Quality control of the iNEXT poised fragment library

Two weeks ago we described fragment libraries built for crystallographic screening, and this week we continue the library theme. A primary challenge after validating a fragment hit is what to do next. You can look for similar compounds, either in-house or from vendors, but ultimately you’ll need to do chemistry. And this might not be trivial: you don’t want to embark on a multistep synthesis if you can avoid it. One solution is to build “poised fragments,” which contain at least one functional group amenable to easy chemistry. Characterization of such a library has just been published (open access) in J. Biomol. NMR. by Harald Schwalbe (Goethe University) and a large multinational group of collaborators.

The researchers are part of the European iNEXT (Infrastructure for NMR, EM, and X-rays for Translational research) consortium. A set of 11,677 commercial fragments was computationally analyzed, and 782 were purchased from several vendors. To efficiently cover chemical space, the researchers took a “minimum fragments and maximum diversity” approach: similarity analysis revealed 391 clusters, most with just 1-3 members. The fragments are mostly rule of three compliant, though a bit on the large side, with 80% of fragments in the 200-250 Da range and an average molecular weight of 220 Da.

Fragments were characterized by NMR and LC-MS. NMR experiments were done in both d₆-DMSO (at 50 mM) and phosphate buffer (at 1 mM). The DMSO solutions were spiked with 10% D₂O; this mixture remains liquid at 4 °C, thus avoiding freeze-thaw problems. Fragment concentrations were established by NMR using either external or internal standards.

Some 30% of fragments did not pass quality control. This probably will not come as a shock to long-time readers, though it is a bit worse than some previous studies. Just like unhappy families, fragments failed QC for a variety of reasons, including impurities, degradation, solubility, and even inconsistency with the expected structures. There were also a couple cases where two fragments were mixed together, suggesting operator error while assembling the library. NMR spectra for various types of QC failures are provided in the paper.

The researchers are honest about deficiencies in the library, noting that it contains PAINful molecules such as catechols and hydroquinones, though one wonders why these were not removed in the first place. Laudably, they provide SMILES for all 782 compounds along with an extensive set of physicochemical calculations (see here – opens as an Excel spreadsheet). Weirdly though, the researchers do not specify which molecules failed QC or which vendors they came from. At first I thought these had been weeded out of the final set, but five examples shown as failures appear in the spreadsheet. The library is sold commercially by Enamine as the DSI-poised Library, and since this library contains only 768 compounds perhaps the most egregious bad actors have been removed. Enamine was rated highly in a reader poll, so presumably the compounds have all passed QC.

So how does the fragment library perform? That – unfortunately – is not addressed in this paper, though the DSI-poised library was among those screened against the SARS-CoV-2 main protease (M^Pro). Have you used it? And if so, has the “poised” nature of the library allowed you to efficiently grow fragment hits? Hopefully these questions will be answered in the literature, if not in the comments.

Fragments vs Mycobacterium tuberculosis InhA

Though this could change in a bad way, tuberculosis is currently the deadliest infectious disease worldwide, causing nearly 1.7 million deaths per year. Multidrug-resistant and extensively drug-resistant strains are widespread. Some approved drugs work by blocking the mycolic acid pathway essential for mycobacterial envelope formation. One member of the pathway, the enzyme InhA, is the target of isoniazid and ethionamide. Both of these molecules are prodrugs, and a major mechanism of resistance shuts down their bioactivation. To sidestep this problem, Mohamad Sabbah, Chris Abell, and collaborators at University of Cambridge and Comenius University in Bratislava have targeted InhA directly, as they describe in a recent open-access J. Med. Chem. paper. (See here for a previous FBLD effort against this target.)

The researchers began with a differential scanning fluorimetry (DSF) screen of 800 fragments, each at 5 mM. Forty-two fragments stabilized InhA by at least 3 °C and were tested at 1 mM in three ligand-based NMR assays: CPMG, WaterLOGSY, and STD. All 18 fragments that hit in at least two of these confirmatory assays were soaked into InhA crystals at 20 mM, yielding 5 hits.

None of the fragments inhibited enzymatic activity at 2 mM, but compound 1 was chosen for optimization based on an attractive growth vector into a hydrophobic region of the binding pocket. The carboxylic acid was replaced with an isosteric sulfonamide to yield compound 6, which has measurable activity. Various substituents were tested around the new phenyl ring, with a significant boost in activity caused by an aminomethyl moiety. Cyclizing the molecule and further medicinal chemistry ultimately led to compound 23, with high nanomolar activity. A crystal structure revealed that compound 23 bound as expected and that the primary amine was making interactions with the enzyme cofactor as well as an ordered water molecule.

Unfortunately, although compound 23 slightly inhibited the growth of M. tuberculosis, it did not inhibit synthesis of mycolic acids, suggesting that activity was through a different mechanism. The researchers suggest that the molecules may not be sufficiently cell permeable or that they are effluxed or metabolized. However, it may be that they just aren’t potent enough. Perhaps further medical chemistry will improve affinity by another couple orders of magnitude and achieve pathway inhibition. Regardless, this is a nice example of a robust biophysical assay cascade followed by fragment growing and structure-based design.

Crystallographic fragment screening roundup

Last year’s poll revealed that crystallography has become the most popular technique for FBLD. Three recent papers discuss some of the whys and hows.

The first publication, published in Structure by Manfred Weiss and collaborators at Helmholtz-Zentrum Berlin, Philipps-Universität Marburg, Freie Universität Berlin, and Lund University, describes the construction of the F2X-Universal Library for crystallographic screening. Starting from 1.4 million commercially available fragments, the researchers filtered out undesirable molecules such as PAINS and then clustered the remaining compounds by similarity. Among clusters with at least 1000 members they chose one fragment from each. The resulting library, sourced from roughly a dozen vendors, contains 1103 fragments individually dissolved in DMSO at 0.5 M concentration. Outside of the Diamond Light Source, screening 1000 crystals is still a sizable effort, so the researchers have also made a 96-membered subset library called F2X-Entry.

F2X-Entry was screened against two targets, the model protein endothiapepsin (EP) and the spliceosomal protein-protein complex Aar2/RNaseH-like domain of Prp8 (AR). Crystals of each were screened with 100 mM fragments and processed with the assistance of PanDDA (described here).

The results were quite impressive: 29 hits for EP, several of which bound in more than one site, and 20 hits for AR. The overall hit rates of 30% and 21% compare favorably with a previous crystallographic screen against EP that yielded a 20% hit rate. The solvent DMSO sometimes doesn’t play well with crystals, but the researchers were able to obtain 72-75% of the original hits when soaked in the absence of solubilizing DMSO.

One bit of data I would have liked to have seen is how the physicochemical properties of the fragment hits compare to the overall library, similar to what was reported here. Both libraries can be accessed by collaborators at the BESSY II synchrotron, and the smaller library can also be accessed through an MTA, so hopefully as these are screened against more targets this information will be published.

A second paper, in ChemMedChem by Gerhard Klebe and collaborators at Philipps-Universität Marburg and AstraZeneca, also discusses crystallographic screening of a 96-compound library, one sold commercially by Jena Biosciences. The target chosen was tRNA guanine transglycosylase (TGT), an enzyme important for the pathogenicity of the sometimes lethal bacteria Shigella. Soaking each of the compounds at 100 mM yielded 8 structures, 5 of which bound in the active site.

The researchers also screened the library using orthogonal methods, SPR and ligand-detected NMR. As in a previous study from the same group, the results were “puzzling”: none of the hits were detected in all three assays. (The first author, Engi Hassaan, presented some of this work at a CHI meeting last year.) Indeed, only one of the crystallographic hits was found among the 10 hits from SPR, and a different crystallographic hit was found among the 22 NMR hits.

Several plausible reasons for the low overlap are provided, including different buffer compositions and concentrations of fragments. Solubility likely played a role: a dozen fragments could not be screened by NMR at 0.2 mM, and 15 fragments interfered with the SPR assay and thus could not be screened. Indeed, some of the fragments are PAINS, including an eyebrow-raising dinitrocatechol, so the fact that they were not observed by crystallography is perhaps unsurprising.

Finally, in a Biomolecules paper, Gerhard Klebe and collaborators screen the same library against human carbonic anhydrase II (hCAII). This resulted in 9 hits – 8 from the library and 1 from a cryoprotectant used at 2.6 M in some experiments. Among the library hits, four bound at the active site, including a couple hydrazides which make interactions with the catalytic zinc ion. Surprisingly, two fragments bind covalently to the N-terminus of  hCAII. One seems to form a formaldehyde-mediated linkage, while the other – an aminomethylheterocycle – has likely oxidized to an aromatic aldehyde that can form an imine linkage.

All of the fragment hits in this case were identified through visual inspection of the electron density maps, and in this system PanDDA was actually not helpful, revealing only three of the fragment hits. The researchers note that they used different soaking times for the different crystals, and suggest that shorter soaking times in particular may not allow time for crystals to equilibrate, as assumed in PanDDA.

So in summary, fragment screening by crystallography is becoming easier and is likely to give high hit rates, particularly when conducted at high concentrations. Also, libraries matter: it is interesting that the F2X-Entry library gave considerably higher hit rates than the second 96-compound library, albeit on different targets. Of course, a fragment hit is only the beginning of a long journey, but at least a structure provides guidance to begin optimization.

Deconstructing an HTS hit for GyrB inhibitors

COVID-19 is deservedly engaging most of our collective mindspace when it comes to infectious diseases. Unfortunately, plenty other threats are out there, including antibiotic-resistant bacteria. A paper recently published in ACS Omega by Fumihito Ushiyama and colleagues at Taisho reports progress in this area.

The researchers were specifically interested in the protein DNA Gyrase B (GyrB), which is essential for bacterial replication (see here for previous work on the same target). A high-throughput screen against the E. coli protein led to a few dozen hits that were validated using a variety of biophysical methods including SPR, isothermal titration calorimetry (ITC), and crystallography. Compound 1 binds in the ATP-binding site, which is also where the natural product inhibitor novobiocin binds. The latter molecule makes an interaction with an arginine residue in the protein, but introducing a carboxylic acid moiety onto compound 1 to make a similar interaction was not successful (compound 8e).

Taking a step back, the researchers stripped compound 1 down to the core fragment 2a, which makes both polar and hydrophobic interactions with GyrB. Unfortunately, this fragment was too weak to show any affinity by ITC, as were 120 related fragments.

Looking closer at the structure of compound 1 bound to the protein revealed a small unfilled hydrophobic pocket near the 2-quinolinone fragment. Making appropriately substituted fragments was “relatively complicated,” and most of them were inactive. However, compound 2d showed binding by ITC as well as excellent ligand efficiency. Growing from this fragment ultimately led to compound 13e, with low nanomolar affinity. In addition to binding, compound 13e is a potent inhibitor of GyrB and is selective against a panel of 96 human kinases. Unfortunately though, it displays only modest antibacterial activity, likely due to efflux.

Nonetheless, this is a nice example of thoughtful structure-based design. In particular, the dramatic boost in potency gained by filling a small pocket (nearly 400-fold from compound 8e to 13e) validates the willingness to explore difficult chemistry rather than sticking with available analogs. The paper ends by noting that optimization is continuing, and I wish them well. By my count only a single fragment-derived antibacterial agent has entered clinical development, and that program is no longer active. We could use more.

BETting on fast follower fragments

A common approach in drug discovery is to improve a previously reported molecule. An example of such a “fast follower” approach has just been published in J. Med. Chem. by Cheng Luo, Bing Zhou, and colleagues at Shanghai Institute of Materia Medica.

The researchers were specifically interested in bromodomains, which recognize acetylated lysine residues in proteins and play major roles in gene expression. In 2018 we described AbbVie’s fragment-based discovery of ABBV-075, which had entered phase 1 clinical trials. Although reasonably selective for BET-family bromodomains, it also strongly inhibits EP300, which could lead to toxicity. Thus, more selective molecules have been sought.

The new paper starts with a thermal shift assay of 1000 fragments against the two separate bromodomains of BRD4, BD1 and BD2. Hits were validated using an AlphaScreen assay. Compound 47 was found to be active against both BD1 and BD2, with high ligand efficiency (all IC₅₀ values shown are for BD1; values for BD2 are similar). Modeling suggested this fragment could be merged with ABBV-075, and indeed the resulting compound 26 was quite potent. (Note: structures of compounds 26 and 38 were originally drawn incorrectly - now fixed.)

Compound 26 was metabolically unstable, but further optimization, aided by crystallography and modeling, ultimately led to compound 38. This molecule has good oral bioavailability in mice and promising pharmacokinetics in both mice and rats. It inhibits the expression of cancer-driving genes such as c-Myc and BCL-2, inhibits the growth of several cancer cell lines, and demonstrated good tumor growth inhibition in a mouse xenograft study. Compound 26 does not inhibit five cytochrome P450 enzymes or hERG. Finally, it is much more selective than ABBV-075 against EP300 and indeed most other bromodomains aside from BET family members. The researchers conclude that “compound 38 is a highly promising preclinical candidate.”

Unfortunately, selectivity for BET-family bromodomains may not be sufficient to avoid toxicity. Indeed, as we described earlier this year, AbbVie has dropped clinical development of ABBV-075 in favor of ABBV-744, which is selective for BD2 over BD1. Whether or not the same could be done for this series, the paper is still another nice example of appending a fragment onto a previously discovered molecule.