MARS-seq2.0: an experimental and analytical pipeline for indexed sorting combined with single-cell RNA sequencing v1

Abstract

Human tissues comprise trillions of cells that populate a complex space of molecular phenotypes and functions and that vary in abundance by 4–9 orders of magnitude. Relying solely on unbiased sampling to characterize cellular niches becomes infeasible, as the marginal utility of collecting more cells diminishes quickly. Furthermore, in many clinical samples, the relevant cell types are scarce and efficient processing is critical. We developed an integrated pipeline for index sorting and massively parallel single-cell RNA sequencing (MARS-seq2.0) that builds on our previously published MARS-seq approach. MARS-seq2.0 is based on >1 million cells sequenced with this pipeline and allows identification of unique cell types across different tissues and diseases, as well as unique model systems and organisms. Here, we present a detailed step-by-step procedure for applying the method. In the improved procedure, we combine sub-microliter reaction volumes, optimization of enzymatic mixtures and an enhanced analytical pipeline to substantially lower the cost, improve reproducibility and reduce well-to-well contamination. Data analysis combines multiple layers of quality assessment and error detection and correction, graphically presenting key statistics for library complexity, noise distribution and sequencing saturation. Importantly, our combined FACS and single-cell RNA sequencing (scRNA-seq) workflow enables intuitive approaches for depletion or enrichment of cell populations in a data-driven manner that is essential to efficient sampling of complex tissues. The experimental protocol, from cell sorting to a ready-to-sequence library, takes 2–3 d. Sequencing and processing the data through the analytical pipeline take another 1–2 d. Introduction The remarkably rich repertoire of transcriptional states of cells within tissues has been studied for many decades. However, only recently have experimental and computational advances in the field of single-cell genomics opened the way for unbiased dissection of tissues into single cells and the de novo characterization of cell types, subtypes, and transcriptional states1–13. Single-cell RNA sequencing (scRNA-seq) is emerging as a key molecular tool for elucidating biological complexity, promising to contribute to a variety of fields in both basic research and medicine14–17. Methods for single-cell genome-wide expression analysis are continuously being developed, offering increasing coverage, precision and throughput1,10,18–31. Here, we describe a robust method for massively parallel scRNA-seq that combines indexed FACS sorting (recording of surface marker levels for each sorted single cell) and robotics with multiple layers of molecular barcoding. Although drop-based and microwell-based methods increased the throughput of scRNA-seq methods17,29,30, the complexity of the cellular hierarchy remains a major challenge for unbiased sampling of mammalian tissues. Human tissues comprise trillions of cells that populate a complex space of molecular phenotypes and functions such that states within this space may vary in their abundance by 4, 5 and up to 9 orders of magnitude32,33. A classic example of this remarkable property of the human cell state space is the difference in abundance between highly abundant red blood cells (N ~ 1013) and rare hematopoietic stem cells (N ~ 103−4)34,35. A well-designed single-cell technology must consider the orders-of-magnitude variation in abundance of human cell lineages and develop experimental, computational and statistical approaches to overcome it. Because cells, and especially combinations of stem cells, progenitors and various types of mature cells, span many orders of magnitude in abundance, relying solely on unbiased sampling to accurately characterize rare populations becomes infeasible, as the marginal utility of collecting more cells diminishes quickly (Fig. 1a). Combining unbiased transcriptional maps with protein markers enables iterative refinement