Detailing hot spots with atomic consensus sites

Practical Fragments has written frequently about hot spots, regions on proteins that are predisposed to bind ligands such as drugs. Determining whether a protein has a hot spot can help prioritize a target for screening, and one of the more established computational approaches to do so is FTMap, which we wrote about most recently just a couple months ago.

 

While FTMap can tell you whether a protein has one or more hot spots, it provides few further details, such as which regions might prefer a hydrogen bond donor or acceptor. This has now been addressed in a new J. Chem. Inf. Mod. paper by Sandor Vajda and collaborators at Boston University, Stony Brook University, and Acpharis. (Diane Joseph-McCarthy presented some of this work at the CHI DDC conference earlier this month.)

 

The original version of FTMap started with a collection of 16 very small molecule probes: these were docked all over a protein, with hot spots being identified as consensus sites where many probes bound. To get more information about each hot spot, the researchers have extended the method – now called E-FTMap – by increasing the number of probes to 119 covering key functional groups. For example, whereas FTMap included dimethyl ether as a probe, E-FTMap also includes 2-methoxypropane, 2-methoxy-2-methylpropane, and tetrahydropyran. If all these probes bind with the oxygen in the same part of the hot spot, this suggests a predilection for a hydrogen bond acceptor, and also provides information about nearby hydrophobic contacts.

 

By using a sufficiently diverse group of virtual probes, E-FTMap is able to more finely detail hot spots, tallying the “atomic consensus sites” within them. This is reminiscent of an approach we wrote about several years ago, though that method used just three different probes.

 

To benchmark E-FTMap, the researchers took 109 fragment-to-lead pairs with published crystallographic information and assessed whether the program could identify interactions that had been experimentally observed. The results were encouraging and far superior to the original version of FTMap. The highest ranked atomic consensus sites generally overlapped with appropriate atoms in fragments and leads. Interestingly, the results for fragments were better than those for leads, and the researchers suggest this is because the fragment “core is responsible for the bulk of the binding free energy in a ligand and that larger ligands bind by forming additional interactions at weaker hot spots that surround the fragment binding site.”

 

Next, E-FTMap was tested against five proteins for which between 31 and 353 fragment-bound crystal structures were available. Here too the program was broadly successful, though some fragments bound regions of the protein that E-FTMap overlooked, particularly in cases where there were conformational changes. This is not surprising given that the program assumes the protein remains rigid. (Other computational approaches such as SWISH, which we wrote about here, are starting to account for protein flexibility.)

 

E-FTMap looks qualitatively at specific atomic interactions, and one question I had was how well the atomic consensus sites matched up with binding affinities of known fragments; perhaps some crystallographically identified fragments bind so weakly one would not expect to find them computationally, as we discussed here and here. This hypothesis might be tested by focusing on comparisons with experimentally characterized fragments with the highest ligand efficiencies.

 

Also, I was struck by the fact that the virtual probes in E-FTMap are roughly the size of MiniFrags or MicroFrags, and I couldn’t help but wonder how well the atomic consensus sites from the virtual screens would correlate with the binding modes of these tiniest of fragments.

 

One nice feature of E-FTMap is that it can be accessed through a simple web server, so if you’re interested in these and other questions you can test it for yourself. If you do, please share your experiences.

Hot spots real and imagined

Practical Fragments has written several times about “hot spots”: regions on proteins where small molecules and fragments readily bind. Knowing whether your target protein has a hot spot can help you decide whether to pursue the target in the first place. A variety of computational approaches have been developed for finding hot spots, most of which start with a crystallographically determined structure. In a new J. Chem. Inf. Mod. paper, Sandor Vajda and collaborators at Boston University and Stony Brook University ask whether computational models of proteins can also be used for one of the more popular methods, FTMap.

 

The researchers started with a set of 62 proteins, each of which had a published crystal structure bound to a fragment (MW < 200 Da) as well as to a larger molecule. The predicted structures of these proteins were then downloaded from the AlphaFold2 (AF2) site, and these models were truncated to correspond to the residues seen in the crystal structures to facilitate comparisons. The computational models were quite similar to the experimental models, particularly when comparing the positions of the peptide backbone atoms which define the overall shape of the proteins.

 

Next, the researchers applied the program FTMap, which computationally explores the surface of proteins using a set of 16 very small probes such as ethanol. Hot spots are regions where lots of probes bind, and the “hotness” of these spots correlates with the number of bound probes. FTMap assessed hotness on the AF2 structures and the crystallographicaly determined structures. (Before running FTMap, the bound ligands in the crystal structures were computationally removed.) Additionally, the researchers ran FTMap on unliganded crystal structures for the 47 proteins where these had been reported.

 

FTMap was broadly successful at finding the hotspots defined by bound fragments, succeeding 77% of the time starting with either the fragment-bound or unliganded structures and 71% starting with the AF2 models. Implementing stricter criteria (demanding the experimental fragment binding site be the top hot spot, for example) reduced the success to 56% for the crystallographic starting points and 47% for the AF2 models.

 

The paper discusses several examples in detail, in particular the two where the AF2 models were most different from the experimental models. Both of these were large, multidomain proteins. When AF2 models of just the ligand-binding domains were used, the models were significantly improved. This seems to be a generally useful hack: generating truncated AF2 models for other proteins also improved the performance of FTMap.

 

The utility of AF2 models for docking has been the subject of some debate, with some arguing that even though the overall protein folds may be accurate, local side chain conformations may be wrong, and a single side chain rotation may make the difference between ligand binding or not. This paper suggests that hot spots are not too sensitive to these subtleties, and that AF2 models can be used for finding hot spots.

Democratizing computational FBLD with BMaps

Computational approaches to FBLD continue to gain in power. For the most part, they require significant knowledge and installation of expensive, customized software. To remedy this, John Kulp, III and colleagues at Conifer Point Pharmaceuticals have introduced a new web-based application, BMaps, which they describe in a recent J. Chem. Inf. Mod. paper.

 

As the researchers note (and appropriately reference), there are more than a dozen virtual fragment-based design tools and another dozen web-based tools. BMaps (for Boltzmann Maps) aims to provide a full range of functions, from visualizing proteins, finding hot spots, docking fragments, and growing them. It also provides information on the energetics of bound water molecules, which as we’ve written can be crucial players in optimizing protein-ligand interactions.

 

Two key techniques used by BMaps are Grand Canonical Monte Carlo (GCMC) simulations and Simulated Annealing of Chemical Potential (SACP). The first entails comprehensive sampling of different fragment conformations on a protein of interest and assessing binding free energy. The second tool “forcefully inserts fragments into all the binding sites of the protein” and then removes them slowly to evaluate which are most difficult to remove, and thus most tightly bound. Together, GCMC-SACP can be used to evaluate fragment binding to any protein uploaded to the site from the protein data bank, AlphaFold, or any other source.

 

One nice feature of BMaps is a repository of several hundred proteins each with more than 100 fragment and water simulations. BMaps also contains a database of more than 4000 fragments, including MiniFrags. Users can import their own fragments or computationally deconstruct larger ligands. The paper itself is quite short, but the supporting information provides more guidance on how to use the software.

 

The researchers “aim to democratize the availability of accurate fragment and water maps,” a laudable goal. Most computational features are available with a free account, though with restrictions on the number of operations per month.

 

BMaps looks quite powerful and easy to use, but I do wish the researchers had included some full case studies, for example those used by the free FastGrow tool we highlighted last year. Try it out and let the community know what you think!

Ensembles of fragment structures guide selectivity

Scientists generally want structural information when a project begins, and ideally that structural information comes from crystallography. Most of us who have been doing drug discovery for a while can remember seeing the first structure of a favorite molecule bound to a target protein and being inspired, reassured, or sometimes confused. But as crystallography becomes increasingly high throughput, it is now not uncommon to obtain dozens or even hundreds of structures. What to do with all this bounty? In a recent open-access J. Chem. Inf. Model. paper, Mihaela Smilova, Brian Marsden, and collaborators at University of Oxford, the Cambridge Crystallographic Data Centre, and Exscientia describe one application.

 

Back in 2016 we wrote about a computational approach called hotspot mapping, which uses three small fragment probes (aniline, cyclohexa-2,5-dien-1-one, and toluene) to virtually explore potential binding sites and map hydrogen bond acceptors, donors, and apolar interactions. The idea was to predict binding sites and the key interactions likely to drive affinity. The new paper focuses not just on affinity, but on selectivity.

 

The approach starts by taking multiple structures of the same protein bound to various ligands, especially fragments. Ligands and water molecules are then removed, and hotspot mapping is conducted for each structure. Then, all the hotspot maps are combined to generate an “ensemble” hotspot map, which in theory should give a more complete picture of potential attractive and repulsive interactions than a single structure.

 

To assess selectivity, the ensemble hotspot map of one protein is “subtracted” from that of another. If the proteins are very closely related, this “selectivity map” might be empty: all the interactions for one protein would be present in the other. But if there are differences, they become very apparent.

 

Several retrospective case studies are provided. In the first, ensemble hotspot maps were generated from the closely related bromodomains BRD1 and BRPF1, using 23 and 26 fragment-bound structures, respectively. The selectivity map clearly shows the potential for a hydrogen bond donor on a ligand to bind to the backbone amide of a serine in BRD1; the corresponding residue in BRPF1 is a proline, incapable of making this interaction. And indeed, an examination of the literature revealed that this interaction had previously been used to generate inhibitors of BRD1 that were 15-fold selective over BRPF1.

 

The kinases p38α and ERK2 are also closely related, but selectivity maps generated from five p38α structures and 17 ERK2 structures revealed a hydrophobic pocket in the former but not in the latter. This pocket had previously been used to generate selective inhibitors of p38α. Similarly, 28 structures of CK2α and 32 structures of PIM1 were used to generate a selectivity map that also revealed a hydrophobic pocket that can form in the former protein and had been used to generate selective inhibitors.

 

Generally, the more structures available, the more informative the selectivity maps are. The researchers note that though they only used five p38α structures, the fragments were chosen to be diverse (and interestingly all of them made interactions in the hydrophobic pocket). Also, while some protein flexibility can improve the maps, too much is a problem. (For the kinases, only DFG-in structures were used, for example.)

 

This method is a nice synthesis of experimental and computational techniques. A skeptic might argue that it doesn’t provide fundamentally new information: in the examples provided, the selectivity features had already been found and exploited by medicinal chemists. But the automated process and the clear output may speed things up, especially for newer targets, and indeed the researchers note that it is being applied in-house at Exscientia.

 

Perhaps most importantly, if you’d like to try it yourself, the code is freely available here. Happy mapping!

Crystallographic vs computational fragment screening

Several recent Practical Fragments posts have touched on crystallographic screening: from ultra-high concentration screening of “MiniFrags,” to an extensive analysis of fragment structures in the protein data bank, to an open-source effort to develop new antibiotics. A new paper in Phil. Trans. R. Soc. A by Tom Blundell and collaborators at University of Cambridge, the Diamond Light Source, University of Oxford, and several other institutes provides a useful synthesis and an interesting comparison with computational approaches.

The researchers were interested in the bacterial protein PurC, also known as SAICAR synthetase, which is essential for purine biosynthesis and is sufficiently different from its human orthologue to be an attractive antimicrobial target. The protein has an extended binding site that can accommodate ATP as well as its substrate CAIR and an aspartic acid. Using a traditional screening cascade, 960 fragments were screened at 5 mM in a thermal shift assay, resulting in 43 hits. Each hit was then soaked at 10 mM into crystals of PurC, resulting in 8 bound structures, all of which occupy the ATP-binding pocket. Isothermal titration calorimetry revealed dissociation constants as good as 178 µM, with a ligand efficiency of 0.39 kcal/mol/atom.

Next, the researchers ran a computational screen, Fragment Hotspot Maps. This confirmed the main fragment-binding site. Indeed, the crystallographically-identified fragments even make the hydrogen-bonding interactions predicted by the model. However, the computational approach also identified three other hot spots, two in the active cleft and one on the rear of the protein. There was also a “warm spot” next to the ATP-binding site. Are these real, or computational artifacts?

To address this question, the researchers screened fragments at a much higher concentration at XChem, and processed the data using the PanDDA software we’ve previously described. They screened two libraries of fragments at 30-50 mM: 125 “shapely” fragments and 768 “poised” fragments designed for rapid follow-up chemistry. The 8 hits from the first crystallographic fragment screen were also included. This exercise yielded structures for 35 fragments, 60% of which bound in the ATP-binding site, including all 8 of the previously identified ones. Most of the other fragments bound in shallow pockets or near crystallographic interfaces; only one of the other hot spots predicted computationally had a bound fragment, and that was present at low occupancy. Some hits made new interactions around the ATP-binding site, but none bound in the predicted warm spot. Unfortunately, the proportions of fragment hits coming from the two libraries are not broken out.

So in summary, both computational and crystallographic screening correctly identified the “hottest” hot spot, but each approach also identified additional sites that were not confirmed by the other. The researchers ask, “are these sites truly hot spots… or are they weak binding sites routinely seen in crystals?”

This is indeed the key question, and it would be interesting to see whether other computational approaches – such as FTMap or SWISH – are able to shed light on the matter.

Fragment hot spots revisited: a public validation set and method

This is the last week for our poll on how much structural information you need to begin optimizing a fragment – please vote on the right-hand side of the page if you haven’t already done so. We’ve recently discussed crystallography and NMR, so this post is focused on computation.

Predicting hot spots – regions on proteins where fragments are likely to bind – is becoming something of a cottage industry (see for example here and here). These can provide some indication as to whether or not a protein is ligandable, and ideally even provide starting points for a lead discovery program. But how should one searching for promising hot spots and binders choose a method, or evaluate a new one? In a recent paper in J. Med. Chem., Marcel Verdonk and colleagues at Astex provide a method as well as a validation set, both of which are freely available.

The validation set consists of 52 high-quality crystal structures pulled from the Protein Data Bank (PDB). These were chosen to be maximally diverse in terms of fragments (41 of them) and proteins (45). The fragments were not taken in isolation; rather, fragments of larger molecules were considered if they bound in the same region of the protein when presented from at least three different ligands. For example, the researchers note that there are no structures in the PDB of resorcinol bound to HSP90A, even though this is a privileged fragment that usually binds in a conserved fashion at the ATP-binding site in the context of a larger molecule.

Fragments chosen for the validation set have at most one rotatable bond and are quite small, just 5 to 12 non-hydrogen atoms. However, as they are culled from larger molecules, some (such as adamantane) are more lipophilic than standard “rule of three” guidelines.

The 52 examples in the test set were divided into 40 hot spots and 12 “warm” spots, depending on the occupancy of the binding site in the protein across the PDB. For example, the canonical purine binding site of kinases is a hot spot, while the nearby chlorophenyl-binding site of the PKA-Akt chimeric kinase is classified as warm.

With this validation set in hand, the researchers tested an in-house developed fragment mapping method called PLImap (which relies on the previously published Protein-Ligand Informatics force field, PLIff) to see how well they could reproduce the bound conformations of the fragments. The results were quite favorable in comparison with other docking methods tested. That’s exciting, and since PLImap is free to download and use (here), it should be a useful tool for modelers everywhere.

But of course PLImap also made mistakes, and some of these were “wrong in an interesting way.” Water often plays a critical role in protein-ligand interactions, but water was not included in the docking. Several cases where PLImap did not choose the experimentally observed conformation of the fragment involved water molecules. For example, PLImap placed the bromodomain-privileged 3,5-dimethoxylisoxazole fragment in the right location but in a flipped orientation, because the highly conserved water was not present.

Perhaps more interestingly, in some cases warm spots were ignored in favor of hot spots. For example, in the case of the PKA-Akt chimeric kinase mentioned above, the chlorophenyl fragment bound not at the chlorophenyl-binding subsite, where it sits in the context of larger ligands, but rather in the “hotter” purine binding subsite. This phenomenon was observed experimentally several years ago by Isabelle Krimm and colleagues; large BCL-xL ligands that were deconstructed into component fragments bound mostly at a single site, rather than the two sites occupied by the larger molecules. It would be fascinating to test this same set of molecules using PLImap.

All of which is to say that, while computational methods continue to make impressive strides, we are still (happily!) some way from getting rid of the experimentalists.

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.

Calculating hotspots in detail

In the eight years since Practical Fragments first started, Moore’s law has held strong and computational power has increased accordingly. Last year we described how tools such as FTMap can be used to identify hot spots – regions on proteins where fragments are most likely to bind. Although FTMap is quite successful at identifying these, it is less able to point to specific interactions (such as hydrogen bond donors or acceptors) that are likely to drive binding. In other words, computational chemists have become adept at identifying where fragments might bind but lag in predicting how. A new paper in J. Med. Chem. by Chris Radoux at the Cambridge Crystallographic Data Centre and collaborators at UCB and the University of Cambridge addresses this challenge.

The approach starts with a set of three simple molecular probes: toluene, to look for hydrophobic interactions; aniline, to look for hydrogen bond acceptors; and cyclohexa-2,5-dien-1-one, to look for hydrogen bond donors. These probes are larger than those (such as ethanol) used in many other programs, the idea being that too-small molecules might find hot spots so small as to be useless. Indeed, with 7 non-hydrogen atoms, these probes are near the low end of the consensus size for fragments.

Calculations are performed on protein structures – either with no ligand bound or with a bound ligand computationally removed – to determine whether each surface atom of the protein is a hydrogen bond donor, acceptor, or hydrophobic, as well as how exposed the particular atom is. The three probes are then mapped onto the proteins to look for favorable interactions. Regions where multiple probes can bind are scored higher, with hotspots defined as those regions of the protein having the highest scores. The type of probe with the highest score also describes what type of interactions are likely to be favorable at various regions within a given hot spot. Although the researchers note that multiple software packages could be used for these calculations, they used a program called SuperStar, and calculations took just a few minutes on an ordinary laptop.

To validate the approach, the researchers used a previously published data set (discussed here) of 21 fragment-to-lead pairs against a variety of proteins for which crystal structures and binding affinities were available. In general, the method was able to identify the fragment binding site quite effectively; the one outright failure was on the fragment with the lowest affinity, which also had poorly resolved electron density in the crystal structure. Importantly, the fragments tended to have the highest scores, with added portions of the leads scoring lower. This data set was used to calibrate the scoring system for identifying hot spots, as well as specific molecular interactions within each hot spot.

Having thus validated the approach, the researchers took a more detailed look at two published fragment-to-lead programs for protein kinase B and pantothenate synthetase. In both these cases, group efficiency analyses had previously been performed to establish which portions of the ligands contributed most significantly to binding. Gratifyingly, the computations correctly predicted these.

Overall this approach appears promising. At a minimum, it is another tool for assessing the ligandability of potential targets. More significantly, by highlighting the hottest bits of hot spots, it could be useful for medicinal chemists trying to optimize and grow fragments and leads. Unfortunately, as currently described, the process will require a skilled modeler. It would be nice if the authors built a simple web-based interface for people to upload pdb files for analysis, as is the case for FTMap. Also, all the data presented are retrospective – a prospective example would be the true test. Does anyone have experience to share?

Review of 2015 reviews

In the Northern Hemisphere the winter solstice has passed but the days are still short, and 2015 is hurtling into history. As we did in 2014, 2013, and 2012, Practical Fragments will spend this last post of the year highlighting notable events as well as reviews we didn't previously cover.

Two major conferences this year were CHI’s Tenth Annual FBDD meeting in San Diego (discussed here and here) and Pacifichem 2015. If you missed these don’t worry – we’ll have an updated list of 2016 events soon.

After a three year drought, two new books published in 2015: Fragment-based methods in drug discovery and Fragment-based drug discovery. And the trend looks set to continue, with a new book edited by Wolfgang Jahnke and me set to publish in early 2016.

In addition to complete books, several book chapters may be of interest to readers, the first being “Fragment-based drug discovery” by Jean-Paul Renaud and NovAliX colleagues, published in Small molecule medicinal chemistry (Wiley). This is a general review of the topic, focused heavily on biophysical techniques, especially SPR, NMR, and native MS. It also includes a couple case studies – one on the clinical compound AT9283 and one on the bromodomain BRD2.

The next three chapters all come from Springer’s massive Methods in molecular biology series. Continuing the biophysical theme is “Biophysical methods for identifying fragment-based inhibitors of protein-protein interactions,” by Michelle Arkin and colleagues at UCSF. This provides background and step-by-step instructions for SPR, differential scanning fluorimetry (DSF), NMR (including STD, WaterLOGSY, and HSQC/HMQC), and X-ray crystallography. A more detailed guide to STD NMR is provided by Hai-Young Kim and Daniel Wyss (Merck) in “NMR screening in fragment-based drug design: a practical guide,” while Byeonggu Han and Hee-Chul Ahn (Dongguk University-Seoul) discuss STD NMR applied to kinases in “Recombinant Kinase Production and Fragment Screening by NMR Spectroscopy.”

Moving on to journals, two reviews focus on protein-protein interactions. The first, by Thomas Magee (Pfizer) in Bioorg. Med. Chem. Lett., briefly touches on challenges and solutions before focusing on several case studies, including navitoclax, Mcl-1, RPA, KRas, Rad, bromodomains, XIAP, HCV NS3, and more. The second, by Chunquan Sheng (Second Military Medical University, Shanghai), Wei Wang (University of New Mexico and East China University of Science and Technology) and colleagues is published in Chem. Soc. Rev. This is much broader, covering not just fragment-based approaches but others as well, and includes 229 references and 21 figures. There’s a lot of good stuff in this paper, but unfortunately the authors do not discuss the numerous false positives that can occur, such as aggregation and PAINS, and some of their examples are artifacts. Caveat lector.

The next two papers focus on specific therapeutic areas. Xinyong Liu and colleagues at Shandong University discuss the application of fragment approaches to HIV targets in Expert Opin. Drug Discov. In addition to recent examples, this covers some of the older literature, as well as less conventional topics such as dynamic combinatorial chemistry. And in Front. Neurol., Jeffry Madura and Christopher Surratt (Duquesne University) discuss the role fragment-based approaches can play in developing drugs that target the central nervous system (CNS). This review is particularly focused on computational methods.

The next three papers continue the computational theme. Dima Kozakov, Adrian Whitty, Sandor Vajda (Boston University) and co-workers have two reviews discussing work we highlighted earlier this year. The first, in Trends Pharmacol. Sci., is an excellent summary of how computational hot spot analysis can predict whether a protein will be ligandable, and includes a number of case studies. The second, a Perspective in J. Med. Chem., is a much more wide-ranging analysis of the approach. This paper also considers difficult targets, some of which may be tackled with larger molecules such as macrocycles, and others of which may simply not be druggable. And in Chem. Biol. Drug Des., Matthew Bartolowits and V. Jo Davisson (Purdue University), focus on “subpockets,” which are essentially the regions surrounding individual amino acid residues in proteins. This paper also includes an extensive list of software tools for analyzing binding sites.

Finally, Chris Murray and David Rees (Astex) have a brief but lively essay in Angew. Chem. Int. Ed. After providing essentially a target product profile for an ideal fragment, they challenge chemists to devise new routes to superior fragments. Although fragments may seem simple, the “precision synthesis” required to elaborate them “is often rate-limiting.” Diversity-oriented synthesis (DOS) is one potential solution, although there does not seem to have been as much activity here as might have been hoped. Some of the problems are prosaic but significant: as we’ve noted, highly water soluble fragments can be hard to isolate. The authors call for new synthetic methodology compatible with small fragments containing diverse hydrogen-bonding functional groups.

And with that, Practical Fragments says farewell for the year. Thanks for reading (and especially for commenting) and may 2016 bring brilliant breakthroughs!

Predicting protein ligandability and conservation of fragment binding modes

Say you have a protein target, and you want to know whether you will be able to find small molecules that bind to it. A fragment screen can give you a good idea as to the likelihood of success: if you find lots of different fragments with high affinities (say, better than < 0.1 mM), your protein is likely to be highly “ligandable.” On the other hand, if you get very few fragments, and most of them are weak (> 1mM), be prepared for a slog.

Of course, it would be even better if you didn’t have to do a physical screen at all, and two recent papers show how a computational approach may be sufficient. The first, by Dima Kozakov, Sandor Vajda, and their collaborators at Boston University and Acpharis is a detailed how-to guide in Nature Protocols. The second, in Proc. Nat. Acad. Sci. USA by Dima Kozakov, Adrian Whitty, and Sandor Vajda and their collaborators at Boston University, Northeastern University, and Acpharis, addresses some interesting questions about fragment binding.

The main program is called FTMap (also highlighted here); it and several related programs are accessible through a free web server. It is remarkably easy to use: just provide a protein data bank (PDB) ID or upload your own structure and away it goes.

The program works by docking a set of 16 virtual probes (such as ethanol, acetonitrile, acetamide – the largest molecule is benzaldehyde) against a protein and looking for “hot spots” where many fragments cluster. Previously the researchers demonstrated that known ligand-binding sites in proteins tend to be computational hot spots, where at least 16 probes bind. (Note that due to their small size, multiple probes of the same type – acetone, for example – can bind within the same hot spot simultaneously.) In other words,

The strongest hot spot tends to bind many different fragment structures, acting as a general “attractor.”

On the other hand, a hot spot with fewer probe molecules is unlikely to have enough inherent binding affinity to bind to ligands with low micromolar or better affinity.

A related program is called FTSite, which focuses on more thoroughly characterizing the best binding sites. Other programs allow for protein side chain flexibility, docking custom probes, or docking against ensembles of protein models such as generated by NMR structural methods.

The PNAS paper goes further to ask about ligand deconstruction. Specifically, why is it that when a larger ligand is dissected into component fragments, sometimes the fragments recapitulate the binding modes seen in the larger molecule, and sometimes they do not? The answer:

Because a substantial fraction of the binding free energy is due to protein-ligand interactions within the main hot spot, a fragment that overlaps well with this hot spot and retains the interacting functional groups will retain its binding mode when the rest of the ligand is removed.

The researchers support this assertion by examining eight literature examples in which structural information was available for fragments and larger ligands (some of which we’ve covered here, here, and here). In cases where the isolated fragments overlapped with 80% of atoms in probe molecules within a given hot spot, the fragment binding mode remained conserved. Also, these fragments tended to have high ligand efficiency values.

This is neat stuff, and it will be fun to see how general it is. I’m especially happy to see that all of the software is free and open access. Even though I’m hardly a computational chemist, I tried playing around with it and found it remarkably fast and easy to use. So if you have a protein with no known ligands, FTMap can find hot spots, and if they’re particularly promising, this should embolden experimental work.

Fragments vs p97

The protein p97 is important in regulating protein homeostasis, and thus a potential anti-cancer target. But this is no low-hanging fruit: the protein has three domains and assembles into a hexamer. Two domains, D1 and D2, are ATPases. The third (N) domain binds to other proteins in the cell. All the domains are dynamic and interdependent. Oh, and crystallography is tough. Previous efforts have identified inhibitors of the D2 domain, but not the others. Not to be put off by difficult challenges, a group of researchers at the University of California San Francisco (UCSF) led by Michelle Arkin and Mark Kelly have performed fragment screening against the D1 and N domains, and report their adventures in J. Biomol. Screen.

Within UCSF, the Small Molecule Discovery Center (SMDC) has assembled a fragment library of 2485 commercial compounds from Life, Maybridge, and Asinex. These have an average molecular weight of 207 Da and 15 heavy atoms, with ClogP ~1.5. The researchers used both biophysical and virtual screening.

For the physical screening, the researchers started with surface plasmon resonance (SPR), with each fragment at 0.25 mM. This resulted in 228 primary hits – a fairly high hit rate. Full dose response studies revealed that 160 of theses fragments showed pathological behavior such as concentration-dependent aggregation or superstoichiometric binding. A further 30 showed weak or no binding, 13 were irreversible, and 5 bound nonspecifically to the reference surface, leaving only 20 validated hits which were then repurchased.

The 228 primary hits were also assessed by STD NMR, each at 0.5 mM when possible (some fragments were not sufficiently soluble). Of these, 84 gave a strong STD signal, and 14 of these were also among the 20 SPR-validated hits.

The 20 repurchased fragments were further tested by both SPR and STD NMR, and 13 of them reconfirmed by both methods. The paper includes a table listing all 20 compounds, and one observation that struck me was the fact that all but one of the hits – which had dissociation constants ranging from 0.14 to 1.7 mM – are larger than the library average. Such results could argue for including larger fragments in libraries, though this goes against both molecular complexity theory as well as extensive experience at groups such as Astex.

Next, the researchers sought to discover information on the binding sites. Three fragments could be competed by ADP, suggesting that they bind in the nucleotide-binding site of D1. To narrow things down further, the researchers turned to ¹³C-¹H-methyl-TROSY NMR, in which specific side chain methyl groups of Ile, Leu, Met, Val, and Ala were labeled, and chemical shifts were examined in the presence and absence of fragments. Two of the proposed nucleotide-binding site fragments showed similar shifts as AMP or ADP, further supporting a common binding mode (the third was too weak to test). This was not an easy experiment: the hexamer has a mass of 324 kDa, well above where most people do protein-detected NMR.

Independent of all the biophysical screens, virtual screens were conducted using Glide XP, which suggested that the nucleotide binding site would be the hottest hot spot. Happily, all three fragments that appear to bind to this site scored highly in the in silico work, with two of these within the top 100 fragments. However, the binding sites for the other ten confirmed fragments remain obscure.

This paper serves as a useful guide for how fragment screening is performed on a tough target in a top-tier research group. Although difficult, it is not impossible to advance fragments in the absence of structure. While it remains to be seen whether that will be the case for any of these fragments, the researchers have provided a wealth of data for those who wish to try.

From fragment to lead: just remove (high energy) water

The proverb "well begun is half done" suggests that getting started comprises much of the work. Such is the case for fragments that bind to “hot spots,” sites on a protein that are particularly adept at binding small molecules and other proteins. Though fragment-to-lead efforts can give impressive improvements in potency, much of the binding energy of the final molecule resides in the initial fragment. In a new paper in ChemMedChem, Osamu Ichihara and colleagues at Schrödinger have asked why.

The researchers examined 23 published fragment-to-lead examples for which crystal structures and affinities of the fragment and lead were available and in which the fragment maintained its binding mode. They then used a computational tool called WaterMap to assess the water molecules displaced by both the initial fragment as well as the optimized molecule. They compared the calculated thermodynamic parameters (free energy, enthalpy, and entropy) of the water molecules displaced by the initial fragment (core hydration sites) or the bits added to it in the lead (auxiliary hydration sites).

When a protein is surrounded by water, water molecules bind just about everywhere. However, some of these water molecules may “prefer” to be in bulk solvent rather than, say, confined within a hydrophobic pocket on the protein. Perhaps not surprisingly, most of the water molecules displaced by ligands turned out to be of this “high-energy” or unstable variety. Also, the researchers consistently found that the core hydration sites were more unstable than the auxiliary hydration sites. In other words, fragments appear to displace the most unstable water molecules. Moreover, most of this higher energy was due to unfavorable entropy.

It is important to note that the focus here is on individual water molecules (or hydration sites) assessed computationally. The researchers are careful to stress that these may not correlate with thermodynamic parameters obtained by isothermal titration calorimetry (ITC). This is because ITC measures the entire system – protein, ligand, and all of the water – and factors such as protein flexibility can confound predictions.

The researchers summarize their findings as follows.

1) The presence of hydrogen bond motifs in a well-shaped small hydrophobic cavity is the typical feature of the hot spot surface  

2) Because of these unique surface features, the water molecules at hot spots are entropically destabilized to give high-energy hydration sites 

3) Fragments recognize hot spots by displacing these high-energy hydration sites

This provides a framework for understanding several phenomena. First, it describes the origin of hot spots. Second, it explains why much of the binding energy of an optimized molecule resides in the initial fragment; additional waters displaced are not as unstable as those displaced by the fragment, so they don’t give you as much bang for your atom. As a corollary, this might help explain the leveling off or decline in ligand efficiency often observed as molecules become larger.

The researchers go on to discuss specific examples of high-energy waters, noting that a water molecule involved in one or more hydrogen bonds may be particularly hard to replace because recapitulating the precise interaction(s) may be difficult. This is especially true for fragment-growing efforts (where one is likely to be limited in the choice of vector and distance) that aim to displace a high-energy water. Thus, the researchers suggest focusing on fragments that themselves displace high-energy waters, rather than trying to displace these later.

This seems like sound advice, but it likely reflects what folks already do. Since fragments that displace high-energy waters are likely to bind most effectively, won’t these be prioritized anyway? Regardless, this is an interesting and thought-provoking paper.

Two types of hot spots

Practical Fragments has previously written about the concept of hot spots – regions on a protein where fragments are particularly prone to bind. It’s always nice to have one of these when starting a program, since it’s a good indication that the protein is ligandable.

But there’s another type of hot spot too. When two proteins interact, they often do so through very large interfaces comprising dozens of amino acid residues. This is daunting from a molecular recognition perspective, but it turns out that most of these residues contribute very little energetically to the binding affinity, as assessed by alanine scanning mutagenesis. The few residues that do matter often cluster into hot spots, which are generally much smaller than the entire interface.

In a new paper in J. Chem. Inf. Model., Sandor Vajda, Adrian Whitty, Dima Kozakov, and colleagues at Boston University ask how these two types of hot spots are related.

The researchers used the program FTMap to look for fragment-binding hot spots on the protein ribonuclease A (RNase A). They found four in the vicinity of the binding site for the protein ribonuclease inhibitor (RNI), three of which had also been shown experimentally to bind small organic (solvent) molecules.

Having shown that FTMap can find fragment hot spots, they next turned to a set of 15 protein-protein complexes for which alanine scanning data were available; amino acid side chains were considered hot spots if mutation to alanine decreased binding by more than 2 kcal/mol. Applying FTMap to the receptor of each protein-protein pair showed that 92% of alanine-scanning hot spot residues map onto FTMap hot spots. Moreover, there were very few false negatives: 92% of “unimportant” residues did not map onto FTMap hot spots. In other words, the two types of hot spots seem to overlap considerably.

Although these results may make sense intuitively, the researchers discuss several reasons why this was not a foregone conclusion. First, a residue identified as a hot spot by alanine scanning might not be important for binding per se, but may instead be important for imposing long-range structure on the protein. Second, alanine scanning only identifies important side chains; backbone atoms are not considered, so a fragment may bind to a hot spot that is invisible by alanine scanning. Finally, hot spots identified by alanine scanning reflect the interactions between two proteins, whereas hot spots identified by fragments (or their virtual equivalents) look at only a single protein. This is important because if a residue protrudes from one protein into a cavity on the other, the protruding residue may be a hot spot for interactions but, because of its geometry, not be a good binding site for small molecules. As the authors put it:

A convex surface site on a protein typically will not bind small molecules strongly no matter how much binding energy the region generates in an interaction with a complementary concave site on its protein binding partner. Thus, observation of a hot spot by alanine scanning mutagenesis does not necessarily imply the existence of a small molecule fragment consensus site at that region.

And of course, as we’ve noted before, proteins can be wriggly little things, with new pockets opening up where you least expect them. So despite all these caveats, it’s reassuring to see that, for the most part, there is some constancy in the hotness of spots.