29 November 2021

DeepFrag: fragment optimization by machine learning

Machine learning is becoming increasingly common in drug discovery. Just a few months ago we highlighted its use to design a library of privileged fragments. However, constructing a library is usually done infrequently (though continued renovation of a library is always a good idea). In two papers from earlier this year, Jacob Durrant and colleagues at University of Pittsburgh use machine learning to tackle the more common task of lead optimization.
The first paper, in Chem. Sci., describes DeepFrag, a “deep convolutional neural network for fragment-based lead optimization.” The researchers started with the Binding MOAD database, a collection of nearly 39,000 high-quality protein-ligand complex structures from the Protein Data Bank. Ligands were computationally fragmented by chopping off terminal appendages less than 150 Da. The fragments were then converted into molecular fingerprints encoding their structures. Meanwhile, the protein region around each ligand was converted into a three-dimensional grid of voxels, akin to how images used for computer vision training are processed.
The researchers describe the goal as follows. “We propose a new ‘fragment reconstruction’ task where we take a ligand/receptor complex, remove a portion of the ligand, and ask the question ‘what molecular fragment should go here.’”
About 60% of the data were used in a training model for the machine learning algorithm. This was then evaluated on 20% of the data and further refined before the final evaluation on the remaining 20% of the data. The details are beyond the scope of this post (and frankly beyond me as well) but DeepFrag recapitulated known fragments about 60% of the time. Importantly, the model worked for diverse types of fragments, including both polar and hydrophobic examples. Even “wrong” answers were often similar to the “correct” responses, for example a methyl group instead of a chlorine atom. In some cases where DeepFrag’s predictions differed from the original ligand the researchers note that these may be acceptable alternatives, a hypothesis supported by subsequent molecular docking studies.
Of course, the goal for most of us is not to recapitulate known ligands but to optimize them, so the researchers applied DeepFrag to crystallographically identified ligands of the main protease from SARS-CoV-2. Many of them docked well, though they have yet to be synthesized and tested.
Laudably, the model and source code have been released and can be accessed here. However, as these require a certain amount of computer savvy to use, Harrison Green and Jacob Durrant have also created an open-source browser app which is described in an open-access application note in J. Chem. Inf. Mod.
The browser app runs entirely on a local computer, without requiring users to upload possibly sensitive data. The application note describes using the app to recapitulate an example from the original paper. It also describes using it on a fragment bound to antibacterial target GyrB, a fragment-to-lead success story we blogged about last year. DeepFrag correctly predicted some of the same fragment additions that were described in that paper.
The app is incredibly easy to use: just load a protein and ligand (from a pdb file, for example) and the structure appears in a viewer. Click the “Select Atom as Growing Point” button, choose an atom, and hit “Start DeepFrag.” The ranked results are provided as SMILES strings and chemical structures, and the coordinates can also be downloaded. You can also delete atoms before growing if you would like to replace a fragment.
In my own cursory evaluation, DeepFrag correctly suggested adding a second hydroxyl to the ethamivan fragment bound to Hsp90 (see here). It did not suggest an isopropyl replacement for the methoxy group, but it did suggest methyl. Trying a newer example unlikely to have been part of the training set did not recapitulate the ethoxy in the BTK ligand compound 18 (see here), but did suggest a number of interesting and plausible rings. Calculations took a few minutes on my aging personal Windows laptop using Firefox.
In contrast to the hyperbolic claims too often seen in the field, the researchers conclude the Chem. Sci. paper modestly: “though not a substitute for a trained medicinal chemist, DeepFrag is highly effective for hypothesis generation.”
Indeed – I recommend playing around with it. We may still be some way from SkyFragNet, but we’re making progress.

22 November 2021

Selective fragments vs GPCRs, guided by modeling

Earlier this year we highlighted a fragment optimization success story against a G protein-coupled receptor (GPCR) which made no use of structural information. Due to the difficulty of crystallizing these membrane-bound proteins, structures have been rare for this large class of drug targets. Advances in crystallography are starting to change that. In a recent open-access Chem. Commun. paper, Jens Carlsson and collaborators at Uppsala University and the US National Institutes of Health make use of the increasing availability of such structures to develop potent, selective inhibitors.
The researchers were interested in A1 and A2A adenosine receptors (A1AR and A2AAR), targets for a variety of ailments from cancer to cardiovascular diseases. (A2AAR was the subject of this blog post a few months ago.) In the current study, the researchers wanted to know whether structures and molecular dynamics (MD) simulations could guide production of selective inhibitors.
Previous computational and experimental work from the authors had yielded compound 1, with low micromolar activity against A1AR and 7-fold selectivity over A2AAR. Crystal structures of both these proteins are available, though not bound to the small molecule. Docking studies suggested that the ligand would make similar interactions to both proteins, but that there might be an opportunity for increased selectivity towards A1AR due to the presence of a smaller threonine residue compared with a methionine in A2AAR. Nine analogs were designed to grow into this lipophilic pocket, and free energy perturbation and MD simulations suggested that they would have improved affinity for A1AR. This turned out to be the case when the molecules were made and tested in radioligand binding assays.

Although compounds 5 and 9 were more potent, selectivity was not improved. MD simulations suggested this might be due to the small size of the fragments, which could be accommodated in A2AAR by slight shifts in the binding modes. To try to anchor compounds within the pocket, the researchers grew off the phenyl ring, leading to molecules such as compound 15. Borrowing from this molecule and compound 9 led to compound 22, the most potent and selective molecule in the series. (A separate effort led to a somewhat weaker but A2AAR-selective ligand.) Both molecules were found to be antagonists when tested in cells, which was expected given that the crystal structures used for modeling were in the inactive conformation.
The correlation between predicted and measured binding energies was respectable, with a mean unsigned error (MUE) of 1.08 kcal/mol and Spearman’s rank correlation coefficient (ρ) of 0.8 for 24 compounds. Selectivity predictions were also impressive at MUE = 0.48 kcal/mol and ρ = 0.85.
This is a nice illustration of using computational methods to improve the affinity of a fragment by more than three orders of magnitude while also increasing selectivity. This particular system is probably on the easier side; we blogged about previous research from this group on A2AAR back in 2013. The researchers note that proteins with larger binding sites and weaker ligands are likely to be more challenging. It will be fun to see efforts towards Class B GPCRs, for example.

15 November 2021

Fragments vs SETD2: a chemical probe

Among the various epigenetic “writers,” only one is capable of trimethylating lysine 36 of histone H3. SET domain-containing protein 2 (SETD2) is thought to be a tumor suppressor, but some evidence suggests it may have the opposite effect in certain cancers. A chemical probe would be useful to resolve these conflicting ideas, and in an (open access) ACS Med. Chem. Lett. paper Neil Farrow and colleagues at Epizyme describe one.
Epizyme has been pursuing epigenetic targets for years and has built a methyltransfersase-biased compound collection. A radiometric screen of this library yielded compound 1 and a related molecule. Both were weak inhibitors, but a co-crystal structure with the enzyme revealed the indole buried deep in the substrate binding pocket. Tweaking this led to compound 4, with low micromolar activity.
Substitution off the indole and phenyl moieties ultimately led to compound 25, with low nanomolar biochemical and cell activity. However, this molecule also had low aqueous solubility and poor pharmacokinetics in mice. Recognizing that the lipophilic and aromatic nature of the molecule were likely responsible, the researchers returned to the initial hit. Replacing the phenyl with a cyclohexyl moiety and making a few more modifications ultimately led to EPX-719.
The pharmacokinetics of EPX-719 in mice are reasonable, and the molecule is >8000-fold selective against a panel of 14 other histone methyltransferases. It is also fairly clean against a panel of 47 off-targets and 45 kinases. EPX-719 showed antiproliferative activity in two multiple myeloma cell lines, and more detailed biological studies are promised in a future paper.
This is a nice hit to lead story. As the researchers note, “close attention to the physical chemical properties of the inhibitors, in particular basicity, lipophilicity, and aromatic character, led to compounds with attractive cellular activities and in vivo exposures.” Interestingly though, the word “fragment” does not appear once in the paper. Although compounds 1 and 4 venture a bit beyond the rule of three, I would argue that starting with small, low affinity binders and focusing closely on molecular properties is the very definition of fragment-based lead discovery.
A quarter-century of FBLD has influenced the scientific zeitgeist, and a fragment by any other name is still as sweet.

08 November 2021

Fragments in the clinic: 2021 edition

Since our last clinical update in 2020 two new fragment-derived drugs, asciminib and sotorasib, have been approved, bringing the total to six.

The current list contains 52 molecules, with 22 approved or in active trials. As always, this table includes compounds whether or not they are still in development (indeed, some of the companies no longer even exist). Because of this, the Phase 1 list contains a higher proportion of compounds that are no longer progressing. 
Drugs reported as still active in clinicaltrials.gov, company websites, or other sources are in bold, and those that have been discussed on Practical Fragments are hyperlinked to the most relevant post. The list is almost certainly incomplete, particularly for Phase 1 compounds. If you know of any others (and can mention them) please leave a comment.


PexidartinibPlexxikonCSF1R, KIT
Amgen KRASG12C
VenetoclaxAbbVie/GenentechSelective BCL-2
Phase 3

Pelabresib (CP-0610)
Phase 2

AT9283 AstexAurora, JAK2
IndeglitazarPlexxikonpan-PPAR agonist
MAK683NovartisPRC2 EED
Navitoclax (ABT-263)AbbottBCL-2/BCLxL
Cullinan Oncology / Wistar
Phase 1

ABBV-744AbbottBD2-selective BET
ABT-518AbbottMMP-2 & 9
AT13148AstexAKT, p70S6K, ROCK
AZD5099AstraZenecaBacterial topoisomerase II
BI 691751Boehringer IngelheimLTA4H
HTL0014242Sosei HeptaresmGlu5 NAM
NavoximodNew Link/GenentechIDO1

With only two phase 3 molecules in active development it may be some time before the next fragment-derived drug is approved. Then again, in 2020 sotorasib was only in phase 2. While long timelines are common in our industry, good drugs can make remarkably rapid progress.

30 October 2021

Asciminib: the sixth fragment-derived drug approved

Yesterday, on October 29, the US FDA approved asciminib (ABL001, from Novartis) for two subsets of patients with chronic myeloid leukemia (CML), making it the sixth fragment-derived drug to reach the market.
In common with the five other approved fragment-based drugs, asciminib is a cancer therapeutic. Like three of them, it is a kinase inhibitor. But there the resemblance ends. As we discussed at length in 2018, asciminib targets not the hinge region of BCR-ABL1, but an allosteric myristoyl-binding pocket on the protein. This unique mechanism of action provides improved selectivity over conventional kinase inhibitors, which could be part of the reason the drug causes fewer side effects than other BCR-ABL1 inhibitors.
Another advantage of targeting the allosteric pocket is to sidestep resistance. One group for which asciminib was approved is for patients with the BCR-ABL1 T315I mutation, which causes resistance to other approved therapeutics. Combining asciminib with other drugs might prevent resistance from emerging in the first place.
The approval of sotorasib in May was a study in speed, with less than three years spent in the clinic. In contrast, asciminib was first dosed in 2014. Even getting there was far from certain: as Wolfgang Jahnke recounted five years ago, the project started as a grass-roots effort and was halted twice. Imatinib, which targets the hinge region of BCR-ABL1, also faced a fraught journey to the clinic before being approved twenty years ago.
These stories of persistence paid off, and today humanity has a new weapon against CML. And this is just the beginning: a dozen clinical trials with asciminib are either announced or in progress. Practical Fragments wishes to offer everyone involved congratulations, luck, and thanks.

25 October 2021

Fragments vs TIM-3

In order to thrive, cancer cells need to evade the immune system. Preventing them from doing so is the goal of cancer immunotherapy. Although it has not entirely lived up to its initial hopes, this promising approach has generated multiple new targets, such as T-cell immunoglobulin and mucin domain-containing molecule 3 (TIM-3), whose upregulation correlates with tumor progression. Several antibodies targeting this protein are working their way through the clinic, but small molecules may have advantages in terms of oral dosing and improved tumor penetration. The discovery of one small molecule binder is reported in a new J. Med. Chem. paper by Stephen Fesik and colleagues at Vanderbilt University.
As is customary for this group, the project began with a two-dimensional (1H/15N HMQC) NMR screen of 13,824 fragments, each at 0.8 mM in pools of 12. This yielded 101 hits, a respectable 0.7% hit rate, and higher than might be expected for this immunoglobulin-like protein. The hits belonged to 11 chemotypes, and 18 had dissociation constants better than 1 mM and ligand efficiencies (LE) better than 0.25 kcal mol-1 per heavy atom. All of the fragments caused similar resonance perturbations, suggesting a common binding pocket, though as specific backbone resonance assignments were not known the exact location was unclear. Compound 1 was pursued due to its (relatively) high affinity, LE, and chemical tractability.
Substitutions off two vectors of the molecule improved affinity, and combining these substituents led to compound 22. This molecule bound sufficiently tightly that NMR could no longer be used to measure the dissociation constant. At this point the researchers were able to solve a crystal structure of the compound bound to TIM-3, revealing that it binds to a protein loop with the tricyclic core sandwiched between two tryptophan residues. The structure also revealed a portion of the molecule that extended toward solvent, and this insight was used to construct a fluorescent probe for use in a fluorescence polarization anisotropy (FPA) competition assay to accurately measure binding of more potent molecules.
With the probe results and crystal structure in hand, the researchers continued to optimize the molecule by growing towards a couple arginine and aspartic acid residues. This led to compound 34, which again started bottoming out the FPA assay and necessitated constructing yet another fluorescent probe. Further optimization using structure-based design ultimately led to compound 38, the most potent molecule in the series. NMR experiments revealed that compound 38 causes a rigidification of the TIM-3 loop where it binds.
And that’s where things stand. Unfortunately no data are presented as to whether compound 38 blocks binding of TIM-3 to its biological partners. The binding site is actually somewhat distant from where natural ligands bind, suggesting that the compounds would likely need to act allosterically. Moreover, the researchers note that many of the compounds are not particularly soluble. Still, whether the compounds move forward or not, this is a nice example of finding fragments that bind to a novel target and using diverse insights to improve them by several orders of magnitude.

18 October 2021

Fragments vs the SARS-CoV-2 main protease – this time by NMR

Last week we highlighted NMR screens against RNA from SARS-CoV-2. As we noted then, much of the drug discovery action has focused on the virus’s main protease, called Mpro or 3CLp. These efforts have included two separate screens by crystallography and/or mass spectrometry. A new (open-access) paper in Angew. Chem. Int. Ed. by Benoit Deprez, Xavier Hanoulle, and collaborators at CNRS and University Lille describes an NMR screen against this same protein.
The main protease is 306 amino acid residues long and forms a 67.6 kDa dimer in solution. Proteins this large are challenging for protein-detected NMR, both because the number of potentially overlapping resonances increases and because of line broadening. Nonetheless, the researchers used several sophisticated NMR techniques to assign more than 60% of backbone resonances as well as quite a few main-chain and side-chain hydrogens to gain information on binding locations.
A library of 960 fragments was purchased from Life Chemicals and Maybridge. These were pooled in groups of five, with each fragment at 377 µM, and screened by Water-LOGSY and, for 427 fluorinated fragments, 19F line broadening and chemical shift perturbation. This exercise yielded 159 hits.
These hits were validated in a protein-detected 1H, 15N TROSY-HSQC experiment, which confirmed 38 fragments, giving an overall hit rate of around 4%, comparable to that seen in the crystallographic fragment screen against Mpro. Also in common with the previous screen is the fact that most of the fragments (32) bind somewhere in the active site, while a few (5) bind at the dimerization interface. Fragment hits tended to be both larger and more lipophilic than those in the overall library.
Earlier this year we highlighted an NMR study of non-covalent fragment hits from the crystallographic fragment screen, which found that only two of them had measurable affinities, and both were weak (KD = 1.6 mM at best). In contrast, one of the new fragments, F01, has a dissociation constant of 73 µM. With a molecular weight of 287 Da and 20 non-hydrogen atoms this is a somewhat portly fragment, but it does have a ligand efficiency of 0.3 kcal mol-1 per heavy atom. It also shows functional activity with an IC50 = 54 µM in a biochemical assay and EC50 = 150 µM in an antiviral assay. The researchers further characterized their molecule crystallographically, which confirmed that it binds to the active site; it makes three hydrogen bond interactions and multiple hydrophobic contacts with the protein.
Although crystallography has been receiving increasing attention among fragment-screening techniques, this paper is a reminder than NMR remains highly relevant, even for larger proteins that crystallize easily. And at the end of the day, it’s not how you screen but what you find and what you do next. Hopefully folks will follow up on F01. While PF-07321332 is making rapid progress in the clinic against this enzyme and the COVID Moonshot effort is moving molecules into animal studies, HIV has taught us that we’ll need multiple small molecule drugs.

11 October 2021

Fragments vs SARS-CoV-2 RNA

The first mention of SARS-CoV-2 on Practical Fragments in early March of last year highlighted a crystallographic fragment screen against the main viral protease. As discussed last week this effort has now led to compounds with nanomolar activity in cells. We’ve also highlighted a separate crystallographic screen against this target as well as a screen against the Nsp3 macrodomain. But proteins are not the only potential viral targets.
A recent (open access) paper in Angew. Chem. Int. Ed. by Harald Schwalbe and a large group of collaborators mostly at Johann Wolfgang Goethe-University focuses not on proteins but on RNA. Harald also presented this work at Discovery on Target last week, where he noted that the effort is part of the COVID19-NMR project, a collaboration of 240 people in 18 countries.
The researchers investigated 15 RNA regulatory elements that are conserved between SARS-CoV-2 and SARS-CoV, ranging from 29-90 nucleotides, as well as 5 larger multielement RNAs (118-472 nucleotides). These were screened against the DSI-poised library (discussed here): 768 fragments designed for rapid follow-up chemistry.
Three different ligand-detected NMR methods were used for screening: chemical shift perturbation (CSP) or line-broadening, waterLOGSY, and T2-relaxation. Fragments were screened at 200 µM in pools of 12 against 10 µM RNA. Compounds that hit in at least two assays were investigated individually.
In total 40 fragments bound to one or more of the 15 shorter RNAs, and an additional 29 fragments bound to the five longer RNAs. Between 5 and 49 hits were found for all but two of the RNAs. Selectivity varied: some fragments bound to just one RNA while one fragment bound to 18 of 20.
Given the negatively-charged phosphate backbone of RNA, it is not surprising that many of the fragment hits are positively charged at physiological pH. Nearly one-third of the 40 hits against the shorter RNAs contain a basic amine; pyrimidine and benzimidazole moieties are enriched, and not one of the hits contain a carboxylic acid. All the hits have at least one aromatic ring and most have two or three, perhaps suggesting intercalation. Moreover, as seen in a previous ambitious RNA screen from the same group, hits tend to have fewer sp3-carbons than non-hits.
The highest affinity fragment had a dissociation constant of just 64 µM but an impressive ligand efficiency of 0.38 kcal/mol/atom. A search of commercial analogs yielded a compound with low micromolar affinity against two RNA targets. In his presentation Harald noted that this series has since been optimized to a 200 nM binder.
This paper is a tour de force, but as I have noted, there remains a dearth of high-affinity specific RNA binders. The researchers also note another potential problem: viral RNA accounts for roughly two-thirds of total RNA in cells infected with SARS-CoV-2. Would this necessitate high concentrations of drug for effective antiviral activity?
Whether or not the work leads to drugs, it should further basic research. Laudably, structures of all the hits and non-hits are provided in the paper, and the extensive supporting information provides more details. Hopefully we will soon see whether fragments poised for ready elaboration really will enable rapid progress against RNA.

04 October 2021

Nineteenth Annual Discovery on Target Meeting

Cambridge Healthtech Institute held its annual Discovery on Target meeting last week. For the first time the event was hybrid, with slightly fewer than half the attendees in Boston and the rest online, and I’m happy to report that it was quite successful. In-person attendees were required to show proof of vaccination against COVID-19, and masks and social distancing guidelines were observed. Ten of the individual tracks were hybrid, while four were virtual only. However, even in these cases it was valuable to attend in person; after one vendor presentation I immediately went from my hotel room to the exhibit hall to find out more.
For many of us this was the first in-person conference we had attended in nearly two years, and the return to some semblance of normalcy. At the same time, the fact that in-person talks were broadcast opened the conference to people unable to travel. One of the most active Q&A participants in one track was in Singapore, despite the 12 hour time difference.
Another nice feature of the virtual or hybrid model is reduction in FOMO; if you find it difficult to choose between the seven concurrent talks you can watch some later. But, as our 2020 poll showed, speakers may be less forthcoming with newer, more speculative results in a recorded format.
With the heavy focus on biology there seemed to be fewer “conventional” fragment stories, though Lars Neumann (Proteros) did discuss the identification and optimization of a kinase inhibitor that does not interact with the hinge region. Novel targets were represented in work from Harald Schwalbe (Johann Wolfgang Goethe University), who described fragment screens against RNA; I’ll post more on this later this month.
We’ve previously discussed the COVID Moonshot Consortium to rapidly discover drugs for SARS-CoV-2. Annette von Delft (Oxford University) provided an update, noting that fragments from a crystallographic screen have been advanced to compounds with mid-nanomolar biochemical and cellular activity. DMPK properties are reasonable, though this is an area of continued optimization. Annette mentioned the goal is to enter clinical development in 2023. Progress has been accelerated by the crowd-sourced nature of the initiative, with nearly 40 groups and 150 individuals working together. She also noted that many of the molecules are active against other coronaviruses.
The main series being advanced by the COVID Moonshot are noncovalent inhibitors of the SARS-CoV-2 main protease MPro. However, covalent molecules against this target are also moving forward. Matthew Reese described Pfizer’s oral PF-07321332, which is currently in several phase 3 trials. The program began on March 16 of last year and the clinical compound was first synthesized just four months later. Clinical trials began in February of this year, a mere 11 months after the program began. This is astonishingly rapid, though the researchers did benefit from previous work on SARS-CoV-1 and even earlier work from the 1990s on rhinovirus inhibitors. It is worth re-reading Glyn Williams’ 2020 discussion of HIV protease inhibitors for more historical context and insights.
Although PF-07321332 did not come from FBLD, fragments capable of forming covalent bonds were well represented. We’ve previously discussed fully-functionalized fragments (FFFs, or PhABits), which in addition to having a photoreactive group also contain an alkyne handle so that any target they bind can be captured and identified. Aarti Kawatkar and Jenna Bradley described using these at AstraZeneca to identify new targets. They’ve constructed a library of just under 500 FFFs and are using these to do phenotypic screening, particularly in hard-to-get cells such as primary tissue samples. They are also making the FFF library available through their open innovation initiative.
Fully functionalized fragments are just one flavor of covalent fragments. Indeed, unlike the light-activated warhead of FFFs, most covalent fragments have a moiety that reacts selectively with amino acid residues such as cysteines. Steve Gygi (Harvard) and Dan Nomura (UC Berkeley) both described covalent screening in cells to identify starting points against challenging targets. The approach is also gaining traction in industry; Heather Murrey described how Scorpion is using covalent fragments, and noted that Vividion (mentioned here) was recently acquired by Bayer for up to $2 billion.
A prominent recent success story from covalent fragments is sotorasib, which was approved earlier this year to treat certain non-small cell lung cancer patients whose tumors carry the G12C mutant form of KRAS. Sotorasib binds to a mostly cryptic pocket, and the protein itself has low ligandability. To improve the odds of finding new fragments, Mela Mulvihill described how she and her colleagues at Genentech have developed antibodies that stabilize the so-called Switch II loop in an “open” conformation more accessible to small molecules. An SPR-based fragment screen in the presence of the antibody led to more than twice as many hits, many of which could bind more tightly than without the antibody. Darryl McConnell (Boehringer-Ingelheim) also described using fragment-based methods to pursue KRAS, including mutants other than G12C.
In addition to inhibitors, Darryl also described bifunctional molecules that selectively cause degradation of KRAS by bringing it to the proteasome via E3 ligases. In his opinion PROTACs are “the best thing since sliced bread.” PROTACs and targeted protein degradation were in fact the subject of two tracks that spread across all three days of the conference, and were also covered in a pre-conference short course taught by Stewart Fisher (C4 Therapeutics) and Alexander Statsyuk (University of Houston). Here too fragments are playing an increasing role; in a second talk Dan Nomura described how he has been using chemoproteomic fragment approaches to identify ligands for E3 ligases.
The recent excitement around PROTACs is probably justified, but as our post last week noted, new technologies are not necessarily fast or inevitable. PROTACs were first described in 2001; Adam Gilbert (Pfizer) puckishly described them as a “20-year overnight success story.” But by the end of this year there will be roughly a dozen PROTACs in the clinic, with more likely to join them soon.
I’ll end on this positive note, but welcome your thoughts on science or experience with hybrid conferences. I look forward to seeing you at one in the near future!

26 September 2021

Success in drug discovery is not necessarily fast or inevitable

The biotech industry rightly prizes speed: every day people die of diseases we are trying to prevent or cure. And developments can indeed happen quickly. Just eight years elapsed from the demonstration that a mutant form of KRAS was druggable to the approval of sotorasib, with less than three of those years spent in the clinic. Even more dramatically, it took less than a year from the first reports of SARS-CoV-2 to develop effective vaccines. But as two recent pieces in Nature Rev. Drug Disc. demonstrate, such speed is not necessarily the norm.
The first, by Asher Mullard, is entitled “FDA approves 100th monoclonal antibody product.” This is a nice review of a remarkably successful therapeutic approach. But this triumph was not a foregone conclusion. Mullard traces the field’s origin to the mid-1970s, and while the first drug was approved in 1986, it took another eight years for the second. The article includes a timeline showing approvals by year, and it is interesting to compare this with FBDD-derived drug approvals since the 1996 publication of the seminal SAR by NMR paper. In the chart below, the first year on the x-axis is for antibody drugs; the second is for FBDD-derived drugs.
A quarter century after work began, new antibody approvals were still uncommon; Mullard notes that “antibody approvals have only been an annual event since 2006.”
Antibody-drug conjugates (ADCs) are an interesting subset that – as their name suggests – comprise an antibody linked to a small molecule, usually a toxin intended to kill cancer cells. Ten of these have been approved in the US, but while the first (gemtuzumab ozogamicin) was approved in 2000 most of the rest are recent, with six of them coming since the beginning of 2019.
By these standards the fact that only five fragment-derived drugs have been approved thus far isn’t surprising. Indeed, antibodies have some advantages: “whereas medicinal chemists can toil for years to find small molecules with activity against a given target, antibody discovery can take a matter of months.” Moreover, as the article continues, success in the clinic is roughly double that of small molecules.
The second article is by Christopher Austin, until recently Director of the National Center for Advancing Translational Sciences at the US National Institutes of Health. Titled “Translational Misconceptions,” it briefly enumerates and debunks false beliefs about translating new discoveries into drugs, which include:
- Translation does not exist 
- Translation is a “thermodynamically favored” process 
- Translation is straightforward and does not qualify as science 
- Translation is a unidirectional process 
- Once an investigational therapy gets into humans for the first time, regulatory approval
  and marketing are all but assured 
- Regulatory approval is the end of the translational process
Those of us in industry would probably dismiss these statements as naïve, but such perceptions are widespread. Indeed, Austin himself acknowledges that he “once believed unquestioningly in all of them.”
Each of these misconceptions invites discussion. To take just the last, the first approved ADC was pulled from the US market in 2010 when confirmatory trials showed that patients on the drug actually did worse than those on placebo. It was reapproved in 2017 after a better dosing schedule was established. In other words, it took 17 years after initial approval to figure out how to effectively use gemtuzumab ozogamicin, and 26 years from the beginning of the project.
Returning to the two successes mentioned at the top of this post reveals that their apparent rapidity does not tell the full story. The Tethering technology that eventually led to sotorasib was initially published more than twenty years ago, and researchers first used mRNA packaged in liposomes to transfect cells way back in 1989.
Amidst rapid visible progress it is easy to lose sight of the fact that much research goes nowhere very slowly. Even when successful, it might take decades to help patients. As Austin concludes, “only by advancing our common understanding of the complexity of translation, translational research and translational science will translational gaps be narrowed and eventually eliminated.”

20 September 2021

Chemical couplets inhibit the GAS41 YEATS domain

Practical Fragments has covered bromodomains extensively, most recently just a couple months ago. But these epigenetic readers are not the only proteins that recognize acetylated lysine residues. Two years ago we highlighted fragment hits against one of the four YEATS domain proteins. A paper recently published in Cell Chem. Biol. by Jolanta Grembecka, Tomasz Cierpicki, and colleagues at University of Michigan tackles another member of the family.
The researchers were interested in the protein GAS41, which is amplified in multiple forms of cancer. The YEATS domain within this protein binds to acylated lysine residues in histone H3 proteins. However, unlike the deep pockets found in bromodomains, the acyl-lysine binding site in the YEATS domain consists of a partially solvent-exposed channel, making it a more challenging site to drug.
Nonetheless, an NMR-based screen of fragments (in pools of 10, each at 500 µM) led to compound 1, which produced multiple chemical shifts in a 1H-15N HSQC experiment. Two different competition assays, fluorescence polarization (FP) and AlphaScreen formats, confirmed that compound 1 could compete with H3-derived peptides. Fragment growing led to compounds 7 and 11. (All IC50 values below are from the FP assay; these are somewhat weaker than those from the AlphaScreen, but they more closely track binding affinities determined by isothermal titration calorimetry.)

A crystal structure of a compound closely related to compound 11 revealed that the molecule nearly fills the small binding channel, suggesting that further gains in affinity would be difficult. Indeed, no GAS41 inhibitors have been previously reported. However, the protein is dimeric, so the researchers decided to dimerize their molecule to bind to two YEATS domains simultaneously. This led to nanomolar molecules such as compound 19.
Not only was compound 19 potent in biochemical assays, it also disrupted binding of GAS41 to acetylated histone proteins in cells. Moreover, the compound inhibited growth of cancer cell lines with amplified GAS41.
This is a nice case study in fragment dimerization, an uncommon but interesting approach. The linking in this case led to a 44-fold improvement in affinity, which though impressive is far from synergistic, and is associated with a considerable loss in ligand efficiency. And although the micromolar potency of compound 19 in cells needs to be improved to generate a chemical probe, let alone a drug lead, these results nonetheless support the notion that targeting GAS41 could be a useful strategy for certain cancers.

13 September 2021

Fragments in the clinic: TAK-020

The fifth fragment-based drug approved, sotorasib, acts by irreversibly binding to its protein target. However, despite increasing interest in covalent inhibitors, they are still rare among fragment-derived clinical candidates. A paper just published in J. Med. Chem. by Mark Sabat and colleagues from Takeda describes a new addition.
The researchers were interested in inhibitors of BTK, a kinase implicated in cancer as well as arthritis. Indeed, the drug ibrutinib, a covalent BTK inhibitor, has been widely credited with driving recent interest in covalent molecules. Back in 2015 we highlighted work from Takeda describing a fragment screen and optimization to a low nanomolar non-covalent lead. In addition to that fragment, the screen produced triazolone-containing compounds 5 and 6. Crystallography revealed that the triazolone moiety interacts with the kinase hinge region through multiple hydrogen bonds. Importantly, triazolones had not previously been observed as hinge binders. Thus the researchers decided to pursue the chemical space opened up by this discovery.

Although compound 6 had higher lipophilicity and lower ligand efficiency than compound 5, it provided better vectors for growing into the binding pocket. Changing the phenyl ring of compound 6 to a pyridyl gave a very slight boost in potency; growing led to low micromolar compound 11. Expanding the pyridyl to an isoquinoline further improved the potency to mid-nanomolar compound 18, which is still fragment-sized and highly ligand efficient. Crystallography suggested how to append an acrylamide warhead to engage cysteine 481 (the target of ibrutinib). This produced TAK-020, which overlays nicely onto compound 6. 
The whole campaign was remarkably fast: just 6 months of synthetic effort from fragment to TAK-020.
In general, the longer an irreversible inhibitor is incubated with a protein, the more potent it will appear, so IC50 values can vary depending on the assay. A more meaningful number is kinact/Ki, which turned out to be 205,000,000 M-1s-1 for TAK-020, more than 10-fold higher than ibrutinib. The compound demonstrated efficacy in a rat model of collagen-induced arthritis and behaved well in various safety assessments including two-week rat and dog toxicology studies.
Based on these results, the compound entered the clinic; the results of a phase 1 trial are reported by Eric Fedyk and colleagues in Clin. Transl. Sci. The compound proved safe and well-tolerated. Although TAK-020 has a relatively short half-life, it led to sustained reduction of BTK activity. (Since irreversible inhibitors by definition remain bound to their targets, their efficacy can be high even if rapidly cleared from the bloodstream)
And there the story ends. ClinicalTrials.gov reports that the two phase 1 clinical trials ended more than four years ago, and TAK-020 does not appear on Takeda’s pipeline. Four BTK inhibitors are already approved, with multiple others at various stages of clinical development, so perhaps the molecule was deprioritized for competitive reasons. The fact that TAK-020 first entered the clinic in March of 2015 is a reminder that what you read in the literature often reflects work completed many years ago. Nonetheless, this is a lovely fragment-to-lead story that is well worth perusing.

06 September 2021

How fragments become leads

Our most recent poll asked how often synthetic challenges had kept researchers from pursuing a particular fragment or had impeded a fragment-to-lead project. Around two-thirds replied sometimes or often. A new open-access paper in Chem. Sci. by Rachel Grainger, Rhian Holvey, and colleagues at Astex does a deep dive into fragment-to-lead chemistry, provides a powerful visual tool, and ends with something of a call to action.
The researchers take as their starting points 131 fragment-to-lead success stories published from 2015 to 2019 and collated in a series of five J. Med. Chem. Perspectives. All of these started with fragments (< 300 Da) for which affinity increased by at least 100-fold, with the resulting leads having affinities of 2 µM or better. As the new paper points out, this could introduce “survivorship bias,” in that less successful projects are not included. However, as the point of the paper is to figure out what works, this likely strengthens the conclusions.
The targets themselves are fairly diverse: 24% kinases, 9% proteases, 36% other enzymes, 11% bromodomains, 14% other protein-protein interactions, and 6% other types of targets. The researchers closely examined how the leads related to the initial fragments. Full details are provided in the Supplementary Material (pdf). The researchers have also constructed a handy interactive viewer you can use to do your own analyses. Here is an overlay of an initial fragment (taupe space-filling) with the final lead (yellow surface).

What are the results? The first observation is that 93% of leads have at least one polar interaction (such as a hydrogen bond) that is conserved from the initial fragment. The most common functional groups making direct contacts to proteins are N-H hydrogen-bond donors (35%) followed by aromatic nitrogen hydrogen-bond acceptors (23%) and carbonyl oxygen hydrogen-bond acceptors (22%).
The second observation is that over 80% of fragments are grown from one or two vectors (examples of one and two, with the second the subject of the figure above). This is perhaps not surprising; growing from three or more vectors would likely result in portly molecules that may be more difficult to advance, venetoclax notwithstanding.
But the really interesting observation is that the majority of growth vectors (~80%) originate from carbon atoms. Moreover, more than half of the bonds formed are carbon-carbon bonds. For the non-chemists in the audience, this is significant because carbon-carbon bond forming reactions are not always straightforward, particularly in the presence of polar moieties.
In the early days of FBLD, one hope was that including functional groups such as amides in a fragment collection would facilitate fragment growing. The new paper suggests that this is naïve: a functional group in a fragment is likely to interact with the protein and so block potential growth vectors. Indeed, only 18% of growth vectors come from N-H groups, despite the fact that these are among the most synthetically accessible.
These findings thus explain why fragment-to-lead efforts can be so challenging. The researchers provide an example of a chemical series they ultimately abandoned due to poor synthetic tractability.
The paper also builds on earlier papers from Astex exhorting chemists to further advance chemical methodology. As they conclude:
An “ideal synthesis” of a lead would allow: (1) site-selective formation of bonds at all growing points of a fragment, (2) whilst being mild enough to be compatible with essential polar functionality, and (3) proceeding with minimal or no need for protecting groups….
We believe that further development of C-H functionalisation that is tolerant to polar fragments has the potential to transform FBDD.
If you’re in academia, this looks like a good opening for a grant proposal!

30 August 2021

Covalent fragments meet high-throughput synthesis and crystallography

Two hot areas of FBLD include covalent fragments (see here for example) and high-throughput crystallography (see here and here). Two new papers add to the excitement but also reveal some of the challenges.
The first, published open access in J. Med. Chem. by Charles Eyermann (University of Cape Town), Christopher Schofield (Ineos Oxford Institute of Antimicrobial Research), Christopher Dowson (University of Warwick) and collaborators at Diamond Light Source focuses on an antibacterial target, the penicillin binding protein PaPBP3 from Pseudomonas aeruginosa.
Some members of the group had previously screened a library of 40 compounds against this protein and identified a single hit that bound covalently. Inspired by this result, they assembled a library of 262 commercial covalent fragments, 152 of which contained boron, an element known to react with serine and threonine hydrolases. These were screened crystallographically at 250 mM, resulting in 34 boron-containing hits “with various levels of electron density observed at the active site.” Many of these seem to have given ambiguous density, and only a handful of fragment structures are reported.
Next, the researchers attempted fragment growing; the team ultimately obtained 10 structures, drawing from the original fragments and the elaborated molecules. Interestingly, these showed three different binding modes: a monocovalent complex with the active-site serine, a dicovalent complex with the active-site serine and a neighboring serine, and a tricovalent complex with these two serine residues and a nearby lysine side chain.
Despite having up to three covalent bonds to the enzyme, even the elaborated molecules showed at best double-digit micromolar inhibition of PaPBP3, and none showed antimicrobial activity. The bond between boron and serine or lysine is reversible, so while disappointing, this weak activity is not entirely surprising. The result is reminiscent of efforts against the SARS-CoV-2 main protease, Mpro or 3CLpro, where many of the crystallographic hits turned out to be very weak binders.
The second paper, published open access in Angew. Chem. Int. Ed. by Alexander Dömling and collaborators at University of Groningen and the Paul Scherrer Institute, also looked at 3CLpro. Here though, rather than starting with commercial fragment libraries the researchers used high-throughput synthesis to make their own.
The Dömling lab has had a long-standing interest in multi-component reactions in which several reagents are combined to generate products. The researchers explored the Passerini reaction (which uses an aldehyde or ketone, a carboxylic acid, and an isocyanide) and the Ugi reaction (which uses all of these as well as an amine). For the carboxylic acid, they chose an electrophile such as acrylic acid. They initially worked out the conditions at 0.5 mmol scale in 96-well plates and then purified the products using chromatography or precipitation. The majority of wells yielded product in good yields and purity.
Next, the researchers turned to high-throughput methods, performing the reactions in 384-well plates using acoustic liquid handling. This is similar to the approach I wrote about here that was used to discover covalent KRAS inhibitors, ultimately leading to the approved drug sotorasib.
The researchers then soaked 181 of their compounds against the SARS-CoV-2 protein 3CLpro, with each compound at 10 mM. Unlike the crude reaction screening described here, the researchers used pure compounds. This effort resulted in five hits, though one of them turned out to bind noncovalently. Three of the molecules had low micromolar activity.
In the end, while both of the papers do report the discovery of covalent modifiers, the functional activities in the first paper are modest, and the active warheads in the second paper (chloroacetamides and acrylates) are likely too reactive to be advanceable. A nice feature of crystallography is that it can provide structural information for extremely weak hits. But as we’ve asked previously, how weak is too weak? Getting more crystal structures more rapidly is one thing, but figuring out which ones are useful requires even more skill, creativity, and luck.

23 August 2021

Fragments vs DYRK1A and DYRK1B: Part 2

Last week we highlighted work out of Vernalis and Servier in which fragment-based methods were used to identify potent and selective inhibitors of DYRK1A and 1B, potential targets for cancer and neurodegenerative diseases. The NMR screens yielded 166 hits, only one of which was advanced in that paper. A second J. Med. Chem. paper by Andras Kotschy and collaborators describes the optimization of another fragment.
Compound 1 is a whoppingly potent fragment with impressive ligand efficiency. If you’ve ever worked on kinases you probably think you know how it binds, as the diaminopyrimidine moiety is a common hinge-binding motif. In fact, crystallography revealed that the molecule binds in a completely different orientation and that the methoxy group makes a single hydrogen bond to the hinge amide NH. Cyclizing the molecule led to compound 10, with a satisfying boost in affinity.

Unfortunately, compound 10 was also a potent inhibitor of the kinase CKD9. To gain selectivity, the researchers took advantage of the fact that one of the backbone carbonyl oxygens in the hinge adopts an unusual orientation in DYRK1A, making room for the methyl group in compound 33. Next, the researchers replaced the benzofuran core for reasons of “synthetic tractability, metabolic stability, and freedom to operate.” This exercise ultimately led to compound 40.
This compound was profiled against 442 kinases and found to be quite selective, with only 8 kinases significantly inhibited at 1 µM. One of these was the related kinase DYRK2, but further growing led to selective compound 58. An overlay of the initial fragment (blue) with compound 58 (gray) reveals how the binding mode has been maintained, in contrast to the series described last week.

Compound 40 had only modest antiproliferative activity against human cancer cell lines that were grown in 2D culture but was more active when the cells were grown in 3D culture. The molecule had good oral bioavailability in mice, and xenograft studies revealed that it inhibited tumor growth, though it was also toxic at higher doses. The researchers do not mention brain penetration, though given the number of hydrogen bond donors I would be surprised if it crosses the blood-brain barrier.
This paper is a nice example of how getting high affinity is often only the beginning of a long journey. In combination with the story from last week it is also a useful reminder of how many starting points a single fragment screen can provide: just two fragments led to two completely independent series. Whether molecules from these series advance to the clinic, they provide useful tools to further understand the biology of DYRK1A.

16 August 2021

Fragments vs DYRK1A and DYRK1B: Part 1

The dual-specificity tyrosine-phosphorylation-regulated kinases 1A and 1B (DYRK1A and DYRK1B) belong to a family of five serine/threonine kinases implicated in several cancers as well as Down’s syndrome and other neurodegenerative diseases. For the latter indications in particular, brain penetration would be essential for any inhibitor, just as in the LRRK2 story last week. In a new (open access) J. Med. Chem. paper, Rod Hubbard and collaborators at Vernalis and Servier describe the discovery of a chemical probe.
The researchers started by testing their in-house library of 1063 fragments in pools of six, each at 500 µM, in three ligand-detected NMR screens. This resulted in a whopping 166 hits. Crystal structures of the eight most ligand-efficient fragments bound to DYRK1A were obtained, including compound 5. Fragment growing led to compound 16, which bound the kinase 200-fold more tightly. 

The crystal structure of compound 16 bound to DYRK1A was compared to structures of other known ligands and suggested the possibility for an alternative binding mode. This led to the synthesis of compound 24, with low nanomolar affinity against both DYRK1A and DYRK1B (only the former is shown in the figure). This compound turned out to be surprisingly unstable in slightly acidic aqueous solution (below pH 5), but replacing the oxygen with a nitrogen fixed this, and further tweaking ultimately led to compound 34.
Compound 34 was profiled at 1 µM against a panel of 442 kinases and found to be fairly selective, with only 15 kinases inhibited by at least 50%. It is orally bioavailable in mice, brain penetrant, and inhibited the proliferation of glioblastoma cells, although the potency was significantly attenuated by serum. In a xenograft study the compound caused tumor growth delays and was well-tolerated.
This is a nice example of fragment-based lead discovery heavily dependent on structural information. Comparing the binding mode of compound 34 (gray) with that of compound 5 (light blue) reveals the significant shift in binding mode of the initial fragment.

The paper is also a useful reminder of how long it can take for industry research to be published. Work began in 2009, and Rod presented some of it at the CHI FBDD conference in 2019. But this is not the end of the DYRK1A story: stay tuned for next week!

09 August 2021

Fragments vs LRRK2 with the help of a surrogate and a magic methyl

Leucine-rich repeat kinase 2 (LRRK2) has been implicated in Parkinson’s disease and has thus long been targeted by drug hunters. Dozens of inhibitors have been approved for other kinases, making the protein class appear “easy”, but LRRK2 is particularly challenging. First, it is a large multidomain protein that has resisted crystallography. Second, an inhibitor for a chronic disease such as Parkinson’s will need to be highly selective. Finally, the fact that LRRK2 is in the brain means that inhibitors will need to cross the treacherous blood-brain barrier (BBB), whose function is to exclude anything unusual. A recent J. Med. Chem. paper by Douglas Williamson and collaborators at Vernalis and Lundbeck addresses the first two of these issues.
The researchers started by screening 1313 fragments (at 200 µM each) against the disease-relevant G2019S mutant of LRRK2. The screen was run using the DiscoveRx KINOMEscan, which relies on displacement of a kinase from an immobilized ligand. Some 80 hits were then triaged in a kinase activity assay.
Rather than banging their heads against the wall that has blocked X-ray structures of LRRK2, the researchers turned to a crystallographic surrogate. The readily crystallizable kinase CHK1 has some similarity to LRRK2, and some inhibitors bind to both kinases. Introducing ten mutations into CHK1 around the ATP-binding site led to a LRRK2 surrogate – an approach we’ve previously mentioned.
Among the fragment hits were adenine (compound 7) and two closely related molecules. Crystallization of compound 7 with wild-type CHK1 and the LRRK2 surrogate revealed two different binding modes. Similarity-based searches of in-house and literature compounds led to molecules with nanomolar potency, and subsequent optimization led to compound 17 (all molecules were tested against both wild-type LRRK2 as well as the G2019S and had similar affinities).

Interestingly, crystal structures of these molecules revealed them to bind more similarly to the complex of compound 7 bound to wild-type CHK1 rather than the surrogate. Adding a methyl group to compound 17 led to compound 18 with a satisfying 100-fold boost in potency: a true magic methyl, as the other enantiomer had slightly worse affinity than compound 17. Crystallography with the surrogate revealed hydrophobic interactions between the methyl and an alanine side chain. It is worth noting that compound 18 is still fragment-sized and yet has high picomolar affinity for LRRK2.
Extensive medicinal chemistry followed, ultimately leading to compound 45. Profiling against 468 kinases in the KINOMEscan assay demonstrated it was quite selective, binding only three other targets with dissociation constants less than 1 µM. The compound was active in cells containing either wild-type or G2019S LRRK2. Unfortunately, while the compound showed good oral bioavailability in dogs, it was not orally bioavailable in rats. Moreover, brain to plasma ratios were low in mice, and the molecule was a substrate for the human protein BCRP, a transporter that pumps small molecules across the BBB.
Although these liabilities halted further work on the series, this is nonetheless a nice fragment to lead story that highlights the utility of crystallographic surrogates. But the different binding modes for the initial fragment are a reminder that multiple binding modes are not uncommon, and it is best to employ crystallography not just early but often, when you can.