Controlling for cellular heterogeneity using single-cell deconvolution of gene expression reveals novel markers of colorectal tumors exhibiting microsatellite instability

Abstract

Approximately 15% of colorectal cancer (CRC) cases present with high levels of microsatellite instability (MSI-H). Bulk RNA-sequencing approaches have been employed to elucidate transcriptional differences between MSI-H and microsatellite stable (MSS) CRC tumors. These approaches are frequently confounded by the complex cellular heterogeneity of tumors. We performed single-cell deconvolution of bulk RNA-sequencing on The Cancer Genome Atlas colon adenocarcinoma (TCGA-COAD) dataset. Cell composition within each dataset was estimated using CIBERSORTx. Cell composition differences were analyzed using linear regression. Significant differences in abundance were observed for 13 of 19 cell types between MSI-H and MSS/MSI-L tumors in TCGA-COAD. This included a novel finding of increased enteroendocrine ( q = 3.71E -06 ) and reduced colonocyte populations ( q = 2.21E -03 ) in MSI-H versus MSS/MSI-L tumors. We were able to validate some of these differences in an independent biopsy dataset. By incorporating cell composition into our regression model, we identified 3,193 differentially expressed genes ( q = 0.05), of which 556 were deemed novel. We subsequently validated many of these genes in an independent dataset of colon cancer cell lines. In summary, we show that some of the challenges associated with cellular heterogeneity can be overcome using single-cell deconvolution, and through our analysis we highlight several novel gene targets for further investigation.

Novel insights into the molecular mechanisms underlying risk of colorectal cancer from smoking and red/processed meat carcinogens by modeling exposure in normal colon organoids

“…We have previously shown that adjusting for cell composition enriches for biological signal and reduces the reporting of DEGs driven by cellular heterogeneity [ 28 ]. Thus, we incorporated cell scores into our regression model and measured the extent of variation in gene expression that could be attributed to cell composition ( Figure 2A ).…”

“…Increasing evidence has demonstrated a strong association between smoking and MSI-H CRC tumor development [ 1 , 2 ]. MSI-H tumors are driven by reduced expression of specific DNA MMR genes [ 28 ], and the average expression of genes within the coral module were reduced in colon organoids exposed to carcinogens. The coral module also contains MLH1 , MSH2 and MSH6 , inherited pathogenic variants for which are well known to be causal in Lynch syndrome.…”

Transcriptomic Response to Calcium in Normal Colon Organoids is Impacted by Colon Location and Sex

“…There is independent epidemiologic evidence that colon location impacts response to different environmental risk factors. These side differences in response to environmental factors may have broader implications, and there is a growing recognition of differences in the molecular phenotypes of tumors arising in the right and left colon (25,45,46). It is important to note that our finding of more DEGs occurring in organoids derived from the right than left colon following increased calcium does not necessarily imply an improved chemopreventive outcome in the right colon.…”

“…Cell-type composition was inferred for samples sequenced in bulk by deconvolution using CIBERSORTx and a curated single-cell RNA-sequencing (scRNA-seq) dataset (24), as reported previously (14,16,17,25). Pre-processing and cell selection of scRNA-seq data used for single-cell deconvolution has been described previously (14).…”

“…Previously, we have shown that variation in cell composition can impact findings from RNA-seq analysis of bulk datasets (14). To address this, we incorporated cell scores into our regression model, as performed previously (14,16,17,25). Expression of 485 genes were significantly associated with the different calcium concentrations in our full analysis, including: CEACAM7 (BH ¼ 4.75E -03 ), KRT80 (BH ¼ 1.43E -07 ), and pleckstrin homology like domain family A member 1 (PHLDA1; BH ¼ 1.48E -03 ; Fig.…”