Inferring Transcription Dynamics By Integrating Single-Molecule RNA Imaging and Pol II ChIP-Sequencing

Single cell gene expression measurements show that individual cells can have significant heterogeneity in mRNA levels. Even one cell grown in culture can develop into a population with strikingly variable gene expression at the single cell level. Single-cell RNA sequencing has enabled characterization of this variability at an unprecedented scale; however, it is largely unknown if this variability has phenotypic consequences such as differential drug sensitivities, proliferation rates, or invasiveness. The primary problem with using single cell RNA sequencing to find these phenotypes is that most experiments capture a large number of variable genes, many of which are thought to be due to experimental or technical “noise” and have no observable association with phenotype. Thus, a huge challenge in this field has been filtering through the “noise” to determining which variable genes are actually leading to different behaviors at the single cell level.

We hypothesized that variable genes associated with different behaviors at the single cell level would show memory of gene expression across cellular division, as genes with this behavior would be part of a coherent “cell state”. Thus, we developed an unbiased method for quantifying single cell gene expression memory by using the design of Luria and Delbrück’s “fluctuation analysis” and combining it with RNA sequencing. Specifically we isolated individual cells, allowed them to expand, and then performed RNA sequencing on each of the samples. For genes with uniform expression, all the samples should have similar RNA sequencing counts; however, for genes that occasionally turn on in a subset of cells, the RNA sequencing counts across all samples show high variability. We applied this method to two melanoma cell lines and one breast cancer cell line, and quantified the memory associated with each gene. From correlation analysis of the most heritable genes, we found distinct subpopulations of cells that are indeed more resistant to targeted therapy and chemotherapy, respectively. Taken together, Luria-Delbruck-seq is a method for quantification of genome-wide transcriptional memory which can enable de novo identification of functionally important rare subpopulations with differential sensitivities to therapy.