Benchmarking native mass spectrometry

Mass spectrometry (MS) is one of the less common tools to find fragments. In the conceptually simplest approach (native mass spectrometry), you incubate your protein with a putative ligand and ionize the mixture. Fragment binding is detected by an increased mass for the complex, and the strength of binding by the ratio of heavier bound complex peak to protein peak. However, the liquid to gas phase transition is a big step, and often the complex does not survive. Aside from more specialized applications of MS (such as here, here, and here) there aren’t many published examples. A recent paper from Federico Sirtori and colleagues at Nerviano and Università degli Studi di Milano in Eur. J. Pharm. Sci. describes fragment screening by native MS in detail.

The researchers used the reliable model protein Hsp90, which was also used in a previous MS study and in benchmarking other techniques. One of the many benefits of Hsp90 is a wealth of well-characterized inhibitors with a range of affinities, and these were used to calibrate the technique. This turned out to be critical: beyond sample preparation itself (beware non-volatile buffer components), all kinds of parameters can be adjusted including various voltages, temperatures, vacuum strength, and ion source. Get one of these wrong and your non-covalent complex either fails to ionize or blows apart.

In addition to using published data on known compounds, the researchers ran both fluorescence polarization (FP) and surface plasmon resonance (SPR) assays to independently determine dissociation constants. Initially the results from MS (a Q-TOF) were quite different, but after optimization the team was ultimately able to find conditions that gave qualitatively as well as quantitatively similar results for ligands with affinities ranging from picomolar to ~100 micromolar.

Thus encouraged, the team embarked on a fragment screening campaign. The Nerviano fragment library consists of 1914 molecules mostly following the rule of 3, though halogenated fragments up to 380 Da are allowed as are compounds with up to 6 hydrogen bond acceptors. The fragments were run in mixtures of 5, with protein at 2.5 µM and each compound at the low concentration of 10 µM. Sample injection and data processing were automated, and the entire screen took 2 days and 2 mg of protein.

Given the low concentration of fragments, the researchers lowered the bar for potential hits, yielding 282 compounds. These were retested individually, yielding 146 confirmed hits that gave signals of 5.2-29.7% bound protein. This is a high hit-rate, particularly given that these binding levels suggest affinities in the 20-179 µM range. Indeed, only 5 fragments could be competed by a high-affinity binder, suggesting either that the others bind outside the active site or are non-specific (false positives). Regarding false negatives, Nerviano reported the results of an NMR fragment screen against Hsp90 last year, and 12 of 14 hits identified there could also be detected by MS. The other two were likely below the detection limit of the MS assay.

Unfortunately, the researchers do not discuss thermodynamics. In theory enthalpic interactions dominate over entropic interactions in the gas phase, but it is unclear whether any of the observed binders were strongly entropy-driven.

In the end, it appears that fragment screening by native MS is workable, but the sensitivity is probably lower than other techniques. Of course, increasing the ligand concentration would increase the sensitivity to weaker binders, but at the cost of more non-specific binding – which is already considerable. Also, Hsp90 is about the friendliest protein one can imagine. I would be reluctant to try this with a more challenging target that lacks good tool ligands. But if you want to give it a go, this paper provides a wealth of information for getting started. And if you have experience with native MS, please share it in the comments.

Fragments vs HSP90: Nerviano’s turn

HSP90, an oncology target, is one of those proteins that seems tailor-made for fragments: it has an active site with a predilection for small molecules, it’s easy to work with, and it crystallizes readily. Indeed, at least two fragment-derived molecules targeting this protein have advanced to Phase 2 clinical trials. In a recent Bioorg. Med. Chem. paper, Elena Casale, Francesco Casuscelli, and colleagues at Nerviano describe their efforts against this target.

The researchers started by identifying a fluorinated probe molecule that they could use in a Fluorine chemical shift Anisotropy and eXchange for Screening (FAXS) assay. This is an NMR-based competition method, in which fragments are screened to find those that displace a known ligand, in this case one that binds in the active site. A total of 1200 fragments were screened in pools of 10, each at the relatively low concentration of 50 micromolar. Nonetheless, 23 hits were found, four of which were characterized crystallographically bound to the protein.

Fragment 3 was among the more interesting, both because of its high ligand efficiency as well as its structural novelty. SAR-by-catalog failed to find anything better from 20 compounds tested, and initial fragment growing also proved disappointing. However, a closer inspection of the crystal structure (cyan) revealed the possibility of linking the fragment to the well-known HSP90 fragment resorcinol. This led to compound 8b, which binds about 5-fold more tightly. Crystallography revealed that the molecule (magenta) also binds as expected.

However, the team wisely chose to test synthetic intermediate 7h (in which the hydroxyl groups were still methylated) and this turned out to be even more active than the designed compound. Since the hydroxyls of the resorcinol are essential for binding in other lead series, the team solved the crystal structure of compound 7h (green) and was surprised to find that it binds in a completely different manner than compound 8b; the ligand essentially flips over.

This discovery led to a change in direction for medicinal chemistry, leading ultimately to the low nanomolar compound 12a. Unfortunately this molecule had only modest cell-based activity and was metabolically unstable.

This is a solid, nuts-and-bolts sort of story. Although it does not conclude with a clinical candidate, it does provide a useful window into how fragment-based methods are applied in industry. It is also a reminder to screen all your intermediates and to remember that even subtle changes to a molecule may have dramatic effects on its binding mode. Those surprising shifts can point the way to promising chemical space.

Capillary Electrophoresis

One of the fun aspects of fragment-based lead discovery is the number of ingenious biophysical methods for finding low-affinity fragments. In a recent issue of J. Biomol. Screen., Carol Austin and colleagues at Selcia describe their approach, capillary electrophoresis, which they term CEfrag.

Capillary electrophoresis itself has been around for quite a while. It involves applying a high voltage across a thin capillary filled with liquid; a solution to be analyzed is injected, and the voltage causes migration of analytes (for example, proteins or small molecules). Analyte movement through the capillary depends on charge and “hydrodynamic radius,” which is a function of molecular size and shape. In the case of CEfrag, the idea is to start with a reporter molecule that can be readily detected, for example via UV absorbance. Under a standard set of conditions, this “probe ligand” will have a characteristic mobility. If an excess of protein that binds to the probe ligand is present in the running buffer, the migration time will shift. If an inhibitor is also present in the running buffer, this will prevent the probe ligand from binding to the protein, also causing the migration time to change. By running different concentrations of inhibitor and measuring the changes in mobility, the inhibition constant can be determined.

The researchers demonstrated their approach using that old work-horse of FBLD, the cancer target Hsp90. The known Hsp90 inhibitor radicicol was used as the probe ligand. A total of 609 fragments were screened individually at an initial concentration of 0.5 mM, yielding 42 fragments that reproducibly inhibited radicicol mobility by 20% or more. This ~7% hit rate is similar to that found by others for this target.

Only 12 of the 42 hits identified by CEfrag were also detected in a confirmatory fluorescence polarization (FP) assay, of which only 5 gave measurable IC₅₀ values. However, FP is not ideal for evaluating fragments. In fact, one of the CE hits that didn’t reproduce by FP was ethamivan, the starting fragment for the program that ultimately led to Astex’s AT13387, now in a phase 2 clinical trial for GIST.

To get a better sense of the quality of the CE hits, the researchers put 6 fragments into crystallography trials: 3 hits from both CE and FP, two from CE alone, and one that hit neither. The negative control didn’t produce a structure, whereas two of the FP-confirmed hits produced co-crystal structures (the one that did not had solubility issues). One of the two CE-only hits (ethamivan) also did.

With a throughput of 100 compounds per day per instrument, this is not a high-throughput method, but it is comparable to many other biophysical approaches. Also, the low protein consumption and ability to use unmodified protein are selling points. Have you tried CEfrag? If so, what do you think?

Experiences in fragment-based drug discovery

This is the title of a new review published in Trends in Pharmacological Sciences by Christopher Murray, Marcel Verdonk, and David Rees of Astex Pharmaceuticals. Although there is certainly no shortage of reviews on fragment-based lead discovery (a situation to which I have admittedly contributed), this one is notable both for its clarity and for being able to draw upon a deep wealth of institutional knowledge.

The review starts by discussing three notable case studies: Plexxikon’s discovery of the mutant B-Raf inhibitor vemurafenib, Astex’s Hsp90 program, and Merck’s BACE program.

Next, the authors describe some key concepts and challenges of FBLD.

Concept 1: Inappropriate physical properties are a major cause of attrition for small-molecule drugs

This should not come as a surprise to readers of this blog; the discovery of compounds with superior properties is one of the key selling points for FBLD. In support, the researchers compare 39 leads against 20 targets from Astex’s fragment-based programs with 335 published HTS-derived leads and 592 oral drugs. The FBLD-derived leads are on average 62 Da smaller and 1 log unit less lipophilic than are the HTS leads, and are much more similar to the oral drugs.

Concept 2: Although weak in potency, fragments actually form high-quality interactions

The position and the orientation of fragments tend to be conserved during the course of optimization (though see here for a notable exception). Of the 39 internal fragment-to-lead programs, roughly 80% of the atoms in the original fragment (which averaged 13 atoms total) were retained in the lead. Moreover, the mean shift in position as judged crystallographically was only 0.79 Å.

Concept 3: LE can be used to judge the relative optimisability of differently sized molecules

I like to think of fragments as ants: small and weak when considered from a human perspective, but impressively strong when considered for their size. Ligand efficiency and its many permutations are tools to assess molecules in a size-appropriate manner.

Concept 4: Relatively small libraries of fragments are required to sample chemical space

There is plenty of theory to support this (see for example here and here). The authors note that a library of 1000 compounds with 12 or fewer heavy atoms would sample ~0.001% of possible molecules with MW < 170 Da, while 1000 compounds with 25 or fewer heavy atoms would sample only 10^-14 percent of the possible larger molecules. But while theory is fine, the real proof is in the number of molecules that have entered the clinic that can trace their origins to small fragment libraries.

Of course, FBLD does have challenges.

Challenge 1: Specialized methods are needed to detect fragment binding

You don’t hunt ants with an elephant gun, and you’ll have a hard time finding fragments using standard procedures. The need for specialized methods was once a major impediment to FBLD, but happily today there are many options, and using two or more of these in combination is the best strategy.

Challenge 2: Efficient optimisation of fragment hits is required

In other words: you’ve found a fragment, now what? Structural biology is extremely helpful to figure out how the fragment binds and suggest what to do next, especially since proteins can be surprisingly flexible: of crystal structures from 25 fragment screens at Astex, 12 proteins showed movement of at least 5.0 Å upon fragment binding.

Of course, it takes more than a crystal structure to advance a fragment, and the challenges can be institutional as much as scientific. But given the proven success of the technique, these are challenges worth facing.

Finally, it’s worth checking out the entire issue of Trends in Pharmacological Sciences, which is devoted to structure-based drug design. There are some nice papers by Zhaoning Zhu on BACE, Tom Blundell and colleagues on protein-protein interactions, Stephen Wasserman and colleagues on high-throughput crystallography, and lots more.

Fragment-based drug discovery and X-ray crystallography

I’m holding in my hands a book of this title, edited by Thomas Davies and Marko Hyvönen and published this year as part of Springer’s Topics in Current Chemistry series. I believe this is the fourth book entirely devoted to fragment-based drug discovery, which shows both the vitality and rapid development of the field.

The book starts with an introduction to fragment-based drug discovery by me. If you’re new to the field, this chapter should serve as a self-contained summary.

In the next chapter Thomas Davies and Ian Tickle describe how FBDD is practiced at Astex, paying particular attention to the use of X-ray crystallography. Notably, researchers from this company “do not consider a fragment hit to be ‘validated’ and suitable as a starting point for medicinal chemistry until it has been observed to bind by crystallography.” This chapter also contains a nice analysis of fragment library design and a couple case studies, including the discovery of the clinical-stage CDK2 inhibitor AT7519.

Rod Hubbard and colleagues at Vernalis and the University of York next describe their efforts to discover Hsp90 inhibitors using a combination of virtual and fragment screening. We’ve covered some of this before (here and here), but it’s nice to see the full story.

The next chapter also focuses heavily on a single target: Daniel Wyss and colleagues at Merck describe their success in discovering BACE inhibitors. This chapter also includes an excellent review of NMR methods for finding fragments.

Michael Hennig and colleagues at Roche (Basel) contrast the various biophysical methods used to discover fragments, with a heavy emphasis on SPR. Crystallography is also covered, in particular co-crystallization of fragments with protein. Co-crystallization is more time-consuming than soaking fragments into preformed crystals, so compound prioritization techniques such as SPR are especially useful.

One of the most promising applications of fragment-based methods is tackling tough targets such as protein-protein interactions, the subject of a chapter by Marko Hyvönen and colleagues at the University of Cambridge. The chapter contains a nice discussion of energetics and hot spots as well as a detailed analysis of methods to find fragments which complements some of the other chapters.

Eddy Arnold and colleagues at Rutgers discuss the use of crystallographic fragment screening against two HIV-1 targets, HIV protease and HIV reverse transcriptase (RT). We’ve previously discussed the former here. In the case of RT, fragments were soaked into crystals in the presence of a high affinity inhibitor, effectively blocking its binding site from fragments. More than 30 fragments were identified binding to multiple other sites on the protein – one fragment bound at 11 distinct sites! Interestingly, the fragments were enriched for halogen-containing molecules. Several also had functional activity with respectable ligand efficiencies. The authors also discuss other published fragment work on HIV RT.

Finally, Didier Rognan at the University of Strasbourg discusses computational approaches to library design, binding site determination, and predicting druggability. Fragment docking is extensively covered, along with a discussion of what factors contribute to success. It seems that docking is particularly good at identifying negatively charged, relatively buried fragments that make similar hydrogen bonds as the substrate. De novo ligand design, both the successes and challenges, is also covered.

Like last year’s book, all the chapters in this one are published online, but it is worth getting a bound copy as it is nicely put together, with color figures liberally integrated throughout rather than banished to plates at the back.

How effectively can fragments sample chemical space?

One of the key advantages of fragment-based drug discovery is that, since there are fewer fragments than lead-sized or drug-sized molecules, it is possible to sample chemical space far more efficiently with fragments than with larger molecules. At least, that’s the theory, but does is it hold true in the real world?

To put it another way, do fragments sample all of the space in which drugs are found? And what kinds of fragments are best for this sampling? In the most recent issue of J. Med. Chem., Stephen Roughley and Rod Hubbard of Vernalis address such questions.

The system they investigate, heat shock protein 90 (Hsp90), is an ideal model system: it is both a popular anti-cancer target as well as structurally tractable, and is thus arguably the most heavily explored single target in terms of fragment-based lead discovery. At least 8 antagonists have entered the clinic, of which at least 2 have come from fragments (see the posts on AT13387, NVP-BEP800/VER-82576, and posts on Evotec compounds discovered by fragment growing or linking.)

Vernalis has had a long-running fragment-based program targeting Hsp90, which has resulted in numerous fragments whose binding modes have been determined by X-ray crystallography. Roughley and Hubbard analyzed these fragments and compared them to published inhibitors. Just 5 distinct fragments can be mapped onto all of the clinical compounds: a handful of fragments effectively samples relevant chemical space. As the authors put it:

For Hsp90 at least, the fragments do cover an appropriate chemical space; what is then important is the imagination of the chemist in evolving the fragments into potent inhibitors.

The second point – about the imagination of the chemist – is critical. Mapping fragments onto elaborated molecules is easier to do retrospectively than prospectively; a cynic could argue that methane is a fragment of just about any drug out there. However, Roughly and Hubbard also point out that, particularly in cases such as this where there are many co-crystal structures, fragments can help identify bioisosteres, including cryptic ones that would not be obvious purely from studying functional SAR.

The paper also addresses the issue of optimal library design, in particular the dilemma of size. Although all five representative fragments were found in an initial set of just 719 fragments, subtle changes can dramatically change the binding mode, an issue we’ve touched on previously. It may not be practical to have multiple similar fragments present in a primary screening library, but testing close analogs after identifying initial fragment hits is likely to be worthwhile.

Finally, one of the concerns about fragment-based approaches is that, if everyone is buying the same set of fragments from the same suppliers and screening them against the same targets, they will end up in the same place – and stumbling over each others’ intellectual property. Reassuringly, this turns out not to be the case:

Even though the various companies discovered rather similar compounds from a fragment screen, exploiting similar binding motifs, there were no exact matches. [Also], the subsequent evolution of the fragments sometimes took very different paths and produced mostly very different chemical leads and candidates.

If this holds true for such heavily mined targets as Hsp90 (and kinases, as discussed previously) it should be even more true for newer classes of targets.

There is a wealth of information in this paper, and it is worth perusing, especially if you find yourself longing for some science over the long holiday weekend.

Hsp90 and fragment linking

There has been no shortage of fragment-based approaches directed toward the anti-cancer target Hsp90, most of which have relied on growing fragments (see here for some impressive recent examples). Researchers from Abbott published a report providing a couple examples of linking fragments against this target a few years back, but in those cases the ligand efficiencies of the linked molecules were dramatically lower than those of the initial fragments. In a recent paper in ChemMedChem, researchers from Evotec describe an example that maintains the ligand efficiency.

The research group had previously conducted a fragment screen against Hsp90, resulting in a number of hits. In the new paper, fragment hits 1 and 2 (see figure) were both found to have fairly low affinities, but were characterized crystallographically. Interestingly, fragment 2 could adopt at least two very different conformations, depending on whether it was co-crystallized in the presence of fragment 1. In the ternary structure, the two fragments come within about 3 Å of each other, and molecular modeling suggested that four atoms should be able to link them.

Gratifyingly, when such a compound was made and tested, it inhibited the enzyme several hundred-fold more tightly than either of the initial fragments. The crystal structure revealed that the compound binds similarly to the ternary structure of Hsp90 and fragments 1 and 2.

The authors note that “the binding free energy of the linked fragment 3c was found to be exactly the sum of those of the original two fragments.” Of course, this is still a long way from an ideal linking situation: as noted earlier this year a good linker should lead to super-additivity (an improvement of ligand efficiency), not just additivity (maintenance of ligand efficiency). Nonetheless, this example is still better than many attempts at linking, which often are less than additive.

Fragments in the Clinic: AT13387

We recently discussed BACE, a target that has been tackled by FBDD due to its intractability to other methods. The subject of this post is quite the opposite: the anticancer target Hsp90 has proven very amenable to a variety of approaches, including fragment methods (see here and here); close to a dozen compounds targeting Hsp90 are in the clinic. Now Astex has detailed their work in this area with two back-to-back papers in a recent issue of J. Med. Chem. describing the discovery of AT13387.

The first paper, by Christopher Murray and colleagues, actually presents the discovery of two separate series of inhibitors. The researchers started with a library of about 1600 fragments and used NMR techniques (water LOGSY) to identify hits against Hsp90. Competition with ADP allowed them to identify molecules that bind to the nucleotide binding site. In all, 125 fragments were taken into crystallography, using both co-crystallography and soaking, resulting in 26 co-crystal structures. Four of these structures are described in some detail, with two leading to potent inhibitors. Throughout the process, isothermal titration calorimetry was used to measure dissociation constants.

In the first series, compound 1 was identified as a weak hit (see Figure 1). Virtual screening led to the purchase of a few variants, including compound 5, with roughly 100-fold improved affinity. Interestingly, the crystal structure of compound 1 bound to Hsp90 showed that the molecule was twisted around the bond connecting the two aromatic rings, despite this not being energetically optimal for the unbound molecule. By substituting the phenyl ring of compound 5 to stabilize this twisted conformation the researchers were able to improve the potency another 20-fold (compound 9), along with a boost in ligand efficiency. Further structural work suggested adding another chlorine to fill a lipophilic site as well as adding a solubilizing group, ultimately leading to compound 14, with low nanomolar binding affinity and low micromolar cell activity.

Figure 1

In the second series, compound 3 (which is actually itself a drug, ethamivan) had only modest ligand efficiency, but crystallography suggested that replacing the methoxy group with something slightly larger and more lipophilic would improve the interactions, a hypothesis borne out by the increased activity of compound 17 (see Figure 2). Increasing the lipophilicity of the amide side chain to take advantage of protein flexibility led to a further two orders of magnitude increase in potency (compound 28). Finally, the researchers were able to use the known binding mode of a natural product to add an additional hydroxyl group, leading to compound 31, with sub-nanomolar affinity (more than a million-fold more potent than the initial fragment!) and mid-nanomolar cell activity.

Figure 2

An impressive feature of both these examples is that, through the use of elegant medicinal chemistry, the researchers were able to improve ligand efficiency throughout the course of affinity improvement. Of course, it helps that they were working on a crystallographically friendly target for which several other groups had published extensive SAR, but these are nonetheless beautiful case studies. As the researchers point out, “in terms of the efficiency of the added groups, the two fragment to lead campaigns… are among the most efficient ever reported.”

But the story doesn’t end there. The second paper, by Andrew Woodhead and colleagues, describes the further optimization of compound 31 to the clinical candidate AT13387. Despite its impressive biochemical and cell potency, compound 31 had only modest activity in a mouse xenograft model, as well as a short plasma half-life. Not surprisingly the hydroxyl groups were found to be points of metabolism, but initial efforts at capping these or changing their electronics either proved detrimental to activity or did not improve the pharmacokinetics. This led to a medicinal chemistry focus on the isoindoline portion of the molecule: a number of positively charged moieties were added at various positions to try to change the overall properties of the molecule. Several substituents were tolerated, and seven related molecules were taken into preclinical candidate selection to look for optimal in vivo properties, solubility, and selectivity against P450 and hERG. AT13387 (see Figure 2) was chosen as the molecule having the best overall profile and entered human clinical trials for solid tumors.

This second paper is a valuable companion to the first: it is particularly notable that, on the simple measures of biochemical and cell potency, AT13387 is no better than compound 31. This emphasizes yet again that affinity is only the first step in drug discovery – it’s a long road from a good lead to the clinic, and an even longer road from there to a marketed drug.

Hsp90 and fragments

Some targets seem particularly amenable to fragment-based approaches. Protein kinases are one example. Another is the N-terminal ATP binding domain of Hsp90, a widely pursued anti-cancer target: at least two fragment-derived compounds against this target are currently in the clinic. At FBLD 2008 in San Diego last year, so many talks discussed this protein that it became a running gag (one speaker promised at the outset not to talk about it, then slipped in a few slides). A recent paper in ChemMedChem provides a particularly clear example of a multidisciplinary fragment-growing approach against this target.

The researchers, mostly from Evotec, started with a high-concentration biochemical displacement assay to screen 20,000 fragments against Hsp90. A relatively potent aminopyrimidine (compound 1, below) was characterized crystallographically, and this structure was then used to run a virtual screen of 3.8 million commercially available molecules using the program GOLD 3.0.1. Some of the resulting hits were purchased and tested, including compound 3, which showed sub-micromolar biochemical activity but no cell-based activity. Subsequent modeling and medicinal chemistry led to compound 19, which, in addition to mid-nanomolar biochemical activity, also displayed submicromolar cell activity in A549 and HCT116 cancer cell lines.


In addition to compound 1, a number of other fragments containing the aminopyrimidine substructure were also identified as hits. This moiety seems to be a privileged pharmacophore for Hsp90: for a fun read, check out this 2007 paper from researchers at Abbott, in which fragments are linked together in a couple different ways as well as grown. As in the more recent paper, the protein displays a remarkable degree of flexibility to accommodate small molecule binders.