How do cryptic pockets form?

Earlier this year we highlighted crystallographic work out of Astex showing that secondary ligand binding sites on proteins are common; in addition to an active site, an enzyme may have several other pockets capable of binding small molecules. Many of these secondary sites are present even in the absence of a ligand. But there are also “cryptic” binding pockets that only appear when a ligand is bound. These are the subject of a new paper in J. Am. Chem. Soc. by Francesco Gervasio and collaborators at University College London and UCB Pharma.

Cryptic pockets are appealing in part because they can salvage an otherwise unligandable target: a featureless flat surface involved in a protein-protein interaction may crack open to reveal a crevasse capable of binding small molecules. Finding these pockets computationally, though, is difficult. In the current paper, the researchers performed molecular dynamics simulations on three different proteins with known cryptic pockets, and the pockets remained mostly closed over hundreds of nanoseconds. Increasing the temperature didn’t help, and even when the simulations were started with structures of the protein-small molecule complexes (with the small molecules removed), the pockets quickly slammed shut. Further calculations suggested that the open forms of the proteins are thermodynamically unstable.

The nice thing about computational approaches is that – unlike Scotty – you can change the laws of physics. In this case, the researchers changed the simulated water molecules to be more attractive to carbon and sulfur atoms in the proteins. (They call this SWISH, for Sampling Water Interfaces through Scaled Hamiltonians). This caused the known cryptic sites to open up during molecular dynamics simulations, even in the absence of ligand.

Next, the researchers added very small fragments (such as benzene), and found that these caused the cryptic pockets to open even further. The researchers speculate that this might reflect how cryptic pockets form in the real world: a ligand could worm its way into a transient pocket, stabilizing it and exposing more room for another ligand (or a different part of the first ligand) to bind.

Of course, just because something shows up in silico doesn’t make it real; how do you avoid false positives? Once the researchers found cryptic pockets using “enhanced” water, they reran simulations using standard parameters to see which pockets remained. The researchers found that subtracting the “density” of fragments bound in a conventional molecular dynamics simulation from the density of fragments in a SWISH simulation causes minor, irrelevant pockets to disappear for their three test proteins, leaving only the known cryptic pockets. Running this subtraction experiment on the protein ubiquitin caused a couple weak superficial pockets to disappear, consistent with the absence of cryptic pockets in this protein.

SWISH is an interesting approach, and I look forward to seeing how it compares with other programs, such as Fragment Hotspots and FTMap. It would also be fun to apply SWISH prospectively to therapeutically important but currently undruggable targets to see whether it is worth taking another look at some of them.

New tools for NMR

As most of you know, Teddy has retired from active blogging, which is unfortunate not just for the loss of his wit but also for the loss of his expertise – particularly regarding NMR. But you blog with the army you have, not the army you want, so I'll take a stab at two recent papers on the subject.

The first, published in J. Med. Chem. by Chen Peng and colleagues at software maker Mestrelab in collaboration with Andreas Lingel and colleagues at Novartis, describes an automated processing program for just about any type of ligand-observed NMR data. After going into some detail on how “Mnova Screen” works, the program was benchmarked on three experimental data sets (on undisclosed proteins) which had previously been processed manually. The first was ¹⁹F data from a collection of 19 mixtures of up to 30 fluorinated compounds each – 551 altogether. Here the program performed quite well, identifying 56 of the 64 hits identified manually and misidentifying only 4 compounds as hits. Most of the false positives and false negatives were close to the predetermined cutoff threshold, which can be set as stringent or lax as desired.

T1ρ and STD NMR experiments on 55 individual protein-compound complexes were also examined, and the results were similarly positive. Of course, single compound experiments are easy to analyze, and the real test was with a set of 1240 compounds in 174 pools. Here the program was not quite as good, missing 16 of the 50 manually identified hits and coming up with 74 hits that had not been identified manually. Although most of these were false positives, closer inspection revealed that 10 of them are probably real. Moreover, some of the “false negatives” should perhaps not have been classified as hits in the first place. Clearly the program isn’t perfect, but it does seem to be a quick way to triage large amounts of data.

Of course, ligand-detected NMR provides at best only limited information on binding modes, which is where the second paper comes in, published in J. Biomol. NMR. by Mehdi Mobli (University of Queensland), Martin Scanlon (Monash University) and collaborators at Bruker and La Trobe University. The researchers were interested in finding inhibitors of the bacterial protein DsbA, and a previous screen had identified a weak fragment that initially proved recalcitrant to crystallography.

One of the best methods to determine the binding mode of a ligand is to look at intermolecular NOEs, NMR signals which only show up when two atoms are in close proximity to one another. In theory you can look at NOEs from ligands to the backbone amide protons in proteins, but this is technically challenging for aromatic ligands, of which there are many. Proteins have plenty of methyl groups – so many in fact that it can be difficult to correctly assign each methyl group to a specific residue, leading some researchers to only focus on isoleucine, leucine, and valine (ILV). However, by carefully studying more than 5000 high-quality protein ligand complexes, the researchers found that looking at all the methyl groups in a protein (ie, including those found in alanine, threonine, and methionine) greatly increases the number of protein-ligand complexes suitable for analysis.

The researchers were able to assign most of the methyl groups in DsbA using several approaches, and this allowed them to identify 11 NOEs between their ligand and ILV methyl groups. Modeling was unable to provide a unique binding mode, but by including 8 more NOEs to threonine and methionine methyl groups a single binding mode for the ligand was determined. Crystallography came through in the end too and confirmed the NMR-derived model.

Teddy would normally end his NMR posts by stating – often forcefully – whether he thought the tools under discussion were practical or not. NMR is one of the most popular methods out there, so new tools are clearly welcome. Since I'm no expert on the subject, I'll ask readers to weigh in – what do you think?

CYP121 revisited: fragmentation approaches

Three years ago we highlighted work out of Chris Abell’s lab at the University of Cambridge targeting CYP121, an important enzyme for the pathogen Mycobacterium tuberculosis (Mtb). Two new papers from his group discuss progress on this target using conceptually similar approaches.

A previous fragment screen had identified some very weak fragments, and merging had led to low-micromolar compound 2 – the starting point for a (free access) J. Med. Chem. paper by researchers at Cambridge, the University of Manchester, the Francis Crick Institute, and São Paulo State University. The researchers used a “retrofragmentation” or deconstruction approach: systematically dissecting the molecule into component fragments (such as compounds 4 and 5) to see which bits were most important. Group efficiency analyses revealed that the two lower aromatic rings were important, while the upper one was much less so.

Crystallography revealed that compound 2 did not make direct interactions with the active-site heme molecule in CYP121, so the researchers sought to create some by growing out from compound 4. This led to a nice increase in affinity (compound 19a). Incorporating the other ring led to compound 25a, with sub-micromolar affinity as measured by isothermal titration calorimetry (ITC). Of course, heme is common to every CYP – including those found in humans – raising the question of selectivity. Happily, compound 25a turned out to be reasonably selective for CYP121 compared with a panel of Mtb and human enzymes.

There’s lots more in this (30 page!) paper, including extensive SAR supported by crystallography, ITC, native mass spectrometry, and an interesting spectroscopic binding assay. But unfortunately, the compounds are not active in a cellular assay, and the researchers are trying to figure out why.

The second (open access) paper also takes a deconstruction approach, this time starting from the substrate cYW. Fragmentation of this and related cyclic dipeptide substrates into amino acid derivatives and analogs led to the testing of 65 commercial compounds in a thermal shift assay, resulting in seven hits that increased the denaturation temperature by more than 1 °C. Compound 1a was the most stabilizing, and a spectroscopic assay suggested interaction with the heme group.

The spectroscopic assay also revealed a high micromolar affinity for the fragment. Attempts to improve this ultimately led to compound 31, with comparable affinity as cYW but with improved ligand efficiency. The thioester could be replaced with only a modest loss in potency, and interestingly the stereochemistry of these molecules did not seem to make a difference. Compound 31 was also reasonably selective for CYP121 in a panel of other CYPs.

Both papers cover lots of ground. Reading some publications you can be lulled into thinking that FBLD is an easy progression of increasingly potent compounds. These examples are useful reminders that many compounds turn out to be dead ends, and that even potent and selective molecules may not have the desired biological effects. Sometimes doing everything right can still leave you short of the goal – at least for a while.

Disrupting constitutive protein-protein interfaces

Protein-protein disruptions are notoriously difficult because the interfaces between proteins tend to be large and flat, with few of the deep pockets where small molecules prefer to bind. That's not to say they're impossible: the second approved fragment-derived drug targets a protein-protein interaction. This interaction, as with most others studied (see here, here, and here, for example), is transient: two proteins come together to transmit a biological signal, then dissociate. But many proteins form constitutive dimers or oligomers, and these tend to be even more challenging to disrupt. This is the class of targets discussed in a paper just published in J. Am. Chem. Soc.

Wei-Guang Seetoh and Chris Abell (University of Cambridge) were interested in the protein kinase CK2, a potential anti-cancer target. The enzyme is a tetramer containing two identical catalytic subunits (CK2α) and two identical regulatory units (CK2β). Previous experiments had shown that introducing mutations into CK2β that disrupted dimer formation decreased enzymatic activity and increased protein degradation. Would it be possible to find small molecules that did this?

Chris Abell is a major proponent of the thermal shift assay, in which a protein is heated in the presence of a dye whose fluorescence changes when it binds to denatured protein. The way this assay is normally conducted, small molecules are added, and if they bind to the protein they stabilize it, thus increasing the melting temperature (see here for an interesting counterexample).For oligomeric proteins, one might expect that anything that disrupts the oligomers would destabilize the proteins, thus lowering the thermal stability, and indeed this turned out to be the case in a couple model systems. Thus, the researchers screened dimeric CK2β against 800 fragments, each at the (very high) concentration of 5 mM. No fragments significantly increased the melting temperature, but 60 decreased the stability by at least 1.5 °C.

Best practice for finding fragments includes using multiple orthogonal methods, so all 60 hits were tested (at 2 mM each) in three different ligand-detected NMR assays: STD, waterLOGSY, and CPMG. Impressively, 40 of these showed binding in all three assays. There was no correlation between the binding affinity and the magnitude of thermal denaturation, which is not surprising because the thermal shift incorporates not just the enthalpy change of ligand binding but also the enthalpy change of protein unfolding. Thus, as the researchers note, “the extent of thermal destabilization cannot be used as a measure of its binding affinity.”

Next, all 40 confirmed fragments were tested at 2 mM to see whether they caused CK2β dimer dissociation, as assessed by native state electrospray ionization mass spectrometry (ESI-MS). 18 fragments shifted the equilibrium to monomeric protein, though interestingly no protein-fragment complexes could be observed. These 18 fragments also decreased dimerization in an isothermal titration calorimetry (ITC) assay.

There is still a long way to go: all the fragments are very weak, and preliminary SAR studies were unable to find analogs with significantly improved activity. Indeed, it is unclear where the fragments bind, or whether the binding site(s) are even ligandable. Still, the combined use of biophysical techniques on a particularly gnarly target make this an interesting study on the frontiers of molecular recognition.