Fragments vs soluble epoxide hydrolase – hundreds of them!

Soluble epoxide hydrolase (sEH) is a potential target for cardiovascular and immune disorders. The enzyme catalyzes the hydrolysis of long chain, lipophilic epoxyeicosatrienoic acids. These bind in a largely hydrophobic “L-shaped” pocket, with the catalytic machinery at the point of the L. Although it is relatively easy to find potent inhibitors of this enzyme, these tend to be greasy, insoluble, and non-druglike. In a recent paper in Bioorg. Med. Chem. Yasushi Amano and colleagues at Astellas describe a fragment-based approach to find better leads.

The researchers started with a high-concentration (up to 2 mM) enzymatic inhibition assay of 4200 fragments, resulting in 307 hits with IC₅₀ values between 700 nM and 1.7 mM. All of these were taken into co-crystallography trials, yielding crystals for about half of them. The other fragments were soaked into apo-crystals of sEH (that is, crystals without bound ligand) to get as many structures as possible. All together 126 crystal structures of fragments bound to sEH were solved, which is all the more impressive considering that there are only three authors on the paper!

Most of the fragments (83) bound at the catalytic site (example shown in green), while 29 bound to one of the lipophilic branches of the L and 9 bound to the other (cyan and magenta). Five fragments bound at two different sites within the enzyme. The researchers discuss ten of the fragments in some detail.

Many of the fragments that bind at the catalytic site contain amides or ureas – moieties in known inhibitors – but some of them (such as compound 3 above) are more unusual. Also, despite the generally lipophilic nature of the binding pocket, many of the fragments – even those that bind in the hydrophobic branches of the L – make hydrogen bonds to the protein or to bound water molecules. This suggests that it should be possible to find potent inhibitors that are less hydrophobic than previously reported molecules. Fragment growing, linking, and merging approaches could all work, and indeed the researchers hint that results of these studies will be reported in future papers.

More importantly, this paper provides a great set of crystallographically validated fragments binding to distinct sites on a well-characterized protein. Helpfully, the researchers have deposited the coordinates of the 10 co-crystal structures discussed in the protein data bank. Moreover, all of the fragments are commercially available. This seems like an ideal model system for validating computational docking and scoring approaches as well as for better understanding the energetics of protein-ligand interactions. If I were an academician working in these areas, I’d jump in feet first!

Fragments vs CHK2: high-concentration screening comes through

Checkpoint Kinase 2 (CHK2) is an oncology target that has been kicking around for years. Its relevance is still debated, so having more small molecule inhibitors would go a long way toward assessing its therapeutic potential. In a recent paper in PLoS One, Rob van Montfort and colleagues at The Institute of Cancer Research (UK) present their fragment-based efforts on CHK2.

The researchers describe the design of their screening library in some detail, starting with a series of typical computational filters on commercially available molecules. Although most of the molecules had MW < 300, molecular weights up to 320 Da were allowed for fragments containing F, Cl, or SO₂ moieties. Also, all fragments were required to have at least 10 heavy atoms, which is on the high-side for a minimum. A total of 1869 fragments were purchased. All of these were analyzed for solubility and purity (by nephelometry and LC-MS, respectively), though unfortunately the researchers do not provide pass rates.

Having assembled the library, the researchers then screened each fragment at 300 micromolar against CHK2 in a biochemical assay (AlphaScreen). This led to 45 hits, but 25 of these showed some interference with the AlphaScreen assay itself. However, the remaining 20 all showed dose-response curves in a different assay format, giving IC₅₀ values from 2.7 to 944 micromolar.

In parallel, the researchers screened CHK2 using a thermal shift assay, with each fragment present at 2 mM. Perhaps not surprisingly given the higher concentrations used, this led to 63 hits.

Where things got interesting – and encouraging – was when the researchers compared hits identified using the two methods. In contrast to others' experiences, there was reasonable overlap; of the 14 hits from both assays, 12 yielded measurable IC₅₀ values when assessed using a microfluidic functional assay. Most of the AlphaScreen hits that didn’t produce thermal shifts came from the set of 25 that had previously been flagged as interfering with the AlphaScreen assay itself, and several were also insoluble. Of the 49 thermal shift hits that did not show up in the AlphaScreen assay, 13 were insoluble. Regarding the remaining 36, the researchers propose that they may bind to CHK2 outside its active site and thus don’t inhibit enzymatic activity.

Next, the researchers attempted to characterize the binding modes of the fragments crystallographically. Of the nine fragments that produced structures, eight came from the set of fragments confirmed using both AlphaScreen and thermal shift. Significantly, the only fragment to yield a structure that was identified solely from the thermal shift assay also produced the worst IC₅₀ value (228 micromolar) and the lowest ligand efficiency. All nine fragments bind to the so-called hinge region of the kinase.

One interesting observation was that, although the library did contain larger molecules, 6 of the 9 fragments characterized crystallographically had MW < 200, and the other 3 were well under 300 Da. This is exactly what you would expect according to the concept of molecular complexity, and suggests that adding larger fragments to your library may actually lower your hit rate (though admittedly it may be a stretch to conclude too much from this one study).

Another interesting note is that co-crystallization was used in all cases. Folks sometimes believe that you need to grow vats of crystals for fragment soaking experiments, but co-crystallizing worked fine here, and in some cases may allow the protein to adopt different conformations than if grown in the absence of small molecules.

Overall, the paper presents some nice starting points against CHK2. Perhaps more important, this is a thorough, well-written, and open access account of fragment screening that is well worth perusing by anyone embarking on a fragment campaign.

PAINS made painless

Practical Fragments has several entries on pan-assay interference compounds, or PAINS: see here for an introduction, here for their (mis)incorporation into a fragment library, here for the sad results of such misincorporations, here for a much longer review, and here for something that will hopefully bring a smile to your face after reading the previous tales of woe.

But artifacts are not only (or even primarily!) restricted to fragments, so Michael Walters at The University of Minnesota has established a blog devoted to PAINS (www.htspains.com).

As with any blog, success depends to a large degree on engagement with the broader community, so please check it out – and leave comments!

Fragments vs DDAH – covalent and noncovalent

Using biochemical assays to find fragments sidesteps the need for expensive biophysical instruments, but is fraught with difficulties. In a recent paper in Bioorg. Med. Chem., Thomas Linsky and Walter Fast at the University of Texas, Austin, describe how they used functional screening to discover fragment inhibitors of dimethylarginine dimethylaminohydrolase (DDAH), an enzyme involved in nitric oxide production.

The researchers used a 4000-member fragment library from ChemBridge, which they screened against two different isoforms of DDAH at 0.1 mM in a biochemical screen. This resulted in 79 hits against the human enzyme and 44 hits against a bacterial (P. aeruginosa) DDAH, 101 in total. 66 of these were then repurchased along with 41 analogs, and tested in a completely different biochemical assay against human DDAH-1; 31 showed >20% inhibition at 0.4 mM, including only 22 of the original hits, suggesting that many of the initial hits were indeed false positives. Further studies showed that 5 of the 31 compounds interfered with this secondary assay (ie, they showed “inhibition” even in the absence of enzyme), suggesting that they were false positives in both the primary and secondary biochemical assays.

DDAH contains an active-site cysteine, and the researchers wanted to exclude molecules that might be generically reactive, so they incubated the remaining 26 compounds with the low-molecular weight thiol glutathione and then retested them; this eliminated another 21 compounds.

Finally, the remaining 5 compounds were examined by mass-spectrometry, and one of these turned out to be a compound other than what was listed on the bottle! This left just 4 legitimate fragment hits.

Two of these compounds were 4-halopyridines, which, although not generally reactive with thiols, could covalently modify the active site cysteine of DDAH (see here for more details). The other two compounds were reversible, competitive inhibitors of human DDAH-1. Although low affinity (Ki values of 0.8 and 1.7 mM), they had respectable ligand efficiencies (0.38 and 0.29 kcal/mol/atom, respectively).

Interestingly, when Linksy and Fast retrospectively analyzed where the four validated fragments came from, they found that only the two halopyridines were detected in the primary screen; the two reversible fragment hits had been purchased for the secondary round of screening as analogs of primary hits.

This is a richly detailed and well-executed example of fragment-based screening in academia. It demonstrates once again that high-concentration biochemical screens can be used to find fragments, but be prepared to wade through a lot of junk: only about 2% of the original hits proved to be legitimate (see here for similar results from Vernalis on a different target). It also illustrates the utility of exploring analogs of initial fragment hits; in this case, even though most of the primary hits didn’t hold up, they nonetheless led to new fragments. Of course, this does raise the question of how spurious primary hits can lead to genuine inhibitors – what do you think?

Fragments vs matrix metalloproteinase-13: avoiding the metal

Matrix metalloproteinases (MMPs), as their name suggests, are metal-dependent proteases that cleave the extracellular matrix. They have been implicated in a wide variety of diseases including cancer and inflammation. MMPs have also been used as model systems to study the effects of fragment linking. (In fact, the first successful example of SAR by NMR was conducted against MMP-3.) Most inhibitors, including those starting from fragments, interact with the catalytic zinc in order to achieve potency. However, with roughly two dozen human MMPs, all dependent on a zinc ion, selectivity has been tricky. A recent paper in J. Med. Chem. mostly from researchers at Boehringer Ingelheim sidesteps this problem nicely.

MMP-13 is one of the more interesting members of the family due to its apparent role in rheumatoid arthritis. The enzyme is crystallographically well-behaved, and has a large substrate-binding pocket (the S1’ pocket) near the catalytic zinc that is dissimilar from other S1’ pockets. Even better, there is an adjacent side pocket (S1’*) that can open when the S1’ pocket is occupied, providing further selectivity.

The researchers started by performing a virtual screen of their entire corporate library to look for fragments that might bind in the S1’ pocket. These were added to an in-house fragment collection, and the combined set of roughly 1000 compounds was screened at 0.5 mM concentration in a biochemical screen. Compounds were also assessed using NMR (saturation transfer difference) and size exclusion chromatography mass-spectrometry. One of the best hits was Compound 1, which came from the virtual screen and was originally made as a synthetic intermediate in a completely different program.

Crystallography revealed that Compound 1 does in fact bind in the S1’ pocket, making several hydrogen bonds, with the amide moiety pointed towards the S1’* portion of the protein. This compound also displayed some selectivity towards two other MMPs. Fragment growing towards the S1’* pocket led to compound 11, with increased potency and ligand efficiency, and ultimately to compound 15, with low nanomolar potency and > 1000-fold selectivity against 9 other MMPs. Crystallography revealed that, as expected, the compound binds with the benzoic acid moiety in the S1’* pocket. And despite the presence of the ethyl ester, compound 15 is orally bioavailable in rats.

The paper gives no indication of where the program is today, but it is another nice example of fragment growing, as well as taking an unconventional approach to achieve selectivity.

Fragments vs PI3 kinase

The phosphatidylinositol-3 kinases (PI3Ks) comprise a family of lipid kinases that are important in a variety of biological pathways and have thus become popular targets for drug-discovery; earlier this year we highlighted a fragment screen out of AstraZeneca against four members of the family. In the most recent issue of Bioorg. Med. Chem. Lett., researchers from Pfizer have published their approach to one of these targets.

Samantha Hughes and colleagues first tested 5960 fragments in a biochemical assay (at 0.5-1.5 mM) to find molecules that inhibited PI3gamma. Hundreds of hits resulted, of which 150 were confirmed in full dose-response curves. These were further triaged using orthogonal methods, ultimately resulting in five X-ray structures of co-complexes, including compound 1, which binds to the hinge region of the kinase. Virtual screening of the larger Pfizer library led to the discovery of additional compounds such as compound 2. Growing compound 1 by adding an acetyl group generated compound 9, improving both potency and ligand efficiency, but synthetic challenges stymied further work. However, a closer examination of the crystal structure of compound 1 suggested a merging strategy with the previously reported compound 10 to generate compound 12, with high potency and ligand efficiency.

Astute (or paranoid) readers will notice that compound 12 contains a Michael acceptor that looks suspiciously reactive. In fact, it is closely related to the notorious rhodanines, many of which form covalent bonds with proteins and, because of the resulting promiscuity, have been accused of “polluting the literature.” Nonetheless, crystallography revealed that the compound binds (noncovalently) to PI3gamma exactly as designed. Moreover, it is fairly selective for other kinases, inhibiting only 3 out of 43 tested at 1 micromolar compound. Compound 12 is also metabolically stable, permeable, and cell active. This is a good example of why it is important not to be overly dogmatic in compound triaging, particularly at an early stage in a project.

Finally, it is worth noting that this paper comes from the storied Sandwich Laboratories, which are being closed down. If there is a silver lining to this tragedy it is that the closure has resulted in a flurry of nice publications from the site. Still, such papers hardly offset the opportunity cost of the drugs that would otherwise have been discovered there – nor the disruption caused to hundreds of scientists. Practical Fragments wishes the best of luck to all of them.

Small but PAINful

We’ve written previously about the phenomenon of compound aggregation, and how this can lead to false positives in high-concentration screening (see also here). Unfortunately, aggregation is not the only thing that can trip you up. Earlier this year Jonathan Baell and Georgina Holloway published a description and systematic catalog of “Pan Assay Interference Compounds,” or PAINS, and Baell has now followed up on this with a Perspective in the October issue of Future Medicinal Chemistry. (For those of you without journal access, he has also summarized some of this material here.)

PAINS are compounds that frequently show up as screening hits, but that act through non-specific mechanisms such as covalent attachment to proteins or generation of hydrogen peroxide. Sometimes it is not the compound itself that is problematic but a breakdown product or leftover reactant; most readers with experience in hit-to-lead discovery will experience a painful sense of déjà vu reading about hits that did not confirm when the compound was resynthesized or repurified. This problem is exacerbated when trifluoracetic acid (TFA, which is toxic to cells and can accelerate compounds’ decomposition) is used in reverse-phase purification of compounds; apparently Pfizer has systematically repurified hundreds of thousands of their screening compounds to remove traces of TFA.

The problem with PAINS is that they may show convincing biochemical and even cell based activity, but mechanistically be useless for further advancement to drugs or even chemical probes. Unfortunately this does not prevent them from being published, where they are picked up by other researchers and identified as “privileged pharmacophores” by computational chemists. This leads to more (wasted) research and more (useless) publications in a vicious circle Baell terms “pollution of the scientific literature.”

Although the focus of the Perspective is high-throughput screening, fragments can be PAINS too, as the sampling below shows.

Compound 1 is obviously a Michael acceptor and compound 2 is unsurprisingly reactive with cysteine, but the problems with compound 3 are less obvious, and compound 4 generates hydrogen peroxide in the presence of reducing agents and oxygen – a mechanism it likely took some time to track down.

One of the fun aspects of reading this paper is that Baell is not shy about calling out high-profile publications that are likely to report false positives, though he readily admits that he too has been misled. The question is what to do: using computational filters to get rid of the worst offenders only eliminates 5-12% of commercial compounds, but more stringent screens can eliminate upwards of 95%!

Baell calls for better stewardship of the scientific literature:

Journals therefore have a responsibility here because if misleading information is published, unintentional though it may be, it is more likely to propagate and be taken as fact by others who may then initiate flawed research projects. Research resources are too precious for this to be acceptable.

I agree whole-heartedly with this, though given what I see published on a weekly basis I despair of this happening any time soon. Ultimately of course the onus is on researchers to carefully follow up and fully understand the mechanism of their hits, particularly those with dubious structures: don’t contribute to the pollution yourself, and don’t let friends (or papers you review) pollute!

Avoiding will-o’-the-wisps: aggregation artifacts in activity assays

The phenomenon of aggregation is the drug hunter’s quicksand. A prerequisite for using biochemical assays to study fragments – or any low-affinity molecules – is an ability to sort activity from artifact. Many small molecules, even bona fide drugs, form aggregates in aqueous solution, and these aggregates can non-specifically interfere with biochemical assays. There are several ways to expose these promiscuous inhibitors (see list below), but even with vigilance, researchers can inadvertently stumble onto a route lit by will-o’-the-wisps. The most recent issue of J. Med. Chem. provides a particularly insidious example from Brian Shoichet, Adam Renslo, and colleagues at UCSF.

The researchers were looking for noncovalent inhibitors of cruzain, a popular protease target for Chagas’ disease. After a virtual screen of commercial lead-like compounds, 17 molecules were purchased and tested in enzymatic assays, and compound 1 (below) inhibited cruzain, albeit weakly. However, the compound looked like the real deal: it showed no time-dependence; it was active in the presence of detergent; and Lineweaver-Burk plots revealed that it was mechanistically competitive.

The researchers thus turned to medicinal chemistry, replacing the ester group of compound 1 with an oxadiazole bioisostere and swapping the aryl group for a substituted pyrazole, ultimately arriving at molecules such as compound 21, more than two orders of magnitude more potent than the starting molecule.



So far, so standard: similar stories appear every week in J. Med. Chem., Bioorg. Med. Chem. Lett., ChemMedChem, and other journals, and it would not have been surprising to see this published with a title like “Discovery of a high affinity inhibitor of cruzain.” Only in this case, the researchers became suspicious: most of the molecules were not active against the targeted protozoa, and many of the dose-response curves had unusually steep Hill slopes, a tell-tale sign of aggregation. Looking more closely at their protocol, the researchers also realized that the concentration of non-ionic detergent in their assays was ten-fold lower than they had thought. D’oh!

A series of tests confirmed that, despite interpretable and rationalizable SAR, the series had been optimized for aggregation-based inhibition: compound 21, with an IC50 of 200 nM in buffer containing 0.001% of the detergent Triton X-100, showed no inhibition whatsoever in 0.01% Triton X-100. The compound also inhibited AmpC beta-lactamase, an enzyme particularly sensitive to aggregators, and this inhibition could be reversed with detergent. Finally, dynamic light scattering (DLS) revealed the presence of particles (or aggregates) in aqueous solutions of compound 21.

But the tale gets even more twisted. Some of the aggregators show legitimate, competitive binding to cruzain under high-detergent conditions, albeit at much higher concentrations (with IC50s above 40 micromolar). Conversely, compound 1 actually shows noncompetitive behavior in low-detergent conditions, though again only at fairly high concentrations. In other words, promiscuous inhibitors can behave legitimately under sufficiently stringent conditions, and legitimate inhibitors can behave promiscuously under less stringent conditions.

What’s especially sobering is how easy this promiscuity would have been to overlook: many molecules with good activity in biochemical assays don’t show any effects in cells, and it is easy to ignore steep slopes in inhibition assays. How many of those “Discovery of a high affinity inhibitor of Hot Target X” papers actually report promiscuous inhibitors? The authors, who have been researching this problem for a long time, end on a justifiably paranoid note:

The cautionary contribution of this study is to point out that even within a clear SAR series, one is never entirely free from the concern that non-stoichiometric, artifactual mechanisms are contributing to the inhibition one observes.

This is a serious problem, both for the researchers doing the original work and for anyone trying to follow up on the results. But one can take precautions, summarized below and described more fully here:

• Add non-ionic detergent to the assay (Triton-X 100, Tween-20, CHAPS, others)
• Increase protein concentration – this should have no effect on genuine binders (within limits)
• Characterize the mechanism of inhibition (competitive, noncompetitive, or uncompetitive): competitive inhibitors are normally not promiscuous
• Centrifuge your samples and retest them – this can sometimes remove aggregators
• Examine your samples with DLS or flow cytometry – aggregators can sometimes be directly observed as 50-1000 nm particles
• Look closely at your dose-response curve - unusually steep slopes can signal aggregation

And of course, biophysical methods such as SPR, NMR, and X-ray crystallography can provide more information than biochemical assays and reveal stoichiometric (and – in the case of SPR – superstoichiometric) binding.

Difficulty sorting true low-affinity binders from false positives stymied fragment-based approaches for decades, and in fact the nature of promiscuous inhibition caused by aggregation wasn’t even characterized until earlier this century. We now have techniques to sort deceptive aggregation from true but faint affinity. Let’s make sure these tools are consistently used.