TY - JOUR
AU - Hibbs, Matthew A.
AB - Abstract Embryonic stem cells (ESCs), characterized by their ability to both self‐renew and differentiate into multiple cell lineages, are a powerful model for biomedical research and developmental biology. Human and mouse ESCs share many features, yet have distinctive aspects, including fundamental differences in the signaling pathways and cell cycle controls that support self‐renewal. Here, we explore the molecular basis of human ESC self‐renewal using Bayesian network machine learning to integrate cell‐type‐specific, high‐throughput data for gene function discovery. We integrated high‐throughput ESC data from 83 human studies (∼1.8 million data points collected under 1,100 conditions) and 62 mouse studies (∼2.4 million data points collected under 1,085 conditions) into separate human and mouse predictive networks focused on ESC self‐renewal to analyze shared and distinct functional relationships among protein‐coding gene orthologs. Computational evaluations show that these networks are highly accurate, literature validation confirms their biological relevance, and reverse transcriptase polymerase chain reaction (RT‐PCR) validation supports our predictions. Our results reflect the importance of key regulatory genes known to be strongly associated with self‐renewal and pluripotency in both species (e.g., POU5F1, SOX2, and NANOG), identify metabolic differences between species (e.g., threonine metabolism), clarify differences between human and mouse ESC developmental signaling pathways (e.g., leukemia inhibitory factor (LIF)‐activated JAK/STAT in mouse; NODAL/ACTIVIN‐A‐activated fibroblast growth factor in human), and reveal many novel genes and pathways predicted to be functionally associated with self‐renewal in each species. These interactive networks are available online at www.StemSight.org for stem cell researchers to develop new hypotheses, discover potential mechanisms involving sparsely annotated genes, and prioritize genes of interest for experimental validation. Stem Cells 2014;32:1161–1172 Embryonic stem cells, Pluripotent stem cells, Biomathematical modeling, Cell signaling, Genomics, Computational biology Introduction Embryo‐derived pluripotent stem cells are a powerful model system for biomedical research and the study of developmental biology. Among the most studied embryo‐derived stem cells are human and mouse embryonic stem cells (hESCs and mESCs). Since first isolated from human embryos in 1998 [1], hESCs have been of particular interest to the research community as a tool for manipulating cell fate, analyzing characteristics of induced pluripotent stem cells (iPSCs) and cancer‐stem‐like cells, and testing potential medical and pharmaceutical applications [2]. All ESCs express common markers of pluripotency, such as POU5F1 (also known as OCT4), SOX2, and NANOG, but there are many known differences between species, including morphological, molecular, and epigenetic characteristics that are reflected in ESCs grown in culture. hESCs are derived from the inner cell mass (ICM) of blastocysts 5–8 days post fertilization and form flat, two‐dimensional round colonies with well‐defined boundaries, whereas mESCs are isolated from the ICM of blastocysts 3.5–4.5 days post fertilization and form tight, dome‐shaped colonies [3]. hESCs respond to cooperative signaling between NODAL/ACTIVIN‐A‐activated fibroblast growth factor (FGF) and transforming growth factor (TGF)‐β signaling pathways to sustain prolonged self‐renewal in vitro [1, 3, 4]. In contrast, mESCs grown in culture require growth factors, leukemia inhibitory factor (LIF) and bone morphogenetic protein 4 (BMP4), to activate JAK/STAT signaling [5-7]. Self‐renewal in mESCs can be boosted by small molecule inhibitors that block differentiation cues from the FGF/ERK signaling cascade and mimic WNT/β‐catenin signaling [8]. Interestingly, the same culture conditions that support self‐renewal in mESCs can drive differentiation in hESCs. For example, hESCs exposed to LIF and BMP4 yield extraembryonic phenotypes, and FGF inhibition promotes neuroectoderm commitment [9]. In general, well‐defined, standard protocols exist for growing mESCs in culture using cell lines of similar genetic backgrounds (predominantly derived from 129S/P/T substrains) [10, 11]. However, hESC lines have been derived from genetically distinct embryos in multiple laboratories, each using different media cocktails and protocols to promote self‐renewal and the pluripotent state; consequently individual hESC lines may respond differently when grown under the same culture conditions [12]. mESCs share epigenetic traits of preimplantation blastocysts and are said to be at the “naive” or primitive developmental ground state of pluripotency, before X‐chromosome inactivation and genomic imprinting [11]. These naive ESCs show no differentiation bias, can self‐renew indefinitely in vitro, and can be expanded clonally without compromising the pluripotent state [13]. In addition, mESCs can be genetically modified, then reintroduced into preimplantation embryos to generate high‐grade chimeric mice [11, 14]. In contrast, hESCs are said to be “primed” for differentiation, and female cells have typically undergone random inactivation of one X chromosome [11, 15]; however, there is variability in imprinted genes and other DNA methylation patterns depending on culture conditions and passage number [15, 16]. Intriguingly, hESCs share many molecular and epigenetic characteristics with mouse pluripotent stem cells isolated from the post‐implantation epiblast (mEpiSCs) [13, 14, 17], leading to the hypothesis that hESCs and mEpiSCs are both “primed.” Although the chimerism assay may not be used with human cells for ethical reasons, studies have shown that lower primate ESCs cannot produce high‐grade chimeras when injected into blastocysts, nor can they contribute to the germline [13, 18], indicating that lineage potential is limited in vivo. While these phenotypic contrasts are clear, the molecular foundations of naive versus primed ESCs have not yet been fully characterized. Understanding the differences between pluripotent cell types is of increasing interest and value as we develop new methods to “reprogram” adult cells to an ESC‐like state for ongoing research or therapeutic applications. In vitro methods used to reprogram different types of human and mouse adult cells into iPSCs vary widely, use different combinations of reprogramming factors (e.g., POU5F1, SOX2, KLF4, KLF2, MYC, and LIN28A) and experimental conditions, and yield ESC‐like cells with different degrees of developmental potential [19-21]. In mouse iPSCs, Nanog is critical for both blocking differentiation and achieving a naive pluripotent state [9]. Human iPSCs have also been shown to require NANOG to block differentiation, but, to date, attempts to derive or reprogram hESCs to a naive state have been unsuccessful [13]. Because naive mESCs and primed hESCs respond to different signaling pathways to sustain and exit the self‐renewing state [13], cross‐species systems‐level analyses of these cell types can reveal molecular details that will assist researchers in assessing the pluripotent state of embryo‐derived cells and reprogrammed adult cells, and reveal novel functional homologs that support self‐renewal and related early developmental processes. While many systems biology studies have been conducted to explore the molecular basis of self‐renewal and pluripotency through gene expression profiles or regulatory networks of transcription factor binding, these efforts have been largely species‐specific, restricted to data from a small number of cell lines, limited to a single experimental platform, and/or focused on a single regulatory characteristic influencing cell fate [3, 9, 22-24]. The goals of this study are to discover potential novel regulators of ESC renewal in hESCs and perform cross‐species comparative analyses to further our understanding of shared and distinct molecular characteristics of hESCs and mESCs. We applied a Bayesian network (Bayes net) machine learning approach based on species‐ and cell‐type‐specific data integration, which we previously applied to mESCs [25], to produce a consensus network that predicts gene function associations in the context of hESC self‐renewal. Bayes nets, a type of supervised machine learning, are particularly useful for gene function discovery as they provide a statistically principled method to model relationships among genes given a solid foundation of biological knowledge [26-31]. To further our understanding of commonalities and differences between human and mouse ESCs, we generated comparative probabilistic networks for ESC self‐renewal using a training set that included examples experimentally validated in hESCs, mESCs, or both species. The results of this study yield insights into novel genes involved in hESC self‐renewal and provide new comparisons of the molecular characteristics of the most widely studied model ESC systems. Materials and Methods Master Gene List and Training Set Construction Using a homology report from Mouse Genome Informatics [32], we created a master gene list of 17,342 protein‐coding mouse and human gene orthologs with one‐to‐one homology associations (Supporting Information Table S1) [33]. Our positive training examples (Supporting Information Table S2) consisted of 2,077 manually curated pair‐wise gene relationships involving 365 genes associated with mouse and/or human ESC self‐renewal based on 108 recent journal articles (Supporting Information Table S3), or annotated to early embryonic developmental signaling pathways by the Kyoto Encyclopedia of Genes and Genomes (KEGG) [34] and WikiPathways [35]. Negative/background training examples were generated in a 1:10 ratio by randomly selecting 20,770 gene pairs not involving any genes in positive training examples (Supporting Information Table S4). Evidential Data Set Collection We assembled a compendium of high‐throughput hESC data, representing 83 independent research studies and consisting of ∼1.8 million data points from 55 hESC lines (though predominantly H1 and H9), collected under 1,100 conditions, using 6 different high‐throughput data types, and encompassing more than 12 billion gene‐pair measurements (Table 1; Supporting Information Table S5). For mESCs, we updated our existing compendium of high‐throughput mESC data [25] to include work from 62 independent research studies, consisting of ∼2.3 million data points from 35 mESC lines (all derived from 129S/P/T substrains), spanning 1,085 conditions, using 5 different high‐throughput data types, and encompassing more than 7 billion gene‐pair measurements (Table 1; Supporting Information Table S6). Summary of integrated data in human and mouse ESC data compendiums Data type . Data sets/platforms . Genes measured . Conditions . Gene pairs . Redundancy . hESC data compendium Gene expression 72/16 1,206,852 597 10,409,092,802 0.00872 Protein‐DNA interactions 28/8 383,784 74 702,255,900 0.02718 Protein–protein interactions 3/3 13,648 3 154,876 0.05047 Epigenetic markers 9/1 152,946 194 1,299,505,689 0.01053 Phylogenetic profiles 1/1 2,391 229 3,267 3.9E‐6 RNAi screens 3/1 42,570 3 44,618,430 0.0195 mESC data compendium Gene expression 61/22 1,028,918 872 5,309,037,873 0.7110 Protein–DNA interactions 102/14 1,311,167 210 1,507,210,159 0.0218 Protein–protein interactions 1/1 207 1 207 2.2E‐06 Phylogenetic profiles 1/1 15,703 1 123,284,253 0.1728 RNAi screens 1/1 16,891 2 131,795,730 0.1954 Data type . Data sets/platforms . Genes measured . Conditions . Gene pairs . Redundancy . hESC data compendium Gene expression 72/16 1,206,852 597 10,409,092,802 0.00872 Protein‐DNA interactions 28/8 383,784 74 702,255,900 0.02718 Protein–protein interactions 3/3 13,648 3 154,876 0.05047 Epigenetic markers 9/1 152,946 194 1,299,505,689 0.01053 Phylogenetic profiles 1/1 2,391 229 3,267 3.9E‐6 RNAi screens 3/1 42,570 3 44,618,430 0.0195 mESC data compendium Gene expression 61/22 1,028,918 872 5,309,037,873 0.7110 Protein–DNA interactions 102/14 1,311,167 210 1,507,210,159 0.0218 Protein–protein interactions 1/1 207 1 207 2.2E‐06 Phylogenetic profiles 1/1 15,703 1 123,284,253 0.1728 RNAi screens 1/1 16,891 2 131,795,730 0.1954 Notes: These data were collected from 83 hESC studies and 62 mESC studies, then standardized and integrated into a pair‐wise format, and used as evidential data to generate predictive hESC‐ and mESC‐specific networks focused on ESC self‐renewal. Data sets were weighted based on the amount of shared mutual information (redundancy) contained in each as compared to all evidential data sets used by the Bayes net. A low mean redundancy indicates the data set is highly unique. Data set contributions to the top 0.001% (1,503) of edges in the hESC and mESC networks are available in Supporting Information Table S10. Abbreviations: ESC, embryonic stem cell; hESC, human ESCs; mESC, mouse ESCs. Open in new tab Summary of integrated data in human and mouse ESC data compendiums Data type . Data sets/platforms . Genes measured . Conditions . Gene pairs . Redundancy . hESC data compendium Gene expression 72/16 1,206,852 597 10,409,092,802 0.00872 Protein‐DNA interactions 28/8 383,784 74 702,255,900 0.02718 Protein–protein interactions 3/3 13,648 3 154,876 0.05047 Epigenetic markers 9/1 152,946 194 1,299,505,689 0.01053 Phylogenetic profiles 1/1 2,391 229 3,267 3.9E‐6 RNAi screens 3/1 42,570 3 44,618,430 0.0195 mESC data compendium Gene expression 61/22 1,028,918 872 5,309,037,873 0.7110 Protein–DNA interactions 102/14 1,311,167 210 1,507,210,159 0.0218 Protein–protein interactions 1/1 207 1 207 2.2E‐06 Phylogenetic profiles 1/1 15,703 1 123,284,253 0.1728 RNAi screens 1/1 16,891 2 131,795,730 0.1954 Data type . Data sets/platforms . Genes measured . Conditions . Gene pairs . Redundancy . hESC data compendium Gene expression 72/16 1,206,852 597 10,409,092,802 0.00872 Protein‐DNA interactions 28/8 383,784 74 702,255,900 0.02718 Protein–protein interactions 3/3 13,648 3 154,876 0.05047 Epigenetic markers 9/1 152,946 194 1,299,505,689 0.01053 Phylogenetic profiles 1/1 2,391 229 3,267 3.9E‐6 RNAi screens 3/1 42,570 3 44,618,430 0.0195 mESC data compendium Gene expression 61/22 1,028,918 872 5,309,037,873 0.7110 Protein–DNA interactions 102/14 1,311,167 210 1,507,210,159 0.0218 Protein–protein interactions 1/1 207 1 207 2.2E‐06 Phylogenetic profiles 1/1 15,703 1 123,284,253 0.1728 RNAi screens 1/1 16,891 2 131,795,730 0.1954 Notes: These data were collected from 83 hESC studies and 62 mESC studies, then standardized and integrated into a pair‐wise format, and used as evidential data to generate predictive hESC‐ and mESC‐specific networks focused on ESC self‐renewal. Data sets were weighted based on the amount of shared mutual information (redundancy) contained in each as compared to all evidential data sets used by the Bayes net. A low mean redundancy indicates the data set is highly unique. Data set contributions to the top 0.001% (1,503) of edges in the hESC and mESC networks are available in Supporting Information Table S10. Abbreviations: ESC, embryonic stem cell; hESC, human ESCs; mESC, mouse ESCs. Open in new tab Construction of Bayes Nets and Inference of Posterior Functional Relationship Scores We trained separate naive Bayes nets using the same gene list and training set, but with species‐specific ESC data compendiums, to predict the probability of functional associations among 150,363,811 protein‐coding gene pairs based on patterns observed in the integrated evidential data. Learning was achieved by computing the posterior probability of a functional relationship between training set gene pairs given all evidential data [36-39] as described previously (equation 1) [25]. P(FR\E1,E2,...En)=1ZP(FR)∏i=1nP(Ei|FR) where FR is a binary hidden variable representing whether a gene pair is functionally related, P(FR = 1) is the predicted probability that a pair is functionally related, Ei represents the evidence score of the gene pair for the ith data set, and Z is a normalization factor. Naive Bayes nets impose a strict assumption of independence between evidence data that is likely violated by many of our input data sets, as such, we performed regularization to quantitatively correct for the amount of redundant information contained in each data set as compared to all other data sets in the compendium (equations 2, 3) as described previously [25]. Sk=1+H(Dk)−1∑i≠kI(Di;Dk) P(FRi,j|E1,E2,...En)=1Z∏k=1nαP[Dk=dk(g1g2)]+αSk−1α+|Dk|αSk−1 where Sk is a heuristic sum of shared information relative to the data set’s entropy used to weight the strength of prior belief in a uniform distribution for the data set, H refers to Shannon entropy, and I(Di;Dk) refers to mutual information. This ultimately results in equation 3 (a variation of equation 1), such that P(FRij|E1, E2, … En) is the predicted probability that there is a functional relationship between genes i and j given evidence in data sets 1 through n, Z is a normalization factor, α is a pseudocount regularization parameter used to modulate the strength of regularization (higher pseudocount values weaken influence of redundant data sets), and Dk is the number of bins used to discretize continuous data values in data set K. A low Sk indicates that the information contained in the data set is highly unique, while a high score indicates that the data sets contained shared (redundant) information. The redundancy score for each data set used to train the Bayesian classifier is listed in Supporting Information Tables S5 (hESC) and S6 (mESC). We conducted performance tests to evaluate effects of regularization and selected a pseudocount value of 70 to regularize mESC evidential data, and a value of 100 for hESC evidential data to achieve similar posterior probability distributions. Computational Performance Assessment To assess biological content and functional relevance of our networks, we used functional genomics tools [32, 40] to evaluate gene ontology (GO) term enrichment [41], validate that known gene pairs were strongly connected in the probabilistic network, and identify novel genes with strong functional linkages supported by evidential data. These computational validations, including standard machine learning metrics and cross validation, show that our networks are highly accurate and a powerful tool for comparative analysis (Supporting Information Fig. S1; additional details in Supporting Information File S1; Table S7). Network Topology and Correlation Analyses To analyze the network topology, we calculated the scaled degree (Ki) for each gene as described previously [25, 42]. To determine genes strongly associated with self‐renewal (rather than the general connectivity measured by degree), we identified a set of “core” self‐renewal genes from our positive training set for each species as those genes tightly connected to other training set genes as described previously (173 mESC genes; 111 hESC genes) [25]. Then for every gene in the genome, we calculated the average posterior probability of functional relationship to these “core” self‐renewal genes, which we refer to as the self‐renewal correlation (SRC) score (Supporting Information Table S8). We used these SRC values to rank genes from most strongly correlated to self‐renewal (ranks near 1) to least correlated to self‐renewal (ranks near 17,342) (Supporting Information Table S9). We produced a differential network by subtracting the posterior edge weights for each gene pair (mESC minus hESC probability) to identify pairs strongly associated in one network, but not the other. Spearman correlation coefficients were used to determine the conservation of edge weights between networks (Supporting Information Table S10). Experimental Validation H1 hESCs were obtained from WiCell Research Institute (Madison, WI; http://www.wicell.org). The hESCs (passages 30–50) were grown on mitotic‐inactivated mouse embryonic fibroblasts (MEFs) in ES medium containing Dulbecco’s modified Eagle’s medium/F‐12 (Invitrogen, Carlsbad, CA; http://www.lifetechnologies.com), 20% knockout serum replacement (KSR, Invitrogen), 0.1 mM nonessential amino acids (Invitrogen), 2 mM l‐glutamine (Mediatech, Inc, Herndon, VA; http://www.cellgro.com/), 0.1 mM β‐mercaptoethanol (Sigma, St. Louis, MO; http://www.sigmaaldrich.com), and 4 ng/ml FGF2 (PeproTech, Rocky Hill, NJ; http://www.peprotech.com) [43, 44]. Embryoid body (EB) differentiation was carried out as described previously [45]. EBs at days 4 and 6 were harvested for RNA isolation. Quantitative reverse transcriptase polymerase chain reaction was performed and adjusted to yield equal amplification of glyceraldehyde‐3‐phosphate dehydrogenase as an internal standard. Further materials and methods details are provided in Supporting Information File S1. Results and Discussion We used a cell‐type‐specific naive Bayes network methodology (Fig. 1) [25] to create separate probabilistic biological networks for hESCs and mESCs focused on self‐renewal and closely related biological processes (e.g., pluripotency and cell fate determination). As input data, our Bayes nets take: (1) a training set of prior knowledge (also known as a gold standard) comprised of protein‐coding gene pairs known to be functionally related (positive training examples) and pairs believed to be unrelated (negative/background training examples), and 2) independent, whole‐genome high‐throughput data sets (observed evidential data). Based on these inputs, Bayes nets “learn” significant patterns in the evidence, assess data reliability, and then probabilistically predict novel relationships among protein‐coding genes based on the most reliable data [46, 47]. (Additional details are given in Supporting Information Fig. S1 and File S1.) The resulting networks consisted of 150,363,811 gene pairs that predicted the strength of functional associations among 17,342 protein‐coding gene orthologs based on patterns observed in integrated genomic data. Computational evaluations showed that these networks were highly accurate (Supporting Information Fig. S1). The topology of both the human and mouse networks show that a subset of genes (roughly 10%, or ∼1,700) are highly connected, as determined by scaled degree (a measure of global connectivity for each gene), while the majority of genes are less well connected. Furthermore, the connectivity profiles are similar between both the human and mouse networks (Supporting Information Figure S2, full results in Supporting Information Table S8). We used GO term enrichment and network connectivity metrics to assess biological content and functional relevance of our these networks, validate that gene pairs known to significantly influence mESC self‐renewal were strongly connected in these probabilistic networks, and identify novel genes with strong functional linkages supported by evidential data (Supporting Information Tables S7, S8). Open in new tabDownload slide Bayes net machine learning methodology for cell type specific comparative networks. To produce reliable, biologically relevant comparative predictive networks for mouse and human embryonic stem cells (hESCs), we adapted our approach for cell‐type‐specific data integration and machine learning [25] as follows: (A): Prepared a master list of protein‐coding mouse and human gene orthologs. (B): Curated a set of training examples focused on ESC self‐renewal in a pairwise format. Positive (known) examples were extracted from recent literature and curated pathways; negative/background examples were randomly generated from the master gene list, excluding genes involved in positive edges. (C): Collected and standardized high‐throughput data from 83 human and 62 mouse ESC studies using diverse high‐throughput data types (Table 1). Used distance/correlation metrics to distill data into a pairwise format. (D): Iteratively tested and validated species‐specific predictive networks for comparative analysis using the same training set, but with species‐specific data compendiums as input. Abbreviations: ESC, embryonic stem cell; hESC, human ESCs; mESC, mouse ESCs. Open in new tabDownload slide Bayes net machine learning methodology for cell type specific comparative networks. To produce reliable, biologically relevant comparative predictive networks for mouse and human embryonic stem cells (hESCs), we adapted our approach for cell‐type‐specific data integration and machine learning [25] as follows: (A): Prepared a master list of protein‐coding mouse and human gene orthologs. (B): Curated a set of training examples focused on ESC self‐renewal in a pairwise format. Positive (known) examples were extracted from recent literature and curated pathways; negative/background examples were randomly generated from the master gene list, excluding genes involved in positive edges. (C): Collected and standardized high‐throughput data from 83 human and 62 mouse ESC studies using diverse high‐throughput data types (Table 1). Used distance/correlation metrics to distill data into a pairwise format. (D): Iteratively tested and validated species‐specific predictive networks for comparative analysis using the same training set, but with species‐specific data compendiums as input. Abbreviations: ESC, embryonic stem cell; hESC, human ESCs; mESC, mouse ESCs. hESC Network Predicts Novel Self‐Renewal Genes Functional associations to known self‐renewal genes (SRC scores, see Materials and Methods section) were calculated for all 17,342 genes in the hESC network and sorted to determine SRC ranks. As expected, many well‐studied genes in our training examples are highly ranked by this metric, including POU5F1, SOX2, and NODAL (Supporting Information Table S8). The top 12 novel genes (not included in the positive training set) identified by SRC ranks were TUBB, HSP90AB1, PELI1, SEMA6A, SLC7A5, RGMB, PCBP1, TUBB6, HSPA4, SFRP1, LRRN1, and PRDM14 (Table 2). Literature validation of these genes confirmed that PRMD14 has recently been shown to play a role in hESC self‐renewal and pluripotency [48, 49], and several have been implicated in cancers, including HSP90AB1, SFRP1, SEMA6A, and PRDM14 [50-53]. TUBB and TUBB6 are β‐tubulin genes, which could play a role in hESC colony morphology as β‐tubulins help define cell shape [54] and could regulate self‐renewal through nuclear positioning [55]. Several other highly ranked genes are known to be involved in embryonic developmental signaling pathways: SFRP1 has been shown to play a key regulatory role in WNT receptor signaling and has been identified as a downstream target of sonic hedgehog (SHH) signaling [56, 57], SEMA6 has been associated with noncanonical WNT receptor signaling and planar cell polarity signaling [52], and RGMB has been shown to play a regulatory role in BMP signaling [58]. In further support of these predictions, we measured the expression of our top novel genes (excluding β‐tubulins) in undifferentiated H1 hESC lines, and during differentiation into EBs. In all cases, expression levels of our novel candidate genes were significantly reduced in the differentiated state compared with the self‐renewing state (Supporting Information Figure S3). Top novel hESC genes most strongly correlated with self‐renewal Gene symbol . Gene name . Rank . SRC (scaled) . TUBB Tubulin, beta class I 1 0.8264 HSP90AB1 Heat shock protein 90 kDa alpha (cytosolic), class B member 1 2 0.8053 PELI1 Pellino homolog 1 (Drosophila) 3 0.7805 SEMA6A Sema domain, transmembrane domain, and cytoplasmic domain, (semaphorin) 6A 4 0.7795 SLC7A5 Solute carrier family 7 (cationic amino acid transporter, y+ system), member 5 5 0.7693 RGMB RGM domain family, member B 6 0.7684 PCBP1 Poly(rC) binding protein 1 7 0.7669 TUBB6 Tubulin, beta 6 class V 8 0.7184 HSPA4 Heat shock 70 kDa protein 4 9 0.7653 SFRP1 Secreted frizzled‐related protein 1 10 0.7587 LRRN1 Leucine‐rich repeat neuronal 1 11 0.7574 PRDM14 PR domain containing 14 12 0.7458 Gene symbol . Gene name . Rank . SRC (scaled) . TUBB Tubulin, beta class I 1 0.8264 HSP90AB1 Heat shock protein 90 kDa alpha (cytosolic), class B member 1 2 0.8053 PELI1 Pellino homolog 1 (Drosophila) 3 0.7805 SEMA6A Sema domain, transmembrane domain, and cytoplasmic domain, (semaphorin) 6A 4 0.7795 SLC7A5 Solute carrier family 7 (cationic amino acid transporter, y+ system), member 5 5 0.7693 RGMB RGM domain family, member B 6 0.7684 PCBP1 Poly(rC) binding protein 1 7 0.7669 TUBB6 Tubulin, beta 6 class V 8 0.7184 HSPA4 Heat shock 70 kDa protein 4 9 0.7653 SFRP1 Secreted frizzled‐related protein 1 10 0.7587 LRRN1 Leucine‐rich repeat neuronal 1 11 0.7574 PRDM14 PR domain containing 14 12 0.7458 Notes: Genes were identified by selecting the top novel genes (those not involved in positive training set edges) rank ordered by our SRC score as described in Materials and Methods section and Supporting Information File S1. Abbreviations: hESC, human embryonic stem cells; SRC, self‐renewal correlation. Open in new tab Top novel hESC genes most strongly correlated with self‐renewal Gene symbol . Gene name . Rank . SRC (scaled) . TUBB Tubulin, beta class I 1 0.8264 HSP90AB1 Heat shock protein 90 kDa alpha (cytosolic), class B member 1 2 0.8053 PELI1 Pellino homolog 1 (Drosophila) 3 0.7805 SEMA6A Sema domain, transmembrane domain, and cytoplasmic domain, (semaphorin) 6A 4 0.7795 SLC7A5 Solute carrier family 7 (cationic amino acid transporter, y+ system), member 5 5 0.7693 RGMB RGM domain family, member B 6 0.7684 PCBP1 Poly(rC) binding protein 1 7 0.7669 TUBB6 Tubulin, beta 6 class V 8 0.7184 HSPA4 Heat shock 70 kDa protein 4 9 0.7653 SFRP1 Secreted frizzled‐related protein 1 10 0.7587 LRRN1 Leucine‐rich repeat neuronal 1 11 0.7574 PRDM14 PR domain containing 14 12 0.7458 Gene symbol . Gene name . Rank . SRC (scaled) . TUBB Tubulin, beta class I 1 0.8264 HSP90AB1 Heat shock protein 90 kDa alpha (cytosolic), class B member 1 2 0.8053 PELI1 Pellino homolog 1 (Drosophila) 3 0.7805 SEMA6A Sema domain, transmembrane domain, and cytoplasmic domain, (semaphorin) 6A 4 0.7795 SLC7A5 Solute carrier family 7 (cationic amino acid transporter, y+ system), member 5 5 0.7693 RGMB RGM domain family, member B 6 0.7684 PCBP1 Poly(rC) binding protein 1 7 0.7669 TUBB6 Tubulin, beta 6 class V 8 0.7184 HSPA4 Heat shock 70 kDa protein 4 9 0.7653 SFRP1 Secreted frizzled‐related protein 1 10 0.7587 LRRN1 Leucine‐rich repeat neuronal 1 11 0.7574 PRDM14 PR domain containing 14 12 0.7458 Notes: Genes were identified by selecting the top novel genes (those not involved in positive training set edges) rank ordered by our SRC score as described in Materials and Methods section and Supporting Information File S1. Abbreviations: hESC, human embryonic stem cells; SRC, self‐renewal correlation. Open in new tab Human and Mouse ESC Networks Are Largely Similar and Significantly Correlated To further evaluate our novel hESC network, we performed a comparative analysis with an updated version of our previous mESC network [25], generated using the same master list of protein‐coding gene orthologs and training set. Differential network analysis showed that the mESC and hESC networks are highly correlated, and most genes share the same functional partners/interactors, as indicated by the posterior probability of edges (Spearman’s rank correlation coefficient ρ = .43, p<1 × 10−300; Supporting Information Fig. 4A). In particular, positive training set edges were significantly correlated between species (Spearman’s rank correlation coefficient ρ = .4435, p = 1.08−93). Our results confirm the conserved roles of genes known to be involved in early developmental transcriptional regulation and stem cell maintenance in both species, including POU5F1/Pou5f1 (hESC rank: 1/mESC rank: 1), SOX2/Sox2 (2/5), and NANOG/Nanog (128/56). The top 1% of genes most conserved in both the mouse and human ESC networks significantly overlap (hypergeometric p = 9.17−15) and were highly enriched for ESC‐related biological processes, including stem cell maintenance (GO:0010074), negative regulation of cell differentiation (GO:0045596), cell fate commitment (GO:0045165), regulation of transcription, DNA‐dependent (GO:0042127), transcription factor activity (GO:0003712) and regulation of gene expression (GO:0010468) using Mus musculus (laboratory mouse) GO annotations (Table 3, Supporting Information Tables S7, S8). We observed some species‐specific annotation biases [33] in that the same set of genes evaluated using Homo sapiens GO annotations indicated enrichment for transcription factor activity, regulation of transcription, and WNT receptor signaling pathway (GO:0016055). Functional similarities and differences between networks Shared . Distinct . Gene (h/m) . Rank (h/m) . hESC gene . Rank (h/m) . mESC gene . Rank (h/m) . POU5F1/Pou5f1 1/1 PELI1 5/8,197 Utf1 7,906/6 SOX2/Sox2 2/5 LRRN1 15/11,119 Tcl1 14,166/17 NODAL/Nodal 12/114 TUBG1 32/7,004 Zfp428 13,991/22 RIF1/Rif1 1,433 RPL10A 37/8,606 Gjb3 13,495/23 PARP1/Parp1 21/97 RPLP2 38/7,284 Tdh 11,393/26 LEFTY2/Lefty2 33/85 TALDO1 63/8,745 Fbxo15 14,921/35 MYC/Myc 40/38 RPS3A 65/11,303 Zfp57 13,784/63 JARID2/Jarid2 46/11 RPL6 101/11,695 Eras 13,432/92 ZFP42/Zfp42 78/27 EIF3A 115/12,160 Dusp27 14,812/134 IFITM1/Ifitm1 107/15 USO1 151/11,350 Enpp3 16,752/169 NANOG/Nanog 128/56 GPM6B 419/13,693 Spats1 15,033/172 Shared . Distinct . Gene (h/m) . Rank (h/m) . hESC gene . Rank (h/m) . mESC gene . Rank (h/m) . POU5F1/Pou5f1 1/1 PELI1 5/8,197 Utf1 7,906/6 SOX2/Sox2 2/5 LRRN1 15/11,119 Tcl1 14,166/17 NODAL/Nodal 12/114 TUBG1 32/7,004 Zfp428 13,991/22 RIF1/Rif1 1,433 RPL10A 37/8,606 Gjb3 13,495/23 PARP1/Parp1 21/97 RPLP2 38/7,284 Tdh 11,393/26 LEFTY2/Lefty2 33/85 TALDO1 63/8,745 Fbxo15 14,921/35 MYC/Myc 40/38 RPS3A 65/11,303 Zfp57 13,784/63 JARID2/Jarid2 46/11 RPL6 101/11,695 Eras 13,432/92 ZFP42/Zfp42 78/27 EIF3A 115/12,160 Dusp27 14,812/134 IFITM1/Ifitm1 107/15 USO1 151/11,350 Enpp3 16,752/169 NANOG/Nanog 128/56 GPM6B 419/13,693 Spats1 15,033/172 Notes: Shared genes strongly associated with ESC self‐renewal were identified by finding the intersection of the highest‐ranking 1% of genes (173 of 17,342) in each network ordered by SRC. Distinct genes were identified by taking the difference of SRC rank for all genes and selecting the top genes most strongly supported in one network, but not the other. Abbreviations: ESC, embryonic stem cell; hESC, human ESCs; mESC, mouse ESCs; SRC, self‐renewal correlation. Open in new tab Functional similarities and differences between networks Shared . Distinct . Gene (h/m) . Rank (h/m) . hESC gene . Rank (h/m) . mESC gene . Rank (h/m) . POU5F1/Pou5f1 1/1 PELI1 5/8,197 Utf1 7,906/6 SOX2/Sox2 2/5 LRRN1 15/11,119 Tcl1 14,166/17 NODAL/Nodal 12/114 TUBG1 32/7,004 Zfp428 13,991/22 RIF1/Rif1 1,433 RPL10A 37/8,606 Gjb3 13,495/23 PARP1/Parp1 21/97 RPLP2 38/7,284 Tdh 11,393/26 LEFTY2/Lefty2 33/85 TALDO1 63/8,745 Fbxo15 14,921/35 MYC/Myc 40/38 RPS3A 65/11,303 Zfp57 13,784/63 JARID2/Jarid2 46/11 RPL6 101/11,695 Eras 13,432/92 ZFP42/Zfp42 78/27 EIF3A 115/12,160 Dusp27 14,812/134 IFITM1/Ifitm1 107/15 USO1 151/11,350 Enpp3 16,752/169 NANOG/Nanog 128/56 GPM6B 419/13,693 Spats1 15,033/172 Shared . Distinct . Gene (h/m) . Rank (h/m) . hESC gene . Rank (h/m) . mESC gene . Rank (h/m) . POU5F1/Pou5f1 1/1 PELI1 5/8,197 Utf1 7,906/6 SOX2/Sox2 2/5 LRRN1 15/11,119 Tcl1 14,166/17 NODAL/Nodal 12/114 TUBG1 32/7,004 Zfp428 13,991/22 RIF1/Rif1 1,433 RPL10A 37/8,606 Gjb3 13,495/23 PARP1/Parp1 21/97 RPLP2 38/7,284 Tdh 11,393/26 LEFTY2/Lefty2 33/85 TALDO1 63/8,745 Fbxo15 14,921/35 MYC/Myc 40/38 RPS3A 65/11,303 Zfp57 13,784/63 JARID2/Jarid2 46/11 RPL6 101/11,695 Eras 13,432/92 ZFP42/Zfp42 78/27 EIF3A 115/12,160 Dusp27 14,812/134 IFITM1/Ifitm1 107/15 USO1 151/11,350 Enpp3 16,752/169 NANOG/Nanog 128/56 GPM6B 419/13,693 Spats1 15,033/172 Notes: Shared genes strongly associated with ESC self‐renewal were identified by finding the intersection of the highest‐ranking 1% of genes (173 of 17,342) in each network ordered by SRC. Distinct genes were identified by taking the difference of SRC rank for all genes and selecting the top genes most strongly supported in one network, but not the other. Abbreviations: ESC, embryonic stem cell; hESC, human ESCs; mESC, mouse ESCs; SRC, self‐renewal correlation. Open in new tab As described in Materials and Methods section, we identified 173 core self‐renewal genes in the mESC network and 111 core self‐renewal genes in the hESC network with a significant overlap of 73 shared genes (Supporting Information Fig. 4B; Supporting Information Table S8; hypergeometric p = 8.56−7). As expected, well‐studied genes known to play a significant role in self‐renewal and pluripotency across species were ranked highest in both networks. These key players included well‐known transcriptional regulators: POU5F1/Pou5f1 (1/1), SOX2/Sox2 (2/5), and NANOG/Nanog (128/56) (Table 3; Supporting Information Table S9). Highly conserved novel genes (not included in training set examples) included PARP1/Parp1 (21/97), a regulator of chromatin structure, transcription, and DNA damage repair that has recently been shown to be required for reprogramming mouse MEFs to a pluripotent state [59, 60]; and IFITM1/Ifitm1 (107/15), an interferon‐induced transmembrane protein identified as a downstream target of WNT receptor signaling during gastrulation involved in somitogenesis and mesoderm formation [61]. In our training examples, we included components of KEGG signaling pathways and WikiPathways known to influence early development and ESC fate, such as FGF and JAK/STAT signaling [13, 35, 62, 63]. However, as these pathways include many homologous components, it is likely that only a subset of documented pathway participants are active in ESCs, and some interactions may be species specific (e.g., LIF‐activated JAK/STAT signaling in mESCs, and NODAL/ACTIVIN‐A‐activated FGF signaling in hESCs). As such, the bulk of training set genes with low SRCs in both networks consist of signaling pathway participants that appear to be more likely involved in signaling in a cellular context other than ESCs, such as FGF22/Fgf22 (15,800/14,199), WNT10A/Wnt10a (12,140/10,554), and INHBA/Inhba (12,583/14,301). Surprisingly, given its role in X‐chromosome inactivation [15, 64], XIST/Xist (13,765/11,727) ranked poorly in both networks. This could be because, while it is measured by most human and mouse microarray platforms, it may not correlate with other specific gene expression regulatory and protein‐DNA binding genes that are strongly associated with self‐renewal. While our compendiums contain extensive high‐throughput data from a broad range of experimental techniques (Table 1), the bulk of our data for both species is derived primarily from gene expression studies (using microarray or RNA‐seq technologies) and from protein‐DNA binding data (from chromatin immunoprecipitation (ChIP) followed by microarray (ChIP‐chip) or sequencing (ChIP‐seq) assays). We show that these data can be highly informative for elucidating protein function in a specific cellular context; however, the scarcity of additional high‐throughput data types, such as protein‐protein‐binding assays performed in ESCs, limited our approach to identifying genes related primarily through expression or direct transcriptional regulation. Predictive Networks Emphasize Important Signaling Pathway Differences Although the mESC and hESC networks are highly correlated, suggesting that gene functional associations in the context of ESC self‐renewal are largely conserved, there are striking differences (Table 3; Supporting Information Table S10). Divergent functional linkages are particularly evident in developmental signaling pathways and our predictive networks recapitulate many known differences in signaling cues that prompt naive mouse versus primed human ESCs to sustain or exit a self‐renewing state. To visualize these differences, we created predictive mESC and hESC developmental signaling pathways using gene elements identified in subsections of KEGG pathways, overlaid with our predicted connection probabilities and functional correlations to self‐renewal (SRCs) for those genes (Figs. 2, 3). Open in new tabDownload slide Predicted fibroblast growth factor (FGF) signaling pathway relationships across species. These network models show the predicted strength of relationships between all known FGF signaling ligand–receptor pairs [65]. Gene nodes are colored by self‐renewal correlation (SRC) score (light gray = weak SRC; dark blue = strong SRC) and edge color/thickness indicates the strength of predicted functional association between ligand and receptor (yellow/thin = low probability; teal/thick = high probability). (A): The mouse ESC (mESC) FGF signaling model showed that the most strongly connected ligand‐receptor pairs with high‐ranking SRCs are Fgf4 and Fgf5, which are both associated with Fgfr1 and Fgfr2, and Fgf15, which is most strongly connected to Fgfr2. In mESCs, Fgf4 is known to activate FGF signaling, Fgf5 is associated with FGF signaling in the late stage blastocyst and epiblast, and Fgf15 has been shown to be involved in early neurodevelopment. Abbreviations: FGF, fibroblast growth factor; hESC, human embryonic stem cells; mESC, mouse ESCs. (B): The human embryonic stem cell (hESC) FGF signaling model shows that the most strongly connected ligand‐receptor pairs with high‐ranking SRCs are FGF2 and FGF19 (the human ortholog of Fgf15), which are most strongly associated with FGFR1 and FGFR3. FGF2 is known to activate FGF signaling in hESCs, while FGF19 has been associated with neuronal development. Open in new tabDownload slide Predicted fibroblast growth factor (FGF) signaling pathway relationships across species. These network models show the predicted strength of relationships between all known FGF signaling ligand–receptor pairs [65]. Gene nodes are colored by self‐renewal correlation (SRC) score (light gray = weak SRC; dark blue = strong SRC) and edge color/thickness indicates the strength of predicted functional association between ligand and receptor (yellow/thin = low probability; teal/thick = high probability). (A): The mouse ESC (mESC) FGF signaling model showed that the most strongly connected ligand‐receptor pairs with high‐ranking SRCs are Fgf4 and Fgf5, which are both associated with Fgfr1 and Fgfr2, and Fgf15, which is most strongly connected to Fgfr2. In mESCs, Fgf4 is known to activate FGF signaling, Fgf5 is associated with FGF signaling in the late stage blastocyst and epiblast, and Fgf15 has been shown to be involved in early neurodevelopment. Abbreviations: FGF, fibroblast growth factor; hESC, human embryonic stem cells; mESC, mouse ESCs. (B): The human embryonic stem cell (hESC) FGF signaling model shows that the most strongly connected ligand‐receptor pairs with high‐ranking SRCs are FGF2 and FGF19 (the human ortholog of Fgf15), which are most strongly associated with FGFR1 and FGFR3. FGF2 is known to activate FGF signaling in hESCs, while FGF19 has been associated with neuronal development. Open in new tabDownload slide Predicted JAK/STAT signaling pathway relationships across species. Analogously to Figure 2, species‐specific predictions are shown for a portion of the Kyoto Encyclopedia of Genes and Genomes pathway map for JAK/STAT signaling focused on the interleukin‐6 (IL‐6) family of cytokines. (A): The mouse ESC predicted JAK/STAT signaling pathway showed a strong connection between Lif and its putative target, Il6st. Furthermore, key known mouse self‐renewal genes in this pathway, including Lif, Il6st, and Stat3 are strongly correlated to other self‐renewal genes as indicated by SRC score. Abbreviations: mESC, mouse embryonic stem cells; hESC, human ESCs. (B): The human embryonic stem cell (hESC)‐predicted JAK/STAT signaling pathway showed that Stat3 is the gene most correlated to self‐renewal, while upstream pathway participants exhibited lower SRC scores and weaker connection probabilities, suggesting that cytokines other than the IL‐6 family, or signaling cross talk, may be required for STAT3 activation in hESCs. Open in new tabDownload slide Predicted JAK/STAT signaling pathway relationships across species. Analogously to Figure 2, species‐specific predictions are shown for a portion of the Kyoto Encyclopedia of Genes and Genomes pathway map for JAK/STAT signaling focused on the interleukin‐6 (IL‐6) family of cytokines. (A): The mouse ESC predicted JAK/STAT signaling pathway showed a strong connection between Lif and its putative target, Il6st. Furthermore, key known mouse self‐renewal genes in this pathway, including Lif, Il6st, and Stat3 are strongly correlated to other self‐renewal genes as indicated by SRC score. Abbreviations: mESC, mouse embryonic stem cells; hESC, human ESCs. (B): The human embryonic stem cell (hESC)‐predicted JAK/STAT signaling pathway showed that Stat3 is the gene most correlated to self‐renewal, while upstream pathway participants exhibited lower SRC scores and weaker connection probabilities, suggesting that cytokines other than the IL‐6 family, or signaling cross talk, may be required for STAT3 activation in hESCs. For example, FGF signaling has been shown to play an important role in regulating ESC self‐renewal and differentiation (as well as myriad other processes) and may be involved in promoting the ESC transition from a naive to primed pluripotent state [65]. Mice and humans have 22 FGF ligands and 5 FGF receptors (FGFRs), each with specific expression patterns that change over time during early development, and different ligand‐receptor pairs have varying functional roles depending on the cellular context. In mESCs, FGF4‐activated ERK signaling promotes differentiation [65], whereas in hESCs, FGF2‐activated ERK signaling sustains self‐renewal [8, 65]. We compared functional associations among FGF ligand–receptor pairs using our hESC and mESC networks and found our predictions clearly recapitulate known differences between species (Fig. 2). The FGF2/Fgf2 (48/14,300) ligand was strongly correlated to self‐renewal in the hESC network only; while FGF4/Fgf4 (3,797/156) and FGF5/Fgf5 (2,965/308) ranked highest in the mESC network only. In both species, these active FGF ligands were strongly linked to the FGFR1/Fgfr1 (126/302) and FGFR2/Fgfr2 (198/379) receptors, while the FGFR3/Fgfr3 (170/6,063) and FGFR4/Fgfr4 (1,191/7,560) receptors were strongly linked only in the human network. In addition, both our hESC and mESC networks showed high‐ranking SRCs for the FGF19/Fgf15 (144/622) ligand; although in humans, it was associated with FGFR1/Fgfr1 (126/302) and FGFR3/Fgfr3 (170/6,063) receptors, while in mice, this ligand was associated primarily with FGFR2/Fgfr2 (198/379). In both species, FGF19/Fgf15 is expressed in the fetal brain and mediates differentiation to the neuroectoderm [66, 67]. We also examined LIF‐activated JAK/STAT signaling, as LIF promotes self‐renewal by activating the JAK/STAT3 and PI3K/AKT signaling pathways in mESCs, but is not required by hESCs to sustain self‐renewal [13, 62]. While STAT3/Stat3 (202/485) is strongly associated with self‐renewal in both species in our networks, it appears that the mode of regulation may be diverged. This difference is clearly supported in our probabilistic pathways focused on JAK/STAT signaling stimulated by the interleukin‐6 (IL‐6) family of cytokines (Fig. 3) [68]. In the mESC network, a strong connection exists between the LIF/Lif (3,457/1,317) cytokine and its well‐characterized signal transducer IL6ST/Il6st (also known as gp130; 9,107/347), which in turn is tightly linked to tyrosine kinase JAK3/Jak3 (10,600/3,126) and then on to STAT3/Stat3 (202/485). These predicted linkages reflect the most well‐documented LIF‐activated IL‐6 JAK/STAT signaling cascade in mESCs [69-71]; however, these strong associations are not present in our hESC network. Our mESC results also showed functional linkages between LIF/Lif and the IFNGR2/Ifngr2 (5,801/946) receptor, which associates most strongly with JAK1/Jak1 (2,789/3,808) and JAK2/Jak2 (6,625/1,989), both of which strongly link to STAT3/Stat3. In contrast, within the confines of this JAK/STAT pathway subset, the hESC network predicts that LIF and the many other IL‐6 cytokines are not strongly associated with self‐renewal (based on SRC scores), and shows only weak connections between LIF and the Il6ST signal transducer [72]. Rather, in hESCs, the cytokines IL11/Il11 (2,939/15,541) and LIF/Lif were moderately associated with the IFNGR1/Ifngr1 (2,701/6,655) and OSMR/Osmr (2,060/8,138) receptors, which were moderately linked to the JAK1/Jak1 (2,789/3,801) and TYK2/Tyk2 (11,394/8,747) tyrosine kinases, respectively (note that while Tyk2 is annotated in the mouse JAK/STAT pathway by KEGG, it is not listed as a participant in the human KEGG pathway). Global expression profiling in mESCs has shown that the interferon gamma receptor IFNGR1/Ifngr1 is regulated by POU5F1/Pou5f1 [73], and although not documented as active in early embryonic development, OSMR/Osmr is predicted to have a strong functional association with POU5F1/Pou5f1 supported by both gene expression and protein‐DNA‐interaction data in the human network. In contrast to these moderate associations with STAT3, our hESC network predicts strong STAT3 associations with members of other signaling pathways, including TCF7L1, FGF2, FGFR1, and BMPR2, suggesting potentially significant cross‐talk and/or alternate modes of STAT3 regulation. However, these strong associations are not observed for Stat3 in our mESC network, where Stat3 is most strongly associated with known LIF and JAK/STAT signaling pathway members, such as Fos, Pim1, and Spry4. Metabolic Differences Between Species Highlighted in Predictive Networks One of the most striking differences between our mouse and human ESC networks concerns threonine catabolism, which is required for mESC self‐renewal, likely through the TDH/Tdh (l‐threonine dehydrogenase) gene, which supports accelerated cell cycle kinetics by catabolizing threonine into glycine and acetyl‐CoA, which is used by the TCA cycle to generate ATP [74, 75]. In our mouse network, TDH/Tdh (11,393/26) has a strong correlation to mESC self‐renewal and is tightly linked to many core self‐renewal genes, including POU5F1/Pou5f1 (1/1), SOX2/Sox2 (2/5), RIF1/Rif1 (14/33), and NR0B1/Nr0b1 (5,583/487). The functional relationship between Tdh and these genes is largely supported by ChIP‐chip binding data from studies investigating the regulatory circuitry of mESCs and microarray data from a study analyzing mESC differentiation (Fig. 4B). In contrast, TDH is not correlated to self‐renewal in our hESC network and not tightly connected to any genes in our positive training set (Fig. 4A). Differential and functional analyses of the mESC and hESC networks did not reveal a direct metabolic equivalent to threonine dehydrogenase in humans. Literature validation showed that the human TDH gene encoding threonine dehydrogenase has been rendered nonfunctional due to three mutations (two AG‐to‐GG splice acceptor mutations in exons 4 and 6, and a nonsense mutation in exon 6) [74]. However, hESCs grow at a slower rate than mESCs (with a doubling time of 35 hours as opposed to every 4–5 hours), and it is not yet known whether the difference in growth rate might be due to the absence of TDH or if there may be some selective advantage for inactivating TDH in humans [74-76]. Open in new tabDownload slide Differences in mouse embryonic stem cell (mESC) and human ESC (hESC) threonine metabolism. We used our StemSight Scout data visualization tool (www.StemSight.org) to create network views centered around l‐threonine dehydrogenase (TDH/Tdh), which supports accelerated cell cycle kinetics in mESCs, but is not functional in hESCs. Node and edge colors are as in Figures 2 and 3, except that edges contained in the positive training set are colored orange. (A): The hESC TDH‐centric network shows that TDH is weakly correlated to genes in our training set and has no strong functional associations to any known self‐renewal genes. (B): The mESC Tdh‐centric network illustrated that Tdh is strongly correlated to self‐renewal genes and had strong predicted functional associations with known self‐renewal genes, including Pou5f1, Sox2, Nr0b1, Klf2, Zfp42, Gdf3, and Fbx015. Open in new tabDownload slide Differences in mouse embryonic stem cell (mESC) and human ESC (hESC) threonine metabolism. We used our StemSight Scout data visualization tool (www.StemSight.org) to create network views centered around l‐threonine dehydrogenase (TDH/Tdh), which supports accelerated cell cycle kinetics in mESCs, but is not functional in hESCs. Node and edge colors are as in Figures 2 and 3, except that edges contained in the positive training set are colored orange. (A): The hESC TDH‐centric network shows that TDH is weakly correlated to genes in our training set and has no strong functional associations to any known self‐renewal genes. (B): The mESC Tdh‐centric network illustrated that Tdh is strongly correlated to self‐renewal genes and had strong predicted functional associations with known self‐renewal genes, including Pou5f1, Sox2, Nr0b1, Klf2, Zfp42, Gdf3, and Fbx015. Comparative Network Analyses Reveal Novel Species‐Specific Differences To discover novel species‐specific differences, we focused on the top 0.001% (1,503) of gene pairs with the greatest difference between the mESC versus hESC networks (Supporting Information Table S10). There were 86 genes involved in edges strongly supported only in the mESC network, including genes annotated to chordate embryonic development (GO:0043009), stem cell maintenance (GO:0019827), negative regulation of cell differentiation (GO:0045596), and regulation of transcription, DNA‐dependent (GO:0042127). In contrast, 179 genes were involved in edges supported only by the hESC network, including genes annotated to WNT Receptor Signaling (GO:0016055), MAPK Cascade (GO:0000165), FGF Signaling (GO:0008543), ATP‐Binding (GO:0005524), and cell cycle regulation (GO:0051726), emphasizing the different set of developmental signaling cues observed in early cell fate decisions in hESCs. These key differences were echoed when we compared the top 1% of novel genes ranked by SRC (the 173 highest ranking genes not included in our positive training set examples). Top‐ranked novel mESC genes were predominantly associated with differentiation and transcriptional regulation processes, whereas the novel hESC genes were largely related to regulation of translation (GO:0006417), microtubule‐based process (GO:0007017), regulation of cell cycle (GO:0051726), cell division (GO:0051301), and cell adhesion (GO:0007155). Although there are known differences in cell cycle controls that permit rapid cell cycling in ESCs, it is not yet known whether the cell cycle regulates pluripotency. Cell cycle length is not believed to determine pluripotency, because different pluripotent cell types divide at different rates [77]; however, the enrichment of cell cycle component genes in our novel hESC gene lists suggest that there may be functional differences in our mouse and human networks that can be mined to improve our understanding of how the core cell cycle machinery adapts to the timing requirements of primed and naive ESCs. Many of the genes most strongly correlated to self‐renewal in the hESC network only (Table 3) are involved in translational regulation and protein synthesis, such as the ribosomal proteins RPL10A/Rpl10a (37/8,606), RPLP2/Rplp2 (38/7,284), RPS3A/Rps3a (65/11,303), and RPL6/Rpl6 (101/11,695), as well as the translation initiation factor EIF3A/Eif3a (115/12,160). Intriguingly, genome‐wide studies have shown that LIN28/Lin28a (62/224), a known regulator of self‐renewal and an RNA‐binding protein [78], targets RNAs in hESCs including ribosomal and translation‐supporting genes, which are important for growth and survival [79]. In contrast, Lin28a targets in mESCs are enriched for translational repressors of membrane‐bound proteins, secretory proteins, and proteins destined for the endoplasmic reticulum/Golgi lumen [80], suggesting alternative regulatory roles for LIN28/Lin28a, despite supporting self‐renewal in both species. Comparative Network Visualization Promotes Novel Gene Discovery To make our probabilistic comparative networks readily available to the stem cell research community, we provide interactive, online visualization resources at www.StemSight.org that can be used to view our hESC and mESC networks independently or comparatively. StemSight Scout enables users to explore subnetworks centered on user‐provided genes of interest, and highlights potentially novel self‐renewal genes by coloring nodes based on their SRC score and illustrates the weight of predicted interactions by coloring edges based on inferred posterior probabilities. Edges included in our training set examples are color‐coded, making it easy to visually segregate novel from known. For positive training set edges, links are provided to the original articles documenting the relationship, while for novel edges, the evidential data supporting that functional relationship is displayed with links to the studies in PubMed. The comparative view enables users to search for interactions among one or more genes in both networks simultaneously, and displays results in a convenient side‐by‐side window for direct comparisons (Fig. 4), which is especially helpful for contrasting connectivity of gene hubs and pathway participants in the hESC versus mESC networks. Conclusions Determining the molecular underpinnings of stem cell self‐renewal and cell fate determination is vital for understanding and refining the production of iPS cells, and has potential implications for targeting cancer stem cells and aging processes in adult stem cells. Our results confirm that multiple signaling pathways contribute to the balance between self‐renewal and differentiation in ESCs, and further suggest that these complex pathways should not be considered monolithically, but recognized as general mechanisms whose details and interactions are only partially understood. For example, while our results indicate that STAT3/Stat3 is strongly associated with self‐renewal in both humans and mice, upstream regulation of signaling appears significantly diverged between species. Similar observations and conclusions can be drawn from our results for various members of WNT, NODAL/ACTIVIN‐A, FGF, and TGF‐β signaling, as well as many other signaling and metabolic pathways. Our computational integration of hundreds of high‐throughput data sets provides an entry point to the greater understanding of these pathways. For example, our evaluation of STAT3 interactions suggests potentially significant crosstalk and alternate modes of human STAT3 regulation. As STAT3/Stat3 is a multifunctional gene implicated in phenotypes and disorders ranging from circadian rhythms to cancer, developing cancer therapies targeting the STAT3 pathway requires investigation of potential off‐target effects, which in turn requires expanding our understanding of its context‐specific regulation. Thus, unraveling the diversity of systems‐level interactions and crosstalk is vital for the use of model organisms, and is likely key for dissecting the cell type specific roles of multifunctional genes within humans. Similarly, the divergent role of TDH/Tdh between species in our networks likely reflects the distinct metabolic mechanisms of these cell types. While Tdh supports increased growth rates in mESCs, it is possible that the increased activity of translational regulatory genes in hESCs (such as the ribosomal and translation initiation genes identified only in our human networks) play a similar role in a human context. By focusing here on human and mouse ESC self‐renewal, we have developed methods for cross‐species and cross‐context comparisons to assist stem cell researchers in evaluating known and novel genes involved in ESC self‐renewal and differentiation. In the future, comparative networks for additional stem cell types, including mEpiSCs, iPSCs, ESCs from other species, as well as adult stem cells, could be used to further elucidate the mechanisms of self‐renewal in general. Primed mEpiSCs would be particularly useful for investigating differences between the naive and primed pluripotent state within mouse (i.e., naive mESCs vs. primed mEpiSCs), as well as for comparing primed pluripotent cells in different species (i.e., primed mEpiSCs vs. primed hESCs). While computational systems biology efforts such as ours begin to scrutinize complex molecular interactions, additional laboratory efforts are required to confirm these hypotheses and to determine precise mechanisms. The data generated by these efforts can be reincorporated into machine learning efforts to refine our training sets, provide additional input evidential data, and ultimately improve our understanding of context‐specific, whole‐genome interactomes. As such, the full results of this study are freely available to the stem cell community at www.StemSight.org, where users can explore our hESC and mESC self‐renewal networks to place their results into a broader context, draw new hypotheses, or prioritize candidate genes. Acknowledgments This work was supported by a Pharmaceutical Researchers and Manufacturers of America (PhRMA) Foundation Fellowship in Informatics (to K.G.D.); the University of Maine Graduate School of Biomedical Sciences; the National Science Foundation Integrated Graduate Education and Research Traineeship (IGERT) program grant 0221625; and partially funded by The Jackson Laboratory; by NIH/National Institute of Arthritis and Musculoskeletal and Skin (NIH/NIAMS) grant R21 AR069981‐01 (Hibbs, Ackert‐Bicknell [to M.A.H.]), and by grant P50 GM076468‐06 (Churchill) from NIH (to M.A.H.). M.A.H. is an Ellison Medical Foundation New Scholar in Aging. Author Contribution K.G.D.: conception and design, financial support, collection and/or assembly of data, data analysis and interpretation, manuscript writing; A.K.S.: collection and/or assembly of data, software development, technical support; H.B.: collection and/or assembly of data; B.K.: collection and/or assembly of data, software development; Z.Z.W.: financial support, data analysis and interpretation; K.Y.: data analysis and interpretation, manuscript editing; M.A.H.: conception and design, financial support, collection and/or assembly of data, data analysis and interpretation, manuscript writing, final approval of manuscript. Disclosure of Potential Conflicts of Interest All other authors indicate no potential conflicts of interest. References 1 Thomson JA , Itskovitz‐Eldor J, Shapiro SS et al. Embryonic stem cell lines derived from human blastocysts . Science 1998 ; 282 : 1145 – 1147 . Google Scholar Crossref Search ADS PubMed WorldCat 2 Gokhale PJ , Andrews PW. The development of pluripotent stem cells . Curr Opin Genet Dev 2012 ; 22 : 403 – 408 . Google Scholar Crossref Search ADS PubMed WorldCat 3 Ginis I , Luo Y, Miura T et al. Differences between human and mouse embryonic stem cells . Dev Biol 2004 ; 269 : 360 – 380 . Google Scholar Crossref Search ADS PubMed WorldCat 4 Brown S , Teo A, Pauklin S et al. Activin/nodal signaling controls divergent transcriptional networks in human embryonic stem cells and in endoderm progenitors . Stem Cells 2011 ; 29 : 1176 – 1185 . Google Scholar Crossref Search ADS PubMed WorldCat 5 Welling M , Geijsen N. Uncovering the true identity of naive pluripotent stem cells . Trends Cell Biol 2013 ; 23 : 442 – 448 . Google Scholar Crossref Search ADS PubMed WorldCat 6 Martin GR . Isolation of a pluripotent cell line from early mouse embryos cultured in medium conditioned by teratocarcinoma stem cells . Proc Natl Acad Sci USA 1981 ; 78 : 7634 – 7638 . Google Scholar Crossref Search ADS PubMed WorldCat 7 Evans MJ , Kaufman MH. Establishment in culture of pluripotential cells from mouse embryos . Nature 1981 ; 292 : 154 – 156 . Google Scholar Crossref Search ADS PubMed WorldCat 8 Greber B , Wu G, Bernemann C et al. Conserved and divergent roles of FGF signaling in mouse epiblast stem cells and human embryonic stem cells . Cell Stem Cell 2010 ; 6 : 215 – 226 . Google Scholar Crossref Search ADS PubMed WorldCat 9 Saunders A , Faiola F, Wang J. Pursuing self‐renewal and pluripotency with the stem cell factor nanog . Stem Cells 2013 ; 31 : 1227 – 1236 . Google Scholar Crossref Search ADS PubMed WorldCat 10 Nichols J , Smith A. The origin and identity of embryonic stem cells . Development 2011 ; 138 : 3 – 8 . Google Scholar Crossref Search ADS PubMed WorldCat 11 Nichols J , Smith A. Naive and primed pluripotent states . Cell Stem Cell 2009 ; 4 : 487 – 492 . Google Scholar Crossref Search ADS PubMed WorldCat 12 Lanza RP . Essent Stem Cell Biol 2006 ; xxxi : 548 . 13 De Los Angeles A , Loh YH, Tesar PJ et al. Accessing naive human pluripotency . Curr Opin Genet Dev 2012 ; 22 : 272 – 282 . Google Scholar Crossref Search ADS PubMed WorldCat 14 Tesar PJ , Chenoweth JG, Brook FA et al. New cell lines from mouse epiblast share defining features with human embryonic stem cells . Nature 2007 ; 448 : 196 – 199 . Google Scholar Crossref Search ADS PubMed WorldCat 15 Tomoda K , Takahashi K, Leung K et al. Derivation conditions impact X‐inactivation status in female human induced pluripotent stem cells . Cell Stem Cell 2012 ; 11 : 91 – 99 . Google Scholar Crossref Search ADS PubMed WorldCat 16 Nazor KL , Altun G, Lynch C et al. Recurrent variations in DNA methylation in human pluripotent stem cells and their differentiated derivatives . Cell Stem Cell 2012 ; 10 : 620 – 634 . Google Scholar Crossref Search ADS PubMed WorldCat 17 Chenoweth JG , McKay RD, Tesar PJ. Epiblast stem cells contribute new insight into pluripotency and gastrulation . Dev Growth Different 2010 ; 52 : 293 – 301 . Google Scholar Crossref Search ADS WorldCat 18 Tachibana M , Sparman M, Ramsey C et al. Generation of chimeric rhesus monkeys . Cell 2012 ; 148 : 285 – 295 . Google Scholar Crossref Search ADS PubMed WorldCat 19 Takahashi K , Yamanaka S. Induction of pluripotent stem cells from mouse embryonic and adult fibroblast cultures by defined factors . Cell 2006 ; 126 : 663 – 676 . Google Scholar Crossref Search ADS PubMed WorldCat 20 Ema M , Mori D, Niwa H et al. Kruppel‐like factor 5 is essential for blastocyst development and the normal self‐renewal of mouse ESCs . Cell Stem Cell 2008 ; 3 : 555 – 567 . Google Scholar Crossref Search ADS PubMed WorldCat 21 Buganim Y , Faddah DA, Jaenisch R. Mechanisms and models of somatic cell reprogramming . Nat Rev Genet 2013 ; 14 : 427 – 439 . Google Scholar Crossref Search ADS PubMed WorldCat 22 Meissner A . Epigenetic modifications in pluripotent and differentiated cells . Nat Biotechnol 2010 ; 28 : 1079 – 1088 . Google Scholar Crossref Search ADS PubMed WorldCat 23 Jaenisch R , Young R. Stem cells, the molecular circuitry of pluripotency and nuclear reprogramming . Cell 2008 ; 132 : 567 – 582 . Google Scholar Crossref Search ADS PubMed WorldCat 24 Chambers I , Tomlinson SR. The transcriptional foundation of pluripotency . Development 2009 ; 136 : 2311 – 2322 . Google Scholar Crossref Search ADS PubMed WorldCat 25 Dowell KG , Simons AK, Wang ZZ et al. Cell‐type‐specific predictive network yields novel insights into mouse embryonic stem cell self‐renewal and cell fate . PLoS One 2013 ; 8 : e56810 . Google Scholar Crossref Search ADS PubMed WorldCat 26 Huttenhower C , Hibbs MA, Myers CL et al. The impact of incomplete knowledge on evaluation: An experimental benchmark for protein function prediction . Bioinformatics 2009 ; 25 : 2404 – 2410 . Google Scholar Crossref Search ADS PubMed WorldCat 27 Hibbs MA . Advanced Bioinformatics Tools and Strategies . In: Principles and Practices of Plant Genomics. Vol. 3 , Enfield, NH: Science Publishers 2010 . 28 Russell S , Norvig P. Artifical Intelligence: A Modern Approach , 3rd ed. Fort Collins, CO : Prentice Hall , 2009 . 29 Charniak E . Bayesian networks without tears . AI Magazine 1991 ; 91 : 50 – 63 . Google Scholar OpenURL Placeholder Text WorldCat 30 Darwiche A . Modeling and Reasoning with Bayesian Networks . Cambridge; New York : Cambridge University Press , 2009 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC 31 Hibbs MA , Myers CL, Huttenhower C et al. Directing experimental biology: A case study in mitochondrial biogenesis . PLoS Comput Biol 2009 ; 5 : e1000322 . Google Scholar Crossref Search ADS PubMed WorldCat 32 Bult CJ , Kadin JA, Richardson JE et al. The Mouse Genome Database: Enhancements and updates . Nucleic Acids Res 2010 ; 38 : D586 – D592 . Google Scholar Crossref Search ADS PubMed WorldCat 33 Dolan ME , Ni L, Camon E, Blake JA. A procedure for assessing GO annotation consistency . Bioinformatics 2005 ; 21 : i136 – i143 . Google Scholar Crossref Search ADS PubMed WorldCat 34 Kanehisa M . The KEGG database . Novartis Found Symp 2002 ; 247 : 91 – 101;discussion 101–103, 119–128, 244‐152 . Google Scholar Crossref Search ADS PubMed WorldCat 35 Kelder T , van Iersel MP, Hanspers K et al. WikiPathways: Building research communities on biological pathways . Nucl Acids Res 2012 ; 40 : D1301 – D1307 . Google Scholar Crossref Search ADS PubMed WorldCat 36 Huttenhower C , Schroeder M, Chikina MD et al. The Sleipnir library for computational functional genomics . Bioinformatics 2008 ; 24 : 1559 – 1561 . Google Scholar Crossref Search ADS PubMed WorldCat 37 Guan Y , Myers CL, Lu R et al. A genomewide functional network for the laboratory mouse . PLoS Comput Biol 2008 ; 4 : e1000165 . Google Scholar Crossref Search ADS PubMed WorldCat 38 Myers CL , Barrett DR, Hibbs MA et al. Finding function: Evaluation methods for functional genomic data . BMC Genom 2006 ; 7 : 187 . Google Scholar Crossref Search ADS WorldCat 39 Huttenhower C , Haley EM, Hibbs MA et al. Exploring the human genome with functional maps . Genome Res 2009 ; 19 : 1093 – 1106 . Google Scholar Crossref Search ADS PubMed WorldCat 40 Dennis G Jr, Sherman BT, Hosack DA et al. DAVID: Database for Annotation, Visualization, and Integrated Discovery Genome Biology 2003 ; 4 : P3 . Google Scholar OpenURL Placeholder Text WorldCat 41 Blake JA , Dolan M, Drabkin H et al. The gene ontology in 2010: Extensions and refinements . Nucl Acids Res 2010 ; 38 : D331 – D335 . Google Scholar Crossref Search ADS PubMed WorldCat 42 Horvath S , Dong J. Geometric interpretation of gene coexpression network analysis . PLoS Comput Biol 2008 ; 4 : e1000117 . Google Scholar Crossref Search ADS PubMed WorldCat 43 Bai H , Chen K, Gao YX et al. Bcl‐xL enhances single‐cell survival and expansion of human embryonic stem cells without affecting self‐renewal . Stem Cell Res 2012 ; 8 : 26 – 37 . Google Scholar Crossref Search ADS PubMed WorldCat 44 Bai H , Gao Y, Arzigian M et al. BMP4 regulates vascular progenitor development in human embryonic stem cells through a Smad‐dependent pathway . J Cell Biochem 2010 ; 109 : 363 – 374 . Google Scholar PubMed OpenURL Placeholder Text WorldCat 45 Bai H , Xie YL, Gao YX et al. The balance of positive and negative effects of TGF‐beta signaling regulates the development of hematopoietic and endothelial progenitors in human pluripotent stem cells . Stem Cells Dev 2013 ; 22 : 2765 – 2776 . Google Scholar Crossref Search ADS PubMed WorldCat 46 Myers CL , Robson D, Wible A et al. Discovery of biological networks from diverse functional genomic data . Genome Biol 2005 ; 6 : R114 . Google Scholar Crossref Search ADS PubMed WorldCat 47 Troyanskaya OG , Dolinski K, Owen AB et al. A Bayesian framework for combining heterogeneous data sources for gene function prediction (in Saccharomyces cerevisiae) . Proc Natl Acad Sci USA 2003 ; 100 : 8348 – 8353 . 48 Tsuneyoshi N , Sumi T, Onda H et al. PRDM14 suppresses expression of differentiation marker genes in human embryonic stem cells . Biochem Biophys Res Commun 2008 ; 367 : 899 – 905 . Google Scholar Crossref Search ADS PubMed WorldCat 49 Chan YS , Goke J, Lu X et al. A PRC2‐dependent repressive role of PRDM14 in human embryonic stem cells and induced pluripotent stem cell reprogramming . Stem Cells 2012 ; 31 : 682 – 692 . Google Scholar Crossref Search ADS WorldCat 50 Cheng Q , Chang JT, Geradts J et al. Amplification and high‐level expression of heat shock protein 90 marks aggressive phenotypes of human epidermal growth factor receptor 2 negative breast cancer . Breast Cancer Res 2012 ; 14 : R62 . Google Scholar Crossref Search ADS PubMed WorldCat 51 Ren J , Wang R, Huang G et al. sFRP1 inhibits epithelial‐mesenchymal transition in A549 human lung adenocarcinoma cell line . Cancer Biother Radiopharm 2013 ; 28 : 565 – 571 . Google Scholar Crossref Search ADS PubMed WorldCat 52 Katoh M . Comparative integromics on non‐canonical WNT or planar cell polarity signaling molecules: Transcriptional mechanism of PTK7 in colorectal cancer and that of SEMA6A in undifferentiated ES cells . Int J Mol Med 2007 ; 20 : 405 – 409 . Google Scholar PubMed OpenURL Placeholder Text WorldCat 53 Zhang T , Meng L, Dong W et al. High expression of PRDM14 correlates with cell differentiation and is a novel prognostic marker in resected non‐small cell lung cancer . Med Oncol 2013 ; 30 : 605 . Google Scholar Crossref Search ADS PubMed WorldCat 54 Lodish HF , Berk A, Kaiser C et al. Molecular Cell Biology , 6th ed. New York, NY : WH Freeman and Co ., 2008 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC 55 Gundersen GG , Worman HJ. Nuclear positioning . Cell 2013 ; 152 : 1376 – 1389 . Google Scholar Crossref Search ADS PubMed WorldCat 56 Kele J , Andersson ER, Villaescusa JC et al. SFRP1 and SFRP2 dose‐dependently regulate midbrain dopamine neuron development in vivo and in embryonic stem cells . Stem Cells 2012 ; 30 : 865 – 875 . Google Scholar Crossref Search ADS PubMed WorldCat 57 Shahi MH , Rey JA, Castresana JS. The sonic hedgehog‐GLI1 signaling pathway in brain tumor development . Expert Opin Ther Targets 2012 ; 16 : 1227 – 1238 . Google Scholar Crossref Search ADS PubMed WorldCat 58 Wu Q , Sun CC, Lin HY et al. Repulsive guidance molecule (RGM) family proteins exhibit differential binding kinetics for bone morphogenetic proteins (BMPs) . PLoS One 2012 ; 7 : e46307 . Google Scholar Crossref Search ADS PubMed WorldCat 59 Doege CA , Inoue K, Yamashita T et al. Early‐stage epigenetic modification during somatic cell reprogramming by Parp1 and Tet2 . Nature 2012 ; 488 : 652 – 655 . Google Scholar Crossref Search ADS PubMed WorldCat 60 Krishnakumar R , Kraus WL. The PARP side of the nucleus: Molecular actions, physiological outcomes, and clinical targets . Mol Cell 2010 ; 39 : 8 – 24 . Google Scholar Crossref Search ADS PubMed WorldCat 61 Lickert H , Cox B, Wehrle C et al. Dissecting Wnt/beta‐catenin signaling during gastrulation using RNA interference in mouse embryos . Development 2005 ; 132 : 2599 – 2609 . Google Scholar Crossref Search ADS PubMed WorldCat 62 He S , Nakada D, Morrison SJ. Mechanisms of stem cell self‐renewal . Ann Rev Cell Dev Biol 2009 ; 25 : 377 – 406 . Google Scholar OpenURL Placeholder Text WorldCat 63 Vallier L , Touboul T, Brown S et al. Signaling pathways controlling pluripotency and early cell fate decisions of human induced pluripotent stem cells . Stem Cells 2009 ; 27 : 2655 – 2666 . Google Scholar Crossref Search ADS PubMed WorldCat 64 Minkovsky A , Barakat TS, Sellami N et al. The pluripotency factor‐bound intron 1 of Xist is dispensable for X chromosome inactivation and reactivation in vitro and in vivo . Cell Rep 2013 ; 3 : 905 – 918 . Google Scholar Crossref Search ADS PubMed WorldCat 65 Lanner F , Rossant J. The role of FGF/Erk signaling in pluripotent cells . Development 2010 ; 137 : 3351 – 3360 . Google Scholar Crossref Search ADS PubMed WorldCat 66 Nishimura T , Utsunomiya Y, Hoshikawa M et al. Structure and expression of a novel human FGF, FGF‐19, expressed in the fetal brain . Biochim Biophys Acta 1999 ; 1444 : 148 – 151 . Google Scholar Crossref Search ADS PubMed WorldCat 67 Borello U , Cobos I, Long JE et al. FGF15 promotes neurogenesis and opposes FGF8 function during neocortical development . Neural Dev 2008 ; 3 : 17 . Google Scholar Crossref Search ADS PubMed WorldCat 68 Kristensen DM , Kalisz M, Nielsen JH. Cytokine signalling in embryonic stem cells . Apmis 2005 ; 113 : 756 – 772 . Google Scholar Crossref Search ADS PubMed WorldCat 69 Okita K , Yamanaka S. Intracellular signaling pathways regulating pluripotency of embryonic stem cells . Curr Stem Cell Res Ther 2006 ; 1 : 103 – 111 . Google Scholar Crossref Search ADS PubMed WorldCat 70 Ying QL , Wray J, Nichols J et al. The ground state of embryonic stem cell self‐renewal . Nature 2008 ; 453 : 519 – 523 . Google Scholar Crossref Search ADS PubMed WorldCat 71 Niwa H , Burdon T, Chambers I et al. Self‐renewal of pluripotent embryonic stem cells is mediated via activation of STAT3 . Genes Dev 1998 ; 12 : 2048 – 2060 . Google Scholar Crossref Search ADS PubMed WorldCat 72 Liu N , Lu M, Tian X et al. Molecular mechanisms involved in self‐renewal and pluripotency of embryonic stem cells . J Cell Physiol 2007 ; 211 : 279 – 286 . Google Scholar Crossref Search ADS PubMed WorldCat 73 Matoba R , Niwa H, Masui S et al. Dissecting Oct3/4‐regulated gene networks in embryonic stem cells by expression profiling . PLoS One 2006 ; 1 : e26 . Google Scholar Crossref Search ADS PubMed WorldCat 74 Wang J , Alexander P, Wu L et al. Dependence of mouse embryonic stem cells on threonine catabolism . Science 2009 ; 325 : 435 – 439 . Google Scholar Crossref Search ADS PubMed WorldCat 75 Shyh‐Chang N , Locasale JW, Lyssiotis CA et al. Influence of threonine metabolism on S‐adenosylmethionine and histone methylation . Science 2013 ; 339 : 222 – 226 . Google Scholar Crossref Search ADS PubMed WorldCat 76 Wang J , Alexander P, McKnight SL. Metabolic specialization of mouse embryonic stem cells . Cold Spring Harbor Symp Quant Biol 2011 , 76 : 183 – 193 . Google Scholar Crossref Search ADS PubMed WorldCat 77 Hindley C , Philpott A. The cell cycle and pluripotency . Biochem J 2013 ; 451 : 135 – 143 . Google Scholar Crossref Search ADS PubMed WorldCat 78 Shyh‐Chang N , Daley GQ. Lin28: Primal regulator of growth and metabolism in stem cells . Cell Stem Cell 2013 ; 12 : 395 – 406 . Google Scholar Crossref Search ADS PubMed WorldCat 79 Peng S , Chen LL, Lei XX et al. Genome‐wide studies reveal that Lin28 enhances the translation of genes important for growth and survival of human embryonic stem cells . Stem Cells 2011 ; 29 : 496 – 504 . Google Scholar Crossref Search ADS PubMed WorldCat 80 Cho J , Chang H, Kwon SC et al. LIN28A is a suppressor of ER‐associated translation in embryonic stem cells . Cell 2012 ; 151 : 765 – 777 . Google Scholar Crossref Search ADS PubMed WorldCat © 2013 AlphaMed Press This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Novel Insights into Embryonic Stem Cell Self‐Renewal Revealed Through Comparative Human and Mouse Systems Biology Networks
JF - Stem Cells
DO - 10.1002/stem.1612
DA - 2014-05-01
UR - https://www.deepdyve.com/lp/oxford-university-press/novel-insights-into-embryonic-stem-cell-self-renewal-revealed-through-YNkeWBGXuP
SP - 1161
EP - 1172
VL - 32
IS - 5
DP - DeepDyve
ER -