Abstract 2495: Does single-cell gene expression reflect RNA abundance?

Abstract Transcriptional dysregulation is a hallmark of cancer that is often studied using single-cell RNA sequencing (scRNA-seq) to capture the transcriptomic profile of individual cells. However, these measurements are not directly representative of gene expression levels due to biased transcript detection which can lead to dramatic misinterpretations of the underlying biology. To demonstrate this phenomenon, we profiled baseline and drug-tolerant persister cells from a patient-derived colorectal cancer xenograft model. Although persister cells are known to be globally hypotranscriptive and exhibit lower total RNA and housekeeping expression by cell number-normalized (CNN) quantitative PCR (qPCR), significantly more transcripts and genes were detected by single-nuclei RNA-seq as compared to baseline. Additionally, we identified this trend in two independent lung adenocarcinoma persister scRNA-seq experiments and noticed an association with higher sequencing read depth per transcript. In all three cases, transcript over-detection could not be ameliorated through filtering, read subsampling, batch correction, or normalizations specifically developed to address read depth bias. As such, we developed a novel computational approach to correct the effects of sequencing read depth bias on scRNA-seq data and recover the relative expression of cells across samples. To benchmark the validity of our approach, we identified five genes predicted to be significantly upregulated from raw and conventionally-normalized counts but downregulated from our depth-normalized counts. Using CNN qPCR, we found decreased abundance of all five targets in colorectal cancer persister cells, in agreement with the depth-normalized analysis approach and results. Beyond the context of persister models, we also examined single-cell ATAC+RNA-seq and spatial transcriptomic data from lymphoma and prostate cancer specimens and demonstrate how depth normalization can be helpful for addressing technical anomalies, such as transcript under-detection due to tissue edge effects. Altogether, sequencing read depth normalization allows us to move beyond the current approach of scaling all cells to the same expression level and achieve more biologically accurate inference from scRNA-seq data. Citation Format: Christie Jay Lau, 1 Liliane Cabral-Fernandes, 1 Yadong Wang, 2 Sumaiyah K. Rehman, 2 Geoffrey Liu, 2 Catherine A. O'Brien, 2 Gregory W. Schwartz2. Does single-cell gene expression reflect RNA abundance [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2495.