CUSTOM-MADE DRUGMost drugs on the market today were found either by chance observation or by systematic screening of large numbers of natural and synthetic substances. In many cases, trial-and-error refinement of substances uncovered by the first two means later led to increased potency or decreased toxicity. Those traditional methods of drug discovery are now being supplemented by a more direct approach, made possible in part by improved understanding of the molecular interactions that underlie diseases.
More and more investigators, including us, are beginning to derive fruitful results from what is called structure-based drug design. Our starting point is not the drug, but its molecular target in the body. We solve the three-dimensional structure of a substance known to participate in some disorder. Then we build a chemical that precisely fits the target and alters its activity. For instance, we might design a compound to block the catalytic site of an enzyme essential to viral replication. We would thereby prevent the virus from reproducing and so would halt the spread of infection.
By somewhat simplistic analogy, standard tactics of drug discovery are akin to making and testing many keys in order to find one that happens to fit a lock of unknown shape. In contrast, prior study of the shape and arrangement of tumblers in a lock would lead to rapid design of an effective key.
In spite of the relative inefficiency of the traditional approaches, they have provided treatments for everything from minor aches to life-threatening illnesses. Today they also benefit from automation, which has markedly speeded up large-scale screening efforts. Yet structure-based methodology can yield promising drugs more quickly and less expensively. Indeed, because the final products are custom-tailored to their targets, they tend to be more potent, more specific and less toxic than remedies discovered in other ways.
Our experience in creating a set of potential drugs -- one of which is now in human trials for psoriasis and a form of Tcell lymphoma -- illustrates the process and the power of structure-based design. Ours is by no means the only product of structure-based technology to have reached an advanced stage of development, however. Another, called captopril, is already widely used for the treatment of hypertension. Several others, produced by various laboratories, are being evaluated in humans for therapy of a host of disorders, including cancer, AIDS, glaucoma and the common cold. A number of additional compounds are in less advanced stages of investigation. Although not every drug that reaches late stages of testing proves useful for therapy, the products that are now in or are moving toward clinical trials are quite impressive.
The scientific concepts underlying all this activity were understood 50 years ago, but their practical application lay beyond the reach of the existing technology. Paul Ehrlich, the German bacteriologist, had long since demonstrated that drugs often induce physiological effects by binding to target structures (receptors) that participate in normal cellular activities. Investigators appreciated as well that the shape of a drug and its chemical composition had to complement that of the binding site on its receptor -- typically a protein. Nevertheless, progress in drug discovery proceeded slowly, by empirical means, until the 1970s.
At that time, new methods became available for obtaining pure samples of many protein targets. Simultaneously, workers improved x-ray crystallography, the one imaging technique then capable of revealing protein structure. In this technique, crystals are bombarded with x-rays. The crystal diffracts the rays, creating a splatter of spots on photographic film or on newer types of electronic detectors. The distribution of atoms within the crystal influences the diffraction patterns that emerge. Hence, with the help of sophisticated computer programs, the patterns can be translated into maps indicating the three-dimensional structure of the protein molecules within the crystal.
Miguel A. Ondetti and David W. Cushman and their colleagues at the Squibb Institute for Medical Research (now Bristol-Myers Squibb) were the first to successfully exploit crystallography for drug development. They did not know the precise architecture of their target: human angiotensin-converting enzyme, a participant in hypertension. But they did know the conformation of a closely related enzyme. Using that information, they came up with captopril in 1975.
Two of us (Montgomery and Bugg) began to collaborate on drug design in the late 1970s. A decade later our program moved to BioCryst Pharmaceuticals, which Montgomery and Bugg helped to establish. BioCryst is among several young companies devoted to structure-based drug design. The pioneer was Agouron Pharmaceuticals in La Jolla, Calif., formed in 1984.
In the 1970s, as now, enzymes were the favored targets because they control important biochemical processes in diseases. Also, one can impair their activity relatively easily, by fitting small molecules into their active (catalytic) sites. Drugs aimed at other kinds of targets -- including nucleic acids (DNA and RNA) and protein receptors for certain hormones -- are under study as well. But in many cases, these targets pose greater design challenges. For example, compounds created to regulate the activity of many hormone receptors would have to be more complex than enzyme inhibitors and would have to form more bonds with the receptors.
One of our major objectives was to design a molecule that would inhibit the enzyme purine nucleoside phosphorylase (PNP). PNP normally operates in the "purine salvage pathway" of cells. It takes up individual nucleosides, which consist of a nitrogenous substance known as a purine base (such as guanine) and a sugar. With the help of a phosphate ion, PNP cleaves the purine from the sugar, giving rise to a free purine base and a phosphorylated sugar. Once the purine is released, the cell can destroy it or recycle it to build any of a number of molecules, such as a building block (nucleotide) of DNA.
NORMAL ACTIVITY OF PNP/ GOAL OF DRUG DESIGN
Unfortunately, PNP can also cleave certain anticancer and antiviral agents that are synthetic mimics of natural purine nucleosides; it can thereby interfere with therapy. One such substance is ddI (2', 3'-dideoxyinosine), which the Food and Drug Administration approved as a treatment for AIDS in 1991. It was our goal to construct an entity that when administered with the nucleoside mimics would inactivate PNP long enough for the anticancer and antiviral agents to accomplish their therapeutic missions. To serve as a drug, our compound would also have to be able to cross membranes to the interior of cells, where PNP does its work. Other investigators had already identified a few inhibitors of PNP, but none that was potent enough to be useful for therapy and also capable of crossing the cell membrane intact.
Shortly after we began this project, we gained even more incentive to design powerful PNP inhibitors. Accumulating evidence indicated that the body needs PNP for the proper functioning of T cells but not other components of the immune system. We, like other research teams, quickly recognized that inhibitors of PNP might selectively suppress the excessive T cell activity associated with an array of autoimmune disorders, such as rheumatoid arthritis, psoriasis, systemic lupus erythematosus, multiple sclerosis and insulin-dependent (juvenile-onset) diabetes.
Having fixed on PNP as our target protein, we followed a systematic strategy for designing inhibitory compounds. In short, we first determined the three-dimensional arrangement of the target's constituent atoms, paying particular attention to the active site. Next we turned to our computers. As we viewed a candidate on a monitor, we worked it into the active site, examining how well the shape and chemical structure of the candidate would complement that of the site. We also used programs to help us estimate the strength of the attractive and repulsive intermolecular forces between a candidate and the active site.
A tight fit is necessary for potency and specificity. A drug that remains bound to its target and inactivates it for a long time can be administered in lower doses than can one that separates from its target rapidly. Further, a substance designed to mesh perfectly with a particular binding site of one protein is unlikely to interact well with any other molecule; therefore, the substance should minimize unwanted interactions and, with them, side effects.
We synthesized only those chemicals that our computer simulations suggested would have greatest affinity for the target. Then we assessed the effects of the drugs on the activity of the target molecule and compared the proposed and the actual fit. Because modeling programs are imperfect, certain compounds we synthesized did not live up to expectations. After exploring the reasons for the successes and the failures, we returned to the computer to propose modifications that might increase the effectiveness of drug candidates.
This iterative strategy -- including repeated modeling, synthesis and structural analysis -- led us to a handful of highly potent compounds that tested well in whole cells and in animals. Had a compound encountered a difficulty in the cellular or animal tests (such as trouble passing through cell membranes), we would have revisited the computer to correct the deficiency. Then we would have cycled a modified drug through the circuit again.
The entire protocol, from choosing the target to creating a drug suitable for clinical trials, can probably be accomplished today in two or three years. But back in the 1970s the first crucial step, determining the structure of the target, proved to the most laborious of all. It occupied our attention and that of a team of crystallographers led by Steven E. Ealick, then at the University of Alabama at Birmingham, through most of the 1980s.
In our case, the stumbling block did not lie with obtaining pure PNP or converting the protein into crystals. Robert E. Parks, Jr., and Johanna D. Stoeckler of Brown University had already isolated the enzyme from human cells. They supplied quantities to William J. Cook, also at Birmingham, who succeeded in preparing the well-ordered crystals required for x-ray studies.
We were also able to demonstrate easily that crystalline PNP is essentially identical to PNP in the body. If the structure were profoundly different, we would have had no justification for basing drug design on the crystal structure. In addition, Jon M. Crate, then a graduate student in Bugg's laboratory, established that a crystal of the protein was able to function normally. It catalyzed the same reaction that PNP induces in living systems.
The real aggravation arose when we proceeded to the detailed structural determination. In the early years we had to depend on an x-ray source that generated relatively low-intensity waves. The resulting low-resolution diffraction patterns enabled us to discern the overall shape of the molecules, but we could not properly place the individual atoms. We eventually filled in the missing details in collaboration with John R. Helliwell of the Daresbury Laboratory Synchrotron Radiation Source in England. The synchrotron emitted the intense x-rays needed for high-resolution imaging. Today greatly improved equipment and more synchrotron facilities are available for protein crystallography.
The x-ray data established that PNP crystals are highly porous, a feature that proved useful for our understanding of the ability of proposed drugs to inhibit PNP activity. We learned, too, that the functional PNP enzyme exists as a trimer: a unit of three joined monomers (single PNP molecules) [see illustration below]. And we showed that the trimer has three identical active sites, one at each junction between monomers. (Hereafter, we will speak as though there was just one active site, formed by two adjacent monomers. )
Information of perhaps greater import emerged from analyses of the complexes formed when synthetic nucleosides, including previously discovered inhibitors, were attached to the active site. This work showed us the shape of the site, which is essentially an irregular indentation on the surface of the enzyme. These investigations additionally revealed the identity of the exact amino acids constituting the active region; such detail was a prerequisite to drug design.
The structural determinations also yielded a surprise. The shape of the enzyme apparently changes when the active region binds another molecule. In other words, the lock-and-key analogy mentioned earlier has a fallacy the shape of the lock is not static, but flexible. For instance, a stretch of amino acids in each PNP monomer forms a loop that serves as a swinging gate. It often covers the nearest active site, but it can move to accommodate a nucleoside or some imitator. Awareness of these conformational changes critically aided our modeling efforts because we could predict which parts of the PNP structure could change shape to interact with a proposed inhibitor.
A clear understanding of our target enabled us to concentrate on assembling inhibitors of PNP. Consequently, in the late 1980s, we formed a design group that included the three of us and Ealick, along with Mark E. Erion and Wayne C. Guida of Ciba-Geigy in Summit, N.J., Y. Sudhakar Babu of BioCryst and John A. Secrist 111 of the Southern Research Institute in Birmingham, Ala. This assembly was rather small compared with the armies of chemists and pharmacologists that traditionally have been required to screen and synthesize drug candidates.
We focused initially on filling the purine binding region of the active site. That done, we planned to attend to the sugar binding region and, finally, the phosphate binding area. We expected that each successive step, which moved us closer toward fully occupying the active region, would enhance the affinity of a drug candidate for the enzyme.
From our crystallographic examinations, we knew that three amino acids in the purine binding pocket of PNP form hydrogen bonds with purines and their mimics. Such linkages are among the strongest reversible chemical bonds that exist. (Hydrogen bonds consist of two atoms, usually two nitrogens or one nitrogen and one oxygen, that share a hydrogen atom.) In proposing candidates for our inhibitor, then, we concentrated on compounds that, at the least, would form hydrogen bonds with the same three amino acids, all of which reside on a single monomer.
We thought we could achieve strong binding by substituting a carbon atom for a specific nitrogen atom in the purine guanine. Guanine consists of a combination of five carbon and four nitrogen atoms, arranged into two adjacent rings. (Each carbon and nitrogen in the rings is assigned a number from one to nine.) From the rings protrude several hydrogen atoms, a single amino group (NH2) and one oxygen atom [see a in box of next figure]. In particular, we favored exchanging a carbon atom for the nitrogen atom that normally occupies position nine. This choice appealed to us because we knew from earlier studies that such a change promotes binding to PNP. Guanine modified in this way is referred to as 9-deazaguanine; the number indicates the site of change, and the term "deaza" means without nitrogen.
We further expected that attaching an amino group to the carbon atom in position eight on our altered guanine molecule would enhance the purine's affinity for PNP. After all, the most potent membrane-permeable inhibitor of PNP available in the 1980s incorporated an amino group at exactly that position.
Taking a stepwise approach, we made one change at a time in the purine part of potential inhibitors. Then we tested the activity of the resulting molecules by examining their ability to block PNP from catalyzing the cleavage of nucleosides in the test tube. As we anticipated, substitution of carbon for the nitrogen in position nine of guanine resulted in an inhibitor that blocked PNP quite well. But, to our disappointment, adding the second change to the first one did not yield the superior inhibitor we expected. In fact, our best hope turned out to be quite a poor performer.
In the absence of detailed structural information, we would have been mystified as to why affixing the amino group to the carbon in position eight proved unhelpful. But crystallography quickly provided the explanation. We separately analyzed the structures of complexes formed by PNP and four different compounds [see box below]. The purine part of the compounds consisted either of pure guanine, the carbon-substituted form (9-deazaguanine), guanine carrying an extra amino group at position eight (8-aminoguanine) or the doubly modified guanine (8-amino-9-deazaguanine).
We saw that one of the three amino acids forming the guanine binding site of PNP -- an asparagine in the 243rd position on the protein chain -- interacts with guanine by forming one strong and one weaker hydrogen bond. At the same time, asparagine establishes a hydrogen bond with a neighboring amino acid, threonine 242. This bond stabilizes the linkage of asparagine to guanine.
The carbon-for-nitrogen switch in the 9-deaza variant favors association with PNP in a fairly straightforward way: it substitutes a strong hydrogen bond for the relatively weak one occurring between asparagine 243 and guanine. Formation of a simple 8-aminoguanine variant leads to tight binding in another way, by giving rise to an extra hydrogen bond between the purine and PNP. Specifically, a direct hydrogen bond arises between the added amino group and threonine 242.
HOW A DESIGN MYSTERY WAS SOLVED.
The combination of the two "improvements" -- the carbon-for-nitrogen substitution and the addition of the amino group to position eight -- was counterproductive because the carbon in position nine prevented the amino group at position eight from forming the extra bond with threonine 242. In fact, it set up an unfavorable, repulsive clash between the threonine and the added amino group.
Based on these findings, we immediately realized that pursuing inhibitors incorporating 8-amino-9-deazaguanine would be futile; 9-deazaguanine itself would be a better choice for the purine component of an inhibitor. This experience underscores the wonderful economy of the structure-based approach. As often happens to more conventional drug designers, we had headed directly into a blind alley. But our access to detailed structural information enabled us to retreat rapidly. Without crystallographic data, we might have pursued a logical but unproductive avenue of research much longer than we did.
The next task was to fill the sugar binding site. The sugar in a nucleoside does not attach to PNP primarily by forming hydrogen bonds. Rather more powerful hydrophobic attractions come into play. The hydrophobic effect is familiar to anyone who has ever watched oil separate out of salad dressing. Oily molecules have little affinity for water and are drawn to one another instead. The sugar binding pocket of PNP consists of three hydrophobic amino acids. Two of these amino acids (a phenylalanine and a tyrosine) come from the same monomer that binds guanine. The third (a phenylalanine) is contributed by the adjacent monomer.
Several known inhibitors carried a benzene group (a ring formed by six carbon atoms and their associated hydrogens) in place of the sugar in nucleosides. (The sugar is attached to position nine of the purine.) Hence, we examined the inhibitory effect of an assemblage consisting of benzene linked to position nine (now a carbon atom) of 9-deazaguanine. The compound worked best when we joined the benzene group to 9-deazaguanine, not directly but through an intermediary: the carbon atom of a methylene group (CH2). Still, we suspected we could do better.
On the computer screen, we saw that the sugar binding pocket could be filled more completely by adding any of several chemical groupings to the benzene ring. In reality, a few additions we tried were not good, for reasons we ascertained by crystallography. Others worked strikingly well, however. The best fit came from adding a chlorine atom to the carbon atom in position three of the benzene ring.
We had completed two thirds of our plan. The final step was adding a group that would interact with the phosphate binding site. We could not use phosphate itself, partly because phosphate containing compounds have difficulty passing through cell membranes intact. Initial modeling studies encouraged us to prepare several structures that failed to improve the binding affinity of our two-part structure. Basically, chemical modifications that seemed reasonable on the computer screen gave rise in reality to substances that could not orient properly in the active site. Although we were disappointed, we were grateful that, once again, crystallography made it possible for us to discern the causes of the failure and to abandon doomed strategies. Indeed, information gleaned from crystallography helped us perform calculations enabling us to arrive at compounds that bound well to PNP.
For instance, calculations based on the crystallographic findings spurred us to add an acetate group (CH2COO-) to the methylene carbon atom that joined 9-deazaguanine to the chlorinated benzene ring. This step positioned the carboxyl segment (COO-) of the acetate group so that it could lock on the phosphate binding site. The carboxyl unit is attracted to a positively charged amino acid in the active site of PNP.
When we tested the ability of the completed molecule to inhibit PNP, the results were gratifying. The compound blocked nucleoside cleavage 100 times more effectively than did any candidates in existence at that time. We were ecstatic. But would this compound -- and others we assembled that were 10 to 20 times more potent than existing inhibitors -- be useful in the body? To answer this question, we studied the ability of several of our molecules to protect the AIDS drug ddI from degradation by PNP in rats. All of them prolonged the half-life of this nucleoside mimic. Recent tests have shown that our PNP inhibitors can protect other nucleoside analogues and can suppress T cell functions in cultured cells and in experimental animals.
Our original goal of developing PNP inhibitors that could prolong the survival of synthetic nucleosides in animals, envisioned more than a decade earlier, had been accomplished. And it had been achieved less than three years after a small group of organic chemists began synthesizing candidates. In traditional practice, the invention of enzyme inhibitors often takes more than 10 years and can cost tens of millions of dollars. We needed to prepare only about 60 compounds in order to identify highly potent inhibitors, a small number in comparison to the hundreds or thousands of candidates that investigators using standard approaches generate.
Researchers at the Washington University School of Medicine have recently completed two combined phase I and phase II clinical trials of one of our best inhibitors: BCX-34. These were small trials that looked at both safety and efficacy. Because the drug performed well against psoriasis and cutaneous T cell lymphoma, extended phase II (efficacy) trials have been planned by BioCryst. They may have already begun by the time these words are printed. Our most potent PNP inhibitor and its immediate relatives have been licensed to Ciba-Geigy for possible treatment of arthritis; these compounds are now being tested in animals.
We and others have had very satisfying experiences with structure-based drug design. Nevertheless, important stumbling blocks must be tackled if this strategy is to live up to its potential. At a fundamental level, the molecular interactions that give rise to disease are not always clear. In such cases, investigators may not be able to pinpoint the best targets to study. Another obstacle relates to the fact that most drug targets are proteins. Although the techniques of molecular biology have made many proteins available in quantity, some of them are still difficult to obtain in pure form. Even where supply is not a problem, solving the structure of a protein continues to be difficult. Crystallography works beautifully if one has well-ordered crystals, but proteins as a group (and nucleic acids, too) are challenging to crystallize in forms suitable for high-resolution x-ray diffraction studies. Membrane-bound proteins, including various hormone receptors, are particularly recalcitrant; they are oily and cling to one another haphazardly.
Many protein crystallographers in pharmaceutical companies have begun to ameliorate the crystallization problem by turning to robotics. Robotic systems can automatically assess thousands of combinations of conditions under which crystallization occurs, suggesting optimal solutions. At the same time, the equipment for obtaining and interpreting x-ray diffraction images is improving, and techniques are being developed for determining the three-dimensional structure of large, noncrystallized proteins in their natural, aqueous environment. A method known as nuclear magnetic resonance spectroscopy (NMR) has been applied to solve the structures of some proteins, but none yet as big as PNP. That limitation may be overcome in the future.
One day researchers may be able to bypass crystallography and NMR altogether, deducing the three-dimensional structure of protein targets directly from their linear amino acid sequence. At the moment, though, the predictive ability of computer programs leaves something to be desired, both for solving the structure of a molecular target and for assessing the fit and attraction between a proposed drug and its target. Predictive accuracy is improving, however, as the theoretical underpinnings of existing programs expand. Eventually, computer analyses may lead workers directly to the best composition of a drug, freeing them from having to synthesize and test less effective, intermediary compounds. The ideal would be to make the best drug on the first try, be it from scratch or through the alteration of an existing substance.
Although this goal remains distant, our development of powerful inhibitors of PNP, and the parallel invention of different enzyme inhibitors by other structural chemists, demonstrates that, perfect or not, structure-based drug design is already quite useful. Among the drug candidates now in clinical trials are two created separately at Abbott Laboratories and at Merck Research Laboratories. These drugs inhibit an enzyme made by the AIDS-causing human immunodeficiency virus (HIV). This enzyme, a protease, is required for the accurate assembly of viral particles and for their spread from cell to cell. The structural approach enabled the drugs to reach human testing in less than four years.
SOME PROMISING DRUGS from structure-based design
Agouron has engineered enzyme blockers now being studied in cancer patients. These drugs inhibit thymidilate synthase, which participates in synthesis of the nucleotides cancer cells need in order to replicate their DNA and proliferate. And an inhibitor developed by Merck is being examined for its ability to combat glaucoma by interfering with the enzyme carbonic anhydrase. Merck is also working on a drug, not yet in clinical trials, to treat emphysema. It shuts off human neutrophil elastase, an enzyme that has been implicated in damaging lung tissue; human neutrophil elastase may also contribute to rheumatoid arthritis and acute respiratory distress syndrome.
Twelve years ago just one pharmaceutical house in the U.S. -- Merck -- had assembled a team of investigators dedicated to structure-based drug design. Today most large drugmakers in this country employ such teams, as do various other companies around the world. Several corporations that rely mainly on this method are emerging on the American pharmaceutical landscape. All indications suggest that structure-based drug design is here to stay and will make a major contribution to the drugs brought to market in the years to come.
CHARLES E. BUGG, WILLIAM M. CARSON and JOHN A. MONTGOMERY collaborate on drug design in Birmingham, Ala. Bugg is director of the Center for Macromolecular Crystallography, associate director of the Comprehensive Cancer Center and professor of biochemistry at the University of Alabama at Birmingham. Carson is director of the Computer Graphics Core Facility at the Center for Macromolecular Crystallography. Montgomery is executive vice president and director of research at BioCryst, a company devoted to structure-based drug design, and distinguished scientist at the Southern Research Institute.
OVERALL STRUCTURE OF PNP
BEST INHIBITOR OF PNP