Drug mechanism‐of‐action discovery through the integration of pharmacological and CRISPR screens

Abstract

Low success rates during drug development are due, in part, to the difficulty of defining drug mechanism-of-action and molecular markers of therapeutic activity. Here, we integrated 199,219 drug sensitivity measurements for 397 unique anti-cancer drugs with genome-wide CRISPR loss-of-function screens in 484 cell lines to systematically investigate cellular drug mechanism-of-action. We observed an enrichment for positive associations between the profile of drug sensitivity and knockout of a drug's nominal target, and by leveraging protein-protein networks, we identified pathways underpinning drug sensitivity. This revealed an unappreciated positive association between mitochondrial E3 ubiquitinprotein ligase MARCH5 dependency and sensitivity to MCL1 inhibitors in breast cancer cell lines. We also estimated drug ontarget and off-target activity, informing on specificity, potency and toxicity. Linking drug and gene dependency together with genomic data sets uncovered contexts in which molecular networks when perturbed mediate cancer cell loss-of-fitness and thereby provide independent and orthogonal evidence of biomarkers for drug development. This study illustrates how integrating cell line drug sensitivity with CRISPR loss-of-function screens can elucidate mechanism-of-action to advance drug development.

iCSDB: an integrated database of CRISPR screens

“…As demonstrated in the case study, it provides an opportunity to identify drug targets to overcome resistance in cancer. Combining the screen data with the drug sensitivity data would open a systematic way for finding the mechanism-of-action of drugs ( 5 ) or for identifying new drug candidates by repurposing ( 22 ). To accommodate such needs, we plan to upgrade iCSDB to include the drug sensitivity data in the GDSC ( 4 , 23 ) and LINCS ( 24 ) databases.…”

“…Similarly, drug sensitivity was measured in these cell lines for 265 drugs to provide the landscape of pharmacogenomic interactions in cancer ( 4 ). Subsequent analysis of combining genetic screening data with drug sensitivity data offered an unprecedented opportunity to decipher the mechanism-of-action of drugs in a genome-wide and unbiased way ( 5 ).…”

The landscape of metabolic pathway dependencies in cancer cell lines

“…Recent developments in large scale CRISPR-based genetic [ 19 , 20 ] and pharmacologic screening [ 21 ] along with large panels of comprehensively characterized cancer cell lines [ 22 ] have proved powerful tools for identification of genes essential for cancer cell survival [ 23 ], elucidation of drug mechanism-of-action [ 19 , 24 , 25 ], and discovery of novel candidate drug targets [ 26 , 27 ]. Furthermore, parallel integration of both pharmacologic and gene loss-of-function data has been used to identify drug mechanism(s) of action [ 25 , 28 – 30 ].…”

Modeling drug mechanism of action with large scale gene-expression profiles using GPAR, an artificial intelligence platform

“…(2) Not all drugs with shared MOA may be highly consistent at transcriptome level, because some MOAs cannot be reflected at human cell transcriptome level [ 20 ] (e.g. anti-virus/bacteria), Gonçalves et al have systemically presented many diffculties in MOA researches, noting that not all drug are significantly correlated to nominal target gene perturbations [ 11 ], and it is therefore necessary to identify whether and the concerned MOAs are directly related to gene expression signatures or not. (3) For most of MOAs, their signatures change with cell types, time points and dosages, but there are still small parts of MOAs (e.g.…”

“…The other widely used metrics to evaluate the perturbation correlations include cosine similarity, Jaccard score and p value of Fisher exact test between DEGs [ 8 , 9 ]. However, drugs with shared MOAs may not present high similarity scores, because, firstly, drug-induced differential expression of the molecular target may be masked by the much larger differential expression of off-target genes [ 10 ], or even not related to nominal target gene perturbations [ 11 ], so limited up/down regulated DEGs may not cover focused MOAs; and secondly, the pairwise similarity evaluation may also be ineffective due to strong interferences such as batch effects [ 12 ] or common responses [ 13 ]. Compared to other methods available, deep learning, as a non-linear method excelling in fitting high-dimensional data, is independent of “feature” (MOA related genes) extracting, because the high layer structure of deep learning could suppress irrelevant variations [ 14 ].…”