Autocorrelation analysis of a phenotypic screen reveals hidden drug activity

Abstract

Phenotype based screening is a powerful tool to evaluate cellular drug response. Using high content fluorescence imaging of simple fluorescent labels and complex image analysis, phenotypic analysis identifies subtle compound-induced cellular changes unique to compound mechanisms of action (MoA). Recently, a screen of 1,008 compounds in three cell lines was reported where phenotype analysis detected changes in cellular phenotypes and accurately identified compound MoA for roughly half the compounds. However, we were surprised that DNA alkylating agents and other compounds known to induce or impact the DNA damage response produced no activity in cells with fluorescently labeled TP53BP1 - a canonical DNA damage marker. We hypothesized that phenotype analysis is not sensitive enough to detect small changes in 53BP1 distribution and analyzed the screen images with autocorrelation image analysis. We found that autocorrelation analysis, which quantifies the clustering of fluorescently-labelled protein within the nucleus, of 53BP1 images from this screen identified higher compound activity for compounds and MoAs known to impact the DNA damage response. These results demonstrate the capacity of autocorrelation to detect otherwise undetectable compound activity and suggest that autocorrelation analysis of specific proteins could serve as a powerful screening tool for drug discovery.