Fragment-based design is one of my favorite approaches in drug discovery. It has everything from very simple conception to sophisticated data analysis. The most importantly, it works as it’s supposed to. So it’s always entertaining to find a paper from the very founders of FBDD, I mean Astex Pharmaceuticals and Harren Jhoti himself.
This new paper deals not so much with the actual drug discovery though. It’s an attempt to summarize the structural data that Astex collected over several campaigns. And it has some really interesting implications and afterthoughts.
When screening fragment libraries using X-ray crystallography, one often sees the molecules binding in ‘unexpected’ places, not where the native ligands bind. These observations bothered the authors quite a while, so they decided to take a closer look at those unexpected binding pockets.
First, they estimated that 67% of target proteins revealed cryptic binding sites. As authors suggest, it’s a lower bound due to limited diversity of chemotypes in their libraries and occlusion of potential binding sites by crystal contacts. But one should keep in mind that the errorbar on this number is quite large as the sample size was only 24 proteins (and this required quite a bit of work).
What about the properties of those unexpected pockets? Authors compared them with ‘expected’ ones by three parameters: i) evolutionary conservation, ii) conformational flexibility of side chains forming the site, and iii) surface polarity.
It turned out that all three parameters are not too different for orthosteric and allosteric sites but different from ‘general protein surface atoms’ (whatever the definition is). The authors put the highest emphasis on the first parameter, evolutionary conservation, because it allows to speculate on some unknown biological functions of new binding pockets. Even setting aside the philosophical argument about “conservation = function”, the result made my eyebrows rise. Because in drug discovery it became almost truism that targeting allosteric pockets has the advantage of lower conservation between related proteins. Hence, allosteric drugs should be more selective than orthosteric ones (kinases and GPCRs are good examples).
The study definitely lacks a cross-correlation between three parameters. I’d bet that there is strong relation between sequence conservation and flexibility of corresponding residues (or flexibility and surface polarity). So chosen parameters could be somewhat redundant, and this could strongly affect the main conclusion. Not to mention the bias from physico-chemical properties of the small-molecule fragments. Yes, rigorous data analysis is overwhelmingly complex.
To resolve all controversies, one could try to ‘reverse-engeneer’ the analysis and from estimated parameters to predict de novo some new allosteric sites. I’d be surprised if nobody tried predicting allosteric pockets before. Quick google search gave me this and this (online tool!). So I played with two PDB structures from the Astex paper and got the following results: for CDK2 (PDB: 5FP5) the picture was pretty similar to Astex’s X-ray data (one binding site was not covered by predicted allosteric pocket), for HSPA2 (PDB: 5FPE) three out of five fragment binding sites were missing.
All in all, the paper brings some new insights in the experimental validation of allosteric pockets. But I wouldn’t buy their main conservation argument. It needs more supporting data.
, and Harren Jhoti “Detection of secondary binding sites in proteins using fragment screening”, PNAS 2015 112 (52) 15910–15915