Eleventh Annual Fragment-based Drug Discovery Meeting

The first major fragment event of 2016, CHI’s Drug Discovery Chemistry, was held last week in San Diego. FBDD was the main focus of one track, and fragments played starring roles in several of the others as well, including inflammation, protein-protein interactions, and epigenetics. Also, for the first time this year the event included a one-day symposium on biophysical approaches, which also included plenty of fragments.

In agreement with our polls, surface plasmon resonance (SPR) received at least a mention in most of the talks. John Quinn (Genentech) gave an excellent overview of the technique, packed with lots of practical advice. At Genentech fragments are screened at 0.5 mM in 1% DMSO at 10°C using gradient injection, which permits calculation of affinities and ligand efficiencies directly from the primary screen. Confirmation of SPR hits in NMR is an impressive 80%. A key source of potential error in calculating affinities is rebinding, in which a fragment dissociates from one receptor and rebinds to another. That problem can be reduced by increasing the flow rate and minimizing the amount of protein immobilized to the surface. Doing so also lowers the signal and necessitates greater sensitivity, but happily the baseline noise has decreased by 10-fold in the past decade.

Some talks focused on using SPR for less conventional applications. Paul Belcher (GE) described using the Biacore S200 to measure fragments binding to wild-type GPCRs. In some cases this provided different hits than those detected against thermally stabilized GPCRs. And Phillip Schwartz (Takeda) described using SPR to characterize extremely potent covalent inhibitors for which standard enzymatic assays can produce misleading results. These screens require exotic conditions to regenerate the chip, so it helps that the SensiQ instrument has particularly durable plumbing.

In theory, SPR can be used to measure the thermodynamics of binding by running samples at different temperatures, but John Quinn pointed out that enthalpic interactions dominate for most fragments, so the extra effort may not be worthwhile. Several years ago many researchers felt that enthalpically driven binders might be more selective or generally superior. Today more people are realizing that thermodynamics is not quite so simple, and Ben Davis (Vernalis) may have put the nail in the coffin by showing that, for a set of 22 compounds, enthalpy and entropy of binding could vary wildly simply by changing the buffer from HEPES to PBS! (Free energy of binding remained the same with either buffer.)

Thermal shift assays (TSA or DSF) continued to be controversial, with Ben finding lack of agreement between the magnitude of the shift and affinity, though there was a correlation with success in crystal trials. In contrast, Mary Harner (BMS) reported good agreement between thermal shift and affinity. She also found that it seemed to work better when the fragments bound in deep pockets than when they bound closer to the surface. However, Rumin Zhang (Merck), who has tested more than 200 proteins using TSA, mentioned that some HCV protease inhibitors could be detected despite the shallow active site. Rumin also pointed out that a low response could indicate poor quality protein – if most of the protein is unfolded it might be fine for biochemical assays but not for TSA. Negative thermal shifts are common and, according to Rumin, sometimes lead to structures, though others found this to be the case less often.

What to do when assays don’t agree was the subject of lively discussion. Mary Harner noted that out of 19 targets screened in the past two years at BMS using NMR, SPR, and TSA, 45% of the BMS library hit in at least 1 assay. However, 68% of hits showed up in only a single assay. Retesting these did lead to more agreement, but even many of the hits that didn’t confirm in other assays ultimately led to leads. All techniques are subject to false negatives and false positives, so lack of agreement shouldn’t necessarily be cause for alarm. Indeed, Ben noted that multiple different soaking conditions often need to be attempted to obtain crystal structures of bound fragments.

Crystallography in general is benefiting from dramatic advances in automation. Jose Marquez described the fully automated system at the EMBL Grenoble Outstation, which is open to academic collaborators. And Radek Nowak (Structural Genomics Consortium, Oxford) discussed the automated crystal harvesting at the Diamond Light Source, which is capable of handling 200 crystals per hour. He also revealed a program called PANDAA (to be released soon) that speeds up the analysis of crystallographic data.

Crystallography was used as a primary screen against KEAP1, as discussed by Tom Davies (Astex). A subset of 330 of the most soluble fragments was tested in pools of four, which revealed several hot spots on the protein. Interestingly, an in-house computational screen had not identified all of these hot spots, though Adrian Whitty (Boston University) noted that they could be detected with FTMap. The fragments themselves bound exceptionally weakly, but intensive optimization led to a low nanomolar inhibitor.

Another case in which extremely weak fragments turned out to be useful was described by Matthias Frech (EMD Serono). A full HTS failed to find any confirmed hits against cyclophilin D, but screening by SPR produced 168 fragments, of which six were characterized crystallographically. Although these were all mM, with unimpressive ligand efficiencies, they could be linked or merged with known ligands to produce multiple leads – a process which took roughly one year from the beginning of the screen. Matthias noted that sometimes fragment efforts are started too late to make a difference, and that it is essential to not be dogmatic.

Huifen Chen discussed Genentech's MAP4K4 program. Of 2361 fragments screened by SPR, 225 had affinities better than 2 mM. Crystallography was tough, so docking was used instead, with 17 fragments pursued intensively for six months, ultimately leading to two lead series (see here and here), though one required bold changes to the core. This program is a nice reminder of why having multiple fragment hits can be useful, as the other 15 fragments didn’t pan out.

Finally, George Doherty (AbbVie) gave a good overview of the program behind recently approved venetoclax, which involved hundreds of scientists over two decades. He also described intensive medicinal chemistry which led to a second generation compound, ABT-731, with improved solubility and oral bioavailability.

We missed Teddy at this meeting, and there is plenty more to discuss, so please add your comments. If you did not attend, several excellent events are still coming up this year. And mark your calendar for 2017, when CHI returns to San Diego April 24-26.

Native mass spectrometry vs SPR

Native state electrospray ionization mass spectrometry (ESI-MS) is, in theory, a fast and easy way to find fragments: just mix protein with fragment, shoot it on the MS, and look for complex. As a bonus, the exact mass tells you whether your fragment is what you think it is (or at least whether it has the right mass). However, published examples are relatively rare, and not always favorable. A new paper in J. Med. Chem. by Tom Peat, Sally-Ann Poulsen, and their colleagues at CSIRO and Griffith University seeks to change this.

The researchers chose the fragment-friendly model protein carbonic anhydrase II (CA II) as their target. They first screened a library of 720 fragments, each at 100 µM, using surface plasmon resonance (SPR). This yielded 7 hits, with affinities ranging from 1.35 to 1280 µM. These seven hits were then assessed by ESI-MS using equimolar concentrations of protein and fragment (10 or 25 µM each). Encouragingly, all seven hits confirmed. Soaking these fragments into crystals of CA II yielded structures for six of them.

This is nice, but of course the real question is how well ESI-MS works as a primary screen. To address this, the researchers chose 70 compounds structurally related to the 7 hits and independently tested these using both SPR and ESI-MS. This yielded 37 hits, of which 24 were detected both by SPR and ESI-MS. In fact, every SPR hit was confirmed by ESI-MS. Of 14 fragments subsequently soaked into crystals of CA II, 7 provided interpretable electron density.

This is impressive, and the researchers note that the level of agreement between SPR and ESI-MS might be better still, since some of the ESI-MS hits did give signals by SPR – they were just weaker than the chosen cutoff (K_D ≤ 3 mM). Thus, in contrast to a paper discussed last year, ESI-MS does seem to be a sensitive detection method. In fact, given the low concentration of fragment needed, the researchers suggest that it could be useful for screening fragments with lower solubilities.

So what’s the secret to success? One difference from some previous reports is that the researchers used a 1:1 ratio of protein to fragment. Others have used excess fragment, which could lead to nonspecific binding and aggregate formation. And of course, CA II is a pretty forgiving model protein. I look forward to seeing ESI-MS used as a primary screen on more difficult targets.

Second fragment-based drug approved

Yesterday the US FDA approved venetoclax (VenclextaTM) for certain patients with chronic lymphocytic leukemia (CLL). This drug, which readers may know more familiarly as ABT-199, was co-developed by AbbVie and Genentech. The drug binds to BCL-2 and blocks its interaction with other proteins.

The first fragment-derived drug approved, vemurafenib, illustrated how quickly FBDD could move: just six years from the start of the program to approval. In contrast, venetoclax is the culmination of a program that has been running for more than two decades; Steve Fesik and his colleagues at Abbott published the X-ray and NMR structure of the protein BCL-xL back in 1996! The original SAR by NMR work was done on this protein, leading to ABT-263, which hits both BCL-xL and BCL-2. Subsequent work revealed that a selective BCL-2 inhibitor might be preferable in some cases, and further medicinal chemistry led to venetoclax.

This drug illustrates the power of fragments to tackle a difficult target by accessing unusual chemical space. It also illustrates creative, fearless, data-driven medicinal chemistry: not only does venetoclax violate the Rule of five, it even contains a nitro group, a moiety red-flagged due to its potential for forming toxic metabolites. This is a useful reminder that in our business rules are more appropriately considered guidelines, to be discarded when necessary.

Clinical results were sufficiently impressive that the drug was given breakthrough status and granted priority review, accelerated approval, and orphan drug designation. The ultimate victory is for the thousands of patients with relapsed CLL who have the 17p deletion on chromosome 17. In the registration trial, 80% of patients showed a partial or complete remission. It is rare to create something that works this well. Congratulations to all who played a role.

Fragments vs histone demethylases: docking and merging

Tweaking epigenetic machinery is increasingly popular as a therapeutic strategy. Epigenetics often involves modification to proteins – such as histones – that interact with DNA. One common type of modification is methylation of lysine or arginine residues. A couple months ago we highlighted how fragment-based approaches were used to discover inhibitors of a methyltransferase, one of many classes of protein-modifying enzymes that underlie epigenetics. Just as methyltransferases put methyl groups on, demethylases take methyl groups off. In a recent paper in J. Med. Chem., Udo Oppermann, Brian Shoichet, and Danica Fujimori and their collaborators at the University of Oxford and UCSF show that demethylases too can be successfully targeted with fragments. What’s more, the work exemplifies concrete contributions of computational approaches to both identify and advance fragments.

The demethylase KDM4C has been implicated in cancer. This enzyme uses iron, the cofactor α-ketoglutarate (α-KG), and oxygen as part of its mechanism. The researchers ran a computational screen (using DOCK 3.6) of more than 600,000 compounds in the ZINC fragment library. Top-scoring hits were triaged on the basis of novelty and good interactions with the iron atom, and 14 fragments were tested in a functional assay. Remarkably, all of them were active, with 7 showing IC₅₀ values < 200 µM!

Several of the top hits were 5-aminosalicylates such as compound 4. Testing 80 commercial analogs led to low micromolar inhibitors, but these could not be further optimized. Moreover, despite the small size and polarity of these compounds, many of them showed signs of aggregation – a reminder that this type of artifact must always be considered.

Unfortunately, crystallography was also not successful for any of the fragments or analogs. But the researchers noticed that, according to the docking results, fragments such as compound 4 could assume two different binding modes: in one, the carboxylate and phenol interacted with the iron atom, while in the other the carboxylate interacted with lysine and tyrosine residues in the protein. This inspired several ideas for fragment merging, leading to molecules such as compound 45. Additional variations led to mid-nanomolar inhibitors such as compound 35.

As expected, these molecules are competitive with the α-KG cofactor (which normally binds to the iron atom) but not with the peptide substrate. Many also showed encouraging selectivity profiles against other demethylases, though no cell data are reported. Finally, crystallography mostly confirmed the predicted binding models for several of the merged compounds, including compound 35.

This is a lovely example of using computational approaches not just for fragment-finding, but for fragment merging as well. As the authors point out, this was done not to showcase computational methods but because crystallography didn’t initially work. Even in the short lifetime of Practical Fragments, in silico methods have made remarkable progress, and this is another milestone. It will be fun to see further optimization of these molecules.

Biophysics: not just for fragments

Biophysics and fragment-based drug discovery go together like Nutella and strawberries. Indeed, SAR by NMR ushered in the dawn of fragment-based methods two decades ago, and most fragment-based programs today make use of NMR, SPR, and/or ITC – not to mention X-ray crystallography. Interestingly, the same is not necessarily true for high-throughput screening (HTS) programs. In a recent paper in Drug Discovery Today, Rutger Folmer makes a strong case for engaging biophysics early and often in HTS. He bolsters his argument with more than 20 examples from internal programs at AstraZeneca.

The first descriptions of using NMR to profile HTS hits were not published until several years after SAR by NMR, but they were rather shocking, with up to 98% of hits failing to confirm. Nor is this merely a historical problem, as discussed here. Aggregators, redox cyclers, generically reactive covalent modifiers – all of these are problems not just in fragment screening but in HTS as well. Sometimes the most potent hits are artifacts, particularly for more difficult targets. The key to triaging out pathological actors is to assess binding and not rely solely on inhibition.

That means bringing biophysics into hit profiling at the earliest stages, before trying to optimize fruitless hits. As Rutger points out, it is often difficult to rally colleagues to look at less active molecules after they have wasted months pursuing more potent dead ends.

And biophysics can make an impact even before running screens. Profiling published tool compounds or in-licensing opportunities with biophysical techniques can reveal unwelcome surprises. Testing the output of early HTS pre-screens (7000-10,000 compounds) before a full HTS (2 million compounds at AstraZeneca) can reveal whether an assay is particularly susceptible to false positives. In some cases this can result in reconfiguring the assay, for example by choosing a different detection technology or modifying the protein construct.

A key element to gaining such benefits is organizational commitment. At AstraZeneca, a biophysicist is assigned to a project team immediately after target selection – well before any screens are run. This seems like prudent practice. How many other organizations are doing this?

An interview with Dr. Saysno

Readers of a certain age may fondly remember the interviews with Dr. Noitall that used to enliven the pages of Science. Sadly, he died a few years back. But his cousin, Dr. Saysno, is still very much alive. Practical Fragments caught up with him at a recent conference in Shutka.

Practical Fragments (PF): Dr. Saysno, you've stated that experts should never be trusted.

Dr. Saysno (DS): Niels Bohr defined an expert as a person who has made all the mistakes that can be made in a very narrow field. If someone has made every possible mistake, how could you possibly trust them?

PF: But don't you think they may have learned from their mistakes?

DS: Balderdash! Hegel was right: the only thing we learn from history is that we learn nothing from history.

PF: What's your opinion of ligand efficiency (LE)?

DS: Ligand efficiency is an abomination! It's mathematically invalid!

Worse, determining the free energy of binding from a dissociation constant is not even wrong: if you change your definition of standard state, you can make ΔG° any positive or negative number you want. Just how relevant do you think your definition of standard state is on the surface of Venus? Or Pluto?

For the same reason, pH is utterly meaningless. You really ought to throw out your pH paper, not to mention your pH meters, since they all assume an arbitrary reference state.

PF: But what about all the researchers who find pH and LE useful?

DS: Usefulness is the last refuge of the scoundrel!

Look, many of the best selling drugs on the market are antibodies, and when you calculate their ligand efficiencies, they are close to zero. How can you have a metric that doesn't work on some of the most important drugs out there?

I only believe in equations that are universal and apply in all situations, unsullied by the physical world. Anything that involves standard states is just mumbo-jumbo.

PF: What do you think of pan-assay interference compounds, or PAINS?

DS: Now that’s a topic that really gets my blood boiling! PAINS were defined on the basis of just six assays. Six assays I tell you!!! [DS vigorously pounds his shoe on the desk.] Just because something hits six assays – or six hundred for that matter – doesn’t mean it will hit the six hundred and first!

PF: But aren't there some chemical substructures that are so generically reactive they should never be used in probes?

DS: Nothing is universal! All molecules are unique, like little snowflakes. If a compound comes up as a hit in your assay, by all means publish it as a chemical probe in the best possible journal, and try to encourage suppliers to start selling it so other people can use and cite your brilliant discovery.

No one has a right to criticize your molecule unless they test it against every single protein in the human body and show that it hits all of them.

When the revolution comes, the imperialist PAINS stooges will be swept into the dustbin of history along with the lackeys of ligand efficiency!

PF: So if you don't trust experts, you don't like metrics, and you can't make generalizations, how can we move forward in science short of deriving every result ourselves from first principles?

DS: That's simple: just ask me!