Abstract

The cell is the fundamental building block of life. With the advent of single-cell RNA sequencing (scRNA-seq), we can for the first time assess the transcriptome of many individual cells. This has profound implications for biological and medical questions and is especially important to characterize heterogeneous cell populations and rare cells. However, the technology is technically and computationally challenging as complementary DNA (cDNA) needs to be generated and amplified from minute amounts of mRNA and sequenceable libraries need to be efficiently generated from many cells. This requires to establish different protocols, identify important caveats, benchmark various methods and improve them if possible. To this end, we analysed amplification bias and its effect on detecting differentially expressed genes in several bulk and a single-cell RNA sequencing methods. We found that correcting for amplification bias is not possible computationally but improves the power of scRNA-seq considerably, though neglectable for bulk-RNA-seq. In the second study we compared six prominent scRNA-seq protocols as more and more single-cell RNA-sequencing are becoming available, but an independent benchmark of methods is lacking. By using the same mouse embryonic stem cells (mESCs) and exogenous mRNA spike-ins as common reference, we compared six important scRNA-seq protocols in their sensitivity, accuracy and precision to quantify mRNA levels. In agreement with our previous study, we find that the precision, i.e. the technical variance, of scRNA-seq methods is driven by amplification bias and drastically reduced when using unique molecular identifiers to remove amplification duplicates. To assess the combined effects of sensitivity and precision and to compare the cost-efficiency of methods we compared the power to detect differentially expressed genes among the tested scRNA-seq protocols using a novel simulation framework. We find that some methods are prohibitively inefficient and others show trade-offs depending on the number of cells per sample that need to be analysed. Our study also provides a framework for benchmarking further improvements of scRNA-seq protocol and we published an improved version of our simulation framework powsimR. It uniquely recapitulates the specific characteristics of scRNA-seq data to enable streamlined simulations for benchmarking both wet lab protocols and analysis algorithms. Furthermore, we compile our experience in processing different types of scRNA-seq data, in particular with barcoded libraries and UMIs, and developed zUMIs, a fast and flexible scRNA-seq data processing software overcoming shortcomings of existing pipelines. In addition, we used the in-depth characterization of scRNA-seq technology to optimize an already powerful scRNA-seq protocol even further. According to data generated from exogenous mRNA spike-ins, this new mcSCRB-seq protocol is currently the most sensitive scRNA-seq protocol available. Single-cell resolution makes scRNA-seq uniquely suited for the understanding of complex diseases, such as leukemia. In acute lymphoblastic leukemia (ALL), rare chemotherapy-resistant cells persist as minimal residual disease (MRD) and may cause relapse. However, biological mechanisms of these relapse-inducing cells remain largely unclear because characterisation of this rare population was lacking so far. In order to contribute to the understanding of MRD, we leveraged scRNA-seq to study minimal residual disease cells from ALL. We obtained and characterised rare, chemotherapy-resistant cell populations from primary patients and patient cells grown in xenograft mouse models. We found that MRD cells are dormant and feature high expression of adhesion molecules in order to persist in the hematopoietic niche. Furthermore, we could show that there is plasticity between resting, resistant MRD cells and cycling, therapy-sensitive cells, indicating that patients could benefit from strategies that release MRD cells from the niche. Importantly, we show that our data derived from xenograft models closely resemble rare primary patient samples. In conclusion, my work of the last years contributes towards the development of experimental and computational single-cell RNA sequencing methods enabling their widespread application to biomedical problems such as leukemia.