Ligand reactivity efficiency (LRE)

As covalent drug discovery continues to rise, the demand for metrics to help guide lead optimization is increasing. Last year we discussed covalent ligand efficiency (CLE). In an open-access paper just published in J. Med. Chem., Benjamin Horning, Brian Cook, and colleagues at Vividion Therapeutics describe ligand reactivity efficiency (LRE). (Benjamin presented LRE at the DDC meeting in 2024.)

 

A key challenge when developing covalent ligands is maximizing specific reactivity towards the target of interest while minimizing intrinsic reactivity towards other proteins; the two types of reactivity are not the same, as we wrote about last year. For molecules that target cysteine residues, intrinsic reactivity is usually determined by assessing reactivity against the small molecule glutathione, which is abundant in cells.

 

For lead optimization more generally, a common metric is lipophilic efficiency (LLE or LipE, see here and here), in which the logP of a molecule is subtracted from the negative log of the IC₅₀ (pIC₅₀). More lipophilic molecules have higher logP values, so maximizing LLE helps to minimize increases in lipophilicity.

 

By analogy, the researchers defined LRE to help minimize increases in intrinsic reactivity. However, distinguishing specific from intrinsic activity is not necessarily straightforward. As we previously discussed, IC₅₀ alone is an inappropriate measurement for covalent inhibitors; the incubation time before the IC₅₀ is measured is an essential variable. The most rigorous value is k_inact/K_I, and although this ratio has been historically time-consuming to determine, we described an easier method earlier this year. Yet an even simpler measurement is the TE_{50(target, 1h)}, the concentration of compound necessary to label 50% of a target after one hour, which is a function of k_inact/K_I. The researchers thus defined LRE as:

 

    LRE = pTE_{50(target, 1h)} – pTE_{50(GSH, 1h)}

 

The variable in the second term, pTE_{50(GSH, 1h)}, is calculated from the reaction rate of the ligand with glutathione; intrinsically reactive ligands have higher rates.

 

In the case of LLE, values above 5 or 6 are generally considered acceptable for advanced leads, and the same is true for LRE. For example, a molecule with TE_{50(target, 1h)} = 10 nM and a (low) GSH reactivity of 0.01 M^-1s^-1 would have an LRE = 6.3. Also analogous to LLE, one can generate plots with pTE_{50(GSH, 1h)} on the x-axis and pTE_{50(target, 1h)} on the y-axis to assess whether LRE values are improving during a lead optimization campaign.

 

In my view, LRE is superior to previously discussed CLE because it explicitly considers the time component. A one hour incubation is practical; a ligand with k_inact/K_I = 10,000 M^-1s^-1 would have TE_{50(target, 1h)} = 19 nM. Also, LRE is more intuitive for medicinal chemists than CLE due to its similarity to LLE.

 

On the minus side, the researchers note that some of the assumptions break down for ligands with high non-covalent affinity (low K_I). Also, some folks may take issue with metrics that take the logarithm of a measurement that has units.

 

The researchers note another alternative metric, the reactivity enhancement factor (REF), which I briefly discussed here. REF is simply the ratio of the specific reactivity to the intrinsic reactivity, which is conceptually simpler to me than LRE. Nonetheless, the researchers state that LRE is commonly used at Vividion, which has put several covalent drugs into the clinic, so clearly it can be useful. Whether REF, LRE, or CLE, ultimately the choice of metric is less important than the ultimate goal: maximizing specific reactivity while minimizing intrinsic reactivity.

Malicious metals muddy fragment-to-lead optimization

Despite the effectiveness of vaccines against SARS-CoV-2, COVID-19 continues to plague us. The handful of approved small molecule drugs target only two proteins and have much room for improvement. One interesting but underexplored target is the nonstructural protein 14 (NSP14), a 3’ to 5’ RNA exonuclease, which is important both for viral replication as well as immune escape. In a new open-access ACS Chem. Biol. paper, Jae Jung, Shaun Stauffer, and colleagues at the Cleveland Clinic describe how their efforts against NSP14 were thwarted.

 

The researchers started with the crystal structures of two fragments that had been identified in a high-throughput crystallographic screen at XChem. They reproduced these in-house, confirming the published structures, and also made and characterized a few analogs. Crystallography demonstrated these bind in a similar manner. Encouragingly, they also showed activity in a biochemical assay.

 

The two published fragments bind next to one another, presenting a good opportunity for fragment merging or linking. The researchers used the computational tool Fragmentstein, which we wrote about here, to design new molecules. Some of these molecules were active in the biochemical assay, and a crystal structure of a merged compound revealed that it bound as expected. Importantly, none of the molecules inhibited an unrelated endonuclease.

 

So far, so good, but the researchers were suspicious about the SAR. For example, changing an isopropyl to a cyclopropyl group weakened the activity  from 2.4 to 150 µM, despite the fact that the moiety is largely solvent exposed. After resynthesizing and more carefully purifying the molecules, the researchers found them to be completely inactive in the biochemical assay.

 

NSP14 contains two catalytic magnesium ions and three structural zinc ions, and the researchers considered the possibility that metal contaminants might have been responsible for the activity. Sure enough, when they screened the Metal Ion Interferences Set (MIIS), which we wrote about here, they found that half a dozen metal ions potently inhibited the assay. They tested whether any of the spuriously active compounds contained palladium and ruled this out, but did not test for other metals. Indeed, metals may not even be to blame: the active molecules all contain thiazoles, and as we discussed in 2022 these can sometimes interfere with assays. What is clear is that the exciting initial activity results were artifacts, and the researchers were sufficiently diligent to figure it out for themselves.

 

One of the most disturbing findings is that the crystal structures looked fine, despite the compounds having no measurable activity. As we’ve written previously, the lack of affinity information is the biggest drawback of fragment screening by crystallography. Perhaps NMR would have been able to invalidate the false-positives, though as we have written both protein-detected and ligand-detected methods can be fooled. As our 2024 poll emphasized, using multiple methods to validate fragment binding is important. And resynthesizing and carefully purifying compounds helps too.

 

These sorts of cautionary tales are not published as often as they should be. Kudos to this team for both warranted skepticism and providing a warning for others.

Selectivity in cells may vary

Last year we celebrated the ten-year anniversary of the Chemical Probes Portal. One of the key requirements for a chemical probe is selectivity, which was set to >30-fold vs related targets when the Portal launched in 2015. For enzymes such as kinases, selectivity is often measured in cell-free assays. A new open-access J. Med. Chem. paper by Matthew Robers, Alison Axtman, and collaborators at Promega and University of North Carolina at Chapel Hill suggests that such data don’t necessarily translate to cellular assays.

 

Kinases are one of the most heavily mined classes of targets this century; five of the eight FBLD-derived approved drugs target kinases. With more than 500 in the human proteome, selectivity has long been a focus. One common method for assessing selectivity in cell-free assays is the Eurofins DiscoverX panel, which currently includes more than 450 kinases. Each kinase has a DNA tag and is paired with a promiscuous high affinity binder attached to a solid support. Test compounds are added, and qPCR is used to assess and quantify which kinases are displaced. The competition assay allows determination of dissociation constants.

 

To measure the binding of compounds to kinases in living cells, the researchers turned to the NanoLuc-bioluminescence resonance energy transfer (NanoBRET) assay. This is also a displacement assay that relies on a bivalent molecule containing a kinase ligand and a fluorophore. Kinases are tagged with NanoLuc, which causes luminescence of the fluorophore when it is in close proximity (ie, bound to the kinase). Ligands that bind to the kinase displace the bivalent molecule, decreasing luminescence.

 

The researchers started with four promiscuous kinase inhibitors, two of which (dasatinib and sorafenib) are approved drugs. They ran these against 240 or 300 kinases in the NanoBRET assay and compared the values with published DiscoverX dissociation constants. Most of the compounds were more potent in the DiscoverX assay than in the cell-based assay, and the researchers suggest several possible reasons. First, the DiscoverX assay is run in the absence of the cofactor ATP, which can compete with ligands that bind to the active site. Second, cell (im)permeability could decrease cellular potency. Finally, most of the DiscoverX kinases are truncated, whereas the NanoBRET kinases are full length.

 

For these and other reasons, it is common for compounds to be less active in cell assays than biochemical or biophysical assays. Surprisingly though, for a few kinases the compounds were actually more potent in the cellular assay than they were in the DiscoverX assay.

 

To extend these findings, the researchers tested additional kinase inhibitors and found that three kinases were particularly susceptible to inhibition in cells. One of these kinases, PIP4K2C, was engaged at mid nanomolar potency by cabozantinib, and the researchers suggest this could be useful for immuno-oncology. More worrisome, several approved drugs bind to the tumor suppressor STK11 in cells, raising the potential that these compounds could inhibit exactly the wrong pathway.

 

On the bright side, the researchers find that some molecules that look moderately selective in the DiscoverX assay are actually quite selective in cell assays, and they propose new chemical probes for the little-studied kinases BRSK1/2 as well the kinases DDR1/2.

 

Kinases are certainly not the only class of targets for which compounds’ performance differs outside vs inside cells; we wrote about covalent WRN inhibitors here. This paper is a good reminder that as useful as cell-free assays are, things can go weird once you go into cells – for better or for worse.

Best practices for applying HDX-MS to FBLD

Among the many biophysical techniques presented at the recent Novalix meeting, hydrogen/deuterium exchange mass spectrometry (HDX-MS) was mentioned only a few times. Practical Fragments last covered it nearly four years ago in the context of a paper that described theoretical challenges and improvements to the method for assessing fragment binding modes. A new open-access Comm. Chem. paper from Tiago Bandeiras, Alessio Bortoluzzi, and collaborators at iBET-Instituto de Biologia Experimental e Tecnológica and Merck KGaA implements some of these suggestions.

 

The researchers find only two examples where HDX-MS had been applied to fragments, one of which we covered here. The new paper focuses mostly on Cyclophilin D (CypD), a mitochondrial protein implicated in several diseases. A screen we wrote about in 2020 described several fragment hits, some of which were crystallographically found to bind in three overlapping sites designated the proline pocket, the aniline pocket, and the pyrazolo pocket.  The researchers chose a binder from each pocket to study as well as two more fragments whose binding sites were not known. All five fragments are extremely weak hits, with at best 7 mM (yes, millimolar) affinity as assessed by SPR, making this a particularly challenging test case.

 

As a reminder, HDX-MS examines the exchange of deuterium from D₂O to protein backbone amides. Nearby ligands can slow this exchange, and mapping the locations and magnitude of these changes reveals where the ligand binds. Optimization experiments were initially run on a compound with a K_D of 44 mM for the aniline pocket. Three protein concentrations were tested. Since the highest (10 µM) yielded the highest number of detected peptides, it was used for subsequent experiments.

 

As suggested in the 2022 paper, compound was added both to the initial protein solution as well as to the deuterium exchange buffer. The fragment was tested at 2.5, 5, and 10 mM, and changes in deuterium uptake were found in all cases, which is remarkable given that the theoretical occupancies range from 5 to 18%. An experiment using substoichiometric concentrations of a high affinity ligand confirmed that 18% protein occupancy is sufficient to generate a reliable signal.

 

Still, given the low occupancy, the researchers used statistical methods to ensure that changes in signals were significant. Mapping those that were onto the structure of the protein confirmed that the fragment bound to the aniline pocket.

 

A second fragment was tested by HDX-MS, and the results confirmed that it binds in the pyrazolo pocket, as previously shown by crystallography. However, for a third fragment that had been shown to bind in the proline pocket, the results suggested instead that it binds in the aniline pocket. The proline pocket exhibited very low levels of deuterium exchange, but when the researchers increased the pH from 7.4 to 9 they did find evidence that the fragment binds here as well.

 

Next, the researchers turned to the two fragments with unknown binding sites. HDX-MS revealed that these bind to the aniline binding site, though with subtly different protection patterns suggesting that they bind in slightly different regions of the pocket.

 

Finally, the researchers used their optimized HDX-MS conditions on a previously identified fragment that binds the kinase FAK with low millimolar affinity. This showed that it binds in the so-called hinge region, in agreement with crystallography. This fragment was also used as a negative control for CypD, showing that it does not bind.

 

This paper is a nice resource for those hoping to apply HDX-MS to fragments. The fact that the binding sites of such weak binders can be determined is quite remarkable. That said, the resolution is not as good as crystallography or protein-detected NMR; for FAK in particular, the ligand reduces deuterium update across a large fraction of the protein surface. And the fact that a proline-pocket binder was initially mapped to the aniline pocket also gives one pause. Perhaps the binding mode in solution really is different from that found crystallographically. It would be interesting to see whether NMR can resolve the conundrum.

Twelfth Novalix Biophysics in Drug Discovery Conference

Last week the Twelfth Novalix Biophysics in Drug Discovery Conference was held for the first time in La Jolla, California. It’s been several years since I wrote about one of these, and I was happy to see that they’ve maintained their reputation for excellent science and convivial conversation. There’s no way to cover the two-dozen talks, but here are a few highlights.

 

One of the things I most enjoy about these meetings is learning about new and emerging methods, and these were well represented. Chris Brosey (AbbVie) discussed time-resolved high-throughput small-angle X-ray scattering (TR-HT-SAXS). As I discussed a couple years ago, the approach can be used to measure the kinetics of protein dimerization in response to fragment-sized ligands.

 

SAXS-based approaches typically require access to a synchrotron, but Takashi Sato (Rigaku) described a related approach, electron density tomography (EDT), using an in-house instrument. Using machine learning, EDT can provide more detailed structural information than standard SAXS, and Takashi provided examples for samples ranging in size from viruses to single-chain variable fragments (scFvs) smaller than 30 kD.

 

Another approach to examine protein complexes is microfluidic diffusional sizing (MDS), described by James Wilkinson of Fluidic Sciences. By assessing the amount of diffusion in disposable chip-based chambers, MDS can determine how the hydrodynamic radius changes in response to ligands. Each chip holds 24 samples, and data can be collected in less than an hour. The minimum observable size change is 5-10% so measuring small molecules directly is unlikely. Still, the technique is useful for observing induced proximity events such as those caused by molecular glues, and it is sufficiently robust that it can be run in pure serum.

 

Among solution-based methods, none have achieved such recent prominence as cryo-EM. Weiru Wang, my colleague at Frontier Medicines, described how this technique was used iteratively to design and characterize bivalent degrader molecules that covalently exploit the E3 ligase DCAF2. (We recently published this work in Structure.)

 

Cryo-EM has revolutionized the types of biological molecules that can be structurally characterized. According to Denis Zeyer (Novalix), the technique accounted for 40% of PDB entries last year. However, despite the “resolution revolution,” most structures are not as detailed as those from X-ray crystallography; Denis noted that only 20% of the new structures were solved to a resolution better than 3 Å, which may have negative implications for machine-learning methods trained on these lower resolution structures.

 

If cryo-EM is the new kid in town, NMR is the grizzled veteran. But proving that it is possible to find new applications for old methods, Matthew Eddy (University of Florida) described a clever ¹⁹F-labeling approach for GPCRs in nanodiscs to quantify the distribution of various states in response to anionic lipids and ligands. This has allowed him to distinguish between antagonists and inverse agonists, which can be difficult using cell-based assays.

 

Turning from solution-based to surface-based methods, SPR has moved into the number two slot for fragment-finding, as we noted in our recent poll. Just as there are new tricks for NMR, the same applies to SPR. Matthew Peterfreund (Bruker Biosensors) described switchSENSE, a fluorescence proximity assay built on an SPR chip that is useful for measuring the binding and kinetics of bifunctional ligands such as PROTACs to two or more proteins. He also introduced the Triceratops SPR#64 instrument, which as its name implies supports 64 sensor spots.

 

Kris Borzilleri (Pfizer) discussed SPR-microscopy (SPRm), which combines an optical microscope with an SPR instrument. This can be used to measure the affinities of ligands binding to receptors in cells grown on SPR chips, and Kris described applications to membrane proteins such as GPCRs and solute carriers. The technique is still quite slow though, at only 10-15 compounds in duplicate per week.

 

Another surface-based approach to screening cells was described by Volker Gatterdam of Lino Biotech. Focal molography relies on changes in diffraction from nanoengineered diffraction gratings, called molograms. Targets, which can include living cells, are immobilized to the molograms, and analyte is flowed over. The instrument contains 64 spots, and assays can be run in complex samples such as tissue lysates.

 

Covalent drug discovery also made an appearance, with talks by Landon Whitby (Lundbeck) and Ben Cravatt (Scripps). Landon provided an overview of chemoproteomics techniques to screen ligands in cells, such as those we wrote about in 2016. Ben continued the theme, including several success stories, and also discussed challenges for finding cryptic ligandable pockets. Despite impressive progress with machine learning, Ben noted that these methods often find only common solutions, while empirical chemoproteomics methods can find rare types of pockets.

 

Of course, as we’ve repeatedly emphasized, biophysics methods are best used in combination, as noted by Daniel Harki (University of Minnesota) and Ann Boriack-Sjodin (Takeda). Daniel presented a screen of 1056 fragments against the cancer target APOBEC3 using NMR and SPR. This yielded just a single validated hit, which interestingly turned out to be the same fragment found against KRAS in a paper we discussed in 2022. And Ann described how biophysics led to multiple clinical compounds against a variety of targets at Epizyme and Accent Therapeutics.

 

I’ll stop here, but please feel free to add your thoughts. And while the date for the next Novalix conference has not yet been scheduled, the location has, with a return to beautiful Strasbourg. Vive la biophysique!

3D fragments vs the histamine H1 receptor

Last month we highlighted a paper from Iwan de Esch and colleagues at Vrije Universiteit Amsterdam about assessing molecular shapeliness. In a new open-access RSC Med. Chem. paper, de Esch and colleagues use a small, shapely library to find and develop potent antagonists of the histamine H₁ receptor (H₁R).

 

The G protein-coupled receptor H₁R is one of four histamine receptors and the target for dozens of antihistamines used for allergic and inflammatory conditions. Thus, it’s well understood and a nice model system.

 

In 2013 we discussed how the de Esch group screened a library of 1010 fragments against several targets, including H₁R. In the new paper, the researchers built a smaller library of just 80 compounds designed to be more shapely, as assessed by Fsp³, plane of best fit (PBF), and principal moment of inertia (PMI). A screen against H₁R using a radioligand displacement assay at 10 and 30 µM yielded a single hit, compound 1a.

 

Compound 1a deviates slightly from the rule of three due to some unusual elements: an azide as a synthetic handle and a butyl moiety to make the compound less explosive. Trimming these features led to compound 3a, a rule-of-three compliant fragment that – while having lower affinity – mostly maintained ligand efficiency and improved lipophilic ligand efficiency (LLE). Fragment growing led to compounds 13a and 13k, and merging with previously reported molecules led to compound VUF26691, with low nanomolar affinity and picomolar antagonistic activity in a cell assay. Throughout the process the cis isomers generally had equal or better potency than the trans isomers, an empirical observation difficult to explain by modeling.

 

This is a nice fragment-to-lead story, and the researchers note that fewer than 40 compounds were synthesized in the journey from compound 3a to VUF26691. Interestingly though, compound 1a was the only hit from the 80 compounds tested, while the hit rate from the larger, flatter library screened in 2013 was nearly three-fold higher, at 3.6%. Although it’s difficult to extrapolate from n=1, the results are consistent with our post last year that more shapely libraries will likely have lower hit rates.

 

Still, this shapely hit makes for a neat scaffold, and if nothing else perhaps it will be easier to establish intellectual property.

Multivalent fragments in the clinic: Muvalaplin

It’s been a couple years since Practical Fragments last updated our “fragments in the clinic” list. Before doing so it makes sense to highlight some of those we’ve missed. Let’s start with an open-access Nature paper from Laura Michael and collaborators at Lilly and Monash University published in 2024. Truth be told I’ve been waiting for a longer discovery paper, but I’ll go with what’s available now.

 

The researchers were interested in lipoprotein(a), or Lp(a), which has been linked to cardiovascular diseases. Lp(a) forms when low-density lipoprotein (LDL) binds to apolipoprotein(a), or apo(a). This is a two step process, in which the ten subtypes of so-called Kringle IV (K_IV) domains in apo(a) bind to lysine residues on LDL, followed by disulfide bond formation between apo(a) and LDL. Blocking the first step in this process should reduce levels of Lp(a).

 

Here's the only description of the initial screen: “Biochemical and biophysical compound screens using purified apo(a) K_IV7-8 protein identified interacting small molecules. Optimization of the initial binding molecules led to…LSN3353871.” Whatever the details, LSN3353871 is unequivocally a rule-of-three compliant fragment. It is also a very ligand-efficient binder with high nanomolar affinity for the K_IV8 domain. LSN3353871 disrupted the formation of Lp(a) in vitro at low micromolar levels and decreased levels of Lp(a) in cynomolgus monkeys when dosed orally.

 

As noted above, the apo(a) protein contains multiple K_IV domains, and a classic method for improving potency is by making dimeric ligands that can bind to two domains simultaneously. The researchers did just this in the form of LSN3441732, which binds to apo(a) and disrupts formation of Lp(a) in vitro at picomolar concentrations.

 

If dimeric ligands are better than monomeric ones, why not go for multimeric ligands? The trimeric molecule LY3473329, or muvalaplin, was synthesized and crystallographically shown to bind to three copies of K_IV8. It blocked formation of Lp(a) in vitro and reduced Lp(a) levels in cynomolgus monkeys.

 

Kringle domains are found not just in apo(a) but also in plasminogen, the zymogen form of plasmin, which is responsible for degrading blood clots. Fortunately, subtle differences between the Kringle domains in apo(a) and human plasminogen provide selectivity for the former protein, especially for multivalent ligands such as muvalaplin, and a phase 1 clinical study showed that Lp(a) could be lowered without affecting plasmin activity.

 

This is a nice application of applying fundamental multivalent principles to develop a potent molecule. It is also another example of a molecule that may not look like a drug but works like one: despite containing four basic nitrogen atoms, three carboxylic acid moieties, and sporting a molecular weight above 700 Da, muvalaplin is orally bioavailable. It is currently in a phase 3 trial in up to 10,450 patients. Cardiovascular disease is the leading cause of death in the developed world, and Practical Fragments wishes luck to everyone involved in these studies.

 

In the meantime, watch for more Practical Fragments posts on new entries to our fragments in the clinic list, which will be updated later this year.

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.”

Fragment merging – and flipping – on the leucine zipper of MITF

Transcription factors can be difficult drug targets, particularly those whose primary structure is a “leucine zipper” in which two α-helices gently coil around each other. Their three-dimensional structure provides few pockets suitable for binding small molecules. In a new (open-access) paper in Nat. Commun., Deborah Castelletti, Wolfgang Jahnke, and a large group of multinational collaborators at Novartis and elsewhere present progress toward one of these, microphthalmia-associated transcription factor (MITF), which has been implicated in melanoma.

 

Most of MITF is believed to be disordered, but the DNA-binding domain (DBD) homodimerizes as a basic helix-loop-helix leucine zipper. Unlike related transcription factors, the helices in MITF contain a small kink that keeps them from heterodimerizing and also creates a small “kink pocket.”

 

The researchers expressed the DNA-binding domain of MITF and screened it using ¹⁹F NMR against the LEF4000 library, which we described here. This yielded just 9 hits that confirmed in protein-observed NMR, a hit rate the researchers note “is amongst the lowest that we have observed across multiple FBS campaigns,” consistent with expectations for a difficult target. Two chemical series, represented by compounds 1 and 2, were prioritized, and analogs from the Novartis compound collection were screened to find more-potent compounds 3 and 4.

 

Crystallography revealed that compounds 3 and 4 both bound in the kink pocket. Excitingly, the binding modes are similar and overlapping, inviting fragment merging. This proved successful, yielding a compound that bound 100-fold more tightly than either fragment. Further optimization ultimately led to compounds 7 and 8, with low or sub-micromolar affinity as assessed by isothermal titration calorimetry (ITC).

 

The bound structures of compounds 7 and 8 were determined by crystallography. Compound 7 (gray, left) superimposes nicely onto compounds 3 (cyan) and 4 (magenta), showing successful fragment merging. Compound 8 (green, right), however, is flipped 180 degrees compared to compound 7, despite having similar structure and affinity. Although surprising, this is not too uncommon; we’ve written about previous flippers here, here, and here.

The MITF homodimer is asymmetric, with one helix kinked and the other straight. NMR experiments and molecular dynamics show that both compounds 7 and 8 slow the interconversion between kinked and straight forms, though it is unclear whether this has functional implications. The compounds do not seem to affect DNA binding, and with at best high nanomolar affinity towards MITF no cell data are reported with the molecules.

 

Nonetheless, the successful identification of ligands against a leucine zipper is exciting. The binding pocket is small; as shown in the figure above, the best compounds already stick out on either side of the helices. Further affinity improvements may be difficult, though perhaps covalent approaches could help. Alternatively, perhaps these molecules could be starting points for induced proximity strategies such as PROTACs. It will be fun to watch this story develop.

How best to assess molecular shapeliness?

The shape of a molecule influences its properties. While this is true on a per-compound level, things get a little more controversial when discussing molecules in general. Back in 2009 researchers argued that “three dimensional” molecules have better drug-like properties, though this assertion has been challenged, repeatedly. But how do you assess the shape of a molecule in the first place? In a recent (open-access) Drug Discov. Today paper, Iwan de Esch and collaborators at Vrije Universiteit Amsterdam compare the main metrics.

 

The researchers focus on three metrics: fraction of sp³-hybridized carbons (F_Csp³), which we wrote about here; plane of best fit (PBF), which we wrote about here; and principal moment of inertia (PMI), which we wrote about here. F_Csp³, which ranges from 0 to 1, is simple to calculate based on the chemical structure alone, while the other two metrics rely on the three-dimensional shape of the molecule, requiring calculations and indeed choices since many molecules can assume multiple conformations. PBF is measured in angstroms with a minimum of 0 Å and no maximum; a protein, for example, could easily have a PBF above 10 Å. PMI is represented by two normalized PMI ratios, and these are often added to give a number (3D Score or ΣNPR) between 1 and 2.

 

The researchers calculated F_Csp³, PBF, and ΣNPR for a set of nearly half a million commercially available fragments which we discussed here; PBF and ΣNPR were calculated based on the single lowest energy conformation for each molecule. As noted above, PBF is somewhat size-dependent. For example, adamantane and buckminsterfullerene have PBF scores of 0.79 and 1.76 Å but identical ΣNPR scores. Nonetheless, the researchers found a correlation between these two metrics, and this correlation increased when PBF was divided by the root of the molecular volume to attempt to normalize for size.

 

In contrast, no correlation was found between F_Csp³ and PMI, making the former “a poor descriptor for predicting 3D molecular shape.” Is there a simple alternative? F_Csp³ only considers carbon atoms, so the researchers proposed F_HAsp³, which includes nitrogen, oxygen, and sulfur atoms. Perhaps not surprisingly, this didn’t improve the correlation.

 

Three years ago we wrote about “spacial scores,” which were developed to assess molecular complexity. The researchers calculated normalized spacial scores (nSPS) for their set of compounds, but these also showed no correlation to PMI.

 

The researchers conclude that, “once corrected for size, PBF captures three-dimensionality similarly to ΣNPR values. However, unlike a PMI analysis, it is not capable of further distinguishing between rod- and disc-shaped molecules, giving PMI a higher resolution in capturing shape diversity.” Interestingly, this is the opposite conclusion of an analysis Teddy wrote about in 2014. My take is that, if you want to assess shapeliness, steer clear of F_Csp³, but both PBF and PMI are fine.

Fragment events in 2026

Lots of interesting events coming up this year - hope to see you at one!

February 17-19:  The Twelfth NovAliX Conference will be held for the first time in San Diego! You can read my impressions of the 2018 Boston event here, the 2017 Strasbourg event here, and Teddy's impressions of the 2013 event here, here, and here. 

 

April 13-16: CHI’s Fragment-Based Drug Discovery turns 21, old enough to legally drink in the US! The longest-running annual fragment event returns as usual to San Diego. This is part of the larger Drug Discovery Chemistry meeting. You can read impressions of the 2025 meeting, the 2024 meeting, the 2023 meeting, the 2022 meeting, the 2021 virtual meeting, the 2020 virtual meeting, the 2019 meeting, the 2018 meeting, the 2017 meeting, the 2016 meeting; the 2015 meeting here, here, and here; the 2014 meeting here and here; the 2013 meeting here and here; the 2012 meeting; the 2011 meeting; and the 2010 meeting. 

May 10-12: Industrial Biostructures of America will be held in Cambridge, MA. The meeting covers all aspects of structural biology including FBLD, membrane proteins, allostery, cryo-EM, machine learning, and more. 

September 14-16: Fragments X, RSC-BMCS Tenth Fragment-based Drug Discovery Meeting, will be held in Cambridge, UK.  You can read my impressions of the 2024 meeting, the 2013 meeting, and the 2009 meeting.

September 28 to October 1: CHI’s Twenty-Fourth Annual Discovery on Target will be held as always in Boston. As the name implies this event is more target-focused than chemistry-focused, but there are always plenty of FBDD-related talks. You can read my impressions of the 2025 meeting, the 2024 meeting, the 2023 meeting, the 2022 meeting, the 2021 meeting, the 2020 virtual meeting, the 2019 meeting, and the 2018 meeting.

 

November 10-12: CHI holds its third Drug Discovery Chemistry Europe in beautiful Barcelona. This will likely include tracks on lead generation, protein-protein interactions, degraders and glues, and machine learning, with multiple fragment talks throughout. 

Know of anything else? Please leave a comment or drop me a note.

A new tool for covalent ligands: kinact/KI made easy with dDRTC

As covalent drug discovery becomes increasingly common, researchers are becoming more rigorous in how they characterize their molecules. The simple IC₅₀ values used for reversible inhibitors are meaningless for irreversible ligands unless the incubation times are also disclosed. And as molecules become more potent during the course of optimization, the incubation time may need to be shortened. An early hit might require treatment overnight to give 50% protein modification, while a potent lead might completely modify the protein in seconds. How do you quantitatively compare these?

 

The most rigorous parameter to characterize irreversible ligands is k_inact/K_I, sometimes called covalent efficiency, which we recently discussed here and here. Unfortunately, determining k_inact/K_I is a pain: it requires running multiple dose-response studies at multiple time points, and is thus typically only done for key compounds. In a new (open-access) Nat. Commun. paper, Robert Everley and colleagues at Frontier Medicines (including yours truly) provide a shortcut.

 

The new method relies on the fact that, especially for low-affinity fragments, much of the data collected in a conventional dose-response time course (DRTC) is redundant, providing little additional value. For example, if a compound at one concentration gives virtually no modification after 8 hours, it also won’t modify after 1 hour. The trick is to collect just the most informative data in a “diagonal” dose-response time course, or dDRTC.

 

I won’t go into the mathematics and full implementation details since the paper is open-access, but suffice it to say that dDRTC lowers the number of required data points by a factor of eight, thus saving both time and reagents – including precious protein.

 

The paper appropriately notes limitations, such as the fact that for compounds with better affinities (K_I < 50 µM), the values derived from dDRTC can underestimate the true k_inact/K_I. However, this situation is uncommon for fragments, and indeed the potencies for even some clinical compounds such as sotorasib and VVD-133214 are largely driven by (specific) k_inact rather than K_I. The paper shows good agreement for k_inact/K_I values determined using dDRTC with those determined using the conventional approach for compounds having k_inact/K_I from 1 to 2000 M^-1s^-1.

 

Perhaps most relevant for this blog, dDRTC is a practical solution for collecting important data. The next time you’re running a covalent program, give it a try!