A slew of sites for fragments in HIV Reverse Transcriptase

The protein HIV-1 reverse transcriptase (RT) has been something of an Achilles heel for HIV; 13 approved drugs inhibit this enzyme! However, HIV is more adaptable than Achilles, and can develop resistance to drugs, creating a need for new molecules. With this in mind, Eddy Arnold and colleagues at Rutgers University performed an extensive fragment campaign against this target; their work was recently published in J. Med. Chem.

The researchers assembled a library of 775 fragments, 500 from Maybridge and most of the rest from Sigma-Aldrich and Acros. These were combined into 143 pools of 4 to 8 fragments, each at 100 mM in DMSO. Crystals of RT grown with the drug rilpivirine were soaked with each of the pools; rilpivirine stabilizes the protein and yields crystals that diffract to high resolution. The researchers also added 80 mM arginine and 6% trimethylamine N-oxide (TMAO) to the soaking solutions; arginine helped solublize some of the more hydrophopic fragments and improved electron density, while TMAO improved diffraction.

Overall, the researchers found 34 fragments that bound to HIV RT, a hit rate just over 4%. Interestingly, halogenated fragments seemed to give a much higher hit rate: 7 of 29 fluorine-containing fragments produced structures, as did 4 of the 17 brominated fragments and one of the two chlorinated fragments. I don’t recall seeing halogens previously over-represented among fragment hits, though last year we did write about halogen-enriched fragment libraries. The sample sizes reported here are small, but if the findings hold up in other studies, fluorine fetishism may be further justified.

But just as interesting as the composition of the fragment hits is the number of binding sites in the protein: 16, with names ranging from the descriptive (“NNRTI Adjacent” and “Incoming Nucleotide Binding”) to the concise (“399”) to the downright thuggish (“Knuckles”). In the case of three of these sites, some of the fragments also inhibited enzymatic activity.

There is a lot of nice information here, and eight co-crystal structures have been deposited in the protein data bank. Still, I am left a bit dizzy at the sheer number of sites. In fact, one fragment (4-bromopyrazole) bound to all of the 16 sites! What are we to make of this – is this a privileged fragment or a promiscuous binder? And as for the sites with no known functional activity, are these useful? What do you think?

Fragment Events in 2013 and 2014

If you missed Fragments 2013 and the Eighth Annual FBDD there are still a few more fragment events this year, and although we're not quite into summer it's not too early to start marking your calendar for 2014!

2013

June 19-21: CHI’s Thirteenth Annual Structure-Based Drug Design will be held in Boston, with several talks on FBLD.

September 3-5: LibPubMedia Conferences is organizing DrugDesign2013 in Oxford, UK, with a focus on fragment- and ligand-based drug design.

September 23: Teddy and I will be teaching a three-hour short course on FBLD in Boston as part of CHI’s 11th Annual Discovery on Target. 

2014

April 23-25: CHI’s Ninth Annual Fragment-Based Drug Discovery will be held in San Diego. You can read impressions of this year's meeting here and here, last year's meeting here, the 2011 meeting here, and 2010 here.

September 21-24: Finally, FBLD 2014 will be held in Basel, Switzerland. This marks the fifth in an illustrious series of conferences, the last of which was in San Francisco in 2012. I believe this will also be the first major dedicated fragment conference in continental Europe. You can read impressions of FBLD 2010 and FBLD 2009.

Know of anything else? Add it to the comments or let us know!

Fragment Mixes for NMR

The number of fragments in a mixture for NMR screening has been the subject of a poll.  Some people have stated that they go much higher than 10 fragments (of course for 19F it is totally different).  What many people who are interested in doing ligand-observed NMR screening, it is daunting to figure out how to compute the mixes.  This paper addresses the issue.  Unlike many current approaches which use the spectra and then deconvolute them, this approach encodes the spectra into "fingerprints" and uses a Monte Carlo algorithm to minimize signal overlap.  The paper itself delves deeply into a lot of computer-ese gobbledygook (e.g. "the knapsack problem, one of the typical, non-deterministic polynomial time (NP-complete) problems") that I don't find interesting at all.  What I do find interesting is that they are targeting mixtures of 5 fragments. Other than that, they go into serious detail about their algorithm and what version was best.  They worked with 342 fragments from their in-house library.

However, after doing the initial POC on these they did not have a library big enough to test for scalability so they generated some virtual libraries: 500, 1000, 3000, and 5000 fragments.  Similar to discussion going on elsewhere, they clustered their fragments as strongly aromatic, strongly aliphatic, or balanced shown here.  As would be expected, library size and peak distribution did not affect the algorithm, but number of fragments per mixture did. As shown here, for the optimized libraries there is less overlap as you increase the number of fragments per mix (for 5 fragments it was ~0% to about 10-20% for 8-10 fragments).  This is a potentially huge increase in efficiency, simply increasing the number of compounds per mix from 5 (our poll found 5-7 to be the median number in mixes) to 10 would half the number of spectra that need to be acquired; hence lowering the potential cost to companies (especially if they are outsourcing (shameless self-promotion)). 

I have spoken to the authors and while, unlike the Beatson, their tool will not be available online, it is being incorporated into an upcoming release of Mnova's software. [Full disclosure: I have a business relationship with Mestrelab.]  Of the other software available, I believe only AMIX (Bruker) has built in screening tools, but I am not sure entirely as I have never used AMIX.  NMRpipe would be the one to be most likely to also have such tools, but their availability would be based upon the kindness of strangers.  Frankie D (Mr. NMRPipe) is at Agilent (nee Varian) now, so maybe vNMRJ will become more utile.  That last major software package from ACDLabs is not geared to this kind of work AFAIK. Additionally, this approach of course is just as "easily" applied to 19F, which could mean a mean increase of compounds from 10-15 to 25-30 routinely. 

I of course will update this if information on other software becomes available in the comments or via email.

[UPDATE #1: Ben Davis (Vernalis) pointed out CCPN has tools for this.  
Anna Vulpetti (Novartis) points out that python scripts for 19F have been published.
Arvin Moser (ACD) points out that ACD does offer screening tools.]

30% of all Posts...

NOTE: Blogger blew up my post when I published it.  I have fixed what I can.  Blogger keeps on blowing up this post after I edit it.   I have removed what I think may have been causing some of the problems.  This post should be considered in "Wiki-ese" as a fragment.   Thankfully, the summary was unaffected. 

As I recently said, GPCRs are a theme around here, so this post will talk about work published last year by the folks at ZoBio and Heptares.  [In terms of full disclosure, I had a business relationship with ZoBio until recently.]  This work is also on STaRs, the stabilized GPCRs developed by Heptares.  I noted my concerns with this approach here.  These two papers focus on A2A GPCRs, while previous posts here were on A1A, A3A, and B1A GPCRs.

In the ACS Chemical Biology paper, the authors are using TINS to screen an antagonist-stabilized A2AR StAR.   The immobilized protein showed a ~50% greater retention in activity after 5 days at 4C compared to the native protein in membranes (60% vs. 30% binding competency).  So, immobilized stabilized protein is more stable than non-immobilized, non-stabilized protein.  They then took a subset (531 compounds) of the ZoBio fragment library picked for maximal chemical and shape diversity and screened using OmpA as the reference protein. As shown in the bucketing below the vast majority of the compounds cluster around a T/R of 1.  This indicates that they have a slight preference for the target or the reference. The used a T/R cutoff of less than 0 e.g

. aromatic and aliphatics.  Additionally, the use of the logarithmic plot for the bucketing obscures the spread around T/R=1.  I have never seen a discussion from the creators of TINS discussing the relative error of the method and how to best evaluate the screening data.  In this case, they chose a 0.7 cutoff because there appears to be a discontinuity in the data there.
They followed up on these (see Table 1 in the paper) as orthosteric hits by observing if they can inhibit binding of an inverse orthosteric agonist in a radiolabeled assay using WT A2AR in HEK membranes; five fragments inhibited binding by >30% at 500uM (see below). 

This data in conjunction with the TINS data shows the compounds bind reversibly with a 1:1 stoichiometry.  These fragments also inhibited A1AR, which would not be unexpected for such small molecules.  However, 3 of these compounds have poor LEAN values >0.3.  This is particularly poor for GPCR targeting compounds.

Four additional fragments either one or the other of the inverse agonist or agonist used.  The two most potent AM appear to have some subtype specificity (A2AR over A1AR).  When they tried to test these compounds in a cell-based assay, toxicity was observed so no data could be collected.

In summary, the authors show that TINS is productive in finding fragments that bind to GPCRs.  However, they have to rely on standard biochemical assays for follow up.  It would have been nice to see at least one other method used to verify the active fragments, like SPR.  What I really like is that they did the biochemical assays against WT, which does not necessarily alleviate my concerns about screening against a mutant.  I would have really liked to see a comparison of the biochemical data in the STaR and WT.

So, while people say 30% of marketed drugs target GPCRs, I can assure you 30% of all of our posts are not about GPCRs.

Reversibly covalent fragments vs kinases

A big problem with small fragments is that they usually have low affinities for their targets; there is only so much binding energy you can pack into a dozen or so heavy atoms. Indeed, it wasn’t until the rise of sensitive biophysical methods such as NMR that fragment-based lead discovery really took off. But what if you could increase the affinity of fragments themselves?

One way to increase affinity is by introducing a covalent bond between the fragment and the protein: an irreversible covalent bond will, by definition, keep a fragment from ever dissociating from the protein. However, with this type of interaction, it may be difficult to distinguish between fragments with different inherent binding energies: iodoacetamide will alkylate every exposed cysteine residue, even though acetamide itself may have no inherent binding affinity. What you really need is a reversible covalent bond: something just strong enough to improve the affinity for the target, but still allow you to define structure-activity relationships among different fragments. This is the principle behind Tethering, which relies on (reversible) disulfide bonds between fragments and the amino acid cysteine. In a recent communication in J. Am. Chem. Soc., Jack Taunton and coworkers at UCSF apply a different chemistry to discover potent and selective kinase inhibitors.

Among the 518 human protein kinases, there are many non-conserved cysteine residues. Indeed, several advanced clinical candidates target a cysteine found just outside the ATP-binding site of certain kinases. These candidates are potent molecules in their own right, with irreversible covalent “warheads” attached to permanently knock out the kinases.

The UCSF researchers instead started with simple fragments (molecular weights between 96 and 250 Da) found in non-covalent kinase inhibitors. Each fragment was derivatized with a cyanoacrylamide moiety that could form a reversible covalent bond with cysteine residues, and these were screened against three of the eleven kinases that contain a cysteine residue at a certain spot within the active site. Remarkably, all showed activity against at least one of the kinases, though there were very different selectivities. Mutation of the targeted cysteine residue dramatically reduced affinity in all but one case, as did removal of the cyanoacrylamide.

Crystal structures of two fragments bound to the C-terminal domain of the kinase RSK2 were determined. In the case of compound 1, the indazole made the expected interactions to the so-called hinge region of the kinase. Interestingly, though, in the case of fragment 8, the azaindole moiety does not bind to the hinge. Instead, the ketone moiety serves as a hydrogen bond acceptor. Overlaying the two fragments suggested that adding an aromatic substituent to the indazole could improve affinity, a hypothesis that was nicely validated by compound 11. Addition of another small moiety gave compound 12, with improved selectivity over the kinases NEK2 and PLK1.

Compound 12 was tested against a panel of 26 kinases, 12 of which have active-site cysteine residues, and was found to be selective for RSK2 against all but NEK2 and PLK1 (and even then, the compound was more than 40-fold selective for RSK2). Crystallography confirmed the binding mode, complete with covalent bond to the cysteine residue, and mass-spectrometry of the denatured protein confirmed that the covalent bond is reversible.

One of the attractive features of the cyanoacrylamides is that they are quite stable, and in fact compound 12 showed respectable cell-based activity against RSK2 as well as the closely related C-terminal domain of the kinase MSK1, for which no inhibitors had previously been reported.

All in all this is a nice approach that should be broadly applicable not just to kinases but to a wide variety of targets. At least some of this technology has been licensed to Principia Biopharma, so it will be fun to watch this story progress.

Fragments in Living Cells

GPCRs seem to be popular around here lately.  In this paper, a group of researchers in the UK developed a report-fluorescent assay to be used on living cells to screen fragments.  Recent advances have been made in structural studies of fragments (see Receptos and Heptares) and with this new assay, the entire suite of experiments for prosecuting FBHG exist for GPCRs.  As the authors point out, and I mentioned here, mucking around with GPCRs with things like detergent stabilization removes ancillary proteins which can provide allosteric interactions. As the authors state:

It is now acknowledged that GPCRs can adopt multiple active conformations as a consequence of protein-protein interactions that can lead to the activation or attenuation of different signaling pathways.  Furthermore, different agonists appear able to bias signaling in favor of a particular downstream pathway, including those that do not involve heterotrimeric G proteins . It is also clear that the binding affinity of antagonists can vary depending on the signaling pathway and agonist that is being studied. These data suggest that intracellular signaling proteins can elicit marked allosteric influences on the binding of both agonists and antagonists to a particular GPCR and as a consequence the cellular context in which binding affinities are measured will have a major impact on drug screening strategies. [Emphasis mine].

To build their assay, they used an existing fluorescent existing amine congener that was commercially available and ensured that it had competitive, antagonistic properties against A3AR (Adenosine-A3 receptor).  It exhibited many desirable characteristics for a high-content screening: including high affinity and slow off-rate.  It was able to quantify agonist displacement from A3AR.  

To test for the ability to detect weak binders, they deconstructed the high affinity A3AR antagonist (1) into its component fragments (2-7) [the affinity values are pKi].  They then acquired a 248 subset of the Maybridge fragment library that was Voldemort Rule compliant.

  They found 38 confirmed actives, with the compounds shown below as the top 6:

To me, the interesting part of this paper is NOT the subsequent SAR and the novel compounds that resulted, but instead the fact that there is now an assay for A3-and A1AR that can reliably detect fragments and support SAR studies.  Of course, one major caveat here is something that I learned from a venerable GPCR chemist at Lilly: the natural ligands of GPCRs are fragments, so of course fragment screening works.  I think most people who work in fragments would be very happy with 1 microM actives from a screen; I think most GPCR chemists would not.  I would also imagine that ligand efficiency is even more important when working in this class of compounds. 

Fragment-based chemogenomics

An article with an intriguing title appeared recently in Drug Discovery Today: Small and colorful stones make beautiful mosaics: fragment-based chemogenomics. Iwan de Esch and colleagues at VU University Amsterdam and IOTA Pharmaceuticals define chemogenomics as:

The discovery of new connections between chemical and biological space leading to the discovery of new targets and biologically active molecules.

Thus, “fragment-based chemogenomics” is:

An approach to accurately characterize protein-ligand binding sites by interrogating protein families with libraries of small fragment-like molecules.

Like the “small, colorful stones” (or tesserae, though presumably not quantum) in a mosaic, fragments can be used to build up an understanding of protein-ligand interactions.

The authors start by constructing a fragment library consisting of 1010 compounds, most of which follow the rule of 3 (or, as Teddy would have it, the Voldemort Rule). An upper limit of 22 non-hydrogen atoms was used, and it looks like the lower limit was 7 atoms (vote on your own lower limit in the box at the right!), with a mean molecular weight of 211 Da. Most of the fragments were originally made as synthetic intermediates, but 117 were purchased specifically to add diversity to the library.

Having assembled the library, the authors then screened it against six targets: 4 GPCRs, 1 ion channel, and 1 kinase. Most of the assays involved displacement of a radioligand; hits were found against all of the targets, with hit rates ranging between 1 and 10%. Perhaps not surprisingly, different proteins preferred fragments with different physicochemical properties: histamine receptors selected polar, positively charged fragments (like histamine itself), while the kinase preferred rigid, hydrophobic, neutral fragments.

Although none of the fragments hit all six targets, a good proportion bound several (up to four). These tended to be larger than average, though no more hydrophobic, in contrast to results from other studies (see here and here).

Some of the most interesting results describe activity “cliffs,” ways to classify SAR observations for the GPCRs. An affinity cliff consists of two closely related fragments, one of which is active, the other of which is not, while a selectivity cliff consists of two fragments, one of which is active for a set of proteins, while the other is selective for one or more. The researchers show several examples where small changes – the introduction of a single heavy atom, the contraction of a ring, or the reversal of an amide bond – ablates activity.

Although crystallographically-enabled fragment optimization is now possible for GPCRs, the activity-guided SAR described here should be accessible to more researchers working on a wider range of proteins, and should prove powerful for tackling targets where structures are still elusive.

Nature recently declared that “‘omics bashing is in fashion,” but I do think there is something here. Whether or not it deserves its own ‘ome is open to debate, so feel free to weigh in!

Fragments vs GPCRs

G protein-coupled receptors, or GPCRs, have been one of the most fruitful areas of drug discovery. Humans have 390 of them throughout the body (plus many more in the nose, where they are essential for smelling), and almost a quarter of new drugs approved in the past decade target GPCRs. Despite these successes, there are plenty of “difficult” GPCRs that have resisted drug-discovery efforts. Since GPCRs are membrane-bound proteins, crystallography has until very recently been all but impossible, making structure-based design and fragment approaches correspondingly more difficult. In a recent issue of J. Med. Chem., John Christopher and colleagues at Heptares Therapeutics describe their success against one member of this target class (see also here for In The Pipeline’s coverage).

Fragments have been screened against other types of membrane proteins using surface plasmon resonance (SPR) and TINS, but one of the particular challenges of GPCRs is that they are generally quite unstable and conformationally flexible. Heptares solves this problem by making a small number of targeted mutations to increase receptor stability and lock the conformation. In this case, the researchers targeted the human β₁-adrenergic receptor (β₁AR); both agonists and antagonists of β-adrenergic receptors are approved drugs.

Approximately 650 fragments were screened using SPR against the stabilized human β₁AR as well as another GPCR, the adenosine A_2A receptor. Selective hits were identified against both targets; among the β₁AR-selective hits were compounds 7 and 8, both with impressive affinities and ligand efficiencies.

Co-crystal structures of various ligands bound to turkey β₁AR (which is identical to its human counterpart in the ligand binding domain) had previously been solved, and molecular modeling of the fragment hits led to the purchase of a set of analogs, which were then tested in a radioligand membrane binding assay. Happily, compounds 19 and 20 both bound with improved affinity over the parent fragments. Crystal structures of these new molecules in complex with turkey β₁AR were also determined, revealing that they do not completely fill the ligand-binding pocket, and suggesting additional modifications to further improve potency and alter their pharmacology.

There are still many unanswered questions. Phenylpiperazines such as these are unusual ligands for β-adrenergic receptors, but they are known to bind other GPCRs, so selectivity will need to be investigated thoroughly. Also, the researchers don’t say whether their molecules are agonists or antagonists, though they suggest the later. Some of this work was publicly presented as early as 2010, so presumably there is plenty more data beyond what’s reported here.

All that said, this is a nice milestone in fragment-based ligand discovery, and it will be fun to see how crystal structures play a role in understanding (and drugging!) this important class of targets.

Fragment merging for renin

Renin, an aspartic protease involved in regulating blood pressure, is one of those drug targets that has been around forever; it took decades before the first direct inhibitor was approved. In a recent paper in J. Med. Chem., Daniel Baeschlin and colleagues at Novartis (where the approved drug was discovered) describe how they’ve taken a fragment-merging approach to look for additional inhibitors of this target.

The researchers started by assembling a small (113 compound) fragment library designed to target aspartic proteases. This was screened against renin by NMR, resulting in hits such as compound 3. Although these were too weak to yield dissociation constants or IC₅₀ values by NMR or biochemical screens, the researchers were able to obtain crystal structures of at least two of these fragments bound to renin, including compound 3. Interestingly, although the amino alcohol moiety of this compound was designed to target the catalytic aspartic acids, this turned out not to be the case. Instead the binding appears to be largely driven by hydrophobic contacts between the tricyclic moiety of the fragment and the so-called S₃-S₁ pocket of the protein.

Along with the fragment effort, the researchers also undertook an HTS screen, resulting in the discovery of compound 5, which itself had come from a 950-compound library targeted towards aspartic proteases. Crystallography revealed that this molecule binds with the diphenylmethane moiety in a similar position as the tricycle of fragment 3, and indeed when a rigid tricyclic framework was grafted onto compound 5 the resulting compound 9 showed a satisfying boost in potency. Further optimization led to a pure enantiomer of compound 12 with low nanomolar potency, good selectivity, moderate oral bioavailability and efficacy in rats, though it did also show some time-dependent CYP3A4 inhibition.

This is really a structure-based design paper, and there is obviously much more detail than would be appropriate here. What caught my eye is that it is a nice example of fragment-assisted drug discovery, in which fragment information is used as one aspect of an overall lead discovery program. In this case a cynic could argue that the only bit that came from the fragment was the tricylic motif. However, given the limitless number of analogs that could be made, such information can be both unexpected and valuable.

Poll: how many atoms are too few?

Last year we asked how large a fragment you would include in your library, but the related question, how low will you go, is also interesting (see poll to right).

Azaindole, with 9 non-hydrogen atoms, has been the starting point for more than one clinical compound, and 5-atom acetohydroxamic acid also figures rather prominently in fragment history, but would you include something this small? How seriously should we take the Rule of 1?

Poll Results - Number of Fragments in the Screen

The poll is closed!  So, the question was

I think it is fair to say that the average number of fragments per fragment mixture for NMR pooling is 6.  I typically say five is the goal and if you can squeeze in more good on ya!  I don't think the limitation is chemical shifts, I think it is solubility.  I would love to hear what others think is the limitation.

Eight Annual Fragment-Based Drug Discovery Meeting (part 2)

The last major fragment event of 2013, CHI’s FBDD, wrapped up earlier this week in San Diego. As with last year this summary is not meant to be comprehensive (and you can also read Teddy’s impressions here.)

The FBDD track is just one of six within the CHI Drug Discovery Chemistry Conference. One indication of the success of the field is its appearance in several of the other tracks: attendees were likely to hear about fragments without going to the FBDD track at all.

In the GPCR track, Robert Cooke from Heptares discussed the application of fragments to the β1-adrenergic receptor (see also here). In the kinase track, Hongtao Zhao presented in silico fragment work (see also here). And in the protein-protein interaction track, David Fry from Roche described a deconstruction of the p53-HDM2 inhibitor RG7112 into its component fragments to see whether the molecule could have been identified from FBLD. RG7112 consists of a central core with four appendages, and although the mono-substituted core was too weak to detect, some of the cores with two substituents could be identified and bound to the protein in the same manner as the parent compound. However, these did tend to be super-sized fragments, with molecular weights in the 300-350 Da range.

Protein-protein interactions were also a theme of Richard Taylor, from the company UCB. They built a sizable fragment library of about 23,000 (mostly commercial) compounds designed to cover molecular frameworks found in known drugs. UCB has invested heavily in SPR technology, and with a stable of four Biacore 3000 instruments could rapidly screen this entire library against a dozen protein-protein interaction targets. Not surprisingly, given the difficulty of this target class, the hit rate was much lower than in conventional fragment screens, averaging just about 1%. What was interesting is that only 964 fragments hit any target – at less than 5%, this is much lower than the roughly 33% hit rate seen in other fragment libraries. Most of these fragments were reasonably specific, though; 908 hit ≤ 8 targets. It will be interesting to see whether anything can be learned about “privileged” protein-protein interaction fragments from this set.

Of course, extracting general trends from collections of fragments is not necessarily straightforward. Teddy has already brought up the difficulties of describing molecular shape; as he pointed out in his presentation, Fsp³ is not the best measure of “three-dimensionality” for several reasons. For example, even toluene has an Fsp³ = 0.14, and while Pete Kenny correctly points out that aromatic molecules do have volume, most chemists would think of this as a very “flat” compound. Principal moment of inertia (PMI) is better, but is harder to calculate. Happily, as Justin Bower described in his presentation, the Beatson Institute is allowing other researchers to use their 3DFIT software to calculate PMI and other properties.

One of the criticisms of 3D fragments is that, as Rod Hubbard pointed out, they can be a “pain in the neck” for chemistry. One solution that researchers at Vernalis took was to do analog work on a simpler molecule, then scaffold-hop back to the original fragment once the SAR was sufficiently understood to justify investment in more challenging chemistry.

Finally, the question of what to do with fragment hits that don’t reproduce in different assays was the topic of at least one breakout discussion and was also extensively discussed by Peter Kutchukian, who presented an analysis of 134 fragment screens using a variety of techniques against 34 different targets at Novartis. Some of this was presented at FBLD 2012, but one interesting finding was that hits from biophysical screens (such as SPR, NMR, or DSF) tended to cluster separately from hits in biochemical assays. Given the number of ways to find fragments, pursuing hits that confirm in both a biochemical and a biophysical method may help to weed out artifacts, though at the risk of increasing false negatives.

Feel free to chime in with your thoughts and impressions, whether or not you were there. And if you are kicking yourself for not attending, next year’s meeting is scheduled to return to San Diego from April 22-25.

What's the Fire behind the Smoke?

Dan and I are here at the CHI FBDD conference, with of course other luminaries.  I am not going to get into details about all the talks, Dan does that much better than I anyways.  I wanted to set up some future blog posts by sharing what I think are some trends.

1.  3D fragments are real and very useful.  However, fSP3 is a horrible way to measure 3D-arity.  Principal Moment of Inertia (PMI) is a much better way to determine this.  This was mentioned in a brilliant talk (watered down from full-fledged rant) by me.  But much better explained by Justin Bower of the Beatson Institute. 

2.  Solubility, Solubility, and Solubility.  Experimentally determined solubility is a much better indicator of fragment quality (for a library) than cLogP, for example. 

3.  The Voldemort Rule (or the rule that shall not be named).  Arbitrary "rules" created as a marketing tool have no place in a discipline that has over a decade of results and empirical evidence.  While Dan is not ready to put a stake in its heart, I am.  Rod Hubbard is.  Who else wants to join Zartler-Dore's Army?

Look for more updates, thoughts, comments, and general frivolity as follow up to the conference.

Fragments in the clinic: AZD5363

As illustrated earlier this year, kinases have been a fertile field for fragments. In a recent issue of J. Med. Chem., Jason Kettle and colleagues at AstraZeneca describe the discovery of AZD5363, a protein kinase B (PKB) inhibitor currently in multiple phase I clinical trials for solid tumors.

The story actually starts a decade ago, with a collaboration between Astex and the Institute of Cancer Research. The two organizations were interested in PKB (also known as Akt1), which has a central role in the PI3K signaling cascade. Virtual, biochemical, and crystallographic screens identified small fragments such as substituted pyrazoles and 7-azaindole (which astute readers will recognize as the starting point for vemurafenib) that bind to the so-called hinge region of PKB. Structure-guided fragment-growing ultimately led to compound 2.

This compound, while potent against PKB, was unselective against the related protein kinase A (PKA), so further crystallographically-enabled medicinal chemistry led to CCT128930, with 30-fold selectivity against PKA. This compound had limited oral bioavailability, so further optimization led to compound 3.

In 2005, Astex partnered this program with AstraZeneca, which is presumably where the current paper picks up. Although compound 3 had good pharmacokinetics and was selective against PKA, it inhibited the kinase ROCK2, which regulates blood pressure; it was also a modest hERG inhibitor. Extensive SAR explorations around the hinge-binding element, the amine, and the aromatic group were not productive, but substitution off the benzylic position was tolerated. Adding a basic substituent dramatically reduced hERG binding, but at the cost of oral bioavailability. However, adding a variety of neutral, polar substituents led ultimately to AZD5363, which has no detectable hERG inhibition, good selectivity against ROCK2, and improved solubility and cell activity.

This paper nicely illustrates some of the challenges in drug discovery: high-affinity molecules were obtained relatively quickly, but these still required a huge amount of effort to achieve selectivity, oral bioavailability, and other properties. Indeed, only three heavy atoms differentiate compound 3 from AZD5363, but it took a heroic effort to get there.

Finally, it is worth noting that this research was done at Alderley Park, which attendees of Fragments 2009 will remember fondly. Sadly, AstraZeneca has announced that they will be closing this site. There are many very talented scientists there, and Practical Fragments wishes all of them the best of luck.

Upcoming Fragment Conference

Next week is the big fragment conference in the US (after a successful european one).  There are lots of interesting things going on.  Dan and I will be co-teaching a short course on fragments on Monday afternoon.  The conference then starts with Protein-Protein Interactions Optimizing Hit to Leads (with the best talk title: To Affinity and Beyond: From Screened Fragments To Optimized Leads With SPR and ITC), followed by Library and computational talk in the afternoon.  Day 2 starts with the ever popular breakfast roundtables, led by Dan, Rod Hubbard, Kevin Burgess, and myself.  The last session of the day will be the case studies.  There are a variety of events happening during the conference (and there are other tracks simultaneously) that Dan and I will update you on right here.  So, if you are going to be in SD we hope to see you. 

Fast, competitive thermodynamic data

The importance of thermodynamics in drug discovery is often debated, with advocates arguing that ligands binding primarily through enthalpic interactions may be superior to those whose binding is driven by entropy. Some publications support these claims, though the data are rather sparse. This is at least partly because thermodynamic measurements are typically done using isothermal titration calorimetry (ITC), which consumes sizable amounts of protein and is not exactly high throughput. In a recent issue of J. Med. Chem., Jose Caaveiro, Kouhei Tsumoto, and their colleagues at the University of Tokyo and GE Healthcare Japan describe a new screening method that could speed things up.

The approach, which is essentially a competition screen, is called single-injection thermal extinction, or SITE. In a conventional calorimetric assay, a ligand is added to the protein in a calorimeter, and the measured heat change upon binding is used to calculate enthalpy. In SITE, a protein is first incubated with the fragment to be tested in the calorimeter, and a known positive control binder is then added. If the fragment binds at the same site as the known binder, addition of the positive control will cause a smaller change in temperature, and this difference should reflect the enthalpy of binding of the fragment compared to the positive control.

To validate the system, the researchers tested the steroid-processing enzyme ketosteroid isomerase (KSI). They first ran an SPR screen of 2000 fragments at 0.2 mM. To weed out false positives, they used two different types of SPR chips, and looked closely at the binding curves; the paper has a nice summary of some of the pathologies that can occur. A total of 129 hits were identified, of which 44 were then tested in SITE and characterized more fully with SPR.

Interestingly, most of the most potent compounds – as assessed by SPR – also gave the strongest signals in SITE, suggesting that these compounds are binding largely through enthalpic interactions. A few of the best compounds were further characterized by conventional ITC, and these did in fact have better enthalpic efficiencies than the positive control (they had better ligand efficiencies too).

It would be interesting to know how SITE behaves with allosteric inhibitors or ligands that bind to different sites on the protein. And of course, the jury is still out on whether enthalpic binders really do make superior leads, and even whether it is possible to use thermodynamics prospectively in lead optimization. But with a 9-fold drop in protein consumption and an increase in speed, this technique may make it easier to get the data to answer these questions.

Everything that can go wrong

Anyone who ventures into fragment screening will encounter problems. Maybe your compound isn’t what you think it is. Or it’s contaminated with something that mucks up your assay. Or it degrades – possibly during the experiment.

Maybe you haven’t removed the PAINS from your screening collection, and you get results like this. Or perhaps you forgot to add detergent, and your hits are all aggregators.

Even if all your compounds are pure and beautiful little three-dimensional works of molecular art, things can go wrong in screening. As part of our occasional “Getting misled” series, we’ve highlighted problems that can arise in NMR and crystallography (here, here, and here). But that doesn’t mean that practitioners of SPR, ITC, thermal shifts, or other assays can relax.

As the old saying goes, forewarned is forearmed, so Ben Davis (of Vernalis) and I have put together a brief review of the many things that can go wrong in fragment-based lead discovery campaigns; thanks also to readers who shared their horror stories. It has just been published as a Digest in Bioorg. Med. Chem. Lett. To maximize readership, Elsevier has agreed to make this open access, at least through 2013. We hope this will be useful for folks starting out in the field, as well as for other researchers, journal editors, and reviewers trying to assess and build on the literature.

Of course, I’m sure we’ve left out lots of things that can go wrong, so feel free to comment here!

One metric to rule them all

Most readers are familiar with ligand efficiency and LLE, and many folks are using LLE_AT as well. However, as we’ve previously noted, a whole cottage industry has been busily devising new metrics, and it sometimes becomes hard to keep them all straight.

To help bring some order to the chaos, researchers at Mordor State College (home of the Sauron Atoms) have developed what they call the Wholly Transcendent Function, or WTF. This metric takes into account binding affinity, number of heavy atoms, ClogP, and molecular topology, but it also includes information about metabolic stability, toxicity, blood-brain barrier penetration, hERG binding, CYP inhibition, and potential for becoming a blockbuster, all encoded into a single number between 0 and 1.

Unfortunately, collecting all the data necessary to calculate WTF is quite a quest, but the gaps are readily filled by guesswork, producing a metric that can banish unnecessary complexity. Moreover, the researchers are confident that improved computational estimates will one day make even more accurate predictions readily available.

Leave Them Asking for More

ret·ro·spec·tive  (rtr-spktv) adj.

1. Looking back on, contemplating, or directed to the past.

2. Looking or directed backward.

3. Applying to or influencing the past; retroactive.

I would add: 4. Looking back on the past, to influence the future.

In this vein, a recent paper by Ferenczy and Keserű in J. Med Chem looks back on hit-lead optimizations derived from fragment starting points.  In this very interesting paper, they look at 145 fragment programs and evaluate the properties of the original hit and then again as it progresses into the lead.  Of the 145 programs, these were aimed at 83 proteins of which 76 are enzymes, 6 are receptors, and 1 is an ion channel.  These programs evolved into leads, tools, and clinical candidates.  The authors set out to answer three questions: 1. do fragments eliminate the risk of property inflation, 2. how do ligand efficiency metrics support fragment optimizations, and 3. what is the impact of detection method, optimization strategy, and company size on the optimization.

Table 2 shows the median the calculated properties for the hits and optimized compounds.  The pIC50 improved by roughly three orders of magnitude, but ligand efficiency (LE) stayed roughly the same.  Log P increased but SILE did not.  SILE was a metric I was not familiar with and is calculated by pIC50/(HAC)^0.3.  SILE is a size-independent metric of ligand efficiency.  I won't attempt to reproduce all the graphs they generated; get the paper.  Some interesting data points: the median fragment hit had 15 heavy atoms (see related poll here), the median size of leads is 28 heavy atoms, and good fraction of hit-lead pairs changed less than 5 heavy atoms (which of course is well known from here). So, what is the answer for their first question?  If you are looking at something like LE, then these hit-lead pairs maintain the efficiency.  If you are looking at logP, then the answer is no, the hit-lead pairs get greasier.  I would really like to see more granularity here (see What's Missing below).  Their major comment here is that SILE and LELP (work by the author's previously reviewed by Dan) are the two best metrics to monitor as hit-lead optimization is underway.  Increases in both metrics correlate with increases in FBDD programs. 

 

They then looked at the  screening method.  The breakout of primary screening (in their definition the first one listed when multiple methods were used) was 38% biochemical, 25% NMR, 18% X-ray, and 11% virtual.  This is an interesting contrast to these results; SPR is not the dominant screening technique (8% tied with MS).  So, does this mean, >40% of practitioners are using SPR, but as a secondary screen?

Table 4 then does a pairwise comparison of the metrics based upon primary screen origin of the fragment (see What's Missing below).  Biochemical screens yield the most potent hits (4.75) while NMR (3.53) has the least potent.  X-ray has the smallest hits( 13 heavy atoms) while virtual screening the largest (17 HA).  I don't think any of this is surprising; the authors point out that hit properties exhibit a significant dependence on the method used. It is noted that optimization tends to diminish differences in hit properties.  Again, I think this is not surprising; thermodynamics and medchem are all the same no matter how big or small the molecule.  They do point out that biochemical based hits preserve their advantage after optimization, primarily relative to NMR.  They posit that the difference is that more potent compounds need less "stuff" to become sufficiently potent, and thus have a better mix of medchem for potency and other property optimization.  Weaker starting points need more bulk, more atoms, to become equipotent with biochemical starting points, they suggest.  Lastly, they show that structural-based optimization efforts are better than those without structural information.  The structural information comes from X-ray primarily (52%) and NMR (10%).  Interestingly, they reference this poll from the blog on whether you need structural information to prosecute fragments.  [As an aside, since I wrote that blog post and it was reference in a paper, do I get to add it to my resume?]

Finally, they break the originating labs into three categories: academic (18%), small/medium enterprises (SME)( 37%), and Big Pharma (45%).  The SME results are superior to those achieved by the academics and Big Pharma.  Their explanation is that SME's tend to be more platform focused and predominantly ensure "structural" enablement of targets.  75% (40/53) optimizations at SMEs used protein structural information, while "only" 62% of those at Big Pharma did.  They do not rule out the differences in target selection at those two different groups of companies.  I would like to propose an alternative hypothesis (tongue-not-entirely-in-cheek): Big Pharma has "old crusty" chemists who don't understand fragments and thus just glom on hydrophobic stuff to increase potency because "that's how we always do it" while SME have innovative chemists.  And of course academia is just making tool compounds and crap.

One thing that I would like to emphasize is that Ro3, metrics, and so on should not be used as hard cutoffs.  As shown in Figure 11, even compounds that are outside the "preferred" space can reach the clinic.  The best way to view them is akin to the Pirate Code; they are more guidelines than rules. 

 

What is Missing?  The supplemental information (which the authors are willing to kindly share) does not break out the targets into specific classes.  However, they do list each target, so it should be easy to add this a data column.  More importantly, they do not break out hit-lead pairs into those that were optimized for use as tools, clinical candidates, and leads (and which of the leads died).  Tools are never supposed to look like leads (but you are lucky if they do), so their inclusion here can be biasing the results.  Although, it is likely that the of the 145 not very many of the examples are strictly tools; it would be nice to know though.   

I am struck by the information denseness of Table 4 and wish that instead of pairwise comparisons, they had instead simply list the hit-lead metrics for each methodology.  I think there is gold to be mined in Table 4 and just like real gold is hard to find.

 

I have not addressed every single point made by the authors.  I, for one, am hoping that they will continue their analyses (especially with an eye to some of What is Missing).  I hope that there will be a significant amount of discussion around these points.  I will make sure we hit on this at the breakfast roundtable at the upcoming CHI FBDD event in SD (I even have the same pithy title as last year!)

Rad fragments

One of the selling points of FBLD is that it can find starting points against challenging targets such as protein-protein interactions. A major reason these targets are so tricky is that they often have large, flat interfaces with few pockets for small molecules to bind. An example is the interaction between the tumor suppressor BRCA2 and the recombinase RAD51, which is mediated in part by the phenyl ring of a phenylalanine residue – a very small moiety even by the standards of fragments. In a paper published recently in ChemBioChem, Marko Hyvönen and colleagues at the University of Cambridge describe how they’ve found fragments that bind to this site.

The researchers started with a microbial RAD51 homolog that had been humanized by mutagenesis; the human protein itself is unstable and difficult to work with. They performed a thermal screen with 1249 fragments. Thermal denaturation has been criticized for producing noisy data, and indeed, 96 fragments produced complex, uninterpretable results. However, the two best fragments both contained an indole core and were confirmed to bind by STD-NMR.

Competition experiments confirmed that these two fragments competed with a short peptide containing the critical phenylalanine, indicating that they bound at the desired spot. ITC revealed that they had dissociation constants around 2 mM. Their binding modes were also confirmed crystallographically.

The researchers then used one of these fragments as a probe in a round of STD-NMR experiments, in which they examined 42 fragments to see whether any of these could compete away the first fragment. This led to two new hits, both slightly more potent than the initial ones.

One of these new fragments was then used as a probe in another round of STD-NMR experiments with 120 fragments chosen as analogs or by in-silico screening. This led to four additional fragments, some of which had sub-millimolar affinities and good ligand efficiencies. All 6 of the new fragments from the two STD-NMR screens were characterized crystallographically and found to bind at the same site as the original indole fragments, though with some subtle differences that could be exploited for further elaboration.

This is a nice, thorough example of fragment discovery in academia. As the authors conclude:

Investment in a platform of orthogonal biophysical assays and screens is crucial for progression into a programme of medicinal chemistry. The elaboration of poorly validated hits not only has a high likelihood of failure, but without a variety of robust assays in place, the risk of being misled by badly behaving compounds increases.

Of course, these are still relatively weak fragments, but I’ve heard one of the authors speak at a conference in which he stated that they’ve been able to advance these to nanomolar leads with cell activity. Stay tuned!