Design vs randomness in lead optimization across a frustrated landscape

The goal of lead optimization is to improve the properties of a molecule such that it can become a drug or a chemical probe. Early in a program, particularly one based on fragments, the focus is often potency, but eventually drug metabolism and pharmacokinetics (DMPK) parameters become important. I always enjoy reading optimization stories and seeing clever use of design principles. But how effective is the design process? Or, to put it another way, how often are improvements just luck? This is the subject of an open-access paper just published in Nature by Bryan Roth, James Fraser, Brian Shoichet, and collaborators at University of California San Franciso, University of North Carolina Chapel Hill, and several other institutes.

 

The goal was to create a “random background” against which to compare directed efforts. To do so, the researchers chose six protein targets with which they had experience: the GPCRs α₂AAR, µOR, and CB2; the transporter SERT; and the soluble enzymes AmpC and Mac1. For each protein, they then chose one to six  ligands, 18 total, with IC₅₀, K_I, or EC₅₀ values ranging from 1 nM to 43 µM. These ‘parent ligands’ were then analyzed systematically to see how they could be modified one atom at a time by adding a methyl group, a halogen, or a phenol, or replacing an aromatic ring carbon with a nitrogen. A decade ago, we wrote about the importance of synthetic tractability, and this was a key consideration here. Ultimately the researchers made 257 molecules, with each parent giving rise to between 5 and 31 analogs. These analogs were then tested in a variety of assays.

 

Breaking things is usually easier than improving them, and that turned out to be the case here: 30% of changes caused a drop in activity of at least ten-fold. Perhaps more surprisingly, 29 of the 257 analogs (11.3%) had at least ten-fold better activity than the parent molecules. Such increases were widespread: only one target (AmpC) did not see a ten-fold increase, and 10 of the 18 parent molecules had more potent analogs. In other words, the odds of improvement for any given chemical series were better than even, even without using any structural insights from the protein.

 

This latest work stands in good company. Back in 2012 we highlighted an analysis of the effects of methyl substituents across thousands of examples covering more than 100 proteins. That work came to similar conclusions, with ten-fold improvements in 8% of cases. A more recent paper looked at 633 published examples where added chlorine atoms improved activity by at least ten-fold. This latest Nature paper finds methyl substituents to be slightly more likely to give ten-fold improvements than chlorine atoms, while adding fluorine was much less likely to yield such improvements, and swapping an aromatic carbon for a nitrogen never did.

 

One might assume that gaining potency by adding a methyl group or a chlorine atom is due solely to increased lipophilicity, but this did not seem to be the case as assessed by calculating LipE/LLE values.

 

Weaker ligands were slightly more likely to be improved by making random changes, which makes sense if you consider that a more potent molecule is likely to be highly complementary to its binding site. Similarly, larger pockets were more likely to see ligand activity improve with random changes.

 

The researchers crystallographically characterized 22 analogs of varying affinities derived from three parent ligands against the protein Mac1 to look for general lessons. Although the analyses proved interesting, “many of the effects of the small perturbation analogs could be explained post hoc from their structures; however, fewer were easily anticipated.”

 

Of course, as noted above, affinity is just one component of a chemical probe, let alone a drug, so the researchers also assessed several in vitro DMPK parameters: microsome stability, plasma stability, plasma protein binding, solubility, hERG inhibition, and membrane permeability. Similar to the activity measurements, some analogs had improvements in one or more properties, while others saw declines. Unfortunately, the analogs with improved potency often had decreased values for in vitro DMPK parameters; none of the 29 analogs with ten-fold improved potency had combined improvements for stability, permeability, and plasma protein binding. As the researchers ruefully conclude, “the changes in in vitro PK were, at best, orthogonal to affinity or potency fold change, and most were, if anything, anti-correlated with it.” This “frustrated landscape” is all too familiar to medicinal chemists.

 

Making and testing all these molecules took a tremendous amount of work (the paper lists over 30 authors), and the researchers naturally wondered whether computational methods would have saved them the effort. They used free energy perturbation (specifically FEP+, from Schrödinger) to calculate affinities and three additional programs to calculate in vitro DMPK parameters. Although the affinity correlations were strong, there was still considerable variation for individual compounds: of 34 analogs predicted to increase in binding energy by more than 1 kcal/mol (about five-fold), only 14 actually did so, while 8 showed decreased activity. Meanwhile, while plasma protein binding and permeability predictions correlated with experimental values, stability and solubility “were essential uncorrelated.”

 

As of now, computers are still no substitute for the good old-fashioned design, make, test cycle. This paper suggests that, in addition to careful design, it may also be worth introducing some randomness into your analogs, particularly when they are easy to make. Sound advice indeed.

Computational prediction of cryptic pockets

Last week we highlighted how NMR-based fragment screening was able to identify cryptic pockets on a small protein. Experimental methods are (still) the gold standard, but they take considerable effort and resources. In a new open-access J. Am. Chem. Soc. paper, Neha Vithani, David LeBard, and collaborators at OpenEye, Bayer, Genentech, and Khartis provide a computational approach.

 

The researchers defined six categories of cryptic pockets: 1) side-chain motion, 2) loop motion, 3) secondary-structure motion, 4) secondary-structure change, 5) interdomain motion, and 6) pre-existing deep pockets with no access to the surface of the protein. They collected a set of 21 proteins with 26 cryptic pockets overall (some proteins contain more than one) whose apo (unbound) and holo (ligand-bound) structures were available in the Protein Data Bank. This “Encore Dataset” contains all six types of cryptic pockets, though some are represented by only two or three examples.

 

To computationally identify cryptic pockets, the researchers used weighted ensemble molecular dynamics (WEMD) simulations on the apo structures. The details are quite technical but involve focusing on “normal modes” of protein movement. To find pockets, xenon is included in the simulation; these large atoms will fill pockets that open up during the course of the simulation, so detecting bound xenon reveals pockets.

 

WEMD successfully identified 24 of the 26 cryptic pockets in the Encore Dataset. Interestingly, both failures were for the same protein, hepatitis C polymerase, one a preexisting deep pocket and the other a secondary-structure change from order to disorder. (This latter category is exemplified in one of the cryptic pockets we discussed last week, in which crystallography revealed that a helix became disordered upon ligand binding.) The researchers hypothesize that these types of order-to-disorder transitions may be particularly difficult to detect computationally because the models are designed to identify secondary structures, thus overstabilizing them.

 

So WEMD is able to correctly predict most cryptic pockets in the Encore Dataset, but does it also predict cryptic pockets that don’t exist? Paul Samulsen famously quipped that economists had “predicted nine out of the last five recessions,” which actually looks pretty good compared to WEMD, which predicted more than 30 potential cryptic pockets on the protein biotin carboxylase. For comparison, the software program fpocket, which we wrote about last year, predicts 13 for this protein.

 

To rank which pockets actually exist, the researchers turned to another computational technique called Target X, which was described in this Research Square article earlier this year. This proved reasonably successful at ranking experimentally observed pockets near the top, with the top two predicted biotin carboxylase pockets corresponding to known pockets.

 

This is an interesting paper, but I am left with more questions. First, while the Encore Dataset is nice, why didn’t the researchers (also) use CryptoSite, a previously published set of 93 proteins with cryptic sites, which we discussed in 2024? More importantly, how can drug hunters decide whether to target a specific cryptic pocket? For instance, does WEMD and Target X allow one to assess whether cryptic pockets can support high-affinity ligands? These would be interesting questions for a follow-up paper.

Merck’s journey from flatland

Back in 2009 we highlighted a paper, “Escape from flatland,” suggesting that molecules with higher fractions of sp³-hybridized (ie, non-aromatic) carbons are more likely to become drugs. This hypothesis has been challenged on methodological grounds, and an analysis last year concluded that “Fsp³ may not have been a useful metric to optimize.” Now a new J. Med. Chem. paper by Olivia Garry and colleagues at Merck weighs in on the debate.

 

The researchers calculated Fsp³ (number of sp³-hybridized carbons divided by total carbons) for millions of small molecules (MW < 1200) synthesized at Merck from 2000 through 2024. Interestingly, mean Fsp³ values dropped from 0.34 to 0.31 between 2000 and 2006 or so, coinciding with the increasing popularity of metal-mediated cross coupling reactions during that period. Since then, Fsp³ has risen to 0.40 in 2024, which could be due to the focus on “escaping flatland” as well as the increasing availability of saturated building blocks and new synthetic methodologies.

 

Consistent with the original publication, the average Fsp³ increases as molecules progress through development, from 0.34 overall to 0.36 for those that go into rat pharmacokinetic studies and to 0.38 for preclinical candidates – close to the 0.39 found for drugs approved since 2009 according to the study last year.

 

A key correlation proposed in the original flatland publication was that solubility increased with Fsp³, and this was observed again here. Of 435,280 compounds having Fsp³ between 0.00-0.70 with measured solubility data, 55% of those with more aromatic character had low solubility, while only 15% of those with Fsp³ > 0.65 were poorly soluble. The correlations held for solubility at neutral pH as well as at pH 2. Molecules with lower Fsp³ also tended to have more aromatic rings, and this has previously been shown to lower solubility.

 

In contrast to solubility, lipophilicity (logD) did not correlate with Fsp³ among 503,780 compounds. Apparent permeability showed a complex and noisy correlation for 16,570 compounds, leading the researchers to conclude that “Fsp³ is not a good parameter to optimize P_app.”

 

Inhibition of three cytochrome P450 enzymes, CYP 3A4, 2C8, and 2C9, was examined for 51,760 compounds. Weak trends were observed for the latter two, with highly aromatic compounds being more likely to inhibit the enzymes, but no trend was seen for CYP 3A4.

 

Inhibition of human ether-a-go-go-related gene (hERG) is a major red flag for compound progression, and here there was a correlation, with higher Fsp³ compounds showing lower inhibition. This trend held for both aminergic and nonaminergic molecules, though as with solubility these trends could be driven by aromatic ring count, which has independently been correlated to hERG inhibition.

 

Since the 2009 flatland publication, many solvents have been spilled to replace phenyl rings with more shapely, non-aromatic moieties. The researchers looked at matched molecular pairs for internal compounds to see what effects these changes had on solubility. Simply saturating a phenyl to a cyclohexyl ring tended to decrease solubility, while the best solubility improvements resulted from cyclopropyl or isopropyl substitutions. But these swaps did little for shapeliness: of the eight substitutions for which sufficient data existed, cyclopropyl and isoprproyl had the lowest increase of average Fsp³. As the researchers note, “using Fsp³ without consideration of other properties is not a good strategy to optimize solubility.”

 

Overall this study suggests that leaving the comforts of flatland may be worthwhile. Molecules with higher Fsp³ values are sometimes more challenging to synthesize, but this itself can lead to new intellectual property space.

 

It’s worth emphasizing that the correlations between Fsp³ and solubility, CYP inhibition, and hERG inhibition are limited. In the end, the researchers recommend “considering Fsp³ in combination with other factors in medicinal chemistry optimization as its standalone effect on specific properties is modest.” Just as general risk factors won’t predict whether any given person will contract a disease, metrics won’t predict whether any given molecule will succeed (or fail) as a drug. 

Vast fields of biologically active chemical space

Exactly 17 years ago, I highlighted a paper from Brian Shoichet’s lab in which he hypothesized that only tiny regions of chemical space are biologically relevant. If so, Practical Fragments quoted, “a major reason why the screening of synthetic compounds ever finds notable hits is that our libraries are biased toward the sort of molecules that proteins have evolved to recognize.” The best scientists continually question their assumptions, and this is exactly what Brian, Bryan Roth, and collaborators at University of California San Francisco and University of North Carolina at Chapel Hill have done in a new open-access J. Med. Chem. paper.

 

The idea that molecules more closely resembling metabolites and natural products would be more likely to be biologically active is reasonable given the promiscuity of many naturally occurring molecules. As the researchers point out, dopamine signals through 5 receptors, while serotonin binds to 14. Nature is the ultimate recycler, constantly reusing chemical motifs.

 

But since 2009 there has been a massive growth of make-on-demand libraries, and these have increasingly diverged from known molecules. Brian has been among the most prolific explorers of this new chemical space, and as he noted in his keynote at DDC 2026, these gargantuan libraries are yielding more hits against more targets.

 

Are these hits more selective? In other words, are molecules that are less “bio-like” likely to bind to fewer targets? This is the question the new paper addresses, focusing on the 5-HT_2A serotonin receptor (5-HT_2AR), the target for a variety of drugs including psychedelics. The researchers have extensively studied this GPCR; we wrote about a successful computational screen here. In the new paper, the researchers compared hits from a previous “make-on-demand” (MoD) library screen of 1.6 billion molecules with a new screen of 3.5 million “in-stock” molecules. As predicted, the in-stock compounds were much more bio-like than the MoD compounds as assessed by several metrics (Tanimoto similarity, Avalon fingerprints, and RDK7 fingerprints, for the cheminformatics aficionados among you).

 

Both libraries were computationally screened using DOCK3.8 against 5-HT_2AR. Of the top scoring hits, 85 molecules from the in-stock library and 384 molecules from the MoD library were tested in a radio-ligand displacement assay. Hit rates were nearly identical at around 24% each. Functional assays revealed the hits to be active, with some acting as agonists while others were antagonists.

 

When tested against related GPCRs, specifically 5-HT_2BR and 5-HT_2CR, the in-stock molecules were more promiscuous than the MoD molecules. But when tested against a panel of 318 GPCRs there were no differences: the in-stock molecules bound on average 1.7% of the receptors while the MoD molecules bound on average 2.1%. The similar promiscuity persisted even when focusing on the 55 aminergic GPCRs (5-HT_2AR is an aminergic GPCR). In fact, the most promiscuous compound of all came from the MoD set and is distinctly not bio-like.

 

Of course, this is just one study, but I suspect it will generalize. In 2009 I questioned whether biologically relevant chemical space is really so sparse. As I wrote then, “consider a vast field of some crop that can only be harvested by night. There are lights scattered haphazardly throughout the field. One might expect that the crops immediately under the lamp posts would be harvested more intensively than crops in darker parts of the field, even if other areas are equally productive. In this scenario, the lamp posts reveal natural products and similar molecules, but much – or even most – of (unlit) chemical space may also be biologically active, it just hasn’t been sampled yet.”

 

There is grandeur in this view of chemical space, which is supported by the new paper. Imagine yourself in the midst of the field on a moonless night. Turn on your headlamp and step forward in any direction. Drug leads glimmer as far as your light can reach.

Twenty-First Annual Fragment-Based Drug Discovery Meeting

Last week some 875 people attended the CHI Drug Discovery Chemistry (DDC) meeting in San Diego. I can’t do justice to the 40 or so presentations I attended over four days but can highlight some of the main themes.

 

Reversible fragments

Membrane targets such as G protein-coupled receptors (GPCRs) pose a challenge for biophysical methods, but three talks presented progress. Matthew Eddy (University of Florida Gainesville) described high-resolution magic angle spinning (HRMAS) NMR, which entails spinning isolated cellular membranes containing GPCRs at high speed (4 kHz!), which miraculously yields sharp NMR signals for bound ligands. Matthew demonstrated applications with the human adenosine A_2A receptor and weak (mM) ligands. He noted that the technique can work with native, poorly expressed proteins, though data collection times can be upwards of 30 minutes.

 

Kris Borzilleri described using ¹⁹F NMR to find ligands against an orphan GPCR at Pfizer. The 2287 fragments screened yielded 87 hits, of which 38 confirmed by SPR. SAR studies eventually yielded low micromolar ligands, but these were difficult to advance in the absence of structure (see here for a more successful example from Merck).

 

Vanessa Porkolab (Eurofins Cerep) described using the Nanotemper Spectral Shift technology to screen 826 fragments against the adenosine A_2A receptor at 300 µM, with a 9.2% hit rate. Many of these ligands stabilized the GPCR in a thermal shift assay and seven were even active (as antagonists) in a cellular assay.

 

Turning to soluble proteins, Paola Di Lello presented a case study from Genentech and Vernalis applying ligand-observed NMR to the protein phosphatase PTPN22. Subsequent protein-observed NMR revealed that most of the 16 validated hits bound to two pockets some distance from the active site. The fragments were optimized to mid-micromolar affinity but showed no functional activity.

 

And Charlotte Hodson presented the eIF4E story from Astex. As we discussed last year, this yielded a low nanomolar ligand that did not have the desired cellular effects. Charlotte noted that subsequent genetic experiments were consistent with the limited efficacy. Still, the target was sufficiently interesting that a chemical probe would have been pursued even knowing it would be high-risk.

 

Covalent ligands

Covalent approaches made appearances throughout the conference. Keriann Backus (UCLA) described chemoproteomic approaches to find cysteine-targeting ligands; she noted that gain of cysteine residues (such as G12C in KRAS) are the most common missense variants in cancer. Keriann also warned how covalent compounds can cause potentially misleading effects in cells, as she described in Nat. Chem. Biol. last year.

 

In 2021 we wrote about the SpotXplorer fragment library from György Keserű (Hungarian Research Centre for Natural Sciences). György has now prepared a PhotoXplorer library, which uses diazirine tags for photochemical screening, which we described here. The new library has produced high hit rates across a variety of targets. György also described a new sulfozone-based photoprobe that is easier to prepare than diazirines.

 

Kelly Craft recounted a DNA-encoded library (DEL) screen at AbbVie against the target BCL2A1, also known as BFL1. This produced an aldehyde-containing low micromolar binder that formed an imine with buried lysine 102. Uncomfortable progressing an aldehyde, the researchers sought to covalently engage cysteine 55, the same cysteine targeted by AstraZeneca, as we wrote about here. The progression included at least one dual-warhead molecule which was crystallographically confirmed to bind both the lysine and cysteine. The effort ultimately yielded cysteine-selective leads.

 

Earlier this year I described the dDRTC method we developed at Frontier Medicines for determining k_inact/K_I, and Svetlana Kholodar presented a nice overview of its scope and utility. My colleague Johannes Hermann spoke in more detail about our covalent technologies, particularly those using AI.

 

Chemical space and the exploration thereof

Brian Shoichet (UCSF) gave an entertaining and wide-ranging account of “directed and random walks in chemical space.” Brian has consistently been on the bleeding edge of high-throughput in silico screens, from 67,000 compounds in 2009 to 138 million molecules in 2019 to 4 billion molecules today. When docking artifacts are avoided (as we discussed here), bigger libraries consistently produce more potent hits for more targets – an observation strikingly consistent with Alex Shaginian’s in 2023 as HitGen expanded their DEL libraries from billions to more than a trillion molecules. Brian is developing methods to computationally screen the >4 trillion make-on-demand molecules now available from companies such as Enamine.

 

Direct-to-biology (DTB) approaches, which rely on microscale chemical reactions screened without purification, have become increasingly popular methods for exploring chemical space. Jack Sadowsky correctly stated that Carmot was the first company formed around this approach; we previously wrote about the role Chemotype Evolution played in the discovery of sotorasib. Jack described how Kimia, which spun out of Carmot, has continued to advance the technology, applying it to find inhibitors selective for single members of closely related kinase families.

 

Allan Jordan described how Sygnature Discovery is applying DTB in a variety of assays including microsome stability and crystallography. (We wrote about crude reaction screening by crystallography earlier this year.) Expanding beyond DTB, Allan called their platform direct-to-discovery, and discussed how it led to a preclinical candidate with STORM Therapeutics in just 18 months.

 

WuXi Apptec is also using DTB. Peichuan Zhang described starting with ligands derived from fragment and DEL screens against the E3 ligase GID4 to make PROTACs to degrade BRD4; DTB was used to explore a wide range of different linkers. And Daniel Blair (St. Jude) described using DTB and affinity selection-mass spectrometry (AS-MS) to find new molecular glues for the oncology target LCK.

 

Computers, DEL, and DTB are not the only way to explore chemical space. Last year we covered Tom Kodadek’s bead-based screening approach at University of Florida Scripps, and Tom presented two talks on the topic, one using macrocycles to find binders to difficult targets such as PTP1B and one using small molecules to find molecular glues.

 

Speaking of PROTACs and glues, plenary keynote speaker Alessio Ciulli (University of Dundee) discussed the “evolution and future of targeted protein degradation.” Alessio noted that there are >25 PROTAC degraders and >10 glues in the clinic, though these collectively target only a small number of E3 ligands, so there is plenty of opportunity for the area to expand.

 

For many of us in industry, drugs represent the most privileged points in chemical space, and these often look quite different than we assume, as Dean Brown (Jnana) noted in his recent analysis of 104 oral small molecule drugs approved by the FDA from 2020 to 2024 (which we mentioned here). Some drugs contain eye-raising moieties such as acetylenes, styrenes, N-O bonds, and nitro groups. Indeed, it is worth remembering that venetoclax, arguably the most successful fragment-derived drug, sports a nitro group.

 

But before getting too complacent, Jonathan Baell (Manas) warned about frequent hitters in libraries of FDA-approved drugs. He notes in Eur. J. Med. Chem. earlier this year that many commercial libraries are actually enriched for molecules that cause spurious biological activity. Jonathan calls on library vendors to remove particularly egregious compounds, though I’d settle for world peace.

 

I’ll close on that pleasant thought, but please feel free to comment. I hope to see you in San Diego next year April 19-22 for the twenty-second iteration of DDC.

xSAR: Crystallographic SAR from crude reactions

Last year we highlighted an example of crystallographic screening of crude reaction mixtures to find inhibitors against the oncology target PHIP(2). Of 957 molecules tested, 22 showed crystallographic binding in two different orientations: 19 in a “lateral” pose and 3 in a “diving” pose. In a new open-access Chem. Sci. article, Philip Biggin and collaborators at Diamond Light Source and University of Oxford try to extract information from both the binders and the non-binders using crystallographic structure-activity relationships, or xSAR.

 

Chemists often think about SAR in qualitative terms: a methyl group here improves affinity, a chlorine atom there reduces it. In xSAR, the researchers sought to take a more quantitative approach. They converted each molecule into “Morgan fingerprints,” a set of more than 2048 binary bits describing structural features such as atom type, hybridization, and connectivity to other atoms within a certain distance. Some bits were found in all binding compounds, and these were referred to as conserved binding bits (CBB), while conserved nonbinding bits (CNB) were found only in non-binding compounds. These bits were then used to calculate Positive and Negative Binding Scores (PBS and NBS); a compound with a PBS of 1 contains all the CBB. Since there were two separate binding modes, the researchers calculated PBS and NBS values for both lateral and diving poses individually as well as for all binders.

 

As the researchers note, false negatives are a likely issue in crude reaction screening for a variety of reasons. To hunt for these, the PBS and NBS values were calculated for all 957 molecules previoulsy tested. A set of 97 pure compounds having mostly high scores were acquired and tested crystallographically, yielding an additional 23 lateral binders and 3 diving binders, more than doubling the initial yield. PBS was particularly informative in this retrospective exercise to recover false negatives, outperforming both NBS as well as other methods such as Tanimoto similarity scores.

 

The researchers also used PBS and NBS scores to search prospectively for new binders in a virtual set of more than 1.7 billion compounds in the Enamine REAL database. After filtering for high PBS/NBS scoring compounds followed by docking, 93 compounds were acquired and tested crystallographically. Interestingly, this yielded a relatively low hit rate of 9 binders, 6 in the lateral pose and three in somewhat different poses. None of the new compounds bound in the diving pose, which the researchers suggest may be due to the small sample size used to calculate PBS and NBS for this binding mode.

 

The 93 new compounds were also tested for binding using grated-coupled interferometry (GCI), and 13 showed measurable affinity, with most better than 50 µM. Two even showed single-digit micromolar affinity, more than an order of magnitude better than the best compound from the screen we discussed last year, and with better ligand efficiencies too. Surprisingly, these two compounds were not hits in the crystallographic screen.

 

This is an interesting paper with a couple important lessons. First, despite the fact that affinity was not used in calculating PBS and NBS, these metrics were nonetheless useful for identifying molecules with better affinity than those in the original training set, arguing for their utility. But perhaps just as importantly, the molecules with the best affinity were missed by crystallographic screening. If anything, this observation only strengthens my conclusion last year that while “there is a strong case for using crystallography first for finding fragments, I am not yet convinced the same applies for optimizing fragments.”

Review of 2025 reviews

Turning and turning in the widening gyre

The falcon cannot hear the falconer…

 

In our small, annual counterweight to Yeats’ “mere anarchy," Practical Fragments looks back on 2025.

 

This year marked the thousandth post on Practical Fragments, a milestone neither Teddy nor I imagined when the blog launched back in 2008. In terms of conferences, I wrote about CHI’s Drug Discovery Chemistry in San Diego here and Discovery on Target in Boston here.

 

For the past decade I’ve participated in annual J. Med. Chem. perspectives covering fragment-to-lead success stories, two of which published this year. The first, spearhead by Rhian Holvey at Astex, covers the year 2023, while Astex’s David Twigg took the lead on covering the year 2024. In addition to the tabular summaries for which these reviews are best known, both also include tables of “near misses,” none of which made the main tables because the starting points were (sometimes just slightly) too large. Four out of six of these heavyweights are covalent fragments, suggesting that the rule of three may need to be relaxed for these. The most recent paper also includes a table showcasing the eight approved FBLD-derived drugs.

 

Two more general publications are also of interest. In a brief (open-access) editorial in J. Med. Chem. Weijun Xu and Congbao Kang at A*STAR summarize fragment-finding methods and approved drugs and discuss future applications of FBLD in PROTACs and targeting RNA. And in Curr. Res. Pharm. Drug Discov., Geoffrey Wells, Exequiel Porta, and colleagues at University College London present a “graphical review” which covers library design, screening strategies, hit validation, fragment optimization, and a few case studies of approved drugs, along with current challenges.

 

In Drug Des. Devel. Ther., Bangjiang Fang and colleagues at Shanghai University of Traditional Chinese Medicine present an open-access bibliometrics analysis of 1301 fragment-based drug design papers published between 2015 and the end of 2024, which includes top ten lists of institutions, authors, and papers as well as keyword trends and analyses. Annual growth has averaged 1.4%, and the field is both global and collaborative, with 35% of publications involving more than one country.

 

Targets

Two reviews focus on oncology. The first, in Bioorg. Chem. by Milind Sindkhedkar and collaborators at Manipal College of Pharmaceutical Sciences and Lupin Ltd., briefly covers the history and practice of FBDD before providing short summaries of seven of the eight approved drugs to come from it. The second, published open-access in Chem. Rev. by Vanderbilt’s Steve Fesik, is a concise and highly readable introduction and account of the author’s groundbreaking work on BCL-2 family proteins, KRAS, and WDR5.

 

A much longer open-access review in Chem. Rev. by Paramjit Arora and colleagues at New York University covers protein-protein interactions (PPIs). Much of the focus is on larger molecules such as cyclic peptides, peptide mimetics, and other macrocycles, but there are summaries of fragment-based approaches against KRAS and 14-3-3 proteins.

 

The topic of 14-3-3 proteins is treated more fully in an open-access Acc. Chem. Res. paper by Michelle Arkin and colleagues at UCSF. While the focus of most efforts against PPIs is to find inhibitors, for 14-3-3 the goal is to find stabilizers, or molecular glues. The Arkin lab and others have been succeeding using various approaches, particularly disulfide tethering. We wrote about these efforts most recently in 2023, and the new review provides a nice update.

 

Fragment finding methods and libraries

Sahra St. John-Campbell and Gurdip Bhalay, both at The Institute of Cancer Research, published a massive open-access perspective on “target engagement assays in early drug discovery” in J. Med. Chem., covering a host of biochemical, biophysical, and cell-based assays. A table lists more than 50 different techniques, almost half of which are applicable to FBLD. Each row shows what characteristic(s) are measured as well as critical requirements for  protein, sample, and equipment. The paper is also beautifully illustrated with dozens of figures: one shows which techniques are most useful for different types of targets, and each method gets its own diagram.

 

A more focused open-access review is provided by Stefanie Freitag-Pohl and colleagues at Durham University in Biophys. Rev. After surveying various biophysical techniques, the researchers focus on spectral shift analysis, and in particular the Dianthus instrument from NanoTemper Instruments. This plate-based, high-throughput microfluidics-free instrument can detect changes in fluorescence caused by environment or temperature. Examples demonstrate affinity measurements across several orders of magnitude, up to double-digit millimolar, and a nice scheme shows use of the Dianthus in a fragment-screening workflow.

 

Moving to specific techniques, Jia Gao, Ke Ruan, and colleagues at University of Science and Technology of China Hefei provide an open-access survey of “the rise of NMR-integrated fragment-based drug discovery in China” in Mag. Res. Lett. After a brief overview of NMR approaches, they cover case studies from China, most of which are focused on fragment screening rather than optimization.

 

A less common biophysical method is native mass spectrometry (nMS), the subject of an open-access opinion in RSC Med. Chem. by Louise Sternicki and Sally-Ann Poulsen at Griffith University. This is a good survey of the approach; we highlighted a more fragment-focused review by the same authors last year.

 

The most common fragment-finding approach, X-ray crystallography, is covered in two open-access reviews. The first, in Acta Cryst. F by Sarah Bowman and collaborators at University of Buffalo and Brookhaven National Laboratory, focuses on critical early stages, from protein characterization to sample preparation and various crystallization approaches. The second, in Curr. Opin. Strut. Biol. by Martin Noble and colleagues at Newcastle University, starts by briefly reviewing crystallographic fragment screening before turning to fragment libraries. The paper includes a nice table summarizing publicly available libraries at major synchrotrons, with the text describing these in more detail.

 

The provider of one of these libraries, EU-OPENSCREEN, is the subject of an open-access review in SLAS Discov. by Robert Harmel and collaborators at EU-OPENSCREEN ERIC and Fraunhover ITMP. As we wrote last year, EU-OPENSCREEN is a broad consortium whose mission is to advance early drug discovery by providing access to technology and expertise. The new paper summarizes the four compound collections, including the European Fragment Screening Library (EFSL), and surveys progress to date. It also lays out ambitious plans, including expanding to >30 sites in nine countries.

 

Computational approaches and cryptic sites

Despite the hype about artificial intelligence in the broader world, AI in fragment-based drug discovery has been less common. In Curr. Opin. Struct. Biol., Woong-Hee Shin and colleagues at Korea University College of Medicine summarize applications to fragment growing, merging, and linking. The open-access paper includes a handy table of 13 programs, and includes GitHub links where available.

 

Cryptic binding sites, defined by Ehmke Pohl and collaborators at Durham University and Cambridge Crystallographic Data Centre “as binding pockets that exist in the ligand-bound state of a protein but not in its apo form,” are the focus of an open-access review in Bioinform. Adv. The researchers cover earlier computational approaches for finding these, especially molecular dynamics (MD) and machine learning (ML). They note that a key challenge for ML is the limited quantity and quality of experimental data: undiscovered cryptic sites would be misclassified as non-binding sites.

 

Yowen Dong, Ge-Fei Hao, and colleagues at Guizhou University review “computational methods for identifying cryptic pockets” in Drug Discov. Today. As with the previous review, these are divided between molecular dynamics and AI-based techniques, which are discussed individually and then compared. The researchers apply six approaches to the model bacterial protein TEM-1 β-lactamase and find that, for this highly studied single protein, the AI-based methods are much faster (seconds instead of days) and just as accurate, though MD-based methods provide more insight into formation mechanisms of cryptic pockets.

 

Covalent ligands

Allosteric sites are an important sub-class of cryptic pockets, and in J. Med. Chem. Jianing Li and colleagues at Purdue University discuss covalent allosteric inhibitors. After briefly discussing advantages of covalent molecules, they review examples targeting protein phosphatases, kinases, and GTPases, such as KRAS.

 

Of course, covalent molecules are not limited to allosteric sites. An open-access review in Bioorg. Med. Chem. Lett. by Walaa Bedewy, John Mulawka, and Marc Adler at Toronto Metropolitan University summarizes published covalent protein ligands, grouping them by target site:  active sites, residues adjacent to an active site, protein-protein interfaces, cofactor binding sites, and allosteric sites.

 

Chem. Rev. published two massive reviews on covalent ligands, each with more than 300 references. The first, by Tomonori Tamura, Masaharu Kawano, and Itaru Hamachi at Kyoto University, covers a wide range of topics, from covalent drugs, to peptide- and protein-based covalent inhibitors, to chemical biology labeling and target engagement strategies, to covalent bifunctional molecules such as PROTACs and radionucleotide-based molecules, and even covalent modification of DNA and RNA. The paper includes 68 figures, many reproduced from the original publications.

 

The second (open-access) Chem. Rev. paper, by Ku-Lung Hsu and colleagues at University of Texas at Austin, focuses on covalent ligands targeting protein residues other than cysteine, particularly lysine and tyrosine; we highlighted some of Hsu’s work recently. The paper also discusses naturally occurring molecules that bind to lysine, such as pyridoxal phosphate and aldose sugars.

 

Methods for finding covalent ligands are the focus on an open-access review in JACS Au by Mengke You, Hong Liu, and Chunpu Li at Shanghai Institute of Materia Medica. Specifically, they review disulfide tethering, activity-based protein profiling (ABPP), covalent DEL, phage and mRNA display, and sulfur(IV) fluoride exchange (SuFEx), with examples for each.

 

The last paper on this topic, in J. Med. Chem., offers a brief but important overview of all covalent FDA-approved small molecule drugs through 2023. Samuel Dalton and collaborators at Isomorphic Laboratories and Merck counted 128 covalent drugs, about 7% of all small molecule drugs. More than half are antibiotics, and more than 85% target serine or cysteine. Only 10% are reversible, but this number is rapidly increasing, with 11 of the 13 reversible covalent drugs approved since 2010. Importantly, the names, chemical structures, indication, target and target residue, warhead, and key references for all the drugs are provided in the supporting information.

 

Miscellaneous

Deconstruction of ligands to smaller fragments that are then “reconstructed” into new leads is a venerable approach in FBLD and the subject of an open-access perspective in J. Med. Chem. by J. Henry Blackwell, Iacovos Michaelidies, and Floriane Gibault at AstraZeneca. Multiple examples dating as far back as the late 1990s are provided, along with appropriate caveats about potential changes in fragment binding modes and protein conformations.

 

Finally, an open-access perspective in J. Med. Chem. by Dean Brown (Jnana Therapeutics) examines the 104 oral drugs approved from 2020 through 2024, including structures, dosing, pharmacokinetics, and safety. Roughly a third of these drugs are dosed more than once per day, and almost a quarter have a black box warning, while 42% have at least one contraindication. Dean warns that “overly prescriptive [development candidate] criteria may inadvertently stifle the development of innovative drugs,” and that it is difficult but important “to be the champion for a compound that others perceive as ‘un-drug like.’” The growing success of covalent drugs illustrates that some organizations are taking this to heart.

 

And that’s it for 2025. Thanks for reading and special thanks for commenting. And in 2026, may the best of us be filled with passionate intensity.

Twentieth-Third Annual Discovery on Target Meeting

The CHI Discovery on Target (DoT) meeting was held last week in Boston. More than 850 people from 24 countries attended, 75% from industry. As usual I’ll just touch on some broad themes.

 

Covalent approaches

Covalent approaches were prominent throughout the conference. One of the very first talks was by Stefan Harry (Harvard/MGH), who described screening 416 cancer cell lines with three reactive “scout probes,” identifying some 6000 cysteine residues that could be covalently liganded. There are some interesting cell and context-dependent differences, and all the data are publicly available and easily searchable through a free online portal called DrugMap. He is now profiling a library of dual-electrophile-containing compounds to identify molecular glues.

 

Knowing which cysteines can be targeted is the first step for covalent drug discovery, and Sherry Ke Li described how she and colleagues at Genentech go about finding ligands. They’ve experimentally determined the reactivity of more than 6400 compounds against free cysteine and used this to train a machine-learning model to predict chemical (as opposed to specific) reactivity. Mass spectrometry (MS) using isolated proteins is the workhorse screening approach, but Sherry also described using variable temperature surface plasmon resonance (SPR) to dissect the individual components of k_inact/K_I.

 

AstraZeneca has also been doing considerable covalent screening, and Hua Xu briefly described the BFL1 story we wrote about here. In addition to pure proteins, they are now also starting to screen their covalent library in cells. Hua also presented earlier work from Pfizer on the discovery of the covalent kinase inhibitor ritlecitinib, which started with a noncovalent binder. Proteomic studies revealed that in addition to the intended target JAK3, it hits other TEC-family kinases too.

 

Adding a covalent warhead to a reversible binder is also the approach taken by MOMA Therapeutics in the discovery of their clinical WRN inhibitor MOMA-341, as presented by Momar Toure. They ended up targeting the same cysteine as Vividion (see here), though the binding mode is somewhat different. MOMA is also pursuing covalent fragment screening using intact protein MS, and Brian Sosa-Alvarado described how they were able to identify nanomolar inhibitors of RAD54L within six months of starting the program, aided by DNA-encoded libraries (DEL).

 

Not everyone is pursuing cysteine: Ken Hsu described how he and his team at University of Texas Austin are using sulfur-triazole exchange chemistry (SuTEx) to target tyrosine residues across the proteome. He noted that although cysteine could react with this warhead, the resulting thiosulfonate would be unstable. This is true in general, but I wonder if, just like the reversible cyanoacrylamide warheads we wrote about more than a decade ago, they could be stabilized within folded proteins.

 

Noncovalent approaches

Covalency is not the only game in town, as exemplified in a talk by Emma Rivers on “integrated hit discovery” at AstraZeneca. I was tickled that she grouped FBLD with HTS as “traditional” approaches, onto which they’ve added DEL and peptide libraries. Importantly, they’re focused on generating and capturing as much high-quality data as possible to enable machine learning – a topic we’ll touch on more below.

 

Nor are proteins the only target; there was a whole track on RNA- and DNA-targeting small molecules, where Benjamin Brigham described the plate-based equilibrium dialysis-based approach taken at Atavistik to screen metabolites and metabolite-like molecules. This led to two fragment-sized hits against RNA encoding SERPINA1, and although the affinities are modest, they do inhibit translation in a cell-free system.

 

At FBLD 2018 Astex presented the first cryo-EM structure of a fragment-protein complex, noting that throughput was an issue. The company has leaned into that challenge and now has three microscopes, including a top-of-the-line Krios, with another on the way. Miguel Zamora-Porras described how they have now solved hundreds of structures. Their Krios can collect data on two compounds per day, and the full cycle time from protein-ligand preparation to structure is about a week. Miguel described how structures of ligands bound to the ion channel protein TRPML1 helped reveal why some were agonists and others antagonists.

 

Data and its discontents

On the subject of structures, Steve Burley (Rutgers) gave an eloquent history and defense of the “RCSB Protein Data Bank: an open access research resource that benefits all humanity.” From its humble beginnings with just seven structures in 1971, the PDB now contains more than 240,000. And these are not just of scientific interest: all 88 of the new molecular entities the FDA approved for oncology between 2010 and 2023 had PDB structures that informed the biology or druggability, and 75% of the efforts involved structure-based design. Steve also mentioned that the question of where and how to store large-scale crystallographic data will be discussed in a meeting sometime in the spring of next year. Finally, Steve is hoping to retire from his position as Director of the RCSB PDB, so if you’re looking to make an impact, please apply.

 

The dramatic advances in protein structure prediction exemplified by AlphaFold would not have been possible without the PDB, but unfortunately the same high-quality information on protein-ligand binding modes and affinities is not available, as noted by Woody Sherman of Psivant. To illustrate the importance of training data, Woody asked ChatGPT to produce a picture of an analog clock set to 6:32. The result? A clock with three hands, one at 10, one at 2, and one at 6, because most images of clocks are set to 10:10.

 

Woody asked whether machine-learning-based docking can extrapolate or just interpolate. Although impressive results have been reported for some protein-ligand complexes, it turns out that there are often similar ligands in the training data. For truly novel ligands, the predictions tend to fall flat. Similarly, allosteric ligands are often (mis)placed into an orthosteric site – just because the model has been trained that that’s where ligands should go. Indeed, although Psivant is heavily invested in computational approaches, Woody mentioned that they often use “wet” approaches for finding initial chemical matter.

 

On the subject of dubious data, Al Edwards (University of Toronto) noted that a third of all immunofluorescence images in the literature use antibodies that give signals in knockout cells. And as we wrote just last month, many reported small molecule “probes” are just as bad. Al is CEO of the Structural Genomics Consortium, whose ambitious Target 2035 aims to find a pharmacological probe for every target in the human genome. As a starter, they’re aiming for 2000 probes in the next five years. They’re using affinity-selected mass spectrometry (ASMS), screening pools of 500 compounds and 8 proteins at a time, and are getting micromolar hits against about 30% of targets. They’re accepting protein submissions, so if you’re looking for starting points against your favorite protein contact them.

 

I’ll end here, but please leave comments. And mark your calendar for Sep. 28 to Oct. 1 next year, when DoT returns to Boston.

Keeping molecular dynamics cool for fragments

Accurately and reliably predicting fragment binding modes would be preferrable to doing messy, expensive, and sometimes tedious experimental work, but we’re not there yet. One of the biggest problems is that, because fragments usually bind weakly to proteins, it is hard to tell which of several possible binding modes is most favorable. In an open-access J. Chem. Inf. Model. paper published earlier this year, Stefano Moro and colleagues at University of Padova report progress.

 

Their approach, called Thermal Titration Molecular Dynamics (TTMD), analyzes short molecular dynamics simulations across increasing temperatures; if the ligand remains bound to the protein, this indicates a more stable binding mode. (It seems a bit like the dynamic undocking we wrote about here.) The researchers had previously reported good results for larger, drug-sized molecules, but not for four fragment-protein complexes.

 

Recognizing the low affinities of fragments, the researchers decided to lower the (virtual) temperatures. Rather than heating from 300 to 450 K, they heated from 73 to 233 K; ie, from just below the boiling point of liquid nitrogen to a moderately cold winter’s day in Minnesota. They first docked fragments using PLANTS-ChemPLP, which is free for academics, and chose the five best-scoring poses for evaluation.

 

Next, the researchers performed TTMD. There are several different ways to assess how well the ligand remains bound to the protein over the course of a molecular dynamics simulation, and four different scoring methods were chosen. When TTMD was tested on the four fragment-protein complexes that had previously failed, at least two of the scoring methods correctly identified the crystallographic binding mode for three of the fragments.

 

Thus encouraged, the researchers tested ten more compounds bound to six new proteins. The results were quite encouraging, with up to 86% of crystallographic binding modes being correctly identified by at least one of the scoring functions in TTMD vs 50% for docking alone. Impressively, two of the examples were MiniFrag-sized, with just 6 or 7 non-hydrogen atoms, yet the crystallographic pose was identified as the lowest energy in all four TTMD scores.

 

This is nice work, but the question arises how these specific ligands and proteins were chosen. Several years ago we highlighted a curated set of 93 protein-ligand structures that were used to benchmark other virtual approaches, and it would be nice to see how TTMD performs on these. Still, TTMD’s performance on its chosen examples is encouraging, and laudably the researchers have made their code freely available. If you try it out, please let us know how it works in your hands.

Targeting SARS-CoV-2 RNA – but not specifically

Last week we highlighted work suggesting that small molecule binding sites in RNA are most likely to be found in complex structures. A new open-access paper in Angew. Chem. Int. Ed. by Harald Schwalbe and collaborators at Goethe University Frankfurt and elsewhere provides both a case in point and an illustration of how difficult it is to target RNA.

 

The researchers had previously screened 15 RNAs from the SARS-CoV-2 virus, an effort we highlighted in 2021. In the new paper, the researchers focus on a portion of the frameshift element, which is important for directing viral replication from either of two partially overlapping open reading frames. The core of this RNA element is a roughly 69-nucleotide-long structure called a pseudoknot. Like most RNA sequences, this one can form multiple structures, including dimers, and the researchers used NMR, small-angle X-ray scattering (SAXS), and native gel electrophoresis to confirm that the construct was behaving as a homogenous monomer, consistent with three previously determined structures.

 

Based on some of the initial fragment hits, the researchers selected 50 similar molecules, of which only 14 were sufficiently soluble for screening. One of the more potent compounds, D05, initially showed promising activity in a ligand-detected NMR assay but turned out to be completely inactive when retested from a fresh stock. It turns out that D05 decomposes to compound 2, which was confirmed as active. Further modification led to compound 4, the most potent compound described. (Dissociation constants were determined by NMR, fluorescence, or both, and the two methods were in good agreement.)

Two-dimensional NMR with isotopically labeled RNA was used to try to determine the location of the binding site(s). Even with access to a 1.2 GHz magnet, the NMR peaks were severely overlapped, so the researchers used segmental isotopic labeling, in which just half of the RNA was labeled at a time. This exercise revealed potentially three different binding sites for compound 2.

 

The researchers also used two different computational approaches, Vina and RLdock, to predict binding sites, each of which could find one or two of the binding sites identified by NMR.

 

Several compounds were tested to see if they could block frameshifting in cell-lysates, and compound 2 showed 40% inhibition at 145 µM.

 

So far so good. But consistent with best practices, the researchers tested compounds 2 and 4 against phenylalanine tRNA. Unfortunately, the two ligands exhibited similar affinities to this control RNA as they did to the SARS-CoV-2 pseudoknot, despite the lack of sequence similarity. This suggests that these ligands bind to RNA nonspecifically. Perhaps this is not surprising given the three binding sites observed in a single 69-mer.

 

In the end, this is a thorough but sobering paper. Despite an impressive screening campaign with multiple biophysical methods, the best ligands seem to have modest affinity and low specificity. Drugging RNA still appears much more difficult than drugging proteins. But for either sort of target, this sort of careful work will be essential to find promising leads.

Identifying ligand-binding pockets in RNA, computationally and experimentally

Most drugs bind to proteins, but RNA provides many interesting targets. Unfortunately, finding drug-like small molecules that bind to RNA is difficult. A new paper in Proc. Nat. Acad. Sci. USA from Kevin Weeks and colleagues at University of North Carolina Chapel Hill provides tools to do so.

 

RNA presents several challenges for drug discovery. First, there are far fewer high-resolution structures than there are for proteins. This is in part due to the second challenge: RNA strands are often wriggly, able to form multiple conformations. And finally, RNA is highly charged and more polar than most proteins, so there are fewer opportunities for the hydrophobic interactions that often provide significant affinity in protein-ligand complexes.

 

These challenges have not deterred intrepid investigators: Practical Fragments first wrote about targeting RNA with fragments way back in 2009. However, examples of high-affinity ligands remain elusive, and in 2023 I wondered whether “most RNA is truly undruggable.”

 

The latest paper leaves me more optimistic. It describes a computational approach to find small-molecule binding sites in RNA. The researchers started with an open-source tool called fpocket, which was built for proteins. The fpocket program places virtual spheres all around a biomolecule, where each sphere contacts the center of four atoms. The size of each sphere depends on local curvature, and clusters of spheres define pockets.

 

To benchmark fpocket on RNA, the researchers first constructed a curated database of drug-like ligands bound to RNA. Of 538 RNA-ligand structures solved at the fairly low bar of ˂ 3.5 Å resolution, only 48 ligands were deemed drug-like by the quantitative estimate of drug-likeness (QED) score. (Although the QED score may be overly restrictive, and many approved drugs have low QED scores, setting a strict threshold means that any pockets identified are likely to be particularly attractive.)

 

Using default (protein-appropriate) parameters, fpocket identified just 63% of known ligand-binding sites in RNA, vs 83% for proteins. Worse, many predicted RNA pockets probably aren’t actually ligandable because they are too exposed to solvent. By tweaking parameters, the researchers improved performance of the program for RNA to 92%, and they also identified several attractive pockets that had previously been missed.

 

When the researchers applied the reparametrized program, redubbed fpocketR, to two bacterial ribosomes, they found several dozen pockets in each, including known antibiotic-binding sites. To assess whether the new pockets could bind fragments, they used an experimental approach called Frag-MaP, which uses fully functionalized fragment (FFF) probes containing a variable fragment, a photoreactive diazirine, and an alkyne. Treating bacterial cells with these FFF probes in the presence of UV light crosslinks them to nearby RNA. Crosslinked probes can then be isolated using click chemistry with the alkyne, and RNA sequencing reveals the sites of modification. Impressively, 89% of ligand binding sites found in the Frag-MaP experiments were predicted by fpocketR.

 

In another validation experiment, fpocketR identified pockets where 7 out of 17 antibiotics bind to bacterial ribosomes. Notably, all but one of the undetected pockets bind antibiotics such as aminoglycosides that don’t appear conventionally drug-like and indeed are not orally bioavailable.

 

Continuing to apply fpocketR to more RNAs led to the identification of dozens of new pockets. Interestingly, most of these pockets occur in complex RNA structures, such as multi-helix junctions or pseudoknots, rather than simpler structures such as bulges and consecutive loops. This could explain the paucity of fragment hits in a study we highlighted in 2023, which focused on simple loops.

 

Now that we know where to find attractive ligand-binding pockets in RNA, hopefully we will be more successful finding high-affinity ligands.

Solving protein-ligand NMR structures without isotopic labeling

Last week we highlighted a protein-detected NMR method that does not require expensive and sometimes difficult isotopic labeling of proteins. However, while that approach is able to provide affinity information, it does not provide structural information. A new (open-access) paper in J. Am. Chem. Soc. by Roland Riek, Julien Orts, and collaborators at the Institute for Molecular Physical Science and the University of Vienna tackles this challenge.

 

The approach builds on NMR Molecular Replacement (NMR²), which we last wrote about here. In NMR², brute force calculations obviate the need for assigning individual NMR peaks to specific protein residues, thereby sidestepping considerable up-front effort. Most of the new paper focuses on applying NMR² to ligand discovery for the oncogenic G12V mutant of KRAS, which I’ll briefly summarize.

 

The researchers start by screening the 890-membered DSI-poised fragment library (in pools of six, with each fragment at 0.6 mM) against KRAS using ligand-detected STD NMR. This produced 133 hits, which were then retested at 1 mM each using [¹⁵N,¹H]-HSQC two-dimensional protein-observed NMR, invalidating about 30% of them. Dose-response titrations were performed on the top 13 hits; all of them were found to be weak binders, with at best low millimolar affinity. NMR² was then used to determine protein-ligand structures for some of these hits. That information guided the design of additional ligands, which had slightly higher affinities.

 

This thorough description of the NMR² workflow should be useful if you’re trying to do this at home. But what really caught my eye was a bit at the very end of the paper describing a new relaxation-filtered NOESY pulse sequence. Specifically, “an inversion recovery pulse block serves as a T₁ filter, followed by a perfect echo sequence and a CPMG without J-modulation, as a T₂ filter.” In essence, the experiment takes advantage of the fact that proteins relax more rapidly than small molecules, so NMR peaks coming from the protein are filtered out. But NMR peaks from protons in the ligand that are in close proximity to protons on methyl groups of the protein are observed, and the intensity of these peaks correlates with the distance between ligand and protein protons. Feeding these distance constraints into NMR² generates a three-dimensional structural model. The researchers compare models generated using NMR² on unlabeled KRAS to those generated using NMR² on labeled KRAS and show that they are roughly similar.

 

This is a neat approach, and it will be interesting to see whether it catches on. According to our poll last year ligand-detected NMR has fallen to fourth place among fragment-finding methods, and protein-detected NMR is in seventh place. Perhaps approaches like this and that described last week will usher in a new era of NMR for FBLD.

Fragments beat lead-like compounds in a screen against OGG1

The twin rise of make-on-demand libraries and speedy in silico docking has supercharged fragment screening and optimization: we’ve written previously about V-SYNTHES, Crystal Structure First and a related method. Another advance is described by Jens Carlsson (Uppsala University) and a large group of multinational collaborators in an (open access) Nat. Commun. paper.

 

The researchers were interested in 8-oxoguanine DNA glycosylase (OGG1), a DNA-repair enzyme and potential anti-inflammatory and anticancer target. They started with a crystal structure into which they docked 14 million fragments (MW < 250 Da) or 235 million lead-like molecules (250-350 Da) from ZINC15. Multiple conformations and thousands of orientations were sampled for each molecule. In all, 13 trillion fragment complexes and 149 trillion lead-like complexes were evaluated using DOCK3.7, a process that took just 2 hours and 11 hours on a 3500 core cluster.

 

After removing PAINS and molecules similar to previously reported OGG1 inhibitors, the top-scoring 0.05-0.07% molecules from each screen were clustered and, after manual evaluation, 29 fragments and 36 lead-like compounds were purchased from make-on-demand catalogs. These were tested at 495 µM (for fragments) or 99 µM (for larger molecules) in a DSF screen. None of the lead-like compounds significantly stabilized the protein, while several fragments did. Four of the fragments were successfully crystallized with OGG1, and in all cases the key interactions predicted in the computational screens were confirmed in the actual crystal structures.

 

Compound 1 showed the greatest stabilization of OGG1 (2.8 ºC) and some inhibition in an enzymatic assay, but not enough to calculate an IC₅₀. Searching for analogs that contained compound 1 as a substructure in the Enamine REAL database of 11 billion compounds produced few hits, but, as before, thinking in fragments proved fruitful. Searching for molecules containing just the core heterocycle and amide (colored blue below) yielded nearly 43,000 possibilities. Docking these and making and testing a few dozen led to compound 5, with mid-micromolar inhibition. Further iterations led to low micromolar compound 7.

At this point the researchers turned from make-on-demand libraries to synthetically accessible virtual libraries to fine-tune the molecule. After docking 6720 virtual molecules, they synthesized and tested 16, of which 12 were more potent than compound 7, with five of them being submicromolar. Compound 23 showed low micromolar activity in two different cell assays and was selective against four other DNA repair enzymes.

 

The same high-throughput docking approach was applied to three other protein targets: SMYD3, NUDT5, and PHIP. In each case crystal structures of bound fragments were available to use as starting points. Multiple compounds with improved docking scores compared to the initial fragments were identified, though no compounds were actually synthesized and tested.

 

The success in finding compound 1 demonstrates experimentally the advantage fragments have in efficiently searching chemical space. The researchers note that 97% of the >30 billion currently available make-on-demand compounds have molecular weights >350 Da, while only 50 million are < 250 Da. Screening all of these fragments in silico is possible; screening everything, less so. Although the fragment hits for OGG1 were weak, this isn’t always the case, as noted here. The fact that fragment 1 could be advanced to a sub-micromolar inhibitor after synthesizing just a few dozen molecules also testifies to the efficiency of in silico approaches.

 

The paper contains lots of useful details and suggestions for streamlining the process and is well worth perusing if you are trying to find hits against a structurally-enabled protein.

Review of 2024 reviews

Long winter is here in the global north, with its dim days and gaping nights. As is tradition, Practical Fragments looks back on the year almost done. 2024 was the best year for conferences since the arrival of COVID. I wrote about CHI’s Drug Discovery Chemistry in San Diego, FBDD-DU in Brisbane, FBLD 2024, and CHI’s Discovery on Target, both in Boston. 

Another tradition is the annual J. Med. Chem. fragment-to-lead success story review; the latest covers the year 2022 and was written by Andrew Woodhead (Astex) and collaborators, including yours truly.

 

For a timely and accessible overview of “how to find a fragment,” look no further than a review of that title in ChemMedChem by Marcio Vinicius Bertacine Dias and collaborators at University of São Paulo and University of Warwick. This covers crystallography, cryo-EM, NMR, SPR, thermal shift, virtual screening, functional screening, ITC, mass spectrometry (including HDX-MS), MST, and BLI, and concludes with a nice comparison table. Some twenty other reviews were also published throughout the year, and these are discussed thematically.

 

Structure-based methods

Three reviews cover NMR. In J. Med. Chem., Janet Caceres-Cortes and colleagues at Bristol-Myers Squibb provide “perspectives on nuclear magnetic resonance spectroscopy in drug discovery.” Ligand- and protein-detected screening are covered thoroughly, with examples such as the discoveries of BI-2852 and venetoclax. Applications beyond hit finding are also discussed, such as the characterization of atropisomers in sotorasib, the identification and characterization of impurities and metabolites, in-cell NMR, and much more.

 

“Perspectives on applications of ¹⁹F-NMR in fragment-based drug discovery” is the title of an open-access review in Molecules by Qingxin Li and CongBao Kang at Guangdong Academy of Sciences and A*STAR, respectively. As we discussed in 2020, fluorine NMR is becoming increasingly common in FBLD, and this paper covers the various methods, including using ¹⁹F-NMR to measure ligand affinity. The authors also include a table summarizing 17 fragment screens that used fluorine NMR.

 

The rise of powerful permanent magnets has enabled low-maintenance benchtop NMR instruments that can be yours for as little as $50,000, compared to upwards of $1 million for a 600 MHz superconducting machine. Although sensitivity is at least 150-fold lower, hyperpolarization techniques such as photo-CIDNP, which we wrote about here, can close the gap. The latest developments are described (open access) in Chemistry–Methods by Felix Torres and collaborators at NexMR and the ETHZ.

 

X-ray crystallography has retained the top position among fragment-finding methods according to our most recent poll. In an open-access Applied Research paper, Daren Fearon, Frank von Delft, and collaborators describe high-throughput crystallographic fragment screening at the Diamond Light Source. As of August 2024 they have collected more than 240,000 academic data sets on hundreds of targets, and the paper distills some of the key lessons, some of which were applied to the COVID Moonshot, which we last wrote about here. The paper also describes future developments and needs, such as how and where to house such massive quantities of data.

 

Another center for high-throughput crystallographic screening is the Helmhotz-Zentrum Berlin (HZB) F2X-Facility at the BESSY II synchrotron, and in an open-access Applied Research paper Manfred Weiss and collaborators provide an overview of workflows and capabilities. One unique offering is the F2X-GO kit, in which F2X fragment libraries (which we wrote about here) as well other supplies are shipped to users to do soaking experiments in their own laboratories prior to shipment to the synchrotron.

 

“Structure-based virtual screening of vast chemical space” is the topic of an open-access review in Curr. Opin. Struct. Biol. by Jens Carlsson (Uppsala University) and Andreas Luttens (MIT). The Enamine REAL collection currently contains 40 billion compounds, and the researchers predict that “make-on-demand chemical libraries will likely reach more than one trillion compounds in the next few years,” which presents both opportunities and challenges, particularly given the existence of the “virtual cheaters” we recently discussed. Machine learning and fragment-based methods such as V-SYNTHES could help.

 

Continuing the virtual theme, Li Wang and collaborators at Nantong University discuss “molecular fragmentation as a crucial step in the AI-based drug development pathway” in an open-access Commun. Chem. paper. This summarizes 15 different computational methods for dissecting larger molecules into fragments, and also includes a list of 11 library vendors.

 

Other methods

Among experimental methods, few can match the throughput of fluorescence techniques, the subject of an open-access Heliyon review by Neelagandan Kamariah and colleagues at inSTEM & NCBS in Bangalore. It has short sections on “fluorescence polarization (FP) and anisotropy (FA), Förster resonance energy transfer (FRET), time-resolved Förster resonance energy transfer (TR-FRET), fluorescence lifetime (FLT), protein-induced fluorescence enhancement (PIFE), fluorescence thermal shift assay (FTSA) and microscale thermophoresis.” It also describes applications to GPCRs, protein-protein interactions, and other biological systems.

 

Native mass spectrometry (nMS), which we wrote about most recently in 2022, is the subject of an RSC Med. Chem. review by Louise Sternicki and Sally-Ann Poulsen at Griffith University. Sally-Ann is a leading expert in nMS, and the paper describes the technique and how it compares to other fragment-finding methods. It also includes a nice table summarizing 17 studies published between 2013-2023 that used nMS for FBLD; a 2013 review of nMS covered earlier examples.

 

Covalent Fragments

Mass spectrometry plays a prominent role in finding covalent fragments, as discussed in an open-access SLAS Discovery review by Simon Lucas and colleagues at AstraZeneca. “Covalent hits and where to find them” also describes other biophysical, biochemical, cellular approaches, and even DEL screening. It also discusses covalent libraries (which we wrote about earlier this month) and successful examples such as the discovery of sotorasib. In my opinion the researchers succeed in their “hope that this review will help serve as a useful roadmap to those seeking to drug the undruggable.”

 

A concise open-access review in Curr. Opin. Struct. Biol. by Katrin Rittinger and collaborators at The Francis Crick Institute and GSK focuses on using covalent fragments to assess target tractability, specifically ligandability and functionality. Target-based, proteome-wide and function-first approaches are summarized, and the researchers also discuss the importance of negative control compounds such as inactive enantiomers.

 

Continuing the theme of “assayability,” Micah Niphakis (Lundbeck) and Ben Cravatt (Scripps) review “ligand discovery by activity-based protein profiling” (ABPP) in a Cell Chem. Biol. paper. Because ABPP is usually conducted in cells or cell lysates, full-length proteins are assayed in their native environment, facilitating the discovery of allosteric ligands as in the case of WRN, which we wrote about earlier this year. The paper summarizes multiple examples of finding covalent ligands for challenging targets, and also highlights future challenges such as increasing throughput and targeting residues beyond cysteine.

 

Reversible covalent inhibitors are the topic of an open-access review by Dustin Duncan and colleagues at Brock University in ACS Chem. Biol. The researchers argue that reversible covalent inhibitors may cause less accumulation of the off-target adducts that could form with irreversible inhibitors. The paper includes a figure showing reversible covalent warheads, details on how to characterize them, a nice summary of general considerations, and success stories for JAK3, BTK, and proteases.

 

Targets

Covalent ligands have been particularly important for cancer targets, as reviewed by Xiaoyu Zhang (Northwestern University) and Ben Cravatt (Scripps) in an open-access Annual Review of Cancer Biology paper. The focus is on use of chemical proteomics “to expand the druggability of cancer proteomes.” The paper presents examples of finding and characterizing covalent ligands for a variety of oncology targets including KRAS^G12C.

 

E3 ligases are briefly mentioned, and this target class is the focus of an Expert Opin. Drug Discov. review by Jongmin Park and colleagues at Kangwon National University. As we’ve discussed recently, the 600 or so human E3 ligases are potentially valuable for targeted protein degradation applications such as PROTACs. The review focuses on “fragment-based approaches to discover ligands for tumor-specific E3 ligases.” In addition to summarizing successes against targets such as BCL6 and XIAP, it includes a list of 113 tumor-specific E3 ligases and another list of 52 E3 ligases that are overexpressed in certain tumors.

 

The “impact of fragment-based drug design on PROTAC degrader discovery” is also the subject of a review in Trends in Analytical Chemistry by Xiaoguang Lei and colleagues at Shenzen Bay Laboratory. Here, the focus is more on using FBLD to discover ligands for target proteins rather than for E3s. For example, the researchers describe how the fragment-derived drug navitoclax was used as a starting point for developing DT2216, a clinical-stage BCL-x_L degrader.

 

As Vicki Nienaber noted more than a decade ago, fragment-based drug discovery is ideally suited for targeting the central nervous system, particularly when combined with a ruthless focus on molecular properties. This is the topic of an open-access review in Front. Chem. by Michael Kassiou and collaborators at University of Sydney, CSIRO, and Vast Bioscience. After a brief summary of FBLD the researchers present case studies published since 2015, the last year this topic was reviewed. We’ve covered quite a few on Practical Fragments, including apoE4, Notum, and PDE10A.

 

Other

We mentioned allostery above, and in an open-access FEBS Open Bio. article Andrea Bellelli and collaborators at Sapienza University of Rome and elsewhere ask “is allostery a fuzzy concept?” Digging into half-century old publications from Jacques Monod, the researchers conclude that the concept was “born with an original sin: two definitions.” Indeed, the first mathematical model did not even apply to monomeric proteins. Most readers of this blog will probably be satisfied with the notion that an allosteric ligand is one that binds outside of an active site, but it is worth remembering that “allostery is an umbrella that covers more than a single reaction mechanism and cannot be defined by a single mathematical expression.”

 

Structure comes up frequently on Practical Fragments, but James Fraser (UCSF) and Mark Murcko (Disruptive Biomedical) remind us in Cell that “structure is beauty, but not always truth.” We’ve written multiple posts about getting misled by crystal structures, and in this brief commentary the authors provide “four harsh truths: a structure is a model, not experimental reality; representing wiggling and jiggling is hard; in vitro can be deceiving; drugs mingle with many different receptors.” They conclude that “truth is a molecule that transforms the practice of medicine.”

 

Stumbling towards truth is a little easier with the help of a good chemical probe, and in Nucleic Acids Res. Paul Workman (Institute of Cancer Research) and collaborators provide an updated description of The Chemical Probes Portal. This free community resource now contains 803 probes against 570 targets, including 28 covalent ligands and 51 degraders. Moreover, 332 of the probes have structurally related negative controls. Importantly, the Portal also includes 258 “Unsuitable” compounds that are insufficiently potent or selective to serve as chemical probes. Checking this list can save you valuable time when reading papers about unfamiliar targets.

 

Finally, a brief open-access interview with Nobel Laureate Katalin Karikó in Issues in Science and Technology is an inspiring reminder that “you learn more from failure,” and that the pleasure of doing science can be its own reward.

 

Thanks for reading. Good luck in 2025, and remember that the sun is always out there, even when you can neither feel nor see it.