Stat and interferon genes identified by network analysis differentially regulate primitive and definitive erythropoiesis
© Greenfest-Allen et al.; licensee BioMed Central Ltd. 2013
Received: 29 November 2012
Accepted: 30 April 2013
Published: 15 May 2013
Hematopoietic ontogeny is characterized by overlapping waves of primitive, fetal definitive, and adult definitive erythroid lineages. Our aim is to identify differences in the transcriptional control of these distinct erythroid cell maturation pathways by inferring and analyzing gene-interaction networks from lineage-specific expression datasets. Inferred networks are strongly connected and do not fit a scale-free model, making it difficult to identify essential regulators using the hub-essentiality standard.
We employed a semi-supervised machine learning approach to integrate measures of network topology with expression data to score gene essentiality. The algorithm was trained and tested on the adult and fetal definitive erythroid lineages. When applied to the primitive erythroid lineage, 144 high scoring transcription factors were found to be differentially expressed between the primitive and adult definitive erythroid lineages, including all expressed STAT-family members. Differential responses of primitive and definitive erythroblasts to a Stat3 inhibitor and IFNγ in vitro supported the results of the computational analysis. Further investigation of the original expression data revealed a striking signature of Stat1-related genes in the adult definitive erythroid network. Among the potential pathways known to utilize Stat1, interferon (IFN) signaling-related genes were expressed almost exclusively within the adult definitive erythroid network.
In vitro results support the computational prediction that differential regulation and downstream effectors of STAT signaling are key factors that distinguish the transcriptional control of primitive and definitive erythroid cell maturation.
KeywordsPrimitive erythropoiesis Definitive erythropoiesis Stat1 Stat3 IFN-signaling Gene-regulatory networks Co-expression network inference
In adult mammals, red blood cells are ultimately derived from hematopoietic stem cells that commit to the erythroid lineage. Erythroid progenitors in the bone marrow give rise to a wave of morphologically identifiable precursors that undergo a limited number of cell divisions in association with macrophage cells. These maturing erythroblasts accumulate hemoglobin, reduce cell size, condense their nucleus and ultimately enucleate to form reticulocytes that are released into the bloodstream. Prior to birth, a similar process of “definitive” red cell production occurs in the fetal liver.
However, the embryo requires red blood cells prior to the formation of the liver. This need is satisfied by the emergence of a transient population of “primitive” erythroid cells from the yolk sac. In the mouse, primitive erythroid progenitors (EryP-CFC) first emerge in the yolk sac beginning at embryonic day 7.5 (E7.5), and generate a wave of maturing primitive erythroblasts that exclusively constitute red cells in the embryo until E12, when the fetal liver begins to release definitive erythrocytes . Primitive erythroblasts progressively undergo nuclear condensation and accumulate increasing amounts of hemoglobin until replication ceases, ultimately reaching steady-state hemoglobin content and a final cell size more than six times that found in adult murine erythrocytes . In the mouse, primitive erythroid precursors primarily express embryonic globins (βH1, ϵy, and ζ), while definitive erythroid cells in the fetal liver and bone marrow express adult globins (β1, β2, α1, and α2) [3, 4]. Despite maturing in the bloodstream, primitive erythroblasts, like their definitive counterparts, ultimately enucleate to form reticulocytes .
Definitive erythropoiesis has been extensively studied and several key transcriptional regulators of erythroid cell maturation have been identified, particularly in the adult erythroid lineage produced in the bone marrow. However, relatively little is known about the regulation of primitive erythropoiesis. Some key transcription factors (TFs) have been identified that regulate the production of both primitive and definitive erythroid cells, including Tal1, Lmo2, Gata1, Gata2, and Klf1[5, 6]. Other key TFs play lineage-specific roles; c-Myb and Gfi1b, for example, preferentially regulate definitive erythropoiesis [7–9]. Likewise, the targeted disruption of the cytokine erythropoietin (Epo) and its receptor (Epor) have revealed an essential role for this pathway in the synthesis of definitive erythrocytes. This is not the case for primitive erythropoiesis, where a reduced population of primitive erythroblasts continues to mature in the bloodstream of Epo- and Epor-deficient embryos [10–12].
Our goal is to utilize knowledge of definitive erythropoiesis to gain further insight into the mechanisms that regulate primitive erythroid maturation and to identify factors that may distinguish the maturation of these two distinct, but closely related erythroid lineages. We employ a network-based systems approach to infer lineage-specific transcriptional regulatory networks from annotated microarray expression data. These data were obtained from primitive erythroid, fetal definitive erythroid and adult definitive erythroid cells isolated from mouse embryos, fetuses, and adult bone marrow, respectively . Five independent samples of primary erythroid precursors at three progressive stages of maturation (proerythroblasts, basophilic erythroblasts, and poly-orthochromatic erythroblasts), as well as reticulocytes, were purified by flow cytometry and used for the analysis of global gene expression on an Affymetrix platform.
Gene-interaction networks inferred from patterns of co-expression have become increasingly popular tools for exploring gene function in biological systems. Such analyses have largely focused on identifying functionally enriched integrated sub-networks of co-expressed genes representing coherent functional units or biological pathways [14–16]. However, the architecture of an interaction network also provides insight into specific gene essentiality in the modeled system. In particular, the topological prominence of a gene or protein in an interaction network may reflect its biological role [17–19], although the association between specific measures of topology (e.g., degree, centrality, or connectivity) and essentiality likely varies .
Here, we applied a three-stage semi-supervised machine learning algorithm to estimate gene essentiality during erythroid precursor maturation. We employed the well-characterized transcriptional control of definitive erythropoiesis to identify topological features of inferred transcriptional regulatory networks and patterns of gene expression during erythroid precursor maturation that characterize known key regulators of red cell differentiation. Using these features, we predicted potential regulators of primitive versus definitive erythropoiesis and these predictions were then validated experimentally. Taken together, our data indicate that differential STAT signaling plays an important role in the regulation of primitive compared to definitive erythropoiesis.
We identified 1,080 potential transcriptional regulators expressed in the microarray expression dataset of erythroid cells using Gene Ontology (GO) annotations (see Materials and Methods). Of this set of potential key factors, 16 were known to play either essential or non-essential roles in the regulation of adult definitive erythropoiesis and were used as a reference dataset for training the machine learning algorithm.
Lineage-specific regulatory networks were assembled by integrating factor co-expression and computational predictions of TF binding based on sequence similarity. Although less than 15% of the potential interactions were realized, the networks did not exhibit scale-free topologies. Networks were overall highly connected, with degree distributions left-skewed and most genes having >400 neighbors (Additional file 1: Figure S1). The full list of inferred interactions comprising these networks can be accessed through interactive search strategies on the ErythronDB website (http://www.cbil.upenn.edu/ErythronDB/search.jsp). No single pattern of expression or standard measure of topological prominence in the estimated regulatory networks characterized the reference gene set, although most were preferentially expressed in the more immature proerythroblast and basophilic erythroblast stages of maturation (Additional file 1: Figure S2).
Measures of expression and network topology that potentially characterize the essentiality of a transcription factor in a gene-regulatory network
number of local networks in which gene is found
number of edges incident to a gene, normalized by size of the neighborhood (k/N)
Proportion Predicted Targets
proportion of incident edges annotated by predicted binding
Weighted Clustering Coefficient*
Weighted Closeness Centrality†
Average expression across replicates
Average expression across replicates
Average expression across replicates
Average expression across replicates
Best-fit vector mapped to the average expression profile
Ratio of maximum to minimum expression during developmental series
During most runs, the GA successfully converged on optimal solutions in less than 100 generations and performed well for both the training (adult definitive) and testing (fetal definitive) erythroid datasets. Various runs were differentiated by the GA parameters (e.g., mutation, cross-over rate, population size). Evolved weights comprising the best solutions found by the GA in each run were ranked by the product of their fitness in both the training and testing datasets and the top 10 solutions are listed in Additional file 2: Table S3. These results are representative of all solutions, which were highly consistent in highlighting measures of global centrality, clustering coefficient, out-degree, and average absolute expression in the basophilic and poly-orthochromatic erythroblast stages as important properties for discriminating key regulators.
Using the Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.7 [21, 22], we annotated and surveyed functional term enrichment in the differentially expressed result set. The 1,080 TFs used to build the interaction networks were employed as the background set for this analysis. Functional enrichment clustering using the DAVID resource resulted in a few significant groups, the most relevant of which included hemopoeisis and erythrocyte homeostasis, embryonic morphogenesis, regulation of cell cycle and cell differentiation, regulation of apoptosis, intracellular signaling, and a variety of signaling pathways (e.g., TPO, TGF-Beta, JAK-Stat, Wnt, ERBB, IL6 signaling). Closer examination of the DAVID results revealed a large degree of overlap in enrichment cluster membership and many of the TFs found in multiple enrichment clusters were known to be involved in or downstream effectors of MAPK signaling. Further scrutiny of the differentially expressed result-set revealed a total of 56 genes (~39%) related to MAPK signaling (Figure 2, indicated by †).
The primitive, fetal definitive, and adult definitive erythroid-specific gene-interaction networks inferred from microarray expression datasets are highly connected and do not exhibit scale-free topologies. The connectivity of the inferred networks is, in part, a reflection of the underlying dataset, which is high quality and exhibits very high fidelity among replicates in the primitive and adult definitive erythroid lineages , effectively yielding stepped expression profiles with four stage-specific plateaus. The connectivity also reflects the underlying biology. By restricting our gene set to transcription factors, we segregated a single cohesive functional sub-network of the genome-wide expression during the terminal maturation of each lineage; i.e., the transcriptional regulation of erythropoiesis.
Annotating network edges with predicted TF binding potentials reduced the connectivity of the co-expression network by introducing directionality. However, the utility of this annotation was limited by the availability of partial weight matrices and binding consensus sequences, which only allowed predictions of targets for a third of the TFs considered in this analysis. These out-directed edges were important for discriminating essential from non-essential regulators (W T ≥ 0.88 in all GA runs, Additional file 2: Table S3), suggesting that integrating further directionality would highlight additional differences among these lineages. The predicted binding may have introduced a bias to the analysis; genes for which binding targets were predicted were more likely to be identified as potential regulators, but only if many of their potential targets were present in the networks. For example, targets were predicted for Foxo3, but <1% of those targets (n = 3) were found in the adult definitive erythropoiesis network (Additional file 1: Figure S2B). The gene still had a relatively high essentiality score within the adult-definitive lineage (S = 1.96), determined by the other properties contributing to the score estimate.
Another limiting factor to this analysis was the use of the Gene Ontology to identify potential regulators. Due to the incompleteness of the annotation, some known, and likely several unknown, factors that play a key role regulating erythropoiesis were removed from consideration. For example, Lmo2, a known transcription factor and essential regulator of erythropoiesis, was filtered from the analysis due to the incompleteness of its GO annotation at the time the analysis was performed.
Despite these limitations, this system provided a rare opportunity to compare a set of closely related regulatory networks underlying the development of phenotypically distinct but functionally equivalent cells within a single organism. The essential regulatory mechanism underlying the fetal and adult definitive erythroid lineages has been well characterized, but comparatively little is known about the regulation of primitive erythropoiesis. The regulatory networks underlying these three erythroid lineages are different. However, they must also possess some commonalities as each results in the synthesis of a cell containing a complex cytoskeletal network, filled with hemoglobin, and devoid of a nucleus and internal organelles. While the timing and identity of essential regulators may vary, it is likely that they regulate the same or a similar suite of down-stream targets. Thus, we hypothesized that the topological and expression properties that characterize the known regulators of definitive erythropoiesis also should characterize equivalent regulators of primitive erythropoiesis; i.e., prior-knowledge about the definitive erythroid lineages could be used to test and validate computational predictions and then to moderate novel inferences about the regulation of the primitive erythroid lineage. With this in mind, the problem of predicting essential regulators of primitive erythropoiesis was considered a good fit for machine learning approaches and a task-specific algorithm was developed.
Our results revealed that key transcription factors in the definitive erythroid lineages could be discriminated by a combination of traits encompassing both the raw expression pattern and the architecture of the computationally inferred gene-interaction network. As expected, given the lack of the scale-free topology in our networks, local degree (k) provided little predictive power for identifying key regulators (W k = 0); however considering specific subsets of these connections (e.g., global connectivity, number of predicted targets) did assist in discriminating the reference gene set. Measures of cohesion (clustering coefficient; W CC = 0.98) and shortest-path centrality (W C = 0.14) were also informative for the highly inter-connected networks (Table 1).
Overall, the estimated essentiality score for a gene in the adult definitive erythroid lineage was not a good predictor of its score in the primitive erythroid lineage. Additionally, known essential and non-essential definitive erythroid regulators were not as well-differentiated in the fetal dataset as in the adult, emphasizing that the majority of genes were not consistently ranked between the lineages. This is not surprising as a subset of these reference regulators (e.g., Klf1, Gfi1b) are known to play different roles in the primitive versus definitive erythroid lineages; thus the scores of individual genes are expected to vary across the lineages and likely reflect the underlying biology. This observation was supported by our analysis; ~57% of the predicted potential key transcriptional regulators of primitive erythropoiesis are differentially expressed in primitive compared to adult definitive erythropoiesis.
The list of putative key transcriptional regulators of primitive erythropoiesis predicted by the GA and found to be differentially expressed between primitive and adult definitive erythropoiesis was enriched in genes activated downstream of MAPK-signaling. This included a striking signature of genes in the EPO signaling pathway, including the STAT-family genes. It has been shown in cell culture that EPO activates Stat1, Stat3, and Stat5a/b. Jak2-mediated phosphorylation of Stat5a/b is a core pathway mediating the EPO effect in erythroid cells; Jak2-deficiency in mice recapitulates the Epo- and Epor-null phenotype with an absolute block in definitive erythroblast production and fetal death by E12.5 . STAT5-deficient fetuses ultimately develop severe anemia and die in the perinatal period , but show no absolute block in definitive erythropoiesis or any known primitive erythroid defect, suggesting that other transcriptional regulators are also involved in mediating this critical signal and supporting our computational prediction of a differential role for STAT-signaling in primitive compared to definitive erythropoiesis.
Stat1 exhibits a pattern of increasing expression during erythroblast maturation specifically in the adult definitive erythroid lineage. Consistent with our computational finding, adult Stat1-null mice exhibit reduced numbers of CFU-E and elevated erythroblast apoptosis . There is no known effect of Stat1 deletion on primitive erythroblasts. Furthermore, Stat1 has been implicated as a necessary downstream mediator of IFNγ in the negative regulation of bone marrow erythropoiesis  and IFNs α, β, and γ have all been shown to negatively regulate definitive erythropoiesis [25, 30–32]. We find that genes involved in interferon-signaling are preferentially expressed in the adult definitive erythroid lineage (Figure 5), including Ifng, downstream apoptotic (Casp1, Tnfsf10) and anti-apoptotic (e.g., Bcl2l, Bcl2) genes, and genes involved in the negative regulation of cell proliferation (e.g., Cxcl10). This differential expression signature finds functional validation in our in vitro studies, which revealed that IFNγ inhibits definitive, but not primitive, erythroblast maturation.
The presence of Stat3 in our list of putative regulators was especially interesting as it is expressed at extremely low levels in the microarray dataset (Figure 3) and was, in fact, filtered out of prior analyses due to its low expression level . This gene was likely identified by the genetic algorithm as a putative regulator based on its topological prominence in the inferred regulatory network, emphasizing the information gained from considering multiple aspects of network topology. Stat3 is preferentially expressed in primitive erythropoiesis, with expression levels increasing gradually during later maturation stages (Figure 3). Although it has been shown that EPO induces tyrosine phosphorylation of Stat3[33, 34] and a potential role for this gene has been inferred in fetal definitive erythropoiesis through pathway analysis , activation of Stat3 is uncommon in hematopoietic cell lines . Here, the computationally predicted functional role for Stat3 in primitive, but not definitive, erythroid cell maturation is validated in vitro. Small molecule inhibition of Stat3 dimerization resulted in reduced numbers of erythroblasts late in the primitive erythroid culture, consistent with the increased expression of Stat3 during late stages of primitive erythroblast maturation.
Although primitive and definitive erythropoiesis share fundamental transcriptional regulators and result in the synthesis of terminally mature enucleated erythrocytes , they are fundamentally different processes. Definitive erythropoiesis in the adult is in steady state, continuously undergoing fine-tuned positive and negative regulation to maintain normal oxygen-carrying capacity. In contrast, primitive erythropoiesis emerges from the yolk sac and must transiently produce exponentially increasing numbers of erythroblasts to fill the newly-formed embryonic vasculature. We have identified the differential usage of Stat1 and Stat3, as well as interferon signaling, as defining characteristics of these lineages that may reflect opposing roles in the regulation of erythroid cell proliferation and survival.
The expression data used in this analysis were obtained from Affymetrix Mouse430_2 chip mRNA expression data from four progressive stages of erythroid maturation, specifically the proerythroblast (P), basophilic erythroblast (B), polychromatic/orthochromatic erythroblast (O), and reticulocyte (R) stages from three erythroid lineages: primitive, fetal definitive, and adult definitive (Array Express E-MTAB-1035) . Five biological replicates were performed for each maturational cell stage. Expression data were gcRMA normalized and MAS5 calls used to flag probe sets as expressed in the dataset only when present in a minimum of 3 out of 5 replicates for at least one maturational stage. Probe sets assigned an absent (A) call and any whose expression did not vary across replicates were also removed. Probe sets were mapped to EntrezGene identifiers and gene-level expression determined as the average across related probe sets.
Predicted transcription factor binding
Potential binding sites were predicted for 352 TFs by matching partial weight matrices (PWMs) to sequences within 1 kb up- or downstream of the promoter regions of all genes expressed in the microarray data. PWMs were obtained from the public version of TRANSFAC© and the freely available JASPAR databases. In addition, the CCNCNCCCN consensus sequence was used to identify potential targets of Klf1, a known key regulator of erythropoiesis . Motif and consensus sequence matching was performed using the Transcription Element Search System (TESS) . A maximum likelihood that a predicted site is a true binding site, or stringency, threshold ≥0.70 was adopted to identify the most likely predicted binding interactions between TFs and potential targets. The stringency of the best scoring match between a motif and matched sequence was used as a measure of binding potential (b ij ) between the transcription factor (i) and predicted target (j).
Within each lineage, Pearson correlation (r) was used as a measure of co-expression between the ordered expression profiles of all expressed gene-pairs across the set of 20 samples (5 replicates per cell stage). The significance of correlations was assessed using a t-statistic and those supported by a p-value ≤ 0.05 were used to estimate an interaction network by drawing edges between all significantly correlated gene-pairs. Self-associations and weak correlations (|r| < 0.50) were dropped. Edges were assigned a base-weight of |r ij |, or the absolute value of the Pearson correlation between factors i and j and then weighted by the estimated binding potential, b ij , between the two genes. Interactions supported solely by co-expression were treated as undirected.
Expression data, profiles, predicted transcription factor binding, and the inferred regulatory networks used in this analysis are all accessible through ErythronDB (http://www.cbil.upenn.edu/ErythronDB), a fully searchable public resource on murine erythrocyte maturation.
Machine learning identification of key regulators
Of genes expressed in the microarray dataset, we identified 1080 as putative transcriptional regulators using the Gene Ontology by selecting genes annotated by the following GO identifiers: GO:0003700, GO:0006350 and GO:0006351. We further identified eleven (11) properties (see Table 1 for a full listing), encapsulating aspects of expression, differential expression, and network topology that provide some insight into both the role and relative importance, or essentiality, of these transcription factors in the study system. Topological properties used in this analysis were chosen to capture multiple aspects of network architecture including local cohesiveness (clustering coefficient), shortest-path lengths (closeness centrality), and global dominance (degree centrality). In addition to these properties, we also considered other measures of dominance (e.g., betweenness centrality, or the number of shortest-paths on which a gene falls), and cohesiveness (e.g., the transitivity ratio, or the relative number of closed triplets associated with a gene), that were more computationally intensive. However, these measures did not well discriminate essential and non-essential regulators in initial trials and so not considered for the final analysis.
Lineage-specific values of each property were calculated for all TFs in expressed in our dataset. Values were then standardized to range from 0 to 1 to account for differences in scaling across the various measures.
It was not computationally feasible to assess the global topological prominence of each transcription factor in the estimated gene-interaction networks. Instead, fully-connected sub-networks for each TF and its neighbors were extracted and the topological properties for all TFs present in these “local-networks” calculated. We hypothesized that a key transcriptional regulator will be central and highly connected to its local network. We further postulated that essential factors should be prominent in the local networks of other key regulators as they likely serve as hubs between the connected sub-networks. Thus, here we take the modal value for each topological measure over all local-networks as an approximate measure of the global essentiality of the TF.
Essential Regulators: factors whose removal results in a complete block on hematopoiesis or erythropoiesis: Tal1, Gata1, Myb. Lmo2 was not included as it was not annotated with the selected GO terms at the time of running the algorithm.
Key Regulators: factors whose removal produces severe defects or prevents terminal differentiation of definitive erythroblasts: Klf1, Gfi1b, Zfpm1, Zbtb7a, Gata2.
Non-essential Regulators: factors that affect stress erythropoiesis or related developmental processes (e.g., megakaryopoiesis) but do not block or otherwise disrupt terminal differentiation of definitive erythroblasts: Foxo3, Ep300, Nfe2, Crebbp, Stat5a, Stat5b, Nfia, Fli1.
The GA was trained on the dataset of expression values and local-network topology derived from the adult definitive erythroid microarray expression dataset. Best solutions were then tested by applying the weighted equation to the properties and network derived from the fetal definitive expression dataset. GA parameters were systematically adjusted (e.g., recombination and mutation rates, population size) and training repeated until the solutions were able to discriminate known regulators in both the training and testing datasets. The weighted ranking equation was then applied to the primitive erythroid dataset to predict novel regulators of that lineage.
Lineage-specific log2 normalized expression profiles were clustered based on pairwise Pearson correlations. Hierarchical clustering and heatmap visualization were generated using GenePattern .
Cross-lineage differential expression
The pairwise cosine similarity was calculated between the adult definitive and primitive erythroid expression profiles of each transcription factor. Similarity values were ranked and genes whose cosine similarity was less than or equal to the median value of the distribution (cos ≤ 0.92) were considered significantly differentially expressed during the maturation of adult definitive compared to primitive erythroid cells.
Erythroid colony-forming assays
Outbred Swiss Webster mice (Taconic, Germantown, NY) were mated overnight and vaginal plugs checked the following morning (E0.3). E8.5 mouse embryos were dissociated with 0.25% trypsin (Worthington Biochemical, Lakewood, NJ, USA) to single cell suspensions and 1/10 yolk sac-equivalents were plated in duplicate in 1 ml IMDM, 1% methylcellulose (Stem Cell Technologies, Vancouver, BC, Canada), 5% PFHM-II (Invitrogen, Carlsbad, CA, USA), 10% serum replacement (Invitrogen), recombinant human erythropoietin (EPO; Amgen, Thousand Oaks, CA, USA) (2 U/ml), SCF (120 ng/ml), 2 mM MTG (Sigma-Aldrich, St. Louis, MO, USA), 2 mM glutamax (Invitrogen). EryP-CFC-derived colonies were counted after 5 days of culture at 37°C and 5% CO2, as described previously . Murine bone marrow was cultured at a density of 4–5 × 104 cells/ml in 1 ml of IMDM, 1% methylcellulose, 5% PFHM-II, 10% plasma-derived serum (PDS; Animal Technologies, Tyler, TX, USA), 20% BIT (Stem Cell Technologies), EPO (0.3 U/ml), 55 μM 2-ME (Gibco), 2 mM glutamax at 37°C and 5% CO2. CFU-E-derived colonies were enumerated at day 2 or 3 of culture.
Erythroblast maturation culture
Dissociated E8.5 embryos were cultured on 0.1% gelatin-coated plastic for 24 hours in primitive erythroid maturation media containing IMDM, 10% serum replacement, 10% PFHM-II, 2 mM glutamax, 150 uM MTG, 1% PDS, and 1 U/ml EPO. After 24 hours, the non-adherent, primitive erythroid cells were transferred to uncoated wells with fresh maturation media and cultured for up to a total of 4 days. Definitive, extensively self-renewing erythroblasts (ESRE) were generated as previously described . ESRE were induced to terminally mature in IMDM, 10% PFHM-II, 5% PDS, EPO (2 U/ml), 150 uM MTG, 2 mM glutamax at 37°C and 5% CO2.
Partial weight matrix
We thank Dr. Paul Kingsley for thoughtful discussions of erythroid gene regulation and Dr. Greg Grant for his insight on the analysis of co-expression networks. This work was supported by funds from the National Institute of Diabetes and Digestive Kidney Diseases (Erythroid Lineage Molecular Toolbox; NIDDK R01 DK 071116), the National Heart Lung and Blood Institute (R01 HL116364), and NRSA Institutional Predoctoral Training Grant T32 HL007152.
- Kingsley PD, Malik J, Fantauzzo KA, Palis J: Yolk sac-derived primitive erythroblasts enucleate during mammalian embryogenesis. Blood. 2004, 104: 19-25. 10.1182/blood-2003-12-4162.PubMedView ArticleGoogle Scholar
- Steiner R, Vogel H: On the kinetics of erythroid cell differentiation in fetal mice. I. Microspectrophotometric determination of the hemoglobin content in erythroid cells during gestation. J Cell Physiol. 1973, 81: 323-338. 10.1002/jcp.1040810305.PubMedView ArticleGoogle Scholar
- Trimborn T, Gribnau J, Grosveld F, Fraser P: Mechanisms of developmental control of transcription in the murine alpha- and beta-globin loci. Genes Dev. 1999, 13: 112-124. 10.1101/gad.13.1.112.PubMedPubMed CentralView ArticleGoogle Scholar
- Kingsley PD, Malik J, Emerson RL, Bushnell TP, McGrath KE, Bloedorn LA, Bulger M, Palis J: "Maturational" globin switching in primary primitive erythroid cells. Blood. 2006, 107: 1665-1672. 10.1182/blood-2005-08-3097.PubMedPubMed CentralView ArticleGoogle Scholar
- Perry C, Soreq H: Transcriptional regulation of erythropoiesis. Fine tuning of combinatorial multi-domain elements. Eur J Biochem. 2002, 15: 3607-3618.View ArticleGoogle Scholar
- Cantor AB, Orkin SH: Transcriptional regulation of erythropoiesis: an affair involving multiple partners. Oncogene. 2002, 21: 3368-3376. 10.1038/sj.onc.1205326.PubMedView ArticleGoogle Scholar
- Mucenski ML, McLain K, Kier AB, Swerdlow SH, Schreiner CM, Miller TA, Pietryga DW, Scott WJ, Potter SS: A functional c-myb gene is required for normal murine fetal hepatic hematopoiesis. Cell. 1991, 65: 677-689. 10.1016/0092-8674(91)90099-K.PubMedView ArticleGoogle Scholar
- Tober J, McGrath KE, Palis J: Primitive erythropoiesis and megakaryopoiesis in the yolk sac are independent of c-myb. Blood. 2008, 111: 2636-2639. 10.1182/blood-2007-11-124685.PubMedPubMed CentralView ArticleGoogle Scholar
- Saleque S, Cameron S, Orkin SH: The zinc-finger proto-oncogene Gfi-1b is essential for development of the erythroid and megakaryocytic lineages. Genes Dev. 2002, 16: 301-306. 10.1101/gad.959102.PubMedPubMed CentralView ArticleGoogle Scholar
- Wu H, Liu X, Jaenisch R, Lodish HF: Generation of committed erythroid BFU-E and CFU-E progenitors does not require erythropoietin or the erythropoietin receptor. Cell. 1995, 83: 59-67. 10.1016/0092-8674(95)90234-1.PubMedView ArticleGoogle Scholar
- Kieran MW, Perkins AC, Orkin SH, Zon L: Thrombopoietin rescues in vitro erythroid colony formation from mouse embryos lacking the erythropoietin receptor. Proc Natl Acad Sci USA. 1996, 93: 9126-9131. 10.1073/pnas.93.17.9126.PubMedPubMed CentralView ArticleGoogle Scholar
- Lin CS, Lim SK, D’Agati V, Costantini F: Differential effects of an erythropoietin receptor gene disruption on primitive and definitive erythropoiesis. Genes Dev. 1996, 10: 154-164. 10.1101/gad.10.2.154.PubMedView ArticleGoogle Scholar
- Kingsley PD, Greenfest-Allen E, Frame J, Bushnell T, Malik J, McGrath KE, Stoeckert CJ, Palis J: Ontogeny of erythroid gene expression. Blood. 2013, 121: e5-e13. 10.1182/blood-2012-04-422394.PubMedPubMed CentralView ArticleGoogle Scholar
- Oldham MC, Steve Horvath S, Geschwind DH: Conservation and evolution of gene coexpression networks in human and chimpanzee brains. Proc Natl Acad Sci USA. 2006, 103: 17973-17978. 10.1073/pnas.0605938103.PubMedPubMed CentralView ArticleGoogle Scholar
- Nayak RR, Kearns M, Spielman RS, Cheung VG: Coexpression network based on natural variation in human gene expression reveals gene interactions and functions. Genome Res. 2009, 19: 1953-1962. 10.1101/gr.097600.109.PubMedPubMed CentralView ArticleGoogle Scholar
- Zhao W, Langfelder P, Fuller T, Dong J, Li A, Hovarth S: Weighted gene coexpression network analysis: state of the art. J Biopharm Stat. 2010, 20: 281-300. 10.1080/10543400903572753.PubMedView ArticleGoogle Scholar
- Gursoy A, Keskin O, Nussinov R: Topological properties of protein interaction networks from a structural perspective. Biochem Soc Trans. 2008, 36: 1398-1403. 10.1042/BST0361398.PubMedView ArticleGoogle Scholar
- Horvath S, Dong J: Geometric interpretation of gene coexpression network analysis. LoS Comput Biol. 2008, 4: e1000117-10.1371/journal.pcbi.1000117.Google Scholar
- Pang K, Sheng H, Ma X: Understanding gene essentiality by finely characterizing hubs in the yeast protein interaction network. Biochem Biophys Res Commun. 2010, 401: 112-116. 10.1016/j.bbrc.2010.09.021.PubMedView ArticleGoogle Scholar
- Zotenko E, Mestre J, O’Leary DP, Przytycka TM: Why do hubs in the yeast protein interaction network tend to be essential: reexamining the connection between the network topology and essentiality. PLoS Comput Biol. 2008, 4: e1000140-10.1371/journal.pcbi.1000140.PubMedPubMed CentralView ArticleGoogle Scholar
- Huang DW, Sherman BT, Lempicki RA: Systematic and integrative analysis of large gene lists using DAVID Bioinformatics Resources. Nature Protoc. 2009, 4: 44-57.View ArticleGoogle Scholar
- Huang DW, Sherman BT, Lempicki RA: Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists. Nucleic Acids Res. 2009, 37: 1-13. 10.1093/nar/gkn923.PubMed CentralView ArticleGoogle Scholar
- Siddiquee KA, Gunning PT, Glenn M, Katt WP, Zhang S, Schrock C, Sebti SM, Jove R, Hamilton AD, Turkson J: An oxazole-based small-molecule Stat3 inhibitor modulates Stat3 stability and processing and induces antitumor cell effects. ACS Chem Biol. 2007, 2: 787-798. 10.1021/cb7001973.PubMedView ArticleGoogle Scholar
- Pang M, Ma L, Gong R, Tolbert E, Mao H, Ponnusamy M, Chin YE, Yan H, Dworkin LD, Zhuang S: A novel STAT3 inhibitor, S3I-201, attenuates renal interstitial fibroblast activation and interstitial fibrosis in obstructive nephropathy. Kidney Int. 2010, 78: 257-268. 10.1038/ki.2010.154.PubMedView ArticleGoogle Scholar
- Means RT, Krantz SB, Luna J, Marsters SA, Ashkenazi A: Inhibition of murine erythroid colony formation in vitro by interferon gamma and correction by interferon receptor immunoadhesin. Blood. 1994, 83: 911-915.PubMedGoogle Scholar
- Halupa A, Bailey ML, Huang K, Iscove NN, Levy DE, Barber DL: A novel role for STAT1 in regulating murine erythropoiesis: deletion of STAT1 results in overall reduction of erythroid progenitors and alters their distribution. Blood. 2005, 105: 552-561. 10.1182/blood-2003-09-3237.PubMedView ArticleGoogle Scholar
- Richmond TD, Chohan M, Barber DL: Turning cells red: signal transduction mediated by erythropoietin. Trends Cell Biol. 2005, 15: 146-155. 10.1016/j.tcb.2005.01.007.PubMedView ArticleGoogle Scholar
- Neubauer H, Cumano A, Müller M, Wu H, Huffstadt U, Pfeffer K: Jak2 deficiency defines an essential developmental checkpoint in definitive hematopoiesis. Cell. 1998, 93: 397-409. 10.1016/S0092-8674(00)81168-X.PubMedView ArticleGoogle Scholar
- Cui Y, Riedlinger G, Miyoshi K, Tang W, Li C, Deng CX, Robinson GW, Hennighausen L: Inactivation of Stat5 in mouse mammary epithelium during pregnancy reveals distinct functions in cell proliferation, survival, and differentiation. Mol Cell Biol. 2004, 24: 8037-8047. 10.1128/MCB.24.18.8037-8047.2004.PubMedPubMed CentralView ArticleGoogle Scholar
- Mizutani T, Tsuji K, Ebihara Y, Taki S, Ohba Y, Taniguchi T, Honda K: Homeostatic erythropoiesis by the transcription factor IRF2 through attenuation of type I interferon signaling. Exp Hematol. 2008, 36: 255-264. 10.1016/j.exphem.2007.11.004.PubMedView ArticleGoogle Scholar
- Yoshida H, Kawane K, Koike M, Mori Y, Uchiyama Y, Nagata S: Phosphatidylserine-dependent engulfment by macrophages of nuclei from erythroid precursor cells. Nature. 2005, 437: 754-758. 10.1038/nature03964.PubMedView ArticleGoogle Scholar
- Libregts SF, Gutiérrez L, de Bruin AM, Wensveen FM, Papadopoulos P, van Ijcken W, Ozgür Z, Philipsen S, Nolte MA: Chronic IFN-γ production in mice induces anemia by reducing erythrocyte life span and inhibiting erythropoiesis through an IRF-1/PU.1 axis. Blood. 2011, 118: 2578-2588. 10.1182/blood-2010-10-315218.PubMedView ArticleGoogle Scholar
- Kirito K, Nakajima K, Watanabe T, Uchida M, Tanaka M, Ozawa K, Komatsu N: Identification of the human erythropoietin receptor region required for Stat1 and Stat3 activation. Blood. 2002, 99: 102-110. 10.1182/blood.V99.1.102.PubMedView ArticleGoogle Scholar
- Cao Y, Lathia JD, Eyler CE, Wu Q, Li Z, Wang H, McLendon RE, Hjelmeland AB, Rich JN: Erythropoietin receptor signaling through STAT3 is required for glioma stem cell maintenance. Genes Cancer. 2010, 1: 50-61. 10.1177/1947601909356352.PubMedPubMed CentralView ArticleGoogle Scholar
- Cokic VP, Bhattacharya B, Beleslin-Cokic BB, Noguchi CT, Puri RK, Schechter AN: JAK-STAT and AKT pathway-coupled genes in erythroid progenitor cells through ontogeny. J Transl Med. 2012, 10: 116-10.1186/1479-5876-10-116.PubMedPubMed CentralView ArticleGoogle Scholar
- Miller IJ, Bieker JJ: A novel, erythroid cell-specific murine transcription factor that binds to the CACCC element and is related to the Kruppel family of nuclear proteins. Mol Cell Biol. 1993, 13: 2776-2786.PubMedPubMed CentralView ArticleGoogle Scholar
- Schug J: Using TESS to predict transcription factor binding. Curr Protoc Bioinform. 2003, 21: 2.6.1-2.6.15.Google Scholar
- Ashyraliyev M, Fomekong-Nanfack Y, Kaandorp JA, Blom JG: Systems biology: parameter estimation for biochemical models. FEBS J. 2009, 276: 886-902. 10.1111/j.1742-4658.2008.06844.x.PubMedView ArticleGoogle Scholar
- Reich M, Liefeld T, Gould J, Lerner J, Tamayo P, Mesirov JP: GenePattern 2.0. Nat Genet. 2006, 38: 500-501. 10.1038/ng0506-500.PubMedView ArticleGoogle Scholar
- Palis J, Robertson S, Kennedy M, Wall C, Keller G: Development of erythroid and myeloid progenitors in the yolk sac and embryo proper of the mouse. Development. 1999, 126: 5073-5084.PubMedGoogle Scholar
- England SJ, McGrath KE, Frame JM, Palis J: Immature erythroblasts with extensive ex vivo self-renewal capacity emerge from the early mammalian fetus. Blood. 2011, 117: 2708-2717. 10.1182/blood-2010-07-299743.PubMedPubMed CentralView ArticleGoogle Scholar
- Watts DJ, Strogatz SH: Collective dynamics of 'small-world' networks. Nature. 1998, 393: 440-442. 10.1038/30918.PubMedView ArticleGoogle Scholar
- Sabidussi G: The centrality index of a graph. Psychometrika. 1966, 31: 58-603.View ArticleGoogle Scholar
- Dijkstra EW: A note on two problems in connexion with graphs. Numerische Mathematik. 1959, 1: 269-271. 10.1007/BF01386390.View ArticleGoogle Scholar
- Siek JG, Lee L, Lumsdaine A: The boost graph library: user guide and reference manual. 2002, USA: Pearson Education, IncGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.