Playing fast and loose with electrostatic anchors on RNA

Two weeks ago we discussed how to find ligand-binding sites in RNA. Last week we wrote about how difficult it is to find good ligands even for good binding sites in RNA. A recent open-access paper in J. Med. Chem. by Christian Kersten and colleagues at Johannes Gutenberg-University explores why targeting RNA is so tough.

 

The researchers were interested in two well-characterized riboswitches, naturally occurring RNA elements that bind to small molecules such as metabolites. Specifically, they chose to study a riboswitch that binds to S-adenosyl methionine (SAM, structure here) and a riboswitch that binds to prequeuosine-1 (PreQ₁) and prequeuosine-0 (PreQ₀). 

Due to the phosphate backbone, RNA is highly negatively charged. The researchers asked whether positively charged moieties on ligands can serve as “electrostatic anchors” to generally improve affinity, and if so whether this can lead to any design principles. Multiple biophysical techniques were used to study the interactions of the two riboswitches with various natural and synthetic ligands: surface plasmon resonance (SPR), isothermal titration calorimetry (ITC), and microscale thermophoresis (MST).

 

In the case of the SAM-VI riboswitch, the researchers compared the binding of SAM with closely related molecules having either one fewer positive charge (S-adenosyl homocysteine, or SAH) or synthetic ligands with the same or one more positive charge than SAM. Not surprisingly, SAM has the highest affinity, binding 20-50 fold more tightly than SAH. Further analysis suggested this is largely driven by an increased association rate, in which the positive charge accelerates the kinetics of binding. The driving energy for binding the ligands is enthalpic, but the favorable electrostatic interactions for more positively charged ligands are largely countered by an entropic penalty.

 

Similarly, the affinity of positively charged PreQ₁ for the PreQ₁ riboswitch is higher than the affinity of neutral PreQ₀, though not dramatically. As in the case of the SAM-VI riboswitch, the association rate of the positively charged ligand is more rapid than that of the neutral ligand. Binding for both ligands is highly enthalpic, with unfavorable entropy.

 

Previous reports had described other synthetic ligands for the PreQ₁ riboswitch, each with between one and three cationic centers. However these ligands showed no binding by ITC, questionable binding by MTC, and non-saturable, non-specific “loose binding” by SPR. Positive charges alone are not sufficient for high affinity, specific binding.

 

So what does it all mean? While adding positive charges can improve affinity of ligands for RNA, the increased affinity is usually not dramatic due to enthalpy-entropy compensation. The researchers note that, even for good ligands, the “thermodynamic binding profiles differ from typical protein-ligand interactions, where enthalpic and entropic contributions are usually more balanced.” Moreover, as we’ve noted, protein ligands often gain significant affinity with entropic gains by displacing "high energy water" molecules, but such opportunities are likely less common on the polar surface of RNA.

 

The affinity and ligand efficiency of PreQ₁ for its riboswitch are impressive, so clearly it is possible for small drug-like ligands to bind tightly to RNA. But this interaction is the product of countless eons of evolution. This careful paper suggests why building similarly effective synthetic ligands for most RNA will be difficult.

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.

Small and simple, but novel and potent

Back in 2012 we wrote about GDB-17, a database of possible small molecules having up to 17 carbon, oxygen, nitrogen, sulfur, and halogen atoms, most of which have never been synthesized. Although novelty isn’t strictly necessary for fragments, as evidenced by the fact that 7-azaindole has given rise to three approved drugs, it’s certainly nice to have. In a new (open-access) J. Med. Chem. paper, Jürg Gertsch, Jean-Louis Reymond, and colleagues at the University of Bern synthesize fragments that had not been previously made and show that they are biologically active.

 

When you start drawing all possible small molecules you get lots of weird stuff, including an explosion of compounds containing multiple three- and four-membered rings, which may be difficult to make. The researchers wisely focused on “mono- and bicyclic ring systems containing only five-, six-, or seven-membered rings.” They further limited their search to molecules containing just carbon and one or two nitrogen atoms (as well as hydrogen, of course). Systematic enumeration led to 1139 scaffolds, ignoring stereochemistry, of which 680 had not been previously reported in PubChem. Out of these, three related scaffolds were chosen for investigation.

 

Computational retrosynthesis was used to devise routes to the three bicyclic scaffolds, and these were successfully synthesized, along with mono-benzylated versions, for a total of 14 molecules (including stereoisomers), all rule-of-three compliant. The online Polypharmacology Browser 2 (PPB2) was used to predict targets, and several monoamine transporters came up as potential hits. The molecules were tested against norepinephrine transporter (NET), dopamine transporter (DAT), serotonin transporter (SERT), and the σ-R1 receptor in radioligand displacement assays. None of the free diamines were active, but several of the benzylated compounds were, in particular compound 1a.

 

Compound 1a was initially made as a racemic mixture, and when the two enantiomers were resolved (R,R)-1a was found to be a mid-nanomolar inhibitor of NET while (S,S)-1a was 26-fold weaker. Compound (R,R)-1a was also a mid- to high nanomolar inhibitor of σ-R1, DAT, and SERT. Pharmacokinetic experiments in mice revealed that the molecule had poor oral bioavailability but remarkably high brain penetration and caused sedation. The researchers conducted additional mechanistic studies beyond the scope of this blog post and conclude that (R,R)-1a could be a lead for “neuropsychiatric disorders associated with monoamine dysregulation.”

 

There are several nice lessons in this paper. First, as we noted more than a decade ago, there is plenty of novelty at the bottom of chemical space. Moreover, and in contrast to our post last week, even small fragments can have high affinities. But novelty comes at a cost: synthesis of compound 2a required eight steps from an inexpensive starting material with an overall yield of just 9%, though this could certainly be optimized. Nonetheless, particularly for CNS-targeting drugs which usually need to be small in order to cross the blood brain barrier, the price might be worth paying.

 

Of course, even within this paper there are hundreds more scaffolds to look at than the three tested, and perhaps the researchers were lucky that their choices were biologically active. As computational methods continue to advance, it will be worthwhile turning them loose on GDB-17.

Crystallography first in fragment optimization: Binding-Site Purification of Actives (B-SPA)

At FBLD 2024, Frank von Delft (Diamond Light Source) announced the ambitious goal of taking a 100 µM binder to a 10 nM lead in less than a week for less than £1000. Fragment to lead optimization usually takes longer, as dozens or even hundreds of compounds need to be synthesized and tested. One way to speed things up is through “crude reaction screening,” otherwise known as “direct to biology,” in which unpurified reaction mixtures are tested directly. In a new (open-access) Angew. Chem. Int. Ed. paper, Frank, John Spencer, and collaborators at University of Oxford, University of Sussex, and Creoptix apply this approach to crystallographic screening.

 

The researchers were interested in the second bromodomain of Pleckstrin Homology Domain-Interacting Protein, or PHIP(2), an oncology target. As we discussed in 2016, they had previously run a crystallographic screen and identified multiple hits, including F709, which, despite having no measurable affinity, had good electron density and multiple vectors for optimization. Six separate libraries based on this fragment were constructed, with between 58 and 1024 targeted small-molecule products per library and up to four steps done without purification.

 

One challenge for crude reaction screening is assessing whether or not a reaction has actually generated product. Typically this is done by analytical liquid chromatography mass spectrometry (LCMS), but analyzing results manually is tedious. Fortunately academics have graduate students and postdocs, and it was presumably these intrepid souls who spent 17 days analyzing the 1876 small-molecule products attempted.

 

I can say from personal experience that spending hours perusing LCMS chromatograms is not enjoyable, so the researchers built an automated tool called MSCheck, which appears to be freely available here. This showed 83% agreement with the manually curated data, and even identified additional true positives that had been missed. All together 1077 of the reaction mixtures had the desired product, with success rates for the various libraries ranging from 39% to 97%.

 

The successful reactions were soaked into crystals and screened, and nearly 90% of these generated usable data. A total of 29 crystals had interpretable density in the ligand binding site: 7 were starting materials and 22 were desired products. Of the products, 19 bound with the piperazine core in a similar position as the initial fragment, while three bound in an alternate manner.

 

Of course, the whole point of this exercise is to find improved binders, so the researchers tested pure versions of each of the 22 crystallographic hits in two different assays. Only compound PHIP-Am1-20 had measurable affinity, with modest ligand efficiency.

This is not the first example of crude reaction screening by crystallography; we wrote about REFiL_x and a related technique in 2020. In one of those papers, the crude reaction mixtures were assessed by SPR as well as crystallography, which revealed that the crystallographic screen missed some binders, and there is no reason to think the same did not happen here. Indeed, molecules that bind tightly in a different conformation may be more likely to shatter the crystal lattice and thus go undetected.

 

The researchers state that for non-crystallographic crude reaction screening “only strong assay readouts are informative.” But is this bug, or a feature? A 2019 publication that used crude reaction screening to identify KRAS ligands (which I wrote about here) used an assay cascade to quickly select the most potent hits. Even the fastest crystallographic screens can’t compete with plate-based assays in terms of speed.

 

Perhaps PHIP(2) is a particularly challenging test case. As we discussed in 2022, multiple computational screens performed poorly in predicting crystallographic binding modes of ligands for this protein. But as I wrote at the time, it may be that many crystallographic ligands are just too weak to be useful.

 

Although there is a strong case for using crystallography first for finding fragments, I am not yet convinced the same applies for optimizing fragments.

From fragment to macrocyclic Ras inhibitors

At the Drug Discovery Chemistry meeting last month chemist John Taylor described efforts against the oncology target RAS. This story was recently published in J. Med. Chem. by John, Charles Parry, and a team of some three dozen collaborators at CRUK Scotland Institute, Novartis, and Frederick National Laboratory for Cancer Research.

 

Practical Fragments has highlighted multiple Ras efforts, including the development and approval of sotorasib, which inhibits the G12C mutant of KRAS. Sotorasib binds in the so-called switch II region, next to the site where the nucleotides GDP and GTP bind. Before the discovery of this site, researchers had identified fragments that bind to a different site, switch I-II. 

 

Most of the ligands that bind to either site only inhibit the off-form of Ras proteins, in which the proteins are bound to GDP. One mechanism of resistance for cancer cells is to increase the amount of protein in the active, or GTP-bound state. Thus, the researchers focused on the oncogenic G12D mutant of KRAS bound to a GTP analog and screened it against 656 fragments using SPR. Ligand-detected NMR confirmed five of the hits, including compound 5.

 

Two dimensional ¹H-¹⁵N HSQC NMR revealed that compound 5 binds in the switch I-II pocket; merging this with a literature fragment generated compound 6. SAR studies led to compound 11, which was characterized crystallographically bound to the protein. The structure suggested trying to make a salt bridge with an aspartic acid residue, leading to compound 13, with sub-micromolar affinity for the inactive form of the protein. A crystal structure of a related compound suggested the possibility of macrocylization, and this turned out to be successful, with compound 21 being the most potent. (All values shown here are determined by NMR or SPR on the G12D KRAS mutant bound to either GDP or the GTP analog GMPPMP.)

 

A number of different macrocycles were made and tested, and all of them were more potent against the inactive than the active form of KRAS. Crystal structures suggested that a glutamic acid side chain adopts a conformation in the the GTP-bound form of KRAS that impedes ligand interactions.

 

Interestingly though, building off the molecules in another direction led to the opening of a small subpocket that had not previously been reported in the literature. Exploiting this “interswitch” region led to compound 36, with a nearly 10-fold preference for the active form of KRAS.

 

Most of the macrocycles in both series were able to block nucleotide exchange in a biochemical assay, meaning they could prevent the exchange of GDP for GTP. A few of the compounds were tested in cell-based assays and could block binding between RAF and multiple Ras isoforms, including two mutants of KRAS as well as wild-type KRAS, HRAS, and NRAS.

 

Unfortunately, and not surprisingly given their high polar surface areas, the compounds had low permeability, high efflux, and high clearance in vitro. Mouse studies on one compound confirmed these liabilities in vivo.

 

Although the compounds could not be advanced, this is still a nice fragment to lead story. The fact that a new pocket could be identified despite so much previous effort on this target is a good reminder that no matter how much you know, there is always room for surprises.

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.

Protein-detected NMR without isotopic labeling

Protein-detected NMR was the first practical approach for finding fragments, and as we noted last week some still consider it the gold standard. As commonly practiced, it requires isotopically labeled protein: at a minimum ¹⁵N and sometimes ¹³C and even deuterium. Making large amounts of labeled protein can be both expensive and difficult. A new ACS Med. Chem. Lett. paper by Andrew Petros and collaborators at AbbVie describes a new approach that avoids this requirement. The method, called 1D-ECHOS, combines two previously described techniques.

 

The first technique addresses the fact that fragment screens typically use much higher ligand concentrations than protein concentrations, and thus the proton signals coming from the ligand can overwhelm those coming from the protein. “1D-diffusion filtered NMR” essentially removes signals coming from small molecules to focus on the protein.

 

When a ligand binds to a protein, the chemical shifts of nearby residues on the protein change, and these peak shifts are most easily observed in two-dimensional (2D) NMR spectra, where each dimension typically corresponds to the signal from a different nucleus, such as ¹H or ¹⁵N. Without isotopic labeling, only protons can be observed, and only in one dimension, so the resulting spectra look like mountain ranges, with overlapping peaks. To facilitate comparison between the two samples (protein with or without ligand), the researchers use a second technique, called Easy Comparison of Higher Order Structure (ECHOS). This allows differences to be expressed as a single “R-score”, where larger numbers indicate more deviation between the two spectra.

 

So, how well does it work? The researchers started by examining a set of 13 hits from a DNA-encoded library against an unnamed 36 kDa protein. Four of these had previously been confirmed to bind using 2D-NMR, and all of these had positive R-scores, while the non-binders had R-scores close to zero. The approach was also faster than a standard HSQC with labeled protein, requiring just 10 minutes rather than 35 minutes.

 

As the researchers note, DEL hits are typically larger and more potent than fragment hits, so they next turned to 11 confirmed MCL-1 binders from a fragment screen we wrote about here. These were tested at 62.5 µM, and the R-scores roughly correlated with their previously measured affinities, which ranged from 20 µM to 500 µM.

 

To try to get more quantitative information, the researchers performed dose-response experiments and plotted the R-score as a function of ligand concentration. This allowed them to extract dissociation constants, which were in good agreement with the known values. For ligands containing tert-butyl groups the 1D-diffusion filter was not fully capable of masking the signal, but this peak could be manually removed from the analysis. The researchers also applied the approach to two additional targets, BRD4 BDII and TNFα, and found good agreement with known ligand affinities. Of course, unlike 2D NMR, 1D-ECHOS does not provide information on where the fragments bind.

 

1D-ECHOS appears to be a practical approach for validating and characterizing fragment binding, but I’m no NMR spectroscopist, so I’ll be interested to hear what experts think.

Twentieth Annual Fragment-Based Drug Discovery Meeting

Last week’s CHI Drug Discovery Chemistry (DDC) meeting was held as usual in San Diego. More than 850 people attended, 96% in person, with 70% from industry and 28% from outside the US. I personally attended more than three dozen talks over the four days and will just touch on some broad themes.

 

Noncovalent approaches

Steve Fesik (Vanderbilt) gave two talks, the first of which was focused on “FBDD tips for success.” This opinionated and entertaining romp revealed lessons learned across several projects on difficult targets such as KRAS. Another holy grail oncology target is MYC, which is largely disordered. A two-dimensional NMR screen against the protein failed to yield any hits, but a screen of the MYC:MAX heterodimer provided hits which have been optimized to high nanomolar potency and are able to block DNA binding.

 

The second talk was focused on E3 ligases, a target class Steve has been pursuing for the past decade. Steve is particularly interested in E3 ligases such as CBL-C, TRAF4, and KLHL12 that are differentially expressed in certain tissues. In the case of KLHL12, which is not found in heart tissue, an NMR-based screen led to fragment hits that were ultimately optimized to mid-nanomolar binders and could be turned into bivalent degraders for Bcl-xL and β-catenin.

 

When asked about his second favorite fragment-finding method after protein-detected NMR, Steve mentioned SPR. The throughput for SPR has historically been modest, but John Quinn (Genentech) described the new Carterra Ultra, which is capable of screening 96 proteins simultaneously while retaining good sensitivity. John screened 3000 fragments at 500 µM against multiple proteins in just two weeks, which provided an immediate assessment of both protein ligandability and fragment selectivity. Interestingly, and in contrast to some other analyses, shapelier fragments had similar hit rates to flatter fragments.

 

Several talks focused on fragment-to-lead success stories, some of which we’ve covered on Practical Fragments, such as RIP2 kinase inhibitors that started from flat fragments and were evolved to more three-dimensional leads as described by Mark Elban (GSK). John Taylor discussed pan-RAS inhibitors discovered at Cancer Research Horizons, the subject of an upcoming post. Andrew Judd (AbbVie) described the discovery of ABBV-973, a potent STING agonist that could be useful for certain types of cancer. And Justyna Sikorska described the discovery of a non-covalent WRN inhibitor at Merck. This is a nice complement to Vividion’s covalent WRN inhibitor, which we wrote about here and which was presented by Shota Kikuchi. Interestingly, structural biology was not enabled until late in this project.

 

One of the earliest arguments for fragment linking was the concept of avidity, and this underlies the basis of a technology discussed by Tom Kodadek and Isuru Jayalath at University of Florida Scripps. The idea is to immobilize fragments onto TentaGel beads, each the size of a red blood cell. These can be screened against multivalent proteins using either simple plate-based assays or FACS, the idea being that even if an individual protein-ligand interaction is weak, a multimeric protein can interact with several ligands on a single bead for enhanced binding. The researchers validated the concept with streptavidin, and also used it to find millimolar binders to the proteasome subunit Rpn13.

 

Last year we wrote about using photoaffinity crosslinking with fully functionalized fragments (FFFs) to identify non-covalent ligands to thousands of proteins in cells, and this was the subject of several talks. Chris Parker (Scripps) has mapped more than 7000 binding sites and described the discovery of an inhibitor against the inflammatory target SLC15A4. Interestingly, the molecule binds what appears to be a disordered region, though Chris speculated that it adopts a more defined structure in cells.

 

Belharra has gone all in on using FFFs, and Jarrett Remsberg and Andrew Wang described the construction of a diverse >11,000-membered FFF library, 88% of which consists of enantiomers. This has been screened against 13 different oncology and immunology cell lines to identify enantioselective or chemoselective hits against >4000 proteins including STAT3, IRF3, and AR.

 

Covalent approaches

The FFF approach uses covalent bond formation to trap a noncovalent ligand, but of course covalent ligands are all the rage these days, as we noted just last week. Dan Nomura (UC Berkeley) described the identification of stereoselective covalent ligands against a disordered region of cMYC that seem to work by destabilizing the protein in cells. Similarly, covalent ligands against the largely disordered AR-V7 also seem to destabilize the protein. It will be interesting to explore the mechanism of these molecules to see whether the proteins are more ordered inside cells.

 

Jin Wang (Baylor College of Medicine) described a chemoproteomic approach called Fragment Probe Protein Enrichment (FraPPE) which entails linking covalent fragments to a desthiobiotin tag. Labeled proteins are then pulled down, proteolyzed, and analyzed by mass spectrometry. In contrast, competition methods such as those described last year pull down labeled peptides after proteolysis. The advantage of FraPPE is that it can capture multiple peptides from each pulled-down protein, leading to fewer false negatives.

 

Of course, not every application of covalent discovery involves chemoproteomics. Joe Patel, who co-organized FBLD 2016, described the Nexo Therapeutics platform. They’ve built from scratch a library of >12,000 fragments, a third of which contain stereocenters. Each member is rule-of-three compliant before adding the warhead, meaning that the final molecules can be larger, which as we noted earlier this month is probably a good idea. To date Nexo has successfully screened more than a dozen targets using intact protein mass spectrometry.

 

The Nexo library targets not only cysteines but other residues as well, and Maurizio Pellecchia (UC Riverside) described using sulfonyl fluorides and fluorosulfates to target histidine residues. He and his group screened a library of 600 fluorosulfate-containing fragments (MW 250-350 Da) against the oncology target MCL1 and found several that stabilized the protein towards thermal denaturation. Crystallography confirmed covalent bond formation.

 

Most covalent fragments are electrophilic so that they can react with nucleophilic protein residues, but as we noted in 2022 it is possible to do the reverse. Megan Matthews (University of Pennsylvania) described how she used chemoproteomics to discover the mechanism of action for hydralazine, a drug that has been used since 1949 to treat hypertension. This fragment-sized (MW 160 Da!) molecule irreversibly alkylates a histidine residue within the active site of the enzyme ADO, a target that has also been implicated in gliobastoma.

 

Plenary Keynotes

The approval of the covalent BTK inhibitor ibrutinib in 2013 arguably marks the start of the modern era of covalent drug discovery, and Chris Helal described Biogen’s efforts against this target using reversible inhibitors, irreversible inhibitors, and degraders. Chris traced the origin of their phase-2 BIIB091 to a collaboration with Sunesis that used Tethering, so perhaps we should include this molecule in our list of fragment-derived clinical compounds.

 

Phil Baran of Scripps, who last spoke at the conference in 2020, gave the secondary plenary keynote. After stating that “medicinal chemists are the backbone of society,” he then detailed multiple examples of how they’ve been doing things wrong. Fortunately, he provided useful chemistry solutions, with “useful” defined as reactions that are operationally simple, have wide scope, and require only readily available reagents. Rather than deploying tedious protecting group installations and deprotections, Phil uses radical chemistry to directly generate carbon-carbon bonds between or within complicated molecules. His goal is to make the chemistry so simple and practical as to be boring, and he illustrated the point by showing his teenage daughter successfully running a reaction.

 

I’ll end here, but please leave comments. And mark your calendar for April 13-16 next year, when DDC returns to San Diego.

A library of covalent fragments vs a library of kinases

Protein kinases have proven to be a fruitful class of targets, as evidenced by more than 80 FDA-approved drugs, five of which came from fragments. Because all protein kinases bind ATP, selectively inhibiting just one of the more than 500 family members can be challenging. This is a bit easier for the 215 protein kinases that contain a cysteine within the ATP-binding pocket capable of reacting with covalent ligands. In a recent (open access) Angew. Chem. Int. Ed. paper, Matthias Gehringer, Stefan Knapp, and collaborators at Johann Wolfgang Goethe-University and Eberhard Karls University Tübingen provide such starting points for dozens of kinases.

 

The researchers built a small library of 47 fragments consisting of six classic hinge-binding moieties such as pyrazole and azaindole coupled through nine aryl linkers at varying positions to an electrophilic acrylamide warhead. Although most of the compounds are rule-of-three compliant, the researchers note they “reside at the upper end of fragments space,” similar to what we discussed last week. Chemical reactivity towards the abundant cellular thiol glutathione was tested and found to be lower than the approved drug afatinib, meaning the fragments might be good starting points for optimization.

 

Each member of the fragment library was screened against 47 different protein kinases chosen to present cysteine residues at a variety of positions around the ATP binding site. Two types of screens were conducted: intact protein mass spectrometry to assess covalent binding and differential scanning fluorimetry (DSF) to assess protein stabilization. Screens were run at fairly high concentrations, 50 µM protein and 100 µM fragment.

 

The results, plotted as a two-dimensional figure with kinases on one axis and compounds on the other, provide a wealth of information. Some compounds hit multiple kinases while others hit few or none. Similarly, some kinases are hit by multiple compounds while others are recalcitrant.

 

A couple more general observations emerged. First, there was little if any correlation between the inherent reactivity of a given fragment (as assessed by reactivity with glutathione) and the number of kinases hit, suggesting that covalent modification was driven by specific interactions rather than nonspecific reactivity. Second, there was also no clear correlation between the ability of a fragment to stabilize a given kinase and the ability of the same fragment to covalently bind to that kinase. This latter observation isn’t surprising, since one could imagine a fragment binding noncovalently to a kinase and stabilizing it without forming a covalent bond.

 

Most proteins contain multiple cysteine residues, and the researchers confirmed that the fragments were covalently modifying the cysteines in the ATP-binding pocket using mutagenesis, trypsin digestion, or, for MAP2K6, RIOK2, MELK, and ULK1, crystallography. The crystal structures were particularly informative in showing hydrogen bond interactions between the covalently-bound fragments and the hinge region.

 

As we’ve noted, the best metric for characterizing irreversible covalent inhibitors is k_inact/K_I, and the researchers determined these for covalent inhibitors of PLK1, PLK3, RIOK2, CHEK2, and CSNK1G2. The values ranged from 2 to 8 M^-1s^-1, comparable to other early covalent fragments.

 

This is a lovely, systematic paper that is in some ways an irreversible complement to a study we wrote about in 2013 focused on reversible covalent kinase inhibitors. The fact that hit rates are relatively high likely reflects the fact that all the fragments contain privileged hinge-binding pharmacophores.

 

Perhaps most importantly, all the data are available in the supporting information. If you’re interested in pursuing any of these 47 kinases, you may find good starting points here.

Do covalent fragments need to be larger?

A few months ago we highlighted work out of AstraZeneca detailing how to build a covalent fragment library. One of the design features was including larger molecules beyond the traditional rule of three (Ro3) criteria. A new open-access paper in J. Med. Chem. by György Keserű and collaborators at the HUN-REN- Research Centre for Natural Sciences and the Weizmann Institute of Science explores “size-dependent target engagement of covalent probes.”

 

The paper starts with a theoretical discussion of covalent inhibitors, focusing on the classic two-step mechanism in which binding of a ligand to a protein is followed by covalent bond formation. These steps are characterized by the inhibition constant (K_I) and the inactivation rate constant (k_inact). The most appropriate way to assess an irreversible covalent inhibitor is by the ratio k_inact/K_I, as we discussed last year.

 

A two-step mechanism is not the only possibility: the researchers also consider a three-step model in which binding of the ligand is followed by a second step, deprotonation of the amino acid nucleophile, before the final bond-forming step.

 

Fragments typically have lower affinities than lead-size or drug-size molecules, and thus k_inact will usually need to be higher for smaller molecules in order to see significant protein labeling. Simulations in which K_I is held constant show that at the high micromolar affinities often seen for fragments, protein modification requires either long incubation times or high reactivities. In addition to these simulations, the researchers also reanalyze publications we’ve previously covered such as this and this to argue that “reactivity contributes to labeling when the effects of other factors cancel out.”

 

Next, the researchers examine the kinase BTK and the oncology target KRAS, both of which have been successfully drugged with covalent molecules, ibrutinib and adagrasib, respectively. They trimmed back these molecules to smaller lead-like and fragment-like molecules and found that while some lead-sized molecules could still label the proteins, this was not the case for the fragment-sized molecules. From this they conclude that “fragment-sized covalent agents do not offer smooth optimization and are not ideal starting points.”

 

Two examples do not a trend make, but the researchers point to other examples in the literature. In 2020 we noted the larger size of covalent KRAS hits, and Vividion’s WRN inhibitor also started from a molecule with a molecular weight of 337 Da, while GSK’s starting point weighs in at 312 Da. The AstraZeneca library we mentioned at the start of this post yielded a hit against BFL1 that also just missed the Ro3 cutoff, coming in at 302 Da.

 

That said, there are counterexamples. Just last month we highlighted a covalent fragment hit that fits comfortably within the rule of three. Fragment-sized covalent hits can be found, but don’t expect them to be common. The alternative approach, screening lead-like compounds, will also likely require screening more compounds due to lower coverage of chemical space. Either way, libraries containing more molecules are likely to be beneficial for finding covalent starting points.

Fragments meet crypto!

Two years ago today, Practical Fragment$ launched a line of non-fungible tokens. Unfortunately, the NFT craze didn't last much longer than that for sea shanties. But another virtual confection, cryptocurrency, appears to be less ephemeral, and today SkyFragNet has announced the launch of FragCoin.

 

One of the problems with crypto is that it uses a huge amount of (arguably) wasted energy. What if the effort to mine crypto could be put to practical use?

 

As we noted six years ago today, SkyFragNet has automated the entire drug discovery pipeline. It is also building a powerful generative AI positronic brain, to be trained using in-house experimental data. But even though the prospective computational docking methods are best in class, they still need to be improved, and for this SkyFragNet is turning to everyone with a computer.

 

To mine new FragCoins, miners will dock fragments against various targets, and these results will be compared with ground truths at SkyFragNet: each successful docking event will initially result in one FragCoin. Think of it as Folding@home but for profit as well as fun.

 

As FragCoins are mined, the docking software will be continually improved. Over time the number of successful docking events required for a FragCoin will increase. 

 

Just like bitcoin, the number of FragCoins will be strictly limited, in this case to 977,468,314, the number of fragments in GDB-13. So don't delay: start docking today!

Fragments and nanodiscs: beware nonspecific binding

Membrane proteins make up roughly a quarter of human proteins, including many important drug targets. Biophysical methods for fragment screening typically require pure, isolated proteins, and removing membrane proteins from their native environment is not always possible. One solution has been to create nanodiscs which, as we described previously, are isolated little membranes containing the protein of interest. These nanodiscs can be immobilized to the sensor chips used for surface plasmon resonance (SPR), one of the most popular fragment finding methods. But in a recent open-access Chem. Biol. Drug Des. paper, Marcellus Ubbink and collaborators at Leiden University and ZoBio show that the precise composition of the nanodiscs can have a profound effect on the results.

 

The researchers chose cytochrome P450 3A4 (CYP3A4) as a model membrane protein. This enzyme metabolizes a large fraction of drugs and has a capacious active site able to bind a wide variety of substrates. Four different lipids were chosen for the nanodisc, all of which contained phosphatidylcholine headgroups and differing hydrophobic tails: POPC, DPPC, DMPC, and DPhPC.

 

Nanodiscs were prepared either with or without CYP3A4 and immobilized to SPR chips. Unlike some membrane proteins, it is possible to isolate and immobilize CYP3A4 in the absence of membranes, though the protein forms physiologically less relevant oligomers.

 

Next, the researchers examined 13 known (non-fragment) CYP3A4 ligands. Unfortunately, most of these bound to the empty nanodiscs, and in some cases more than ten ligands bound to a single empty nanodisc. This nonspecific binding correlated with lipophilicity, with only the three least lipophilic molecules showing no binding to empty nanodiscs. One of these was the antifungal drug fluconazole, with a clogP = 0.4. Happily though SPR studies using either free or nanodisc-bound CYP3A4 yielded dissociation constants of 10-20 µM, consistent with published values.

 

Thus encouraged, the researchers screened a diverse set of 140 fragments at 250 or 500 µM against empty and CYP3A4-loaded nanodiscs using SPR. Just as with the larger molecules, there was a good correlation between cLogP and nonspecific binding to empty nanodiscs. Fragments that bound to one type of nanodisc (POPC, for example) also tended to bind nonspecifically to other types of nanodiscs (DPPC, DMPC, and DPhPC). Fewer fragments bound nonspecifically to DMPC nanodiscs than to the others, suggesting this may be the best lipid to use.

 

Fragment hits were defined as those binding to CYP3A4-containing nanodiscs more than they bound empty nanodiscs (or, for isolated CYP3A4, the unmodified SPR chip). Hit rates varied dramatically, from 9 of 140 fragments tested against CYP3A4 in POPC nanodiscs to 33 of 140 tested against CYP3A4 in DMPC nanodiscs. There were also 33 hits against free CYP3A4, 11 of which were unique. However, all 11 of these are somewhat lipophilic (average cLogP ~2.3) and most also bound significantly to empty nanodiscs. The researchers suggest that these bind “aspecifically” to CYP3A4.

 

A Venn diagram of all the hits shows only two that bind to free CYP3A4 as well as all nanodiscs containing CYP3A4, and the researchers highlight these two as the most promising. Unfortunately these are not further characterized.

 

Near the beginning of the paper, the researchers note that very few fragment screens have been conducted against membrane proteins incorporated into nanodiscs. This analysis suggests why this is so. If you use nanodiscs, make sure to consider different types of ligands. And look carefully for nonspecific binding.

Fragments vs eIF4E: a chemical probe

Cancer cells are known for growing and multiplying quickly, and to do so they need to produce large amounts of protein. The rate determining step in protein translation happens early, when ribosomes are recruited to the 5’-end of mRNA by the eukaryotic initiation factor 4F (eIF4F) complex. This complex has long been a target for drug discovery, and in a recent open-access Nat. Comm. paper Paul Clarke, Andrew Woodhead, Caroline Richardson, and collaborators at Institute of Cancer Research and Astex describe a chemical probe. (Andrew spoke about this program last year at FBLD 2024.)

 

The eIF4F complex includes three core proteins, confusingly named eIF4E, eIF4G, and eIF4A. eIF4E binds to the 5’cap of mRNA and recruits eIF4G. Blocking the interaction of eIF4E either with mRNA or eIF4G could in principle shut down protein synthesis, but intensive efforts by multiple groups have struggled: the mRNA binding site is very polar, and disrupting protein-protein interactions is tough. Thus, the researchers took a fragment approach.

 

Developing a form of eIF4E suitable for fragment screening was itself a challenge because the protein mostly exists as part of a complex in cells and the native monomer is unstable. After making more than two dozen different constructs, the researchers developed a stable, soluble form that could be crystallized. This construct was screened against a library of 1371 fragments in pools of four, each at 500 µM, using CPMG NMR followed by crystallography, leading to 50 hits. A few bound at the mRNA cap-binding site but most bound to a previously unreported “site 2,” which is near where eIF4G binds.

 

One of these, compound 1, has a reasonable ligand efficiency despite its low affinity as assessed by ITC. The phenol appeared to be making no interactions and so was removed. Adding a fluorine usefully enforced the twisted biaryl conformation and filled a small dimple; fragment growing then led to mid micromolar compound 3. Further growing to pick up additional lipophilic and polar contacts eventually led to compound 4, with low nanomolar affinity. Understanding the importance of negative controls for chemical probes, the researchers also switched the stereochemistry at the benzylic carbon to produce compound 5, which has >30-fold lower affinity for eIF4E. 

 

Crystallography revealed that binding of compound 4 to eIF4E causes conformational changes that should impair binding of the protein to eIF4G. Experiments in cell lysates bore out this hypothesis. Moreover, compound 4 also inhibited protein translation in cell lysates at low micromolar concentrations, while compound 5 did not.

 

Unfortunately, these observations did not extend to intact cells. A cellular thermal shift assay (CETSA) demonstrated that compound 4 did stabilize eIF4E in cells with an EC₅₀ = 2 µM, consistent with binding. But it was much less effective at blocking the interaction with eIF4G in cells, even at high concentrations, and showed no inhibition of protein translation.

 

To understand why, the researchers conducted a series of targeted protein degradation and genetic rescue experiments that are beyond the scope of this blog post. The upshot is that eIF4G binds to several regions of eIF4E, and that while compound 4 disrupts binding to the “non-canonical binding site”, it does not block binding to the “canonical binding site,” and thereby does not block overall complex formation. Why there should be a difference between intact cells and cell lysates is not obvious to me, but perhaps the more dilute conditions of cell lysates play a role, as seen for a paper we discussed last year.

 

One interesting feature of this story is that the initial fragment makes no polar interactions with the protein; all of the polar interactions in compound 4 were added during optimization. This is quite the opposite of ASTX660, where all the polar interactions in the final clinical compound came from the initial fragment. Indeed, a 2021 analysis of fragment to lead successes found that fewer than one in ten retained no polar interaction from the initial fragment.

 

This paper also illustrates the gap that can occur between research and publication; a couple of the authors listed as affiliated with Astex left in 2017. But better late than never, and this study nicely integrates fragment-based lead discovery with elegant biology. Compound 4 should be a useful tool for further exploring the nuances of eIF4E.

Crude success against the SARS-CoV-2 main protease: From covalent fragment to noncovalent lead

With increased throughput and reliability of biophysical and other methods, finding fragments against most targets is now fast and easy. Advancing these fragments to leads, not so much. In a new open-access Angew. Chem. Int. Ed. paper, Jacob Bush and collaborators at GSK, University of Strathclyde, and the Francis Crick Institute provide a case study for how to accelerate the process.

 

Almost exactly five years ago we highlighted early efforts against the main protease (M^pro) from SARS-CoV-2. This target turned out to be a good choice, as demonstrated by the rapid discovery and approval of the drug nirmatrelvir. M^pro is a cysteine protease and thus ideally suited for covalent fragment screening.

 

In the new paper, the researchers screened a library of 219 chloroacetamide-containing fragments (each at 5 µM) individually against 0.5 µM protein for 16 hours at 4 ºC and then analyzed them by intact protein mass spectrometry. Six of these gave at least 75% modification, and further characterization found that the most potent, compound 2, had a k_inact/K_I = 170 M^-1s^-1. This (and the other hits) also inhibited the protein in an enzymatic assay, and additional chemoproteomic experiments revealed that compound 2 could bind to the active site cysteine of M^pro in living cells with surprising selectivity; just 11 targets were more strongly engaged than M^pro.

 

To optimize compound 2, the researchers turned to crude reaction screening, also known as direct-to-biology or D2B. As we described here and here, this entails running reactions at small scale and testing them directly, without purification. To validate the approach, the researchers synthesized a subset of the original 219 chloroacetamides in 384-well plates. HPLC studies confirmed the desired product as the major component for 43 of the 69 attempted syntheses; only four failed. Importantly, there was a good correlation in activity between the crude reaction mixtures and the pure molecules.

 

Next, the researchers synthesized a new D2B library of 193 molecules related to compound 2. HPLC analysis of the crude products showed a 77% success rate, with just nine outright failures. The library was screened against M^pro for 1 hour (as opposed to 16 hours in the first screen), resulting in 14 hits. The best of these, compound 7a, was such a rapid modifier that the a k_inact/K_I could not be easily calculated, but it showed nanomolar activity in the enzymatic assay. It was also more selective than compound 2 in cell-based experiments.

 

Chloroacetamides are not considered advanceable as drugs, so the researchers sought to remove the warhead, initially by replacing it with the simple acetamide in compound 12. Although this molecule showed almost no activity in the enzymatic assay, the researchers coupled a diverse set of 146 carboxylic acids to the amine building block and screened the crude reaction mixtures in a functional assay at 50 µM to identify seven molecules that gave nearly complete inhibition, with compound 13 being the most potent. A second D2B library of analogs around compound 13 was screened at 1 µM, leading to the mid-nanomolar compound 14.

 

This is a nice illustration of the power of crude reaction screening to rapidly identify new chemical matter. It is true that M^pro is quite ligandable; we wrote about other non-covalent fragment success stories here and here. However, as we discussed here, D2B can be applied to more challenging targets. The supporting information in the new paper should be particularly valuable for those hoping to try the approach themselves.

 

At FBLD 2024 Frank von Delft set a goal of taking a “100 µM binder to a 10 nM lead in less than a week for less than £1000.” We’re not there yet, but developments in D2B are moving us forward.

Fishing for pearls more efficiently with a new NMR method

NMR is the most venerable approach for finding fragments, and ligand-detected NMR is still among the more popular methods. But the amount of protein required for a full fragment library screen can be a limitation, particularly for more challenging targets. A new paper in Angew. Chem. Int. Ed. by Alvar Gossert and collaborators at ETH Zürich, Bruker, and Karlsruhe Institute of Technology provides a new, less protein-intensive approach.

 

I’ll preface the next paragraph by admitting that not only am I no spectroscopist, I don’t even play one on TV. So, spectroscopy-savvy readers, please feel free to provide more details in the comments, especially if I get something wrong. For fellow non-spectroscopists, the takeaway is that clever NMR tricks increase sensitivity.

 

PEARLScreen, short for Perfect Echo for Advanced Relaxation-based Ligand Screen, is related to the classic Carr-Purcell-Meiboom-Gill (CPMG, or T_1ρ) method, which we wrote about most recently here. As in that older method, PEARLScreen relies on the decrease in signal intensity of a ligand that binds to a protein. This is due to slower tumbling of the bound ligand, resulting in faster relaxation of excited protons (see here). Lengthening the time between excitation and measurement should in theory boost contrast between bound and free ligands, but various technical challenges impede this in practice. PEARLScreen overcomes these challenges using “a perfect echo pulse train with water suppression by excitation sculpting.” In addition to lengthening the relaxation delay, PEARLScreen also allows exchange broadening to occur between the ligand and protein, further increasing sensitivity.

 

The researchers simulated multiple conditions to optimize various parameters, and then experimentally tested PEARLScreen on four different proteins with three types of NMR instruments, starting with a standard high-end 600 MHz.

 

The first protein-ligand pair was trypsin binding to a known benzamidine fragment. This interaction was detectable using a standard T_1ρ experiment with 200 µM ligand and 20 µM protein. Using PEARLScreen, the researchers could reduce the protein concentration to 1 µM while maintaining similar signal to noise .

 

Next, they screened 94 fragments in pools of 8 against three different proteins: PPAT, Abl, and FKBP. In all cases PEARLScreen was more sensitive than T_1ρ, allowing screening at 2.5 µM rather than 20 µM protein. PEARLScreen was also more sensitive than the two other most common ligand-detected NMR methods, STD and WaterLOGSY.

 

We wrote recently about benchtop NMR, and the researchers found that PEARLScreen was also more sensitive than a T_1ρ experiment on an 80 MHz instrument, though the difference was not as dramatic as on the 600 MHz machine. On the other hand, on a 1.2 GHz instrument PEARLScreen was so sensitive that the researchers could screen mixtures of 16 fragments with just 1 µM protein.

 

This is a neat paper, which confidently concludes that “due to the superior sensitivity of the PEARLScreen compared to all established screening experiments at standard fields, we expect it to become the standard experiment for ¹H-detected ligand screening.” We look forward to hearing how it performs for others.

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.

A fragment prodrug discovered in a phenotypic screen

Glioblastoma multiforme (GBM) is a particularly nasty type of brain tumor with few drug options aside from the DNA alkylating agent temozolomide (TMZ), which is toxic and not particularly effective. Drugs fail for multiple reasons, among them the difficulty many small molecules have crossing the blood-brain barrier. A recent Nature paper by Luis Parada and a large group of collaborators at Memorial Sloan Kettering Cancer Center and elsewhere describes a promising new approach.

 

The researchers screened >200,000 molecules (not necessarily fragments) against primary GBM cells to look for compounds that reduced viability. Generically toxic molecules are so common that they (literally) grow on trees, so hits were counter-screened against mouse embryonic fibroblasts to looks for molecules that selectively killed GBM cells. This led to a rule-of-three compliant compound the researchers dubbed gliocidin.

 

Figuring out how gliocidin works turned out to be a complicated quest, starting with a genome-wide CRISPR-Cas9 screen to look for genes that either protected or sensitized cells to gliocidin. Subsequent work, including knocking out specific genes of interest and LC-MS/MS studies of metabolites, revealed that gliocidin leads to inhibition of a protein called inosine monophosphate dehydrogenase 2 (IMPDH2), which is necessary for guanine synthesis.

 

However, gliocidin is not a direct inhibitor of IMPDH2. Rather, it is is essentially a "pro-prodrug". Gliocidin is first converted into gliocidin-monocucleotide by the enzyme NAMPT (a target we wrote about back in 2014), and subsequently converted to gliocidin-adenine dinucleotide (GAD) by the enzyme NMNAT1. Cryo-EM showed that GAD binds at the NAD⁺ cofactor binding site of IMPDH2, blocking enzyme activity.

In addition to being a DNA-alkylating agent, TMZ induces NMNAT1 expression, thereby increasing conversion of gliocidin to GAD. Consistent with this, the combination of gliocidin and TMZ was more effective than either agent alone in mouse xenograft models. This is a lovely paper that reads like a detective story, and I’m only able to scratch the surface in a brief blog post. It also has multiple lessons for FBDD.

 

First, as expected given its molecular properties, gliocidin has excellent brain penetration. Vicki Nienaber argued in 2009 that FBDD may be ideally suited for finding molecules that can cross the blood-brain barrier, and gliocidin is a case in point.

 

Second, this paper answers emphatically in the affirmative the question we posed in 2022: “Is phenotypic fragment screening worthwhile?”

 

Third, this is another example of in situ inhibitor assembly to generate an analog of NAD⁺; we wrote about a small fragment targeting a different protein here. Given that fragments are the size of many metabolites, fragments as prodrugs could be a productive area of research.

 

But such a prodrug approach is not without risks. In that 2014 post about NAMPT inhibitors, I noted that some molecules had poorly characterized off-target activities, which could perhaps now be explained through this type of in situ activation. The new paper found that GAD does not inhibit two different NAD⁺ or NADPH-dependent enzymes, but hitting off-target enzymes will be something to watch for during optimization. I look forward to following this story.