A Proof-of-Concept Study: Reference Transcriptome Choice as a Hidden Analytical Variable in Breast Cancer RNA-Seq Analysis

Source abstract

Reference genome and transcriptome are fundamental components of RNA-seq analysis and can influence transcript quantification and downstream differential expression results. Although GRCh38 is the current human reference build, GRCh37 remains in use in research and clinical bioinformatics workflows. The extent to which reference transcriptome choice affects transcriptomic inference in breast cancer remains insufficiently characterized. We performed a comparative analysis of publicly available RNA-seq data comprising two MCF7 breast cancer cell-line samples and two healthy breast tissue samples from the European Nucleotide Archive (PRJEB4829). Identical preprocessing, transcript quantification, and differential expression workflows were applied using GRCh37- and GRCh38-based reference transcriptomes. Reference transcriptome-dependent variability was further examined by comparing the same samples quantified under the two reference transcriptomes. Concordance of differentially expressed genes and pathway enrichment was assessed, and sensitivity to reference transcriptome choice was evaluated for genes included in the MammaPrint and Naderi expression signatures and selected hereditary cancer gene panels. The MCF7-versus-normal-tissue comparison identified 15,950 differentially expressed genes using GRCh37 and 16,348 using GRCh38, with 13,899 genes shared between the two analyses. The most significantly enriched Hallmark and KEGG pathways were also concordant between annotations. Principal component analysis showed that the dominant pattern of variation in the MCF7-versus-normal-tissue comparison was associated with biological status. In contrast, when the same MCF7 samples were quantified using different reference transcriptomes, reference transcriptome choice represented a major source of variation. Expression of genes comprising the evaluated prognostic signatures and hereditary cancer panels showed no significant differences in median expression between reference transcriptomes. However, individual cancer-relevant genes, including PTEN, RAD51D, NANOG, and SOX11, demonstrated reference transcriptome-dependent expression differences. In this proof-of-concept analysis, reference transcriptome choice produced measurable differences in RNA-seq quantification and differential expression, while broad biological patterns and the aggregate expression of established gene signatures were relatively stable. Reference transcriptome-dependent variability became particularly apparent when biological heterogeneity was limited, suggesting that reference transcriptome choice can act as a hidden analytical variable in transcriptomic studies. These findings support explicit reporting and version control of reference annotations and highlight the need for validation in larger, independently replicated breast cancer cohorts before drawing clinical conclusions.