Help develop new antibiotics from fragments!

The state of antibiotic drug discovery is – to put it mildly – dangerously poor. Not only do you have all the challenges inherent to drug discovery, you’re dealing with organisms that can mutate more rapidly than even the craftiest cancer cells. And then there’s the commercial challenge: earlier this month the biotech company Achaogen filed for Chapter 11 bankruptcy, less than a year after winning approval for a new antibiotic.

As Douglas Adams’s Golgafrincham learned, complacency about microbial threats is suicidal. But what can any one of us do? Chris Swain, whom we’ve previously highlighted on Practical Fragments, is involved with a consortium of researchers called Open Source Antibiotics. Their mission: “to discover and develop new, inexpensive medicines for bacterial infections.” And they are asking for our help. More on that below.

The researchers initially chose to focus on two essential enzymes necessary for cell wall biosynthesis, MurD and MurE, both of which are highly conserved across bacteria and absent in humans. They conducted a crystallographic fragment screen of both enzymes at XChem, soaking 768 fragments individually at 500 mM concentration. As we’ve written previously, you’ll almost always get hits if you screen crystallographically at a high enough concentration.

For MurD, four hits were found, all of which bind in the same pocket (in separate structures). Interestingly, this pocket is not the active site, but adjacent to it. The binding modes of the fragments are described in detail here, and the researchers suggest that growing the fragments could lead to competitive inhibitors. The fragments also bind near a loop that has been proposed as a target for allosteric inhibitors, so growing towards this region of the protein would also be an interesting strategy.

MurE was even more productive, with fragments bound at 12 separate sites. (Though impressive, that falls short of the record.) Some of these sites are likely artifacts of crystal packing, or so remote from the active site of the enzyme that they are unlikely to have any functional effects. However, some fragments bind more closely to the active site, and would be good candidates for fragment growing.

If this were a typical publication one might say "cool," and hope that someone picks up on the work sometime in the future. But this, dear reader, is different.

The researchers are actively seeking suggestions for how to advance the hits. Perhaps you want to try running some of these fragments through the Fragment Network? Or do you have a platform, such as “growing via merging,” AutoCouple, or this one, that suggests (and perhaps even synthesizes) new molecules? Perhaps you want to use some of the fragments to work out new chemistry? The consortium has a budget to purchase commercial compounds, and will also accept custom-made molecules. In addition to crystallography, they have enzymatic assays, and are building additional downstream capabilities.

The Centers for Disease Control identifies antibiotic resistance as one of the most serious worldwide health threats. Some have called for a global consortium—modeled after the International Panel for Climate Change—to tackle the problem. But in the meantime, you can play a role yourself. If you would like to participate, you can do so here. The bugs are not waiting for us – and they are already ahead.

A bestiary of fragment hits

What do fragment hits look like, and how do they bind? Fabrizio Giordanetto, David Shaw, and colleagues at D. E. Shaw Research were interested in these questions, and their answers are provided in a recent J. Med. Chem. paper (open access, and also covered well by Derek Lowe).

The researchers started by searching the protein data bank (PDB) for the word “fragment” and selecting higher resolution structures (at least 2.5 Å) with a ligand containing 20 or fewer non-hydrogen atoms. Those of you who have done bibliometric searchers will appreciate that a lot of manual curation is required, and the initial list of 5115 complexes was ultimately winnowed down to 489, with 462 unique fragments and 126 unique proteins, about two-thirds solved to ≤2.0 Å resolution.

In contrast to a previous study, only a minority (18%) of proteins contained more than one binding site, suggesting that secondary (possibly allosteric) sites may be less common than hoped.

As to the fragments themselves, 21 bound in more than one pocket (not necessarily on the same protein), including the universal fragment 4-bromopyrazole. The fragments ranged in size between 6 and 20 non-hydrogen atoms, with 81% having 10 to 16, consistent with our poll last year. Given these sizes, it is perhaps not surprising that the vast majority of fragments conformed to the rule of three.

Roughly two thirds of the fragments were uncharged, while 22% contained a negative formal charge (usually a carboxylic acid) and 11% contained a positive formal charge such as an aliphatic amine. Interestingly, more than 90% of the fragment hits were achiral. Since our 2017 poll found that most fragment libraries contain chiral compounds, these results might suggest lower hit rates for these compounds. Fragment hits also tended to have lower Fsp³ scores than those in a popular commercial library, which is consistent with the observation that “less shapely” fragments give higher hit rates.

Digging more deeply into the chemical structures themselves, nearly a third of the fragments contained a phenyl ring, while 6% contained a pyridine and 5% contained a pyrazole. Thiophenes, indoles, indazoles, piperidines, furans, and pyrrolidines were present in 2-3% of fragment hits. But there was also plenty of diversity: more than half of fragments contained a unique ring system.

So that’s what fragment hits look like. How do they bind? Deeply, for starters: about three quarters of the fragment hits buried more than 80% of their solvent-accessible surface area, and 21 fragments were completely engulfed within a protein.

Not surprisingly, more than 90% of complexes showed at least one polar interaction, such as a hydrogen bond or a coordination bond to a metal ion. Many complexes contained more than one, and one had seven! Interestingly, these polar interactions also tended to be buried. Nitrogen and oxygen atoms from the fragments were equally likely to form hydrogen bonds. Interactions with bound water molecules were considered for high-resolution (≤1.5 Å) structures, and nearly half of these contained a water molecule with at least two hydrogen bonds to the protein and one to a fragment.

Beyond these conventional sorts of polar interactions, there were also less traditional interactions such as arene-mediated contacts, which occurred in nearly half of cases. As we’ve noted, these are often under-appreciated but can be useful for improving affinity. The subject of halogen bonds came up recently, but these turned out to be quite rare, appearing in just 3% of cases. Sulfur-mediated contacts and carbon hydrogen bonds were more common, appearing in 11-12% of complexes, respectively.

All of this has important implications for fragment library design. As the researchers note, this set of 462 fragments could be used as the basis for a library, and laudably all the structures are provided in the supporting information. Generalizing beyond these specific molecules, roughly a quarter of the atoms in the fragment hits are polar (nitrogen or oxygen) and thus more likely to form classic hydrogen bonds. The researchers “strongly suggest” maintaining this ratio in designing new fragments.

The researchers also suggest presenting “a minimum set of individual polar pharmacophoric elements, as opposed to distributing several pharmacophores on a given fragment,” which is essentially the minimal pharmacophore strategy described here.

The one category of data I would have liked to see was affinity. Many binding measurements were probably not reported, and experimental error can be particularly confounding for weaker interactions, but even a subset of the data should allow some conclusions about the strength of various molecular interactions. Hopefully this will be the basis of a follow-up publication.

Third fragment-based drug approved!

Last Friday the US FDA approved erdafitinib (Balversa) for certain bladder cancers with FGFR2 or FGFR3 mutations. Although the fragment-to-lead story has yet to be published, those of you who were fortunate enough to attend Fragments 2019 last month heard some of it from Harren Jhoti.

Congratulations to the folks at Astex and J&J for a new tool in the campaign against cancer!

And earlier in the pipeline, several more drugs have entered the clinic starting from fragments, taking the number above 45.

Fourteenth Annual Fragment-based Drug Discovery Meeting

CHI’s Drug Discovery Chemistry (DDC) meeting took place last week in San Diego. I think this was the largest yet, with >825 attendees, a third from outside the US, and nearly 70% from industry. The initial DDC meeting in 2006 had just four tracks, of which FBDD is the only one that remains. This one had nine tracks and four one-day symposia, so it was obviously impossible to see everything. Like last year, I’ll just stick to broad themes.

Success Stories

As always, clinical compounds received deserved attention. Among two I’ve covered recently, Paul Sprengeler described eFFECTOR’s MNK1/2 inhibitor eFT508, while Wolfgang Jahnke discussed Novartis’s allosteric BCR-ABL1 inhibitor ABL001. As previously mentioned, ABL001 is a case study in persistence: the project started in stealth mode and was put on hold a couple times until seemingly intractable problems could be overcome.

Another story of persistence, albeit with a less happy outcome, was presented by Erik Hembre, who discussed Lilly’s BACE1 program. Teddy wrote about their first fragment-derived molecule to enter the clinic, LY2811376, back in 2011. Unfortunately this molecule showed retinal toxicity in three-month animal studies, so the researchers further optimized their molecule to LY2886721, which made it to phase 2 studies before dropping out due to elevated liver enzymes. Reasoning that a more potent molecule would require a lower dose and thus lower the risk of toxicity, the researchers used structure-based drug design to get to picomolar LY3202626, which also made it to phase 2 before being scuttled due to the apparent invalidation of BACE1 as an Alzheimer’s disease target.

Talks on BCL2 and MCL1 inhibitors from Vernalis, AstraZeneca, and Servier all involved fragments in some capacity, but unfortunately they were in the protein-protein interaction track which was held concurrently with the FBDD session I was chairing. Suffice it to say you can expect to hear more about the phase 1 compounds AZD5991 and S654315.

A few earlier-stage success stories included Till Maurer’s discussion of the Genentech USP7 program (see here), Santosh Neelamkavil on Merck’s Factor XIa inhibitors, and Rod Hubbard on Vernalis DYRK1A, PAK1, and LRRK2 inhibitors. We have previously written about how displacing “high-energy” water molecules can be useful, and this tactic was used by Sven Hoelder at the Institute of Cancer Research for their BCL6 inhibitors. Last week we highlighted halogen bonds, which proved important for transforming molecules that simply bind to MEK1 to molecules that bind and inhibit the protein, as described by AstraZeneca’s Paolo Di Fruscia.

Methods

The MEK1 story Paolo told began with a very weak (0.45 mM) fragment that the team was able to advance to 300 nM in the absence of structure, though they did eventually obtain a crystal structure that supported further optimization. On the topic of crystallography, Marc O’Reilly discussed the Astex MiniFrag approach, which we recently wrote about here. Only a couple of these fragments contain a bromine atom, but Marc did mention that, of the 10,051 X-ray complexes solved at Astex, a number show halogen bonds, including some to the hinge region in kinases.

At FBLD 2018 Astex’s Chris Murray showed the first cryo-EM structure of a fragment bound to a protein, and Marc confirmed that they have now obtained structures of fragments bound to two targets, with fragments as small as 120 Da and resolution as good as 2.3 Å. They are increasing automation, with turnaround times of less than 24 hours in some cases. Santosh also mentioned that Merck is applying cryo-EM to fragments.

Frank McCormick (UCSF) highlighted multiple fragment-finding methods used to discover inhibitors against RAS family proteins, which are responsible for more than a million cancer deaths each year. In addition to stalwarts such as crystallography and NMR, these include less common methods such as Tethering and the second harmonic generation (SHG) approach for detecting conformational changes used by Biodesy. RAS was reported as a cancer driver almost forty years ago, but only now are the first direct inhibitors entering the clinic – a testimony to both the challenging nature of the target and how far we’ve come.

SHG and Tethering were also highlighted elsewhere: Charles Wartchow described how SHG identified 392 hits from a collection of 2563 fragments against an E3 ligase bound to a target protein at Novartis, while Michelle Arkin described her use of Tethering at UCSF to find molecules that could stabilize a complex of 14-3-3 bound to a specific client protein (see here).

An effective sponsored talk was presented by Björn Walse of SARomics Biostructures and Red Glead Discovery, who described weak affinity chromatography (WAC). Once they saw the schedule for DDC, they looked for a target that would be presented shortly before their presentation, and chose the protein USP7 as a test case. Beginning in January, they screened a library of 1200 fragments to obtain 34 hits, of which 7 confirmed in a thermal shift assay. This led to an SAR-by-catalog experiment, and 11 of the 31 fragments tested showed activity, as did a Genentech positive control compound.

All methods can generate false positives and false negatives (see for example here and here), some of which were described in an excellent talk by Engi Hassaan of Philipps University. Engi discussed how improving the sensitivity of an STD assay by decreasing salt concentration identified more fragments that had previously been found by crystallographic screening. She also presented a case study of how introducing a tryptophan residue into a small protein to facilitate purification led to problems down the road when the tryptophan side chain blocked a key pocket in the crystal lattice. Gregg Siegal (ZoBio) also highlighted a case where a fragment bound to the dimer interface in a crystal structure, whereas in solution the fragment bound to the active site, as observed by NMR.

Finally, among computational methods, Pawel Sledz (University of Zurich) gave a nice overview of the SEED and AutoCouple methods, while Paul Hawkins (OpenEye) described rapid searching of more than 10 billion chemical structures using ROCS (rapid overlay of chemical features). SkyFragNet is looking closer with each passing year.

There is much more to say, so please feel free to comment. Several good events are still coming up this year, and mark your calendar for 2020, when DDC returns to San Diego April 13-17!

Helpful halogens in fragment libraries

A couple weeks ago we highlighted a small fragment collection (MiniFrags) designed for crystallographic screening. We continue the theme this week with two more papers on the topic, with an emphasis on halogens.

The first, published in J. Med. Chem. by Martin Noble, Michael Waring, and collaborators at Newcastle University, describes a library of “FragLites.” These small (< 14 non-hydrogen atom) fragments are designed to explore “pharmacophore doublets,” such as a hydrogen bond acceptor (HBA) next to a hydrogen bond donor (HBD). For example, the universal fragment 5-bromopyrazole contains an HBA separated by one bond from an HBD. The researchers constructed a set of compounds with either two HBAs or an HBA and an HBD separated by 1 to 5 bonds. Importantly, all compounds also contained either a bromine or iodine atom, the idea being that anomalous dispersion could be used to help identify the fragments using crystallography. A total of 31 FragLites are described, with between 1 to 9 examples for each type of connectivity.

As a test case, these were screened against the kinase CDK2, which has previously been screened crystallographically. FragLites were soaked into crystals at 50 mM, and 9 of the FragLites were found to bind in a total of 6 sites, 4 of which had not been previously observed. The anomalous signal provided by the halogens was important: when the researchers used only normal scattering they identified just 10 of the 16 binding events even when using the powerful PanDDA background correction method. The anomalous signal also helped clarify the binding modes.

The ATP-binding site is where 7 of the 9 FragLites bound, with all but one of them making hydrogen bonding interactions to the hinge region. While not surprising, this does demonstrate that the FragLites can be used experimentally to identify the best binding site. Interestingly, (2-methoxy-4-bromophenyl)acetic acid bound in the active site as well as three other secondary sites; one of these sites hosted three copies of the ligand! It will be interesting to see whether this fragment is generally promiscuous in other proteins too.

As the researchers note, the composition of the FragLite library can be optimized. For example, both of the HBA-HBD fragments with 1-bond separation were identified as hits, while only 3 of the 9 HBA-HBD fragments with 2-bond separation were. Is this due to the choice of fragments, the target tested, or both? The approach is conceptually similar to the Astex minimal pharmacophore concept, so it will be useful to include other types of pharmacophores too (a single HBA or HBD, for example).

A related paper was published in Front. Chem. by Frank Boeckler and colleagues at Eberhard Karls Universität Tübingen. Long-time readers will recall his earlier halogen-containing library designed for identifying halogen bonds: favorable interactions between halogens and Lewis bases such as carbonyl oxygen atoms. Perhaps because they have relatively stringent geometric requirements (2.75 – 3.5 Å, and a bond angle of 155-180°), halogen bonds are often ignored; the FragLite paper doesn’t even mention them.

The new Boeckler paper describes the construction of a library of 198 halogen-containing fragments, all of which are commercially available and relatively inexpensive. Most of these are rule-of-three compliant, though quite a few also contain more than three hydrogen bond acceptors. Also, given that each fragment contains a halogen, the molecular weights are skewed upward. Solubility was experimentally determined for about half of the fragments, but the highest concentration tested was only 5 mM, and even here several were not fully soluble.

Although no screening data are provided, the researchers note that their “library is available for other working groups.” In the spirit of international cooperation, I suggest a collaboration with the FragLite group!

Machines, fixing human disease

Last year we highlighted the secretive juggernaut DREADCO's move into drug discovery. Today they announced the launch of their new division SkyFragNet (not to be confused with the European graduate training program FragNet). Its audacious mission: “to eradicate human disease."

SkyFragNet will automate every aspect of drug discovery. The approach starts with a powerful docking method, in which all 166 billion members of GDB-17 will be docked against a target of interest. Synthetic schemes for the virtual hits will be computationally generated, and the compounds will be synthesized using automated flow synthesis and mass-directed purification.

Fragment hits that confirm in a panel of biophysical techniques will then undergo computational-based growing; SkyFragNet incorporates the latest AI algorithms to maximize the likelihood of success. As with the fragments, designed molecules will be synthesized and tested, first in biochemical and then in cell-based assays.

Although the folks at Mordor State College are trying to make animal testing obsolete, SkyFragNet will still rely on pharmaokinetic and pharmacodynamic studies. However, they have built a fully mechanized vivarium run entirely by robots - think of The Matrix but with mice in place of humans.

Finally, compounds that make it through this gauntlet will be scaled up under GMP conditions (automated, of course) for clinical trials. It remains to be seen how many compounds SkyFragNet will take into the clinic, or whether the success rates will be higher than those of their human counterparts.

Of course, with all this power comes enormous responsibility. If things go wrong, hopefully DREADCO will have the wisdom to Terminate the program. Eradicating human disease could be done in two very different ways.

Tiny fragments at high concentrations give massive hit rates

Screening fragments crystallographically is becoming more common, especially as the process becomes increasingly automated. Not only does crystallography reveal detailed molecular contacts, it is unmatched in sensitivity. At the FBLD 2018 meeting last year we highlighted work out of Astex taking this approach to extremes, screening very small fragments at very high concentrations. Harren Jhoti and colleagues have now published details (open access) in Drug Discovery Today.

The researchers assembled a library of 81 diminutive fragments, or “MiniFrags”, each with just 5 to 7 non-hydrogen atoms. Indeed, the fragments adhere more closely to the “rule of 1” than the “rule of 3.” Because the fragments are so small, they are likely to have especially low affinities: a 5 atom fragment with an impressive ligand efficiency of 0.5 kcal mol^-1 per heavy atom would have a risibly weak dissociation constant of 14 mM. In order to detect such weak binders, the researchers screen at 1 M fragment concentrations, almost twice the molarity of sugar in soda! Achieving these concentrations is done by dissolving fragments directly in the crystallographic soaking solution and adjusting the pH when necessary. Although this might mean preparing custom fragment stocks for each protein, it avoids organic solvents such as DMSO, which can both damage crystals and compete for ligand binding sites.

As proof of concept, the researchers chose five internal targets they had previously screened crystallographically under more conventional conditions (50-100 mM of larger fragments). All targets diffracted to high resolution, at least 2 Å, and represented a range of protein classes from kinases to protein-protein interactions. The hit rates were enormous, from just under 40% to 60%, compared to an average of 12% using standard conditions.

Astex has previously described how crystallography often identifies secondary binding sites away from the active site, and this turned out to be the case with MiniFrags: an average of 10 ligand binding sites per protein. In some cases protein conformational changes occurred, which is surprising given the small size and (presumably) weak affinities of the MiniFrags.

All this is fascinating from a molecular recognition standpoint, but the question is whether it is useful for drug discovery. The researchers go into some detail around the kinase ERK2, which we previously wrote about here. MiniFrags identified 11 ligand-binding sites, several of which consist of subsites within the active site. Some of the MiniFrags show features previously seen in larger molecules, such as an aromatic ring or a positively charged group, but the MiniFrags also identified new pockets where ligands had not previously been observed. The researchers argue that these “warm spots” could be targeted during lead optimization.

One laudable feature of the paper is that the chemical structures of all library members are provided in the supplementary material. Although it would be easy to recreate by purchasing compounds individually, hopefully one or more library vendors will start selling the set. If MiniFrag screening is standardized across multiple labs, the resulting experimental data could provide useful inputs for further improving computational approaches, as well as providing more information for lead discovery.

Better properties from fragments: c-Abl kinase activators

Last year we described the discovery of asciminib, an allosteric inhibitor of the kinase BCR-Abl that binds in the enzyme’s myristoyl-binding pocket. As we also highlighted nearly a decade ago, molecules that bind in this pocket can either inhibit or activate the enzyme. Although inhibitors have the most obvious therapeutic potential as anti-cancer agents, activators of the ubiquitously expressed c-Abl protein could potentially treat chemotherapy-induced neutropenia. In a recent J. Med. Chem. paper, Sophie Bertrand and coworkers at GlaxoSmithKline describe their efforts in this area.

The researchers started with a high-throughput screen of 1.3 million compounds. Among the hits was fragment-sized compound 2, which showed good binding and activation in biochemical assays but only modest activity in cells. Building off the left side of the molecule improved biochemical potency, but cell activity still lagged. SAR studies on the dichlorophenyl moiety suggested that this hydrophobic group was probably optimal, and a crystal structure of an analog bound to the enzyme confirmed this. Replacing the central thiazole with other aromatic rings also did little to improve cell activity.

The researchers acknowledge “that the chemistry strategy was largely pursuing compounds with rather poor physical properties,” notably low solubility, high lipophilicity, and high aromatic character. As co-author Robert Young has noted previously, physical properties matter. Happily, a fragment screen identified compound 28.

Adding the acetyl group from the HTS hit generated compound 29, with improved activity compared to the fragment. Moreover, this molecule had better solubility and permeability compared to the more lipohilic, thiazole-containing compound 2. Compound 29 also showed significantly improved activation of c-Abl in a cellular assay. Crystallography revealed that it bound in a similar fashion as compound 2, but with a twisted, more “three-dimensional” shape.

Further optimization, in part informed by previous work done on the thiazole series, ultimately led to compound 52, the most active compound synthesized. Another molecule in the pyrazoline series showed good pharmacokinetic properties in mice. Unfortunately, in vivo efficacy studies had to be halted early due to unexpected (and not clearly understood) toxicity.

This paper nicely illustrates several points. First, the power of fragment-assisted drug discovery, in which information from both HTS and FBLD is combined for lead optimization. Second, the inherently fuzzy line between FBLD and other discovery approaches: had compound 28 been tested in the HTS collection, it likely would have been a hit. Third, the importance of physicochemical properties. And finally, the inadequacy of potency and physicochemical properties alone to produce a developable compound. You can optimize your molecule to the best of your ability but still be sideswiped by nasty surprises such as toxicity. It is helpful to be clever in drug discovery, but you need to be lucky too.

Targeting RAS via PDEδ: another protein-protein interaction

Last week we highlighted molecules that inhibit the interaction between oncogenic RAS proteins and an activator protein, SOS1. This week continues the subject of fragments and RAS, but with a different protein-protein interaction, described in a recent paper in Eur. J. Med. Chem. by Min Huang, Naixia Zhang, Bing Ciong, and colleagues at Shanghai Institute of Materia Medica.

The researchers were interested in the protein PDEδ, which binds to lipidated RAS proteins and helps shuttle them to the plasma membrane. Blocking this protein-protein interaction could interfere with RAS signaling. PDEδ was screened against just 535 fragments using two ligand-observed NMR techniques (STD and CPMG), yielding five hits. Crystallography revealed that compound 1-H9 bound at the site where RAS normally binds. Other groups had previously identified molecules that bind in this same region, and the researchers used this information to grow their fragment to compound 16, with low micromolar activity.

Interestingly, a crystal structure of compound 16 showed that the binding mode had flipped relative to the initial fragment: the isobutyl group, which had been designed to replace the isopropylthio group, was binding in a region of the protein previously unoccupied by the fragment. Further growing led to compound 40, with mid-nanomolar potency in a biochemical assay.

Unfortunately, compound 40 and other molecules in the series had at best only modest activity in a viability assay of cells dependent on PDEδ. This result is in contrast to a previously reported PDEδ inhibitor, and the researchers suggest that the difference could be due to off-target activity of that molecule. Indeed, a third group has reported that inhibition of PDEδ would need to be nearly complete to be pharmacologically useful. As the researchers conclude somewhat optimistically, “all these complexities of PDEδ-associated proteins may impose a challenge and opportunity for PDEδ-targeted anticancer drug discovery.” While it is easier to see the challenges than the opportunities, this is nonetheless a nice example of using fragments to target a protein-protein interaction.

Stabilizing and destabilizing SOS1-RAS interactions

Last week we highlighted an example of fragments stabilizing a protein-protein interaction. This week continues the theme, with a paper published in Proc. Nat. Acad. Sci. USA by Roman Hillig, Benjamin Bader, and colleagues at Bayer.

The protein of interest was KRAS, inhibitors of which have long been sought as anti-cancer agents (see here and here for previous fragment efforts). KRAS binding to GTP activates cell survival and proliferation pathways. Guanine nucleotide exchange factors (GEFs) such as the proteins SOS1 and SOS2 facilitate the exchange of GDP for GTP. While inhibitors of this interaction would seem an obvious goal, other researchers had discovered molecules that stabilize the interaction, so the team looked for these too.

An STD-NMR screen of 3000 fragments (in pools of 8, each at 200 µM) yielded 310 hits, of which 97 bound to the complex of a mutant form of KRAS (G12C) and SOS1, but not to either isolated protein. Crystallography was attempted on 42 of these molecules, resulting in 13 structures. All compounds bound in a small hydrophobic pocket on SOS1, near where KRAS binds. Interestingly, two of these, including compound F1, stabilized the interaction between KRAS and SOS1, as assessed by 2-dimensional protein-observed NMR, SPR, and a biochemical assay. The remaining fragments bound to the complex but neither stabilized nor destabilized it. Unfortunately, efforts to improve the affinity of F1 proved unsuccessful.

Meanwhile, the researchers conducted an HTS screen of more than 3 million molecules, which they validated in a variety of biochemical and biophysical assays. Compound 1 passed all of them, and crystallography revealed that the naphthyl moiety binds in the same hydrophobic pocket of SOS1 as compound F1. Unlike the fragment, however, compound 1 inhibits the interaction of KRAS^G12C and SOS1. Structural analysis suggests that this is in part steric: one of the methoxy groups would clash with KRAS. Also, binding of compound 1 causes a conformation change in a critical tyrosine side chain of SOS1 that normally interacts with KRAS. Interestingly, the fragment F1 also interacts with this residue, but enforces a conformation similar to what it adopts when bound to KRAS, thus explaining the stabilization of the complex caused by F1.

Those of you who have worked on kinases will immediately recognize the quinazoline core of compound 1, and indeed this molecule inhibits kinases such as EGFR with nanomolar potency. This activity would make cell assays difficult to interpret, so the researchers added a methyl group to prevent interaction with the hinge region of kinases. Other changes improved the solubility, but only marginally improved the affinity of the best molecule, compound 17.

With two separate series, both of which bind in the same region, the researchers tried merging F1 and compound 17, ultimately leading to BAY-293, with low nanomolar affinity as assessed by isothermal titration calorimetry and functional activity in disrupting the KRAS-SOS1 interaction. Crystallography confirmed that the molecule binds as designed, with the amine group from F1 making similar interactions. BAY-293 was also active in a variety of cell-based assays, and should be a good chemical probe for better understanding the complexities of KRAS signaling.

Superficially BAY-293 bears more resemblance to its HTS parent than its fragment parent, and perhaps this story is best described as an example of fragment-assisted drug discovery. It is also a nice reminder that sometimes subtle chemical changes can make the difference between activation, disruption, or simple binding with no functional activity.

Stabilizing protein-protein interactions

Despite the fact that the second FDA-approved fragment-derived drug targets a protein-protein interaction (PPI), these types of targets have a well-earned reputation for being difficult. Most researchers try to disrupt PPIs. An alternative is to stabilize PPIs. This is not as crazy as it sounds: rapamycin, tafamidis, and PROTACs all stabilize PPIs. In a paper just published in J. Am. Chem. Soc., Michelle Arkin, Christian Ottmann, and collaborators at UCSF, Eindhoven University of Technology, Novartis, and the University of Duisburg-Essen bring fragments to bear on the problem.

The researchers were interested in the protein 14-3-3δ, a “hub” protein that binds to more than 300 other proteins (not all at the same time). One of these is estrogen receptor α (ERα): binding prevents the transcription factor from dimerizing and binding to DNA. The natural product fusicoccin A (FC-A) binds at the interface of 14-3-3δ and ERα and stabilizes that interaction, thereby inhibiting the growth of breast cancer cells. Because FC-A is a structurally complex natural product, the researchers sought fragments that would have a similar effect. They used Tethering, in which reversible disulfide bond formation stabilizes a protein-ligand complex, allowing its identification (see here and here). Specifically, fragments that bind near a cysteine residue are resistant to reduction, and the extent of binding can be detected by mass spectrometry.

The 14-3-3δ protein conveniently contains a cysteine residue in the vicinity of the ERα binding groove; the researchers used this native protein and also created two additional mutant proteins in which the native C38 cysteine was removed and new cysteine residues were introduced nearby. These three proteins were then screened against a library of 1600 disulfide-containing fragments under mildly reducing conditions in the presence or absence of a phosphopeptide derived from ERα. Most of the hits against the native protein were weak, but several hits against the N42C mutant were both resistant to reduction and also bound preferentially to the 14-3-3δ/ERα peptide complex compared to 14-3-3δ alone. Thus, ERα could enhance the binding of fragments to 14-3-3δ.

Next, the researchers used a fluorescently labeled peptide derived from ERα to show that one fragment could improve the apparent dissociation constant for the peptide and 14-3-3δ about 40-fold, from 1.3 µM 32 nM. Crystallography revealed that the cooperative fragments bound at the PPI interface, as expected given the location of the cysteine residues. The cooperative fragments placed a phenyl group in close proximity to a valine residue from the ERα peptide.

The researchers then examined the selectivity of one of their stabilizing fragments for other 14-3-3δ client proteins. In the case of a phosphopeptide derived from TASK3, which has a similar sequence to that of the ERα peptide, the fragment also showed cooperative binding. However, two peptides from other client proteins competed with the fragment for binding, and crystal structures revealed that the binding modes would be incompatible.

This is a nice illustration of site-directed fragment discovery to identify fragments that can modulate protein function in a more sophisticated manner than simple inhibition. One of the nice features of Tethering is that – like crystallography – it is able to identify extraordinarily weak binders. Unfortunately, this sometimes makes the hits challenging to advance: NMR experiments do show binding between a non-disulfide-containing derivative of one of the fragments and the 14-3-3δ/ERα peptide complex, but at high concentrations. It will be interesting to see whether this can be built into a potent non-covalent binder, and/or whether other types of covalent modifiers will be able to produce useful chemical probes for this target.

Metal-binding fragments vs GLO1

Practical Fragments has occasionally highlighted examples of metal-binding fragments. Strong interactions between low-molecular weight compounds and zinc, iron, or magnesium ions in metalloproteins makes for impressive ligand efficiencies. Unfortunately, some metal binders are PAINS and thus likely to inhibit a variety of targets; for others, the pharmacokinetic properties are not characterized. In a new J. Med. Chem. paper, Abraham Palmer, Seth Cohen, and colleagues at University of California San Diego describe a metallophilic molecule with in vivo efficacy.

The researchers were interested in glyoxalase 1 (GLO1), a zinc-dependent enzyme that catalyzes the clearance of the reactive metabolite methylglyoxal (MG). Although cytotoxic, MG may also have antidepressant effects. Thus, the researchers sought to find an inhibitor of GLO1.

They started by screening a library of 240 metallophilic fragments in a functional assay at 200 µM; more than 50 hits produced at least 50% inhibition. A second screen at 50 µM yielded 25 hits, including 8-MSQ.

Initial SAR studies revealed that both nitrogen atoms were essential for activity, suggesting a bidentate binding mode to the active-site zinc. Researchers at Chugai had previously reported a crystal structure of a very different molecule bound to GLO1, and this structure was used to model the binding mode of 8-MSQ. This exercise suggested growing from the sulfonamide, leading to compound 23. Incorporating information from other GLO1 inhibitors ultimately led to compound 60, with high nanomolar activity.

Those of you who have worked on drugs targeting the central nervous system may be concerned that compound 60 tends towards the large and lipophilic. However, when tested in mice at 12.5 mg/kg, it achieved a concentration of roughly 30 µM in the brain after two hours. Moreover, brain MG levels were increased 11-fold. Finally, mice dosed with compound 60 spent less time immobile in the forced swim test, a behavioral test used in rodent models of depression.

Overall, then, it seems that compound 60 has on-target activity in the brain and produces behavioral effects consistent with antidepressant activity. No selectivity data are provided, and because it could well be hitting other targets it is probably premature to use this as a chemical probe. Also, whether increasing the level of a toxic metabolite is a viable treatment for depression is likely to be hotly debated. Still, given the paucity of effective treatments for this widespread and devastating disease, it is nice to see researchers exploring bold mechanisms.

What will you do with hundreds of thousands of new ligands?

Ten years ago we highlighted a paper out of Brian Shoichet’s group in which 137,639 commercially available fragments were screened against the anti-bacterial target AmpC β-lactamase, resulting in a couple dozen weak hits, one of which was ultimately optimized to a picomolar covalent inhibitor. As evidenced by the devices in our pockets, computers have improved over the past decade. This is beautifully illustrated in a paper just published in Nature by Brian Shoichet, Bryan Roth, John Irwin, and an international team of collaborators at UCSF, UNC Chapel Hill, and labs in China, Ukraine, and Latvia.

Rather than limiting themselves to commercially available compounds, the researchers turned to a virtual set of make-on-demand molecules available from Enamine. These are built from 70,000 building blocks using 130 different two-component chemical reactions; 350 million molecules are currently available, with one billion expected by next year. The molecules exist virtually in the ZINC database, but can be physically ordered from Enamine as well. According to the Methods section of the paper, 93% of compounds ordered were successfully synthesized and delivered within six weeks.

The researchers screened 99 million virtual molecules using the program DOCK3.7. On average, 280 conformations of each molecule were fit into the active site in 4054 orientations. The top million compounds were then grouped by scaffold, and only molecules that differed considerably from known AmpC ligands and commercial compounds were considered further. Fifty one compounds were actually made and tested, of which five were active with affinities between 1.3 and 400 µM. Next, 90 analogs of these were synthesized, and more than half were active; the best came in at 77 nM, among the most potent non-covalent AmpC inhibitors ever reported. Crystal structures of several ligands from different scaffolds showed good agreement with the docking predictions.

For a test case against a very different binding pocket, the researchers turned to the D₄ dopamine receptor, against which they screened 138 million molecules in silico: 70 trillion different complexes, a process that took just 1.2 days using 1,500 cores. As with AmpC, the top hits were clustered, and anything resembling commercially available or known ligands was discarded. Of 549 compounds purchased and tested, 81 had K_i values of 8.3 µM or better. Many of the molecules were also active in functional assays, including full and partial agonists and even a couple antagonists. One molecule, a 180 pM agonist, was 2500-fold selective against the related D₂ and D₃ dopamine receptors. By way of comparison, in work published by some of the researchers just two years ago, the best hit from 600,000 commercial compounds was a 260 nM agonist which required three rounds of medicinal chemistry optimization to get to 4 nM.

How well did the hit rates correlate with the docking scores? The researchers separated the molecules screened against the D₄ receptor into a dozen “bins” and randomly chose 444 molecules from across the bins to make and test. Happily, the hit rates did in fact vary by score: among top bins, hit rates were 22-26%, dropping to 12% in the middle, and 0% at the bottom. Based on these numbers (and considerably more sophisticated analyses, including Bayesian statistics), the researchers suggest that the library of 138 million molecules contains more than 453,000 D₄ receptor ligands in more than 72,600 scaffolds with inhibition constants of at least 10 µM, and perhaps 158,000 with K_i values of 1 µM or better. These may well be conservative estimates, as they assume no hits among poorer-scoring molecules.

In a human-machine head-to-“head” contest, the researchers chose 124 of the top-ranked molecules manually and another 114 based on docking scores alone. Reassuringly, carbon-based systems held out over silicon, with hit rates for both sets around 24% but the human-chosen molecules typically having higher affinities, including the 180 pM winner. But while human performance will likely remain steady for the near future, machines will continue to improve.

On an academic level, the approach described in this paper could allow empirical tests of the molecular complexity hypothesis. It would be fascinating to see whether hit rates are higher for smaller molecules than larger ones, though of course smaller ligands are likely to have lower affinities and are thus less likely to be among the top hits. As in a previous analysis from Astex, one would need to compare hit rates among molecules with equal numbers of non-hydrogen atoms.

On a practical level, ultra-large library docking could be a game-changer for targets that have been structurally characterized. If the method proves generalizable, the question a decade hence may not be how to find hits, but rather how to choose between hundreds of thousands of them.

Taking a step towards STEP activators

Most drugs – and small molecule modulators in general – inhibit something, often an enzyme. Enzyme activation, on the other hand, is rare; we’ve highlighted just a few cases on Practical Fragments over the past decade. A new example is described in J. Med. Chem. by Christofer Tautermann and collaborators at Boehringer Ingelheim and the Beckman Research Institute of the City of Hope.

The researchers were interested in the protein tyrosine phosphatase non-receptor type 5 (PTPN5), also known as striatal-enriched protein tyrosine phosphatase (STEP). As its name suggests, this enzyme is found in the brain, and has been implicated in multiple neuropsychiatric disorders. However, phosphatases are tough targets due to their small, polar active sites. The problem is exacerbated for CNS targets, because negatively charged molecules have a hard time crossing the blood-brain barrier. Thus, the researchers sought allosteric modulators.

They began with a screen of 3083 fragments using STD NMR, differential scanning fluorimetry, and microscale thermophoresis. Validation of the several hundred hits by 2-dimensional NMR confirmed just seven, and comparison of the protein chemical shifts with those caused by a non-specific active site binder (sodium vanadate) suggested that compound 2 bound outside the active site. Crystallography confirmed this, revealing that the compound binds on the “back side” of the protein, about 20 Å from the catalytic pocket. The affinity was extraordinarily weak, with no functional activity, so the researchers used NMR to drive the SAR. Ultimately this led to fragment-sized BI-0314, with measurable affinity by isothermal titration calorimetry (ITC). Crystallography revealed that it binds in the same pocket as compound 2.

Surprisingly, far from being an inhibitor, BI-0314 actually showed activation of the enzyme in functional assays, increasing the activity by up to 60% at 0.5 mM. Careful mechanistic analysis revealed that this was due to an increase of k_cat, while the K_M for substrate was mostly unchanged. Molecular dynamics simulations suggested that BI-0314 increases the rigidity of the enzyme, and also stabilizes the active conformation. As expected of an allosteric modulator, the molecule was selective for STEP, with no activity (activating or inhibitory) for a couple other phosphatases.

As it turns out, the researchers were actually interested in STEP inhibitors, so they didn’t pursue BI-0314 further. As they note, there is still much to be done to generate a useful chemical probe, in particular improving potency. Laudably, the researchers are making BI-0314 available to other researchers free of charge. Perhaps someone else will be able to take this forward, as we’ve seen for other published fragments. And indeed, as researchers at Novartis have shown with asciminib, the transition from an allosteric binder with no functional activity to an inhibitor is possible – perhaps the same will hold true for an activator. If you are interested in STEP, you now have a new site to explore, and even a well-characterized starting point.

Readers beyond bromodomains: Fragments vs YEATS

Epigenetic readers recognize modified amino acids in histone proteins to cause changes in gene expression. Readers containing bromodomains, which recognize acetylated lysine residues, have received particular attention, and fragment-based approaches have led to at least a couple bromodomain inhibitors entering clinical development. But the numerous bromodomains are not the only epigenetic readers to recognize acetylated lysine residues. In a recent paper in J. Med. Chem., Apirat Chaikuad, Stefan Knapp, and collaborators at Goethe-University Frankfurt and University of Oxford describe their efforts targeting a different family.

YEATS domains are present in four human proteins, three of which have been linked to cancer. Unlike bromodomains, YEATS domains recognize lysine residues modified with acyl derivatives beyond acetyl, such as propionyl, butyryl, and crotonyl. The biological significance of these modifications is not clear, and no inhibitors of these proteins had been reported when the work began.

The researchers focused on the oncogenic eleven-nineteen-leukemia protein (ENL). They solved the first apo crystal structure of ENL (ie, without a bound ligand), which revealed that although the binding pocket was pre-formed, there was some flexibility in the side chain residues. They also noted distinct differences in how the acylated lysine is recognized, including the absence of an asparagine residue that is conserved in all bromodomains, and a more-open pocket that can accommodate larger acyl chains.

Next, the researchers chose a set of nineteen fragments containing a central amide bond to mimic acetylated lysine. None of these showed activity in a thermal shift assay, but when the ligands were soaked (at 5-40 mM) into crystals of ENL, electron density consistent with binding was observed for ten of them, and two could be modeled with some confidence. (For the other nine compounds, the crystals no longer diffracted.) These two fragments also showed binding by isothermal titration calorimetry (ITC). This is a useful reminder of the need for orthogonal assays, and the power of crystallography to detect weak hits. Compound 19, a rather super-sized fragment, was similar to compounds identified in a high-throughput screen that the researchers reported here and here.

Using this information, the researchers made a handful of analogs and found that compound 20 had high nanomolar affinity as assessed by ITC. Like last week’s story, this effort could probably be considered more fragment-assisted than fragment-based. But whatever the precise genealogy, hopefully molecular descendants of compound 20 will help to elucidate the biological poetry of the YEATS domains.

Fragments vs PI3Kδ via deconstruction and regrowth

Ligand deconstruction, in which a larger molecule is dissected into component fragments that are subsequently optimized, can be useful for developing new chemical series. This is nicely illustrated in a paper recently published in J. Med. Chem. by Kenneth Down and colleagues at GlaxoSmithKline.

The researchers were interested in phosphoinositide 3-kinase δ (PI3Kδ), a popular target for a variety of indications from oncology to inflammation. They had already developed GSK2292767 as a clinical candidate, but they wanted a backup with a different chemotype. Crystallography revealed that the indazole moiety was interacting with the hinge region of the protein. Trimming off the top of the molecule (compound 4) led to a loss of both potency and specificity against three related members of the lipid kinase family, not surprising given the fact that indazole is a privileged fragment for kinases in general.

To generate a new series, the researchers sought to replace the indazole hinge binder using modeling and previously published information. Starting with a selection of more than 30 possible hinge binders, they synthesized 324 molecules and found that compound 11 was more potent and ligand efficient than compound 4, as well as reasonably selective against other PI3K isoforms. Growing this fragment-sized molecule led to compound 16, with low nanomolar potency against PI3Kδ, greater than 100-fold selectivity against three related PI3K isoforms and 29 additional kinases, good permeability, and activity in a cellular assay.

The careful observer will note that the dihydropyran hinge binder in compound 11 is shorter than the indazole in compound 4, and indeed crystal structures of compounds 11 and 16 complexed to PI3Kδ revealed that the pyridine sulfonamide fragment is shifted in the active site compared to the original drug molecule, accommodated by various conformational shifts in the protein.

This paper is a good illustration of what has been called fragment-assisted drug discovery. Nowhere in the article do the researchers use the phrase “fragment-based,” though they do refer to the pyridine sulfonamide as a “privileged fragment.” In the end, the proof of practicality is in the chemical matter, so we’ll need to wait until more is revealed about this series.

(Not) getting misled by crystal structures: part 5 – conformational heterogeneity

It’s been a while since the last installment in our “getting misled” series. One of the key issues with crystallography is that ligands are almost always modeled as binding in a single conformation. This does not necessarily reflect reality, as we discussed here. Indeed, as described here and here, subtle changes can cause ligands to dramatically change their binding modes, which could reflect the fact that the initial ligand itself had multiple binding modes, and the change simply shifts the equilibrium. In an effort to proactively seek out disparate binding conformations, Henry van den Bedem and a group of collaborators from Stanford, UCSF, Schrödinger, and Université Paris-Saclay have created a new program, which they describe in J. Med. Chem. (See here for In The Pipeline’s discussion.)

The open-source program, called qFit-ligand, starts with an existing protein-ligand structure and an electron density map. It first breaks the ligand into rigid fragments (such as rings) and rotatable bonds. Each rigid group is then allowed to move around and rotate to fit the density. Of course, this might entail the rest of the molecule moving as well to avoid bumping into the protein; up to five positions are stored for each rigid group. Combinations that best match the electron density are retained: for a ligand with three rigid groups, 15 conformations would be considered. Importantly, the entire process is automated.

The researchers validated qFit-ligand against a set of 73 reasonably high-resolution structures from the protein data bank (PDB) that had included two different binding conformations; they started with just one of the reported conformations and used their program to find the second. qFit-ligand was very effective at identifying cases where a terminal portion of the molecule had flipped or rotated, though less so for more difficult cases such as displacement of the entire ligand.

Next, the researchers turned to the D3R dataset of 145 high-quality, manually curated crystal structures, where qFit-ligand correctly identified 7 of the 10 structures with alternate conformations, and even identified an alternative conformation for a ligand that had not previously been detected.

The researchers then examined a large set of crystal structures that had been flagged as potentially dubious, and found several could be improved by including alternate conformations. Similarly, an examination of all 126 crystal structures of BRD2-4 bromodomain-ligand complexes in the PDB revealed that 12 almost certainly had previously undetected alternate binding conformations; another 24 likely did.

qFit-ligand strikes me as a powerful tool for getting beyond the static picture usually presented by crystallography. Because the program is automated, the researchers note, it should be complementary to high-throughput approaches such as PanDDa (which we described here). Of course, using qFit-ligand effectively assumes that everyone is aware of the potential for both false positives and negatives. As the researchers conclude, “communication between structural biologists, computational chemists, and medicinal chemists remains a requisite for successful, rational design.”

Fragment events in 2019

Happy New Year! Lots of exciting events scheduled this year, many of them in the first half.

March 20-22: Although not exclusively fragment-focused, the Sixth NovAliX Conference on Biophysics in Drug Discovery will have lots of relevant talks, and will be held in the nice city of Nice Cannes. You can read my impressions of the 2018 event here, the 2017 Strasbourg event here, and Teddy's impressions of the 2013 event here, here, and here.

March 24-26: The Royal Society of Chemistry's Fragments 2019 will be held in the original Cambridge. This is the seventh in an esteemed conference series that alternates years with the FBLD meetings. You can read my impressions of Fragments 2013 and Fragments 2009.

April 9-10: CHI’s Fourteenth Annual Fragment-Based Drug Discovery, the longest-running fragment event, will be held in San Diego. You can read impressions of the 2018 meeting here, the 2017 meeting here, the 2016 meeting here; the 2015 meeting here, here, and here; the 2014 meeting here and here; the 2013 meeting here and here; the 2012 meeting here; the 2011 meeting here; and 2010 here. Also, Ben Davis and I will teach an FBDD short course on April 8, and it comes with dinner!

April 28-May 1: If you're looking for an even more intensive course, the 15th EFMC Short Course on Medicinal Chemistry is entitled "Small Becomes Big in Medicinal Chemistry: Fragment-based Drug Discovery." This will be held near Leiden, and as the number of participants is limited to 35, you should register early.

May 19-21: Structures are often critical for FBLD. The Second Annual Industrial Biostructures America Conference, which will be held in La Jolla, CA, is sponsored by Proteros and will cover a range of structural techniques.

September 1-4: BrazMedChem2019 will be held in the Brazilian city of Pirinopolis, and I know there will be some FBLD-relevant content.

November 12-15: The third FBDD Down Under will take place in Melbourne, and given the success of the first, it should be excellent.

Fourth Quarter: If you can't make it to Nice, NovAliX will also be holding a biophysics meeting for the first time in Japan, likely Osaka or Tokyo, sometime between mid-October and early December. Stay tuned for further details.

Know of anything else? Add it to the comments or let us know!