MicroRNA and gene expression patterns in the differentiation of human embryonic stem cells

The expression of hES-specific markers was assessed by immunofluorescence and flow cytometry. Our results revealed that over 90% of the hES cells were positive for Oct4, Nanog, Sox2, Tra-1-81, and Ssea4, but negative for Ssea1, suggesting that most of the hES cells were in an undifferentiation state.

Global miRNA expression was analyzed among the 10 samples from 3 undifferentiated hES cell lines, 6 samples from EB and 5 samples from adult cell via a microarray platform (Gene Expression Omnibus accession number GSE12229). Unsupervised hierarchical clustering analysis separated the samples to three major groups: the hES cells, embryoid body (EB), and adult cells (Figure 1). Without statistic stratification, signature miRNAs specific for hES were distinguishable from EB and adults cell suggesting a diverse biological entity and fundamental difference in miRNA expression patterns.

We identified 104 miRNA differentially expressed by the hES, EB and adult cell types (F-test, P < 0.01, FDR < 0.05). These included 38 miRNA upregulated in hES cells, 31 upregulated in EB cells, and 35 upregulated in adult cells (Figure 2). The 20 miRNAs most highly expressed in hES cells, EB, and adult cells respectively were shown in additional file 1. MiR-302a, miR-302b, miR-302c, miR-302d, miR-367, and miR-200c were increased in hES and have previously been reported to be hES-specific [16, 17]. Moreover, the upregulation of these miRNAs in hES was confirmed by qRT-PCR with high correlation (R² = 0.65–0.9, data not shown).

Most miRNAs that are organized in clusters in close proximity on a chromosome have similar expression levels, indicating the possibility of transcribed in polycistronic fashion under the same promoter [16, 23]. From our data, the expression of miR-302a, miR-302b, miR-302c, miR-302d and miR-367, which are co-located in a cluster on chromosome 4 were highly correlated (R² = 0.78–0.98). Likewise, miR-200c and miR-141 located in a cluster on chromosome 12 were also highly correlated (R² = 0.94). Our results also confirmed other miRNAs that are upregulated in hES cells such as miR-299-3p, miR-369-3p, miR-96 and miR-372[16, 17, 24, 25]. However, miR-371, which is located in the same cluster with miR-372, was not discovered to be upregulated in hES cells by our results. Another member in this cluster, miR-373, was found to be upregulated in EB by our results, which was consistent with a recent report [26]. The differences among these studies may be attributed to the different cell lines tested or the different technical platforms used in assessing miRNA expression.

Most interestingly, 21 miRNAs located in a cluster on chromosome 19 exhibit similar expression levels. A portion of this large cluster has previously been found to be primate-specific and placenta-associated [27, 28]. Among these miRNAs, miR-518b, miR-518c, miR-519b, miR-519c, miR-520a, miR-520c, miR-520e, miR-520g, and miR-524* are over-expressed in undifferentiated hES cells [24, 26, 29]. Besides these 9 miRNAs, we also identified 12 more miRNAs in this cluster; they were miR-515-5p, miR-517a, miR-517b, miR-517c, miR-519e, miR-520b, miR-520d, miR-520f, miR-520h, miR-521, miR-525-3p, and miR-526b*. The similar expression levels of these miRNAs imply that they may share functional similarity.

We identified three miRNA clusters that were upregulated in embryoid body (EB). One was the Oncomir cluster consisting of miR-17-5p, miR-20a, miR-18a, miR-19a, miR-19b, and miR-92a located on chromosome 13. The second was located on chromosome 7 and includes miR-25, miR-93 and miR-106b. The third was located on chromosome X and includes miR-106a, miR-363, and miR-20b. We also identified EB upregulated miRNAs that have not been previously reported such as miR-130a, miR-301a, and miR-135, miR-190, miR-30c, and miR-30e.

A maternally-expressed imprinted miRNA cluster on chromosome 14 [30] was upregulated in adult cells. This cluster included miR-495, miR-376a, and miR-369-5p. In addition, we identified 8 miRNAs of the let-7 family that upregulated in adult cells, whose expression was detected in hES cells [16, 26], and was upregulated at the end of embryonic development [31].

We assessed global hES gene expression profiles on 3 separate passages of cells from 3 different hES cell lines, EB samples at 3 different time points, and 5 types of adult cells, HUVEC, HMVEC, UASMC NHA, and LFB using a custom spotted oligonucleotide microarray (Gene Expression Omnibus accession number GSE12228). Unsupervised hierarchical clustering using filtered genes classified the samples into three groups: the hES group, EB group and adult cell group. This clustering analysis also identified one node containing the hES cell markers POU5F1 (OCT4), LEFTY1, TDGF1 and DPPA4 (Figure 3).

We identified 776 genes differentially expressed among hES, EB and adult cell types (F-test, cut-off p < 0.005, FDR < 0.05). Hierarchical clustering analysis of these genes also divided the samples into three groups, hES, EB, and adult cells, and divided the genes into 4 major nodes (Figure 4). The node containing 226 genes that were upregulated in hES cells (node B) included the previously identified hES markers OCT4, TDGF1, LEFTY1, DNMT3B, GAL, DPPA4, UGP2, TERF1, GABRB3, CD24, FAM46B, SALL4, TCEA1, ZNF398, NODAL, and ACVR2B [32‐35]. The node containing genes upregulated in EB (node C) included the genes HAND1, HOXA1, HOXB2, MSX1, MSX2, MEIS1, FGF9 and FREM1 which are involved in morphogenesis and development [36‐39]. This node also included transcription factors GATA5, ELF3, MSRB2, MIER1, XRCC6 and ZFHX3 which are related to development. A node containing a small number of genes that were upregulated both in EB and in hES cells (node A) included GLI1, ISL1, CRABP1, and KRT9. Of note is that GLI1 activation is required in sonic hedgehog (Shh) signalling pathway [40], which is essential in regulating development, stimulation of the Shh pathway also results in the upregulation of GLI1 in hES cells [41], suggesting that Gli1 plays an important role in embryological development and hES cell differentiation.

The mRNAs that are predicted to be targets of specific miRNAs are expressed at significantly lower levels [42, 43]. This is likely caused by miRNA-mediated destabilization of target mRNA. To determine whether the negative correlation between miRNA and gene expression levels actually reflected miRNA-target relationships in hES cells, we calculated the correlation coefficients between the expression levels of hES upregulated miRNAs and the levels of their predicted targets. The predicted targets for each miRNA were downloaded from miRNAMap2.0 [44] and their expression value were extracted from our gene expression microarray data. To avoid random correlation, we calculated the correlation coefficients between miRNA expression levels and randomly-selected non-target genes of the same number. In general, the expression levels of miRNAs were both positively and negatively correlated with their predicted targets for all the miRNAs analyzed. However, we still observed a preponderance of negative correlation over positive correlation between some specific miRNAs and their targets. The distribution of the correlation coefficients for miR-302c-target genes was shifted toward the negative side compared to that of the miR-302c-non-target genes. This was also true for the miR-520b-target genes. The mean of the correlation coefficients between the two sets, targeted and non-targeted genes, was significantly different (p = 0.0003 for miR-302c and p = 0.049 for miR-520b) (Figure 5).

Using qRT-PCR we found that the expression levels of miR-302b, miR-302c, miR-367, miR-200c, miR-519b, and miR-520b were much higher in hES cells than in either EB or adult cells (Figure 6, panel A). The difference in the expression of miR-200c, miR-302b, and miR-367 between hES cells and EB, and between hES cells and adult cells was significant (P < 0.05). The difference in miR-302c expression between hES cells and adult cells was also significant (P < 0.05). In particular, the expression of miR-519b was 8-fold greater in hES cells than in EB cells and it was not even detected in adult cells. The expression of miR-520b was 26-fold greater in hES cells then in EB cells (P < 0.05) and it was detected only in two types of adult cells HMVEC and HUVEC.

Differences in the expression of EB signature miRNA were also confirmed by qRT-PCR. The expression of miR-106a, miR-106b, miR-17-5p, miR-92, miR-93, miR-130a, miR-20a and miR-190 were much higher in EB than in either hES cells or adult cells (Figure 6, panel B). For miR-106b, miR-92, miR-93, miR-130a and miR-190, the difference in their expression between EB and hES cells and between EB and adult cells were significant (P < 0.05). The difference in expression of miR-17-5p between EB and hES cells, and of miR-20a between EB and adult cells were also significant (P < 0.05).

We also confirmed that let-7b, let-7i, miR-221, miR-222 and miR-181a were much more highly expressed in adult cells (Figure 6, panel C). The differences in expression of these miRNA expression between adult cells and hES cells and between adult cells and EB were significant (P < 0.05). Of note, the expression levels of let-7b and let-7i were much higher in hES cells than in EB, and this result was consistent with both our microarray results and a previous report [26], indicating that the let-7 family plays important role in the maintaining hES cells function although their expression level was much lower than in adult cells. We also confirmed that miR-222 was more highly expressed in adult cells, although it was reported to be enriched in hES cells [17]. Actually, miR-222 was also expressed in multiple adult cell lines [16], forebrain and midbrain [45], and hippocampus [46]; it was upregulated in the differentiation process of undifferentiated hES cells to neural progenitor cells and then declined upon further differentiation [25]; it was also downregulated in erythropoietic culture of cord blood CD34+ progenitor cells [47].

Selected differentially expressed genes identified by microarray analysis were also validated via qRT-PCR. Markers for hES cells, POU5F1 (OCT4), LEFTY1 and TDGF1 were highly expressed in hES cells (Figure 7). The expression of OCT4 by hES cells was upregulated by 12-fold, LEFTY1 by 70-fold and TDGF1 by 19-fold compared to EB. Compared to adult cells, expression of OCT4 in hES was increased by 4,324-fold, LEFTY1 by 769-fold, and TDGF1 by 2,443-fold. We did not find that Nanog was upregulated in hES cells by microarray analysis, and by qRT-PCR its expression was increased by only 3-fold in hES cells compared to EB cells and 25-fold to adult cells. This discrepancy may have resulted from the fact that microarray platform is less sensitive than qRT-PCR. Analysis by qRT-PCR confirmed that both HAND1 and GATA5 were upregulated in EB, but were not detected in adult cells; HAND1 was only expressed in 1 hES sample, GATA5 expression was increased by 37-fold in EB cells compared to hES cells. The expression level of NFIB was much higher in adult cells than in either hES cells or EB (Figure 7) which was also consistent with the microarray results.

Among the miRNAs upregulated in hES cells, we observed 7 miRNAs were located in the miR-302 cluster and 21 miRNAs were located in miR-520 cluster. Most of these miRNAs had highly similar sequences at the 5' end seed region. In particular, miR-302a, miR-302b, miR-302c, miR-302d, miR-519b, miR-519c, miR-520a, miR-520b, miR-520c, miR-520d, and miR-520e had a consensus seed sequence: AAGUGC (Figure 8, panel A). To infer the function of these miRNAs, we predicted 2,436 targets for the miR-302 cluster and 4,691 targets for the miR-520 cluster by querying the public database miRNAMap 2.0 http://​mirnamap.​mbc.​nctu.​edu.​tw, and 2,284 target genes were shared by both clusters suggesting functional similarity. Gene Ontology (GO) enrichment analysis confirmed that the inferred functions of miRNAs within the miR-302 and miR-520 clusters were overlapping based on their involvement in cell growth, negative regulation of cellular metabolic process, negative regulation of transcription, and small GTPase mediated signal transduction. To visualize the functions of these miRNA targeted genes, a binary (red indicate participate in the functional category and green indicate not) heatmap was used to indicate functional commonality among all miRNAs in miR-302 and miR-520 clusters. MiR-520b, miR-302b, miR-302c, miR-302d, miR-519c, miR-520a and miR-302a were clustered closely base on the 48 GO terms analyzed. Interestingly, out of 48 functional categories analyzed, 6 related to chromatin structure were identified in this cluster, which included histone modification, covalent chromatin modification, establishment and or maintenance of chromatin architecture, chromosome organization and biogenesis, and chromatin modification (Figure 8, panel B).

Human embryonic stem cell lines WA09 (H9), TE06 (I6), and BG01v from WiCell Research Institute (Madison, WI), Technion-Israel Institute of Technology (Haifa, Israel) and ATCC (Manassas, VA) were cultured on mitotically inactivated mouse embryonic fibroblast (MEF) feeders using DMEM/F12 medium optimized for human ESC culture (GlobalStem Inc, Rockville, MD) supplemented with 20% knockout serum replacement and 4 ng/ml bFGF (both from Invitrogen, Gaithersburg, MD). Culture medium was changed daily and subculturing was performed every 4–6 days by collagenase IV (1 mg/ml) (Invitrogen, Gaithersburg, MD) digestion and mechanical disruption. The undifferentiation state of hES cells was determined by immunofluorescence detection of Pou5f1 (Oct4), Ssea4 (Millipore, Billerica, MA), Nanog (BD Bioscience, San Jose, CA), Sox2 (R&D Systems Inc. Minneapolis, MN), Tra-1-81 (Abcam, Cambridge, MA) and negative marker Ssea1 (Abcam, Cambridge, MA). The percentage of hES cells positive for Pou5f1 (Oct4), Sox2 and Ssea4 was measured by flow cytometry (FCM).

For embryoid body (EB) differentiation, hES cells were detached with collagenase IV and the cell aggregates were briefly triturated then cultured in ultra low attachment plates (Corning Inc, Corning, NY) for up to 14 days in maintenance medium. The medium was changed every three days.

The tested adult cells were Human Microvascular Endothelial Cells (HMVEC), Human Umbilical Vein Endothelial Cells (HUVEC), Umbilical Artery Smooth Muscle Cells (UASMC), Normal human astrocytes (NHA) (all from Lonza Inc, Walkersville, MD), and Lung Fibroblasts (LFB) (ATCC). All of the adult cells were cultured according to manufacturer's protocol.

A miRNA probe set was designed using mature antisense miRNA sequences (Sanger data base, version 9.1) consisting of 827 unique miRNAs from human, mouse, rat and virus plus two control probes. The probes were 5' amine modified and printed in duplicate on CodeLink activated slides (General Electric, GE Health, NJ, USA) via covalent bonding at the Infectious Disease and Immunogenetics Section of the Department of Transfusion Medicine (DTM) (Clinical Center, NIH, Bethesda, MD). 4 μg total RNA isolated by using Trizol reagent (Invitrogen, Gaithersburg, MD) was directly labelled with miRCURY™ LNA Array Power Labelling Kit (Exiqon, Woburn, MA) according to manufacture's procedure. The total RNA from Epstein-Barr virus (EBV)-transformed lymphoblastoid cell line was used as the reference for the miRNA expression array assay. The test sample was labelled with Hy5 and the reference with Hy3. After labelling, the sample and the reference were co-hybridized to the miRNA array at room temperature overnight in the present of blocking reagents as previously described[80] and the slides were washed and scanned by GenePix scanner Pro 4.0 (Axon, Sunnyvale, CA, USA).

Total RNA was extracted using Trizol reagent and the RNA quality was tested with the Agilent Bioanalyzer 2000 (Agilent Technologies, Santa Clara, CA). The RNA was amplified into antisense RNA (aRNA) as previously described[80]. Total RNA from PBMCs pooled from six normal donors was extracted and amplified into aRNA to serve as the reference. Both reference and test aRNA were directly labelled using ULS aRNA Fluorescent Labelling kit (Kreatech, Salt Lake City, UT) with Cy3 for reference and Cy5 for test samples. Whole-genome human 36K oligo arrays were printed in the Infectious Disease and Immunogenetics Section of Transfusion Medicine, Clinical Center, NIH (Bethesda, MD) using a commercial probe set which contains 35,035 oligonucleotide probes, representing approximately 25,100 unique genes and 39,600 transcripts excluding control oligonucleotides (Operon Human Genome Array-Ready Oligo Set version 4.0, Huntsville, AL). The design is based on the Ensemble Human Database build (NCBI-35c), with a full coverage on NCBI human Refseq dataset (04/04/2005). Hybridization was carried out at 42°C for 18 to 24 hours and the arrays were then washed and scanned on a GenePix scanner Pro 4.0 at variable photomultiplier tube to obtain optimized signal intensities with minimum (< 1% spots) intensity saturation.

The resulting gene expression data files were uploaded to the mAdb database and further analyzed using BRBArrayTools developed by the Biometric Research Branch, National Cancer Institute http://​linus.​nci.​nih.​gov/​BRB-ArrayTools.​html. Briefly, the raw data set was filtered according to standard procedure to exclude spots with minimum intensity and size. Then, the filtered data were normalized using Lowess Smoother. For miRNA array, the signal intensities were extracted via the R programming language (version 2.6.0, http://​cran.​r-project.​org/​) and the libraries provided by the Bioconductor project[81]. The background-subtracted data were then subject to variance stabilization normalization[82] and imported into BRBArray Tools http://​linus.​nci.​nih.​gov/​BRB-ArrayTools.​html. Differentially expressed miRNAs or genes were identified using F-tests with a P-value cutoff of 0.01 (miRNA) or 0.005 (gene); P-values were adjusted for multiple comparisons by False Discovery Rate < 0.05. Clustering and visualization of expression profiles was preformed with Cluster and Treeview software http://​rana.​lbl.​gov/​EisenSoftware.​htm[83]. The correlation between miRNA and target genes was performed using CCA package http://​cran.​r-project.​org/​web/​packages/​CCA/​index.​html[63]; for comparison, the expression level of non-target genes of the same number was also correlated with the miRNA expression. Density plot of correlation coefficient distribution was generated in R environment.

For validation of microarray data, differentially expressed genes were detected by using the pre-designed TaqMan^® Gene Expression Assays (Applied Biosystems, Foster City, CA). Differentially expressed miRNAs were measured by TaqMan microRNA Assays as previously reported [84]. The differences of expression were determined by relative quantification method; the Ct values of the test genes or miRNAs were normalized to the Ct values of endogenous control (RNU48 for miRNA and 18s rRNA for mRNA). The fold change was calculated using the equation 2^-ΔΔCt.

Gene target prediction was performed by querying the miRNA Database miRANDA [85] and RNAhybrid [86] through a miRNA gateway miRNAMap 2.0 http://​mirnamap.​mbc.​nctu.​edu.​tw[44]. Gene annotations were conducted using web-based tools Database for Annotation, Visualization and Integrated Discovery (DAVID, http://​david.​abcc.​ncifcrf.​gov/​) [87] or High-throughput GOminer http://​discover.​nci.​nih.​gov/​gominer/​htgm.​jsp[88]. The significantly (P < 0.05) enriched genes involved in biological process for miRNA targets were extracted; and heatmap was created using R 2.6.0.