29 March 2021

Fragment chemistry roundup part 5

Last week we highlighted three chemistry-focused papers, and this week we’ve got three more. And to add a third three, all these papers address “three-dimensional” fragments.
The first, in ACS Med. Chem. Lett. by Brian Cox (University of Sussex), Philip Cox (AbbVie) and colleagues, describes using photochemical [2 + 2] cycloadditions to generate bridged pyrrolidine fragments, which can be further diversified.

The researchers analyzed 54 products by principal moments of inertia (PMI) and plane of best fit (PBF), which revealed them to be quite shapely, more so than AbbVie’s Rule of Three collection. Interestingly, amide derivatives with aromatic substituents were often shapelier despite being less saturated, and the researchers thus caution against using Fsp3 as a proxy for shapeliness – a point Teddy made several years ago.
The next paper, open access in ChemComm, also deals with cyclobutane-containing molecules. However, rather than building them from scratch in cycloaddition reactions, David Spring and collaborators at University of Cambridge and California State Polytechnic University Pomona use palladium-catalyzed C-H arylation to functionalize the rings. Lactonization and further derivatization generates a range of molecules.

The researchers generated a virtual library of 90 scaffolds that were rule-of-three compliant and quite shapely. It would be interesting to explore this chemistry on the bridged pyrrolidines of the previous paper – perhaps deliberately “losing control,” as discussed last week.
Finally, in Chem. Eur. J., Peter O’Brien (University of York) and a large group of collaborators from academia and industry describe (also open access) the “design and synthesis of 56 shape diverse 3-D fragments.” Because of their prevalence in drugs, pyrrolidines and piperidines were chosen as targets. The researchers specifically set out to make diverse molecules that would be shapelier (as assessed by PMI) than members of typical libraries. In considering three-dimensionality, the researchers considered not just the lowest energy conformation, as is typically done, but also other conformations with energies up to 1.5 kcal/mol higher; these would be present roughly 8% of the time at 37 °C. Some of the molecules are shown.

A PMI analysis of these fragments revealed them to be more three-dimensional than representatives of six commercial fragment libraries. In fact, although three of the commercial libraries are touted as being “3D,” a PMI analysis revealed them to “have only a marginally better 3-D profile compared to the standard 2-D rich commercial fragment libraries.” As in the paper discussed above, there was no correlation between Fsp3 and PMI.
Despite being relatively simple, 42 of these molecules had been previously unreported. In addition to their shapeliness, they also adhered to the rule of three, with an average ClogP of just 0.54. Moreover, 52 of the fragments were stable for >6 weeks in DMSO, 48 were stable in aqueous buffer for >24 hr, and 40 of them were soluble at >0.5 mM in buffer.
Most of these fragments are available for screening at the Diamond XChem facility. It will be interesting to see what kinds of hit rates they produce, and whether they generate superior leads. As we noted last year, the majority of drugs are not particularly shapely. Still, it is fun to explore new regions of chemical space, and these three papers are good starting points.

22 March 2021

Fragment chemistry roundup part 4

Last week’s post on diversity-oriented synthesis (DOS) reminds me that it has been nearly a year since we’ve done a post on fragment chemistry. Since then, several interesting papers have appeared. These next two posts will cover them.
The point of DOS is to generate lots of analogs from a small number of starting materials in a controlled fashion. But as any chemist knows, reactions often go out of control. In ChemMedChem, John Spencer (University of Sussex) and collaborators at several institutions have decided to turn lemons into lemonade by deliberately losing control, though they are quick to emphasize “in the selectivity (not health and safety) context.”
The researchers focused on C-H bond activation, a handy class of reactions in which a carbon-hydrogen bond is broken in order to generate a new molecule. C-H bonds are of course ubiquitous in drugs, and chemists normally try to selectively activate just one. Here, the researchers focused on Ru/Pd-catalyzed photochemical arylation in the presence of alcohols, which can further react with certain substrates. For example, reacting 2-phenylpyridine with a 4-fluorophenyldiazonium salt in methanol led to five products (some of which breach the rule of three).

These products, and those of other similar reactions, were screened crystallographically against an enzyme called NUDT7, resulting in one hit.
On a related chemical subject, Quentin Lefebvre and colleagues at SpiroChem explore photoredox-nickel dual catalyzed N-arylation reactions in Beilstein J. Org. Chem. In 4 days they tested 29 combinations of amines with various aryl halides, 15 of which gave products; examples are shown to the right.
Given that SpiroChem is a chemical vendor, expect to see more of these sorts of molecules in their catalog.
Finally, in Chem Sci., Nicholas Turner (University of Manchester - corrected) and collaborators at Keele University ask whether it is “time for biocatalysis in fragment-based drug discovery.” Biocatalysis involves using enzymes to run reactions. Despite stunning advances in synthetic organic chemistry, Nature is still the master, so why not work together?
The researchers review examples where biocatalysis could be used to generate new fragments or elaborate fragment hits. Importantly, enzymes can perform selective reactions even in the presence of multiple reactive centers that would normally need to be protected in conventional synthesis. Moreover, researchers are increasingly engineering enzymes to increase their substrate scope, efficiency, or completely alter the reactions performed. Some of the products are illustrated here.
I confess I haven’t done much biocatalysis, largely due to unfamiliarity but also because enzymes for organic synthesis don’t seem to be as widely offered by vendors as – for example – transition metal catalysts. Perhaps there is a market opportunity for fragment libraries designed for enzyme-mediated elaboration?
A decade ago, the main challenge of fragment-based drug discovery was finding fragments. Now it is elaborating them. It is nice to see solutions accumulating.

15 March 2021

Fragments from DOS, advanced

DOS – diversity-oriented synthesis – intentionally generates a disparate set of compounds from a small number of starting materials in just a few synthetic steps. The idea is that if any of these turn out to be hits, it will be straightforward to make analogs. Since figuring out what to do with a fragment is a common bottleneck, DOS-derived fragments could help. An open-access paper published in Chem. Sci. by David Spring (University of Cambridge) and collaborators from several institutions demonstrates how to use DOS to move fragment hits forward.
The researchers had previously disclosed (also open access) a rule-of-three-compliant DOS library of 40 compounds derived from racemic α-methyl propargylglycine. In the current paper, these molecules were screened crystallographically at 500 mM on the Diamond Light Source XChem platform against three protein targets.
The first, penicillin binding protein 3 (PBP3) from P. aeruginosa, is a classic antibiotic target. A single hit, compound 1, was identified. Interestingly, this turned out to be a covalent modifier, with the catalytic serine opening up the lactone. The researchers made 10 analogs exploring four different vectors, each in five synthetic steps using cheap reagents (< £3 per gram). These were screened crystallographically and six bound; one example is compound 6.
The next protein screened, cleavage factor 25kDa (CFI25), is a subunit of the pre-mRNA cleavage factor Im. (No, I hadn’t heard of it either.) A crystallographic screen yielded two hits, one of which – compound 19 – was elaborated into 14 analogs. This provided some preliminary SAR around the phenyl ring as well as a surprise: compound 31 bound to a different region of the protein.
Finally, the researchers screened activin A, a member of the transforming growth factor β superfamily. Compound 40 was elaborated into 14 analogs exploring four vectors, and compound 42 was found to bind in a similar manner.
There are several take-aways from this paper. First, DOS libraries can be remarkably diverse: compounds 1, 19, and 40 are all quite different from one another. Second, although I hesitate to discuss hit rates from such a small library, it is encouraging that hits were found at all even from fairly shapely fragments. These are also the first reported small molecule binders for CFI25 and activin A. Laudably, all the structures have been deposited in the protein data bank, extensive details are provided in 185 pages of supplementary material, and the library itself is available for screening at XChem.
One downside to crystallographic screening is that affinities are not part of the package, and some hits may be so weak as to be difficult to advance. But the researchers note they are further characterizing the compounds in the hope of producing more potent analogs. Although Teddy’s deadline for demonstrating a highly ligand-efficient molecule from DOS has long passed, hopefully the Safran Zunft Challenge will soon be met.

08 March 2021

Fragments in the clinic: LYS006

Practical Fragments’ first list of fragment-derived clinical compounds, published in 2009, listed just 17 molecules. The current count is approaching 50. One of the original compounds, DG-051, targeted the enzyme leukotriene A4 hydrolase (LTA4H). That molecule did not advance, but in a recently published J. Med. Chem. paper Christian Markert and colleagues at Novartis describe the discovery of a superior clinical compound.
LTA4H catalyzes the rate-determining step in the biosynthesis of leukotriene B4, a pro-inflammatory lipid implicated in multiple diseases. The researchers performed a differential scanning fluorimetry (DSF) screen of the enzyme against a library of 1800 fragments, 350 of which had been selected by in silico screening. This exercise yielded 14 hits that stabilized the protein against thermal denaturation, including compounds 1 and 2. Crystallography revealed that they both bind in the hydrophobic substrate pocket.

Merging compounds 1 and 2 led to compound 3, with low nanomolar potency. However, this lipophilic amine had poor stability in rat liver microsomes and also inhibited hERG and a couple CYP450s. Adding a carboxylic acid (compound 13) fixed these problems, though the molecule did bind to the dopamine transporter and had low solubility. Interestingly, the two enantiomers of compound 13 have very similar affinities, and crystallography revealed they could each bind in a similar fashion. Further optimization of the lipophilic tail ultimately led to LYS006. The crystal structure of this molecule overlays nicely with the initial fragments.
Much of the paper is devoted to characterization of LYS006, which appears to be a remarkably selective molecule. It does not bind or inhibit > 150 GPCRs, hERG, CYP450s, or a panel of metalloproteases. Oral bioavailability and pharmacokinetics are good in mouse, rat, and dog, and the molecule achieves essentially complete target inhibition at low nanomolar plasma levels. Moreover, LYS006 showed efficacy in preclinical efficacy studies. The drug is currently in four phase 2 clinical trials for ulcerative colitis, inflammatory acne, NASH, and hidradenitis suppurativa. (One of these began in 2018, yet only now are we finding out the origins of LYS006; this illustrates the difficulty of maintaining an up-to-date list of fragment-derived drugs.)
This is a beautiful drug discovery story with several lessons. First, like AZD5363 and many other examples, enzymatic potency was achieved relatively quickly; the bulk of the effort was focused on improving other properties. Second, the final molecule is not necessarily “surprising”: it contains the same biaryl-ether pharmacophore found in previous clinical compounds. Yet fragments along with careful medicinal chemistry allowed the researchers to obtain a best-in-class inhibitor.
Finally, this effort is a useful reminder that persistence can pay off: although LTA4H has been the target of drug discovery for decades, no inhibitors have yet been approved. Hopefully LYS006 will succeed.

01 March 2021

Fragments vs MEK1: allosteric binders

MEK1 is a central player in the MAP kinase signaling cascade, which is often dysregulated in cancer. As such the enzyme has been the focus of considerable research and the target of four approved drugs. Interestingly, these drugs bind not to the hinge region targeted by most kinase inhibitors but rather to an allosteric pocket adjacent to the ATP binding site. The drugs also look somewhat alike. Seeking something completely different, Paolo Di Fruscia, Fredrik Edfeldt, Helena Käck, and colleagues at AstraZeneca turned to fragments. They have recently published their results in ACS Med. Chem. Lett.
As we discussed in 2016, the AstraZeneca fragment library is quite large at 15,000 molecules. The researchers used a computational screen to narrow this down to a more manageable 1000 compounds for ligand-detected NMR screening. AMP-PNP, a nonhydrolyzable version of ATP, was included to block the hinge region, biasing the screen for fragments that bind the allosteric site. (See here for earlier work looking for ATP-competitive molecules.) A total of 142 fragments were identified and further characterized by SPR, and 46 showed dissociation constants better than 1 mM and similar affinities in both the presence and absence of AMP-PNP, suggesting they do indeed bind in the allosteric site.
Crystallography was attempted on all the fragments, but only two produced structures. Reassuringly, both bound in the allosteric site. But with only limited structural information, the researchers tested analogs of the fragment hits within their corporate collection. This identified compound 10, which is more potent than initial fragment 3. Moreover, compound 10 lends itself well to library synthesis.

All library members were initially made and tested as racemates. When the two enantiomers of the best hit were separated, compound 23 was found to be a sub-micromolar binder, roughly 100-fold better than the other enantiomer. At this point the researchers finally obtained a crystal structure of compound 23, confirming that it did bind in the allosteric pocket. Compound 23 is also still fragment-sized, just three heavy atoms larger than compound 3.
The astute reader will notice that the word “inhibitor” has not appeared until now, and indeed despite the encouraging affinity no mention is made in the paper of inhibition – a rather important feature! At a conference in 2019 Paolo did describe further optimization to a functional molecule, so hopefully we will see a second publication detailing this work.
Like the NPBWR1 story last month, this is another nice example of advancing fragments in the absence of structural information. It is also a good case study of fragments yielding completely different chemical matter in a crowded field.