Progression of RNA-sequencing to single-cell applications

Abstract

New methods enable new discoveries. My time as a PhD student has run in parallel with the maturation of the RNA-seq method, and I have used it to discover basic properties of gene expression and transcriptomes. My part has been bioinformatics – the computer analysis of biological data.

RNA-seq quantifies gene expression for all genes in one experiment, allowing discoveries without prior knowledge, as opposed to single-gene hypothesis testing. When I started my PhD, this was done by microarray followed by qRT-PCR validation, which can be arduous. In contrast to microarrays, RNA-seq quantifies expression with little ambiguity of which gene each expression value corresponds to, and in absolute terms. But at the time, data analysis of RNA-seq was full of unknowns and there were little software available. Nowadays, partly the result of my work, the data analysis is much less complicated, and RNA-seq can be performed on diminutive samples, down to single cells, which was not viable using microarrays.

My first study (Paper I) used one of the very first RNA-seq datasets to study general features of transcriptomes, such as mean mRNA length (~1,500 nt) and the number of genes expressed per tissue (~13,000). I also found special features of some tissues: the liver transcriptome is dominated by a few highly expressed gene, brain expresses especially long mRNAs and testis expresses many more genes than other tissues. Following this tissue RNA-seq study, I evaluated a new library preparation method for single-cell RNA-seq (Paper III), developed before the prevalence of single-cell RNA-seq. I used technical replicates to show that the method was accurate and reliable for the more highly expressed genes at single-cell RNA levels, and with input RNA amounts corresponding to >50 cells it produced as good quality data as bulk RNA-seq. Then the method was applied on melanoma cells isolated from human blood, and I listed surface antigen genes that distinguished these circulating tumour cells from other cells in the blood.

This single-cell RNA-seq method was then applied on pre-implantation embryo cells (Paper IV). Using first-generation crosses between two mouse strains, I could separate the expression from the maternal and the paternal copies of the genes. I found that 12-24% of the genes express only one of their two copies in any given cell, in a random manner that affects almost all the expressed genes. I also found that the two copies are expressed independently from each other.

Finally, I studied Sox transcription factors during neural development (Paper II), combining RNA-seq and microarray data for different cell types with ChIP-seq data for transcription factor binding and histone modifications. I found that Sox proteins bind to the enhancers active in the stem cells where the Sox proteins are active, but also to enhancers specific to subsequent cells in ii development. I also found that different Sox factors bind to much the same enhancers, and that they can induce histone modifications.

In conclusion, my work has advanced the RNA-seq method and increased the understanding of transcriptional regulation and output.