DRUG DESIGN – Fragment-Based Drug Design: Delivering Drugs That Hit Multiple targets, Leveraging Insights From Systems Biology


Everyone says life is complicated, but systems biologists have a unique perspective thanks to computer models that probe the dense interconnections of the body’s myriad cells and tissues. The field is exploding. At last count, the National Institute of General Medical Sciences was sponsoring more than a dozen systems biology centers around the country. Harvard Medical School is taking the next step: an initiative in “systems pharmacology” that puts chemists and biologists together with clinicians and computer scientists to view how drugs perform in living systems.

These many-faceted efforts to understand illness in the context of networked cellular systems will eventually yield fresh insights about drug targets. And, in the spirit of systems biology, the science will also show us how the targets interact along different metabolic pathways involved in complex diseases, such as diabetes,cancer, and autoimmune disorders.

This sophisticated view of living systems has huge implications for drug development. I believe the next generation of pharmaceuticals will have to hit more than one target simultaneously in order to fully exploit the knowledge gained from systems biology. Currently, the most effective strategy to achieve this rational design goal is computational fragment-based drug design (CFBDD). This in silico approach is not only well synchronized with the objectives of systems biology, but also sets up important signposts for the new discipline of systems pharmacology.

Based on reliable predictions of where, and with what binding affinity, small chemical fragments – organic-based moieties commonly found in the drug discovery process – interact with disease targets, such as enzymes and cell receptors, drug designers can assemble new drug-like molecules. The fragments, once linked together synthetically, are much more active biochemically. And the entire progression from fragment identification through lead optimization can be performed using predictable software models. Delivering on this vision is fraught with obstacles, given the difficulty of modeling molecular interactions. But computational chemists must take on the challenge. At BioLeap, we are refining the concepts and methods that will allow CFBDD to be the standard approach of early stage small-molecule drug design.

Today’s Tough Development Environment

Combination therapy – administering multiple small-molecule drugs – is the current standard for modulating multiple disease targets simultaneously. However, this approach faces numerous hurdles, such as matching the pharmacokinetics of multiple targeted drugs, coping with drug interactions, and navigating new regulatory obstacles. A second approach is to screen each target with the same chemical collection using high throughput screening (HTS) and looking for single compounds that hit each member of the target panel. But based on what we know today, the odds of success using traditional methods to pinpoint a single molecule that hits four or five different targets are slim, and attempts to do so are bound to be very expensive.

Even if these approaches started to yield significant results, aided by advances in pharmacogenomics, there are structural and economic issues that could stand in the way. In a sweeping study titled Beyond Borders: Global Biotechnology Report 2012, consultants Ernst & Young note that the rush of innovations in drug development hasn’t lowered the cost, increased the pace, or appreciably raised the number of new drugs reaching the market each year. It still takes more than a decade and upward of $1 billion to create a novel drug, including the cost of efforts that fail, E&Y contends. Despite the new technologies: “drug development is still linear, slow, inflexible, expensive, and siloed.”

The litany of familiar industry woes is even longer than this. According to the Tufts Center for the Study of Drug Development, just 3 of 10 compounds sold by pharma companies bring revenues that match or exceed development costs – and many of those will soon be off patent. London-based consultants EvaluatePharma reckon that between 2011 and 2018, more than $290 billion in prescription drug sales are at risk from the patent cliff. Meanwhile, the drug industry is shedding research jobs at an alarming pace, and funding hurdles are crimping the ability of biotech start-ups to deliver game-changing ideas. According to BioWorld, public and private biotech firms raised 40% less capital in the first half of 2012, compared with the same period in 2011.

How CFBDD Facilitates Polypharmacy

No single drug development strategy or breakthrough can address all the issues we have just enumerated. Yet there are reasons to believe that a fragment-based approach can produce superior results that genuinely make a difference. The reason has to do with the intrinsic virtues of fragments. It’s not uncommon for small chemical probes to bind to more than one target. The diverse clustering of fragments finds the protein hot spots or sites of high interaction energy.1 Finding and utilizing these sites is an important step in multi-targeting drugs. Also, by building up from small fragments, you’re able to explore new chemistry and new chemical spaces that effectively map whole, and multiple, proteins. In short, by rationally combining the fragments on multiple proteins, you maximize the likelihood of generating a multi-talented drug.

While these are still early days for CFBDD, the methodology is already proving its worth. By using predictive models, we can weed out the drug candidates most likely to fail before spending time and money synthesizing and testing them. Ultimately, the strategy of targeting multiple proteins with a single compound will eliminate the hazards of multi-drug interactions, avoid the hit-ormiss pitfalls of HTS, and simplify the regulatory review when compared with combination therapies. The result is true polypharmacy: a single small molecule drug that hits multiple targets simultaneously.

Two years ago, Drug Discovery Today published a sweeping review of the fragmentbased design field, spotlighting advances at Abbott Laboratories, AstraZeneca, Roche, and Eli Lilly, as well as Vertex and other biotech startups.2 In many cases, the research involved exploiting fragment binding data derived using x-ray, surface plasmon resonance, nuclear magnetic resonance (NMR), and other traditional tools. In just the 2 years since this review appeared, we’ve seen fresh evidence that the bulk of fragmentbased drug design can, indeed, be executed in silico. At BioLeap, which currently possesses the only industrial-scale rational CFBDD platform, we had 45 projects on 70 protein structures underway in 2011. In just the past 3 months, we have run no fewer than 12,000 fragment-based Monte Carlo simulations and thousands of quantum mechanical calculations to further refine the assessment of ligand binding when highly polar and/or charged interactions are involved.

A Path Strewn With Obstacles

As the CEO of BioLeap, my commitment to this approach owes much to the pioneering work of Frank Guarnieri and John Kulp. More than 10 years ago, these two scientists recognized that to be adequately predictive, any model for designing drugs from chemical fragments needed to be thermodynamically rigorous. Unfortunately, at the time, methods for computing the binding affinity of a drug to a target protein (free energy) involved trade-offs: the methods were either accurate, but too computationally intensive and hard to use, or they relied too heavily on approximation and ceased to be predictive.3 Guarnieri had developed a method that approached this issue very differently, called simulated annealing of chemical potential in grand canonical Monte Carlo simulations.4 This represented a significant advance in closing the gap between accuracy and practicality. Kulp and colleague Richard Bryan, meanwhile, developed a robust, chemist-friendly platform that would one day make CFBDD a practical methodology for drug design.

Even with these advances, however, the path forward has been strewn with obstacles. Historically, the field of computational chemistry developed in support of the screenand- correlate paradigm of drug discovery. Capabilities such as docking, virtual screening, cheminformatics, chemical database similarity searches, etc., helped streamline and facilitate the assessment of compounds from libraries and the synthetic expansion of screening hits. These capabilities were an operational improvement. And yet, they didn’t fundamentally provide information leading to new ideas or insights that would result in higher success rates and higher quality molecules in discovery programs. Thus, the initial enthusiasm for computational chemistry waned and disappointment spread as expectations for high impact were not met.

CFBDD diverges from traditional computational chemistry, mobilizing a different set of capabilities targeted at a different paradigm. While it is a potent source of new information about proteinligand binding interactions and chemical novelty, early CFBDD approaches from the 1990s suffered from an excessively high rate of false positives, limiting the technology’s utility. There were two basic issues: the reliability of the fragment binding predictions as a starting point, and the fragment linking problem. Specifically, even if you know the location and binding affinity of the component fragments, it’s difficult to tell whether ligands assembled by linking those fragments together will precisely follow suit. The fragment poses may lead you to a strong hypothesis about the location and binding affinity of the completed ligand. However, when fragments are linked, charges in the resulting ligand are often redistributed, and the binding pose can shift to accommodate a trade-off in the binding of each component. Further, the bonds impose restrictions on motion, which changes the entropy of the fragments.

The Challenges of Ranking Ligands

New algorithms are needed to solve the fragment-linking problem. For instance, in the computer model, one or more fragments could be constrained to move in certain ways, subject to forces binding them to other fragments or ligands. This would provide a characterization of the free energy that includes changes in configurational entropy caused by the limited range of motion under the bonded constraint. Working with an approach like this, BioLeap has found that the interconnected fragments can now explore somewhat different poses. Each fragment is annealed, and the sum of the lowest free energies of the components is used for ranking.

In addition to the fragment-linking problem above, we have identified several other challenges in ranking ligands. One is the change in solvation energy between the unbound and bound states (ΔGs). The industry now has access to new, highperformance, empirical solvation models, which when combined with fragment-linking algorithms, may improve the ligand ranking.5 One also needs to closely consider conformational stress upon ligand binding due to the difference in energy between the lowest energy unbound conformation and the bound conformation. Tools are needed that can quickly identify the global minimum conformation when the high number of rotatable bonds makes systemic exploration impractical. More extensive, and automated, use of quantum mechanics should be a priority in order to further refine the assessment of ligand binding when highly polar and/or charged interactions are involved. This is “the new frontier” of molecular interactions. By combining fragment-linking algorithms, salvation ΔGs, quantum mechanics, and conformational analysis, researchers will be able to make significant progress in ranking compounds over the sum-of-fragment free energies of previous methods. In a recent blinded computational challenge hosted by OpenEye Scientific, BioLeap demonstrated how incorporating these aspects in CFBDD significantly improves affinity ranking and pose predication, which could have a transformative impact on fragment-based drug discovery and drug discovery in general.6

It is clear that computational advances throughout the past decade have brought us much closer to accurately predicting binding affinity between, say, a ligand and an enzyme. Though more work is necessary, the CFBDD approach now seems to be the most effective way to design new molecules. Consider the limitations of other fragmentbased approaches, such as experiments that rely on x-ray crystallography or NMR. If you are working with a fragment molecule that isn’t water soluble, for example, it is difficult to use crystallography or NMR to test whether it binds to a protein in solution. Nor is it easy to pinpoint where the fragment binds with the highest affinity and orientation.

High-Affinity Binding & Low Molecular Weight

CFBDD solves this dilemma. Our models reveal not only the highest-affinity binding site on the protein, but every site where the fragment binds with any affinity at all. We can give the medicinal chemist something called a fragment map displaying how thousands of different fragments might bind to a protein target, and what the orientation and relative affinity is at each site. Chemists, armed with computational data, can now use an automated search to find high-affinity fragments in the right location and start linking up the fragments in any fashion they chose.

In the course of designing new drugs, digital chemists mull the problem from many angles. For drug selectivity, they’ll typically want a higher binding affinity to one protein and not others. The chemists will be aiming for a low molecular weight, and will want something that has reasonable solubility in water. These and other considerations affect the choice of fragments, and may lead the chemist to settle for a weaker binding affinity as a trade-off. By turning these matters into conscious choices, our technology eliminates hunt-and-peck chemistry. One can evaluate countless potential molecules, operating at minimal cost until a decision is made as to what molecules to synthesize. Out of 200 designed candidates, we might make 20.

The technology I’ve described has other virtues. With HTS, the molecules you cull are likely to be large – 400 molecular weight or more. By building up from small fragments, we design significantly smaller compounds with higher selectivity. The result may be a molecular weight of 225, or even lower. Due to this context-dependent, custom-made approach, the molecules are not only smaller, but have better pharmacological properties. A typical HTS-derived drug candidate with a molecular weight of 400 will have features that don’t serve your therapeutic needs, or might be toxic. You’ll have to cut out the bad bits, while adding in the good bits. Why not assemble the molecule you want from the start?


In the course of this short review, I don’t mean to imply there is only one way to harness computing in drug discovery. Different companies are exploring a variety of information-based strategies. Some biopharma companies search databases for molecules that closely resemble existing drugs with a proven track record. With this Google-like approach, there’s no need to examine the binding sites or contemplate basic biophysics. That’s very helpful if you have a target or pathway that is already well characterized. However, I submit that the approach doesn’t help when you have a new target for which there is no historic database (welcome to the new terrain in systems biology!).

Due to the revolution in this field, we are moving into an era in which there will be many new targets and clusters of targets at the juncture of different disease pathways. Understanding what goes on in an organism based on computer models remains a tall order. And that understanding only points us to places in the disease process where we ought to intervene – we still have to design the molecules that affect the enzymes, receptors, and other targets to cure the disease. With marvelously good luck, scientists have already discovered a few medicines, such as the cancer drug Gleevec, that act on multiple targets simultaneously. With fragment-based drug design, our mission is to create many more.


1. Kulp JL, 3rd, et al. Diverse fragment clustering and water exclusion identify protein hot spots. J Am Chem Soc. 2011; 28:10740-10743.
2. de Klo GE, et al. Transforming fragments into candidates: small becomes big in medicinal chemistry. Drug Discov Today. 2009;13-14:630- 646.
3. Guvench O, et al. Computational evaluation of protein–small molecule binding. Curr Opin Struct Biol. 2009;1:56-61.
4. Guarnieri F, et al. Simulated annealing of chemical potential: a general procedure for locating bound waters. application to the study of the differential hydration propensities of the major and minor grooves of DNA. J Am Chem Soc. 1996:35:8493-8494.
5. Cramer CJ, et al. A universal approach to solvation modeling. Acc Chem Res. 2008;6:760-768.
6. Kulp JL, 3rd, et al. A fragment-based approach to the SAMPL3 challenge. J Comput-Aided Mol Des. DOI: 10.1007/s10822-012-9546-1, in press.

David Pompliano, PhD
BioLeap, Inc.

Dr. Pomplianohas has 19 years of leadership experience in the pharmaceutical industry with a strong track record in joint venture management, international business development, and driving strategic change in research and development. Most recently, he served as Vice President, Worldwide Basic Head, Vaccines and Infectious Diseases Franchise for Merck Research Laboratories, where he created the global antimicrobial research strategy and led the research and business development teams that integrated internal discovery activities with external shared-risk venture partners. Prior to Merck, he was Vice President, Head of Biology, Microbial Musculoskeletal and Proliferative Diseases Center for Excellence in Drug Discovery at GlaxoSmithKline. As a drug hunter, Dr. Pompliano has contributed to the discovery or development of four marketed drugs (Tykerb, Promacta, Votrient, and Altabax) and four drugs that have reached clinical proof-of- concept. He has published more than 50 research papers and is co-inventor on three patents. Dr. Pompliano earned his PhD in Chemistry from Stanford University and was an NIH Postdoctoral Fellow at Harvard University.