Random clonal expansion as a limiting factor in transplantable in vivo CRISPR/Cas9 screens

Transplantable in vivo CRISPR/Cas9 knockout screens, in which cells are transduced in vitro and inoculated into mice to form tumours in vivo, offer the opportunity to evaluate gene function in a cancer model that incorporates the multicellular interactions of the tumour microenvironment. In this study, we sought to develop a head and neck squamous cell carcinoma (HNSCC) tumour xenograft model for whole-genome screens that could maintain high gRNA representation during tumour initiation and progression. To achieve this, we sought early-passage HNSCC cell lines with a high frequency of tumour initiation-cells, and identified the pseudodiploid UT-SCC-54C line as a suitable model from 23 HNSCC lines tested based on a low tumourigenic dose for 50% takes (TD50) of 1100 cells in NSG mice. On transduction with the GeCKOv2 whole-genome gRNA library (119,461 unique gRNAs), high (80-95%) gRNA representation was maintained in early (up to 14 d) UT-SCC-54C tumours in NSG mice, but not in UT-SCC-74B tumours (TD50=9200). However, loss of gRNA representation was observed in UT-SCC-54C tumours following growth for 38-43 days, which correlated with a large increase in bias among gRNA read counts due to stochastic expansion of clones in the tumours. Applying binomial thinning simulations revealed that the UT-SCC-54C model would have 40-90% statistical power to detect drug sensitivity genes with log2 fold change effect sizes of 1-2 in early tumours with gRNA libraries of up to 10,000 gRNAs and modest group sizes of 5 tumours. In large tumours, this model would have had 45% power to detect log2 fold change effect sizes of 2-3 with libraries of 2,000 gRNAs and 14 tumours per group. Based on our findings, we conclude that gRNA library size, sample size and tumour size are all parameters that can be individually optimised to ensure transplantable in vivo CRISPR screens can successfully evaluate gene function.