Mapping early human blood cell differentiation using single-cell proteomics and transcriptomics.

Single-cell RNA sequencing (scRNA-seq) has facilitated the characterization of cell state heterogeneity and recapitulation of differentiation trajectories. However, the exclusive use of messenger RNA (mRNA) measurements comes at the risk of missing important biological information. We leveraged recent technological advances in single-cell proteomics by mass spectrometry (scp-MS) to generate an scp-MS dataset of an in vivo differentiation hierarchy encompassing >2500 human CD34+ hematopoietic stem and progenitor cells. Through integration with scRNA-seq, we identified proteins important for stem cell function, which were not indicated by their mRNA transcripts. Further, we showed that modeling translation dynamics can infer cell progression during differentiation and explain substantially more protein variation from mRNA than linear correlation. Our work offers a framework for single-cell multiomics studies across biological systems. Editor's summary: Single-cell analysis is increasingly popular in biomedical research, but most single-cell studies are focused on transcriptomics. These messenger RNA (mRNA) analyses allow for the classification of cell types and changes in the cells over time, but they do not directly reflect cellular functions, which depend on proteins, and the abundance of a given mRNA often does not correlate with that of its corresponding protein. To help address this knowledge gap, Furtwängler et al. developed a mass spectrometry–based method for performing single-cell proteomics. They demonstrated their method on bone marrow samples from healthy human donors, providing biological insights into the process of differentiation of hematopoietic stem and progenitor cells. — Yevgeniya Nusinovich INTRODUCTION: Our ability to characterize complex biological systems at single-cell resolution has facilitated the in-depth resolution of cellular subsets, states, and differentiation trajectories. Given that such studies, to a large extent, have relied on single-cell RNA sequencing (scRNA-seq), it is currently unknown whether emerging technologies such as single-cell proteomics by mass spectrometry (scp-MS) could uncover additional systems information, which would have broad implications for our understanding of fundamental biological processes. RATIONALE: Proteins are the primary executioners of genetic information and because there is a far-from-ideal correlation between mRNA and the proteins they encode, we hypothesized that scp-MS could complement scRNA-seq in the characterization of complex biological systems. In this work, we therefore generated a scp-MS dataset of the human hematopoietic stem and progenitor cell (HSPC) hierarchy and integrated it with a matched scRNA-seq dataset, allowing us to follow the two modalities across hematopoietic differentiation. RESULTS: We performed scp-MS and scRNA-seq on the human CD34+ HSPC hierarchy and demonstrated that the proteome data resolved HSPC heterogeneity in accordance with accompanying flow cytometry data. Using a variational autoencoder deep learning tool, we integrated the transcriptomic and proteomic modalities resulting in a joint embedding. Trajectory analysis demonstrated that, unlike the individual modalities, the integrated dataset more accurately predicted differentiation end states, highlighting the complementarity of the two modalities and their combined advantage in the characterization of hematopoietic differentiation. We also compared the concordance of mRNA and protein information across hematopoietic differentiation and found them to correlate well in more differentiated progenitor populations. By contrast, the two modalities correlated poorly in the most immature stem and progenitor subsets. We also identified proteins whose abundances were better explained by other proteins (such as protein complex partners) rather than their cognate mRNAs. To showcase the ability of scp-MS to uncover proteins of functional importance in stem cells, we selected proteins whose abundances were enriched in stem cells at the protein but not mRNA level and functionally inactivated them using gene editing in human HSPCs. As examples, proof-of-concept functional analyses demonstrated that inactivation of TALDO1, a key enzyme in the pentose phosphate pathway, and H1F0, a histone 1 linker protein associated with quiescent cells, yielded in vitro phenotypes consistent with their functional roles in blood stem cells. Finally, to model the temporal dynamics of mRNA and protein across differentiation, we developed single-cell protein velocity (scProtVelo), which explains observed mRNA and protein abundances across time through learned kinetic parameters. In contrast to corresponding analyses relying solely on mRNA, scProtVelo correctly resolved the information flow in several trajectories, further highlighting the power of combining mRNA and protein level data at the single-cell level. CONCLUSION: We have demonstrated the potential of scp-MS both alone and in combination with scRNA-seq for uncovering biological insights even in an extensively studied system such as the human HSPC compartment. As scp-MS technology continues to develop with respect to sensitivity, throughput, and possibilities for assessing post translational modifications, our understanding of the proteome at single-cell resolution will continue to improve, with far-reaching consequences for biomedical research. From single-cell proteomics (scp-MS) to multimodal analysis.: (1) Application of scp-MS workflow and (2) subsequent integration of scRNA-seq and scp-MS data enable (3) the comparison of mRNA and protein abundance during differentiation at single-cell resolution. (4) Identification and validation of proteins of functional importance and (5) recovery of temporal dynamics through a translation model. [Figure partially created with BioRender.com] [ABSTRACT FROM AUTHOR]