04 December 2023

Screening tough proteins by SPR

Surface-plasmon resonance (SPR) is among the most popular methods for finding fragments. However, as we have noted, SPR can be very prone to operator error and misinterpretation. In a recent (open access) SLAS Discovery paper, U. Helena Danielson (Uppsala University) and a who’s-who team of biophysicists from across Europe provide experimental strategies for screening difficult proteins.
The researchers chose five different proteins, some of which were screened in two or three different forms for a total of nine protein constructs. Six of these were screened against their FL1056 library, a custom-built 1056-member library which include molecules from the FragNet program. The library includes a number of “three dimensional” molecules as assessed by principal moment of inertia (PMI). The other library, FL90, is a small set of commercially available fragments we highlighted here.
Before screening compounds against proteins, the researchers conducted a “clean screen.” This involved injecting fragments (at 500 µM each) over the sensor surface using the same buffer that would be used in the actual screen to pre-identify fragments that stick to the surface. This typically disqualified about 1% of fragments, though for one set of conditions the number was closer to 3%.
That work done, the researchers turned to the actual screens. After proteins were immobilized on the sensor chips, the fragments were typically screened at a single concentration of 250 µM each. The threshold for the initial hit cutoff was set low, often around 10% of the library, to minimize false negatives. Subsequent follow-up studies at varying concentrations were used for confirmation. This led to a significant winnowing, with the final number of confirmed hits between 0.5 and 7% of the library.
The proteins themselves were intentionally chosen to present various difficulties. Acetylcholine binding protein (AChBP, which we wrote about here) forms a large (125 kDa) pentameric complex with multiple binding sites. Lysine demethylase 1 (LSD1) is a multidomain, cofactor-dependent protein that requires a partner protein, CoREST, for activity. LSD1 was screened in the presence or absence of CoREST. Farnesyl pyrophosphate synthase (FPPS, which we wrote about here) is a target for cancer and osteoporosis, and the microbial forms are targets for trypanosomiasis drugs. Human as well as Trypanosoma cruzi and Trypanosoma brucei proteins were screened. Protein tyrosine phosphatase 1B (PTP1B, which we recently wrote about here) is a difficult enzyme with a couple allosteric sites. The C-terminal region is intrinsically disordered, and the protein was screened with or without this region. Finally, human tau is both intrinsically disordered and prone to aggregation. As we noted earlier this year it is of interest due to its potential role in Alzheimer’s disease.
Happily, hits were identified against all the proteins, some with ligand efficiency values above 0.5 kcal/mol per heavy atom. The chemical structures for selected hits are shown, and the researchers do appropriately caution that validating them using orthogonal (non-SPR) methods is essential before further studies.
I do wish the researchers had noted whether shapely hits were enriched or depleted among the confirmed hits. To my eye most seemed fairly flat, and some seemed dubiously PAINS-like, including an eyebrow-raising dinitro-catechol. Nonetheless, the paper is a nice summary of multiple SPR campaigns. If you’re about to embark on one yourself, it is worth carefully perusing.

27 November 2023

Beware of fused tetrahydroquinolines

Practical Fragments has written frequently about pan-assay interference compounds, or PAINS. These molecules contain substructures that frequently show up in hits that tend not to be advanceable, often wasting considerable effort. One criticism of the PAINS concept is that the original definitions were based on a limited number of screens in one assay format. In a new (open-access) J. Med. Chem. paper, Alison Axtman and collaborators at University of North Carolina Chapel Hill, Emory University, and Oxford University characterize one class of PAINS in more detail.
The researchers focused on fused tetrahydroquinolines, or THQs. Of the 51 molecules containing this substructure in the original 2010 PAINS paper, 34 hit in at least one of the assays, and one hit all six. At the time Jonathan Baell and Georgina Holloway noted that “it is not clear for some PAINS, such as the fused tetrahydroquinolines, what the relevant mechanisms of interference may be.” 

The new paper notes that fused THQs are common in screening libraries, with more than 15,000 commercially available. They also frequently show up as hits: the researchers summarize more than two dozen examples against a wide variety of targets including phosphatases, kinases, protein-protein-interactions, and more. In most cases the hits are modestly active, with low to mid micromolar IC50 values, though a few are high nanomolar. 
Promiscuity per se is not necessarily bad. Just last week we noted that the 7-azaindole fragment was the starting point for three approved drugs. However, despite showing up as hits in so many screens, only one peer-reviewed paper reports a crystal structure of a fused THQ bound to a protein, and the researchers note that “no optimized chemical probes or approved drugs contain the chemotype.”
Importantly, fused THQs hit in a variety of different assay formats, including spectrophotometric, chemiluminescent, SPR, and radiochemical. Thus, these are not merely problematic in the AlphaScreen format studied in the original PAINS paper.
So what’s going on? The researchers found that, while molecules containing the fused THQ core were initially colorless, they darkened when dissolved in DMSO or chloroform, turning purple within 72 hours. Interestingly, the reaction seems to be light-dependent: solutions stored in the dark remained colorless. Thin layer chromatography and NMR showed new species forming, and mass spectrometry revealed oxidation with loss of two or four hydrogen atoms. The isolated double bond in the cyclopentene ring seemed to be the culprit, as the saturated analog was stable. Indeed, all of the hits shown in the paper contain the double bond, so fused THQs that lack this feature may be fine – if they ever show up in your assay.
It is still not clear exactly how the decomposition products light up so many assays, but in general it’s a good idea to steer clear of molecules that fall apart in ambient light, unless you’re trying to make a photosensitizer.
The researchers conclude that “it is tragic to continue to watch groups invest time and resources in dead-end hit-to-lead campaigns, and the medicinal chemistry community will benefit everyone if the cycle stops.”
This concludes our public service announcement.

20 November 2023

Capivasertib: the seventh approved fragment-derived drug

On Thursday last week the FDA approved capivasertib for certain breast cancer patients. This marks the seventh fragment-derived drug to be approved. It is also the first approved drug targeting the kinase AKT.
Practical Fragments first wrote about capivasertib, then called AZD5363, way back in 2013, where we described the decade-long odyssey from fragment to drug. Interestingly that fragment, 7-azaindole, was also the starting point for two other approved drugs, pexidartinib and vemurafenib. As we noted at the time, “high-affinity molecules were obtained relatively quickly, but these still required a huge amount of effort to achieve selectivity, oral bioavailability, and other properties.”
What happened next is a poster child to counter one of the false beliefs Christopher Austin noted as being widespread outside industry: “Once an investigational therapy gets into humans for the first time, regulatory approval and marketing are all but assured.”
Capivasertib entered the clinic in 2010 in the first of more than 30 studies listed on ClinicalTrials.gov to date. One challenge was finding patients that would benefit sufficiently to offset a long list of side effects, including diarrhea and glucose fluctuations. In the end, the current approval is in combination with fulvestrant for “adult patients with hormone receptor (HR)-positive, human epidermal growth factor receptor 2 (HER2)-negative locally advanced or metastatic breast cancer with one or more PIK3CA/AKT1/PTEN-alterations, as detected by an FDA-approved test, following progression on at least one endocrine-based regimen in the metastatic setting or recurrence on or within 12 months of completing adjuvant therapy.”
Needless to say, these were not the first patients tested. Use of genetic testing to match patients with a drug likely to help them is not routine even today, let alone in 2010. Managing side effects also required figuring out how much of the drug to dose and how often. But additional combination trials are ongoing. Perhaps, as with venetoclax, capivasertib will eventually prove to be useful for a wider range of patients.
The first marketed fragment-derived drug, vemurafenib, sprinted from program initiation to approval in just six years. Capivasertib took twenty. As we previously noted, success in drug discovery is not necessarily fast or inevitable. Every year more than 40,000 people die of breast cancer in the US alone, but the death rate has slowly been declining. Hopefully the introduction of capivasertib will continue to reduce this.
Congratulations to all the researchers at AstraZeneca, Astex, and the Institute for Cancer Research for participating and persisting in this 20-year marathon to bring a new treatment to people with cancer.

13 November 2023

An update on the COVID Moonshot

On March 18, 2020, a group called the COVID Moonshot released crystal structures of 71 fragments bound to the SARS-CoV2 Mpro protein. The same day, they launched an online crowdsourcing initiative seeking ideas for how to advance these fragments, none of which had activity in an enzymatic assay. The results of this experiment in open science have just been published in Science, appropriately open-access.
Within the first week, the group received more than 2000 submissions. Ultimately more than 20,000 molecules were submitted, and all of these were evaluated in “alchemical free-energy calculations.” These are computationally intensive, requiring ~80 GPU hours per compound, so the consortium used the volunteer-based distributed computing network Folding@home. Compounds were evaluated not just for potency but also synthetic accessibility, and those that passed were synthesized at Enamine and tested in various functional assays.
In addition to accepting submissions for how to advance fragments, a core group of researchers proposed their own ideas. Interestingly, at least in the early stages of the project, this elite group did no better at coming up with more potent or synthetically accessible molecules, despite being intimately involved with the project. This finding validates the open-sourcing of ideas from the larger scientific community.
Ultimately more than 2400 compounds were synthesized, and more than 500 crystal structures were determined. All experimental results were posted online to help guide the synthesis of additional compounds. Speed was consistently prioritized, not just with high-throughput crystallography but also high-throughput chemistry and "direct-to-biology" screening of crude reaction mixtures.
The paper highlights one lead series, which originated from a community submission (TRY-UNI-714a760b-6, itself fragment-sized) inspired by merging overlapping fragments. This mid micromolar inhibitor was ultimately optimized to MAT-POS-e194df51-1, with mid-nanomolar activity in both biochemical and cell assays. (Despite a chloroacetamide in one of the original fragments and a nitrile in the final molecule, which is the warhead found in the approved covalent Mpro inhibitor nirmatrelvir, MAT-POS-e194df51-1 is non-covalent.) 

The molecule is potent against known SARS-CoV-2 variants, including recent ones such as Omicron. A crystal structure of the final molecule also overlays remarkably well onto the initial fragments.
The paper notes that there is still considerable work to do, particularly optimizing the pharmacokinetics to lower clearance and improve bioavailability. These efforts can take vast sums of time and money, and the lead series has been adopted by the Drugs for Neglected Diseases initiative for further development. Although a handful of drugs are already approved against SARS-CoV-2, there is room for improvement: Derek Lowe posted a vivid personal account of his experience on nirmatrelvir here.
When we wrote about the COVID Moonshot in March of 2020, we correctly predicted that vaccines would be approved before drugs from this effort emerged. Fortunately, our warning that “there will be a SARS-CoV-3” has not proven correct – yet. But open science endeavors such as the COVID Moonshot will help us prepare for this eventuality. We may not have made it to the moon yet, but perhaps we’ve learned how to leave Earth’s orbit.

06 November 2023

Finding weak fragments for membrane proteins with WAC

Last week we wrote about NMR, one of the most popular fragment-finding methods. This week we turn to a less common technique: weak affinity chromatography, or WAC. As we’ve written previously, WAC involves immobilizing a protein of interest in a chromatography column and flowing a ligand-containing solution through the column. If the ligand interacts with the protein, its elution time will be delayed in proportion to its affinity. In a new (open-access) Molecules paper, Claire Demesmay and collaborators at Universite Claude Bernard Lyon and Ecole Supérieure de Biotechnologie de Strasbourg extend the technique to membrane proteins.
Membrane proteins are themselves tricky to study, since removing them from their membranes often denatures them. One trick is to use nanodiscs, which are tiny lipid bilayer islands surrounded by proteins that keep them soluble in water. These scaffolding proteins can also be biotinylated so that the nanondiscs can be attached to streptavidin, which itself can be linked to a surface or matrix. Each nanodisc holds one or at most a few membrane proteins.
When we first wrote about WAC in 2011 the technique used standard HPLC columns, which required non-negligible amounts of protein. Here, the technique has been miniaturized to use glass capillaries with volumes of less than 1 microliter, requiring only a few tens of picomoles of protein. The researchers fill the capillaries with a bio-compatible polymer, functionalize it with streptavidin, and then capture biotinylated nanodiscs containing the membrane protein of interest.
A long-recognized challenge with WAC is nonspecific binding of the fragments to the column or matrix. Here, the researchers chose a filling (or monolith) that is more hydrophilic (for aficionados, they picked poly(DHPMA-co-MBA)) and found it superior to the previous polymer both with regards to capacity and non-specific binding.
Another challenge with WAC is detecting low-affinity binders: because interactions with the protein are weak, the shift in retention time is harder to detect. One solution is to pack more protein in the column, and the researchers develop a clever way of doing this with a “multilayer grafting” approach in which successive injections of streptavidin and nanodiscs more effectively fill the capillary. The combination of a more hydrophilic filling and multilayer grafting increased the column capacity for nanodiscs by three-fold.
The researchers tested their approach on the adenosine-A2A receptor (AA2AR), which has frequently been used as a model GPCR. Two previously reported weak ligands, both with affinities around 0.2 mM, could be detected, and competition with an orthosteric binder revealed that they were binding specifically.
This is a nice, how-to guide for performing WAC on membrane proteins, and the paper includes detailed equations for calculating affinities from differences in retention times. I look forward to seeing the technique used in de novo screens.

30 October 2023

NMR for SAR: All about the ligand

In last week’s post we described a free online tool for predicting bad behavior of compounds in various assays. But as we noted, you often get what you pay for, and computational methods can’t (yet) take the place of experimentation. In a new (open-access) J. Med. Chem. paper, Steven LaPlante and collaborators at NMX and INRS describe a roadmap for discovering, validating, and advancing weak fragments. They call it NMR by SAR
Unlike SAR by NMR, the grand-daddy of fragment-finding techniques which involves protein-detected NMR, NMR for SAR focuses heavily on the ligand. The researchers illustrate the process by finding ligands for the protein HRAS, for which drug discovery has lagged in comparison to its sibling KRAS.
The researchers started by screening the G12V mutant form of HRAS in its inactive (GDP-bound) state. They screened their internal library of 461 fluorinated fragments in pools of 11-15 compounds (each at ~0.24 mM) using 19F NMR. An initial screen at 15 µM protein produced a very low hit rate, so the protein concentration was increased to 50 µM. After deconvolution, two hits confirmed, one of which was NMX-10001.
The affinity of the compound was found to be so low that 1H NMR experiments could not detect binding. Thus, the researchers kept to fluorine NMR to screen for commercial analogs. They used 19F-detected versions of differential line width (DLW) and CPMG experiments to rank affinities, and the latter technique was also used to test for compound aggregation using methodology we highlighted in 2019. Indeed, the researchers have developed multiple tools for detecting aggregators, such as those we wrote about in 2022.
Ligand concentrations were measured by NMR, which sometimes differed from the assumed concentrations. As the researchers note, these differences, which are normally not measured experimentally, can lead to errors in ranking the affinities of compounds. The researchers also examined the 1D spectra of the proteins to assess whether compounds caused dramatic changes via pathological mechanisms, such as precipitation.
The researchers turned to protein-detected 2D NMR for orthogonal validation and to determine the binding sites of their ligands. These experiments revealed that the compounds bind in a shallow pocket that has previously been targeted by several groups (see here for example). Optimization of their initial hit ultimately led to NMX-10095, which binds to the protein with low double digit micromolar affinity. This compound also blocked SOS-mediated nucleotide exchange and was cytotoxic, albeit at high concentrations.

I do wish the researchers had measured the affinity of their molecules towards other RAS isoforms as this binding pocket is conserved, and inhibiting all RAS activity in cells is generally toxic. Moreover, the best compound is reminiscent of a series reported by Steve Fesik back in 2012.
But this specific example is less important than the clear description of an NMR-heavy assay cascade that weeds out artifacts in the quest for true binders. The strategy is reminiscent of the “validation cross” we mentioned back in 2016. Perhaps someday computational methods will advance to the point where “wet” experiments become an afterthought. But in the meantime, this paper provides a nice set of tools to find and rigorously validate even weak binders.

23 October 2023

A Liability Predictor for avoiding artifacts?

False positives and artifacts are a constant source of irritation – and worse – in compound screening. We’ve written frequently about small molecule aggregation as well as generically reactive molecules that repeatedly come up as screening hits. It is possible to weed these out experimentally, but this can entail considerable effort, and for particularly difficult targets, false positives may dominate. Indeed, there may be no true hits at all, as we noted in this account of a five-year and ultimately fruitless hunt for prion protein binders.
A computational screen to rapidly assess small molecule hits as possible artifacts would be nice, and in fact several have been developed. Among the most popular are computational filters for pan-assay interference compounds, or PAINs. However, as Pete Kenny and others have pointed out, these were developed using data from a limited number of screens in one particular assay format. Now Alexander Tropsha and collaborators at University of North Carolina Chapel Hill and the National Center for Advancing Translational Science (NCATS) at the NIH have provided a broader resource in a new J. Med. Chem. paper.
The researchers experimentally screened around 5000 compounds, taken from the NCATS Pharmacologically Active Chemical Toolbox, in four different assays: a fluorescence-based thiol reactivity assay, an assay for redox activity, a firefly luciferase (FLuc) assay, and a nanoluciferase (NLux) assay. The latter two assays are commonly used in cell-based screens to measure gene transcription. The thiol reactivity assay yielded around 1000 interfering compounds, while the other three assays each produced from 97 to 142. Interestingly, there was little overlap among the problematic compounds.
These data were used to develop quantitative structure-interference relationship (QSIR) models. The NCATS library of nearly 64,000 compounds was virtually screened, and around 200 compounds were tested experimentally for interference in the four assays, with around half predicted to interfere and the other half predicted not to interfere. The researchers had also previously built a computational model to predict aggregation, and this – along with the four models discussed here – have been combined into a free web-based “Liability Predictor.”
So how well does it work? The researchers calculated the sensitivity, specificity, and balanced accuracy for each of the models and state that “they can detect around 55%-80% of interfering compounds.”
This sounded encouraging, so naturally I took it for a spin. Unfortunately, my mileage varied. Or, to pile on the metaphors, lots of wolves successfully passed themselves off as sheep. Iniparib was recognized correctly as a possible thiol interference compound. On the other hand, the known redox cycler toxoflavin was predicted not to be a redox cycler – with 97.12% confidence. Similarly, curcumin, which can form adducts with thiols as well as aggregate and redox cycle, was pronounced innocent. Quercetin was recognized as possibly thiol-reactive, but its known propensity to aggregate was not. Weirdly, Walrycin B, which the researchers note interferes with all the assays, got a clean bill of health. Perhaps the online tool is still being optimized.
At this point, perhaps the Liability Predictor is best treated as a cautionary tool: molecules that come up with a warning should be singled out for particular interrogation, but passing does not mean the molecule is innocent. Laudably, the researchers have made all the underlying data and models publicly available for others to build on, and I hope this happens. But for now, it seems that no computational tool can substitute for experimental (in)validation of hits.

16 October 2023

Spacial Scores: new metrics for measuring molecular complexity

Molecular complexity is one of the theoretical underpinnings for fragment-based drug discovery. Mike Hann and colleagues proposed two decades ago that very simple molecules may not have enough features to bind tightly to any proteins, whereas highly functionalized molecules may have extraneous spinach that keeps them from binding to any proteins. Fragments, being small and thus less complex, are in a sweet spot: just complex enough.
But what does it mean for one molecule to be more complex than another? Most chemists would agree that pyridine is more complex than methane, but is it more complex than benzene? To decide, you need a numerical metric, and there are plenty to choose from. The problem, as we discussed in 2017, is that they don’t correlate with one another, so it is not clear which one(s) to choose. In a new (open access) J. Med. Chem. paper, Adrian Krzyzanowski, Herbert Waldmann and colleagues at the Max Planck Institute Dortmund have provided another. (Derek Lowe also recently covered this paper.)
The researchers propose the Spacial Score, or SPS. This is calculated based on four molecular parameters for each atom in a given molecule. The term h is dependent on atom hybridization: 1 for sp-, 2 for sp2-, 3 for sp3-hybrized atoms, and 4 for all others. Stereogenic centers are assigned an s value of 2, while all other atoms are assigned a value of 1. Atoms that are part of non-aromatic rings are also assigned an r value of 2; those that are part of an aromatic ring or linear chain are set to 1. Finally, the n score is set to the number of heavy-atom neighbors.
For each atom in a molecule, h is multiplied by s, r, and n2. The SPS is calculated by summing the individual scores for all the atoms in a molecule. Because there is no upper limit, and because it is nice to be able to compare molecules of the same size, the researchers also define the nSPS, or normalized SPS, which is simply the SPS divided by the number of non-hydrogen atoms in the molecule. Although SPS can be calculated manually, the process is tedious and the researchers have kindly provided code to automate the process. Having defined SPS, the researchers compare it to other molecular complexity metrics, including the simple fraction  of sp3 carbons in a molecule, Fsp3, which we wrote about in 2009. 
The researchers next calculated nSPS for four sets of molecules including drugs, a screening library from Enamine, natural products, and so-called “dark chemical matter,” library compounds that have not hit in numerous screens. The results are equivocal. For example, the nSPS for dark chemical matter is very similar to that for drugs. On the other hand, natural products tend to have higher nSPS scores than drugs, as expected. Interestingly, the average nSPS score for compounds in the GDB-17 database, consisting of theoretical molecules having up to 17 atoms, is also quite high.
The researchers assessed whether nSPS correlated with biological properties, and found that compounds with lower nSPS tended to have lower potencies against fewer proteins, as predicted by theory. That said, this analysis was based on binning compounds into a small number of categories, and as Pete Kenny has repeatedly warned, this can lead to spurious trends.
The same issue of J. Med. Chem. carries an analysis of the paper by Tudor Oprea and Cristian Bologa, both at University of New Mexico. This contextualizes the work and confirms that drugs do not seem to be getting more complex over time, as measured by nSPS. This may seem odd, though Oprea and Cristian note that by “normalizing” for size, nSPS misses the increasing molecular weight of drugs.
This observation also raises other questions, such as the fact that SPS explicitly excludes element identity. Coming back to benzene and pyridine, both have identical SPS and nSPS, which does not seem chemically intuitive. One could quibble more: why square the value of n in the calculation of SPS? Why allow s to be only 1 and 2, as opposed to 1 and 5?
In the end I did enjoy reading this paper, and I do think having some metric of molecular complexity might be valuable. I’m just not sure where SPS will fit in with all the existing and conflicting metrics, and how such metrics can lead to practical applications.

09 October 2023

Fragments finger the BPTF PHD Finger

Plant homeodomain (PHD) fingers, despite their name, are found in nearly 300 human proteins. They are small (50-80 amino acid) domains that typically recognize post-translational modifications such as trimethylated lysine residues in histones. The PHD finger in BPTF is implicated in certain types of acute myeloid leukemia. However, because of the large number of PHD fingers as well as their small binding sites, few attempts have been made to develop corresponding chemical probes. (Indeed, the only mention of them on Practical Fragments was in 2014.) In a just-published ACS Med. Chem. Lett. paper, William Pomerantz and collaborators at University of Minnesota and St. Jude Children’s Research Hospital report the first steps.
The researchers started by screening a library of 1056 fragments (from Life Chemicals) against the BPTF PHD finger using ligand-observed (1H CPMG) NMR. Fragments were at 100 µM in pools of up to five. This gave a preliminary hit rate of 5.7%, but only ten compounds (<1%) reproduced when compounds were repurchased and retested individually.
These ten fragments were next tested by SPR (at 400 µM), which confirmed six of them. Also, all ten CPMG hits were tested in an AlphaScreen assay in which they competed with a known peptide binder. This confirmed nine, including the six that confirmed by SPR.
Interestingly, the most potent fragment in the AlphaScreen assay was the starting point for the KRAS inhibitor we highlighted last year. However, this fragment did not show binding to the BPTF PHD finger by SPR, and the researchers had identified the 2-aminothophene substructure as a hit against an unrelated protein. Whether this fragment is privileged or pathological may be context dependent.
This and the top three fragments that confirmed in all assays were used as starting points for SAR by catalog, and a handful of analogs were purchased. The researchers also resynthesized two of the compounds. Oddly, resynthesized F2 turned out to be three-fold more active in the AlphaScreen assay than the commercial material. One analog, compound F2.7, showed mid-micromolar activity.

Docking and two-dimensional protein-observed (1H,15N HSQC) NMR experiments suggest that most of the fragments bind in the “aromatic cage” which normally recognizes methylated lysine residues, but F2 may bind in an adjacent region. Both subpockets were also identified as being ligandable using the program FTMap.

This paper is a nice example of using orthogonal methods to find and carefully validate fragments against an underexplored class of targets. The researchers conclude by stating that “these hits are suitable for further SAR optimization and development into future methyl lysine reader chemical probes.” I look forward to seeing more publications.

02 October 2023

Discovery on Target 2023

Last week CHI’s Discovery on Target was held in Boston. This was the Twentieth Anniversary edition, though oddly last year also claimed to be the twentieth. Regardless, attendance surpassed pre-pandemic levels, with some 1200 attendees, 90% of them in person. Eight or nine concurrent tracks over the course of three days competed with one another, while a couple pre-conference symposia and a handful of short courses were held before the main event. Outside obligations kept me from seeing many talks, including plenary keynotes by Jay Bradner (Novartis), Anne Carpenter, and Shantanu Singh (both at the Broad Institute), but most of these were recorded and will be made available for a year, and I look forward to watching them. Here I’ll just touch on a few of the fragment-relevant talks I was able to attend.
“Protein Degraders and Molecular Glues” was a popular track during all three days of the main conference, and in a featured presentation Steve Fesik (Vanderbilt) described how he is using NMR-based FBLD to identify tissue-specific E3 ligases and β-catenin degraders. In the case of β-catenin, a difficult oncology target, a fragment screen identified a 500 µM hit that was optimized to 10-20 nM. This has no functional activity on its own, but combining it with a ligand for an E3 ligase to generate a bivalent PROTAC causes degradation of the protein. Steve is currently optimizing the pharmaceutical properties of these molecules.
One exciting application for PROTACs is tissue-specific targeted protein degradation, which could avoid systemic toxicity for proteins such as Bcl-xL. Steve said that for the past five years he has been pursuing ligands against E3 ligases preferentially expressed in certain tissues, and he presented brief vignettes for three of them. These came from an initial list of 20 E3 targets, but many of them turned out to be too difficult to express.
Steve typically screens a library of nearly 14,000 fragments, large according to our recent poll, but this has proven fruitful as only about 10% of proteins he has screened have turned out to be “teflon.” He noted the odd little fragment hit that proved so impactful to the KRAS program we highlighted last year as being something that might have been excluded from a smaller library.
We wrote last week about ligands for the E3 ligase DCAF1, and Rima Al-Awar (Ontario Institute for Cancer Research) described another series. She also described ligands against the oncology target WDR5, a target Steve Fesik has pursued as well.
Continuing the theme of targeted protein degradation, Jing Liu described Cullgen’s discovery of fragment-sized ligands for a broadly-expressed E3 ligase which could be an alternative to CRBN-targeting ligands when resistance (inevitably) arises. Although he did not specify the E3 ligase, Cullgen has filed a patent application for ligands targeting DCAF1.
Rounding out targeted protein degradation, Kevin Webster, my colleague at Frontier Medicines, described the discovery of covalent ligands for the E3 ligase DCAF2 (or DTL) using chemoproteomics and a variety of other techniques including cryo-electron microscopy. Consistent with Steve’s comments, considerable effort went into successfully obtaining a soluble, well-behaved protein.
The late Nobel laureate Sydney Brenner said that “progress in science depends on new techniques, new discoveries, and new ideas, probably in that order.” Harvard’s Steve Gygi, one of Frontier’s Scientific Advisory Board members, described multiple new techniques in a featured presentation focused on cysteine-based profiling. These included multiplexed methods to more rapidly find covalent ligands for targets across the proteome. A just-released mass spectrometry instrument made by Thermo Fisher called the Astral further accelerates the process with order-of-magnitude improvements in both speed and sensitivity compared to existing machines.
The cell-based covalent screening described by Steve Gygi is very powerful, but so is investigating a single protein, as demonstrated by the discovery of sotorasib. AstraZeneca did early work on covalent screening (which Teddy noted in 2015), and they have continued to build their platform, as described by Simon Lucas. The company has around 12,000 covalent fragments, some beyond the rule of three, with molecular weights between 200 and 400 Da and logP between 0 and 4. More than 90% are acrylamides, a clinically validated warhead, and the researchers are careful to avoid particularly reactive molecules that would be non-specific.
In contrast to the electrophilic fragments that comprise most covalent libraries, Megan Matthews (University of Pennsylvania) is exploring nucleophilic fragments for “reverse polarity activity-based protein profiling,” as we highlighted last year. This has led to the discovery of unusual post-translational modifications. For example, the sequence of the protein SCRN3 suggests that it should be a cysteine hydrolase, but the purified protein has no cysteine hydrolase activity, and in cells the N-terminal cysteine is processed to form a glyoxylyl moiety.
Finally, Alex Shaginian provided an overview of DNA-encoded library screening (DEL) at HitGen. The company currently has 1.2 trillion compounds spread across more than 1500 libraries, and an obvious question is whether this is overkill. Alex noted that one protein has been screened three times over the course of several years. In the original screen, a modest (30 µM) hit was found from 4.2 billion compounds screened. A later screen of 130 billion compounds produced nothing new, but a more recent screen of 1 trillion compounds led to four mid-nanomolar series. As Steve Fesik noted, screening larger libraries, whether experimentally or computationally, really can be helpful, especially for the hardest targets.
Despite only attending half the conference this post is getting long, but for those of you who were there, which talks would you recommend watching?

25 September 2023

Fragments vs DCAF1: a new tool for targeted protein degradation

Targeted protein degradation (TPD) goes beyond merely inhibiting a protein; it takes a protein out of commission entirely. This is frequently done using a bivalent ligand: one part binds to the protein of interest, while the other part binds to an E3 ligase, which ubiquitinates the protein of interest, targeting it for destruction in the proteasome. Human cells have hundreds of E3 ligase proteins, some of which may work better in certain situations, such as specific cell compartments or tissues. In a recent ACS Med. Chem. Lett. paper, Anna Vulpetti and colleagues at Novartis describe progress against DCAF1.
DCAF1 is one component of the Cullin4-RING E3 ubiquitin ligase complex. The C-terminus of the protein contains a WD40 repeat (WDR) domain, which in this case consists of seven “blades” arranged around a central cavity, or “donut hole”. WDR domains are relatively common, and indeed we wrote about a previous Novartis effort that identified chemical probes against another WDR domain in the protein EED. In the new work, the researchers took 21 EED binders and screened them using both protein-detected and ligand-detected NMR against DCAF1, identifying two hits. Crystallography revealed that compound 1 binds in the central cavity, which previous computational screening had suggested would be ligandable
Next, the researchers screened 30 related compounds from within Novartis. Two of them, including compound 4, had improved affinity (as assessed both by NMR and SPR) and could be characterized crystallographically. In addition to binding in the central cavity, these compounds also bound to a site in the blade region, which the researchers wanted to avoid. Adding a piperazine to compound 4 both improved affinity and disrupted binding to the blade region; further optimization and growing to better fill the central cavity led to compound 13, the most potent molecule in the paper.
A crystal structure of a closely related molecule reveals that the acetyl group is near the entrance to the donut hole, providing an easy synthetic attachment point to construct bivalent degraders. A separately published preprint revealed this to be successful, with degraders of BRD9, multiple tyrosine kinases, and BTK.
There are several takeaways from this nice fragment to lead story. First, despite the fact that compound 1 is clearly fragment-sized (albeit a bit too lipophilic to be fully rule-of-three compliant), the word fragment never appears in the article. FBLD has become so routine that researchers may not even mention it, which does mean that our list of fragment-derived drugs is destined to be incomplete.
Second, although DCAF1 and EED share less than 25% sequence similarity, screening EED hits turned out to be successful, which could argue for screening specific subsets of fragments (for example kinase-focused or, in this case, WDR-focused). On the other hand, compound 1 binds in a different manner to DCAF1 than it does to EED. Indeed, compound 1 actually binds in two different orientations to DCAF1, consistent with its low affinity. The researchers mention a paper published earlier this year that reports a successful DEL screen against the target. Perhaps DCAF1 is just very ligandable, and a naïve fragment screen would have worked just as well as the pre-selected set.
Finally, the fact that this program yielded bivalent degraders suggests that many E3 ligases might be coopted for drug discovery. The field of targeted protein degradation is just getting started.

18 September 2023

Fragments vs hIL-1β: Growing into a cryptic pocket to inhibit a protein-protein interaction

Protein-protein interactions have a well-deserved reputation for being difficult to drug with small molecules. This is particularly true for cytokine-receptor pairs, which are involved in a host of extracellular signaling functions. Human interleukin-1β (hIL-1β) plays a key role in inflammation by binding to its receptor IL-1R1. Biologics such as anakinra and canakinumab have been approved as drugs, but apart from some very low affinity fragments no small molecule inhibitors are known. In a new (open access) Nat. Commun. paper, Frédéric Bornancin, and collaborators at Novartis and University of Leicester report the first.
The researchers started by screening the 3452-compound LEF4000 library, which we described here, using 19F-NMR. After confirmation using protein-observed 2D NMR just a single super-sized fragment hit remained, consistent with the difficulty of the target. The individual enantiomers of this racemic compound were studied, and only (S)-1 was found to be active. Further characterization revealed that, despite weak affinity, this compound had both slow association and dissociation rates. More on that below.
Fragment growing in multiple directions led to mid-micromolar compounds such as 11 and 12. Combining elements from these molecules ultimately led to compound (S)-2, with low micromolar affinity as assessed by SPR
Compound (S)-2 specifically blocked the binding of hIL-1β with its receptor IL-1R1, but did not inhibit the binding of the related cytokine hIL-1α to IL-1R1. Even better, the compound blocked IL-1R-mediated signaling in cells at low micromolar concentrations in two different assays. The similar activity in biochemical and cell assays is likely due to the fact that the compound only needs to act at the cell surface, so permeability is not an issue, in contrast to our post last week.
A crystal structure of (S)-2 bound to hIL-1β revealed important interactions between the protein and both the phenol and lactam nitrogen, two contacts that were maintained during fragment optimization. The structure explains why only the (S)-enantiomer is active, as maintaining these contacts would cause clashes for the other enantiomer.
The structure also explains the mechanism of inhibition. (S)-2 binds to a cryptic pocket that forms in a region of hIL-1β important for interacting with IL-1R1, and formation of the pocket involves a loop movement that would be incompatible with the protein-protein interaction. The researchers argue convincingly that that the compound stabilizes the cryptic pocket, which naturally exists as a minor population within solution. This also explains the slow kinetics, which would be expected if the compound essentially has to wait until the cryptic pocket opens before it can bind.
There is still a long way to go to a drug. Not only is the affinity of (S)-2 modest, the two carboxylic acid moieties and the phenol are likely to impede oral bioavailability. Nonetheless, this is a lovely paper, and the researchers point out that cryptic pockets frequently involve “large movements of secondary structural elements” that could block biological function. Indeed, this is the case for approved drugs such as sotorasib. Don’t give up just because your protein of interest appears like a featureless billiard ball: there may well be opportunities hidden just beneath the surface.

11 September 2023

Fragments vs malarial DHFR

Malaria continues to be a worldwide scourge, with some quarter billion cases last year. A seventy-year-old drug called pyrimethamine targets the dihydrofolate reductase (DHFR) enzyme from Plasmodium falciparum, but resistance mutations have rendered this molecule mostly useless. An analog called P218 was developed to overcome this resistance and completed a handful of phase 1 clinical trials, but unfortunately the human pharmacokinetics were found lacking. In a new RSC Med. Chem. paper, Marie Hoarau and colleagues at the National Center for Genetic Engineering and Biotechnology in Thailand describe their efforts to improve this molecule.
The researchers recognized that the phenyl propanoate moiety of P218 was a metabolic liability and sought a replacement. They screened a library of 1163 fragments (from Key Organics) at 1 mM using a thermal shift assay. This resulted in 64 hits, 52 of which confirmed by SPR. Of these, 22 showed some level of inhibition at 0.5 mM against mutant PfDHFR.
Among the hits, five were “bi-aromatic carboxylates,” such as compound 136. These were prioritized because, while reminiscent of the phenyl propanoate in P218, they had fewer rotatable bonds. Some of them also showed slow off-rates by SPR, though in my opinion the sensorgrams look suspicious, perhaps due to excessive protein loading on the chip. (For example, the Kd for compound 136 calculated from the on and off rates comes in at 160 nM, unrealistically potent given that it shows only 20% enzymatic inhibition at 0.5 mM. Note – all values here and in the figure are for the mutant form of the enzyme.)

SAR by catalog was used to find additional analogs, such as compound AF10, which showed measurable inhibition of the enzyme. Next, the researchers tested hits in the presence of a pyrimidine fragment (L4) derived from P218, known to bind nearby. Compound AF10 showed greater inhibition than would be expected by simple additivity, perhaps suggesting some preorganization of the binding site, as in a different example discussed here.
Molecular modeling was used to link the carboxylate fragments with L4, and eight were made and tested. All inhibited both wild type and mutant PfDHFR, and compound 8 showed good selectivity over human DHFR too. A crystal structure confirmed that it bound as predicted. From a fragment-linking perspective, the sub-nanomolar affinity of compound 8 is impressively better than would be expected given the weak affinities of L4 and AF10.
Unfortunately, despite similar in vitro potency against the isolated enzymes, compound 8 and the other molecules tested showed “disappointing” activity against Plasmodium falciparum carrying either wild-type or mutant DHFR, roughly 100- to 1000-fold less potent than P218. The researchers suggest solubility may be a factor.
This paper is a useful reminder of the dramatic disconnects often seen between enzymatic and cell activity. Nonetheless, it is another good example of using fragment-based methods to replace one portion of an existing molecule.

04 September 2023

Fragment screening on a benchtop NMR

Practical Fragments has been on an NMR theme for the last two weeks, and this post continues that trend. One of the main barriers to entry for NMR methods is the instrument itself: not only are the machines large, requiring a good size room, the price starts at several hundred thousand dollars. Then there is the maintenance, which includes regular refills of liquid helium, which is both costly and often scarce. And if the helium runs out, your precious superconducting magnet “quenches”, which looks like this.
Large magnets such as those in 600 MHz instruments are unlikely to change until room temperature superconductors become a reality. Less powerful permanent magnets are available though, and you can purchase a benchtop 80 MHz machine for less than $100,000. But the low sensitivity requires very high concentrations of sample, too high for fragment screening. Unless, that is, you could increase the sensitivity. This has now been described in a new (open-access) Angew. Chem. Int. Ed. paper by Felix Torres, Roland Riek, and collaborators at the Swiss Federal Institute of Technology, Bruker, and NexMR.
The somewhat complicated method is called photochemically induced dynamic nuclear polarization (photo-CIDNP), which we wrote about in June. As the name suggests, this involves light excitation of a photosensitizer molecule which can then increase sensitivity for detecting other small molecules, particularly when they are not bound to proteins. Weirdly and fortuitously, photo-CIDNP theory predicts that polarization transfer is actually higher at lower magnetic fields, making it ideal for benchtop NMR.
The researchers first tested three fragments, each at 500 µM, using 25 µM fluorescein as the photosensitizer. Just 3 minutes of measurements each gave very clear spectra after light irradiation at 450 nm. In the absence of light it would take between 22 hours and 10 years to achieve comparable signal-to-noise enhancement.
Next, the researchers screened 32 fragments from their custom-designed "NMhare1.0 library” we previously described, which contains molecules suitable for photo-CIDNP. As before they used the protein PIN1 (at 10 µM) and collected data for 3 minutes per sample. Six compounds had reduced polarization in the presence of protein, four of which had been previously detected as binders and validated using a 600 MHz NMR. Of the two new hits, one confirmed using protein-detected NMR while the other did not.
To explore the limits of sensitivity, the researchers conducted a series of experiments lowering the concentrations of protein and small molecules. One of the compounds could be detected at concentrations as low as 250 nM and quantified at 1 µM in just 3 minutes. At 50 µM this compound clearly showed binding to 5 µM protein, despite having an affinity in the low millimolar range.
This is a fun paper, and I particularly like the fact that it expands fragment screening to an instrument previously not thought to be suitable. As we wrote previously, one limitation of photo-CIDNP is that only some molecules are able to be photo-sensitized. A solution would be to find one such ligand and then run a displacement assay to see whether a second ligand could compete with it, akin to what has been done for fluorine NMR. I look forward to seeing how this technique develops.

28 August 2023

Affinity measurements in a single NMR tube?

Last week we highlighted a ligand-detected NMR method to measure affinities of protein-ligand interactions. That technique, R2KD, requires preparing multiple NMR samples with the ligand at different concentrations. In a new open-access paper published in J. Am. Chem. Soc., Serena Monaco and collaborators at University of East Anglia and Universidad de Sevilla describe a method that can be done in a single NMR tube.
The researchers have actually combined two methods, chemical shift imaging (CSI) and Saturation Transfer Difference (STD) NMR, to create imaging STD NMR. We’ve written previously about STD NMR, which relies on the transfer of magnetization from an irradiated protein to a bound ligand. In CSI, chemical shift information is recorded at multiple slices along the length of an NMR tube. Normally the solution in an NMR tube is homogenous and so the chemical shifts would be identical at the bottom and top of the NMR tube. Here, though, the researchers create concentration gradients by carefully pipetting a solution containing ligand on top of a solution containing protein and allowing the ligand to diffuse the length of the NMR tube.
Like all things NMR-related, the mathematics get a bit complicated. One important factor is the rate of diffusion for a given small molecule. This “diffusion coefficient” can be experimentally measured by creating a concentration gradient in the absence of protein and measuring the ligand concentration at various positions in an NMR tube after a given length of time (typically more than 12 hours). Diffusion is dependent on molecular weight, so it is also possible to calculate the diffusion coefficient, and in fact the researchers found that the calculated values matched the experimental values for three different small molecules.
Knowing the diffusion coefficient helps establish the maximum ligand concentration to use and the ideal diffusion time. The researchers examined three different protein-ligand pairs, all of which had weak affinities, with KD values from 0.2 to 2 mM. Measuring STD signals at different slices along the NMR tube effectively yields STD signals at different concentrations of ligand, and fitting this to an equation allows calculation of the dissociation constant. For the three model systems the affinities agreed with literature values, which had been determined using ITC or WAC.
One nice feature of imaging STD NMR is that it can identify non-specific binding. This is because STD signals vary depending in part on how close a proton on the ligand is to the protein, resulting in different STD signals for different protons for specific binders. If this “epitope pattern” is lost at higher concentrations, this suggests non-specific binding, where the ligand can bind in random orientations to multiple sites on the protein. The researchers demonstrated this for one of their model systems: tryptophan binds specifically to bovine serum albumin with a dissociation constant of 0.2 mM, but above 1 mM or so the epitope disappears, suggesting non-specific binding.
Imaging STD NMR does have some limitations. For one thing, it requires a high initial concentration of ligand: 30 mM in the case of tryptophan, and even higher for the other two ligands. Most small molecules are nowhere near this soluble in water. The researchers suggest that ligands could be dissolved in DMSO and placed on the bottom of the NMR tube, with the protein solution gently layered on top. They show that the concentration gradients develop in a similar manner as a fully aqueous system, but acknowledge that high DMSO concentrations may not play well with most proteins.
Also not stated is the sensitivity of the method for higher affinity binders. Last week’s R2KD could measure affinities as tight as 10 µM, but it is unclear how much below 200 µM imaging STD NMR can go.
Finally, as we noted in 2019, STD effects are remarkably complex and not well-correlated with affinity. In particular, binding kinetics can play a role in the strength of the signal. It would have been nice to see more than three protein-ligand pairs tested.
All that said, this is an intriguing approach. Laudably, the researchers provide extensive supporting information, including mathematical derivation of the fitting equations, a spreadsheet, NMR pulse sequences, and macros. I’ll be curious to see how it works for others.

21 August 2023

Ligand-observed NMR – quantitatively

Ligand-observed NMR is one of the most popular fragment-finding methods. Among its strengths is the ability to find extraordinarily weak fragments that most other techniques would miss. However, common ligand-observed NMR methods such as STD are not quantitative: they can tell you that a fragment binds, but not how tightly. In a new open-access J. Med. Chem. paper Manjuan Liu and colleagues at the Institute of Cancer Research provide an easy solution.
The approach is based on an NMR phenomenon called transverse relaxation (see here), which describes how atomic nuclei return to their ground state after being excited by a radiofrequency pulse in a magnetic field. The transverse relaxation rate R2 for a given nucleus depends on the tumbling speed of the molecule in which it is contained: small molecules tumble rapidly and have small R2 values, while larger molecules tumble slowly and have larger R2 values. When a small molecule binds to a protein its tumbling speed slows and its R2 increases. The R2 values can be measured experimentally using a Carr-Purcell-Meiboom-Gill (CPMG) pulse sequence.
This is all fairly standard for NMR spectroscopists, and in fact CPMG is widely used to find fragments. Liu and colleagues proposed that, by measuring the change in R2 with changing concentrations of small molecule, they would be able to extract the dissociation constant (Kd). The theory gets a little hairy (14 equations), and the analysis depends on non-linear regression curve fitting, but this is easily done using modern analytical software. The technique is called R2KD.
The experiment itself is straightforward. The ligand alone is prepared at two different concentrations; these are used to determine the R2 values of the free ligand. Another eight samples contain protein and various concentrations of ligand. The R2 values are measured and fit to an equation to extract the dissociation constant. An initial test case with a known 50 µM ligand for the protein BCL6 was encouraging, giving a Kd of 53 to 78 µM for four different protons on the ligand. The accuracy could be further improved by using a “global fit” with all the data rather than analyzing each NMR peak in isolation.
Next, seven ligands against three proteins were analyzed using R2KD and compared with their literature values. Here too the results were in agreement, mostly within a factor of two. The lower limit for sensitivity is dependent on the NMR signal for the ligand; below concentrations of about 20 µM the experiments become impractically long. The upper limit is dictated by the solubility of the ligand. The researchers could reliably measure dissociation constants around 1 mM and suggested that with a sufficiently soluble ligand even weaker ligands could be measured.
The R2KD experiment requires that the protein concentration be less than about 20% of the lowest ligand concentration. (That said, protein concentrations up to 35 µM gave reasonable results.) Preserving protein is usually a goal, so lower concentrations (single-digit micromolar) are desirable from both a practical and theoretical standpoint.
Finally, the researchers demonstrated the application of R2KD to assess 10 fragment hits from a 1000-compound screen against the E3 ligase complex CRBN/DDB1, one of the most popular targets for PROTACs. The hits had dissociation constants ranging from 70 to 1200 µM, and the R2KD values were similar to those found in a fluorescence polarization (FP) assay, though for the most part the affinities from R2KD were higher. In particular, two compounds with essentially no activity in the biochemical assay came in at sub-millimolar by R2KD, which may speak to the insensitivity of the FP assay.
Overall this is a lovely and, as befits this blog, practical paper, and I hope R2KD becomes widely adopted. With a sweet spot for Kd values of 10-1000 µM the technique fills an important niche: biochemical assays are well-suited for tighter binders but less reliable at millimolar ligand concentrations. As crystallography becomes increasingly popular as a primary screen, I could imagine R2KD being used to rank the resulting fragment hits.

14 August 2023

Stabilizing protein-protein interactions: part 3 (fragment linking)

Stabilizing protein-protein interactions is becoming increasingly popular, and not just for PROTACs. Nearly three years ago we highlighted the use of crystallographic screening to find fragments that could stabilize interactions between the adapter protein 14-3-3δ and peptides derived from p53, a prominent cancer target. After noting how much work lay ahead, we ended the post with, “expect a part 3!” This has now been published (open access) in Angew. Chem. by Adam Renslo, Luc Brunsveld, Michelle Arkin, Christian Ottmann, and collaborators at UCSF and Eindhoven University of Technology.
In addition to crystallographic fragment screening, the researchers had previously performed a disulfide Tethering screen on the 14-3-3δ protein, which we described here. The fragments from the two screens bound next to one another, so the researchers decided to link them. They started by solving the crystal structure of compound 1 disulfide-bonded to 14-3-3δ in the presence of fragments from the crystallographic screen as well as a peptide derived from estrogen receptor alpha (ERα, another anti-cancer target). These co-structures guided the synthesis of new linked molecules, and these were soaked into crystals of 14-3-3δ and the ERα peptide. Compound 6 gave strong electron density and overlayed nicely on the initial fragments.
To determine whether the linked molecule could stabilize the 14-3-3δ/ERα complex, the researchers developed a fluorescence anisotropy assay with a dye-labeled peptide from ERα. Some of the linked molecules produced an increase in anisotropy, suggesting stabilization of the 14-3-3δ/ERα complex, but when the researchers ran the important control of repeating the experiment in the absence of 14-3-3δ they found that several molecules still increased anisotropy, which could be due to aggregation. (Adam published a nice early paper on aggregation and is thus particularly attuned to the dangers.)
Fortunately, some of the molecules passed this control, and with a robust crystallography system the researchers were able to use structure-based design to improve them, ultimately arriving at compound 24, which increased the affinity of the 14-3-3δ/ERα complex by 25-fold. It was also quite specific towards ERα, and did not increase the affinity of nine peptides from from other proteins for 14-3-3δ. The researchers attribute this selectivity to the fact that most other peptides would sterically clash with compound 24. (Not reported was the peptide from p53, which would be interesting.)
This is a nice paper on several levels. In addition to selectively stabilizing a therapeutically relevant protein-protein interaction, this is a rare example of starting with a covalent fragment and developing a non-covalent binder. (For another, see here.) Also, this is a good example of fragment linking, which is often challenging.
There is still a long way to go. The most potent molecules all contain amidine moieties, whose high polarity is a liability for cell permeability, let alone oral bioavailability. Moreover, the affinity of compound 24 is still quite weak, with a low ligand efficiency.
That said, with a wealth of structural and biological understanding I am optimistic further progress can be made, perhaps by rebuilding the covalent linkage to the protein, as was the case of sotorasib or this more recent paper from the UCSF team. I look forward to part 4!