Clonal phylogenies inferred from bulk, single cell, and spatial transcriptomic analysis of cancer

Abstract

AbstractEpithelial cancers are typically heterogeneous with primary prostate cancer being a typical example of histological and genomic variation. Prostate cancer is the second most common male cancer in western industrialized countries. Prior studies of primary prostate cancer tumor genetics revealed extensive inter and intra-patient tumor heterogeneity. Recent advances have enabled extensive single-cell and spatial transcriptomic profiling of tissue specimens. The ability to resolve accurate prostate cancer tumor phylogenies at high spatial resolution would provide tools to address questions in tumorigenesis, disease progression, and metastasis. Recent advances in machine learning have enabled the inference of ground-truth genomic single-nucleotide and copy number variant status from transcript data. The inferred SNV and CNV states can be used to resolve clonal phylogenies, however, it is still unknown how faithfully transcript-based tumor phylogenies reconstruct ground truth DNA-based tumor phylogenies. We sought to study the accuracy of inferred-transcript to recapitulate DNA-based tumor phylogenies.We first performed in-silico comparisons of inferred and directly resolved SNV and CNV status, from single cancer cells, from three different cell lines. We found that inferred SNV phylogenies accurately recapitulate DNA phylogenies (entanglement = 0.097). We observed similar results in iCNV and CNV based phylogenies (entanglement = 0.11). Analysis of published prostate cancer DNA phylogenies and inferred CNV, SNV and transcript based phylogenies demonstrated phylogenetic concordance. Finally, a comparison of pseudo-bulked spatial transcriptomic data to adjacent sections with WGS data also demonstrated recapitulation of ground truth (entanglement = 0.35). These results suggest that transcript-based inferred phylogenies recapitulate conventional genomic phylogenies. Further work will need to be done to increase accuracy, genomic, and spatial resolution.