Studying hematopoiesis using single-cell technologies

Several single-cell whole genome amplification (WGA) methods have been developed to amplify the rare genomic DNA. Degenerate-oligonucleotide PCR was used for analyzing copy number variation in cancer cells [ 29]. Another well-known WGA method was multiple displacement amplification (MDA), which utilized random primers and bacteriophage polymerase to achieve high-coverage single-cell exome sequencing [ 30, 31]. Multiple annealing and looping-based amplification cycles (MALBAC) have also been developed to reduce the bias in nonlinear genome amplification process. MALBAC achieves both high-coverage and uniform amplification. It can be applied to detect both copy number variations (CNVs) and single nucleotide polymorphisms (SNPs) in single-cell genome [ 32].

Single-cell transcriptome analysis remarkably serves as a powerful tool for studying cellular heterogeneity and lineage hierarchy (Fig.  1). There are several available methods: single-cell qPCR [ 16], single-cell microarray analysis [ 33], and single-cell RNA-seq [ 34, 35]. After single-cell isolation from complex tissue, the first challenge is to amplify the small amount of RNA, which is about 10 pg per cell. Four mainstream strategies are used: multiplexed RT-PCR, polyA tailing followed by second-strand synthesis [ 19], template switching, and in vitro transcription (IVT) [ 21]. Multiplexed RT-PCR is used in single-cell qPCR experiment [ 16]. Single-cell qPCR does not need to sequence the sample. It is convenient for detection of dozens of genes. PolyA tailing method was used in single-cell microarray and Tang-seq studies. Smart-seq and Smart-seq2 amplification is a widely used approach for the full-length mRNA analysis of single cells [ 22, 36, 37]. It uses the template-switching-based protocol to append a primer binding site on the 3′ end of the cDNA. cDNA is then amplified by PCR and sequenced by Illumina sequencing platform. The mRNA coverage of Smart-seq is between 10 and 20%. IVT used in CEL-seq and MARS-seq accomplishes a linear amplification of RNA using T7 promoter and RNA polymerase [ 21, 38]. The unique molecular identifiers (UMIs) are designed for reducing the amplification bias [ 39]. They enable the absolute counting of mRNA molecules in the single cell when mRNA capture efficiency and the sequencing depth are good enough. The low coverage of mRNA is a common problem for all existing methods.

Recently, application of single-cell transcriptomic analysis has rapidly spread to many areas such as early embryonic development [ 16, 40– 44], cellular reprogramming [ 18, 45], human breast cancer [ 46], metastatic melanoma [ 47], circulating tumor cells [ 48], olfactory neurogenesis [ 49], early embryo development [ 50], neuronal cell heterogeneity, and immune cell pathogenicity [ 51– 53]. These applications demonstrate the broad applicability of single-cell transcriptomic analysis.

Traditional single-cell protein analysis depends on fluorescence flow cytometry [ 54]. The development of mass flow cytometry notably increased multiplexity by isotope label on antibodies [ 55]. This method resolved the spectral overlap problem in fluorescence flow cytometry and can detect more than 30 parameters simultaneously. The idea has also been used in multiplexed ion beam imaging (MIBI) [ 56], which is capable of analyzing up to 100 targets at the same time in the tissue sections. Recent advances in microfluidic chips also enabled multiplexed analyses for quantitative single-cell proteomics [ 57, 58]. All existed methods only allow detection of limited kinds of protein. A whole proteome analysis approach remains to be developed.

Single-cell epigenomic technologies are becoming more and more accessible. Single-cell reduced representation bisulfite sequencing (scRRBS) and single-cell 5hmC-sequenceing were applied to investigate DNA methylation [ 59– 61]. Single-cell chromatin immunoprecipitation sequencing (ChIP-seq) [ 62] and single-cell Hi-C [ 63] have been developed to profile chromatin structure in single cells. Single-cell chromatin accessibility methods, such as single-cell assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq), have been used to investigate cell-to-cell variation in mammalian regulatory elements [ 64]. Large-scale profiling of single-cell chromatin accessibility landscape can be achieved by combining cellular indexing and ATAC-seq [ 65].

Single-cell capturing is a challenge. However, we have seen the significant progress of platform developments in recent years (Table  2). When samples are rare, mouth pipetting and laser capture microdissection (LCM) [ 66] are good choices to isolate single cells. But when dealing with large number of cells, throughput becomes the bottleneck. FACS played an importing role in scaling up single-cell collection efficiency, but library generation remains to be labor intensive and costly. Very recently, lots of other convenient methods have been invented (Fig.  2). Some of them are directly linked with combinatorial indexing and sequencing library generations, which greatly facilitated high-throughput single-cell analysis without requirements for expensive instruments.

Single-cell imaging enables phenotypic characterization of single cells, while preserving their spatial information. DEPArray has image-based single-cell sorting function. Besides, time-lapse microscopy allowed continuous live image on single cells. This method provides important information of the dynamic cell fate decision process during blood cell differentiation [ 75, 76]. Hardware and software requirements for setting up the single-cell long-term imaging system have also been thoroughly discussed [ 77].

Single-cell transplantation is a powerful approach to verify the stem cell identity. Single-cell transplantation of hematopoietic stem cell (HSC) purified by different surface marker proved their multilineage reconstitution function [ 5, 78]. Recently, Notta et al. revealed CD49f as a human HSC marker. Flow-sorted single cell based on CD49f and mitochondrial dye rhodamine-123 displayed robust chimerism even 20 weeks after transplantation [ 79].

The next-generation sequencing platforms generate massive data set. The large amount of cell and gene dramatically increases the dimension and complexity of the data. Traditional computational methods are no longer suitable. Several single-cell computational methods have been developed. Here, we briefly introduce the workflow of single-cell data analysis.

Blood cell production depends on HSC’s self-renewal and multilineage differentiation abilities. However, hematopoietic stem cells are heterogeneous in differentiation behavior. Single-cell transplantation is the most definitive assessment of HSC functional heterogeneity. Classical single-cell transplantation experiment showed that HSC defined by mouse homolog of CD34 reconstituted the lymphohematopoietic system for more than 3 months in mice. Highly purified mouse HSCs based on the expression of CD34 demonstrate variability in self-renewal potential and multilineage differentiation potential [ 5]. A recent research utilized single-cell transplantation assay to analyze phenotypic long-term HSC systematically. Donor-derived contribution to the circulating white blood cells showed at least four distinct patterns. They provide solid evidence that primitive hematopoietic cells can maintain distinct repopulation properties upon serial transplantation in vivo [ 97]. Using similar approaches, Morita et al. found that in the HSC subset, single cells behave differently based on their CD150 expression. Decreased expression of CD150 appears to be associated with reduced erythroblast/megakaryocyte differentiation potential. The balanced long-term repopulating cells are enriched in the CD150 intermediate subpopulation [ 78]. To gain deeper insight into the regulatory program of mouse HSCs, Wilson et al. linked single-cell functional assays with flow cytometric index sorting and single-cell gene expression assays. They identify key molecules that associate with long-term durable self-renewal and provide a single-cell molecular dataset that can be further analyzed regarding HSC heterogeneity [ 98].

In the classical model of hematopoiesis, an organized hematopoietic lineage tree starts with multipotent HSC, and then followed by oligopotent and unipotent progenitors. However, recent single-cell results challenged the classic model and proposed that traditional hematopoietic progenitor types are very heterogeneous [ 99, 100]. Guo et al. used 280 multiplexed qPCR assays to analyze over 1500 single mouse hematopoietic cells [ 17]. The analysis revealed dramatic heterogeneity within all of the classically defined progenitor types, such as HSC, MPP, CMP, and CLP. The comprehensive data revealed a revised hierarchy of hematopoietic stem cell differentiation in which megakaryocytic and erythroid (MegE) lineage was the first to branch from hematopoietic stem cells (Fig.  3b). More recently, Notta et al. combined single-cell gene expression analysis and single-cell functional assay to study human hematopoiesis, and their results challenge the classical human hematopoietic hierarchy model [ 100] (Fig.  3d). They found that the cell hierarchy differed from fetal stage to adult stage. In fetus, multipotent, oligopotent, and unipotent progenitors are all can be seen, while only multipotent and unipotent progenitor stages were observed in adult. They show that megakaryocytic lineage can derive from HSC and multipotent progenitors in fetus but only branch from HSC in adult. Franziska et al. combined index sorting with MARS-seq to analyze mouse bone marrow CMP. The remarkable data set revealed seven transcriptionally distinct subpopulations within CMP cells. These subpopulations showed unexpected priming towards seven differentiation fates but no progenitors with a mixed state. The findings challenged traditional common myeloid progenitors (CMP) defined by cell surface markers and built a single-cell reference for studying mouse myeloid differentiation [ 99] (Fig.  3c). Kristiansen et al. analyzed the differentiation process from fetal liver HSCs to B-1a/B-2 B cells and provided novel insights into the B cell lineage development [ 101]. Macaulay et al. applied single-cell RNA-seq to study thrombocyte lineage commitment in zebra fish. They placed all data points into a continuum to form a refined lineage pathway [ 102]. Another work combined single-cell transcriptional profile and immune-phenotype to clarify differentiation pathway in megakaryocyte-erythroid progenitors (MEP) [ 103].

Single-cell mass cytometry is also a widely used method in hematopoietic differentiation pathway study. A recent primer article detailed the development of mass cytometry and related data analysis methods [ 104]. Bendall et al. measured 34 parameters for human bone marrow samples. Using the data, they constructed a minimal spanning tree-based hematopoietic differentiation pathway. The research also revealed system-wide signaling responses among traditional cell subsets [ 55]. Single-cell mass cytometry was also used to map cell cycle phases for human hematopoietic cells [ 105]. The method allowed deep profiling of major phases of the cell cycle simultaneously in single cells. Single-cell mass cytometry data can be aligned onto a unified trajectory. This approach accurately predicts the stepwise human B cell developmental path de novo. The trajectory revealed virtually all the cellular states of early B cells differentiated from hematopoietic stem cell [ 106].

Altogether, these findings challenged the current step-by-step bifurcation hierarchy in hematopoietic system. Transcriptional heterogeneity among progenitors suggested that cell differentiation may proceed in a more sophisticated way. Single-cell analysis helped to rebuild hematopoietic lineage hierarchy and identify markers of previously unknown subpopulations.

How do blood cells make their lineage decisions? Single-cell analysis helped to resolve this question with higher precision. Pina et al. used single-cell qPCR to define primary multipotent self-renewing cells and early erythroid-committed cells [ 107]. Their data suggested an uncoordinated molecular transition between self-renewal and committed states. They also found dissociation between self-renewal potential and transcriptome-wide activation in lineage program. Multipotent cells are unlikely to change into the committed state under independent activation by individual regulators. A very recent work aimed to capture mixed-lineage states in mouse hematopoietic stem and progenitor cells. With single-cell transcriptomic data, they provided evidence that mixed-lineage intermediates manifested concurrent expression of hematopoietic stem cell/progenitor and myeloid progenitor cell genes [ 108].

Single-cell imaging and tracking methods are also powerful tools for examining blood generation and lineage determination. Eilken et al. used time-lapse microscopy to achieve continuous long-term single-cell observation and detected hemogenic endothelial cells giving rise to blood cells [ 75]. Hoppe et al. applied time-lapse imaging and single-cell tracking to explore co-regulation of transcription factors GATA1 and PU.1 in differentiation dynamics of single HSCs [ 76]. They found that the ratio of PU.1 and GATA1 is not a key regulator for HSPC lineage decision, which challenged the old view on the early hematopoietic lineage separation.

The explosion of genomic data enabled generation of regulatory networks in various biological systems. However, network build with data from population of cells are intrinsically flawed, because the fundamental unit of gene regulation is single cell, rather than heterogeneous population. Regulatory model at single-cell level starts to emerge in recent years. In 2013, Moignard et al. analyzed 18 key hematopoietic transcription factors in hundreds of blood stem cells and progenitor cells [ 109]. They revealed factor interaction between Gata2, Gfi1, and Gfi1b. In their model, Gfi1 represses Gata2, whereas Gata2 activates Gfi1b. Gata2 functions in a regulatory loop to modulate Gfi1/Gfi1b cross-antagonism during HSC entry into the lympho-myeloid lineages. In the same year, Guo et al. build a remarkably similar single-cell model to explain early megakaryocyte and erythroid lineage regulation, in which Gata2 primes MegE fate and represses lympho-myeloid fate [ 17]. In 2015, Moignard et al. introduced another strategy using diffusion maps to analyze single-cell qPCR data. Based on state transition graphs from 3934 cells in the mouse embryo, they generated a transcriptional regulatory network model to explain the whole blood development process [ 28]. Thanks to the emergence of high-throughput single-cell analysis method, we are able to elucidate gene regulation mechanisms for the first time at the single-cell level.

It has been a debate whether intrinsic cell changes or variations in composition contributes to the systematic HSC aging. Single-cell analysis helped to provide a deeper look into this question. Grover et al. analyzed old and young HSCs transcriptomes at the single-cell level. They identified significantly increased molecular platelet priming and functional platelet bias in the aged HSCs [ 110]. They observed that loss of the FOG-1 transcription factor associated with HSC platelet programming increased lymphoid output. Thus, increased platelet bias is a key process during HSCs ageing. Kowalczyk et al. compared cells from young and old mice by single-cell RNA-seq. They found lower frequency of cells in G1 phase among old long-term HSCs. Moreover, old short-term (ST) HSCs resemble young long-term (LT) HSCs, suggesting that they exist in a less differentiated state [ 111]. Single-cell analyses of HSC aging process demonstrate that both compositional changes and intrinsic, population-wide changes contribute to phenotypic aging.

The origin of HSCs in mammalian systems has long been a mystery. The classical study identified the primitive type of hematopoietic stem cell in yolk sac (YS-HSC). But definitive hematopoiesis is maintained by HSC originated within the aorta-gonad-mesonephros (AGM) region of the embryo [ 112, 113]. Definite hematopoiesis produces HSCs with multilineage potential and long-term reconstitution ability. A recent work adopted single-cell RNA-seq to analyze endothelial cells and pre-HSCs in the mouse AGM region. Zhou et al. demonstrated that pre-HSCs have unique features in transcription factor network, signaling pathway, and unique cell cycle status at the single-cell level. They identified new surface markers for pre-HSC isolation and revealed the importance of mTOR in regulating the stepwise generation of HSCs in vivo (Fig.  3e). Transplantation of isolated pre-HSCs demonstrates strong self-renewal capability. New marker of pre-HSCs would provide new insights into HSC development and push forward future clinical applications of HSCs [ 114, 115]. Thus, single-cell analysis show great convenience and precision in tracing the origins of HSCs. Future efforts may focus on the transition of HSC heterogeneity during developmental stage. Besides, the origin of human hematopoietic stem cells also deserves further investigation.