Not all aromatics stink the same

A couple years ago we highlighted research suggesting that the more aromatic rings in a molecule, the less “developable” it is likely to be. In the February issue of Drug Discovery Today, the same researchers have now published an update in which they dig into the data in more depth and find that not all aromatic rings are created equal.

As before, the researchers turned to the GlaxoSmithKline internal database of tens of thousands of compounds to correlate chemical features with a variety of measured properties that have an impact on drug development, including solubility, logD, human serum albumin binding, inhibition of several cytochrome P450 isozymes, and hERG inhibition. What they found is summarized in the figure:

In short, while an increase in the number of all-carbon aromatic rings (carboaromatics) had a serious negative effect on nearly all parameters, an increase in the number of heteroaromatic rings was much less problematic. All-carbon aliphatic rings were relatively benign (albeit also relatively rare), while heteroaliphatic rings actually improved most of the properties with the exception of hERG inhibition (and this was only a problem with charged molecules).

One point that was unaddressed in the previous paper was whether an increasing number of aromatic rings is problematic in and of itself, or if this is merely a proxy for larger molecules. In this paper, the authors probed this question directly by examining the properties of molecules with similar molecular weights and lipophilicities but different numbers of aromatic rings. Significantly, the deleterious effects of aromatics appear relatively independent of both size and lipophilicity.

The authors also analyzed ring counts in 1200 oral drugs and found that, while the number of carboaromatic and aliphatic rings has remained relatively constant over time, the number of heteroaromatic rings has roughly doubled from the 1960s to today.

These results provide more support for making sure that fragment libraries contain a good assortment of aliphatics - particularly heteroaliphatics. Aromatics are still very useful of course: as the researchers note, there are thousands of commercially available aromatics, many robust chemistries exist for modifying them, and aromatics provide rigid scaffolds. Thus, fragment libraries should still include a fair share of these moieties, but it is probably worth cutting the number of carboaromatics in favor of more heteroaromatics.

Updated: Fragment-based events in 2011

The first few fragment events of 2011 are almost upon us, but so far it's looking like the second half of the year is completely empty. If you know of anything let us know and we'll get the word out.

February 21-22: SMi’s 10th Annual Advances & Progress in Drug Design Conference will be held in London, and has a number of fragment-based talks. There is also a separate half-day post-conference workshop on the topic on February 23.

February 23-25: CHI’s Molecular Medicine Tri-Conference will again be held in my beautiful city of San Francisco, with a program on medicinal chemistry that includes a fragment section. Notes on last year’s meeting can be found here.

March 7-8: Fragments 2011, the Third RSC-BMCS Fragment-based Drug Discovery meeting, will be held in Stevenage, UK. This is a biennial event; the last was in Alderley Park in 2009 (you can read about it here and here).

April 12-13: CHI’s Sixth Annual Fragment-Based Drug Discovery will be held in San Diego. I will be helping teach one of two pre-conference short courses on the topic on April 11. You can read impressions of last year’s meeting here.

June 8-10: CHI's Eleventh Annual Structure-Based Drug Discovery will be held in Cambridge, MA, with a full session on FBDD on June 9.

Finally, if anyone attends one of these and wants to write a summary please let us know and we'll post it.

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.

Paying the fee for ligand efficiency

Practical Fragments has highlighted a couple cases in which larger molecules have been deconstructed into fragments and then analyzed for binding (see for example here and here). In the most recent issue of J. Med. Chem., Peter Brandt, Matthis Geitmann, and U. Helena Danielson of Beactica have applied this strategy to inhibitors of HIV reverse transcriptase (HIV-1 RT). They also delve into the theory and energetics.

The researchers dissect three non-nucleoside reverse transcriptase inhibitors (NNRTIs) into a total of 21 commercially available “fragments”. Each of these was then tested for binding using SPR (see also this paper for a detailed account of how they perform these screens, and this one for discovery of new fragments against this target). If the binding energies of the fragments were evenly distributed across the entire parent NNTRIs, most of the fragments would be predicted to be sub-millimolar. In fact, most of them were much worse: only 9 showed any evidence at for binding, and only 3 were fragment-sized (the other six had molecular weights above 300 Da).

This sort of result – that fragments of larger molecules bind less effectively than predicted – has now been seen several times, and the researchers asked why. One issue is that when a molecule binds to a protein it loses translational and rotational entropy, and this imposes an energetic cost. This “fee” is, unfortunately, hard to estimate, and complicated by the fact that there may be further energetic costs if the protein itself is flexible (as in the case of HIV-1 RT). The authors provide a nice review of the literature, where values range from 2.5 to a whopping 16 kcal/mol (see here for more discussion on this). When they (admittedly arbitrarily) subtracted 7.0 kcal/mol, the agreement between expected and observed binding of their fragments improved.

However, as the researchers acknowledge, this model still assumes that the binding energy is equally distributed over the entire parent molecule – in other words, it ignores the existence of hot spots. The fact that hot spots exist probably accounts for the decrease in maximum observed ligand efficiency with an increase in the number of heavy atoms:

Once [the hot spot] is occupied, larger molecules need also to interact with other parts of the ligand binding pocket. Hence, a decrease in ligand efficiency will be observed for larger molecules.

True, and to complicate things even more, different proteins will have hot spots of different sizes and “temperature” – or perhaps none at all. This variation calls into question the utility of using notions such as fit quality or %LE, which attempt to normalize ligand efficiency for the size of the ligand. The problem is that different proteins are likely to have different maximal affinity ligands; kinases tend to have high-affinity binding sites where high ligand efficiency can be achieved, while for protein-protein interactions the ligand binding site is likely to be larger and the ligand efficiencies lower. Thus, one-size fits all metrics could prove too stringent – or not stringent enough.

Fragments in Nature

The most recent issue of Nature has a brief but trenchant summary of fragment-based screening (FBS) by Abbott’s Phil Hajduk, of SAR by NMR fame. This is the first half of a drug discovery forum comparing FBS with diversity-oriented synthesis, or DOS, covered by Warren Galloway and David Spring of the University of Cambridge.

Hajduk summarizes the advantages of FBS:

Fragment libraries are more diverse, synthetic resources are used more efficiently and the leads identified from FBS are more likely to yield drug candidates that have optimal physico-chemical properties.

He also points out that fragment-based approaches have led to a number of drugs in the clinic.

In the spirit of “vigorous debate,” Hajduk also takes aim at DOS. In comparison with fragment-based approaches, which start with small libraries of small fragments, DOS generally makes use of larger libraries of structurally diverse molecules which are usually drug-sized and are often inspired by natural products. However, Hajduk alleges that:

Most compounds in DOS libraries would be excluded from many corporate screening collections because of their poor physico-chemical properties.

I don’t know about “most”, but I will say that many DOS compounds look suspiciously like PAINS. Still, DOS does have at least one strength: FBS is generally limited to well-characterized systems with purified proteins, whereas DOS libraries can be used in complex phenotypic assays where the target may not be known. Whether these will ultimately yield new drugs remains to be seen.

What do you think?

18 PI3K fragments

As we’ve noted before, kinases are a fertile field for fragment finding, but most of the targets have been protein kinases. Lipid kinases such as the phosphatidylinostide 3-kinases (PI3Ks), which mediate signal transduction by transferring a phosphate group to lipids, are also popular targets for a variety of diseases, but less has been disclosed about their suitability for fragment-based lead discovery. A paper in a recent issue of Bioorg. Med. Chem. Lett. remedies that.

Fabrizio Giordanetto and colleagues at AstraZeneca started with a homology model of p110beta (no crystal structure of this enzyme has been reported). They then used commercial software to dock 183,330 fragments selected from their corporate collection. All fragments that made at least two hydrogen bonds with the protein were organized into clusters of similar molecules and representatives of each cluster were visually inspected. This led to the selection of 210 fragments to be screened against the protein, of which 18 showed measurable activity. Structures of these fragments are provided in the paper; they range from kinase workhorses such as compound 1 to known PI3K motifs such as compound 10 to more unusual molecules such as compound 18. These hits were also tested on other members of the PI3K family, and while most showed activity across the board, others (such as compound 18) showed some selectivity.

There are some interesting structures in here; if I were starting a PI3K program I would definitely take a close look at them. Although the researchers have likely developed some of these into attractive leads, one of the virtues of fragments is that they are often so protean that different teams can start with the same fragment and end up in very different places.

Fragment linking in crystallo

Of the many ways to link fragments, one of the most intriguing is when the protein itself catalyzes or templates the assembly of two fragments (see for example here and here). The latest example of such target-directed fragment linking was published in last month’s issue of J. Appl. Cryst.

The researchers, led by Isao Tanaka at Hokkaido University, were interested in ligating fragments together in protein crystals. They first took crystals of the model protein trypsin and soaked these with an “anchor molecule,” in this case one of two benzamidine-containing aldehydes (benzamidines are classic trypsin binders). The crystals were then transferred to a second solution containing a “tuning molecule,” each of which contained either an aminooxy or hydrazine moiety that could react covalently with the aldehyde of the anchor molecule. Finally, the crystals were analyzed by X-ray and structures of any bound ligands solved.

A total of 33 different tuning molecules were examined, and two of these produced clear electron density in the active site showing that ligation with the anchor molecule had occurred (for example ALD2 and OXA9). Three others produced structures that suggested some disorder in the binding mode of the tuning molecule, and a fourth showed an assembled product that extended from the active site to a second trypsin molecule in the crystal lattice.

A study similar to this was published a number of years ago, but in that case it was not clear whether the ligation occurred in the crystal or in solution. In the present case, soaking pre-assembled molecules into the crystals produced inferior electron density to the two-step process. More excitingly, time-resolved experiments actually showed structural snapshots of the complex forming, both in the active site (which occurred in under a minute) as well as at the dimer interface (which took over an hour).

Unfortunately, the assembled products are not notably better binders than the initial fragment. The authors attribute this to the fact that their library of tuning molecules was very small. However, it is also possible that the approach selects not for the best binders but for those that can best form complexes within a fairly rigid crystal lattice. As we’ve seen before, protein crystals are far from physiological. It will be interesting to see whether in-crystal chemical ligation can generate superior binders.

Ligand efficiency in action

At an introductory talk I was giving recently on FBLD, someone asked how useful ligand efficiency (LE) really is. A paper published online in J. Med. Chem. by Daisuke Tanaka and colleagues at Dainippon Sumitomo Pharma illustrates how the metric can guide medicinal chemistry to superior molecules.

The enzyme soluble epoxide hydrolase (sEH) is a potential anti-inflammatory target. Inhibitors have been reported, but these tend to be quite lipophilic, so the researchers sought to find smaller, less hydrophobic inhibitors that bound with high-affinity to the target. A virtual screen against multiple ligand-bound crystal structures led to the selection of 735 diverse compounds which were tested in a biochemical assay, resulting in 68 compounds with IC50 values better than 1 micromolar. Most of these were relatively hydrophobic amides or ureas.

After removing known chemotypes and obviously unattractive molecules, the researchers were left with 42 compounds. They decided to eliminate compounds with MW > 380 or logP > 3.5, leaving 17 compounds; as might be expected, these smaller molecules turned out to be the most ligand-efficient of the bunch. Despite being one of the weakest hits identified, fragment-like compound 1 was the most ligand efficient and was chosen for lead optimization. A crystal structure of this compound bound to sHE guided parallel synthesis of 155 analogs, all with low molecular weights. Many of these compounds were very potent, and compound 11 not only showed a nice improvement in affinity but also demonstrated good ADME properties.

The authors conducted a retrospective analysis using some of the other ligand-efficiency-like indices such as %LE and fit-quality, which allow looser standards for affinity as molecule size increases. Interestingly, compound 1 was not an obvious starting point using these metrics. It is impossible to say what would have happened had these measures been prioritized over ligand efficiency, but the success of the simpler LE suggests that taking size into account would not have been useful, and could even have been misleading.

This paper is not a traditional fragment paper in which a low affinity lead is optimized; the initial hit was already quite potent. Rather, as the authors note, this is a nice example of “fragment-inspired medicinal chemistry, in which the essence and advantages of FBDD are faithfully respected.” It also provides another example of how focusing on ligand efficiency, rather than just potency, can lead to attractive chemotypes.

Flattening fragments

If you want to visualize and manipulate crystallographic data there are plenty of options, such as PyMOL and OpenAstexViewer. However, powerful molecular graphics programs such as these have a learning curve associated with them, and sometimes it’s nice to get a simple two-dimensional image. In the December issue of ACS Medicinal Chemistry Letters, Katrin Stierand and Matthias Rarey describe software they’ve developed to do this.

PoseView, which can be used online free of charge, automatically generates a two-dimensional view of a protein-ligand complex and labels five different types of interactions: hydrogen bonds, metal interactions, pi-cation interactions, pi-pi stacking, and undirected hydrophobic contacts.

The researchers tested their software by using it to draw nearly every protein-ligand interaction in the Protein Data Bank (PDB) in which the ligand has 5 to 80 atoms, excluding crystallization buffer components and things like heme or iron-sulfur clusters. For the remaining 200,000+ complexes, PoseView successfully generated plots for 85% of them; interestingly, most of the failures resulted because there were no apparent interactions between the protein and “ligand.” Of the successes, 80% produced high-quality figures in which there were no overlapping features – not bad considering the challenge of reducing a three-dimensional model to a two-dimensional image.

The paper includes some interesting data about protein-ligand complexes in general. For example, over 90% of complexes in the PDB have less than 11 directed interactions (ie, interactions other than hydrophobic contacts), with the largest number having only a single directed interaction. It would be interesting to see how these statistics apply to fragment-sized ligands.

PoseView is incredibly easy to use: just enter a pdb code or upload a structure, and the program spits out an image. I tried it on the very first fragment I discovered, and the result looks quite attractive:

Of course, there is always room for improvement: for example, it would be nice if you could change fonts or colors (the output is a pdf). Still, this is a useful addition to the toolbox of modeling programs.

And with this post, Practical Fragments says goodbye to 2010 and wishes everyone a happy holiday season. We look forward to more exciting developments in 2011 – and if we miss any, please send them our way!

SAR by Crystallography – minus the “A”

Practical Fragments has highlighted at least a couple papers (here and here) where fragments were identified by crystallography but where functional activity was not reported. Indeed, most experienced fragment hunters will be able to point to cases where they observed well-defined electron density but were unable to measure inhibition. Although this is often a source of frustration, researchers from Johnson and Johnson decided to avoid measuring activity altogether until after a couple cycles of crystallography-guided library design, as they describe in a paper published this month in J. Med. Chem.

The researchers were interested in the protein ketohexokinase, an ATP-dependent enzyme that is a potential target for metabolic diseases. They assembled a primary fragment screening library of 900 compounds with the following characteristics:

• 6-15 non-hydrogen atoms
• < 3 hydrogen bond acceptors
• < 3 hydrogen bond donors
• < 2 rings
• No unspecified chiral centers
• No unpleasant functionalities (peroxides, acyl halides, etc.)
• Prioritization given to molecules found in known pharmaceutical compounds (similar to the Fragments of Life)

This is a fairly standard set of criteria for assembling a fragment library. The researchers broke with convention by pooling fragments into groups of 5, with the members of a given fragment pool as structurally similar as possible. Normally when researchers pool libraries for crystallography-based fragment screening the idea is that each member of a given pool will be structurally unique to facilitate identification. In this case, though, the researchers were interested in observing general features of how fragments bind to the target protein.

Crystals of hexokinase were soaked with the pools, and 60 of these pools yielded electron density in the active site – a very high hit rate of 30%. A number of generalizations could be made about what molecular features were preferred in different parts of the binding site, and this information was used to build a secondary library of about 350 compounds based on 6 scaffolds; the idea was to merge or grow fragments found in the primary screen. These compounds were similarly pooled and screened crystallographically but not functionally, yielding hits from 4 of the scaffolds. The structures of these hits were then used to generate a third library of several hundred compounds around at least 4 separate scaffolds. These compounds were then tested individually to see if they could inhibit ketohexokinase, and were also characterized crystallographically. Gratifyingly, a number of these were active at submicromolar potency, including compound 6.

In addition to showing decent potency against ketohexokinase, compound 6 exhibited respectable drug-like properties as well, including lack of activity against the CYP450 enzymes, high oral bioavailability, and good selectivity against receptors, ion channels, and protein kinases, though other ribokinases were not tested.

This is an interesting and potentially useful approach, but it would be fun to see how it compares with a more “traditional” strategy, where individual fragments are advanced. It should be relatively straightforward to assay fragments from the first and second library for affinity; was fragment 3 (precursor to compound 6) one of the most potent or ligand efficient? Or was the information from fragments with no detectable activity instrumental in delivering the best compounds?

Fragment-based job listings

The fragment-based community is very diverse, with practitioners coming from biochemistry, biophysics, chemistry, computational chemistry, and many other disciplines. Communicating job openings to such a fragmented community can be difficult, but we think Practical Fragments can provide a common place for employers who want to build up fragment expertise of any flavor. If you have an opening you are trying to fill, please add it as a comment to this post. These comments will then be listed in the “Job Openings” link under the “Links of Utility” heading on the right side of the page, so they will always be easy to find.

(Note: this is for positions in fragment-based drug discovery only. There is also a LinkedIn group that lists jobs with a broader but overlapping focus.)

Hsp90 and fragment linking

There has been no shortage of fragment-based approaches directed toward the anti-cancer target Hsp90, most of which have relied on growing fragments (see here for some impressive recent examples). Researchers from Abbott published a report providing a couple examples of linking fragments against this target a few years back, but in those cases the ligand efficiencies of the linked molecules were dramatically lower than those of the initial fragments. In a recent paper in ChemMedChem, researchers from Evotec describe an example that maintains the ligand efficiency.

The research group had previously conducted a fragment screen against Hsp90, resulting in a number of hits. In the new paper, fragment hits 1 and 2 (see figure) were both found to have fairly low affinities, but were characterized crystallographically. Interestingly, fragment 2 could adopt at least two very different conformations, depending on whether it was co-crystallized in the presence of fragment 1. In the ternary structure, the two fragments come within about 3 Å of each other, and molecular modeling suggested that four atoms should be able to link them.

Gratifyingly, when such a compound was made and tested, it inhibited the enzyme several hundred-fold more tightly than either of the initial fragments. The crystal structure revealed that the compound binds similarly to the ternary structure of Hsp90 and fragments 1 and 2.

The authors note that “the binding free energy of the linked fragment 3c was found to be exactly the sum of those of the original two fragments.” Of course, this is still a long way from an ideal linking situation: as noted earlier this year a good linker should lead to super-additivity (an improvement of ligand efficiency), not just additivity (maintenance of ligand efficiency). Nonetheless, this example is still better than many attempts at linking, which often are less than additive.

Fragment-based conferences in 2011

As 2010 winds to a close, it’s time to start planning for 2011. As in previous years, fragment events seem to cluster in the first half of the year, so start planning now!

February 21-22: SMi’s 10th Annual Advances & Progress in Drug Design Conference will be held in London, and has a number of fragment-based talks. There is also a separate half-day post-conference workshop on the topic on February 23.

February 23-25: CHI’s Molecular Medicine Tri-Conference will again be held in my beautiful city of San Francisco, with a program on medicinal chemistry that includes a fragment section. Notes on this past year’s meeting can be found here.

March 7-8: Fragments 2011, the Third RSC-BMCS Fragment-based Drug Discovery meeting, will be held in Stevenage, UK. This is a biennial event; the last was in Alderley Park in 2009 (you can read about it here and here). Registration is now open, as is a call for posters through January 31.

April 12-13: Cambridge Healthtech Institute’s Sixth Annual Fragment-Based Drug Discovery will be held in San Diego, with at least one pre-conference short course on the topic on April 11. You can read impressions of this past year’s meeting here.

Know of anything else? Organizing a fragment event? Let us know and we’ll get the word out.

More techniques: NovAliX and Graffinity combine MS and SPR

We’ve written previously about Graffinity, which uses surface plasmon resonance (SPR) to find fragments, and NovAliX, which initially focused on mass spectrometry. The two companies have been collaborating since last year, and this has apparently been a fruitful partnership: this month NovAlix acquired a majority ownership stake in Graffinity (click here for press release). Earlier this year NovAlix also purchased an NMR-focused company, and they have already added crystallography expertise. Finding fragments effectively requires using a range of orthogonal technologies, and this latest move gives NovAlix a full suite of biophysical techniques.

Fragments vs PDK1

Kinases have been a particularly productive target class for fragment-based drug discovery (and drug discovery in general), with nearly half of reported FBDD-derived clinical candidates targeting kinases. The latest dispatch from this field can be found in the November issue of ACS Medicinal Chemistry Letters.

In this paper, Jeffrey Axten and colleagues at GlaxoSmithKline describe their use of fragment screening to identify inhibitors of PDK1, a popular anti-cancer target. They started by assembling a library of fragments biased towards the purine-binding site of kinases, and tested 1065 of these in a biochemical screen at 400 micromolar concentration. Of these, 193 inhibited activity at least 60% and were further characterized; 89 had IC50 values better than 400 micromolar. A set of 36 of these, chosen on the basis of ligand efficiency and chemical tractability, were chosen for follow-up.

Saturation transfer difference (STD) NMR was used to confirm which fragments bound to PDK, which cut the number of hits in half. X-ray crystallography experiments were started before NMR and performed on 7 fragments; only the fragments that were confirmed by NMR gave interpretable data. One of these was the aminoindazole compound 8 (see figure).

A substructure search was conducted to find more elaborated molecules within the corporate screening collection, leading to compound 19, which has sub-micromolar potency. This compound also showed some signs of selectivity for PDK1 over other kinases. Although the paper stops here, Jeffrey Axten gave a nice presentation at FBLD 2010 in which he discussed subsequent medicinal chemistry that ultimately led to novel, high picomolar inhibitors of PDK1.

There are at least two lessons from this story. First, the significant attrition from the biochemical screen again emphasizes the need for orthogonal methods of fragment validation. Second, even though the fragment identified has been around the block with respect to kinases (as of last year, the aminoindazole substructure had appeared in over 70 kinase patents), skillful medicinal chemistry can still get you to novel compounds.

Pin1 revisited

Earlier this year we highlighted a paper from Vernalis that described the use of NMR methods to discover inhibitors of the anti-cancer target Pin 1. In a recent issue of Bioorg. Med. Chem. Lett. the same team now reports a second series of compounds that inhibit this protein, also discovered and advanced through FBLD. The two papers together provide some interesting lessons.

Rather than using NMR, the researchers identified the second series of compounds with an inhibition assay. After screening 900 fragments at 2 mM, they obtained 40 hits, including 3 compounds previously discovered by NMR. Disturbingly though, follow-up NMR experiments confirmed binding for only 2 of the 37 new hits, suggesting that the remaining compounds may act through pathological mechanisms (see also here). Still, two hits are better than none, and the binding mode of one of the fragments (compound 3 in figure) was determined by X-ray crystallography. Several analogs of this were purchased and tested, and compound 5 was found to have an improved potency and ligand efficiency.

At this point chemistry entered the picture. The researchers synthesized several analogs of compound 5, guided by crystallography and modeling. This led to compound 10e and eventually to compound 20, with sub-micromolar biochemical activity and measurable cell activity.

Ultimately though, as in the previous Pin1 series, even this modest cellular potency was gained at the cost of unacceptable increases in size and hydrophobicity. This brings up an interesting question: at what point do you declare a target undruggable? The authors note that “the nature of the Pin1 active site makes it difficult to optimise hits into drug-like molecules.”

Fragment-based approaches can sometimes deliver inhibitors to challenging targets where HTS has failed. However, if the inhibitors can’t be transformed into drugs, is finding them actually a good thing? Researchers are getting better at improving potency at the same time as ligand efficiency for some targets, but ultimately getting to clinical candidates for harder targets may come down to how many resources you are willing to throw at a project: molecules such as ABT-263, though far from rule-of-5 compliant, are doing well in the clinic, but only after the investment of dozens if not hundreds of people-years.

Ligand efficiency hot spots

Hot spots are regions of a protein with a particular predilection for binding to small molecules – thermodynamic sinkholes, so to speak. Discovering one of these can get you to potent molecules very quickly. In an effort to better understand hot spots, Iwan de Esch and colleagues at VU University in Amsterdam and collaborators at Beactica have deconstructed a potent ligand for nicotinic acetylcholine binding protein (AChBP), a model protein for ligand-gated ion channels, which are implicated in a variety of neurological diseases. They report their results in a recent issue of J. Med. Chem.

The researchers started with the previously reported quinuclidine compound 6 (see figure) and fragmented this into 20 analogs. They tested these in a surface plasmon resonance (SPR) assay as well as in a more conventional radioligand binding assay; the agreement between these very different assay formats was excellent, further validating the utility of SPR as a useful tool for discovering fragments.

Not surprisingly, some of the fragments have higher ligand efficiencies than the larger, more potent molecule, suggesting that there is a hot-spot that recognizes the core fragment 25 (which is structurally related to nicotine). This concept of “group efficiency” has been described previously, and can be useful for optimizing fragments. For example, compound 22, without a basic nitrogen atom, has the lowest ligand efficiency in the bunch; presumably, simply adding the nitrogen would give a sizable boost in potency.

However, one needs to be cautious. The researchers use computer docking to develop models of how each of these fragments bind, but as we have seen before, isolated fragments do not always recapitulate the binding modes of fully elaborated molecules. Still, particularly in the absence of structure (as is the case with many ion channels), exercises such as this could provide useful ideas for what to do with fragment hits.

Fragment linking for specific Bcl-2 inhibitors

One of the most well-known examples of a fragment-based program that has yielded a clinical compound is Abbott’s Bcl-2 effort: ABT-263 is currently in over a dozen trials for various cancers. However, this molecule hits several proteins in the Bcl-2 family, and a more specific inhibitor of Bcl-2 alone may have lower toxicity. Phil Hajduk and colleagues have used SAR by NMR to do this, as they report in the latest issue of Bioorg. Med. Chem. Lett.

The researchers started with a protein-detected NMR screen of 17,000 compounds; compound 1 (see figure) was found to be fairly potent, and also roughly 20-fold selective for Bcl-2 over the related protein Bcl-xL. Hajduk’s team did not have a crystal structure at this point, but they were able to use NMR to determine that the compound lies in a large hydrophobic groove. This is the same groove found to bind biaryl acids in earlier work, so the researchers screened a set of 70 of these to see if they could bind in the presence of compound 1. Interestingly, compound 6 was equally potent in the presence or absence of compound 1, suggesting that both fragments could bind simultaneously.

NMR was then used to determine the ternary structure of fragments related to compound 1 and compound 6 bound to Bcl-2, and linking these led to compound 25, with high nanomolar potency. Although this represents a good boost in potency, the binding energies were not additive (let alone synergistic). An NMR structure of one of the linked molecules revealed that, although it binds in the same groove as its component fragments, its position is shifted, and also that one of the protein side chains moves to deepen a hydrophobic pocket. Additional chemistry to fill this pocket led to compound 29, with 40 nM biochemical potency and measurable cell activity, as well as greater than 1000-fold selectivity for Bcl-2 over Bcl-xL and at least 28-fold specificity over other Bcl-2 family members.

This is a nice example of starting with a modestly selective fragment (albeit a jumbo-sized one) and, through fragment linking, increasing both the potency and specificity towards the target protein.

Small but PAINful

We’ve written previously about the phenomenon of compound aggregation, and how this can lead to false positives in high-concentration screening (see also here). Unfortunately, aggregation is not the only thing that can trip you up. Earlier this year Jonathan Baell and Georgina Holloway published a description and systematic catalog of “Pan Assay Interference Compounds,” or PAINS, and Baell has now followed up on this with a Perspective in the October issue of Future Medicinal Chemistry. (For those of you without journal access, he has also summarized some of this material here.)

PAINS are compounds that frequently show up as screening hits, but that act through non-specific mechanisms such as covalent attachment to proteins or generation of hydrogen peroxide. Sometimes it is not the compound itself that is problematic but a breakdown product or leftover reactant; most readers with experience in hit-to-lead discovery will experience a painful sense of déjà vu reading about hits that did not confirm when the compound was resynthesized or repurified. This problem is exacerbated when trifluoracetic acid (TFA, which is toxic to cells and can accelerate compounds’ decomposition) is used in reverse-phase purification of compounds; apparently Pfizer has systematically repurified hundreds of thousands of their screening compounds to remove traces of TFA.

The problem with PAINS is that they may show convincing biochemical and even cell based activity, but mechanistically be useless for further advancement to drugs or even chemical probes. Unfortunately this does not prevent them from being published, where they are picked up by other researchers and identified as “privileged pharmacophores” by computational chemists. This leads to more (wasted) research and more (useless) publications in a vicious circle Baell terms “pollution of the scientific literature.”

Although the focus of the Perspective is high-throughput screening, fragments can be PAINS too, as the sampling below shows.

Compound 1 is obviously a Michael acceptor and compound 2 is unsurprisingly reactive with cysteine, but the problems with compound 3 are less obvious, and compound 4 generates hydrogen peroxide in the presence of reducing agents and oxygen – a mechanism it likely took some time to track down.

One of the fun aspects of reading this paper is that Baell is not shy about calling out high-profile publications that are likely to report false positives, though he readily admits that he too has been misled. The question is what to do: using computational filters to get rid of the worst offenders only eliminates 5-12% of commercial compounds, but more stringent screens can eliminate upwards of 95%!

Baell calls for better stewardship of the scientific literature:

Journals therefore have a responsibility here because if misleading information is published, unintentional though it may be, it is more likely to propagate and be taken as fact by others who may then initiate flawed research projects. Research resources are too precious for this to be acceptable.

I agree whole-heartedly with this, though given what I see published on a weekly basis I despair of this happening any time soon. Ultimately of course the onus is on researchers to carefully follow up and fully understand the mechanism of their hits, particularly those with dubious structures: don’t contribute to the pollution yourself, and don’t let friends (or papers you review) pollute!

FBLD 2010

The last major fragment event of this year is over, but it ends on a high note: FBLD 2010 has remained true to its predecessors in bringing together a great group of fragment enthusiasts in a Gordon-Conference-like environment. With 30 talks and even more posters I won’t attempt to be comprehensive or even representative, but will instead just pick out a few themes. Those of you who were there, please chime in with your own observations.

One of the themes was the shape of chemical space, and what makes a good binder. Jean-Louis Reymond, who has been systematically enumerating all stable molecules containing carbon, nitrogen, oxygen, and a few other atoms, has already published up to 13 heavy atoms but is now expanding his analysis to molecules containing up to 17. The issue of whether more attention should be given to three-dimensional fragments was discussed, with Ken Brameld reporting that fragments in crystal structures at Roche and the protein data bank contain fewer “flat” compounds than does the ZINC database of commercial molecules. However, analyzing 150,000 molecules with MW < 300 that had been screened in 40-100 high-throughput screens at Roche did not show any shape differences between the 50,000 molecules that showed up in at least one screen and those that didn’t. Interestingly, this ratio also came up in a talk by Tony Giannetti of Genentech, who said that across 13 screens 36% of their fragments hit at least one protein, while the rest didn’t hit any. Vernalis has found similar results; is there any way to enrich for the productive binders?

While FBLD 2009 had a strong computational theme, a major thrust of this conference was using biophysics to detect and confirm fragment binding. Tony discussed best-practices in SPR, and noted that since small molecules are “brighter” in NMR assays than proteins it is possible to find even very tiny fragments, including a 6 heavy-atom compound with a Kd of 600 micromolar. Tony also described the use of SPR for weeding out badly behaved compounds. Spookily, he noted that promiscuity is a function of compound, protein, and buffer, so it is not possible to weed out bad actors in a library before screening: one compound that was promiscuous against 8 targets bound legitimately and gave a crystal structure with a ninth. Adam Renslo of the University of California San Francisco described how easy it is to be misled by such phenomena. Glyn Williams described how Astex uses biophysical techniques to detect problem compounds, and noted that oxidizers can be particularly insidious – a trend that will likely continue as people explore novel heterocycles.

Glyn also presented a fascinating if slightly depressing discussion of ligand efficiency. As many have found, it can be challenging to maintain ligand efficiency during the course of fragment optimization. Yet even this goal is too modest. A fragment pays about 4.2 kcal/mol in binding energy when it binds to a protein due to loss of rotational and translational entropy; since this enropy cost is only paid once, atoms added to this molecule do not have this liability . Thus, merely maintaining ligand efficiency means that the atoms being added are binding less efficiently. This point was also emphasized by Colin Groom of the Cambridge Crystallographic Data Centre.

Membrane proteins are increasingly being targeted by fragment-based methods, as recently discussed on this site, and both Gregg Siegal of ZoBio and Rebecca Rich of the University of Utah presented progress against GPCRs.

There was general agreement that many approaches can find fragments, and that using several orthogonal methods is a good way to separate the true binders from the chaff, but a continuing challenge is what to do next. The last two sessions were devoted to chemical follow-up strategies and success stories. Some of these have been at least partially covered on Practical Fragments (for example here, here, and here) but there were a number of unpublished examples too – we’ll try to discuss these individually as they emerge.

If you missed this or the previous two conferences you’ll have another chance in 2012, when the meeting will be held in my fair city of San Francisco. And if you can’t wait that long, there are at least two fragment conferences scheduled for next year – details to come shortly.