Fragments vs SARS-CoV-2: the whole proteome

More than three years have passed since SARS-CoV-2 first entered human airways, and it looks set to stay despite the rapid development of remarkably protective vaccines. Drugs such as nirmatrelvir are effective, but as with any infectious agent we’ll need lots more to counteract inevitable mutational resistance. Practical Fragments has previously discussed virtual and experimental screens against individual SARS-CoV-2 viral proteins and RNA. In Angew. Chem., Harald Schwalbe at Goethe University Frankfurt and more than six dozen collaborators in ten countries describe (open access) the results of NMR screens against most of the viral proteome.

 

The researchers, all part of the COVID19-NMR project, used ligand-detected NMR methods to screen the DSI-PL library against 25 of the 28 viral proteins. As we’ve written previously, the DSI-PL library consists of 768 diverse fragments designed for rapid chemistry follow-up. Fragments were mostly screened in mixtures of twelve, with spectra visually inspected to identify hits for confirmation.

 

Fragments were classified as binders if they passed any of these four criteria: “chemical shift perturbations (CSPs) or severe line broadening, sign change in the waterLOGSY (wLOGSY), STD signal or significant decrease of signal intensity in a T₂-relaxation experiment.”

 

A total of 311 hits were identified, with between 2 and 154 hits per protein. In some cases multiple forms of the protein were screened. For example, three forms of the main protease (called by various groups nsp5, Mpro, and CLpro) were screened, yielding from 12 to 78 binders, only 8 of which were common to all three screens. One of the protein constructs forms the biologically relevant dimer (the others are monomers), and the researchers suggest this could account for the differences. True, but I suspect many of the “hits” against some of these proteins are artifacts or non-specific binders: researchers at Vernalis, for example, prioritize fragments that hit in two or three different NMR assays over those that hit in just one.

 

Crystal structures were available for 18 of the proteins screened, and these were computationally analyzed using FTMap to identify between one and three potential small-molecule binding hot spots on each protein. FTMap uses 16 very small probe molecules (such as benzene and urea) to interrogate the protein surface, and a comparison between the NMR hits with those from FTMap was comfortingly good. For example, a protein with a hot spot preferring benzene and urea also bound a fragment containing those moieties. While this by no means proves that the fragments are binding at a given hot spot, it is suggestive.

 

Not surprisingly, most of the fragments are weak binders: titration experiments revealed that five of ten tested had dissociation constants > 5 mM, though one came in at double-digit micromolar. This result is consistent with work last year that found that most of the crystallographic hits against Mpro were also weak binders, and also consistent with an independent NMR study of Mpro.

 

Despite these limitations, this campaign provides multiple starting points to develop chemical probes. Laudably, the chemical structures of all the DSI-PL library compounds and the targets hit by each are provided in the supporting information. Last week we highlighted how fragment hits against the SARS-CoV-2 Nsp3 macrodomain were advanced to sub-micromolar inhibitors. The Angew. Chem. work provides fragment starting points against two dozen more targets.