Redefining normal breast cell populations using long noncoding RNAs

Single-cell transcriptomics (scRNAseq) has emerged as a powerful tool to assess the transcriptome of individual cells, revealing new and rare cell types and improving the reconstruction of lineage hierarchies. However, a major limitation of current studies is that they only quantify sequencing reads that map to annotated isoforms, which represent roughly one third of all human transcripts. Since the vast majority of unannotated genes are long noncoding RNAs (lncRNAs), these remain largely unexplored. The human breast is a complex organ that harbours different cell populations. Interestingly, different mammary cell populations give rise to different breast tumour subtypes and the cell-of-origin determines the tumour molecular characteristics and clinical outcomes. Using deep bulk RNAseq of normal breast epithelial cells we discovered >13,000 lncRNAs (being 95% unannotated) and mapped their expression levels in scRNAseq, showing they perform better than protein-coding genes at clustering the different cell types. On average,each cell expressed 900 lncRNAs and 4,000 protein-coding genes. LncRNAs had significantly higher cluster specificity levels and were expressed in less cells than their protein-coding counterpart, which is in line with the view of lncRNAs being highly cell type-specific and their apparent lower expression levels being a result of bulk RNAseq estimates. Indeed, when investigating the expression of lncRNAs at cellular level, we confirmed it to be comparable to that of protein-coding genes. We conducted a thorough assessment of these lncRNAs, their expression levels in individual cells and across populations and their correlation with breast cancer subtypes. On average, each cell population has nearly 300 lncRNA markers, from which at least 30 (10%) can be used to classify breast cancer tumours in their different subtypes. Notably, using their predicted protein-coding targets or annotation, we were able to link several specific lncRNAs to tumour subtypes and cancer hallmarks.