NMR screening in drug discovery Jonathan MM NMR methods in drug discovery have traditionally been used drugs. Many pharmaceutical industry research groups have to obtain structural information for drug targets or embarked on drug discovery programs using NMR screen get-ligand complexes Recently, it has been shown that ing. Although there have been few articles published to NMR may be used as an alternative approach for identification date describing such techniques, it is clear from presenta of ligands that bind to protein drug targets, shifting the tions and discussions at scientific meetings that mphasis of many NMR laboratories towards screening and development of screening tools and protocols has become design of potential drug molecules, rather than structural a priority for these groups. In this review, studies published characterization or otherwise publicly disclosed since 1996 will be dis cussed. Methods developed for NMR screening, as wel application of these methods to specific targets, will be ertex Pharmaceuticals Incorporated, 130 Waverly Street, Cambridge, addressed. Papers covering general methods for detection of binding by NMR, using either relaxation or diffusion Current Opinion in Biotechnology 1999, 10: 54-58 based methods, are beyond the scope of this short review ttp: //biomednet. com/elecref/0958166901000054 The concept of using NMR spectroscopy to detect binding C Elsevier Science Ltd ISSN 0958-1669 of ligands to a protein is well established in the literature Abbreviations [1-4]. Detection of binding by Nmr may be achieved in HTS several ways, using techniques such as chemical shift per MSCS ystal structures turbation, differential line broadening, the transferred NOE lOE, and diffusion-based methods. All of the above tech SAR activity relationships niques have been used by various groups in differin implementations of NMR screening. These methods fall Introduction into two general categories: those that monitor NMR sig- The process of preclinical drug discovery traditionally nals from the protein, or those that monitor the ligands involves two processes: lead generation and lead optimiza- These differing strategies are discussed belo tion. In the former process, molecules are identified that bind a drug target and inhibit in uitro activity, whereas in the Monitoring the protein signals- SAR by NMR latter, potential drug molecules are optimized with respect Structure activity relationships(SAR)by nmR is the fir to in vitro potency and other important parameters reflecting method for NMR screening disclosed in the literature, pre bioavailability, pharmacokinetic or toxicological properties. posed by the group at Abbott Laboratories [5]. SAr by NMR may best be described as a method for totally NMR In structure-based drug design programs, NMR methods are driven ligand design. Sar by NMR relies on a fragment ypically used during lead optimization for characterization based approach, wherein a large library(103-104)of small of the structure and dynamics of the drug target. Such stud- molecules is screened using 2D'H-15N spectra of the tar ies may include solution structure determination of proteins get protein as a readout. From changes in protein amide or protein-ligand complexes, or the rapid determination of chemical shifts, one can readily identify whether binding ligand structures in a protein-ligand complex using either has occurred from one or more components in a mixture of isotope editing/filtering or transferred nuclear Overhauser compounds. As individual peaks in the spectrum are reviewed elsewhere in this issue(see Roberts, pp 42-eare effect(NOE)techniques. These structural methods assigned to specific residues in the primary sequence,the affected peaks in the 2D spectra indicate at what site(s)on the protein the ligand is binding. After deconvolution of Alternatively, it has been shown recently that NMR may the mixture has revealed the identity of the compound(s) also be used for lead generation in drug discovery programs. binding the target, a second screen of close analogs of the In these applications, the process of NMR screening may be first 'hit is performed to optimize the binding affi thought of as occurring in two steps. First, NMR methods the first site. In order to identify small molecules that bind are used to detect weak binding of small molecule scaffolds at a proximal site, the screen is then repeated with the first c a target. Subsequently, the binding information is used to site saturated If small molecule fragments are identified esign much tighter binding inhibitors, or drug leads which occupy several neighboring subsites, one can then Although it has been known for many years that NMR is a incorporating the small molecule fragments with variou sensitive method for detecting binding of a small molecule linking groups. The set of linked compounds is then to a protein [1-3l, only recently have these NMr tech- assayed. If linked effectively, resulting compounds may niques been applied in a systematic way to screen libraries have affinities for the target that are even stronger than the of compounds against a target for the purpose of designing products of the binding constants of the individual
NMR screening in drug discovery Moore 55 unlinked fragments. This synergism arises due to an addi of FKBP-12. Although the compounds designed are of no tional entropic contribution from linking. For example, therapeutic value, the results suggested the methodology linking of two millimolar binders may result in a submicro- could indeed work and set the stage for screening and molar affinity binder, or combination of three millimolar inhibitor design of two additional targets of much greater affinity fragments with the appropriate linkers may result therapeutic interest, stromelysin [12"I and the E2 protein in a compound of nanomolar or higher affinity from human papilloma virus(HPV E2)[13.] SAR by NMR may be thought of as an experimental coun- For stromelysin, SAR by NMR screening resulted in the terpart to a number of de novo computational approaches to design of two very potent enzyme inhibitors ligand design [6-8 Using e or more of these approach. ( 25 nanomolar). These inhibitors contained a common es, small molecule fragments are docked into sites on the hydroxamate moiety linked to a set of novel biphenyl scaf. nked together to form 'virtual folds. The latter scaffolds appeared to account for most of inhibitors. Although such computational techniques are the binding affinity. For matrix metalloproteinases such as rapid, accounting for phenomena such as ligand flexibility, stromelysin, as well as FKBP-12, there exists a wealth of induced fit, or solvent-protein interactions within these published information regarding inhibitor design, and it is protocols may be problematic. X-ray crystallographic(mul- clear this thinking influenced the approaches the authors tiple solvent crystal structures [MSCSD)[9, 10] and NMr took in these studies. For example, the scaffold used in the [11 studies of enzymes on-aqueous solvents have FKBP-12 study, a bicyclic system consisting of a piece provided experimental support for computational inyl moiety and trimethoxy phenyl,is ell-known approaches to ligand design by attempting to characterize building block for high affinity FKBP-12 binders [14, P1 hall solvent binding sites in enzymes. SAR by NMR, In the stromelysin study, the authors chose to pursue however, provides advantages over strictly computational fusions containing acetohydroxamic acid out of necessity pproaches or hybrid computational experimental as it was known that acetohydroxamic acid containing com pproaches such as MSCS In the Sar by NMR method, pounds can inhibit the protease activity of stromelysin the fragments to be linked are determined experimentally, [15, 16]; however, acetohydroxamic acid bound with an and multiple fragments may be screened in mixtures for a affinity of 17 mM. With such a low affinity, it is question- single target, whereas the MSCS method requires X-ray able whether this fragment would have been discovered rystal structures to be solved in multiple solvents. More independently in the course of a normal screen importantly, the solvent small molecule 'fragments'are necessarily limited in diversity What are the possibilities for ligand design by sar by nmr in the case of a target for which little is known regarding Although SAR by NMR is a significant improvement over inhibitor design: The Abbott study with the HPV E2 protein purely computationally based approaches, one should also [13] addresses this question in what one might consider the note that there are limitations associated with the method. most interesting of the three Abbott papers. In this work For example, because Sar by NMR entails iterative fragments were discovered that bound regiospecifically ands were then dreds of milligrams of i'N labeled protein is required. This deemed most relevant, the DNA-binding site. Ultimately requirement could be problematic and very expensive to inhibitors of DNA binding in the low micromolar range were overcome for drug targets that can not be expressed to high synthesized based on the screening hits. Although these levels in bacterial cells. Furthermore, identification of inhibitors were much weaker than those obtained in the binding sites for screening compounds requires that NMr FKBP-12 and stromelysin studies, the work described is assignment and structure determination for the protein be important because lead compounds were obtained for a tar- completed before screening commences, Because nmr get for which little chemistry was known. Furthermore, these structure determination is currently limited to proteins leads are in the affinity range where further optimization under-30 k Da, the number of potential drug targets that ght be possible with additional medicinal or combinatori hay be approached using SAr by NMR is small. Despite al chemistry cffort. This work, as well as the stromelysin these limitations, sar by NMr has been successful in study, provide extensive experimental detail on how the generating leads for several low molecular weight protein screens were carried out and thus are of greatest interest to targets, as described below Applications of saR by NMR Monitoring the ligand signals: line broadening Thus far, the Abbott group has reported in the literature transferred NOEs and diffusion-basel adening, three different applications of the sar by nmr method to screening methods protein drug targets. The first paper 15] demonstrated proof Several additional NMR methods have been described f of concept with a well-characterized model system, the screening of small molecule binding to target proteins FK506 binding protein FKBP-12. In this work, it was These techniques employ NMR line broadening, trans shown that several fragments with micromolar and millimo- ferred NOEs, and pulsed field gradient diffusion lar affinities could be linked to form a nanomolar inhibitor measurements. There are numerous advantages to observing
the ligand signals rather than protein signals. Simultaneous 1D pulsed field gradient NMR methods. Lin and Shapiro observation of signals from all small molecules in a mixture [18] first reported the DECODES experiment, an experi allows the straightforward identification of the specific com- ment designed for mixture analysis of small molecule pools ponent(s) binding the target protein without a need for from combinatorial chemistry reactions. In this experi- deconvolution. More importantly, observing the ligands ment, mixtures of compounds that could not be identified directly obviates the need for isotopically labeled protein, using standard 2D techniques alone could be assigned by and allows one to examine targets with no molecular weight acquiring a series of 2D spectra at different pulsed field mit; both severe limitations of the sar by nmr method. gradient strengths, effectively separating the original 2D spectra into a third dimension modulated by molecular dif- The transferred NOE experiment is a well established fusion. This methodology was extended to detect method for determining the bound conformation of a ligand intermolecular interactions between small molecules and hich undergoes rapid exchange with a target molecule. small molecule receptors 19, 20 Due to the high ligand When a small molecule binds a large molecule, such as a concentrations required(20 mM), however, these tech even transiently, the numerical sign and magni- niques will have limited utility for screening of biological des of the NOEs change. The magnitudes of the NOEs targets. Another publication by the Novartis group [21] are dependent on the proton-proton internuclear distances describes an alternative method for examination of small and may be used to calculate approximate interproton dis- molecule interactions with proteins. The pulsed sequence tance restraints. Although the NOE information is most combines diffusion editing, to detect ligand binding, with often used to determine the structure of the ligand when 1.C isotope filtering, which removes background signals bound to the target, for screening purposes it provides a very from the protein. Although this experiment is simple to sensitive and highly reliable diagnostic for binding as well. carry out and does not rely on obtaining difference spectra veral groups have reported using the transferred Noe the method will have limited utility for higher throughput method as a screening tool. Meyer et al.[17.I demonstrated NMR screening of mixtures, due to the high costs of that this method could be used to screen a mixture of preparing the requisite amounts of i.C labeled protein oligosaccharides for binding to Aleuria aurantia agglutinin In this study, it was found that despite the complexity of the A more practical implementation of such methods is 2D spectra for two mixtures of oligosaccharides containing described by Hajduk et al. 22. In this paper, the authors six and fifteen components, a single binding ligand could be show how NMR relaxation and diffusion, well known tech readily identified and characterized. Other groups are also niques for the study of macromolecule-small molecule using the transferred NOE technique routinely for screen- interactions, may be used to screen large libraries of small ing purposes. Kline and co-workers at Lilly Research molecules against a macromolecular target. The ID meth Laboratories repo fragments derived from biotin to the throughput as experiment time is greatly reduced relative using transferred NOE methods to ods described in this paper promise much greater examine bind protein streptavidin (A Kline, personal communication). to the 2D transferred NOE or 2D 1H-15N heteronuclear ur SHAPES strategy for NMR screening, discussed below, correlation experiments used in the sar by NMR also makes extensive use of transferred NOE spectra to approach. Additional advantages of these 1D methods are identify binding components in mixtures that they do not require the use of isotopically labeled pro In and n ay be applied to targets in a much Other groups have reported on methods for screening mix- molecular weight range. Drawbacks to these methods d on relaxation and diffusion editing. Like the include the additional processing and errors associate NOE, other NMR observables that exhibit a marked use of nmr difference methods, and, for the use of line dependence on molecular mass include the transverse broadening methods, decreased sensitivity for detection of elaxation time (T2)and the translational diffusion coeffi- weaker binding compounds compared to the transferred ent (D). In relaxation editing, binding of a smal NOE experiment. It should also be noted that the group at molecule to a large protein will cause the ligand 'T2 values Bristol Myers Squibb(Wallingford, CT) also reported at a to decrease and the signals to broaden. The degree of recent meeting the use of diffusion-based techniques for broadening observed is dependent on, and increases with, high-throughput screening(DJ Detlefson et al. Abstract the affinity of the ligand for the macromolecule. This W/Th P147, 39th Experimental Nuclear Magnetic broadening is easily measured by comparing ID NMR Resonance Conference, Asilomar, CA, 1998) spectra of the ligand with and without the protein present In diffusion editing, a similar approach is used. A small The SHAPES strategy molecule diffuses in solution several orders of magnitude In our laboratory, we have developed a strategy for NMR ter than a large protein. If a small molecule binds to a screening that, like the above methods, relies on monitor macromolecule, its diffusion coefficient will be reduced ing of ligand signals to determine which compound in a compared with that for the free ligand. The extent of the mixture binds a drug target Fejzo et al. Abstract MIIA-4 reduction will be dependent on the affinity of the small XVIllth International Conference on Magnetic Resonance molecule for the target. Changes in the diffusion proper- in Biological Systems, Tokyo, Japan, 23-28 August 1998) ties of a small molecule may be measured using standard Our primary motivation for developing these methods is
NMR screening in drug discovery Moore 57 target size; our in-house drug targets are mostly very large signals via line broadening, transferred NOEs, or diffusion by nmr standards. The NMR screening methods we cur- based methods, appear to be better suited for screening of rently employ, however, differ from the sar by NMr larger targets. It is probable that hybrid approaches com method both in implementation and philosophy. Our bining the two types of methods will be of utility in some method, the SHAPES strategy, uses standard iD line drug design programs. Although many groups are known to broadening and 2D transferred NOE measurements to be working in the field, the literature describing these detect binding of a limited ( 200)but diverse library of methods is very thin. It is probable that some pharmaceuti low molecular weight, soluble scaffolds to a potential drug cal firms will prefer to retain their methods as trade secrets target. These scaffolds are derived largely from shapes, or or wait to patent their methods before publication frameworks, most commonly found in known therapeutic Hopefully, as methods evolve more groups will be willing agents 123) and as such represent approximations tosuc- and able to publish their findings. For now, however, in the cessful regions of diversity space enduring battle between industrial scientists and their legal Following screening, weak binding( Kd-HM-mM)scaf departments, the lawyers appear to be winni folds. most of which would be missed in a standard Acknowledgements synthesis of combinatorial libraries, and bias the first com- contributions to the SHAPES project. pounds that undergo enzyme-directed high-throughput References and recommended reaa\,of Papers of particular interest, published within the annual perio drug targets indicate that HTs hit rates for compounds have been highlighted as selected based on NMR SHAPES hits may be higher by of special interes fourfold or more than in a control set of randomly selected es of outstanding interest compounds(CA Lepre, J Fejzo, Jw Peng, JM Moore, 1. Campbell ID, Dwek RA: Biological Spectroscopy. Menlo Park unpublished data). Data from SHAPES hits may also be used advantageously in computational efforts, such as virtual 2. Dwek RA: Nuclear Magnetic Resonance in Biochemistry. Oxford screening, where databases of molecules, either real or virtu cred accordin their synthetic ac 3. James TL: Nuclear Magnetic Resonance in Biochemistry. New York: shape, flexibility, strain energy, similarity to known leads, pharmacophores, docking scores, or other properties relevant 4, NiF: Recent developments in transferred NOE methods, Prog vity, such as lipophilicity or metabolic stability complete description of the SHAPES strategy 5. Shuker SB, Hajduk PJ, Meadows RP, Fesik SW: Discovering and in drug design will be presented elsewhere high-affinity ligands for proteins: SAR by NMR. Science 1996 274:1531-1534 SaR by NmR and the shaPes strategy are similar in that Murcko MA: Recent advances in ligand design methods. In they are both protocols for drug design using NMR, rather Boyd DB. New York: Wiley and Sons: 1997: 1-66 咖如时m如bR22,数电mmm10 NMR are in their basic objectives. SaR by NMR is a very 8. Murcko MA: An introduction to de novo ligand design In Practical elegant but resource intensive method of designing very Application of tight binding(nanomolar affinity) inhibitors using NMR as New York: Marcel Dekker, inc: 1997: 305-354 the principal technology. Alternatively, the go I of sh 9. Mattos C, Ringe D: Locating and characterizing binding sites or proteins. Nat Biotechnol 1996. 14: 595-599 creening and the subsequent follow-up experiments is to establish one or more lead classes of micromol g X, Jeffery CJ, Mattos C, Petsko GA An experimental approach to mapping th compounds that may be further optimized using more effi crystalline proteins. J Phys nt, high-throughput methods, such as HTS and 1. Liepinsh E, Otting G: Organic solvents identify specific ligand combinatorial chemistry. We believe the latter methods may. binding sites on protein surfaces. Nat Biotechnol 1997,15: 264-268 be better suited for exploring the larger regions of chemical An NMR study to examine the binding of solvent molecules to lysozyme Ity space necessary to maximIze in otro potency. Conclusion eakly to sites formed due to crystal packing NMR methods detect NMR screening methods show promise as a highly useful occupancy problems of MSCS path to drug discovery. Although the NMR methods used 12. Hajduk P). Sheppard G, Nettesheim DG, Olejniczak ET, Shuker SB. screening are not novel, application of these methods for HTS of compound libraries forms the basis for several 1195818-582 unique approaches to ligand design. SAR by NMR, despite Application of the SAR by NMR method to design two potent (<25 nM) its limitations, appears to be the method of choice for drug inhibitors of stromelysin. This work provides a step by targets <30 kDa. Other approaches that monitor the ligand ing directly in the field
13. Hajduk PJ, Dinges J, Miknis GF, Merlock M, Middleton T, Kempf DJ 19 MJ, Wareing JR: Screening mixtures by affinity covery of lead inhibitors that block DNA binding of 20. Lin ML M, Wareing JR: Diffusion- edited NMR-affinit NMR for 1:53495250 portance because lead pounds were obtained for a target for which little chemistry was known. 21. Gonnella N, Lin M, Shapiro MJ, Wareing JR, Zhang X tope-filtered affinity NMR. J Magn Reson 1998, 131: 336-33B 14. Arm tad Des aa s the is iand st ruct re f no saancercs ic 22. Hajduk PJ, Olejniczak ET, Fesik Sw: One-dimensional rela immunosuppressant FK506. Acta Crystallogr 1995, D51: 522-528. 5. Hagmann WK, Lark MW, Becker JW: Inhibition of matrix hem199631:231-240 6. Wahl RC, Dunlap RP: Biochemistry and inhibition of collagenase interactions, may be used to screen large libraries of small molecules against 23. Bemis Gw, Murcko MA: The properties of known drugs. 1 ctivity by NMR. Eur Biochem 1997 Molecular frameworks, j Med Chem 1996. 39: 2887-2893 lent example of how the transferred NoE experiment may be used 24. Walters WP, Stahl MT, Murcko MA: Virtual screening- an overview Drug Discov Today 1998, 3: 1 60-178 hardes were screened against Aleuria aurant ingle binding ligand from each mixture could be readily identified and characterized Patent 8. Lin M, Shapiro M: Mixture analysis in combina HV. Saunders jo Application of diffusion- resolved NMR spectro essive compounds. US Patent 1993 1996,61:76177619