Covalent fragments in the clinic: VVD-133214

Back in 2016 we highlighted a paper describing chemoproteomic screening of covalent fragments. That technology formed the basis of Vividion, which was acquired by Bayer in 2021. Now, a paper just published in Nature by Matthew Patricelli, Todd Kinsella, and collaborators at Vividion, Roche, and Universitat Autònoma de Barcelona describes one of the fruits to come from this platform.

 

The work stems from another promising recent approach to find oncology targets, synthetic lethality: searching for proteins that are essential in certain types of cancer cells but dispensable for normal cells, which might mean reduced toxicity. WRN is a DNA helicase that can clean up secondary DNA structures caused by expanded TA-dinucleotide repeats found in cancer cells with microsatellite instability (MSI), which is caused by mutations in DNA repair genes. Previous research had shown that knocking out WRN caused double-stranded DNA breaks and cell death in MSI-high (MSI-H) cancer cells but not normal cells, which do not have so many expanded TA-dinucleotide repeats. This has set off an industry-wide search for WRN inhibitors.

 

The researchers screened several thousand fragment electrophiles against cell lysates and found that some, such as VVD-109063, modified C727 of WRN. Although this cysteine is located some distance from the ATP binding site, functional activity studies with the pure protein found that the molecule was an inhibitor.

 

Optimization of VVD-109063 and related molecules found inconsistencies between results in lysates and intact cells. Some engaged C727 better in intact cells than lysates, others worse. Differences in cell permeability were ruled out by the fact that a cysteine on an unrelated protein was liganded to a similar extent in cell lysates and intact cells. The researchers speculated that, because cell lysates are diluted, they have lower ATP concentrations, and sure enough some molecules were less active in the presence of ATP while others were more active.

 

The team decided to focus on the second class. Optimization ultimately resulted in the clinical candidate VVD-133214. (Unfortunately details are not given; the paper does say these will be provided elsewhere).

 

A crystal structure of VVD-133214 confirmed covalent binding to C727, with the molecule in a hydrophobic pocket in a flexible “hinge region” of the protein. This causes a conformational rearrangement into a “closed” form, which presumably affects the catalytic activity of the helicase. Surprisingly, there are no hydrogen bonds between WRN and VVD-133214. This is highly unusual: a paper we discussed in 2021 found >90% of fragment-derived leads had at least one polar contact.

 

The k_inact/K_i value is reported as being 4848 M^-1s^-1, which is on the low side for clinical-stage irreversible inhibitors. Like sotorasib, its potency seems driven by k_inact, with the K_i being greater than > 15 µM. Consistent with this low inherent affinity, the molecule was inactive against the C727A mutant enzyme.

 

Much of the paper focuses on the biology, which is interesting but beyond the scope of this post. Suffice it to say that VVD-133214 is cytotoxic in MSI-H cells, where it causes G2 arrest and DNA damage, but inactive in microsatellite stable (MSS) cells. Oral dosing led to tumor regression in several MSI-H mouse models, including patient-derived xenografts.

 

This is a nice paper, though I look forward to a full account of the medicinal chemistry. In particular, vinyl sulfones are generally considered quite reactive, and I know of only one other clinical-stage molecule with this warhead. Presumably the cyclopropyl substituent was added at least in part to sterically block access to the electrophile.

 

Also, while the paper refers to VVD-133214 as “clinical-stage,” it appears neither on clinicaltrials.gov nor on Vividion’s website. The Roche website lists RG6457 as a phase 1 WRN covalent inhibitor partnered with Vividion, so perhaps this is the same molecule.

 

The paper ends by mentioning another clinical-stage WRN inhibitor from a different company, this one noncovalent. It notes that “this presents a rare opportunity to compare two small molecule oncology drugs targeting the same protein by different mechanisms,” and that using both could be useful in combating resistance. Practical Fragments wishes luck to these – and other drugs targeting WRN – helping patients quickly.

The limits of published data

Machine learning (or, for investors, artificial intelligence) has received plenty of attention. To be successful you need lots of data. If you’re trying to, say, train a large language model to write limericks you’ve got oodles of text to draw from. But to train a model that can predict binding affinities you need lots of measurements, and they must be accurate. Among public repositories, ChEMBL is one of the most prominent due to its size (>2.4 million compounds) and quality, achieved through manual curation. But even here you need to be cautious, as illustrated in a recent open-access J. Chem. Inf. Model. paper by Gregory Landrum and Sereina Riniker (both at ETH Zurich).

 

The researchers were interested in the consistency of IC₅₀ or K_i values for the same compound against the same target. They downloaded data for >50,000 compounds run at least twice against the same target. Activity data were compared either directly or after “maximal curation,” which entailed removing duplicate measurements from the same paper, removing data against mutant proteins, separating binding vs functional data, and several other quality checks. They used a variety of statistical tests (R², Kendall τ, Cohen’s κ, Matthew’s correlation coefficient) to evaluate how well the data agreed, the simplest being the fraction of pairs where the difference was more than 0.3 or 1 log units, roughly two-fold or ten-fold differences.

 

The results were not encouraging. Looking at IC₅₀ values, 64% of pairs differed by >0.3 log units, and 27% differed by more than 1 log unit. In other words, for more than a quarter of measurements a molecule might test as 100 nM in one assay and >1 µM in another.

 

Of course, as the researchers note, “it is generally not scientifically valid to combine values from different IC₅₀ assays without knowledge of the assay conditions.” For example, the concentration of ATP in a kinase assay can have dramatic effects on the IC₅₀ values for an inhibitor. Surely K_i values should be more comparable. But no, 67% of pairs differed by >0.3 log units and 30% differed by >1!

 

The situation improved for IC₅₀ values using maximal curation, with the fraction of pairs differing by >0.3 and >1 log units dropping to 48% and 13%. However, this came at the expense of throwing away 99% of the data.

 

Surprisingly, using maximal curation data for K_i data actually made the situation worse. Digging into the data, the researchers found that 32 assays reporting K_i values for human carbonic anhydrase I, all from the same corresponding author, include “a significant number of overlapping compounds, with results that are sometimes inconsistent.” Scrubbing these improved the situation, but 38% of pairs still differed by >0.3 log units, and 21% differed by >1 log unit.

 

This is all rather sobering, and suggests there are limits to the quality of available data. As we noted in January there are all kinds of reasons assays can differ even within the same lab. Add in subtle variations in protein activity or buffer conditions and perhaps we should not be too surprised at log-order differences in experimental measurements. And this assumes everyone is trying to do good science: I’m sure sloppy and fraudulent data only make the situation worse. No matter how well we build our computational tools, noisy data will ensure they often differ from reality, whatever that may be.

Detailing hot spots with atomic consensus sites

Practical Fragments has written frequently about hot spots, regions on proteins that are predisposed to bind ligands such as drugs. Determining whether a protein has a hot spot can help prioritize a target for screening, and one of the more established computational approaches to do so is FTMap, which we wrote about most recently just a couple months ago.

 

While FTMap can tell you whether a protein has one or more hot spots, it provides few further details, such as which regions might prefer a hydrogen bond donor or acceptor. This has now been addressed in a new J. Chem. Inf. Mod. paper by Sandor Vajda and collaborators at Boston University, Stony Brook University, and Acpharis. (Diane Joseph-McCarthy presented some of this work at the CHI DDC conference earlier this month.)

 

The original version of FTMap started with a collection of 16 very small molecule probes: these were docked all over a protein, with hot spots being identified as consensus sites where many probes bound. To get more information about each hot spot, the researchers have extended the method – now called E-FTMap – by increasing the number of probes to 119 covering key functional groups. For example, whereas FTMap included dimethyl ether as a probe, E-FTMap also includes 2-methoxypropane, 2-methoxy-2-methylpropane, and tetrahydropyran. If all these probes bind with the oxygen in the same part of the hot spot, this suggests a predilection for a hydrogen bond acceptor, and also provides information about nearby hydrophobic contacts.

 

By using a sufficiently diverse group of virtual probes, E-FTMap is able to more finely detail hot spots, tallying the “atomic consensus sites” within them. This is reminiscent of an approach we wrote about several years ago, though that method used just three different probes.

 

To benchmark E-FTMap, the researchers took 109 fragment-to-lead pairs with published crystallographic information and assessed whether the program could identify interactions that had been experimentally observed. The results were encouraging and far superior to the original version of FTMap. The highest ranked atomic consensus sites generally overlapped with appropriate atoms in fragments and leads. Interestingly, the results for fragments were better than those for leads, and the researchers suggest this is because the fragment “core is responsible for the bulk of the binding free energy in a ligand and that larger ligands bind by forming additional interactions at weaker hot spots that surround the fragment binding site.”

 

Next, E-FTMap was tested against five proteins for which between 31 and 353 fragment-bound crystal structures were available. Here too the program was broadly successful, though some fragments bound regions of the protein that E-FTMap overlooked, particularly in cases where there were conformational changes. This is not surprising given that the program assumes the protein remains rigid. (Other computational approaches such as SWISH, which we wrote about here, are starting to account for protein flexibility.)

 

E-FTMap looks qualitatively at specific atomic interactions, and one question I had was how well the atomic consensus sites matched up with binding affinities of known fragments; perhaps some crystallographically identified fragments bind so weakly one would not expect to find them computationally, as we discussed here and here. This hypothesis might be tested by focusing on comparisons with experimentally characterized fragments with the highest ligand efficiencies.

 

Also, I was struck by the fact that the virtual probes in E-FTMap are roughly the size of MiniFrags or MicroFrags, and I couldn’t help but wonder how well the atomic consensus sites from the virtual screens would correlate with the binding modes of these tiniest of fragments.

 

One nice feature of E-FTMap is that it can be accessed through a simple web server, so if you’re interested in these and other questions you can test it for yourself. If you do, please share your experiences.

Nineteenth Annual Fragment-Based Drug Discovery Meeting

Last week the CHI Drug Discovery Chemistry (DDC) meeting was held in San Diego. This was the largest ever, with more than 900 participants, 95% of whom attended in person, up from 87% last year. I won’t attempt to cover all fourteen tracks but will just touch on some of the main themes.

 

Computational approaches

All four days of the conference featured dedicated sessions on machine learning and artificial intelligence, but since I was in other sessions I don’t know how relevant they were to FBLD. If you attended an interesting talk please let me know so I can watch it on-demand.

 

Among computational talks I did see, Antonina Nazarova (University of Southern California) provided an update on V-SYNTHES, which we first wrote about here. This synthon-based screening approach now covers 36 billion molecules and has been tested against eight different proteins, four of which yielded nanomolar hits when tested experimentally.

 

Computational methods have historically treated proteins as rigid, though many targets are anything but. Diane Joseph-McCarthy (Boston University) described an improvement to the pocket finding approach FTMap, called FTMove, to incorporate molecular dynamics by starting with an ensemble of different crystal structures. A further advance is E-FTMap, which expands the number of virtual probes from 16 to 119 to more finely assess ligandable sites.

 

Benjamin Walters (Genentech) described using protein dynamics to find cryptic pockets using ESP, or Experimental Structure Prediction. In this approach, experimental data from hydrogen-deuterium exchange (HDX) or chemical shift perturbations (CSPs) are used to constrain multiple parallel computational simulations, leading to better models than flexible docking, even for weak fragments.

 

Experimental approaches

Protein-detected NMR was the first practical fragment-finding method, and Steve Fesik (Vanderbilt) described using SAR by NMR to find fragments binding to the papain-like protease of SARS-CoV-2. These have been advanced to molecules with nanomolar affinity and activity in cell-based assays.

 

Andreas Lingel described the new fluorine-containing fragment library built at Novartis and how ¹⁹F NMR was used to generate inhibitors of IL-1β. We wrote about that success last year, noting that the initial fragment hit was “super-sized,” and Andreas confirmed that for trifluoromethyl-containing fragments the upper molecular weight limit was relaxed to 350 Da.

 

Sriram Tyagarajan (Merck) presented a crystallographic screen against the neurodegeneration target TTBK1 which yielded hits at 15 sites. Several potential allosteric sites were identified, but fragment growing and linking were not successful, leading them to a quick (3 month) no-go decision on the protein.

 

Virgil Woods (City University of New York) described using crystallographic screening to find hits against the challenging phosphatase PTP1B both under conventional cryogenic temperatures as well as at room temperature. As we noted about related work, there was a surprisingly poor overlap between the two sets of hits, and some fragments bound in a different manner at different temperatures.

 

Integrating FBDD and DNA-encoded libraries (DEL) for lead generation was the topic of Chaohong Sun’s talk. She noted that of some two dozen targets at AbbVie screened by both methods, 60% found hits from both, 10% found only fragment hits, and 5% found only DEL hits, with a quarter of the targets producing no hits. Hits from both approaches can be combined, as we noted here. Chaohong also noted that for both FBDD and DEL, high quality protein is essential for successful screens.

 

Covalent approaches

Covalent approaches to drug discovery are becoming ever more acceptable as more covalent drugs are approved. Understanding these in depth was the focus of Micah Niphakis (Lundbeck), who characterized 22 approved drugs containing 18 different warheads. The stability in buffer, liver microsomes, and hepatocytes varied dramatically, though more recently approved drugs tended to be more stable. Chemoproteomic studies revealed many off-targets in cells; for example, all the kinase inhibitors tested hit BTK to some extent even when this was not the primary target. The fact that the drugs are (mostly) safe and well-tolerated is a useful reminder that just because we can detect something doesn’t mean it is relevant.

 

Henry Blackwell described building a 12,000-member covalent fragment library at AstraZeneca. Due to the presence of a warhead, they relaxed rule of three parameters, with MW ranging from 250-400 Da and ClogP from 0-4. Henry also discussed the successful use of this library to identify covalent hits against the anticancer target BFL1 that were optimized to k_inact/K_I ~ 7000 M^-1s^-1. This accomplishment is all the more impressive given that screens using ASMS, DSF, ¹⁹F NMR, and SPR had all failed to yield validated hits.

 

We recently wrote about electrophilic MiniFrags, and György Keserű (Research Center for Natural Sciences, Hungary) described screening these against HDAC8 and the main protease from SARS-CoV-2. He also mentioned that the set is available for purchase from Enamine, so you can try it yourself against your favorite target.

 

As covalent modifiers become more common we will see new metrics for characterizing them, as illustrated by Benjamin Horning’s (Vividion) presentation, “Ligand Efficiency Metrics in Covalent Drug Discovery.” He described Ligand Reactivity Efficiency (LRE), defined as pTE₅₀(target, 1 hr) – pTE₅₀(glutathione, 1 hr), where TE is the target (or glutathione) engagement. LRE is analogous to LLE but focused on reactivity rather than lipophilicity. Despite my post last week, the metric could be useful, and I look forward to seeing what Dr. Saysno and friends will make of it.

 

Most covalent modifiers bind to a target and remain intact, but Nir London (Weizmann Institute) has developed Covalent Ligand Directed Release (CoLDR), in which a portion of the small molecule leaves; applications include release of fluorescent or chemiluminescent probes. Useable warheads include α-substituted methacrylamides and sulfamate acetamides.

 

Although more recent covalent drugs have targeted cysteine residues, there is growing interest in other amino acid side chains. Nir mentioned that thio-methacrylate esters can react with lysine residues, thought the kinetics are slow. And Carlo Ballatore (University of California San Diego) described hydroxy-naphthaldehyde fragments that bound reversibly to a lysine on the vascular target KRIT1.

 

Both plenary keynote speakers focused heavily on covalent chemistry. Dan Nomura (UC Berkeley) described using chemoproteomics approaches to find covalent molecules that could inhibit, degrade, or change the cellular localization of myriad proteins.

 

Finally, K. Barry Sharpless (Scripps), one of only five people to have been awarded the Nobel Prize twice, gave a rich description of sulfur (VI) fluoride exchange chemistry (SuFEx), which included drawing chemical structures on a flip chart. He presented the discovery of a fluorosulfate that is bactericidal against multiple resistant forms of Mycobacterium tuberculosis. Interestingly, the molecule works by modifying a catalytic serine residue which then cyclizes to form a β-lactam. His passion for chemistry is obvious, but he also has personal reasons for pursuing the second most deadly infectious disease: his brother died of tuberculosis before effective drugs were developed. And with the rise of extensively drug resistant TB, we’ll need new ones.

 

I’ll end on that note, but please leave comments. And mark your calendar for April 14-17 next year, when DDC returns to San Diego.

Personality tests for molecules

Long-time readers of Practical Fragments will be familiar with various metrics for measuring molecules, such as LE, LLE, and WTF. But these are all hard-edged, numerical constructs. Some folks argue that we should take a softer, more nuanced approach. This call has been heeded by Katharine Bigg and Isabel Myerrors in the form of a “Myerrors-Bigg” Type Indicator, or MBTI.

 

The MBTI consists of a series of questions which rank a molecule into four dimensions: Extroversion/Introversion, Sociable/Nonsociable, Flat/Three-dimensional, and Pretty/Janky. Defining molecules as extroverts may sound strange, but it really just comes down to a question of molecular recognition: we’ve noted that 4-bromopyrazole seems to bind to just about every protein and is thus an Extrovert while other compounds, being Introverts, fall into the category of “dark chemical matter” and never come up in screens.

 

As for the other dimensions, Practical Fragments has written previously about (Non)Sociable fragments as well as Flat fragments. This leads to the last dimension. Claims that beauty is in the eye of the beholder are undermined by the rigorous process of the MBTI, which places molecules such as curcumin squarely in the Janky category while approved drugs are self-evidently Pretty. Thus, toxoflavin is an ESFJ, while sotorasib is an INTP.

 

The utility of the MBTI remains to be established, but this has not stopped companies everywhere from applying it in their acquisition and evaluation processes. And other tests, such as the Decagram of Personality and the Big Six Personality Traits, are also becoming popular. Which do you prefer?