LETTERS

Concurrent nucleation of 16S folding and induced fit in30S ribosome assemblyTadepalli Adilakshmi1{, Deepti L. Bellur2{ & Sarah A. Woodson1

Rapidly growing cells produce thousands of new ribosomes eachminute, in a tightly regulated process that is essential to cellgrowth1,2. How the Escherichia coli 16S ribosomal RNA and the20 proteins that make up the 30S ribosomal subunit can assemblecorrectly in a few minutes remains a challenging problem, partlybecause of the lack of real-time data on the earliest stages of assem-bly. By providing snapshots of individual RNA and protein inter-actions as they emerge in real time, here we show that 30S assemblynucleates concurrently from different points along the rRNA.Time-resolved hydroxyl radical footprinting3 was used to mapchanges in the structure of the rRNA within 20 milliseconds afterthe addition of total 30S proteins. Helical junctions in eachdomain fold within 100 ms. In contrast, interactions surroundingthe decoding site and between the 59, the central and the 39domains require 2–200 seconds to form. Unexpectedly, nucleo-tides contacted by the same protein are protected at different rates,indicating that initial RNA–protein encounter complexes refoldduring assembly. Although early steps in assembly are linked tointrinsically stable rRNA structure, later steps correspond toregions of induced fit between the proteins and the rRNA.

It has been previously shown that hierarchical addition of ribo-somal proteins to the 16S rRNA produces cooperativity4 that is due toprotein induced structural changes in the 16S rRNA, rather than todirect contacts between proteins5. Because the rRNA becomes morestructured as proteins join the complex, assembly is coupled to thefolding pathway of the rRNA6,7. The simplest kinetic model for 30Sassembly is that regions of the 16S rRNA contacted by primaryassembly proteins fold first, whereas helices stabilized by tertiaryassembly proteins fold last. If assembly is strictly sequential, eachsubdomain of the rRNA will fold within a distinct time, producinga limited set of intermediate complexes.

Alternatively, the rate of protein binding may initially depend on thestochastic probability of forming locally stable rRNA and protein inter-actions, with progression to later intermediates depending on propaga-tion of this conformational order to neighbouring regions in the rRNA.If more than one region of the naked rRNA can fold, assembly isexpected to nucleate from many places at once, producing an ensembleof reconstitution intermediates and multi-stage assembly kinetics8.

To visualize the intermediates of 30S ribosome assembly, the struc-ture of the 16S rRNA was probed by time-resolved X-ray hydroxylradical footprinting (Fig. 1). The extent of RNA cleavage correlateswith backbone exposure9. Thus, this method probes individual ter-tiary contacts in the rRNA as well as protein interactions that bury therRNA backbone. Previous efforts to map the conformational changesin the 16S rRNA during 30S ribosome assembly used low temper-ature or subsets of proteins to stall assembly at specific stages10,11. Wetook advantage of the exceptional time resolution of synchrotron

X-ray footprinting (,10 ms) to resolve the very early stages of assem-bly in real time, without needing to stall the reaction.

At the start of the experiment, native 16S rRNA from E. coli wasprefolded at 42 uC in standard reconstitution buffer (see Methods;Fig. 1a) because this pretreatment improved the quality of recon-stitution (Supplementary Fig. 2). Assembly of 30S ribosomal sub-units was initiated at 30 uC by mixing the rRNA with native total 30Sproteins (TP30) using a rapid quench apparatus (see SupplementaryMethods). The rRNA–protein mixture was exposed to a white syn-chrotron X-ray beam for 10 ms to generate hydroxyl radicals and tocleave the exposed parts of the rRNA backbone3. The extent of cleav-age at each position in the 16S rRNA was analysed by primer exten-sion and compared to parallel reactions on naked 16S rRNA andnative 30S ribosomes (Fig. 1b).

Many regions of the prefolded 16S rRNA were strongly protectedfrom cleavage, indicating that the rRNA has native tertiary structurein the absence of proteins. This is consistent with chemical footprint-ing experiments5,10,12 and with neutron scattering data showing thatthe deproteinized 16S rRNA has roughly the same dimensions as itdoes in the 30S ribosome13.

After a 3 min incubation with 30S proteins, the cleavage patternwas similar to that of native 30S ribosomes (Fig. 1b), and correlatedwell with the solvent accessibility of the C49 atoms predicted fromcrystal structures of the Thermus thermophilus 30S (ref. 14) or E. coli70S ribosomes15. Reconstitution of 30S subunits was confirmed bysedimentation velocity and activity assays (Supplementary Fig. 2).Further rearrangements after 3 min may be needed to produce fullyactive subunits, as the temperature-dependent transition from the RIreconstitution intermediate to the RI* activated reconstitution inter-mediate late in assembly is slow at 30 uC (ref. 16), and is associatedwith changes in the accessibility of the nucleotide bases rather thanthe backbone11. Nonetheless, most of the expected RNA and RNA–protein contacts formed within 3 min.

The kinetics of rRNA backbone protection showed the presence ofmany early assembly intermediates. Some nucleotides in the 16SrRNA were completely protected from hydroxyl radical cleavagewithin the 20 ms dead time of our experiment, whereas othersrequired 1–3 min to be fully protected (Fig. 1c and SupplementaryTable 1). For many nucleotides, a partial burst of protection in thefirst 50–100 ms was followed by slower saturation of the contact overthe next few minutes (Fig. 1c and Supplementary Fig. 3).

The multiphasic folding kinetics strongly indicate that individual30S complexes take different routes to the final structure, as sug-gested by previously reported rates of protein binding17 and by thekinetic partitioning of simple ribozymes among parallel folding path-ways18. In kinetic footprinting experiments on the Tetrahymena ribo-zyme, partial protection of the RNA backbone at intermediate times

1T.C. Jenkins Department of Biophysics, and 2Program in Cell, Molecular and Developmental Biology and Biophysics, Johns Hopkins University, 3400 North Charles Street, Baltimore,Maryland 21218-2685, USA. {Present address: Weis Center for Research, Geisinger Medical Center, 100 North Academy Avenue, Danville, Pennsylvania 17822, USA (T.A.);Department of Molecular Genetics and Cell Biology, The University of Chicago, 920 East 58th Street, Chicago, Illinois 60637, USA (D.L.B).

Vol 455 | 30 October 2008 | doi:10.1038/nature07298

1268

©2008 Macmillan Publishers Limited. All rights reserved

a b

c

16S rRNA

Total 30S protein (TP30)

1.0nt 398

nt 263

nt 617–618

Time (s)

0.8

0.6Y

0.4

0.2

00.01 0.1 1 10 100

330 mM KCI20 mM MgCI2

42 °C

Prefold Assemble Cleave

30 °C

X-rays

20 ms

TP30Time (s)

30S UCAG

1,170–1,171

1,179–1,181

1,185–1,194

1,198–1,200

1,202–1,203

1,215–1,220

1,222

1,224–1,228

1,230–1,234

1,236–1,239

–RN

A0.

02

180

0–180 srRNA

fragments

15 min

(S13, S19)

(S3, S5, S10, S14)

(S9)

(S14)

(S14)

(S19)

(S9)

(S7)

Figure 1 | Time-resolved X-ray footprinting of 30S ribosome assembly.a, Native 16S rRNA and total protein from 30S subunits (TP30) were mixedwithin 5–10 ms and irradiated with a synchrotron X-ray beam to cleave theRNA at exposed riboses. b, 16S fragments were analysed by primerextension. The cleavage pattern shown was from 0.02 to 180 s after TP30addition. 30S, triplicate controls on native 30S subunits; RNA, prefolded 16SrRNA; UCAG, dideoxy sequence ladders; 2, untreated RNA. The primer

annealed after nucleotide 1,257. c, The relative saturation (Y) of eachprotection versus assembly time, fitted to single or double exponential rateequations (Supplementary Methods). Filled circles, nucleotide (nt) 398;filled squares, nucleotides 617–618; filled triangles, nucleotide 263; opencircle and plus symbol, the average of 30S controls. Further data are shown inSupplementary Fig. 3 and Supplementary Table 1.

RNA–RNA contact:

RNA–protein contact:

700

a b

650

600

450

9501,250

1,000

1,050

1,150

1,100

300

200

750

550

350

150

100

5′

3′

10 1,400

1,450

50

1,300

Rate constant

>– 20 s–1

2–20 s–1

0.2–2 s–1

0.01–0.2 s–1

ND

900

800

850

250

Figure 2 | Simultaneous folding of 16S domains. a, Protected nucleotides(740 positions) were clustered according to the rate constant for backboneprotection and coloured as indicated in the key. Where the amplitude of theinitial burst phase is #60%, the slower rate constant is used. Grey (120positions) indicates the rate constant was undetermined (ND) owing to apause in reverse transcription or to weak protection. Natural adenine

methylation hampered quantitative analysis of residues in the decoding site.Circles, RNA–RNA contact; squares, RNA–protein contact; open circles,predicted C49 accessible surface area . 4 A2; solid symbols, buried C49 incrystal structures. b, Three-dimensional ribbon of E. coli 16S rRNA (PDBaccession 2avy)15 coloured as in a, viewed from the 50S interface. SeeSupplementary Fig. 4 for further views of each domain.

NATURE | Vol 455 | 30 October 2008 LETTERS

1269

©2008 Macmillan Publishers Limited. All rights reserved

was explained by the contemporaneous formation of folding inter-mediates with different structures19. We observed no exposure of 16Sresidues that might indicate the disappearance of non-native assem-bly intermediates over time. However, such interactions may havebeen too dynamic or heterogeneous to produce a distinct footprint.

To visualize the assembly pathway of the rRNA, the observed rateconstants for making individual backbone contacts were clustered,and the clusters were then projected onto the secondary and tertiarystructure of the mature 16S rRNA (Fig. 2). This locates the interactionsformed at each stage, even though the intermediate structures maydiffer from the mature structure. Nucleotides with similar rates ofbackbone protection did not map to single domains, but were distri-buted throughout the 16S rRNA. In general, nucleotides that wereprotected in the first 20–50 ms (red; Fig. 2) were partially folded inthe naked rRNA or were protected by a local structure such as a kink ora helical junction. Helices contacted by the primary assembly proteinsS4 (59 domain), S7 (39 domain), S8 and S15 (central domain) were alsoprotected within 50 to 100 ms after the proteins and rRNA were mixed.Thus, each domain assembles independently and simultaneously.

In contrast, the messenger RNA decoding site and long-rangeinteractions between domains required the longest time to form(blue; Fig. 2). For example, 16S helix H21 extends from the centraldomain to wrap around the body of the 30S subunit, where it con-tacts proteins S4 and S16. Interactions with H21 required 1 min tosaturate completely (0.02–0.1 s21). The slow appearance of rRNAbackbone contacts at G530 (H18), and the central pseudoknot(H2) in the decoding site, is consistent with previous equilibriumstudies demonstrating that structural changes in the central pseudo-knot are linked to the transition from the RI to RI* intermediates latein 30S assembly11.

The interplay between rRNA interactions that are intrinsicallystable and those that are protein-dependent is illustrated by the kin-etics of RNA and RNA–protein interactions in the body of the 30Sribosome (Fig. 3a). H44 (39 minor domain) lies along one face of thebody, packing against the stable tertiary structure formed by H7–H10and H14 (ref. 20). Interactions with the distal tip of H44 are mediatedby protein S20, which inserts its carboxy-terminal a-helix behind H8and H14 in the 59 domain21. Many RNA tertiary interactions in thebody of the 30S ribosome form within 20 ms (red; Fig. 3a), whereasinteractions with S20 form in 0.5 to 3 s (green), and long-range con-tacts with H44 form in ,10 s (blue; Fig. 3a). The multi-stage assem-bly of these RNA and protein interactions is consistent with theresults of hydroxyl radical probing from Fe-EDTA complexes teth-ered to S20 (ref. 22).

We next addressed whether the rates of RNA–protein interactionscorrelated with the position of each protein in the assembly map.Protections arising from direct protein–rRNA contacts were iden-tified from crystallographic structures of the 30S ribosome21 andprevious footprinting of individual proteins23 (see SupplementaryMethods). Four of the six primary assembly proteins (S4, S7, S8 andS15) protected a segment of their binding site during the first 50 msof assembly. This was only true of three of the nine secon-dary assembly proteins (S16, S9 and S10) and none of the tertiaryassembly proteins (Supplementary Table 2). Thus, primary assemblyproteins more frequently bind to the rRNA early in assembly, inagreement with pulse-chase measurements of protein bindingkinetics17.

Unexpectedly, nucleotides contacted by a single protein were pro-tected from hydroxyl radical cleavage with different rate constants(Fig. 4). For example, protein S4 binds a five helical junction in the 59domain, and initiates 30S assembly together with protein S7 (ref. 24).The interactions between S4 and 16S H17 (nucleotides 436–441)saturated in 20–50 ms, whereas those in H3, H16 and H18 formedin 0.3–2 s, and those with H21 (nucleotide 620) appeared over 8–10 s(Fig. 4a). S7 binds a complex helical junction in the 16S 39 domain(Fig. 4b)21. H43 (nucleotides 1,369–1,377) was protected in 20–50 ms, whereas contacts with H29, H41 and the distant H37 were

protected more slowly. Thus, S4 and S7 engage their binding sites instages rather than in one step. Similar behaviour was observed forproteins S8, S9, S10, S15 and S16 (Supplementary Figs 4 and 5).

The unexpectedly complex kinetics of the rRNA–protein interac-tions can be explained by the formation of encounter complexes thatslowly reorganize into the final complex. For S4, the variable kineticsof rRNA backbone protection may be due to co-folding of the proteinand the rRNA25. H17, which is protected rapidly, contacts the well-folded C-terminal domain of S4 (ref. 21). In contrast, H18 and thetop of H16, which are protected at an intermediate rate, contact theS4 amino-terminal domain, which is disordered in solution26.

If different regions of the 16S rRNA fold and interact with theribosomal proteins simultaneously, the implications for the hier-archy of protein association represented by the assembly map remaina challenge. The association rates suggested by the fastest rates ofRNA backbone protection approach the diffusion-controlled limit(108 M21 s21). However, the slowest protection rates are mostsimilar to the protein binding rates measured by pulse-chase mass

a

b

H14

H44

H8

H41

H39

S10

H9

S20

S7

S17

S9

Figure 3 | Stepwise assembly of RNA and protein interactions. a, ProteinS20 (yellow ribbon) contacts the 30S body in the 59 domain (grey) earlierthan helix H44 in the 39 minor domain (pink). 16S nucleotides are colouredas in Fig. 2. b, Proteins S7 (yellow) and S9 (green) protect a segment of theirbinding site immediately (red), whereas nucleotides at the interface betweenthe subdomains are protected slowly (blue).

LETTERS NATURE | Vol 455 | 30 October 2008

1270

©2008 Macmillan Publishers Limited. All rights reserved

spectrometry17 (Supplementary Table 2). Our data indicate that slowforming interactions in the rRNA, many of which are indirectly sta-bilized by bound proteins, determine the rate of protein addition andthe hierarchy of assembly.

This idea is illustrated by interactions of protein S7, which is neces-sary for stable binding of other proteins to the 16S 39 domain24,27.Residues indirectly protected by S7 (ref. 23), which overlap the bind-ing sites of the secondary assembly proteins S9, S13 and S19, foldslowly (Supplementary Fig. 5), explaining why these proteins join30S complexes more slowly than S7 (ref. 17). However, residues inH34 that directly contact S9 and S10 are fully protected in 50 ms(Fig. 3b and Supplementary Fig. 5). Thus, S7, S9 and S10 all engagetheir binding sites in the rRNA early in assembly, but productiveincorporation of S9 and S10 presumably waits for conformationalchanges in the 16S rRNA induced by S7. An important question iswhether early stochastic interactions between the ribosomal proteinsand the rRNA backbone bias the ensemble of rRNA conformationstowards the native state, thus accelerating self-assembly, or whetherthese interactions delay the search for the native conformation.

Stable structure in the 16S rRNA allows the concurrent nucleationof assembly from many points along the rRNA, resulting in the see-mingly chaotic but rapid appearance of native interactions through-out the complex. Previous studies indicated that ribosome assembly isnot completely cooperative28, demonstrating the need for several nu-cleation sites2. The lack of complete cooperativity, and the differences

between the time-dependence of 16S folding and the assembly maprevealed by kinetic footprinting of the nucleotide bases10, support theconclusion that assembly proceeds in parallel through intermediateswith different subsets of proteins6,8. In vivo, assembly of the pre-rRNAduring transcription is more cooperative and involves specific access-ory factors6,8. Co-transcriptional assembly may simplify the pathwayby limiting the number of intermediates that can be populated.Nonetheless, we expect stochastic fluctuations among alternativeintermediates to contribute to the early pathway of ribosome assemblyin vivo.

METHODS SUMMARY

Native E. coli 16S rRNA was prefolded in 80 mM K-cacodylate, pH 7.5, 330 mM

KCl, 20 mM MgCl2 for 15 min at 42 uC, then mixed with an equal volume of E.

coli TP30 in the same buffer at 30 uC, using a Kin-Tek rapid quench apparatusfitted with an X-ray flow cell. Between 20% and 40% of the input RNA was

cleaved after 10 or 20 ms exposure to a white light synchrotron beam (X28C,

National Synchrotron Light Source (NSLS), Brookhaven National Laboratory).

The cleaved RNA was analysed by primer extension with reverse transcriptase.

The increase in relative backbone protection (Y) after the addition of TP30 was

fitted to rate equations to obtain the observed rate constants and amplitudes for

the reaction (see Supplementary Fig. 3 and Table 1). Individual protections were

assigned to specific RNA or RNA–protein interactions on the basis of crystal-

lographic structures (Protein Data Bank accessions 1j5e (ref. 14), 2avw and 2avy

(ref. 15)); see Supplementary Methods for details.

Received 26 March; accepted 30 July 2008.Published online 10 September 2008.

1. Warner, J. R., Vilardell, J. & Sohn, J. H. Economics of ribosome biosynthesis. ColdSpring Harb. Symp. Quant. Biol. 66, 567–574 (2001).

2. Nierhaus, K. H. The assembly of prokaryotic ribosomes. Biochimie 73, 739–755(1991).

3. Ralston, C. Y. et al. Time-resolved synchrotron X-ray footprinting and itsapplication to RNA folding. Methods Enzymol. 317, 353–368 (2000).

4. Held, W. A., Mizushima, S. & Nomura, M. Reconstitution of Escherichia coli 30 Sribosomal subunits from purified molecular components. J. Biol. Chem. 248,5720–5730 (1973).

5. Stern, S., Powers, T., Changchien, L. M. & Noller, H. F. RNA–protein interactions in30S ribosomal subunits: folding and function of 16S rRNA. Science 244, 783–790(1989).

6. Culver, G. M. Assembly of the 30S ribosomal subunit. Biopolymers 68, 234–249(2003).

7. Williamson, J. R. After the ribosome structures: how are the subunits assembled?RNA 9, 165–167 (2003).

8. Noller, H. F. & Nomura, M. in Escherichia Coli and Salmonella Typhimurium, Cellularand Molecular Biology (ed. Neidhardt, F. C.) 104–125 (American Society forMicrobiology, 1987).

9. Tullius, T. D. & Greenbaum, J. A. Mapping nucleic acid structure by hydroxylradical cleavage. Curr. Opin. Chem. Biol. 9, 127–134 (2005).

10. Powers, T., Daubresse, G. & Noller, H. F. Dynamics of in vitro assembly of 16 SrRNA into 30 S ribosomal subunits. J. Mol. Biol. 232, 362–374 (1993).

11. Holmes, K. L. & Culver, G. M. Mapping structural differences between 30S ribosomalsubunit assembly intermediates. Nature Struct. Mol. Biol. 11, 179–186 (2004).

12. Adilakshmi, T., Ramaswamy, P. & Woodson, S. A. Protein-independent foldingpathway of the 16S rRNA 59 domain. J. Mol. Biol. 351, 508–519 (2005).

13. Ramakrishnan, V. Distribution of protein and RNA in the 30S ribosomal subunit.Science 231, 1562–1564 (1986).

14. Wimberly, B. T. et al. Structure of the 30S ribosomal subunit. Nature 407,327–339 (2000).

15. Schuwirth, B. S. et al. Structures of the bacterial ribosome at 3.5 A resolution.Science 310, 827–834 (2005).

16. Traub, P. & Nomura, M. Structure and function of Escherichia coli ribosomes. VI.Mechanism of assembly of 30 s ribosomes studied in vitro. J. Mol. Biol. 40,391–413 (1969).

17. Talkington, M. W., Siuzdak, G. & Williamson, J. R. An assembly landscape for the30S ribosomal subunit. Nature 438, 628–632 (2005).

18. Pan, J., Thirumalai, D. & Woodson, S. A. Folding of RNA involves parallelpathways. J. Mol. Biol. 273, 7–13 (1997).

19. Laederach, A., Shcherbakova, I., Liang, M. P., Brenowitz, M. & Altman, R. B. Localkinetic measures of macromolecular structure reveal partitioning among multipleparallel pathways from the earliest steps in the folding of a large RNA molecule. J.Mol. Biol. 358, 1179–1190 (2006).

20. Cate, J. H., Yusupov, M. M., Yusupova, G. Z., Earnest, T. N. & Noller, H. F. X-raycrystal structures of 70S ribosome functional complexes. Science 285,2095–2104 (1999).

21. Brodersen, D. E., Clemons, W. M. Jr, Carter, A. P., Wimberly, B. T. &Ramakrishnan, V. Crystal structure of the 30 S ribosomal subunit from Thermus

a

b

c

H21

H4

H41

S7

S7S4

H43

0.01 0.1 1 10 100

1.0

0.8

Y Y0.6

0.4

0.2

0

H29

H37

H3

S4

H18

H18

H16

H17H21

H41

H43

H37

Time (s)

d

0.01 0.1 1 10 100

1.0

0.8

0.6

0.4

0.2

0

Time (s)

H17

H16

Figure 4 | Ribosomal proteins interact with the rRNA in stages.a, b, Kinetics of direct rRNA backbone protection by proteins S4 (magenta)and S7 (yellow), coloured as in Fig. 2. Schematic symbols are also as in Fig. 2.c, d, Progress curves for protection of individual residues in contact with S4and S7; for clarity, only fitted curves are shown (see data in SupplementaryFig. 5). The relative saturation (Y) of each protection versus assembly time isshown. See Supplementary Methods for definition of RNA–protein contacts.

NATURE | Vol 455 | 30 October 2008 LETTERS

1271

©2008 Macmillan Publishers Limited. All rights reserved

thermophilus: structure of the proteins and their interactions with 16 S RNA. J.Mol. Biol. 316, 725–768 (2002).

22. Dutca, L. M. & Culver, G. M. Assembly of the 59 and 39 minor domains of 16Sribosomal RNA as monitored by tethered probing from ribosomal protein S20. J.Mol. Biol. 376, 92–108 (2008).

23. Powers, T. & Noller, H. F. Hydroxyl radical footprinting of ribosomal proteins on16S rRNA. RNA 1, 194–209 (1995).

24. Nowotny, V. & Nierhaus, K. H. Assembly of the 30S subunit from Escherichia coliribosomes occurs via two assembly domains which are initiated by S4 and S7.Biochemistry 27, 7051–7055 (1988).

25. Powers, T. & Noller, H. F. A temperature-dependent conformationalrearrangement in the ribosomal protein S4?16 S rRNA complex. J. Biol. Chem. 270,1238–1242 (1995).

26. Sayers, E. W., Gerstner, R. B., Draper, D. E. & Torchia, D. A. Structural preorderingin the N-terminal region of ribosomal protein S4 revealed by heteronuclear NMRspectroscopy. Biochemistry 39, 13602–13613 (2000).

27. Samaha, R. R., O’Brien, B., O’Brien, T. W. & Noller, H. F. Independent in vitroassembly of a ribonucleoprotein particle containing the 39 domain of 16S rRNA.Proc. Natl Acad. Sci. USA 91, 7884–7888 (1994).

28. Dodd, J., Kolb, J. M. & Nomura, M. Lack of complete cooperativity of ribosomeassembly in vitro and its possible relevance to in vivo ribosome assembly and theregulation of ribosomal gene expression. Biochimie 73, 757–767 (1991).

Supplementary Information is linked to the online version of the paper atwww.nature.com/nature.

Acknowledgements We thank R. Moss, A. Cukras, L. Cochella and R. Green forhelp with ribosome preparation and peptidyl transferase assays, P. Fleming for helpwith Calc-Surf software, and S. Gupta, M. Sullivan and M. Brenowitz for help withX-ray footprinting. This work was supported by the National Institutes of Health(NIH; GM60819). The NSLS X28C and the Center for Synchrotron Biosciences aresupported by NIH P41-EB0001979.

Author Contributions T.A. performed the experiments, analysed the data andprepared the figures; D.L.B. analysed protections in the 39 minor domain; andS.A.W. prepared the figures and wrote the paper.

Author Information Reprints and permissions information is available atwww.nature.com/reprints. Correspondence and requests for materials should beaddressed to S.A.W. (swoodson@jhu.edu).

LETTERS NATURE | Vol 455 | 30 October 2008

1272

©2008 Macmillan Publishers Limited. All rights reserved