Fragment docking: it’s all about ligand efficiency

A widespread belief holds that it is more difficult to computationally dock fragment-sized molecules than lead-sized or drug-sized molecules. But is this really true? And if so, why? These questions are tackled by Marcel Verdonk and colleagues at Astex in a recent paper in J. Med. Chem.

The researchers examined 11 targets for which they had multiple crystal structures of each with bound fragments (which contained up to 15 non-hydrogen atoms) and larger molecules (which contained at least 20 non-hydrogen atoms); these crystal structures were the “correct” structures against which computational models could be judged. A total of 106 fragments and 100 larger molecules were then docked against their target proteins using a variety of different methods.

Surprisingly, the overall results were not overly impressive (<70% correct depending on methodology – often much less). But even more surprisingly, there was no difference between the success rates of the fragments and that of the larger molecules. However, the reasons for the poor performance were different. In the case of fragments, the problem was often that the scoring function didn’t recognize the correct solution; the energetics were just too subtle. In the case of the larger molecules, though, the problem was more often one of sampling: the docking program failed to produce the conformation of protein or ligand that corresponded to the correct solution, so it had no opportunity to score it. Potency made no difference: high-affinity compounds fared just as poorly as lower affinity compounds. What did make a difference, though, was ligand efficiency: compounds with high ligand-efficiency (> 0.4 kcal/mol/atom) were docked with considerably greater success than those with lower ligand efficiencies. As the authors point out, this makes sense intuitively:

High LE compounds form high-quality interactions with the target, which should make it easier for a docking program (both from a scoring and search perspective) to dock these compounds correctly.

So the next time you see a computational model of a protein-ligand complex, you might want to take a closer look at ligand efficiency to get a sense of how trustworthy the structure might be.

Journal of Computer-Aided Molecular Design 2011 Special FBDD Issue

The most recent issue of J. Comput. Aided Mol. Des. is entirely devoted to fragment-based drug discovery. This is the second special issue they’ve dedicated to this topic, the first one being in 2009.

Associate Editor Wendy Warr starts by interviewing Sandy Farmer of Boehringer Ingelheim. There are many insights and tips here, and I strongly recommend it for a view of how fragment-based approaches are practiced at one large company. A few quotes give a sense of the flavor.

On corporate environment:

In most cases, the difference between success and failure has little to do with the process and supporting technologies (they work!), but rather much more to do with the organizational structure to support FBDD and the organizational mindset to accept the different risk profile and resource model behind FBDD.

On success rates:

We have found that FBDD has truly failed in only 2-3 targets out of over a dozen or so.

On cost:

FBDD must be viewed as an investment opportunity, not a manufacturing process. And the business decisions surrounding FBDD should factor that in. FBDD is more about the opportunity cost (of not doing it) than the “run” cost (of doing it).

On expertise:

Successful FBDD still requires a strong gut feeling.

On small companies:

In the end, FBDD will always have a lower barrier to entry than HTS for a small company wanting to get into the drug-discovery space.

The key to success for such companies is to identify or construct some technology platform.

There’s a lot of other really great content in the issue, much of which has been covered in previous posts on fragment library design, biolayer interferometry, LLEAT, and companies doing FBLD. The other articles are described briefly below.

Jean-Louis Reymond and colleagues have two articles for mining chemical structures, one analyzing their enumerated set of all compounds having up to 13 heavy atoms (GDB-13), the other focused on visualizing chemical space covered by molecules in PubChem. They have also put up a free web-based search tool (available here) for mining these databases.

Roland Bürli and colleagues at BioFocus describe their fragment library and its application to discover fragment hits against the kinase p38alpha. A range of techniques are used, with reasonably good correlation between them.

Finally, M. Catherine Johnson and colleagues present work they did at Pfizer on the anticancer target PDK1 (see here and here for other fragment-based approaches to this kinase). NMR screening provided a number of different fragment hits that were used to mine the corporate compound collection for more potent analogs, and crystallography-guided parallel chemistry ultimately led to low micromolar inhibitors.

Designing fragment libraries

The topic of fragment library design is similar to foundation construction: most people don’t give it much thought, but any organization that doesn’t take it seriously could quickly find itself on shaky ground. Three recent papers cover different aspects of this topic.

The first paper, published in J. Comput. Aided Mol. Des. by researchers at Pfizer, describes the design and construction of their Global Fragment Initiative (GFI), a 2,885 fragment library meant to be broadly applicable to any target using any screening method (NMR, X-ray, SPR, MS, and biochemical assays). Most of these fragments came from commercial or in-house collections, but 293 were synthesized specifically for the library. All compounds were filtered to remove reactive or otherwise undesirable moieties. Interestingly, a large number of cationic and anionic molecules were included, based on the observation that many approved drugs are charged. Also, roughly a quarter of the compounds contained at least one chiral center.

Potential library members were put through a rather more rigorous selection than the standard Rule of 3 (for example, cLogP < 2.0). Molecular complexity was explicitly considered, and overly complex fragments were excluded. Fragments were also analyzed by 2D and 3D similarity and chosen to maximize diversity, though with the criterion that close analogs were available either in-house or commercially. Compounds were also chosen to allow rapid chemical elaboration. Finally, compounds were evaluated by NMR for purity and solubility at 1 mM in aqueous buffer and 50-100 mM in DMSO.

The bulk of the library (excluding custom-synthesized fragments) has been screened against at least 13 targets in 8 different protein families, mostly by NMR and biochemical assays, resulting in hit rates between 2.8 – 13%. Only one fragment hit all 13 targets, while 766 hit only one; in total 33% of the fragments hit one or more of the targets, a fraction eerily similar to that seen at Genentech and Vernalis. Overall this is a thorough, information-dense paper, and well worth reading if you are considering building or expanding a fragment library.

One of the most productive first steps you can take after identifying a fragment hit is to test close analogs or larger molecules that contain the fragment. Of course, it is easier to buy compounds than to make them, so a fragment library that effectively samples commercial compounds is likely to be useful. This “SAR by catalog” approach is the topic of the second paper, also in J. Comput. Aided Mol. Des., from Rod Hubbard and colleagues at Vernalis and the University of York.

The researchers analyzed catalogs of available compounds from each of three vendors (Asinex, Maybridge, and Specs). Filtering out undesirable functionalities and binning the molecules by size left 5600-7700 fragment-sized molecules and 28,600-252,000 larger molecules per vendor. Compound properties of the fragment sets (MW, polar surface area, number of hydrogen bond donors and acceptors, etc.) are summarized for each of the vendors, similarly to Chris Swain’s analysis. Six different algorithms were then tested to find sets of 200 fragments that would best represent the entire collection. In accordance with Murphy’s Law, the most complicated algorithm proved to be the most effective; it involves an iterative selection procedure with precisely defined similarity criteria. Still, this algorithm is not too difficult to implement, and it should prove a useful tool for selecting fragments from larger sets of commercial or in-house compounds.

Finally, a chapter by James Na and Qiyue Hu at Pfizer in a recent volume of Methods in Molecular Biology gives a broad overview of fragment library design. In addition to general considerations, the paper succinctly summarizes the design of the Pfizer Global Fragment Initiative as well as an earlier fragment library designed specifically for NMR screening. A more lengthy but instructive description of several Vernalis fragment libraries is also provided, as are some of the screening results. Finally, a nice table summarizes fragment libraries from more than a dozen companies.

First fragment-based drug approved

Today marks history with the first FDA approval of a drug to come out of fragment-based screening. The drug is branded as Zelboraf (vemurafenib), but readers of this blog are probably more familiar with its previous name of PLX4032. Although widely expected to be approved, the FDA acted more than two months ahead of schedule. The drug targets a mutant form of BRAF and has received widespread media coverage because of dramatic clinical results showing that it extends life for patients with a particularly deadly form of skin cancer. FiercePharma has an article with links to several others.

The drug was discovered at Plexxikon and developed in partnership with Roche; Plexxikon was acquired earlier this year by Daiichi Sankyo. The PLX4032 story is a case study in how rapidly fragments can enable a program: initiated in Februrary 2005, it took just six years to reach approval. It’s also an example of starting with a profoundly unselective fragment and winding up with a very selective drug (see here for early discovery and here for characterization of PLX4032).

Although I claim no prescience, I did state back in 2008 that it would be nice if a fragment-based drug would be approved by 2011. But more importantly, it is worth pausing to remember that this is a victory not just for the field of fragment-based drug discovery, but for those patients afflicted with metastatic melanoma. In the end, that’s what this is all about.

Fragment selectivity

A constant debate in fragment-based lead discovery is whether to focus on fragments that are selective for the target of interest. Because fragments have lower complexity than larger molecules they are likely to be less specific – that is, after all, one of the main arguments for why a small set of fragments can explore more chemical space than a much larger set of lead-like molecules. But does it make sense to prioritize those fragments that are more selective? In a recent issue of J. Med. Chem. Paul Bamborough and colleagues at GlaxoSmithKline address this question experimentally.

The broad family of kinases was chosen for the investigation. Protein kinases in particular have been a rich field for drug development, including fragment-based methods. The researchers assembled a library of 1065 commercially available fragments, most of which were designed to bind to the so-called “hinge” region of protein kinases where the substrate ATP binds. Of these fragments, 936 passed quality-control and maintained stability over the course of the year-plus study.

The researchers screened these fragments at 0.4 or 0.667 mM against a panel of 30 kinases using several different assay formats: FP (fluorescence polarization), IMAP (immobilization metal affinity phosphorylation), LEADseeker (a scintillation proximity assay), and TR-FRET (time-resolved fluorescence resonance energy transfer). Various experiments suggested that FP was most susceptible to assay artifacts, though the results were still usable.

17 of the fragments screened were chosen based on common fragment motifs in the literature. One example is adenine, a fragment of ATP, which of course is used by all kinases. Despite this universality, adenine actually showed surprising specificity, inhibiting some kinases strongly and not inhibiting others at all. The same goes for other hinge-binding fragments that we’ve seen before (such as indazole). On the other hand, biaryl urea fragments designed to bind to the less-conserved adaptive pocket of kinases were quite selective, hitting just 2 kinases strongly.

Of course, especially for kinases, the trick is not getting fragment hits but in figuring out which ones to pursue. Ligand efficiency is often used to prioritize fragments, but is this necessarily a good idea? The researchers compared published high-affinity inhibitors of several kinases with fragments contained within these inhibitors and found that the fragments often would not have stood out above the pack when compared solely on the basis of ligand efficiency. Even spookier, many of the most ligand-efficient fragments appear to be assay artifacts.

What about selectivity? Are non-selective fragments bound to become non-selective leads? The authors present one example of a rather non-selective fragment that could be optimized to a highly selective molecule; PLX4032, which started life as a promiscuous azaindole, is another example.

These are just anecdotes though, so to get a broader handle on this question the authors examined a set of 577 lead-like compounds that had been screened against 203 kinases. This led to a list of 592 matched pairs of lead-like compounds and fragment substructures (most of which are likely hinge-binders) which could be analyzed for selectivity. The results recapitulate a smaller, earlier study performed with a very different data set. As Bamborough et al. put it:

It is not uncommon to find selective lead-sized compounds based upon unselective fragments. Equally, unselective leadlike compounds are frequently based upon selective fragments. It seems that the property of selectivity need not be maintained between fragments and their related lead-sized molecules.

On one level, Bamborough’s study is a bit discouraging: fragment selectivity should be used cautiously if at all in prioritizing fragments. Even ligand efficiency should not be gating; last year we discussed how a fragment with relatively modest ligand efficiency was transformed into the clinical-stage (and more ligand efficient) Hsp90 inhibitor AT13387. Other factors, such as structural novelty or how amenable a fragment will be to further elaboration, are just as if not more important for choosing fragments. All of which serves to reemphasize the fact that drug discovery is less a series of hard and fast rules than a loose system of guidelines and hunches. This lack of predictability is part of what makes the process so frustrating – and fun.

Fragment-based events in 2011 and 2012

If you missed the fragment events earlier this year there is still one late addition to the calendar as well as some webinars. And it’s not too soon to be thinking about 2012!

August 16: Is your travel budget limited? Emerald Biosciences is putting together a series of free webinars related to FBLD on August 16, September 20, October 18, and November 15.

October 21: Zenobia Therapeutics is putting together a FBLD conference in San Diego. Although just one day, there is a nice lineup of speakers, so try to make it if you can.

2012

March 19-23: Keystone Symposium: Addressing the Challenges of Drug Discovery – Novel Targets, New Chemical Space and Emerging Approaches will be held in Tahoe City, CA. Although not exclusively devoted to fragments, there are many speakers I look forward to hearing.

April 17-19: Cambridge Healthtech Institute’s Seventh Annual Fragment-Based Drug Discovery will be held in San Diego. You can read impressions of this past year’s meeting here and 2010’s here.

September 23-26: FBLD 2012, the fourth in an illustrious series of conferences, will be held in my fair city of San Francisco. This should be a biggy – the first such event in the Bay Area (and the weather in September is usually decent too). You can read impressions of FBLD 2010 and FBLD 2009.

Know of anything else? Organizing a fragment event? Let us know and we’ll get the word out.

Ligand efficiency metrics poll results

Poll results are in, and not surprisingly, ligand efficiency (LE) comes out on top, with 86% of respondents using the metric. What was a surprise to me is how many folks use ligand lipophilic efficiency (LLE) (46%). Coming in a distant third at 15% is LLEAT, but given that this metric was just reported it has a pretty strong showing, and I wouldn't be surprised to see this increase. Binding efficiency index (BEI) comes in fourth with 12% of the vote, and Fsp3 is tied with "other" with 8% of the vote. The other metrics only received one or two votes each.
Since people could vote on multiple metrics, there were more responses than respondents. Subtracting those who voted for "none" leaves 124 data points, suggesting that the average researcher is using 1.9 of these metrics (though unfortunately we don't have information on the median user).

Finally, for the 5 of you who selected "other", what else is out there that we've left out?