|Home | About | Journals | Submit | Contact Us | Français|
Mesenchymal stem cells (MSCs) isolated from various tissues share common phenotypic and functional properties. However, intrinsic molecular evidence supporting these observations has been lacking. Here, we unravel overlapping genome-wide promoter DNA methylation patterns between MSCs from adipose tissue, bone marrow, and skeletal muscle, whereas hematopoietic progenitors are more epigenetically distant from MSCs as a whole. Commonly hypermethylated genes are enriched in signaling, metabolic, and developmental functions, whereas genes hypermethylated only in MSCs are associated with early development functions. We find that most lineage-specification promoters are DNA hypomethylated and harbor a combination of trimethylated H3K4 and H3K27, whereas early developmental genes are DNA hypermethylated with or without H3K27 methylation. Promoter DNA methylation patterns of differentiated cells are largely established at the progenitor stage; yet, differentiation segregates a minor fraction of the commonly hypermethylated promoters, generating greater epigenetic divergence between differentiated cell types than between their undifferentiated counterparts. We also show an effect of promoter CpG content on methylation dynamics upon differentiation and distinct methylation profiles on transcriptionally active and inactive promoters. We infer that methylation state of lineage-specific promoters in MSCs is not a primary determinant of differentiation capacity. Our results support the view of a common origin of mesenchymal progenitors.
Most human tissues contain populations of stem or progenitor cells. Multipotent cells isolated from adipose tissue, bone marrow, or skeletal muscle harbor mesenchymal stem cell (MSC) characteristics in vitro, such as plastic adherence, proliferation capacity, clonogenicity, immunophenotype, and ability to differentiate into several cell types (De Ugarte et al., 2003b ; Delorme et al., 2006 ; Kern et al., 2006 ; Peault et al., 2007 ; da Silva et al., 2008 ). Adipose stem cells (ASCs) and bone marrow (BM) MSCs express many similar surface markers (De Ugarte et al., 2003a ; Kern et al., 2006 ; da Silva et al., 2008 ), similar gene expression profiles (Boquest et al., 2005 ; Shahdadfar et al., 2005 ; Pedemonte et al., 2007 ) and adipogenic, osteogenic, and chondrogenic differentiation potential (De Ugarte et al., 2003a ; Kern et al., 2006 ). Satellite cells isolated from skeletal muscle can differentiate into myocytes, adipocytes, and osteocytes in vitro and their descendants, muscle progenitor cells (MPCs), can undergo multiple divisions before terminal myogenic differentiation (Peault et al., 2007 ). The overall resemblance of these progenitor cells suggests that they are of related ontogeny. Interestingly, cells called pericytes, with surface markers and multilineage differentiation capacity common to MSCs, have independently been shown to reside within the perivascular compartment of fat, bone marrow, muscle, and other tissues (Dellavalle et al., 2007 ; Crisan et al., 2008 ; Zannettino et al., 2008 ). These observations together raise the hypothesis of a common perivascular origin of MSCs (Crisan et al., 2008 ).
A common ontogeny of MSCs would predict that progenitor cells from various tissues exhibit some “intrinsic” similarity; however, there is currently no strong molecular evidence supporting this view. In an attempt to address this issue, we recently showed by bisulfite genomic sequencing that DNA methylation patterns in a handful of lineage-specific promoters in ASCs, BMMSCs, and MPCs were similar (Sørensen et al., 2009 ). Cytosine methylation in CpG dinucleotides constitutes a developmentally regulated epigenetic mark aiming at silencing genes whose expression is no longer required during development (Jaenisch and Bird, 2003 ). DNA methylation is carried out by DNA methyltransferases and is a reversible process, although mechanisms of active DNA demethylation remain incompletely unraveled (Ooi and Bestor, 2008 ). Promoter DNA methylation is not always associated with transcriptional repression. This relationship depends on promoter CpG content, with methylated high CpG promoters being usually inactive, whereas methylated low CpG promoters can either be active or inactive (Weber et al., 2007 ). Interestingly, differentiation of mouse embryonic stem (ES) cells into neuronal progenitors and subsequently into neurons has been shown to be accompanied by few DNA methylation changes, most of which occur during the first step of differentiation (Meissner et al., 2008 ; Mohn et al., 2008 ). The promoter methylation states of progenitor cells isolated from primary human tissues and the extent to which these are altered upon lineage-specific differentiation, however, remain uncharacterized.
Here, we surveyed and characterized DNA methylation profiles of all human RefSeq promoters in relation gene expression and differentiation, in adipose tissue-, bone marrow-, and skeletal muscle-derived mesenchymal progenitors, as well as in bone marrow-derived hematopoietic progenitors. Our data lend molecular support to the view of a common origin of mesenchymal precursors. The results also suggest an epigenetic programming of MSC differentiation potential by coenrichment in trimethylated lysine 4 and 27 on histone H3 over an unmethylated DNA background in promoters of lineage-specification genes.
ASCs were purified from the stromal vascular fraction of liposuction material from three donors and cultured (Boquest et al., 2005 ) as a pool. BMMSCs were isolated from marrow aspirates from two donors and cultured as described previously (Shahdadfar et al., 2005 ). CD34+ hematopoietic progenitor cells (HPCs) were isolated from bone marrow (Steidl et al., 2004 ). MPCs (CC-2580; Lonza, Allendale, NJ) were cultured in SkGM skeletal muscle medium (Lonza). Human neuronal progenitor cells (NPCs) were as described previously (Donato et al., 2007 ). Transient-amplifying keratinocyte precursor cells (KPCs) were isolated from epidermal sheets obtained from human neonatal foreskin biopsies with >99% purity based on CD45−/CD71bri/α6 integrinbri marker expression, as described previously (Li and Kaur, 2005 ). KPCs were flash frozen and used in uncultured state. Isolation, culture, and banking of ASCs, BMMSCs, and HPCs were done according to protocols approved by the Regional Committee for Ethics in Medical Research for Southern Norway (approval S-06387a and S-07043a).
ASCs were cultured to confluence in DMEM/F-12 medium containing 10% fetal calf serum and stimulated for 3 wk with 0.5 mM 1-methyl-3 isobutylxanthine, 1 μM dexamethasone, 10 μg/ml insulin, and 200 μM indomethacin (Boquest et al., 2005 ). Cells were stained with Oil Red-O to visualize lipid droplets. For myogenic differentiation, MPCs at 70% confluence were cultured for 6 d in DMEM containing 2% horse serum (Sørensen et al., 2009 ). Nuclei were stained with hematoxylin and eosin.
Genomic DNA was bisulfite-converted using MethylEasy (Human Genetic Signatures, Sydney, Australia) and amplified by polymerase chain reaction (PCR; Supplemental Table S1). PCR products were cloned into bacteria and sequenced as described previously (Noer et al., 2006 ). CpG methylation information is shown for approximately five bacterial clones.
MeDIP was performed in duplicate from 4 μg of DNA as described previously (Weber et al., 2007 ), with minor modifications (Sørensen and Collas, 2009 ). In brief, genomic DNA was treated with 30 μg/ml RNase A for 2 h at 37°C, diluted to 200 μl, and fragmented to ~200–800 base pairs, with enrichment in ~400-base pair fragments, by sonication on ice. Sonicated DNA was ethanol-precipitated using glycogen as a carrier and dissolved in 60 μl of MilliQ H2O (Millipore, Billerica, MA). Four micrograms of sonicated DNA was diluted in 450 μl of TE buffer (10 mM Tris-HCl, pH 8.0, and 1 mM EDTA), denatured for 10 min in boiling water, and immediately chilled on ice for 10 min. Fifty-one microliters of 10× immunoprecipitation (IP) buffer (1×: 140 mM NaCl, 10 mM Na-phosphate, pH 7.0, and 0.05% Triton X-100) and 10 μl of 5-methylcytosine antibodies (MAb-5MCYT; Diagenode, Liège, Belgium) were added and incubated for 2 h at 4°C on a rotating wheel. Prewashed Dynabeads M-280 sheep anti-mouse immunoglobulin (Ig)G (Invitrogen, Oslo, Norway) in 40 μl of 1× IP buffer was added and incubated for 2 h at 4°C on a rotator. Samples were collected by magnetic separation, washed, and immune complexes were digested with proteinase K for 3 h at 50°C. DNA was extracted with phenol-chloroform isoamylalcohol, ethanol precipitated, and dissolved in 15 μl of H2O overnight. Input DNA was fragmented and treated as described above except that no immunoprecipitation step was performed. Precipitated and input DNA was amplified using the WGA-2 Whole Genome Amplification kit (Sigma-Aldrich, St. Louis, MO) and cleaned up using the MinElute PCR purification kit (QIAGEN, Hilden, Germany).
For PCR assays, amplified and purified MeDIP and input DNA were diluted to ~25 ng/μl, and 1 μl was amplified by PCR using primers listed in Supplemental Table S1. PCR conditions were 95°C for 3 min and 30 cycles of 95°C for 30 s, 60°C for 30 s, 72°C for 30 s, followed by 7 min at 72°C. PCR products were visualized in a 1% agarose gel stained with ethidium bromide.
For hybridization to microarrays, input and MeDIP DNA fragments were labeled with Cy3 and Cy5, respectively, and hybridized on Roche-Nimblegen human HG18 RefSeq Promoter arrays (C4226-00-01; Nimblegen, Madison, WI). Signal intensity data were centered on zero using NimbleScan (Johnson et al., 2008 ). From scaled log2 MeDIP/Input ratios, a 750-base pair window was placed around each consecutive probe and a one-sided Kolmogorov–Smirnov (K-S) test was applied to determine whether probes were drawn from a significantly more positive distribution of intensity log2 ratios than those in the rest of the array. Resulting score for each probe was the P-value from the windowed test around that probe. Using NimbleScan, methylated peak data were generated from P values by searching for at least 2 probes with a P-value cut-off of 0.01 or less. Data were viewed using Nimblegen SignalMap and deposited under NCBI GEO GSE19795.
Correlation of log2 MeDIP/Input DNA ratios between replicates were computed using values from MaxTen calculations as described previously (O'Geen et al., 2006 ). This algorithm scores each promoter by finding the highest average log2 ratio among 10 consecutive probes per tiled region. Plotted MaxTen values were the average values from both MeDIP replicates for each cell type. Metagene calculations of average methylation enrichment over the tiled region were performed as described previously (Dahl et al., 2009 ) by using genes with a high probability of enrichment (K-S ≤0.05).
ChIP was performed essentially as described previously (Dahl and Collas, 2007 ). In brief, cells were cross-linked with 1% formaldehyde for 8 min, lysis buffer was added to ~120 μl, and samples were incubated for 5 min on ice. Cells were sonicated to produce fragments of ~400 base pairs. After centrifugation, the supernatant was collected, chromatin was diluted to 0.5 A260 units, and 100 μl was incubated with 2.4 μg of antibody coupled to magnetic Dynabeads protein A (Invitrogen) for 2 h at 4°C. After washing the ChIP material, 5 μg of RNAse A was added to the ChIP samples, DNA was eluted with 1% SDS and 50 μg/ml proteinase K for 2 h at 68°C, and DNA was dissolved in 10 μl of MilliQ water. ChIP DNA was analyzed by hybridization to the same promoter arrays as those used in MeDIP experiments. Antibodies to H3K9me3 were from Diagenode (pAb-056-050), H3K27me3 was from Millipore (07-449 [note: ex-Upstate catalog no. 05-851]), and H3K4me3 was from Abcam (Cambridge, United Kingdom; Ab8580).
ChIP and input DNA were amplified using the WGA4 kit (Sigma-Aldrich), cleaned up as described above, and eluted in 30 μl of MilliQ water. ChIP and input DNA fragments were labeled with Cy5 and Cy3, respectively, and hybridized to promoter arrays described above. Data were analyzed using NimbleScan (Johnson et al., 2008 ) and are accessible under NCBI GEO GSE17053. Peaks were detected by searching for at least four probes with a signal above a cut-off value using a 500-base pair sliding window. Ratio data were randomized 20 times to evaluate probability of false positives, and each peak was assigned a false discovery rate of 0.1 or less. Metagene assembly was done from genes with identified peaks as described previously (Dahl et al., 2009 ).
Gene ontology (GO) term enrichments within a target gene set were calculated using Bioconductor GOstats (Falcon and Gentleman, 2007 ). GOstats identifies functional terms for selected genes and provides a significance of enrichment for a term by giving a p value indicating the probability that the identified term is enriched among the target genes relative to what would be expected by chance based on the number of genes in the genome that belong to this term.
RNA was isolated using RNeasy Mini kit (QIAGEN). Biotin-labeled cRNA (1.5 μg) was hybridized onto Illumina Human-6 v2 Expression BeadChips (Illumina, San Diego, CA). Data were analyzed with Bioconductor (www.bioconductor.org). Present/absent calling relied on a classification based on detection p values calculated by Illumina Beadstudio software. Genes with detection p values ≤0.01 were classified as present, those with p values >0.05 were absent, and the rest were marginal. Microarray expression data are accessible in NCBI GEO database under accession GSE17053.
We addressed the epigenetic relationship, at the DNA methylation level, between progenitor cells isolated from human adipose tissue, bone marrow, and skeletal muscle by MeDIP-chip mapping of promoter DNA methylation profiles in ASCs, BMMSCs, MPCs, and HPCs (Figure 1A). Immunocaptured DNA fragments enriched in 5-methylcytosine were hybridized on promoter arrays tiling −2 kb to +0.5 kb relative to the transcription start site (TSS) of ~27,000 human promoters at 100-base pair resolution. Correlation analysis of log2 MeDIP/Input ratios for each cell type revealed high reproducibility between replicates (Supplemental Figure S1A).
Validation of the MeDIP approach was done at several levels. MeDIP-chip data were corroborated by bisulfite sequencing of randomly chosen promoters (Figure 1, B and C), by published bisulfite sequencing data for all cell types examined here (Noer et al., 2006 ; Sørensen et al., 2009 ) and by MeDIP-PCR single-gene analysis (Figure 1, D and E). MeDIP-PCR data were in addition verified for additional ASC, BMMSC, and MPC donors (Supplemental Figure S2). MeDIP-chip further corroborated published MeDIP-PCR data for methylated and unmethylated promoters in human fibroblasts (Weber et al., 2007 ) (Supplemental Figure S1, B and C). MeDIP-PCR also confirmed hypomethylation of the housekeeping UBE2B promoter and methylation of the H19 imprinting control region (H19ICR) reported previously in fibroblasts (Weber et al., 2007 ) (Figure 1E). Lastly, the proportions of methylated genes detected by MeDIP-chip in ASCs and BMMSCs (19% of 17,790 RefSeq genes in both cell types) were similar to those detected earlier by combined bisulfite restriction analysis (17 and 16%, respectively) among ~170 genes (Dahl et al., 2008 ), and methylation patterns reported for those genes, validated by bisulfite sequencing (Dahl et al., 2008 ), were corroborated by MeDIP-chip data.
We identified with high significance (K-S test p value ≤0.01 for detection of methylation “peaks”) >3300 promoters hypermethylated relative to genome-average methylation in ASCs and BMMSCs, 2630 in MPCs, and 3902 in HPCs (Figure 2A). These made up 15–22% of all RefSeq promoters represented on the array (Figure 2A). Hybridization patterns (Figure 2B) and calculated MaxTen values of methylation intensity for all promoters (Figure 2C) revealed high similarity and overlap between ASCs, BMMSCs, and MPCs. Intersect analysis of promoters with at least one hypermethylation peak showed that ASCs and BMMSCs shared 2486 hypermethylated genes (74% of all hypermethylated genes in these cell types; Figure 2, C and D). ASCs and BMMSCs, respectively, shared 1944 (57%) and 2053 (61%) hypermethylated genes with MPCs (Figure 2, C and D). We also identified a core of 1755 hypermethylated genes common to ASCs, BMMSCs, and MPCs, representing 52–66% of all hypermethylated genes in these cell types (Figure 2D). Another 20–30% was methylated in two of three cell types, whereas 15–20% was methylated only in one cell type (Figure 2E). These data indicate a high similarity of promoter DNA methylation patterns in progenitor cells from adipose tissue, bone marrow and skeletal muscle.
To determine whether the hypermethylated gene core was specific to mesenchymal progenitors, we also examined BM-derived CD34+ HPCs. We found that 91% of the 1755 core hypermethylated genes also were hypermethylated in HPCs, whereas HPCs contained 2302 hypermethylated genes that distinguished them from mesenchymal progenitors considered as a whole (Figure 3A). Moreover, 30–50% of genes found to be hypermethylated in ASCs, BMMSCs, or MPCs only (Figure 2D, crescents) were also hypermethylated in HPCs (Supplemental Figure S3). Lists of these genes hypermethylated in ASCs, BMMSCs, and MPCs are provided in Supplemental Table S2.
These results collectively indicate that promoter methylation profiles are similar but not identical among ASCs, BMMSCs, and MPCs, highlighting an intrinsic epigenetic identity between these mesenchymal progenitors. The majority of these genes are also hypermethylated in HPCs, which also contain an additional large set of hypermethylated genes.
To address the biological significance of the hypermethylated genes revealed by MeDIP-chip, we identified GO terms enriched among these genes (Figure 3B and Supplemental Table S3). Interestingly, genes hypermethylated in MSCs as a whole were enriched in signaling and developmental functions pertaining to early fetal development. Genes hypermethylated in HPCs were enriched in signaling functions linked to sensory perception, whereas genes hypermethylated in both MSCs and HPCs were associated with reproduction processes in addition to signaling, transcription regulation, and metabolic functions (Figure 3B). This finding corroborated the differential epigenetic programming of the germline and the soma shown previously by MeDIP-chip using similar promoter arrays (Weber et al., 2007 ). GO analysis therefore suggests that hypermethylation targets developmental functions disabled at the progenitor stage examined here, as well as late differentiation-associated functions.
A randomly chosen subset of early developmental genes identified above was shown to also be hypermethylated in NPCs and KPCs (Figure 3C), indicating that hypermethylation of these genes can occur in precursors of both mesodermal and ectodermal origin. Nonetheless, among the genes examined some (TBX3, ALX4, and PAX5) were hypomethylated in NPCs and/or KPCs (yet were as expected from our MeDIP-chip data hypermethylated in ASCs and BMMSCs) (Figure 3C), a pattern that may be linked to their role in neurogenesis and keratinocyte function (Asbreuk et al., 2002 ; Norhany et al., 2006 ; Pillai et al., 2007 ).
MeDIP-chip and bisulfite sequencing data have shown that in vitro differentiation of mouse ES cells into neuronal progenitors and subsequently into neurons is accompanied by remarkably few methylation changes, most of which occur during the first step of differentiation (Meissner et al., 2008 ; Mohn et al., 2008 ). This predicts that at least in this in vitro model, methylation patterns of differentiated cells would be established at the progenitor stage. To address this issue in primary progenitors, ASCs were differentiated in vitro into adipocytes and MPCs were differentiated into multinucleated myocytes. Differentiation was assessed by formation of Oil Red-O–positive lipid inclusions in adipocytes (Figure 4A), formation of multinucleated myocytes (Figure 4A), and up-regulation of lineage-specific genes in microarray expression analyses (Supplemental Table S4).
Promoter methylation changes after differentiation distinguished adipocytes from ASCs and myocytes from MPCs (Figure 4B). Nonetheless, most (~80%) hypermethylated promoters in undifferentiated cells remained hypermethylated (Figure 4C), suggesting that methylation states in differentiated cells are largely established at the progenitor stage. In addition, 29% of all methylated promoters identified in adipocytes were hypermethylated after ASC differentiation, whereas 15% of the methylated promoters in ASCs were hypomethylated (Figure 4, C and D). Similar observations were made after MPC differentiation (Figure 4, C and D). Thus, ASC and MPC differentiation is accompanied by methylation changes leading to an increase in the number of hypermethylated promoters in differentiated cells (p < 10−4; chi-square test with Yates' correction).
These data are consistent with enhanced transcriptional restrictions by DNA methylation as cells differentiate. To address the lineage specificity of these methylation changes, we cross-examined genes methylated in adipocytes and myocytes. Twenty percent of genes hypermethylated after differentiation were common to both cell types (Figure 4E). A subset of these genes was involved in stimulation-dependent changes in metabolism, consistent with differentiation induction (Supplemental Figure S4). Eighty percent of the hypermethylated genes, however, were cell type specific (Figure 4E). GO term enrichment analysis indicates that these were involved in the regulation of nuclear assembly, nuclear-cytoplasmic transport, and G-protein signaling in adipocytes (consistent with the completion of nuclear reorganization taking place during the formation of mature adipocytes), and in cell–cell interaction and exocytotic and sensory perception functions in myocytes (Supplemental Figure S4 and Supplemental Table S5). The reduced overlap of hypermethylated genes between adipocytes and myocytes, compared with ASCs and MPCs, reflects a greater epigenetic divergence between the two differentiated cell types than between their respective undifferentiated counterparts.
We next determined the extent to which differentiation-induced changes in promoter methylation reflected transcriptional changes. We first assessed the proportion of expressed genes in ASCs, BMMSCs, and MPCs by using Illumina expression arrays by defining present (expressed), marginal (weakly expressed), and absent (not expressed) cells. In each cell type, 54–57% of the hypermethylated genes were detected as expressed or weakly expressed. These percentages were similar to the proportion of expressed RefSeq genes detected in these cell types irrespective of methylation state (Supplemental Figure S5). Thus, promoter methylation is compatible with transcriptional activity (also see Weber et al., 2007 ).
We next determined transcriptional states associated with promoter hypo- or hypermethylation resulting from differentiation. After adipogenic differentiation, we found 702 genes overexpressed or induced (Supplemental Table S4), 645 of which were included on the Nimblegen platform. Of these, 102 (16%) were hypermethylated in undifferentiated ASCs. Among those methylated genes, 15 became demethylated, whereas 87 retained their methylation state. After myogenic differentiation of MPCs, 444 genes were overexpressed or induced (Supplemental Table S4), 417 of which were covered on the Nimblegen platform. Among those, 49 (12%) were hypermethylated in undifferentiated MPCs. Among those methylated genes, 13 became demethylated, whereas 36 retained their methylation state. These results indicate that the majority of genes up-regulated after MSC differentiation are DNA hypomethylated in undifferentiated cells. Moreover, among hypermethylated genes, only a quarter or less undergo methylation change.
We next addressed whether methylation occurred in distinct regions relative to the TSS in expressed versus nonexpressed genes in ASCs, BMMSCs, and MPCs. To this end, we determined average methylation by computing metagene profiles for all hypermethylated promoters. These profiles were distinct for transcriptionally active and inactive promoters (Figure 5A and Supplemental Figure S6). In all cell types, the amplitude of methylation enrichment was greater on promoters of expressed genes than nonexpressed genes (p values from Welsh two-sample t tests for methylation intensity amplitude in ASCs: p < 2.2 × 10−16; BMMSCs: p = 1.34 × 10−14; and MPCs: p = 3.04 × 10−3): enrichment was stronger on active promoters but sharply decreased to genome-average or below immediately 5′ of the TSS. In contrast, on inactive promoters, maximum enrichment was lower but was more widely spread by an additional 500-1500 base pairs to include the TSS, as determined by extension of the width at half-maximal enrichment (Figure 5, A and B, and Supplemental Figure S6). These data indicate that the profile of methylation coverage distinguishes promoters of expressed and nonexpressed genes. Nevertheless, the density of methylated CpGs was lower at the TSS than upstream in both expressed and repressed genes, corroborating recent genome-scale bisulfite sequencing data (Lister et al., 2009 ).
The relationship between promoter DNA methylation and gene activity has been shown to depend on CpG content (Weber et al., 2007 ). Thus, we asked whether methylation enrichment detected in the tiled regions in progenitor cells was influenced by promoter CpG content. Previous classification of human RefSeq promoters based on CpG density revealed a bimodal distribution from observed/expected CpG ratios, identifying high (HCP), intermediate (ICP), and low (LCP) CpG promoters (Weber et al., 2007 ). We applied the algorithm of Weber et al. (2007) to the tiled regions (−2.5 to +0.5 kb relative to the TSS) of all RefSeq promoters represented on the array, and we identified 11511 HCPs, 3173 ICPs, and 3246 LCPs; these numbers were comparable with those of Weber et al. (2007) .
In all cell types examined, CpG methylation targeted a higher proportion of ICPs relative to the proportion of ICPs in the genome (Figure 6A; p < 10−4; chi-square test with Yates' correction), at the expense of HCPs whose proportion was reduced among methylated promoters (p < 10−3 to 10−4). Methylation did not preferentially target LCPs except in hematopoietic progenitors where methylated LCPs were enriched (p = 0.0005). Thus, CpG methylation targets a higher proportion of intermediate to low CpG promoters compared with their proportions in the genome, in consistency with the enhanced protection of CpG islands against methylation (Weber et al., 2007 ; Irizarry et al., 2009 ; Straussman et al., 2009 ).
Having established that methylation differentially affects promoters with distinct CpG contents, we determined whether the nature of methylation changes (hypo- or hypermethylation) after adipogenic or myogenic differentiation differed between promoter classes. To this end, methylation changes identified in Figure 4D were reanalyzed for HCPs, ICPs, and LCPs. Figure 6B shows that in ASCs, hypomethylated genes were enriched in HCPs (p = 0.0003; Fisher's exact test) relative to the total number of hypomethylated genes, at the expense of ICPs (p < 10−4) and LCPs (p = 0.033). Furthermore, there was an enrichment of hypermethylated genes in LCPs (p = 0.0004), whereas HCPs and ICPs were not affected (p > 0.5). In MPCs, we also detected a trend in enrichment of hypomethylated genes in HCPs (p = 0.077) and an enrichment of hypermethylated genes in LCPs (p = 0.05) without significantly affecting HCPs and ICPs (Figure 6B). We concluded that differentiation-elicited hypomethylation predominantly affected methylated HCPs, whereas hypermethylation preferentially concerned LCPs.
Our previous bisulfite sequencing results suggested no predictability of MSC differentiation capacity based on the methylation state of a few lineage-specific promoters (Sørensen et al., 2009 ). Using our MeDIP-chip data, we extended our analysis of ASCs, BMMSCs, MPCs, and HPCs to 200 lineage-priming genes (including 50 HCPs and 150 non-HCPs) linked to differentiation into mesodermal, endodermal, and ectodermal lineages (Supplemental Table S6). These included 107 lineage-priming genes recently reported to be expressed at least at some level in BMMSCs (Delorme et al., 2009 ). We detected 57 hypermethylated promoters (28.5%) in at least one cell type, of which 8 (4%) were hypermethylated in all cell types. Methylation of these promoters did not occur in any particular developmental lineage for a given cell type, and we did not observe any significant difference in the proportion of hypermethylated promoters between cells, including HPCs. In fact, most promoters specifying mesodermal (adipogenic, osteogenic, chondrogenic, myogenic, and vascular), endodermal (pancreatic, hepatic) and ectodermal (neurogenic, skin) differentiation were not hypermethylated (Supplemental Figure S7 and Supplemental Table S6). These findings confirm the absence of relationship between methylation state and differentiation capacity of MSCs.
The lack of straightforward relationship between promoter DNA methylation and MSC differentiation capacity prompted the interrogation of additional epigenetic states on promoters. We examined by ChIP-on-chip in ASCs and in relation to DNA methylation the promoter enrichment profiles for trimethylated lysine 4 of histone H3 (H3K4me3), a transcriptionally permissive modification; H3K27me3, a Polycomb-mediated transcriptionally repressive mark; and H3K9me3, a mark of heterochromatin associated with repressed promoters (Kouzarides, 2007 ).
We identified 3362 promoters enriched in H3K4me3 and 2321 enriched in H3K27me3 (Figure 7A). GO term enrichment analysis showed that H3K4me3-marked genes were associated with transcription regulation, macromolecule synthesis, and metabolic processes, whereas H3K27me3-marked genes were distinctively enriched in developmental, differentiation and signaling functions (Figure 7B and Supplemental Table S7). Moreover, 25% of H3K4me3 promoters were coenriched in H3K27me3 (Figure 7, A and C) and displayed largely overlapping average enrichment profiles for these modifications, in contrast to promoters exclusively harboring either mark (Figure 7D). Although we do not have formal proof that these modifications co-occupy individual promoters, the metagene profiles together with previous sequential ChIP results (Noer et al., 2009 ) suggest that they might. GO terms enriched among H3K4/K27me3-enriched genes pertained to transcription regulation, development, differentiation, and cell adhesion (Figure 7B and Supplemental Table S7). These functional groups were similar to those of H3K4/K27me3 “bivalent” genes in ESCs (Bernstein et al., 2006 ; Pan et al., 2007 ; Zhao et al., 2007 ), in hematopoietic progenitors (Cui et al., 2009 ) and in embryos (Dahl et al., 2010 ). These findings extend the concept that H3K4/K27me3 coenrichment marks developmentally important promoters in stem and progenitor cells.
We next examined histone modifications associated with DNA methylated promoters (exemplified in Figure 7E). We found that of DNA methylated promoters, 22% were enriched in H3K4me3, 17% were enriched in H3K27me3, and <7% were trimethylated on H3K9 (Figure 7, A and C). These proportions were notably lower than those of H3K4me3-, H3K27me3-, and H3K9me3-enriched promoters among all modified RefSeq promoters (respectively, 37%, 24 and 17%; data not shown; p < 0.001; chi-square test with Yates' correction). Thus, DNA methylation and H3K4, K9, or K27 trimethylation seem to be largely exclusive at least in the promoter regions examined. H3K4/K27me3 coenrichment occurs mainly on weakly or unmethylated promoters (Figure 7C), a configuration reminiscent of the DNA hypomethylated state of developmentally regulated bivalent promoters in ES cells (Fouse et al., 2008 ; Mohn et al., 2008 ).
Nonetheless, a nonnegligible proportion of DNA methylated promoters was found to be enriched in H3K4me3 or H3K27me3 (Figure 7, A and C). These genes were enriched linked to transcription regulation, metabolic and synthetic processes (H3K4me3), early development, and differentiation (H3K27me3), or transcription and differentiation (H3K4/K27me3). These functional categories were similar to those defined by H3K4 or H3K27 methylation alone (Supplemental Table S7) and were not altered by DNA methylation states. Moreover, we found that the majority (80 to >90%) of H3K27me3- or H3K9me3-enriched genes were not expressed, whereas 60% of H3K4me3 genes were expressed (data not shown). These percentages were similar among DNA methylated genes and among all RefSeq genes bearing these marks (data not shown); thus, DNA methylation does not confer additional repressive effect on promoters harboring any of these histone modifications.
Our earlier data outlined distinct average DNA methylation enrichment profiles on the promoters of expressed versus nonexpressed genes (Figure 5A). To start addressing the biological significance of this observation, we examined histone modifications patterns on these promoters (Supplemental Figure S8). We first noted that only 23 and 28% of DNA methylated expressed and nonexpressed promoters, respectively, were enriched above genome-average level in any of the histone marks examined (see above). Second, of the active DNA-methylated promoters coenriched in trimethylated H3K4, K9, or K27, 85% were enriched in H3K4me3 alone (74%) or together with H3K27me3 (11%) (Figure 8A, left). A minority harbored H3K9me3 (2%) or H3K27me3 (13%) only, as expected. Inactive DNA-methylated promoters, in contrast, were predominantly enriched in H3K27me3 only (43%) or together with H3K4me3 (25%), or in H3K4me3 only (25%) (Figure 8A, right). H3K9me3 enrichment accounted again for only a minor proportion of these DNA-methylated repressed promoters (7%) at least within the tiled region (Figure 8A, right).
We next determined whether these proportions were different from those among all expressed or repressed RefSeq promoters that are also modified. Figure 8B shows that histone modification enrichment patterns on these promoters were very similar to those of DNA-methylated expressed or repressed promoters. Thus, differential histone modification marking of the promoter of expressed or repressed genes is independent of their DNA methylation state in the genomic regions examined. Lastly, we showed, however, that these histone modification patterns were clearly distinct form those of all RefSeq promoters, regardless of transcriptional status (Figure 8C). Therefore, in addition to the profile of DNA methylation, epigenetic markings such as trimethylated H3K4 and K27, and to a lesser extent K9, delineate distinct chromatin states on a subset of transcriptionally active versus inactive promoters (Figure 9A). Of these modifications, however, only H3K9me3 seems to be differentially enriched on DNA hypermethylated versus hypomethylated repressed promoters (p < 0.01; chi-square test; Figure 8, A and B).
A compilation of phenotypic, transcriptomic, and functional evidence argues that MSCs isolated from various adult tissues may be similar, although not identical (De Ugarte et al., 2003b ; Delorme et al., 2006 ; Kern et al., 2006 ). The finding that pericytes contain cells with MSC properties has raised the hypothesis that MSCs originate from a common perivascular niche within their respective tissues (Crisan et al., 2008 ; da Silva et al., 2008 ; Zannettino et al., 2008 ). In an analysis of all annotated RefSeq promoters, we show here a similarity of DNA methylation patterns in mesenchymal progenitors from adipose tissue, bone marrow, and skeletal muscle. We propose that a core of hypermethylated genes constitutes a common intrinsic epigenetic marking of ASCs, BMMSCs, and MPCs, lending support to the functional resemblance of MSCs identified in various tissues and to the view of a common origin of MSCs.
Not all pericytes, however, are MSC ascendants because pericytes also include hematopoietic precursors (Kiel and Morrison, 2006 ). This implies that differentially programmed progenitors coexist with MSCs in specific compartments. We find that 90% of the hypermethylated genes common to MSCs are also hypermethylated in HPCs. These genes are associated with regulation of development, transcription, signaling, and metabolic functions, arguing that promoter methylation contributes to repressing a common array of a wide range of functions in these precursor cells. HPCs also harbor another 2300 hypermethylated genes not identified in MSCs, suggesting that they are more epigenetically distant from MSCs than MSC types are from one another. The endoderm, mesoderm, and ectoderm specification and differentiation functions of genes hypermethylated in HPCs are consistent with an additional developmental restriction of HPCs relative to MSCs.
Does promoter methylation in progenitor cells reflect lineage programming, or pathways and processes no longer enabled at this stage of differentiation? We have recently proposed that strong methylation of lineage-specification promoters may impose a restriction on differentiation capacity (e.g., adipogenic and myogenic potential in HPCs, or endothelial potential of ASCs), whereas hypomethylation seems to have no prediction value on differentiation potential (Boquest et al., 2007 ; Sørensen et al., 2009 ). The present results establish that most endodermal, mesodermal, and ectodermal lineage-specific promoters are hypomethylated, even though differentiation into some of these lineages cannot be achieved by the cell types examined here (Kern et al., 2006 ). Thus, promoter methylation state may constitute a “ground state” program of gene activation potential, with strong methylation being repressive and hypomethylation being potentially permissive (Figure 9B). This model is compatible with a lineage-priming model of MSC differentiation (Boquest et al., 2006 ; Delorme et al., 2009 ).
The core of methylated genes identified in progenitor cells suggests that methylation is established before organogenesis and reflects inaccessibility to developmental programs no longer enabled. These genes are DNA methylated without any of the histone modifications examined here, or are coenriched in the repressive H3K27me3 (Figure 9A). This view is supported by ES cell differentiation studies showing that de novo DNA methylation occurs on pluripotency-associated loci when cells lose pluripotency, whereas subsequent terminal differentiation is accompanied by surprisingly only few methylation changes (Meissner et al., 2008 ; Mohn et al., 2008 ). Similarly, we show here that the majority of hypermethylated promoters in undifferentiated adipogenic or myogenic progenitors retain their methylation state after differentiation (Figure 9C). This indicates that the promoter methylation patterns of differentiated cells are already largely established at the progenitor stage.
A comprehensive methylation analysis recently identified tissue-specific differentially methylated regions located far from promoters or genes and suggested to undergo methylation changes during development (Irizarry et al., 2009 ; Straussman et al., 2009 ). Thus, promoter methylation is unlikely to be the primary determinant of differentiation programming in the soma; it may, however, be involved in, or result from, additional developmental restrictions upon terminal differentiation. Indeed, it is able to distinguish distantly related cell types such as gametes versus somatic cells (Weber et al., 2007 ), or as shown in this study, differentiated adipocytes versus myocytes, or hematopoietic versus mesenchymal progenitors.
What, then, determines MSC differentiation programs? In undifferentiated ES cells, promoters of early differentiation genes are often DNA hypomethylated and cooccupied by transcriptionally permissive H3K4me3 and repressive H3K27me3, creating a temporarily repressive chromatin state (Azuara et al., 2006 ; Bernstein et al., 2006 ; Mikkelsen et al., 2007 ; Fouse et al., 2008 ; Meissner et al., 2008 ; Mohn et al., 2008 ). Differentiation resolves this “bivalency” by removing trimethylation on H3K27 while maintaining H3K4me3 on expressed genes, whereas genes that remain or become repressed retain H3K27me3. In MSCs, these genes are often DNA methylated in the presence or absence of H3K27me3 (Figure 9A). Lineage-specific promoters involved in terminal differentiation are, however, DNA methylated in ES cells (Fouse et al., 2008 ) but are for the most part unmethylated in MSCs (this study). The view of lineage priming by promoter DNA hypomethylation and co-occupancy by H3K4me3 and H3K27me3 can now be extended to tissue-specific progenitors, including HPCs (Cui et al., 2009 ) and adipose-derived MSCs (this study; Figure 9A).
This and previous studies (Weber et al., 2007 ) indicate that promoter DNA methylation poorly correlates with promoter activity. However, although promoters of active genes can also be methylated, promoters of inactive genes seem to be more prone to a spreading of DNA methylation, particularly over the TSS (Figure 5). In addition, relative to the rest of the region examined, the density of methylated CpG is lower at the TSS regardless of promoter activity, in consistency with earlier findings (Weber et al., 2007 ; Lister et al., 2009 ; Straussman et al., 2009 ). Although constitutively unmethylated CpG islands may be protected from methylation by sequence determinants, the mode of recognition of absence of methylation at the TSS remains currently unknown (Straussman et al., 2009 ). It will be interesting to determine whether this is related to chromatin structure and particularly to the existence of unstable nucleosomes around TSSs, notably among expressed genes (Jin and Felsenfeld, 2007 ; Henikoff, 2008 ; Zilberman et al., 2008 ; Jin et al., 2009 ).
We thank Dr. Jan Brinchmann (Oslo University Hospital, Oslo, Norway) for generous gift of HPCs and BMMSCs, Dr. Joel Glover (University of Oslo, Oslo, Norway) for neuronal precursors, Dr. Pritinder Kaur (PeterMac Callum Cancer Centre, Melbourne, Australia) for keratinocytes progenitor cells. We thank Kristin Vekterud for expert technical assistance. This work was supported by the Research Council of Norway and a doctoral stipend from the University of Oslo (to B.M.J.).
This article was published online ahead of print in MBoC in Press (http://www.molbiolcell.org/cgi/doi/10.1091/mbc.E10-01-0018) on April 21, 2010.