Deciphering transcriptional and functional heterogeneity in hematopoiesis with single-cell genomics

Abstract

Purpose of review:

Single-cell genomic approaches have uncovered cell fate biases and heterogeneity within hematopoietic subpopulations. However, standard single-cell transcriptomics suffers from high sampling noise, which particularly skews the distribution of lowly expressed genes, such as transcription factors. This might preclude the identification of rare transcripts that define cell identity and demarcate cell fate biases. Moreover, these studies need to go hand in hand with relevant functional assays to ensure that observed gene expression changes represent biologically meaningful alterations.

Recent findings:

Single-cell lineage tracing and functional validation studies have uncovered cell fate bias within transcriptionally distinct hematopoietic stem and progenitor subpopulations. Novel markers identified using these strategies have been proposed to prospectively isolate functionally distinct subpopulations, including long-term hematopoietic stem cells for ex vivo applications. Furthermore, the continuous nature of hematopoiesis has prompted the study of the relationship between stochastic transcriptional noise in hematopoietic transcription factors and cell fate determination.

Summary:

An understanding of the limitations of single-cell genomic approaches and follow-up functional assays is critical to discern the technical and biological contribution of noise in hematopoietic heterogeneity, to identify rare gene expression states, and to uncover functionally distinct subpopulations within hematopoiesis.

Introduction

Of the 330 x 10⁹ cells that are estimated to turn over daily in the human body, nearly 90% correspond to blood cells (1). Hematopoiesis, the process of blood formation that is necessary to compensate for this regular loss of blood cells, is regulated by a complex set of external signaling cues and intrinsic fate determinants. It is also the paradigm of mammalian tissue homeostasis, sustained by hematopoietic stem and progenitor cells (HSPCs) over the lifetime of an organism. Transcription factors (TFs) have a central role in regulating hematopoiesis, particularly by impacting HSPC self-renewal and differentiation through the orchestration of coordinated expression of target genes as lineage commitment initiates and progresses (2). Single-cell genomics has provided unprecedented insights into the regulatory heterogeneity in individual cells, and multiple methods have been developed to predict upstream regulators of the transcriptional programs of specific cell states along the course of differentiation (3). The source of the transcriptional and cell fate heterogeneity in HSPCs described by these methods is multifactorial, with contributions of intrinsic transcriptional stochasticity, external signals, and technical sources (4).

In this review, we analyze recent advances in our understanding of the transcriptional determinants of cell fate. Given the ubiquitous use of single-cell RNA sequencing (scRNA-seq) to probe hematopoiesis, we highlight important sensitivity considerations on the use of these methods to identify functionally relevant subpopulations and their regulators. Insufficient transcriptome coverage exacerbates the issues due to sparse sampling of genes with low expression yet critical functions, such as TFs. This can lead to incorrect assessment of the specificity of markers or to underdetection of true determinants of cell fate. We underscore the impact of intrinsic stochastic transcriptional noise in regulators of hematopoiesis. Finally, we review several reports of novel regulators that characterize functionally relevant hematopoietic subpopulations, such as long-term reconstituting hematopoietic stem cells (LT-HSCs), while highlighting some of their limitations. While scRNA-seq continues to reveal unparalleled insights into heterogeneity in hematopoiesis, the transcriptome may not always reflect important functional differences within subpopulations. This further underscores the critical need to marry single-cell genomics with functional assays to gain valuable biological insights.

Transcriptional and fate heterogeneity

Understanding the molecular basis of cell fate determination has been a long-standing question in the study of hematopoiesis (5). Recent advances in scRNA-seq have allowed massively parallel profiling of the transcriptome of tens and even hundreds of thousands of single cells (6–9). Applied to the study of hematopoiesis, scRNA-seq has been used to obtain representations of cellular state and the factors underlying it (10). In particular, it has challenged the traditional view of hematopoiesis as a collection of discrete cell states with particular transition points (Figure 1A). This notion, derived from the characterization of cell populations using a limited set of surface markers, has evolved into a continuous model of states along hematopoietic differentiation (Figure 1B) (11). Such gene expression heterogeneity is present across all canonical hematopoietic cell types, including HSPCs and lineage-restricted progenitors (10,12).

Figure 1: Single-cell genomics have reshaped the conceptual model of hematopoiesis.

(A) Classical hierarchical model of hematopoiesis, characterized by presence of discrete cell states and a stepwise progression along differentiation. (B) Continuous model of hematopoiesis uncovered by single-cell genomics, characterized by the lack of transitions in cell states along differentiation. The inset highlights how specific hematopoietic stem and progenitor cell subpopulations can be defined by overlapping single-cell lineage tracing or functional validation information over single-cell transcriptomic representations of the progenitor compartment. This allows to define candidate clusters (e.g. as myeloid progenitors, lymphoid progenitors, etc.) based on their fate bias in in vitro or in vivo assays.

Does transcriptional heterogeneity in stem or progenitor populations also reflect differences in cell fate choice and functional potential? Weinreb et al. barcoded murine hematopoietic progenitors using heritable expressed lentiviral constructs, allowed cells to divide, and profiled clones using scRNA-seq immediately after barcoding or after differentiation in vitro or in vivo (13**). This allowed the authors to associate each clone with their fate, to identify differentially expressed genes shared by progenitors with a similar fate, and to estimate the predictability of terminal fate choice from scRNA-seq gene expression measurements in progenitor populations. Notably, the most informative gene sets for fate prediction were differentially expressed genes in progenitors with a shared fate (38% and 20% more informative that a random size-matched gene set for in vitro and in vivo experiments, respectively). In contrast, TFs were only 10% and 3% more informative than the random gene set in vitro and in vivo. It is likely that additional factors not captured by scRNA-seq, such as chromatin potential or protein expression also contribute to fate determination (14). This notion is further reinforced by the fact that HSPC heterogeneity is not merely stochastic, since it appears to be propagated in serial transplantation experiments (4).

However, it is also possible that the low expression of TFs, together with the limited sensitivity of shallow sequencing scRNA-seq (i.e. the median number of unique transcripts detected per cell in each sample ranged between 1720 and 6429) contributed to this result (13**). Despite its ability to detect large-scale changes in transcriptional programs, low depth scRNA-seq might fail to distinguish genes with low expression that are functionally required in a cell from those that are not expressed (15). Indeed, a comparison of four scRNA-seq datasets profiling HSPCs showed that most abundant transcripts were the main determinants of single cell cluster assignment (16**). In contrast, critical hematopoietic genes measured by scRNA-seq, such as Gata1, Cebpa, Runx1, Myb, Zfpm1, Meis1, Mpo, and Gypa had a negligible influence. However, at a functional level, many of these key TFs and other factors have been shown to be critical determinants of cell fate (11).

The main contributor of zero counts in scRNA-seq data is the sequencing depth per cell, which directly contributes to the total number of unique transcripts detected per cell (17). An empirically determined threshold on the minimal number of unique transcripts detected per cell attempts to mitigate this issue (15,18). As a consequence, the resulting per gene count distributions more closely resemble those from single-molecule RNA fluorescence in situ hybridization (smRNA FISH), considered to be the gold standard to recapitulate ground truth RNA distributions (Figure 2A) (15). Moreover, recent advances in single-cell genomics have achieved significantly increased molecular recovery without compromising cell number throughput, allowing for the profiling of hundreds of thousands of unique transcripts in a cell with deep sequencing coverage (19–21). These considerations of sequencing depth and transcript coverage are critical to faithfully probe the complete transcriptional landscape of hematopoiesis, particularly given the now-routine use of scRNA-seq to profile subpopulations within the hematopoietic hierarchy. Without adequate functional assays following transcriptomic studies, technical noise creates the risk of misassignment of the specificity of certain markers to isolate subpopulations of interest.

Figure 2: Probing the regulators of hematopoiesis at single-molecule resolution.

(A) Single-molecule RNA fluorescent in situ hybridization (smRNA FISH) is considered the gold standard to recover ground truth transcript abundances in cells. Left, the schematic represents an idealized smRNA FISH experiment in which a gene with high expression (gene A, in green) and a gene with low expression (gene B, in red) are measured using fluorescent probes. Right, idealized representation of the distributions recovered by smRNA FISH and single-cell RNA sequencing (scRNA-seq) of these two genes. Without stringent filtering on the per cell transcriptome coverage, the skew in the distribution created by scRNA-seq due to sampling noise is particularly pronounced on lowly expressed transcripts (i.e. gene B) (1). (B) Left, simplified Gata1, Gata2 and Pu.1 minimal regulatory network, as well as the regulatory relationships between its components. Right, idealized representation of the transcript abundance of the three master regulators Gata1, Gata2 and Pu.1 in hematopoietic stem and progenitor cells (HSPCs) across time, displaying episodes of transcriptional bursting. Wheat et al. showed that despite their antagonistic roles, these transcription factors are often co-expressed in HSPCs.

Intrinsic transcriptional noise in the regulators of hematopoiesis

As a result of the stochastic activation and inactivation of promoters, the transcription of genes is thought to occur in “bursts”, with frequent episodes of monoallelic expression (22,23). Beyond the aforementioned technical factors, this phenomenon makes mammalian gene expression inherently noisy and further increases the challenge to define transcriptionally homogeneous populations, even when cells may truly be in the identical cell state (24).

Wheat et al. leveraged smRNA FISH on murine HSPCs to shed light into how robust hematopoietic cell fates arise within such intrinsically noisy biological systems (16**). To this end, they undertook characterization of the stochastic transcriptional noise present for three central regulators of hematopoiesis: Pu.1/Spi1, Gata1, and Gata2. The PU.1 (SPI1) transcription factor is a negative regulator of erythroid differentiation and is required for terminal myeloid differentiation (25). Conversely, GATA1 and GATA2 play crucial roles in erythroid differentiation, and its mutations have been linked to several hematological disorders (26–28). Together, these TFs have been conceptualized as giving rise to a minimal regulatory network, given their opposing effects on erythroid and granulocyte/monocyte cell fate commitment (29,30).

Despite their antagonistic roles, smRNA FISH measurements showed that stochastic transcriptional bursting in HSPCs and common myeloid progenitors often resulted in co-expression of Pu.1 and the Gata TFs (Figure 2B) (16**). Time-lapse microscopy tracking of individual HSPC clones showed that their progeny tended to be in related transcriptional states, suggesting a certain degree of transcriptional priming. However, the dynamics of the system were best explained by a model in which stochastic and reversible transitions occurred between states defined by the expression of the aforementioned TFs (i.e. Gata1/2^hi, Gata2^hi, Pu.1^hi states, and a low expression state for the three TFs). This suggested that stochastic processes dominated by intrinsic noise could underlie the seemingly deterministic behaviors of hematopoiesis. As such, intrinsic noise derived from transcriptional bursting would facilitate the transcriptional plasticity required for balancing differentiation and self-renewal in stem cells. Although this work reported an approach with high molecular sensitivity, defining cell states with only three transcription factors could not fully predict the past or future states of a cell. Future studies that characterize stochastic transcriptional noise in thousands of genes with single-cell resolution might uncover how stochastic transcriptional noise operates more globally during and plays a role in the process of hematopoiesis (31). The insights uncovered by these studies also warrant caution against the definition of discrete subpopulations based on the expression of a handful of genes (or often a single one) in scRNA-seq, especially given the stochastic transitions between related states.

Functional characterization of subpopulations defined using single-cell transcriptomics

Continuous and stochastic models of regulation based on expression dynamics are also influencing new approaches to isolate functionally relevant HSPC populations. Figure 3 illustrates the general workflow from identification of candidate subpopulations using single-cell genomics, to flow cytometric isolation, and functional validation. As such, devising relevant functional validation studies and understanding their limitations is a critical step in this process, since it determines the contexts to which conclusions from markers or regulators can be extended.

Figure 3:

Workflow for the identification of novel markers and regulators with single-cell genomics and functional validation studies.

LT-HSCs have been operationally defined based on their ability to give rise to multiple lineages for more than 16 weeks upon primary transplantation and at least in a subsequent secondary round of transplantation (4). In recent years, panels of surface antigens have been proposed to isolate human LT-HSCs from ex vivo cultures, such as EPCR (CD201) and ITGA3 (CD49c) (32,33).

To simplify such complex antibody panels, Lehnertz et al. devised a reporter of HLF expression to isolate functional human LT-HSCs (34*). The HLF transcription factor is strongly and characteristically expressed in immunophenotypically defined LT-HSCs (34*). Furthermore, HSPCs from Hlf-knockout mice exhibit a reduced ability to reconstitute hematopoiesis upon serial transplantation (35,36). Moreover, HLF is among the six TFs that were used to reprogram committed mouse progenitors into induced hematopoietic stem cells (37). To this end, a fluorescent reporter gene was edited in the endogenous HLF locus using CRISPR genome editing in human cord blood HSPCs (34*). Reporter-positive cells expressed characteristic surface antigens of LT-HSCs. Upon transplantation into immunodeficient mice, reporter-positive cells indeed showed high reconstitution capacity and multipotency. This is the first report of transgenic labeling of human HSCs, demonstrating its potential to use the transcriptome to isolate functionally meaningful subpopulations.

Single-cell transcriptomics and endogenous reporters were also leveraged to identify the expression of the Tcf15 TF as characteristic of mouse LT-HSCs (38*). Rodriguez-Fraticelli et al. used single-cell transcriptomics and lentivirally-barcoded hematopoietic progenitors during long-term bone marrow reconstitution. This allowed the authors to characterize the transcriptional signatures of low-output HSC clones (i.e. clones with a low ratio of committed progenitors relative to HSCs). Tcf15 was one of the TFs that characterized low-output HSCs compared to their high-output counterparts. Of note, the transcriptional signature of low-output HSCs in this study resembled that of published immunophenotypically-defined murine LT-HSCs. One of the hits of an in vivo CRISPR screen of genes that affected HSC output was Tcf15, whose knockout resulted in loss of immunophenotypic LT-HSCs and impaired long-term engraftment potential in secondary transplantation. In turn, overexpression of Tcf15 caused a 20.8-fold increase in the frequency of LT-HSCs, and reporter-positive cells showed higher reconstitution activity than negative cells upon transplantation. Taken together, this work supports a critical role of Tcf15 in long-term repopulation potential of murine LT-HSCs.

Using single-cell transcriptomics, flow cytometry, and functional validation studies, Amann-Zalcestein et al. identified a subpopulation within mouse lymphoid-primed multipotent progenitors (LMPPs) that harbors almost fully restricted lymphoid potential (39*). As a whole, LMPPs have been shown to be heterogeneous in their fate bias, comprising progenitors biased towards lymphoid, myeloid, and dendritic cell fates (40). This heterogeneity was also present transcriptionally, with lymphoid-like and myeloid-like clusters. The Dach1 TF was differentially expressed in the myeloid-associated cluster, positively correlated with myeloid and stem-like genes, and negatively correlated with lymphoid genes. An endogenous Dach1 reporter showed high expression in LT-HSCs, heterogeneity in its expression within the LMPP compartment, and negative expression within the more committed common lymphoid progenitors. This prompted the hypothesis that reporter-negative LMPPs from Dach1 reporter mice identified a transcriptionally distinct subpopulation with restricted lymphoid potential, termed lymphoid-primed progenitors (LPPs). Indeed, LPPs produced more T and B cells per clone, and had minimal myeloid potential in vitro and in vivo compared to Dach1 reporter-positive cells. Notably, this population was not identifiable using standard HSPC markers or previously reported markers of lymphoid priming (39*).

Taken together, these studies exemplify the power of coupling single-cell transcriptomics with functional validation studies for the study of regulation in hematopoiesis. At the same time, they also highlight limitations present with current approaches. For instance, studies that require transplantation assays to assess clonogenic capacity assume that the regulation of hematopoiesis in native and post-transplantation hematopoietic states may be similar. However, major differences exist between the two (i.e. normal hematopoiesis displays low individual HSC contribution, while post-transplantation hematopoiesis is dominated by a few HSC clones) (41). Moreover, reporter expression may not always faithfully mimic endogenous expression, especially if post-transcriptional regulation of the endogenous transcript has a strong influence in determining protein levels, as we discuss in the final section.

Future perspectives: charting heterogeneity in hematopoiesis across the central dogma

Single-cell transcriptomics have allowed unprecedented insights into the transcriptional heterogeneity in hematopoiesis, uncovering previously unnoticed functionally distinct subpopulations and regulators. Some of these markers hold unique translational potential to prospectively isolate clinically relevant progenitors for ex vivo applications. However, in the future, it will be crucial to tease out the degree of stochasticity transfer across the central dogma in hematopoiesis, from RNA to protein. This is critical because reporter expression may not always faithfully mimic endogenous expression of the gene under study. Moreover, the delay created by nuclear export has been postulated as a buffer for transcriptional noise in RNA transcripts, which together with the longer half-lives of proteins compared to their transcriptional counterparts, contributes to smaller variance in their abundance across single cells (42,43). Joint RNA and protein analysis in single cells holds the promise to uncover transcriptional regulation using correlations between TFs and putative target genes at the RNA level, and post-transcriptional regulation in the joint distributions of RNA and protein of a given gene (18). A recent example is the finding that BCL11A, a master regulator of fetal hemoglobin switching, is regulated at the level of messenger RNA translation in hematopoietic development by the RNA-binding protein LIN28B (44,45). Moreover, posttranscriptional regulatory mechanisms may vary between distinct stages of development, as well, adding another layer of complexity (46). As such, recent advances allow single-cell measurements of protein abundance, joint measurements of intracellular transcription factors and single-cell transcriptomes, joint chromatin and single-cell RNA profiling, joint chromatin and protein measurements, and joint profiling of histone modifications and single-cell transcriptomes (18,47–52). Although these techniques will provide novel insights into additional dimensions of cell state, they will need to go hand in hand with relevant functional validation assays to unequivocally define clinically- and biologically-relevant cell states and their regulators.

Supplementary Material

Supplemental Video File

Supplementary video file: deciphering transcriptional and functional heterogeneity in hematopoiesis with single-cell genomics. We highlight the key points of the review, including the need to couple relevant functional assays with single cell genomics, sensitivity considerations of the use of single-cell transcriptomics to uncover regulators of hematopoiesis, as well as the potential of nascent technologies to shed insights into additional dimensions of cell identity.

Key points.

• Single-cell transcriptomics have uncovered gene expression patterns that suggest cell fate biases within hematopoietic subpopulations. However, such observations require validation with relevant functional assays to ensure that the observed gene expression differences are biologically meaningful.

• Standard single-cell transcriptomics suffers from high sampling noise, which particularly skews the distribution of lowly expressed genes, such as transcription factors. This might preclude the identification of rare transcripts that define cell identity and demarcate cell fate biases.

• RNA levels are inherently noisy due to stochastic transcriptional bursting, and subject to post-transcriptional regulation. This might reduce the specificity of reporters of endogenous gene expression to prospectively isolate functionally distinct populations.

• Insights into additional dimensions of cell identity, such as single-cell measurements of chromatin potential or protein abundance, will provide a more complete picture of the molecular determinants of fate and function in hematopoiesis.