Keeping molecular dynamics cool for fragments

Accurately and reliably predicting fragment binding modes would be preferrable to doing messy, expensive, and sometimes tedious experimental work, but we’re not there yet. One of the biggest problems is that, because fragments usually bind weakly to proteins, it is hard to tell which of several possible binding modes is most favorable. In an open-access J. Chem. Inf. Model. paper published earlier this year, Stefano Moro and colleagues at University of Padova report progress.

 

Their approach, called Thermal Titration Molecular Dynamics (TTMD), analyzes short molecular dynamics simulations across increasing temperatures; if the ligand remains bound to the protein, this indicates a more stable binding mode. (It seems a bit like the dynamic undocking we wrote about here.) The researchers had previously reported good results for larger, drug-sized molecules, but not for four fragment-protein complexes.

 

Recognizing the low affinities of fragments, the researchers decided to lower the (virtual) temperatures. Rather than heating from 300 to 450 K, they heated from 73 to 233 K; ie, from just below the boiling point of liquid nitrogen to a moderately cold winter’s day in Minnesota. They first docked fragments using PLANTS-ChemPLP, which is free for academics, and chose the five best-scoring poses for evaluation.

 

Next, the researchers performed TTMD. There are several different ways to assess how well the ligand remains bound to the protein over the course of a molecular dynamics simulation, and four different scoring methods were chosen. When TTMD was tested on the four fragment-protein complexes that had previously failed, at least two of the scoring methods correctly identified the crystallographic binding mode for three of the fragments.

 

Thus encouraged, the researchers tested ten more compounds bound to six new proteins. The results were quite encouraging, with up to 86% of crystallographic binding modes being correctly identified by at least one of the scoring functions in TTMD vs 50% for docking alone. Impressively, two of the examples were MiniFrag-sized, with just 6 or 7 non-hydrogen atoms, yet the crystallographic pose was identified as the lowest energy in all four TTMD scores.

 

This is nice work, but the question arises how these specific ligands and proteins were chosen. Several years ago we highlighted a curated set of 93 protein-ligand structures that were used to benchmark other virtual approaches, and it would be nice to see how TTMD performs on these. Still, TTMD’s performance on its chosen examples is encouraging, and laudably the researchers have made their code freely available. If you try it out, please let us know how it works in your hands.

Which cryptic sites are ligandable, and why?

Many interesting proteins have flat, featureless surfaces, lacking the deep pockets in which small molecules usually bind. But structures can be deceptive: crevasses can open unexpectedly, revealing “cryptic sites” for ligands. Or not – just because a site is available does not mean it is ligandable (able to bind to ligands with high affinity). A new (open access) paper in Drug Disc. Today by Sandor Vajda and collaborators at Boston University and Stony Brook University asks “which cryptic sites are feasible for drug targets?” (Sandor presented some of this at FBLD 2024 last month.)

 

To get started, the researchers turned to the aptly named CryptoSite, a previously published list of 93 proteins where unexpected pockets had been found. Each protein has at least two published crystal structures, one in the apo form and one with a ligand bound to the (no longer) cryptic pocket. Cryptic sites form primarily through two mechanisms. In the first, amino acid side chains move aside, opening a pocket. In the second, larger motions occur in protein loops or secondary structural elements, such as alpha helices, creating pockets.

 

Of the 18 cases for which cryptic sites formed primarily through the movement of side chains, ten had published affinities for the ligands, and all of these were weak, with the best being low micromolar. In contrast, of the 27 cryptic sites created by loop movements for which affinity information was available, all but two were nanomolar binders. From this evidence, the researchers suggest that cryptic sites formed only by the motion of side chains are not sufficient to support high affinity ligands. Why?

 

The researchers note that side chain motions occur very rapidly, on a timescale of 10^-11 to 10^-10 seconds, much faster than ligand binding, which at its fastest is 10^-8 seconds. Thus, “a fast-moving side chain that spends a substantial fraction of time in the pocket interacting with the other residues competes with ligands for binding and, hence, acts as a competitive inhibitor.” This intuitive picture is supported in the paper by mathematical simulations.

 

In contrast, loop movements occur on 10^-9 to 10^-6 second timescales, while the movements of secondary structure elements are even slower. Thus, a ligand could bind while the cryptic site is open, and, like a wrench in a machine, keep it open.

 

This finding is important. As the researchers point out, the molecular dynamics calculations frequently used to find cryptic pockets are typically run at short timescales likely to miss loop movements. Other computational methods used to assess ligandability may also suffer; the researchers note that their program FTMap, which we’ve written about here and here, overestimates the ligandability of cryptic sites created by side chain movements.

 

Of course, just because a cryptic site is created by loop movements does not mean it is ligandable, as we discussed for interleukin-1β. And the researchers acknowledge that covalent inhibitors might be able to take advantage of less traditionally ligandable sites, cryptic or otherwise. Certainly this has been the case for KRAS. I’m confident that many more examples will be forthcoming.

Fragments in the clinic: Lirafugratinib

With crystal structures of protein-ligand interactions becoming increasingly accessible, it is easy to forget that proteins do not exist as the static structures seen on page or screen. Indeed, back in 2018 we quoted Karplus quoting Feynman that “everything that living things do can be understood in terms of the jiggling and wiggling of atoms,” and even the smallest proteins have lots of atoms. In an open-access paper published in Proc. Nat. Acad. Sci. USA earlier this year, Heike Schönherr, David Shaw, and collaborators at Relay Therapeutics, D.E Shaw Research, Pharmaron, and Columbia University take advantage of these movements.

 

The researchers were interested in finding selective inhibitors of fibroblast growth factor receptor 2 (FGFR2), which is activated in many cancers. The four members of the FGFR family are so closely related that finding selective inhibitors is difficult. Inhibiting FGFR1 can lead to hyperphosphatemia, while inhibiting FGFR4 can cause diarrhea, side effects seen with the approved fragment-derived drug erdafitinib.

 

Although the structures of FGFR1 and FGFR2 are very similar, extended (25 µs) molecular dynamics simulations revealed that the so-called P-loop of the proteins behaved differently: in FGFR1 it became disordered, while in FGFR2 it remained more rigid. The researchers sought to take advantage of these differences with a covalent inhibitor.

 

The researchers started with a non-selective hinge-binding fragment, compound 1. Adding an acrylamide warhead led to a nanomolar inhibitor with modest selectivity for FGFR2. (All IC₅₀ values are measured after 30 minute incubations.) Growing the molecule into the so-called back pocket of the kinase led to compound 5, with nearly 100-fold selectivity for FGFR2 over FGFR1. 

 

 

The path from compound 5 to lirafugratinib (also called RLY-4008) looks straightforward but was anything but. First, the aryl acrylamide was a metabolic liability, so the researchers attenuated the reactivity by adding a methyl group. Mechanistic studies with this molecule revealed that while it had only a slightly better affinity (K_I) for FGFR2 than FGFR1, it had a k_inact value about 15-fold higher for FGFR2. Molecular dynamics studies suggested that the relevant cysteine in FGFR1 is locked in a position too far from the acrylamide to react, while the corresponding cysteine in FGFR2 may be able to more closely approach the acrylamide warhead.

 

Further optimization, guided by extended molecular dynamics simulations, led eventually to lirafugratinib with ~250-fold selectivity for FGFR2 over FGFR1 and >5000-fold selectivity over FGFR4. Remarkably, the noncovalent version of lirafugratinib, compound 11, shows dramatically lower affinity for both FGFR1 and FGFR2 and very little selectivity between them. The ligand seems to assume a different binding mode after covalent bond formation, which could explain these differences in selectivity.

 

Mouse studies of lirafugratinib showed tumor stasis or regression without increased serum phosphate levels. More importantly, early clinical data has shown “minimal hyperphosphatemia and diarrhea.”

 

This is a lovely example of structure and dynamics-based design (SDBD?). Commonly cited advantages of covalent drugs include improved potency and extended pharmacological effects, but this work shows that they can also achieve remarkable selectivity between closely related proteins, even when both proteins contain cysteine residues in the same location. Moreover, an open-access paper in Cancer Discov. that dives more deeply into the biology shows that lirafugratinib is selective across the kinome, inhibiting just two of 468 kinases other than FGFR2 by >75% at 500 nM.

 

The next time you’re trying to find a selective inhibitor for one member of a protein family, it may be worth taking a covalent approach, and paying close attention to dynamics along the way.

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 A_2A adenosine receptors (A₁AR and A_2AAR), targets for a variety of ailments from cancer to cardiovascular diseases. (A_2AAR 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 A_2AAR. 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 A_2AAR. 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 A_2AAR 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 A_2AAR-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 A_2AAR 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.

Antifreeze opens cryptic pockets, experimentally and computationally

Imagine an alien species sees a single photo of a human. They would have no idea how our arms and legs move, or that our mouths can open and close. So it is with protein crystal structures: even multiple static images often fail to show possible conformations. Pockets open and close in unexpected places, and these can be critical for drug discovery. But how do you find these “cryptic” pockets? Harsh Bansia, Suryanarayanarao Ramakumar, and collaborators at Indian Institute of Science, Bengaluru and Pennsylvania State University provide a new approach in J. Chem. Inf. Mod.

 

The researchers were studying a bacterial xylanase called RBSX and had mutated a tryptophan residue to an alanine. When they solved the crystal structure, they found that the mutation had created a surface pocket that was filled with a molecule of 1,2-ethanediol (EDO). EDO is an ingredient in antifreeze because of its ability to prevent ice formation, and this property also makes it a common cryoprotectant in crystallography. The EDO molecule was making both van der Waals contacts as well as a hydrogen bond with the protein. The researchers found similar results when they used propylene glycol. (See here for a related discussion of MiniFrags, the smallest of which are the size of propylene glycol.)

 

To see whether water could make these same interactions, the researchers determined another crystal structure in the absence of EDO. Surprisingly, a phenylalanine side chain rotated and closed the pocket. Had this been the only structure solved, the possibility of pocket formation would not have been suspected.

 

Next, the researchers conducted molecular dynamics simulations. Starting from the closed state, the pocket remained occluded by the phenylalanine, giving no hint of its potential presence. Starting from the open state and removing EDO, the pocket also rapidly closed. In other words, in the absence of a ligand, the pocket appears to collapse in upon itself.

 

Importantly, these observations are not limited to a mutant bacterial protein. The researchers looked at published crystal structures of four unrelated proteins with known cryptic pockets and found that EDO could bind in all of them. They also ran molecular dynamic simulations on two proteins in which EDO was included as a virtual cosolvent. For both NPC-2 and IL-2, addition of EDO was able to open up cryptic pockets that had been previously found using other molecules; we’ve discussed earlier computational work on IL-2 here.

 

This is a nice example of following up on an unexpected observation, and is well-suited for further study. For example, it would be interesting to do a systematic study of EDO and propylene glycol binding sites throughout the entire protein data bank. For those of you doing molecular dynamics or crystallography, it may be worth adding EDO – virtually or experimentally – to see if it reveals any surprises in your favorite proteins.

Merging two of the same fragments for FABP4

The fatty acid binding proteins (FABPs) are a family of 10 proteins that – as their name suggests – shuttle fatty acids around cells. FABP4 has been implicated in a host of diseases, from atherosclerosis to nonalcoholic steatohepatitis. A recent paper in J. Med. Chem. by Yechun Xu and collaborators mostly at Shanghai Institute of Materia Medica describes how a fragment led to a compound with in vivo efficacy. It is a lesson in both recognizing and capitalizing on the fact that fragments often have multiple binding modes.

The researchers screened just 500 fragments, each at 1 mM, looking for displacement of a fluorescent ligand. Two hits were identified, of which compound 1 was by far the most potent. The researchers characterized the binding mode using crystallography, which itself was challenging because the protein co-purified with bound fatty acids. They had to denature the protein, strip fatty acids, and then refold it to obtain the apo form. When they were finally able to determine the crystal structure, they were surprised to find that compound 1 adopted three different binding modes under two different conditions (pH 6.5 and 7.5). These experimental results were supported by molecular dynamics calculations.

It is not uncommon for fragments to assume different binding modes. Indeed, the 7-azaindole fragment that led to vemurafenib, pexidartinib, and other clinical compounds has been found to bind in multiple orientations. In this case, the researchers recognized that the three binding modes put the two phenyl rings in three positions, suggesting that grafting a third phenyl ring onto compound 1 could improve affinity. This proved successful, and the resulting compound 3 had an affinity more than two orders of magnitude better as assessed both in the displacement assay and by isothermal titration calorimetry. Crystallography revealed that the molecule bound as expected.

Further structure-based design ultimately led to compound 17, with low nanomolar affinity. This molecule is also active in a cellular assay and has surprisingly good pharmacokinetic properties in mice. Given these encouraging results, the researchers tested whether the molecule could protect mice from multiorgan damage promoted by inflammatory lipopolysaccharides. The results were positive.

Unfortunately, compound 17 does show low micromolar activity against FABP3, whose inhibition would likely cause cardiac toxicity. Still, this is a nice example of fragment “self-merging”. Although merging two different fragments is common, merging a fragment onto itself is relatively rare, and – as shown here – not necessarily easy. It is an approach worth keeping in mind the next time you encounter a fragment with multiple binding orientations.

Looks can be deceiving: Getting misled by crystal structures – part 3

It’s been a while since we’ve touched on some of the hazards of interpreting crystal structures (see here, here, and here). In a recent issue of J. Comput. Aided Mol. Des., Alpeshkumar Malde and Alan Mark of the University of Queensland, Australia describe some mishaps taken from the literature, and how molecular modeling could have avoided them.

The authors start by noting that although protein structure determination using crystallography has been highly optimized, small molecule ligands are a different matter. Part of the problem is that small molecules may show more disorder than protein side chains, thus making it more challenging to fit the model into the observed electron density. Moreover, the parameters for refining protein structures do not always transfer to small molecules: electrostatic interactions are frequently ignored, as are alternative conformations.

As an example, the authors revisit the structure of noradrenochrome bound to an enzyme that synthesizes adrenaline. A racemic mixture of the ligand was used during crystallization, and when the crystal was solved at modest resolution it was possible to fit the ligand within the electron density in eight different orientations – four for each enantiomer. Despite this ambiguity, only a single structure was deposited in the protein data bank (pdb). Malde and Mark ran molecular dynamics (MD) simulations and free energy calculations and found that this structure is likely incorrect: it is higher energy than other structures and binds in a different orientation than the natural ligand, whose structure had previously been solved. In fact, the conformation suggested by MD is the opposite enantiomer from that deposited in the pdb and rotated 180 degrees.

In another example, the authors examine a high-resolution structure of a pyrazole-containing compound bound to the kinase CDK2. Pyrazoles can adopt two different tautomers in which the hydrogen is on either of two adjacent nitrogens, and in this particular case the original paper suggested that both tautomers were present in equal amounts, and both were deposited in the pdb. However, computations suggested that one tautomer is 7 kJ/mol higher energy than the other, and Malde and Mark suggest that in fact probably just a single tautomer is present in the structure.

Finally, the authors describe cases where a primary amide or primary sulfonamide group is in the wrong orientation. In most cases it is difficult to distinguish between a nitrogen and oxygen atom on the basis of electron density alone, and given that there are about 1000 ligands containing a -CONH2 group and about 200 containing a -SO2NH2 there are probably many mistakes.

The authors acknowledge that the examples they present are relatively simple, and one could argue that some of them would have been caught if they were critical structures in a lead optimization program. Nonetheless, the fact that they weren’t suggests that one must always be on guard, particularly in virtual screening where dozens or hundreds of structures are used in an automated fashion to develop or validate docking algorithms. Malde and Mark also note that, in the case of fragment screening with very small low-affinity ligands, one needs be especially cautious.

There is something extremely attractive about a crystal structure: it looks so real that it is easy to lose sight of the fact that it is just a model. Checking one’s assumptions with a bit of computation can prevent costly mistakes.