A new strategy for improved secondary screening and lead optimization using high-resolution SPR characterization of compound–target interactions

JOURNAL OF MOLECULAR RECOGNITIONJ. Mol. Recognit. 2005; 18: 273–281Published online in Wiley InterScience (www.interscience.wiley.com). DOI:10.1002/jmr.744

Review

A new strategy for improved secondary screeningand lead optimization using high-resolution SPRcharacterization of compound–target interactions

Walter Huber*F. Hoffman-La Roche Ltd, Basel, Switzerland

Biophysical label-free assays such as those based on SPR are essential tools in generating high-quality dataon affinity, kinetic, mechanistic and thermodynamic aspects of interactions between target proteins andpotential drug candidates. Here we show examples of the integration of SPR with bioinformatic approachesand mutation studies in the early drug discovery process. We call this combination ‘structure-basedbiophysical analysis’. Binding sites are identified on target proteins using information that is eitherextracted from three-dimensional structural analysis (X-ray crystallography or NMR), or derived from apharmacore model based on known binders. The binding site information is used for in silico screening of alarge substance library (e.g. available chemical directory), providing virtual hits. The three-dimensionalstructure is also used for the design of mutants where the binding site has been impaired. The wild-typetarget and the impaired mutant are then immobilized on different spots of the sensor chip and theinteractions of compounds with the wild-type and mutant are compared in order to identify selective bindersfor the binding site of the target protein. This method can be used as a cost-effective alternative to high-throughput screening methods in cases when detailed binding site information is available. Here, we presentthree examples of how this technique can be applied to provide invaluable data during different phases ofthe drug discovery process. Copyright # 2005 John Wiley & Sons, Ltd.

Keywords: surface plasmon resonance; biosensor; protein interaction analysis; drug discovery; direct binding assay;Biacore

Received 15 March 2005; revised 4 April 2005; accepted 2 May 2005

INTRODUCTION

The identification of high-quality hits and lead compoundsis crucial in the drug discovery process. An understanding ofstructure–activity relationships and mechanistic aspects ofaction are key in selecting the best chemical class for furtheroptimization. It is therefore essential to obtain sufficienthigh-quality data on affinity, kinetic, mechanistic and ther-modynamic aspects of an interaction between potential drugcandidates and their targets.

High-throughput screening assays are routinely used forthe identification of compounds which bind to the target,although a very large proportion of the hits identified arefalse positives that do not bind to the target binding site(Xiang et al., 2003; Gribbon and Sewing, 2003). Com-pounds produce false positive results for a number ofreasons, including interference with the labelling method(fluorescence quenching), binding to the substrate or general(promiscuous) protein binding. There are also situations

when high-throughput screening (HTS) is not applicable,such as when new target proteins have been identified fromtheir similarity with established target proteins, but nonatural binders (e.g. substrate) have been identified, thusmaking an inhibition-based labelling approach impossible.Similarly, HTS may be hindered if the substrate is known,but is not stable. This is often the case for isomerases, andwas also true for the hydroxymethyl-pterin pyrophosphoki-nase (HPPK) study we report here.

Biophysical binding assays are essential tools in generat-ing label-free, high-quality data on the interaction betweentarget proteins and potential drug candidates. Owing to theirmoderate throughput and/or high target protein and com-pound consumption, these assays are generally only suitedto characterization of a restricted number of compounds andare therefore not suited to systematic secondary screening ofhits from HTS. Label-free biophysical screening methodsinclude isothermal titration calorimetry (ITC), analyticalultrafiltration, nuclear magnetic resonance, mass spectro-metry, size exclusion chromatography, electrophoresis andbiosensors.

Direct binding assays based on surface plasmon reso-nance (SPR) protein interaction analysis technology (Fig. 1)have several advantages in comparison with other biophy-sical screening methods. The major advantages are lowtarget consumption (a few mg protein/microplate com-pounds), the possibility to simultaneously use reference

Copyright # 2005 John Wiley & Sons, Ltd.

*Correspondence to: W. Huber, Hoffman-La Roche Ltd, Pharmaceutical Re-

search, Discovery Chemistry, CH-4070 Basel, Switzerland.

E-mail: [email protected]

Abbreviations used: AUC, analytical ultra centrifugation; DHNA, dihydro-

neopterin-aldolase; DPP-IV, dipeptidylpeptidase IV; HPPK, hydroxymethyl-

pterin pyrophosphokinase; HTS, high-throughput screening; ITC, isothermal

titration calorimetry; SBBA, structure-based biophysical analysis; SPR, surface

plasmon resonance.

proteins (to discriminate between selective and promiscuousbinders) and that it resolves affinity into on/off-rate con-stants, improving structure–activity relationship studies.The study of direct binding between low molecular weightcompounds and target molecules was developed by HelenaDanielson at Uppsala University, Sweden in collaborationwith scientists at Biacore, AB (Markgren et al., 1998, 2000,2001, 2002; Alterman et al., 2001; Hamalainen et al., 2000;Karlsson et al., 2000). The use of optical biosensors in drugdiscovery has recently been reviewed (Cooper, 2002,2003a,b; Myszka and Rich, 2000; Rich and Myszka,2004; Karlsson, 2004; Lofas, 2004). Direct binding SPRdata on affinity, kinetics and thermodynamics have beenpublished for a number of different drug target systems,including HIV-1 protease (Shuman et al., 2003, 2004a,b),

CMV protease (Geitmann and Danielson, 2004), HIV TAR-RNA (Davis et al., 2004), p38 MAP kinase (Davidson et al.,2004; Bukhtiyarova et al., 2004; Casper et al., 2004),oestrogen receptor alpha and beta (Rich et al., 2002),carbonic anhydrase (Myszka, 2004), PPAR (Yu et al.,2004), DNA/telomerase (Tedious et al., 2004; Carrasco etal., 2002, 2003; Harrison et al., 2003; Bailly et al., 2003),MDM2-p53 (Vassiljev et al., 2004), (pp60)Src (Lange et al.,2003) and VEGF (Ueda et al., 2003).

Despite these reports, the true value of resolving affinityinto kinetic constants has not been fully recognized. This ismainly because kinetic data have not previously been avail-able for larger series of lead compounds and for differentscaffold classes. A study of the literature reveals, however,that concrete examples of a critical role for the kineticresolution of drug–target binding properties have been avail-able for more than a decade. In the study of saxitoxins (Hallet al., 1990), the authors found that it was not possible tounderstand structure–activity relationship before they hadresolved the affinity into binding kinetics. They showed thatan increased charge state of the toxins correlated withincreases in on-rate. Elg et al. (1997) found that, forthrombin inhibitors, the on-rate correlated much better thanaffinity with the in vivo pharmacological effect in a rat bloodcoagulation model. This means that, for thrombin inhibitors,using affinity as a measure of binding strength is misleadingfrom the critical perspective of pharmacological effect.Markgren et al. (2002) showed in their structure–activityrelationships (SAR) study for HIV-1-protease that scaffoldsclustered in the on/off-rate/KD-map and that the affinity ofsome of the more structurally rigid cyclic scaffold werelimited by the instability of the complex formed (rapid off-rate). They could also show that compounds with virtuallyidentical affinities could have off/on-rate ratios that differedby three orders of magnitude. Kroogsgardt et al. (2004)showed that off-rate and heat capacity are responsible for theT cell activation, not the affinity.

In this paper, we show three examples in which bindingsite-modified proteins were used as reference targets indirect binding SPR assays aimed at identifying compoundswith true target protein binding site specificity. Thismodification is achieved either by point mutations (hydro-xymethyl-pterin pyrophosphokinase, HPPK; dihydroneop-terin-aldolase, DHNA) or by chemically blocking the activesite with covalent inhibitors (dipeptidylpeptidase IV, DPP-IV). The examples presented represent different steps of thedrug discovery process. The HPPK example shows the valueof the technique for hit finding, since no HTS could beperformed for this target due to the lack of a stable substrate.The process applied in this case is called ‘structure-basedbiophysical analysis’ (SBBA), an approach which was firstapplied to the study of DNA gyrase (Boehm et al., 2000).SBBA is based on the use of structural information forsetting up a focused library and for directing mutations ofthe binding site of the wild-type (wt) target protein, enablingthe simultaneous use of the wt and the mutant for parallelanalysis in SPR studies. Two additional examples (DPP-IVand DHNA) are also presented, in which SBBA is used forlead selection and optimization during later stages of the drugdiscovery process. For DPP-IV, kinetic data for 160 com-pounds from three structural classes are used for the selec-tion/prioritization of scaffolds during lead optimization. For

Figure 1. Biacore systems monitor molecular interactions inreal-time using a label-free detection method. (a) In the SBBAassays, wt and binding site mutant target proteins are immobi-lized onto different regions of a sensor surface, while com-pounds are injected in solution and flow over the sensorsurface. When injected compounds bind to the wt protein, thisresults in an alteration in refractive index that is proportional tothe change in mass concentration at the surface. (b) Using thephenomenon of surface plasmon resonance (SPR), thesechanges are detected in real time and data is presented as asensorgram (SPR response plotted against time). These plotsdisplay the formation and dissociation of complexes over theentire course of an interaction, with the kinetics (association anddissociation rates) revealed by the shape of the binding curve.The sensorgram provides real-time information about the entireinteraction, with binding responses measured in resonanceunits (RU). Binding responses at specific times during the inter-action can also be selected as report points.

274 W. HUBER

Copyright # 2005 John Wiley & Sons, Ltd. J. Mol. Recognit. 2005; 18: 273–281

DHNA, SPR-data are used for analysing mechanistic aspectsof inhibition and biological effects.

EXPERIMENTAL

The SPR studies were performed with Biacore 2000(DHNA), Biacore 3000 (HPPK) and Biacore S51 (DPP-IV) instruments. The target proteins were immobilized usingstandard amine coupling using EDC/NHS coupling chem-istry and CM5 sensor chips, or using biotinylated targetproteins and streptavidin-coated SA sensor chips. HBS-EPor phosphate buffer saline pH 7.4 were used as running andsample buffers. The DMSO content in sample and runningbuffers varied between 2 and 5%, as appropriate.

Structure-based biophysical analysis

In bioinformatic approaches, three-dimensional structuralinformation from X-ray crystallography or NMR and muta-tion studies are used to identify a binding site on the targetprotein. Binding site analysis leads to the generation of a‘pharmacophore hypothesis’, which in principle outlinespotential interactions between the binding site and a smallligand molecule. This pharmacophore hypothesis is used foran in silico pre-screening of large libraries to establish afocused library of a few hundred compounds. This stepcircumvents the need for an HTS assay.

The pharmacophore hypothesis can also be used to designmutants with altered binding sites. In general, these involvepoint mutations producing a change in one or two of theamino acids located in the binding site. These amino acidsare assumed to be important in any proposed bindingmechanism, or pharmacophore hypothesis, and differentbinding patterns are expected for wild-type and mutantproteins. Once the focused library has been established, itcan be tested with a biophysical screen using the wild typeand mutant proteins.

SBBA of HPPK—virtual screening and SPR

As no biological assay was available for HPPK, a focusedlibrary was generated, using in silico screening, and a directbinding assay had to be developed to screen these com-pounds for HPPK binding. The X-ray structure of a proteinsubstrate analogue complex was determined in-house (Hen-nig et al., 1999) and the identification of the binding site andgeneration of a pharmacophore hypothesis was straightfor-ward. The most prominent features of the binding siteincluded polar interactions between the pterin ring of thesubstrate and asparagine 56, serine 43, lysine 44 and leucine46, and non-polar interactions with phenylalanine 123 andtyrosine 54 (Fig. 2).

Two-hundred and fifty compounds were selected from anin-house library. These fully or partly matched the bindingpattern identified through the pharmacophore hypothesisthat was developed from structural information.

A mutant form of the protein designed to have impairedbinding capability was generated by exchanging asparagine56 with alanine. While the binding site had been success-

fully altered, and binding affinity for the substrate analoguewas fully destroyed by this single point mutation, proteinconformation remained virtually unchanged, as confirmedby circular dichroism spectroscopy.

The binding activity profiles of the wt and mutantproteins were elucidated using SPR. Proteins were immo-bilized on a Biacore sensor chip using standard aminecoupling chemistry. The saturation signal for the bindingof small molecules (molecular weight, MW �400 Da) wasapproximately 25 resonance units (RUs). Comparing thissaturation response with the amount of protein immobilizedindicated that 30% of the immobilized wt protein remainedactive. No binding was observed for the substrate analogueto the immobilized mutant protein.

Figure 3 shows binding curves of three representativesfrom the 250 compounds. This analysis enables binding andnon-binding compounds to be distinguished. Compoundsthat bind show a time-dependent increase in response duringthe association phase when the surface is in contact with thecompound solution. The response decreases during the

Figure 2. The pharmacophore hypothesis model for HPPK andthe pterin ring of the substrate analogue showing the non-polarinteractions with phenylalanine 123 and tyrosine 54 (left) and thehydrogen bonding pattern (right). This model was used forvirtual screening of the internal compound library.

Figure 3. (a) Sensorgrams for three compounds (A, B, C) andfour report points (1, 2, 3, 4) used for data reduction. Report point1 is the baseline level immediately before the injection of thecompounds, point 2 the binding level in the end of the associa-tion phase, point 3 after 10 s of dissociation and point 4 after20min of dissociation. (b) For screening purposes it is moreconvenient to use bar graphs where the report points for thecompounds are plotted. Compound A shows a rapid associationto and dissociation from the target and shows therefore signalsfor report points 2 and 3. Compound B does not bind to thetarget. Compound C depicts a strong binder with significantsignals for report points 2–4.

SECONDARY SCREENING AND LEAD OPTIMIZATION USING HIGH-RESOLUTION SPR 275


dissociation phase, when the surface is again in contact withpure buffer. The dissociation phase also gives a roughestimate of the stability of the complex. In order to facilitatethe rapid data processing required for the daily screening ofhundreds of compounds, the real-time binding responseswere evaluated using a limited number of discrete ‘reportpoints’. These are software-predetermined time points usedto identify specific signs of binding activity [Fig. 3(b)].Figure 4(a) compares binding to the wt and mutant proteinfor 16 out of the 250 compounds. Of these, eight bind to wtHPPK but only four of them bind to the mutant withsignificantly different affinity [Fig. 4(b)]. These four repre-sentatives can be classified as selective binders since theyinteract with the target protein at the active site. This is animportant advantage of the assay method employed, sinceselectivity is generally a more valuable parameter for com-pound selection and development than absolute target bind-ing affinity that takes no account of the degree of specificity.

Some 250 compounds belonging to 10 different pharma-cophore/scaffold classes were screened in this project. Thedirect binding assay identified a total of 15 compounds thatshowed selective binding to the HPPK binding site. Thesehits belong to four different classes of compounds and theiraffinities, determined by SPR, were between 400 nM and200 mM. Although these hits are still far from being drugcandidates, they are selective target binders and representideal starting points for the synthesis of derivatives withimproved binding activity, bioavailability and metabolicprofiles and decreased toxicity.

Validation of DPP-IV lead series using high-resolutionkinetic analysis and modified binding site controls

SPR was also used to validate leads for DPP-IV (Thomaet al., 2003). In this assay, DPP-IV was immobilized on the

sensor surface and the binding and dissociation of inhibitorcompounds was monitored in real time by SPR.

Active site point mutants were prepared and used to probefor site specificity. For this project, another strategy wasapplied. The active site serine was used for the binding of aphosphinate ester (Augustyns et al., 1999). The covalentbinding of this ester to the serine is irreversible and blocksthe active site. The flow system design of the Biacoreinstrument allows this blocking reaction to be carried outon proteins already immobilized to a specific region of thesensor surface, enabling parallel analysis of blocked andunmodified targets in the same assay.

The outcome of such an experiment is shown in Fig. 5.The unmodified protein yields information on reversibilityof binding along with kinetic data, with all interactionsmonitored in real time. From the time course of thesereactions it is possible to extract the relevant kinetic data(association rate constant, kon, and dissociation rate con-stant, koff) and calculate the equilibrium binding constant,KD (KD¼ koff/kon). The relevance of this high-resolutionbinding characterization was clearly validated by the lack ofresponse seen with the modified protein, showing that thekinetic data obtained for wt DPP-IV represents compoundbinding to the active site (Fig. 5).

We routinely use this setup for further characterization ofactive inhibitors synthesized during the hit expansion step.The compounds could be assigned to three different struc-ture classes, represented in the scheme below. In general, KD

values correlate well with IC50 values and confirm thebiological results. The significant additional value of char-acterizing lead compounds with an alternative technique isreadily demonstrated by looking at a so-called on/off/KD

plot (Markgren et al., 2002), in which the two rate constantsare plotted against each other and the diagonal iso-affinitylines correspond to the equilibrium dissociation constantvalue, KD (Fig. 6). Important information can be obtainedfrom such plots. The plot in Fig. 6 clearly indicates cluster-ing of the compounds in three major areas (A, B, C). Theseclusters contain mainly compounds from three differentstructural classes, cyanopyrolidines (A), benzoquinolizin(B) and pyrrolidinones (C) (Scheme 1). The cyanopyrroli-dines that cluster in the upper left corner exhibit fastassociation and slow dissociation rate constants. The com-pounds with highest affinities (KD�1 nM) have on-ratesaround 106

M� 1 s� 1 and off-rates around 10� 3 s� 1. This

is probably the class from which a potential drug might

Figure 4. (a) Screening results for 16 compounds binding to thewt HPPK (left bar) and to the impaired mutant (right bar). Reportpoint 2 from the end of the association phase (see Fig. 2) isshown. (b) The difference between the wt and the mutant. Fourcompounds (2, 3, 6, 14) show selective binding to the wt HPPK.Compound 5 indicates promiscuous binding as it binds twice thetheoretical saturated 1:1 binding level of �25 RU.

Figure 5. Sensorgrams showing a concentration series (7.8–125nM) of an inhibitor binding to wt (left) and to blocked DPP-IV (right) showing selective, concentration-dependent and satu-rated binding to the binding site of the wt. The black lines showsthe results from global fitting to a 1:1 bindingmodel. No bindingto the blocked target was detected.

276 W. HUBER


emerge if all other prerequisites for developing a successfulcompound are fulfilled. The benzoquinoline class in theupper right corner comprises compounds with fast on-ratesbut, in comparison to the first class, also fast off-rates. Thesecompounds can be optimized mainly through structuralmodifications that slow down the dissociation process

because it appears from these two compound classes that anincrease in KD is mainly obtained by a decrease in koff.

The third cluster (C) is somewhat diffuse and is mainlylocated in the lower left corner, indicating very slow on- andoff-rates. Since such extremely slow on-rates have notpreviously been observed for low molecular weight ligands,additional analytical ultracentrifugation (AUC) experimentswere performed to analyse protein, inhibitor and protein/inhibitor preparations. Solutions of these ligands showedvery high sedimentation coefficients, indicating that thesecompounds exist mainly as very large aggregates and not asmonomers. These aggregates have molecular weights largerthan 106 Da, which explains why slow association is seen inthe SPR experiments. The slow association is due to the lowamount of monomer in solution and to the fact that there isno real driving force for the molecules to leave theseaggregates and to bind to the immobilized protein as amonomer. The kinetic of this process is very slow and isassociated with a high activation barrier.

The ligands from cluster C behave totally differentlywhen the protein is present in solution. When a DMSOsolution of these ligands is dispensed into an aqueoussolution containing the protein, no sedimentation of largecompound aggregates or loss of material was seen in theAUC experiment. However, independently from the

Figure 6. On/off/KD-map for 160 compounds binding to DPP-IV. (a) Association rate constant (kon)plotted vs. dissociation rate constant (koff) and the affinity, equilibrium dissociation constant (KD)added as diagonal iso-affinity lines. The three structural classes (A¼ cyanopyrrolidines,B¼benzoquinolizines, C¼pyrrolidinones) cluster in different positions. Compounds 1 and 2 arehighlighted to illustrate that compounds with almost identical affinities can exhibit very differentkinetic properties. (b) Kinetic binding profiles of compounds 1 and 2. Although their affinities differby only around 2-fold, compound 2 exhibits a much more rapid kinetic profile, with on- and off-rates around 100-fold faster.

Scheme 1. DPP-IV binders.

Scheme 2. DHNA binders.



concentration ratio of protein to ligand, all of the ligand wasfound to be bound to the protein. Obviously the presence ofprotein in solution reduces the concentration of free ligandsuch that no aggregates can be formed. Stoichiometricfactors of up to 13 are determined. It is not surprising thatenzymatic activity is suppressed for a protein that is fullycovered with an unspecifically binding compound. Thecompounds appear as inhibitors in the enzymatic assay,but the inhibitory effect may not be due to the binding of themolecules to the active site. Consequently such compoundsare not drug-like and the substance class C was rejected as apotential lead compound. Compounds exhibiting such be-haviour have been classified recently in the literature as‘promiscuous inhibitors’ or ‘frequent hitters’ (Ryan et al.,2003). These types of compounds are one of the mainsources of false positives found in high-throughput screens.Since the behaviour of this compound class is independentof the protein (it is, for instance, similarly solubilized byhuman serum albumin), it will probably be active in everyprotein assay with a similar format.

Mechanistic aspects of inhibition of DHNA

The SBBA strategy was also used to design suitable mutantproteins for testing binding site activity in the case ofdihydroneopterin-aldolase (DHNA). DHNA is another pro-tein of the bacterial folate pathway. The three-dimensionalstructure of DHNA, which was obtained in-house, showsthat this aldolase exists as an octamer (Hennig et al., 1998).This quaternary structure was confirmed in solution byAUC. The protein is only active in this octameric form. Acrystal structure of a substrate–protein complex enabled thedesign of mutants with impaired binding sites. Two mutantswere constructed, with the glutamic acid 74 replaced byarginine or alanine. These mutations eliminated bindingactivity but did not change the folding or physico-chemicalproperties of the protein.

In the course of lead selection, structure–activity analysisshowed very different activities for a series of compoundswith similar structures (especially those with similar phar-macophores). L-Biopterin, for example, is a highly activecompound, while D-neopterin is not (Fig. 7). The parts ofthe molecules responsible for the polar interactions with theprotein are virtually identical in both compounds. In theprocess of optimizing such compounds for high activity, it isessential to know why such minor structural changes caninfluence the activity of a lead compound to such an extent.This problem was tackled using SPR assays, with thealdolase biotinylated and immobilized on a streptavidin-coated sensor chip surface. This enabled full regeneration ofthe surface and immobilization of a new batch of aldolasebefore each binding experiment. The experimental protocolincluded capture of the biotinylated aldolase, injection ofthe analyte solution, and regeneration of the surface withethanolamine solution and short pulses of 50 mM sodiumhydroxide and 3 M sodium chloride. This regeneration stepremoved all the aldolase from the surface without destroy-ing the binding capacity of the streptavidin. After regenera-tion, fresh aldolase could be captured on the sensor chip.

The octameric form is only stable at salt concentrationabove 1 M and at basic pH (>8.0), conditions that are often

found in bacteria. This experimental setup using capturedtarget allows the evaluation of varying salt or pH on thecleavage of the octamer. The amount of protein loss due tothe cleavage of the octamer could be measured directly.

The reason why one compound is active and the otherinactive becomes obvious when comparing sensorgrams.For the active compound, a significant amount of aldolase islost from the surface, a phenomenon shown by AUCexperiments to be due to cleavage of octamers into tetra-mers. Such protein loss is not observed with the inactivecompound. In independant experiments using tetramericaldolase, it was shown that both compounds bind to thealdolase. The degree of octamer cleavage is concentration-dependent. Using the mutant protein with an impairedbinding site as a reference, it was demonstrated that thecompound must bind to this site to cleave octamericaldolase. Figure 7 shows the concentration-dependent clea-vage of the wt protein by L-biopterin, the active compound.No cleavage is observed when the binding site is impaired,as in the mutant protein E74R-aldolase, or when wt aldolaseis contacted with D-neopterin, the inactive compound ininhibition experiments.

In the case of DHNA, structure–activity interpretation hasto consider areas not directly involved in forming polarinteractions to binding site amino acids. It has to considerinhibitor moieties that protrude from the binding site toinfluence intramolecular interactions responsible for theformation of the octamer. This conclusion is confirmed byanother series of compounds, again with identical moietiesprotruding into the binding site but with different fragmentsat the side. All bind to aldolase, but only those that cleavethe octamer in a concentration-dependent manner are activein the biological assay.

DISCUSSION

In our laboratory the SPR- direct binding assays have beenroutinely used for 10 years and assays have been developed

Figure 7. Dose–response curve of biologically active L-biopterininjected over wt DHNA (solid circle), E74R-mutated DHNA (opencircle) and D-neopterin in contact with wt DHNA (solid squares).The dose-dependent decrease in signal shows that L-biopterincleaves the octameric wt DHNA but not the mutant. D-neopterindoes not cleave the octameric wt DHNA.

278 W. HUBER


for about 20 different low molecular weight compoundprotein target systems. The main use has been as a second-ary screen for validation of hits from HTS. We have foundthat a high percentage of hits are false positives, e.g. they donot bind to the target or they are promiscuous binders. Thisdoes not mean that HTS is unreliable, but rather that itshould be used in combination with a biophysical secondaryscreening method that identifies the true binders. The use ofstructural information about the binding site and pharmo-cophore hypothesis, as in SBBA, gives a significant im-provement because an impaired mutant can be used for theidentification of compounds that bind selectively to thetargeted binding site. The use of impaired mutants asreferences significantly reduces the number of false posi-tives. The selection process is further improved becausepromiscuous binders can be identified by the fact that theygive signals far higher than expected from 1:1 binding.

We found that the extra work needed for preparing mutanttarget proteins is well worth the trouble and we nowroutinely use them when screening focused libraries orfragment libraries as well as in hit validation using directSPR binding. The design of such mutants is based on thethree-dimensional structure information and the respectivepharmacophore hypothesis or binding model derived there-from. Generally amino acids that are involved in polarinteractions are replaced by amino acids that have altereddonor/acceptor properties, have the opposite charge or havehigher spatial demand and block the site by steric hindrance.It is important to check that the mutation has little or noinfluence on the protein conformation and the physico-chemical properties. When using an irrelevant proteinsuch as for instance HSA that is not closely related to thetarget protein such as a mutant, these indispensable proper-ties of an ideal control are lost.

The examples described here show that increased infor-mation achieved by resolving the affinity into binding rateconstants significantly improves the understanding of struc-ture–activity relationships. In the DPP-IV example, wefound that only one of the three scaffolds provided adequatebinding properties. In addition a second scaffold possessedproperties that justified further attempts at improving thekinetic behaviour of the compounds. The third structuralclass was not considered for further optimization attemptsbecause of its promiscuous binding properties.

In the case of DHNA structural information on ligandbinding site and binding model was not sufficient to under-stand inhibitory activity. The Biacore experiments clearlyshowed that efficient inhibitors must be able to bind to theactive site and to cleave the octamer.

The major advantages with SPR direct binding assays incomparison with other biophysical screening methods arebinding kinetic information and the very low consumptionof the target molecule. The throughput fits well in asecondary screening situation when hits from HTS or virtualscreens are going to be validated. The target consumption ismuch lower than with the other biophysical screeningmethods. For example with the same amount used todetermine the affinity using ITC, NMR or AUC, an SPRassay can determine the affinities for 100–1000 substances.If the affinities of a few compounds are going to bemeasured then AUC and ITC are very useful methodswith very short assay development times. However, these

methods do not give the kinetic information useful for leadoptimization and information on promiscuous binding,which is readily obtained by using reference proteins inSPR assays.

The approach described here is a powerful method forthe identification of inhibitors aimed at a well-definedactive binding site on the target protein. It may not, however,be equally applicable to all drug discovery scenarios.Allosteric inhibitors, for example, bind to regions that arespatially distinct from the active site and may be missedaltogether using impaired binding site mutant as a referenceprotein.

There are also some complications with SPR directbinding assay-based secondary screening methods, whichare related to the reuse of the target protein. The SPR assaysneed reasonably pure and active proteins as the detectionprinciple is related to detection of mass measured as changein refractive index. Therefore care is needed in findingproper immobilization conditions and chemistries. In mostcases simple amine coupling using EDC/NHS couplingchemistry works. However, there are proteins which areunstable in acidic conditions which are used in the pre-concentration step. This problem can be minimized bymixing the target with the immobilization buffer immedi-ately before the injection on the sensor chip. The use of aninhibitor or substrate to block the binding site has also beenshown to increase the activity of the immobilized protein(Casper et al., 2004). Additionally, a number of otherchemical and capturing-based immobilization options arealso available.

In secondary screening there is little need for regenera-tion experiments since low affinity compounds dissociatevery rapidly. During lead optimization, the need for properregeneration methods is greater as the compounds haveslower dissociation rates. When the protein is unstable andstrong regeneration conditions cannot be used, increasingthe assay temperature to increase the dissociation rate andusing longer dissociation times are often effective. Anothermethod is to use capture-based immobilization, whereregeneration breaks the binding between the capturingmolecule and the target (although this does increase targetconsumption).

CONCLUSION

Structure-based biophysical analysis is a powerful approachto the selection of lead compounds with potential for drugdevelopment. The use of mutated target proteins as refer-ence in the SPR studies further improves the quality of theinformation obtained. The structure activity data, strength-ened by the resolution of affinities into kinetic parameters,can provide researchers with the knowledge they need toreliably select the most promising compounds for furtherdevelopment. In our experience, SBBA is a powerful toolfor improving the drug discovery process.

Acknowledgements

The author thanks Dr G Franklin and Dr M. Hamalainen from Biacore ABfor support and valuable discussions.



REFERENCES

Alterman M, Sjobom H, Safsten P, Markgren P-O, Danielson UH,Hamalainen MH, Lofas S, Hulten J, Classon B, SamuelssonB, Hallberg A. 2001. P1/P1 modified HIV-1 protease inhibitorsas tools in two new sensitive surface plasmon resonanceassays. Eur. J. Pharm. Sci. 13: 203–212.

Augustyns K, Bal G, Thonus G, Belyaev A, Zhang XM, BollaertW,Lambeir A-M, Durinckx C, Goossens F, Haemers A. 1999. Theunique properties of dipeptidyl-peptidase IV (DPP IV/CD26)and the therapeutic potential of DPP IV inhibitors. Curr. Med.Chem. 6: 311–327.

Bailly C, Carrasco C, Joubert A, Bal C, Wattez N, Hildebrand MP,Lansiaux A, Colson P, Houssier C, Cacho M, Ramos A, BranaMF. 2003. Chromophore-modified bisnaphthalimides: DNArecognition, topoisomerase inhibition, and cytotoxic proper-ties of two mono- and bisfuronaphthalimides. Biochemistry42: 4136–4150.

Boehm HJ, Boehringer M, Bur D, Gmuender H, Huber W, KlausW, Kostrewa D, Kuehne H, Luebbers T, Meunier-Keller N,Mueller F. 2000. Novel inhibitors of DNA gyrase: 3D structurebased biased needle screening, hit validation by biophysicalmethods, and 3D guided optimization. A promising alter-native to random screening. J. Med. Chem. 43: 2664–2674.

Bukhtiyarova M, Northrop K, Chai X, Casper D, Karpusas M,Springman E. 2004. Improved expression, purification, andcrystallization of p38alphaMAP kinase. Protein Expr Purif 37:154–161.

Carrasco C, Rosu F, Gabelica V, Houssier C, De Pauw E, Garbay-Jaureguiberry C, Roques B, Wilson WD, Chaires JB, WaringMJ, Bailly C. 2002. Tight binding of the antitumor drugditercalinium to quadruplex DNAChembiochem 3: 1235–1241.

Carrasco C, Joubert A, Tardy C, Maestre N, Cacho M, Brana MF,Bailly C. 2003. Chromophore-modified bisnaphthalimides:DNA recognition, topoisomerase inhibition, and cytotoxicproperties of two mono- and bisfuronaphthalimides.Biochemistry 42: 11751–11761.

Casper D, Bukhtiyarova M, Springman EB. 2004. A Biacorebiosensor method for detailed kinetic binding analysis ofsmall molecule inhibitors of p38alpha mitogen-activatedprotein kinase. Anal. Biochem. 325: 126–136.

Cooper MA. 2002. Optical biosensors in drug discovery. Nat.Rev. Drug Discov. 1: 515–528.

Cooper MA. 2003a. Biosensor profiling of molecular interactionsin pharmacology. Curr. Opin. Pharm. 3: 557–562.

Cooper MA. 2003b. Label-free screening of bio-molecular inter-actions. Anal. Bioanal. Chem. 377: 834–842.

Davis B, Afshar M, Varani G, Murchie AI, Karn J, Lentzen G,Drysdale M, Bower J, Potter AJ, Starkey ID, Swarbrick T,Aboul-ela F. 2004. Rational design of inhibitors of HIV-1 TARRNA through the stabilisation of electrostatic ‘hot spots’. J.Mol. Biol. 336: 343–356.

ElgM, Gustafsson D, Deinum J. 1997. The importance of enzymeinhibition kinetics for the effect of thrombin inhibitors in a ratmodel of arterial thrombosis. Thromb. Haemostas. 78: 1286–1292.

DavidsonW, Frego L, Peet GW, Kroe RR, Labadia ME, Lukas SM,Snow RJ, Jakes S, Grygon CA, Pargellis C, Werneburg BG.2004. Discovery and characterization of a substrate selectivep38alpha inhibitor. Biochemistry 43: 11658–11671.

Geitmann M, Danielson UH. 2004. Studies of substrate-inducedconformational changes in human cytomegalovirus pro-tease using optical biosensor technology. Anal. Biochem.332: 203–214.

Gribbon P, Sewing A. 2003. Fluorescence readouts in HTS: nogain without pain? Drug Discov. Today 8(22): 1035–1043.

Hall S, Strichartz G, Moczydlowski E, Ravindran A, Reichardt PB.1990. The saxitoxins: sources, chemistry, and pharmacol-ogy. In Marine Toxins Origin, Structure and Molecular Phar-macology, Hall S, Strichartz G (eds). ACS SymposiumSeries. American Chemical Society: Washington, DC; 29–65.

Harrison RJ, Cuesta J, Chessari G, Read MA, Basra SK, ReszkaAP, Morrell J, Gowan SM, Incles CM, Tanious FA, Wilson

WD, Kelland LR, Neidle S. 2003. Trisubstituted acridinederivatives as potent and selective telomerase inhibitors.J. Med. Chem. 46: 4463–4476.

Hennig M, D’Arcy A, Hampele IC, Page MG, Oefner C, Dale GE.1998. Crystal structure and reaction mechanism of 7,8-dihy-droneopterin aldolase from staphylococcus aureus.Nat. Struct. Biol. 5: 357–362.

Hennig M, Dale GE, D’Arcy A, Danel F, Fischer S, Gray CP,Jolidon S, Muller F, Page MG, Pattison P, Oefner C. 1999.The structure and function of the 6-hydroxymethyl-7,8-dihy-dropterin pyrophosphokinase from Haemophilus influen-zae1. J. Mol. Biol. 287: 211–219.

Hamalainen MD, Markgren P-O, Schaal W, Karlen A, Classon B,Vrang L, Samuelsson B, Hallberg A, Danielsson UH. 2000.Characterization of a set of HIV-1 protease inhibitors usingbinding kinetic data from a biosensor-based screen. J. Bio-mol. Screen. 5: 353–360.

Karlsson R. 2004. SPR for molecular interaction analysis, a re-view of emerging application areas. J. Mol. Recognit. 17:151–161.

Karlsson R, Kullman-Magnusson M, Hamalainen MD, RemaeusA, Andersson K, Borg P, Gyzander E, Deinum J. 2000.Biosensor analysis of drug target interactions—direct andcompetitive binding assays of interactions between throm-bin and thrombin inhibitors. Anal. Biochem. 278: 1–13.

Krogsgaard M, Prado N, Adams EJ, He XL, Chow DC, Wilson DB,Garcia KC, Davis MM. 2004. Evidence that structural rear-rangements and/or flexibility during TCR binding can con-tribute to T cell activation. Mol. Cell 12: 1367–1378.

Lange G, Lesuisse D, Deprez P, Schoot B, Loenze P, Benard D,Marquette JP, Broto P, Sarubbi E, Mandine E. 2003. Require-ments for specific binding of low affinity inhibitor fragmentsto the SH2 domain of (pp60)Src are identical to those forhigh affinity binding of full length inhibitors. J. Med. Chem.46: 5184–5195.

Lofas S. 2004. Optimizing the hit-to-lead process using SPRanalysis. Assay Drug Devl. Technol. 2: 407–416.

Markgren P-O, Hamalainen MD, Danielsson UH. 1998. Screeningof compounds interacting with HIV-1 proteinase using op-tical biosensor technology. Anal. Biochem. 265: 340–350.

Markgren P-O, Hamalainen MD, Danielsson UH. 2000. Kineticanalysis of the interaction kinetics between HIV-1 proteaseand inhibitors using optical biosensor technology. Anal.Biochem. 279: 71–78.

Markgren P-O, Hamalainen MD, Danielsson UH. 2001. Determi-nation of interaction kinetic constants for HIV-1 proteaseinhibitors using optical biosensor technology. Anal. Bio-chem. 291: 207–218.

Markgren P-O, Schaal W, Hamalainen MD, Karlen A, Hallberg A,Samuelsson B, Danielson UH. 2002. Relationships betweenstructure and interaction kinetics for HIV-1 protease inhibi-tors. J. Med. Chem. 45: 5430–5439.

Myszka DG. 2004. Analysis of small-molecule interactions usingBiacore S51 technology. Anal. Biochem. 329: 316–323.

Myszka DG, Rich RL. 2000. Implementing surface plasmon re-sonance biosensors in drug discovery. Pharm. Sci. Technol.Today 3: 310–317.

Noble CG, Barnard FM, Maxwell A. 2003. Quinolone–DNA inter-action: sequence-dependent binding to single-strandedDNA reflects the interaction within the gyrase–DNA com-plex. Antimicrob. Agents Chemother. 47: 854–862.

Rich RL, Myszka DG. 2004. Why you should be using more SPRbiosensor technology. Drug Discov. Today Technol. 1: 301–308.

Rich RL, Hoth LR, Geoghegan KF, Brown TA, LeMotte PK,Simons SP, Hensley P, Myszka DG. 2002. Kinetic analysisof estrogen receptor/ligand interactions. Proc. Natl Acad.Sci. 99: 2562–2567.

Ryan AJ, Gray NM, Lowe PN, Chung C. 2003. Effect of detergenton ‘promiscuous’ inhibitors. J. Med. Chem. 46: 3448–3451.

Shuman CF, Markgren P-O, Hamalainen MD, Danielsson UH.2003. Elucidation of HIV-1 protease resistance by character-

280 W. HUBER


ization of interaction kinetics between inhibitors and enzymevariants. Antiviral Res. 58: 235–243.

Shuman CF, Hamalainen MD, Danielsson UH. 2004a. Kinetic andthermodynamic characterization of HIV-1 protease inhibi-tors. J. Mol. Recognit. 17: 1–14.

Shuman CF, Vrang L, DanielssonUH. 2004b. Improved structure–activity relationship analysis of HIV-1 protease inhibitorsusing interaction kinetic data. J. Med. Chem. 47: 5953–5961.

Tedious FA, Hamelberg D, Bailly C, Czarny A, Boykin DW,WilsonWD. 2004. DNA sequence dependent monomer-dimer bind-ing modulation of asymmetric benzimidazole derivatives. J.Am. Chem. Soc. 126: 143–153.

Thoma R, Loffler B, Stihle M, Huber W, Ruf A, Hennig M. 2003.Structural basis of proline-specific exopeptidase activity asobserved in human dipeptidyl peptidase-IV. Structure 11:947–959.

Ueda Y, Yamagishi T, Samata K, Ikeya H, HirayamaN, TakashimaH, Nakaike S, Tanaka M, Saiki I. 2003. A novel low molecularweight antagonist of vascular endothelial growth factorreceptor binding: VGA1155.Mol. Cancer Ther. 2: 1105–1111.

Vassilev LT, Vu BT, Graves B, Carvajal D, Podlaski F, Filipovic Z,Kong N, Kammlott U, Lukacs C, Klein C, Fotouhi N, Liu EA.2004. In vivo activation of the p53 pathway by small-mole-cule antagonists of MDM2. Science 303: 844–848.

Yu C, Chen L, Luo H, Chen J, Cheng F, Gui C, Zhang R, Shen J,Chen K, Jiang H, Shen X. 2004. Binding analyses betweenHuman PPAR gamma-LBD and ligands. Eur. J. Biochem. 271:386–397.

XiangW, Glickman JF, Bowen BR, SillsMA. 2003. Comparison ofassay technologies for a nuclear receptor assay screenreveals differences in the sets of identified functional an-tagonists. J. Biomol. Screen. 8: 381–392.



Documents

A new strategy for improved secondary screening and lead optimization using high-resolution SPR characterization of compound–target interactions