Genome-wide profiling of histone H3K27 acetylation featured fatty acid signalling in pancreatic beta cells in diet-induced obesity in mice
Epigenetic regulation of gene expression has been implicated in the pathogenesis of obesity and type 2 diabetes. However, detailed information, such as key transcription factors in pancreatic beta cells that mediate environmental effects, is not yet available.
To analyse genome-wide cis-regulatory profiles and transcriptome of pancreatic islets derived from a diet-induced obesity (DIO) mouse model, we conducted chromatin immunoprecipitation coupled with high-throughput sequencing (ChIP-Seq) of histone H3 lysine 27 acetylation (histone H3K27ac) and high-throughput RNA sequencing. Transcription factor-binding motifs enriched in differential H3K27ac regions were examined by de novo motif analysis. For the predicted transcription factors, loss of function experiments were performed by transfecting specific siRNA in INS-1, a rat beta cell line, with and without palmitate treatment. Epigenomic and transcriptional changes of possible target genes were evaluated by ChIP and quantitative RT-PCR.
After long-term feeding with a high-fat diet, C57BL/6J mice were obese and mildly glucose intolerant. Among 39,350 islet cis-regulatory regions, 13,369 and 4610 elements showed increase and decrease in ChIP-Seq signals, respectively, significantly associated with global change in gene expression. Remarkably, increased H3K27ac showed a distinctive genomic localisation, mainly in the proximal-promoter regions, revealing enriched elements for nuclear respiratory factor 1 (NRF1), GA repeat binding protein α (GABPA) and myocyte enhancer factor 2A (MEF2A) by de novo motif analysis, whereas decreased H3K27ac was enriched for v-maf musculoaponeurotic fibrosarcoma oncogene family protein K (MAFK), a known negative regulator of beta cells. By siRNA-mediated knockdown of NRF1, GABPA or MEF2A we found that INS-1 cells exhibited downregulation of fatty acid β-oxidation genes in parallel with decrease in the associated H3K27ac. Furthermore, in line with the epigenome in DIO mice, palmitate treatment caused increase in H3K27ac and induction of β-oxidation genes; these responses were blunted when NRF1, GABPA or MEF2A were suppressed.
These results suggest novel roles for DNA-binding proteins and fatty acid signalling in obesity-induced epigenomic regulation of beta cell function.
The next-generation sequencing data in the present study were deposited at ArrayExpress.
Dataset name: ERR2538129 (Control), ERR2538130 (Diet-induced obesity)
Repository name and number: E-MTAB-6718 - RNA-Seq of pancreatic islets derived from mice fed a long-term high-fat diet against chow-fed controls.
Dataset name: ERR2538131 (Control), ERR2538132 (Diet-induced obesity)
Repository name and number: E-MTAB-6719 - H3K27ac ChIP-Seq of pancreatic islets derived from mice fed a long-term high-fat diet (HFD) against chow-fed controls.
KeywordsEpigenetics Fatty acid oxidation Glucose intolerance High-fat diet Histone acetylation Insulin secretion Next-generation sequencing Obesity Pancreatic islets Transcriptome Type 2 diabetes
Chromatin immunoprecipitation coupled with high-throughput sequencing
The Database for Annotation, Visualization and Integrated Discovery
Forkhead box A1
GA repeat binding protein α
Gene Expression Omnibus
Glucose-stimulated insulin secretion
Genome-wide association study (studies)
Histone H3 lysine 27 acetylation
HNF1 homeobox A
Hypergeometric Optimization of Motif EnRichment
Intraperitoneal glucose tolerance test
Kyoto Encyclopedia of Genes and Genomes
v-maf musculoaponeurotic fibrosarcoma oncogene family protein K
Myocyte enhancer factor 2A
Nuclear respiratory factor 1
Regulatory factor X, 7
High-throughput RNA sequencing
Spatial clustering for identification of ChIP-enriched regions
Transcription start site
Type 2 diabetes and obesity are chronic metabolic disorders that affect an increasing number of people globally [1, 2]. Genome-wide association studies (GWAS)  have highlighted the role of cis-regulatory elements in various disease processes by combining knowledge of comprehensive epigenetic profiling [4, 5, 6]. However, given the modest effect of identified susceptibility variants, achieving a full understanding of a disease remains challenging .
There is clear evidence supporting the role of environmental factors in type 2 diabetes [1, 2, 7]. Obesity, along with reduced physical activity, is regarded as being responsible for the recent pandemic . Insufficient insulin secretion is essential in the development of type 2 diabetes, and a variety of environmental factors as well as genetic predisposition play important roles in the modification of beta cell function .
Much remains unknown, especially in humans, about the molecular characteristics of the pancreatic islets of Langerhans under various physiological and pathological conditions . Several studies have provided new insights through more comprehensive analyses [4, 9, 10, 11]. Among assays for epigenomic profiling, histone H3 lysine 27 acetylation (H3K27ac), a marker of active regulatory regions , is recognised as being valuable not only in annotating human variants  but also in providing information on diseased tissue from both humans and mice [13, 14].
Changes in epigenetic modifications and transcriptional regulation are essential for phenotypic adaptation of cell types to environments, including high-fat diet (HFD) [15, 16, 17]. Since some DNA-binding proteins play a role in triggering epigenetic reprogramming in specific genomic regions , we hypothesised that the corresponding sequence motifs could be enriched in cis-regulatory regions with local epigenetic alterations in response to environmental stimuli. In the present study, by examining genome-wide H3K27ac signatures and whole transcriptome, we aimed to identify sequence motifs enriched in genomic regions exhibiting HFD-driven epigenetic changes in mouse pancreatic islets. In addition, we investigated the involvement of DNA-binding proteins in the regulation of gene expression and pancreatic beta cell function during the development of obesity and glucose intolerance.
Mice and metabolic assessment
Male C57BL/6JJcl mice were purchased from CLEA Japan (Tokyo, Japan). Half of the mice were switched from a standard chow diet (CE-2; CLEA Japan) to an HFD (D12492, 60% fat; Research Diets, New Brunswick, NJ, USA) at 10 weeks of age. The samples were not randomised by statistical computing. Mice were weighed to the nearest 0.1 g. Blood glucose was determined by tail-vein bleeding. Following fasting of the mice overnight, an intraperitoneal glucose tolerance test (IPGTT) was conducted using a glucose dose of 1 g/kg body weight. Plasma was separated by centrifugation of whole blood at an ambient temperature of 22 ± 2°C and frozen in liquid nitrogen for measurement of insulin levels using a Mouse Insulin ELISA Kit (Morinaga Institute of Biological Science, Kanagawa, Japan). All animal procedures at the National Center for Global Health and Medicine were approved by the Institutional Animal Care and Use Committee. Blinding for experimenters was not carried out. We did not lay down any criteria for inclusion and exclusion of any data, since all the results supported the main findings of the study. For details, see electronic supplementary material (ESM) Methods.
Isolation of pancreatic islets
High-throughput RNA sequencing
Total RNA was extracted from pancreatic islets using Trizol (Thermo Fisher Scientific, Waltham, MA, USA). Library preparation was performed using the TruSeq Stranded Total RNA Sample Prep Kit (Illumina, San Diego, CA, USA) and high-throughput RNA sequencing (RNA-Seq) was performed using the Illumina HiSeq 2000 platform with 100 bp paired-end sequencing. See ESM Methods for details.
Analysis of differential gene expression and functional annotation
Analyses were performed using DESeq 1.32.0  with an iterative pipeline (iterative differentially expressed gene elimination strategy [iDEGES]/DESeq) , both of which were R/Bioconductor packages (https://www.bioconductor.org/). Reads per million (RPM)-normalised read counts allowed the visualisation of global changes in gene expression as an MA plot. For those genes with highest variance (p < 0.1), functional annotation was analysed using the Database for Annotation, Visualization and Integrated Discovery (DAVID) 6.8 (https://david.ncifcrf.gov/, accessed 15 Feb 2018) . See ESM Methods for details.
Chromatin immunoprecipitation, high-throughput sequencing and analysis of differential sequencing signals
Chromatin preparation for chromatin immunoprecipitation (ChIP) was performed as described in  with modifications. A list of the primary antibodies used is provided in ESM Table 1. ChIP libraries were prepared using NEBNext ChIP-Seq Library Prep Master Mix Set for Illumina (New England BioLabs, Ipswich, MA, USA) and were sequenced on a HiSeq 2000 system (Illumina). The processed data were uploaded to the Integrative Genomics Viewer (IGV) browser 2.3.75 (http://software.broadinstitute.org/software/igv/) . Spatial clustering for identification of ChIP-enriched regions (SICER) v1.1  was used to compare differentially enriched regions between experimental and control groups using the command SICER-df-rb.sh. Running this command enabled identification of significantly enriched H3K27ac regions in each library, merging of all overlapping intervals and determination of the significance of changes by comparing the signal intensity of ChIP coupled with high-throughput sequencing (ChIP-Seq) in HFD samples with that in controls on each merged H3K27ac region. To identify DNA sequence motifs enriched in merged H3K27ac regions with HFD-induced changes, we used Hypergeometric Optimization of Motif EnRichment (HOMER) v4.6 (http://homer.ucsd.edu/homer/)  for de novo motif detection using the whole genome as the background by running the command findMotifsGenome.pl. This command is a tool used to identify binding motifs enriched in the genomic intervals without prior knowledge of the transcription factor binding sites. First, analyses were performed according to the default parameters for base redundancy and motif length (8, 10 or 12 bp). Subsequently, the identified motifs were checked for homology with known motif matrices and the best match result was presented as a likely candidate, which we listed in our tables. See ESM Methods for details.
Correlating differential H3K27ac signals with differential gene expression
Each H3K27ac region was mapped to the closest gene based on distance to the transcription start site (TSS), as described . Overall cis-regulatory activity of individual genes was defined as increased, decreased or unchanged by counting the number of regions showing increased H3K27ac relative to regions showing a decrease. DESeq2 1.20.0  enabled comparison of log2 (fold change) between categories. For Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis , we selected the 610 most upregulated and 174 most downregulated genes, which also showed overall increases and decreases in cis-regulatory activity, respectively. See ESM Methods for details.
Cell culture and insulin secretion measurements in INS-1 cells and pancreatic islets
INS-1 cells (a rat beta cell line) were gift from C. B. Wollheim (Lund University; University of Geneva) and N. Sekine (University of Geneva), and were cultured as described . Batch insulin release was estimated by incubating in 3 mmol/l glucose, 25 mmol/l glucose or 30 mmol/l KCl for 60 min. For pancreatic islets isolated from 54-week-old mice, insulin secretion in 2.8 mmol/l glucose, 16.7 mmol/l glucose or 60 mmol/l KCl in 30 min batch incubations was measured as described . Total insulin content was calculated by summating the insulin secreted into the medium with the amount of insulin measured by whole-cell extraction with acid ethanol for 24 h at 4°C. Insulin secretion was expressed as a percentage of the total insulin content. For INS-1 cells, insulin was detected using a Mouse Insulin ELISA Kit (AKRIN-011 T; Shibayagi, Gunma, Japan). Mycoplasma contamination was not tested. See ESM Methods for details.
siRNA knockdown of NRF1, GABPA and MEF2A followed by palmitate treatment
Lipofectamine 2000 (Thermo Fisher Scientific) was used for reverse transfection of specific siRNAs in INS-1 cells (ESM Table 2). Palmitate (Wako Pure Chemical Industries, Osaka, Japan), sodium palmitate (Sigma-Aldrich, St Louis, MO, USA) or vehicle treatment began 36 h after siRNA-mediated knockdown, and treated cells were used for the following experiments after incubation for 48 h. See ESM Methods for details.
Quantitative real-time PCR
Quantification of cDNA or ChIP-DNA samples was conducted via standard quantitative PCR (qPCR) methods using SYBR Green (Thermo Fisher Scientific). The following genes were tested: Acaa2, Gabpa, Mef2a, Nrf1 and Slc25a20 for rat cDNA; Acaa2, Cpt1a, Elovl6, Fads2, Mki67, Scd2, Slc25a20 and Top2a for mouse cDNA and Acaa2 and Slc25a20 for rat ChIP-DNA. A list of primers is provided in ESM Table 3. See ESM Methods for details.
We used R or GraphPad Prism 7.04 (GraphPad Software, San Diego, CA, USA) for statistical analyses. The two-sided unpaired t tests, two-sided paired t tests, or Mann–Whitney U tests were applied. Error bars in the Figures represent SD.
Previous ChIP-Seq data in human or mouse tissues were obtained from Gene Expression Omnibus (GEO) and processed as described above (summarised in ESM Table 4). The UCSC liftOver tool and Ensembl Compara cross-species resources (https://www.ensembl.org/info/genome/compara/index.html) were applied for conversion of genomic positions between reference genomes mm9, hg19 and rn6. Data in the present study were deposited at ArrayExpress under accession number E-MTAB-6718 (RNA-Seq) and E-MTAB-6719 (ChIP-Seq).
Body weight and glucose tolerance in mouse model of diet-induced obesity
Impact of long-term HFD feeding on pancreatic islet transcriptome
To clarify the global relationships between epigenome and transcriptome in the pancreatic islets after HFD feeding for an extended period of 27 weeks, we used a next-generation sequencing platform (Fig. 1a and ESM Fig. 2). RNA-Seq analyses were performed by generating approximately 31.7–35.1 million high-quality paired-end sequencing reads (ESM Table 5).
To identify genes differentially expressed between DIO and control mice, we chose DESeq  with an iterative pipeline  for read count normalisation and gene ranking. The analysis revealed that most genes showed relatively limited dynamics with respect to the read count ratio (ESM Fig. 3a). Nevertheless, analysis of the top-ranked 303 upregulated genes (p < 0.1) (ESM Fig. 3b and ESM Tables 6, 7) using DAVID  revealed enrichment of terms that were linked to islet functions important for adaptive responses to HFD, including the endoplasmic reticulum  (ESM Fig. 3b and ESM Table 7).
We also performed the same analysis on the top-ranked 269 downregulated genes (p < 0.1) (ESM Fig. 3c and ESM Tables 6, 8). Interestingly, the enriched terms were mainly related to epigenetic regulation of gene transcription, which may suggest the involvement of epigenetic change in pancreatic islets following prolonged HFD feeding (ESM Fig. 3c and ESM Table 8).
Genome-wide changes in H3K27ac associated with differential gene expression under HFD feeding
SICER also detected changes in the ChIP-Seq signal in individual merged regions . We found 13,369 increased and 4610 decreased H3K27ac regions in the islets derived from DIO mice (Fig. 2a, ESM Fig. 2 and ESM Table 12), with median fold changes in normalised enrichment of 1.30 (first quartile 1.20, third quartile 1.41) and 0.73 (first quartile 0.67, third quartile 0.80), respectively. Compared with unchanged and decreased H3K27ac, we found that increased H3K27ac showed a greater distribution in the promoter-proximal regions (Fig. 2b and ESM Fig. 4a–d). Of all the H3K27ac regions, increased H3K27ac accounted for 68.1% (6943/10,188) from −5 kb to +5 kb relative to the TSS (Fig. 2b), while it accounted for less than 22.0% (6420/29,132) if located >5 kb relative to the TSS (Fig. 2b).
To connect epigenetic changes with the transcriptome, a precise regulatory map of the cell types is vital . In the present study, however, we assigned a single gene to each merged H3K27ac region using only genomic positioning information (termed a ‘single nearest gene association rule’ in the Genomic Regions Enrichment of Annotations Tool [GREAT] ). A similar approach was also employed in a previous study that successfully analysed a mouse model of Alzheimer’s disease . Accordingly, expression of 9391 and 2794 genes were assumed to be associated with increased and decreased H3K27ac, respectively (ESM Table 6 and ESM Fig. 5a).
As 13.3% (1632/12,283) of all H3K27ac-harbouring genes revealed regulatory regions consisting of both increased and decreased H3K27ac (ESM Fig. 5c and ESM Table 6), we investigated how changes in individual gene expression arose according to the number of increased H3K27ac sites relative to those decreased (Fig. 2c,d and ESM Fig. 5b). Compared with genes containing equal numbers of increased and decreased H3K27ac regions, the median log2 fold change in gene expression was significantly larger when the increased H3K27ac was superior in number in the cis-regulatory domain (p = 2.6 × 10−8; Fig. 2d), whereas it was significantly smaller when the number of increased H3K27ac regions was less than the decreased ones (p < 2.2 × 10−16; Fig. 2d).
De novo motif analysis of HFD-responsive H3K27ac regions revealed enrichment of novel DNA-binding sites
De novo motif discovery by HOMER in HFD-responsive H3K27ac regions
NRF1 and GABPA are involved in local epigenomic regulation in INS-1 cells of the palmitate-induced expression of Acaa2 and Slc25a20 for β-oxidation
KEGG pathway enrichment of differentially expressed genes that were accompanied by H3K27ac change in the same direction
Fatty acid metabolism
Acaa2, Pecr, Hadh, Acat1, Tecr, Cpt1a
Biosynthesis of unsaturated fatty acids
Scd2, Fads2, Elovl6
For epigenetic analysis, we measured H3K27ac enrichment using qPCR primer sets for cis-regulatory elements of interest (Fig. 4, ESM Fig. 7 and ESM Table 3). Within the regions corresponding to the islet-merged H3K27ac regions spanning the TSS, H3K27ac enrichment in Acaa2 and Slc25a20 was lower than the control by 37.6% (p < 0.05; Fig. 5d and ESM Fig. 8b) and 28.8% (p < 0.05; Fig. 5h and ESM Fig. 8d) upon knockdown of NRF1 and GABPA, respectively. As observed in quantitative RT-PCR analysis of Acaa2 or Slc25a20 transcripts (Fig. 5b,f), ChIP-qPCR showed that the increase in H3K27ac in response to palmitate was blunted by knockdown of NRF1 and GABPA, respectively (Fig. 5d,h and ESM Fig. 8b,d), although the fold change in H3K27ac enrichment before and after palmitate treatment was similar regardless of whether NRF1 or GABPA was suppressed by siRNA. H3K27ac of Acaa2 increased in scrambled-siRNA-treated cells by 2.43-fold (p < 0.01), whereas it increased in Nrf1-siRNA-treated cells by 2.57-fold (p < 0.01) (Fig. 5d and ESM Fig. 8b). Although palmitate increased H3K27ac in Slc25a20 by 1.46-fold in control cells (p < 0.05; Fig. 5h and ESM Fig. 8d), a significant increase was not observed when GABPA was suppressed (Fig. 5h and ESM Fig. 8d).
MEF2A regulated fatty acid β-oxidation genes, Acaa2 and Slc25a20, similarly to NRF1 and GABPA
Further, we tested whether MEF2A regulated Acaa2 and Slc25a20. Rather than a localised peak, the TSS regions of Acaa2 and Slc25a20 genes revealed modest and relatively broad signals in previously deposited MEF2A ChIP-Seq data (ESM Fig. 7). Treatment of INS-1 cells with Mef2a siRNA resulted in a reduction in mRNA level of 18.8% (p < 0.01; Fig. 5i). The downregulation of Acaa2 and Slc25a20 was not significant following MEF2A knockdown (Fig. 5j,k). However, MEF2A knockdown resulted in 26.7% (p < 0.05) and 30.1% (p < 0.01) decrease in expression of Acaa2 and Slc25a20, respectively, in the presence of palmitate treatment (Fig. 5j,k). Taken together, these results suggest the important roles of the transcription factor motifs and their binding transcription factors in the regulation of genes and epigenome for fatty acid β-oxidation.
Knockdown of binding transcription factors for enriched motifs resulted in increased basal insulin secretion in INS-1 cells
In the present study, we show that the histone modification H3K27ac characteristic of active cis-regulatory regions changes purely through environmental effects by comparing samples derived from genetically identical C57BL/6J mice fed an HFD. Despite heterogeneity in the cellular composition, the global gene expression profile in pancreatic islets can be largely explained by that in beta cells , thus facilitating the analysis of whole islets and further examination of molecular mechanisms using INS-1 cells.
Previous large-scale analyses have demonstrated the correlation between changes in H3K27ac and gene expression . Although transcription factors serve as direct transcriptional activators , they may also be involved in inducing epigenetic alterations in specific regions of the genome . Consistent with this, our analyses revealed that environmental changes resulted in genome-wide epigenetic alterations, which correlated with differential gene expression. Furthermore, by de novo motif analysis, we identified enrichment of sequence motifs in H3K27ac regions showing changes in response to HFD, among which transcription factors such as NRF1, GABPA and MEF2A were the most enriched binding motifs. This led to the identification of potential target genes, Acaa2 and Slc25a20, which are important for fatty acid β-oxidation. Our in vitro analysis using INS-1 cells produced results consistent with such in vivo findings. Suppression of NRF1, GABPA or MEF2A by siRNA resulted in blunted responses of both H3K27ac gain surrounding the TSS and gene induction to palmitate treatment, especially in the case of genes involved in fatty acid metabolism. Therefore, the environmental factors, including fat-rich diet, shaped epigenetically induced specific pathways that were generated by the contribution of specific transcription factors.
NRF1, GABPA and MEF2A are transcription factors that have been linked to mitochondrial regulation ; however, their molecular function in pancreatic beta cells has not been well characterised. The knockdown experiments using INS-1 cells showed that changes in the secretory profile occurred in response to different testing solutions. This was in marked contrast to the reduced in vitro glucose responsiveness of pancreatic islets derived from 54-week-old hyperinsulinaemic DIO mice with mild glucose intolerance (ESM Fig. 9), in which the gene expression profile was similar to that of 37-week-old DIO mice (ESM Fig. 10). The epigenomic state should reflect the increased transcription factor activity (Table 1); thus, it could be possible that changing NRF1 activity could be useful in modulating pancreatic beta cell function. Besides this, the importance of the NRF1 gene was shown by a recent GWAS demonstrating the association of an NRF1 locus to the childhood obesity-related trait .
We also identified enrichment of a binding motif for MAFK in decreased H3K27ac regions. MAFK has been reported to be a negative regulator of pancreatic beta cell function and inhibition of endogenous small-MAFs improved glucose tolerance in DIO mice . Therefore, our findings of decreased H3K27ac regions could be responsible, at least in part, for mild glucose intolerance in DIO mice.
Our study supports the potential use of epigenetic analyses, such as H3K27ac, in a wide variety of applications for the research of human complex diseases. In this respect, the application of transcription factor-mediated repression of the epigenome has potential as an effective intervention against obesity and type 2 diabetes in the future.
We thank C. B. Wollheim (Lund University, Lund, Sweden; University of Geneva, Geneva, Switzerland) and N. Sekine (University of Geneva) for providing INS-1 cells. We appreciate the assistance given by D. Suzuki, K. Nagase, N. Ishibashi (Lab Managers), H. Shiina and T. Shibuya (Administrative Assistants) (Department of Metabolic Disorder, National Center for Global Health and Medicine). We would like to thank Editage (www.editage.jp) for English language editing.
TN and KY conceived this study. TN and HU performed the experiments. TN performed the computational analyses. TN and KY wrote the manuscript. TN, HU, NF, MK, TU, MH, WN and KY analysed the data, interpreted the results and contributed to discussions. The manuscript was critically reviewed, revised and given final approval by all co-authors. TN and KY are the guarantors of this work.
This work was supported by Japan Society for the Promotion of Science (JSPS) Grants-in-Aid for Scientific Research (KAKENHI), a grant from the National Center for Global Health and Medicine, the Japan Diabetes Foundation (to TN) and JSPS KAKENHI and a grant from the National Center for Global Health and Medicine (to KY). The study sponsors were not involved in the design of the study; the collection, analysis, and interpretation of data; writing the report; or the decision to submit the report for publication.
Duality of interest
The authors declare that there is no duality of interest associated with this manuscript.
- 1.Caspard H, Jabbour S, Hammar N, Fenici P, Sheehan JJ, Kosiborod M (2018) Recent trends in the prevalence of type 2 diabetes and the association with abdominal obesity lead to growing health disparities in the USA: an analysis of the NHANES surveys from 1999 to 2014. Diabetes Obes Metab 20(3):667–671. https://doi.org/10.1111/dom.13143 CrossRefPubMedGoogle Scholar
- 16.Malmgren S, Spégel P, Danielsson AP et al (2013) Coordinate changes in histone modifications, mRNA levels, and metabolite profiles in clonal INS-1 832/13 β-cells accompany functional adaptations to lipotoxicity. J Biol Chem 288(17):11973–11987. https://doi.org/10.1074/jbc.M112.422527 CrossRefGoogle Scholar
- 17.Siersbæk M, Varticovski L, Yang S et al (2017) High fat diet-induced changes of mouse hepatic transcription and enhancer activity can be reversed by subsequent weight loss. Sci Rep 7(1):40220. https://doi.org/10.1038/srep40220