In vivo single-cell CRISPR uncovers distinct TNF-α programs in tumor evolution

The tumor evolution model posits that malignant transformation is preceded by randomly distributed driver mutations in cancer genes, which cause clonal expansions in phenotypically normal tissues. Although clonal expansions can remodel entire tissues1-3, the mechanisms behind why only a small number of clones transform into malignant tumors remain enigmatic. Here, we develop an in vivo single-cell CRISPR strategy to systematically investigate tissue-wide clonal dynamics of the 150 most frequently mutated squamous cell carcinoma genes. We couple ultrasound-guided in utero lentiviral microinjections, single-cell RNA sequencing and guide capture to longitudinally monitor clonal expansions and document their underlying gene programs at single-cell transcriptomic resolution. We uncover a TNF-α signaling module, dependent on TNF receptor 1 and involving macrophages, that acts as a generalizable driver of clonal expansions in epithelial tissues. Conversely, during tumorigenesis, the TNF-α signaling module is downregulated. Instead, we identify a subpopulation of invasive cancer cells that switch to an autocrine TNF-α gene program, associated with epithelial-mesenchymal transition. Finally, we provide in vivo evidence that the autocrine TNF-α gene program is sufficient to mediate invasive properties and show that the TNF-α signature correlates with shorter overall survival in human squamous cell carcinoma patients. Collectively, our study demonstrates the power of applying in vivo single-cell CRISPR screening to mammalian tissues, unveils distinct TNF-α programs in tumor evolution and highlights the importance of understanding the relationship between clonal expansions in epithelia and tumorigenesis. To longitudinally monitor the function of cancer driver genes, we chose the 150 most frequently altered genes in human head and neck (HNSCC) and skin squamous cell carcinomas (sSCC), comprising 134 mutated genes and 16 copy number variations (CNVs). We selected 3 sgRNAs per gene and included 50 non-targeting control sgRNAs, resulting in a CROP-seq library of 500 sgRNAs used for high-titer lentivirus production. Lentiviral library was delivered into E9.5 embryos using ultrasound-guided in utero microinjection to infect the single-layered surface ectoderm, resident immune cells and melanoblasts of constitutively Cas9-expressing mouse embryos and introduce loss-of-function mutations in each of the selected cancer genes. At postnatal day 4 (P4) and 60 (P60), epidermal tissues are collected, sorted for mCherry-positive infected cells, and subjected them to single-cell RNA sequencing (scRNA-seq) to assess gene expression profiles and read out the corresponding sgRNA identity (via dial-out PCR). At P60, chemical carcinogenesis was induced by DMBA/TPA for 12 weeks, then from 30 of thus obtained skin tumors (from 2 animals) single-cell suspensions were sorted for mCherry-positive and negative cells by FACS, and finally sequenced for single-cell transcriptomes. Visium spatial transcriptomics was additionally performed on 4 other DMBA/TPA-induced tumors.