of its 1H NMRspectra (Supplementary Figs S9 and S10). Evaporation of themother liquor afforded the opposite diastereomer rich in(–)-DH2BNaB

2.(–)-DMEphþ with 29% d.e.The CD spectra of both diastereomers showed almost mirror

image Cotton effects in their patterns in the range of 240–330 nmexcept for the intensities because the CD signals due to the(–)-DMEphþ moiety are negligible (Supplementary Fig. S8). Thehelix senses of the diastereomers were assigned to be right- andleft-handed for (þ)-DH2BNaB

2.(–)-DMEphþ and (–)-DH2BNaB2.

(–)-DMEphþ, respectively, on the basis of their CD patterns relativeto those of (þ)- and (–)-DH1BNaB

2.Naþ. The relationship betweenthe handedness of the hexaphenol-based helicate and the CD pat-terns of these had been established in an X-ray single crytsallo-graphic study (unpublished observations).

The resulting diastereomeric double-stranded helicates were suc-cessfully converted to the corresponding enantiomers by exchangingthe optically active ammonium cation (–)-DMEphþ with achiraltetrabutylammonium bromide (TBAmmþ.Br2) (Supplementary

tBu

tBuA

tBuB

tBuAtBuB

O

tBu

O

tBu

O

tBu

O

tBu

O

tBu

O

tBu

O

tBu

OBB Na

–tBu

O

tBu

O

tBu

O

tBu

O

tBu

O

tBu

O

tBu

O

tBu

OB B

2–

N

O

O

N

O

O

O

DH2BNaB– DH2BB

2–[2.2.1]

[2.2.1] Na+ [2.2.1]

a

b c

DH2BNaB–

H4L2

DH2BB2–

OOB

A

B

C

(ppm)1.4 1.3 1.2 1.1 1.0 0.9 13.0 Å

(±)-DH2BB2–·(BMAmm+)2

~180°

BB

Figure 3 | Synthesis and characterization of the dinuclear helicate. a, Schematic illustration of the removal of the central Naþ ion from DH2BNaB2.Naþ by

cryptand [2.2.1]. b, 1H NMR spectra of H4L2 (2) (top), DH2BNaB2.Naþ (middle), and DH2 2

BB.(Naþ, [2.2.1])2 (bottom) in CD3CN. The 1H NMR signals of

the t-Bu groups of DH2 22BB did not show upfield shifts. c, The capped-stick representations of the crystal structure of (+)-DH2 22

BB.(BMAmmþ)2 (side view

(bottom) and top view (top)). All the t-Bu groups are pointing outwards and are therefore remote from benzene rings in this structure, which explains the

lack of upfield shifts of the 1H NMR signals of the t-Bu groups as there is no ring current effect. The two terminal benzene rings of each strand are twisted by

�1808, and the distance between the two boron atoms is 13.0 Å, indicating that the length of helicate DH2 22BB is extended by almost twofold. All the

hydrogen atoms, solvent molecules, and countercations are omitted for clarity.

ARTICLES NATURE CHEMISTRY DOI: 10.1038/NCHEM.649






Figs S11 and S12). Thus, the (þ)-DH2BNaB2.TBAmmþ bearing only

the helical chirality, without any other chiral factors, was obtained.The Cotton effects and absorption spectra are perfectly identical tothose of (þ)-DH2BNaB

2.(–)-DMEphþ, indicating that the helicateis inert to changes in stereochemistry during the cation exchangeprocess, and its enantiomeric excess (e.e.) (helical sense-excess)can be estimated to be 93% (Supplementary Fig. S8). The other dia-stereomer, (–)-DH2BNaB

2.(–)-DMEphþ with 29% d.e., was sim-ilarly converted to the corresponding enantiomer,(–)-DH2BNaB

2.TBAmmþ, of which the d.e. was determined to be29% on the basis of its relative CD intensity to that of the oppositeenantiomer (Supplementary Fig. S8). Furthermore, neither the racemi-zation nor decomposition of the complex (þ)-DH2BNaB

2.TBAmmþ

took place on heating in acetonitrile at 60 8C for 9 days and at 80 8Cfor 24 h. The extended helicate (þ)-DH2BB

22.TBAmmþ.Naþ, 221(see below) was also stable and maintained its optical activitywithout racemization after heating at 60 8C for 9 days. In contrast,when 10 mol% of trifluoroacetic acid was added to an acetonitrilesolution of (þ)-DH2BNaB

2.TBAmmþ, the Cotton effects comple-tely disappeared within 1 h and there was a negligible change inthe absorption spectra. This indicates that racemization tookplace, probably through an acid-catalysed B–O bond cleavage andreformation of the spiroborate groups.

We expected that the weakly bound central Naþ ion could beremoved from the DH2BNaB

2 helicate with the assistance of crownethers or cryptands to give the divalent cationic helicate DH2BB

22

(Fig. 3a). The first attempt using 2 equivalents of 15-crown-5 etherfailed, but cryptand [2.2.1], which is known to bind cations morestrongly than its crown ether analogue, successfully trapped theNaþ ion. On the addition of 2 equivalents of cryptand [2.2.1] toa CD3CN solution of DH2BNaB

2.Naþ, the 1H NMR signals assignedto Naþ, [2.2.1] (this denotes that the cryptand [2.2.1] includes a

Naþ ion) appeared and those of the DH2BNaB2.Naþ were drastically

changed (Fig 3b and Supplementary Figs S19 and S20). In particular,the two t-Bu signals showed large downfield shifts of Dd≈ 0.52 and0.14 ppm, which indicates that the t-Bu groups are no longer underthe ring current of the aromatic rings of the other strand.

When a tenfold excess of BMAmmþ.Br2 was added to the sol-ution, the countercation exchange readily proceeded to yieldDH2BB

22.(BMAmmþ)2, of which single crystals were graduallyformed from the solution. The X-ray single crystal analysis ofDH2BB

22.(BMAmmþ)2 confirmed that the Naþ ion is indeedremoved from the centre of the DH2BNaB

2 helicate by the cryptandand that the resultant boron helicate, DH2BB

22.(BMAmmþ)2, hasa dianionic nature with two countercations, BMAmmþ, in theouter space (Fig. 3c and Supplementary Fig. S13). The crystal struc-ture accounted for the absence of the ring current effect on the t-Bu groups in the 1H NMR spectrum and the absence of interstrandcross peaks in the rotating-frame nuclear Overhauser effect spec-troscopy (ROESY) spectra (Supplementary Figs S16–S18). Thebinding constant of DH2BB

22 to the Naþ ion was determined tobe 2.68 × 106 M21 in acetonitrile at 25 8C by the competitivebinding titration experiment between DH2BNaB

2 and dicylohex-ano-18-crown-6 ether (DC18C6) using 1H NMR spectroscopy(Supplementary Fig. S22).

Of particular interest is that the length of the boron helicateDH2BB

22 is significantly extended—by almost twofold and with aB–B distance of 13.0 Å—when compared with DH2BNaB

2 (6.0 Å).This noticeably large extension is most likely attributable to theenhanced electrostatic repulsion between the two anionic spiroboratemoieties due to the absence of the central Naþ ion. Close inspection ofthe crystallographic data of the DH2BNaB

2 and DH2BB22 reveals that

the torsion angles between the benzene rings of the spiroborate moi-eties are inverted during the extension–contraction process, while

Δ(M

–1 c

m–1

)(10

5 M–1 cm

–1)

400200 250 300

0

1

2

3

40

200

400b

a

–200

–400

Wavelength (nm)

350

(+)-DH2BNaB–·TBAmm+

+ [2.2.1]

+ NaPF6

Na+

B [2.2.1]

DH2BNaB– DH2BB

2–

6.0 Å

Na

13.0 Å

B BB

– 2–

Figure 4 | Anisotropic twisting motion of the helicate. a, Schematic representation of anisotropic twisting motion observed in a right-handed double helical

helicate. b, CD and absorption spectra (CH3CN, 25 8C) of (þ)-DH2BNaB2.TBAmmþ (93% d.e.) (red). After the addition of 1.5 equivalents of cryptand [2.2.1]

(blue), (þ)-DH2BNaB2 was quantitatively converted to the extended helicate (þ)-DH2 22

BB , which gave a completely different CD spectrum to that of

(þ)-DH2BNaB2, reflecting the difference in their helical structures (blue). The addition of 1.5 equivalents of NaPF6 reverted (þ)-DH2 22

BB back to the contracted

helicate, (þ)-DH2BNaB2, quantitatively and without racemization, as is apparent from the complete recovery of both CD and absorption spectra (dashed black).






maintaining their spirochirality (Supplementary Table S3). In accord-ance with the extension, the helicate unwinds with a twisting anglebetween the terminal benzene rings of each tetraphenol strand of�1808, demonstrating that the extension is coupled with the twistingmotion in one direction, although the direction of twisting could notbe defined because the DH2BNaB

2 complex used was racemic. Addinga small excess of NaPF6 to the solution of DH2BB

22 readily and quan-titatively regenerated the trinuclear helicate, DH2BNaB

2, as evidencedby its 1H NMR spectral change. Analogous trinuclear helicatesembracing Liþ, Kþ or NH4

þ ions in the centre were obtained by theaddition of LiPF6, KPF6 or NH4PF6, respectively, instead of NaPF6.

In order to realize the anisotropic right- or left-handedtwisting motion, we next used the right-handed double helicalDH2BNaB

2.TBAmmþ. A unidirectional rotary motion has beenachieved with an optically active helical alkene25 bearing two stereo-centres that are essential to control the diastereomeric helical chir-ality, leading to the unidirectional behaviour of the molecularmotor. The pure enantiomers of helical alkenes with no stereogeniccentres will lose their optical activity owing to racemization, result-ing in bidirectional rotating.

The addition of a slight excess of [2.2.1] to a solution of the right-handed double helical (þ)-DH2BNaB

2.TBAmmþ with 93% e.e.brought about complete removal of the central Naþ ion from thehelicate, as confirmed by its 1H NMR spectrum (SupplementaryFig. S21). The CD spectra of the extended helicate (þ)-DH2 22

BB isquite different from that of the contracted helicate (þ)-DH2BNaB

2,reflecting the difference in their helical structures (Fig. 4). Onfurther addition of NaPF6, the extended helicate can quantitativelyrevert back to the contracted helicate, as demonstrated by the recov-ery of the initial absorption and CD spectra. This complete recoveryof the CD spectrum unambiguously indicates the dynamic exten-sion and contraction motions that proceed without any racemiza-tion or breaking and reformation process of the spiroborategroups. It should be noted that both helicates DH2BNaB

2 andDH2 22

BB showed no spectral change within the temperaturerange of þ70 to –10 8C (Supplementary Fig. S23), indicating thatthe spiroborate helicates DH2BNaB

2 and DH2 22BB most likely

exist as single species—the contracted form with a sandwichedNaþ ion and the extended form, respectively—irrespective oftemperature. This extension–contraction cycle can be repeatedseveral times by the sequential addition of [2.2.1] and NaPF6 inan alternating manner. Similarly, Liþ and Kþ ions also triggeredan anisotropic extension/contraction motion, as supported by theCD and absorption spectral changes (Supplementary Fig. S24).

In order to study the kinetics of the extension event ofDH2BNaB

2 induced by cryptand [2.2.1] and the contraction eventof DH2 22

BB triggered by Naþ ions, stopped-flow CD measure-ments in acetonitrile at 22 8C were performed and the CD intensitychanges at 240 nm were followed (Supplementary Fig. S25).Interestingly, the extension event of DH2BNaB

2 induced by cryptand[2.2.1] was found to take place much more slowly than thecontraction event of DH2 22

BB triggered by Naþ ions; the extensionevent took approximately 5 sec to finish. The data for the extensionwere fitted to an equation based on pseudo-second order kineticsby the nonlinear least-squares curve-fitting method to yield a rateconstant kext of (5.38+0.05) × 103 M21 s21. In contrast, thecontraction event was too fast to follow under the present exper-imental conditions, indicating that it takes place within thedead time of the apparatus (7.8 msec), which would give a kcontmuch higher than 4.8 × 105 M21 s21. The slower rate of theextension process could be attributed to the steric hindrancegenerated by the cryptand [2.2.1]; the contraction process, in con-trast, involves Naþ cations solvated by acetonitrile molecules.This suggests that the extension and contraction rates can becontrolled by selecting Naþ ion trapping reagents (cryptand[2.2.1] in this case).

The understanding gained in this study will serve as a startingpoint for future nanoscale mechanical device applications thatincorporate molecules undergoing a specific double helical twistingmotion although maintaining their one-handedness.

MethodsSynthesis of (+++++)-DH2 2

BNaB.Na1. To a solution of 2 (100 mg, 134 mmol) in

1,2-dichloroethane (18.0 ml) was added a solution of NaBH4 in ethanol (44.6 mM,3.0 ml, 134 mmol) under an argon atmosphere. After stirring at 80 8C for 20 h,the mixture was cooled to ambient temperature to form a white precipitate,which was removed by filtration. The filtrate was evaporated to dryness, and1,2-dichloroethane (2 ml) was added to the residual viscous oil to form a whiteprecipitate. This was collected by filtration, washed with 1,2-dichloroethane(2 × 2 ml), and dried in vacuo to afford (+)-DH2BNaB

2.Naþ as a white solid in22% yield. Melting point . 300 8C; ESI-MS (CH3CN, negative): m/z¼ 1,530[M–Na]2, 753 [M–2Na]2. Analysis calculated for C104H108B2Na2O8: C, 80.40;H, 7.01; found, C, 80.40; H, 6.89. For 1H and 13C NMR data, seeSupplementary Information.

Synthesis of (+++++)-DH2 2BNaB

.BMAmm1. To a solution of (+)-DH2BNaB2.Naþ

(10.0 mg, 6.4 mmol) in CH3CN (10 ml) was added a tenfold molar excess ofbenzyltrimethylammonium bromide (BMAmmþ.Br2, 14.7 mg, 64.0 mmol).The solution was allowed to stand at ambient temperature for a few days toform colourless crystals, which were collected by filtration to give(+)-DH2BNaB

2.BMAmmþ in 42% yield. High resolution mass spectrometry(HRMS) (negative mode ESI, CH3CN/CHCl3 (1/1, v/v)): m/z calculated for[M–BMAmmþ]2, 1,529.8156; found, 1,529.8134. For 1H NMR data, seeSupplementary Information.

Optical resolution of (+++++)-DH2 2BNaB

.Na1. To a solution of (+)-DH2BNaB2.Naþ

(200 mg, 0.129 mmol) in CH3CN (168 ml) was added a tenfold molar excess of(–)-N-dodecyl-N-methylephedrinium bromide ((–)-DMEphþ.Br2) (553 mg,1.29 mmol). The mixture was stirred at room temperature for 36 h to form a whiteprecipitate. After filtration, the collected white precipitate was washed withCH3CN (3 × 5 ml), and then dried in vacuo to give the diastereomeric salt,(þ)-DH2BNaB

2.DMEphþ (93% d.e.) as a white solid in 38% yield (45.8 mg). Thefiltrate was evaporated in vacuo to give a mixture of the other diastereomeric salt,(–)-DH2BNaB

2.DMEphþ (29% d.e.) and an excess of ammonium salt(–)-DMEphþ.Br2 (691 mg). (þ)-DH2BNaB

2.DMEphþ: HRMS (negative mode ESI,CH3CN/CHCl3 (1/1, v/v)): m/z calculated for [M–DMEphþ]2, 1,529.8156; found1,529.8108. For 1H NMR data, see Supplementary Information.

Conversion of (1)-DH2 2BNaB

.DMEph1 to (1)-DH2 2BNaB

.TBAmm1 (93% e.e.).To a solution of (þ)-DH2BNaB

2.DMEphþ (93% d.e.) (10.0 mg, 5.3 mmol) inCH3CN (15 ml) was added a 100-fold molar excess of tetra-n-butylammoniumbromide (TBAmmþ.Br2) (171 mg, 530 mmol). The mixture was stirred at roomtemperature for 24 h, and the solution was evaporated to dryness. The residue wastitrated with CH3CN (2 ml), filtrated, washed with CH3CN (0.5 ml), and driedin vacuo to give the enantiomeric salt, (þ)-DH2BNaB

2.TBAmmþ (93% e.e.), as awhite solid in 51% yield (4.8 mg). HRMS (negative mode ESI, CH3CN/CHCl3 (1/1,v/v)): m/z calculated for [M–2TBAmmþ]2, 1,529.8156; found 1,529.8107. For1H NMR data, see Supplementary Information.

Conversion of (1)-DH2 2BNaB

.TBAmm1 to (1)-DH2 22BB

.TBAmm1.Na1 , 221.To a solution of (þ)-DH2BNaB

2.TBAmmþ (93% e.e., 0.347 mg, 0.196 mmol) inCD3CN (0.7 ml) was added a solution of [2.2.1] in CD3CN (10 mM, 29.4 ml,0.294 mmol) to yield (þ)-DH2 22

BB.TBAmmþ.Naþ, 221 quantitatively. HRMS

(negative mode ESI): m/z calculated for [M–TBAmmþ – Naþ]2, 753.4129; found753.4125. For 1H NMR data, see Supplementary Information.

Received 5 October 2009; accepted 23 March 2010;published online 2 May 2010

References1. Urry, D. W. Molecular machines: How motion and other functions of living

organisms stem from reversible chemical changes. Angew. Chem. Int. Ed. Engl.32, 819–841 (1993).

2. Fyfe, M. C. T. & Stoddart, J. F. Synthetic supramolecular chemistry. Acc. Chem.Res. 30, 393–401 (1997).

3. Collin, J.-P., Dietrich-Buchecker, C., Gavina, P., Jimenez-Molero, M. C. &Sauvage, J.-P. Shuttles and muscles: Linear molecular machines based ontransition metals. Acc. Chem. Res. 34, 477–487 (2001).

4. Kinbara, K. & Aida, T. Toward intelligent molecular machines: Directed motionsof biological and artificial molecules and assemblies. Chem. Rev. 105,1377–1400 (2005).

5. Jousselme, B. et al. Crown-annelated oligothiophenes as model compounds formolecular actuation. J. Am. Chem. Soc. 125, 1363–1370 (2003).







6. Jimenez, M. C., Dietrich-Buchecker, C. & Sauvage, J.-P. Towards syntheticmolecular muscles: Contraction and stretching of a linear rotaxane dimer.Angew. Chem. Int. Ed. 39, 3284–3287 (2000).

7. Liu, Y. et al. Linear artificial molecular muscles. J. Am. Chem. Soc. 127,9745–9759 (2005).

8. Fang, L. et al. Acid-base actuation of [c2]daisy chains. J. Am. Chem. Soc. 131,7126–7134 (2009).

9. Bell, T. W. & Jousselin, H. Self-assembly of a double-helical complex of sodium.Nature 367, 441–444 (1994).

10. Jung, O.-S., Kim, Y. J., Lee, Y.-A., Park, J. K. & Chae, H. K. Smart molecularhelical springs as tunable receptors. J. Am. Chem. Soc. 122, 9921–9925 (2000).

11. Yashima, E., Maeda, K. & Sato, O. Switching of a macromolecular helicity forvisual distinction of molecular recognition events. J. Am. Chem. Soc. 123,8159–8160 (2001).

12. Barboiu, M. & Lehn, J.-M. Dynamic chemical devices: Modulation ofcontraction/extension molecular motion by coupled-ion binding/ph change-induced structural switching. Proc. Natl Acad. Sci. USA 99, 5201–5206 (2002).

13. Pengo, P. et al. Quantitative correlation of solvent polarity with the alpha -/310-helix equilibrium: A heptapeptide behaves as a solvent-driven molecular spring.Angew. Chem. Int. Ed. 42, 3388–3392 (2003).

14. Berni, E. et al. Assessing the mechanical properties of a molecular spring. Chem.Eur. J. 13, 8463–8469 (2007).

15. Kim, H.-J., Lee, E., Park, H.-S. & Lee, M. Dynamic extension-contraction motionin supramolecular springs. J. Am. Chem. Soc. 129, 10994–10995 (2007).

16. Percec, V., Rudick, J. G., Peterca, M. & Heiney, P. A. Nanomechanical functionfrom self-organizable dendronized helical polyphenylacetylenes. J. Am. Chem.Soc. 130, 7503–7508 (2008).

17. Chalmers, R., Guhathakurta, A., Benjamin, H. & Kleckner, N. Ihf modulation oftn10 transposition: Sensory transduction of supercoiling status via a proposedprotein/DNA molecular spring. Cell 93, 897–908 (1998).

18. Schoenauer, R. et al. Myomesin is a molecular spring with adaptable elasticity.J. Mol. Biol. 349, 367–379 (2005).

19. Nishinaka, T., Doi, Y., Hara, R. & Yashima, E. Elastic behavior of reca-DNAhelical filaments. J. Mol. Biol. 370, 837–845 (2007).

20. Eelkema, R. et al. Molecular machines: Nanomotor rotates microscale objects.Nature 440, 163 (2006).

21. Goto, H., Furusho, Y. & Yashima, E. Supramolecular control of unwinding andrewinding of a double helix of oligo-resorcinol using cyclodextrin/adamantanesystem. J. Am. Chem. Soc. 129, 109–112 (2007).

22. Goto, H., Furusho, Y. & Yashima, E. Double helical oligo-resorcinols specificallyrecognize oligosaccharides via heteroduplex formation through noncovalentinteractions in water. J. Am. Chem. Soc. 129, 9168–9174 (2007).

23. Goto, H., Katagiri, H., Furusho, Y. & Yashima, E. Oligoresorcinols fold intodouble helices in water. J. Am. Chem. Soc. 128, 7176–7178 (2006).

24. Katagiri, H., Miyagawa, T., Furusho, Y. & Yashima, E. Synthesis and opticalresolution of a double helicate consisting of ortho-linked hexaphenol strandsbridged by spiroborates. Angew. Chem. Int. Ed. 45, 1741–1744 (2006).

25. Koumura, N., Zijistra, R. W. J., Van Delden, R. A., Harada, N. & Feringa, B. L.Light-driven monodirectional molecule rotor. Nature 401, 152–155 (1999).

AcknowledgementsThis work was supported in part by Grant-in-Aids for Scientific Research from the JapanSociety for the Promotion of Science (JSPS) and for Scientific Research on Innovative Areas,“Emergence in Chemistry” (21111508) from the MEXT. We acknowledge Y. Kondo and K.Nagamori of JASCO for their help in the measurements of the stopped-flow CD spectra.

Author contributionsY.F. and E.Y. designed and directed the project and K.M. performed the experiments. K.M.,Y.F. and E.Y. analysed and discussed the results and co-wrote the manuscript.

Additional informationThe authors declare no competing financial interests. Supplementary information andchemical compound information accompany this paper at www.nature.com/naturechemistry. Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/. Correspondence and requests for materials should be addressedto Y.F. and E.Y.



http://www.nature.com/sifinder/10.1038/nchem.649

http://www.nature.com/cifinder/10.1038/nchem.649



http://npg.nature.com/reprintsandpermissions/







Direct transformation of graphene to fullereneAndrey Chuvilin1,2*, Ute Kaiser1, Elena Bichoutskaia3, Nicholas A. Besley3 and Andrei N. Khlobystov3*

Although fullerenes can be efficiently generated from graphite in high yield, the route to the formation of thesesymmetrical and aesthetically pleasing carbon cages from a flat graphene sheet remains a mystery. The most widelyaccepted mechanisms postulate that the graphene structure dissociates to very small clusters of carbon atoms such as C2,which subsequently coalesce to form fullerene cages through a series of intermediates. In this Article, aberration-correctedtransmission electron microscopy directly visualizes, in real time, a process of fullerene formation from a graphene sheet.Quantum chemical modelling explains four critical steps in a top-down mechanism of fullerene formation: (i) loss ofcarbon atoms at the edge of graphene, leading to (ii) the formation of pentagons, which (iii) triggers the curving ofgraphene into a bowl-shaped structure and which (iv) subsequently zips up its open edges to form a closedfullerene structure.

Over the past two decades, many different models have beenproposed to explain the formation of fullerene from graphite.The generally accepted mechanisms can be categorized into

four major groups according to the exact route leading to the full-erenes: the ‘pentagon road’1–3, the ‘fullerene road’4, ‘ring coalesc-ence’5,6 and the ‘shrinking hot giant model’7,8. All of these can beclassified as bottom-up mechanisms, because fullerene cages areconsidered to be formed from atomic carbon or small clusters ofcarbon atoms. Although there is a large body of experimental evi-dence supporting bottom-up mechanisms9–11, it is almost entirelybased on mass spectrometry and its variants, which analyseonly those species present in the gas phase. These experimentsprovide no direct structural information about the precursorsof fullerenes and do not allow fullerene formation to be followedin situ. Any process of fullerene assembly on the surface ofgraphite, for example, would be overlooked by the traditionalexperimental methodology.

An atomically thin single sheet of graphite, so-called graphene12,represents an ideal viewing platform for molecular structures usingtransmission electron microscopy (TEM), because it provides arobust and low-contrast support for molecules and other nanoscalespecies adsorbed on the surface. Under TEM observation whileexposed to an 80-keV electron beam (e-beam), the edges of the gra-phene sheet appear to be continuously changing in shape (Fig. 1a;see also Supplementary Video). The high energy of the e-beam,when transferred to the carbon atoms of the graphene, can causefragmentation of large sheets of graphene into smaller flake-likestructures (Fig. 1c). The flakes adsorbed on the graphene substratecan be visualized and their further transformations readily observedin TEM. The final product of these transformations is often a perfectfullerene molecule (Fig. 1b).

The sequence shown in Fig. 1c–h presents a typical transform-ation route for an individual graphene flake, which changes itsshape under the influence of the e-beam, becoming increasinglyround (Fig. 1c–e). The contrast of its edges gradually increases(Fig. 1f), indicating that the edges of the flake come progressivelyout of plane and rearrange into the spheroidal shape of a fullerene(Fig. 1g). The experimentally observed images can be relatedthrough TEM image simulation to models of a graphene flake(Fig. 2b′), curved graphene intermediates (Fig. 2d′ and e′), and

the resultant fullerene molecule (Fig. 2f ′ ) adsorbed on the graphenesubstrate. Once fullerene formation is complete, the moleculeappears to roll back and forth on the underlying graphene(Fig. 1h). This is possible, because the energy of the van derWaals interaction of a fullerene with the substrate is significantlyreduced compared to a flat graphene flake13,14 due to the reducedsurface area in contact with the underlying graphene sheet.

Loss of carbon atoms at the edge of graphene is a key initial step inthe graphene-to-fullerene transformation. Carbon atoms at the edgeof a graphene flake are labile, because only two bonds connect themto the rest of the structure. Our density functional theory (DFT)calculations (for details see the Methods section) show that theloss of a carbon atom at the zigzag edge of a small graphene flakeand the subsequent relaxation of the structure require 5.4 eV.As expected, it is approximately one-third less than the energy ofcarbon atom loss from the middle of the flake, which is estimatedto be 7.4–7.6 eV (refs 15–17). Indeed, recent TEM experimentshave demonstrated that the edge atoms of graphene can bechipped away, one by one, by the e-beam18,19. Followingthe removal of one or several carbon atoms, the graphene edgeundergoes structural reconstruction, normally leading to the moststable zigzag configuration18. The loss of carbon atoms at the edgeand the subsequent reconstruction do not cause any significantchanges to the structure of a large graphene sheet18,19

(Supplementary Fig. S1). However, small fragments of graphene,as our extended observations demonstrate (SupplementaryFig. S2), undergo drastic structural transformations under thee-beam, leading to the formation of fullerene cages (Fig. 1c–h).These observations provide direct evidence (unlike massspectrometry) that a fullerene can be formed directly from graphenewithout the need for dissociation to small carbon clusters as isinherent to other mechanisms of fullerene formation1,2,20.

The experimental TEM images provide compelling evidence forthe graphene-to-fullerene transformation. However, the exactpathway of this process can be best explained through quantumchemical modelling of the key stages of the process (Fig. 2a–f).The structures used for modelling represent one possible flakethat is as close as possible to the experimental observations(Fig. 1c–h). Induced by the high-energy e-beam, the initial loss ofcarbon atoms at the edge destabilizes the structure (Fig. 2b),

1University of Ulm, Central Facility of Electron Microscopy, Electron Microscopy Group of Materials Science, Albert Einstein Allee 11, 89069 Ulm, Germany,2IKERBASQUE, Basque Foundation for Science, E-48011, Bilbao, Spain, 3School of Chemistry, University of Nottingham, University Park, NottinghamNG7 2RD, UK. *e-mail: [email protected]; [email protected]








1 nm

a

c d e f g h

b

1 nm

Figure 1 | Experimental TEM images showing stages of fullerene formation directly from graphene. a, The black arrow indicates a double layer of graphene,

which serves as the substrate. The white arrow indicates a strip of graphene (monolayer) adsorbed on this substrate. The dashed white line outlines a more

extended island of graphene mono- or bi-layer, which has its edges slightly curved on the left side. b, The final product of graphene wrapping: a fullerene

molecule on the surface of graphene monolayer (carbon atoms appear as black dots). c–h, Consecutive steps showing the gradual transformation of a small

graphene flake (c) into fullerene (g,h). The graphene lattice is filtered out of images c–h for clarity. (See Supplementary Video for a demonstration of the

dynamics of the entire process.).

+0.201 eV/atom +0.197 eV/atom +0.178 eV/atom

+0.059 eV/atom

–0.261 eV/atom

+0.1

+0.2

+0.3

–0.1

–0.2

–0.3

E(eV/atom)

(a)

Graphene

(b)

b’ e’d’ f’

Etchedgraphene

(c)

Flat graphenewith pentagons

(d)Curved

graphenewith pentagons (e)

(f)

Bowl-shapedstructure

Fullerene

a b c d e f

Figure 2 | Quantum chemical modelling of the four critical stages of fullerene formation from a small graphene flake. a–d, Loss of carbon atoms at the

edge (a � b); formation of pentagons (b � c); curving of the flake (c � d); formation of new bonds, leading to zipping of the flake edges (d � e). e, Top

and side views of a bowl-shaped intermediate structure. Stabilization energies (in eV per carbon atom) of the intermediate structures and resultant fullerene

C60 (f), relative to the flat defect-free flake of graphene shown in a, are presented pictorially and graphically. b′,d′–f′, Top views of the graphene flake (b′),

curved graphene intermediates (d′,e′) and the fullerene C60 molecule (f′) adsorbed on the underlying graphene substrate and simulated TEM images

corresponding to each structure, showing how they would appear in TEM experiments.






because it increases the number of dangling bonds at the edge. Theformation of pentagons at the edge (Fig. 2c) and subsequent curvingof the flake are thermodynamically favourable processes, becausethey bring covalently deficient carbon atoms close to one another(Fig. 2d), thus enabling them to form bonds. New carbon–carbonbonds are then formed, leading to zipping of the flake edges andreducing the number of dangling bonds. This has a profound stabi-lizing effect on the structure (Fig. 2e). In a similar way, the bowl-shaped structure (Fig. 2e) can evolve further by losing carbonatoms from its remaining open edge through e-beam etching,forming more pentagons and curling until the structure is suffi-ciently small to close up into a cage (Fig. 2f ). Fullerene is themost stable configuration for a finite number of sp2 carbon atomsbecause the molecular cage has no open edges (that is, etching isprohibited), and all the carbon atoms form three bonds. If the struc-ture of the newly formed fullerene does not correspond to the moststable isomer, its structure can ‘anneal’ by means of a series ofStone–Wales rearrangements1 that are facilitated by the e-beam inTEM or by an add-atom under the standard fullerene synthesisconditions21,22.

A theoretical study exploring the possibility of the transform-ation of a graphene sheet into a fullerene confirms that the for-mation of defects at the edge of graphene is the crucial step in theprocess20. The structural defects considered in this earlier studyare based on a series of rearrangements that give rise to pentagonalrings. However, the energy barrier for such rearrangementsappeared to be extremely high, making this pathway plausibleonly at extremely high temperatures (3,500 K)20, significantlyhigher than the temperatures used in fullerene production. Our cal-culations, however, show that the loss of the outermost carbonatoms in a graphene flake provides a viable route for fullereneformation under realistic experimental conditions.

The initial size of the graphene flake is important, because itdetermines the size of the fullerene cage that can be formed. If theflake is too large, in the region of several hundreds of carbonatoms, there will be a significant energetic penalty during thecurving step (Fig. 2d) associated with the van der Waals interactionsbetween the underlying graphene sheet and the flake. Its edges willcontinue to be etched until the flake reaches a size that enables thethermodynamically driven formation of fullerene described above.On the other hand, the transformation of very small flakes (lessthan 60 carbon atoms) into fullerenes will be suppressed by exces-sive strain on C–C bonds imposed by the high curvature of smallfullerene cages and the violation of the isolated pentagon rule in full-erenes smaller than C60. Indeed, our experiments indicate that full-erene cages formed directly from graphene have a relatively narrowrange of diameters averaging �1 nm, which corresponds to 60–100carbon atoms (Supplementary Fig. S2). This observation is in agree-ment with the consistent observation of a disproportionately highabundance of C60 and C70 fullerenes found in the differentmethods of fullerene production.

Our in situ TEM experiments correlated with quantum chemicalmodelling demonstrate that a direct transformation of flat graphenesheets to fullerene cages is possible. Etching of edge carbon atoms bythe e-beam facilitates the formation of curved graphene fragments,which continue to be etched until it becomes possible for them tozip up into a fullerene. Previous studies have suggested that apiece of graphene of limited size may not be the most stable allo-trope of carbon23–25, particularly under e-beam radiation, so thelatter stages of this thermodynamically driven process shouldoccur with similar ease to the formation of C60 fullerene from care-fully designed polyaromatic molecules26. Once the edges are sealed,no further carbon atoms can be lost, and the newly created fullereneremains intact under the e-beam.

Could these TEM observations be relevant to real-life fullereneproduction methods, such as arc discharge or laser ablation of

graphite? On the one hand, conditions inside the transmission elec-tron microscope are quite different (for example, no direct heating,high vacuum, no buffer gas), but on the other hand, the e-beamsupplies the energy required to break chemical bonds (just likea laser beam), and the underlying graphite substrate attached to aTEM sample holder absorbs the excess energy released on theformation of new bonds (just like a helium buffer-gas). Indeed, amass spectrometry study27 has shown that C60 and C70 can formalmost exclusively from graphite under 10-keV e-beam radiationin vacuum—conditions very similar to those in TEM—but nomechanism explaining fullerene formation was suggested at thetime. The top-down mechanism of fullerene formation proposedin our study does not exclude the bottom-up models, because differ-ent mechanisms may co-exist under the same experimental con-ditions (Supplementary Fig. S3). For example, in arc discharge,some fullerenes may form in the gas phase from C2 fragmentsand some may form on the surface of graphite electrodes fromsmall flakes of graphene. A laser ablation study28 has shown thatthe yield of fullerene is critically dependent on the orientationand quality of the graphitic surface, which confirms that theformation of fullerene directly on the graphite surface is certainlynot unique for TEM experiments and may have relevance forpreparative methods of fullerene production. We hope that ourstudy will stimulate a reassessment of the current understandingof how fullerenes are formed.

MethodsThin graphite flakes were prepared from spectroscopically pure graphite by grindingin an agate mortar under a layer of ethanol. The dispersion was treated as prepared inan ultrasonic bath and deposited onto a holey carbon TEM grid. Stacks of graphenewith thicknesses varying between one and several layers were observed.

Images were acquired using a Titan 80-300 instrument (FEI) equipped with animaging spherical aberration (Cs) corrector. We used an accelerating voltage of80 kV and Cs optimized Scherzer conditions29 (Cs value, þ20 mm; defocus,23 nm), so the atoms were imaged dark. The exposure time was 1 s per frame, withan interval of 4 s between the frames in one sequence (Supplementary Video).Images of one sequence were aligned by cross-correlation and low-pass-filtered fornoise reduction. The filtering did not affect the final resolution because of significantover-sampling of the original images.

TEM image simulations were performed using MUSLI code30. Coherentaberrations corresponding to those in the experimental images were used.Parameters for the dumping envelope were as follows: focal distance, 1.5 mm(tabulated value for Titan 80-300); coefficient of chromatic aberration, 1.4 mm(measured experimentally); energy spread of the electron source, 0.2 eV (measuredexperimentally); stability of high tension, 1 × 1026 (tabulated value for Titan80-300); stability of objective lens current, 3 × 1027 (fitted by simulations);convergence semi-angle, 0.5 mrad (this parameter does not measurably influenceaberration-corrected imaging). Thermal vibrations were treated using the frozenphonons approach, with 100 phonon configurations averaged for every image at acorresponding Debye–Waller factor of 0.005 nm2. The sampling rate was0.017 nm pixel21. Images were calculated at an electron dose of 1 × 106 e2 nm22

and further processed using the same routine as for the experimental images(see above).

In theoretical calculations of the geometries and energies of the intermediates,the description of a small graphene flake, which initially contained 117 carbon atomsand was subsequently reduced to 86 and 84 atoms, was based on the DFT formalismusing the Q-CHEM quantum chemistry package31 with the B3LYP exchange-correlation functional32 and 6-31G* basis set.

Dispersive van der Waals interactions between the graphene flake and theunderlying graphene sheet contribute up to �30% reduction in its stabilizationenergy, depending on the size and curvature of the fragment. The energy of thesenon-covalent interactions was estimated using the empirical Girifalco potential,which has been successfully applied to describe the interactions between graphiticnanostructures13.

Received 5 November 2009; accepted 19 March 2010;published online 9 May 2010

References1. Smalley, R. E. Self-assembly of the fullerenes. Acc. Chem. Res. 25, 98–105 (1992).2. Goroff, N. S. Mechanism of fullerene formation. Acc. Chem. Res. 29,

77–83 (1996).3. Kroto, H. W. & McKay, K. The formation of quasi-icosahedral spiral shell

carbon particles. Nature 331, 328–331 (1988).






4. Heath, J. R. Synthesis of C60 from small carbon clusters: a model based onexperiment and theory. ACS Symp. Ser. 481, 1–23 (1991).

5. Hunter, J. M., Fye, J. F., Roskamp, E. J. & Jarrold, M. F. Annealing carbon clusterions—a mechanism for fullerene synthesis. J. Phys. Chem. 98, 1810–1818 (1992).

6. Rubin, Y., Kahr, M., Knobler, C. B., Diederich, F. & Wilkins, C. L. The higheroxides of carbon C8nO2n (n¼3–5): synthesis, characterization and X-ray crystalstructure. Formation of cyclo[n]carbon ions Cn

þ (n¼18, 24), Cn2 (n¼18, 24, 30),

and higher carbon ions including C60þ in laser desorption Fourier transform

mass spectrometric experiments. J. Am. Chem. Soc. 113, 495–500 (1991).7. Irle, S., Zheng, G., Wang, Z. & Morokuma, K. The C60 formation puzzle ‘solved’:

QM/MD simulations reveal the shrinking hot giant road of the dynamicfullerene self-assembly mechanism. J. Chem. Phys. B 110, 14531–14545 (2006).

8. Huang, J. Y., Ding, F., Jiao, K. & Yakobson, B. I. Real time microscopy, kineticsand mechanism of giant fullerene evaporation. Phys. Rev. Lett. 99,175503 (2007).

9. Yannoni, C. S., Bernier, P. P., Bethune, D. S., Meijer, G. & Salem, J. R. NMRdetermination of the bond lengths in C60. J. Am. Chem. Soc. 113, 3190–3192 (1991).

10. Hawkins, J. M., Meyer, A., Loren, S. & Nunlist, R. Statistical incorporation ofcarbon-13 13C2 units into C60 (buckminsterfullerene). J. Am. Chem. Soc. 113,9394–9395 (1991).

11. Ebbesen, T. W., Tabuchi, J. & Tanigaki, K. The mechanistics of fullereneformation. Chem. Phys. Lett. 191, 336–338 (1992).

12. Geim, A. K. & Novoselov, K. S. The rise of graphene. Nature Mater. 6,183–191 (2007).

13. Girifalco, L. A. & Hodak, M. Van der Waals binding energies in graphiticstructures. Phys. Rev. B 65, 125404 (2002).

14. Ulbricht, H., Moos, G. & Hertel, T. Interaction of C60 with carbon nanotubesand graphite. Phys. Rev. Lett. 90, 095501 (2003).

15. El-Barbary, A. A., Telling, R. H., Ewels, C. P., Heggie, M. I. & Briddon, P. R.Structure and energetics of the vacancy in graphite. Phys. Rev. B 68,144107 (2003).

16. Saito, M., Yamashita, K. & Oda, T. Magic numbers of graphene multivacancies.Jpn J. Appl. Phys. 46, L1185–L1187 (2007).

17. Carlsson, J. M. & Scheffler, M. Structural, electronic and chemical properties ofnanoporous carbon. Phys. Rev. Lett. 96, 046806 (2006).

18. Girit, C. O. et al. Graphene at the edge: stability and dynamics. Science 323,1705–1708 (2009).

19. Jia, X. et al. Controlled formation of sharp zigzag and armchair edges in graphiticnanoribbons. Science 323, 1701–1705 (2009).

20. Lozovik, Y. E. & Popov, A. M. Formation and growth of carbon nanostructures:fullerenes, nanoparticles, nanotubes and cones. Uspekhi Fizicheskikh Nauk 167,751–774 (1997).

21. Eggen, B. R. et al. Autocatalysis during fullerene growth. Science 272,87–90 (1996).

22. Ioffe, I. N. et al. Fusing pentagons in a fullerene cage by chlorination: IPR D2C76rearranges into non-IPR C76Cl24. Angew. Chem. Int. Ed. 48, 5904–5907 (2009).

23. Ugarte, D. Curling and closure of graphitic networks under electron-beamirradiation. Nature 359, 707–709 (1992).

24. Burden, A. P. & Hutchison, J. L. An investigation of the electron irradiation ofgraphite in a helium atmosphere using a modified electron microscope. Carbon35, 567–578 (1997).

25. Fuller, T. & Banhart, F. In situ observation of the formation and stability of singlefullerene molecules under electron irradiation. Chem. Phys. Lett. 254,372–378 (1996).

26. Otero, G. et al. Fullerenes from aromatic precursors by surface-catalysedcyclodehydrogenation. Nature 454, 865–868 (2008).

27. Bunshah, R. F. et al. Fullerene formation in sputtering and electron beamevaporation processes. J. Phys. Chem. 96, 6866–6869 (1992).

28. Xie, Z.-X. et al. Formation and coalescence of fullerene ions from direct laservaporization. J. Chem. Soc. Faraday Trans. 91, 987–990 (1995).

29. Scherzer, O. The theoretical resolution limit of the electron microscope. J. Appl.Phys. 20, 20–29 (1949).

30. Chuvilin, A. & Kaiser, U. On the peculiarities of CBED pattern formationrevealed by multislice simulation. Ultramicroscopy 104, 73–82 (2005).

31. Shao, Y. et al. Advances in methods and algorithms in a modern quantumchemistry program package. Phys. Chem. Chem. Phys. 8, 3172–3191 (2006).

32. Becke, A. D. A new mixing of Hartree–Fock and local density-functionaltheories. J. Chem. Phys. 98, 1372–1377 (1993).

AcknowledgementsThis work was supported by the Engineering and Physical Sciences Research Council(Career Acceleration Fellowship to E.B., grant no. EP/C545273/1 to A.N.K.), the EuropeanScience Foundation, the Royal Society, the DFG (German Research Foundation) and theState Baden-Wurttemberg within the SALVE (Sub Angstrom Low Voltage ElectronMicroscopy) project and by the DFG within Collaborative Research Centre (SFB) 569.

Author contributionsA.C. conceived, designed and carried out experiments. U.K. contributed to thedevelopment of the experimental methodology and the discussion of the results. E.B. andN.A.B. performed theoretical modelling and contributed equally to this work. A.N.K.proposed the mechanism and wrote the original manuscript. All authors discussed theresults and commented on the manuscript.

Additional informationThe authors declare no competing financial interests. Supplementary informationaccompanies this paper at www.nature.com/naturechemistry. Reprints and permissioninformation is available online at http://npg.nature.com/reprintsandpermissions/.Correspondence and requests for materials should be addressed to A.C. and A.N.K.











Lattice-strain control of the activity in dealloyedcore–shell fuel cell catalystsPeter Strasser1,2*, Shirlaine Koh2, Toyli Anniyev3,4, Jeff Greeley5, Karren More6, Chengfei Yu2,

Zengcai Liu2, Sarp Kaya3,4, Dennis Nordlund4, Hirohito Ogasawara3,4, Michael F. Toney3,4

and Anders Nilsson3,4

Electrocatalysis will play a key role in future energy conversion and storage technologies, such as water electrolysers, fuelcells and metal–air batteries. Molecular interactions between chemical reactants and the catalytic surface control theactivity and efficiency, and hence need to be optimized; however, generalized experimental strategies to do so are scarce.Here we show how lattice strain can be used experimentally to tune the catalytic activity of dealloyed bimetallicnanoparticles for the oxygen-reduction reaction, a key barrier to the application of fuel cells and metal–air batteries. Wedemonstrate the core–shell structure of the catalyst and clarify the mechanistic origin of its activity. The platinum-richshell exhibits compressive strain, which results in a shift of the electronic band structure of platinum and weakeningchemisorption of oxygenated species. We combine synthesis, measurements and an understanding of strain from theory togenerate a reactivity–strain relationship that provides guidelines for tuning electrocatalytic activity.

Electrocatalytic energy-conversion processes are expected to playa major role in the development of sustainable technologies thatmitigate global warming and lower our dependence on fossil

fuels. More specifically, fuel cells that use polymer electrolyte mem-branes (electrochemical energy-conversion devices) are potentiallyuseful as power sources in the transportation sector, one of thelargest emitters of greenhouse gases and consumers of fossil fuels.Efficient and stable fuel-cell electrocatalysts, however, are largelyunavailable1–3, and so fundamental progress in the design of thesecatalysts is needed.

The ultimate goal in catalytic design is to have complete syntheticcontrol of the material properties that determine the reactivity4,5.Catalysts that consist of two metals (bimetallic) allow greater reactivityand more flexible design, and recent studies have focused on these cat-alysts6–14. There are three fundamental effects in bimetallic catalysis:ensemble, ligand and geometric. Ensemble effects arise when dissim-ilar surface atoms, individually or in small groups (ensembles), take ondistinct mechanistic functionalities, as demonstrated for palladiumatom pairs on gold for a gas-phase catalytic reaction7,15. Ligandeffects are caused by the atomic vicinity of two dissimilar surfacemetal atoms that induces electronic charge transfer between theatoms, and thus affects their electronic band structure. Finally, geo-metric effects are differences in reactivity based on the atomic arrange-ment of surface atoms and may include compressed or expandedarrangements of surface atoms (surface strain)16. Ligand and geometriceffects1,2,8,15,17–20 and, in some cases, all three effects15, are generallysimultaneously present and coimpact the observed catalytic reactivity.To date, however, no effective strategy to isolate and tune strain effectsin electrocatalytic systems has been achieved. Considering the dimen-sions across which ensemble, geometric strain and ligand effects areeffective, only geometric strain can impact surface reactivity over

more than a few atomic layers. Hence, a catalyst structure that consistsof a few atomic monometallic layers, supported on a substrate withdifferent lattice parameters (a core–shell structure), should isolate geo-metric strain effects. If the amount of strain in these structures canbe controlled, we could use the rich effects of bimetallic catalysis totune surface catalytic reactivity continuously.

The preferential dissolution (removal) of the electrochemicallymore reactive component from a bimetallic alloy (precursor) thatconsists of a less reactive (here Pt) and more reactive metal (hereCu)21–23 is commonly referred to as ‘dealloying’. We showed thatdealloyed Pt–Cu nanoparticles have uniquely high catalytic reactivityfor the oxygen-reduction reaction (ORR) in fuel cell electrodes24–28,which is of tremendous importance as it occurs at the cathode ofvirtually all fuel cells29,30, with pure Pt as the preferred catalyst.The ORR is sluggish, which is why significant amounts of Ptmetal are required in fuel cells, which makes them prohibitivelyexpensive. Dealloyed Pt catalysts, however, meet and exceed thetechnological activity targets in realistic fuel cells24, as shown inSupplementary Fig. S1. Owing to their reactivity, dealloyed Pt cata-lysts can reduce the required amount of Pt by more than 80%.Despite such importance, the mechanistic origin of their enhancedreactivity remains poorly understood. To realize similar propertyimprovements in related electrocatalytic systems, the principlesthat underlie their performance must be elucidated. Here, wedemonstrate that the concept of strain tuning the catalytic proper-ties, introduced above, provides such a unifying principle.

ResultsWe studied six different Pt–Cu alloy nanoparticle precursors andtheir corresponding dealloyed counterparts. We used alloy precur-sors with various initial atomic Pt/Cu ratios and preparation, in

1The Electrochemical Energy, Catalysis, and Materials Science Laboratory, Department of Chemistry, Chemical Engineering Division, Technical UniversityBerlin, 10623 Berlin, Germany, 2Department of Chemical and Biomolecular Engineering, University of Houston, Houston, Texas 77204, USA, 3StanfordInstitute of Materials and Energy Sciences, SLAC National Accelerator Laboratory, Menlo Park, California 94025, USA, 4Stanford Synchrotron RadiationLightsource, SLAC National Accelerator Laboratory, Menlo Park, California 94025, USA, 5Center for Nanoscale Materials, Argonne National Laboratory,Argonne, Illinois 60439, USA, 6Materials Science & Technology Division,’ Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831-6064, USA.

*e-mail: [email protected]

ARTICLESPUBLISHED ONLINE: 25 APRIL 2010 | DOI: 10.1038/NCHEM.623






particular the annealing temperature. We focused on two sets ofthree Pt/Cu ratios (Pt25Cu75, Pt50Cu50 and Pt75Cu25); one set wasannealed at 800 8C and the other at 950 8C. To obtain the activedealloyed catalysts, the precursors were subjected to an identicalelectrochemical dealloying protocol in which Cu was removed pre-ferentially from the precursor particles.

We first address the structure of the Pt–Cu bimetallic precur-sors and the corresponding dealloyed catalysts. These form face-centred cubic disordered alloys31. Figure 1a,b shows energy-dispersiveelemental colour-map overlays for Pt and Cu, acquired using aprobe-corrected scanning transmission electron microscope(STEM), for the Pt25Cu75 800 8C alloy precursor and the dealloyedcatalyst, respectively; the mean particle size prior to dealloying(Fig. 1a) was around 4.5 nm (see Supplementary Fig. S2e), but thedealloyed, catalytically active form (Fig. 1b) exhibited a decreasein the average particle size to �3.4 nm (Supplementary Fig. S2e).The corresponding Pt25Cu75 950 8C alloy precursor and dealloyedcatalysts had average particle sizes of 6.0 nm and 5.1 nm, respect-ively. As Fig. 1b shows, for the dealloyed catalyst, Cu is confinedto the centre of the majority of the dealloyed Pt–Cu nanoparticles.The dealloyed particles exhibit a distinct Pt-enriched layer (blue inFig. 1b) on the surfaces of the alloy Pt–Cu cores (pink in Fig. 1b).A corresponding energy-dispersive spectroscopy (EDS) line profiletaken across the diameter of typical (�4 nm) dealloyed nano-particles confirmed the presence of a �0.6 nm, Pt-enriched layeron the surface of the dealloyed particles (Fig. 1c). Hence, the Ptand Cu elemental maps and line profiles shown in Fig. 1 provideevidence for the formation of a core–shell structure in the nano-particles after dealloying, in which a Pt-enriched shell surrounds aPt–Cu core.

Furthermore, aberration-corrected, high-angle annular dark-field(HAADF) STEM images indicate a change in the Pt and Cu distri-butions within the nanoparticles, from a uniform Pt–Cu alloy(Fig. 2a) to a morphology suggestive of a core–shell structure (Fig. 2b)32.

Further evidence comes from X-ray photoelectron spectroscopy(XPS) data, which show a large enhancement in the surfaceconcentration of Pt (Table 1). Based on the STEM, EDS and XPSdata, we estimate the thickness of the Pt shell to be 0.6–1.0 nm,which corresponds to three or more Pt-rich layers, consistent withestimates from anomalous small-angle X-ray scattering of similarmaterials33. The propensity for Cu to dissolve means that dealloyed

a b

0

20

40

60

80

100

0.5 1 1.5 2 2.5 3 3.5 4 4.5

CuPt

Inte

nsity

(c.

p.s.

)

Distance (nm)

c

10 nm 10 nm

Figure 1 | Elemental maps and line profiles of Pt–Cu bimetallic nanoparticle precursors and dealloyed active catalysts. a,b, High-resolution EDS (HR-EDS)

elemental maps of a Pt25Cu75 bimetallic nanoparticle alloy precursor (a) and of the active electrocatalyst after Cu dealloying from the precursor (b). Pt is

given in blue, Cu in red and pink domains indicate well-alloyed Pt–Cu domains. The precursor alloy was prepared at 800 8C and is supported on carbon

black of high surface area. The dealloyed catalyst was obtained by voltammetric cycling of the precursor between 0.05 V and 1.2 V (with respect to a

reversible hydrogen electrode) in 0.1 M HClO4 solution. Pt–Cu alloy domains are discernible in the centres of the dealloyed particles with a Pt-enriched

surface. EDS elemental maps were acquired using a probe-corrected STEM and were extracted from spectrum images acquired from an area that measured

40 nm × 40 nm, with a beam size of �0.2 nm and a step (pixel) size of �0.2 nm. c, HR-EDS line profile across an individual �4 nm diameter dealloyed

Pt–Cu alloy active-catalyst particle. Data were acquired using a probe-corrected STEM with a probe diameter �0.2 nm and step size �0.2 nm. The thickness

of the Pt-enriched particle shell was �0.6 nm, as shown by the dashed lines. c.p.s.¼ cycles per second.

2 nm

2 nm

b

a

Figure 2 | HAADF-STEM images of Pt–Cu bimetallic nanoparticle

precursors and dealloyed active catalysts. a,b, Sub-angstrom resolution

images of individual (�4 nm) Pt–Cu nanoparticles that show a typical Pt–Cu

precursor alloy nanoparticle (a) and a typical Pt–Cu dealloyed nanoparticle

(b). The dealloyed nanoparticle exhibits an outer Pt-enriched shell (outer

shell images brightly) and a Pt–Cu alloy core (which images less brightly

than the shell). Contrast variations in the HAADF-STEM images result from

the atomic number (Z) difference between Pt and Cu (commonly referred to

as Z-contrast imaging). The thickness of the Pt-enriched particle shell in (b)

is �0.6 nm.






Pt50Cu50 and Pt75Cu25 nanoparticles possess a similarly Pt-enrichedshell structure. From these combined, consistent results, we con-clude that the geometry of the dealloyed Pt-enriched shell on thePt–Cu core leads to the high electrocatalytic activity. Specifically,we hypothesize that lattice strain in the Pt shell controls thesurface catalytic reactivity.

We address our lattice-strain hypothesis using anomalous X-raydiffraction (AXRD), which permits independent measurements ofthe average lattice parameter and average composition of the scatter-ing nanoparticles before and after dealloying. In AXRD, diffractionpatterns are collected at a number of X-ray energies near the X-rayabsorption edge of an element of interest (here Cu). Figure 3a shows(111) diffraction profiles of a Pt25Cu75 precursor at several energies.The scattering power of Cu drops near the Cu absorption edge, andso the diffracted intensity of the alloy shows a characteristicdecrease, as indicated by the arrow in Fig. 3a. In AXRD, peakpositions yield average lattice parameters for the nanoparticles,and the relationship between scattering intensity and energy(Supplementary equation (S2)) provides the chemical compositionof the scattering alloy phase. As illustrated in Fig. 3b, this modelequation for diffraction intensity (black line) was fitted to the inte-grated (111) peak intensities (red circles), and the compositions xPtand xCu were determined. Supplementary Fig. S3a provides a directcomparison of the chemical compositions, xPt, of the alloy precur-sors derived by AXRD and by inductively coupled plasma opticalemission spectroscopy.

Given the core–shell nature of the dealloyed particles, the latticestrain in the Pt shell is most relevant for surface catalysis. To esti-mate the lattice parameter in the particle shell, we approximatedthe structure of the dealloyed particles by a simple two-phasecore–shell model, as shown schematically in Fig. 4a. We assumeda pure Pt shell with lattice parameter ashell that surrounded a

Pt–Cu alloy core with lattice parameter acore. Using the AXRD-derived nanoparticle compositions and lattice-parameter data34–36,our core–shell model allowed the determination of ashell. Thestrain, s(Pt), in the particle shell relative to bulk Pt, is given byequation (1), where aPt is the bulk Pt lattice parameter.

s(Pt) = ashell − aPt

aPt

× 100 (1)

Figure 4b shows ashell as a function of precursor composition andpreparation temperature, and s(Pt) is shown in SupplementaryFig. S4. Foremost, the data show that for all catalysts the lattice par-ameter of the Pt shell, ashell , is smaller than that of pure Pt (dottedline, Fig. 4b) and so is strained compressively. With increasing Cu inthe alloy precursor and with higher preparation temperature, ashellbecomes smaller and the magnitude of s(Pt) larger. The observedsynthesis–strain trends are understood within our core–shellmodel. The lattice mismatch between the Pt shell and the Pt–Cucore causes a reduced Pt–Pt interatomic distance in the shell. Thericher in Cu the particle core, the smaller its lattice parameter,and hence the more strain induced in the shell. A similar argumentholds for the increased strain in the high-temperature materials; forthese, the bimetallic precursor phase is alloyed more uniformly withless residual, unalloyed Cu (ref. 25), which effectively makes thealloy phase richer in Cu. Both experimental and computationalwork suggest that lattice contraction induced by particle size(surface stress) in small Pt particles only becomes significantbelow a particle diameter of 2.5 nm (refs 37,38) and can thereforebe ruled out as source of strain for the dealloyed core–shell particles.

To gain insight into how the strain of the dealloyed catalysts affectsthe catalytic surface reactivity, we studied the electronic band struc-ture. The d-band model developed by Nørskov and co-workers hasbeen successful in relating the adsorption properties of rate-limitingintermediates in catalytic processes to the electronic structure of thecatalyst39–41. For simple adsorbates, such as the ORR intermediatesO and OH, this can be understood in a simple electron-interactionmodel in which the adsorbate valence p-level forms bonding and anti-bonding states with the metal d-band40,41. Population of any antibond-ing state leads to Pauli repulsion, and the bond strength is therebyweakened. A downward shift of the d-band pulls more of the anti-bonding states below the Fermi level, which results in increasing occu-pation and weaker adsorbate bonding.

2.6 2.8 3.0 3.2 3.4 3.6

(200)

8,800 eV8,890 eV8,980 eV8,990 eV (Cu edge)9,060 eV9,150 eV

Inte

nsity

(a.

u.)

Q (A–1)

(111)

8,800 8,900 9,000 9,100

800

900

1,000

|F|2

Energy (eV)

Fit

Experiments

a b

Figure 3 | AXRD-based structural and phase-composition analysis of Pt–Cu bimetallic nanoparticle precursors and dealloyed nanoparticle catalysts.

a, AXRD intensity profiles of a Pt25Cu75 alloy precursor as function of the scattering vector Q. Diffraction profiles were taken as a function of the X-ray

energy across the X-ray absorption edge of Cu at 8,990 eV. Scattering intensities of Bragg reflections (shown are the (111) and (200) reflections) decrease

(arrow) as the Cu absorption edge energy is approached. The extent of intensity decrease correlates with the Cu content of the scattering phase and can be

used to extract composition. b, Fit (black line) of the relation (Supplementary equation (S2)) between the square of the scattering amplitude |F|2 and incident

X-ray energy to experimental values (red circles) to extract the molar fractions of Pt and Cu in the scattering phase. AXRD provided the lattice parameter of

the scattering phase (structural information) as well as its actual composition (chemical information). a.u.¼ arbitrary units.

Table 1 | Composition depth profiles for Pt25Cu75.

Photoelectron energy (eV)

264 616 1,133 1,480 8,000

Pt25Cu75 precursor (Pt at %) 8 12 19 18 31Pt25Cu75 dealloyed (Pt at %) 84 68 55 56 59

Composition depth profiles for Pt25Cu75 nanoparticles annealed at 950 8C before and afterdealloying. Increasing photon energy correlates with increased probing depth. Pt at%¼ platinumatomic per cent.






We extended and applied these ideas to single-crystal Pt surfaces bypreparing and characterizing bimetallic single-crystal model surfacesthat consist of atomic layers of Pt with various thicknesses grown ona Cu(111) substrate. This model mimics the structural and electronicenvironment of Pt layers that surround a particle core with signifi-cantly smaller lattice parameters, similar to the dealloyed Pt–Cu cata-lysts42. Figure 5a shows the photoelectron spectra of the valence bandof Pt in Pt(111) and for five monolayers (5 ML) of Pt grown onCu(111), measured at grazing electron emission to enhance thesurface sensitivity. There is no detectable Cu signal in the 5 ML Pt spec-trum, and it thereby represents pure Pt with no ligand effect from theunderlying Cu substrate. Based on low-energy electron diffraction(LEED) probing of the lattice parameter during the growth of Pt onCu(111), 5 ML Pt should give about 2.5+0.3% compressive strain,as shown in Supplementary Fig. S6. The spectrum of Pt on Cu(111)shows a much broader d-band compared to that of the Pt(111)surface, and the d-band centre is downshifted from 2.87 eV to 3.26 eVbelow the Fermi level. The broadening is related directly to the com-pressive strain, because the electronic state overlap between the metalatoms increases with shorter interatomic distances; furthermore,keeping the d-occupancy constant for a pure metallic system givesrise to a downward shift of the d-band centre43.

Next, we used X-ray spectroscopy to monitor directly the positionand atom-specific occupation of the oxygen 2p and Pt 5d antibondingstates projected onto the oxygen atom40,41. Figure 5b shows the K-edgeX-ray emission spectra (XES) and X-ray absorption spectra (XAS) ofatomic oxygen adsorbed on thin films of Pt on Cu(111) and Pt(111) toshow the occupied and the unoccupied electronic states, respectively.For oxygen on Pt(111), we observed a broad occupied bonding state inthe XES spectrum and an intense resonance related to the antibondingstate in the XAS spectrum. For the two Pt films on Cu(111), whichcorrespond to strains of around 2.8+0.3% and 3.3+0.3% (ref. 42),we observed a decrease in the intensity of the antibonding resonancewith increasing strain. Of primary interest is that, for the film withmaximum strain, the antibonding resonance vanishes in the XASspectrum and is resolved in the XES spectrum, which directly indi-cates that the antibonding state is occupied fully with a peak around1.5 eV below the Fermi level.

Pt shell

Pt–Cu core

acore

ashell

12

a b

25 50 753.70

3.75

3.80

3.85

3.90

3.95

Uni

t cel

l par

amet

er, a

shell (

Å)

Precusor Cu content (%)

Bulk Pt

Figure 4 | A simple structural two-phase core-shell model for the dealloyed nanoparticles and evaluation of their lattice parameters. a, Scheme of a

simple two-phase structural model for the dealloyed state of a bimetallic particle. Pure Pt layers surround an alloy particle core (ashell and acore represent the

mean lattice parameters in shell and core, respectively). Deviations of ashell from the bulk unit-cell parameter of pure Pt (aPt) imply strain, s(Pt). Shell and core

regions possess a face-centred cubic structure and are assumed to be structurally and compositionally uniform. Using a simple linear model (Supplementary

equation (S3)), ashell can be estimated without further assumptions about the volume of the core or shell. As Pt–Cu alloys exhibit unit-cell parameters that are

smaller than those of pure Pt, the model predicts ashell , aPt; that is, compressive strain in the shell. Unlike the model, a real core–shell particle shows a

gradient in lattice parameters, with strain in the surface layer (point 1) being less than in layers near the core (point 2). b, Determination of ashell for dealloyed

Pt–Cu bimetallic particles, plotted as a function of the alloy precursor Cu atomic composition at precursor annealing temperatures of 950 8C (blue) and

800 8C (red). Dealloyed catalysts derived from precursors richer in Cu or precursors annealed at higher temperatures display a smaller lattice parameter in

the shell. All alloys show lattice parameters below that of pure bulk Pt (dotted line). Error bars refer to resulting uncertainty involved in obtaining ashell from

experiments and relations (Supplementary equations (S2), (S3), (S17)).

10 5 0 –5

Binding energy (eV)

2.6 ML

Pt(111)

3.5 ML

Pt/Cu(111)

0%

XAS

Inte

nsity

(a.

u.)

Inte

nsity

(a.

u.)

Strain

3.3%

2.8%

XES Adsorbed oxygen

Bonding Antibonding

b

8 6 4 2 0

Binding energy (eV)

Cu(111)

5 ML Pt/Cu(111)Pt(111)

e–

15°

a

Figure 5 | Surface-science XAS and XES studies of single-crystal model

systems that mimic dealloyed bimetallic core–shell structures.

a, Photoelectron spectra of the valence-band region for Cu(111), Pt(111) and

5 ML of Pt on Cu(111) measured with a photon energy of 620 eV and with

a 158 grazing electron-emission angle. The inelastic background has been

subtracted. The spectra of the 5 ML Pt on Cu(111) is dominated completely

by the emission from Pt because no sharp Cu d-band emission is seen at

2.3 eV from the underlying substrate (similar spectra are also found for 3 ML

of Pt). b, In-plane polarized oxygen K-edge XAS and normal emission oxygen

K-edge XES of 0.2 ML of oxygen chemisorbed on 3.6 ML and 2.6 ML films

of Pt on Cu(111) and on bare Pt(111). The XAS were normalized with respect

to the high-energy region, where the cross-section is dominated by atomic

effects related to the photoionization continuum, whereas the absolute

intensity scaling between the XES and XAS spectra is arbitrary. The energy

scale is with respect to the oxygen 1s binding energy, which represents the

Fermi level38. The strain was estimated based on the Pt coverage and using

the LEED data shown in Supplementary Fig. S6.






To quantify the relationship between surface strain, O and OHbinding energies and the catalytic ORR reactivity, we carried outdensity functional theory (DFT) calculations to predict the changes inthe Pt–O surface bond energy for a strained Pt(111) model surface.Our computations showed a linear relationship between lattice strainand the adsorbate bond energy, consistent with the experimental X-ray spectroscopy data and previous computational analyses16,44. Bycombining the strain-bond energy relationships with a microkineticmodel, originally developed by Hammer and Nørskov39, for the electro-reduction of oxygen18,45, we derived a ‘volcano’ relation between the pre-dicted ORR rate and the strain (the dashed line in Fig. 6). The volcanoshape implies that compressive strain first enhances the overall ORRactivity by reducing the binding energy of intermediate oxygenatedadsorbates and, thereby, lowers the activation barriers for proton- andelectron-transfer processes. Beyond a critical strain, however, thebinding becomes too weak and the catalytic activity is predicted todecrease because of an increased activation barrier for either oxygen dis-sociation or the formation of a peroxyl (OOH) intermediate39.

We also plotted (Fig. 6) the experimentally measured ORR electro-catalytic activities for our Pt–Cu catalysts as a function of s(Pt) (elec-trochemical currents are reported in Supplementary Table S1). Wedid not, however, observe a decrease in the experimental activityvalues on the left side of the volcano curve, as predicted by theory;this is probably related to compressive-strain relaxation in the Ptshells. Pt atoms adjacent to the Pt–Cu cores (Fig. 4a, point 2) adopta lattice parameter closer to that of the cores32, but outer Pt shellatoms (Fig. 4a, point 1) relax towards the lattice constant of bulk Pt(ref. 42). Hence, the surface strain is less than that represented byashell, which is an average strain in the Pt shell; if the surface strainwas plotted in Fig. 6, we would expect a shift of all experimentaldata points to the right. Furthermore, it is difficult to prepare dealloyed

nanoparticles of sufficiently high surface strain to access the truemaximum of the volcano. When the strain passes a critical point,surface relaxation probably relieves further strain and thereby limitsthe accessibility of the high-strain side of the volcano. Figure 6 alsoreveals that the set of dealloyed nanoparticle catalysts prepared atthe higher annealing temperature exhibits reduced activity at com-parable lattice strain in the particle shells, presumably because ofdifferences in the mean particle size. Hence, our strain-related con-clusions generally refer to particles of comparable size.

DiscussionDealloyed fuel-cell catalysts show unprecedented electrocatalyticactivity for the electroreduction of oxygen, yet we did not have afundamental understanding of the mechanistic origin of the cataly-tic enhancement. Here, we clarify the origin, on an atomic scale, ofthe exceptional electrocatalytic activity of dealloyed Pt–Cu nanopar-ticles. We present microscopic and spectroscopic evidence for theformation of a Pt–Cu alloy core–shell nanoparticle structure usingSTEM elemental maps (Fig. 1), XPS depth profiling (Table 1) andanomalous small-angle X-ray scattering33. Given the thickness ofthe pure Pt shell and considering the limited range of ligandeffects46, we conclude that compressive-strain effects rather thanligand effects are responsible for the exceptional reactivity of theparticle surface. This is in contrast to other ORR electrocatalystsystems, such as Pt monolayer20 or Pt skin17 catalysts, in whichstrain and ligand effects are always convoluted. Using X-ray diffrac-tion, we measured and quantified the presence of compressive latticestrain in the Pt shells of the dealloyed particles.

To further corroborate the lattice-strain hypothesis in core–shellstructures, we studied a surface-science core–shell model system.Our goal was to verify experimentally the predicted effects on theband structure for compressively strained Pt layers. In contrast toprevious reports of correlations in reactivity-band structure19, inwhich band-broadening and band-centre shifts were adopted fromcomputational predictions, we demonstrated experimentally a con-tinuous change of the oxygen 2p and Pt 5d antibonding state fromabove to below the Fermi level as additional compressive strain wasapplied, which resulted in a weakening of the adsorbate bond. Thisrepresents the first direct experimental confirmation of the compu-tational prediction of band shifts of adsorbate-projected band struc-ture. Finally, we correlated experimental synthesis–strain–activitydata of dealloyed core–shell particles (Fig. 6). The resultingactivity–strain relationships provide experimental evidence thatthe deviation of the Pt-shell lattice parameter from that of bulkPt, that is the lattice strain in the shell, is the controlling factor inthe catalytic enhancement of dealloyed Pt nanoparticles; in particu-lar, these relationships are consistent with computational predic-tions that compressive strain enhances ORR activity.

In conclusion, a coherent picture of the origin of the exceptionalelectrocatalytic reactivity for the ORR of dealloyed Pt–Cu particles isnow established. Strain forms in Pt-enriched surface layers (shells)that are supported on an alloy particle core with a smaller latticeparameter. The compression in the shell modifies the d-band struc-ture of the Pt atoms, and thereby weakens the adsorption energy ofreactive intermediates compared to unstrained Pt and results in anincrease in the catalytic reactivity, consistent with DFT-based pre-dictions. A unique feature of the class of dealloyed catalysts is theexperimental control over the extent of dealloying (shell thickness)and the alloy core composition (the upper limit for strain in theshell). The noble and the non-noble constituents can be adjustedin the alloy precursor, such that both expansive and compressivestrain can be achieved to control the strengthening or weakeningof surface bonds. This enables continuous tuning of catalytic reac-tivity. We demonstrate explicitly such strain-related tuning for theORR, and this phenomenon will probably offer control over theactivity of other important electrocatalytic reactions that require

–4 –3 –2 –1 0

–0.02

0.00

0.02

0.04

0.06

0.08

0.10

0.12

Pt25Cu75

Pt25Cu75

Pt50Cu50

Pt50Cu50

Pt75Cu25

Pt75Cu25

Strain (ashell–aPt)/aPt (%)

Pt

OR

R a

ctiv

ity (

eV)

Figure 6 | Experimental and predicted relationships between

electrocatalytic ORR activity and lattice strain. The experimental ORR

activity (in units of kT ln( js,alloy/js,Pt), T¼ 298 K), based on relative surface

area, of two families of dealloyed Pt–Cu bimetallic core–shell nanoparticles

plotted as a function of strain, s(Pt), in the particle shell (red and blue

triangles denote dealloyed Pt–Cu precursors prepared at annealing

temperatures of 800 8C and 950 8C, respectively). The ORR activity is

proportional to the logarithm of the ratio of the ORR current density of the

dealloyed particles to that of pure Pt nanoparticles; this quantity is the

effective difference in ORR activation energies (‘reaction driving force’). The

experimental curves are upper bounds of the strain in the surface layer. The

dashed line is the DFT-predicted, volcano-shaped trend of the ORR activity

for a Pt(111) single-crystal slab under isotropic strain. Moderate compressive

lattice strain is predicted to enhance the rate of ORR catalysis. Absolute and

relative values are provided in Supplementary Table S1. Vertical error bars

refer to experimental uncertainty involved in measuring catalytic activities

from electrochemical systems; horizontal errors bars refer to errors involved

in obtaining ashell from experiments and modelling.






modification of the adsorption energy of reactive intermediates,such as the electrooxidation of small organic molecules, includingethanol, methanol and related species.

MethodsSynthesis. Pt–Cu binary electrocatalyst precursors were synthesized using a liquidmetal–salt precursor impregnation method followed by freeze-drying and thermalannealing. This method was applied previously to the synthesis of binary25,47 andternary24 Pt alloys. The nanoparticle-catalyst precursors were prepared by addingappropriate stoichiometric amounts of a solid Cu precursor (Cu(NO3)2

.2.5H2O,Sigma-Aldrich) to weighted amounts of commercial Pt electrocatalyst powder, of30 weight per cent platinum nanoparticles supported on carbon with a high surfacearea (TEC10E30E from Tanaka Kikinzoku).

Electrochemical measurements. The voltammetric response of the electrocatalystswas measured first during the initial three cyclic voltammetry (CV) scans at100 mV s21 to obtain the initial rapid Cu-dissolution profiles. The catalysts werefurther pretreated using 200 CV scans between 0.05 V and 1.0 V at a scan rate of500 mV s21, during which a large amount of Cu was lost from the alloynanoparticles. Thereafter, the Pt electrochemical surface area (Pt-ECSA) wasdetermined by cycling the treated catalysts at 100 mV s21 between 0.05 V to 1.2 Vand integrating the Faradaic charge associated with stripping of underpotentiallydeposited hydrogen (Supplementary Fig. S1a). Pt-ECSA measurements using COstripping resulted in comparable values of the surface area.

Linear-sweep voltammetry (LSV) measurements were conducted under an oxygenatmosphere by sweeping the potential from 0.06 V anodically to the open-circuitpotential (�1.0 V) at a scan rate of 5 mV s21 (Supplementary Fig. S1b). The ORRactivities of the dealloyed, activated catalysts were corrected for mass-transportlimitation using equation (5) in Gasteiger et al.3. Specific activities for mass and surfacearea were established at 900 mV at room temperature (Supplementary Fig. S1c,d).

Electron microscopy (STEM and EDS). EDS imaging was carried out in STEMmode at 200 kV with a Philips CM200FEG equipped with an EDAX detector/pulseprocessor and an Emispec Vision system (Supplementary Fig. S2). Pt-L and Cu-Kelemental maps were extracted from the spectral data and Pt–Cu (blue and red,respectively) colour map overlays were produced (Fig. 1a,b). High-angle annulardark-field (HAADF) STEM images of individual Pt–Cu nanoparticles were recordedat sub-angstrom resolution using a JEOL 2200FS Cs-corrected STEM (CEOShexapole aberration-corrector) operated at 200 kV.

Anomalous X-ray diffraction. Synchrotron-based XRD was used to characterizePt-alloy electrocatalyst precursor powders as well as electrochemically treatedactivated catalyst films using X-ray energies from 8,900 eV, through the CuK-adsorption edge (8,979 eV) to 9,150 eV. Diffraction measurements wereconducted at the Stanford Synchrotron Radiation Lightsource (SSRL) beamline 2-1.A detailed description of the analysis of the AXRD results is provided in theSupplementary Information.

Computational methods. Computational analysis was carried out using DACAPO(ref. 48), a total-energy calculation code. All calculations were performed on afour-layer slab with a 2 × 2 unit cell. Full relaxation of the oxygen adsorbate andof the first two metal layers was allowed. For strained Pt slabs, uniform expansion(or contraction) of the Pt(111) lattice was allowed in all three Cartesian directions,and no corrections to the interlayer Pt distance were included. This model providesa reasonable representation of the compression found in the Pt-base metal alloysthat form the substrate of the Pt samples.

X-ray photoelectron, X-ray emission and X-ray absorption spectroscopy. XPS,XES and XAS measurements were performed in an ultrahigh vacuum end-station with abase pressure better than 10–10 torr at beamline 13-2 at SSRL, which contains anelliptically polarized undulator that allows control of the direction of the photon Evector about the propagation direction. An electron energy analyser (VG-ScientaSES-100 or R3000), mounted perpendicular to the incoming light, was used for theXPS measurements. This was also equipped with a partial-yield detector for X-rayabsorption measurements. Samples were mounted on a rotatable sample rod at grazingangle (�58 for the photon incidence angle) with respect to the incoming light. Theindependent rotation of the sample and the photon polarization (E vector) allowed forselection of arbitrary angles between the E vector and the sample surface and any choiceof detection angle with respect to the sample surface. Composition of the catalysts wasdetermined by measuring the ratio of Pt 4f to Cu 3p XPS intensities normalized to theirrespective subshell photoionization cross-sections49 for different photon energies. Thekinetic energy of the photoelectron defines the inelastic mean free path and the probingdepth of the analysis. We varied the photoelectron kinetic energy by changing theincident photon energy to obtain the composition at different probing depths (seeTable 1). The estimated probing depths are 0.6 nm, 1 nm, 1.5 nm, 1.8 nm and 7 nm atphoton energies of 250 eV, 620 eV, 1,130 eV, 1,480 eV and 8,000 eV, respectively.Further experimental details are provided in the Supplementary Information.

Received 27 October 2009; accepted 11 March 2010;published online 25 April 2010

References1. Stonehart, P. Development of alloy electrocatalysts for phosphoric acid fuel cells

(PAFC). Appl. Electrochem. 22, 995–1001 (1992).2. Mukerjee, S. & Srinivasan, S. in Handbook of Fuel Cells—Fundamentals,

Technology and Applications Vol. 2. (eds Vielstich, W., Gasteiger, H. A.& Lamm, A.) 502–519 (John Wiley, 2003).

3. Gasteiger, H. A., Kocha, S. S., Sompalli, B. & Wagner, F. T. Activity benchmarksand requirements for Pt, Pt-alloy, and non-Pt oxygen reduction catalysts forPEMFCs. Appl. Catal. B 56, 9–35 (2005).

4. Ertl, G., Knozinger, H., Schueth, F. & Weitkamp, J. Handbook of HeterogeneousCatalysis (Wiley-VCH, 2008).

5. Somorjai, G. A. Introduction to Surface Chemistry and Catalysis (Wiley, 1994).6. Sinfelt, J. H. Bimetallic Catalysts: Discoveries, Concepts, and Applications

(Wiley, 1983).7. Maroun, F., Ozanam, F., Magnussen, O. M. & Behm, R. J. The role of atomic

ensembles in the reactivity of bimetallic electrocatalysts. Science 293,1811–1814 (2001).

8. Rodriguez, J. A. & Goodman, D. W. The nature of the metal–metal bond inbimetallic surfaces. Science 257, 897–903 (1992).

9. Rodriguez, J. A. Physical and chemical properties of bimetallic surfaces. Surf. Sci.Rep. 24, 223–287 (1996).

10. Greeley, J., Norskov, J. K. & Mavrikakis, M. Electronic structure and catalysis onmetal surfaces. Annu. Rev. Phys. Chem. 53, 319–348 (2002).

11. Hammer, B. & Nørskov, J. K. Electronic factors determining the reactivity ofmetal surfaces. Surf. Sci. 343, 211–220 (1995).

12. Ruban, A., Hammer, B., Stoltze, P., Skriver, H. L. & Nørskov, J. K. Surfaceelectronic structure and reactivity of transition and noble metals. J. Mol. Catal. A115, 421–429 (1997).

13. Chen, S. et al. Enhanced activity for oxygen reduction reaction on Pt3Conanoparticles: direct evidence of percolated and sandwich segregation structures.J. Am. Chem. Soc. 130, 13818–13819 (2008).

14. Chen, S. et al. Origin of oxygen reduction reaction activity on ‘Pt3Co’nanoparticles: atomically resolved chemical compositions and structures. J. PhysChem. C 113, 1109–1125 (2009).

15. Chen, M., Kumar, D., Yi, C.-W. & Goodman, D. W. The promotional effect ofgold in catalysis by palladium–gold. Science 310, 291–293 (2005).

16. Mavrikakis, M., Hammer, B. & Norskov, J. K. Effect of strain on the reactivity ofmetal surfaces. Phys. Rev. Lett. 81, 2819–2822 (1998).

17. Stamenkovic, V. R. et al. Improved oxygen reduction activity on Pt3Ni(111) viaincreased surface site availability. Science 315, 493–497 (2007).

18. Stamenkovic, V. et al. Changing the activity of electrocatalysts for oxygenreduction by tuning the surface electronic structure. Angew. Chem. Int. Ed. 45,2897–2901 (2006).

19. Kibler, L. A., El-Aziz, A. M., Hoyer, R. & Kolb, D. M. Tuning reaction rates bylateral strain in a palladium monolayer. Angew. Chem. Int. Ed. 44,2080–2084 (2005).

20. Zhang, J., Vukmiovic, M. B., Xu, Y., Mavrikakis, M. & Adzic, R. R. Controllingthe catalytic activity of platinum-monolayer electrocatalysts for oxygen reductionwith different substrates. Angew. Chem. Int. Ed. 44, 2132 (2005).

21. Oppenheim, I. C., Trevor, D. J., Chidsey, C. E. D., Trevor, P. L. & Sieradzki, K.In situ scanning tunneling microscopy of corrosion of silver–gold alloys. Science254, 687–689 (1991).

22. Erlebacher, J., Aziz, M. J., Karma, A., Dimitrov, N. & Sieradzki, K. Evolution ofnanoporosity in dealloying. Nature 410, 450–453 (2001).

23. Renner, F. U. et al. Initial corrosion observed on the atomic scale. Nature 430,707–710 (2006).

24. Srivastava, R., Mani, P., Hahn, N. & Strasser, P. Efficient oxygen reduction fuelcell electrocatalysis on voltammetrically de-alloyed Pt–Cu–Co nanoparticles.Angew. Chem. Int. Ed. 46, 8988–8991 (2007).

25. Koh, S. & Strasser, P. Electrocatalysis on bimetallic surfaces: modifying catalyticreactivity for oxygen reduction by voltammetric surface de-alloying. J. Am.Chem. Soc. 129, 12624–12625 (2007).

26. Strasser, P. in Handbook of Fuel Cells: Advances in Electrocatalysis, Materials,Diagnostics and Durability Vol. 5 & 6. (eds Vielstich, W., Gasteiger, H. A. &Yokokawa, H.) 30–47 (Wiley, 2009).

27. Strasser, P., Koh, S. & Greeley, J. Voltammetric surface dealloying of Pt bimetallicnanoparticles: an experimental and DFT computational analysis. Phys. Chem.Chem. Phys. 10, 3670–3683 (2008).

28. Koh, S., Hahn, N., Yu, C. & Strasser, P. Effects of compositions and annealingconditions on the catalytic activities of Pt–Cu Nanoparticle electrocatalysts forPEMFC. J. Electrochem. Soc. 155, B1281–B1288 (2008).

29. Vielstich, W., Lamm, A. & Gasteiger, H. (eds) Handbook of Fuel Cells—Fundamentals, Technology, and Applications (Wiley, 2003).

30. Janik, M. J., Taylor, C. D. & Neurock, M. First-principles analysis of the initialelectroreduction steps of oxygen over Pt(111). J. Electrochem. Soc. 156,B126–B135 (2009).

31. Subramanian, P. R. & Laughlin, D. E. in Binary Alloy Phase Diagrams 2nd edn,Vol. 2 (ed. Massalski, T. B.) 1460–1462 (ASM International, 1990).






32. Wang, J. X. et al. Oxygen reduction on well defined core shell nanocatalysts:particle size, facet and Pt Shell thickness effects. J. Am. Chem. Soc. 131,17298–17302 (2009).

33. Yu, C., Koh, S., Leisch, J., Toney, M. T. & Strasser, P. Size and compositiondistribution dynamics of alloy nanoparticle electrocatalysts probed byanomalous small angle X ray scattering (ASAXS). Faraday Discuss. 140,283–296 (2008).

34. Cullity, B. D. & Stock, S. R. Elements of X ray Diffraction 3rd edn (PrenticeHall, 2001).

35. Pecharsky, V. & Zavalij, P. Y. Fundamentals of Powder Diffraction and StructuralCharacterization of Materials (Springer, 2003).

36. DeGraef, M. & McHenry, M. E. Structure of Materials: An Introduction toCrystallography, Diffraction, and Symmetry (Cambridge Univ. Press, 2007).

37. Klimenkov, M. et al. The structure of Pt-aggregates on a supported thinaluminum oxide film in comparison with unsupported alumina: a transmissionelectron microscopy study. Surf. Sci. 391, 27–36 (1997).

38. Cammarata, R. C. Surface and interface stress effects in thin films. Prog. Surf. Sci.46, 1–38 (1994).

39. Hammer, B. & Nørskov, J. K. Why gold is the noblest of all metals. Nature 376,238–240 (1995).

40. Nilsson, A. et al. The electronic structure effect in heterogeneous catalysis. Catal.Lett. 100, 111–114 (2005).

41. Nilsson, A., Pettersson, L. G. M. & Nørskov, J. K. Chemical Bonding at Surfacesand Interfaces (Elsevier, 2008).

42. Fusy, J., Meneaucourt, J., Alnot, M., Huguet, C. & Ehrhardt, J. J. Growth andreactivity of evaporated platinum films on Cu(111): a study by AES, RHEED andadsorption of carbon monoxide and xenon. Appl. Surf. Sci. 93, 211–220 (1996).

43. Bligaard, T. & Nørskov, J. K. in Chemical Bonding at Surfaces and Interfaces (edsNilsson, A., Pettersson, L. G. M. & Nørskov, J. K.) Ch. 4 (Elsevier, 2008).

44. Lischka, M., Mosch, C. & Gross, A. Tuning catalytic properties of bimetallicsurfaces: oxygen adsorption on pseudomorphic Pt/Ru overlayers. Electrochim.Acta 52, 2219–2228 (2007).

45. Norskov, J. K. et al. Origin of the overpotential for oxygen reduction at a fuel-cellcathode. J. Phys. Chem. B 108, 17886–17892 (2004).

46. Schlapka, A., Lischka, M., Groß, A., Kasberger, U. & Jakob, P. Surface strainversus substrate interaction in heteroepitaxial metal layers: Pt on Ru(0001). Phys.Rev. Lett. 91, 016101 (2003).

47. Mani, P., Srivastava, S. & Strasser, P. Dealloyed Pt–Cu Core–Shell Nanoparticleelectrocatalysts for use in PEM fuel cell cathodes. J. Phys. Chem. C 112,2770–2778 (2008).

48. Hammer, B., Hansen, L. B. & Nørskov, J. K. Improved adsorption energeticswithin density-functional theory using revised Perdew–Burke–Ernzerhoffunctionals. Phys. Rev. B 59, 7413–7421 (1999).

49. Yeh, J. J. & Lindau, I. Atomic Data and Nuclear Data Tables 32, 1–155 (1985).

AcknowledgementsThis project was supported by the Department of Energy, Office of Basic Energy Sciences,under the auspices of the President’s Hydrogen Fuel Initiative. Acknowledgment is alsomade to the National Science Foundation (grant #729722) for partial support of thisresearch. P.S. acknowledges support from the Cluster of Excellence in Catalysis (UNICAT)funded by the German National Science Foundation (Deutsche Forschungsgemeinschaft)and managed by the Technical University Berlin, Germany. Portions of this research werecarried out at the Stanford Synchrotron Radiation Lightsource, a national user facilityoperated by Stanford University on behalf of the US Department of Energy, Office of BasicEnergy Sciences. Use of the Center for Nanoscale Materials was supported by the USDepartment of Energy, Office of Science, Office of Basic Energy Sciences, under contractNo. DE-AC02-06CH11357. We acknowledge computer time at the Laboratory ComputingResource Center (LCRC) at Argonne National Laboratory, the National Energy ResearchScientific Computing Center (NERSC) and the EMSL, a national scientific user facilitysponsored by the Department of Energy’s Office of Biological and Environmental Researchand located at Pacific Northwest National Laboratory. Microscopy research supported byORNL’s SHaRE User Program, which is sponsored by the Scientific User Facilities Division,Office of Basic Energy Sciences, US Department of Energy. The authors thank L. Petterssonfor reading the manuscript.

Author contributionsP.S., M.F.T., J.G. and A.N. designed the research and co-wrote the paper, S.K., T.A., K.M.,C.Y., Z.L., S.K., D.N. and H.O. performed the experiments and analysed the data, and J.G.performed the theoretical calculations.

Additional informationThe authors declare no competing financial interests. Supplementary informationaccompanies this paper at www.nature.com/naturechemistry. Reprints and permissioninformation is available online at http://npg.nature.com/reprintsandpermissions/.Correspondence and requests for materials should be addressed to P.S.










Assembly of a metal–organic frameworkby sextuple intercatenation of discreteadamantane-like cagesXiaofei Kuang1, Xiaoyuan Wu1, Rongmin Yu1, James P. Donahue2, Jinshun Huang1

and Can-Zhong Lu1*

Metal–organic frameworks form a unique class of multifunctional hybrid materials and have myriad applications, includinggas storage and catalysis. Their structure is usually achieved through the infinite coordination of metal ions andmultidentate organic ligands by means of strong covalent bonds. Threaded molecules such as catenanes and rotaxaneshave largely been restricted to comprising components of two-dimensional interlocking rings or polygons. There are veryfew examples of the catenation of polyhedral cages. Although it has been postulated that the infinite extendedarchitecture can be obtained from the polycatenation of a discrete cage based on such threading, this has not beendocumented to date. Here we describe an infinite three-dimensional metal–organic framework composed of catenatedpolyhedral cages, in which the framework is achieved by mechanical interlocking of all of the vertices of the cages. Thethree-dimensional polycatenated framework shows twofold self-interpenetration in its crystal packing. The penetration ofpolycatenanes creates nanosized voids into which the Keggin polyoxometalate anions are perfectly accommodatedas counteranions.

Interpenetrated and interlocked assemblies such as catenanes androtaxanes, in which components are held together by mechanicallinkages rather than by chemical covalent bonds, can form proto-

types for molecular machines1,2, molecular motors3 and molecularknots4, all of which have potential applications in information pro-cessing and storage5, molecular electronics6 and light-driven mol-ecular machines7. Because of their ubiquity in biological systems,distinctive non-covalent mechanical interactions and uniquedynamic behaviour8, catenanes and rotaxanes have been thesubject of extensive interest in supramolecular chemistry9–11. Inmost cases, catenanes (such as organic species12,13, proteins14 andsynthetic DNA assemblies15) have been predominantly limited intheir use to components of two-dimensional molecular rings16. In1999, Fujita and colleagues reported pioneering studies on theassembly of a 2[catenane] with two discrete, interlocking three-dimensional metal–organic cages17. By 2008, only one furtherexample, by Hardie and colleagues, had been reported18. However,both supramolecular architectures were confined to two interlock-ing three-dimensional discrete polyhedral cages.

Recently, several groups have studied supramolecular assemblieswith regular three-dimensional prismatic and polyhedral struc-tures19–25. In principle, each vertex of a three-dimensional poly-hedral molecule is a potential linking point for the formation of acatenane. Thus, multiple catenations of polyhedral molecules maylead to the formation of three-dimensional infinite polycatenanes.This mechanism has been recognized as a basis on which extendedarchitectures might be constructed26. Despite the fact that boththree-dimensional infinite metal–organic frameworks and polycate-nanes have been intensely investigated for nearly two decades, noexample of a three-dimensional, infinite polycatenane fabricatedby the intercatenation of discrete polyhedral molecular componentshas been documented. The construction of polyhedral, three-dimensional, periodic extended architectures from a discrete

polyhedral molecular motif by means of polycatenation has there-fore remained an intriguing challenge for chemists. Here, wereport the synthesis and structure of a new member of the poly-catenanes, {[Ag2(trz)2][Ag24(trz)18]}[PW12O40]2 (1) (trz¼ 1,2,4-triazole), which, to the best of our knowledge, represents the firstexample of a three-dimensional infinite polycatenane that hasbeen assembled directly from discrete polyhedral nanocages.

Results and discussionColourless or pale-coloured crystals of compound 1 were obtainedfrom the reaction of the trilacunary Keggin [A-PW9O34]9– anion,Ag(O2CCH3) and 1,2,4-triazole in water under hydrothermal reac-tion conditions. The formula of the product was determined to be{Ag26(trz)20}[PW12O40]2 on the basis of the combined results ofX-ray single-crystal structure analysis, thermogravimetric analysis(TGA) and elemental analysis. It is noteworthy that there is a struc-ture transformation from the trivacant [A-PW9O34]9– species to thesaturated Keggin [PW12O40]3– motif. The phase purity of theproduct was confirmed by powder X-ray diffraction, and its crystal-linity was stable over a period of several months at ambient tempera-ture in air.

The asymmetric unit of compound 1 is composed of one-twelfthof a [PW12O40]3– polyanion, half each of two independent Agþ

cations, half of an independent triazole ligand, one-fourth of asecond independent triazole ligand, and a third disordered Agþ

cation that is connected to the second triazole ligand. A third tri-azole anion could not be located, possibly due to a highly disorderedand statistical arrangement of the third Agþ cation to which it coor-dinates, which is a situation imposed by the very high symmetry ofthe total structure in the crystalline state. Despite the complicationsposed by disorder, crystals of compound 1 were strongly diffracting.The amount of the third Agþ cation and the third triazole anion wasestablished by elemental analysis to be one-twelfth of the

1State Key Laboratory of Structural Chemistry, Fujian Institute of Research on the Structure of Matter, Chinese Academy of Sciences, Fuzhou, Fujian350002, China, 2Department of Chemistry, Tulane University, 6400 Freret Street, New Orleans, Louisiana 70118, USA. *e-mail: [email protected]







independent composition of the asymmetric unit. The average ofthe Ag–N bond distances (2.11(4) Å) between the third Agþ

cation and the second triazole ligand is comparable to the otherAg–N bond lengths in the structure.

The structure of compound 1 comprises a polyoxometalate-containing metal–organic framework composed of catenatedmetal–organic cages, which can be described overall as two inter-penetrating frameworks based on the primitive cubic geometry.Each framework is constructed from nodes consisting of a metal–organic adamantane-like cage that is catenated by six othersthrough its vertices. The existence of the adamantane-like nano-cages in compound 1 was established unambiguously by X-raysingle-crystal analysis. Each cage is composed of twenty-four Agþ

cations, twelve m3 1,2,4-triazole ligands and six m2 1,2,4-triazoleligands, and is formulated as {Ag24(trz)18}6þ.

The several levels of structural complexity of compound 1 are bestappreciated by reversing the description summarized in the previousparagraph and proceeding in a bottom-up fashion while referring toFigs 1 to 5. The structure of the adamantane-like nanocage can beunderstood from Fig. 1. Three Agþ cations are first linked by three tri-azole ligands in a pyrazole-like bridging mode to give a neutral trigonal[Ag3(trz)3] fragment with D3 local symmetry (Fig. 1a). The triazoleligands deviate slightly from the triangular plane defined by thethree Agþ ions. The non-bonding Ag...Ag distances are 3.5852(1) Å.A similar Ag3(trz)3 unit has been reported in the compound of[Ag5(trz)4]2[Ag2(Mo8O26)].4H2O, in which the trigonal units act asbidentate ligands27. In contrast, each Ag3(trz)3 unit in compound 1behaves as a hypothetical tridentate ligand connecting to three Agþ

ions and leading to the formation of a [Ag3(Ag3(trz)3)]3þ cationicunit (Fig. 1b). Subsequently, another six triazole ligands, whichadopt the pyrazole-like coordination mode, and four[Ag3(Ag3(trz)3)]3þ cationic units are connected to generate a perfectlyenclosed adamantane-like metal–organic {Ag24(trz)18}6þ nanocage(Fig. 1c). In the nanocage, each Agþ ion is coordinated in near-linear fashion by two triazole ligands. The nanocage shows Td sym-metry, and its structure can be best visualized when superimposedover an octahedron (Fig. 1c). The m2 1,2,4-triazole ligands define thesix vertices of the octahedron, and the [Ag3(Ag3(trz)3)]3þ cationicunits and windows alternately occupy the eight faces of the octahedron.

A geometrically related adamantane-like complex was firstreported by Saalfrank and colleagues in 198819. Since then, severalsimilar three-dimensional cages have been reported20–25. In theseearlier examples, the frameworks were constructed from ten mol-ecular species, that is, four tripodal organic ligands and six metalunits, and represented as M6L4. The nanocage in compound 1can also be represented as an M6L4 species if the neutral trigonal[Ag3(trz)3] fragment is considered to be a tridentate ligand (L)and the combination of the apical triazole ligand of the octahedralnanocage with its two bonding Agþ ions (trzAg2)þ as an end-capped transition-metal building block (M) (Fig. 1). The non-bonding Ag...Ag distance in the hypothetical metal units is4.1367(1) Å, which is significantly longer than those within the tri-angular Ag3(trz)3 fragment (L). Although several geometricallyrelated M6L4 adamantane-like coordination molecular nanocageshave been reported previously, in which the L ligands are pre-synthesized or are known triangular tridentate ligands, the cagereported in this work shows new features. It is the first example inwhich the trigonally symmetric building element (tridentate ligand)of a polyhedron is assembled in situ from simple starting componentsinstead of from a pre-designed tridentate ligand. Second, it is anadamantane-like structure constructed from a considerably greaternumber of constituent parts than any previous example28.

The most aesthetically pleasing structural feature of 1 is itsunique polycatenation between discrete octahedral {Ag24(trz)18}6þ

nanocages. As illustrated in Fig. 2a–c, each octahedral nanocage iscatenated by six others through all its six vertices. Variouslycoloured adamantane-like nanocages are shown catenating acentral, black-coloured nanocage (Fig. 2c) at each of the six verticesthat define the octahedron illustrated in Fig. 1c. It is noteworthy thatthe overall architecture is produced with mechanical linkages rather

+

4 + 6

3 Ag

3 3Ag++HN N

N N N

N

NN NN

N

N

Ag

Ag

Ag

a

b

c

Figure 1 | Schematic representation of the details of the adamantane-like

{Ag24(trz)18}61 nanocage. a, Three Agþ cations and three triazole ligands

form trigonally symmetric Ag3(trz)3 fragments. b, The Ag3(trz)3 fragments

are further coordinated with three Agþ ions to form the extended trigonal

[Ag3(Ag3(trz)3)]3þ cationic unit. c, Four of these units assemble with six

additional triazole ligands through Ag–N bonds, generating the adamantane-

like {Ag24(trz)18}6þ nanocage. Six triazoles are located at the vertices of the

adamantane-like cage. Two of these act as threefold bridging ligands and the

remaining four act as twofold bridging ligands. The twofold bridging ligands

each also coordinate a further Agþ ion. These additional Agþ ions are

disordered and are omitted for clarity. In the molecular triangle [Ag3(trz)3]

fragment, the Ag–N bonds are 2.105(16) Å in length, and the N–Ag–N

angles are 172.0(9)8. The Ag–N bonds between the m-Ag ions and the

triazole on the vertices of the nanocage as well as the molecular triangle

[Ag3(trz)3] are 2.11(2) Å and 2.19(3) Å long, respectively. The N–Ag–N

angles are 171.1(9)8.






than by common coordination bonding interactions. The propa-gation of the sextuple polycatenation of each discrete adaman-tane-like cage generates an unprecedented three-dimensionalextended polycatenated polyhedral architecture. The unique three-dimensional structure of the discrete, robust and spacious{Ag24(trz)18}6þ nanocage makes possible a continuous catenationbetween the nanocages in three perpendicular directions, thusforming the exquisitely beautiful three-dimensional polycatenatedextended structure of compound 1.

From a topological viewpoint, if each {Ag24(trz)18}6þ nanocage isconsidered to be a six-connected node, with the catenation as linkagebetween nodes, the structure of 1 can be described as a NaCl-type a-Po topological framework (Fig. 2d). To the best of our knowledge, thethree-dimensional infinite polycatenated structure of compound 1represents the first example of a zero- to three-dimensional polycate-nated system, namely, one in which the catenation of individualmolecules or ionic species (zero-dimensional) has directly produceda three-dimensional polycatenated assembly.

Some of the detailed aspects of the interactions between adjacent{Ag24(trz)18}6þ nanocages in the polycatenanes are pertinent tounderstanding its formation. A view of two interlocking{Ag24(trz)18)}6þ nanocages is shown in Fig. 3. The interpenetratingcorners are disposed orthogonally to one another and overlap by4.0255(1) Å. It is particularly interesting that the two m3 triazoles inthe interpenetrated corner and the one m2 triazole from the adjacentcatenating cage are arranged in a face-to-face-to-face manner. The

separation between the centroids of the adjacent aromatic rings is3.63 Å and the mean interplanar distance is 3.4864 Å, indicatingweak p–p interactions between the triazole ligands of the two inter-locking cages29. Furthermore, weak argentophilic interactionsbetween the interpenetrating corners are observed30, as illustrated bythe proximity of pink and orange rods at the catenated vertex inFig. 3. The Ag...Ag distances between the corners are 3.018(3) Å,which is shorter than the sum of their van der Waals radii. Thus,weak p–p and argentophilic interactions likely have critical roles ininitiating, propagating and stabilizing the polycatenated structure.

Yet another remarkable feature of the structure of compound 1 isthat two independent but identical NaCl-type a-Po topological poly-catenated extended frameworks interpenetrate one other to form ahighly ordered supramolecular aggregate (Fig. 4). The void space inthe single polycatenated framework is so large that two identicalthree-dimensional polycatenated frameworks interpenetrate oneanother to achieve efficient packing (Supplementary Fig. S1). Thewindow-to-window arrangement of the {Ag24(trz)18}6þ nanocagesbetween the interpenetrating frameworks creates nanosized pores inwhich [PW12O40]3– counteranions are located (Fig. 5). The arrange-ment of the [PW12O40]3– polyanions around a {Ag24(trz)18}6þ nano-cage is shown in Fig. 5a. A polyanion resides in each of the fourwindows of the nanocage. Weak coordination bonding interactionsbetween the nanocage and the Keggin anion are observed in com-pound 1. The shortest Ag...O distances between the silver ions in thetrigonal Ag3(trz)3 unit and the m2-O groups of the Keggin anions

a b

d

c

e

Figure 2 | Schematic representation of the overall structure of the polycatenanes in 1 from a discrete {Ag24(trz)18}61 nanocage to three-dimensional

infinite polycatenation. a, Discrete {Ag24(trz)18}6þ nanocage. b, Simplified structure of the {Ag24(trz)18}6þ nanocage. In proceeding from a to b, the

{Ag24(trz)18}6þ nanocage is represented more simply by depicting each of the four threefold symmetric Ag3(trz)3 units as a vertex. c, View of a

{Ag24(trz)18}6þ nanocage (black) catenated by six others through its vertices. These six interlocking cages are shown in different colours, although they are

crystallographically equivalent. d, Ball-and-cylinder model representation of c. Each nanocage is viewed as a six-connected node, with the catenation as

linkage between nodes. The polycatenated nanocages (c) can be rendered as a ball-and-cylinder model in which the ball represents the octahedral nanocage

and the cylinder represents the catenation as linkage. The nanocages in c are represented by correspondingly coloured balls in d, and the catenanes are

represented by correspondingly coloured cylinders. e, Schematic view of the extended architecture of d. The NaCl-type a-Po topology represents the

polycatenated three-dimensional infinite extended framework in 1.






are 2.6737 Å. If the weak Ag...O coordination interactions are con-sidered, each [PW12O40]3– acts as an interpenetrating unit toconnect two {Ag24(trz)18}6þ nanocages from different polycatenatedframeworks through six m2-O atoms (Fig. 5b). The entire supramole-cular architecture of the interpenetrating polycatenated cationic frame-work and the Keggin polyoxometalate anions is shown in Fig. 5c.Furthermore, evidence of an anion template effect is obvious in com-pound 1. Without the presence of the Keggin polyoxometalate, the

self-assembly of Agþ cations and 1,2,4-triazole leads to the formationof a two-dimensional [Ag(trz)]1 network31. Therefore, the Kegginpolyanions function not only as counteranions but also as templatesto help direct the formation of the polycatenated framework in com-pound 1.

In conclusion, the self-assembly of {[Ag2(trz)2][Ag24(trz)18]}[PW12O40]2 demonstrates that the construction of infinite three-dimensional polycatenated frameworks can be realized by the self-intercatenation of discrete polyhedral building blocks. The structureof compound 1 shows first that an adamantane-like nanocage canbe assembled from simple triazole ligands and does not require a

Figure 3 | Supramolecular architecture of adjacent interlocking

{Ag24(trz)18}61 nanocages. X-ray crystallographic analysis shows that two

identical nanocages have interlocked with one another through their vertices;

each cage is shown in a different colour. The interpenetrating corners are

disposed orthogonally to one another and overlap by 4.0255(1) Å. The two

m3 triazoles in the interpenetrated corner and the one m2 triazole from the

adjacent catenating cage are arranged in a face-to-face-to-face manner with

the centroid of the aromatic rings, threaded by thin red string. The

separation between the centroids of the adjacent aromatic rings is 3.63 Å,

and the mean interplanar distance is 3.4864 Å, indicating weak p–p

interactions between the triazole ligands of the two interlocking cages. The

Ag...Ag distances between the corners are 3.018(3) Å, indicating weak

argentophilic interactions between the interpenetrating corners, as illustrated

by the proximity of the pink and orange rods at the catenated vertex.

a b

Figure 4 | Topological views of the twofold interpenetration of the

polycatenation in 1. a, Two independent NaCl-type a-Po topological

polycatenaned frameworks further interpenetrate one another, with the ball

representing the cage and the cylinder representing the catenation. The two

independent, equivalent and interpenetrating frameworks are distinguished

here by different colours. b, Schematic view of the twofold interpenetrating,

infinite extended polycatenated framework.

a

c

b

Figure 5 | Arrangement of polyoxometalates and {Ag24(trz)18}61

nanocages. a, View showing four polyoxometalate species residing in four

windows of the adamantane-like nanocage. b, View showing the

supramolecular architecture of the interlocked cages from the catenated

framework and the cage from the interpenetrating framework. The

polyoxometalate anion is located in a nanosized pore created by the twofold

interpenetration of the polycatenated framework in 1. c, X-ray single-crystal

structure of {[Ag2(trz)2][Ag24(trz)18]}[PW12O40]2, composed of two

interpenetrating frameworks and polyoxometalate species. The two

independent interpenetrating frameworks are distinguished by different

colours, for clarity. Each framework is constructed from the polycatenation of

a discrete adamantane-like cage, which is mechanically interlocked with six

others through its six vertices.






designed, pre-synthesized tridentate ligand, and second that poly-hedral nanocages can interlock each other by means of mechanicallinkages through all of their vertices to produce multiply catenatedinfinite extended structures. The successful synthesis and structuralcharacterization of compound 1 provides new insights for the designof complex types of metal–organic frameworks built from simple,readily available components.

MethodsSynthesis of {[Ag2(trz)2][Ag24(trz)18]}[PW12O40]2 (1). A mixture ofNa9[A-PW9O34].7H2O (0.338 g, 0.13 mmol), CH3COOAg (0.108 g, 0.65 mmol)and 1,2,4-triazole (0.121 g, 1.75 mmol) was dissolved in 10 ml distilled water atroom temperature. The pH value of the mixture was adjusted to �7.3 with 2.0 MNaOH, and the suspension was placed in a Teflonw-lined autoclave and kept underautogenous pressure at 160 8C for 4 days. After slowly cooling to room temperaturefor another 4 days, highly pure and slightly green octahedral crystals were filteredand washed with distilled water (48% yield, based on tungsten).

Crystallographic study of 1. Diffraction data for 1 were collected on a Saturn 70charge-coupled device diffractometer equipped with confocal-monochromated MoKa radiation (l¼ 0.71073 Å) at room temperature. The CrystalClear program wasused for absorption correction. The structure was solved by direct methods and refinedon F2 by full-matrix, least-squares methods using the SHELXL-97 program package.Crystal data are as follows: Ag26C40H40N60O80P2W24, Mr¼ 9,920.28, cubic, spacegroup Pn–3m, a¼ b¼ c¼ 19.3329(5) Å, V¼ 7,225.9(3) Å3, Z¼ 2, rcalcd¼

4.559 g cm23, m¼ 22.564 mm21, F(000)¼ 8,736, GOF¼ 1.042. A total of 19,675reflections were collected, 1,205 of which were unique. R1¼ 0.0791, wR2¼ 0.2356 for110 parameters and 1,163 reflections. R1¼ 0.0771, wR2¼ 0.2306 for data with I .

2s(I). CCDC reference no. 745116. (See Supplementary Information for details.)

Received 11 September 2009; accepted 2 March 2010;published online 11 April 2010

References1. Balzani, V., Credi, A. & Venturi, M. Molecular Devices and Machines—Concepts

and Perspectives for the Nanoworld (Wiley-VCH, 2008).2. Balzani, V., Credi, A., Raymo, F. M. & Stoddart, J. F. Artificial molecular

machines. Angew. Chem. Int. Ed. 39, 3348–3391 (2000).3. Hernandez, J. V., Kay, E. R. & Leigh, D. A. A reversible synthetic rotary

molecular motor. Science 306, 1532–1537 (2004).4. Wang, L., Vysotsky, M. O., Bogdan, A., Bolte, M. & Bohmer, V. Multiple

catenanes derived from calix[4]arenes. Science 304, 1312–1314 (2004).5. Serreli, V., Lee, C.-F., Kay, E. R. & Leigh, D. A. A molecular information ratchet.

Nature 445, 523–527 (2007).6. Flood, A. H., Stoddart, J. F., Steuerman, D. W. & Heath, J. R. Whence molecular

electronics? Science 306, 2055–2056 (2004).7. Balzani, V., Credi, A. & Venturi, M. Light powered molecular machines. Chem.

Soc. Rev. 38, 1542–1550 (2009).8. Mullen, K. M. & Beer, P. D. Sulfate anion templation of macrocycles, capsules,

interpenetrating and interlocked structures. Chem. Soc. Rev. 38,1701–1713 (2009).

9. Lehn, J.-M. Toward self-organization and complex matter. Science 295,2400–2403 (2002).

10. Reinhoudt, D. N. & Crego-Calama, M. Synthesis beyond the molecule. Science295, 2403–2407 (2002).

11. Ikkala, O. & Brinke, G. Functional materials based on self-assembly of polymericsupramolecules. Science 295, 2407–2409 (2002).

12. Collier, C. P. et al. A [2]catenane-based solid state electronically reconfigurableswitch. Science 289, 1172–1175 (2000).

13. Stoddart, J. F. The chemistry of mechanical bond. Chem. Soc. Rev. 38,1802–1820 (2009).

14. Wikoff, W. R. et al. Topologically linked protein rings in the bacteriophageHK97 capsid. Science 289, 2129–2133 (2000).

15. Rybenkov, V. V., Ullsperger, C., Vologodskii, A.V. & Cozzarelli, N. R.Simplification of DNA topology below equilibrium values by type IItopoisomerases. Science 277, 690–693 (1997).

16. Fuller, A.-M. L., Leigh, D. A., Lusby, P. J., Slawin, A. M. Z. & Walker, D. B.Selecting topology and connectivity through metal-directed macrocyclizationreactions: a square planar palladium [2]catenate and two noninterlockedisomers. J. Am. Chem. Soc. 127, 12612–12619 (2005).

17. Fujita, M., Fujita, N., Ogura, K. & Yamaguchi, K. Spontaneous assembly of tencomponents into two interlocked, identical coordination cages. Nature 400,52–55 (1999).

18. Westcott, A., Fisher, J., Harding, L. P., Rizkallah, P. & Hardie, M. J. Self-assemblyof a 3-D triply interlocked chiral [2]catenane. J. Am. Chem. Soc. 130,2950–2951 (2008).

19. Saalfrank, R. W., Stark, A., Peters, K. & Schnering, H. G. The first ‘adamantoid’alkaline earth metal chelate complex: synthesis, structure, and reactivity. Angew.Chem. Int. Ed. Engl. 27, 851–853 (1988).

20. Beissel, T., Powers, R. E. & Raymond, K. N. Symmetry-based metal complexcluster formation. Angew. Chem. Int. Ed. Engl. 35, 1084–1086 (1996).

21. Fujita, M., Tominaga, M., Hori, A. & Therrien, B. Coordination assemblies froma Pd (II)-cornered square complex. Acc. Chem. Res. 38, 371–380 (2005).

22. Leininger, S., Olenyuk, B. & Stang, P. J. Self-assembly of discrete cyclicnanostructures mediated by transition metals. Chem. Rev. 100, 853–908 (2000).

23. Seidel, S. R. & Stang, P. J. High-symmetry coordination cage via self-assembly.Acc. Chem. Res. 35, 972–983 (2002).

24. Cotton, F. A., Murillo, C. A. & Yu, R. Deliberate synthesis of the preselectedenantiomer of an enantiorigid molecule with pure rotational symmetry T.Dalton Trans. 3161–3165 (2005).

25. Tranchemontagne, D. J., Ni, Z., O’Keeffe, M. & Yaghi, O. M. Reticular chemistryof metal–organic polyhedra. Angew. Chem. Int. Ed. 47, 5136–5147 (2008).

26. Carlucci, L., Ciani, G. & Proserpio, D. M. Polycatenation, polythreading andpolyknotting in coordination network chemistry. Coord. Chem. Rev. 246,247–289 (2003).

27. Zhai, Q.-G., Wu, X.-Y., Chen, S.-M., Zhao, Z.-G. & Lu, C.-Z. Construction ofAg/1,2,4-triazole/polyoxometalates hybrid family varying from diversesupramolecular assemblies to 3-D rod-packing framework. Inorg. Chem. 46,5046–5058 (2007).

28. Fujita, M. et al. Self-assembly of ten molecules into nanometre-sized organichost frameworks. Nature 378, 469–471 (1995).

29. Lu, J. et al. Octapi interactions: self-assembly of a Pd-based [2]catenane drivenby eightfold p interactions. J. Am. Chem. Soc. 131, 10372–10373 (2009).

30. Codina, A. et al. Do aurophilic interactions compete against hydrogen bonds?Experimental evidence and rationalization based on ab initio calculations. J. Am.Chem. Soc. 124, 6781–6786 (2002).

31. Zhang, J.-P., Lin, Y.-Y., Huang, X.-C. & Chen, X.-M. Copper (I) 1,2,4-triazolatesand related complexes: studies of the solvothermal ligand reactions, networktopologies, and photoluminescence properties. J. Am. Chem. Soc. 127,5495–5506 (2005).

AcknowledgementsThis work was supported by the 973 Key Program of the Ministry of Science andTechnology of China (2006CB932904, 2007CB815304 and 2010CB933501), the NationalNatural Science Foundation of China (20873150, 50772113, 20821061 and 20973173), theChinese Academy of Sciences (KJCX2-YW-M05, 319) and the National ScienceFoundation (CHE-0845829; J.P.D.).

Author contributionsX.K. and C.Z.L. conceived and designed the experiments. X.K. performed the experimentalwork. X.K., X.W., R.Y., J.H. and C.Z.L. analysed the X-ray structural data and interpretedthe results. C.Z.L. was responsible for the overall design, direction and supervision of theproject. X.K., R.Y., J.P.D. and C.Z.L. co-wrote the paper.

Additional informationThe authors declare no competing financial interests. Supplementary informationaccompanies this paper at www.nature.com/naturechemistry. Reprints and permissioninformation is available online at http://npg.nature.com/reprintsandpermissions/.Correspondence and requests for materials should be addressed to C.Z.L.










Direct detection of CH/p interactions in proteinsMichael J. Plevin1,2,3*, David L. Bryce4* and Jerome Boisbouvier1,2,3*

XH/p interactions make important contributions to biomolecular structure and function. These weakly polar interactions,involving p-system acceptor groups, are usually identified from the three-dimensional structures of proteins. Here, nuclearmagnetic resonance spectroscopy has been used to directly detect methyl/p (Me/p) interactions in proteins at atomicresolution. Density functional theory calculations predict the existence of weak scalar (J) couplings between nucleiinvolved in Me/p interactions. Using an optimized isotope-labelling strategy, these J couplings have been detected inproteins using nuclear magnetic resonance spectroscopy. The resulting spectra provide direct experimental evidence ofMe/p interactions in proteins and allow a simple and unambiguous assignment of donor and acceptor groups. The use ofnuclear magnetic resonance spectroscopy is an elegant way to identify and experimentally characterize Me/p interactionsin proteins without the need for arbitrary geometric descriptions or pre-existing three-dimensional structures.

The three-dimensional structures of biomacromolecules aredependent on an array of weak non-covalent interactions.As well as stabilizing secondary, tertiary and quaternary struc-

ture in proteins and nucleic acids, such interactions can also playessential roles in macromolecular function. The family of hydrogen-bond-like interactions includes examples in which a delocalizedsystem of sp2-hybridized covalent bonds can act as an acceptorgroup1,2. These so-called XH/p interactions, in which X is anatom capable of forming a covalent bond with hydrogen and prefers to the delocalized p-electrons, have long been proposed tocontribute to biomolecular structure and function1.

XH/p interactions have been well studied in small-moleculesystems2,3. Spectroscopic investigations of benzene and water4 orammonium5 have revealed intermolecular hydrogen-bond-likeinteractions in which the p-electrons of benzene act as an acceptorgroup. XH/p interactions are best classified somewhere betweenconventional hydrogen bonds and weaker interactions dominatedby dispersion. The hydrogen-bond-like nature of XH/p interactionsis largely dependent on the polarity of the donor group. When X isnitrogen or oxygen, the interaction energy is dominated by an elec-trostatic term. However, when X is carbon, the interaction has amore dispersive character, and is thus less analogous to classicalhydrogen bonds. XH/p interactions in simple model systems,such as between benzene and methane (CH/p), ammonia(NH/p) or water (OH/p), are predicted to have interaction ener-gies between 1 and 4 kcal mol21 (ref. 6). CH/p interactions, in par-ticular, are considered ‘borderline’ cases under most definitions ofthe hydrogen bond. XH/p interaction energies are dependent ondonor–acceptor group geometry. Polydentate donor groups suchas water, methane or ammonia preferentially form interactions inwhich the XH vector is pointing directly at the centre of the ringand is collinear with the ring normal6,7.

Experimental or theoretical approaches applicable for studyingnon-covalent bonds in smaller molecules cease to be practicalwhen studying larger systems containing multiple examples. Inmore complicated molecules such as proteins or nucleic acids,XH/p interactions are usually identified from a three-dimensionalstructural model. Cyclic aromatic moieties and other p-richsystems found in proteins, nucleic acids or other biomolecules canall act as acceptor groups in XH/p interactions. Surveys of the

protein three-dimensional structure database (PDB) have proposedthat substantial numbers of XH/p interactions exist in nature8–10.

A major advance in the study of classical hydrogen bonds camewith the discovery that scalar (J) couplings exist between acceptorand donor nuclei and that they can be detected using high-resolution nuclear magnetic resonance (NMR) spectroscopyexperiments11–13. This approach has allowed the unambiguousidentification of donor and acceptor groups and the study ofindividual hydrogen bonds in proteins and nucleic acids. J couplingsare detectable across many biologically relevant hydrogen bonds14,including weak CaHa...O¼C interactions between b-strandsin proteins15.

In this Article, direct experimental evidence of CH/p inter-actions between side-chain methyl groups and aromatic amino-acid residues in proteins is presented. The high concentration ofaromatic and methyl groups in the hydrophobic cores of structuredproteins and macromolecular interfaces suggests that methyl/p(Me/p) interactions play important structural or functional roles.A geometric analysis of a database of high-resolution protein struc-tures shows a preference for methyl groups to be located above aro-matic rings. Hybrid density functional theory (DFT) calculationspredict the existence of weak J couplings between nuclei involvedin Me/p interactions in proteins. With the aid of an optimizedisotope-labelling strategy, these weak proton–carbon and carbon–carbon couplings were used in two heteronuclear NMR experimentsthat directly identified the donor and acceptor groups involved in aMe/p interaction.

ResultsOccurrence of Me/p interactions in proteins. Despite severalgeneral studies looking at XH/p interactions, the propensity formethyl groups to locate above aromatic rings in proteins has notyet been investigated. Three geometric parameters were used toidentify putative Me/p interactions in a database of high-resolution X-ray structures (Fig. 1a): d, the distance betweenthe donor carbon atom and the centre of the acceptor ring; u,the angle between the ring normal and a vector connecting themethyl carbon atom and the centre of the ring; and w,the angle between the C–H and ring centre-H vectors. Thevalues used here to define Me/p interactions (d , 4.3 Å; u , 258;

1CEA, Institut de Biologie Structurale Jean-Pierre Ebel, Grenoble, France, 2CNRS, Institut de Biologie Structurale Jean-Pierre Ebel, Grenoble, France,3Universite Joseph Fourier, Institut de Biologie Structurale Jean-Pierre Ebel, Grenoble, France, 4Department of Chemistry and Centre for Catalysis Researchand Innovation, University of Ottawa, Ottawa, K1N 6N5, Canada. *e-mail: [email protected]; [email protected]; [email protected]









w . 1208) are similar to those from previous database studies ofXH/p interactions and are generally less conservative than thosedescribing a standard hydrogen bond8–10. It should be noted thatthese geometric cutoffs, although consistent with earlier studies,are somewhat arbitrary. A total of 1,014 putative Me/pinteractions were identified from a database of 183 high-resolution X-ray structures, giving an overall average of fourputative interactions per 100 residues. When considering themultiple donor groups in isoleucine, leucine and valine residuesand the sizes of aromatic side-chains, �3% of methyl groups and15% of aromatic rings are involved in Me/p interactionsmatching the search parameters used (see Supplementary TablesS1 and S2).

An analysis of the distribution of methyl–aromatic distances, d,shows a maximum density at 4.2 Å (Fig. 1b). A plot of u versus dshows a grouping of points towards the limit permitted before vander Waals clashes occur (Fig. 1c). Steric hindrance produces a dis-tribution in which shorter methyl–ring distances are associatedwith methyl groups located directly above the ring (that is, u ,

108). However, it is noteworthy that the bottom-right quadrant ofFig. 1c—that is, values of d and u where no steric clashes occur—is largely unoccupied. This observation suggests that methylgroups sitting more directly above an aromatic ring tend to belocated at distances consistent with XH/p interactions.

DFT prediction of J couplings between nuclei involved in Me/pinteractions. A series of DFT calculations was performed to assesswhether J couplings exist between donor and acceptor nucleiinvolved in Me/p interactions. A small model system consisting

of toluene and ethane was used to explore the variation ofhpJXMeCaro couplings, where X refers to the carbon or hydrogen ofthe donor methyl group, within the free parameters describing thegeometry of the Me/p interaction. The size of the hpJCMeCarocoupling is unsurprisingly dependent on the distance between theethane and aromatic moieties (Fig. 2a). At distances greater than5.0 Å, the coupling becomes negligible. The location of the donormethyl group above the aromatic ring greatly affects the sizeof the coupling. With the methyl group directly above the ring(u¼ 08), hpJCMeCaro is equal for all aromatic carbons (0.22 Hz ford¼ 3.66 Å; Fig. 2b). At larger values of u, the hpJCMeCarocouplings increase specifically for the two carbon nuclei broughtcloser by the displacement of the donor carbon. Maximumcouplings (0.7 Hz) are observed when the methyl group is directlyabove the ring carbons (u¼ 20.78, d¼ 3.9 Å). The angle made bythe methyl C–H bond with the ring normal has less influence onthe magnitude of the coupling (Fig. 2c).

When considering hpJHMeCaro couplings it is important to takeaccount of the effect of rotation of the methyl donor groupprotons with respect to the ring. Rapid transitions between thethree rotameric states of the methyl group will average the threeindividual hpJHMeCaro couplings at each site, resulting in alower overall value of hpJHMeCaro. As with hpJCMeCaro couplings,the donor–acceptor group distance affects the magnitude of thecoupling (Supplementary Fig. S1). The effect of varying u or won hpJHMeCaro is similar to that seen for hpJCMeCaro.

In the majority of cases presented in Fig. 2, the isotropichpJCMeCaro coupling constant is dominated by the Fermi-contactcoupling mechanism. For hpJHMeCaro couplings, paramagnetic

θ

d

Hφ

C 50

40

30

20

10

0

θ (d

eg)

Distance, d (Å) Distance, d (Å)4.0 6.02.04.0 6.02.0

0.4

0.0

Den

sity

of M

e/π

inte

ract

ions

(a.u

.)

0.8

a b c

Figure 1 | Analysis of Me/p interactions in a database of 183 three-dimensional protein structures (resolution, < 2.0 Å). a, Schematic showing the three

parameters (d, u and w) used to describe Me/p interactions. b, Histogram of Me/p distances for six-carbon-ring systems (Phe, Tyr and Trp; for any Me/p

pair satisfying u, 50.08). The frequency of distances was corrected for the increase in volume at larger values of d to give the density value plotted.

c, Scatter plot of d versus u for six-carbon-ring systems. Turquoise points, w, 1208; black points, 1208 ,w, 1808. The red line approximates the steric

hindrance limit.

θ (deg)Distance, dCMeCaro (Å) φ (deg)

0 12 240.0

0.4

0.8

3.2 4.0 4.8 180 160 140

hπJ C

MeC

aro

(Hz)

hπJ C

MeC

aro

(Hz)

hπJ C

MeC

aro

(Hz)

θ

0.0

0.4

0.8

0.0

0.4

0.8

φ

d

H

H

Ha b c d

Figure 2 | DFT analysis of hpJCMeCaro couplings in Me/p interactions in a model system consisting of toluene and ethane. a–c, Plots showing the variation

of hpJCMeCaro as a function of d (a), u (b) and w (c). The subject parameters in panels a–c and the colour coding of the aromatic carbon atoms are drawn in d.

The orientation of the ethane C–C bond is defined by the C(ethane)–C(ethane)–H-centroid dihedral angle, which is 1808. Viewed from above, the C–C

bond bisects the bond between the green and red carbon atoms. In a, u and w were fixed at 08 and 1808, respectively, and only d was varied. In b and c,

starting values of d¼ 3.66 Å, u¼08 and w¼ 1808 were used. Variation of u in b was achieved by a translation of the donor group such that the value of u

increases from 08 as the ethane moiety moves towards the red and green carbon atoms. Variation of w (c) was achieved by rotation of the C–H bond around

the donor proton. See also Supplementary Fig. S1.






and diamagnetic spin–orbital terms are not negligible, butpartially cancel to leave the Fermi-contact term dominant(Supplementary Fig. S2).

Two proteins, the third domain of protein-G (GB3) and ubiqui-tin, for which high-precision solution and X-ray structures are avail-able, were analysed for potential Me/p interactions. The three-dimensional structure of ubiquitin16,17 revealed a potential Me/pinteraction involving Y59 and L50 (d¼ 3.6 Å; u¼ 5.68; w¼140.68). The local structure shows the d2 methyl group of L50sitting directly above the aromatic ring of Y59. In the three-dimen-sional structure of GB318,19 a possible Me/p interaction involvingthe g2 methyl group of V54 and the five-membered ring of W43(d¼ 4.0 Å; u¼ 16.78; w¼ 112.68) closely matches the search cri-teria. Both ubiquitin-L50 d2 and GB3-V54 g2 methyl groups haveupfield-shifted proton resonance frequencies that are indicative ofclose proximity to an aromatic moiety.

The magnitude of hpJXMeCaro couplings between nuclei involvedin these two putative Me/p interactions was evaluated using DFTcalculations performed at the B3LYP/6-311þþG** level. In ubiqui-tin, hpJXMeCaro couplings for the L50/Y59 interaction vary, dependingon the pair of nuclei considered. Both hpJCMeCaro and hpJHMeCaro coup-lings range between 0.1 and 0.2 Hz (Supplementary Fig. S3). TheV54/W43 interaction in GB3 has smaller hpJHMeCaro couplings(0–0.1 Hz), and the maximum hpJCMeCaro coupling is 0.07 Hz(Supplementary Fig. S3).

Direct detection of Me/p interactions by NMR spectroscopy.Long-range HCC and 1H,13C-heteronuclear multiple quantumcoherence (HMQC) NMR experiments were designed to permitthe transfer of magnetization between nuclei involved in Me/pinteractions via hpJXMeCaro couplings (Supplementary Fig. S4). Theexamples of Me/p interactions identified in ubiquitin and GB3involve proS methyl groups of Leu or Val. A recently reported isotopelabelling scheme was used to prepare uniformly [2H,13C,15N]-labelledprotein samples with proS-specific protonation (and [13C]-labelling)of Leu and Val methyl groups20 (Supplementary Fig. S5) to ensuremaximal experimental sensitivity during the long transfer periodsrequired to evolve hpJXMeCaro couplings.

Long-range HCC spectra of U-[2H,13C,15N], Leu/Val-[13C1H3]proS-labelled ubiquitin showed crosspeaks originatingfrom Me/p interactions (Fig. 3). At 20.2 ppm, the chemical shiftof L50 d2-protons, two correlations are observed that resonate inthe aromatic region of the carbon spectrum (Fig. 3a,b) with chemi-cal shifts that have previously been assigned to Cd1/d2 and C11/12

nuclei of Y59 (ref. 17). These correlations support the existenceof the L50/Y59 Me/p interaction identified in the three-dimen-sional structure of ubiquitin. Two-dimensional spectra ofU-[2H,13C,15N], Leu/Val-[12C1H3]proS-labelled ubiquitin acquiredwith a long-range 1H,13C-HMQC experiment also yielded cross-peaks arising from the L50/Y59 interaction (Fig. 4).

No correlations between the V54 g2 methyl group of GB3 andthe aromatic carbon nuclei in W43 were observed, which is consist-ent with the small couplings predicted by DFT (Supplementary FigsS3,S6). However, this interaction was observed in long-range1H,13C-HMQC spectra of U-[2H,13C,15N], Leu/Val-[12C1H3]proS-labelled GB3. In these data, correlations between V54 g2 methylproton nuclei and C12, Cj2, Cd1, Cd2 and Ch2 aromatic carbonnuclei of W43 were detected (Supplementary Fig. S7).

In HCC and HMQC spectra, crosspeaks were observed betweennuclei not predicted to participate in ‘standard’ Me/p interactions.For example, a correlation was observed in HCC spectra of ubiquitinbetween the L67 d2 methyl group and the Cd nuclei of F45

0.01.0

100

120

140

160

1H (ppm)

13C

(pp

m)

L50 Hδ2*L67 Hδ2*

F45 Cδ1/δ2

Y59 Cε1/ε2

Y59 Cδ1/δ2

L50

Y59

L67

F45

a b c

0.31 Hz 0.22 Hz 0.1 Hz

×2

Figure 3 | Exploitation of weak hpJCMeCaro couplings in proteins by NMR spectroscopy. a, Aromatic region of a long-range two-dimensional HCC spectrum

of U-[2H,13C,15N], Leu/Val-[13C1H3]proS-labelled ubiquitin. A one-dimensional 1H spectrum is shown above to demonstrate the efficiency of the selective

labelling scheme. One-dimensional traces taken from the 13C-dimension are shown. b,c, Selected images of the three-dimensional structure of ubiquitin

(1ubq) showing L50/Y59 (b) and L67/F45 (c) Me/p interactions, with the magnitude of experimental couplings annotated on the figure. 1H and 13C

assignments of ubiquitin are taken from BMRB entry 6427 (ref. 17).

L50

Y59

13C (ppm)

100120140160

Cε1/ε2Cδ1/δ2

Cξ

0.11 Hz

0.07 Hz

0.12 Hz

1H ppm:–0.213

Figure 4 | Exploitation of weak hpJHMeCaro couplings in proteins by NMR

spectroscopy. One-dimensional 13C trace (bottom panel) from the aromatic

region of a long-range two-dimensional 1H,13C HMQC spectrum of U-

[2H,13C,15N], Leu/Val-[12C1H3]proS-labelled ubiquitin. Correlations between

L50-Hd21/d22/d23 and Y59-Cd1/d2, C11/12 and Cj are observed (bottom panel)

for the three-dimensional structure of the L50/Y59 Me/p interaction (top

panel). 1H and 13C assignments of ubiquitin are taken from BMRB entry

6427 (ref. 17).






(Fig. 3a,c). The three-dimensional structure of ubiquitin suggests apossible Me/p interaction involving F45 and L67, although the geo-metry (d¼ 4.7 Å; u¼ 50.38; w¼ 106.78) does not match the orig-inal search parameters. Likewise, HCC spectra of U-[2H,13C,15N],Leu/Val-[13C1H3]proS-labelled GB3 showed a correlation betweenthe L5 d2 methyl group and the Cd1/d2 nuclei of F3021

(Supplementary Fig. S6). The same interaction was observed in along-range HMQC spectrum of U-[2H,13C,15N], Leu/Val-[12C1H3]proS-labelled GB3, as well as an additional correlation invol-ving F30 Cg (Supplementary Fig. S7). These correlations match aputative Me/p interaction identified in the three-dimensional struc-ture (d¼ 4.0 Å; u¼ 16.78; w¼ 112.68). The HMQC spectrum ofGB3 also shows a correlation between the L5 d2 methyl group theand Cd1/d2 nuclei of Y33 (Supplementary Fig. S7). Again, thethree-dimensional structure of GB3 reveals a putative Me/p inter-action (d¼ 4.6 Å; u¼ 60.28; w¼ 129.88). These data demonstratethat the L5 d2 methyl group participates in Me/p interactionswith two aromatic groups simultaneously.

Comparison of theoretical and experimental values. Severalcrosspeaks were observed by NMR spectroscopy that correspondto methyl/aromatic interactions that do not match the geometriesused in the original search criteria. In each case, DFT calculationsrevealed that these interactions have similar-sized couplings to‘classical’ Me/p interactions (Supplementary Fig. S8).

The magnitude of hpJXMeCaro couplings can be directly quantifiedusing the long-range HCC and HMQC experiments introducedhere. The spectra presented show that it is possible to observe andquantify hpJCMeCaro couplings of 0.06 Hz in small proteins (corre-sponding to an experimental S/N ≈ 3) when a suitable isotope-lab-elling scheme has been used. However, owing to the more favourablerelaxation properties of methyl protons, it is possible to detecthpJHMeCaro couplings as small as 0.03 Hz (corresponding to anexperimental S/N ≈ 3) using a [1H,12C]-methyl-labelled sampleand sensitive band-selective, optimized flip-angle, short transient(SOFAST)-HMQC experiments. Each experimental hpJXMeCarocoupling corresponds to interactions between different pairs ofnuclei, the positions of which are averaged by ring flipping of thearomatic acceptor group and rapid rotation of the methyl donorgroup. In contrast, DFT calculations report hpJXMeCaro couplingsfor a single static conformation. To take into account the extremelyshort lifetime of interactions between each pair of atoms, the DFTcalculated couplings have been averaged assuming a simple two-or three-step fast jump model for ring flipping and methyl rotation,respectively. For Me/p interactions in GB3, excellent agreement wasobserved between the experimental NMR values of both hpJCMeCaroand hpJHMeCaro couplings and the theoretical values predicted byDFT (root mean-squared deviation (rmsd)¼ 0.06 Hz; Fig. 5). Itshould be noted that an ultrahigh-resolution, NMR-refined X-raystructure of GB3 (PDB code: 2oed)19 was available for the calcu-lation of precise hpJXMeCaro couplings.

The agreement between NMR and DFT data for Me/p inter-actions in ubiquitin is lower (rmsd¼ 0.1 Hz). This disparity reflectslocal structural uncertainty between the X-ray (1ubq16) and solutionstructures (1d3z17) used for DFT calculations. A detailed compari-son of these two three-dimensional structures revealed small, butnon-negligible, variations in the orientations of the donor andacceptor groups of both of the Me/p interactions described in ubi-quitin. For example, the parameters of the L50/Y59 Me/p inter-action calculated from the X-ray structure of ubiquitin (d¼ 3.6 Å;u¼ 5.68; w¼ 140.68) differ from the average across the NMRensemble (d¼ 3.17 Å; u¼ 8.578; w¼ 134.08). These local structuraldifferences affect the value of the couplings calculated by DFT. Thermsd between hpJXMeCaro couplings calculated from the NMRensemble and those determined from the X-ray structure is0.29 Hz (Supplementary Fig. S9).

A small number of couplings calculated to have magnitudesabove the detection threshold were not observed. A 0.48-HzhpJCMeCaro coupling is predicted for the L5/Y33 interaction inGB3, but no correlation was detected experimentally(Supplementary Figs S7,S8). Furthermore, the experimentally quan-tified couplings between the L5 d2 methyl group and the Cd2 nucleiof Y33 are lower than those predicted by DFT (rmsd¼ 0.1 Hz). Inthe crystal structure of GB318, Y33 is located at an intermolecularinterface and forms two potential crystal contacts. This is not thecase for either W43 or F30. Analysis of the x1 rotameric state ofY33 in solution was not possible due to degeneracy of the b

protons, which is likely due to side-chain flexibility22. The samestudy reported a good agreement between the x1 angles of W43and F30 calculated in solution and from the X-ray structure.Excluding couplings from the L5/Y33 interaction reduces thermsd between experimental and DFT couplings for GB3 to0.01 Hz. These observations suggest that the local structure of Y33in solution may be different from that reported in the original1.1-Å X-ray structure and a subsequent NMR-refined model19.The analysis presented here demonstrates that quantification ofhpJXMeCaro couplings is particularly sensitive to small structuralchanges and that these coupling values could potentially be usedto finely optimize side-chain positions in the hydrophobic coresof proteins.

DiscussionXH/p interactions in biomolecules have received considerableattention and their existence and role have been the subject ofmuch debate1,23,24. From the first three-dimensional structuralstudies of proteins and nucleic acids, there has been suspicionthat interactions involving aromatic acceptor groups might consti-tute an important stabilizing force for biomolecular structure.However, in larger biomolecules there is only anecdotal, structure-based evidence of these interactions8–10. Despite the small size ofhpJXMeCaro couplings, the spectra presented clearly demonstratethat it is possible to use NMR spectroscopy to study Me/p inter-actions in proteins at atomic resolution. The high sensitivity of

0.0

0.2

0.4

Ubiquitin GB3

δ ε δ/εδ ε ξ

50/59 67/45

δ/ε

ε2 η2ξ2 δ2 δ1

5/30 54/43

hπJCMeCaro (DFT)hπJCMeCaro (exp)

hπJHMeCaro (DFT)hπJHMeCaro (exp)

hπJ X

MeC

aro

(Hz)

Protein:

Me/π pair:

Aromatic C:

5/33

δγ δ

Figure 5 | Plot comparing calculated and observed hpJCMeCaro andhpJHMeCaro couplings. Experimental hpJCMeCaro couplings (black circles) were

quantified from long-range two-dimensional HCC spectra of ubiquitin

(L50/Y59 and L67/F45) and GB3 (L5/F30). Experimental hpJHMeCaro

couplings (red circles) were quantified from long-range two-dimensional

HMQC spectra of ubiquitin (L50/Y59) and GB3 (L5/F30 and V54/W43).

Calculated hpJXMeCaro couplings (red and grey bars) for each interaction were

determined using hybrid DFT protocols from the three-dimensional

structures of ubiquitin (1ubq) and GB3 (2oed). Standard deviations (error

bars) for each experimentally quantified coupling were extracted from the

spectral noise.






the labelling and spectroscopic approaches also enabled the detec-tion of crosspeaks corresponding to coupled nuclei that do notconform to a ‘canonical’ Me/p description.

High-level ab initio analysis of CH/p interactions betweenmethane and benzene6,7,25 have shown that the interaction is stron-gest when a CH bond is pointing directly at the centre of the aro-matic ring (that is, u¼ 08 and w¼ 1808). Changing the value of wsuch that the interaction becomes tridentate only slightly reducesthe interaction energy. Lateral displacement of the donor group(that is, increasing u) also does not greatly affect the interactionenergy. Only when the donor–acceptor distance increases does theoverall interaction energy diminish. Thus, although a classicalXH/p geometry (u¼ 08 and w¼ 1808) is favoured, the locationof the carbon determines the interaction energy rather than thelocation of the proton. These previously reported ab initio resultsseem to match well the NMR spectroscopy results presentedhere. That is, crosspeaks were detected in two-dimensional NMRdata both for Me/p interactions with ‘classical’ geometry (forexample, L50/Y59 in ubiquitin) and for interactions that falloutside the ‘classical’ XH/p cone (for example, L5/F30 in GB3).It is important to note that the detected J couplings do not directlyreport interaction energies. Although the magnitude of a coupling isproportional to the degree of overlap between electronic wavefunc-tions, the electron correlation indicated by J couplings can be attrac-tive or repulsive. This said, the data from previously publishedab initio studies6,7,25 and those presented here suggest that thecone commonly used to identify XH/p interactions8–10 does notoffer a full description of interactions involving apolar donorgroups. Studies of the electrostatic potential of aromatic groups inproteins show clear negative regions towards the centre of aromaticring systems26. XH/p interactions involving polar donor groupsseem to be well described by a cone centred on this negativelycharged point. However, for XH/p interactions in which thedonor group is apolar and where the interaction has a significantlylower electrostatic component, it seems necessary to consider amuch broader volume above and below the whole aromatic ring.Nishio and colleagues proposed a geometric definition for CH/pinteractions that stated that the CH donor group must be locatedabove the p-system and preferably above one of the sp2-hybridizedatoms27,28. This broader description, which does not focus on themost negatively charged region of the ring, seems to be largely con-sistent with the CH/p interactions for which hpJXMeCaro couplingscould be experimentally detected.

Me/p and XH/p interactions have been shown to be a commonand functionally important feature in structured proteins. The datapresented here represent the first direct detection of individualXH/p interactions in proteins. Using NMR spectroscopy it hasbeen possible to unambiguously identify donor and acceptorgroups and to experimentally characterize these weak interactions,on an individual basis, and without a priori knowledge of thethree-dimensional structure of the protein.

MethodsDatabase analysis of Me/p interactions. A database of high-resolution three-dimensional protein structures (X-ray only; Supplementary Table S2) was preparedand refined against sequence redundancy (,25%), resolution (,2.0 Å) and R-factor(,0.25) using PISCES29. Protons were added to each of 183 structures using theHGEN module of CCP4 (ref. 30). Putative Me/p interactions were identified usingthe following definition of an Me/p interaction: d , 4.3 Å; u , 258; w . 1208(ref. 10). Definitions of d, u and w are given in Fig. 1a. Distributions of methyl–aromatic group distances (six-carbon rings only) were measured within a conedefined by d , 7.0 Å; u , 508. The steric hindrance limit shown in Fig. 1c wasestimated by a grid search using an idealized six-carbon ring (bond length¼ 1.4 Å)acceptor group.

DFT calculations. All DFT calculations of indirect nuclear spin–spin (J) couplingconstants were carried out using Gaussian 03, revision C.02 (ref. 31). Three CPUswere used per job. Calculations of J couplings were performed using the hybridB3LYP functional and the 6-311þþG** basis set on all atoms. The geometries of

ethane and toluene, and the positions of protons (including for calculations of proteinmodel systems), were optimized at the same level of theory before the calculation of Jcouplings. hpJXMeCaro coupling values were calculated from 1ubq (ubiquitin) and 2oed(GB3). More detail is provided in the Supplementary Information.

Sample preparation. Ubiquitin was expressed from cultures of E. coli BL21(DE3)transformed with a pET21b vector with ampicillin resistance32. Following cell lysis,soluble protein was passed over a Q-sepharose ion exchange resin. The flow-throughwas further purified by size-exclusion chromatography and fractions containingubiquitin were identified by SDS–polyacrylamide gel electrophoresis (SDS-PAGE).Samples of GB3 were prepared as previously reported19. Final NMR samples ofubiquitin (50 mM tris-HCl, pH 8.0) and GB3 (50 mM sodium phosphate, pH 5.5) of�2 mM protein were prepared in D2O and supplemented with sodium azide andprotease inhibitors.

Isotope labelling. All protein samples were prepared in standard M9 minimal mediacontaining [15N]-labelled ammonium chloride (Cambridge Isotope Laboratories,Inc. (CIL)), [2H,13C]-d6-glucose (CIL) and 99.85% D2O (CIL). Specific labelling ofleucine d2 and valine g2 methyl groups was achieved by supplementing the growthmedium with 300 mg of acetolactate 1 h before induction. ProS-specific [1H,12C] or[1H,13C] labelling of Leu and Val methyl groups was achieved by supplementing themedium with either 2-[12C1H3]methyl-4-[12C2H3] acetolactate or 2-[13C1H3]methyl-4-[12C2H3] acetolactate20. Both varieties of acetolactate wereprepared in house. ProR/S nomenclature of Leu and Val methyl groups followsIUPAC/IUBMB/IUPAB guidelines33.

NMR spectroscopy. NMR spectra were recorded on an 800 MHz VarianDirectDrive spectrometer equipped with pulsed-field gradients and a cryogenicallycooled triple-resonance probe head. Detailed information regarding the long-rangeHCC and HMQC experiments can be found in Supplementary Fig. S4. Two-dimensional long-range HCC spectra were collected with 144 (t1) × 1,200 (t2)complex points with acquisition times of 1.8 ms (t1) and 60 ms (t2). A total of 160(ubiquitin) or 224 (GB3) scans were recorded per complex t1 increment with arecycle delay of 1.5 s. Two-dimensional long-range HMQC spectra were collectedwith 320 (t1) × 2,600 (t2) complex points with acquisition times of 4.0 ms (t1) and100 ms (t2). A total of 128 scans were recorded per complex t1 increment with arecycle delay of 1.0 s. Total experimental times were between 10 and 20 h.

All NMR data were processed with NMRPipe34 and analysed using NMRDrawor AZARA (www.bio.cam.ac.uk/azara). Peak intensities and noise estimates wereextracted using nmrDraw.

Quantification of experimental hpJXMeCaro couplings. Experimental hpJCMeCaroand hpJHMeCaro couplings were determined from the ratio of signal intensity betweencorresponding peaks in transfer and reference spectra (see SupplementaryInformation for further explanation). DFT-calculated hpJXMeCaro couplings wereobtained for all pairs of nuclei from the donor and acceptor groups involvedin Me/p interactions. Under experimental conditions, aromatic Cd1/d2, C11/12

carbon (Phe and Tyr) and methyl protons were considered to have degenerateresonance frequencies due to ring flipping or methyl group rotation. To take intoaccount the multiplicity of nuclei involved in each detected NMR magnetizationtransfer, DFT-calculated couplings for each pair of nuclei were suitably combined togive predicted values that were directly comparable to the experimental valuesobtained from NMR spectra.


References1. Perutz, M. F. The role of aromatic rings as hydrogen-bond acceptors in

molecular recognition. Phil. Trans. R. Soc. Lond. A 345, 105–112 (1993).2. Desiraju, G. R. & Steiner, T. The Weak Hydrogen Bond (Oxford Univ.

Press, 1999).3. Meyer, E. A., Castellano, R. K. & Diederich, F. Interactions with aromatic rings

in chemical and biological recognition. Angew. Chem. Int. Ed. 42,1210–1250 (2003).

4. Suzuki, S. et al. Benzene forms hydrogen bonds with water. Science 257,942–945 (1992).

5. Rodham, D. A. et al. Hydrogen bonding in the benzene–ammonia dimer. Nature362, 735–737 (1993).

6. Tsuzuki, S., Honda, K., Uchimaru, T., Mikami, M. & Tanabe, K. Origin of theattraction and directionality of the NH/p interaction: comparison with OH/pand CH/p interactions. J. Am. Chem. Soc. 122, 11450–11458 (2000).

7. Tsuzuki, S., Honda, K., Uchimaru, T., Mikami, M. & Tanabe, K. The magnitudeof the CH/p interaction between benzene and some model hydrocarbons. J. Am.Chem. Soc. 122, 3746–3753 (2000).

8. Burley, S. K. & Petsko, G. A. Amino–aromatic interactions in proteins. FEBSLett. 203, 139–143 (1986).

9. Steiner, T. & Koellner, G. Hydrogen bonds with p-acceptors in proteins:frequencies and role in stabilizing local 3D structures. J. Mol. Biol. 305,535–557 (2001).



www.bio.cam.ac.uk/azara




10. Brandl, M., Weiss, M. S., Jabs, A., Suhnel, J. & Hilgenfeld, R. C–H...p-interactions in proteins. J. Mol. Biol. 307, 357–377 (2001).

11. Dingley, A. J. & Grzesiek, S. Direct observation of hydrogen bonds in nucleicacid base pairs by internucleotide 2JNN couplings J. Am. Chem. Soc. 120,8293–8297 (1998).

12. Pervushin, K. et al. NMR scalar couplings across Watson–Crick base pairhydrogen bonds in DNA observed by transverse relaxation-optimizedspectroscopy. Proc. Natl Acad. Sci. USA 95, 14147–14151 (1998).

13. Cordier, F. & Grzesiek, S. Direct observation of hydrogen bonds in proteins byinterresidue 3hJNC’ scalar couplings. J. Am. Chem. Soc. 121, 1601–1602 (1999).

14. Grzesiek, S., Cordier, F., Jaravine, V. & Barfield, M. Insights into biomolecularhydrogen bonds from hydrogen bond scalar couplings. Prog. Nucl. Magn. Reson.Spectrosc. 45, 275–300 (2004).

15. Cordier, F., Barfield, M. & Grzesiek, S. Direct observation of Ca–Ha...O¼Chydrogen bonds in proteins by interresidue h3JCaC’ scalar couplings. J. Am.Chem. Soc. 125, 15750–15751 (2003).

16. Vijay-Kumar, S., Bugg, C. E. & Cook, W. J. Structure of ubiquitin refined at 1.8 Åresolution. J. Mol. Biol. 194, 531–544 (1987).

17. Cornilescu, G., Marquardt, J. L., Ottiger, M. & Bax, A. Validation of proteinstructure from anisotropic carbonyl chemical shifts in a dilute liquid crystallinephase. J. Am. Chem. Soc. 120, 6836–6837 (1998).

18. Derrick, J. P. & Wigley, D. B. The third IgG-binding domain from streptococcalprotein G. An analysis by X-ray crystallography of the structure alone and in acomplex with Fab. J. Mol. Biol. 243, 906–918 (1994).

19. Ulmer, T. S., Ramirez, B. E., Delaglio, F. & Bax, A. Evaluation of backbone protonpositions and dynamics in a small protein by liquid crystal NMR spectroscopy.J. Am. Chem. Soc. 125, 9179–9191 (2003).

20. Gans, P. et al. Stereospecific isotopic labeling of methyl groups for the NMRstudies of high molecular weight proteins. Angew. Chem. Int. Ed. 49,1958–1962 (2010).

21. Nadaud, P. S., Helmus, J. J. & Jaroniec, C.P. 13C and 15N chemical shiftassignments and secondary structure of the B3 immunoglobulin-bindingdomain of streptococcal protein G by magic-angle spinning solid-state NMRspectroscopy. Biomol. NMR Assign. 1, 117–120 (2007).

22. Miclet, E., Boisbouvier, J. & Bax, A. Measurement of eight scalar and dipolarcouplings for methine–methylene pairs in proteins and nucleic acids. J. Biomol.NMR 31, 201–216 (2005).

23. Mitchell, J. B., Nandi, C. L., McDonald, I. K., Thornton, J. M. & Price, S. L.Amino/aromatic interactions in proteins: is the evidence stacked againsthydrogen bonding? J. Mol. Biol. 239, 315–331 (1994).

24. Weiss, M. S., Brandl, M., Suhnel, J., Pal, D. & Hilgenfeld, R. More hydrogenbonds for the (structural) biologist. Trends Biochem. Sci. 26, 521–523 (2001).

25. Tsuzuki, S., Honda, K., Uchimaru, T., Mikami, M. & Fujii, A. Magnitude anddirectionality of the interaction energy of the aliphatic CH/p interaction:significant difference from hydrogen bond. J. Phys. Chem. 110,10163–10168 (2006).

26. Dougherty, D. A. Cation–p interactions in chemistry and biology: a new view ofbenzene, Phe, Tyr and Trp. Science 271, 163–168 (1996).

27. Umezawa, Y. & Nishio, M. CH/p interactions in the crystal structure of class IMHC antigens and their complexes with peptides. Bioorg. Med. Chem. 6,2507–2515 (1998).

28. Umezawa, Y. & Nishio, M. CH/p interactions as demonstrated in the crystalstructure of guanine-nucleotide binding proteins, Src homology-2 domains andhuman growth hormone in complex with their specific ligands. Bioorg. Med.Chem. 6, 493–504 (1998).

29. Wang, G. & Dunbrack, R. L. Jr. PISCES: a protein sequence culling server.Bioinformatics 19, 1589–1591 (2003).

30. CCP4. The CCP4 suite: programs for protein crystallography. Acta Crystallogr. D50, 760–763 (1994).

31. Frisch, M. J. et al. Gaussian 03, Revision C.02. (Gaussian, 2003).32. Sass, J. et al. Purple membrane induced alignment of biological macromolecules

in the magnetic field. J. Am. Chem. Soc. 121, 2047–2055 (1999).33. Markley, J. L. et al. Recommendations for the presentation of NMR structures of

proteins and nucleic acids. IUPAC-IUBMB-IUPAB Inter-Union Task Group onthe Standardization of Data Bases of Protein and Nucleic Acid StructuresDetermined by NMR Spectroscopy. J. Biomol. NMR 12, 1–23 (1998).

34. Delaglio, F. et al. NMRPipe: a multidimensional spectral processing systembased on UNIX pipes. J. Biomol. NMR 6, 277–293 (1995).

AcknowledgementsThe authors wish to thank O. Hamlin and P. Gans for providing labelled acetolactate,B. Brutscher, J.-P. Simorre and D. Marion for a critical reading of the manuscript, I. Ayalafor help in preparing protein samples, and the Partnership for Structural Biology for accessto integrated structural biology platforms. The clone of GB3 was kindly provided by A. Baxand that of ubiquitin by S. Grzesiek. M.J.P. acknowledges funding from L’Association pourla Recherche sur le Cancer and the EU (FP7-PEOPLE-IRG-2008), J.B. acknowledgesfunding from Agence Nationale de la Recherche, Human Frontiers Science Programme andCentre National de la Recherche Scientifique, and D.L.B. acknowledges the NaturalSciences and Engineering Research Council of Canada and the High-Performance VirtualComputing Laboratory.

Author contributionsAll authors conceived and devised the experiments, and co-wrote the manuscript. M.J.P.prepared samples. M.J.P. and J.B. recorded and analysed the NMR data. D.L.B. performedand analysed the DFT calculations.

Additional informationThe authors declare no competing financial interests. Supplementary informationaccompanies this paper at www.nature.com/naturechemistry. Reprints and permissioninformation is available online at http://npg.nature.com/reprintsandpermissions/.Correspondence and requests for materials should be addressed to M.J.P., D.L.B. and J.B.












Geometry-controlled kineticsO. Benichou1*, C. Chevalier1, J. Klafter2, B. Meyer1 and R. Voituriez1

It has long been appreciated that the transport properties of molecules can control reaction kinetics. This effect can becharacterized by the time it takes a diffusing molecule to reach a target—the first-passage time (FPT). Determining theFPT distribution in realistic confined geometries has until now, however, seemed intractable. Here, we calculate this FPTdistribution analytically and show that transport processes as varied as regular diffusion, anomalous diffusion, anddiffusion in disordered media and fractals, fall into the same universality classes. Beyond the theoretical aspect, this resultchanges our views on standard reaction kinetics and we introduce the concept of ‘geometry-controlled kinetics’. Moreprecisely, we argue that geometry—and in particular the initial distance between reactants in ‘compact’ systems—canbecome a key parameter. These findings could help explain the crucial role that the spatial organization of genes has intranscription kinetics, and more generally the impact of geometry on diffusion-limited reactions.

It is generally known that reaction kinetics can be influenced by thetransport properties of the reactants1–3. In fact, the transport step,before reactants meet and eventually react, can even be limiting

in the case of confined systems such as cells or cell subdomainswhere a small number of reactants are involved4–9. In such systems,the first step in estimating the kinetics of reactions consists of evalu-ating the properties of the first encounter between reactants.Quantitatively, this amounts to calculating the distribution of thetime it takes a diffusing molecule to reach a target site—the first-passage time (FPT) distribution. Although this quantity is wellknown in quasi-one-dimensional or unconfined geometries (seeref. 10 for a review), determining the FPT distribution seems to beintractable in the realistic situation where the diffusing molecule isconfined within a finite domain11. A first estimate of the effect ofgeometrical parameters of confinement on this search time is givenby the mean of the FPT. This has recently been calculated, and alinear scaling with the volume has been demonstrated12–15.However, as soon as several timescales are involved, the kineticscan not be determined by the mean of the FPT only, and the entiredistribution is needed16,17.

Gene transcription provides an extremely important example—which we shall repeatedly invoke in what follows—of reactions invol-ving a small number of (or even single) reactants confined within asmall (microsized) domain, and whose kinetics must be preciselyregulated to fulfill vital cell functions. Interest in the question ofhow geometrical parameters impact on the kinetics of such transcrip-tional reactions and how they could act as regulatory factors hasrecently increased, mainly because of new experimental tools thatenable the observation of the real-time production of proteins atthe single molecule level4. These techniques, which give access tothe spatial organization of the genetic material, have revealed strongcorrelations between the spatial locations of successively activatedgenes, both for prokaryotes18 and eukaryotes19. Indeed, it has beenfound that successively activated genes are often colocalized, that islocated in the very same regions. These observations raise the ques-tion of the importance of geometrical parameters in transcriptionkinetics, which has remained so far widely unanswered.

In the broader context of chemical reactions in confinement, weare interested here in several questions. (i) How does the FPT distri-bution depend on the volume of the confining domain? (ii) How

does it depend on the initial position of the diffusing molecule?(iii) Is this geometric dependence an important factor that couldpotentially control the kinetics?

Note that the influence of confinement and crowding effects onbiochemical reactions has already been studied (see refs 20, 21 forreviews) on the basis of a thermodynamical treatment of reactionkinetics. Although this approach is well suited to the case of alarge number of reactants, it does not provide the dependence ofthe kinetics on the geometrical parameters mentioned above(such as the initial position of the reactants), which involve theindividual nature of the reactants and their dynamical properties.

In this work, we calculate analytically the distribution of the FPTat a target T for a diffusing particle released at a starting point S (seeFig. 1) and quantitatively answer questions i–iii above. We highlightuniversal laws of the FPT distribution as functions of the volumeN of the confining domain and of the distance between S andT (ST ; r), and show that two regimes emerge. More precisely, wefind that the key criterion is the compact versus non-compactnature of the diffusion process, to be defined mathematicallybelow. In the non-compact case, which physically corresponds toa diffusing molecule that ‘sparsely’ explores its environment andleaves unvisited regions (typically a molecule diffusing in a dilutesolution), we show that the kinetics are widely independent of thestarting point. In the contrasting compact case of a diffusing moleculethat ‘densely’ explores its environment (for example a molecule in avery crowded medium), the position of the starting point stronglyinfluences the search time of the target, which leads us to introducethe concept of ‘geometry-controlled kinetics’. In the context of genetranscription, this result implies that the kinetics of activation of agene T by a transcription factor can be orders of magnitudes fasterif the transcription factor is released from a site that is colocalizedwith (that is, in the vicinity of) T, as compared with the case wherethe transcription factor is released from a remote site (Fig. 1).

ResultsWe consider a Markovian random walker of position r(t),whose dynamics are characterized by the dimension of the walkdw, defined by the scaling of the mean squared displacementkr 2(t)l/ t2/dw . The walker is confined in a domain of N sites withreflecting walls. Additionally, we assume that the medium is of

1UPMC Univ Paris 06, CNRS-UMR 7600 Laboratoire de Physique Theorique de la Matiere Condensee, 4 Place Jussieu, F-75005 Paris, France, 2School ofChemistry, Raymond and Beverly Sackler Faculty of Exact Sciences, Tel Aviv University, Tel Aviv, Israel; and Institute for Advanced Studies (FRIAS),University of Freiburg, 79104, Freiburg, Germany. *e-mail: [email protected]







fractal type, so that its characteristic linear size R scales as R/N1/df , where df is the fractal dimension22. We are interested in thedistribution P(TTS) of the time it takes a walker starting from thesite S to reach for the first time the target site T located at a distancer from S.

We start from the backward equation satisfied by the probabilityP(TTS¼ t) in discrete time t (ref. 10)

P(TTS = t) =∑

j

wjSP(TTj = t − 1) (1)

obtained by partitioning over the first step of the walk, where wij isthe transition probability from site j to site i. It is shown in theSupplementary Information that this equation, after Laplace trans-form, leads to the following hierarchy satisfied by the moments kTTj

n lof the FPT:

kTnTSl = 1

WstatT

∑j

∑n

k=1

(n

k

)(−1)k+1

× [(HTT − HTS)Wstatj + (HjS − HjT )Wstat

T ]kTn−kTj l

(2)

Here

Hji =∑1

n=0

(Wji(n) − Wstatj ) (3)

where Wji(n) denotes the propagator of the walk, that is, the prob-ability to be at site j at step n starting from site i, and Wj

stat is theprobability of being at site j in the stationary state.

At this stage, the hierarchy of equation (2) remains formal as itinvolves the unknown function Hji, and does not allow an explicitdetermination of the FPT distribution. However, this difficulty canbe circumvented by taking the large volume limit and by consideringthe rescaled time u = TTS/kTlT , where kTlT =

∑S kTTSlWstat

S is themean FPT to the target site rT, averaged over the initial position.Note that here we implicitly assume kTlT (as well as its disorderaverage in the case of disordered systems to be discussed below)to be finite. Actually, a detailed analysis of equation (2) inthe Supplementary Information shows that the distribution of the

rescaled variable u takes the following general form in the largevolume limit:

GTS(u) = (1 −PTS)d(u) +PTSc(u) (4)

where the Dirac d function corresponds to trajectories hitting thetarget without reaching the boundary within a time of order rdw ,which is much smaller than kTlT . The geometrical factor 1 2PTScan be interpreted as the weight of these trajectories. Similarly, thecontribution PTSc(u) accounts for trajectories reaching the boundarybefore the target. Note that the dependence on the starting point liesentirely in the geometrical factor PTS, whereas the time dependence iscontained in the scaling function c. The geometrical factor PTS andthe rescaled variable u are explicitly determined in the SupplementaryInformation and their scaling with the volume N and the distancer are obtained under the standard scale-invariance assumptionof the unconfined propagator Wij

1(n) / n−df /dw f(|ri 2 rj|/ n1/dw )(ref. 22). Actually, the FPT distribution given by equation (4) fallsinto a few universality classes, defined according to a purely geo-metrical criterion as detailed below.

In the case of non-compact exploration, defined here by dw , df(ref. 23), where the mean number of distinct sites visited by thewalker in the absence of confinement grows linearly with thenumber of steps, we find:

kTlT = HTT/WstatT / N

PTS =kTTSlkTlT

/ 1 − a1r

( )df−dw

c(u) = e−u

⎧⎪⎪⎪⎨⎪⎪⎪⎩ (5)

where a is a lattice-dependent constant of order 1. Note that thelinear dependence on N of the scaling variable kTlT is the same asfound in ref. 13 for the mean FPT kTTSl, and in particular doesnot depend on the dimensions df and dw. This general resultincludes the special case of regular diffusion in three dimensions(for which dw¼ 2 and df¼ 3) that is given in ref. 24. Strikingly,the exponential form of C hold for any dimensions such thatdw , df. Note that although in the limit of r . step size the FPT dis-tribution is a mere exponential of weight, it departs significantlyfrom this distribution if the starting point is close to the target.

In the case of compact exploration, dw . df (ref. 23), where themean number of distinct sites visited by the walker in the absenceof confinement grows slower than linearly with the number ofsteps, further hypotheses on the unconfined propagator are neededto estimate the relative importance of the terms involved in equation(2). Making use of the O’Shaughnessy–Procaccia operator25 to evalu-ate the large volume behaviour of Hij in equation (2), we find that theFPT distribution obeys the generic form of equation (4) with

kTlT = HTT/WstatT / Ndw/df

PTS =2d2

f

dw(df + dw)kTTSlkTlT

/rR

( )dw−df

c(u) = 2df dw

d2w − d2

f

G(n)G(2 − n)

∑1

k=0

(ak2 )

3−2nJn(ak)J1−n(ak)

e− a2

kdwdf

2(d2w − d2

f )u

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎪⎩

(6)

where u . 0, n¼ df/dw and a0 , a1 , , . . . stand for the zeros of theBessel function J2n. Strikingly, the scaling with N of the scaling vari-able u is no longer given by the mean FPT kTTSl, which clearly indi-cates that several timescales are involved in the problem. Theinterplay between these timescales leads to a non-trivial family ofuniversal non-exponential scaling functions, parametrized by dw anddf. The geometrical factor strongly depends on the source–target

S1

S2

T

Figure 1 | First-passage time distribution (FPT) and geometry-controlled

kinetics. Is the initial position of the particle an important parameter of

reaction kinetics? We show quantitatively that in the case of non-compact

exploration (for example, for dilute solutions), the kinetics turns out to be

widely independent of the starting point (S1 or S2), whereas in the compact

exploration case (for example, for crowded environments), the position of

the starting point strongly influences the search time of the target, leading

to ‘geometry-controlled kinetics’. This result in particular implies that the

kinetics of activation of a gene T by a transcription factor can be orders of

magnitudes faster if the transcription factor is released from a site S ; S2,

which is colocalized with T, as compared to the case where the transcription

factor is released from a remote site S ; S1.






distance r, and in particular is very small for r ≪ R, in contrast to thenon-compact situation. We add that the marginal case dw¼ df ,which is compact according to the definition given in ref. 23, corre-sponds to an exponential scaling function c as given by equation (5)with logarithmic corrections in the scalings of kTlT and PTS with rand N (see Supplementary Information).

Equations (5) and (6) fully define universality classes of FPT distri-butions in confinement. Additional comments are as follows.(i) Whereas the linear scaling with N of the first moment is universal,the scaling of higher moments differ in compact and non-compactcases. In particular, this scaling implies that whereas the reduced var-iance of the FPT is always of order one in the non-compact case, in thecompact case it reads (kTTS

2 l 2 kTTSl2)/kTTSl

2/ (R/r)dw−df , so verylarge fluctuations occur for r ≪ R. (ii) Remarkably, the FPT distri-bution is entirely determined as soon as the mean kTTSl of the FPTis known (as well as its average over the starting point kTlT )26–30,even if the distribution is not exponential. (iii) For specific cases, thismean FPT can be calculated exactly, which provides a fully explicitexpression of the FPT distribution (as used in Figs 2b, 3a,b). (iv) Inall cases it can be calculated in the large volume limit using recentresults13, leading to the scaling in the geometrical parameters givenin equations (5) and (6).

We note that our approach covers in particular the important caseof subdiffusion31, which is characterized by a sublinear dependenceof the mean squared displacement with time (that is, dw . 2).Subdiffusion is widespread in complex crowded environments suchas biological cells32,33, and might physically originate from a fewclasses of models based on different underlying microscopic mechan-isms34. Importantly, subdiffusive processes can be either compact ornon-compact, which will prove below to be the relevant criterion inthe context of reaction kinetics. The FPT distribution for one class ofmodels for subdiffusion, which rely on spatial inhomogeneities22 asexemplified by diffusion in fractals, is directly given by equations (5)and (6). Another class of models stems from large trapping times,leading to the case of infinite kTlT , which we have discarded so far.Whereas the quenched version of this type of model becomes quiteinvolved in the case of broad distribution of trapping times over thedisorder, the FPT distribution in the annealed case—the continuoustime-random walk model (CTRW), which is a standard randomwalk with random waiting times, drawn from a distribution f(t) (refs31,35–37)—is straightforwardly deduced from equations (5) and (6).In this case, the Laplace transform of the FPT distribution reads:PCTRW(s) = P( f (s)), where P(s) is the generating function (discreteLaplace transform) of the FPT distribution of the underlying discretetime-random walk, which is determined in equations (5) and (6).

These analytical results are validated by Monte Carlo simulationsand exact enumeration methods, applied to various models, whichillustrate the universality classes defined above. These schematicmodels have been widely used to describe transport in disorderedmedia16,22—for example in the case of exciton trapping on percola-tion systems38 or anomalous diffusion in biological cells39,40—as afirst step to account for geometrical obstruction and binding effectsinvolved in real crowded environments20,21. (i) The non-compactand marginal cases (see Fig. 2) are exemplified by: regular diffusionon three-dimensional (3D) and two-dimensional (2D) cubic lattices;diffusion on a 3D percolation cluster above criticality; and diffusion indisordered systems such as the random barrier model (namely a sym-metric random walk on a 3D cubic lattice with transition rates G

drawn from the normalized distribution r(G)/ G2a) and therandom trap model (namely a symmetric random walk on a 3Dcubic lattice with frozen waiting times ti at each site drawn fromthe normalized distribution r(t)/ t2(1þa)). (ii) The compact case(see Fig. 3) is exemplified by diffusion on deterministic fractals suchas a Sierpinski gasket and T-graph and on a critical percolationcluster, as defined in Supplementary Fig. S1. Figures 2 and 3 revealan excellent quantitative agreement between the asymptotic analytical

0.01

0.1

1 100 × 20 × 20, T(20,5,10), S(70,15,10)60 × 40 × 20, T(4,4,4), S(56,36,16)80 × 80 × 10, T(25,25,5), S(55,55,5)200 × 150, T(4,4), S(2,4)100 × 80, T(54,40), S(56,40)Theory (Eq. 5)

0.008

0.04

0.2

1α = 0.8, T(2,2,2), S(28,28,28)

α = 0.5, T(2,2,2), S(3,3,3)

α = 0.2, T(10,10,10), S(20,20,20)

α = 0.2, T(2,2,2), S(3,3,3)

0.008

0.04

0.2

1

5

α = 2, 100 × 20 × 10, T(2,2,2), S(98,18,8)

α = 1.5, 100 × 80 × 10, T(60,60,5), S(20,20,5)

α = 1.5, 100 × 80 × 60, T(3,3,3), S(4,4,4)

α = 1, 100 × 20 × 20, T(50,10,10), S(40,8,8)

Random clustersand S/T pairs

c

e

d

b

Rescaled time (θ)

ψ(θ)

4 5

10 2 3 4 5

10 2 3

0.01

0.1

1ψ(θ)

ψ(θ)

ψ(θ)

10 2 3 4 5

Rescaled time (θ)10 2 3 4 5

Theory (Eq. 5)

Theory (Eq. 5)

Theory (Eq. 5)

0 1 2 3 4 5Rescaled time (θ)

Rescaled time (θ)

Rescaled time (θ)

0.01

0.1

1

Res

cale

d F

PT

dis

trib

utio

n ψ(θ)

2D and 3D simple lattices (5 pairs)3D supercritical percolation (5 pairs)Random barrier model (4 pairs)Random trap model (4 pairs)Theory (Eq. 5)

a

Figure 2 | Universal FPT distribution in the non compact and marginal

cases. The simulated distribution GTS(u) divided by the weight PTS is plotted

against the universal theoretical prediction c(u) (equation (5)). The collapse

of various examples onto a single master curve shows the universality of the

result. a, All non-compact and marginal cases (plotted independently in b–e)

collapse onto a single exponential master curve. b, Regular diffusion on a 3D

cubic lattice and 2D square lattice for various rectangular domains (of sizes

L1 × L2 × L3 and L1 × L2) and source–target pairs S(x,y,z) and T(x,y,z) whose

rectangular coordinates are indicated in the legend inset. Here u and PTS

are calculated using exact results for kTTSl and kTlT given in ref. 24.

c–e, Examples of disordered systems. Here, kTTSl and kTlT are evaluated

numerically. c, Diffusion on a 3D percolation cluster above

criticality embedded in a 30× 30 × 30 rectangular domain with link

probability p¼0.4. d, 3D random barrier model (see text) embedded in a

30 × 30 × 30 rectangular domain. e, 3D random trap model (see text)

embedded in various rectangular domains.






predictions and the numerical simulations, even for systems of smallsize. We emphasize that despite their very different nature, all thesemodels fall into the above defined universality classes.

DiscussionWe now discuss important implications of these results for reactionkinetics, using the ubiquitous example of a target search process

involving an immobile target B and searcher particles A (ref. 1).When only a small number of A are involved in the reaction, as isthe case in a microdomain in a biological cell, this reaction has tobe described at the single molecule level6 and can be quantified bythe survival probability of a particle A, S(t) = 1 −

∑tt′=0 P(t′),

which gives the probability that A has not reacted with B untiltime t. The quantity S(t) depends on the initial position of the mol-ecule A and is explicitly determined using equations (5) and (6),which indicate that such reactions in confinement obey universalkinetic laws, depending on the non-compact versus compactnature of the underlying transport process.

In the non-compact case, which corresponds qualitatively to tra-jectories leaving many sites unvisited, as in the case of a 3D mediumdilute enough to lead to regular diffusion, for any r significantlygreater than a typical molecular length, the geometrical factor PTSis close to 1, and the dependence on the initial position is lost.We therefore recover a simple first-order decay of the survival prob-ability, which depends on the volume of the confining domain only,and not on the initial position of the reactant. In this case of non-compact exploration, the initial position is not an important par-ameter of the kinetics (see Fig. 4), except in specific cases involvingreturn times, such as recombination reactions.

On the contrary, in the compact case, which means that eachvisited site is on average oversampled, geometrical factors dominate.This is typically the case of a crowded medium described to a firstapproximation as a fractal structure where the available space fordiffusion is restricted. Here, the temporal evolution of S(t) stronglydepends on the starting position. S(t) drops to small values—indi-cating that the reaction occurred with high probability—on a time-scale that depends on the volume, but also critically on the startingposition of the reactant. This timescale ranges according toequations (4) and (6) from rdw (for starting positions such thatr ≪ R) to Rdw (for starting positions such that r ≃ R), which inpractice can span several orders of magnitude (see Fig. 4). Inthese types of ‘geometry-controlled reactions’ (not to be confusedwith ‘fractal-like reactions’2), spatial organization of reactantsplays a crucial role, which can be quantified by our approach.

We stress that the decisive criterion leading to geometry-con-trolled kinetics is not the subdiffusive versus diffusive nature ofthe transport process, but rather the compact versus non-compactnature. We expect this effect to impact a wide class of reactionsinvolving either an inhomogeneous initial concentration of reactants,such as the speckled distribution experimentally realized in ref. 41, ora small number of particles, such as biochemical reactions in cell sub-domains. Notably, in the context of gene colocalization, our resultsgive access to the kinetics of elementary steps of activation by tran-scription factors. As an illustrative example, let us consider twogenes A and B, which share a common transcription factor (forexample the genes sog and zen of the Drosophila genome, whichare both targeted by Dorsal42). Experimental results concerning sub-diffusive motion of tracer particles in the nucleus32,43,44 on the onehand and observations of a fractal organization of the chroma-tin on the other hand44–46, provide the estimates 2 ≤ dw ≤ 3 anddf ≃ 2.4, and therefore suggest that both compact and non-compact exploration cases could occur. Relying on the analysis ofthe survival probability developed previously, we find that in the caseof compact exploration (with, for example, df¼ 2.4, and dw¼ 3 as inref. 43) that the typical time needed for the transcription factor toreach gene B starting from a gene A colocalized with B (typicallyrAB¼ rcoloc ≤ 100 nm, which is the size of a transcription factory19)can be (rremote/rcoloc)

dw ≃ 103 times faster than for a remote gene A(typically rAB¼ rremote ≃ 1 mm, which is the order of magnitude ofa nucleus radius). This is in strong contrast with the case of non-compact exploration (with for example df¼ 2.4 and dw¼ 2)32

where the typical activation time of B starting from A has the sameorder of magnitude for A either colocalized with B or remote. This

Res

cale

d F

PT

dis

trib

utio

n ψ(θ)

Theory (Eq. 6)N = 123N = 366N = 1,095

0.2 0.4

1

1.5

Sierpinski gasketT-graph

0.6 0.8

0.1

1

10

1

Theory (Eq. 6)N = 82N = 244N = 730

0 2 43

0.1

1

10

a

c

b

20 × 20 × 20 domains30 × 30 × 30 domains40 × 40 × 40 domains

Theory (Eq. 6)0.1

1

10

ψ(θ)

ψ(θ)

Rescaled time (θ)

10 2 43

Rescaled time (θ )

10 2 43

Rescaled time (θ)

Figure 3 | Universal FPT distributions in the compact case. The simulated

distribution GTS(u) divided by the weight PTS is plotted against the universal

theoretical prediction c(u) (equation (6)). The collapse for different system

sizes N shows the universality of the results. a,b, Examples of deterministic

fractals. Diffusion on a Sierpinski gasket (with a target at the apex site) (a)

and on a T graph (with a target at the centre) (b). Here, exact expressions

are used for calculating kTTSl and kTlT . The inset shows that the scaling

function c weakly depends on the dimensions df and dw. c, Diffusion on a

3D critical percolation cluster (random fractal) embedded in rectangular

domains of sizes (L1 × L2 × L3), as indicated in the inset. Here, kTTSl and

kTlT are evaluated numerically, and average over pairs of points is

performed, to fulfill the scale-invariance hypothesis of the propagator

(see text before equation (5)).






suggests that gene colocalization is highly favourable for transcriptionkinetics, but only when it is geometrically controlled (that is, in thecase of compact exploration), which makes the experimental charac-terization of the nature of transport in the nucleus a major issue.

To conclude, we calculated analytically the FPT distribution of adiffusing particle to a target, and showed that transport processes asvarious as regular diffusion, anomalous diffusion, diffusion in disor-dered media and in fractals fall into the same universality classes.Our results put forward that geometry, and in particular the initiallocalization of reactants, can become a key parameter of reaction kin-etics in confinement. In particular, this regime of ‘geometry-controlledkinetics’ could be relevant to transcription kinetics and could helpunderstand the crucial role of spatial organization of genes.

Received 30 November 2009; accepted 9 March 2010;published online 18 April 2010

References1. Rice, S. Diffusion-Limited Reactions (Elsevier, 1985).2. Kopelman, R. Fractal reaction kinetics. Science 241, 1620–1626 (1988).3. Shlesinger, M. F., Zaslavsky, G. M. & Klafter, J. Strange kinetics. Nature 363,

31–37 (1993).4. Yu, J., Xiao, J., Ren, X., Lao, K. & Xie, X. S. Probing gene expression in live cells,

one protein molecule at a time. Science 311, 1600–1603 (2006).5. Elf, J., Li, G.-W. & Xie, X. S. Probing transcription factor dynamics at the single-

molecule level in a living cell. Science 316, 1191–1194 (2007).6. Schuss, Z., Singer, A. & Holcman, D. The narrow escape problem for diffusion in

cellular microdomains. Proc. Natl Acad. Sci. USA 104, 16098–16103 (2007).7. Loverdo, C., Benichou, O., Moreau, M. & Voituriez, R. Enhanced reaction

kinetics in biological cells. Nature Phys. 4, 134–137 (2008).8. Mirny, L. Biophysics: Cell commuters avoid delays. Nature Phys. 4,

93–95 (2008).9. Benichou, O., Loverdo, C., Moreau, M. & Voituriez, R. Optimizing intermittent

reaction paths. Phys. Chem. Chem. Phys. 10, 7059–7072 (2008).10. Redner, S. A Guide to First Passage Time Processes (Cambridge University

Press, 2001).11. Shlesinger, M. F. Mathematical physics: first encounters. Nature 450, 40–41 (2007).12. Condamin, S., Benichou, O. & Moreau, M. First-passage times for random walks

in bounded domains. Phys. Rev. Lett. 95, 260601 (2005).13. Condamin, S., Benichou, O., Tejedor, V., Voituriez, R. & Klafter, J. First-passage

times in complex scale-invariant media. Nature 450, 77–80 (2007).14. Benichou, O. & Voituriez, R. Narrow-escape time problem: Time needed for a

particle to exit a confining domain through a small window. Phys. Rev. Lett. 100,168105 (2008).

15. Benichou, O., Meyer, B., Tejedor, V. & Voituriez, R. Zero constant formulafor first-passage observables in bounded domains. Phys. Rev. Lett. 101,130601 (2008).

16. Bouchaud, J.-P. & Georges, A. Anomalous diffusion in disordered media:Statistical mechanisms, models and physical applications. Phys. Rep. 195,127–293 (1990).

17. Kurzynski, M., Palacz, K. & Chelminiak, P. Time course of reactions controlledand gated by intramolecular dynamics of proteins: Proc. Natl Acad. Sci. USA 95,11685–11690 (1998).

18. Kolesov, G., Wunderlich, Z., Laikova, O. N., Gelfand, M. S. & Mirny, L. A. Howgene order is inuenced by the biophysics of transcription regulation. Proc. NatlAcad. Sci. USA 104, 13948–13953 (2007).

19. Fraser, P. & Bickmore, W. Nuclear organization of the genome and the potentialfor gene regulation. Nature 447, 413–417 (2007).

20. Zimmerman, S. B. & Minton, A. P. Macromolecular crowding: Biochemical,biophysical, and physiological consequences. Ann. Rev. Biophys. Biom.22 (1993).

21. Zhou, H.-X., Rivas, G. & Minton, A. P. Macromolecular crowding andconfinement: Biochemical, biophysical, and potential physiologicalconsequences. Ann. Rev. Biophys. 37, 375–397 (2008).

22. Ben-Avraham, D. & Havlin, S. Diffusion and Reactions in Fractals andDisordered Systems (Cambridge Univ. Press, 2000).

23. de Gennes, P. G. Kinetics of diffusion-controlled processes in dense polymersystems. i. nonentangled regimes. J. Chem. Phys. 76, 3316–3321 (1982).

24. Condamin, S., Benichou, O. & Moreau, M. Random walks and brownianmotion: a method of computation for first-passage times and related quantitiesin confined geometries. Phys. Rev. E 75, 021111 (2007).

25. O’Shaughnessy, B. & Procaccia, I. Diffusion on fractals. Phys. Rev. A 32,3073–3083 (1985).

26. Montroll, E. W. Random walks on lattices. iii. calculation of first-passage timeswith application to exciton trapping on photosynthetic units. J. Math. Phys. 10,753–765 (1969).

27. Kozak, J. J. & Balakrishnan, V. Analytic expression for the mean time toabsorption for a random walker on the sierpinski gasket. Phys. Rev. E 65,021105 (2002).

28. Agliari, E. Exact mean first-passage time on the t-graph. Phys. Rev. E 77,011128 (2008).

29. Haynes, C. P. & Roberts, A. P. Global first-passage times of fractal lattices. Phys.Rev. E 78, 041111 (2008).

30. Tejedor, V., Benichou, O. & Voituriez, R. Global mean first-passage times ofrandom walks on complex networks. Phys. Rev. E 80, 065104 (2009).

31. Metzler, R. & Klafter, J. The random walk’s guide to anomalous diffusion: afractionnal dynamics approach. Phys. Rep. 339, 1–77 (2000).

32. Wachsmuth, M., Waldeck, W. & Langowski, J. Anomalous diffusion of uorescentprobes inside living cell nuclei investigated by spatially-resolved uorescencecorrelation spectroscopy. J. Mol. Biol. 298, 677–689 (2000).

33. Golding, E. & Cox, E. Physical nature of bacterial cytoplasm. Phys. Rev. Lett. 96,981102 (2006).

34. Condamin, S., Tejedor, V., Voituriez, R., Benichou, O. & Klafter, J. Probingmicroscopic origins of confined subdiffusion by first-passage observables. Proc.Natl Acad. Sci. USA 105, 5675–5680 (2008).

35. Montroll, E. W. & Weiss, G. H. Random walks on lattices. ii. J. Math. Phys. 6,167–181 (1965).

Rescaled time (θ)

Sur

viva

l pro

babi

lity

Sur

viva

l pro

babi

lity

0.0001 0.001 0.01 0.10

0.2

0.4

0.6

0.8

1

r = 45r = 23r = 12r = 7r = 4r = 2Theory

Rescaled time (θ)

a b

0.0001 0.001 0.01 0.1 1 101 100

0.2

0.4

0.6

0.8

1

r = 38r = 20r = 10r = 5r = 2Theory

Figure 4 | Reaction kinetics as quantified by the survival probability S(t), plotted for different source–target distances r. a,b, The non-compact case (a) is

exemplified by 3D regular diffusion and the compact case (b) by diffusion on a Sierpinski gasket (with a target at the apex). The theoretical prediction for

S(t) is obtained from equations (5) and (6). Here u and PTS are calculated using exact results for kTTSl and kTlT. Quantitatively, the typical reaction time ttyp

can be defined, for example, through the median S(ttyp)¼ 1/2, indicated by the dotted line. In the non-compact case, ttyp weakly depends on the initial

position of the reactant, which therefore is not an important parameter of the kinetics. On the contrary, in the compact case, ttyp runs over several orders of

magnitude depending on the initial position, which, in turn, controls the kinetics.






36. Shlesinger, M. F. Asymptotic solutions of continuous-time random walks. J. Stat.Phys. 10, 421–434 (1974).

37. Condamin, S., Benichou, O. & Klafter, J. First-passage time distributions forsubdiffusion in confined geometry. Phys. Rev. Lett. 98, 250602 (2007).

38. Parson, R. P. & Kopelman, R. Percolative versus homogenous energy transportkinetics: time-resolved donor and acceptor uoresence of isotopic mixednaphthalene crystals. Chem. Phys. Lett. 87, 528–532 (1982).

39. Saxton, M. J. A biological interpretation of transient anomalous subdiffusion.ii. reaction kinetics. Biophys. J. 94, 760–771 (2008).

40. Malchus, N. & Weiss, M. Elucidating anomalous protein diffusion in livingcells with fluorescence correlation spectroscopy—facts and pitfalls.J. Fluoresc. (2009).

41. Monson, E. & Kopelman, R. Observation of laser speckle effects and nonclassicalkinetics in an elementary chemical reaction. Phys. Rev. Lett. 85, 666–669 (2000).

42. Markstein, M., Markstein, P., Markstein, V. & Levine, M. S. Genome-wideanalysis of clustered dorsal binding sites identifies putative target genes in thedrosophila embryo. Proc. Natl Acad. Sci. USA 99, 763–768 (2002).

43. Platani, M., Goldberg, I., Lamond, A. I. & Swedlow, J. R. Cajal body dynamicsand association with chromatin are atp-dependent. Nat. Cell Biol. 4,502–508 (2002).

44. Bancaud, A. et al. Molecular crowding affects diffusion and binding of nuclearproteins in heterochromatin and reveals the fractal organization of chromatin.EMBO J. 28, 3785–3798 (2009).

45. Lebedev, D. V. et al. Fractal nature of chromatin organization in interphasechicken erythrocyte nuclei: Dna structure exhibits biphasic fractal properties.FEBS Lett. 579, 1465–1468 (2005).

46. Lebedev, D. et al. Structural hierarchy of chromatin in chicken erythrocyte nucleibased on small-angle neutron scattering: Fractal nature of the large-scalechromatin organization. Crystallogr. Rep. 53, 110–115 (2008).

AcknowledgementsSupport from ANR grants DYOPTRI and DYNAFT is acknowledged.

Author contributionsAll authors contributed equally to this work.

Additional informationThe authors declare no competing financial interests. Supplementary informationaccompanies this paper at www.nature.com/naturechemistry. Reprints and permissioninformation is available online at http://npg.nature.com/reprintsandpermissions/.Correspondence and requests for materials should be addressed to O.B.










Efficient stereo- and regioselective hydroxylationof alkanes catalysed by a bulky polyoxometalateKeigo Kamata1,2, Kazuhiro Yonehara1, Yoshinao Nakagawa1, Kazuhiro Uehara1,2

and Noritaka Mizuno1,2*

Direct functionalization of alkanes by oxidation of C–H bonds to form alcohols under mild conditions is a challenge forsynthetic chemistry. Most alkanes contain a large number of C–H bonds that present difficulties for selectivity, and theoxidants employed often result in overoxidation. Here we describe a divanadium-substituted phosphotungstate thatcatalyses the stereo- and regioselective hydroxylation of alkanes with hydrogen peroxide as the sole oxidant. Both cyclicand acyclic alkanes were oxidized to form alcohols with greater than 96% selectivity. The bulky polyoxometalateframework of the catalyst results in an unusual selectivity that can lead to the oxidation of secondary rather than theweaker tertiary C–H bonds. The catalyst also avoids wasteful decomposition of the stoichiometric oxidant, which can resultin the production of hydroxyl radicals and lead to non-selective oxidation and overoxidation of the desired products.

The selective transformation of inert C–H bonds of alkanes intouseful functional groups has attracted much attention becausealkanes are less expensive and more readily available than the

current petrochemical feedstocks1–3. The catalytic hydroxylation ofalkanes under mild conditions remains a major challenge in indus-trial and synthetic chemistry4–7. Among various oxidants, molecularoxygen is used widely in homogeneous and heterogeneous industrialprocesses. However, the scope of homogeneous aerobic oxidationreactions is narrow because autoxidation reactions are only compa-tible with the production of terephthalic acid and cyclohexanone.Although hydrogen peroxide (H2O2) is an economically and envir-onmentally desirable oxidant in comparison with peracids8, organichydroperoxides9, N-oxides10,11 and iodosobenzene12, the combi-nation of metal complexes with H2O2 often generates hydroxylradicals. The disadvantages of H2O2-based systems are:

† selectivity for alcohols is low because of overoxidation;† non-productive decomposition of H2O2 leads to low efficiencies

of H2O2 utilization;† indiscriminate attack of hydroxyl radicals results in low stereo-

specificity and regioselectivity.

Therefore, efficient H2O2-based catalytic oxidation systems werelimited to a few examples of organocatalysts13 and transition-metalcatalysts, such as iron14–18, vanadium19–21, osmium22 and manga-nese23,24, and the development of efficient, stereospecific and regio-selective hydroxylation of various kinds of alkanes with H2O2 undermild conditions is in great demand.

Natural enzymes (for example, methane monooxygenases andfatty-acid desaturases) use molecular recognition, such as sizeand/or shape selectivity and substrate orientation, to achieve highchemo-, regio- and stereoselectivities25. On the basis of these strat-egies, manganese- and iron-based biomimetic selective oxidationswere studied24–26. However, these systems need a directing car-boxylic acid group in the substrate to achieve the high selectivity,and also susceptibility of the organic ligands to oxidative self-degra-dation has limited their usefulness. In addition, the activated C–H

bonds (that is, adjacent to a heteroatom, a p-system and/or anelectron-rich tertiary C–H bond) are hydroxylated selectivelyin most metal-catalysed oxidation systems. Polyoxometalates(transition-metal oxygen anion clusters) were applied in variousfields, such as structural chemistry, analytical chemistry, surfacescience, medicine, electrochemistry, photochemistry and catalysis27–31.Our approach to the design of active catalysts for the regioselectivehydroxylation of alkanes is to create strong electrophilic oxidantswith high steric hindrance based on the concepts of:

† synthesis of a divanadium site for the cooperative activationof H2O2;

† control of the nature of the oxidant by changing the hetero atom;† steric protection of the catalyst by using a rigid and bulky all-

inorganic polyoxometalate ligand32–35.

In this paper, we report the efficient, stereospecific and regio-selective hydroxylation of alkanes with H2O2 catalysed by a divana-dium-substituted phosphotungstate, [g-H2PV2W10O40]32 (1a).High selectivity for alcohols, efficiency of H2O2 utilization and stereo-specificity were observed. This system also showed specific regioselec-tivity for secondary alcohols in the oxidation of some cycloalkanesthat have both secondary and tertiary C–H bonds. This study pro-vides the first example of a synthetic catalyst that can achieve specific,regioselective, H2O2-based hydroxylation of secondary C–H bonds,even in the presence of more reactive tertiary C–H bonds.

ResultsCompound 1 was synthesized by the cation-exchange reaction of thecaesium salt of deprotonated [g-PV2W10O40]52 with an alkyl-ammonium and subsequent protonation with perchloric acid. AnX-ray crystallographic structural analysis of anion 1a was carriedout on the tetraethylammonium salt derivative. 1a had twovanadium atoms in the g-Keggin-type polyoxotungstate structure(Fig. 1a). The bond-valence sums of vanadium (5.20), tungsten(5.98–6.12) and phosphorous (5.17) indicated that the anion wascomposed of vanadium(V), tungsten(VI) and phosphorous(V) ions.

1Department of Applied Chemistry, School of Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan, 2Core Research forEvolutional Science and Technology, Japan Science and Technology Agency, 4-1-8 Honcho, Kawaguchi, Saitama, 332-0012, Japan.

*e-mail: [email protected]







The bond-valence sum of atom O1 (1.34) was lower than those ofthe other oxygen atoms (1.71–2.08), which indicates that O1 andO1′ are protonated (hydroxo ligands). The 31P NMR spectrum inCD3CN showed a single line at 213.8 ppm. The 51V NMR spectrumin CD3CN gave a single line at 2578 ppm, which shows two equiv-alent vanadium atoms. The 1H NMR spectrum in CD3CN gave aline at 7.0 ppm (2H per anion), which can be assigned to thebis(m-hydroxo) groups. All NMR data were consistent with thecharacterization of 1a in CD3CN as a single species of formula[g-H2PV2W10O40]32.

The oxidation of various cyclic alkanes with H2O2 catalysed by1a was investigated (Table 1). The efficiency of H2O2 utilizationwas more than 80% in each case (Table 1, entries 1–6). The oxidationof cyclic alkanes 2–6 proceeded selectively to give the correspondingalcohols 8–12 (≥98% selectivity) without significant formation ofketones, carboxylic acids and products of cleaved C–C bonds(Table 1, entries 1–5). Under stoichiometric conditions (Table 1)the selectivity of the oxidation of cyclohexane 2 (50 mM) to givecyclohexanol 8 was 87–98% (with a conversion range of 1.1–8.2%),a value larger than or comparable to those of the hydroxyl radical(�50%) (ref. 36), Fe(III)-porphyrin/iodosobenzene (83–94%)(ref. 36) and H2O2-based vanadium (50–90%) (refs 19–21), iron(0–81%) (refs 14–18), manganese (�50%) (ref. 23) and osmium(70%) (ref. 22) systems. Secondary alcohols react very poorly withthe bis(m-hydroxo) divanadium site in [g-H2SiV2W10O40]42

because of the steric crowding between the polyoxometalate frame-work and substrates. Therefore, this high selectivity with 2 resultsfrom suppression of the oxidation of 8 by the steric hindrance of1a. Adamantane (5) was hydroxylated preferentially at the electron-rich tertiary C–H bonds (Table 1, entry 4). The selectivity ratio oftertiary/secondary (38/28) C–H activation normalized to thenumber of C–H bonds was 18. This value is comparable to thoseof cytochrome P450 and haem catalysts (6�48) and much larger

than those of hydroxyl (2) and t-butoxy radical (10) systems36.Also, acyclic n-hexane (7) was hydroxylated to give the correspondingalcohols (13a–13c) with ≥96% selectivity (Table 1, entry 6).

This system was applied on a larger scale (scaled up 33-fold) tohydroxylations of cycloalkanes; 0.29 g of 8 (85% isolated yield basedon H2O2), 0.37 g of cyclooctanol (9) (80%) and 0.47 g of cyclodode-canol (10) (77%) were isolated with 99% purity according to 1HNMR spectroscopy (Table 1, entries 1–3). The turnover frequencyand efficiency of H2O2 utilization for the oxidation of 2 were710 h21 and 90%, respectively, the highest values among thosereported for H2O2-based catalytic systems so far:

† [Fe(TPA)(CH3CN)2](ClO4)2, 17 h21, 43% (ref. 14) (TPA¼tris(2-pyridylmethyl)amine);

† [(N4Py)Fe(CH3CN)](ClO4)2, 89 h21, 44% (ref. 15) (N4Py¼N,N-bis(2-pyridylmethyl)-N-bis(2-pyridyl)methylamine);

† [Fe(CF3SO3)2((S,S,R)-mcpp)], 296 h21, 66% (ref. 18) ((S,S,R)-mcpp¼N,N′-dimethyl-N,N′-bis[(R)-[4,5]-pineno-2-pyridylmethyl][(1S,2S)-1,2-cyclohexanediamine]);

† [V(O)(Cl)(PBHA)2], 19 h21, 8% (ref. 19) (PBHA¼N-phenylbenzohydroxamate);

† [(VO)4(hptb)2(H2O)2(m-O)](ClO4)4, 98 h21, 12% (ref. 20) (hptb¼N,N,N,N′-tetrakis(2-benzimidazolylmethyl)-2-hydroxo-1,3-diaminopropane);

† VOSO4/HNO3, 9 h21, 11% (ref. 21).

Although the turnover frequencies of [Fe(III)2OL2(NO3)2(CH3OH)2](NO3)2 (785 h21) (ref. 17) (L¼ 2,6-bis(N-methylbenzi-midazol-2-yl)pyridine), [Mn2O3(TMTACN)](PF6)2/oxalic acid(1,500 h21) (ref. 23) (TMTACN¼ 1,4,7-trimethyl-1,4,7-triazacyclo-nonane) and [(n-C4H9)N][Os(N)Cl4]/FeCl3/CH3COOH (840 h21)(ref. 22) were higher than that of 1a, the selectivity to give 8(17–68%) was lower than that of 1a.

CH3CN/t-BuOH (0.67/1.33 ml)333 K, 120 minutes

[(n-C4H9)4N]4[γ-HPV2W10O40] (1.3 mM), HClO4 (1.3 mM)24% H2

18O2 (50 mM; 18O content: 90%) 80% yield98% selectivity18O content: 90 ± 1%

18OH

82 (9.7 M)

V101' V101

O1'

O1

O101' O101

O3'

O2'

O2

O3

V

HO

VOH

V

HO

VO

H

VO

VO

H2O

H2O

H2O

+ H2O

H2O

H2O2

H2O2

R1

CR3

R2

H

R1

CR3

R2

OH

Step 3Step 1

Step 2

1a

1b1c

[γ-PV2W10O38(OH)2]3−

[γ-PV2W10O38(OH)(OOH)]3−[γ-PV2W10O38(O2)]3−

O

a

c

b

Figure 1 | Structure and reactivity of a divanadium-substituted phosphotungstate catalyst for the selective oxidation of alkanes. a, X-ray crystal structure

of 1a (the tetraethylammonium counterion is omitted for clarity). Yellow, grey, green and red circles represent vanadium, tungsten, phosphorus and oxygen,

respectively. b, Proposed mechanism for the hydroxylation of alkanes with H2O2 catalysed by 1a. c, Hydroxylation of cyclohexane catalysed by 1a using

H218O2 gave cyclohexanol in 80% yield. The 18O content of the alcohol product (90+1%) remained constant throughout the reaction, which shows that all

the oxygen atoms in the product originated from the H2O2 oxidant.












http://www.nature.com/compfinder/10.1038/nchem.648_comp13a

http://www.nature.com/compfinder/10.1038/nchem.648_comp13c









The regioselectivity for the hydroxylation of cycloalkanes 14–19,with both secondary and tertiary C–H bonds, was investigated(Table 2). For all substrates, stereospecific hydroxylationswere observed. The high stereospecificity is comparable to those(92–99%) of stoichiometric organic oxidants, such as perfluorodi-alkyldioxiranes, perfluorodialkyloxaziridines and p-nitroperoxy-benzoic acid, for which a concerted mechanism mediated by anelectrophilic oxidant has been proposed37–39. In systems catalysedby organometallic complexes that do not have bulky ligands and instoichiometric systems with dioxirane complexes, the activatedtertiary C–H bonds of cycloalkanes with both secondary and tertiaryC–H bonds are hydroxylated selectively to give the correspondingtertiary alcohols (Table 3 and Supplementary Tables S1 and S2)40–47.However, the unactivated secondary C–H bonds were hydroxylatedselectively with product ratios (28 alcohols)/(38 alcohols) thatranged from 78/22 to .99/,1 for the cycloalkanes listed inTable 2. Such specific selectivities are different from those of radicalreactions and are observed for the reaction of platinum(II)complexes with alkanes via oxidative addition, in which stericfactors play an important role in determining the unusual selectiv-ity48. For the oxidation of trans-1,2-dimethylcyclohexane (14), selec-tivity for the secondary alcohols was 90% (Table 2, entry 1) andthe value was higher than that (80%) of the sterically hindered met-alloporphyrin Mn(II)(TPFPP)(ClO4)/m-chloroperoxybenzoic acid(m-CPBA) (TPFPP¼meso-tetrakis(pentafluorophenyl)porphinato)system8 and much higher than those of H2O2-based oxidationsystems, such as [(N4Py)Fe(CH3CN)](ClO4)2 (,1%) (ref. 15),[(n-C4H9)4N][Os(N)Cl4]/FeCl3/CH3COOH (69%) (ref. 22) and[Fe(TPA)(CH3CN)2](ClO4)2 (50%) (ref. 14). In addition, for theoxidation of 14 and trans-decalin (15), with two adjacent tertiary

C–H groups, the selectivities for trans-3,4-dimethylcyclohexanol(20c) and trans-2-decalinol (21a) were 86% and 93%, respectively(Table 2, entries 1 and 2), and mixtures of various secondary alcoholsand the corresponding ketones have been obtained for the iron- andosmium-based systems14,15,22. Such high regioselectivities for thesecondary alcohols in the presence of the more electron-rich tertiaryC–H bonds have not been reported before. Also, monoalkyl-substi-tuted cyclohexanes, such as methylcyclohexane (17), ethylcyclohex-ane (18) and t-butylcyclohexane (19), hydroxylated selectively togive the corresponding secondary alcohols with ≥81% selectivity(Table 2, entries 4–6). 5 is an exception to this pattern of oxidationof cyclic alkanes that have both secondary and tertiary C–H bonds.The six-membered rings of both 5 and the cyclic alkanes 14–19have chair conformations, but the tertiary C–H bonds of 5 are atthe equatorial positions and the tertiary C–H bonds of the cyclicalkanes 14–19 are at the axial positions. Therefore, the steric hin-drance around the tertiary C–H bonds of 5 is much smaller thanthose of the cyclic alkanes 14–19. Such a difference results in thelow selectivity for secondary alcohol in the oxidation of 5.

The competitive oxidation between 2 and cyclohexane-d12 showsan intermolecular kinetic isotope effect (the ratio of the hydrogenand deuterium reaction rate constants, kH/kD) of 3.2, larger thanthose (1�2) associated with radical chain autoxidations36. Theaddition of the radical scavenger 2,6-di-t-butyl-4-methylphenol(five equivalents with respect to 1a) did not affect the reactionrate, selectivity and total yield for the oxidation of 5. All the data,including the alcohol/ketone ratio for cyclohexane oxidation,38/28 for the oxidation of 5, regioselectivity, stereospecificity andkinetic isotope effect show the formation of a non-free-radical,electrophilic, metal-based oxidant in 1a.

Table 1 | H2O2 hydroxylation of alkanes catalysed by 1a.

Entry Substrate Time (h) Yield (%) Product (selectivity (%)) H2O2 efficiency (%)

1*

2

1 92 (85)OH

8 (98)

94

2†

3

1 84 (80)OH

9 (99)

84

3‡

4

1 79 (77)OH

10 (98)

80

4§

5

2 98OH

OHOH

OH

11a (82) 11b (15) 11c (3)

98

5

6

2 80 OH

OH12a (94) 12b (6)

80

6‖7

4 56OH

OH

OH

(2) (26)13a(66)13b

13c

56

Reaction conditions: [(n-C4H9)4N]4[g-HPV2W10O40] (1.3 mM), HClO4 (1.3 mM), substrate (2.5 M), CH3CN/t-BuOH (0.67/1.33 ml), 30% aqueous H2O2 (50 mM), 333 K. Yield (%)¼ products (mol)/initialH2O2 (mol) × 100. H2O2 efficiency (%)¼ (alcohols (mol)þ 2 × ketones (mol))/consumed H2O2 (mol) × 100. The values in parentheses are isolated yields. *2 (4.7 M), cyclohexanone (2% selectivity).†Cyclooctanone (2% selectivity). ‡Cyclododecanone (2% selectivity). §5 (0.3 M). ‖7 (7.5 M), 342 K, 2-hexanone (4% selectivity).

































The reactivity of 1a in dichloroethane with H2O2 (90–95%aqueous solution) at 253 K was investigated. The cold-spray ioniz-ation mass spectrum of 1a in 1,2-dichloroethane shows that themost intense parent-ion peak (centred at m/z 3,583) has an iso-topic distribution that agrees with the pattern calculatedfor [((n-C4H9)4N)4H2PV2W10O40]þ and a fragment peak(centred at m/z 3,565) with an isotopic distribution that agreeswith the pattern calculated for [((n-C4H9)4N)4PV2W10O39]þ

(Supplementary Fig. S3). On addition of H2O2, a new peakcentred at m/z (3,583þ 16) appeared and agrees with the patterncalculated for [((n-C4H9)4N)4HPV2W10O39(OOH)]þ. The 1HNMR spectrum of 1a in dichloroethane showed a signal at5.48 ppm from the hydroxyl proton of the (VO-(m-OH)2-VO)core in 1a (Supplementary Fig. S1a). Two new 1H NMR signalsappeared at 9.49 and 5.31 ppm with intensity ratios of 1:1 on theaddition of H2O2 (Supplementary Fig. S1b). The intensity of thesignal at 5.48 ppm decreased and the sum of the three signal inten-sities remained constant. The chemical shift of 9.49 ppm is close tothose of [g-SiV2W10O38(OH)(OOH)]42 (9.45 ppm) and otherhydroperoxide species such as H2O2 (8.74 ppm), t-butyl hydro-peroxide (8.84 ppm) and cumen hydroperoxide (8.95 ppm) (refs33,34). Therefore, the 1H NMR signals at 9.49 and 5.31 ppm canbe assigned to the protons of the hydroperoxo and hydroxo

groups in [g-PV2W10O38(OH)(OOH)]32 (1b), respectively. Onenew 51V NMR signal appeared at –539 ppm on the addition ofH2O2, and the intensity of the signal of 1a at –574 ppm decreased,although the sum of the two signal intensities remained almost constant(Supplementary Fig. S2). The oxo� hydroperoxo transformationsproduced 15–30 ppm (from [HxMoO2(O2)2](2–x)– (0 ≤ x ≤1) to[MoO(O2)2OOH]2

22) and 32 ppm (from [g-SiV2W10O38(OH)2]42 to[g-SiV2W10O38(OH)(OOH)]42) downfield shifts in the 95Mo and51V NMR signals, respectively33,34,49. The downfield shift of35 ppm in our experiments is in accordance with the oxo � hydro-peroxo transformation. Detection of 1b by 183W NMR spectroscopywas unsuccessful because the solubility of 1b in 1,2-dichloroethaneis very low, below the detection limit of 183W nuclei. In CH3CN/t-BuOH (volume/volume¼ 1/2), a new signal at –630 ppm (1c)was observed in addition to the two signals of 1a and 1b(Supplementary Fig. S4). The effects of H2O2 and H2O on the for-mation of 1b and 1c were investigated (Supplementary Fig. S5). Theratio of [1b]/[1a] was proportional to [H2O2]/[H2O], which indi-cates the reversible formation of 1b. The ratio of [1c]/[1a] was pro-portional to [H2O2]/[H2O]2, which suggests the successivedehydration of 1b to form 1c. The two vanadium atoms in 1c areequivalent because only one 51V NMR signal was observed for 1c.In addition, 1c was formed by the dehydration of 1b. Therefore,

Table 2 | Regioselective H2O2 hydroxylation of alkanes to give secondary alcohols catalysed by 1a.

Entry Substrate Time (h) Yield (%) Product (selectivity (%)) [288888alcohols]/[388888 alcohols]

1

14

1 59

OH

OHHO

(10) (86)(4)

20a 20b 20c

90/10

2*H

H15

1 51H

H

OH

(93)

21a

.99/,1

3† H

H16

1 72 H

H

H

H

OH

H

OH

OH

(22) (36) (40)

22a 22b 22c

78/22

4‡

17

1 75 OH

OHOH

OH

(19) (6) (44) (24)

23a 23b 23c 23d

81/19

5§

18

2 64 OH

OH

OH

OH

(3) (7) (53) (25)

OH

(4)

24a 24b 24c 24d 24e

97/3

6‖

19

2 67

OH

OH

(63) (24)

25a 25b

.99/,1

Hydrolysis reaction conditions: [(n-C4H9)4N]4[g-HPV2W10O40] (1.3 mM), HClO4 (1.3 mM), substrate (2.5 M), CH3CN/t-BuOH (0.67/1.33 ml), 30% aqueous H2O2 (50 mM), 333 K. Yield (%)¼ products(mol)/initial H2O2 (mol) × 100. *Trans-2-Decalone (7% selectivity). †Cis-2-Decalone (1% selectivity). ‡3-Methylcyclohexanone (4% selectivity) and 4-methylcyclohexanone (2% selectivity).§3-Ethylcyclohexanone (6% selectivity) and 4-ethylcyclohexanone (2% selectivity). ‖3-Butylcyclohexanone (10% selectivity) and 4-butylcyclohexanone (3% selectivity).






the m-h2:h2-peroxo species [g-PV2W10O38(O2)]32 (1c) is a poss-ible active species similar to that used in epoxidation catalysed by[g-H2SiV2W10O40]42 catalysed epoxidation (refs 33,34). Kineticstudy of the 1a-catalysed hydroxylation revealed first-order depen-dencies of the reaction rate on the concentrations of 1a (0.25–1.25 mM), 5 (35–250 mM) and H2O2 (15–80 mM), and an inverselysecond-order dependency on the concentration of water (190–530 mM). These kinetic results suggest that [g-PV2W10O38(O2)]32

(1c) is the active species for the hydroxylation and that the reactionof an alkane with 1c is the rate-determining step (Fig. 1b). Thesteric effect of the all-inorganic, rigid and bulky polyoxometalate fra-mework in 1c leads to a specific regioselectivity for the hydroxylationof alkyl-substituted cycloalkanes.

To investigate the origin of the oxygen atom incorporated intothe alcohol, hydroxylation of 2 with H2

18O2 (24% solution, 90%enriched) catalysed by 1a was carried out. This gave 8 (98% selectiv-ity) in 80% yield with an 18O content of 90+1% and the 18Ocontent in the alcohol (18O-labelled alcohol/total alcohol ratio)did not change during the reaction (Fig. 1c), which shows that allthe 18O atoms incorporated into the alcohol originated from H2

18O2.

ConclusionsThe bis(m-hydroxo) divanadium-substituted phosphotungstate 1areacted easily with H2O2 to produce a highly active, sterically hin-dered oxidant species. The strong electrophilic oxidants with highsteric hindrance hydroxylated alkanes with high selectivity for thealcohols, complete stereospecificity and specific regioselectivity.Such unique properties of all-inorganic molecular polyoxometalatesare promising as homogeneous and heterogeneous catalysts forstereo- and shape-selective H2O2-based hydroxylation of C–H bonds.

MethodsSynthesis and characterization of divanadium-substituted phosphotungstate(1a). The caesium salt of deprotonated divanadium-substituted phosphotungstateCs5[g-PV2W10O40] was synthesized according to published literature procedures50

and characterized by infrared spectroscopy. The tetra-n-butylammonium salt of themonoprotonated derivative [(n-C4H9)4N]4[g-HPV2W10O40] was prepared using acation-exchange reaction. Sodium metavanadate (1.2 mmol) was dissolved in120 ml of hot water. On cooling, the pH of the solution was adjusted to 2.0 with 3 MHCl. Cs5[g-PV2W10O40].6H2O (3.8 g, 1.1 mmol) was dissolved in the solutionand the insoluble materials removed by filtration. Tetra-n-butylammonium bromide

(1.8 g, 5.6 mmol) was added with vigorous stirring. The precipitate was collectedby filtration, washed with 400 ml of water and dried in vacuo. Recrystallizationfrom acetone/ether gave analytically pure orange crystals of [(n-C4H9)4N]4[g-HPV2W10O40].H2O, with a yield of 60%. 51V NMR (CD3CN), –581 ppm;1H NMR (CD3CN), 4.38 ppm (1H); 31P NMR (CD3CN), –14.1 ppm. Analyticalcalculation for [(C4H9)4N]4[HPV2W10O40].H2O was C, 21.4; H, 4.12; N, 1.56; P,0.86; V, 2.83; W, 51.1; found was C, 21.3; H, 3.96; N, 1.61; P, 0.84; V, 2.92; W, 49.0.Infrared (KBr) 1,096, 1,062, 1,039, 1,001, 952, 870, 803, 752, 534, 489, 399, 358, 333,282, 256 cm21. The tetraethylammonium (Et4N) salt of diprotonated derivative[(C2H5)4N]3[g-H2PV2W10O40] (Et4N)3

.1a) was prepared by the reaction of[g-HPV2W10O40]42 with an acid followed by a cation-exchange reaction. Perchloricacid (0.1 mmol) and tetraethylammonium bromide (76 mg, 0.36 mmol) dissolvedin CH3CN (0.5 ml) was added to the CH3CN solution (10 ml) of [(n-C4H9)4N]4[g-HPV2W10O40].H2O (216 mg, 60 mmol). Diethyl ether (20 ml) was added to theclear yellow solution and the resulting precipitate collected by centrifugation. Thecrude product was dissolved in CH3CN (10 ml), followed by the addition of1,4-dioxane (10 ml). Standing the solution in an open vessel for a few hours gaveyellow crystals of [(C2H5)4N]3[H2PV2W10O40].3C4H8O2, with a yield of 120 mg(63%). 51V NMR (CD3CN), –578 ppm; 1H NMR (CD3CN), 6.95 (2H, V-OH-V), 3.59(24H, 1,4-dioxane), 3.19 (24H, cation), 1.23 ppm (36H, cation); 31P NMR (CD3CN),–13.8 ppm. Analytical calculation for [(C2H5)4N]3[H2PV2W10O40].3C4H8O2 was C,13.2; H, 2.65; N, 1.29; P, 0.95; V, 3.17; W, 56.3; found was C, 13.6; H, 2.73; N, 1.37; P,0.95; V, 3.15; W, 58.6. Infrared (KBr): 1,119 (1,4-dioxane), 1,094, 1,060, 1,040, 1,011sh,972, 957, 871 (1,4-dioxane), 803, 716, 606, 585, 537, 490, 413, 397, 358, 333, 311,279, 255 cm21. Crystal data (153 K): C24N3O40PV2W10; formular weight 2,941.62,orthorhombic, space group Pbcn (No. 60), a¼ 15.5486(2) Å, b¼ 21.1547(2) Å,c¼ 23.1083(2) Å, V¼ 7600.92(14) Å3, Z¼ 4, Dcalcd¼ 2.571 g cm23, m (Mo Ka)¼15.39 cm21, R1¼ 0.0761 (I . 2s(I)), wR2¼ 0.2302 (all 10,413 data) and 194parameters were used for refinement. CCDC 746253 contains the supplementarycrystallographic data, which can be obtained free of charge through www.ccdc.cam.ac.uk/data_request/cif.

Typical procedure for the catalytic oxidation of alkanes. The catalytic reactionswere carried out with a glass tube that contained a magnetic stir bar. The catalyst,solvent and substrate were charged in the reaction vessel. For 1a-catalysed reactions,diprotonated 1a was prepared in situ by the reaction of monoprotonated[(n-C4H9)4N]4[g-HPV2W10O40] with one equivalent of perchloric acid (70%aqueous solution) because of the low solubility of (Et4N)3

.1a in the solvent. Theformation of 1a was confirmed by 51V NMR spectroscopy (–578 ppm). The reactionwas initiated by the addition of 30% aqueous H2O2. The reaction solution wasanalysed periodically by gas chromatography, gas chromatography–massspectroscopy and NMR spectroscopy. Ce3þ/4þ titration showed that no H2O2remained after the reaction. All products are known and identified by comparison oftheir 1H and 13C NMR signals with the literature data.

Received 1 December 2009; accepted 22 March 2010;published online 2 May 2010

Table 3 | Regioselectivity for the oxidation of 14.

OHOH

HO

O

O

+ + + +

26 27 28 29 3014

Entry System Yield (%) Selectivity (%) [288888 alcohols]/[388888 alcohols]

[cis-26]/[trans-26]

Eox

(%)Ref.

26 27 28 29 30 Others

1 1a/H2O2 (this work) 59 10 4 86 – – – 90/10 0/.99 59 –2 CH3ReO3/H2O2 20 – – – – – .98* 0/.99 – 39 403† [Fe(II)(TPA)(CH3CN)2](ClO4)2/H2O2 23 50 50 – – – 50/50 0/.99 23 144‡ [(N4Py)Fe(II)(CH3CN)](ClO4)2/H2O2 2 .99 – – – – – 0/.99 58/42 2 155§ [Mn(IV)2O3(TMTACN)2](PF6)2/peracetic acid 7 33 20 47 – 67/33 9/91 10 416‖ [Mn(II)(BQEN)(CF3SO3)2]/peracetic acid 48 38 62 – 62/38 0/.99 .48 427},# Co(III)(TPFPP)(CF3SO3)/m-CPBA 51 84 16 – – 16/84 0/.99 51 438},** Mn(II)(TPFPP)(ClO4)/m-CPBA 45 20 80 – – – 80/20 0/.99 45 89†† (n-Bu4N)[Os(VIII)(N)O3]/Fe(III)Cl3/Cl2PyO 66 30 27 27 15 – 69/31 ,5/.95 76 1010},§§ Fe(III)(TPFPP)Cl/PhI(OAc)2 27 37 63 – – – 63/37 0/.99 27 4511 Cr(VI)O2(OAc)2/H5IO6 18 .99 – – – – – 0/.99 0/.99 18 4612‖‖ Methyl(trifluoromethyl)dioxirane/TFP 89 90 – – 6 4 – 10/90 0/.99 98 3813 Dimethyldioxirane 45 .99 – – – – – 0/.99 0/.99 45 4714 Perfluoro cis-2-n-butyl-3-n-propyloxaziridine 67 .99 – – – – – 0/.99 0/.99 67 39

The yield (%) for the regioselectivity is (26þ 27þ 28þ 29þ 30þ others) (mol)/oxidant used (mol) × 100; selectivity (%)¼ product (mol)/total products (mol) × 100; Eox (%)¼ (26þ 27þ 28þ 29 × 2þ30 × 2þ others) (mol)/oxidant used (mol) × 100. *7-Hydroxy-2-octanone (.99% selectivity). †TPA¼ tris(2-pyridylmethyl)amine, 27þ 28 (50% selectivity). ‡N4Py¼N,N-bis(2-pyridylmethyl)-N-bis(2-pyridyl)methylamine. §TMTACN¼ 1,4,7-trimethyl-1,4,7-triazacyclononane, 27þ 28 (20% selectivity) and 29þ 30 (47% selectivity). ‖BQEN¼N,N′dimethyl-N,N′-bis(8-quinolyl)ethane-1,2-diamine, 27þ28þ 29þ 30 (62% selectivity). }TPFPP¼meso-tetrakis(pentafluorophenyl)porphinato. #27þ 28 (16% selectivity). **27þ 28 (80% selectivity). ††Cl2PyO¼ 2,6-dichloropyridine N-oxide. §27þ 28 (63%selectivity). ‖‖TFP¼ 1,1,1-trifluoropropanone.






www.ccdc.cam.ac.uk/data_request/cif

www.ccdc.cam.ac.uk/data_request/cif





References1. Jia, C., Kitamura, T. & Fujiwara, Y. Catalytic functionalization of arenes and

alkanes via C–H bond activation. Acc. Chem. Res. 34, 633–639 (2001).2. Labinger, J. A. & Bercaw, J. E. Understanding and exploiting C–H bond

activation. Nature 417, 507–514 (2002).3. Schroder, D. & Schwarz, H. Gas-phase activation of methane by ligated

transition-metal cations. Proc. Natl Acad. Sci. USA 105, 18114–18119 (2008).4. Punniyamurthy, T., Velusamy, S. & Iqbal, J. Recent advances in transition metal

catalyzed oxidation of organic substrates with molecular oxygen. Chem. Rev.105, 2329–2363 (2005).

5. Cavani, F. & Teles, J. H. Sustainability in catalytic oxidation: an alternativeapproach or a structural evolution? ChemSusChem 2, 508–534 (2009).

6. Mizuno, N. Modern Heterogeneous Oxidation Catalysis (Wiley, 2009).7. Backvall, J.-E. Modern Oxidation Methods (Wiley, 2004).8. Nam, W., Ryu, J. Y., Kim, I. & Kim, C. Stereoselective alkane hydroxylations

by metal salts and m-chloroperbenzoic acid. Tetrahedron Lett. 43,5487–5490 (2002).

9. Foster, T. L. & Caradonna, J. P. Fe2þ-catalyzed heterolytic RO–OH bondcleavage and substrate oxidation: a functional synthetic non-heme ironmonooxygenase system. J. Am. Chem. Soc. 125, 3678–3679 (2003).

10. Yiu, S.-M., Wu, Z.-B., Mak, C.-K. & Lau, T.-C. FeCl3-activated oxidation ofalkanes by [Os(N)O3]2. J. Am. Chem. Soc. 126, 14921–14929 (2004).

11. Wang, C., Shalyaev, K. V., Bonchio, M., Carofiglio, T. & Groves, J. T. Fastcatalytic hydroxylation of hydrocarbons with ruthenium porphyrins. Inorg.Chem. 45, 4769–4782 (2006).

12. Traylor, T. G., Hill, K. W., Fann, W.-P., Tsuchiya, S. & Dunlap, B. E. Aliphatichydroxylation catalyzed by iron(III) porphyrins. J. Am. Chem. Soc. 114,1308–1312 (1992).

13. Litvinas, N. D., Brodsky, B. H. & Du Bois, J. C–H hydroxylation using a heterocycliccatalyst and aqueous H2O2. Angew. Chem. Int. Ed. 48, 4513–4516 (2009).

14. Kim, C., Chen, K., Kim, J. & Que, L. Jr Stereospecific alkane hydroxylation withH2O2 catalyzed by an iron(II)-tris(2-pyridylmethyl)amine complex. J. Am.Chem. Soc. 119, 5964–5965 (1997).

15. Roelfes, G., Lubben, M., Hage, R., Que, L. Jr & Feringa, B. L. Catalytic oxidationwith a non-heme iron complex that generates a low-spin FeIIIOOH intermediate.Chem. Eur. J. 6, 2152–2159 (2000).

16. Chen, M. S. & White, M. C. A predictably selective aliphatic C–H oxidationreaction for complex molecule synthesis. Science 318, 783–787 (2007).

17. Wang, X., Wang, S., Li, L., Sundberg, E. B. & Gacho, G. P. Synthesis, structure,and catalytic activity of mononuclear iron and (m-oxo)diiron complexes with theligand 2,6-bis(N-methylbenzimidazol-2-yl)pyridine. Inorg. Chem. 42,7799–7808 (2003).

18. Gomez, L. et al. Stereospecific C–H oxidation with H2O2 catalyzed by achemically robust site-isolated iron catalyst. Angew. Chem. Int. Ed. 48,5720–5723 (2009).

19. Si, T. K., Chowdhury, K., Mukherjee, M., Bera, D. C. & Bhattacharyya, R.Homogeneous selective peroxidic oxidation of hydrocarbons using anoxovanadium based catalyst. J. Mol. Catal. A 219, 241–247 (2004).

20. Suss-Fink, G., Cuervo, L. G., Therrien, B., Stoeckli-Evans, H. & Shul’pin, G. B.Mono and oligonuclear vanadium complexes as catalysts for alkane oxidation:synthesis, molecular structure, and catalytic potential. Inorg. Chim. Acta 357,475–484 (2004).

21. Kirillova, M. V. et al. Group 5–7 transition metal oxides as efficient catalysts foroxidative functionalization of alkanes under mild conditions. J. Catal. 248,130–136 (2007).

22. Yiu, S.-M., Man, W.-L. & Lau, T.-C. Efficient catalytic oxidation of alkanes byLewis acid/[OsVI(N)Cl4]2 using peroxides as terminal oxidants. Evidence for ametal-based active intermediate. J. Am. Chem. Soc. 130, 10821–10827 (2008).

23. Shul’pin, G. B. et al. Oxidations by the system ‘hydrogen peroxide–[Mn2L2O3][PF6]2 (L¼1,4,7-trimethyl-1,4,7-triazacyclononane)–carboxylicacid’. Part 10: Co-catalytic effect of different carboxylic acids in the oxidation ofcyclohexane, cyclohexanol, and acetone. Tetrahedron 64, 2143–2152 (2008).

24. Das, S., Incarvito, C. D., Crabtree, R. H. & Brudvig, G. W. Molecular recognitionin the selective oxygenation of saturated C–H bonds by a dimanganese catalyst.Science 312, 1941–1943 (2006).

25. Mas-Balleste, R. & Que, L. Jr Targeting specific C–H bonds for oxidation. Science312, 1885–1886 (2006).

26. Bhyrappa, P., Young, J. K., Moore, J. S. & Suslick, K. S. Dendrimer-metalloporphyrins: synthesis and catalysis. J. Am. Chem. Soc. 118,5708–5711 (1996).

27. Pope, M. T. Heteropoly and Isopoly Oxometalates (Springer, 1983).28. Okuhara, T., Mizuno, N. & Misono, M. Catalytic chemistry of heteropoly

compounds. Adv. Catal. 41, 113–252 (1996).29. Hill, C. L. Special thematic issue on polyoxometalates. Chem. Rev. 98,

1–390 (1998).30. Neumann, R. Polyoxometalate complexes in organic oxidation chemistry. Prog.

Inorg. Chem. 47, 317–370 (1998).

31. Kozhevnikov, I. V. Catalysts for Fine Chemical Synthesis, Volume 2, Catalysis byPolyoxometalates (Wiley, 2002).

32. Kamata, K. et al. Efficient epoxidation of olefins with ≥99% selectivity and use ofhydrogen peroxide. Science, 300, 964–966 (2003).

33. Nakagawa, Y., Kamata, K., Kotani, M., Yamaguchi, K. & Mizuno, N.Polyoxovanadometalate-catalyzed selective epoxidation of alkenes withhydrogen peroxide. Angew. Chem. Int. Ed. 44, 5136–5140 (2005).

34. Nakagawa, Y. & Mizuno, N. Mechanism of [g-H2SiV2W10O40]42-catalyzedepoxidation of alkenes with hydrogen peroxide. Inorg. Chem. 46,1727–1736 (2007).

35. Kamata, K., Hirano, T., Kuzuya, S. & Mizuno, N. Hydrogen-bond-assistedepoxidation of homoallylic and allylic alcohols with hydrogen peroxide catalyzedby selenium-containing dinuclear peroxotungstate. J. Am. Chem. Soc. 131,6997–7004 (2009).

36. Costas, M., Chen, K. & Que, L. Jr Biomimetic nonheme iron catalysts for alkanehydroxylation. Coord. Chem. Rev. 200–202, 517–544 (2000).

37. Schneider, H.-J. & Muller, W. Mechanistic and preparative studies on the regio-and stereoselective paraffin hydroxylation with peracids. J. Org. Chem. 50,4609–4615 (1985).

38. Mello, R., Fiorentino, M., Fusco, C. & Curci, R. Oxidations bymethyl(trifluoromethyl)dioxirane. 2. Oxyfunctionalization of saturatedhydrocarbons. J. Am. Chem. Soc. 111, 6749–6757 (1989).

39. DesMarteau, D. D., Donadelli, A., Montanari, V., Petrov, V. A. & Resnati, G.Mild and selective oxyfunctionalization of hydrocarbons byperfluorodialkyloxaziridines. J. Am. Chem. Soc. 115, 4897–4898 (1993).

40. Bianchini, G. et al. Efficient and selective oxidation of methyl substitutedcycloalkanes by heterogeneous methyltrioxorhenium–hydrogen peroxidesystems. Tetarahedron 62, 12326–12333 (2006).

41. Smith, J. R. L. & Shul’pin, G. B. Efficient stereoselective oxygenation of alkanesby peroxyacetic acid or hydrogen peroxide and acetic acid catalysed by amanganese(IV) 1,4,7-trimethyl-1,4,7-triazacyclononane complex. TetrahedronLett. 39, 4909–4912 (1998).

42. Nehru, K. et al. A highly efficient non-heme manganese complex in oxygenationreactions. Chem. Commun. 4623–4625 (2007).

43. Nam, W., Kim, I., Kim, Y. & Kim, C. Biomimetic alkane hydroxylation bycobalt(III) porphyrin complex and m-chloroperbenzoic acid. Chem. Commun.1262–1263 (2001).

44. Costas, M. & Que, L. Jr Ligand topology tuning of iron-catalyzed hydrocarbonoxidations. Angew. Chem. Int. Ed. 41, 2179–2181 (2002).

45. In, J.-H., Park, S.-E., Song, R. & Nam, W. Iodobenzene diacetate as an efficientterminal oxidant in iron(III) porphyrin complex-catalyzed oxygenationreactions. Inorg. Chim. Acta 343, 373–376 (2003).

46. Lee, S. & Fuchs, P. L. Chemospecific chromium[VI] catalyzed oxidation of C–Hbonds at –40 8C. J. Am. Chem. Soc. 124, 13978–13979 (2002).

47. Murray, R. W., Jeyaraman, R. & Mohan, L. Chemistry of dioxiranes. 4. Oxygenatom insertion into carbon–hydrogen bonds by dimethyldioxirane. J. Am.Chem. Soc. 108, 2470–2472 (1986).

48. Shilov, A. E., & Shul’pin, G. B. Activation of C–H bonds by metal complexes.Chem. Rev. 97, 2879–2932 (1997).

49. Nardello, V., Marko, J., Vermeersch, G. & Aubry, J. M. 90Mo NMR and kineticstudies of peroxomolybdic intermediates involved in the catalyticdisproportionation of hydrogen peroxide by molybdate ions. Inorg. Chem. 34,4950–4957 (1995).

50. Domaille, P. J. & Harlow, R. L. Synthesis and structural characterizationof the first phosphorus-centered Baker–Figgis g-dodecametalate:g-Cs5[PV2W10O40].xH2O. J. Am. Chem. Soc. 108, 2108–2109 (1986).

AcknowledgementsWe are grateful to K. Yamaguchi and K. Yonehara for discussions. This work was supportedby the Core Research for Revolutional Science and Technology program of the JapanScience and Technology Agency, the Global COE Program Chemistry Innovation throughCooperation of Science and Engineering, the Development in a New Interdisciplinary FieldBased on Nanotechnology and Materials Science Programs and a Grant-in-Aid forScientific Research from the Ministry of Education, Culture, Science, Sports andTechnology of Japan.

Author contributionsK.K. and N.M. conceived and designed the experiments. K.K., K.Y. and Y.N. carried out theexperiments. K.U. analysed the crystallographic data. K.K. and N.M. co-wrote the paper.

Additional informationThe authors declare no competing financial interests. Supplementary informationand chemical compound information accompany this paper at www.nature.com/naturechemistry. Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/. Correspondence and requests for materials should beaddressed to N.M.













Using first principles to predict bimetallic catalystsfor the ammonia decomposition reactionDanielle A. Hansgen, Dionisios G. Vlachos* and Jingguang G. Chen*

The facile decomposition of ammonia to produce hydrogen is critical to its use as a hydrogen storage medium in a hydrogeneconomy, and although ruthenium shows good activity for catalysing this process, its expense and scarcity are prohibitive tolarge-scale commercialization. The need to develop alternative catalysts has been addressed here, using microkineticmodelling combined with density functional studies to identify suitable monolayer bimetallic (surface or subsurface)catalysts based on nitrogen binding energies. The Ni–Pt–Pt(111) surface, with one monolayer of Ni atoms residing on aPt(111) substrate, was predicted to be a catalytically active surface. This was verified using temperature-programmeddesorption and high-resolution electron energy loss spectroscopy experiments. The results reported here provide aframework for complex catalyst discovery. They also demonstrate the critical importance of combining theoretical andexperimental approaches for identifying desirable monolayer bimetallic systems when the surface properties are not a linearfunction of the parent metals.

The ammonia decomposition reaction has recently been subjectto an increasing level of attention due to the possibility ofammonia being used as a hydrogen storage medium in a poss-

ible hydrogen economy. Ammonia can be liquefied easily at a pressureof 8 atm at 293 K, leading to high energy densities. It is readilyavailable because of its use in fertilizers, and is a CO-free source ofhydrogen. Experimental studies of single-metal catalysts have shownthat Ru is the most active decomposition catalyst1–3, but it is expensiveand limited in supply; hence there is a need to develop either lessexpensive alternatives or catalysts with higher activity.

Ammonia decomposition proceeds by means of dehydrogenation,

NH3 −−−−−�−H

NH2 −−−−−�−H

NH−−−−−�−HN

followed by recombination of N and H to form N2 and H2, respect-ively. It has been shown that the heat of nitrogen chemisorption is agood descriptor for ammonia synthesis and decomposition4,5. Thebinding energy of the nitrogen atom to the surface must be strongenough for dehydrogenation of the NHx species to occur, but suffi-ciently weak that the nitrogen recombines to desorb from the surfaceto complete the catalytic cycle. This trade-off leads to a volcano-typerelationship between nitrogen binding energy and ammonia decompo-sition activity. Although Ru has the optimal heat of chemisorptionamong single metals, it is possible that bimetallic catalysts withhigher activities might exist. Encouragingly for the ammonia synthesisreaction, a bimetallic catalyst (CoMo) has been found that is moreactive than Ru. However, because the activity for the CoMo catalystdecreases significantly in the presence of ammonia, with concen-trations as low as 5%, its use as a decomposition catalyst is restricted4,5.

The CoMo bimetallic catalyst for the synthesis reaction was pre-dicted through the concept of Periodic Table interpolation, in whichthe binding energy of a mixed metal (alloyed) surface is taken as alinear combination of the binding energies of the parent metals. Ametal with a high nitrogen binding energy (Mo) and one with alow binding energy (Co) were chosen to give a surface with an inter-mediate binding energy. Although this specific alloy catalyst may bestable4, for many bimetallic alloys, one metal often segregates to the

surface either to minimize the surface energy of the alloyed systemor due to adsorbate-induced reconfiguration6–11. This results insurface properties that are vastly different from the alloyed surface,and makes predictions from the periodic table interpolationmethod invalid.

Monolayer bimetallic catalysts consist of a monolayer of anadmetal in the top layers of a host metal12–14. These surfaces canbe used to represent the segregated surface of an alloy or can beused to model core–shell bimetallic nanoparticles. The admetalcan be on the surface of the host metal, giving rise to the surfaceconfiguration, or below the surface layer, forming the subsurfaceconfiguration. These two configurations have been shown, bothexperimentally and through density functional theory (DFT) calcu-lations, to have properties that differ drastically from one anotherand from the parent metals; these bimetallic surfaces are also verydifferent from the corresponding alloyed system, where thetopmost surface layer is intermixed with the two parent metals.Typically, the monolayer bimetallic surfaces have binding energiesthat are greater or less than both of the parent metals, dependingon the configuration, and are not a function of the parent metalbinding energies12,13. Owing to this nonlinear behaviour, there iscurrently no method to rationally design these novel catalysts.

In this study, we present full microkinetic models for NH3decomposition on various single-metal catalysts, including Co, Pt,Pd, Ni, Ru, Rh, Ir, Re and Mo. These models include adsorbate–adsorbate interactions, which have been shown to have a significanteffect on calculated surface coverages, the rate-determining step andcatalytic activity15. The results, combined with DFT data, are used topredict suitable monolayer bimetallic (surface or subsurface)systems, based on nitrogen binding energies, for the ammoniadecomposition reaction. These bimetallic surfaces are then testedexperimentally for their activity towards ammonia decompositionusing temperature-programmed desorption (TPD) and high-resolution electron energy loss spectroscopy (HREELS) to validatethe model predictions. To the best of our knowledge, our approachrepresents the first time that full microkinetic models and DFT pre-dictions, together with experimental verification, have been com-bined to identify novel catalyst formulations and surface structures.

Center for Catalytic Science and Technology, Department of Chemical Engineering, University of Delaware, Newark, Delaware 19716, USA.

*e-mail: [email protected]; [email protected]








ResultsComputational results. DFT calculations were performed to obtainthe interaction energies for nitrogen and hydrogen adsorbates onthe surface. The binding energies were determined at coverages of1/9 to 1 monolayer (ML), which were then plotted as a functionof coverage. Although not completely linear, approximating thedata with a linear function is adequate for trends and screeningstudies as performed here. The binding energies extrapolated tothe zero coverage limit (QA(0)) and the interaction parameters(IP), the slope of the line, for each metal studied are listed inTable 1. Plots of the binding energies and the linear fits can befound in the Supplementary Information. The zero coveragebinding energy and the interaction parameter were used toapproximate the binding energy at any surface coverage withinthe microkinetic models through the following equation:

QA(uA) = QA(0) + IP uA (1)

where uA is the coverage of adsorbate A.The interaction parameters for hydrogen are low, at approxi-

mately 21 kcal mol21 (ML hydrogen)21. This may be a result ofthe small size of the hydrogen atom, the low binding energy or acombination of both. Because the hydrogen interaction parametersare small, they have very little effect on the reaction mechanism andpredicted ammonia conversions. The nitrogen interaction par-ameters, on the other hand, are much greater, ranging from227.0 to 242.6 kcal mol21 (ML nitrogen)21 (depending on themetal), and have a significant effect on conversion.

For each metal, the microkinetic model was used to calculate theconversion at the reactor exit. The overall ammonia decompositionreaction was modelled with 12 elementary reaction steps, with noassumption of a rate-determining step. The elementary reactionsteps are as follows:

NH3 + ∗ � NH∗3 (R1)

NH∗3 � NH3 + ∗ (R2)

NH∗3 + ∗ � NH∗

2 + H∗ (R3)

NH∗2 + H∗ � NH∗

3 + ∗ (R4)

NH∗2 + ∗ � NH∗ + H∗ (R5)

NH∗ + H∗ � NH∗2 + ∗ (R6)

NH∗ + ∗ � N∗ + H∗ (R7)

N∗ + H∗ � NH∗ + ∗ (R8)

N∗ + N∗ � N2 + 2∗ (R9)

N2 + 2∗ � N∗ + N∗ (R10)

H∗ + H∗ � H2 + 2∗ (R11)

H2 + 2∗ � H∗ + H∗ (R12)

where ∗ represents an adsorbed surface species. The coverage-dependent atomic binding energies were used to calculate the mol-ecular heats of chemisorption (QNHx) and the activation barriers ofthe elementary reactions using the bond-order conservationmethod16, resulting in activation barriers that were coverage-dependent. Pre-exponentials for the elementary reaction stepswere taken from a previous literature study in which the pre-exponentials were fit to Ru experimental data with constraints onthe overall entropic consistency15.

Figure 1 shows the predicted conversion of ammonia versus thenitrogen heat of chemisorption (QN(0)) at a reactor temperature of850 K. Among the single-metal catalysts studied, Ru was found tohave the highest activity, consistent with experimental data1–3. Themodel results (circles) for the full microkinetic library are in goodagreement with an extensive experimental study of 13 metals from

Ganley and colleagues1 (triangles). Figure 1 reveals a volcanorelationship, and shows that the heat of nitrogen chemisorption isa good descriptor to identify surfaces with desirable catalytic activity(other possible descriptors are listed in Supplementary Section 4).This conclusion is consistent with previous studies on ammoniasynthesis and decomposition reactions4,5, although in the currentstudy repulsive adsorbate–adsorbate interactions were accountedfor and no assumptions of the rate-determining step or surface cov-erages were made, as is frequently done in previous literaturestudies. Through the models, a maximum activity (peak of thevolcano curve) is predicted to be at a nitrogen heat of chemisorptionof �134 kcal mol21.

To determine the kinetically significant reaction steps, a sensi-tivity analysis was performed. The pre-exponentials of each

Table 1 | Monometallic binding energies and interactionenergies.

Metal Zerocoverage Nbindingenergy

Calculated Ninteractionparameter

Zerocoverage Hbindingenergy

Calculated Hinteractionparameter

Pd 110.1 242.0 67.0 21.6Pt 107.4 231.7 63.3 21.4Ir 117.3 232.8 62.1 22.7Ni 122.3 242.6 64.4 0.1Rh 125.4 239.5 64.8 20.5Co 125.4 234.0 64.8 21.6Ru 141.6 236.9 66.6 21.4Re 152.0 228.6 71.6Mo 154.3 227.0 69.8

Binding energies extrapolated to the zero coverage limit and the calculated interaction parameterscan be used to estimate the atomic binding energies at any surface coverage through equation(1). The negative sign indicates that the interactions are repulsive. All energies are in kcal mol21.

0.01

0.1

1

10

100

100 110 120 130 140 150 160

Con

vers

ion

(%)

(kcal mol–1)

Pt IrPd CoRh

RuNi Re Mo

TO

F (s

–1)

Figure 1 | Ammonia decomposition volcano curve. Ammonia conversion

calculated from microkinetic modelling (circles, left axis) at 850 K for

various transition-metal catalysts and experimental supported catalyst

turnover frequencies (TOF) (triangles, right axis) from Ganley and

colleagues1 at 850 K plotted against the nitrogen binding energies (QN(0)).

The dotted line is calculated by assuming average interaction energies (see

Table 1) and a correlation between nitrogen and hydrogen binding energies

to aid in determining the volcano maximum. The nitrogen binding energy is

found to be a good descriptor for this reaction. The peak of the volcano

curve is at a nitrogen binding energy of �134 kcal mol21, and this value is

used to identify bimetallic surfaces with desirable catalytic activity for the

ammonia decomposition reaction.






forward and reverse elementary reaction step pair (Ai) (that is, R1and R2, R3 and R4, and so on) were perturbed simultaneously,thus ensuring thermodynamic consistency. The normalized sensi-tivity coefficient for each reaction pair (NSCi) was calculated using

NSCi =D(ln X)D(ln Ai)

(2)

where X is the conversion at the end of the reactor. The largest (inabsolute value) model responses among the reaction pairs indicatekinetically significant reaction steps17,18 (see SupplementaryInformation). A sensitivity analysis performed on each of the sur-faces shows the rate-determining step to be the removal of thesecond hydrogen (from the NH2; reaction R5) for surfaces with anitrogen binding energy less than 125 kcal mol21. For surfaceswith higher nitrogen binding energies, the removal of the first andsecond hydrogens (R3 and R5) and nitrogen desorption (R9) arekinetically significant, as shown in Fig. 2. At the peak of thevolcano curve (�134 kcal mol21), the removal of the second hydro-gen is the most significant elementary reaction step, although theremoval of the first hydrogen and nitrogen desorption are bothkinetically significant. Interestingly, the dominant surface coveragechanges from the left to the right leg of the volcano curve(Supplementary Fig. S4). This analysis underscores the fact thatthe kinetically significant step and dominant coverage may be chan-ging along a volcano curve, the maximum in the volcano curve maybe a result of multiple physical mechanisms, and the importance ofperforming a full microkinetic analysis rather than assuminga priori a rate-determining step and a dominant surface species.

Because the nitrogen binding energy is a good activity descriptor,a DFT search was performed to identify catalyst surfaces that havea similar binding energy to the optimal value of �134 kcal mol21.Pt-based monolayer bimetallic surfaces were the focus of thisstudy. These surfaces have been shown to form the surface(M–Pt–Pt) and subsurface (Pt–M–Pt) configurations, both

experimentally and through DFT calculations11,12,19,20. Binding ener-gies were calculated at the 1/9 ML coverage, which is a good approxi-mation to the binding energies extrapolated to zero coverage. Table 2shows the calculated nitrogen binding energies for several surface andsubsurface configurations. Also included are the metal–nitrogen bondlengths on each surface. The binding energies for the subsurface con-figurations are lower than both the parent metals due to a broadeningof the d band, whereas the binding energies for the surface configur-ations are stronger than both the parent metals, due to a contractionof the d band21. For the subsurface configurations, the Pt–N bondlengths were similar to Pt(111), with only a slight lengthening ofthe bond. Pt–Ni–Pt was the only subsurface configuration in whichthere was a shortening of the bond compared to the Pt(111)surface. For the surface configurations, the nitrogen surface bondsshow much more variation, which is probably a result of the differ-ences in the metals to which the nitrogen is bound.

The nitrogen binding energies vary from 71 to 207 kcal mol21 byadding a second metal to the surface or subsurface layer of the Pthost. Based on the theoretical predictions in Fig. 1, the activity ofthese bimetallic surfaces should also follow a volcano relationship.The Ni–Pt–Pt(111) bimetallic surface has a nitrogen bindingenergy of 130.7 kcal mol21, slightly lower than that of Ru, and isa potentially active catalyst (see the maximum in Fig. 1). The sub-surface configuration, Pt–Ni–Pt(111), and the parent metals,Pt(111) and Ni(111), are expected to have lower activities becauseof the weaker nitrogen binding energies (Table 2). The Ni/Pt bime-tallic system was chosen as a test system to be studied experimentallyin the current paper to assess the model predictions. Table 2 alsoidentifies the Co–Pt–Pt(111) and Fe–Pt–Pt(111) surfaces asadditional promising systems. The activity of these additionalsurfaces will be evaluated in future studies.

TPD studies of ammonia decomposition. We tested the DFT/microkinetic model predictions using the Ni–Pt bimetallicsystems. The Ni–Pt–Pt surface, together with Pt–Ni–Pt, Pt(111)and a Ni(111) film, were tested for their activity towardsammonia decomposition through TPD experiments. DepositingNi at room temperature leads to the Ni–Pt–Pt surface, whereasdepositing at 600 K leads to the Pt–Ni–Pt configuration22.

0

0.1

0.2

0.3

0.4

0.5

0.6

100 110 120 130 140 150 160

Nor

mal

ize

d se

nsiti

vity

coe

ffic

ient

(kcal mol–1)

Pt IrPd CoRh

RuNi Re Mo

NH3* + *NH2* + *

NH2* + H*

NH* + H*

2N* N2 + 2 *

Figure 2 | Normalized sensitivity coefficients of the kinetically significant

elementary reaction steps for each monometallic metal. Pre-exponentials of

the forward and reverse reaction pairs were perturbed simultaneously at a

temperature of 850 K and the response of conversion was monitored. The

reaction pairs that have the highest normalized sensitivity coefficient are

kinetically significant reaction steps. The sensitivity analysis shows that there

are multiple kinetically significant reaction steps and their sensitivity changes

across the volcano curve. Only reaction pairs that have a normalized

sensitivity coefficient above 0.01 for any metal are shown.

Table 2 | Library of DFT binding energies and bond lengthsof nitrogen atoms at a 1/9 ML coverage on variousmonolayer bimetallic surfaces.

Configuration Metal (111)surface

Nitrogen bindingenergy (kcal mol21)

Bond lengthdM–N (Å)

Subsurface Pt–Ti–Pt 70.7 1.975Pt–V–Pt 81.0 1.975Pt–Cr–Pt 76.3 1.965Pt–Mn–Pt 77.6 1.968Pt–Fe–Pt 78.4 1.969

Single metal Pt–Co–Pt 83.4 1.964Pt–Ni–Pt 87.5 1.941Pt 102.1 1.954Ni 113.8 1.770Ni–Pt–Pt 130.7 1.761

Surface Co–Pt–Pt 126.5 1.780Fe–Pt–Pt 134.1 1.864Mn–Pt–Pt 207.2 1.854Cr–Pt–Pt 188.3 1.894V–Pt–Pt 188.1 1.876Ti–Pt–Pt 176.1 1.918

By adding a metal to the Pt(111) surface, in either the surface or subsurface configuration, thenitrogen binding energy is modified, creating a large range of binding energies on the bimetallicsurfaces. The binding energies and bond lengths on the Pt(111) and Ni(111) surfaces are alsoincluded for comparison. The different colours in the structures indicate the different metals.






Depositing at least 5 ML of Ni achieves a surface with chemicalproperties similar to a Ni(111) surface23.

Ammonia (3 L, equal to 1 × 1028 torr for 300 s, where L indi-cates Langmuir) was dosed onto each of the four surfaces at350 K. Figure 3 shows the desorption spectra of the decompositionproduct, nitrogen, for each of the surfaces. The m/z¼ 14 AMU wasused to monitor N2 desorption (N atoms from N2 cracking) to elim-inate the overlap between N2 and any background CO adsorbed tothe surface (both seen at m/z¼ 28 AMU). Detection of the peak at626 K confirms that the Ni–Pt–Pt surface is active towardsdecomposition. The absence of any peaks on the other three sur-faces shows that these surfaces are not active towards decompositionat these dosing conditions. The results for Pt(111) are consistentwith previous results, which showed that Pt(111) does not decom-pose ammonia under 400 K (ref. 24). The inactivity of thePt(111), thick Ni(111) film and the Pt–Ni–Pt surfaces can be attrib-uted to the low binding energy of nitrogen on these surfaces, con-firming the predictions from the microkinetic models. TPDresults following the desorption of hydrogen from the four surfacesas well as ammonia dosed at low temperatures are provided in theSupplementary Information.

Because the Ni–Pt–Pt surface showed decomposition activity, thesurface was exposed to 3 L of ammonia at various temperaturesfrom 150 to 425 K to determine the onset temperature for decompo-sition (Fig. 4). Nitrogen coverages on the Ni–Pt–Pt surface werequantified by comparing the integrated nitrogen peak area to thearea produced by a saturation coverage of CO (uCO¼ 0.68) on thePt(111) surface after taking into consideration the different sensi-tivity factors for N2 and CO, as described previously for other reac-tions on Ni/Pt(111) (ref. 25).

The nitrogen coverage achieved at each dosing temperature is com-pared in the inset of Fig. 4. The coverage stays constant at 0.07 ML upto 300 K. At 325 K, the coverage begins to increase and continues toincrease up to 375 K, indicating decomposition occurring duringdosing at a phenomenally low temperature of �325 K. Above adosing temperature of 375 K, the coverage stays constant because

the surface is saturated with nitrogen at a coverage of �0.3 ML.A longer exposure of 6 L at 400 K was also performed (not shown),resulting in identical desorption spectra, confirming that saturationcoverage was achieved. The saturation value is consistent with thesaturation coverage of 0.28 ML on the Ru(0001) surface26.

Interestingly, the nitrogen desorption peak is constant at �630 Kover all nitrogen coverages. This is surprising, because nitrogen des-orption is a second-order reaction, and a peak shift towards lowertemperatures at higher coverages is expected. The absence of thisshift indicates that surface reconstruction may be occurring.Reconstruction of a Ni(111) surface to pseudo-Ni(100) has beenreported in the presence of strong binding adsorbates such as N,C and S (ref. 27). Because the surface layer in Fig. 4 is Ni, asimilar transition is likely to be occurring.

In a study of ammonia decomposition on Ru(0001), an exposureof 3,500 L of ammonia was dosed at 500 K to achieve a nitrogen sat-uration coverage26. In comparison, a saturation coverage is achievedwith 3 L at 375 K on the Ni–Pt–Pt surface. The significantly lowerdosing temperature and ammonia exposure clearly indicate thatthe overall dehydrogenation barrier, which was shown through sen-sitivity analysis to be a kinetically significant reaction step, is muchlower for the bimetallic surface.

The other kinetically significant reaction step for ammoniadecomposition was found to be the recombinative nitrogen deso-rption (Fig. 2). Nitrogen on Ru(0001) was shown to have a peak des-orption temperature ranging from 770 K at low coverages to 680 Kat a saturation coverage of 0.28 ML, using the same heating rate of3 K s21 used in this study26. The lower desorption temperature onNi–Pt–Pt is comparable to the highly stepped Ru(109) and under-coordinated Ru(1121) and Ru(1010) surfaces, which have peakdesorption temperatures at a saturation coverage of �600 K (at2 K s21) (ref. 28), �610 K (ref. 29) and �620 K (ref. 30), respect-ively. Furthermore, the desorption peaks from Ni–Pt–Pt are muchnarrower, with a temperature span of 560–725 K, compared to550–900 K from the Ru(0001) surface. Our results indicate thatnot only is ammonia dehydrogenation on Ni–Pt–Pt facile, butalso nitrogen desorption is relatively fast.

To support the DFT predictions and TPD experiments, HREELSwas used to investigate the decomposition of ammonia on the

Inte

nsity

(a.

u.)

800700600500400

Temperature (K)

400 K

375 K

350 K

340 K

325 K

300 K

250 K

N2 (14 AMU)

N = 0.27

N = 0.29

N = 0.18

N = 0.14

N = 0.11

N = 0.07

N = 0.07

630 K0.4

0.3

0.2

0.1

0.0

400300200

Figure 4 | TPD spectra of nitrogen desorption from a Ni–Pt–Pt surface

after dosing 3 L ammonia at the specified temperature. The nitrogen

coverages (uN) resulting from ammonia decomposition are indicated. The

inset shows the nitrogen coverage as a function of the dosing temperature.

The saturation coverage of nitrogen on this surface is �0.3 ML, as indicated

by the coverage plateau from the inset. The onset temperature for the sharp

increase in the N2 peak area demonstrates that the Ni–Pt–Pt surface is active

towards the decomposition of ammonia at temperatures as low as �325 K.

Inte

nsity

(a.

u.)

800700600500400

Temperature (K)

Ni(111) film

Ni-Pt-Pt

Pt-Ni-Pt

Pt(111)

N2 (14 AMU)626 K

Figure 3 | Ammonia decomposition on different Ni–Pt surfaces. TPD results

of nitrogen desorption from the decomposition of ammonia on Pt(111),

Pt–Ni–Pt, Ni–Pt–Pt and a Ni(111) film. 3 L of ammonia was dosed at 350 K, then

the temperature was ramped at a heating rate of 3 K s21 and the desorption

of nitrogen was monitored using a mass spectrometer. The Ni–Pt–Pt surface

is the only one that shows activity towards ammonia decomposition under

these conditions, as indicated by the peak at 626 K. This confirms model

predictions of the Ni–Pt–Pt surface being active towards ammonia

decomposition based on an optimal nitrogen binding energy and the other

three surfaces being inactive due to a nitrogen binding energy that is too low.






Ni–Pt–Pt surface. The HREELS results and the vibrational assign-ment of possible reaction intermediates are provided in theSupplementary Information. The most important observationfrom the HREELS results is that, furthermore to adsorbed NH3,partially decomposed intermediates including NH2 and atomicnitrogen are produced on Ni–Pt–Pt at 350 K, supporting the TPDdetection of the N2 gas-phase product from the decomposition ofammonia on the Ni–Pt–Pt surface.

DiscussionOur experimental findings support the proposed computationalframework in which potential (complex) bimetallic catalysts can effec-tively be screened using a library of binding energies of bimetalliccatalysts from DFT calculations guided from full, first-principles-based microkinetic models. The Ni–Pt–Pt bimetallic surface waspredicted to be very active for the ammonia decomposition reaction;this was confirmed by experimental measurements using TPD andHREELS. The TPD experiments indicate that a Ni–Pt–Pt catalystmay be more active than Ru based on a lower nitrogen desorptiontemperature and a remarkably low dehydrogenation barrier, bothof which were determined to be kinetically significant reactionsteps. For this reaction to take place under steady-state conditions,both NHx decomposition and nitrogen desorption must occur.Our results show that NHx decomposition occurs at temperaturesas low as 325 K and the onset for nitrogen desorption is at 560 K.Therefore, steady-state decomposition activity should be seen attemperatures below 600 K.

As the activity is highly dependent on the location of the Ni atoms,the stability of the surface is very important. Menning and Chen haveshown that the relative thermodynamic stability of the surface andsubsurface structures depends on the adsorbate coverage and thePauling electronegativity of the adsorbate9. For oxygen, which has asimilar Pauling electronegativity as nitrogen (3.44 for oxygen and3.04 for nitrogen), the surface configuration was shown to be morestable than the subsurface configuration through DFT and was con-firmed experimentally. Exposure to oxygen caused the subsurfaceNi to reconstruct to the surface configuration (Ni–Pt–Pt) under con-ditions of both low pressure (1× 1027 torr O2) and atmosphericpressure9,31,32. A similar stability was predicted for the Ni–Pt–Ptsurface with nitrogen adsorbates9, and this surface is expected to bestable under reaction conditions.

Previous studies within our group have shown a remarkable cor-relation between the activity on single-crystal surfaces, polycrystal-line films and supported bimetallic catalysts for reactions such ascyclohexene hydrogenation20,33,34. This, combined with strongagreement of the microkinetic models with the supported mono-metallic catalysts of Ganley and colleagues,1 gives reason to expectthat the activity of the supported bimetallic catalysts for theammonia decomposition reaction will also follow the predictedtrends of the microkinetic models on the single-crystal surfaces.

The ammonia decomposition reaction was used in the currentstudy as a test system due to the relative simplicity of the moleculeand the decomposition reaction pathways. This methodology forpredicting active monolayer bimetallic surfaces should also be appli-cable to more complex chemistries, although care must be taken tochoose bimetallic systems that are thermodynamically stable underreaction conditions.

MethodsTheoretical predictions. DFT calculations were performed using the Viennaab initio Simulation Package (VASP) version 4.6 (refs 35,36). A plane-wave basis setwas used with an energy cutoff of 396 eV, together with ultrasoft Vanderbiltpseudopotentials37 and the PW-91 functional38. A 3 × 3 × 1 Monkhorst Packk-point grid39 was used for all slab calculations. All calculations were performed onthe close-packed surfaces: fcc(111) for Pd-, Pt-, Ir-, Ni-, Rh- and Pt-based monolayerbimetallic surfaces; hcp(0001) for Co, Ru and Re; and bcc(110) for Mo. Latticeconstants that were previously optimized for the PW-91 functional were used40. Forthe Pt-based monolayer bimetallic surfaces, the Pt lattice constant of 4.01 Å was

used. A 3 × 3 supercell of four layers and a vacuum region equivalent to seven metallayers were used to decouple consecutive slabs. All calculations were performed spin-polarized, with the bottom two layers frozen to represent the bulk structure and thetop two metal layers allowed to relax. An electron smearing parameter of 0.2 eV wasapplied and the convergences were set to 1 × 1024 for the self-consistent electronicminimization loop and to 1 × 1023 for the ionic relaxation loop.

DFT was used to calculate the binding energies of atomic species (N and H) onthe single-metal and bimetallic surfaces using

Q(adsorbate) =E(slab) + X ∗ E(adsorbate) − E(slab+X adsorbates)

X(3)

where E(adsorbate) is the energy of the atom in vacuum, E(slab) is the energy of the bareslab without adsorbates, E(slabþX adsorbate) is the energy of the slab with the adsorbatesbound to the surface, and X is the number of adsorbates within the supercell. Atomicnitrogen and hydrogen binding energies were calculated at varying coverages,specifically at 1/9, 2/9, 1/3, 2/3 and 1 ML. The adsorbate configurations used areshown in the Supplementary Information.

For the microkinetic models, a fixed-bed reactor with a length of 0.5 cm and adiameter of 0.32 cm was modelled. A flow rate of 70 s.c.c.m. and a catalytic surfacearea of 12,000 cm2 cm23 were used.

Experimental methods. A two-level stainless steel ultrahigh vacuum (UHV)chamber, with a typical base pressure of �4 × 10210 torr, was used for the TPDexperiments25. A heating rate of 3 K s21 was used for preparation methods and allTPD experiments. Anhydrous ammonia, of 99.99% purity, was used and the puritywas checked in situ with the mass spectrometer before experiments each day.Ammonia was dosed in the background for all experiments. After dosing, thetemperature was held constant for 10 min to allow the residual ammonia in theUHV background to be pumped away before ramping the temperature.

The Pt(111) surface was cleaned by cycles of Neþ bombardment at 600 K. Thecrystal was then heated to 890 K in 3 Langmuir (1 L¼ 1 × 1026 torr s) oxygen, of99.998% purity, to burn off any remaining surface carbon. The crystal was thenheated to 1,100 K and held at this temperature for 5 min to anneal the crystal surface.This procedure was repeated until surface cleanliness was verified by Augerelectron spectroscopy.

The Ni/Pt(111) bimetallic surfaces were prepared for this study by depositing Nithrough physical vapour deposition on the clean, freshly annealed Pt(111) surface.Three Ni-containing surfaces were created: the surface monolayer (Ni–Pt–Pt), thesubsurface monolayer (Pt–Ni–Pt) and a thick Ni(111) film that mimics a Ni(111)surface when over 5 ML are deposited on the Pt(111) surface23. Using preparationprocedures described previously22,25, the Pt–Ni–Pt surface was created by depositingNi at 600 K until a Ni(849 eV)/Pt(241 eV) Auger ratio of 1.0 was achieved. For theNi–Pt–Pt surface, Ni was deposited for the same amount of time at 300 K, resultingin an Auger ratio of approximately 1.5. The thick Ni(111) film was created bydepositing Ni until an Auger ratio above 3.0 was achieved.

Received 31 December 2009; accepted 12 March 2010;published online 25 April 2010

References1. Ganley, J. C., Thomas, F. S., Seebauer, E. G. & Masel, R. I. A priori catalytic

activity correlations: the difficult case of hydrogen production from ammonia.Catal. Lett. 96, 117–122 (2004).

2. Choudhary, T. V., Sivadinarayana, C. & Goodman, D. W. Catalytic ammoniadecomposition: COx-free hydrogen production for fuel cell applications. Catal.Lett. 72, 197–201 (2001).

3. Yin, S. F. et al. Investigation on the catalysis of COx-free hydrogen generationfrom ammonia. J. Catal. 224, 384–396 (2004).

4. Jacobsen, C. J. H. et al. Catalyst design by interpolation in the periodic table:bimetallic ammonia synthesis catalysts. J. Am. Chem. Soc. 123,8404–8405 (2001).

5. Boisen, A., Dahl, S., Norskov, J. K. & Christensen, C. H. Why the optimalammonia synthesis catalyst is not the optimal ammonia decomposition catalyst.J. Catal. 230, 309–312 (2005).

6. Campbell, C. T. Bimetallic surface-chemistry. Annu. Rev. Phys. Chem. 41,775–837 (1990).

7. Tao, F. et al. Reaction-driven restructuring of Rh–Pd and Pt–Pd core–shellnanoparticles. Science 322, 932–934 (2008).

8. Ruban, A. V., Skriver, H. L. & Norskov, J. K. Surface segregation energies intransition-metal alloys. Phys. Rev. B 59, 15990–16000 (1999).

9. Menning, C. A. & Chen, J. G. General trend for adsorbate-inducedsegregation of subsurface metal atoms in bimetallic surfaces. J. Chem. Phys. 130,174709 (2009).

10. Kitchin, J. R., Reuter, K. & Scheffler, M. Alloy surface segregation in reactiveenvironments: first-principles atomistic thermodynamics study of Ag3Pd(111)in oxygen atmospheres. Phys. Rev. B 77, 075437 (2008).

11. Ma, Y. G. & Balbuena, P. B. Pt surface segregation in bimetallic Pt3M alloys: adensity functional theory study. Surf. Sci. 602, 107–113 (2008).






12. Chen, J. G., Menning, C. A. & Zellner, M. B. Monolayer bimetallic surfaces:experimental and theoretical studies of trends in electronic and chemicalproperties. Surf. Sci. Rep. 63, 201–254 (2008).

13. Pallassana, V., Neurock, M., Hansen, L. B., Hammer, B. & Norskov, J. K.Theoretical analysis of hydrogen chemisorption on Pd(111), Re(0001) and Pd-ML/Re(0001), Re-ML/Pd(111) pseudomorphic overlayers. Phys. Rev. B 60,6146–6154 (1999).

14. Rodriguez, J. A. & Goodman, D. W. The nature of the metal metal bond inbimetallic surfaces. Science 257, 897–903 (1992).

15. Mhadeshwar, A. B., Kitchin, J. R., Barteau, M. A. & Vlachos, D. G. The role ofadsorbate2adsorbate interactions in the rate controlling step and the mostabundant reaction intermediate of NH3 decomposition on Ru. Catal. Lett. 96,13–22 (2004).

16. Shustorovich, E. & Sellers, H. The UBI-QEP method: a practical theoreticalapproach to understanding chemistry on transition metal surfaces. Surf. Sci. Rep.31, 5–119 (1998).

17. Campbell, C. T. Finding the rate-determining step in a mechanism—comparingDeDonder relations with the ‘degree of rate control’. J. Catal. 204,520–524 (2001).

18. Campbell, C. T. Micro- and macro-kinetics: their relationship in heterogeneouscatalysis. Top. Catal. 1, 353–366 (1994).

19. Menning, C. A., Hwu, H. H. & Chen, J. G. Experimental and theoreticalinvestigation of the stability of Pt–3d–Pt(111) bimetallic surfaces under oxygenenvironment. J. Phys. Chem. B 110, 15471–15477 (2006).

20. Humbert, M. P. & Chen, J. G. Correlating hydrogenation activity with bindingenergies of hydrogen and cyclohexene on M/Pt(111) (M¼Fe, Co, Ni, Cu)bimetallic surfaces. J. Catal. 257, 297–306 (2008).

21. Kitchin, J. R., Norskov, J. K., Barteau, M. A. & Chen, J. G. Role of strain andligand effects in the modification of the electronic and chemical properties ofbimetallic surfaces. Phys. Rev. Lett. 93, 156801 (2004).

22. Kitchin, J. R. et al. Elucidation of the active surface and origin of the weak metal–hydrogen bond on Ni/Pt(111) bimetallic surfaces: a surface science and densityfunctional theory study. Surf. Sci. 544, 295–308 (2003).

23. Khan, N. A., Hwu, H. H. & Chen, J. G. Low-temperature hydrodesulfurization ofthiophene on Ni/Pt(111) bimetallic surfaces with monolayer Ni coverage.J. Catal. 205, 259–265 (2002).

24. Sun, Y. M., Sloan, D., Ihm, H. & White, J. M. Electron-induced surfacechemistry: production and characterization of NH2 and NH species on Pt(111).J. Vac. Sci. Technol. A 14, 1516–1521 (1996).

25. Skoplyak, O., Barteau, M. A. & Chen, J. G. Reforming of oxygenates for H2production: correlating reactivity of ethylene glycol and ethanol on Pt(111) andNi/Pt(111) with surface d-band center. J. Phys. Chem. B 110, 1686–1694 (2006).

26. Dietrich, H., Jacobi, K. & Ertl, G. Coverage, lateral order and vibrations of atomicnitrogen on Ru(0001). J. Chem. Phys. 105, 8944–8950 (1996).

27. Gardin, D. E., Batteas, J. D., Vanhove, M. A. & Somorjai, G. A. Carbon, nitrogen,and sulfur on Ni(111)—formation of complex structures and consequences formolecular decomposition. Surf. Sci. 296, 25–35 (1993).

28. Kim, Y. K., Morgan, G. A. & Yates, J. T. Site-specific dissociation of N2 on thestepped Ru(109) surface. Surf. Sci. 598, 14–21 (2005).

29. Dietrich, H., Jacobi, K. & Ertl, G. Decomposition of NH3 on Ru(11(2)–1). Surf.Sci. 352, 138–141 (1996).

30. Dietrich, H., Jacobi, K. & Ertl, G. Vibrations, coverage, and lateral order ofatomic nitrogen and formation of NH3 on Ru(10(1)–0). J. Chem. Phys. 106,9313–9319 (1997).

31. Menning, C. A. & Chen, J. G. Thermodynamics and kinetics of oxygen-inducedsegregation of 3d metals in Pt–3d–Pt(111) and Pt–3d–Pt(100) bimetallicstructures. J. Chem. Phys. 128, 174709 (2008).

32. Menning, C. A. & Chen, J. G. Regenerating Pt–3d–Pt model electrocatalyststhrough oxidation–reduction cycles monitored at atmospheric pressure. J. PowerSources 195, 3140–3144 (2010).

33. Lu, S. L. et al. Low temperature hydrogenation of benzene and cyclohexene: acomparative study between gamma-Al2O3 supported PtCo and PtNi bimetalliccatalysts. J. Catal. 259, 260–268 (2008).

34. Humbert, M. P., Murillo, L. E. & Chen, J. G. Rational design of platinum-basedbimetallic catalysts with enhanced hydrogenation activity. ChemPhysChem 9,1262–1264 (2008).

35. Kresse, G. & Furthmuller, J. Efficient iterative schemes for ab initio total-energycalculations using a plane-wave basis set. Phys. Rev. B 54, 11169–11186 (1996).

36. Kresse, G. & Furthmuller, J. Efficiency of ab initio total energy calculations formetals and semiconductors using a plane-wave basis set. Comput. Mater. Sci. 6,15–50 (1996).

37. Vanderbilt, D. Soft self-consistent pseudopotentials in a generalized Eigenvalueformalism. Phys. Rev. B 41, 7892–7895 (1990).

38. Perdew, J. P. et al. Atoms, molecules, solids and surfaces—applications of thegeneralized gradient approximation for exchange and correlation. Phys. Rev. B46, 6671–6687 (1992).

39. Monkhorst, H. J. & Pack, J. D. Special points for Brillouin-zone integrations.Phys. Rev. B 13, 5188–5192 (1976).

40. Pseudopotential Library. Center for Atomic-Scale Materials Design. https://wiki.fysik.dtu.dk/dacapo/Pseudopotential_Library.

41. Catlett, C. et al. TeraGrid: analysis of organization, system architecture,and middleware enabling new types of applications, in HPC and Gridsin Action (Grandinetti, L. ed.) ‘Advances in Parallel Computing’ series(IOS Press, 2007).

AcknowledgementsThis research was supported by the Office of Basic Energy Sciences, Department of Energygrants DE-FG02-06ER15795 and DE-FG02-00ER15104. The DFT calculations wereperformed using the TeraGrid resources provided by the University of Illinois NationalCenter for Supercomputing Applications (NCSA)41.

Author contributionsD.A.H. and D.G.V. designed and developed the microkinetic models. D.A.H. and J.G.C.designed and developed the UHV experiments. D.A.H. performed and analysed allmodelling and experimental work. All authors contributed to writing the paper.

Additional informationThe authors declare no competing financial interests. Supplementary informationaccompanies this paper at www.nature.com/naturechemistry. Reprints and permissioninformation is available online at http://npg.nature.com/reprintsandpermissions/.Correspondence and requests for materials should be addressed to D.G.V. and J.G.C.



https://wiki.fysik.dtu.dk/dacapo/Pseudopotential_Library

https://wiki.fysik.dtu.dk/dacapo/Pseudopotential_Library









Nucleophilic catalysis of acylhydrazoneequilibration for protein-directed dynamiccovalent chemistryVenugopal T. Bhat1†, Anne M. Caniard1†, Torsten Luksch2, Ruth Brenk2, Dominic J. Campopiano1* and

Michael F. Greaney1*

Dynamic covalent chemistry uses reversible chemical reactions to set up an equilibrating network of molecules atthermodynamic equilibrium, which can adjust its composition in response to any agent capable of altering the free energyof the system. When the target is a biological macromolecule, such as a protein, the process corresponds to the proteindirecting the synthesis of its own best ligand. Here, we demonstrate that reversible acylhydrazone formation is aneffective chemistry for biological dynamic combinatorial library formation. In the presence of aniline as a nucleophiliccatalyst, dynamic combinatorial libraries equilibrate rapidly at pH 6.2, are fully reversible, and may be switched on or offby means of a change in pH. We have interfaced these hydrazone dynamic combinatorial libraries with two isozymes fromthe glutathione S-transferase class of enzyme, and observed divergent amplification effects, where each protein selectsthe best-fitting hydrazone for the hydrophobic region of its active site.

Dynamic covalent chemistry (DCC) uses reversible chemicalreactions to set up equilibrating assemblies of molecules atthermodynamic equilibrium1–4. The resultant dynamic com-

binatorial library (DCL) is responsive to the addition of a template,which will selectively amplify the best binding compounds from theequilibrium distribution. The essence of the concept lies in the sub-sequent adjustment of the DCL equilibrium, which will expressmore of the best binding compounds at the expense of the poorerones. A DCL is thus adaptive and capable of evolutionary behaviour,whereby individual components are either amplified or reduced inresponse to template-directed binding events. These concepts havebeen applied to diverse problems in biological and medicinal chem-istry5–11, synthetic receptor–ligand interactions12–16, self-replica-tion17–19, complex molecule synthesis20–22 and materialsscience23,24. Taken together, they represent the best characterizedexamples to date of systems chemistry, which looks to synthesizecomplex molecular networks and study their properties and behav-iour in macrocosm, rather than as a sum of their individualcomponents25,26.

We are interested in DCC systems that use a biological molecule,such as a protein, to template assemblies of small molecules atdynamic equilibrium27. Here, the DCC experiment provides amethod for discovering, studying and ranking novel proteinligands, concepts fundamental to medicinal chemistry. In theseterms, the DCC process bridges the gap between targeted chemicalsynthesis of drug candidates and their biological binding assay,meshing the two processes into a single step in which the structureof the biological target directs the assembly of its own best inhibitorin situ.

A particular challenge for DCC in biological systems lies in theimplementation of a suitable reversible reaction that can operateeffectively under the physiological conditions required by the bio-template. Lehn has defined two limiting cases for DCL construction:

adaptive and pre-equilibrated DCC28. The adaptive DCL representsthe ideal scenario, where the DCL chemistry is fully compatible withthe biological target and the ensuing binding events control theevolution of the DCL composition. Pre-equilibrated DCL refers tothe cases where the reversible chemistry used to constitute the

NH2

R CHO

N

R

R

R

pH < 7

R CHOpH < 4

no biologicalapplications

Unstableimine DCL

NaBH3CN

NH

R

R

Stableacylhydrazone

DCL

Static amine library

a

AnilinepH = 6,

biologicalapplications

O

NHNH2R

O

NH

RN R

b

Figure 1 | Transimination reactions for DCC. a, Imine DCLs: reversible

addition of amines to aldehydes gives unstable imines that cannot be

isolated or analysed directly, necessitating an in situ reduction step. The

resultant static library of amines may or may not share the binding profile of

the imine precursors. b, Acyl hydrazone DCLs: reaction of aldehydes with

hydrazides gives acylhydrazones that have good stability and are amenable

to analysis. Equilibration requires acidic conditions that are incompatible with

biological targets—a nucleophilic catalyst such as aniline may enable DCL

formation at biocompatible pH.

1EastChem, School of Chemistry, University of Edinburgh, King’s Buildings, West Mains Road, Edinburgh EH9 3JJ, UK, 2College of Life Sciences, Universityof Dundee, James Black Centre, Dow Street, Dundee DD1 5EH, UK; †These authors contributed equally to this work. *e-mail: [email protected];[email protected]








DCL is not compatible with the biological target, meaning that theDCL and the target must be separated in some manner. This resultsin static libraries in which the molecular recognition events thatcontrol DCL composition are lost. Given the challenges associatedwith conducting fast, freely reversible chemistry under physiologicalconditions, it is not surprising that methods for true adaptive DCLgeneration are limited, with the majority of successful systems usingsulfur-based transformations such as disulfide bond formation orthiol conjugate addition29–32. The development of new methodsfor adaptive DCLs is thus central to the application of DCC to bio-logical systems, as the chemistry will define the target scope andrange of available DCL components.

The reversible formation of C¼N imine-type linkages emergedearly on as a DCL-forming reaction33. The ready availability ofdiverse carbonyl and amine building blocks, plus the extensive pre-cedent of imine formation in biochemical systems, makes it an idealcandidate reaction. However, the inherent instability of imines inaqueous solution presents serious analytical and isolation problemsin the DCC context. The solution to this in the field of biologicalDCC has been to construct pseudo-adaptive DCLs where theimine linkage is reduced in situ to an amine with an externalhydride source. The resulting library contains static amine com-ponents that can correspond to the imines in binding affinity,

although both false-positive and false-negative results are possible.In addition, the introduction of an in situ reduction step complicatesthe DCL equilibration and makes it difficult to distinguish betweengenuine thermodynamic selection of the best binders and selectionof those compounds that are kinetically favoured.

An advance on simple imine formation in DCC came from theSanders group, who introduced acylhydrazones as reversible lin-kages34. The reaction has proven to be an excellent balancebetween facile reversibility and product stability; the acylhydrazoneproducts formed are stable to analysis and isolation, and the reac-tion has very good equilibration properties, as is made evident byits application to a large number of elegant abiological DCCstudies subsequently reported by the Sanders group35–37. It hasnot, however, been generally possible to apply this reaction directlyto adaptive biological DCC systems because of the acidic pHrequired for reversibility to occur in a reasonable timeframe(pH , 4)38,39. A single elegant study from Poulsen has shown thatslow equilibration of acylhydrazones, taking one week at pH 7.2,can be accelerated in the presence of the enzyme carbonic anhy-drase, enabling in situ identification of binders using mass spec-trometry40. We were keen to apply this proven reaction to ourDCC studies of enzymes, and reasoned that it could be harnessedas a powerful tool for biological investigation if a suitable catalyst

CHO

NO2

Cl

N

O O O

S

O

S

O

O O

-BuO

NH

OMeOH2N HO

N

HN

ClNO2

Aniline pH 6.2

NHNH2 NHNH2 NHNH2 NHNH2

NHNH2NHNH2NHNH2

O

NHNH2

1 3

2a 2b 2c 2d 2e

2f 2g 2h 2i 2j

O

NHNH2O

O

NHNH2S

R

OR

O

NHNH2

2

a

b

c d

3a

3i 3b 3g 3e

3f

3h 3c3j

Abs

orba

nce

(mA

U)

Abs

orba

nce

(mA

U)

30

50

2010

018 20 22 24 26 28 30 32

= 1 h

= 48 h

= 5 days

= 0 h

= 2 h

= 6 h

01020304050

01020304050

Time (min)

18 20 22 24 26 28 30 32Time (min)

18 20 22 24 26 28 30 32

18 20 22 24 26 28 30 32

18 20 22 24 26 28 30 32

18 20 22 24 26 28 30 32

3d

401

50

10152025

51015202530

0

0

51015202530

3a3h

3i 3f

3b 3g3e

3j

3c

3d

Figure 2 | Aniline-catalysed acylhydrazone formation. a, Aldehyde equilibration with hydrazide to form an acylhydrazone. b, Hydrazide components of the

ten-membered DCL. c, DCL established in the absence of aniline. Conditions: aldehyde (5mM), hydrazides (20 mM each) in NH4OAc buffer (50 mM, pH¼

6.2) containing 15% DMSO. d, DCL established in the presence of aniline (10 mM).






could be found to accelerate the equilibration. Nucleophilic catalysisof semicarbazone formation using aniline derivatives was estab-lished in classic work from Jencks in the 1960s, and recentlyapplied to hydrazone and oxime formation in peptide ligationsystems by Dawson41–44. We reasoned that an additive such asaniline could promote the equilibration of acylhydrazides and alde-hydes at pH values closer to the physiological window required bybiological targets in DCC (Fig. 1).

We began by reacting aldehyde 1 (Fig. 2), related to the knownglutathione S-transferase (GST) substrate chlorodinitrobenzene(CDNB, see below), with an excess of the ten aryl hydrazides2a–2j at room temperature. The hydrazides were chosen to randomlydisplay aryl and heteroaryl groups and featured eight acyl and twosulfonyl hydrazides (2d and j). Equilibration at pH 6.2 wasslow, and only two of the ten possible hydrazones could be observedby high-performance liquid chromatography (HPLC) after 1 h(Fig. 2c). Notably, there was a significant amount of free aldehyde 1present throughout the reaction, despite the presence of excessamounts of the ten different hydrazides. Equilibrium was not com-plete after 48 h, and required incubation for a further 5 days untilthe library composition reached a steady-state composition withsignals for each of the ten hydrazones 3a–3j being clearly identified.In contrast, repeating the experiment in the presence of excessaniline produced a far higher rate of equilibration. A distribution ofacylhydrazones was observed after initial mixing and HPLC sampling,and complete equilibration of the ten components was observed afterjust 6 h (Fig. 2d). Aldehyde 1 could not be detected following initialmixing, indicating that it was continually being sequestered as an acyl-hydrazone component, reflecting the faster exchange processes oper-ating in the presence of aniline (see Supplementary Information for astudy on the effect of varying aniline concentration on rates of hydra-zone formation). We demonstrated the reversibility of the DCL bygenerating it from a different starting composition, hydrazone 3gplus the nine other hydrazides and aniline. An identical equilibriumdistribution to Fig. 2 was observed, indicating true thermodynamicequilibrium. A second control experiment confirmed the reversibilityof the DCL through the addition of excess hydrazide 2b to the pre-equilibrated DCL, which resulted in a large amplification of the corre-sponding acylhydrazone 3b (see Supplementary Information).

Having established that aniline could act as an effective nucleo-philic catalyst for hydrazone DCC formation at both a pH and time-frame reasonable for biomolecule stability, our next step was tointroduce proteins to the DCL. Our target chosen for DCC interrog-ation was the GST enzyme superfamily45. The GSTs are responsiblefor cell detoxification, catalysing the conjugation of glutathione(GSH) to a wide variety of xenobiotic electrophiles, thereby

protecting the cell from cytotoxic and oxidative stress. We havepreviously developed thiol conjugate addition DCLs directedtowards GST inhibition, and successfully interfaced the enzymewith small molecules so that it controlled library evolution27. TheGSTs are well suited to exploration using DCC methods, beingwell-characterized, robust proteins having nascent medicinal chem-istry application46,47. There are relatively few ligands reported in theliterature for GST binding—a plus point, as it would enable us to useDCC as a genuine discovery tool for new binding motifs, rather thanas a proof-of-principle process for confirming the binding ability ofknown ligands. The cytoplasmic GSTs are inherent dimers withactive sites composed of residues from both monomers, bifurcatingbetween a highly conserved G-site, which binds the endogenousligand GSH, and an H-site, which binds hydrophobic substratesfor GSH conjugation (Fig. 3). This bisubstrate architecture isparticularly appropriate for DCC interrogation, given that themethod essentially uses a reversible linkage to couple two sets offragment structures together48. Furthermore, within the GST super-family, the large, heterogeneous H-sites are functionally evolvedto accommodate many different hydrophobic substrates forconjugation, a classically difficult architecture to investigate usingorthodox structure-based drug-design methods.

We prepared two recombinant GST isozymes as targets, SjGSTfrom the helminth worm Schistosoma japonicum, a drug target intropical disease49, and hGST P1-1, a human isoform that has beentargeted in the treatment of chemotherapy drug resistance50. Aninitial control experiment with SjGST established that the enzymeretained GSH conjugation activity in the presence of aniline (upto 20 mM). The acylhydrazone DCL prepared in Fig. 2 was theninterfaced with the two protein targets and amplification wasmeasured (Fig. 4). Both DCLs demonstrated strikingly clear ampli-fication of hydrazone components; thiophene acylhydrazone 3g wasselected by SjGST and t-butylphenyl hydrazone 3c by hGST P1-1.

Synthesizing the DCL in the presence of bovine serum albumin(BSA, 1 equiv.) as a control experiment produced no measurableamplification of any component, indicating the GST enzymes asbeing responsible for component amplification. We further demon-strated that amplified components were bound in the targetH-region of the active site of the enzyme by performing conjugationexperiments with GSH. Conjugation of GSH to the aryl chloridegroup in hydrazones 3 by means of SNAr substitution is a slow reactionat pH 6.2, taking several days. In the presence of catalytic amounts ofSjGST, however, rapid formation of the SNAr conjugation adduct forhydrazone 3g was observed at pH 6.2. The amplified hydrazone canthus act as a substrate for SjGST and binds in the targeted H-site.

The amplified hydrazones were re-synthesized and assayedagainst both GSTs and found to be inhibitors of GSH conjugationof CDNB, but poor solubility prevented the determination of accu-rate IC50 values at the higher concentrations necessary to assay weakbinding compounds. To solve this problem, and simultaneouslyincrease the potency of our DCL components, we conjugatedGSH to aldehyde 1 using an SNAr reaction. We anticipated thatthe highly soluble GSH tripeptide motif would act as an ‘anchor’at the G-site, enabling exploration of the H-site with assorted hydra-zide fragments. This approach, in which a known enzyme–substrateinteraction is used for inhibitor discovery, is well exemplified in clas-sical medicinal chemistry drug design, GST inhibition51 and DCCmethods. The IC50 value for SjGST inhibition of the anchored frag-ment 4 was measured in the CDNB conjugation assay as 280 mM.

Initial DCC experiments using GS-conjugated aldehyde 4 andthe same ten hydrazides used previously confirmed the utility ofaniline as a nucleophilic catalyst (Fig. 5). Equilibration was completein 6 h, compared to 4 days in the absence of aniline, and each of theten acylhydrazones were clearly identified by liquid chromatography-mass spectrometry (LC-MS) (see Supplementary Information).As before, clear amplifications could be observed for both GST

Figure 3 | Structure of GST illustrating H- and G-sites. Grey, monomer 1;

yellow, monomer 2; green, G-site; red, H-site.





http://www.nature.com/compfinder/10.1038/nchem.658_comp2j

http://www.nature.com/compfinder/10.1038/nchem.658_comp2d






http://www.nature.com/compfinder/10.1038/nchem.658_comp3g

http://www.nature.com/compfinder/10.1038/nchem.658_comp2b











targets: in each case the same hydrazide fragment was selected as thebest binder, thiophene (5g) for SjGST and t-butylphenyl (5c).Both components were amplified to over 300% of their con-centrations in the blank DCL, at the expense of nearly all other com-peting hydrazones. Also of note, the anisyl sulfonylhydrazone 5junderwent �100% amplification using hGST P1-1 as the onlyother positively selected component. The most significantreductions in equilibrium concentrations occurred for 5b, f and i(SjGST) and 5f, g and i (hGST P1-1).

The GST-directed DCLs were synthesized with the protein presentfrom the beginning of the experiment, that is, in the presence of alde-hyde 4 and the ten hydrazides 2a–2j. To verify that the amplificationresults were not due to a kinetic selection by means of target-acceler-ated synthesis, we added SjGST to the pre-equilibrated DCL. Thesame equilibrium distribution was achieved as is shown in Fig. 5,with hydrazone 5g strongly amplified, indicating that the amplifiedcomponents are the result of genuine thermodynamic selection.Further controls involved a BSA control experiment, which was nega-tive, and DCL synthesis in the presence of a large excess of the non-selective GST inhibitor ethacrynic acid. Component amplificationwas completely suppressed for both SjGST and hGST P1-1 DCLs,indicating that the GST active site is saturated by the ethacrynicacid and cannot influence the DCL equilibrium composition.

We completed our protein-directed DCL studies by preparing acatalytically inactive SjGST mutant. It was of interest to see whethera functionally disabled enzyme would exert the same control andselectivity on DCL composition as the wild-type enzyme. The con-served Tyr 7 active site residue is known to play a critical role inGSH conjugation for the Sj class of GSTs, stabilizing the GSH thio-late anion through H-bonding from the phenol group, withenzymes lacking this residue being catalytically inactive52. We pre-pared a Y7F mutant of SjGST, in which the crucial tyrosine

residue is replaced with phenylalanine. We observed essentiallyzero activity with this mutant in CDNB conjugation when com-pared with the wild-type SjGST. However, SjGST Y7F provedequally effective in controlling DCL composition, showing a clearpreference for the same thiophene derivative 5g as was amplifiedby the wild-type SjGST (see Supplementary Information).

Biological assay was then performed to establish whether the bestbinding compounds in the GST-directed DCLs were also the bestinhibitors of the GST enzyme. To fully explore the isozyme-specificamplification effects of the two DCLs, we separately synthesizedhydrazone conjugates 5a–5j for study. We first confirmed that theamplified ligands 5c and 5g bound to SjGST and hGST P1-1 by iso-thermal calorimetry (ITC) (see Supplementary Information). Wethen studied their inhibitory activity towards SjGST and hGSTP1-1 using the CDNB conjugation assay. The IC50 values wereslightly higher for all hydrazones against hGST P1-1 compared toSjGST (data ranging from 59 to 126 mM and 22 to 63 mM, respect-ively; see Supplementary Information). For each isozyme, the DCCamplified hydrazone was the most active; thiophene 5g had thelowest IC50 value (22 mM) among all the library members againstSjGST, and t-butylphenyl 5c had the lowest value among the fourconjugates tested against hGST P1-1 (57 mM). The DCL hydrazoneselection process has successfully extended inhibitor structure in theGST H-site, increasing potencies by sixfold for hGST P1-1 (331 to57 mM ) and by over tenfold for SjGST (279 to 22 mM) relative tothe starting anchored aldehyde 4.

Steady-state kinetic studies on the two amplified DCL com-ponents 5c (hGST P1-1) and 5g (SjGST) confirmed the expectedcompetitive inhibition profile, with both compounds binding tothe GST active sites. It was interesting to note slightly higher Kivalues for both compounds when assayed against CDNB, a substratefor the H-site of the enzyme, relative to the endogenous G-site

3gCl

NO2

N

HN

O

S

3c

ClNO2

N

HN

O

Abs

orba

nce

(mA

U)

Abs

orba

nce

(mA

U)

Abs

orba

nce

(mA

U)

0

10

5

15

2025

30a

b

c

0102030405060

0

5

10

15

20

20 22 24 26 28 30 32

Time (min)

Time (min)

Time (min)

20 22 24 26 28 30 32

20 22 24 26 28 30 32

Blank

SjGST

hGST P1-1

Figure 4 | GST-templated DCLs. a, DCL hydrazone composition in the absence of any target (blank). b, When the DCL is constituted in the presence of

SjGST, the thiophene hydrazone 3g is clearly amplified. c, Changing the target protein to hGSTP1-1 produces a different distrubution, in which the

t-butylphenyl derivative 3c is amplified. Targeted DCL conditions: GST (1 equiv.), aldehyde (5 mM), hydrazides (20mM) and aniline (10 mM) in NH4OAc

buffer (50 mM, pH¼ 6.2) containing 15% DMSO for 16 h.







http://www.nature.com/compfinder/10.1038/nchem.658_comp5f

http://www.nature.com/compfinder/10.1038/nchem.658_comp5i

http://www.nature.com/compfinder/10.1038/nchem.658_comp5f


http://www.nature.com/compfinder/10.1038/nchem.658_comp5i




















ligand GSH. The affinity of the two hydrazone conjugates towardsboth GST G-sites was relatively close (data ranging from 5.25 to7.19 mM), as would be expected for two compounds sharing acommon GSH-tagged nitrobenzene fragment.

To obtain some molecular insight into the selectivity of our iso-zymes towards the two hydrazone inhibitors 5c and g, we carriedout a molecular modelling study. We surveyed the available GSTstructures in the protein data bank (PDB) and retrieved those thatcontained a bound GSH-based ligand. The binding sites of thesestructures, together with the bound ligands, were aligned, andit became evident that the glutathione portions overlaid well,

being bound in very similar conformations in the G-sites (Fig. 6a).In contrast, the conjugate parts of the various ligands showed greatdiversity in their conformations within the H-site, an unsurprisingresult given the respective functions of the G- and H-sites. Detailedanalysis of the superimposed crystal structures identified the GSHconjugate of 1,2-epoxy-3-( p-nitrophenoxy)propane (EPNP) (6)bound to cGST M1-1 (PDB code 1c72)53 as the ligand that projectedfunctionality into the H-site with the most similar geometry to theenergy-minimized structure of hydrazone 5g (Fig. 6b).

Analysis of the GST–EPNP complex shows the EPNP moietyorienting towards R107 and Q165 in the H-site of the enzyme

200

0

50

100

150

Abs

orba

nce

(mA

U)

Time (min)18 20 22 24 26 28 30

200

0

50

100

150

Abs

orba

nce

(mA

U)

Time (min)18 20 22 24 26 28 30

200

0

50

100

150

Abs

orba

nce

(mA

U)

Time (min)

18 20

5a0

5

10

15

20

25

% A

rea

30

35

40

45Blank

Templated by SjGST

Templated by hGST P1-1

5b 5gHydrazone

5e 5j 5d 5c5h 5i+5f

22 24 26 28 30

Blank

SjGST

hGST P1-1

5a

5h

2j

5i + f5b

5g

5e 5j 5d5c

GS

NO2

4

CHO

R NHNH2

O

2a–j

GS

NO2

N

HN R

O

5a–j

AnilinepH 6.2

GST template

a

b

c

5gGS

NO2

N

HN

O

S

5cGS

NO2

N

HN

O

Figure 5 | GST-templated DCLs of GSH conjugates. a, Acyl hydrazone DCL based on GSH-conjugated aldehyde 4 (GS¼ S-linked glutathione). b, DCL

hydrazone composition in the absence of target (blank), in the presence of SjGST and in the presence of hGSTP1-1. DCL conditions: GST (1 equiv.), aldehyde

(5 mM), hydrazides (20mM) and aniline (10 mM) in NH4OAc buffer (50 mM, pH¼ 6.2) containing 15% DMSO for 16 h. c, Changes in DCL component

concentration for blank, SjGST and hGST P1-1 DCLs. The error bars represent the standard deviation over three experiments.











(Fig. 7c). The side chains of R107, F110, Q165, Q166 and F208define the pocket that confines the EPNP moiety. On this basis,we could generate a binding model for SjGST with thiophene hydra-zone 5g and for hGST P1-1 with t-butyl hydrazone 5c (Fig. 7). Theinteractions in the generated binding modes for SjGST in complexwith 5g (Fig. 7a) and for hGST P1-1 in complex with 5c (Fig. 7b)between the glutathione moiety and the proteins are identical tothose reported in previous publications54,55. We predict that thehydrazone group of 5g forms hydrogen bonds to R103 and Q204in SjGST, and equivalent interactions are observed for 5c incomplex with hGST P1-1. Residue V161 in SjGST and I161 inhGST P1-1 make hydrophobic interactions in our models withthe ligands 5c and 5g.

As expected, the sub-pockets of both isoforms accommodatingthe hydrazones are rather hydrophobic, and complement the hydro-phobic hydrazones amplified from the DCL. The thiophene hydra-zone fits easily in the SjGST binding pocket, with only minor side-chain adjustments necessary (root mean square deviation (RMSD)0.3 Å between model and crystal structure template), whereas thet-butylphenyl group would lead to a steric clash and wouldrequire some degree of induced fit in order to bind. Induced fit isalso required to accommodate this ligand in the hGST P1-1pocket, but in that case the binding mode could be stabilized byadditional lipophilic interactions of the t-butyl group with Y103,H162 and I161. It is worth noting that Chern and colleagues havereported that mutations in that region of the H-site had a greatimpact on EPNP binding as a substrate, with mutation of cGSTM1-1 Q165 to leucine (V161 in SjGST and I161 in hGST P1-1)

reducing k catEPNP by 59%, although Km showed only small

changes53. Because the amino acids in the equivalent pocket ofSjGST and hGST P1-1 are not highly conserved, these residueshave such a great influence on ligand binding that it is likely thatthese amino-acid exchanges across the isoforms are critical in deter-mining ligand selectivity.

ConclusionsWe have demonstrated that reversible synthesis of acylhydrazonescan be compatible with protein targets by using aniline as a nucleo-philic catalyst. The many advantages of this DCC tool (ready avail-ability of easily customized building blocks, good kinetic andthermodynamic properties leading to ease of analysis, good biologi-cal compatibility forming amide-like linkages) may now be realizedwith biological targets. Most importantly, the acylhydrazone DCLsare truly adaptive, allowing amplification effects to be simply anddirectly related to structures present at equilibrium.

The GST enzyme proved extremely effective as a DCL template,with two isozymes from the GST family smoothly integratingwith the small molecule assemblies and strongly amplifying thebest binding components. The selected hydrazones showedincreased inhibitory activity of over one order of magnitude fromthe starting GSH-tagged benzaldehyde 4, validating the approachin the context of protein–ligand discovery. Interestingly, a single,small DCL composed of only ten members displays isozymeselectivity according to which variant of the GST enzyme is usedas the template.

The study at hand has been deliberately confined to a smallnumber of DCL components so as to thoroughly characterize equi-librium distributions and quantify amplifications with the aniline-catalysed hydrazone method. In principle, much larger hydrazoneDCLs may be accessed to thoroughly explore chemical space, bothwithin the GST H-site and for other biological targets9,15. It maynot be possible, or even desirable, to accurately characterize theequilibrium distribution of such complex DCLs, but this will notbe necessary if one simply seeks to identify prominently amplifiedcomponents from a ligand discovery perspective.

To gain insight into isoform selectivity, we found that eachamplified molecule could be effectively docked into its respectiveGST H-site, although the fine structural features of the SjGSTversus hGST P1-1 H-site that discriminate between thiophenehydrazone 5g and t-butylphenyl hydrazone 5c are unclear at thepresent time. Structural determination of the complexes of variousGST:GS–hydrazone conjugates will be needed for a deeper under-standing of the factors that control H-site selectivity. Work in this

K45N54

Q67

R103

Q204V161

V162

V106

W41

K44 Q51

Q64

R100

N204

Y103H162

I161

W38

K49

N58

Q71

R107

N208

Q166

Q165F110

W45

a b c

Figure 7 | Molecular modelling of amplified DCL components with the GST active site. a, Model of 5g bound to SjGST. b, Model of 5c bound to hGST P1-1.

The binding pocket surfaces are shown in light blue and key amino acids as blue sticks. The ligands are represented in salmon pink, with atoms coloured by

type. Hydrogen bonds of the conjugated ligand parts are shown as yellow dotted lines. c, The EPNP–cGST M1-1 crystal structure (PDB code 1c72). The

binding pocket surface is shown in raspberry pink and key amino acids as red sticks. The ligand is represented in green, with atoms coloured by type.

Hydrogen bonds of the conjugated ligand parts are shown as yellow dotted lines.

GS

O

OHNO2

NO26 5g

GS

N

HN

O

S

a b

Figure 6 | GST ligands. a, Superposition of a selection of GST ligands from

the PDB. b, Conformation of the GST-bound EPNP ligand 6 as found in the

crystal structure of cGST M1-1 (PDB code 1c72, green carbon atoms), relative

to the energy-minimized structure of compound 5g (pink carbon atoms).





















area, together with applications of acylhydrazone DCC to other bio-logical targets, is the subject of our current research.

MethodsAniline catalysis of reversible hydrazone formation. The ten hydrazides 2a–j(10 × 5 ml, 10 mM, DMSO), aldehyde 1 (2 ml, 10 mM, DMSO) and aniline (10 ml,1 M, DMSO) were added to a mixture of DMSO (93 ml) and ammonium acetatebuffer (845 ml, 50 mM, pH 6.2). The DCL was allowed to stand at room temperaturewith occasional shaking, and was monitored periodically by HPLC to establish theblank composition until the relative populations of the hydrazones became constant.The pH of all samples was raised to 8 by the addition of NaOH (15 ml, 1 M,aqueous). LC-MS verified that each of the expected hydrazones was present in theDCL (HPLC conditions: column, Luna 5 m C18(2), 30 mm × 4.6 mm, and Luna 5 m

C18(2), 50 mm × 4.6 mm, in sequence; flow rate, 1 ml min21; wavelength, 254 nm;temperature, 23 8C; gradient, H2O/MeCN (0.01% TFA) from 95% to 80% over6 min, then to 45% over 30 min, and eventually to 5% over 5 min) (Fig. 2d). TheDCL was then re-synthesized in the absence of aniline, and the HPLC traces atdifferent time intervals were compared (Fig. 2c).

Templated DCL aldehyde 1. SjGST (111 ml, 180 mM, in potassium phosphatebuffer 0.1 M, pH 6.8), the ten hydrazides 2a–j (10 × 5 ml, 10 mM, DMSO), aldehyde1 (2 ml, 10 mM, DMSO) and aniline (10 ml, 1 M, DMSO) were added to a mixture ofDMSO (93 ml) and ammonium acetate buffer (734 ml, 50 mM, pH 6.2). The DCLwas allowed to stand at room temperature, with occasional shaking, for 12 h. The pHof the sample was raised to 8 by the addition of NaOH (15 ml, 1 M, aqueous), andthe protein was removed by ultrafiltration using a 10,000 MWCO filter (Vivaspin).HPLC analysis was performed and the traces were compared with the blankcomposition (HPLC conditions: column, Luna 5 m C18(2), 30 mm × 4.6 mm, andLuna 5 m C18(2), 50 mm × 4.6 mm, in sequence; flow rate, 1 ml min21; wavelength,254 nm; temperature, 23 8C; gradient H2O/MeCN (0.01% TFA) from 95% to 80%over 6 min, then to 45% over 30 min, and eventually to 5% over 5 min).

DCL composition was identical, regardless of whether the SjGST was presentfrom the beginning or added after pre-equilibration, but equilibration took morethan 24 h in the latter case.

For the hGST P1-1 templated library, the ten hydrazides 2a–j (10 × 5 ml, 10 mM,DMSO), aldehyde 1 (2 ml, 10 mM, DMSO), aniline (10 ml, 1 M, DMSO) and hGSTP1-1 (100 ml, 200 mM, in potassium phosphate buffer 0.1 M, pH 6.8) were added to amixture of DMSO (93 ml) and ammonium acetate buffer (734 ml, 50 mM, pH 6.2).After equilibration for 12 h, the DCL was analysed using HPLC. Control experimentswere performed using the same equivalents of BSA in place of GST.

Conjugate DCLs. To establish the blank DCL composition, the ten hydrazides 2a–j(10 × 5 ml, 10 mM, DMSO), aldehyde 5 (5 ml, 10 mM, aqueous) and aniline (10 ml,1 M, DMSO) were added to a mixture of DMSO (96 ml) and ammonium acetatebuffer (839 ml, 50 mM, pH 6.2). The DCL was allowed to stand at room temperature,with occasional shaking, and was monitored periodically by HPLC to establish theblank composition until the relative populations of the hydrazones became constant.The pH of all the samples was increased to 8 by the addition of NaOH (15 ml, 1 M,aqueous). LC-MS verified that each of the expected hydrazones was present in theDCL (Fig. 5) (HPLC conditions: column, Luna 5 m C18(2), 50 mm × 4.6 mm, andLuna 5 m C18(2), 250 mm × 4.6 mm, in sequence; flow rate, 1 ml min21;wavelength, 254 nm; temperature, 23 8C; gradient H2O/MeCN (0.01% TFA) from95% to 5% over 40 min).

For re-synthesizing the DCL in the presence of the protein SjGST (278 ml,180 mM, in potassium phosphate buffer 0.1 M, pH 6.8), the ten hydrazides 2a–j(10 × 5 ml, 10 mM, DMSO), aldehyde 5 (5 ml, 10 mM, DMSO) and aniline (10 ml,1 M, DMSO) were added to a mixture of DMSO (96 ml) and ammonium acetatebuffer (561 ml, 50 mM, pH 6.2). The DCL templated by hGST P1-1 was synthesizedby adding the ten hydrazides 2a–j (10 × 5 ml, 10 mM, DMSO), aldehyde 5 (5 ml,10 mM, DMSO), aniline (10 ml, 1 M, DMSO) and hGST P1-1 (250 ml, 200 mM, inpotassium phosphate buffer 0.1 M, pH 6.8) to a mixture of DMSO (96 ml) andammonium acetate buffer (589 ml, 50 mM, pH 6.2). The DCLs were allowed to standat room temperature for 12 h, after which the pH was raised to 8 by the addition ofNaOH (15 ml, 1 M). The protein was filtered off using a centrifuge filter of MWCO10,000 followed by analysis of the filtrate by HPLC using conditions similar to thoselisted above.

Molecular modelling. To establish ligand alignment, the superposition of GSTligands was carried out using Relibaseþ 3.0.0 (ref. 56). A search was first performedto find binding sites that share a sequence identity between 40 and 100% with thetarget GST crystal structure 1m9a. The resulting 38 structures with bound ligandwere superimposed by using binding site residues only. Finally, the ligands from thesuperimposed structures were extracted and visually analysed.

To carry out a binding mode prediction with Moloc57, the SjGST crystalstructure (PDB code 1m9a SjGST – S-hexyl–GSH complex) and the hGSTP1-1–GSH complex crystal structure (PDB code 6gss)54 were used as startingconformations for binding mode generation. The glutathione groups of thesynthesized ligands were mapped onto the glutathione groups of the ligands boundto the crystal structures. The hydrophobic hydrazone groups of the synthesizedligands were oriented towards the cavity, lying at the end of the S-hexyl site, as

observed for the EPNP ligand bound to cGSTM1-1 (PDB code 1c72). In the nextstep, the protein in complex with the modelled ligand was minimized, consideringthe ligand as fully flexible. For the protein all residues were kept rigid, except for theamino acids that define the pocket at the end of the S-hexyl site (R103, V106, V161,V162, Q204 for SjGST and R100, Y103, I161, H162, N204 for hGST P1-1).

Received 7 December 2009; accepted 30 March 2010;published online 16 May 2010

References1. Lehn, J.-M. Dynamic combinatorial chemistry and virtual combinatorial

libraries. Chem. Eur. J. 5, 2455–2463 (1999).2. Rowan, S. J., Cantrill, S. J., Cousins, G. R. L., Sanders, J. K. M. & Stoddart, J. F.

Dynamic covalent chemistry. Angew. Chem. Int. Ed. 41, 898–952 (2002).3. Corbett, P. T. et al. Dynamic combinatorial chemistry. Chem. Rev. 6,

3652–3711 (2006).4. Ladame, S. Dynamic combinatorial chemistry: on the road to fulfilling the

promise. Org. Biomol. Chem. 6, 219–226 (2008).5. Erlanson, D. A. et al. Site-directed ligand discovery. Proc. Natl Acad. Sci. USA 97,

9367–9372 (2000).6. Corbett, A. R., Cheeseman, J. D., Kazlauskas, R. J. & Gleason, J. L.

Pseudodynamic combinatorial libraries: a receptor-assisted approach for drugdiscovery. Angew. Chem. Int. Ed. 43, 2432–2436 (2004).

7. Hochgurtel, M. et al. Target-induced formation of neuraminidase inhibitorsfrom in vitro virtual combinatorial libraries. Proc. Natl Acad. Sci. USA 99,3382–3387 (2002).

8. Zameo, S., Vauzeilles, B. & Beau, J. M. Dynamic combinatorial chemistry:lysozyme selects an aromatic motif that mimics a carbohydrate residue. Angew.Chem. Int. Ed. 44, 965–969 (2005).

9. McNaughton, B. R., Gareiss, P. C. & Miller, B. L. Identification of a selectivesmall-molecule ligand for HIV-1 frameshift-inducing stem-loop RNA from an11,325 member resin bound dynamic combinatorial library. J. Am. Chem. Soc.129, 11306–11307 (2007).

10. Gareiss, P. C. et al. Dynamic combinatorial selection of molecules capable ofinhibiting the (CUG) repeat RNA-MBNL1 interaction in vitro: discovery of leadcompounds targeting myotonic dystrophy (DM1). J. Am. Chem. Soc. 130,16254–16261 (2008).

11. Scott, D. E., Dawes, G. J., Ando, M., Abell, C. & Ciulli, A. A fragment-basedapproach to probing adenosine recognition sites by using dynamiccombinatorial chemistry. ChemBioChem 10, 2772–2779 (2009).

12. Eliseev, A. V. & Nelen, M. I. Use of molecular recognition to drive chemicalevolution 1. Controlling the composition of an equilibrating mixture of simplearginine receptors. J. Am. Chem. Soc. 119, 1147–1148 (1997).

13. Hioki, H. & Still, W. C. Chemical evolution: a model system that selects andamplifies a receptor for the tripeptide (D)Pro(L)Val(D)Val. J. Org. Chem. 63,904–905 (1998).

14. Wietor, J. L., Pantos, G. D. & Sanders, J. K. M. Templated amplification of anunexpected receptor for C-70. Angew. Chem. Int. Ed. 47, 2689–2692 (2008).

15. Ludlow, R. F. & Otto, S. Two-vial LC-MS identification of ephedrine receptorsfrom a solution phase dynamic combinatorial library of over 9,000 components.J. Am. Chem. Soc. 130, 12218–12219 (2008).

16. Turega, S. M., Lorenz, C., Sadownik, J. W. & Philp, D. Target-driven selection ina dynamic nitrone library. Chem. Commun. 4076–4078 (2008).

17. Xu, S. & Giuseppone, N. Self-duplicating amplification in a dynamiccombinatorial library. J. Am. Chem. Soc. 130, 1826–1827 (2008).

18. Sadownik, J. W. & Philp, D. A simple synthetic replicator amplifies itself from adynamic reagent pool. Angew. Chem. Int. Ed. 47, 9965–9970 (2008).

19. Nguyen, R., Allouche, L., Buhler, E. & Giuseppone, N. Dynamic combinatorialevolution within self-replicating supramolecular assemblies. Angew. Chem. Int.Ed. 48, 1093–1096 (2009).

20. Chichak, K. S. et al. Molecular Borromean rings. Science 304, 1308–1312 (2004).21. Lam, R. T. S. et al. Amplification of acetylcholine-binding catenanes from

dynamic combinatorial libraries. Science 308, 667–669 (2005).22. Au-Yeung, H. Y., Pantos, G. D. & Sanders, J. K. M. Dynamic combinatorial

synthesis of a catenane based on donor–acceptor interactions in water. Proc. NatlAcad. Sci. USA 106, 10466–10470 (2009).

23. Tauk, L., Schroder, A. P., Decher, G. & Giuseppone, N. Hierarchical functionalgradients of pH-responsive self-assembled monolayers using dynamic covalentchemistry on surfaces. Nature Chem. 1, 649–656 (2009).

24. Fujii, S. & Lehn, J.-M. Structural and functional evolution of a library ofconstitutional dynamic polymers driven by alkali metal ion recognition. Angew.Chem. Int. Ed. 48, 7635–7638 (2009).

25. Kindermann, M., Stahl, I., Reimold, M., Pankau, W. M. & von Kiedrowski, G.Systems chemistry: kinetic and computational analysis of a nearly exponentialorganic replicator. Angew. Chem. Int. Ed. 44, 6750–6755 (2005).

26. Ludlow, R. F. & Otto, S. Systems chemistry. Chem. Soc. Rev. 37, 101–108 (2008).27. Shi, B., Stevenson, R., Campopiano, D. J. & Greaney, M. F. Discovery of

glutathione S-transferase inhibitors using dynamic combinatorial chemistry.J. Am. Chem. Soc. 128, 8459–8467 (2006).






















28. Ramstrom, O. & Lehn, J.-M. In situ generation and screening of a dynamiccombinatorial carbohydrate library against concanavalin A. ChemBioChem 1,41–48 (2000).

29. Otto, S., Furlan, R. L. E. & Sanders, J. K. M. Dynamic combinatorial libraries ofmacrocyclic disulfides in water. J. Am. Chem. Soc. 122, 12063–12064 (2000).

30. Nicolaou, K. C. et al. Target-accelerated combinatorial synthesis and discovery ofhighly potent antibiotics effective against vancomycin-resistant bacteria. Angew.Chem. Int. Ed. 39, 3823–3828 (2000).

31. Milanesi, L., Hunter, C. A., Sedelnikova, S. E. & Waltho, J. P. Amplification ofbifunctional ligands for calmodulin from a dynamic combinatorial library.Chem. Eur. J. 12, 1081–1087 (2006)

32. Shi, B. & Greaney, M. F. Reversible Michael addition of thiols as a new tool fordynamic combinatorial chemistry. Chem. Commun. 886–888 (2005).

33. Huc, I. & Lehn, J.-M. Virtual combinatorial libraries: dynamic generation ofmolecular and supramolecular diversity by self-assembly. Proc. Natl Acad. Sci.USA 94, 2106–2110 (1997).

34. Cousins, G. R. L., Poulsen, S. A. & Sanders, J. K. M. Dynamic combinatoriallibraries of pseudo-peptide hydrazone macrocycles. Chem. Commun.1575–1576 (1999).

35. Furlan, R. L. E., Ng, Y. F., Otto, S. & Sanders, J. K. M. A new cyclic pseudopeptidereceptor for Liþ from a dynamic combinatorial library. J. Am. Chem. Soc. 123,8876–8877 (2001).

36. Roberts, S. L., Furlan, R. L. E., Cousins, G. R. L. & Sanders, J. K. M. Simultaneousselection, amplification and isolation of a pseudo-peptide receptor by animmobilised N-methyl ammonium ion template. Chem. Commun.938–939 (2002).

37. Liu, J. Y., West, K. R., Bondy, C. R. & Sanders, J. K. M. Dynamic combinatoriallibraries of hydrazone-linked pseudo-peptides: dependence of diversity onbuilding block structure and chirality. Org. Biomol. Chem 5, 778–786 (2007).

38. Bunyapaiboonsri, T. et al. Dynamic deconvolution of a pre-equilibrated dynamiccombinatorial library of acetylcholinesterase inhibitors. ChemBioChem 2,438–444 (2001).

39. Bunyapaiboonsri, T., Ramstrom, H., Ramstrom, O., Haiech, J. & Lehn, J.-M.Generation of bis-cationic heterocyclic inhibitors of Bacillus subtilis HPrkinase/phosphatase from a ditopic dynamic combinatorial library. J. Med.Chem. 46, 5803–5811 (2003).

40. Poulsen, S. A. Direct screening of a dynamic combinatorial library using massspectrometry. J. Am. Soc. Mass Spectrom. 17, 1074–1080 (2006).

41. Cordes, E. H. & Jencks, W. P. Nucleophilic catalysis of semicarbazone formationby anilines. J. Am. Chem. Soc. 84, 826–831 (1962).

42. Dirksen, A., Dirksen, S., Hackeng, T. M. & Dawson, P. E. Nucleophilic catalysisof hydrazone formation and transimination: implications for dynamic covalentchemistry. J. Am. Chem. Soc. 128, 15602–15603 (2006).

43. Dirksen, A. & Dawson, P. E. Rapid oxime and hydrazone ligations with aromaticaldehydes for biomolecular labeling. Bioconjugate Chem. 19, 2543–2548 (2008).

44. Rodriguez-Docampo, Z. & Otto, S. Orthogonal or simultaneous use of disulfideand hydrazone exchange in dynamic covalent chemistry in aqueous solution.Chem. Commun. 5301–5303 (2008).

45. Hayes, J. D., Flanagan, J. U. & Jowsey, I. R. Glutathione transferases. Annu. Rev.Pharmacol. Toxicol. 45, 51–88 (2005).

46. Mahajan, S. & Atkins, W. M. The chemistry and biology of inhibitors andpro-drugs targeted to glutathione S-transferases. Cell. Mol. Life Sci. 62,1221–1233 (2005).

47. Li, W.-S. et al. Overcoming the drug resistance in breast cancer cells byrational design of efficient glutathione S-transferase inhibitors. Org. Lett. 12,20–23 (2010).

48. Murray, C. W. & Rees, D. C. The rise of fragment-based drug discovery. NatureChem. 1, 187–192 (2009).

49. Jao, S. C., Chen, J., Yang, K. & Li, W. S. Design of potent inhibitors forSchistosoma japonica glutathione S-transferase. Bioorg. Med. Chem. 14,304–318 (2006).

50. Tew, K. D. Glutathione-associated enzymes in anticancer drug-resistance.Cancer Res. 54, 4313–4320 (1994).

51. Lyon, R. P., Hill, J. J. & Atkins, W. M. Novel class of bivalent glutathione S-transferase inhibitors. Biochemistry 42, 10418–10428 (2003).

52. Andujar-Sanchez, M. et al. Crystallographic and thermodynamic analysis of thebinding of S-octylglutathione to the Tyr 7 to Phe mutant of glutathioneS-transferase from Schistosoma japonicum. Biochemistry 44, 1174–1183 (2005).

53. Chern, M. K. et al. Tyr115, Gln165 and Trp209 contribute to the 1,2-epoxy-3-(p-nitrophenoxy)propane-conjugating activity of glutathione S-transferasecGSTM1-1. J. Mol. Biol. 300, 1257–1269 (2000).

54. Oakley, A. J. et al. The structures of human glutathione transferase P1-1 incomplex with glutathione and various inhibitors at high resolution. J. Mol. Biol.274, 84–100 (1997).

55. Cardoso, R. M. F., Daniels, D. S., Bruns, C. M. & Tainer, J. A. Characterization ofthe electrophile binding site and substrate binding mode of the 26-kDaglutathione S-transferase from Schistosoma japonicum. Proteins 51,137–146 (2003).

56. Bergner, A., Gunther, J., Hendlich, M., Klebe, G. & Verdonk, M. Use of relibasefor retrieving complex three-dimensional interaction patterns includingcrystallographic packing effects. Biopolymers 61, 99–110 (2001).

57. Gerber, P. R. & Muller, K. MAB, a generally applicable molecular-force field forstructure modeling in medicinal chemistry. J. Comput. Aided Mol. Des. 9,251–268 (1995).

AcknowledgementsThe authors would like to thank EastChem for the award of a studentship to V.T.B. and theMarie Curie Early Stage Training Network (Syn4chembio) and School of Chemistry atEdinburgh for awarding a studentship to A.M.C. R.B. is supported by an EC SeventhFramework Programme (FP7/2007-2013) under grant agreement no. 223461. M.F.G. is anEngineering and Physical Sciences Research Council (EPSRC) Leadership Fellow. Theauthors thank A. Cooper (University of Glasgow) for ITC measurements and helpfuldiscussions. N. Petitjean is thanked for the synthesis of hydrazone–GSH conjugates.

Author contributionsV.T.B., A.M.C., D.J.C. and M.F.G. conceived and designed the experiments, V.T.B. andA.M.C. performed the experiments, and T.L., R.B. and A.M.C. carried out molecularmodelling. All authors discussed the results and co-wrote the manuscript.

Additional informationThe authors declare no competing financial interests. Supplementary information andchemical compound information accompany this paper at www.nature.com/naturechemistry. Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/. Correspondence and requests for materials should be addressedto D.J.C. and M.F.G.














Electrochemistry through glassJeyavel Velmurugan, Dongping Zhan† and Michael V. Mirkin*

In this Article we have used new approaches to investigate a well-known chemical process, the propagation ofelectrochemical signals through a thin glass membrane. This process, which has been extensively studied over the lastcentury, is the basis of the response of a potentiometric glass pH sensor; however, no amperometric glass sensors haveyet been reported because of its high ohmic resistance. Voltammetry at nanoelectrodes has revealed that water moleculescan diffuse through nanometre-thick layers of dry glass and undergo oxidation/reduction at the buried platinum surface.After soaking for a few hours in an aqueous solution, voltammetric waves of other redox couples, such as Ru(NH3)6

31/21,could also be obtained at the glass-covered platinum nanoelectrodes. This behaviour suggests that the nanometre-thickinsulating glass sheath surrounding the platinum core can be largely converted to hydrated gel, and electrochemicalprocesses occur at the platinum/hydrogel interface. Potential applications range from use in nanometre-sized solid-statepH probes and determination of the water content in organic solvents to glass-modified voltammetric sensorsand electrocatalysts.

An intriguing aspect of nanoelectrochemical experiments isthe possibility of using a nanometre-sized electrode toobserve processes and phenomena that are not accessible

using macroscopic probes1. Examples include single molecularevents2, unusual transport phenomena3 and electrical double-layereffects4. The subject of this article—electrochemistry throughglass—may sound like an oxymoron, because glass is commonlyused as an insulating material in electrode fabrication. However,�100-nm-thick layers of glass have been found to be sufficientlyconductive for electrochemical measurements.

The propagation of an electrical signal through glass membraneshas been extensively studied since the beginning of the twentiethcentury because of its relevance to the potentiometric glass pH elec-trode5. Because the potential of the glass electrode is a linear functionof solution pH, it is intuitive to assume that the voltage drop across themembrane is determined by proton transfer. However, numerousexperiments using the 3H isotope and other methods have shownthat protons do not cross the glass membrane6,7. The potentiometricresponse of the pH electrode originates in the ion-exchange equili-brium on the glass surface, and the diffusion of protons and wateris essentially confined to a nanometre-thick surface layer of hydratedgel. This gel forms on both sides of the membrane when it is soakedfor several hours in acidic aqueous solution8. This response mechan-ism was established for electrodes with micrometre-thick sensingmembranes (typical thickness, �100 mm). The behaviour ofnanometre-thick glass layers, however, is substantially different.

We used nanometre-sized platinum electrodes to investigate thepermeability of glass in aqueous and non-aqueous solutions. Anelectrode of this type with a conductive core radius of a ≥ 5 nmcan be produced by pulling a platinum microwire into a borosilicateglass capillary with the help of a laser pipette puller. After pulling,the metal wire is completely sealed into the glass, and its nano-metre-sized tip can be exposed by gentle polishing under videomicroscopic control9. The geometry of the polished nanoelectrodeswas characterized by a combination of voltammetry, scanning elec-tron microscopy (SEM) and scanning electrochemical microscopy(SECM). It was shown that the effective radius value (a) determinedfrom steady-state voltammetry is close to the geometric radius of theconductive disk. The absence of detectable solution leakage through

the glass seal was also demonstrated9. By selecting appropriatepulling parameters, the thickness of the glass at the tip could bevaried between a few tens of nanometres and several micrometres.Pulled platinum probes encased in submicrometre-thick glasswere used in the experiments described in this Article.

ResultsFigure 1a shows cyclic voltammograms (CVs) obtained for twoglass-covered platinum nanoelectrodes in 0.5 M H2SO4 solution.Curve 1, obtained for an electrode with a thicker glass sheath, exhi-bits very low background current and no cathodic or anodic waveswithin a wide potential window (+3 V). This response could beexpected, because the conductive platinum core is completelyburied in borosilicate glass, which is often used as an insulatingmaterial in the fabrication of electrodes. Curve 2 was obtainedwith the platinum electrode buried inside a significantly thinnerglass sheath (right image in Fig. 1b). This CV shows well-definedanodic and cathodic currents at electrode potentials ofE ≥þ1.5 V and E ≤ 20.5 V versus a Ag/AgCl reference, respect-ively. These currents, which increase exponentially with appliedpotential and can reach relatively high values (nA), are clearly dueto the electrochemical oxidation/reduction processes. The onlyredox species present in our system that can be reduced or oxidizedwithin the above potential range are water and molecular oxygen.The current in curve 2 (Fig. 1a) was essentially unaffected byoxygen removal, thus suggesting that the anodic and cathodic pro-cesses are the oxidation and reduction of water. The differencebetween the onset potentials of the cathodic and anodic waves incurve 2 (�2 V) is somewhat larger than the theoretical minimumvoltage required for water electrolysis (1.23 V).

To eliminate the possibility that water molecules diffuse to theplatinum surface through microscopic cracks or pinholes in theglass sheath, we obtained voltammograms of hydrophilic(Ru(NH3)6

3þ; curve 1 in Fig. 1c) and relatively hydrophobic (forexample, ferrocenemethanol, FcCH2OH; not shown) redox speciesat glass-covered platinum electrodes. The complete absence of areduction wave observed with Ru(NH3)6

3þ concentrations as highas 20 mM suggests that the platinum core is completely coveredby glass. The sensitivity of this test to extremely small pinholes in

Department of Chemistry and Biochemistry, Queens College—CUNY, Flushing, New York 11367, USA; †Present address: Department of Chemistry, XiamenUniversity, Xiamen 361005, China. *e-mail: [email protected]







a glass insulator can be seen from equation (1) for the diffusionlimiting current to a disk-shaped electrode:

id = 4FDac∗ (1)

where F is the Faraday constant, D¼ 6.5 × 1026 cm2 s21 is thediffusion coefficient of Ru(NH3)6

3þ (ref. 9) and c* is its concen-tration in solution. Assuming that a 1-nm-radius platinum disk isexposed to the solution due to the presence of a pinhole, one

obtains id¼ 5 pA for c*¼ 20 mM. Curve 2 in Fig. 1c is anexample of such pinhole detection. The well-defined reductionwave of Ru(NH3)6

3þ corresponds to a value of a as small as0.7 nm. The actual defect size may be somewhat larger, becausethe measured current is affected by the diffusion of redox speciesthrough the pore in the glass film. However, the smallest currentmeasurable by our instrument (≤50 fA) is about two orders of mag-nitude lower than that in Fig. 1c, and therefore practically anymicroscopic defect in glass should be detectable.

The currents in Fig. 1a cannot be attributed to diffusion ofsodium in glass, because water has to physically cross the membraneto be reduced or oxidized at the platinum surface. To further provethis point we monitored the generation of hydrogen (Fig. 2) andoxygen (not shown) at sufficiently high negative and positive poten-tials. Biasing a glass-covered nanoelectrode at 2900 mV in 0.5 MH2SO4 produced �300–500 pA current of hydrogen reduction(Fig. 2a, inset). After �5 min, the pressure of the generated hydro-gen became high enough to break the insulating sheath and exposethe platinum surface to the solution, resulting in a dramatic increasein cathodic current. Figure 2b shows that the reduction ofRu(NH3)6

3þ, which did not occur at this electrode before the hydro-gen evolution experiment (curve 1), produced a pronounced vol-tammetric wave after the rupture of the insulating glass (curve 2).

Figure 3 shows the response of a glass-covered nanoelectrode towater dissolved in an aprotic organic solvent (1,2-dichloroethane,DCE). The voltammogram obtained with no water added to twicedistilled DCE is essentially flat and contains neither anodic(Fig. 3a) nor cathodic (Fig. 3b) waves. In contrast, several anodicand cathodic waves appear in the curves obtained when differentwater concentrations are added to the DCE (cH2O; from 1 to130 mM, as shown by the colour code in Fig. 3a,b). The height ofall waves increases with cH2O. A more prominent anodic peak isobserved at �3.3 V, which in fact represents two closely spacedpeaks, as can be seen in the yellow curve. The dependence of thispeak current (ip) on cH2O is essentially linear (Fig. 3c).

The behaviour of the glass-covered electrode changes dramati-cally after soaking in acidic solution for a few hours, which isknown to result in glass swelling and the formation of a hydrogelsurface layer. Figure 4a shows CVs of 0.5 M H2SO4 at a glass-covered nanoelectrode before (1) and after (2) it was kept overnightin 6 M HCl solution. In contrast to curve 1, curve 2 exhibits

–6

–4

–2

0

2

–2,000 –1,000 0 1,000 2,000

E versus Ag/AgCl (mV)


i (pA

)i (

pA)

2

1

1.23 m152 nm

505 nm

–300

–200

–100

0

100

200

300a

c

b

–3,000 –2,000 –1,000 0 1,000 2,000 3,000

1

2

Figure 1 | Characterization of glass-covered nanoelectrodes. a, Voltammograms obtained in 0.5 M H2SO4 solution at platinum nanoelectrodes with thicker

(trace 1) and thinner (trace 2) insulating glass sheaths. b, SEM images of thick glass (left) and thin glass (right) electrodes. c, Voltammograms of 20 mM

Ru(NH3)6Cl3 in 1 M KNO3 at two thin glass covered electrodes without (1) and with (2) a nanometre-sized pinhole. Potential sweep rate, n¼ 500 mV s21.

–9.0

–6.0

–3.0

0.0

–600 –400 –200 0 200 400


i (nA

)

2

0

20

40

60

a

b

0 50 100 150 200 250 300t (s)

i (nA

)

i (nA

)

0

2

4

6

0 100 200 300t (s)

1

Figure 2 | Hydrogen evolution by means of through-glass electrolysis of

water. a, Dependence of current on time. The electrode potential was

2900 mV. The inset shows the initial portion of the curve (up to t ≈ 320 s,

when the insulating sheath was broken) at higher current sensitivity.

b, Voltammograms of 20 mM Ru(NH3)63þ obtained before (1) and after (2)

the hydrogen evolution transient shown in a.






well-defined hydrogen, double-layer and oxygen regions. This curveis quite similar to voltammograms of water oxidation/reductionobtained at macroscopic platinum electrodes (inset in Fig. 4a;ref. 10). The current was stable in time and reproducible; the twoconsecutive potential cycles shown in Fig. 4a produced essentiallyidentical responses.

After the acid treatment, a glass-coated electrode responds toredox species other than water. Figure 4b shows a voltammogramof Ru(NH3)6

3þ, which (unlike curve 1 in Fig. 1c obtained at a dryglass-coated electrode) contains both cathodic and anodic waves.The voltammograms of anionic IrCl6

32 (Fig. 4c) and neutralFcCH2OH (Fig. 4d) are not so well shaped.

Figure 5a shows a voltammogram of copper electrodeposition inhydrated glass. The shape of the curve, with a characteristic cathodicpeak appearing after the potential sweep reversal and a sharp anodicpeak of copper stripping, is typical of metal nucleation/growth pro-cesses11. In a chronoamperometric experiment (Fig. 5b), the electrodepotential was stepped to 2600 mV to deposit a larger amount ofcopper. The cathodic current was almost constant during the first8 s (the inset in Fig. 5b) and then increased sharply by a factor of.400 (the highest current of 10 nA in Fig. 5b corresponds to the over-flow of the potentiostat amplifier). The growth of copper beyond thehydrogel limits and the formation of a micrometre-sized metal elec-trode were confirmed voltammetrically (not shown), and the depos-ited copper was observed by optical microscopy and SEM.

The potentiometric response of the hydrated glass nanoelectrodeto solution pH is shown in Fig. 6. After soaking a glass-covered plati-num electrode overnight in 6 M HCl, its potential was measured in

ten buffer solutions of different pH. The linear pH dependence(r2 . 0.99) of the electrode potential over the range of pH from1 to 10 exhibits a slightly sub-Nernstian slope of 52 mV pH21.

DiscussionUsing nanometre-sized probes, we were able to observe oxidation/reduction reactions at electrodes buried inside borosilicate glass,which should not have been possible according to conventionalwisdom. Nanometre-thick layers of glass are sufficiently permeableto observe the oxidation/reduction of water in either aqueous ororganic media. From Fig. 1a it can be seen that the through-glassoxidation/reduction of water occurs with a total tip diameter of�150 nm (curve 2), but is not observed when the diameter is�1 mm (curve 1). The rupture of the insulating sheath by evolvedhydrogen (Fig. 2) or oxygen unequivocally confirms the diffusionof water molecules through the film, and a very sensitive voltam-metric test (Fig. 1c) provides strong evidence against the possibilityof microscopic pores or cracks in the glass layer. So far, no otherelectroactive species—either hydrophilic or hydrophobic, ionic orneutral—have been found to cross the dry glass barrier.

Voltammetry in organic solutions (Fig. 3) provides additionalevidence for molecular diffusion through glass (as opposed to sol-ution permeation through defects), because organic solution doesnot spontaneously enter nanometre-sized holes in hydrophilicglass12. Importantly, no waves appeared within the entire potentialrange from 23 to þ3 V when no water was added to distilledDCE. The dependence of the peak current on cH2O for the peakoccurring at �3.3 V (Fig. 3c) is essentially linear for the entirerange of concentrations from 1 to 130 mM (the latter correspondsto water-saturated DCE13). The linearity of the calibration curveand the detection limit of �0.5 mM attained without any optimiz-ation suggest that voltammetry at glass-covered nanoelectrodes maybecome an alternative to the well-known Karl Fisher titration tech-nique14 for the determination of water in organic solvents. Obviousadvantages of the nanoelectrochemical approach include an extre-mely small sample size and the fact that additional reagents neednot be used.

A hydrated gel layer can be formed by soaking a glass-coverednanoelectrode in acidic solution. Unlike conventional glass pHsensors, for which the thick (�0.1-mm) membrane remainsmostly dry when immersed in an aqueous solution, the �100-nm-thick glass covering our electrodes seems to be largely (if not com-pletely) converted to hydrated gel. The voltammograms of wateroxidation/reduction before and after the formation of hydratedgel are completely different. The hydrogen and oxygen evolutioncurrents in Figs 1 and 2 were produced by the diffusion of watermolecules through the thin layer of dry glass. The currents corre-sponding to these processes can be recorded immediately afterimmersing a glass-covered nanoelectrode in aqueous solution. Theresponse is stable and essentially time-independent on the timescaleof several minutes, which is too short for the slow process of glassswelling. In contrast, hydrated gel is an aqueous environment, inwhich a monolayer of adsorbed water forms on the platinumsurface. Accordingly, curve 2 in Fig. 4a exhibits characteristic vol-tammetric peaks corresponding to adsorption/desorption of hydro-gen and oxygen, similar to those obtained at macroscopic platinumelectrodes in acidic solutions (cf. the inset in Fig. 4a). Such peaks arenot present in curve 1, obtained at a ‘dry’ glass electrode.

Ru(NH3)63þ/2þ species, which were completely blocked by dry

glass (curve 1 in Fig. 1c), yielded a pair of well-defined cathodicand anodic peaks at a hydrated glass electrode (Fig. 4b). Anincreased peak separation, DEp¼ 95 mV, points to a significantresistance of the hydrated glass layer that varies for different electro-des (as does the DEp value). The resistive potential drop apparentlydepends on the thickness of the hydrogel and the surface area ofthe platinum exposed to it. In Fig. 4b, the �35 mV increase in

0

50

100

150

200

a

b

c

0 1,000 2,000 3,000E (mV)

E (mV)

i (pA

)i (

pA)

130 mM50 mM20 mM10 mM5 mM1 mM0 mM

R 2 = 0.9755

0

50

100

150

200

0 50 100 150

i (p

A)

cH2O (mM)

–50

–25

0

–2,700 –1,700 –700

130 mM

50 mM

10 mM

5 mM

Figure 3 | Oxidation/reduction of water in DCE at glass-covered

electrodes. a,b, Anodic (a) and cathodic (b) voltammograms obtained in

DCE solutions containing different concentrations of water. n¼ 500 mV s21.

c, Dependence of the second anodic peak current in a on cH2O. The

supporting electrolyte was 0.1 M tetrabutylammonium perchlorate.






peak separation (the peak separation expected for a reversible CVunaffected by the resistive potential drop is �59 mV) correspondsto the glass layer resistance of ≥100 MV.

An unusual combination of the low (pA) current typical ofnanoelectrodes and a peak-shaped CV that is normally obtainedat macroscopic electrodes can be attributed to considerable viscosity(and, thus, low diffusion coefficients of Ru(NH3)6

3þ/2þ), whichresults in non-steady-state diffusion within a thin layer of hydratedgel. The time at which the diffusion in a thin layer approaches asteady state is of the order of d2/D (ref. 15), where d is the layerthickness. Assuming a film thickness of 100 nm and noting theexperimental timescale in Fig. 4b of �1 s, the apparent D value is≤1 × 10210 cm2 s21.

The hydrated glass film exhibits permselectivity; in comparisonwith cationic Ru(NH3)6

3þ, voltammograms of anionic IrCl632

(Fig. 4c) and the neutral, more hydrophobic FcCH2OH (Fig. 4d)exhibit larger peak separations and less-defined faradaic waves.Electrode surface modification by nanometre-thick hydrated glassmay provide new opportunities for sensor preparation, protectionof electrocatalysts from fouling and inhibitors, and other electroche-mical applications. Another interesting possibility—electrodeposition

of metals within the glass matrix—is suggested by the nucleation andbulk deposition of copper in hydrated gel (Fig. 5).

The pH response of nanometre-thick borosilicate glass is due tothe formation of the hydrogel layer over a few hours of treatmentwith acidic solution. The high resistance of the hydrated gel andvery slow diffusion within it are consistent with the developmentof a membrane potential across this layer. The strong dependenceof the nanoelectrode potential on pH and excellent linearity of thecalibration curve (Fig. 6) suggest that these electrodes can serve asall-solid-state pH nanoprobes. Electrodes containing a buffer sol-ution (presently the most common type of pH sensors16) havemany disadvantages, including storage problems, pressure andtemperature dependence, mechanical instability and relativelylarge size and fabrication cost17. Numerous efforts to producesolid-state pH electrodes based on iridium oxide18, conductivepolymer composites19 and other sensing strategies have met withlimited success.

The main advantages of a glass-coated platinum pH electrode—its microscopic size and biocompatibility—make it potentiallyuseful for cell biology applications and other experiments in smallvolumes. The attainable tip size (tens of nanometres) is comparable

–50

–25

0

25

50a b

–600 –400 –200 0 200 400


i (pA

)

i (nA

)

i (pA

)

200

0

5

10

0 2 4 6t (s)

t (s)

8

0

100

0 2 4 6 8

Figure 5 | Electrodeposition of copper in hydrated glass. a,b, CV (a) and chronoamperogram (b) of copper deposition on the platinum nanowire buried in

hydrated glass from 20 mM CuSO4 solution. In a, n¼ 50 mV s21. In b, the electrode potential was stepped to 2600 mV versus Ag/AgCl. The inset shows

the initial portion of the current transient at a higher current sensitivity.

–150

–100

–50

0

50

100

–600 –400 –200 0

E (mV)

i (pA

)

–100

–50

0

50

100

300 400 500 600 700 800


i (pA

)

–200

–100

0

100

200

–100 0 100 200 300 400 500

E versus Ag/AgCl (mV)i (

pA)

–200

–100

0

100

200

a

c

b

d

–1,000 0 1,000 2,000


i (pA

)

1

2

1.21.00.80.60.40.20.0–0.2–1.5–1.0–0.50.00.51.01.5

E (mV)

i (m

A)

Figure 4 | Effect of acid pre-treatment of glass-encased nanoelectrodes on their voltammetric responses. a–d, CVs obtained in 0.5 M H2SO4 solution

before (1) and after (2) an electrode was soaked overnight in 6 M HCl (a), and CVs of 20 mM Ru(NH3)6Cl3 in 1 M KNO3 (b), 15 mM K3IrCl6 in 0.2 M KCl

(c) and 2 mM FcCH2OH in 0.25 M KCl (d) at the glass-covered electrode after pre-treatment with 6 M HCl. n¼ 500 (a) and 50 (b–d) mV s21.

The inset in a shows a CV obtained in 0.5 M H2SO4 with a macroscopic platinum working electrode (5.47 cm2) and saturated calomel electrode reference;

n¼ 200 mV s21 (ref. 10).






to or smaller than that of the existing nanopipette-based potentio-metric sensors20. It may also be used as a scanning probe for pHmicroscopy21, where its small size can help to significantlyimprove spatial resolution. At the same time, glass nanosensorsare not likely to replace conventional pH electrodes in routineanalytical measurements because of their fragility andlimited lifetime.

The borosilicate glass used in this work may not be the bestmaterial for a pH sensor because its composition is different fromconventional pH glass (for example, Corning 0150 glass), and itwould not be suitable for the fabrication of macroscopic pHprobes. A relatively slow response (minutes) can probablybe improved by using more suitable glass for the preparation ofpH nanosensors.

MethodsChemicals. Ferrocenemethanol (FcCH2OH, 97%; Aldrich) was recrystallized twicefrom acetone. Hexaammineruthenium (III) chloride (99%) was obtained from StremChemicals. KNO3, Li2SO4 and KCl (99þ%, Aldrich) were used as supportingelectrolytes. Aqueous solutions were prepared from deionized water (Milli-Q,Millipore). Twice-distilled HPLC grade DCE (Sigma-Aldrich) was used to prepareorganic solutions. Potassium hexachloroiridate (III) (Alfa Aesar), H2SO4 (Aldrich)and CuSO4 (Mallinckrodt) were used as received.

Electrodes. To prepare a ‘dry’ glass-covered nanoelectrode, an annealed 25-mmplatinum wire (Goodfellow) was pulled into a borosilicate capillary (Drummond;OD, 1.0 mm; ID, 0.2 mm) under vacuum with the help of a Sutter P-2000/G laserpipette puller, as described previously9,22. To form hydrated gel, a glass-coveredplatinum nanoelectrode was soaked in 6 M HCl solution overnight. As electrodeswere removed from the HCl solution they were rinsed with water beforemeasurements. To improve reproducibility, such electrodes were stored in aqueoussolution between experiments. However, it was found that the hydrated gel remainson the electrode surface, even after being kept for several days in an oven at �100 8C.A two-electrode configuration was used for voltammetry and chronoamperometry,with either a commercial Ag/AgCl reference electrode or a Ag quasi-reference(DCE solutions).

Instrumentation and procedures. Cyclic voltammograms were obtained usingeither an EI-400 bipotentiostat (Ensman Instruments) or a BAS-100Belectrochemical analyser (Bioanalytical Systems). All experiments were carried out atroom temperature (23+2 8C) inside a Faraday cage. Unless otherwise specified, CVsobtained at glass-covered electrodes show the second or subsequent potential cycles,which are essentially indistinguishable from one another (steady-state response), butare different from the first potential sweep.

pH measurements were carried out in the following solutions: HCl/KCl (pH 1),phthalate buffers (pH 2–4), acetate buffers (pH 3–6), phosphate buffers (pH 7–8)and carbonate buffers (pH 9–10).

SEM images were obtained using a field-emission scanning electron microscope(Zeiss Supra 55 VP) with no conductive coating applied to the nanoelectrodes.


References1. Murray, R. W. Nanoelectrochemistry: metal nanoparticles, nanoelectrodes and

nanopores. Chem. Rev. 108, 2688–2720 (2008).2. Fan, F.-R. F. & Bard, A. J. Electrochemical detection of single molecules. Science

267, 871–874 (1995).3. Smith, C. P. & White, H. S. Theory of the voltammetric response of electrodes of

submicron dimensions. Violation of electroneutrality in the presence of excesssupporting electrolyte. Anal. Chem. 65, 3343–3353 (1993).

4. Sun, P. & Mirkin, M. V. Electrochemistry of individual molecules in zeptolitervolumes. J. Am. Chem. Soc. 130, 8241–8250 (2008).

5. Bach, H., Baucke, F. K. G. & Krause, D. (eds) Electrochemistry of Glasses andGlass Melts, Including Glass Electrodes (Springer, 2001).

6. Haugaard, G. The mechanism of the glass electrode. J. Phys. Chem. 45,148–157 (1941).

7. Schwabe, K. & Dahms, H. Permeability of the glass electrode to hydrogen ionswith the aid of tritium tagging. Monatsber. Deutschen Akad. Wissen. 1,279–282 (1959).

8. Vetter, K. J. Electrochemical Kinetics: Theoretical and Experimental Aspects(Academic Press, 1967).

9. Sun, P. & Mirkin, M. V. Kinetics of electron transfer reactions at nanoelectrodes.Anal. Chem. 78, 6526–6534 (2006).

10. Attard, G. S. et al. Mesoporous platinum films from lyotropic liquid crystallinephases. Science 278, 838–840 (1997).

11. Fletcher, S. et al. The response of some nucleation/growth processes totriangular scans of potential. J. Electroanal. Chem. 159, 267–285 (1983).

12. Shao, Y. & Mirkin, M. V. Fast kinetic measurements with nanometer-sizedpipets. Transfer of potassium ion from water into dichloroethane facilitated bydibenzo-18-crown-6. J. Am. Chem. Soc. 119, 8103–8104 (1997).

13. Masterton, W. L. & Gendrano, M. C. Henry’s Law studies of solutions of water inorganic solvents. J. Phys. Chem. 70, 2895–2898 (1966).

14. Harris, D. C. Quantitative Chemical Analysis 6th edn, 397(W. H. Freeman, 2002).

15. Hubbard, A. T. & Anson, F. C. The theory and practice of electrochemistry withthin layer cells, in Electroanalytical Chemistry (ed. Bard, A. J.) Vol. 4, 129–214(Marcel Dekker, 1970).

16. Vonau, W., Gabel, J. & Jahn, H. Potentiometric all solid-state pH glass sensors.Electrochim. Acta 50, 4981–4987 (2005).

17. Kreuer, K.-D. Solid potentiometric pH electrode. Sens. Actuat. B 1,286–292 (1990).

18. El-Giar, E. E.-D. M. & Wipf, D. O. Microparticle-based iridium oxideultramicroelectrodes for pH sensing and imaging. J. Electroanal. Chem. 609,147–154 (2007).

19. Malkaj, P., Dalas, E., Vitoratos, E. & Sakkopoulos, S. pH electrodes constructedfrom polyaniline/zeolite and polypyrrole/zeolite conductive blends. J. Appl.Polym. Sci. 101, 1853–1856 (2006).

20. Bakker, E. & Pretsch, E. Nanoscale potentiometry. Trends Anal. Chem. 27,612–618 (2008).

21. Horrocks, B. R. et al. Scanning electrochemical microscopy 19. Ion selectivepotentiostatic microscopy. Anal. Chem. 65, 1213–1224 (1993).

22. Shao, Y. et al. Nanometer-sized electrochemical sensors. Anal. Chem. 69,1627–1634 (1997).

AcknowledgementsThe authors gratefully acknowledge support from the National Science Foundation (CHE-0645958) and a grant from PSC-CUNY. The authors would like to thank H. Gafney, F.Laforge and A. Bard for helpful discussions and J. Morales (CCNY electron microscopyfacility) for his help with SEM imaging.

Author contributionsJ.V. performed the experiments. D.Z. conceived the experiments and developed analyticaltools for nanoelectrode characterization. M.V.M. conceived and designed the experiments,analysed data and wrote the paper.

Additional informationThe authors declare no competing financial interests. Reprints and permission information isavailable online at http://npg.nature.com/reprintsandpermissions/. Correspondence andrequests for materials should be addressed to M.V.M.

E = –51.861pH + 591.73R2

= 0.9948

0

200

400

600

0 2 4 6 8 10pH

E (m

V)

Figure 6 | Potentiometric response of a hydrated glass nanoelectrode to

solution pH. Potential was measured versus a Ag/AgCl reference electrode.

The glass nanoprobe was transferred sequentially between ten different

buffer solutions, and each measurement was taken after the potential

stabilized to within+1 mV.








Enhancement of anhydrous proton transport bysupramolecular nanochannels in comb polymersYangbin Chen1, Michael Thorn2, Scott Christensen3, Craig Versek2, Ambata Poe1, Ryan C. Hayward3*,

Mark T. Tuominen2* and S. Thayumanavan1*

Transporting protons is essential in several biological processes as well as in renewable energy devices, such as fuel cells.Although biological systems exhibit precise supramolecular organization of chemical functionalities on the nanoscale toeffect highly efficient proton conduction, to achieve similar organization in artificial systems remains a daunting challenge.Here, we are concerned with transporting protons on a micron scale under anhydrous conditions, that is proton transferunassisted by any solvent, especially water. We report that proton-conducting systems derived from facially amphiphilicpolymers that exhibit organized supramolecular assemblies show a dramatic enhancement in anhydrous conductivityrelative to analogous materials that lack the capacity for self-organization. We describe the design, synthesis andcharacterization of these macromolecules, and suggest that nanoscale organization of proton-conducting functionalities isa key consideration in obtaining efficient anhydrous proton transport.

Efficient and selective transport of protons is critical both inbiological contexts1 and in materials for renewable energy2.In biological systems, nature has optimized proton conduction

on a nanometre scale by using secondary and tertiary structures ofproteins to arrange precisely the appropriate side chains of aminoacids, for example in the membrane protein M2 (refs 3–5).Although control of proton transfer on this scale is adequate formost biological processes, it is essential that efficient proton conduc-tion be obtained on a micron scale for clean-energy applications6,7.

In hydrogen fuel cells, for example, after oxidation of molecularhydrogen at the anode, the resulting protons must be transportedacross a selective membrane to reach the cathode and completethe conversion of chemical energy into electrical energy. Theproton conductivity of this membrane, often called the proton-exchange membrane or the polymer electrolyte membrane (PEM),has been one of the bottlenecks to achieving affordable fuel-celltechnology. Nafion, a poly(tetrafluoroethylene)-based polymerwith sulfonic acid groups arranged at intervals along the backbone,is one of the most widely used materials for this membrane8. Thekey to proton transport in Nafion is thought to be nanochannelsof sulfonic acid groups, through which ‘hydrated’ protons canpass efficiently9–11. Although a good proton conductor for hydratedprotons, Nafion suffers from poor conductivity in unassisted protontransfer, that is Grotthuss or anhydrous proton transfer12,13, whichresults in low conductivities at temperatures above the boilingpoint of water. PEMs with high proton conductivities at tempera-tures of 120–200 8C are desirable, because operation at higher temp-eratures can increase fuel-cell efficiency, reduce cost, simplify heatmanagement and provide better tolerance of the catalysts againstpoisoning14.

One approach to address this issue is to use amphoteric func-tional groups that allow anhydrous proton transport15,16, forexample imidazole, which is a common motif in biological protontransport in the form of the amino acid histidine. Several groupshave studied synthetic polymers that contain such amphoteric func-tional groups as candidates for high-temperature proton transfer17–22.Although a number of interesting candidate materials were

identified, one avenue that was not explored in these anhydrousproton-conducting systems is the role of supramolecular organiz-ation in nanoscale ion-conducting channels. This is surprisingbecause, in the context of hydrated proton-conducting systems,such as Nafion9–11

, and sulfonated block copolymers12,23–26, as wellas lithium-ion conducting supramolecular assemblies27–29, it iswell-established that the formation of nanoscale domains enrichedin the ion-conducting materials is critical to the resulting macro-scopic ionic conductivity.

In this paper, we describe the molecular design and synthesis of anovel class of comb polymers with amphoteric proton-transferfunctionalities that can self-assemble into organized supramolecularstructures. We also show that very subtle changes in the monomerand analogous polymer provide solid-state structures that lacksuch nanoscale organization. By comparing these polymers, weshow that the self-assembled structures yield a dramatic increasein proton conductivity (by as much as three orders of magnitude),presumably because of the locally increased concentration ofproton-transport functionalities within the nanophase-separateddomains. These results suggest that a careful consideration ofmacromolecular architecture and nanoscale assembly is critical tooptimizing anhydrous proton transport in new materials for PEMs.

ResultsTo prepare polymers that form supramolecular assemblies withproton-transporting functionalities concentrated within nanoscaledomains, we made use of comb polymer architectures (Fig. 1).One of our groups recently used this architecture to prepare amphi-philic comb polymers by attaching lipophilic and hydrophilic func-tionalities at the meta-positions of the benzene ring of each styrenicrepeat unit30,31. Such polymers were shown to form assemblies of amicelle type in aqueous milieu and of an inverse-micelle type inapolar organic solvents. Thus, we hypothesized that similar poly-mers would also form nanoscale assemblies in the melt state. Forthis purpose, we designed a series of styrenic comb polymers inwhich one of the meta-positions contained a polar N-heterocyclicfunctionality capable of proton transport, and the other contained

1Department of Chemistry, 2Department of Physics, 3Department of Polymer Science and Engineering, University of Massachusetts, Amherst,Massachusetts 01003, USA. *e-mail: [email protected]; [email protected]; [email protected]









a non-polar alkyl chain to drive phase segregation. For example,polymer 1 (Fig. 1a) contains non-conducting decyl chains and theproton-conducting heterocyclic functionality benzotriazole.Although Liu et al.18 reported that triazole appreciably enhancesproton conductivities with respect to imidazole (also benzimidazolehas been studied20 as a proton-transporting functionality), we arenot aware of any reports of benzotriazole used as a protonconductor. To test our premise that nanoscale phase segregationof 1 would facilitate proton transport, we synthesized theanalogous polymer 2 that has no alkyl chains and therefore didnot undergo nanoscale assembly. Syntheses of these polymers areexemplified with polymer 1 in Fig. 1b.

To determine proton-conductivity values, polymer films weredrop-cast from solution onto a hole in a piece of Kapton tape andsubsequently sandwiched between two electrodes to allow character-ization by impedance spectroscopy, as described previously19. TheKapton tape determined the thickness of the polymer film, whichwas therefore constant at 125 mm for all impedance measurements.Separate thermogravimetric analyses (TGAs) were conducted toverify that all polymers reported in this study were thermallystable up to at least 200 8C, which was the highest temperatureinvestigated in the impedance measurements. Conductivities ofthe polymer samples were measured through several heating–cooling cycles (40–200 8C) under high vacuum and were found tobe consistent from cycle to cycle, which eliminates any possibleeffects of residual solvent on the performance of these polymers.

Proton conductivities for 1 and 2 measured as a function oftemperature between 40 and 200 8C are shown in Fig. 2a. Bothpolymers show qualitatively similar non-Arrhenius increases inconductivity with temperature that are typical for anhydrousproton-conducting polymers. However, the conductivity of 1 ranges

from 6 × 1026 S cm21 at ambient temperature to 1.3 × 1023 S cm21

at 200 8C, at least two orders of magnitude larger than the conduc-tivity of 2 across the same temperature range, which varies from4× 1029 S cm21 at ambient temperature to 1.2 × 1025 S cm21 at200 8C. As a benchmark, Nafion membranes show room-temperatureconductivities of 1022 to 1021 S cm21 when fully hydrated32, but atlow humidity (below 5%) their conductivities were reported as 1027

to 1025 S cm21 (ref. 33). (Under our experimental conditions,measured conductivities of Nafion were below the noise floor of themeasurements of �1029 S cm21.) Thus, although the conductivitiesof our materials remain significantly below those of Nafion underideal conditions, the dramatic increase in conductivity from 2 to 1suggests an important design principle for optimizing proton transportunder anhydrous conditions.

At first, it may seem surprising that the addition of a decyl chainto each repeat unit of a polymer could boost proton conductivity bytwo orders of magnitude. After all, the average density of proton-transporting groups is lowered by the presence of the decyl chain;the benzotriazole unit makes up only 23 weight per cent (wt%) of1 as compared to 34 wt% of 2. However, we hypothesized that thedecyl chain renders the mixing of 1 with the amphoteric hetero-cycles incompatible, and so drives 1 to self-assemble and formnanoscale domains that contain enhanced local concentrations ofbenzotriazole, and thereby facilitates proton transport.

To test this hypothesis, we characterized the structures ofthese polymers using small-angle X-ray scattering (SAXS). Asshown in Fig. 2b, polymer 1 gave rise to scattering peaks thatindicate self-assembled nanostructures, but the control polymer2 yielded a completely featureless pattern that indicates a homo-geneous phase-mixed structure. The first-order scattering peakfrom 1 falls at q*¼ 1.47 nm21, which corresponds to a real-space

O O

n

n-C10H21

n-C10H21 n-C10H21

n-C10H21n-C10H21

NH

O

N

N

HN

1

O

n

NH

O

N

N

HN

2

OH

OHHOi. K2CO3, 18-crown-6

NaI, n-C10H21Br

NaI, Br(CH2)6CO2Et CH3PPh3Brii. K2CO3, 18-crown-6

O O

OH

O

O

O OO

O

i. PCC

ii. KO-t-Bu

3 4 5

O

N125 °C

O O

n

O

O

6

i. Ester hydrolysis

ii. Et3NCl O

O

then

NN

HN

H2N

O O

n

NH

O

N

N

HN

1

a

b

Figure 1 | Structures and synthesis of benzotriazole-based polymers. a, Benzotriazole-based polymers that only differ by the inclusion of a decyl chain.

b, Synthetic scheme for benzotriazole-based polymers. This synthesis was modified slightly to prepare proton-conducting polymers without decyl chains and

those that contain imidazole in place of benzotriazole. Details are given in the Supplementary Information. PCC¼ pyridinium chlorochromate.






distance of d¼ 4.3 nm, and a faint, although clearly resolvable,second-order peak at

p4q*. Although the structure cannot be deter-

mined unambiguously from these data, the presence of only asecond-order peak suggests a lamellar structure with a repeatspacing of 4.3 nm. We estimate the fully stretched length of amonomer, from the tip of the decyl chain to the benzotriazolegroup, as �3 nm, and thus the observed repeat spacing is consistentwith ‘back-to-back’ stacking of polymer chains with some interdigi-tation of the decyl chains and/or benzotriazole group and spacer.A schematic of this proposed structure is shown in Fig. 2c, in whichtwo repeating units arranged with the benzotriazoles head-to-headallow hydrogen bonding, with the alkyl groups pointed away fromeach other. Analysis of the first-order peak revealed a width (full-width

at half-maximum, Dq) of 0.6 nm21 for polymer 1, which yielded acorrelation length of 2p/Dq ≈ 10 nm, indicating that the size ofordered domains is small, with positional correlations that extendonly over several repeat units.

As demonstrated by Ikkala and co-workers34, self-assembly intoanisotropic nanostructures yields orientation-dependent conduc-tivity, and McGrath and co-workers have shown that continuity ofnanoscale domains is critical to efficient proton transport in sulfo-nated polymers35. For our polymers, although the limited length ofordering precludes any considerations of the effects of orientation ordimensionality of nanoscale domains on conductivity, the nanoscaleorganization provided by supramolecular assembly clearly enhancesanhydrous proton conductivity by at least two orders of magnitude

2.0 2.2 2.4 2.6 2.8 3.0 3.210–9

10–2

10–3

10–4

10–5

10–6

10–7

10–8

227 181 144 111 84 60 39

Con

duct

ivity

σ (

S c

m–1

)

1,000/T (K–1)

a Temperature (°C)

1 2

1 2 3 4 50.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

2.2

2.4

b

I(q)

(a.

u.)

q (nm–1)

12

q* = 1.47 nm–1

4q*

Decyl chain

Spacer

Benzotriazole

c

N

N

HN

O

O

Polymer backbone

HN

N

N O

O

O4

O

45

5

Figure 2 | Conductivity and SAXS results for benzotriazole polymers. a, Proton conductivity over a wide temperature range, in which the polymer that

contains the decyl chain exhibits a much higher conductivity than that of its counterpart. b, SAXS profiles of benzotriazole polymers that indicate ordering of

the alkylated polymer. Curves are shifted vertically for clarity. I(q) is the azimuthally-averaged scattering intensity as a function of scattering wave-vector (q).

c, An illustration of the proposed structure of 1, with two units arranged with the benzotriazoles head-to-head, which thus allows for hydrogen bonding.

a.u.¼ arbitrary units.






compared with that of the phase-mixed control polymer. The exactmechanism of enhancement is currently unclear, although wespeculate that self-organization leads to interconnected channelsof locally enriched benzotriazole concentrations that facilitateproton hopping through the membrane. To test for the possibilityof disordering or nanoscale structural transitions at high tempera-ture (which, in some cases, have yielded dramatic changes in ionicconductivity27), we carried out variable-temperature SAXS measure-ments over the range 40–200 8C. Although the intensity of the

first-order scattering peak decreased continuously with increasingtemperature, its position and width remained nearly constant,which indicates that a similar level of nanoscale organization waspresent in these materials over the entire temperature range of inter-est (see Supplementary Information for details).

An additional factor to consider when the proton conductivitiesof two polymer chains that bear the same functional group are com-pared is the glass-transition temperature (Tg), because the mobilityof the polymer chain is well-known to influence the rate of protontransport36. To test whether the difference in conductivity observedbetween 1 and 2 simply reflects a decrease in Tg because the decylchain is present, we carried out differential scanning calorimetryexperiments. As summarized in Table 1, the Tg values of 1 and 2were 55 8C and 67 8C, respectively. The modest difference in Tgbetween these polymers suggests that the mobility of the polymerbackbone is not a major contributor to the difference in proton con-ductivities observed. As relatively high conductivity values of 1 wereachieved at temperatures well above Tg, the mechanical properties ofthis polymer at such temperatures are not well-suited for applicationas PEMs. Although components with higher Tg values or semicrys-talline components need to be incorporated to provide materials

Table 1 | Properties of polymers 1, 2, 7 and 8.

Polymer Decompositiononset (88888C) (5%weight loss)

Tg

(88888C)N-heterocycleweight fraction(%)

Molecularweight (Mn)

1 218 55 23 25,0002 233 67 34 25,0007 225 61 14 23,0008 221 71 20 24,000

Glass-transition temperatures, decomposition onset temperatures, N-heterocycle weight fractionsand molecular weight of polymers 1, 2, 7 and 8. Mn¼ number-average molecular weight.

d

a b

78

c78

q* = 1.64 nm–1

3q*

2.0 2.2 2.4 2.6 2.8 3.0 3.210–10

10–2

10–3

10–4

10–5

10–6

10–7

10–8

10–9

227 181 144 111 84 60 39

Con

duct

ivity

(

S c

m–1

)

1,000/T (K–1)

Temperature (°C)

1 2 3 4 50.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

I(q)

(a.

u.)

q (nm–1)

Decyl chain

Spacer

Imidazole

O O

n

n

n-C10H21 NH

ONH

N

ONH

ONH

N

7

8

Figure 3 | Results for imidazole-based polymers a, Structures of imidazole polymers that only differ by the inclusion of a decyl chain. b, Proton conductivity

over a wide temperature range, which again demonstrates that the polymer with an added decyl chain has a much higher conductivity. c, SAXS profiles of

imidazole polymers, which show order in polymer 7 and a disordered polymer 8. Curves are shifted vertically for clarity. d, An illustration of the proposed

structure of 7, which self-assembles into cylindrical domains.






with adequate mechanical properties, we emphasize that the poly-mers described here provide proof-of-concept results and demon-strate the importance of nanoscale organization in anhydrousproton conductivity.

If our hypothesis and conclusions are correct, this moleculardesign strategy should also work for other amphoteric heterocycles.To test the generality of our molecular design, we prepared polymersthat contained imidazole as a proton-transporter functionality,which was also shown to be capable of anhydrous proton trans-fer16–19 and has a very different dissociation constant (pKa) valuein the protonated form as compared to that of benzotriazole. Inanalogy to 1 and 2, we synthesized polymers 7 and 8 with imidazolemoieties (Fig. 3a). Alternating current impedance measurementsrevealed that polymer 7, with its decyl chain, exhibits dramaticallyhigher conductivity than that of the corresponding controlpolymer 8, in this case by more than three orders of magnitude(Fig. 3b). The morphologies of these polymers were investigatedusing SAXS and revealed that 7 gave two well-defined scatteringpeaks, which indicates the presence of self-assembled nanostruc-tures, but 8 showed no signs of structure (Fig. 3c). The second scat-tering peak for 7 falls at a position of

p3q*, which clearly indicates a

non-lamellar structure and suggests a hexagonal symmetry thatprobably corresponds to a structure of hexagonally packed cylinders(Fig. 3d). Tg values determined for these polymers (Table 1) are verysimilar, which once again reveals that mobility of the polymer back-bone is not a significant factor in the difference in proton conduc-tivity of three orders of magnitude. We also tested randomcopolymers synthesized from a monomer disubstituted with decylgroups and another monomer disubstituted with N-heterocycles(1:1 ratio). These random copolymers also provided some extentof phase separation, but with nanostructures organized morepoorly than those of the comb polymers and with only a single scat-tering peak for each. The conductivities of these random copoly-mers are generally significantly greater than those of theunorganized control homopolymers 2 and 8, although somewhatless than those of the comb polymers 1 and 7 (see SupplementaryInformation). This further supports our conclusion that phase sep-aration on a nanoscale is tied directly to the efficiency ofproton transport.

In summary, we have designed, synthesized and characterized anew class of comb polymers for anhydrous proton transport. Wehave shown that:

† styrenic comb polymers that contain incompatible functionalitiesat opposite faces of the monomer units provide ordered nano-structures through self-assembly in the melt state;

† styrenic polymers that contain a non-conducting decyl group anda proton-conducting functionality on the meta-positions of thephenyl ring exhibit conductivities two-to-three orders of magni-tude greater than those of polymers that contain only the con-ducting functionality, despite the lower overall content ofproton transporter in the former;

† polymer backbone mobility is not a major contributor to theobserved differences in proton conductivity in these systems;

† the high conductivities observed for the decyl-functionalized poly-mers correlate with the ability to form organized lamellar or hexa-gonal nanostructures that consist of domains with locally highconcentrations of proton conductors that facilitate transport;

† this molecular design strategy works for two different proton-transfer functionalities with substantially different pKa values,which suggests that the importance of nanochannel formationin proton conduction is a general phenomenon.

Our work here indicates that careful consideration of polymerarchitecture and nanoscale morphology is a key element in thedesign of efficient anhydrous PEMs.

MethodsTGA was carried out using a TA Instruments TGA 2950 thermogravimetric analyserwith a heating rate of 10 8C min21 from room temperature to 500 8C under nitrogen.

Glass-transition temperatures were obtained by differential scanning calorimetryusing a TA Instruments Dupont DSC 2910. Samples were analysed at a heating rateof 10 8C min21 from 0 8C to 150 8C under a flow of nitrogen (50 ml min21).

Electrochemical impedance data were obtained using a Solartron 1287potentiostat and 1252A frequency response analyser in the range 0.1 Hz to 300 kHz.Measurements were conducted under vacuum at temperatures between 40 8C and200 8C with a sinusoidal excitation root-mean-square voltage of 0.1 V. The samplethickness and contact surface area were controlled by a 125 mm thick Kapton tapewith a 0.3175 cm diameter hole.

SAXS measurements were carried out on an in-house beamline using a Rigakurotating anode source to generate Cu Ka radiation (wavelength l¼ 0.154 nm).Scattering patterns were collected on an image plate positioned a distance of 500 mmfrom the sample. All samples yielded isotropic patterns, and thus data wereintegrated to yield plots of intensity as a function of the magnitude of the scatteringvector, q¼ (4p/l)sin(u), where 2u is the total scattering angle. The actual scatteringangles were calibrated using the known reflection from silver behenate.

Received 16 November 2009; accepted 16 March 2010;published online 25 April 2010

References1. Williams, R. J. P. Proton circuits in biological energy interconversions. Annu.

Rev. Biophys. Biophys. Chem. 17, 71–97 (1988).2. Kreuer, K. D. Proton conductivity: materials and applications. Chem. Mater. 8,

610–641 (1996).3. Sass, H. J. et al. Structural alterations for proton translocation in the M state of

wild-type bacteriorhodopsin. Nature 406, 649–653 (2000).4. Schnell, J. R. & Chou, J. J. Structure and mechanism of the M2 proton channel of

influenza A virus. Nature 451, 591–560 (2008).5. Stouffer, A. L. et al. Structural basis for the function and inhibition of an

influenza virus proton channel. Nature 451, 596–600 (2008).6. Carrette, L., Friedrich, K. A. & Stimming, U. Fuel cells – fundamentals and

applications. Fuel Cells 1, 5–39 (2001).7. Steele, B. C. H. & Heinzel, A. Materials for fuel-cell technologies. Nature 414,

345–352 (2001).8. Mauritz, K. A. & Moore, R. B. State of understanding of Nafion, Chem. Rev. 104,

4535–4585 (2004).9. Diat, O. & Gebel, G. Proton channels. Nature Mater. 7, 13–14 (2008).10. Schmidt-Rohr, K. & Chen, Q. Parallel cylindrical water nanochannels in Nafion

fuel-cell membranes. Nature Mater. 7, 75–83 (2008).11. Elliott, J. A., Hanna, S., Elliott, A. M. S. & Cooley, G. E. Interpretation of the

small-angle X-ray scattering from swollen and oriented perfluorinated ionomermembranes. Macromolecules 33, 8708–8713 (2000).

12. Hickner, M. A., Ghassemi, H., Kim, Y. S., Einsla, B. R. & McGrath, J. E.Alternative polymer systems for proton exchange membranes (PEMs). Chem.Rev. 104, 4587–4612 (2004).

13. Rikukawa, M. & Sanui, K. Proton-conducting polymer electrolyte membranesbased on hydrocarbon polymers. Prog. Polym. Sci. 25, 1463–1502 (2000).

14. Li, Q., He, R., Jensen, J. O. & Bjerrum, N. J. Approaches and recent developmentof polymer electrolyte membranes for fuel cells operating above 100 8C. Chem.Mater. 15, 4896–4915 (2003).

15. Kreuer, K. D. Fast proton conductivity: a phenomenon between the solid and theliquid state? Solid State Ionics 94, 55–62 (1997).

16. Kreuer, K. D., Fuchs, A., Ise, M., Spaeth, M. & Maier, J. Imidazole and pyrazole-based proton conducting polymers and liquids. Electrochim. Acta 43,1281–1288 (1998).

17. Scharfenberger, G. et al. Anhydrous polymeric proton conductors based onimidazole functionalized polysiloxane. Fuel Cells 6, 237–250 (2006).

18. Zhou, Z., Li, S. W., Zhang, Y. L., Liu, M. L. & Li, W. Promotion of protonconduction in polymer electrolyte membranes by 1H-1,2,3-triazole. J. Am.Chem. Soc. 127, 10824–10825 (2005).

19. Granados-Focil, S., Woudenberg, R. C., Yavuzcetin, O., Tuominen, M. T. &Coughlin, E. B. Water-free proton-conducting polysiloxanes: a study on theeffect of heterocycle structure. Macromolecules 40, 8708–8713 (2007).

20. Persson, J. C. & Jannasch, P. Intrinsically proton-conducting benzimidazoleunits tethered to polysiloxanes. Macromolecules 38, 3283–3289 (2005).

21. Shogbon, C. B., Brousseau, J.-L., Zhang, H., Benicewicz, B. C. & Akpalu, Y.Determination of the molecular parameters and studies of the chain conformationof polybenzimidazole in DMAc/LiCl. Macromolecules 39, 9409–9418 (2006).

22. Subbaraman, R., Ghassemi, H. & Zawodzinski, T. A. Jr 4,5-Dicyano-1H-[1,2,3]-triazole as a proton transport facilitator for polymer electrolyte membrane fuelcells. J. Am. Chem. Soc. 129, 2238–2239 (2007).

23. Won, J. et al. Fixation of nanosized proton transport channels in membranes.Macromolecules 36, 3228–3234 (2003).

24. Serpico, J. M. et al. Transport and structural studies of sulfonated styrene–ethylene copolymer membranes. Macromolecules 35, 5916–5921 (2002).






25. Shi, Z. & Holdcroft, S. Synthesis and proton conductivity of partially sulfonatedpoly([vinylidene difluoride-co-hexafluoropropylene]-b-styrene) blockcopolymers. Macromolecules 38, 4193–4201 (2005).

26. Rubatat, L., Shi, Z., Diat, O., Holdcroft, S. & Frisken, B. J. Structural study ofproton-conducting fluorous block copolymer membranes. Macromolecules 39,720–730 (2006).

27. Cho, B. K., Jain, A., Gruner, S. M. & Wiesner, U. Mesophase structure –mechanical and ionic transport correlations in extended amphiphilic dendrons.Science 305, 1598–1601 (2004).

28. Kishimoto, K. et al. Nano-segregated polymeric film exhibiting high ionicconductivities. J. Am. Chem. Soc. 127, 15618–15623 (2005).

29. Wanakule, N. S. et al. Ionic conductivity of block copolymer electrolytes in thevicinity of order–disorder and order–order transitions. Macromolecules 42,5642–5651 (2009).

30. Savariar, E. N., Aathimankandan, S. V. & Thayumanavan, S. Supramolecularassemblies from amphiphilic homopolymers: testing the scope. J. Am. Chem.Soc. 128, 16224–16230 (2006).

31. Basu, S., Vutukuri, D. R. & Thayumanavan, S. Homopolymer micelles inheterogeneous solvent mixtures. J. Am. Chem. Soc. 127, 16794–17695 (2005).

32. Wintersgill, M. C. & Fontanella, J. J. Complex impedance measurements onNafion. Electrochim. Acta 43, 1533–1538 (1998).

33. Sanders, E. H. et al. Characterization of electrosprayed Nafion films. J. PowerSources 129, 55–61 (2004).

34. Ruotsalainen, T. et al. Structural hierarchy in flow-aligned hexagonally self-organized microphases with parallel polyelectrolytic structures. Macromolecules36, 9437–9442 (2003).

35. Roy, A. et al. Influence of chemical composition and sequence length on thetransport properties of proton exchange membranes. J. Polym. Sci. B 44,2226–2239 (2006).

36. Schuster, M. F. H. & Meyer, W. H. Anhydrous proton-conducting polymers.Annu. Rev. Mater. Res. 33, 233–261 (2003).

AcknowledgementsThis work was supported by the National Science Foundation through the Fueling theFuture Center for Chemical Innovation at the University of Massachusetts Amherst(CHE-0739227). We thank W. de Jeu for discussions on the X-ray scattering results.

Author contributionsS.T. and Y.C. conceived the molecular design. S.T., R.H. and Mark T. planned the project.Y.C, Michael T., S.C. and C.V. carried out the experiments and analysed the data. Y.C. andA.P. synthesized the discussed compounds, Michael T. and C.V. measured ionicconductivities, and S.C. performed SAXS. Results were discussed by R.H., Mark T. and S.T.All authors contributed to writing the manuscript.

Additional informationThe authors declare no competing financial interests. Supplementary information andchemical compound information accompany this paper at www.nature.com/naturechemistry. Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/. Correspondence and requests for materials should be addressedto R.C.H., M.T.T. and S.T.















in your element

Although it has been known for almost two centuries, lithium is suddenly making the news: it is the primary

ingredient of the lithium-ion batteries set to power the next generation of electric vehicles and, as such, could become as precious as gold in this century1. It is also non-uniformly spread within the Earth’s crust, sparking rumours that Andean South American countries could soon be the ‘new Middle-East’. Together, these factors set the scene for controversial debates about the available reserves2–4 and the anticipated demands1: if all cars are to become electric within 50 years, fears of a crunch in lithium resources — and thus a staggering price increase such as that faced today with fossil fuels — are permeating.

With its atomic number of 3, lithium is located in the top left corner of the periodic table. It was Johann August Arfvedson, one of Jöns Jakob Berzelius’s students, who first detected its presence in 1817 while analysing the mineral petalite (LiAlSi4O10), itself discovered in 1800. Berzelius called this new element lithos (Greek word for stone).

Lithium, whose silvery-white colour tarnishes on oxidation when exposed to air, is the most electropositive metal (−3.04 V versus a standard hydrogen electode), the lightest (M = 6.94 g mol–1) and the least dense (ρ = 0.53 g cm–3) solid element at room temperature, and is also highly flammable. Owing to this high reactivity, lithium is present only in compounds in nature — either in brines or hard rock minerals — and must be stored under anhydrous atmospheres, in mineral oil or sealed evacuated ampoules.

Their particular physical, chemical and electrochemical properties make lithium and its compounds attractive to many fields. Apart from the recent advent of lithium-based batteries, lithium niobate (LiNbO3) is an important material in nonlinear optics. Engineers use lithium in high-temperature lubricants, to strengthen alloys, and for heat-transfer applications. It is also

widespread in the fine chemical industry, as organo-lithium reagents are extremely powerful bases and nucleophiles used to synthesize many chemicals. Its effect on the nervous system has also made lithium attractive as a mood-stabilizing drug, and in nuclear research tritium (3H) is obtained by irradiating 6Li. Annual demand has therefore grown by 7–10%, currently reaching about 160,000 tons of lithium carbonate (Li2CO3) per year — about 20–25% of which is for the battery sector.

Energy storage, which should help mitigate the issues of pollution, global warming and fossil-fuel shortage, is becoming more important than ever, and Li-ion batteries are now the technology of choice to develop renewable energy technology and electric vehicles. They typically consist of a Li-containing positive electrode and a Li-free negative electrode, separated by a Li-based electrolyte. From simple calculations, assuming a one-molar Li-based electrolyte and a 3.6 V LiMPO4 electrode (where M is Fe or Mn), the demand is estimated to be about 0.8 kg Li2CO3 per kWh — and this number is not expected to decrease with recently developed batteries such as lithium–air or lithium–sulfur, which need an excess of lithium at the negative electrode to function properly. The fact that tritium might also be used with deuterium for nuclear fusion could increase demands.

Extracting lithium from hard rocks is laborious and expensive, however, and most of that produced (roughly 83%) at present comes

from brine lakes and salt pans: salty water is first pumped out of the lake into a series of shallow ponds, then concentrated using solar energy into a lithium chloride brine, which is subsequently treated with soda to precipitate Li2CO3. Considerable amounts of lithium are present in sea water, but its recovery is trickier, and highly expensive.

It is extremely difficult to estimate the world’s lithium reserves1–3 — a debate typically fed by investors and venture capitalists. The present production of Li2CO3 is about half what would be needed to convert the 50 million cars4 produced every year into ‘plug-in hybrid electric vehicles’ (with an electric motor powered by a 7 kWh Li-ion battery and a combustion engine). The demand becomes astronomic if we consider full electric vehicles — which require an on-board battery of 40 kWh. These numbers bring fears of a potential Li shortage in a few decades, painting a dim picture.

This alarming global situation will hopefully drive researchers to investigate new battery technologies5 and loosen our dependence on lithium. Fortunately, the situation improves if one also considers recycling — the low melting point (180 °C) of lithium metal and the very low water solubility of its fluoride, carbonate and phosphate salts make its recovery quite easy. Combining further brine exploitation with an efficient recycling process should be enough to match the demands of a ‘propulsion revolution’ that would solely rely on Li-ion cells, lessening geopolitical risks. ❐

JEAN-MAriE TArAscON is at the Laboratory of reactivity and solid-state chemistry, University of Picardie Jules Verne, F-80039 Amiens, France. e-mail: [email protected]

References1. Greene, L. Batteries & Energy Storage Technology 37–41

(Spring issue, 2009)2. Tahil, W. The Trouble With Lithium (Meridian International

Research, 2006); http://go.nature.com/jhDqLH3. Tahil, W. The Trouble With Lithium2 (Meridian International

Research, 2008); http://go.nature.com/AWITRo4. http://www.worldometers.info/cars/5. Armand, M. & Tarascon, J. M. Nature 451, 652–657 (2008).

Is lithium the new gold?Jean-marie tarascon ponders on the value of lithium, an element known for about 200 years, whose importance is now fast increasing in view of the promises it holds for energy storage and electric cars.

HUuo He Be B C N O F Ne Na Mg Al Si P S Cl Ar K CaLi

nchem_.680_JUN10.indd 510 7/5/10 11:54:02



http://www.worldometers.info/cars/

Documents

Nature Chemistry Volume 02 No 06 Pp425-510