Objective

Cancer drug discovery programs aimed at blocking cancer cell proliferation fail too often. Three lines of evidence typically support selection of a target. First, that the target’s expression level is altered or the target is mutated in cancers. Second, that genetic manipulation of the target inhibits proliferation. Third, that a small molecule that affects the target produces the desired phenotype. Despite the importance of small molecule support for selection of a target, there is a problem- largely unrecognized- inherent in the use of small molecules to support a target’s validity: the kinds of compounds available prior to screening and optimization almost always have multiple effects, meaning that the level of support they provide can be quite weak. As an illustration of the issue, attention has recently returned to the Warburg effect- the fact that many cancer cells obtain most of their ATP from glycolysis even when oxygen is available- as a potential antiproliferative target. Until relatively recently, the only inhibitors were unspecific agents like 2-deoxyglucose (2-DG) and 3-bromopyruvate, which have multiple effects on glycolysis and also have effects apart from glycolysis inhibition that could play important roles in inhibiting growth. Recently screens have been conducted to identify blockers of specific glycolytic enzymes, but questions about theglycolysis inhibition in their effects on proliferation- when those have even been reported- remain. We surmised that it might be possible to provide robust small-molecule support for the hypothesis that glycolysis is a good antiproliferative target by examining the effects of a panel of structurally and mechanistically diverse glycolysis-inhibiting compounds rather than focusing on a single compound. We reasoned that any one compound that inhibits glycolysis could of course reduce proliferation via off-target effects, However, if inhibiting glycolysis is indeed a good way to prevent cell growth, then if we can identify a sufficient number of compounds that inhibit glycolysis and all are antiproliferative, it could provide strong support for the idea that effects on glycolysis are responsible for effects on proliferation. In contrast, if even one effective glycolysis inhibitor were to fail to block proliferation, it would falsify the mechanism. We screened the NCI’s Mechanistic Set III for glycolysis inhibitors using a novel multi-read plate reader assay that we developed using an intramolecular FRET sensor for ATP expressed stably in K562 cells, looking for metabolically-active compounds. Our results point to a novel way to provide small molecule support during target validation. A library should be identified that is likely to have both an appropriate number of antiproliferative compounds as well as a sufficient diversity of scaffolds that it will yield enough hits on the target of interest. It should be screened for effects on the target and for effects on proliferation. A threshold for considering hits effective should be established. If any compound that inhibits the target fails to inhibit proliferation the target is likely invalid. The level of support provided when all compounds that block the target effectively inhibit proliferation can be estimated using the binomial distribution to calculate the likelihood that the same number of antiproliferative compounds would be drawn by chance. We screened entire pathways with hit rates of ~ 1%. In a screen focused on a single target, it might be necessary to screen ~10,000 compounds to identify 5-10 active inhibitors, which, depending on the overall number of phenotypic positives in the library, would be sufficient to produce a reasonable degree of support for the target. This might be well worth the effort if it improved the target selection process and reduces the failure rate of drug discovery effort. The strategy could also be applied in to target validation for other phenotypes such as cellular differentiation or a

Methods

We surmised that it might be possible to provide robust small-molecule support for the hypothesis that glycolysis is a good antiproliferative target by examining the effects of a panel of structurally and mechanistically diverse glycolysis-inhibiting compounds rather than focusing on a single compound. We reasoned that any one compound that inhibits glycolysis could of course reduce proliferation via off-target effects, However, if inhibiting glycolysis is indeed a good way to prevent cell growth, then if we can identify a sufficient number of compounds that inhibit glycolysis and all are antiproliferative, it could provide strong support for the idea that effects on glycolysis are responsible for effects on proliferation. In contrast, if even one effective glycolysis inhibitor were to fail to block proliferation, it would falsify the mechanism. We screened the NCI’s Mechanistic Set III for glycolysis inhibitors using a novel multi-read plate reader assay that we developed using an intramolecular FRET sensor for ATP expressed stably in K562 cells, looking for metabolically-active compounds.

Conclusion

Our results point to a novel way to provide small molecule support during target validation. A library should be identified that is likely to have both an appropriate number of antiproliferative compounds as well as a sufficient diversity of scaffolds that it will yield enough hits on the target of interest. It should be screened for effects on the target and for effects on proliferation. A threshold for considering hits effective should be established. If any compound that inhibits the target fails to inhibit proliferation the target is likely invalid. The level of support provided when all compounds that block the target effectively inhibit proliferation can be estimated using the binomial distribution to calculate the likelihood that the same number of antiproliferative compounds would be drawn by chance. We screened entire pathways with hit rates of ~ 1%. In a screen focused on a single target, it might be necessary to screen ~10,000 compounds to identify 5-10 active inhibitors, which, depending on the overall number of phenotypic positives in the library, would be sufficient to produce a reasonable degree of support for the target. This might be well worth the effort if it improved the target selection process and reduces the failure rate of drug discovery effort. The strategy could also be applied in to target validation for other phenotypes such as cellular differentiation or acquisition of pluripotency.