Fragments vs Lp-PLA2 – less greasily

Human lipoprotein-associated phospholipase A2 (Lp-PLA₂) is an enzyme involved in lipid metabolism that is implicated in multiple diseases, from atherosclerosis to Alzheimer’s. Because the natural substrates are lipophilic phospholipids, it is no surprise that reported inhibitors are also large and hydrophobic. A case in point is darapladib: with a molecular weight of 667 Da and a clogP of 8.3 this is a poster child for molecular obesity – and it also failed in phase 3 clinical trials. A new paper in J. Med. Chem. by Alison Woolford (Astex), Joseph Pero (GlaxoSmithKline) and colleagues describes an effort to discover less lipophilic inhibitors.

The researchers performed a screen of 1360 fragments using thermal shift and ligand-detected NMR and a smaller screen of 150 fragments using crystallography. This yielded 34 fragments that were ultimately characterized crystallographically; screening commercial and in-house collections for related fragments yielded another 16. Interestingly, rather than clustering at a single hot spot, these fragments bound to different regions of the extended active site, with some – such as fragment 6 – a full 13 Å from the catalytic center. This is reminiscent of a fragment campaign against soluble epoxide hydrolase, another enzyme with a long, hydrophobic active site.

In addition to binding in an interesting site, fragment 6 also has good affinity and ligand efficiency. Moreover, its binding site overlaps partly with that of fragment 5. Thus, the researchers merged the two fragments together, resulting in compound 7, with submicromolar activity.

Further structure-guided optimization, which included growing into a polar region of the protein, ultimately led to compound 16, with low nanomolar potency.

Compound 16 has a molecular weight of 411 Da and a clogP of 3.4 and is correspondingly reasonably soluble (> 0.3 mM). Whereas darapladib showed a dramatic 700-fold potency decrease upon addition of human plasma – presumably due to nonspecific binding to other proteins – the decrease in potency for compound 16 is only 13-fold. Indeed, though darapladib is a picomolar binder, compound 16 is slightly more potent in plasma.

Unfortunately, compounds in this series turned out to have high clearance in rats, proving once again that lead optimization is often a frustrating game of whack a mole. Still, the fact that the researchers were able to develop smaller, more soluble inhibitors of an enzyme with such a lipophilic substrate gives hope that the game is perhaps winnable.

19F-NMR-guided fragment linking on BACE1

Fragment growing has been the dominant strategy of most of the recent posts involving lead optimization, consistent with our poll results. However, fragment linking can be powerful too, as illustrated by the recent approval of venetoclax, which was derived from fragment linking. A recent paper in J. Med. Chem. by Brad Jordan and colleagues at Amgen provides another nice case study.

Amgen researchers had previously used fragment growing to discover inhibitors of BACE1, an Alzheimer’s target which has been heavily tackled by fragments. However, the most potent molecules in the series also inhibited the related aspartic protease cathepsin D (CatD), which could cause serious side effects. The researchers sought to gain selectivity by building inhibitors to occupy the so-called S3subpocket of BACE1. To do so, they used ¹⁹F-NMR to find fragments that would bind to BACE1 in the presence of a “blocking compound” that filled most of the active site but not the S3subpocket. This led to the discovery of seven fragments, the most potent being compound 3. Interestingly, this fragment only bound in the presence of the blocking compound as assessed both by NMR and SPR. Also, it could be competed by a compound that binds in the S3subocket.

Having thus identified a fragment that bound in the presence of one of their inhibitors, the researchers used interligand NOE (ILOE) to determine how the two compounds bind relative to one another. This supported the idea that compound 3 binds in the S3subpocket, and also suggested how the fragment could be linked onto the existing lead series, exemplified by compound 5. Just four compounds were designed and synthesized, and all of them were more potent than either of the starting points, with compound 9 being the best. More importantly, this compound also proved to be ~2000-fold selective for BACE1 over CatD in enzymatic and cell-based assays.

Despite the excellent (high picomolar) affinity of compound 9 for BACE1, this is actually about 25-fold worse than would be predicted by a simplistic additivity of binding energies – a not uncommon occurrence when linking molecules. Still, with its combined used of multiple NMR techniques and structure-based design to solve a specificity challenge, this paper is worth perusing.

Covalent fragments writ large

We’ve written previously about irreversible covalent fragment-based lead discovery. The nice thing about irreversible inhibitors is that they have an infinite no off-rate: once they bind and react with a target, that protein is permanently out of action. A paper published today in Nature by Keriann Backus, Benjamin Cravatt, and colleagues at Scripps Research Institute takes this approach to a whole new level.

The researchers assembled a library of just over 50 fragments containing cysteine-reactive electrophiles, such as chloroacetamides and acrylamides; the average molecular weight was 284 Da. These were then screened against human cells or cell lysates using a proteomic approach called isotopic tandem orthogonal proteolysis-activity based protein profiling (isoTOP-ABPP). This technique, previously developed by the Cravatt laboratory, uses mass spectrometry to differentiate contents of treated and untreated cells and identify specific regions of proteins that are modified.

In all, 758 cysteine residues in 637 different proteins were found to be modified by at least one of the fragments. These included targets (such as BTK) with known covalent drugs as well as many proteins with no small molecule inhibitors. Even more exciting, this set included some particularly challenging classes of proteins, such as transcription factors and various adapter and scaffolding proteins. Most proteins only had a single modified cysteine, and these were not necessarily in the active site (see also here). Happily, computational docking did a good job of (retrospectively) predicting the modified cysteine residues.

The fragments themselves ranged significantly in how many cysteines they modified, from < 0.1% to > 15%, with a median of 3.8%. Interestingly, the correlation with intrinsic electrophilicity – as measured by reaction with the small molecule thiol glutathione – was fairly weak. This suggests that the fragments are modifying proteins based on other properties, such as specific interactions between fragment and protein.

The initial studies were done using cell lysates at high (500 µM) fragment concentrations. Follow-up studies in whole cells using 50-200 µM fragment gave similar results, with 64% of the cysteines from the lysate experiments reacting with the same fragments in cells, even at the lower concentrations. Interestingly though, four fragment-cysteine interactions were found only in cells and not in lysates.

One class of proteins you might expect reactive fragments to hit are cysteine proteases, such as the caspases, and indeed one chloroacetamide-containing fragment reacted with the active site cysteine of caspase-8 (CASP8). Surprisingly though, this fragment showed only marginal activity in an inhibition assay, and subsequent experiments revealed that it is selective for the inactive zymogen (or proenzyme) form of the protein, thereby preventing activation. This fragment does not react with the related caspases 2, 3, 6, or 9, though it does hit CASP10. Modest modifications led to a compound that was also selective for CASP8 over CASP10. These two molecules were used to show that both CASP8 and CASP10 appear to be essential for extrinsic apoptosis in primary human T cells, but not in the immortalized Jurkat T-cell line.

Of course, it will be essential to rigorously characterize any covalent molecules used as probes. Chloroacetamides are well-known electrophiles – so well known in fact that they are generally excluded from screening libraries, including those that helped define the original PAINS filters. A single digit percentage hit rate means that any given covalent fragment could easily hit hundreds of proteins. The researchers here do careful control experiments – such as using an inactive enantiomer and extensive proteomic analyses – but someone less careful could easily mislead themselves and others. Done rigorously, though, this is an exciting approach that may well increase the number of ligandable targets.

Fragments vs MetAP2: reversible inhibitors

Methionine aminopeptidases, or MetAPs, cleave the N-terminal methionine residue from newly translated proteins. The human enzyme MetAP2 is a potential target for obesity, as demonstrated by the impressive clinical results of beloranib. But this drug hasn't been approved, and patients have died while taking it. Beloranib is an irreversible inhibitor that may also hit other targets, so researchers at Takeda California have been seeking non-covalent inhibitors. They report their results in two recent papers in Bioorg. Med. Chem. Lett.

In the first paper, Zacharia Cheruvallath and colleagues describe a biochemical fragment screen of ~5000 fragments (11-19 non-hydrogen atoms) conducted at 0.1 mM. This produced an impressive number of hits (110 compounds with > 20 % inhibition), which were triaged based on both ligand efficiency and LLE, ultimately yielding 16 interesting fragments. In particular, fragment 6 is remarkably potent.

Crystallography was not successful for any of the fragments. Undeterred, the researchers performed classic “SAR by catalog” (and corporate collection) to develop a binding model. This quickly revealed that the hydroxyl group was unnecessary. It also suggested that one of the indazole nitrogen atoms might be interacting with an active site metal ion, and that the bromine might be pointing towards a hydrophobic pocket where the side chain of the methionine substrate normally binds. Growing led to compound 16, and a closely related compound was characterized crystallographically bound to the protein, confirming the model. Further optimization led to compound 38, with low nanomolar potency in both biochemical and cell-based assays, excellent selectivity against a panel of >100 other targets, good oral bioavailability, and reasonable pharmacokinetics. This compound caused dose-dependent weight loss in a mouse model of obesity.

The second paper, by Christopher McBride and colleagues, involved more dramatic changes to the fragment. The indazole 4 is very potent, but indazoles are quite common in the literature, so the researchers sought to scaffold-hop to a novel core. This led them to design compound 6’, and using some of the SAR from the previous series ultimately led to compound 10. As with compound 38 above, this compound showed good cell-based activity, acceptable pharmacokinetics, oral bioavailability, and a clean profile against > 100 off-targets at 10 µM. It also showed measurable weight loss in a rodent model of obesity.

The question sometimes arises as to how many fragment hits are necessary for a program to move forward. These two papers show that a single fragment can be elaborated to two very different lead series with animal efficacy. In contrast to some of our recent posts, these efforts did not initially require crystallography. There are many ways to advance fragments, and no single technique is essential.

Fragments vs Dengue virus polymerase

Dengue fever, evocatively called “breakbone fever” for the severe pain it can inflict, is caused by a mosquito-borne virus that infects hundreds of millions of people each year. There are no approved antiviral treatments. Two papers from researchers at the Novartis Institute for Tropical Diseases and the University of Texas Galveston provide some promising early leads.

The first, in J. Biol. Chem., by Christian Noble, Pei-Yong Shi, and collaborators, describes a crystallographic screen of 1408 fragments against Dengue virus RNA-dependent RNA polymerase (DENV RdRp), which is highly conserved among the four serotypes of Dengue virus. Crystals were soaked in pools of eight fragments, with each present at only 0.625 mM, ten to one hundred times lower than other recent crystallographic screens. Perhaps because of this low concentration, only a single hit was identified – compound JF-31-MG46. The crystal structure revealed that the molecule binds in the “palm subdomain” of the protein, which is analogous to a druggable site on the hepatitis C virus protein.

Surface plasmon resonance (SPR) showed that this fragment had a dissociation constant of 0.21 mM against RdRp from serotype 3 and 0.61 mM against RdRp from serotype 4, suggesting weak but real binding. Isothermal titration calorimetry (ITC) was not successful, perhaps because of compound solubility, but replacing the terminal phenyl group with a thiophene led to more potent compounds which could be characterized both by SPR and ITC. The compounds were also active in an enzymatic assay, with IC₅₀ values comparable to their affinities.

The second paper, by Fumiaki Yokokawa and collaborators and published in J. Med. Chem., describes the optimization of these fragments. Fragment growing was performed to try to displace a bound water molecule, resulting in the low micromolar compound 17. Compounds that contain carboxylic acids often have low cell permeability, so several bioiosteres were tested to try to replace this moiety, and compound 23 showed increased affinity. However, this compound was still quite polar, showed poor permeability, and no cell activity. Adding a lipophilic substituent and decreasing the acidity led to compound 27, with nanomolar affinity and enzymatic inhibition of all four Dengue virus serotypes. Importantly, this compound also showed low micromolar activity against all four serotypes in cell assays.

The J. Med. Chem. paper notes that a high-throughput screen against RdRp had been plagued with false positives. One validated low micromolar hit was optimized to nanomolar potency, but this was very lipophilic and displayed no cell activity. It is interesting that the fragment-derived leads initially displayed no cell activity for the opposite reason: they were too polar. This is a useful reminder that physicochemical properties matter. The successful optimization of the fragment-derived series suggests that it can be easier to make leads more lipophilic than less.

Fragment library vendors - 2016 version

It's been two years since we last updated our list of commercial fragment libraries, and there have been several changes. The prompt for updating the list is a new Perspective published in J. Med. Chem. by György M. Keserű & György G. Ferenczy (Hungarian Academy of Sciences), Mike Hann & Stephen Pickett (GlaxoSmithKline), Chris Murray (Astex), and me. This covers all aspects of fragment library design, so definitely check it out.

One table in the Perspective compares various libraries, both commercial and proprietary. One of the manuscript reviewers asked if we could evaluate the various vendors, particularly given some negative experiences with commercial compounds. Such direct criticism (and praise!) can be awkward in the peer-reviewed literature, but is more acceptable in an online forum - think of Yelp for library suppliers. Please comment (anonymously if desired) if you've had experiences, positive or negative, with these vendors, and please feel free to add any we omitted.

Note that this list only includes companies that sell their libraries (as opposed to just using them internally).

ACB Blocks: 1280 compounds, 19F NMR-oriented, RO3 compliant, predicted to be soluble, purity >96%

Analyticon: 213 compounds, fragments from nature, RO3 compliant, high solubility, purity >95%

Asinex: >22,000 compounds

ChemBridge: >7000 compounds, RO3 compliant with predicted solubility; minimum purity 90% by 1H NMR

ChemDiv: >4000 3D fragments

Enamine: Multiple subsets including >18,000 RO3 compliant, ~1800 "Golden", and >126,000 with < 20 heavy atoms. Also separate fluorinated, brominated, sp3-rich, and covalent subsets.

InFarmatik: 1700 member consolidated library with different subsets (3D, GPCR, kinase)

IOTA: 1500 diverse, mainly RO3 compliant fragments

Integrex: 1500 compounds with diversity in shape and chemical structure, RO3 allowing one violation

Key Organics: ~26,000 compounds total with multiple subsets including 1166 with assured solubility and RO3 compliant as well as brominated, fluorinated, and CNS-directed fragments

Life Chemicals: 31,000 fragments of which 14,000 are RO3 compliant; also fluorinated, brominated, covalent, Fsp3-enriched, and covalent subsets

Maybridge: >30,000 fragments in total. The 2500 Diversity collection is guranteed soluble at 200 mM in DMSO and 1 mM in PBS.  NMR spectra are available (in organic solvent). It is available in many formats, from powder to DMSO-d6 solution. A smaller 1000-fragment subset is also available.

Otava: >12,000 fragments with various subsets including fluorinated, brominated, and metal-chelating

Prestwick: 910 mainly derived from drugs, RO3 compliant

Timtec: 3200 compounds, structurally diverse with predicted high solubility

Vitas-M: ~19,000 fragments, RO3 compliant

Zenobia:  968 fragments from different design paradigms, cores from drugs, higher Fsp3, flexible cores