Proteomic/Transcriptomic Analysis of Erythropoiesis

Abstract

Purpose of review:

Erythropoiesis is a hierarchical process by which hematopoietic stem cells give rise to red blood cells through gradual cell fate restriction and maturation. Deciphering this process requires the establishment of dynamic gene regulatory networks (GRNs) that predict the response of hematopoietic cells to signals from the environment. While GRNs have historically been derived from transcriptomic data, recent proteomic studies have revealed a major role for post-transcriptional mechanisms in regulating gene expression during erythropoiesis. These new findings highlight the need to integrate proteomic data into GRNs for a refined understanding of erythropoiesis.

Recent findings:

Here, we review recent proteomic studies that have furthered our understanding of erythropoiesis with a focus on quantitative mass spectrometry approaches to measure the abundance of transcription factors and cofactors during differentiation. Furthermore, we highlight challenges that remain in integrating transcriptomic, proteomic and other omics data into a predictive model of erythropoiesis, and discuss the future prospect of single cell proteomics.

Summary:

Recent proteomic studies have considerably expanded our knowledge of erythropoiesis beyond the traditional transcriptomic-centric perspective. These findings have both opened up new avenues of research to increase our understanding of erythroid differentiation while presenting new challenges in integrating multiple layers of information into a comprehensive gene regulatory model.

Introduction

Erythropoiesis is a gradual process leading to the formation of red blood cells from hematopoietic stem cells (HSCs) [1,2,3]. This process is driven by combinations of transcription factors (TFs) that establish lineage-specific gene expression programs while simultaneously inhibiting non-erythroid hematopoietic lineages [4,5]. The role of TFs in erythropoiesis can be best understood in the context of gene regulatory networks (GRNs) that integrate multiple types of information to describe and predict the response of hematopoietic cells to signals from the environment, leading to the production of red blood cells in sufficient quantities to respond to the demands of the organism [6*]. While invaluable to our understanding of erythropoiesis, GRNs have so far been derived almost exclusively from transcriptomic data [7,8] which has limited their usefulness because proteins (not transcripts) are the drivers of transcriptional mechanisms. Importantly, there is only a limited correlation between the abundances of mRNA and their corresponding proteins [9,10,11,12,13,14,15,16*,17*,18]. Furthermore, we recently showed that master regulators of erythropoiesis (e.g. GATA1, KLF1, KLF3, GFI1B, TAL1) display large discrepancies in their protein versus mRNA dynamics [17*]. Several, non-mutually exclusive post-transcriptional mechanisms can explain these discrepancies. For example, ribosome profiling experiments showed that GATA1 protein production is regulated for a large part at the translational level [15] whereby a short and unstructured 5’UTR transcript renders GATA1 translation particularly sensitive to changes in ribosome levels [19**,20*]. Translational efficiency in erythropoiesis can also be modulated by RNA binding proteins such as RBM38 that promotes translation of select erythroid transcripts [21] or LIN28B that inhibits translation of the gamma-globin repressor BCL11A [22]. Furthermore, protein levels can be directly controlled through interaction with other proteins as recently shown for KLF1 which is protected from proteasomal degradation by the G protein pathway suppressor 2 (GPS2) protein [23*]. While multiple post-transcriptional mechanisms are likely to participate in regulating protein levels, it has been shown that the rate of protein translation in erythroid progenitors is greatly elevated compared to other hematopoietic cells [24*] suggesting that translation plays a particularly important role during erythropoiesis. Further supporting this possibility, mass spectrometry analyses have identified a global decrease in ribosomal proteins in Diamond-Blackfan anemia (DBA) [19**], a genetic disease with mutations in ribosomal proteins genes whereby patients display a selective defect in erythropoiesis [25].

Taken together, the above studies underline a critical need for direct protein measurements to better understand the regulation of erythropoiesis. Here, we review recent proteomic studies that have furthered our understanding of erythropoiesis with a focus on quantitative mass spectrometry approaches. We also provide a perspective on current challenges and exciting future opportunities in the field.

Text of review

1). Proteomic studies in erythropoiesis: quantitative mass spectrometry approaches

Mass spectrometry (MS)-based proteomics is a powerful technique for studying erythropoiesis because it is relatively unbiased, and thousands of proteins can be identified in a typical analysis of samples derived from whole cells or subcellular compartments. Important for studying dynamic processes such as erythropoiesis, MS can also measure quantitative changes in protein levels using either relative or absolute quantification approaches. In the following paragraphs we highlight studies that have used relative or absolute MS-based quantification strategies to provide important insights in our understanding of erythropoiesis.

a). Relative MS protein quantification approaches in erythropoiesis

In a relative quantification experiment, changes in the abundances of proteins across samples (for example at different time-points during erythropoiesis, or between healthy and disease cells) are determined. Relative quantification can be achieved by label–free or isotopic labeling based-approaches. In the label-free approach, each sample is analyzed separately by MS and the relative levels of the proteins in the samples are determined by comparing either the peptide ion signals or the number of identified peptide spectra corresponding to each protein in the different samples. Quantification can also be performed through differential isotopic labeling of proteins or peptides which are then combined prior to MS analysis during which relative protein abundances are determined by comparing the ion signals of isotopically labelled sibling peptides (or their fragment ions). In one of the very first applications of the isotopic labeling quantitative MS approach in 2004, we used the isotope-coded affinity tag (ICAT) method to identify a switch in the proteins that interact with the TF MafK from corepressors to coactivators, an event critical for ß-globin gene activation during erythroid differentiation [26]. More recently, a number of laboratories have used the iTRAQ isobaric labeling technique for erythropoiesis studies. For example, the Xu lab identified mitochondrial factors that play critical roles in erythroid specification [15] as well as proteins that bind to the ß-globin locus [27*]. Also using iTRAQ, the Mayeux lab compared the protein content of pyrenocytes and reticulocytes, revealing that proteasome particles largely segregate to the reticulocyte [14]. We have also used iTRAQ to obtain a global view of quantitative changes in nuclear protein levels at multiple time points along the erythroid lineage from hematopoietic stem/progenitor cells [17*]. Finally, using a slightly different isobaric labeling approach (“tandem mass tags” or TMT [28]), the Frayne lab identified differentially expressed proteins between adult and cord blood red blood cells [29], the Sankaran lab found a global reduction in ribosome proteins levels in erythroid progenitors from patients with Diamond-Blackfan anemia [19**] and the Blanc lab identified dexamethasone targets that regulate erythroid progenitors’ proliferation [30]. Thus, proteomic studies based on relative quantification of proteins have played a major role in revealing molecular processes that are important for erythropoiesis in health and disease.

b). Absolute MS protein quantification approaches in erythropoiesis: different approaches

Despite their usefulness in highlighting important aspects of protein dynamics, MS approaches that rely on relative protein quantification do not provide information on protein copy number and thus have several limitations. First, they do not inform us on protein availability in a specific sample (e.g. we do not know how many molecules of TFs are present in a cell compared to the number of DNA binding sites). Furthermore, they do not measure protein stoichiometry, information that is critical for understanding protein complex composition and formation, or competition between TFs that drive cell fate decision in erythropoiesis. Finally, relative quantification has limited utility for quantitative modeling of GRNs. Absolute measurements, on the other hand, permit the determination of the number of molecules of each protein in a sample and have therefore an extended range of applications (see examples below).

Absolute quantification of proteins by MS can be achieved by two main approaches: 1) by using isotopically labeled peptide standards or 2) by label-free methods (for review, [13]). The first approach, stable isotope dilution (SID), employs stable isotope labeled (SIL) peptides corresponding to peptides derived from proteins of interest which are spiked into the biological sample at known concentrations. Thus, the absolute concentration of the proteins of interest are determined by comparing the ion signals produced by the SIL peptides to the signals produced by the sample peptides. This approach is considered the gold standard for protein quantification by MS because each peptide is quantified by using its own internal control (i.e. an isotopically labeled version of itself). Because the ion signal that each peptide produces in the mass spectrometer can be affected by factors such as amino acid sequence and sample composition, the SIL peptide approach is less prone to quantification inaccuracies due to these issues. On the other hand, because the SIL approach is typically based on a few peptides per protein, quantification can be affected by issues such as PTMs or differences in digestion efficiencies. This approach is commonly used in conjunction with targeted MS approaches, such as selected reaction monitoring (SRM), which are well-suited for reproducible and quantitative measurements of a defined set of analytes over a wide dynamic range of abundances [31,32,33]. However due to the cost of synthetic SIL peptides, this approach is typically used for quantification of relatively small numbers (10–50) of proteins. Alternative approaches for production of SIL peptides such as recombinant expression of artificial proteins composed of target peptides or full-length proteins can reduce the cost of SIL peptides [34,35,36,37,38]. In addition, chemical tagging approaches that permit sample multiplexing such TMT or iTRAQ, can be combined with spiked-in peptide/protein standards and high throughput MS approaches, to improve the throughput of the analysis [39, 40].

Label-free approaches permit quantification of large numbers of proteins in a high-throughput manner. Since they do not require SIL peptides, label-free approaches are not as expensive as the SID approach, and they are less time-consuming. Label-free quantification is based on either the number of identified peptides corresponding to a protein or the ion signals that the peptides corresponding to a protein produce in the mass spectrometer [41]. To estimate the absolute abundance of each protein, the label-free values can be compared to standards of known abundance [11,41] or to the total protein amount of the sample. In this later method, called the total protein approach (TPA) [42], the copy number for each protein is calculated from the ratio of each protein’s total peptide ion signal divided by the total ion signal of the measured proteome with knowledge of the proteins molecular weight, Avogadro’s number and by the protein content of a single cell. In an extension of this approach, called the “proteomic ruler,” the ion signals derived from the histone proteins are used as a proxy for the mass of DNA. This value allows estimation of cell number which in turn allows scaling the protein abundances by cell number. The authors found that the accuracy of TPA was best for proteins quantified with more than a few peptides. Overall, label-free methods provide a rapid and cost-effective approach to estimate absolute abundances for large numbers of proteins without the need for SIL peptides Care must be taken however for proteins quantified by only a few peptides, which typically corresponds to less abundant proteins due to the potential for different peptides to produce different MS signals per mole. Furthermore, potential changes in cellular histone content should be taken into consideration when using the proteomic ruler method [43,44].

c). Insights from absolute quantification of proteins using label-free MS approaches

Label-free MS approaches were used in several large-scale studies of erythropoiesis [14,16*,45*]. All these studies measured proteins from whole cell lysates (made of a mixture of cytoplasmic and nuclear proteins) that have been prepared from either purified or synchronised cell populations ranging from erythroid progenitors to terminally differentiated orthochromatic erythroblasts. Overall, these studies identified and quantified between 6,000 and 8,000 total proteins at different stages of differentiation, including proteins from the nucleus, the cytoplasm, the plasma membrane (including solute carrier transporters of metal ions), mitochondria and even secreted proteins. The data provide a previously unappreciated scaling of these various classes of factors. Overall, these global analyses revealed a gradual progression of protein clusters across the different stages of erythropoiesis and identified a decrease in the total amount of cellular proteins during terminal erythroid differentiation. Interestingly, comparative analyses between protein levels and transcripts revealed a global dampening of quantitative variations at the proteome level suggesting that much of transcriptional noise is buffered at the protein level. Finally, comparative analyses between the proteome of primary erythroid cells with that of cell lines commonly used for studying erythroid differentiation allowed the authors to pinpoint the exact stage of differentiation at which these cell lines are arrested, showing that global comparative analyses of proteins are useful to identify specific stages of differentiation [16*]. While the conclusions from these studies are mostly in agreement, we note that Gautier et al. did not detect a significant decrease in overall cellular histone contents at the end of erythroid differentiation [14,16*] whereas Karayel et al. were able to detect this change [45*] first suggested in previous studies [43,44]. This inconsistency remains to be addressed. While these data provide valuable information to detect global changes in multiple classes of proteins during erythropoiesis, the limited coverage of low abundance proteins such as TFs makes this approach less appropriate for incorporation into GRNs.

d). Insights from absolute quantification of transcription-related proteins using SRM-based MS with SIL peptides

To identify quantitative changes in the stoichiometry of transcription factors and cofactors during human erythropoiesis, we extracted nuclear proteins at multiple stages of erythropoiesis from hematopoietic stem/progenitor cells and we used SRM-based targeted mass spectrometry coupled with the spiking of known amounts of isotopically labelled peptides to quantify over one hundred proteins, including transcription factors, cofactors, chromatin remodeling enzymes and subunits of the general transcription machinery [17*,33*,46]. These analyses led to several major findings. First, we found that the correlation between protein and RNA levels is the lowest at the early stages of differentiation in hematopoietic stem/progenitor cells [17*], consistent with the previous finding that these cells have a low protein translation rate [47]. Furthermore, this result is in agreement with other MS-based studies using purified hematopoietic progenitor populations of both human [18] and mouse [48] origin. Importantly, we were able to measure for the first time, changes in the stoichiometry between antagonistic TFs in hematopoietic progenitors, showing that the lineage specifying TFs KLF1 and FLI1 are both present at about 3,000 copies per cell in hematopoietic progenitors and that KLF1 protein gradually becomes more abundant as the cells progress along the erythroid lineage [46]. Interestingly, our data showed major discrepancies in the dynamics of transcripts versus proteins during the course of erythropoiesis, leading to reconsideration of some previously accepted models that were based on transcripts measurements. For example, while the switch from GATA2 to GATA1 genomic binding that occurs at the proEB stage has traditionally been thought to be driven by GATA1 becoming more abundant than GATA2, this view is not compatible with the proteomic data. Indeed, at the protein level, GATA1 is more abundant than GATA2 from the early stages of hematopoiesis. Thus, GATA2 is able to bind to its target genes despite an excess of GATA1, and the GATA2/GATA1 switch is likely to be mediated by a decrease in GATA2 protein rather than just a competition with GATA1 [17*].

Perhaps the most important finding from our study was the unexpected large difference in abundances between transcriptional co-repressors and co-activators. Specifically, we observed that in the nucleus, co-repressors are on average 2 orders of magnitude more abundant than co-activators with TFs being present at intermediate levels (Figure 1a). Interestingly, these differences exist at the protein, not the RNA level and can be at least partially explained by the instability of co-activator proteins [17*]. This finding has major implications for our understanding of transcriptional regulation during erythropoiesis in general by implying that DNA binding TFs must compete with each other to recruit rare co-activators in the context of a highly repressive nuclear environment (Figure 1b). Furthermore, a high co-repressor / co-activator ratio is likely to play an important role in maintaining oligopotency and regulating cell fate decisions in progenitors by preventing inappropriate expression of genes from mutually exclusive lineages (Figure 2). Consistent with this, one of the most abundant factors we quantified (i.e. CHD4, over 400,000 copies per nucleus) is a subunit of the NuRD co-repressor complex that plays a critical role in repressing the fetal g-globin gene [49] and suppressing transcriptional noise [50]. It is likely that the repressive nuclear environment is a general principle of cellular differentiation.

Figure 1. Stoichiometry of different classes of proteins in the nucleus.

(a) Abundance of different classes of proteins in the nucleus, including co-activators, transcription factors and co-repressors

(b) The high abundance of co-repressors facilitates their recruitment to target genes whereas transcription factors must compete to recruit rare co-activators.

Figure 2. Proposed model of cell fate decision facilitated by the stoichiometry between transcription factors, co-repressors and co-activators.

This model is based on the high abundance of co-repressors (not shown), and the low abundance of co-activators in the nucleus as determined by quantitative mass spectrometry. In the multipotent progenitor cell (top row), the low abundance of co-activators coupled to the large excess of co-repressors prevent high level activation of genes from mutually exclusive lineages A, B and C. As the progenitors differentiate (second row), the abundance of lineage-specifying transcription factors (colored circles) increases, which facilitates the recruitment of co-activators to lineage-specific genes at the expense of genes from other lineages.

Finally, it should be noted that even though co-activators are rare in the nucleus, their protein levels can vary dramatically during erythropoiesis [17*] suggesting that they may have specific roles at different stages of differentiation. Consistent with this, a number of studies found that histone-modifying enzymes, including the histone acetyltransferase MOF [51**], the histone H3K36 methyltransferase NSD1 [52], the H3K27 methyltransferase PRC2 [53] or the H4K20 methyltransferase SETD8 [54] have erythroid-specific functions. While our study provided quantitative information on a diverse set of representative subunits from transcriptional co-regulatory complexes, comprehensive quantification of complex subunits remains to be performed in a systematic manner to assess potential changes in stoichiometry during erythropoiesis.

2). Proteomic studies in erythropoiesis: future prospects and challenges

Despite the considerable amount and variety of proteomic data available, additional measurements are necessary to fully comprehend gene regulation in erythropoiesis. These measurements ought to be quantitative and aim to include all expressed proteins and their proteoforms, including post-translational modifications (PTMs) [55*]. Indeed, while a global phospho-proteome study has been published in erythroid cells [45*], additional PTMs such as acetylation, sumoylation and others remain to be systematically identified. Furthermore, a recent study highlighted the effect of trace metals such as zinc in modifying the proteome of erythroid cells [56], an interesting area to pursue. Given the complexity of the proteome (i.e. >100M predicted proteoforms derived from 83,000 transcripts and 20,000 genes [55*], and the interactions amongst these proteins, much remains to be done to reveal the full proteome and its changes during erythropoiesis.

Beside generating new proteomic data, two major challenges remain: 1) integrating these data within other data types such as DNA binding (e.g., by chromatin-immunoprecipitation) into a comprehensive and predictive GRN, and 2) moving mass spectrometry-based proteomics into the single cell arena.

a). Data integration and gene regulatory networks.

Using quantitative protein measurements by SRM and ordinary differential equations, we recently built the first temporal GRN for erythroid commitment that integrates quantitative changes in protein and mRNA abundances of key TFs [17*]. Notably our model accurately recapitulated a number of previously known regulatory links, including the cross-antagonisms between lineage specifying (LS)-TFs in their correct sequential order. Importantly the model revealed that these cross-antagonisms are quantitatively imbalanced and that these imbalances become more pronounced with time, further supporting the continuous and quantitative nature of cell fate decisions. Despite this initial success, much remains to be done in this area, including integration of information about co-activators and co-repressors, PTMs, protein-protein interactions as well as protein-DNA and protein-RNA interactions into a refined model of erythropoiesis. The numerous challenges that remain in data integration and quantitative modeling of GRNs have been reviewed recently [6,57].

b). Single cell proteomics: are we there yet?

In the past 5 to 6 years, single cell transcriptomics has revolutionized our understanding of hematopoiesis and erythropoiesis [5]. However, single cell proteomic technology has not kept pace with transcriptomics, which is largely due to a lack of sensitivity of current protein-based technologies. And while antibody-based single cell approaches such as CyTOF have the sensitivity to permit measurements of low abundance proteins like TFs, they are limited to proteins for which suitable antibodies are available. Furthermore, antibody-based approaches do not provide information about absolute protein amounts unless calibrated with standards of known abundance [46].

Importantly, recent studies suggest that single cell proteomics by MS (sc-MS) may soon be an achievable goal even for low abundance proteins. For example, recent advances in MS instrumentation, sample processing, and experimental design have allowed the identification of ~1000 protein groups from single cells using both label-free [58] and isobaric labeling approaches [59*,60**]. Furthermore, ~6000 protein groups have been identified in samples derived from a few hundred cells [61]. Notably, the Slavov laboratory recently demonstrated that a handful of TFs could by identified and quantified by sc-MS in macrophages [60**]. Finally, temporal trajectories were recently established from sc-MS data of murine macrophages during lipopolysaccharide stimulation [62]. Based on these new exciting results, it is entirely conceivable that, with further advances in sample processing, experimental design, and computational approaches, MS-based proteomics will soon permit large-scale proteome measurements in single cells, opening new doors to understanding transcriptional regulation of erythropoiesis and other dynamic cellular processes.

Conclusion

Recent proteomic studies have considerably expanded our knowledge of erythropoiesis beyond the traditional transcriptomic-centric perspective. These findings have both opened up new avenues of research to increase our understanding of erythroid differentiation while presenting new challenges in integrating multiple layers of information into a comprehensive gene regulatory model.

Key Points.

• Major discrepancies between mRNA and protein abundances for master regulators of erythropoiesis

• Absolute quantification of proteins by mass spectrometry uniquely provides information of about the stoichiometry and dynamics of native proteins in erythropoiesis

• Direct quantitative protein measurements (copy number per cell) provide key information for building quantitative models of gene regulatory networks in erythropoiesis

• The nucleus is a highly repressive environment with corepressors that are two orders of magnitude more abundant than coactivators, likely a general principle of transcriptional regulation during cellular differentiation.