One of the problems faced by drug discovery scientists running a screening programme is the correlation of hits (or the lack of correlation) from one assay technique compared to another, run against the same drug target. Most often this occurs as part of screening cascade, when you move from one primary screening method to a secondary confirmation method and a vast majority of hits don’t repeat or have different ranked orders of potency. This can be for good reasons, and why the secondary technique is being used as a counter screen to remove false positives.
This process can also be used to validate a new screening methodology by comparing the correlation of actives from screening file when it is run with two differing techniques against the same target. This latter example was described in a recent publication
This paper has also been described on the Practical Fragments blog. Please read for a far more detailed description and many other interesting articles.
In this publication the authors from two separate groups ran a NMR and SPR fragment screen on the same target (HIV-1 integrase core domain) with the same fragment file (about 500 compounds). Once primary actives had been identified, the authors used X-ray crystallography as further evidence of binding. The results from this experiment showed that the NMR technique identified 62 primary hits of which 15 generated protein/fragment complexes. The SPR generated 16 primary hits which led to six clear protein/fragment complexes. When the fragments/crystal complexes were compared to one another there was no correlation between the different assay methods.
There were some differences between the two techniques which might account for the differences, for example the total amount of DMSO present varied in each technique, which could affect solubility of compounds, and the final screening concentration of the SPR (500µM) was less than that of the NMR (1mM) which would suggest some of the weaker compounds would be more difficult to find in the SPR. In addition to this, the conditions used for crystallisation are bespoke for each protein. This would favour a certain set of physiochemical properties which might assist the formation of a crystal complex for a certain sub-population of fragments. All these factors could affect the amount of hits identified for each method.
Another critical difference it seems between the two techniques is that the SPR method used a reference protein (AMA1) as control. Hits were identified from SPR, only if the binding response was twice the level with HIV-1 integrase compared to the AMA1. When comparing the NMR actives, 8 compounds which gave crystal structures did not show this fold binding difference in the SPR (even though they bound to the HIV-1 integrase and therefore were eliminated as potential hits from the SPR hit list.
When the 5 SPR hits which gave crystal structures were run as singletons with NMR, 4 of them gave a weak response and it was suggested that the pooling of the fragments (10 compounds per sample) made the identification of these compounds indistinguishable from the background signal.
So overall both methods did find active fragments, however importantly each method did not find all the active fragments from the collection.
The different methods of analysis and screening had an impact on the population of the hits that were found for each of the techniques. Maybe once a reasonable subset of the active fragments has been identified, the use of structural similarities searches could be of assistance to locating other fragments from the file, to ensure that as few as possible active fragments are not missed.
From this review it does not look like one method has any greater performance than another in identifying active fragments, however it seems care needs to be taken with the analysis.