feature · 2015. 11. 12. · harren jhoti, founder and chief scientific officer, left his position...

7
FEATURE Protein structure information has tradition- ally played only a supporting role in many pharmaceutical drug discovery programs. With huge investments in ‘high-throughput’ screening and combinatorial chemistry tech- nologies, many pharmaceutical companies aren’t willing to put structural information front and center in the drug discovery process. But recent technological advance- ments are transforming the efficiency of structural analysis—traditionally a slow and laborious process—and many more compa- nies are targeting drugs against intracellular enzymes (such as nonreceptor tyrosine kina- ses and phosphatases), which are more amen- able to a structure-based approach, rather than membrane proteins (such as G protein– coupled receptors (GPCRs) and ion chan- nels). As structural information slowly gains influence, a new generation of companies is intent on finding original approaches for using structure in drug discovery. The problem with proteins Proteins are among the most unpredictable molecules in nature. As genes are translated into proteins, an essentially digital store of information becomes a three-dimensional language of folds and motifs, helices and sheets that we only dimly understand. Yet for chemists hoping to create the best drug for a specific target, it can be a great help—per- haps even a necessity in some cases—to see the target in question, atom for atom, in three dimensions. Better yet is to see how drug leads dock into that target, offering a visual guide to how they might be improved. There is nothing new about the pursuit of structure per se. The first structure-based drug hit the US market in 1981: Capoten (captopril) from Bristol-Myers Squibb (New York, NY, USA), the first angiotensin-con- verting enzyme (ACE) inhibitor. Although the ACE structure was not known at the time, key modifications to the drug were made based on the structure of carboxypeptidase A 1 , a protein that had similar catalytic capa- bilities and was thus presumed to have simi- lar structure. The first wave of structure-aided drug design companies appeared in the mid- to late-1980s, hoping to use rapidly improving methods for determining protein structures to get an early look at their targets and to use them as a guide for drug design—a strategy that met with some success. Agouron (San Diego, CA, USA) created the HIV protease inhibitor nelfinavir mesylate (Viracept) and was eventually acquired by Warner-Lambert (now part of Pfizer, New York, NY, USA). Vertex Pharmaceuticals (Cambridge, MA, USA) brought two HIV drugs, amprenavir (Agenerase) and fosamprenavir (Lexiva), to market, and has a pipeline of compounds in clinical and preclinical development. Bio- Cryst (Birmingham, AL, USA) has had a more difficult road, but is launching its first phase II trial this year, an inhibitor of purine phosphorylase (PNP) to treat T cell–medi- ated diseases. Their approach is often called ‘rational’ drug design as a contrast with blind screen- ing approaches that hope to find molecular ‘hits’ essentially by luck. But that doesn’t mean these companies started their drug dis- covery programs with structure in hand. “It has never been that way in practice anywhere that I’m aware of,” says Mark Murcko, vice president and Chief Technology Officer at Vertex. “In practice, a project team gets started and uses whatever approaches they can to begin to design molecules.” Part of that philosophy can be attributed to deliberate strategic uses of different discov- ery technologies, but in the past it also had to do with the slow pace of structure determi- nation. For instance, BioCryst chairman and CEO Charles Bugg notes efforts to determine the structure of PNP, which began in the 1970s, took five or six years.“Now they could probably do it in a week,” he says. Industrial revolution Yet the process of protein structure solution is becoming speedier and more amenable to automation, not because of any one break- through but rather a convergence of new technologies. Advances in molecular biology make even difficult proteins easy to express in usable quantities. Scientists can quickly engi- neer proteins to meet crystallization condi- tions, removing or altering troublesome regions through random mutagenesis or more directed approaches. X-ray beamlines are stronger and better focused, allowing the use of smaller crystals, while detectors are more sensitive and capture clearer data than ever before—turning this slow, laborious art into something that proponents of the field like to call ‘high-throughput X-ray crystal- lography.’ The net effect is that for many kinds of drug targets, structure can become less of a sideline to drug discovery and more of an integral part of the process. In 1999–2000, when capital was flowing into the biotech sector faster than you could raise the rent on laboratory space, a group of companies were formed with the idea of using protein structure differently than their first-generation predecessors. Each has a slightly different strategy, but they all have something in common: they are taking advantage of this ‘high-throughput’ struc- ture determination, whether by crystallogra- phy or predictive modeling, particularly the possibilities offered by rapid, iterative solutions of structures bound to various ligands. Structure-aided drug design’s next generation Karl A Thiel Has structural bioinformatics advanced enough to form the core of a drug discovery program? A new generation of companies exploiting structure-focused technologies is counting on it. Karl Thiel is a freelance writer based in Portland, Oregon, USA. email: [email protected] NATURE BIOTECHNOLOGY VOLUME 22 NUMBER 5 MAY 2004 513 © 2004 Nature Publishing Group http://www.nature.com/naturebiotechnology

Upload: others

Post on 05-Nov-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: FEATURE · 2015. 11. 12. · Harren Jhoti, founder and chief scientific officer, left his position as head of Glaxo We llcome’s (now GlaxoSmithKline, Uxbri-dge, UK) Structural Biology

F E AT U R E

Protein structure information has tradition-ally played only a supporting role in manypharmaceutical drug discovery programs.With huge investments in ‘high-throughput’screening and combinatorial chemistry tech-nologies, many pharmaceutical companiesaren’t willing to put structural informationfront and center in the drug discoveryprocess. But recent technological advance-ments are transforming the efficiency ofstructural analysis—traditionally a slow andlaborious process—and many more compa-nies are targeting drugs against intracellularenzymes (such as nonreceptor tyrosine kina-ses and phosphatases), which are more amen-able to a structure-based approach, ratherthan membrane proteins (such as G protein–coupled receptors (GPCRs) and ion chan-nels). As structural information slowly gainsinfluence, a new generation of companies isintent on finding original approaches forusing structure in drug discovery.

The problem with proteinsProteins are among the most unpredictablemolecules in nature. As genes are translatedinto proteins, an essentially digital store ofinformation becomes a three-dimensionallanguage of folds and motifs, helices andsheets that we only dimly understand. Yet forchemists hoping to create the best drug for aspecific target, it can be a great help—per-haps even a necessity in some cases—to seethe target in question, atom for atom, in threedimensions. Better yet is to see how drugleads dock into that target, offering a visualguide to how they might be improved.

There is nothing new about the pursuit ofstructure per se. The first structure-baseddrug hit the US market in 1981: Capoten

(captopril) from Bristol-Myers Squibb (NewYork, NY, USA), the first angiotensin-con-verting enzyme (ACE) inhibitor. Althoughthe ACE structure was not known at the time,key modifications to the drug were madebased on the structure of carboxypeptidaseA1, a protein that had similar catalytic capa-bilities and was thus presumed to have simi-lar structure.

The first wave of structure-aided drugdesign companies appeared in the mid- tolate-1980s, hoping to use rapidly improvingmethods for determining protein structuresto get an early look at their targets and to usethem as a guide for drug design—a strategythat met with some success. Agouron (SanDiego, CA, USA) created the HIV proteaseinhibitor nelfinavir mesylate (Viracept) andwas eventually acquired by Warner-Lambert(now part of Pfizer, New York, NY, USA).Vertex Pharmaceuticals (Cambridge, MA,USA) brought two HIV drugs, amprenavir(Agenerase) and fosamprenavir (Lexiva), tomarket, and has a pipeline of compounds inclinical and preclinical development. Bio-Cryst (Birmingham, AL, USA) has had amore difficult road, but is launching its firstphase II trial this year, an inhibitor of purinephosphorylase (PNP) to treat T cell–medi-ated diseases.

Their approach is often called ‘rational’drug design as a contrast with blind screen-ing approaches that hope to find molecular‘hits’ essentially by luck. But that doesn’tmean these companies started their drug dis-covery programs with structure in hand. “Ithas never been that way in practice anywherethat I’m aware of,” says Mark Murcko, vicepresident and Chief Technology Officer atVertex. “In practice, a project team getsstarted and uses whatever approaches theycan to begin to design molecules.”

Part of that philosophy can be attributed todeliberate strategic uses of different discov-

ery technologies, but in the past it also had todo with the slow pace of structure determi-nation. For instance, BioCryst chairman andCEO Charles Bugg notes efforts to determinethe structure of PNP, which began in the1970s, took five or six years.“Now they couldprobably do it in a week,” he says.

Industrial revolutionYet the process of protein structure solutionis becoming speedier and more amenable toautomation, not because of any one break-through but rather a convergence of newtechnologies. Advances in molecular biologymake even difficult proteins easy to express inusable quantities. Scientists can quickly engi-neer proteins to meet crystallization condi-tions, removing or altering troublesomeregions through random mutagenesis ormore directed approaches. X-ray beamlinesare stronger and better focused, allowing theuse of smaller crystals, while detectors aremore sensitive and capture clearer data thanever before—turning this slow, laborious artinto something that proponents of the fieldlike to call ‘high-throughput X-ray crystal-lography.’ The net effect is that for manykinds of drug targets, structure can becomeless of a sideline to drug discovery and moreof an integral part of the process.

In 1999–2000, when capital was flowinginto the biotech sector faster than you could raise the rent on laboratory space, agroup of companies were formed with theidea of using protein structure differentlythan their first-generation predecessors. Eachhas a slightly different strategy, but they allhave something in common: they are takingadvantage of this ‘high-throughput’ struc-ture determination, whether by crystallogra-phy or predictive modeling, particularly the possibilities offered by rapid, iterativesolutions of structures bound to various ligands.

Structure-aided drug design’s next generationKarl A Thiel

Has structural bioinformatics advanced enough to form the core of a drug discovery program? A new generation of companiesexploiting structure-focused technologies is counting on it.

Karl Thiel is a freelance writer based inPortland, Oregon, USA.email: [email protected]

NATURE BIOTECHNOLOGY VOLUME 22 NUMBER 5 MAY 2004 513

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

ureb

iote

chno

logy

Page 2: FEATURE · 2015. 11. 12. · Harren Jhoti, founder and chief scientific officer, left his position as head of Glaxo We llcome’s (now GlaxoSmithKline, Uxbri-dge, UK) Structural Biology

F E AT U R E

At Astex Technologies (Cambridge, UK),rapid X-ray crystallography became the gate-way to a new means of building molecules.Harren Jhoti, founder and chief scientificofficer, left his position as head of GlaxoWellcome’s (now GlaxoSmithKline, Uxbri-

dge, UK) Structural Biology and Bioinfor-matics group in 1999 with the idea of “doingmore with protein structure than just simplyoptimizing leads”—something he says“makes up well over 80% of the use of pro-tein structure in a pharmaceutical company.”

Astex’s approach not only takes advantage ofhigh-throughput crystallography but also cir-cumvents some shortcomings of an oh-so-1990s technology, high-throughput screening.

Instead of finding the kind of large molec-ular weight, high-affinity molecules that

514 VOLUME 22 NUMBER 5 MAY 2004 NATURE BIOTECHNOLOGY

The awkward truth of the matter is that we don’t know exactly why proteins fold up the way they do. Electrostatic forces from the charge-carrying ends of proteins and their side chains, as wellas hydrophobic forces, play the main roles, and in living systemsinteractions with other proteins also play an important role. Yetexactly how these factors so consistently and quickly determine the final shape of a protein is enigmatic.

In the late 1960s, Columbia University biologist Cyrus Levinthalobserved that one would expect an unfolded amino acid chain toachieve a stable three-dimensional shape by finding the lowestenergy state. Yet by sampling all the degrees of freedom in therotational atomic bonds throughout an amino acid chain, even atthe fastest conceivable speed physics would allow, it would takelonger than the age of the universe for even a relatively simpleprotein to find a stable conformation4. Instead, of course, evencomplex proteins fold up in seconds or less. Proteins obviouslydon’t find energy minima through trial and error. And yet they dofind a structure with incredible consistency: X-ray crystallography is only possible because pure proteins always form the same shapeand can thus be crystallized into uniform lattices of atomiccoordinates.

For over 35 years, scientists have pondered ‘Levinthal’sParadox,’ proposing that proteins don’t necessarily find globalenergy minima but rather maintain some energy wells protectedfrom biological perturbations, and simply sacrifice energy effi-ciency to other functions. “Proteins are obviously not optimizedfor stability,” probably because they need to unfold in variouscircumstances, says Krzysztof Fidelis, director of the ProteinStructure Prediction Center at the Lawrence Livermore NationalLaboratory (Livermore, CA, USA).

The enigmatic behavior of proteins has made computationalefforts at predicting what shape an amino acid sequence willassume in the real world rather challenging. Of course, that hasn’tstopped intrepid scientists around the world from trying to designpredictive models anyway. Fidelis is an organizer of the Critical

Assessment of Techniques for Protein Structure Prediction (CASP)experiment, a biannual event in which entrants informally competeto predict a series of unpublished structures purely by computa-tional techniques. He says the prediction methods used fall intothree main categories: comparative modeling, fold recognition andab initio folding.

The first of these looks for sequence similarity between anunknown protein and a protein with known structure, assuming thatindividual features with the same or similar sequence will form thesame or similar shapes, and that proteins in the same family willshare general structural features. Fold recognition is based on thefact that many folds are broadly conserved across proteins of widelydivergent sequence—with a relatively small number, just 20 or so,making up a significant percentage of known protein topology. Foldrecognition is a sort of distant homology analysis focused ondiscovering known folding patterns in sequence data.

That doesn’t help for predicting some entirely novel folds or flexiblefeatures like loops, however. Thus some predictive models have usedan approach often called ab initio folding—based on the principlethat structure can be deduced entirely from sequence and thathomologous models are not necessary. Ab initio methods couldinclude physics-based approaches that seek to determine how atomicbonds rotate, attracted or repelled by charge, in three-dimensionalspace, for instance. But in practice, Fidelis says, most current abinitio methods (now often called “new fold prediction”) really comedown to homology analysis of small sequence fragments, wherestructure is built up out of these small predictions into a larger model.

Although the purpose of CASP is not to find a winner, Fidelisnotes that for the past three CASP events, the most ‘consistentlyinteresting’ performance has come from a system called ROSETTA,developed in the laboratory of David Baker (see http://depts.wash-ington.edu/bakerpg/) at the University of Washington (Seattle, WA, USA).

Sequences for the next CASP event will be published in May, notesFidelis, and results will be discussed this December in Gaeta, Italy.

a b c

Figure 1 Screening by X-ray crystallography. The stages in optimizing binding of a drug lead to a kinase active site are shown. (a) An empty pocket. (b) Amillimolar (mM) fragment ‘hit bound’. (c) A nanomolar ‘lead’ developed by iterating around the initial hit using information on the three-dimensional shapeand chemical nature of the active site. (Images courtesy of Astex Technology, Cambridge, UK)

Box 1 Computer protein origami

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

ureb

iote

chno

logy

Page 3: FEATURE · 2015. 11. 12. · Harren Jhoti, founder and chief scientific officer, left his position as head of Glaxo We llcome’s (now GlaxoSmithKline, Uxbri-dge, UK) Structural Biology

F E AT U R E

come out of typical screening methods, Astexworks with chemical ‘fragments’ with molec-ular weights in the 0.1–0.2 kDa range. Theapproach allows them to screen “a massiveamount of chemical space very efficiently,”says Jhoti, without being tied down to thespecific chemical scaffolds that often limitthe diversity in chemical libraries.

But ‘hits’ in this approach bind weakly,often in the millimolar range, and would be undetectable through most bioassays.“We needed a technology that would be ableto detect the binding of these fragments,”says Jhoti, and the answer was X-ray crystal-lography. By solving co-crystals of targetbound to fragments, Astex researchers cansee where the molecules have attached to thetarget and how they might be turned into realdrugs. “We’re really using X-ray crystallo-graphy as a screening method,” says Jhoti(Fig. 1).

Structural GenomiX (SGX, San Diego, CA,USA), which also uses a fragment-basedapproach in its drug discovery programs,uses its dedicated beamline at ArgonneNational Laboratory’s (Chicago, IL, USA)

Advanced Photon Source to solve co-crystals,refine, and solve again. “The era of FedExcrystallography has arrived,” says CEOTimothy Harris, meaning that the companycan set up experiments at the synchrotron,refine compounds back in San Diego basedon the results, and quickly send back newcrystals for another round. “The whole pur-pose of organizing the way we have is that it’sreal-time, and [structure determination] hasto be co-incident with the time that it takesfor other parts of the drug discovery processto take place.”

It is difficult to put a number on the compounds in the clinic, or even on themarket, that owe part of their genesis tostructural-based drug design—often becauseno mention is made of the use of struct-ural information, traditionally employedduring the lead optimization process. It isclear, however, that the use of structuralinformation is not limited to specialist companies.

“Half of our programs use structure as alead optimization technology,” says JamesSummers, divisional vice president for

advanced technology at Abbott Laboratories(Abbott Park, IL, USA). And for a ‘significant’number of programs, he says, structure isused as a discovery technology—eitherthrough a fragment-based crystallizationapproach comparable to what is being done at Astex and Structural GenomiX, orthrough a technology invented at Abbottcalled structure-activity relationship bynuclear magnetic resonance (SAR by NMR),which looks at binding of fragments to pro-teins in solution2.

Putting your best structure forwardFor the younger companies, structure hasbeen given an even greater emphasis and hasallowed them to avoid expenditures in otherareas. At Syrrx (San Diego, CA, USA), “wewon’t start chemistry on a target until wehave a structure,” says Ken Goodwill, directorof business development. Belief in the valueof structure guides the selection of targets,too. Syrrx is pursuing validated targets forwhich other pharmaceutical companies maybe developing drugs, but for which no crystalstructures exist. “We intentionally go afterones that are more difficult because no oneelse has gotten to them,” says Goodwill.Syrrx’ development programs are currentlybased around dipeptidyl peptidase 4 (DP4), aprotease implicated in diabetes, and histonedeacetylases, potential oncology targets. Thecompany is planning its first clinical trial fora DP4 inhibitor later this year.

“Syrrx has not invested a lot in high-throughput screening, and we still manage tobe very successful. We don’t think it’s neces-sary,” says Goodwill. Removing the emphasisfrom broad screening of chemically diverselibraries could prove the biggest departure ofthese young structure-aided design compa-nies from conventional discovery approaches.

The strategy emphasizes both what com-panies believe are advantages of new struc-

NATURE BIOTECHNOLOGY VOLUME 22 NUMBER 5 MAY 2004 515

Box 2 The flipside: ‘anti-targets’

Both Cengent (in its previous incarnation as Structural Bioinformatics) and StructuralGenomiX started out life hoping to raise substantial revenue by selling subscriptions totheir protein structure databases, an echo of the late-1990s strategy of companies likeIncyte (Palo Alto, CA, USA) and Celera (Rockville, MD, USA), which created privategenomic databases for subscription access. The idea of a commercial database forproteins never really got off the ground, in part because there is little likelihood that a structural database of the human proteome can ever be ‘complete’ in the same way as a genomic database. Private industry has essentially ceded the broad databaseapproach to the government, which is funding the Protein Structure Initiative (PSI)through the National Institute of General Medical Science (Bethesda, MD, USA), with the goal of solving 10,000 diverse structures in ten years.

But there are already hints of how the PSI and related efforts may guide drug design in the future. Cengent, for example, queries its own computational ‘structure bank,’ whichincludes 6,000 unique, computationally predicted structures and another 2,000 novelstructures taken from the public Protein Data Bank, in the lead discovery phase. They not only look for information on what kinds of molecules will bind to a target, but also seekout what Cengent’s CEO Maggio calls ‘anti-targets’—proteins with similar structures thatchemists should design away from in order to avoid unintentional interactions andpotential side effects (Fig. 3).

Indeed, the use of structure as a guide for what a drug shouldn’t bind with is not uniqueto Cengent—Astex researchers, for instance, have solved four human cytochrome P450enzyme isoforms5, which are involved in metabolizing many drugs and are often the sourceof drug-related side effects or toxicity. They have used these structures to help themselvesand partners steer molecules away from interactions that might cause unwanted sideeffects.

Another example of this approach is found at Abbott, where researchers use proprietarystructures of human serum albumin as a sort of negative template for design—making surethat drugs in the bloodstream won’t bind to serum albumin and thus never make it to theirintended target. Structural information helps fine tune medicinal chemistry not onlytoward what will make a drug bind successfully to its target but to how it can avoid otherpotential obstacles.

Figure 2 ACTOR robot for high throughput X-raycrystallography. (Graphic courtesy of AbbottLaboratories)

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

ureb

iote

chno

logy

Page 4: FEATURE · 2015. 11. 12. · Harren Jhoti, founder and chief scientific officer, left his position as head of Glaxo We llcome’s (now GlaxoSmithKline, Uxbri-dge, UK) Structural Biology

F E AT U R E

ture-based technologies and some of the per-ceived shortcomings of older approaches.Astex’ Jhoti asserts that a lot of drug compa-nies were “seduced by potency” in the 1990s,

using screening assays designed to find mole-cules with the highest binding affinity fortheir targets and as a result fishing out high-molecular-weight, heavily decorated com-

pounds that didn’t always leave a lot of roomfor medicinal chemists to work.

Murcko says that for some of Vertex’ pro-grams, the timely availability of structural

516 VOLUME 22 NUMBER 5 MAY 2004 NATURE BIOTECHNOLOGY

Table 1 Structural-aided and in silico drug design companies

Company (location) Focus Contact/comment Year founded

3-Dimensional Pharmaceuticals Structure-based drug design http://www.jnj.com 1993

(Exton, PA, USA) Acquired by Johnson & Johnson,

January 2003, for $88 million

7TM Pharma Focus on GPCR targets, including http://www.7tm.com 2000

(Copenhagen, Denmark) computational structure prediction

Affinium Pharmaceuticals Structure solution and http://afnm.com 2000

(Toronto, ON, Canada) homology analysis Founded as Integrative Proteomics,

from the merger of Chalon Biotech and

Borealis Biosciences. Changed name in

February 2002.

AlgoNomics Rationally designed vaccines based on http://www.algonomics.com 1999

(Ghent, Belgium) structural genomic data.

Astex Technology High-throughput X-ray crystallography http://www.astex-technology.com 1999

(Cambridge, UK) to guide ‘fragment’ best lead discovery

BioCryst Pharmaceuticals Structure-based drug design, focused http://www.biocryst.com 1986

(Birmingham, AL, USA); on PNP inhibitors

NASDAQ: BCRX

Cengent Therapeutics Structure-based drug design and http://www.cengent.com 1997

(San Diego, CA, USA) contract crystallography A merger of Structural Bioinformatics and

Geneformatics in January 2003. (Geneformatics

previously acquired Structure Function Genomics

LLC; Structural Bioinformatics had acquired some

assets of Molecular Applications Group.)

Concurrent Pharmaceuticals De novo drug design and http://www.concurrentpharma.com 2001

(Fort Washington, PA, USA) computational screening

Crystal Genomics High-throughput structure solution, http://crystalgenomics.com 2000

(Taejon, Korea) X-ray and NMR

deCode Genetics Structure solution via Emerald http://www.emeraldbiostructures.com. DeCode acquired 1997

(Reykjavik, Iceland); BioStructures group MediChem Life Sciences in March 2002, which in turn

NASDAQ: DCGN had acquired Emerald BioStructures in May 2000. The

structures group is in Bainbridge Island, WA, USA.

Pintex Pharmaceuticals Structure-based design around http://www.pintexpharm.com 1999

(Watertown, MA, USA) the Pin1 enzyme

Predix Pharmaceuticals De novo prediction of three-dimensional http://www.predixpharm.com 2000

(Ramat Gan, Israel) structure of G-coupled protein receptors Founded as BioIT; merged with Physiome Sciences in

August 2003

Rib-X Pharmaceuticals Focus on ribosomal crystallography for http://www.rib-x.com 2001

(New Haven, CT, USA) novel anti-infectives

Structural GenomiX High-throughput 3-dimensional protein http://www.stromix.com 1999

(San Diego, CA, USA) structure solution Acquired Prospect Genomics

Syrrx (San Diego, CA, USA) High-throughput crystal structure solution http://www.syrrx.com 2000

TRIAD Therapeutics NMR systems to build focused http://www.triadtherapeutics.com 1998

(San Diego, CA, USA) compound libraries

Vertex Pharmaceuticals ‘Rational drug design,’ including http://www.vpharm.com 1989

(Cambridge, MA, USA); structure solution

NASDAQ: VRTX

Xencor (Monrovia, CA) Structure-based protein engineering http://www.xencor.com 1997

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

ureb

iote

chno

logy

Page 5: FEATURE · 2015. 11. 12. · Harren Jhoti, founder and chief scientific officer, left his position as head of Glaxo We llcome’s (now GlaxoSmithKline, Uxbri-dge, UK) Structural Biology

F E AT U R E

information has moved it from a lead opti-mization tool to means for refining drugscreening—by allowing researchers to selectsmaller libraries more likely to contain inter-esting hits. Astex’ Jhoti suggests that a similarchange in emphasis was taking place when hewas at Glaxo Wellcome. In the late 1990s, hesays, more emphasis was placed on gettingstructure earlier in the process and “movingaway from diversity-based high-throughputscreening into focused screening or generat-ing focused libraries.”

In silico drug designBut by deciding to forego industrial-scalescreening altogether, the younger structure-aided companies are often relying muchmore heavily on computational tools.Indeed, although there may be relatively fewexamples of truly de novo drug design, wherechemists invent a drug molecule almost atomby atom to fit a target, the advent of rapidstructure determination has brought somecompanies very close. At Cengent Thera-peutics (San Diego, CA, USA), much of thediscovery and design process takes place in silico before any laboratory work begins.

Cengent specializes in computationalstructure prediction—an approach that hasgrown in accuracy and acceptance in recentyears but still faces a good deal of skepticismamong traditional structural biologists whouse X-ray crystallography as a gold standard.But Edward Maggio, Cengent’s CEO, says thecompany can often get “close to crystallo-graphic” accuracy of the features they aremost interested in for about 550 families ofglobular proteins, including most phos-phatases, kinases and proteases.

The company uses dynamic modeling—predicting how a protein structure moves inspace over time—and a virtual compoundlibrary to find the scaffolds for initial hits, allbefore beginning any wet laboratory work.By the time it comes to screen real molecules,Maggio says libraries are small and hit ratesare on the order of one in ten, rather than theone in thousands or tens of thousands thatcould be expected with broader screening.Cengent researchers still use crystal structuredata when it comes to optimizing com-pounds, says Maggio, but the need for thisslower and more expensive approach is lim-ited by the initial in silico work.

Genomic gnosticsOne potential shortcoming to structure-based drug design is that it theoreticallyplaces a limit on the kinds of targets a com-pany can pursue. After all, if you can’t get astructure of your target, then you can’t use astructure-guided approach. With the grow-ing importance of kinases and other globularproteins as drug targets, that may not bemuch of a limitation, but many of the mostimportant drugs on the market work bymodulating membrane-bound or mem-brane-integrated targets, which are notori-ously difficult to solve experimentally.GPCRs, a particularly important class ofintegral membrane proteins, are the target ofabout half of all commercially available phar-maceuticals. But they have proven almostimpossible to isolate and crystallize, bothbecause they are hard to express in sufficientquantity and also because the natural confor-mation of a GPCR—with seven transmem-brane loops diving in and out of the cellsurface—is tough to mimic in crystallizationconditions when no membrane is present.

Only one GPCR structure has ever beenexperimentally determined, for bovine

NATURE BIOTECHNOLOGY VOLUME 22 NUMBER 5 MAY 2004 517

Table 2 Software and service providers for structure-based and in silico drug design

Company (lcation) Focus Contact/comment Year founded

Accelrys (a subsidiary of Leading supplier of commercial http://www.accelrys.com 1984

Pharmacopeia; San Diego, software for structure solution Formerly Molecular Simulations (MSI), renamed June 2001.

CA, USA); NASDAQ: ACCL MSI acquired by Pharmacopeia in 1998. Subsequently acquired

Synopsis Scientific Systems, Oxford Molecular and GCG.

BioSolveIT Software for molecular docking and other http://www.biosolveit.de 2001

(Sankt Augustin, Germany) in silico design applications

Bio-Xtal (Gif sur Yvette, France) Proteomic services company— http://www.bioxtal.com 1998

including contract crystallography

BSI Proteomics Contract and internal structure solution, http://www.bsiproteomics.com 1997

(Gaithersburg, MD, USA) focus on membrane bound proteins.

De Novo Pharmaceuticals Markets in silico drug design software— http://www.denovopharma.com 1999

(Cambridge, UK) recently terminated internal discovery

programs

IBM (Raleigh, NC, USA); ‘Blue Gene’ project aimed at de novo http://www.research.ibm.com/bluegene/ 1911

NYSE: IBM prediction of protein structure from $100M project begun in December 1999; full Blue Gene

sequence machine will be complete in 2005

Inpharmatica (London, UK) Biopendium database links sequence, http://www.inpharmatica.co.uk 1998

structure, and function for gene products

in the public domain

MolSoft (San Diego, CA, USA) Molecular modeling tools and structure http://www.molsoft.com 1994

prediction software. Founded as Biosoft; name changed in 1995

ProCeryon Biosciences Protein folding recognition software and http://www.proceryon.com 1999

(Salzburg, Austria) other bioinformatics tools and databases.

Including ProHit.

Tripos (St. Louis, MO, USA); Three-dimensional visualization software http://www.tripos.com 1979

NASDAQ: TRPS for lead discovery and optimization

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

ureb

iote

chno

logy

Page 6: FEATURE · 2015. 11. 12. · Harren Jhoti, founder and chief scientific officer, left his position as head of Glaxo We llcome’s (now GlaxoSmithKline, Uxbri-dge, UK) Structural Biology

F E AT U R E

rhodopsin3, a feat that was possible in partbecause researchers were able to isolate largequantities of the protein from cow eyeballsreadily available at slaughterhouses.

But a couple of young companies are taking the growing knowledge of proteinstructure a step further into virtuality, doingstructure-based drug design of GPCRs—not through traditional structure determina-tion, but purely through computational prediction.

The claim that a computer program canaccurately predict structures of proteins thathave never been experimentally solved may at first smack of a sort of Gnosticism.But the idea is not so far-fetched. AlthoughMichael Kauffmann, CEO of Predix Phar-maceuticals, (Ramat Gan, Israel) acknowl-edges that his research team can never really‘know’ their predictions are accurate, hestresses that they can check how knownGPCR-binding drugs interact with their pre-dictive models and thus get a proxy measureof structural accuracy.

And while membrane-bound proteinstructures are hard to solve experimentally,Krzysztof Fidelis, director of the ProteinStructure Prediction Center at the LawrenceLivermore National Laboratory (Livermore,CA, USA), points out that they are poten-tially easier targets for computer predictionsthan globular proteins.

Fidelis is one of the organizers of theCritical Assessment of Techniques for ProteinStructure Prediction (CASP) experiment, inwhich researchers try to solve unknownstructures computationally (Box 1). Fidelisnotes that whereas current methods lack theability to accurately predict structure of glob-ular proteins from amino acid sequencealone, the outcome of prediction is poten-tially much more favorable for membraneproteins, including many features of GPCRs.

Often one can identify from sequence thelocation of the alpha helices—the undulatingsection of the protein that dips in and out ofthe membrane—and where they intersect thecellular membrane. Predictive techniques forglobular proteins are greatly enhanced by awealth of experimental data that can be usedfor homology modeling. “If we had the sameamount of experimental data for GPCRs,” sayFidelis, “we would have solved this [structureprediction] problem a long time ago.”

Kaufmann notes that many GPCR-tar-geted drugs bind right where the alphahelices meet the membrane, so this is wherethey’ve concentrated their energies. Predix’computational technique, PREDICT (whichhas never been part of the CASP event), gen-erates over a billion possible structureswithin the confines of the membrane envi-ronment, then narrows those down byassuming that, all other things being equal,

the fold will go to the lowest energy state. Thefinal model is not necessarily the lowestenergy state but the best low-energy optionthat also accounts for chemical and biophysi-cal properties of the membrane (for example,loops can only be so close together). The finalchoice is usually made by docking knownmolecules into the model—thus, the com-pany works with GPCRs that have been validated as drug targets but are onlyaddressed by one or two commercial drugs,leaving the company room to improve on thecompetition through structure-guided medi-cinal chemistry.

Predix’ lead product PRX-00023, an antag-onist of 5-hydroxytryptamine-1A (5HT-1A,or serotonin) intended to treat anxiety andother neuropsychiatric disorders, entered aphase 1 clinical trial in February, makingPREDICT the most ‘virtual’ discovery pro-gram to ever produce a clinical candidate.However, other companies have based opti-mization programs on protein structure pre-diction. Abbott’s drug Atrasentan, forinstance, currently in phase 3 trials for thetreatment of certain prostate cancers, targetsthe endothelin-A receptor, a GPCR. Accord-ing to Summers, this compound was opti-mized on the basis of a homology model ofthe target, predicted using theoretical GPCRstructures adjusted for the endothelin-Areceptor sequence.

The scariest form of flatteryYet if the new emphasis on structure bringsthe results these companies are hoping for,what is to stop better-financed imitatorsfrom duplicating their approach? Most ofthese small biotech companies are, at best,only now readying for their initial clinicalstudies and face a long expensive roadthrough a hostile financing environment.How will they survive against deep-pocketedcompetitors?

“Eventually you’ll see an evolution insideof big pharma where programs won’t get thatfar down the pike unless there’s a structureassociated with the program,” says JohnMendlein, chairman and CEO of Affinium(Toronto, ON, Canada). And if that shifttakes place, the burgeoning new field ofstructure-aided drug design companies mayshift some more, too. Already there has beenconsiderable consolidation in the industrysubsector (Tables 1 and 2), and there arehints that if some big pharma companiesdecide to commit more to structural bioin-formatics, they’ll bring it in house.

Johnson & Johnson’s (J&J; New Bruns-wick, NJ, USA) recent acquisition of 3-Dimensional Pharmaceuticals (3DP; Yardley,

518 VOLUME 22 NUMBER 5 MAY 2004 NATURE BIOTECHNOLOGY

Figure 3 Anti-targeting approach to drug dosing. Side-by-side surface views of the electrostatic chargedistribution for the drug target PTP-1B, co-crystallized with an active-site binding ligand (upper leftcorner), and five structurally similar proteins (‘anti-targets’). Comparative views of additional physicaland biochemical properties—dynamic flexibility, hydrophobicity or strain energy—together provide anunderstanding of subtle structural differences between drug target and anti-targets. (Graphic courtesyof Cengent Therapeutics, San Diego, California, USA.)

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

ureb

iote

chno

logy

Page 7: FEATURE · 2015. 11. 12. · Harren Jhoti, founder and chief scientific officer, left his position as head of Glaxo We llcome’s (now GlaxoSmithKline, Uxbri-dge, UK) Structural Biology

F E AT U R E

PA, USA) for $88 million has already causedsome minor ripples—J&J’s research agree-ment with Cengent has been terminatedsince the 3DP acquisition, and it’s not clearthat they will continue a current relationshipwith Astex. Amgen (Thousand Oaks, CA,USA), which normally relies heavily on collaboration, got acquisitive when it came to building out its structure-aideddesign capabilities around protein kinases.In 2000, they bought Kinetix Pharma-ceuticals (Boston, MA, USA) for $170 mil-lion. Acquisition may not be the worst fatefor some young structure-aided design com-panies, but major pharmaceutical companiescould also decide to expand structural biol-ogy groups organically.

In 1999, as a new generation of structure-based drug design companies were forming,Abbott implemented major improvements toits own structural biology capabilities, takingadvantage of many of the same technologydevelopments that the new companies wereleveraging. “We put into place robotic sys-tems that automate just about every phase ofthe crystallography process,” says JamesSummers, divisional vice president ofAdvanced Technology at Abbott. Perhaps

most significant, he says, is a robot calledACTOR (automated crystal transport, orien-tation, and retrieval; Fig. 2). “It allows you tomount crystals in the X-ray beam and keepthem at liquid nitrogen temperature, alignthem perfectly with the X-ray, and thenrepeat that process over and over for multiplecrystals.” (Interestingly, both Abbott’sACTOR and a crystallization robot fromSyrrx won R&D Magazine ‘R&D 100’ awardsin 2002.)

Yet structure determination, in particulargetting from a target sequence to a usablecrystal, is still as much an art as a science.When it comes to solving a difficult butpotentially important target, larger compa-nies may not have any advantage over spe-cialists, where pure dedication to getting ananswer may trump financial resources.

And structure-aided design has a prag-matic advantage over some other technology‘platforms’: it is focused on the identificationand refinement of drug leads that can bepatented and ultimately put in the clinic. Ifthe approach works, the early movers maynot be able to block imitators, but theyshould at least have an attractive pipeline of drug candidates, which is where the rub-

NATURE BIOTECHNOLOGY VOLUME 22 NUMBER 5 MAY 2004 519

ber meets the road for investors and partnersalike.

Syyrx’ Goodwill believes that the shifttoward structure at larger companies willcome slowly, if at all, because of the invest-ment they have made in high-throughoutscreening. Adds Astex’ Jhoti: “I think that inthe 1990s, the big pharma companies learnedthat you don’t always have to buy a technol-ogy and bring it in house to get its full poten-tial. Sometimes you can actually destroy whatyou bought.”

“Besides,” he adds,“if a technology is beingreplicated, that’s a sign we were successful.”

1. Goldstein, J.L. Remarks at the Awards Ceremony, AlbertLasker Award for Clinical Research, 1999.http://www.laskerfoundation.org/awards/library/1999remarksgsclin2.shtml

2. Shuker, S.B. et al. Discovering high-affinity ligands forproteins: SAR by NMR. Science 274, 1531–1534(1996).

3. Palczewski, K. et al. Crystal structure of rhodopsin: a Gprotein–coupled receptor. Science 289, 739–745(2000).

4. Levinthal, C. How to fold graciously. in MössbauerSpectroscopy in Biological Systems. Proceedings of aMeeting Held at Allerton House, Monticello, Illinois.(eds. DeBrunner, J.T.P. & Munck, E.) 22–24(University of Illinois Press, Chicago, IL, USA, 1969).

5. Williams, P. et al. Crystal structure of humancytochrome P450 2C9 with bound warfarin. Nature424, 464–468 (2003).

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

ureb

iote

chno

logy