The DNA methylome changes more during postnatal ISC maturation than during differentiation of ISCs
We focused mainly on ISCs from the colon since this region of the gut is the primary site of several human diseases, including inflammatory bowel disease and cancer. Although Lgr5+ cells are known to exhibit stem cell properties before birth and drive the dynamics of developing crypts in the small intestine [32], their ontogeny has not been examined in the developing colon. We, therefore, characterized the temporal emergence of Lgr5+ colonic ISCs by fluorescence activated cell sorting (FACS) at embryonic and fetal stages using knock-in Lgr5-EGFP-CreER mice [4]. Lgr5-EGFP+ colonic ISCs are detectable at embryonic day 14.5 (E14.5) and become an appreciable subpopulation by E18.5 (Additional file 1: Figure S1).
To determine whether the ISC population undergoes epigenetic changes during early postnatal life, we performed unbiased DNA methylation mapping by whole-genome bisulfite sequencing (WGBS) [33] in Lgr5-EGFP+ colonic ISCs at birth [postnatal day 0 (P0)] and at the end of the suckling period (P21). Since colonic ISCs are continually undergoing differentiation to replace the short-lived population of differentiated epithelial cells, we analyzed the DNA methylomes of sorted epithelial cell adhesion molecule (EpCAM)-positive and Lgr5-enhanced green fluorescent protein (EGFP)-negative population (the descendants of ISCs) [34–36] to identify DNA methylation changes that correlate with differentiation (differentiated cells versus ISCs). We found that at both ages, global methylation levels were significantly lower in differentiated cells relative to ISCs (P < 0.0001; Additional file 1: Figure S2). Across the 25 million CpG sites analyzed, average CpG methylation levels were 73.9 % and 70.3 % at P0, and 73.1 % and 70.0 % at P21, for ISCs and differentiated cells, respectively. We compared specific genomic regions that underwent methylation changes either during maturation from P0 to P21 [maturation-associated differentially methylated regions (mDMRs)] or during differentiation from ISCs to epithelial cells [differentiation-associated differentially methylated regions (dDMRs)]. Our analysis revealed that there are more mDMRs than dDMRs, particularly at CGI-associated genomic regions (Fig. 1a). Interestingly, at non-CGIs, mDMRs frequently lose methylation, but at CGIs, mDMRs predominantly gain methylation (Fig. 1a). There was no significant enrichments of a gene region for non-CGI associated DMRs as compared to known genes, regardless of mDMR or dDMR, methylation gains or losses (Fig. 1b). The pattern was completely different, however, at CGI-associated mDMRs; methylation gains at CGI mDMRs were strongly associated with the gene body (introns and exons, excluding the first and last exons) and 3′ end of genes (last exon and 3′ UTR) (Fig. 1b). Together, these results demonstrate that changes in genomic DNA methylation patterns are more dynamic during postnatal ISC maturation than during differentiation of ISCs to epithelial cells.
3′ CGI-associated mDMRs are functionally implicated in development and intestinal maturation
To evaluate the functional significance of mDMRs, we performed DAVID Gene Ontology (GO) analyses to identify statistically overrepresented biological processes (see Additional file 2: Tables S1 and S2 for lists of CGI and non-CGI mDMR genes). This analysis revealed multiple developmental processes significantly associated with mDMR genes showing methylation gains at non-promoter CGIs (Additional file 2: Table S3 contains a list of all significantly enriched GO terms). For example, genes with mDMRs in the gene body or 3′ CGI are enriched for embryonic organ development, intracellular signaling cascade, regulation of transcription, and cell morphogenesis involved in differentiation (Additional file 1: Figure S3). Of particular interest is the glycosphingolipid biosynthetic process, which is intimately linked to intestinal maturation by modulating mucus barrier function; glycosphingolipids may additionally modulate receptors for toxins, virus, and bacteria [37, 38]. Indeed, a careful examination of the gene list revealed a cluster of 3′ CGI genes responsible for glycan biosynthesis, including glycosidase (Net37), glycosyltransferases (A3galt2, B4galnt1, B4galnt4, and Gal3st1) and other related enzymes (Fkrp and Phospho1). When we compared the gene lists for 3′ CGI-associated mDMRs in ISCs and their differentiated progeny during the suckling period, we found, as expected, that most (60 %) of the mDMR genes in differentiated cells are found in ISCs (Additional file 1: Figure S4a). Since differentiated cells are more exposed to the intestinal lumen than are the ISCs in the crypts, mDMRs unique to differentiated cells may reflect responses to the very different luminal environment at P21 relative to P0. Not surprisingly, only 7 % (41/517) of mDMR genes in the differentiated cells overlap with P21 dDMR genes.
Methylation gains at 3′ CGIs of glycosylation genes correlate tightly with transcriptional activation in the developing ISCs
To gain insight into the function of the methylation changes at mDMRs, we used RNA-Seq to comprehensively examine the relationship between ISC mDMRs and gene expression. We found that 3′ CGI methylation was positively correlated with gene expression (Additional file 1: Figure S4b), suggesting that postnatal establishment of DNA methylation in ISCs plays a functional role in intestinal maturation. To validate our genome-wide results, we focused on ten candidate genes, including five with mDMRs at non-CGIs and five with mDMRs at CGIs. All the non-CGI mDMR genes (Rnf43, Zbtb22, Fam109a, Adamtsl5, and Mif) showed DNA methylation loss in the ISCs during the suckling period. They were selected because their mDMRs are proximal to CGIs, in regions known as CGI shores, which have been shown to exhibit tissue-specific methylation correlated with gene expression [39]. Notably, Rnf43 is a stem cell E3 ligase which acts as a negative regulator of Wnt signaling [40]. All the CGI mDMR genes (B4galnt1, Net37, Lpar5, Fkrp, and Phospho1) showed gain of methylation during ISC development. Among them, four (B4galnt1, Net37, Fkrp, and Phospho1) encode enzymes that affect glycosylation, and one (Lpar5) is involved in sodium and water absorption in the intestine [41].
To assess the correlation between DNA methylation and gene expression over an extended time-series, we analyzed DNA methylation and gene expression quantitatively in both ISCs and differentiated epithelial cells at E18.5, P0, P21, P100 (young adult), and P300 (old adult). In all cases, the results (Fig. 2) not only confirmed our WGBS findings but also demonstrated that ISC epigenetic regulation extends beyond the suckling period. Interestingly, we found no significant correlation between methylation changes at non-CGI genes and gene expression during the suckling period, but over the extended time course, expression increases lagged behind methylation decreases (Fig. 2a; Additional file 2: Table S4). Conversely, at all five of the 3′ CGI genes examined we found a positive temporal correlation between 3′ CGI methylation and transcriptional activation across development (Fig. 2b; Additional file 2: Table S4). To study at higher resolution the developmental dynamics of DNA methylation in these regions, we performed methylation analysis of 3′ CGI genes at P4, P7, P11, P16, and P18. The 3′ CGI methylation increases for several glycosylation genes (B4galnt1, Net37, and Fkrp) were most dramatic during the first week of postnatal life (Additional file 1: Figure S5), suggesting that some developmental signal coincident with parturition drives these changes. Most importantly, the nearly identical DNA methylation dynamics in ISCs and differentiated epithelial cells, from fetal to adult life (Fig. 2a, b), further support the conclusion from our genome-wide analyses that differentiation of ISCs does not entail widespread changes in DNA methylation.
DNMT1-mediated DNA methylation is essential for postnatal intestinal development
To further address the functional role of DNA methylation in intestinal development, we performed genetic studies by targeting genes responsible for the establishment and maintenance of DNA methylation. Dnmt3a and Dnmt3b are known as de novo methyltransferases (to establish methylation patterns during development) and Dnmt1 is considered a maintenance methyltransferase (to propagate established methylation patterns in daughter cells during mitosis) [42]. By examining the mRNA expression patterns of these three Dnmts from P0 to P300, we found that Dnmt3b expression is essentially undetectable in either ISCs or differentiated epithelial cells (Fig. 3a, green lines). Dnmt3a expression is intermediate and relatively constant (Fig. 3a, red lines), whereas Dnmt1 expression is the strongest in both ISCs and differentiated epithelial cells and increases progressively with age (Fig. 3a, blue lines). Based on these results, we focused our functional analyses on Dnmt3a and Dnmt1. We derived Dnmt3a
f/f
; Villin-Cre and Dnmt1
f/f
; Villin-Cre mouse lines to ablate individual Dnmt genes in an intestinal epithelial cell-specific fashion. In both Dnmt3a and Dnmt1 homozygous mutant mice, we detected a nearly 90 % reduction in Dnmt3a and Dnmt1 expression in neonatal intestines, respectively (Fig. 3b; Additional file 1: Figure S6). At P0, Dnmt3a and Dnmt1 homozygous mutant pups were indistinguishable from their control littermates in terms of body weight and gross appearance. By P7, however, Dnmt1 homozygous mutant mice showed significantly reduced body weight and shortened intestinal length compared with their littermate controls (Fig. 3c). Most (~80 %) of Dnmt1 homozygous mutant mice died around the time of weaning (P21). Survivors had crypts containing cells that had escaped Dnmt1 deletion. In contrast to the Dnmt1 mutants, no obvious changes were detected in Dnmt3a homozygous mutant mice (Additional file 1: Figure S7). To better characterize the postnatal lethal phenotype of Dnmt1 mutants, we performed histological and immunohistochemical analyses. These demonstrated that at P7, loss of Dnmt1 results in morphological changes characterized by villus-atrophy and vacuolated cells in both small intestine (Additional file 1: Figure S8) and colon (Fig. 3d). In addition, the Dnmt1 mutant colons showed fewer goblet cells (indicated by Alcian blue staining), and a decrease in Ki-67+ proliferative cells (Fig. 3d). These alterations were not apparent at P0 or P3 (Additional file 1: Figure S9), indicating that the observed epithelial deterioration occurs postnatally. Notably, signs of severe mucosal damage, including epithelial pseudostratification and adenoma-like lesions, were occasionally observed in P7–P21 Dnmt1 homozygous mutant mice (scored by pathologist; Additional file 1: Figure S8b). These histopathological changes potentially indicate indirect consequences of epithelial injury or inflammation.
To investigate the direct impact of Dnmt loss on postnatal epigenetic regulation of gene expression, we focused on the five previously characterized 3′ CGI mDMRs, and performed quantitative analyses of DNA methylation and gene expression in colonic epithelial cells at P7. Dnmt3a deletion caused significant methylation decreases for B4galnt1, Net37, and Lpar5 (Fig. 3e; Additional file 1: Figure S10), but these were only modestly associated with expression (Fig. 3f). Dnmt1 deletion, however, resulted in greater methylation decreases in four of the five genes (Fig. 3e; Additional file 1: Figure S10), causing stronger changes in expression (Fig. 3e). In all cases, loss of methylation was correlated with reduced expression, consistent with the notion that 3′ CGI methylation acts as a transcriptional activator. For Net37, the degree of expressional reduction far exceeded the alteration of methylation, reflecting broader impact of Dnmt1 loss on additional regulatory regions, consistent with previous observations [7, 9]. Together, these results suggest a regulatory function of 3′ CGI methylation in postnatal mouse intestinal development, providing the first in vivo evidence to support the involvement of Dnmt1 in controlling postnatal epigenetic regulation and epithelial maturation.
Dnmt1-mediated DNA methylation is required for the control of adult ISC homeostasis
The data above on pan-intestinal epithelial cell deletion (by Villin-Cre) raised an intriguing possibility that Dnmt1-mediated postnatal epigenetic mechanisms could be involved in the regulation of adult ISC function. To test this, we derived a Dnmt1
f/f
; Lgr5-EGFP-CreER mouse line, in which deletion of Dnmt1 can be induced in adult ISCs by tamoxifen (referred to hereafter as Dnmt1
ISCKO). We used two types of controls: the Dnmt-wild type; Lgr5-EGFP-CreER mice with the same tamoxifen treatment, and the Dnmt1
f/f
; Lgr5-EGFP-CreER mice without tamoxifen treatment. Based on real-time RT-PCR analysis of Dnmt1 expression in Lgr5-EGFP+ cells, we estimated that tamoxifen treatment induces around 80 % ablation of Dnmt1 in both small intestine and colon (Fig. 4a). Using confocal immunofluorescent analyses for Dnmt1 and EGFP, we analyzed Dnmt1 protein expression in Lgr5+ crypt-based columnar (CBC) cells. Tamoxifen-injected wild-type control mice strongly expressed Dnmt1 in the Lgr5-EGFP+ CBCs (white arrowheads, top panel in Fig. 4b). Expression of Dnmt1 was also detected in villus epithelial cells, consistent with our previous real-time RT-PCR results (Fig. 3a). In small intestines of Dnmt1
ISCKO mice, co-staining of EGFP and Dnmt1 showed decreased Dnmt1 expression in Lgr5-EGFP+ ISCs (solid line marks the mutant crypt, Fig. 4b). It is known that Cre recombinase is expressed mosaically in the intestinal epithelium of Lgr5-EGFP-CreER mice [4]. Indeed, in the same Dnmt1
ISCKO mice, some Lgr5-EGFP-negative crypts, which are Dnmt1 wild type as a result of no recombination, retained strong Dnmt1 staining (dotted line circles the wild-type crypt with white arrowheads pointing to the Dnmt1+ cells, Fig. 4b, bottom panel). Similar mosaic deletion patterns were observed in colons of Dnmt1
ISCKO mice (solid circles indicate mutant crypts, and dashed circles indicate wild-type crypts with arrowheads pointing to Dnmt1+ cells, Fig. 4c). Therefore, our RT-PCR and immunostaining results for Lgr5+ cells suggest efficient but incomplete ablation of Dnmt1 from the adult ISCs. Notably, compared with the well-defined triangle shapes of wild-type CBCs (arrowheads in Fig. 4b), mutant CBCs in Dnmt1
ISCKO mouse crypts showed altered morphologies with most cells aggregating into clumps (solid circles, Fig. 4b; Additional file 1: Figure S11), suggesting dysregulated crypt homeostasis. Our observations on Paneth cells also supported this notion. In small intestines of Dnmt1
ISCKO mice, Lysozyme+ Paneth cells lost their clear compartmentalization with ISCs, and some were mislocalized in the villus epithelium (arrowheads, bottom panel in Fig. 4d). Normally, Paneth cells are localized in the crypt (top panel in Fig. 4d).
In parallel, we derived a Dnmt3a
f/f; Lgr5-EGFP-CreER mouse line and applied the same strategy to evaluate the functional consequences of induced Dnmt3a deletion in adult ISCs (Dnmt3a
ISCKO). After tamoxifen administration, we confirmed the high-efficiency of gene ablation in Lgr5-EGFP+ ISCs from Dnmt3a
ISCKO animals (Additional file 1: Figure S12a). In contrast to Dnmt1
ISCKO, Dnmt3a
ISCKO small intestine and colon exhibited no obvious morphological abnormalities (Additional file 1: Figure S12b). Lysozyme staining revealed normal Paneth cell localization (top panel in Fig. 4d). Overall, these results were consistent with our observation in the Dnmt3a
f/f; Villin-cre mice.
Finally, to determine whether dysregulation of postnatal epigenetic mechanisms might be involved in the observed phenotypic defects of Dnmt1-deficient ISCs, we analyzed DNA methylation and gene expression of three genes (B4galnt1, Net37, and Lpar5) that normally undergo increases in DNA methylation at 3′ CGIs. All three exhibited significantly reduced DNA methylation in the ISCs after disruption of Dnmt1 but not Dnmt3a (Fig. 4e; Figure S13). As in the Villin-Cre experiments (Fig. 3e, f), hypomethylation at the 3′ CGIs led to reduced mRNA expression (Fig. 4f). These results provide further evidence that Dnmt1 is a prominent player in control of the postnatally programmed DNA methylation in ISCs, while Dnmt3a is somewhat dispensable. Collectively, these data indicate that Dnmt1-mediated developmentally programmed epigenetic mechanisms are involved in the regulation of adult ISC function.
Microbial exposure guides developmental epigenetics in the colonic epithelium
Our discovery of postnatal epigenetic regulation in the ISC population highlights a potential critical ontogenic window in which the infant microbiome could influence intestinal development. To test this, we profiled DNA methylation in genetically identical C57 wild-type mice under either conventional (CNV) or germ-free (GF) conditions. We collected the colons at E15.5, P0, P21, and P100, and used the unfractionated tissues for methylation analysis. We quantitatively measured DNA methylation of over 90 CpG sites spanning 15 genes identified by WGBS as showing postnatal methylation changes in ISCs. In addition, to rule out the possibility that microbiome effects on DNA methylation were simply due to the changes of global methylation, we analyzed the methylation of two generic repetitive elements (Line1 and IAP). Remarkably, in CNV mice, an unsupervised hierarchical clustering based purely on DNA methylation yielded a perfect correspondence with stage of development (Fig. 5a). In GF mice, however, these developmental changes in DNA methylation were markedly dysregulated, at both P21 and P100 (Fig. 5b). The most pronounced effect was in a large block of regions normally targeted for methylation increases from P0 to P21; in GF mice many of these changes did not occur, even by P100 (Fig. 5b, indicated by purple boxes). Additionally, we found similar methylation profiles of newborn mice (P0) in both CNV and GF conditions, consistent with these epigenetic changes being induced by colonization. It is worth mentioning that the macronutrient composition of the GF mouse diets was similar to that of the control CNV mouse diet (Additional file 2: Table S5). Further, we observed essentially the same results for GF mice housed at different facilities. Interestingly, we found significantly increased mRNA levels of Dnmt1 in the GF mice at P21 and P100 (Additional file 1: Figure S14), indicating that the methylation defects under the GF condition are not due to insufficient Dnmt1 activity.
To test whether the effects of the microbiome on DNA methylation might be mediated by altered cellular composition, and also to evaluate effects on gene expression, we obtained another cohort of mice at weaning under either CNV (N = 5) or GF (N = 5) conditions. Rather than studying whole colon, we isolated intestinal epithelial cells and assessed DNA methylation and expression of five 3′ CGI-associated genes. In agreement with our previous results, the GF condition significantly reduced the developmentally programmed methylation at all genes analyzed (Fig. 6a). We observed significantly reduced gene expression for the two genes with the greatest methylation decrements in GF mice (B4galnt1 and Phospho1; Fig. 6b). Thus, the alterations in developmental epigenetics we observed in the whole colon of GF mice (Fig. 5b) occur within colonic epithelial cells.
Finally, to address the direct link between postnatal DNA methylation and the gut microbiome, we performed fecal microbiota transplant (FMT) experiments to conventionalize the GF mice at age P25. Based on our previous results, we examined 3′ CGI methylation of B4galnt1 and Phospho1 in colonic epithelial cells at P100 under FMT (N = 3) compared with GF (N = 3) and CNV (N = 2) conditions. As shown in Additional file 1: Figure S15, reestablishing commensal microbiota significantly increases DNA methylation at multiple CpG sites of 3′ CGIs. Consistently, restored DNA methylation at the 3′ CGI of the Phospho1 gene in FMT mice leads to significantly increased gene expression.