ACTAUNIVERSITATIS

UPSALIENSISUPPSALA

2013

Digital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Science and Technology 1087

Fragment Based Drug Discoverywith Surface Plasmon ResonanceTechnology

HELENA NORDSTRÖM

ISSN 1651-6214ISBN 978-91-554-8775-1urn:nbn:se:uu:diva-209136

Dissertation presented at Uppsala University to be publicly examined in C2:301, BMC,Uppsala University, Husargatan 3, Uppsala, Friday, 29 November 2013 at 10:15 for the degreeof Doctor of Philosophy. The examination will be conducted in English. Faculty examiner:Professor Gregg Siegal (University of Leiden).

AbstractNordström, H. 2013. Fragment Based Drug Discovery with Surface Plasmon ResonanceTechnology. Digital Comprehensive Summaries of Uppsala Dissertations from theFaculty of Science and Technology 1087. 60 pp. Uppsala: Acta Universitatis Upsaliensis.ISBN 978-91-554-8775-1.

Fragment based drug discovery (FBDD) has been applied to two protease drug targets,MMP-12 and HIV-1 protease. The primary screening and characterization of hit fragments wereperformed with surface plasmon resonance -technology. Further evaluation of the interactionwas done by inhibition studies and in one case with X-ray crystallography. The focus of the twoprojects was different.

Many MMP inhibitors contain a strong zinc chelating group, hydroxamate, interacting withthe catalytic zinc atom. This strategy may be the cause for the low specificity of MMP inhibitors.Using FBDD we found a fragment with an unusual strong affinity for MMP-12. An inhibitionassay confirmed that it was an inhibitor but indicated a stoichiometry of 2:1. Crystallographydata revealed that an adduct of the fragment was bound in the active site, with interactionsboth with the catalytic zinc and the S1’ pocket. This may present a new scaffold for MMP-12inhibitors.

For HIV-1 protease the focus was on identifying inhibitors not sensitive to current resistancemutations. A fragment library for screening with SPR-technology was designed and used forscreening against wild type enzyme and three variants with resistance mutations. Many of thehits were promiscuous but a number of fragments with possible allosteric inhibition mechanismwere identified.

The temperature dependency of the dissociation rate and reported resistance mutations wasstudied with thermodynamics. A good, but not perfect correlation was found between resistanceand both the dissociation data and the free energy for dissociation compared to data from wildtype enzyme. However, the type of mutation also influenced the results. The flap mutation G48Vdisplayed thermodynamic profiles not completely correlating with resistance. It was found thatdissociation rate and thermodynamics may complement each other when studying resistance,but only one of them may not be enough.

Keywords: Fragment based drug discovery, SPR biosensor, matrix metalloproteinase-12,HIV-1 protease, resistance

Helena Nordström, Department of Chemistry - BMC, Biochemistry, Box 576, UppsalaUniversity, SE-75123 Uppsala, Sweden.

© Helena Nordström 2013

ISSN 1651-6214ISBN 978-91-554-8775-1urn:nbn:se:uu:diva-209136 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-209136)

To my family

List of Papers

This thesis is based on the following papers, which are referred to in the text by their Roman numerals.

I Gossas T., Nordström, H., Xu, M.-H. Sun, Z.-H. Lin, G.-Q.

Wallberg, H., Danielson U. H. (2013) The advantage of biosen-sor analysis over enzyme inhibition studies for slow dissociat-ing inhibitors – characterization of hydroxamate-based matrix metalloproteinase-12 inhibitors. Med. Chem. Comm. 4(2): 432-442

II Nordström, H., Gossas,T., Hämäläinen, M., Källblad, P., Nys-tröm, S., Wallberg, H., Danielson, U. H. (2008) Identification of MMP-12 inhibitors by using biosensor-based screening of a fragment library. Journal of Medicinal Chemistry. 51(12): 3449-3459

III Nordström, H., Derbyshire, D., Gossas, T., Danielson U. H. Characterization of fragment hits interacting with MMP-12 re-veals a dimerised adduct of a fragment hit in the active site. Manuscript

IV Elinder, M., Geitmann, M., Gossas, T., Källblad, P., Winquist, J., Nordström, H., Hämäläinen, M., Danielson, U. H (2011) Experimental validation of a fragment library for lead discovery using SPR biosensor technology. Journal of Biomolecular Screening. 16(1): 15-25

V Nordström, H., Winquist, J., Geitmann, M., Gossas, T., Solbak, S., Homan, E., Dinér, P., Hämäläinen, M., Danielson, U. H. Identification of fragments for design of HIV-1 pro-tease inhibitors with allosteric mechanisms and new re-sistance profiles. Manuscript

VI Nordström, H., Lindgren, M., Önell, A., Shuman, C., Dan-ielson, U. H. From temperature dependence of drug-target inter-actions to thermodynamics and HIV drug resistance. Manuscript

Reprints were made with permission from the respective publishers.

Papers not included in this thesis Lundquist, A., Hansen, S. B., Nordström, H., Danielson, U. H., Ed-wards, U. K. (2010) Biotinylated bilayer disks as model membranes in biosensor analyses. Analytical Biochemistry. 405(2): 153-159 Elinder, M., Nordström, H., Geitmann, M., Hämäläinen, M., Vrang, L., Öberg, B., Danielson, U. H. (2009) Screening for NNRTIs with Slow Dissociation and High Affinity for a Panel of HIV-1 RT Variants. Jour-nal of Biomolecular Screening. 14(14): 395-403:

Contents

Introduction ................................................................................................... 11

Matrix metalloproteinase 12 ......................................................................... 13 Structure and function .............................................................................. 13 MMP-12 as a drug target .......................................................................... 14 Inhibitors for MMPs ................................................................................. 14 MMP-12 studied with SPR-biosensor technology ................................... 15

Human immunodeficiency virus-1 protease ................................................. 16 Structure and function .............................................................................. 16 HIV-1 protease as a drug target ................................................................ 17 Resistance mutations ................................................................................ 18 HIV-1 protease studied with SPR-biosensor technology ......................... 18

Molecular recognition ................................................................................... 19 Kinetics .................................................................................................... 19 Thermodynamics ...................................................................................... 19

Methods ........................................................................................................ 23 SPR-based biosensor interaction analysis ................................................ 23

Kinetics with SPR-biosensor technology ............................................ 24 Thermodynamics with SPR-biosensor technology .............................. 25

Inhibition analysis with fluorescence resonance energy transfer ............. 25 Optimization of dissociation rates and enthalpies .................................... 25

Fragment based drug discovery .................................................................... 27 History ...................................................................................................... 27 Conceptual framework ............................................................................. 28

Molecular libraries for fragment based drug discovery ....................... 28 Analysis of fragment hits ..................................................................... 28 Turning hits into leads – fragment expansion strategies ...................... 29 False positives - promiscuous inhibitors .............................................. 30 Orthogonal methods............................................................................. 31

Present investigations .................................................................................... 32 Discovery of MMP-12 inhibitors – papers I, II and III. ........................... 32

Paper I: Slow dissociation rate for hydroxamate-based inhibitors. ..... 32 Paper II: Identification of MMP-12 fragment inhibitors. .................... 33

Paper III: Characterization of fragment hits for MMP-12 ................... 35 Conclusions for MMP-12 inhibitor discovery – papers I, II and III......... 37 Discovery of drugs against resistance HIV-1 – papers IV, V and VI ...... 38

Paper IV: Experimental validation of a fragment library. ................... 38 Paper V: Discovery of fragments with inhibition potential for resistant HIV-1 protease. ..................................................................... 40 Paper VI: Effect of HIV-1 protease mutations on the temperature dependence of inhibitor dissociation rates ........................................... 43

HIV-protease resistance drug discovery – papers IV, V and VI .............. 47

Summary and future perspectives ................................................................. 49

Svensk sammanfattning (Summary in swedish) .................................... 51 Enzymer i läkemedelsutveckling .............................................................. 51

Läkemedelsutveckling ......................................................................... 51 MMP-12 – ett målenzym för behandling av kronisk obstruktiv lungsjukdom ........................................................................................ 51 HIV-proteas – ett målenzym för behandling av AIDS ........................ 52 Min forskning ...................................................................................... 52

Slutsatser .................................................................................................. 53

Acknowledgements ....................................................................................... 54

References ..................................................................................................... 56

Abbreviations

ADME Adsorption, distribution, metabolism, excretion AIDS Autoimmune deficiency syndrome COPD Chronic obstructive pulmonary disease CMC Critical micelle concentration clogP Computed octanol-water partition coefficient ΔG The change in Gibbs free energy ΔH The change in enthalpy ΔS The change in entropy DMSO Dimethyl sulfoxide EE Enthalpic efficiency FBDD Fragment based drug discovery FRET Fluorescence resonance energy transfer HAART Highly active antiretroviral therapy HTS High throughput screening ITC Isothermal titration calorimetry KD Dissociation constant LE Ligand efficiency LLE Lipophilic ligand efficiency MS Mass spectrometry nHA Number of heavy atoms (non-hydrogen atoms) NMR Nuclear magnetic resonance PSA Polar surface area R Ideal gas constant SAR Structure-activity relationships SPR Surface plasmon resonance - T Absolute temperature expressed in Kelvin

11

Introduction

The development of drugs is very time consuming and expensive. The caus-es and possible solutions to this problem are continuously discussed in the pharmaceutical society. One aspect thought to play a role is how efficiently compounds with suitable characteristics can be identified for a certain dis-ease and target. Pharmaceutical industry has during the past decades relied on high throughput screening (HTS) of libraries with lead like compounds. This was thought to be the answer to efficient drug development, and has generated many successful drugs. However, there are problems with this strategy. The leads tend to become too large and too hydrophobic upon op-timization. One explanation is that HTS hits typically have a relatively high molecular weight, and the size is often further increased to enhance affinity. In addition, it is often easier to add a hydrophobic then a polar moiety to increase affinity since hydrophobic groups do not have the angle and bond length restrictions as polar groups do for binding to residues with corre-sponding chemical properties. Consequently, the molecules resulting from this type of optimization often have problems with solubility and oral bioa-vailability. The optimization of ADME-properties (adsorption, distribution metabolism excretion) and avoidance of toxicity problems are additional issues to address, making the lead generation phase challenging.

Fragment based drug discovery (FBDD) has been introduced as a method to overcome some of the problems associated with the conventional HTS approach, and that strives to decrease molecular size of leads. This is achieved by screening of libraries with small molecules (“fragments”) in-stead of libraries with larger lead like molecules. By expanding and optimiz-ing hits from the fragment library, it is possible to obtain smaller inhibitors with a better fit in an active or allosteric site on the target, active or alloster-ic, that reduces the activity.

Another recent trend in drug discovery is not only to optimize compounds with respect to affinity, but to focus on the kinetics and thermodynamics of the interaction [1-4]. The kinetics of the interaction is very important since it provide information about the actual interaction mechanisms. In addition it has been shown that the residence time, the time length of the interaction, reflects the biological effect of the drug [4, 5]. This is connected to the kinet-ic dissociation rate of the complex. For some diseases a long residence time is important while for others it might be detrimental.

12

Similarly, thermodynamic properties influence the characteristics of the interaction. By optimizing the leads by adding hydrophobic groups, the en-tropic contribution (∆S) of the free energy (∆G) of the interaction is typically increased. It has therefore been suggested that it is more favorable to identify compounds that have a large enthalpy (∆E) contribution of the interaction and to focus in maintaining this high throughout the optimization process. The enthalpy reflects the energy in the number of hydrogen bonds broken and formed in the interaction between inhibitor and target.

Other essential information on the drug-target complex is where and how the fragment or inhibitor interacts with the enzyme. X-ray crystallography and NMR are potent methods for answering these structural questions.

This thesis is focused on two protease drug targets, matrix metallopro-teinase-12 and HIV-1 protease and how we can describe the nature of the drug-target interaction by studying the kinetics and inhibition. For this pur-pose surface plasmon resonance (SPR)-biosensors and inhibition studies with fluorogenic substrate have been applied. With these methods we have performed fragment-based screening, resolved the kinetics and thermody-namics of drug-target interactions.

13

Matrix metalloproteinase 12

Matrix metalloproteinase 12 (MMP-12) belongs to the large family of en-zymes involved in tissue remodeling and regulation [6]. Remodeling of the extracellular matrix is needed both in normal physiological processes like. embryonic development and in abnormal pathogenic conditions e.g. inflam-mation and invasion of cancer cells. Furthermore, MMPs are also involved in signal cascades, by interacting with inflammatory chemokines and thereby up-regulate or down-regulate inflammation [7].

MMPs hydrolyze connective tissue in the extracellular matrix and are therefore often classified on the basis of this function. For example MMPs consists of collagenases, gelatinases, stromelysins, matrilysin, metalloelas-tase, membrane-type MMPs and non-classified MMPs [8]. An important structural feature of MMPs is a catalytic zinc-atom in the active site coordi-nated by three histidines. As many other proteases, MMPs are expressed in a latent pro-form in which the amino-terminal have a cysteine sulphydroxyl group which blocks the active site and interacts with the catalytic zinc-atom. This propeptide is proteolytically removed during secretion.

Structure and function The enzyme has 5 metal atoms: a catalytic zinc atom, and a zinc and 3 calci-um atoms with structural roles (Figure 1). The three calcium sites on the enzyme have different affinities for calcium, one with high affinity with a KD of 0.1 mM, one with intermediate affinity with KD of 4.6 mM and one with very low affinity with a KD of >100 mM [9]. This has been suggested to be of relevance for the regulation of MMP-12 since inflammation increases the calcium concentration in a limited area [10].

14

Figure 1. Structure of MMP-12. Zinc atoms are shown in green and calcium atoms in blue. The inhibitor batimastat is bound in the active site. Illustration based on the PBD structure IJK3

MMP-12 as a drug target An important function for MMP-12 is the destruction of elastin in the

lung alveolar wall, why the action of MMP-12 has a substantial influence on lung health and disease. It plays a major role in the development of emphy-sema and chronic obstructive pulmonary disease (COPD) [6, 7]. Primary caused by cigarette smoke. Knock-out studies in mice have shown that mice without MMP-12 did not develop emphysema after smoke exposure [11]. Inhibition of the activity of MMP-12 is therefore hypothesized to be a good strategy to treat COPD.

Inhibitors for MMPs Many MMP inhibitors include a hydroxamate group. It binds to the catalytic zinc and forms hydrogen bonds with the catalytic Glu219 and the non-catalytic Ala182.

Figure 2. Binding of the hydroxamate group (black) to residues in the active site of matrix metalloproteinase (grey).

R

Ala182

Glu219

Zn2+

15

The hydroxamate group is a very strong zinc-chelator which is the basis for their low selectivity [6]. There is therefore an ongoing search after more specific MMP inhibitors. One way may be to use a weaker zinc-chelating group with for example a carboxylate as a scaffold. Another strategy is to find a non-zinc interacting inhibitor.

MMP-12 studied with SPR-biosensor technology MMP-12 hydrolyzes peptide bonds. This can be seen from the SPR-signal of the biosensor immobilized with MMP-12. The enzymes immobilized nearby are degrading one another. This auto-proteolysis gives a negative baseline drift in the sensorgrams. When immobilizing the enzyme the surface had to be cross-coupled, thereby linking the individual molecules to each other to decrease the auto-hydrolysis. Furthermore, the enzyme was found to be sen-sitive to low calcium concentrations. As discussed above the calcium site with lowest affinity in the enzyme had a KD of >100 mM. This may have been the reason why the optimal calcium concentration in the assay buffer was as high as 200 mM for best surface functionality. Furthermore, when immobilizing the enzyme on the sensor chip 10 mM calcium had to be pre-sent in the immobilization reagents to preserve functionality.

16

Human immunodeficiency virus-1 protease

The human immunodeficiency virus (HIV) is transmitted via blood and causes acquired immunodeficiency syndrome (AIDS), by infecting cells important for the function of the immune system. Untreated, the infection leads to AIDS after 10-15 years. However, if treated the infection is no long-er a certain path to death. A complete cure is not yet available but with anti-viral therapy the development of AIDS is postponed. Inhibitors against HIV-protease are an important component of HAART (highly active antiretroviral therapy). HIV-protease has a high mutation rate during viral replication. Many inhibitors have been design against this enzyme and when used in clinic, resistance mutations for the inhibitor have rapidly been selected for. New inhibitors are constantly needed to battle resistance.

Structure and function HIV is a retrovirus that utilizes the replication machinery of the host cell for reproduction of new viruses. HIV-protease cleaves the viral Gag and Gag-Pol polyproteins at nine sites resulting in proteins needed for the formation of infectious viral particle in a process known as maturation. The proteins include matrix, capsid and nucleocapsid, and the enzymes reverse transcrip-tase, protease and integrase [12, 13].

HIV-1 protease is a homo dimer with a monomeric weight of 11 kDa. It is an aspartic protease with the characteristic catalytic triad of Asp-Thr-Gly.

The structure of the protease dimer most exposed to resistance mutations are the active site, the flaps covering the active site and the dimerization domain Figure 3. The active site is comprised by both residues from the central part of the active site cavity and the flaps. These residues are forming a highly hydrophobic active site (Arg8, Leu23, Asp25, Gly27, Ala28, Asp29, Asp30, Val32, Lys45, Ile47, Met46, Gly48, Gly49, Ile50, Phe53, Leu76, Thr80, Pro81, Val82, Ile84). The flaps are formed by β-hairpin struc-tures of each monomer (residues 43-58). This flap domain is highly flexible and allows entrance to the active site and also interacts with the substrate or inhibitor when bound in the active site. The backbone movement of the flaps is reported to be in a sub-ns timescale of the protease in solution [14].

17

Figure 3. Structure of the HIV-1 protease dimer, based on the PDB entry 2IEN. The flaps are here in the closed conformation and the inhibitor darunavir has been re-moved for clarity. Residues often mutated and studied in the current thesis are the flap residue G84 (red), active site residues V82 (green) and I84 (blue) and the dimer-ization domain residue L90 (orange). The monomers are shown in gray and light blue.

HIV-1 protease as a drug target All HIV-protease inhibitors, used in clinic bind to the active site. The first generation inhibitors (saquinavir, indinavir, ritonavir and nelfinavir) are pep-tidic compounds mimicking the transition state of the substrate [15]. The second generation inhibitors (amprenavir, lopinavir, atazanavir, tipranavir and darunavir) have a gradually reduced backbone structure. In the structure of tipranavir even the hydroxyl ethylene core present in all other inhibitors was substituted. The hydroxyl group interacts with the catalytic aspartates (Asp25, Asp25’) and mimics the reaction intermediate [13].

Three of the five most flexible residues in the flap tips are highly con-served (G49, G51 and G52) indicating the importance of the function of the flaps for the activity of the protease. This suggests that the flap domain of the enzyme is a good drug target [14]. Activity of the enzyme may be re-duced by restriction of the movements of the flaps in either closed or open conformation. Favoring the closed conformation of the flap domain has been seen by X-ray crystallography to be a promising mechanism for allosteric regulation of the protease activity for two low molecular weight compounds [16].

18

Resistance mutations HIV-protease has a high mutation rate during viral replication. Many inhibi-tors have been design against this enzyme and when used in clinic, resistance mutations for the inhibitor have rapidly been selected for. New inhibitors are constantly needed to battle resistance. The HIV-1 protease is a highly plastic protein. Of the 99 residues in the monomer resistance mutations have been reported for one third of them. This suggests that the enzyme is tolerant against mutations, and can be active even upon multiple mutations.

The mechanism for resistance have been studied with X-ray crystallog-raphy [13]. The flap mutation G48V and the active site mutations V82A and I84V, are examples of mutation causing reduced interaction with inhibitor. The reduced interaction is attributed to loss of hydrogen bonds or van der Waals interactions. For V82A, a shift in the main chain for residues 80-82 has been seen for several inhibitors as an adaptation to maintain the interac-tion with the inhibitor. The L90M mutation is located in the dimerization domain distant from the active site. Even if distant, the mutation has an ef-fect on the active site, reducing affinity of the catalytic Asp25 for the inhibi-tor.

A nomenclature for the resistance mutation is presented by the Interna-tional Antiviral Society–USA [17], with mutations that give resistance on their own defined as major mutations for a specific inhibitor. Mutations of-ten occurring under treatment with the inhibitor but cannot give resistance on their own are called minor. All of the mutations mentioned above are either a major or minor resistance mutation for the inhibitors studied in present the-sis.

HIV-1 protease studied with SPR-biosensor technology HIV-1 protease has auto-proteolysis just as the other protease in the present thesis, MMP-12. This was overcome by cross-coupling of the immobilized enzyme surface. The baseline drift was stabilized. The original assay was developed by Markgren et al. [18].

19

Molecular recognition

The active site of an enzyme is not a static box but a dynamic cavity. The amino acids are flexible and the flexibility of the enzyme mostly involves conformational changes of the whole enzyme during ligand binding and during catalysis. To find the most efficient inhibitors it is important to char-acterize inhibitors and hits with respect to kinetics and thermodynamics of the drug-target complex.

Kinetics The reversible interaction between two molecules for example an inhibitor and an enzyme drug target, can schematically be described as

Scheme 1

The affinity of the ligand for the enzyme is expressed as the equilibrium dissociation constant, KD (Equation 1) or equilibrium association constant, KA (Equation 2). When the ligand is an inhibitor KD is named Ki.

Equation 1

Equation 2

Fragments and many low molecular weight molecules have simple one-step reversible interaction kinetics and display low affinity, high KD. However, even if the interaction is reversible, the dissociation rate might be very slow. This is the case for many inhibitors since slow dissociation rate is often de-sirable, more about this below.

Thermodynamics The binding strength between two molecules can be described by the change of the Gibbs free energy of the interaction, at standard conditions, ΔG°. This

E + I E Ikon

koff

20

energy can be divided upon the two separate thermodynamic parameters, enthalpy (ΔH°) and entropy (ΔS°) (Equation 3) (Figure 4).

° ° ° Equation 3 The energy of the interaction is coupled to the kinetic dissociation constant at equilibrium, KD by ° Equation 4 and

° ° Equation 5

where T is the absolute temperature in Kelvin (K) and R is the gas constant. Equation 5 is known as the van’t Hoff equation.

Figure 4. Energy diagram of an inhibitor (I) interacting with an enzyme (E) accord-ing to a simple 1-step mechanism (Scheme 1). ΔG is the Gibbs free energy and ΔGon and ΔGoff the activation energies for the association and dissociation event respec-tively.

The enthalpy change in a reaction primarily refers to the heat change and the entropy change to the randomness, the rotational and translational degrees of freedom in the system. Using a schematic description of the binding event in biochemical terms, the enthalpy describes the energy needed for forming and breaking hydrogen bonds and van der Waals interactions. The entropy, ΔS, of the interaction describes the degrees of freedom in the molecular com-plex. However, this includes the degrees of freedom in the whole system, including the solvent molecules. The system strives after increased entropy. When a molecule interacts with another, the degrees of freedom decrease for

Reaction coordinate

Ener

gy

ΔGE + I

E I

E I‡

ΔGoffΔGon

21

the interacting molecules, but since the highly ordered solvent molecules arranged around hydrophobic and polar parts of the macromolecules are broken, the solvent molecules gain degrees of freedom. This may result in an overall beneficial change of entropy for the interaction [19].

If ΔH and ΔS are temperature independent, Equation 5 describes the ener-gy in the binding event correctly. However, if the binding is associated with conformational changes of the enzyme, for example, deviations may occur. In those cases, the equation is expanded with a heat capacity term giving a more correct representation [20].

° ° ° °

Equation 6 T0 is a selected reference temperature. Thermodynamic parameters of an interaction are often measures by isother-mal titration calorimetry (ITC). With this method the enthalpy of the interac-tion is measured directly, and the entropy, heat capacity (Cp) and dissocia-tion constants (KD) are derived from this. The thermodynamic parameters can also be calculated from kinetic constants measured at different tempera-tures, using for example SPR-biosensor technology (see below). If using a well-defined robust experimental method the values derived from ITC and SPR will not differ significantly [21, 22].

The van’t Hoff equation (Equation 5) can be used to calculate thermody-namic parameters from equilibrium dissociation constants, KD. A plot of lnKD against 1/T gives a straight line with a slope of ΔH°/R and an intercept of ΔS°/R. If the line is not straight it is better to use a non-linear equation (Equation 6). With ITC it is not possible to estimate the activation energies of the association and dissociation phases, since it does not have kinetic resolution but involves measurement at equilibrium. For the purpose of ana-lyzing the thermodynamics of the association and dissociation event the Ar-rhenius empirically derived equation (Equations 7 and 8) and the Eyring equation based on transition-state theory (Equations 9 and 10) can be used [23]. Equation 7 The Arrhenius equation expressed in a linear form

∗ Equation 8

22

Where Ea is the activation energy and k is the rate constant for the associa-tion or dissociation, kon or koff.

Equation 9

The Eyring equation expressed in a linear form

∗ Equation 10

Where kB is the Boltzmanns constant and h is the Planck constant.

23

Methods

SPR-based biosensor interaction analysis In the SPR biosensor instruments produced by Biacore (now GE-healthcare), the detection of interaction between biomolecules is detected by surface plasmon resonance. This is a phenomenon that occurs in a thin layer of gold between two layers with different refractive index, glass and sample solution [24]. Monochromatic and polarized light is reflected on the gold surface under total internal reflection (Figure 5). Even if the light is reflected an electromagnetic field component, the evanescent wave penetrates the gold surface to the sample solution. The reflected light intensity goes through a minimum at a certain angle producing a sharp dip in the reflected light. This angle is the SPR angle and it is proportional to the mass in the solution sam-ple. A molecule has been immobilized on the carboxy methylated dextran surface on the opposite side of the gold layer. When another molecule, free in solution is injected over the surface and binds to the immobilized mole-cule, the mass increases and the angle of the dip changes. The interaction is monitored in real time.

Figure 5. Schematic picture of SPR-detection in biosensors. Monochromatic and polarized light is reflected on the gold layer of the biosensor chip with immobilized molecules, proteins. The SPR angle changes when another molecule binds to the immobilized molecule.

The change of the SPR angle is viewed over time as Resonance units (RU) against time (Figure 6).

24

Figure 6. Sensorgram of a high affinity inhibitor interacting with HIV-1 protease.

Kinetics with SPR-biosensor technology Rate constants kon and koff, and equilibrium constants KA and KD can be ob-tained by a fit of kinetic models for sensorgrams from a concentration series to an equation [25] (Figure 7). For the simplest case with a 1:1 interaction, the interaction can be described as (Equation 11):

Equation 11

Were R is the response in RU, dR/dt is the change of response over time, Rmax is the maximum response and C is the concentration of analyte (the molecule injected cross the surface with the immobilized molecule). Howev-er, other models are required when there is e.g. mass transport limitation, heterogeneous surface or conformation changes involved in the interactions.

Figure 7. Ideal sensorgrams from a 1:1 interaction.

-5

0

5

10

15

20

25

30

35

40

-50 0 50 100 150 200 250 300 350 400

Re

spo

nse

(R

U)

Time (s)

Dissociation phaseAssociation phase

Injection start

Injection end

-5

0

5

10

15

20

25

30

35

0 100 200 300 400 500 600 70 0

Tim e s

Res

pons

e

RU

Rmax

0.010 µM

0.040 µM

0.16 µM

2.6 µM

10 µM

0.64 µM

25

Thermodynamics with SPR-biosensor technology Thermodynamic parameters can be calculated from the affinity and rate con-stants of the interactions. For determination of equilibrium thermodynamic parameters a van’t Hoff analysis is made (Equation 5). When analyzing the association and dissociation phase the. Arrhenius and/or the Eyring equation can be used (Equations 8 and 10).

Inhibition analysis with fluorescence resonance energy transfer Resonance energy transfer is used to monitor changes in the distance be-tween molecules. This method was first introduced by Theodor Förster and the technique was named after him, ‘Förster resonance energy transfers’ [26].

In the form the technique is used for proteases, energy is transferred from a molecule with high intrinsic fluorescence to a quenching molecule that absorbs energy in the emission wavelength of the first molecule [27]. The donor and quencher molecules are connected through a peptide sequence with a cleavage site. No fluorescence is emitted when this substrate molecule is intact, but when the protease cuts the peptide sequence the quencher and donor are separated and light is emitted and can be detected in a fluorometer.

Many activity and inhibition assays for the study of kinetic constants for proteases like Km, kcat, Ki and also IC50 use the concept of resonance energy transfer [28] In this thesis the technique has only been used for evaluation of IC50-values. The fraction of activity for the enzyme is measured for samples in a concentration series of inhibitors. The data is plotted against the inhibi-tor concentration and the IC50-value is where the activity is 50% of the non-inhibited activity. These curves can also be used to evaluate the stoichiome-try of the interaction. For this purpose the fraction of activity data was fitted to equation 12 in paper III. The original equation was developed by A. V. Hill to describe the binding of oxygen to hemoglobin [29]. However, the Hill equation is only valid for estimation of the actual number of binding sites when there is marked positive cooperativity between the ligand interactions. However, the Hill coefficient can be used as an indicator of stoichiometry of the interaction. In this thesis, the name slope factor is preferred over Hill coefficient.

Optimization of dissociation rates and enthalpies The correlation between increased dissociation rates of inhibitors and re-sistance mutations, were initially found for HIV-1 protease by Shuman et al.

26

2003 [30]. However, in a later study involving kinetic constants and Ki and their correlation with inhibitory effect in a viral replication assay, koff was not found to be the most critical agent [31]. Instead the highest correlation between ED50 and the free concentrations of virus particles was found with KD.

Later, Copeland et al. 2006 argued that the dissociation rate of the binary inhibitor-target complex correlated with the biological effect of the inhibitor [4]. They introduced the term residence time and argued that long dissocia-tion time was essential for the duration of pharmacological effect and target selectivity. Of course this is not valid for all cases. For some inhibitors a long residence time may be toxic, for example for antipsychotic and throm-bosis drugs [32]. In the year 2008, ideas about optimization lead compounds with respect to enthalpy was put forward by Freire [1]. Freire meant that the increased enthalpy had occurred seemingly unintentionally during drug de-velopment. He exemplified his statement with inhibitors to HIV-protease and HMG-CoA reductase and argued that since increasing the enthalpy of a lead is difficult, molecules should be selected with respect to their enthalpic properties already in the beginning of the drug discovery process. By others, there were also attempts to correlate a favorable enthalpy with a long disso-ciation rate of drug-target complexes [5].

During later years the drug discovery community have begun following up, modified and improved the appealing simple and straight forward ideas of optimization regarding both enthalpy and dissociation rate [2, 3, 5]. How-ever these metrics might not be useful in all cases. One example of drug discovery where there was no connection between favorable enthalpy and long dissociation time was illustrated by Lange et al. [33]. For the com-pounds in the SAR-series the inhibitors with slowest dissociation rate were not the ones with most beneficial enthalpy.

27

Fragment based drug discovery

Fragment based drug discovery has gained increased popularity during the last decade. This approach is becoming important for pharmaceutical com-panies as a complement to high throughput screening (HTS), the previous state of the art method. In HTS, large libraries of drug-like compounds (MW 250-600 Da are screened for interaction to a drug target [34]. The com-pounds are complex and to cover a larger part of chemical space the libraries have to consist of many molecules. In fragment based screening, the libraries consist of small compounds (MW < 300 Da) generating hit compounds with weak interactions to the target, but the binding efficiency relative to the size of molecule is high. That is, a larger portion of the fragment participates in target binding compared to that for large drug-like compound. The fragments are later optimized to larger molecules, leads. Fragment libraries can be small since the small compounds with low molecular complexity have better coverage of chemical space compared to the HTS-libraries.

History The original idea of combining weak interacting compounds into a larger compound with stronger affinity is attributed to William Jencks [35]. He wrote about the theoretical concept of fragment based drug discovery three decades ago but the idea could not be realized until experimental techniques sensitive enough enabled efficient screening of weakly interacting com-pounds. The first company to show the utility of the concept was Abbott Laboratories in 1996 [36]. The detection of weak interactions and binding sites for small organic molecules to a 15N-labeled protein (FK506) was demonstrated. They called their technique SAR by NMR (structure-activity relationships by nuclear magnetic resonance).

One of the advantages of the technique was that it was possible to use high concentrations of the organic molecules, without interference with background noise. This is important since weak interactions might not be detectable at low concentrations. After this, several companies for example Astex, began exploring the possibility to use X-ray crystallography for frag-ment based drug discovery. Also Abbott Laboratories [37] came up with a crystallographic screening method, a few years after SAR by NMR. The crystallographic methods often apply a soaking technique, in which crystals

28

of the protein target is soaked with a fragment solution. Compounds interact-ing within a defined area, often the active site, are identified by analysis of an electron density map. After NMR and X-ray crystallography, SPR-based biosensor techniques began to be used. SPR-biosensor has become a very popular high throughput technique in fragment based drug discovery [38], since it is easy to use, requires low amounts of reagents and gives infor-mation of the affinity and kinetics of the interactions. SPR-biosensors is typically used in combination with other techniques for FBDD.

The above three techniques can be regarded historically as those first ap-plied to and still dominate FBDD. However, in recent years many other promising techniques have emerged [39], both for screening and for hit vali-dation purposes. Worth to mention are ITC, thermal shift, mass spectrometry (MS) and in silico docking.

Conceptual framework The history of FBDD is not only dependent on the new techniques but just as much the development of the conceptual framework. Cornerstones for this are how to develop the fragment libraries e.g. according to the rule of three, development of metrics for analyzing and evaluating the potentials of the fragment hits e.g. ligand efficiency, and finally how to evolve hits to leads.

Molecular libraries for fragment based drug discovery Lipinski proposed a ‘rule of five’ for description of the properties molecules required to produce lead compounds suitable for design of orally bioavaila-ble drugs [42]. These rules are not applicable to fragments since fragments are much smaller in size than HTS leads. A ‘rule of three’ was proposed by Congreve et al. from Astex Technology [43]. These rules propose that the molecular weight should be < 300 Da, the number of hydrogen bond donors as well as hydrogen bond acceptors should be ≤ 3, the computed partition coefficient (clogP) should be ≤ 3, and the number of rotatable bonds ≤ 3. Furthermore, the polar surface area (PSA) should be ≤ 60Å2. This has since then been used as a guideline for fragment properties when designing librar-ies. Properties have been added or removed depending on the purpose of the library, but the guideline rules are still valid.

Analysis of fragment hits A useful metric for analyzing hits from a fragment screen is ligand efficien-cy (LE) [44]. It relates binding energy to the size of the fragment. The basic concept is to normalize the free energy of ligand binding with respect to the number of heavy atoms in the ligand (non-hydrogen atoms), LE = ΔG/nHA.

29

This can also be written as LE = (-RTlnKD)/nHA, were R is the ideal gas constant, T the absolute temperature and KD the dissociation constant. If KD is not known, IC50 may be used for comparison of hits identified with the same assay. Additional types of metrics related to LE have been developed all focusing on other characteristics of importance for drugs. Worth mention-ing are lipophilic ligand efficiency (LLE)[45] and enthalpic efficiency (EE)[2]. LLE relates the potency of the hit to its lipophilicity obtained by subtracting the partition coefficient from the logarithm of the inhibition con-stant, LLE = pIC50 (or pKi) – clogP. This is relevant since too high lipo-philicity may result in multiple binding to the target, toxicity, poor solubility and poor metabolic clearance. The enthalpic efficiency relates the thermody-namic enthalpy parameter to the size of the hit, EE = ΔE/nHA. The enthalpy reflects the number/strength of hydrogen bonds in the interaction of hit to target. For further information on other useful metrics in drug discovery see the review by Hann and Keserü [46].

Turning hits into leads – fragment expansion strategies In the first FBDD article involving SAR by NMR, fragments binding in proximal parts of the enzyme were chemically linked to each other (Figure 8) [36]. Though it is a simple concept, linking two fragments binding in proximal but not overlapping sites has proven to be difficult, since finding links with the right length, rigidity and bond angles not reducing the ligand affinity is problematic [47]. In the merging strategy, fragments interacting in the same site are structurally overlaid and merged to provide a higher affini-ty compound. An alternative strategy involves the expanding/growing of fragments similar to the one used for evolution by HTS. Fragments are op-timized of adding molecular groups giving the scaffold new and improved properties.

Figure 8. Schematic presentation of hit optimization to lead molecule in FBDD.

Protein and fragments Fragments bound to protein Lead compound

Linking/ growing /merging

Fragment screening

30

False positives - promiscuous inhibitors In FBDD high concentrations of fragments is often necessary to use to detect low affinity interactions. These high concentrations may favor false posi-tives. These non-well behaved hits are called promiscuous and have an abil-ity to inhibit a broad spectrum of different protein classes. Some of them often turn up frequently in different screening campaigns and have been named Pan Assay Interference Compounds (PAINS) [46] or frequent hitters [47].

A compound screening library is designed to contain molecules that may be made into leads with beneficial ADME properties, low toxicity and high oral bioavailability. A promiscuous fragment might be an unsuitable starting point for drug design, since it can be difficult, but not impossible, to expand and make changes to the compound to get properties suitable for a drug.

False positive hits or leads can have properties like aggregation, protein reactivity or interference with assay signaling. For paper I-V in this thesis the most important aspects are aggregation and interference with assay sig-naling.

Recognition of promiscuous binders SPR-biosensors The response in the surface plasmon resonance based technology is depend-ing on the mass density on the sensor surface. Promiscuity based on aggrega-tion of the inhibitor can most often be detected as having higher signal than the expected maximal response (Rmax) [48, 49]. The signal cannot represent a 1:1 interaction based on comparison with a positive control incorporated in the assay. If the stoichiometry is ≥5:1 Giannetti et al. has identified the inter-action as super stoichiometric and argues that aggregates of fragments binds to the target. If the stoichiometry is between 1:1 and 5:1 the fragment inter-action might be worth following up since fragments may bind to several sites on the target enzyme. Even two fragments in the same binding site is not an unrealistic scenario. It has been seen with X-ray crystallography [50] and in paper III.

Further recognition signs of promiscuous inhibitors are that the interac-tion sensorgrams do not indicate any saturation of the response, and some might have very slow dissociation. Generally the promiscuous inhibitors interact with many different types of proteins, why it may be advisable to incorporate nonrelated enzymes in your assay, but even if the fragment in-teracts with many different enzymes it might do so with different mecha-nisms. For example fragment interaction might show slow dissociation for one enzyme and fast for another one as further discussed in paper IV.

In many cases aggregation can be mitigated if using detergent in the buff-er. Some aggregates are sensitive to detergent but some are not.

31

Inhibition assay - FRET In fluorescent based inhibition assays promiscuous interacting compounds can often be recognized by a steep slope of the activity-concentration curve when determining the IC50-value [51]. A steep slope might be attributed to compound interacting with several sites on the target enzyme, to high en-zyme concentration relative KD in the experiment, inhibition caused by mi-celle formation of the inhibitor or fluorescent inhibitor.

If too high enzyme is used the IC50-value will be linearly increased with increasing enzyme concentration. If KD is not known enzyme concentration could be varied until stable IC50 is reached.

Solubility is often problematic in screening regimes. Aggregating com-pounds might form micelles and micelles can inhibit the activity of the en-zyme by enveloping the protein, shielding it from the substrate. In most cas-es micelles are sensitive to detergent why adding detergent to the experiment may completely abolish the inhibition. Detergents used for this might be tween-20, CHAPS or triton X-100 [52, 53].

High intrinsic fluorescence of a compound may mask its true inhibitory effect [54]. The signal from the fluorescent substrate is drowned by the sig-nal from the compound. At higher concentration of compound, no activity can be seen and the results are misinterpreted as inhibition.

Orthogonal methods One way to overcome high portions of false positives is to use orthogonal methods for validation. Validation of a hit in one method by another strengthens the hit. However, different methods can generate completely different hits [55]. A hit may still be a valid hit even if it doesn’t turn up as a hit in an orthogonal assay. The reason for this is that different methods put different restrains on the target and fragment due to different conditions.

Fragment libraries should always be screened with different methods. The techniques complement each other and thereby increase the information of the interaction. Kinetic, thermodynamic and structural information all gives valuable insights of where on the target the fragment interacts, the binding strength and mechanism, an essential step on the way to a new potent drug.

32

Present investigations

Discovery of MMP-12 inhibitors – papers I, II and III. In this collection of papers we show for the first time how fragment based drug discovery can be performed using SPR-based biosensor technology. MMP-12 was the target of choice since there was a need to identify new classes of MMP inhibitors. Most compounds targeting this family of en-zymes and tested in the clinic failed due to toxic effects. This was thought to be due to poor specificity, caused by the strong zinc chelating properties of the hydroxamate group that was an essential feature of the clinical candi-dates. If finding fragments interacting efficiently with the S1’ pocket instead of the catalytic zinc, specificity might be improved. The three papers repre-sent different stages in the drug discovery process. That is, 1) identification of desired properties of inhibitors, 2) finding hits, and 3) hit validation.

Paper I: Slow dissociation rate for hydroxamate-based inhibitors. In this study, the advantage of the time-resolved data obtained from SPR-biosensor analyses, compared to enzyme inhibition studies, was demonstrat-ed. Even if the kinetic constants cannot be quantified, a qualitative analysis of the sensorgrams gives important information on the interaction character-istic for a specified interaction. This information can be of help for avoiding pitfalls resulting from sub-optimally performed enzyme inhibition measure-ments. This was demonstrated with MMP-12 and hydroxamate-based inhibi-tors.

Results paper I The hydroxamate of inhibitors containing this group chelates the catalytic zinc in the active site of the enzyme. SPR-biosensor analysis of the interac-tion revealed that it had a very slow dissociation rate. By substituting the hydroxamate group in one inhibitor by a carboxylate group, the dissociation rate was increased, demonstrating that the hydroxamate group alone was responsible for this kinetic behavior (Figure 9). This is information is very useful when designing an enzyme inhibition experiment since standard ini-tial rate analyses are based on steady-state measurements that rely on an equilibrium with fast on- and off-rates. Also progress curve analysis without accounting for the very slow dissociation rate, e.g. without pre-incubation of

33

the inhibitor and enzyme reaction solution is problematic since the inhibition may be interpreted as time-dependent and due to a slow association rate. Such a conclusion would be erroneous for hydroxamate inhibitors since the SPR-biosensor interaction study shows a fast on-rate. However, even with the knowledge about the slow dissociation rates for these compounds it was difficult to perform inhibition measurements that gave reliable data.

Figure 9. Interaction between MMP-12 and two inhibitors differing only in the N-terminal group, i.e. having a hydroxamate or carboxylate zinc chelating group.

Paper II: Identification of MMP-12 fragment inhibitors. In this paper the design of a fragment screen for MMP-12 using SPR-biosensor technology was shown. The aim of the screen was to find active site interacting fragments that did not interact with a reference enzyme con-taining a catalytic zinc atom.

Results paper II A library of 245 compounds partly designed to target MMPs was screened against MMP-12. A Biacore A100 instrument was used since it enables the use of four flow cells with four different enzymes in parallel in a single screening experiment.

MMP-12 is an inherently unstable enzyme since it is subject to auto-proteolysis and requires high concentrations of Ca2+-ions in order to be sta-ble. For this reason, screening was performed with three surfaces containing variants of the enzyme: 1) wild type, 2) wild type stabilized with the inhibi-tor ilomastat during immobilization, and 3) an inactive E219A mutant, with the catalytic glutamate exchanged for alanine. Carbonic anhydrase II was used as a reference enzyme in the assay since it has a catalytic zinc atom in the active site like MMP-12.

-1 5

-1 0

-5

0

5

1 0

1 5

-5 0 5 0 15 0 25 0 35 0 45 0

Time (s)

Re

sp

on

se

(RU

) H ydroxam ate

C arboxylate

OH NH

O

N

NH

ONH

O

NH

O

O

NHO

OH

N

34

Fragments with a pronounced binding to the reference enzyme were de-fined as non-specific or zinc interacting fragment and discarded. In order to discriminate active site binding from non-active site binding fragments, a competition experiment was performed. It involved the injection of frag-ments over the enzyme surfaces together with the active site binding inhibi-tor ilomastat. The interactions were compared with those in the absence of ilomastat. A total of 11 fragments were classified as selective and active site binding, and were subsequently followed up in a screen involving a concen-tration series of the fragments. Five fragments with a non-linear response-concentration relationship were defined as hits.

The inhibitory potencies of the selected hits were examined with an inhi-bition assay using a fluorescent substrate. All five fragments inhibited the activity of MMP-12, albeit three with very modest inhibitory potencies. The IC50-values could be determined for the other two compounds, and the corre-sponding ligand efficiencies LEIC50) were calculated. Compound B6 had an IC50-value of 355 ± 14 µM and a ligand efficiency of 0.4 kcal/mol non-hydrogen atoms, while the corresponding values for B7 were 0.29 ± 0.01 µM and 0.7 kcal/mol non-hydrogen atoms.

Information of the interaction based on sensorgram curve shape. Sensorgrams are very information rich and a visual inspection can reveal important features of the interaction mechanisms.

Figure 10. Sensorgrams for B7 and B6 interacting with wt MMP-12 stabilized with ilomastat during immobilization (a, c) and inactive E219A mutant (b, d). Fragments were injected in concentration series from 0-500 µM in two-fold dilution.

35

The auto-proteolysis of MMP-12 can be seen as a negative drift in the base-line in the sensorgrams in Figure 10 (a, c). For B6 the drift is disrupted dur-ing injection of the compound (Figure 10(c)). This indicates that the com-pound is indeed an inhibitor, with an ability to prevent autoproteolysis. Fur-thermore, B6 does not bind to the stable mutant E219A Figure 10(d), sug-gesting that the catalytic glutamic acid interacts with B6. Taken together, it appears that fragment B6 interacts with the active site of MMP-12.

B7 also disrupts baseline drift (Figure 10(a)), both during association and dissociation, and the signal seems to have reached saturation even at the lowest concentrations. Moreover, the fragment appears to have a high affini-ty for MMP-12. In contrast to B6, B7 interacts with the E219A mutant, but with different kinetics to the interaction with wild type MMP-12 (Figure 10(b)). There seems to be a least two interactions since the signal from the association phase does not reach saturation and the dissociation seems to be composed of two dissociation events with different rates, a fast event in the immediate beginning, and a slower event at the end. The sensorgrams can be a result of a promiscuously interacting fragment, with slow dissociation and no saturation of the signal. The B7 sensorgrams for the stable mutant had a suspiciously promiscuous appearance, but since B7 competed with the active site inhibitor ilomastat it may be a fragment with specific interaction with the active site of MMP-12.

Paper III: Characterization of fragment hits for MMP-12 The interactions between MMP-12 and fragment hits B7 and B6 from paper II were examined further with an inhibition assay and X-ray crystallography. To gain more knowledge about the nature of the interactions, analogues of the fragments were also studied with the inhibition assay. The overall aim was to obtain a preliminary SAR and to find possible ways to expand the fragment hits to successful leads.

Results paper III B7 and analogues thereof had different inhibition kinetics compared to B6. Equation 12 could only be fitted to B7 data if a slope factor (S) of two was used in the equation. For B6 this was not required and S was set to one. This indicated that B6 interacted with a stoichiometry of 1:1, and B7 with a stoi-chiometry of 2:1 (Figure 11).

Equation 12 vi = activity in presence of inhibiting fragment

36

v0 = non-inhibited activity [I] = fragment concentration S = slope factor

Figure 11. Fragments B7 and B6 inhibiting the activity of MMP-12. Equation 1 was fitted to the data by non-linear regression analysis with a slope factor (S) of 1 (left) or 2 (right). The fragments ranges of concentration were 0.049-4.2 µM (B7) and 5.2-330 µM (B6). The enzyme and substrate concentrations were 4.8 nM and 24 µM, respectively.

All compounds, except two of the B7-analogues, appeared to have the same binding stoichiometry as B7 in the MMP-12 inhibition assay. The main simi-larity between B7 and the B7-analogues was the hydantoin-motif (also named imidazolidin-2, 4-dione) which has a zinc-chelating function. None of the analogues had better inhibitory potencies than B7. An analysis of the correlation between the structures and inhibitory effects of the analogues indicated structural modifications that should be avoided in further expan-sion of the fragments. Substitutions on the phenyl-group of the molecule did not have any significant effect on the slope factor. This suggests that both the hydantoin- and the phenyl-motifs are important for the interactions.

Crystallographic analysis of the B7-enzyme complex gave an unexpected explanation for the 2:1 stoichiometry. Instead of the expected binding to two different sites, it was found that two fragments had fused and bound to the active site as an adduct. The hydantoin on the first phenylhydantion was che-lated to the catalytic zinc atom, and the phenyl part of the second phenylhy-dantoin was positioned in the S1’ pocket.

[I] µM

0.1 1

vi/v

0

0

0.5

1

[I] µM

0.1 1

vi/v

0

0

0.5

1

[I] µM

10 100

vi/v

0

0.5

1

[I] µM

10 100

vi/v

0

0.5

1

S = 1

S = 1 S = 2

S = 2

37

Conclusions for MMP-12 inhibitor discovery – papers I, II and III This work has been important for the discovery of new MMP-12 inhibitors while also demonstrating that it is possible to use SPR-biosensor technology for fragment based drug discovery also for unstable enzyme that are difficult to work with. The biosensor method proved to be valuable for fragment screening and gave a large numerical dataset that was easily interpreted, while also enabling a more detailed analysis of interactions by simply exam-ining sensorgrams. Significant information about the interactions was ob-tained, but was varied with the design of the assay. For example, indications of specificity and active site interaction can be obtained by the use of a ref-erence enzyme (here carbonic anhydrase II), and with competition experi-ments using an active site inhibitor (here ilomastat).

The work also highlights the risk of using an enzyme with an active site mutation for increased stability. The mutation may change the interaction kinetics so much that the results are completely misinterpreted. In this case the stable mutant gave higher and more robust signals than the unstable wild type MMP-12 in the screening assay. However, the immobilization stabi-lized wild type and the mutated variant gave completely different kinetics for B7 and B6. If only the stable mutant had been used, B7 would most likely have been considered promiscuous and B6 would not have been identified at all.

The three studies also demonstrate the importance of using orthogonal techniques. In paper II, SPR biosensor analysis indicated that the fragments interacted with MMP-12. It also provided further information about the bind-ing site for the fragment and the affinity for the interaction. The FRET-based inhibition assay in paper III validated the hits and enabled the determination of IC50-values and the binding stoichiometry. This was important for ration-alizing the interaction, with two types of interactions for B7 and one for B6. X-ray crystallography revealed that an unexpected adduct was formed by two B7 fragments and that it interacted with the active site of MMP-12. The importance of orthogonal methods was demonstrated also in paper I. With-out knowing about the very slow dissociation rate of hydroxamates it was not possible to design inhibition experiments or interpret the resulting data.

Another important aspect that was revealed by these studies is the similar-ity in the binding of fragment B7 to MMP-12 and inhibitors containing a hydroxamate group. For B7 the dissociation rate was slower than expected for low affinity fragments, although not as slow as for the extremely slowly dissociating hydroxamate inhibitors. B7 was shown to interact with the cata-lytic zinc in a similar manner as hydroxamate inhibitors. Strong zinc chelat-ing inhibitors seem to result in interactions with very slow dissociation rates, as seen I paper I.

38

Even if slow dissociation is important for the biological effect of a drug, strong zinc chelating inhibitors should perhaps be avoided for MMPs since this is not only a large family of enzymes, but there are also many other metallo enzymes with a catalytic zinc atom. Specificity may thus be an issue for this motif. Although B7 did not interact with the reference enzyme, the interaction characteristics between B7 and hydroxamate inhibitors are very similar. If B7 or the B7 adduct are used for further fragment evolution, it is probably important to perform specificity studies as part of analogue evalua-tion.

Discovery of drugs against resistance HIV-1 – papers IV, V and VI There is a constant need for new AIDS drugs since resistance develops rap-idly against drugs in the clinic. The problem of drug resistance has therefore been in focus in these papers. In paper IV, a fragment library was designed and validated for screening using an SPR-biosensor based approach. It was later screened against HIV-protease in paper V. In this paper, the experience of fragment screening with SPR-biosensor technology gained in paper II was exploited. It served as the basis for the screening strategy for wild type HIV-1 protease and three variants with resistance mutations. The aim was to iden-tify hits suitable for development into leads with sustainable inhibition prop-erties also for resistant variants. In paper VI the resistance characteristics were further explored by determination and analysis of the thermodynamic profiles of the dissociation phase of inhibitors interacting with HIV-1 prote-ase.

Paper IV: Experimental validation of a fragment library. An important component in any fragment-based drug discovery campaign is access to a high quality fragment library. The fragments in the library must have a potential to be optimized without risking poor pharmacokinetic prop-erties or toxicity. In this paper, the design of a fragment library and its suita-bility for SPR-biosensor based screening was demonstrated. In the valida-tion, the solubility and tendencies for promiscuous interactions of fragments were the major concerns.

Results paper IV A fragment library essentially obeying the rule of three was designed, but with some additional criteria were included to reduce the number of reactive groups, oral bioavailability, and to ensure that the fragments had appropriate structural frameworks suitable for optimization. Special care was taken to

39

select compounds that fit SPR-biosensor technology based analysis. Com-pound solubility was considered important since fragments need to be screened at high concentrations to ensure identification of low affinity inter-actions.

A selection of 930 fragments was selected from 4.6 million commercially available compounds. To validate the characteristics of the fragment library, a set of three enzymes identified as drug targets, HIV-1 protease, human α-thrombin, carbonic anhydrase II and a non-target, human serum albumin, were used. The drug targets belonged to different classes of enzymes; HIV-1 protease is an aspartic protease, thrombin a serine protease and carbonic anhydrase a metalloenzyme. The same buffer and fragment preparations could be used for the four proteins, enabling the library to be screened against all 4 proteins in parallel. The compounds were screened at a single concentration of 200 µM in phosphate buffer, with 5% DMSO and 0.05 % Tween-20.

In order to identify fragments with unsuitable characteristics for screen-ing, the sensorgrams and interaction features were carefully scrutinized. However, no solubility problems could be noticed for the fragments. This was an analysis that was regarded as particularly relevant since aggregation leading to precipitation can be identified by SPR biosensor analysis as une-ven sensorgrams and the presence of spikes. Less severe aggregation leading to micelle formation is seen as much higher interaction levels than expected for a 1:1 interaction or even a 5:1 interaction. The interactions were there-fore also analyzed with respect to stoichiometry and dissociation rates. Inter-actions with a stoichiometry of ≥ 5:1 were identified as super-stoichiometric, a threshold first used by Giannetti el al. [48]. Of the 930 compounds ana-lysed, 26 were considered as super-stoichiometric, at least for one enzyme. Only 3 compounds interacted super-stoichiometrically with HSA, 7 with carbonic anhydrase and 2 with thrombin. For HIV-1 protease 22 compounds were identified as super-stoichiometric, that is as much as 85% of all super-stoichiometrically interacting compounds. It may be attributed to the rela-tively poor functionality of the sensor surfaces, probably resulting from par-tially degraded and unfolded enzyme.

Slowly dissociating fragments were identified by dividing the interaction signal late in the dissociation phase with a signal point in the immediate beginning of the dissociation. The detection limit was set to > 0.8. With this method, 35 compounds were identified as slowly dissociating for at least one enzyme. For HSA, 6 fragments were defined as slowly dissociating, the cor-responding values were 23 for thrombin, 22 for carbonic anhydrase, and 10 for HIV-1 protease. Only 2 fragments did not dissociate at all from all 4 proteins.

The results from the validation study were encouraging since no insoluble fragments could be identified and the number of potential promiscuously interacting fragments was low.

40

Paper V: Discovery of fragments with inhibition potential for resistant HIV-1 protease. The aim of this paper was to find fragments interacting with wild type HIV-1 protease and three variants with resistance mutations: G48V, V82A and I84V. For this purpose, the fragment library studied in paper IV was used.

Results paper V As in paper IV, the library was screened simultaneously against all 4 enzyme variants and using the same buffer components. Instead of using a single concentration of each fragment, the fragments were screened in concentra-tion series, ranging from 0 to 500 µM with two fold dilution, giving six con-centrations for each fragment. This approach enabled the estimation of the dissociation constant, KD, and the maximum binding level, Rmax, by fitting the data of the concentration series to a Langmuir interaction model by non-linear regression (Equation 13). A linear term (k) was incorporated in the equation accounting for weaker ‘non-specific’ interactions. The concentra-tion is designated c, while m is the offset.

∗ ∗ Equation 13

The estimated KD was used to filter the data in the beginning of the analysis step, and later to calculate ligand efficiency (LE). Of those samples having sufficiently high signals, fragments with KD-values up to 10 mM were se-lected. Furthermore, only fragments interacting with wild type enzyme and 2 or 3 of the enzyme variants were identified as interesting hits. This resulted in 47 fragments that were followed up in a FRET-based enzyme inhibition assay. Hits were selected in this assay on the basis of two criteria, >40% inhibition of wild type enzyme, and low intrinsic fluorescence (Figure 12).

Figure 12. Inhibition of HIV-1 protease in the presence of 200 µM fragment and 5 µM substrate for 10 fragments selected as hits. The selection threshold of 40% is marked. Results from interactions with the enzyme variants are shown in black (wt), white (G48V), dark grey (V82A), and light grey (I84V). The estimation of fragment fluorescence is expressed as % of non-inhibited signal (bold black line).

-10 0

0

1 00

2 00

3 00

4 00

-2 0

0

2 0

4 0

6 0

8 0

1 00

1 20

1 2 3 4 5 6 7 8 9 10

Flu

ore

sc

en

ce

[%]

%In

hib

itio

n

41

The 10 fragments identified as hits were re-screened over two surfaces, with unmodified wild type enzyme, and wild type enzyme with the active site blocked with darunavir. This inhibitor was judged to be suitable for blocking the active site since it has an exceptionally slow dissociation rate, 7.8 ⋅ 10-7 s-1 [56]. None of the fragments appeared to compete with da-runavir.

The IC50-values of the fragment interactions with wild type were deter-mined with the same FRET-based assay as used for estimation of inhibitory effect as a single concentration. A steep slope was a characteristic often seen in the IC50-curves, indicating that inhibition may be a result of aggregate formation. To examine if this was the case, the IC50-values for the five frag-ments with highest inhibition potencies were determined also with the addi-tion of 0.5 x CMC (∼0.003%) of Tween-20 in the buffer. The IC50-values were increased and the slope factors decreased for some hits, confirming that inhibition by aggregation was a realistic interpretation (Figure 13 and Figure 14.). With detergent present, the IC50 for hits 2 and 5 was increased to more than 1000 µM and the slope factor was lowered from 2 to 1, indicating a stoichiometry of 1:1. For hits 9 and 10, the IC50-values had the same magni-tude and the slope factor remained between 1 and 1.5 whether detergent was present or not in the buffer. Hit 1 didn’t seem to be very sensitive to deter-gent either. The IC50-value was only increased by one order of magnitude, from 13 µM to 160 µM and the slope factor remained the same. This hit seems to have a multisite interaction mechanism. The results thus suggested that hits 2 and 5 exerted their effect by an aggregation based inhibition mechanism, while hits 9 and 10 did not and maybe hit 1 neither.

Figure 13. Inhibition plots of fragment hits 2 and10, interacting with wild type HIV-1 protease. Data were determined with 0.5 x CMC Tween-20 in the assay buffer (open circles) and without detergent (black squares). The fraction of inhibition vs. concentration data were fitted to a non-linear regression four-parameter model.

Inhibitor [µM]10 100 1000

vi/

v0

0

0.2

0.4

0.6

0.8

1

1.2

Inhibitor [µM]10 100 1000

vi/

v0

0

0.2

0.4

0.6

0.8

1

1.2Hit 2 Hit 10

42

Figure 14. IC50-values and slope factors for the five hits with highest inhibition potency in the primary screen. The values were determined from measurements with and without detergent in the buffer, i.e. 0.5 x CMC Tween-20 in the assay buff-er (open circles) and no detergent (black squares). Values with an asterisk have an IC50 over 1000 µM. The IC50-curves were fitted to a non-linear regression four-parameter model.

Analogues of hits 1, 2, 3, 5, 6, 7, 9 and 10 were bought from commercial vendors and analogues for hits 4 and 8 were synthesized in house. The frag-ment analogues were screened together with the original hits using SPR-biosensor technology. The results were analyzed with the same method as in the primary screen. Only hit 2 was re-identified as a hit. Analogues of frag-ments 1, 2, 4 and 5 were identified as hits. These were the three compounds with suspected aggregation-based inhibition mechanism (2, 4 and 5), and fragment 1 which had a probable multisite interaction. A more thorough investigation of the inhibition mechanism for these analogues is in progress.

The work resulted in three hits from the primary screen, 1, 9 and 10. For the wild type enzyme, the ligand efficiency based on IC50-values were 0.30, 0.42 and 0.40 kcal/mol, and 0.35, 0.51 and 0.55 kcal/mol based on KD-values from the fragment screen. Hits from the secondary screen have not been sufficiently evaluated yet. However, initial results have shown that only one analog of hits 1 and 2 and four analogs of hit 4 may still be identified as hits. It should be noticed that both analogue hits from 1 and 2 has been identified as PAINS (Pan Assay Interference Compounds) in the work by Baell and Holloway [46]. Even if these fragments have turned up frequently as hits for different classes of enzymes, and in that respect could be termed promiscu-ous binders, it may be risky to discard a hit when the inhibition mechanism and the site of interaction to the enzyme is not known. X-ray crystallography may be the best way to investigate the binding site and binding mode.

43

Paper VI: Effect of HIV-1 protease mutations on the temperature dependence of inhibitor dissociation rates The work in paper VI is an extension of the article published 2004 by Shu-man et al. [57], where the interaction between HIV-1 protease and seven inhibitors were kinetically and thermodynamically characterized. The same inhibitors were used in the current study, but the enzyme panel was extended and contained three HIV-1 proteases variants with resistance mutations in addition to the wild type enzyme. The focus in this study was to create a simplified thermodynamic screening method that contained reliable de-scriptors and that accurately described trends. Furthermore, the nature of the resistance patterns and the possibility of describing thermodynamic data in a simple way were of interest. Only the thermodynamics and kinetics of inhib-itor dissociation was examined since resistance is primarily reflected by in-creased dissociation rates [30].

Results paper VI The interaction between seven inhibitors and four HIV-1 protease variants were studied at six temperatures between 10 and 35°C. The data could not be fitted to a 1:1 kinetic model but a heterogeneous surface model was applica-ble. However, this model describes two independent parallel reactions and yields two association and dissociation rates. The analysis method was there-fore simplified by first setting the response level after injection stop for the two parallel reactions to be fixed and equal and then only using the fast dis-sociation rate for describing the temperature effects on the interactions. This gave a satisfyingly good correlation of dissociation rates of 0.93 between the simplified model and a more biphasic model. The dissociation rates for the different temperatures were used for estimation of the thermodynamic con-stants of the interactions. The data gave in general the same trends as seen in the article by Shuman et al. (Figure 15).

Figure 15. Thermodynamic parameters for interactions between wild type HIV-1 and 5 inhibitors evaluated in paper VI (light grey) and by Shuman et al. [57] (dark grey).

Some mutations are more critical for resistance against a certain inhibitor than others and are classified as signature mutations. Another nomenclature dividing the resistance mutation into major and minor mutations has been suggested by the International Antiviral Society–USA [17]. This classifica-

75

80

85

90

95

Amp Ind Lop Nelf Rito Saq

ΔGo

ff(k

J m

ol-1

)

0

10

20

30

40

50

Amp Ind Lop Nelf Rito Saq

ΔHo

ff(k

J m

ol-1

)

0

20

40

60

80

100

Amp Ind Lop Nelf Rito Saq

-TΔS

off

(kJ

mo

l-1)

44

tion defines mutations that give resistance for an inhibitor on their own as major mutations, while minor mutations are mutations that often arise during treatment with a certain inhibitor but do not give resistance on their own.

Table 1. Major and minor mutations for HIV-protease inhibitors in the clinic, freely compiled from Johnson et al. [17]. In the original text the resistance mutations were associated with different drug regimens containing or not containing ritonavir as a booster. Critical resistance mutations for ritonavir are collected from Clemente et al. and Molla et al. [58, 59] and marked by an asterisk.

Inhibitor Major Minor

Amprenavir - V82A, L90M

Indinavir

V82A

L90M

Ritonavir

V82A*, L90M*

Saquinavir

G48V, L90M

V82A

Atazanavir

- G48V,V82A, L90M

Nelfinavir

L90M V82A

Lopinavir V82A L90M

An investigation of the correlations between the temperature dependen-

cies and resistance profiles for the inhibitors was performed in two steps. First, the dissociation rates for all inhibitor-enzyme variants at different tem-peratures were compared in Figure 16. The influence of the temperature on the dissociation rates was found to be higher for enzyme variants with re-sistance mutations compared to wild type enzyme. This agreed well with major and minor mutations for the inhibitors in most cases, but not for the G48V enzyme variant. G48V is a major mutation for saquinavir and a minor mutation for atazanavir. For these two inhibitors, the difference in koff over the temperature range in Figure 16 is higher for this enzyme than for the other enzymes. However, the columns for amprenavir, indinavir ritonavir and nelfinavir are also high for this enzyme, even if the resistance mutation is not critical for these inhibitors.

45

Figure 16. Dissociation rates at different temperatures for interaction between seven inhibitors and four enzyme variants of HIV-1 protease. (1) amprenavir, (2) indinavir, (3) ritonavir, (4) nelfinavir, (5) atazanavir, (6) saquinavir and (7) lopinavir.

The thermodynamic parameters for the dissociation event of the inhibitor-enzyme pairs were estimated, followed by a comparison of the parameters from the interactions with wild type enzyme to enzyme variants with re-sistance mutations (Figure 17). The values for Gibbs free energy indicate less stable inhibitor-enzyme complexes when resistance mutation has been introduced, except for the saquinavir-V82A interaction. The later agrees with the article of Shuman et al. [30], where the complex of V82A-saquinavir had dissociation rates of the same magnitude as the wild type complex, even slightly slower. As for temperature dependence, the results from the free energy of the complexes also agree well with known resistance profiles for the inhibitor-enzyme pairs, except for the enzyme with G48V mutation.

No correlations between the temperature dependence of dissociation rates and enthalpies or entropies could be seen. However there was a certain cor-relation between the type of mutation and the enthalpy and entropy compo-nent of the dissociation energy. The dissociation enthalpy (ΔH) for the en-zyme with the G48V flap mutation was most often slightly higher than for other three enzyme variants (Figure 18). It was the other way around for the entropy component (-TΔS) which was lower than for the other enzymes. This could be interpreted as the G48V mutation restricts motion of the flaps, allowing easier entrance and exit of inhibitors and solvent molecules in the active site. Even if the enzyme-inhibitor binding is stronger, the possibility for exchange of molecules in the active site is increased.

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0 1 2 3 4 5 6 7 8

k off

1(s

-1)

Wild-type

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0 1 2 3 4 5 6 7 8

k off

1(s

-1)

G48V

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0 1 2 3 4 5 6 7 8

k off

1(s

-1)

V82A

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0 1 2 3 4 5 6 7 8

k off

1(s

-1)

L90M

35°C

10°C

Tem peraturerange

46

Figure 17. Change in free energy of dissociation (ΔΔGoff1) between inhibitors inter-acting with wild type (ΔGoff1wild-type) and mutant enzyme (ΔGoff1mutant), according to ΔΔGoff1 = ΔGoff1wild-type - ΔGoff1mutant. The bars represent the enzymes with different resistance mutations: G48V (black), V82A (light grey), and L90M (dark grey). Crit-ical resistance mutations for the inhibitors are marked by a black dot.

Figure 18. Enthalpy and entropy of dissociation for seven inhibitors interacting with wild type HIV-1 protease (grey dotted line, white diamonds) and three enzyme vari-ants with resistance mutations G48V (black line, filled squares), V82A (grey dotted line, light grey triangles) and L90M (grey dotted line, dark grey circles).

In this case the type of mutation has a lager impact on the enthalpy and en-tropy than the type of inhibitor. For ΔG and the temperature dependence, the type of inhibitor in combination with resistance mutation is most important.

-6.0

-5.0

-4.0

-3.0

-2.0

-1.0

0.0

1.0

2.0

Amprenavir Indinavir Ritonavir

Saquinavir

Atazanavir Nelfinavir LopinavirΔΔ

Go

ff1

(kJ

mo

l-1)

51015202530354045

∆Hoff1(kJm

ol-1)

404550556065707580

-T*∆

S off1(kJm

ol-1)

47

HIV-protease resistance drug discovery – papers IV, V and VI In paper IV a fragment library was designed and validated for SPR-based fragment screening. This library was then used in paper V to detect frag-ments interacting with, and having enough inhibitory potency for both wild type enzyme and resistance mutants. In paper IV, the 10 hits found in the primary screen and hits 1, 2, 3, 7 and 8 from paper V were identified as hav-ing a super-stoichiometric interaction for wild type HIV-1. Hit 2 also dis-played a slow dissociation for this enzyme. These findings were essentially supported for hits 1 and 2 in paper V, where hit 2 had the characteristics of an aggregation-based inhibition mechanism and hit 1 seemed to interact with several targets.

For hits 4 and 5, the results from paper IV were not supported by the stud-ies in paper V. These two fragments seemed to have an aggregation-based inhibition mechanism in paper V, but had no such indications in paper IV. Fragments displaying a super-stoichiometric interaction for a certain enzyme may raise the suspicion of micelle formation. However, this was not seen for fragments 4 or 5 in paper IV. This might be explained by the low binding capacity of the HIV-protease surface, probably resulting from partially de-graded and unfolded enzyme.

Hits identified in paper V were often promiscuous, showing super-stoichiometric interactions with HIV-1 protease. The reason for this behavior may be an aggregation-based inhibition mechanism or multisite interactions. The identification of these compounds as hits may be explained on the basis of the selection criteria. Only compounds interacting with wild type enzyme and 2 or 3 mutant variants were chosen, and the estimation of affinities ac-counted for non-specific interactions. These criteria may have resulted in an enrichment of promiscuous interacting fragments.

No active site interacting hits were found. This leads to the suspicion that low affinity compounds interacting with the highly dynamic active site of HIV-1 protease might be difficult to detect [14, 60], at least when using SPR-biosensor technology. In the X-ray crystallography based HIV-1 frag-ment screen published by Perryman et al. [61], this problem was not seen since they stabilized the enzyme by using an active site inhibitor when crys-tallizing the protease. The group found two fragments with allosteric regula-tion of the activity of HIV-protease [16]. The difficulties in detecting frag-ments interaction with highly flexible targets is highlighted in a paper by Brandt et al. [62]. In this paper non-nucleoside reverse transcriptase inhibi-tors (NNRTIs) were deconstructed into fragments and evaluated for binding. It was found that it was difficult to identify the fragments as hits for HIV-1 RT. The inhibitor had to be full size in order to induce the conformational change required for inhibition. This insight might be applicable to HIV-1 protease also since HIV protease inhibitors interact both with the highly

48

flexible lid of the entrance to the active site as well as the flexible active site residues.

A parallel to this reasoning can be seen in the results from Figure 18 in paper VI, showing the enthalpy and entropy component of the free energy of dissociation. The G48V mutation in the flaps of the protease has a different thermodynamic profile compared to the wild type, the active site mutation or the dimerization border mutation. The G48V mutation might thus have a significant effect on the flexibility of the flaps. This was not seen as a lower or higher hit rate for this enzyme variant compared to the others in paper V. However, of the 10 hits from the primary screen none of the fragments were seen to compete with darunavir for binding in the active site, indicating inhi-bition of allosteric sites on the enzyme. All 10 hits may interact in a less flexible area of the enzyme.

49

Summary and future perspectives

Fragment based drug discovery is a new approach in the quest for more se-lective drugs with higher resistance thresholds. Surface plasmon resonance technology is well suited to be adopted both for screening in the beginning of drug discovery campaigns, and for the more detailed characterization studies in lead optimization. The papers in this thesis have been produced during the years from when SPR-biosensor technology in FBDD was seen with suspicion to now when it has become one of the most used techniques. The method is easy to use and adaptable to fulfill the unique environmental needs of many drug targets. However, fragment based drug discovery pushes the technique to the limit. But since the sensitivity of the instruments has increase during the past years, even very low affinity interactions can be detected.

The papers presented in this thesis give examples of how to use the meth-od in order to set up new assays for fragment screening, analyze the results and characterize hits and leads (paper II and V). Special concern has been devoted to how to detect potentially problematic compounds and the follow up of potential hits before they are dismissed due to unexpected or non-typical interaction characteristics. The apparent problematic interaction might reveal a valuable hit (paper III). Furthermore, no FBDD technique should be used alone, but always together with an orthogonal technique. Surface plasmon resonance has a central role since it can be used to analyze kinetic and mechanistic properties of the interaction.

For a more complete understanding of the interaction it is vital to know the characteristics of potential target binding sites and which residues are involved. Are polar or hydrophobic interactions involved in the binding? Does the compound have inhibition potency, or does the compound interact with a site that does not affect the enzymatic activity? Techniques to answer these questions may be X-ray crystallography, NMR and inhibition assays. Future crystallographic studies of complexes between HIV-1 protease and the potentially allosteric inhibitors found in paper V will hopefully reveal the location of the binding sites and the binding modes of the fragments. Allo-steric inhibitors against HIV-1 protease are expected to benefit the struggle against AIDS by bringing a new category of inhibitors with different proper-ties into the arsenal.

New inhibitors for MMP-12 are also of interest since MMP-12 has been recognized as a drug target for COPD. The results of the work in paper III

50

will be useful for the development of new inhibitors against this target. The phenylhydantoin adduct may be used as a starting compound for design of new inhibitors with high selectivity.

The best way to optimize hits and leads in drug discovery is still under evaluation. The most used strategies involve optimizing compounds for high affinity, but slow dissociation rates and enthalpy driven interactions have more recently become commonly used optimization criteria. However, as seen in paper VI the picture may be more complicated than presumed by the advocates of these approaches. There are still questions concerning how to best use our tools of analysis for a deeper understanding of the biological response to a drug to be solved. Nature is complicated; I look forward to future publications in this area of drug discovery.

51

Svensk sammanfattning (Summary in swedish)

Enzymer i läkemedelsutveckling De allra flesta läkemedel riktas mot enzymer. Enzymer är protein som fun-gerar som kroppens katalysatorer. Reaktioner som skulle ta väldigt lång tid okatalyserade blir mycket snabbare med hjälp av enzymer. Våra kroppar skulle inte fungera om vi inte hade enzymer. Dessa används i alla kemiska reaktioner i våra kroppar som exempelvis vid nedbrytning av mat till energi, vid tankeprocesser i hjärnan, vid rörelser av muskler i armar, ben, hjärta och tarmar med mera.

Den här avhandlingen handlar om läkemedelsutveckling mot enzymerna matrix metalloproteas-12 (MMP-12) och HIV-proteas. Bådas övergripande funktion är att klippa sönder andra protein.

Läkemedelsutveckling Läkemedelsutveckling är en mycket resurskrävande och dyr process. Därför behövs metoderna utvecklas och effektiviseras. I de tidiga faserna identifie-ras ett målenzym vars aktivitet ska inhiberas (hämmas) för att bota eller mildra sjukdomen. Därefter börjar sökandet efter en molekyl med inhibe-ringspotential mot enzymet. Interaktionen mellan målenzymet och stora samlingar av molekyler, så kallade bibliotek, undersöks. Lovande substanser identifieras. Även identifieringsmetoderna behöver förbättras. Identifierade substanser utvecklas vidare för att öka den inhiberande förmågan.

MMP-12 – ett målenzym för behandling av kronisk obstruktiv lungsjukdom MMP-12 tillhör en stor familj matrix metalloproteaser (MMP) som bryter ner olika komponenter i bindväv. Bindväv består till exempel av kollagen, gelatin och elastin. Bindväv finns mellan våra organ och mellan olika delar inom organ. Det stöder, separerar och sammanlänkar olika vävnader. Exem-pel är senor, blodkärl och brosk. MMP enzymer behövs för vävnadsomsätt-ningen i kroppen exempelvis då kroppen växer eller vid sårläkning.

52

MMP-12 har stor inverkan på lungornas hälsa. Enzymet bryter ned elastin i cellväggarna hos lungornas alveoler. Det har visat sig att MMP-12 kan vara ett bra målenzym for behandling av sjukdomen Kronisk Obstruktiv Lung-sjukdom (KOL). Kol utvecklas främst genom lång tids exponering mot ciga-rettrök. Cigarettröken orsakar inflammation i lungvävnaden. MMP-12 aktiv-eras och lungorna bryts ner och förlorar sin syreupptagningsförmåga. Djur-försök på möss har indikerat att inhibering av MMP-12s katalytiska förmåga kan förhindra nedbrytningen av alveolerna.

HIV-proteas – ett målenzym för behandling av AIDS HIV-viruset invaderar celler tillhörande immunförsvaret. Det är ett så kallat retrovirus vilket använder värdcellerna för tillverkning av nya viruspartiklar. Värdcellen förmås att syntetiserar molekylära komponenter, (aminosyraked-jor) som bygger upp virusets protein. Aminosyrakedjor framställs som sedan klipps av på specificerade ställen av HIV-proteaset så att nya virala proteiner kan bildas och sättas samman till nya virus.

Läkemedel som inhiberar HIV-protease ingår i den arsenal av olika typer av läkemedel en HIV-drabbad person får för att förhindra utvecklingen av AIDS. Ett stort problem är att resistans mot läkemedlen uppkommer väldigt fort. Detta orsakar ett kontinuerligt behov av nya läkemedel.

Min forskning I den här avhandlingen har en metod använts där bibliotek med mycket små molekyler som ger svaga interaktioner med målenzymet, genomsökts efter intressanta molekyler. Tidigare genomsökningsmetoder har använt moleky-ler som varit lika stora som färdiga läkemedel och bundit starkt till må-lenzymet. Nackdelen med den metoden är att vidareutvecklingen av lovande molekyler har tenderat till att ge läkemedelskandidater med bieffekter. För-hoppningarna är att optimeringen av de små molekylerna ska ge läkemedels-kandidater med mindre biverkningar. Lovande resultat har redan visats för läkemedelsutveckling med denna metod.

Den teknik som använts för att genomsöka dessa bibliotek med små mo-lekyler är baserad på biosensorer där detektionen av interaktioner sker ge-nom fenomenet ytplasmonresonans. Målenzymet sätts fast på ytan av bio-sensorn och de små molekylerna flödas över ytan, en efter en. Om en mole-kyl binder ger det en signal som detekteras och analyseras för att hitta loven-de molekyler.

Delarbeten I artikel I beskrivs identifieringen av lämpliga egenskaper hos inhibitorer för MMP-12. Inhibitorer med förmåga att binda starkt till zink atomen i må-

53

lenzymets aktiva yta bedömdes vara olämpliga. I artikel II, genomsöks ett bibliotek av små molekyler mot MMP-12 och i artikel III analyseras de lo-vande föreningarna ytterligare. I dessa analyser identifierades en ny förening som kan vara en bra utgångsstruktur vidareutveckling till ett läkemedel.

Den första artikeln om HIV-1 proteas, artikel IV, behandlar genomsök-ningen av ett nytt bibliotek och detta biblioteks sammansättning. I artikel V används biblioteket för genomsökning. Metoden som utvecklades tog hänsyn till resistans hos HIV-proteaset. De lovande föreningar som hittades uppvi-sade en annan inhiberingsmekanism än de läkemedel mot HIV som används inom vården. Dessa kan vara utgångspunkter för en ny typ av inhibitorer för HIV. Slutligen, i artikel VI undersöks hur interaktionen med läkemedel på-verkas av resistenta varianter av HIV-proteaset. Vi visade att resistans mins-kade stabiliteten hos inhibitor-proteas komplexet. Inhibitorn lossnade från proteaset snabbare om proteaset hade resistensmutationer. En undersökning gjordes om denna stabilitetsförsämring beskrevs bäst genom dissociations-hastigheten eller den sammanhållande energin inom komplexet.

Slutsatser Resultaten i artiklarna kommer att underlätta de inledande faserna inom lä-kemedelsutvecklingen. Detta är ett steg i riktningen mot snabbare utveckling av effektiva läkemedel till en lägre kostnad. Resultaten visar också på vikten av förbättrade metoder för att välja ut lovande molekyler. Den sista artikeln är ett inlägg i den pågående diskussionen om vilka egenskaper en lovande molekyl ska ha för att bedömas vara värd att vidareutveckla.

54

Acknowledgements

First of all, I would like to thank my supervisor Helena Danielson for all your support, optimism and inspiration. It is hard to find a better, more com-petent supervisor and scientific mentor.

I would also like to thank my co-supervisor Gunnar Johansson who has giving me helpful and inspirational discussions on thermodynamics. Our lunches have been a joy. I have met many nice and funny persons during my time in the research group, both past and present members of the HD group, PhD-students and master students. The list on all people who have enlightened my days at the institution is very long, and all of them deserve special thanks. It has been a pleasure to work with you! I would like to thank my roommates: Malin - for being a caring person and great friend and having lots of energy in the works we have done together. Johan - for always lending a helping hand, being a great support in theoreti-cal discussions and a splendid co-author. Eldar - someone you always can count on. Angelica - no one can tell a story like you. Sofia - for your skills in arranging social events for the group. Christian and Dirk - for being so nice and easy to talk with and for reminding me of lunch time. Hilde - a girl with a warm heart. Other past and present members of the HD group Sara - It was great working side by side with you in the lab. Thank you for all your help and friendship. Thomas, Matthis and Maria - for the work on articles we have done together and for the fun we have had on’ fika’ and disputation dinners. Göran – for giving a friendly and enjoyable atmosphere to the group Past and present members of Mikael Widerstens and Bengt Mannerviks groups. For all the reviving lunches with lots of laughter Diana, Ann, Cissi, Åsa, Emil, Birgitta, Natalia and Malena also for helping me out with teaching,

55

Gun- I have enjoyed your company both in the laboratory and on disputation dinners. Francoise- for help on administrative issues and helpful advices concerning teaching. Ylva for joining me at the institution working nights and weekends.during the thesis work and for helping me with the ‘Swedish summary’. My co-authors from outside of the HD group: Markku Hämäläinen al-ways generously sharing his knowledge about the SPR-biosensor technology and its applications. Dean Derbyshire - for your wonderful work with the X-ray structure in pa-per III. Hans Wallberg, Per Källblad, Susanne Nyström – for your work with the papers involving MMP-12. Per Dinér, Annica Önell and Cynthia Shu-man – for your work with papers on HIV-1 protease. Vänner, även om det kan gå lång tid mellan de gånger vi träffas, tänker jag ofta på er Åsa, Tobias, Iris, Linda, Malin Mina föräldrar för ert stöd och kärlek genom hela livet. Ni har låtit mig utvecklas till den jag är och har alltid ställt upp för mig. Min syster Cia, du har alltid kommit till min hjälp när jag behövt dig. En toppensyrra! Min familj, mina barn Eira och Egil för era skratt och gränslösa kärlek. Ni får mig att se framåt. Håkan för att du har fått stå ut med mycket under min doktorandtid, främst vid avhandlingsarbetet. Jag älskar er.

56

References

1. Freire, E., Do enthalpy and entropy distinguish first in class from

best in class? Drug Discov Today, 2008. 13(19-20): p. 869-74. 2. Ladbury, J.E., G. Klebe, and E. Freire, Adding calorimetric data to

decision making in lead discovery: a hot tip. Nat Rev Drug Discov, 2010. 9(1): p. 23-7.

3. Ferenczy, G.G. and G.M. Keseru, Thermodynamics guided lead discovery and optimization. Drug Discov Today, 2010. 15(21-22): p. 919-32.

4. Copeland, R.A., D.L. Pompliano, and T.D. Meek, Drug-target residence time and its implications for lead optimization. Nat Rev Drug Discov, 2006. 5(9): p. 730-9.

5. Tummino, P.J. and R.A. Copeland, Residence time of receptor-ligand complexes and its effect on biological function. Biochemistry, 2008. 47(20): p. 5481-92.

6. Hu, J., et al., Matrix metalloproteinase inhibitors as therapy for inflammatory and vascular diseases. Nat Rev Drug Discov, 2007. 6(6): p. 480-98.

7. Churg, A., S. Zhou, and J.L. Wright, Series "matrix metalloproteinases in lung health and disease": Matrix metalloproteinases in COPD. Eur Respir J, 2012. 39(1): p. 197-209.

8. Morales, R., et al., Crystal structures of novel non-peptidic, non-zinc chelating inhibitors bound to MMP-12. J Mol Biol, 2004. 341(4): p. 1063-76.

9. Gossas, T. and U.H. Danielson, Characterization of Ca2+ interactions with matrix metallopeptidase-12: implications for matrix metallopeptidase regulation. Biochem J, 2006. 398(3): p. 393-8.

10. Brown, E.M. and R.J. MacLeod, Extracellular calcium sensing and extracellular calcium signaling. Physiol Rev, 2001. 81(1): p. 239-297.

11. Hautamaki, R.D., et al., Requirement for macrophage elastase for cigarette smoke-induced emphysema in mice. Science, 1997. 277(5334): p. 2002-4.

12. Kaplan, A.H., et al., Partial inhibition of the human immunodeficiency virus type 1 protease results in aberrant virus

57

assembly and the formation of noninfectious particles. J Virol, 1993. 67(7): p. 4050-5.

13. Weber, I.T. and J. Agniswamy, HIV-1 Protease: Structural Perspectives on Drug Resistance. Viruses, 2009. 1(3): p. 1110-36.

14. Freedberg, D.I., et al., Rapid structural fluctuations of the free HIV protease flaps in solution: relationship to crystal structures and comparison with predictions of dynamics calculations. Protein Sci, 2002. 11(2): p. 221-32.

15. Wensing, A.M., N.M. van Maarseveen, and M. Nijhuis, Fifteen years of HIV Protease Inhibitors: raising the barrier to resistance. Antiviral Res, 2010. 85(1): p. 59-74.

16. Tiefenbrunn, T., et al., Small Molecule Regulation of Protein Conformation by Binding in the Flap of HIV Protease. ACS Chem Biol, 2013.

17. Johnson, V.A., et al., Update of the drug resistance mutations in HIV-1: March 2013. Top Antivir Med, 2013. 21(1): p. 6-14.

18. Markgren, P.O., M. Hamalainen, and U.H. Danielson, Kinetic analysis of the interaction between HIV-1 protease and inhibitors using optical biosensor technology. Anal Biochem, 2000. 279(1): p. 71-8.

19. Kawasaki, Y. and E. Freire, Finding a better path to drug selectivity. Drug Discov Today, 2011. 16(21-22): p. 985-90.

20. Zhukov, A. and R. Karlsson, Statistical aspects of van't Hoff analysis: a simulation study. J Mol Recognit, 2007. 20(5): p. 379-85.

21. Horn, J.R., et al., Van't Hoff and calorimetric enthalpies from isothermal titration calorimetry: are there significant discrepancies? Biochemistry, 2001. 40(6): p. 1774-8.

22. Wear, M.A. and M.D. Walkinshaw, Thermodynamics of the cyclophilin-A/cyclosporin-A interaction: a direct comparison of parameters determined by surface plasmon resonance using Biacore T100 and isothermal titration calorimetry. Anal Biochem, 2006. 359(2): p. 285-7.

23. Winzor, D.J. and C.M. Jackson, Interpretation of the temperature dependence of rate constants in biosensor studies. Anal Biochem, 2005. 337(2): p. 289-93.

24. Biacore, Biacore sensor surface handbook-version AA. 2003. 25. Biacore, BIAapplications Handbook, version AB. 1998: Biacore AB. 26. Copeland, R.A., Enzymes. 2000, Wiley-VCH. 27. Piston, D.W. and G.J. Kremers, Fluorescent protein FRET: the

good, the bad and the ugly. Trends Biochem Sci, 2007. 32(9): p. 407-14.

28. Matayoshi, E.D., et al., Novel fluorogenic substrates for assaying retroviral proteases by resonance energy transfer. Science, 1990. 247(4945): p. 954-8.

58

29. Weiss, J.N., The Hill equation revisited: uses and misuses. FASEB J, 1997. 11(11): p. 835-41.

30. Shuman, C.F., et al., Elucidation of HIV-1 protease resistance by characterization of interaction kinetics between inhibitors and enzyme variants. Antiviral Res, 2003. 58(3): p. 235-42.

31. Shuman, C.F., L. Vrang, and U.H. Danielson, Improved structure-activity relationship analysis of HIV-1 protease inhibitors using interaction kinetic data. J Med Chem, 2004. 47(24): p. 5953-61.

32. Copeland, R.A., The dynamics of drug-target interactions: drug-target residence time and its impact on efficacy and safety. Expert Opin Drug Discov, 2010. 5(4): p. 305-10.

33. Lange, J.H., et al., Biophysical and physicochemical methods differentiate highly ligand-efficient human D-amino acid oxidase inhibitors. Eur J Med Chem, 2011. 46(10): p. 4808-19.

34. Rees, D.C., et al., Fragment-based lead discovery. Nat Rev Drug Discov, 2004. 3(8): p. 660-72.

35. Jencks, W.P., On the attribution and additivity of binding energies. Proc Natl Acad Sci U S A, 1981. 78(7): p. 4046-50.

36. Shuker, S.B., et al., Discovering high-affinity ligands for proteins: SAR by NMR. Science, 1996. 274(5292): p. 1531-4.

37. Nienaber, V.L., et al., Discovering novel ligands for macromolecules using X-ray crystallographic screening. Nat Biotechnol, 2000. 18(10): p. 1105-8.

38. Hennig, M., A. Ruf, and W. Huber, Combining biophysical screening and X-ray crystallography for fragment-based drug discovery. Top Curr Chem, 2012. 317: p. 115-43.

39. Davis, B.J. and D.A. Erlanson, Learning from our mistakes: the 'unknown knowns' in fragment screening. Bioorg Med Chem Lett, 2013. 23(10): p. 2844-52.

40. Lipinski, C.A., et al., Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv Drug Deliv Rev, 2001. 46(1-3): p. 3-26.

41. Congreve, M., et al., A 'rule of three' for fragment-based lead discovery? Drug Discov Today, 2003. 8(19): p. 876-7.

42. Hopkins, A.L., C.R. Groom, and A. Alex, Ligand efficiency: a useful metric for lead selection. Drug Discov Today, 2004. 9(10): p. 430-1.

43. Leeson, P.D. and B. Springthorpe, The influence of drug-like concepts on decision-making in medicinal chemistry. Nat Rev Drug Discov, 2007. 6(11): p. 881-90.

44. Hann, M.M. and G.M. Keseru, Finding the sweet spot: the role of nature and nurture in medicinal chemistry. Nat Rev Drug Discov, 2012. 11(5): p. 355-65.

45. Scott, D.E., et al., Fragment-based approaches in drug discovery and chemical biology. Biochemistry, 2012. 51(25): p. 4990-5003.

59

46. Baell, J.B. and G.A. Holloway, New substructure filters for removal of pan assay interference compounds (PAINS) from screening libraries and for their exclusion in bioassays. J Med Chem, 2010. 53(7): p. 2719-40.

47. Roche, O., et al., Development of a virtual screening method for identification of "frequent hitters" in compound libraries. J Med Chem, 2002. 45(1): p. 137-42.

48. Giannetti, A.M., B.D. Koch, and M.F. Browner, Surface plasmon resonance based assay for the detection and characterization of promiscuous inhibitors. J Med Chem, 2008. 51(3): p. 574-80.

49. Elinder, M., et al., Experimental validation of a fragment library for lead discovery using SPR biosensor technology. J Biomol Screen, 2011. 16(1): p. 15-25.

50. May, P.C., et al., Robust central reduction of amyloid-beta in humans with an orally available, non-peptidic beta-secretase inhibitor. J Neurosci, 2011. 31(46): p. 16507-16.

51. Shoichet, B.K., Interpreting steep dose-response curves in early inhibitor discovery. J Med Chem, 2006. 49(25): p. 7274-7.

52. Ryan, A.J., et al., Effect of detergent on "promiscuous" inhibitors. J Med Chem, 2003. 46(16): p. 3448-51.

53. Feng, B.Y. and B.K. Shoichet, A detergent-based assay for the detection of promiscuous inhibitors. Nat Protoc, 2006. 1(2): p. 550-3.

54. Jadhav, A., et al., Quantitative analyses of aggregation, autofluorescence, and reactivity artifacts in a screen for inhibitors of a thiol protease. J Med Chem, 2010. 53(1): p. 37-51.

55. Wielens, J., et al., Parallel screening of low molecular weight fragment libraries: do differences in methodology affect hit identification? J Biomol Screen, 2013. 18(2): p. 147-59.

56. Dierynck, I., et al., Binding kinetics of darunavir to human immunodeficiency virus type 1 protease explain the potent antiviral activity and high genetic barrier. J Virol, 2007. 81(24): p. 13845-51.

57. Shuman, C.F., M.D. Hamalainen, and U.H. Danielson, Kinetic and thermodynamic characterization of HIV-1 protease inhibitors. J Mol Recognit, 2004. 17(2): p. 106-19.

58. Molla, A., et al., Ordered accumulation of mutations in HIV protease confers resistance to ritonavir. Nat Med, 1996. 2(7): p. 760-6.

59. Clemente, J.C., et al., Comparing the accumulation of active- and nonactive-site mutations in the HIV-1 protease. Biochemistry, 2004. 43(38): p. 12141-51.

60. Heaslet, H., et al., Conformational flexibility in the flap domains of ligand-free HIV protease. Acta Crystallogr D Biol Crystallogr, 2007. 63(Pt 8): p. 866-75.

60

61. Perryman, A.L., et al., Fragment-based screen against HIV protease. Chem Biol Drug Des, 2010. 75(3): p. 257-68.

62. Brandt, P., M. Geitmann, and U.H. Danielson, Deconstruction of non-nucleoside reverse transcriptase inhibitors of human immunodeficiency virus type 1 for exploration of the optimization landscape of fragments. J Med Chem, 2011. 54(3): p. 709-18.

Acta Universitatis UpsaliensisDigital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Science and Technology 1087

Editor: The Dean of the Faculty of Science and Technology

A doctoral dissertation from the Faculty of Science andTechnology, Uppsala University, is usually a summary of anumber of papers. A few copies of the complete dissertationare kept at major Swedish research libraries, while thesummary alone is distributed internationally throughthe series Digital Comprehensive Summaries of UppsalaDissertations from the Faculty of Science and Technology.

Distribution: publications.uu.seurn:nbn:se:uu:diva-209136

ACTAUNIVERSITATIS

UPSALIENSISUPPSALA

2013