I attended RSC Fragments 2024 (Hinxton, 4–5 March 2024), a conference dedicated to fragment-based drug discovery. The various talks were really good, because they gave overviews of projects involving teams across long stretches of time. As a result there were no slides discussing wet lab protocol optimisations and not a single Western blot was seen. The focus was primarily either illustrating a discovery platform or recounting a declassified campaign. The latter were interesting, although I’d admit I wish there had been more talk of organic chemistry —there was not a single moan/gloat about a yield. This top-down focus was nice as topics kept overlapping, namely:

• Target choice,
• covalents,
• molecular glues,
• whether to escape Flatland,
• thermodynamics, and
• cryptic pockets

Target identification by druggability

I will start from the start of a drug discovery campaign: target identification. In most talks a slide or two are dedicated to convincing the audience why a given protein is worth targeting. Target identification is traditionally all-too-often done by a clinician or a geneticists, who picks a target based on a cartoon diagram of a pathway, where arrows activate protein by cartoon physics and not thermodynamics. In the academic setting, it all too frequently is a case of stamp-collecting paralogues across a family. A countertrend is making druggability of a protein (ease of chemical matter to bind) take over this step. A few start-ups are know for pushing this. In RSC Fragments there were several cases. Dan Erlanson (Frontier Medicines), presented his “Druggability Atlas”, a map of the covalently druggable proteome derived from a combinatorial scaffold/warhead MS covalent screen of cell lysates. Jonathan Pettinger (GSK) presented his cysteinome with over 200k targets, along with an interesting note that stereoselective interactions (only one isomer was found) give credibility to the binding event. Exscientia is known to have a strong target discovery program, but unfortunately there was a last minute substitution in the schedule.

In the camp of the classical approach, i.e. choosing a target based on solid reasoning, a talk stood out for its precise target and the challenges that presents. Matthew Calabrese (Pfiser) sought an activator for a specific isoform (technically paralogue) of a component of AMPK, the gamma-3 component based on its tissue expression in GTex (an RNA tissue/isoform expression database, which I have found invaluable many times). This resulting in the creation of a lead compound that used a cryptic pocket by breaking a salt bridge stabilised by compensatory interactions in a manner analogues to the phosphorylated state (active) of the protein. These interaction were missing in the other paralogues, thus achieving specificity. This presented issues as the rodent orthologues of this paralogue differed. Drugs of the future will need to be more precise and targetted and whereas some pathways may present an easy target, the majority will require campaigns with such levels of complications.

Covalents

Everyone who works with covalents inexplicably introduces these compounds as pariahs shunned by med chemists —yet, a high fraction of the talks involved them, and, given that covalent warheads come in a range of reactivity, the self-labelled–maverick chemists are not using maleimide or iodoacetamide fragments, but acrylamides, isoxazoles, disulfides and so forth. The poster child for covalents is normal penicillin G, but for Fragments 2024 it was rightfully Sotorasib, a recently FDA-approved KRas G12C inhibitor that is both a product of fragment-based drug-discovery and an acrylamide.

Covalent fragments have the advantage that they stay even after the protein is denatured/proteolyzed, which is good for non-native MS. Covalent drugs have the advantage that they can compete against much larger tight-binding protein partners.

Binding first, bonding second

A very reactive haloacetamide will have lots of off-target effects, hence the favouring of less reactive warheads. However, the moderate reactivity means that k_on/k_off needs to be good in the first place. Dan Erlanson (Frontier Medicines) spoke about kinetics involved, namely binding (inhibition constant, K_I) vs bonding (inactivation constant, K_Inact) and how the former is many magnitudes better. A further interesting observation by him was that a covalent ligand on an E3 ligase would in effect reprogram it to target the (un)desired protein target.  

Parenthetically, In XChem even if the soak is at room temperature there’s generally a relatively lower “hit rate” for covalent libraries. In some cases that is due to a variety of factors (such non-physiological pH, default inactive conformation etc.) causing the catalytic triads to not be reaction-ready (protonated acid and deprotonated nucleophile), but generally the warheads are not aggressive as their name suggests and binding comes first.

Targeting beyond cysteine

A frequent idle discussion point in the conference was covalents targeting beyond cysteine. My PhD was on the catalytic promiscuity in a PLP-dependent enzyme, where the actual star of the show was the cofactor, PLP, which can form Schiff base with amine substates and form stable reversible carbocations adducts, so I was very pleased to see a Schiff base make it to a talk and actually be the case example for a non-cysteine covalent.

Peter Cossar (Eindhoven University of Technology) presented a molecular glue using a halobenzaldehyde (paper), which targeted a lysine in a positively charged region in a hub protein, 14-3-3, at the interface with its many partners. The nice part was that the ligand favoured the binding of one specific partner, Pin1, but not the others thus achieving specificity.

Molecular Glues

This was one of a series of talks and posters about molecular glues, namely compounds that help two protein bind, either by strengthening an existing one or forming a de novo interaction. The superstars of this category are obviously Pʀᴏᴛᴀᴄs, but there are many examples of successful molecular glues, such as hydroxytamoxifen, rapamycin and fusicoccins. Molecular glues and covalents were frequent two overlapping topics (e.g. the aforementioned reprogrammed E3 ligase).

Thermodynamics

Throughout the talks there were several ITC and SPR results. The primary value given is the binding constant, which used to determine if to progress a design or not. However, these give a lot more information. With ITC one can split the Gibbs free energy into enthalpy an entropy. With SPR one can split into the rates ranging from a simple single on rate and single off rate to a complex diagram involving different states. This allows one to describe the binding mode of a ligand: enthalpic binders (interaction driven) or entropic binder (desolvation without rigidification), and whether binding occurs by conformational selection or causes induced fit.

One attendee asked if these descriptors are important in driving lead progression. The ensuing multiway discussion attempted to justify why this interesting information was useful, but tragically fell short. In a previous blog post I discuss the kinetics involved in binding, so I am keen on this. One can certainly drive a car without knowing how it works, but as soon as a fault occurs this quickly breaks down: In my opinion, I believe this to be true here.

A great talk on the physics of binding was by Gerald Platzer (Universität Wien), who showed 15N and 1H NHR chemical shift upon binding. Including C–H··pi dispersion forces yielding a measurable enthalpic contribution (cf paper).

Escape from Flatland

Escape from Flatland is a concept, wherein more 3D molecules are advocated because synthetic drugs are very flat, yet protein pockets are 3D and natural products are very 3D. That is not to say folk want to mimic natural products that look like nightmarish (eg. aconitine) and have consumed whole PhDs for total synthesis. The aim is using more compounds like piperidine, piperazine, morpholine, bicyclopentane, norborane, adamantane, and so forth. I am a big fan of Escape from Flatland, so I was delighted how much attention it got, although it was quite clear there are limits as the results were lukewarm at best.

It should be said that nobody is advocating that drugs be like natural products, which take several synthetic chemistry PhD theses for its total synthesis. Instead what is sought is occasional 3D moieties that can better attain certain enthalpically strong interactions. In the various talks and posters in Fragments 2024, there were several such examples. These were in lead optimisation, wherein functional groups needed to be optimally positioned to engage with tight waters or residues.

I should note that this mimics what is seen in the literature for examples where Escape from Flatland is advocated. Namely, a part of a flat compound is replaced with a 3D part and there is a good improvement. A caveat is that these papers do skip any pharmacokinetics and plants have been trying to kill herbivores for millions of years using high Fsp3 compounds, while the herbivores have been evolving in turn to defend themselves, so 3D compounds often do not fair well in cytochrome assays.

3D compounds often bring with them the headache of chirality. An interesting observation was that the binding of one isomer, but not the other was a helpful trick to help find worthwhile interactions.

Parenthetically, there was a discussion about the libraries stocked at Diamond XChem. It is true that the Edelris Keymical library, as Chris Swain (Cambridge MedChem Consulting) noted is discontinued. However, there are several libraries available that are 3D in nature, such as the York3D library and the EUbOPEN extension to the Enamine DSiPoised library designed by Adam Nelson (paper). In these libraries, the ~200 Da compounds often bind strongly via one part of the molecule and expand outwards, sometimes unfruitfully in the solvent, but other times in worthwhile vectors. Anecdotally in XChem, the 3D fragments, which like most fragments have two or three parts explore novel regions of protein cavities, by binding in a well explored themodynamically favourable pocket, but extending into a novel vector inaccessible via planar fragments. Sometimes the 3D part are themselves constructive in elucidating useful tight-binding chemistry: in one case there was a lovely atypical carboxylic acid bioisostere from the York3D library (mimicking the native substrate, a peptide C-terminus), whereas the DSiPoised failed to provide anything similar.

In Fragments 2024, the jury was out whether the combinatorial explosion was worth it at the fragment screening stage. The vendors had some beautiful compounds on display, but they were for fragment elaboration more so that screening. 

Other

Astex mini-frags

In terms of screening libraries, the Astex minifrags made a few appearances. These are 1M compounds in PBS buffer that effetely make it a mixed solvent crystallisation. They are very small, so there will not be any piggybacking sidegroups with moderate gains. This means that everything that binds is forming tight interactions, but it does also mean more effort is required to grow the hit.

Cryptic pockets

Another topic was cryptic pockets. A common question asked by students is why do a fragment screen, why just do docking or given a static snapshot of a protein drawn in a compound. It is a great question as the answer is an ad hoc lecture in molecular mechanics, with a potential digression into hobby horse dressage or spherical cows. Docking does not have great accuracy as it is a 3D snapshot of a 4D event with a flexible protein and an interplay of orbitals (e.g. π interactions are dispersion forces driven by the quadrupole moment of the π system). In a fragment screen pockets do occasionally more and open up cryptic pockets, ie. sidechain are re-orientated in a stable conformation, which differed from the native crystal form (which is is nice but not alway true, cf James Fraser and Mark Murcko paper).

Fragments, best strategy? (Yes)

The conference was dedicated to fragment based drug discovery and there was an overview on the success of the method by Dan Erlanson (cf his blog post), and an analysis by Chris Swaine. The former pointed out that many FDA drugs are in effect fragments, so counting these numbers could be larger depending on definitions.

The consensus was that fragment based drug discovery has shown its potential and actually comparison with historical data is unfair. Current targets are harder as easy hanging fruits have already been picked, therefore biochemically tricky. Natural products from traditional medicine may have had its heyday in the past, with penicillin G, aspirin, taxol and so forth, but have reached their limits after all natural products evolved to kill / incapacitate predators not cure them of certain cellular defects, for example taxol is a tubular inhibitor evolved against fungi not as a chemotherapy agent.

A competing technology is DEL, DNA-encoded libraries, wherein a combinatorial library is made and barcoded with DNA. There was a talk about an intersection of this and fragment screening. Lucie Guetzoyan (Vernalis) presented PAC-FragmentDEL (paper), with a DNA barcode, diazirine group (UV crosslinking), and a combination of linker and fragment libraries.

Overall, it was clear that fragment screening is a great approach for quickly zeroing in chemical matter to explore. It was the topic of the conference so one would expect success stories presented, but what was great was the fact that there were so many of them and in a large variety of approaches.

Conclusion

The major strength of the conference was that it was a series of overviews of drug discovery pipelines, spanning years of work, using different approaches and methodologies all centred around FBDD as discussed above and enumerated in the further reading footnote. Not everything was covered obviously and, for anyone considering future attendance, I have listed areas that were not covered in a footnote below in an attempt to help frame relevance. However, the tight focus on FBDD is great and it showcased what happens in drug discovery campaign —I for one will take note of several details presented.

Lastly, I would like to conclude with thanking the Rosetrees Fund whose grant allowed me to attend.

Footnote: Further reading

In an attempt to write this post thematically as a discussion as opposed to a laundry list recapping the talks, I have skipped many talks and not referenced papers. Here are some papers mentioned in the talks:

• https://pubs.acs.org/doi/abs/10.1021/acsmedchemlett.6b00464 — Structure Based Design of Non-Natural Peptidic Macrocyclic Mcl-1 Inhibitors
• https://pubs.rsc.org/en/content/articlelanding/2022/md/d2md00197g — PAC-FragmentDEL – photoactivated covalent capture of DNA-encoded fragments for hit discovery
• https://europepmc.org/article/MED/27050677 — Small molecules, big targets: drug discovery faces the protein-protein interaction challenge.
• https://www.cell.com/cell-chemical-biology/pdfExtended/S2451-9456(21)00058-1 — A small-molecule inhibitor of the BRCA2-RAD51 interaction modulates RAD51 assembly and potentiates DNA damage-induced cell death
• https://www.biorxiv.org/content/10.1101/2023.03.22.533679v1 — Selective inhibitors of the Aurora A-TPX2 protein-protein interaction exhibit in vivo efficacy as targeted anti-mitotic agents
• https://www.nature.com/articles/s42004-023-00926-1 — Deconstructing allostery by computational assessment of the binding determinants of allosteric PTP1B modulators
• https://pubs.acs.org/doi/abs/10.1021/jm980299k — Structural Optimization Affording 2-(R)-(1-(R)-3,5-Bis(trifluoromethyl)phenylethoxy)-3-(S)-(4-fluoro)phenyl-4- (3-oxo-1,2,4-triazol-5-yl)methylmorpholine, a Potent, Orally Active, Long-Acting Morpholine Acetal Human NK-1 Receptor Antagonist
• https://pubs.acs.org/doi/full/10.1021/acscentsci.3c01102 — Inhibitors of the Elastase LasB for the Treatment of Pseudomonas aeruginosa Lung Infections
• https://www.sciencedirect.com/science/article/pii/S0960894X22004243 —
• Discovery of a selective c-MET inhibitor with a novel binding mode
• Author links open overlay panel
• https://www.nature.com/articles/nchembio.1081 — Discovery of an allosteric mechanism for the regulation of HCV NS3 protein function
• https://pubs.acs.org/doi/10.1021/acs.jmedchem.1c02175 — Improved Binding Affinity and Pharmacokinetics Enable Sustained Degradation of BCL6 In Vivo
• https://www.nature.com/articles/s41586-020-2925-1 — Small-molecule-induced polymerization triggers degradation of BCL6
• https://www.pnas.org/doi/abs/10.1073/pnas.1812963116 — Discovery of potent SOS1 inhibitors that block RAS activation via disruption of the RAS–SOS1 interaction
• https://pubs.acs.org/doi/10.1021/jacs.1c03035 — Reversible Covalent Imine-Tethering for Selective Stabilization of 14-3-3 Hub Protein Interactions

Lastly, in my slowness in publishing this post, Dan Erlanson has posted a great review on the conference.

Footnote: one’s background

I would like to give an unsolicited piece of advice to more junior academics: do not get impostor syndrome around medicinal chemists, based on conversations at the conference.

In The Big Bang Theory sitcom, the protagonist, Sheldon, a physicist, denigrates a geologist: this behaviour has a grain of behavioural truth and applies here. In various interactions, I saw this toxic pattern: someone apologising to me for not being a chemist, someone asking if I was a chemist or even saying it was nice to talk to a chemist, or a PhD student admitting being lost with the compound names. I did my PhD ten years ago in biochemistry, but I work in drug discovery (cheminformatics) and have learnt as I went along the names of many exceptions to Hantzsch–Widman nomenclature, which is not hard. Likewise, most med-chemists, did not do a PhD in MedChem, they most likely worked on a very specific reaction. So my advice is:

• do not feel threatened by Sheldon mocking your rocks
• Learn the names of the compounds —slow and steady is fine. Do remember nobody is IUPAC strict about the names.
• Do ask for clarifications but without apologising.

Footnote: not covered

As mentioned in the lead, the conference was shy on certain aspects. Not everything can be covered at a conference and those are mega-conferences with concurrent sessions,15-minute talks, and running for 16 hours, which few people like. Therefore the under-represented topics were:

• There were very few cheminformatics posters. UK-QSAR is in April so the close timing may be a factor.
• There were some computational chemistry works presented, most notably a talk by Julien Michel (University of Edinburgh).
• Deep learning made an appearance in only two posters, while in the panel discussion all the panelists admitted to being lost on the topic —RSC AI in Chemistry 2024 ironically will be in the same venue.
• There were two talks were in house synthesis was described, but no cases were to make a given compound some novel organic chemistry was performed. The latter is a curious absence as in my naïve understanding as leads progress, more fanciful chemistry may come into play to attain the perfect compound. Whereas it seemed that CROs played a part throughout.
• Interestingly, the biochemistry was under-represented as a whole, but there was a universal omission of construct design, which is a source of a lot of headaches and in one or two cases I got the feeling that some woes were from that.
• There was understandably but regrettably not much on Fragment library design. Obviously, the libraries used are often industrial secrets. Universally, a major problem is however what would be the selection criteria. Historically the compounds in a generic library were picked manually by a chemist based on personal preferences. In a more scientific context, it becomes problematic balancing the various details. The DSiPosed library routinely used at XChem was designed for elaborability by using SMARTS patterns, but these do not actually reflect the number of expansion of the compounds and actually results in a lot of repetition as shown by a recent analysis.