# paper.pdf ## Metadata - PDFFormatVersion=1.4 - Language=en-US - IsLinearized=false - IsAcroFormPresent=false - IsXFAPresent=false - IsCollectionPresent=false - IsSignaturesPresent=false - Author=Elisabetta Manduchi - CreationDate=D:20260427170157+05'30' - Creator=Springer - Custom.CrossMarkDomains[1]=springer.com - Custom.CrossMarkDomains[2]=springerlink.com - Custom.doi=10.1038/s42255-026-01495-y - ModDate=D:20260427170313+05'30' - Subject=Nature Metabolism, doi:10.1038/s42255-026-01495-y - Title=Epigenetic adaptation of beta cells across lifespan and disease - dc:format=application/pdf - dc:identifier=doi:10.1038/s42255-026-01495-y - dc:publisher=Springer US - dc:description=Nature Metabolism, doi:10.1038/s42255-026-01495-y - dc:title=Epigenetic adaptation of beta cells across lifespan and disease - dc:creator=Elisabetta Manduchi,Hélène C. Descamps,Jinping Liu,Jonathan Schug,Tong Da,Deeksha Lahori,Hilana El-Mekkoussi,Michelle Lee,Eseye Feleke,Diana Bernstein,Chengyang Liu,Ali Naji,Benjamin Glaser,Klaus H. Kaestner,Dana Avrahami - prism:doi=10.1038/s42255-026-01495-y - prism:url=http://dx.doi.org/10.1038/s42255-026-01495-y - prism:aggregationtype=journal - prism:publicationname=Nature Metabolism - prism:copyright=© 2026, The Author(s) - prism:issn=2522-5812 - pdfx:doi=10.1038/s42255-026-01495-y - pdfx:crossmarkdomains=springer.comspringerlink.com - crossmark:majorversiondate=2010-04-23 - crossmark:crossmarkdomainexclusive=True - crossmark:doi=10.1038/s42255-026-01495-y - crossmark:crossmarkdomains=springer.comspringerlink.com - xmp:creatortool=Springer - xmp:modifydate=2026-04-27T17:03:13+05:30 - xmp:createdate=2026-04-27T17:01:57+05:30 - xmp:metadatadate=2026-04-27T17:03:13+05:30 - xmprights:marked=True - jav:journal_article_version=VoR - xmpmm:documentid=uuid:0a7deda7-8112-4177-a6a6-a88526f2fa67 - xmpmm:instanceid=uuid:9b4d688f-f320-4e7b-9aff-450a5c35d83b - xmpmm:renditionclass=default - xmpmm:versionid=1 - xmpmm:history=converteduuid:ab685219-d6da-4c2d-bfea-978de3915ce1converted to PDF/A-2bpdfToolbox2026-04-27T17:03:11+05:30 - pdfaid:part=2 - pdfaid:conformance=B - sn:authorinfo=Elisabetta Manduchihttp://orcid.org/0000-0002-4110-3714Hélène C. Descampshttp://orcid.org/0000-0001-5591-5839Chengyang Liuhttp://orcid.org/0000-0001-9812-1676Benjamin Glaserhttp://orcid.org/0000-0003-4711-5000Klaus H. Kaestnerhttp://orcid.org/0000-0002-1228-021XDana Avrahamihttp://orcid.org/0000-0002-2910-2452 - pdfaextension:schemas=http://ns.adobe.com/xap/1.0/mm/xmpMMXMP Media Management SchemainternalUUID based identifier for specific incarnation of a documentInstanceIDURIinternalThe common identifier for all versions and renditions of a document.OriginalDocumentIDURIinternalA reference to the original document from which this one is derived. It is a minimal reference; missing components can be assumed to be unchanged. For example, a new version might only need to specify the instance ID and version number of the previous version, or a rendition might only need to specify the instance ID and rendition class of the original.DerivedFromResourceRefinternalAn unordered array of references to resources that were incorporated, by inclusion or reference, into this document.IngredientsBag ResourceRefIdentifies a portion of a document. This can be a position at which the document has been changed since the most recent event history (stEvt:changed). For a resource within an xmpMM:Ingredients list, the ResourceRef uses this type to identify both the portion of the containing document that refers to the resource, and the portion of the referenced resource that is referenced.http://ns.adobe.com/xap/1.0/sType/Part#stPartParthttp://ns.adobe.com/pdf/1.3/pdfAdobe PDF SchemainternalA name object indicating whether the document has been modified to include trapping informationTrappedTexthttp://ns.adobe.com/xap/1.0/t/pg/xmpTPgXMP Paged-TextinternalXMP08 Spec: An ordered array of plate names that are needed to print the document (including any in contained documents).PlateNamesSeq TextinternalA structure containing the characteristics of a font used in a documentFontsBag FontA structure containing the characteristics of a font used in a document.http://ns.adobe.com/xap/1.0/sType/Font#stFntFonthttp://ns.adobe.com/pdfx/1.3/pdfxpdfxinternalID of PDF/X standardGTS_PDFXVersionTextinternalConformance level of PDF/X standardGTS_PDFXConformanceTextinternalCompany creating the PDFCompanyTextinternalDate when document was last modifiedSourceModifiedTextexternalMirrors crossmark:CrosMarkDomainsCrossMarkDomainsSeq TextinternalMirrors crossmark:DOIdoiTexthttp://www.aiim.org/pdfa/ns/id/pdfaidPDF/A ID SchemainternalPart of PDF/A standardpartIntegerinternalAmendment of PDF/A standardamdTextinternalConformance level of PDF/A standardconformanceTexthttp://crossref.org/crossmark/1.0/crossmarkcrossmarkinternalCrossMarkDomainsCrossMarkDomainsSeq TextinternalCrossmarkDomainExclusiveCrossmarkDomainExclusiveTextinternalUsual same as prism:doiDOITextexternalThe date when a publication was published.MajorVersionDateTexthttp://prismstandard.org/namespaces/basic/2.0/prismPrismexternalThe aggregation type specifies the unit of aggregation for a content collection. Comment PRISM recommends that the PRISM Aggregation Type Controlled Vocabulary be used to provide values for this element. Note: PRISM recommends against the use of the #other value currently allowed in this controlled vocabulary. In lieu of using #other please reach out to the PRISM group at info@prismstandard.org to request addition of your term to the Aggregation Type Controlled Vocabulary. aggregationTypeTextexternalCopyright copyrightTextexternalThe Digital Object Identifier for the article. The DOI may also be used as the dc:identifier. If used as a dc:identifier, the URI form should be captured, and the bare identifier should also be captured using prism:doi. If an alternate unique identifier is used as the required dc:identifier, then the DOI should be specified as a bare identifier within prism:doi only. If the URL associated with a DOI is to be specified, then prism:url may be used in conjunction with prism:doi in order to provide the service endpoint (i.e. the URL). doiTextexternalISSN for an electronic version of the issue in which the resource occurs. Permits publishers to include a second ISSN, identifying an electronic version of the issue in which the resource occurs (therefore e(lectronic)Issn. If used, prism:eIssn MUST contain the ISSN of the electronic version. See prism:issn. issnTextexternalTitle of the magazine, or other publication, in which a resource was/will be published. Typically this will be used to provide the name of the magazine an article appeared in as metadata for the article, along with information such as the article title, the publisher, volume, number, and cover date. Note: Publication name can be used to differentiate between a print magazine and the online version if the names are different such as “magazine” and “magazine.com.” publicationNameTextexternalThis element provides the url for an article or unit of content. The attribute platform is optionally allowed for situations in which multiple URLs must be specified. PRISM recommends that a subset of the PCV platform values, namely “mobile” and “web”, be used in conjunction with this element. NOTE: PRISM recommends against the use of the #other value allowed in the PRISM Platform controlled vocabulary. In lieu of using #other please reach out to the PRISM group at prism-wg@yahoogroups.com to request addition of your term to the Platform Controlled Vocabulary. urlTexthttp://springernature.com/ns/xmpExtensions/2.0/snSpringer Nature ORCID SchemaexternalAuthor information: contains the name of each author and his/her ORCID (ORCiD: Open Researcher and Contributor ID). An ORCID is a persistent identifier (a non-proprietary alphanumeric code) to uniquely identify scientific and other academic authors.authorInfoBag AuthorInformationSpecifies the types of author information: name and ORCID of an author.http://springernature.com/ns/xmpExtensions/2.0/authorInfo/authorAuthorInformationGives the name of an author.nameTextGives the ORCID of an author.orcidURIhttp://www.niso.org/schemas/jav/1.0/javNISOexternalValues for Journal Article Version are one of the following: AO = Author’s Original SMUR = Submitted Manuscript Under Review AM = Accepted Manuscript P = Proof VoR = Version of Record CVoR = Corrected Version of Record EVoR = Enhanced Version of Recordjournal_article_versionClosed Choice of Text ## Contents ### Page 1 Nature Metabolism | Volume 8 | April 2026 | 941–956 941nature metabolismhttps://doi.org/10.1038/s42255-026-01495-yResourceEpigenetic adaptation of beta cells across lifespan and diseaseElisabetta Manduchi 1,2,3 , Hélène C. Descamps 1,2,3 , Jinping Liu1,2,3 , Jonathan Schug1,2,3 , Tong Da1,2,3 , Deeksha Lahori1,2,3 , Hilana El-Mekkoussi1,2,3 , Michelle Lee1,2,3 , Eseye Feleke4,5 , Diana Bernstein1,2 , Chengyang Liu 3,6 , Ali Naji3,6 , Benjamin Glaser 5 , Klaus H. Kaestner 1,2,3 & Dana Avrahami 4,5Although the prevalence of type 2 diabetes (T2D) increases with age, most adults maintain normoglycaemia despite rising insulin resistance owing to the adaptive capacity of pancreatic beta cells to meet increased metabolic demand. However, persistent insulin resistance can lead to beta cell dysfunction and T2D onset. Here we show the mapping of genome-wide DNA methylation (DNAm) patterns and the epigenomic basis of beta cell adaptations by leveraging cell-type-specific methylome data from the Human Pancreas Analysis Program. In healthy donors, we identify progressive age-related demethylation enriched in cis-regulatory elements at beta cell identity and function genes. By contrast, alpha cells show the opposite trajectory, with subtle, age -rel ated hyp ermethylation. In T2D beta cells, but not alpha cells, we observed further demethylation compared to healthy controls, underscoring a unique capacity of beta cells to respond to changes in metabolic demand. Together, our findings suggest that DNAm remodelling in healthy beta cells reflects a long-term adaptation to metabolic demand, which, in T2D, is accelerated as part of a compensatory response that ultimately fails under sustained insulin resistance. Ageing is commonly described as the gradual decline in physiologi- cal functions and metabolic regulation, a process observed across the animal kingdom. However, although physiological deterioration predominantly occurs in the later stages of life, recent studies suggest that the ageing process can also include phases of physiological adap- tation and even functional improvement. During human adulthood, many physiological functions remain stable for decades, reflecting ongoing adaptation. For instance, older adults maintain the ability to improve cardiovascular fitness and cardiac output in response to aerobic exercise 1 , enhance endothelial function to reduce arterial stiff- ness2 and sustain daily energy expenditure 3 . Pancreatic beta cells have a crucial role in regulating glucose homoeostasis by secreting insulin in response to elevated blood glu- cose levels. Failure to adequately compensate for insulin resistance can lead to persistent hyperglycaemia and, eventually, T2D4 . Although insulin resistance and elevated blood glucose levels tend to worsen with age 4–6 , most older individuals maintain normoglycaemia. This suggests that beta cells, which are long-lived and rarely proliferate in adulthood 7–10 , must sustain function over decades by continually adapting to fluctuating metabolic demands. Indeed, beta cells can adapt during physiological insulin-resistant states, such as pregnancy and puberty, by expanding their mass and enhancing their capacity toReceived: 1 July 2025 Accepted: 2 March 2026Published online: 24 April 2026Check for updates1 Institute of Diabetes, Obesity, and Metabolism, Perelman School of Medicine, The University of Pennsylvania, Philadelphia, PA, USA. 2 Department of Genetics and Institute for Diabetes, Obesity, and Metabolism, University of Pennsylvania Perelman School of Medicine, Philadelphia, PA, USA. 3 The Human Pancreas Analysis Program (RRID:SCR_016202), Philadelphia, PA, USA. 4 Department of Developmental Biology and Cancer Research, The Hebrew University Hadassah Medical School, Jerusalem, Israel. 5 Department of Endocrinology and Metabolism, Hadassah Medical Center, Faculty of Medicine, Hebrew University of Jerusalem, Jerusalem, Israel. 6 Department of Surgery and Institute for Diabetes, Obesity, and Metabolism, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA. e-mail: kaestner@upenn.edu; dana.tzfati@mail.huji.ac.il ### Page 2 Nature Metabolism | Volume 8 | April 2026 | 941–956 942Resource https://doi.org/10.1038/s42255-026-01495-yidentity differences, quality control clustering of all 154 HPAP methyl- ome samples shows clear separation between alpha, beta and exocrine cells regardless of sex, ancestry, age, diabetes status and assay type, with the outliers in each branch corresponding to samples with a low purity score (Extended Data Fig. 1). To characterize the cell-type-specific methylation landscapes of pancreatic alpha and beta cells in our cohort, we compared methylomes from 14 CTL donors with high sample purity in both cell types (average (±s.d.) purity scores of 86 ± 6% and 91 ± 7% for alpha and beta cells, respectively; see Supplementary Table 1). The donors in this compari- son represented an age range from 26 to 58 years. Principal component analysis (PCA) of methylation levels revealed that the cell identity of the sample accounted for 44% of the variance (Fig. 1a). Comparison of methylation profiles between alpha and beta cells identified 78,389 differentially methylated regions (DMRs) with a false discovery rate (FDR) of <0.01 and |delta| > 10% (Fig. 1b), whose distribution of average DNAm differences is shown in Fig. 1c. We next focused on the 7,522 DMRs with the strongest effect sizes (|delta| > 40%) and analysed the signals from published islet chro- matin immunoprecipitation sequencing (ChIP-seq) and alpha and beta single-nucleus assay for transposase-accessible chromatin with sequencing (snATAC-seq) experiments across 10 kb regions centred at these DMRs (see Methods). This analysis revealed that regions hypomethylated in both beta and alpha cells were enriched for active chromatin marks and depleted of repressive marks (Fig. 1d). Consist- ently, hypomethylated regions in beta and alpha cells overlapped with open chromatin in each respective cell type (Fig. 1e). Finally, analysis of published TF binding data for whole islets32 revealed that binding sites for FOXA2, MAFB and NKX2.2, which are expressed in both alpha and beta cells, are mostly hypomethylated in both cell types, whereas binding sites for NKX6.1 and PDX1, which are specific to beta cells, are predominantly hypomethylated in beta cells (Fig. 1e). These findings further demonstrate that alpha-cell-specific and beta-cell-specific hypomethylated regions are enriched in cell-type-specific regulatory elements, extending prior observations33 . To test which gene sets are associated with hypomethylation in alpha and beta cells, we applied a hypergeometric test using DMRs selected with thresholds of FDR < 0.01 and |delta| > 40%, considering genes annotated as DMR-distal, as described in Methods. This analysis showed that although common biological pathways important for both cell types, such as regulation of secretion and exocytosis, are associated with hypomethylation in both alpha and beta cells, other gene sets are specifically associated with hypomethylation in either cell type (Fig. 1f). For example, genes critical for beta cell function, such as ADCYAP1, IAPP, INS, ABCC8 and SYT7 (‘BETA_GENES’ in Fig. 1f and Supplementary Table 9), as well as beta-cell-specific TFs includingPDX1, NKX6.1, MNX1, PAX6 and SIX3 (‘GO_ENDOCRINE_SYSTEM_DEVEL- OPMENT’ in Fig. 1f and Supplementary Table 9) are associated with hypomethylation in beta cells. By contrast, hypomethylation in alpha cells is associated with alpha cell function genes such as GCG, DPP4,FAP, FEV and ADRB1 (Beta-1 Adrenergic Receptor), the latter known to stimulate glucagon secretion under adrenergic stress 34 (‘ALPHA_GENES’ in Fig. 1f and Supplementary Table 10).Age-progressive DNA methylation loss in beta cell regulatory regionsBeta cells are long-lived cells that must continually adapt to fluctuating metabolic demands35 . These ongoing adaptations may be supported by the accumulation of epigenetic changes over time. Unlike gene expression, which is highly variable across individuals and becomes increasingly heterogeneous with age 36 , DNAm provides a more stable epigenetic signature that reflects cellular identity and differentiation state. Although age-related DNAm changes have been documented in blood and brain tissues37,38 , they have not been studied in human beta cells. To address this knowledge gap, we analysed DNAm profiles from secrete insulin11,12 . Our prior transcriptomic analysis of human beta cell ontogeny supports the notion that beta cells activate genes critical to their function to sustain enhanced activity with age13 . This finding is further corroborated by a recent report14 that highlights a beneficial adaptive response in beta cells throughout adulthood, associated with enhanced glucose-stimulated insulin secretion (GSIS), that persists at least until age 60 (ref. 14). DNAm, a relatively stable epigenetic mark, exhibits cell-specific patterns in regulatory regions, helping to maintain cell-type-specific expression programmes across multiple cell generations and through- out life 15 . Despite its stability, DNAm patterns are altered gradually over the human lifespan, with changes at specific sites serving as an epigenetic clock and providing one of the most accurate predictors of chronological age16 . Although numerous studies have shown DNAm alterations in the context of cancer 17 , far fewer have explored how factors related to metabolic diseases impact DNAm and, more specifi- cally, how these methylation changes may contribute to the develop- ment of T2D. Previous studies have reported DNAm alterations in the promoters and enhancers of genes associated with insulin secretion in whole islets of deceased organ donors with T2D18–20 . However, as the composition of human islets varies dramatically both within and between individuals10 , with alterations in the alpha-to-beta cell ratio in patients with T2D owing to reduced beta cell mass 21,22 or possibly alpha cell expansion 23 , it is essential to analyse T2D methylomes in a cell-type-specific manner. This is further underscored by the distinct global methylation patterns between alpha and beta cells24,25 and by the unique DNAm alterations observed in beta cells across the human lifespan and in T2D, as described below. In this work, we leveraged the islet molecular phenotyping data generated by the Human Pancreas Analysis Program (HPAP; https:// hpap.pmacs.upenn.edu) 26,27 . To investigate epigenetic adaptation in beta cells across the human lifespan, we analysed genome-wide DNAm patterns in sorted beta and alpha cell populations from control (CTL) donors of various ages (here and throughout, ‘control’ refers to non-diabetic, autoantibody-negative individuals). This analysis revealed age-associated demethylation in beta cells, particularly in regulatory regions near islet transcription factor (TF) binding sites. Although these regions exist in a hypomethylated state in beta cells, the ongoing process of maintaining that state may require active demeth- ylation, which counteracts maintenance methylation28,29 . This process may help maintain enhancer activity and sustain stable expression of key beta cell function genes to preserve cell identity throughout the long lifespan of beta cells. By contrast, alpha cells exhibited a subtler and opposite age-related trend, with a predominance of hypermethyla - tion even in active regulatory regions. Interestingly, we noted a beta cell-specific additional demeth- ylation process in T2D compared to CTL, suggestive of an intensified compensatory response. By contrast, alpha cells from diabetic donors exhibited minimal methylation changes compared to healthy donors. Our findings suggest that in healthy individuals, the beta cell epig- enome supports lifelong adaptation to metabolic demand. In T2D, how- ever, enhanced demethylation probably reflects mounting stress on beta cells as diabetes progresses, representing an intensified attempt to maintain beta cell function that ultimately becomes unsustainable.ResultsDNA methylation patterns distinguish alpha and beta cell regulatory landscapesDNAm has a crucial role in gene silencing, which is essential for maintaining cell type identity and cellular expression programmes throughout the lifespan and across cell divisions15 . This stable meth- ylation pattern is established by regulated processes of methylation and demethylation, and it is preserved by the activity of maintenance methyltransferase (DNMT1) as well as de novo methyltransferases such as DNMT3A30,31 . To confirm that our dataset captures robust cell ### Page 3 Nature Metabolism | Volume 8 | April 2026 | 941–956 943Resource https://doi.org/10.1038/s42255-026-01495-yb d e f–20 0 20 40 –50 –25 0 25 50 PC1 (44.3%) PC2 (5.8%) % methylation betaa–10 10 Number of DMRs per bin –30–50 30 50 % methylation difference (beta – alpha) H3K27ac H3K4me1 H3K27me3 snATAC α FOXA2 MAFB NKX2.2 NKX6.1 PDX1snATAC βcAlpha Beta BETA_ATAC ALPHA_GENES BETA_GENES GO_REGULATION_OF_SECRETION GO_ENDOCRINE_SYSTEM_DEVELOPMENT GO_CELL_CYCLE GO_EXOCYTOSIS GO_REGULATION_OF_HORMONE_LEVELS HALLMARK_APOPTOSIS GO_INFLAMMATORY_RESPONSE GO_STRESS_RESPONSE KEGG_TYPE_II_DIABETES_MELLITUS 0 5 10 –log10(P value)P = 0.05 15 20 Hypo in alpha Hypo in beta 25 H3K27ac islets 0 1 2 3 4 5 H3K4me1 0 1 2 3 4 H3K27me3 0 0.05 0.1 0.15 DMRs Alpha.hyper Beta.hyper snATAC alpha snATAC beta FOXA2 islets MAFB islets NKX2.2 islets NKX6.1 islets PDX1 islets 0 200 400 600 800 01,000 500 1,000 0 2 4 6 8 0 2 4 6 0 2 4 6 8 0 2 4 6 0 2 4 6 Alpha.hyper Beta.hyper DMRs0 80 60 40 20 0 % methylation alpha 0 20 40 60 80 100 0 5,000 10,000 15,000Fig. 1 | DNAm distinguishes alpha and beta cells in association with their cell-type-specific regulatory landscapes. a, Scatterplot of the first two principal components from PCA of DNAm CTL alpha and beta cells, showing separation by cell type along PC1. b, Methylation scatterplot. Smooth scatterplot of beta versus alpha cell methylation values (as %) in CTL donors for 78,389 DMRs with an FDR < 0.01 and |delta | >10%. c , Effect size distribution, showing the difference in average methylation between alpha and beta cells from the CTL donor DMRs shown in b. x axis, % difference (beta – alpha); y axis, number of DMRs per bin. d,e, Enrichment heatmaps. Enrichment plot of signal intensity for islet histone marks (ChIP-seq 32,67 ) (d) and chromatin accessibility (snATAC-Seq, HPAP) and TF binding sites (whole islet, ChIP-seq 32 ) (e) across top alpha-beta DMRs (FDR < 0.01, |delta| > 40%). Each row shows the signal over the 10 kb region centred at one of 7,522 DMRs. Top, regions hypomethylated in beta cells; bottom, regions hypomethylated in alpha cells. f, Gene set enrichment results for hypomethylated DMRs in alpha (green) and beta (blue) cells, based on DMR- distal genes (FDR < 0.01, |delta| > 40%, CTL alpha vs beta). The x axis displays −log10 of the unadjusted P values from a one-sided hypergeometric test. ### Page 4 Nature Metabolism | Volume 8 | April 2026 | 941–956 944Resource https://doi.org/10.1038/s42255-026-01495-ypancreatic beta cells collected from 21 HPAP CTL donors across a broad age spectrum from 8 to 64 years old. UXM-estimated purity scores had a mean (±s.d.) of 91 ± 6%. Moreover, based on the UXM-estimated cell type composition (see Methods), there was no strong associa- tion between age and the proportion of cells of any given type (maxi- mum R2 = 0.1), including monocytes and macrophages (R2 = 0.048). Supplementary Table 2 provides the characteristics of the samples used in this comparison. PCA of methylation levels indicated that PC2 was significantly correlated with age (R2 = 0.62, P < 0.001) (Fig. 2a). In this comparison we analysed age as a numeric covariate, and to meas- ure effect size, we defined delta as the difference in average methyla- tion between the group of older donors aged 39–58 years and that of younger donors aged 18–35 years (see Methods). This comparison identified 8,748 DMRs at the thresholds of FDR < 0.01, |delta| > 5%, with a striking 96% (8,394 DMRs) showing hypomethylation with age (Fig. 2b). Although the magnitude of methylation changes was generally modest, with most DMRs showing |delta| < 30% (Fig. 2c), the differences were consistent and statistically significant. Therefore, despite the small effect size, ageing in beta cells is associated with a robust and widespread hypomethylation signature. Next, we analysed the signals from islet ChIP-seq and beta snATAC-seq experiments across 10 kb regions centred at these age-related beta cell DMRs (see Methods) (Fig. 2d,e). Our analysis revealed that regions hypomethylated with age are predominantly located in beta-cell-specific active regulatory chromatin domains, as evidenced by significant enrichment for H3K27ac, H3K4me1, beta-ATAC regions and binding sites for islet and beta-cell-specific TFs such asNKX6.1 and PDX1. To validate these findings, we assessed the average methylation levels within previously identified islet active enhancer regions32 , which revealed subtle progressive demethylation with age (Fig. 3a). The distribution of slopes from linear fits of methylation versus age in these enhancer regions, representing the rate of change–20 –10 0 10 20 30 20 40 60 Age 20 40 60 PC2 (8.5%) Age (years)a b c d e% methylation younger % methylation older % methylation difference (older – younger) Number of DMRs per bin Islet H3K27Ac Islet H3K4me1 Islet H3K27me3 snATAC beta FOXA2 Islets MAFB Islets NKX2.2 Islets NKX6.1 Islets PDX1 Islets 100 80 60 40 20 0 20 40 60 80 2,000 1,000 0 –60–50 –40 –30 –20 –10 10 20 30 3,000 H3K27ac islets H3K4me1 H3K27me3 DMRs Hyper.young Hypo.young DMRs Hyper.young Hypo.young snATAC beta FOXA2 islets MAFB islets NKX2.2 islets NKX6.1 islets PDX1 islets 00 1 2 3 4 5 2 4 60 1 2 3 4 5 0 1 2 3 4 0 0.05 0.1 0.15 0 500 1,000 0 2 4 6 8 0 2 4 6 0 2 4 6 8Fig. 2 | Progressive loss of DNAm at regulatory regions in beta cells with age (part I). a, PCA of DNAm in CTL beta cells. Scatterplot of PC2 versus age, illustrating a strong correlation (R2 = 0.62, two-sided Pearson correlationP = 2.3 × 10−5 ) between the two variables. b, Methylation scatterplot. Smooth scatterplot of beta cell methylation values (as %) in older (39–58 years old) versus younger (18–35 years old) CTL donors across 8,748 age-associated DMRs (FDR < 0.01 and |delta | >5%), illustrating age-related demethylation. c , Effect size distribution. Distribution of average methylation differences between older and younger donors for the DMRs shown in b. x axis, % difference (older – younger);y axis, number of DMRs per bin. d,e, Enrichment heatmaps. Enrichment plot of signals from islet histone marks (d) and chromatin accessibility (snATAC-Seq, beta cells) and TF binding (ChIP-seq, whole islets) (e) across 8,748 age-associated DMRs (FDR < 0.01 and |delta | >5%) in beta cells from CTL donors. Each row shows the signal over the 10 kb region centred at one DMR. Top, regions hypomethylated in older donors; bottom, hypomethylated in younger donors. ### Page 5 Nature Metabolism | Volume 8 | April 2026 | 941–956 945Resource https://doi.org/10.1038/s42255-026-01495-yof DNAm per year, is clearly shifted towards negative values (Fig. 3b). Although the absolute values of the slopes are modest, they reflect a consistent directional trend, with 82% of the 28,813 fits exhibiting nega- tive slopes and relatively strong negative associations, as indicated by the corresponding Pearson correlation coefficients (median Pearson correlation coefficient = −0.35; Fig. 3c) and P values (median P = 0.06; Fig. 3d). Notably, enhancer regions can be actively demethylated by TET enzymes, which also help to protect against age-related functional decline in other tissues such as the brain28,39 . To explore the mechanisms driving age-associated demethylation in active enhancers, we examined methylation levels at TF binding sites of NKX6.1 and PDX1, which are enriched in beta cell enhancers40 , focus- ing on those showing the greatest variation across age. We observed progressive demethylation with increasing donor age at these sites (Fig. 3e,i), and the distribution of slopes from linear fits of methylation versus age was clearly shifted toward negative values, shown for PDX1in Fig. 3f and for NKX6.1 in Fig. 3j. For regions with negative slopes, the median Pearson correlation was –0.56 with a median P value of 0.004 for PDX1 (Fig. 3g,h), and –0.55 with a median P value of 0.005 for NKX6.1 (Fig. 3k,l). We next asked whether age-related DMRs in beta cells were enriched near genes associated with specific biological pathways or gene sets. Using DMRs filtered by FDR < 0.01 and |delta| >10% and considering DMR-distal genes (as defined in Methods), we found that hypomethylated regions in older donors were enriched in beta-cell-specific regulatory regions, such as beta cell ATAC peaks, and near genes involved in beta cell function, cell cycle regulation and, notably, DNA damage response (DDR) and T2D-associated genes, whereas the disallowed gene set showed no enrichment (Fig. 3m and Supplementary Table 11). To explore potential links to beta cell het- erogeneity, we assessed enrichment for subtype gene signatures41,42 and found that genes characteristic of the ST8SIA1-positive and βHI subpopulations were enriched among beta cell age-related hypo- methylated DMRs (Fig. 3m and Supplementary Table 11). However, this enrichment could be influenced by age-related changes in the proportion of these subpopulations, as our methylomes represent a bulk signal integrating inputs from diverse beta cell populations. Gene set enrichment analysis (GSEA) of HPAP single-cell RNA sequencing (scRNA-seq) data from donors in a similar age range revealed upregu- lation of the p53 pathway in older individuals, including increased expression of CDKN2A, CDKN1A and CDKN2B (Extended Data Fig. 2a and Supplementary Table 12), supporting progressive activation of stress response and senescence programmes in beta cells with age. In line with previous reports 14 , GSEA also showed age-associated downregulation of the HALLMARK beta cell gene set as a whole (Extended Data Fig. 2b and Supplementary Table 13), although most individual genes were not significantly downregulated (Extended Data Fig. 2c). The mild downregulation of this gene set is unlikely to compromise beta cell identity or function, as key functional pathways such as hormone secre- tion and oxidative phosphorylation remain unchanged with age, along with the expression of major beta cell identity genes such as PDX1 andNEUROD1 (Extended Data Fig. 2c). To further explore whether the observed demethylation with age was directly correlated with adaptation, we analysed the same 21 beta cell methylomes for correlation with body mass index (BMI) and with GSIS (measured by the ratio of glucose-stimulated [B] ver- sus basal [A] insulin secretion; for details see Methods). The R2 for BMI versus age in this dataset was 0.18. The correlations between the first five beta cell methylation-based PCs and BMI were much weaker (R2 ranging from 0.009 to 0.21) than the noted strong correlation between PC2 and age (R2 = 0.62). A BMI-based DMR analysis with the same thresholds as those used for age identified 599 regions, and only 47 of the 8,394 age-demethylated DMRs were among these. As for GSIS, the R2 for the B/A ratio versus age was negligible (0.0018). The correlations between the first five beta cell methylation-based PCs and the B/A ratio were much weaker (R2 ranging from 0 to 0.18) than the noted strong correlation between PC2 and age. A GSIS-based DMR analysis with the same thresholds as those used for age identified only 20 regions, none of which were among the 8,394 age-demethylated DMRs. These observations, together with heatmaps of the age-hypomethylated regions in healthy beta cells ordered by age, BMI and GSIS (Extended Data Fig. 3), indicate that BMI and GSIS are not the main drivers of the age-demethylated DMRs identified above.In contrast to beta cells, alpha cells show progressive DNA hypermethylation with ageFor alpha cells, we analysed 14 samples from CTL donors spanning an age range from 18 to 58 years. UXM-estimated purity scores had a mean (±s.d.) of 87 ± 5%. Based on the UXM-estimated cell type com- position (see Methods), there was no association between age and the proportion of cells of any given type (maximum R2 = 0.22), including monocytes and macrophages (R2 = 0.074). Supplementary Table 3 pro- vides the characteristics of the samples used in this comparison. PCA of methylation levels indicated that PC2 was significantly correlated with age (R2 = 0.77, P < 0.001) (Fig. 4a). We analysed the CTL alpha cell age comparison in the same manner as described above for beta cells using the same thresholds of FDR < 0.01 and |delta| > 5% and identified 1,466 DMRs, of which, surprisingly, 1,344 (92%) were hypermethyl- ated with age, although the magnitude of methylation changes was modest, with most DMRs showing |delta| < 20% (Fig. 4b–c). A subset of age-dependent hypermethylation occurred in active regulatory domains, but a higher proportion occurred in repressed regions that are not bound by TFs, as illustrated by the enrichment heatmaps of signals from the islet ChIP-seq and alpha snATAC-seq experimentsFig. 3 | Progressive loss of DNAm at regulatory regions in beta cells with age (part II). a, Enhancer methylation heatmap. Heatmap of average beta cell methylation at islet active enhancer regions 32 (rows) across 21 HPAP CTL donors (columns), ordered by age, illustrating progressive demethylation with age. Methylation values are shown as fractions between 0 and 1. b, Linear fit slopes for enhancers. Violin plot of the slopes of linear fits to methylation versus age, corresponding to the regions in a. Mean slope = −0.0012; median slope = −0.0008. c,d, Pearson correlations for enhancers. Boxplot of Pearson correlation coefficients (c) and corresponding P values (d) for the 23,705 regions with negative methylation–age slopes from a. Boxplot parameters: middle line, median; box upper (resp. lower) bound, 3 rd (resp. 1st ) quartile; upper/lower whisker extends to the largest/smallest value within 1.5× the interquartile range (IQR); individual points, outliers. e, PDX1 binding sites methylation heatmaps. Heatmap of average beta cell methylation at PDX1 binding sites 32 , showing the 500 peaks with the greatest variability across CTL donors, ordered by age. Note the demethylation trend with age. f, Linear fit slopes for PDX1 binding sites. Violin plot of the slopes of linear fits to methylation versus age, corresponding to the regions in e. g,h, Pearson correlations for PDX1 binding sites. Boxplot of Pearson correlation coefficients (g) and corresponding P values (h) for the 447 regions with negative methylation–age slopes from e. Boxplot parameters: middle line, median; box upper (resp. lower) bound, 3 rd (resp. 1st ) quartile; upper/lower whisker extends to the largest/smallest value within 1.5× IQR; individual points, outliers. i, NKX6.1 binding sites methylation heatmaps. Heatmap of average beta cell methylation at NKX6.1 binding sites 32 , showing the 500 peaks with the greatest variability across CTL donors, ordered by age. Note the demethylation trend with age. j, Linear fit slopes for NKX6.1 binding sites. Violin plot of the slopes of linear fits to methylation versus age, corresponding to the regions ini. k,l, Pearson correlations for NKX6.1 binding sites. Boxplot of Pearson correlation coefficients (k) and corresponding P values (l) for the 448 regions with negative methylation versus age slopes from i. Boxplot parameters: middle line, median; box upper (resp. lower) bound, 3 rd (resp. 1st ) quartile; upper/ lower whisker extends to the largest/smallest value within 1.5× IQR; individual points, outliers. m, Gene set enrichment results. Gene set enrichment for hypomethylated DMRs in beta cells, based on DMR-distal genes (FDR < 0.01 and |delta | >10%) from the CTL beta age comparison. The x axis displays −log 10 of the unadjusted P values from a one-sided hypergeometric test. ### Page 6 Nature Metabolism | Volume 8 | April 2026 | 941–956 946Resource https://doi.org/10.1038/s42255-026-01495-yover the 10 kb regions centred at these DMRs (Extended Data Fig. 4; see Methods for details). We next examined whether age-related DMRs in alpha cells were enriched near specific pathways or gene sets. Using an FDR < 0.01 and |delta| > 10% cutoff, we did not detect enrichment for specific pathways or gene sets in association with either hypomethylated or hypermethyl- ated regions with age in alpha cells. Analysis of pseudo-bulk alpha cell gene expression data across age did not reveal a correlation with gene silencing or reduced expression with age, possibly because of the low effect size and the higher degree of variability of transcriptomic versus CpG methylation data. To further explore the discrepancy between alpha and beta cells, we compared their methylation levels across ages at binding sites for FOXA2 and NKX2.2, two TFs expressed at similar levels in both cell types 32,43–46 . We used the same number of available samples (13) for each cell type (mostly common donors and, when not available, aged-matched donors) and, for each cell type separately, considered the 500 binding sites whose methylation varied most across ages; that is, focusing on sites most affected by age in that cell type. As illustrated in Fig. 4d,e and Extended Data Fig. 5, we observed subtle progressing demethylation of these regions with age in beta cells, whereas the trend in alpha cells was opposite. The distribution of slopes from linear fitsMeth. vs age slope <0 Pearson correlation 0 –0.25 –0.75 –0.50P value of Pearson correlation 0 0.1 0.2 0.3 0.4 0.5 Meth. vs age slope <0a b d e f g h i j0 0.005 –0.005 –0.100 Slope Methylation vs age Meth. vs age slope <0 PDX1 most var. peaks in beta cellsDNA methylation at islet enhancers –0.004 0.004 0 –0.1200 –1.00 Slope Methylation vs age Meth. vs age slope <0 0 –0.25 –0.50 –0.75 Pearson correlation 0 0.1 0.2 0.3 0.4 0.5ck lMeth. vs age slope <0 Slope Methylation vs ageP value of Pearson correlation P value of Pearson correlation 0 –0.005 0.005 –0.010 –0.50 0 0.1 0.2 0.3 0.4 0.5 Meth. vs age slope <0 Pearson correlation 0 –0.25 –0.75mNKX6.1 most var. peaks in beta cells60 10 0 0.2 0.4 0.6 0.8 1 Age Methylation 60 10 0 0.2 0.4 0.6 0.8 1 Age Methylation 60 10 0 0.2 0.4 0.6 0.8 1 Age Methylation BETA_GENES ST8S_POS_SUBTYPE BETA_HI_SUBTYPE GO_INSULIN_SECRETION KEGG_TYPE_II_DIABETES_MELLITUS GO_ENDOCRINE_SYSTEM_DEVELOPMENT GO_CELL_CYCLE GO_EXOCYTOSIS GO_REGULATION_OF_HORMONE_LEVELS GO_CELLULAR_RESPONSE_TO_DNA_DAMAGE_STIMULUS 0 1 2 3 4 –log10 (P value) 5 6 7 8 9 BETA_ATAC ### Page 7 Nature Metabolism | Volume 8 | April 2026 | 941–956 947Resource https://doi.org/10.1038/s42255-026-01495-y–30 –20 –10 0 10 20 20 30 40 50 60 Age (yr) Age 20304050 *** 0 0.25 0.50 0.75 Slope>0 Slope<0 Methylation vs age |Pearson correlation|P value of Pearson correlation *** –0.01 0 0.01 Methylation vs age Slope Cell Alpha Beta FOXA2 most var. peaks in alpha cells Age 50 20 0 0.2 0.4 0.6 0.8 1 0 20 40 60 80 100 0 80 60 40 20 100a b d e f g h cFOXA2 most var. peaks in beta cells Age Age 50 20 0 0.2 0.4 0.6 0.8 1 *** 0 0.1 0.2 0.3 0.4 0.5 Slope>0 Slope<0 Methylation vs age % methylation older PC2 (11 %) % methylation younger –30 –20 –10 10 20 30 40 600 400 200 0 Number of DMRs per bin % methylation difference (older – younger)Fig. 4 | Alpha cells show modest age-related DNA hypermethylation, opposite to the trend in beta cells. a, PCA of DNAm in CTL alpha cells. Scatterplot of PC2 versus age, showing a strong correlation (R2 = 0.77, two-sided Pearson correlationP = 3.8 × 10−5 ). b, Methylation scatterplot. Smooth scatterplot of alpha cell methylation (%) in older (39–58 years old) versus younger (18–35 years old) CTL donors across 1,466 age-associated DMRs (FDR < 0.01, |delta| > 5%), illustrating subtle hypermethylation in older donors. c, Effect size distribution. Distribution of average methylation differences (older – younger) for the DMRs shown in b. xaxis, % difference (older – younger); y axis, number of DMRs per bin. d,e, FOXA2binding sites methylation heatmaps. Heatmaps of average CpG methylation at FOXA2 binding sites 32 in alpha (d) and beta (e) cells. Shown are the 500 most variable peaks (per cell type) across 13 age-matched donors. f, Linear fit slopes for FOXA2 binding sites. Violin plots of slopes from linear fits of methylation versus age at the regions shown in d and e, for alpha and beta cells. Asterisks indicate a two-sided Wilcoxon P < 0.001 (in this case, P < 2.22 × 10−16 ). g,h, Pearson correlations for FOXA2 binding sites. Boxplots of absolute Pearson correlation coefficients (g) and correlation P values (h) for methylation versus age at the 332 regions with slope >0 in alpha cells and the 410 regions with slope <0 in beta cells from f. Boxplot parameters: middle line, median; box upper (resp. lower) bound, 3 rd (resp. 1st ) quartile; upper/lower whisker extends to the largest/smallest value within 1.5× IQR; individual points, outliers. Asterisks indicate a two-sided Wilcoxon P < 0.001 (P = 7.1 × 10−10 in both cases). ### Page 8 Nature Metabolism | Volume 8 | April 2026 | 941–956 948Resource https://doi.org/10.1038/s42255-026-01495-y0 1 2 3 4 0 1 2 3 40 2 4 6 0 1 2 3 40 2 4 60 200 400 600 800 0 200 400 600 0 1 2 3 4 0 1 2 3 4 0 0.05 0.1 0.15 0.2e f a0 20 40 60 80 100 0 20 40 60 80 100b cNumber of DMRs per bin 6,000 4,000 2,000 0 –10 10 30% methylation control % methylation T2D –30 % methylation difference (T2D – control)dH3K27ac H3K4me1 H3K27me3 FOXA2 MAFB NKX2.2 NKX6.1 PDX1 snATACβ ctl snATACβ T2D H3K27ac islets H3K4me1 H3K27me3 DMRs snATAC beta CTL snATAC beta T2D FOXA2 islets PDX1 isletsNKX6.1 isletsNKX2.2 islets MAFB islets DMRs CTL.hyper T2D.hyper BETA_GENES BETA_ATAC BETA_HI_SUBTYPE GO_INSULIN_SECRETION FATTY_ACID_BETA_OXIDATION GO_REGULATION_OF_SECRETION –log 10(P value) 0 1 2 3 4 5CTL T2D 20 0 –20 PC1 (11.6%)CTL.hyper T2D.hyperFig. 5 | Beta cells exhibit accelerated DNA demethylation in T2D (part I).a, PCA of DNAm in beta cells from CTL and T2D donors. Boxplots of the PC1 stratified by condition for 12 CTL and seven T2D donors. Boxplot parameters: middle line, median; box upper (resp. lower) bound, 3 rd (resp. 1st ) quartile; upper/lower whisker extends to the largest/smallest value within 1.5× IQR; each donor is represented by a point. Asterisk indicates a two-sided WilcoxonP < 0.05 (P = 0.03 in this case). b, Methylation scatterplot. Smooth scatterplot of beta cell methylation (%) in T2D versus CTL donors across 9,296 DMRs (FDR < 0.05, |delta| > 5%), illustrating subtle hypomethylation in T2D. c , Effect size distribution. Distribution of average methylation differences (T2D – CTL) for the DMRs shown in b. x axis, % difference (T2D – CTL); y axis, number of DMRs per bin. d,e, Enrichment heatmaps. Enrichment plot of signals from islet histone marks (d) and chromatin accessibility (snATAC-Seq, beta cells) and TF binding (ChIP-seq, whole islets) (e) across 9,296 DMRs (FDR < 0.05, |delta | >5%) from the T2D versus CTL beta cell comparison. Each row shows the signal over the 10 kb region centred at one DMR. Top, regions hypomethylated in T2D; bottom, hypomethylated in CTL beta cells. f, Gene set enrichment results. Gene set enrichment for hypomethylated DMRs based on DMR-distal genes (FDR < 0.05, |delta| > 10%) from the T2D versus CTL beta cell comparison. The x axis displays −log 10 of the unadjusted P values from a one-sided hypergeometric test. ### Page 9 Nature Metabolism | Volume 8 | April 2026 | 941–956 949Resource https://doi.org/10.1038/s42255-026-01495-yof methylation versus age in these regions is shifted towards positive values in alpha cells and negative values in beta cells, with the shift more evident at FOXA2 binding sites than at NKX2.2 binding sites in alpha cells (Fig. 4f and Extended Data Fig. 6a). Moreover, the regions with negative slopes in beta cells correspond to significantly better linear fits than the regions with positive slopes in alpha cells (Fig. 4g,h and Extended Data Fig. 6b,c). Together, these findings demonstrate that despite being closely related cell types residing within the same functional unit, beta and alpha cells undergo opposing age-related DNA methyla- tion trajectories. Beta cells exhibit progressive hypomethylation at regulatory regions, particularly at TF binding sites, potentially reflecting their adaptive, metabolically responsive nature, whereas alpha cells do not show this demethylation process but instead display gradual hypermethylation, predominantly in non-active chromatin regions.T2D is associated with premature hypomethylation in beta cellsA recent study using medium-throughput DNAm and RNA-seq analy- sis of whole human islets reported DNAm changes associated with impaired insulin secretion and reduced expression of beta-cell-related genes in T2D donors 20 . However, given that islet cell composition varies between individuals10 , and T2D islets often exhibit altered alpha-to-beta cell ratios owing to beta cell loss21,22 or possibly alpha cell expansion23 , whole-islet analysis can obscure cell-type-specific epigenetic signals. This is especially relevant given the marked differences in DNAm profiles between alpha and beta cells and the opposing age-related methylation changes between the two cell types reported above. To capture beta-cell-specific changes in T2D, we analysed beta cell sam- ples from 12 CTL and seven donors with T2D, with similar ages rang- ing from 33 to 64 years. All samples had high UXM-estimated purity, with a mean (±s.d.) of 91 ± 7% for CTL and 89 ± 4% for T2D donors (see Supplementary Table 4 for characteristics). PCA of methylation levels revealed an association between PC1 and disease status (two-sided Wilcoxon P = 0.04; Fig. 5a). Using FDR < 0.05 and |delta| > 5% thresholds, we identified 9,296 DMRs, the majority of which (8,619; 93%) were hypomethylated in T2D, with most showing modest differences (|delta| < 20%) (Fig. 5b,c). About half of the hypomethylated regions in T2D were enriched for H3K4me1, marking primed enhancers, while a subset also carried H3K27ac, suggesting an active regulatory state. The remaining hypo- methylated regions were enriched for the repressive mark H3K27me3 (Fig. 5d). Some hypomethylated regions also showed enrichment for TF binding sites, particularly those bound by FOXA2, NKX2.2 andPDX1 (Fig. 5e). We then examined DMR-distal genes to DMRs with FDR < 0.05 and |delta| > 10% and found enrichment in T2D-associated demeth- ylation for genes involved in beta-cell-related pathways, mirroring the pattern observed in age-associated hypomethylation (Fig. 5f and Supplementary Table 14). In addition, these regions were enriched for the βHI beta cell subpopulation gene set41,42 , whose relative proportion increases in T2D, suggesting that this enrichment may partly reflect over-representation of this subtype. Notably, T2D hypomethylated DMRs showed no enrichment near disallowed genes 47 (Fig. 5f and Supplementary Tables 8 and 14), which were also not upregulated in our dataset. By contrast, no significant pathway enrichment was observed for hypermethylated regions, consistent with a less specific, stochastic process. To assess the potential functional consequences of these meth- ylation changes, we examined gene expression in beta cells from CTL (n = 13) and T2D (n = 15) donors of similar age using HPAP scRNA-seq data and found that T2D-associated hypomethylated DMRs did not show a direct correlation with upregulation of their associated genes. In fact, only 1.6% (about 20 genes) of the 1,212 genes associated with hypomethylated DMRs in T2D were differentially expressed (mostly upregulated), and none are known beta cell identity or function genes, suggesting a limited direct role of T2D-associated demethylation in regulating gene expression levels. Analysis of additional scRNA-seq datasets comparing T2D and CTL beta cells 48,49 likewise showed no enrichment of genes upregulated in T2D near hypomethylated DMRs. However, although core beta cell identity genes such as PDX1, NKX6.1,MAFA, ABCC8 and KCNJ11 maintain stable expression levels in T2D donors (Extended Data Fig. 7a), we observed reduced expression of broader functional pathways, including oxidative phosphoryla- tion, glycolysis and protein secretion (Supplementary Table 15 and Extended Data Fig. 7b). Genes from these pathways, which are largely not linked to hypomethylation, do not maintain stable expression (Extended Data Fig. 7c), supporting intact core identity program but compromised beta cell function in T2D. Next, we compared average beta cell methylation levels at PDX1binding sites using seven CTL and seven T2D donors matched for age, selecting the 500 most variable sites across age within each condition. This analysis revealed that although demethylation at these sites tends to progress with age in CTL donors (Fig. 6a), as was shown above using a larger number of samples (Fig. 3e), this trend is less evident in beta cells from donors with T2D (Fig. 6b). The distribution of slopes from linear fits of methylation versus age at these regions was signifi- cantly more negative in CTL donors than those with T2D (Fig. 6c). The Pearson correlation coefficients for regions with negative slopes were also significantly more negative in CTL donors (Fig. 6d), and their associated P values trended lower (Fig. 6e). Although this analysis is based on a relatively small number of age-matched donors and should be interpreted with caution, the consistent differences in slope direction and correlation values across matched CTL and T2D samples support a more robust age-related demethylation process in non-diabetic donors at these regions. The lack of a clear age-related demethylation trajectory in T2D donors probably reflects already reduced methylation levels in beta cells early in life, possibly driven by chronic stress during diabetes progression. To illustrate this observa- tion, we plotted the median methylation levels of the same PDX1 bind- ing sites in T2D and CTL donors matched for age. As shown in Fig. 6f, focusing on the peaks in the fifth quintile of the median methylation levels (~55–65%), T2D donors consistently exhibit lower methylation than age-matched controls across the age range. This pattern sup- ports the idea that demethylation at these sites occurs earlier in life in T2D beta cells. Similar trends were observed at FOXA2 and NKX2.2binding sites in the fifth quintile (Extended Data Fig. 8), further sup- porting this observation. We explored the possibility of a genetic link by examining the T2D-associated single nucleotide polymorphisms (SNPs) identified in previous work 50 . GREGOR enrichment analysis did not detect a particularly significant enrichment of these SNPs and their proxies in the beta cell hypomethylated DMRs in T2D from our results (P = 0.11). Finally, we performed an analysis of DNAm changes in alpha cells from nine T2D and 13 control donors matched for age (33–63 years) and with high UXM-estimated purity (mean ± s.d., 87 ± 5% for CTL; 86 ± 8% for T2D; Supplementary Table 5). Despite the larger number of alpha cell samples compared to the beta cell analysis, we identified signifi- cantly fewer DMRs (1,574) between T2D and CTL alpha cells using the same thresholds (FDR < 0.05 and |delta| > 5%) (Fig. 7a) and comparable to the numbers of DMRs in shuffled data (Supplementary Table 16). These DMRs showed a roughly equal distribution of hypomethylation and hypermethylation (Fig. 7b). Most were marked by the repres- sive histone modification H3K27me3 (Fig. 7c), while hypermethyl- ated regions in T2D were also somewhat enriched for NKX2.2 binding (Fig. 7d). Neither hypomethylated nor hypermethylated regions were associated with specific biological pathways. Altogether, our data demonstrate that beta cells in T2D exhibit an accelerated demethylation pattern resembling age-related changes, predominantly in TF binding sites, a phenomenon not observed in alpha cells, and whose functional consequences remain to be investigated. ### Page 10 Nature Metabolism | Volume 8 | April 2026 | 941–956 950Resource https://doi.org/10.1038/s42255-026-01495-yPDX1 most variable peaks in CTL Group CTL T2D Age 60 35 0 0.2 0.4 0.6 0.8 1e f a b c dAge (years) Median methylation level Slope Methylation vs age CTL T2D CTL T2D Methylation vs age Slope<0 Slope<0 Methylation vs age Slope<0 Slope<0 PDX1 most variable peaks in T2D Group CTL T2D Age 60 35 0 0.2 0.4 0.6 0.8 1 0.01 *** *** *** –0.01 –0.02 40 50 60 70 0.61 0.60 0.59 0.58 0.57 0.56 0.55 0 Pearson correlation 0 –0.50 –0.75 –0.25P value of Pearson correlation 0.5 0.3 0.2 0.1 0 0.4 CTL T2D CTL T2DFig. 6 | Beta cells exhibit accelerated DNA demethylation in T2D (part II). a,b, PDX1binding sites methylation heatmaps. Heatmaps of average beta cell methylation atPDX1 binding sites32 in CTL (a) and T2D (b) donors. Shown are the 500 most variable peaks (separately, for CTL and T2D) across seven age-matched donors. c, Linear fit slopes for PDX1 binding sites. Violin plots of slopes from linear fits of methylation versus age at the regions shown in a and b, for beta cells from CTL and T2D donors. Asterisks indicate a two-sided Wilcoxon P < 0.001 (in this case, P = 2.8 × 10−9 ).d,e, Pearson correlations for PDX1 binding sites. Boxplots of Pearson correlation coefficients (d) and corresponding P values (e) for methylation versus age at regions (398 for CTL and 310 for T2D) with slope of <0 from a and b. Boxplot parameters: middle line, median; box upper (resp. lower) bound, 3rd (resp. 1st ) quartile; upper/ lower whisker extends to the largest/smallest value within 1.5× IQR; individual points, outliers. Asterisks indicate a two-sided Wilcoxon P < 0.001 (P = 6.5 × 10−5 in both cases). f, Methylation versus age in PDX1 binding sites. Median methylation values across the PDX1 binding sites in the fifth quintile of the median methylation levels for seven CTL (blue) and seven T2D (red) age-matched donors. ### Page 11 Nature Metabolism | Volume 8 | April 2026 | 941–956 951Resource https://doi.org/10.1038/s42255-026-01495-yDiscussionThis study leverages comprehensive human islet cell methyl- ome datasets generated by HPAP, a unique resource that provides high-resolution, cell-type-specific epigenomic profiles from human donors across a range of ages and diabetes status 26,27 . Although previ- ous studies have examined DNAm in whole islets, this dataset enables analysis of epigenetic changes in purified alpha cell and beta cell popu- lations. In an accompanying paper, our colleagues51 analysed DNAm from sorted human alpha and beta cells and, similar to our findings, identified thousands of DMRs between the two cell types. Remarkably, 66% of the genes associated with DMRs in our study were also associ- ated with DMRs in theirs, despite the differences in experimental and bioinformatics methodologies used, underscoring the robustness of this epigenetic mark. Using a subset of HPAP donors with alpha and beta cell samples of the highest purity, we traced DNAm changes associated with age and T2D. Our analysis revealed that beta cells, but not alpha cells, from healthy donors undergo progressive demethylation with age at active regulatory regions, particularly at TF-bound cis-regulatory elements. This finding is consistent with our previous study in mice40 , which suggested a conserved adaptive process that supports beta cell function with age. Owing to the limited age range and smaller number of T2D donors, we could not perform an age-related DNAm analysis in T2D. However, a similar demethylation pattern was also noted in T2D compared to CTL donors, suggesting that T2D beta cells may acquire a hypomethylated state earlier in life, consistent with a previous study showing that premature DNA demethylation in peripheral blood leuco- cytes is associated with increased risk of T2D 52 . Although the changes0 200 400 600 –30 –20 –10 10 20 300 20 40 60 80 100 0 20 40 60 80 100a b c d% methylation control % methylation difference (T2D – control)Number of DMRs per binIslet H3K27Ac Islet H3K4me1 Islet H3K27me3 snATAC alpha CTR FOXA2 Islets MAFB Islets NKX2.2 Islets NKX6.1 Islets PDX1 Islets % methylation T2D snATAC alpha T2D 0 1 2 3 H3K27ac islets H3K4me1 H3K27me3 DMRs snATAC alpha CTL snATAC alpha T2D FOXA2 islets MAFB islets MKX2.2 islets NKX6.1 islets 4 0 0 200 400 600 800 0 200 400 600 0 2 4 0 2 46 60 1 2 3 4 PDX1 islets 0 1 2 3 4 0 1 2 3 4 1 2 3 4 0 0.05 0.15 0.2 CTL.hyper T2D.hyper DMRs CTL.hyper T2D.hyper 0.1Fig. 7 | Alpha cells show minimal DNA methylation differences in T2D.a, Methylation scatterplot. Smooth scatterplot of alpha cell methylation (%) in T2D versus CTL donors across 1,574 DMRs (FDR < 0.05, |delta| > 5%), illustrating subtle, bidirectional methylation changes in T2D. b, Effect size distribution. Distribution of average methylation differences (T2D – CTL) for the DMRs shown in a. x axis: % difference (T2D – CTL); y axis, number of DMRs per bin.c,d, Enrichment heatmaps. Enrichment plots of signals from islet histone marks (c) and chromatin accessibility (snATAC-seq, alpha cells) and TF binding (ChIP-seq, whole islets) (d) across 1,574 DMRs (FDR < 0.05, |delta| > 5%) from the T2D versus CTL alpha cell comparison. Each row shows the signal over the 10 kb region centred at one DMR. Top, regions hypomethylated in T2D; bottom, hypomethylated in CTL alpha cells. ### Page 12 Nature Metabolism | Volume 8 | April 2026 | 941–956 952Resource https://doi.org/10.1038/s42255-026-01495-yare modest in magnitude, they are statistically significant and consist- ently enriched in functionally relevant regions. The enrichment of hypomethylated regions in regulatory elements suggests a regulated and potentially adaptive process, rather than stochastic drift, targeting loci essential for beta cell identity and function. DNAm patterns in differentiated cells are generally stable and maintained by the coordinated activity of the maintenance methyl- transferase DNMT1 together with DNMT3A, which, although tradition- ally described as a de novo methyltransferase, has been recently shown to also contribute to maintenance methylation30,31,53–55 . However, these patterns can be actively remodelled by TET enzymes, which oxidize 5-methylcytosine (5mC) to 5-hydroxymethylcytosine (5hmC). This intermediate can either initiate active or passive demethylation or persist as a stable epigenetic mark, especially in the context of ageing and disease56 . Importantly, TET-mediated demethylation accumulates at active regulatory regions and TF binding sites 28,57 , where it may counteract DNMT activity and protect enhancers from repressive methylation28,29,58 . In beta cells, such mechanisms could help sustain enhancer activity and support functionality throughout life. Our find- ings support this idea, as we show that beta-cell-specific gene sets are preferentially hypomethylated, and that core identity genes such asPDX1, MAFA, FOXA2 and ABCC8, which are associated with hypomethyla- tion, maintain stable expression levels throughout life and in T2D. By contrast, other pathways not enriched near hypomethylated regions, such as those involved in mitochondrial metabolism, oxidative stress defence and secretory processes, tend to show reduced expression in T2D, suggesting that in beta cells, demethylation in regulatory regions primarily serves to preserve stable expression of the identity gene pro- gram rather than drive broad transcriptional activation. Alternatively, or in addition, the observed demethylation may result from rare repli- cation events triggered by increased metabolic demand throughout life and under diabetic conditions, during which 5hmC-modified sites are not recognized by DNMT1, leading to a loss of CpG methylation in the daughter cells. However, alpha cells, which have been reported to proliferate at a higher rate in T2D23 , do not show substantial hypometh- ylation, suggesting that replication-dependent passive demethylation alone is insufficient to explain the beta-cell-specific pattern. Clarifying the relative contributions of these mechanisms will require profiling steady-state 5hmC levels in alpha and beta cells from non-diabetic and diabetic donors. Finally, the expression levels of DNMT1, DNMT3A, DNMT3B and the TET enzymes (TET1–3), which collectively regulate DNA methylation state, remain unchanged with age or in T2D, but are lower in alpha cells than in beta cells. Beta cell gene expression is influenced by conditions such as endo- plasmic reticulum stress, oxidative stress and metabolic workload, and gene activation and repression are often reversible, supporting the adaptive capacity of beta cells until chronic stress overwhelms this capacity, leading to failure and progression to T2D59 . By contrast, DNAm, being a relatively stable epigenetic mark, may not respond to transient metabolic cues or drive rapid changes in gene expression. Indeed, in our study, most age-related and disease-associated dem- ethylated regions did not strongly correlate with increased expression of nearby genes. Nonetheless, the specific demethylation process observed in beta cells but not alpha cells supports the idea that this represents a cell-type-specific epigenetic adaptation. In beta cells, demethylation may serve to preserve enhancer activity and sustain cellular responsiveness over time, rather than directly modulate gene expression acutely. Based on the fact that we did not detect significant overlap between BMI-demethylated and age-demethylated DMRs, this adaptation is mainly associated with ageing, rather than metabolic stress. Although the functional significance of this age-demethylation remains to be fully elucidated, its cell-type specificity highlights a unique epigenetic adaptive response in this long-lived, metabolically responsive cell type. However, in the context of T2D, this mechanism may ultimately fail, leading to impaired function despite preserved expression of core identity genes. Future studies should explore whether targeted modulation of the DNAm machinery, particularly TET activity, can enhance beta cell resilience during ageing and diabetes.MethodsHuman pancreatic isletsPancreatic islets were procured by the HPAP consortium under the Human Islet Research Network (https://hirnetwork.org) with approval from the University of Florida Institutional Review Board (201600029) and the United Network for Organ Sharing. A legal representative for each donor provided informed written consent before organ retrieval, and the data have been deidentified. The protocols used for the human islet sorting, whole genome bisulfite sequencing (WGBS), scRNA-seq and snATAC-seq assays used in this work are available on the PancDB website at https://hpap.pmacs.upenn.edu/explore/workflow/islet- molecular-phenotyping-studies (see Kaestner lab protocols). In addi- tion to libraries generated with WGBS, the methylome data used in this work also comprises libraries generated using enzymatic methyl sequencing (EM-seq) following these steps: 100 ng DNA input was sheared to 300 bp and cleaned with the Zymo research kit (D4014). Quality control was performed on a BioAnalyzer (Agilent), and EM-seq libraries were generated (New England Biolabs, E7120/E7140). All kits were used following the manufacturer’s instructions.Data analysisMethylome. All available (154) HPAP methylome samples, regardless of cell type or donor disease state, were pre-processed together (see Extended Data Fig. 9 for workflow outline). Adaptors were trimmed from paired-end reads with cutadapt (v.4.4) 60 . Trimmed reads were aligned to the human hg38, Lambda and pUC19 genomes using bwa-meth (v.0.2.7) 61 . The resulting sam files were converted to bam files using samtools (v.1.14) 62 , then duplicated reads were marked by sambamba (v.0.8.1) 63 and the resulting bam files were sorted and indexed with samtools. We used wgbstools (v.0.2.0) 24 to generate preliminary pat, beta and tab files from the hg38 alignments while removing reads with low mapping quality, duplicated or not mapped in a proper pair (using -F 1796 -q 10). We then leveraged the HPAP whole genome sequencing data to identify the C > T and G > A variants at CpG positions for each donor and subtracted them from the tab files, also using bedtools (v.2.26.0)64 . Using wgbstools again, we generated beta files from the filtered tab files, then computed average coverage, gen- erated bigwig tracks and segmented the genome using these filtered beta files. In general, the segmentation from wgbstools identifies non-overlapping continuous regions of highly correlated CpG sites similarly methylated in each sample but possibly covarying across conditions. We derived the segmentation from all samples with an average coverage above the median (22×) and required a minimum of four CpGs per region and a max region size of 2,000 bp. This segmenta- tion (2,061,859 regions; median size, 607 bp; median number of CpGs, 8) was then applied to all samples and used in subsequent analyses. For each sample, we estimated purity using the UXM fragment-level deconvolution algorithm24 (v.0.1.0) with reference Atlas.U25.l4.hg38. full.tsv (from https://github.com/nloyfer/UXM_deconv) restricted to the markers for pancreatic cell types, endothelial cells and monocytes and macrophages. Before applying UXM, we refined the atlas for each donor by filtering out CpG positions where that donor had C > T or G > A variants. We leveraged methylSig (v.1.11.0)65 for differential methylation analyses. We performed several comparisons, including (1) alpha versus beta cells in CTL donors; (2) age (as a numeric covariate) separately in alpha and (3) in beta cells of CTL donors; and (4) T2D versus CTL donors separately for alpha and (5) beta cells. For each comparison, we selected relevant samples with sufficiently high average coverage (~20× or more) and estimated purity (~65% or more). For comparisons between T2D and CTL donors, we selected donors in similar age ranges in the two ### Page 13 Nature Metabolism | Volume 8 | April 2026 | 941–956 953Resource https://doi.org/10.1038/s42255-026-01495-ygroups. The characteristics of the samples used in each comparison are summarized in Supplementary Tables 1–5. For each comparison, we removed CpGs covered by fewer than ten reads in any of the samples and then aggregated methylation fractions by segmentation regions and removed any region covered by fewer than ten reads in any of the samples. We focused on regions that were not on the X and Y chromo- somes, with an average methylation difference of at least 5% between the groups being compared. In the case of the age comparisons (where age was treated as a numeric covariate), the two groups used for this filtering were those of donors aged 39–58 years and 18–35 years. To find DMRs, we used the general model test with DSS and a design dependent on the comparison of interest. In comparison (1), we paired by donor; in comparisons (3) to (5), we included ancestry and assay type (WGBS or EM-seq) in the model. In comparison (2), we included assay type in the model but not ancestry, given that the latter was confounded with age. For each comparison, DMRs for further examination were defined based on thresholds for the FDR and effect size; note that effect size here refers to the absolute value of delta, the difference of methylation levels between the two groups as estimated by methylSig (meth_diff). We set basic thresholds of 0.05 for the FDR and 5% for the effect size. But we made either or both of these thresholds more stringent for those comparisons where we expected less subtle differences (also based on inspection of PC and effect size plots), to focus on the top DMRs. The thresholds chosen are listed in Supplementary Table 6. To functionally annotate DMRs, we defined the following sets of features based on the Gencode Human v43 transcript models: • 1,500 bp upstream of the transcription start site (TSS); • 1,501–5,000 bp upstream of TSS; • 5,001–50,000 bp upstream of TSS; • 1,000 bp downstream of TSS; and • 1,001–50,000 bp downstream of TSS and downstream of the transcription end site. We used bedtools (v.2.26.0)64 to mark the features from each of the above sets overlapping with each DMR. To explore enrichment of signals from alpha, beta and islet-specific snATAC-seq and ChIP-seq experiments, we leveraged EnrichedHeatmap (v.1.32.0) 66 . We downloaded the bigwig files with accessions listed in Supplementary Table 7 from the Common Metabolic Disease Genome Atlas (https://cmdga.org). For H3K27me3, we used data from a previous publication 67 , lifted over to hg38 with LiftOver 68 . In addition, we gener- ated bigwig files for chromatin accessibility in alpha and beta cells from HPAP snATAC-seq experiments in CTL and T2D donors as described below. Our plots also leveraged pheatmap (v.1.0.12) (https://github. com/raivokolde/pheatmap) and ggplot2 (v.3.5.1)69 . The principal com- ponent analyses leveraged PCATools (v.2.14.0) (https://bioconductor. org/packages/PCAtools) with removeVar = 0.1. The analyses of beta cell methylomes from healthy donors with respect to BMI and GSIS were done in the same manner as the age analy- ses, using methylSig and with assay type and ancestry in the model. BMI (resp. GSIS) was used as a numeric covariate, and the effect size was based on its top and bottom quartiles. GSIS was measured by the insulin ratio B/A derived from the individual perifusion data obtained from PancDB as follows. The numerator, B, was the maximum insulin secretion value after stimulation with G16.7. The denominator, A, was the average insulin secretion over the time points in the ‘plateau’ amino acid stimulation phase (AAM, before G3 stimulation). For the latter, outlier time points at the beginning of the AAM phase were manually filtered out donor by donor. As a quality control check, for each of the alpha and beta cell age or T2D versus CTL comparison, we randomly shuffled the sample labels 100 times and ran the same analysis on the shuffled data (summary counts are provided in Supplementary Table 16). For the age compari- sons in both cell types and the beta cell T2D comparison, the number of DMRs identified in shuffled data was much reduced (almost null in the age comparisons) and did not follow the imbalance in hypomethylation versus hypermethylation observed in the unshuffled data.scRNA-seq. For each donor islet sample, we obtained counts with CellRanger (v.7.1.0) 70 and used Seurat (v.4.9.9.9041) 71 (complemented by SoupX (v.1.6.2) 72 and scDblFinder (v.1.12.0) 73 ) for clean-up, normali- zation, pre-processing and data integration. Cells were annotated with two approaches: scSorter (v.0.0.2)74 and manual cluster annotation based on selected pancreatic cell markers. Only cells for which the two approaches agreed were assigned to a final cell type. For each cell type, we used all available relevant donors for a com- parison of interest, with at least 100 cells of that type and processed with SC3Pv3 chemistry. We generated pseudo-bulk counts and used DESeq2 (v.1.38.1) for differential expression analyses 75 . For the age comparison, we did not treat age as a numeric covariate as we did for DMRs, but we compared age group categories (18–35 vs 39–58 years old) based on DESeq2 recommended guidelines.snATAC-Seq profiles in alpha and beta cells. Chromatin accessibility profiles for alpha and beta cells were generated using HPAP snATAC-seq datasets. Each dataset was first analysed using Signac (v.1.7.0)76 and Seurat (v.4.4.0)71 . A standard processing pipeline was applied (https:// stuartlab.org/signac/articles/pbmc_vignette). Nuclei with fewer than 1,000 or more than 100,000 reads in peaks, a TSS enrichment score of <1 or a nucleosome signal of >2 were filtered out. Each nucleus was annotated by mapping to an internal multiome reference dataset, and nuclei with a mapping score of >0.8 were kept. Then, to generate cell-type-specific tracks in each disease group, we used samples from CTL donors with age ≥30 years and samples from T2D donors. With cell types annotated from the previous steps, fragments associated with cells from each donor and each cell type (alpha and beta) were merged to generate pseudo-bulk tracks; tracks with fewer than 30 cells were removed. A bed file including the start, end and chromosome information of each read was created for each donor and each cell type. Bam files were constructed using bedToBam (v.2.30.0) 64 with a genome size file provided as part of the GRCh38 reference – 2020-A-2.0.0 available on the 10× Genom- ics website (https://cf.10xgenomics.com/supp/cell-arc/refdata- cellranger-arc-GRCh38-2020-A-2.0.0.tar.gz). Bam files were indexed using samtools (v.1.11 index) 62 . BAMscale (v.0.0.5) 77 was used to gen - erate scaled coverage tracks, with parameter -t 4. Lastly, to generate cell-type-specific tracks per disease group, cell-type-specific tracks were merged across donors of the same condition using bigWigMerge (v.2) and the output was sorted and converted to a bigwig file using bedGraphToBigWig (v.2.8)78 .Gene set analyses. To analyse enrichment of gene sets associated with DNAm changes, we focused on a gene set collection comprising gene sets obtained from MSigDB 79 and custom gene sets defined in a previous publication 13 . The hypothesis-driven gene set collection is described in Supplementary Table 8. For these analyses, we used the top DMRs (in terms of effect size) to reduce noise and focus on the most robust enrichment. For the comparison between alpha and beta cells, we used a |delta| > 40% cutoff owing to the large inherent differences between these two cell types, whereas for all other com - parisons (age and T2D), we used a consistent |delta| > 10% cutoff. We used a hypergeometric test implemented in the Genomica (v.3.0.0) software 80 , considering gene sets with P < 0.01 and FDR < 0.05 to be significant. For the association of genes to DMRs in likely enhancers, we used protein-coding genes in the unions of the following DMR annotation files defined above: 1,501–50,000 bp upstream of TSS or 1,001–50,000 bp downstream of TSS and downstream of the transcrip- tion end site. We refer to these genes as ‘DMR-distal’. To analyse enrichment of gene sets associated with mRNA expres- sion, we used GSEA79,81 (v.4.3.3) on the DESeq2 normalized scRNA-seq ### Page 14 Nature Metabolism | Volume 8 | April 2026 | 941–956 954Resource https://doi.org/10.1038/s42255-026-01495-ypseudo-bulk counts. The gene sets input in these analyses were the Hall- mark gene sets from MSigDB, to cover the main biological programmes.T2D GWAS SNP enrichment analyses. To examine the 558 T2D-associated SNPs identified in a previous publication50 versus the 8,619 beta cell demethylated DMRs in T2D, we used GREGOR (v.1.4.0)82 after lifting these regions over to hg19. We used default parameter settings, a threshold of r2 = 0.7 for proxies and EUR as the reference population (given that most of the donors in this comparison were from this ancestry).Reporting summaryFurther information on research design is available in the Nature Portfolio Reporting Summary linked to this article.Data availabilityRaw data and metadata related to this work are available on the PancDB website https://hpap.pmacs.upenn.edu and on dbGaP (https://www. ncbi.nlm.nih.gov/gap) with accession no. phs002465 (v.2). With the PancDB website interactive analysis link, it is also possible to explore Genome Browser tracks for our methylation data, as well as the results of islet perifusion experiments used in our insulin secretion ratio cal- culations. Finally, the result files for the methylome comparisons in this work and the scRNA-seq DESeq2 results are available in the Supplementary Data.Code availabilityThe study was conducted using only publicly available software as out- lined above; no new code was generated. The list of software packages and versions used is also available in the Nature Portfolio Reporting Summary linked to this article.References1. Al-Khelaifi, F. et al. Metabolic GWAS of elite athletes reveals novel genetically-influenced metabolites associated with athletic performance. Sci Rep. 9, 19889 (2019). 2. Kresnajati, S. et al. Changes in arterial stiffness in response to various types of exercise modalities: a narrative review on physiological and endothelial senescence perspectives. Cells 11, 3544 (2022). 3. Pontzer, H. et al. Daily energy expenditure through the human life course. Science 373, 808–812 (2021). 4. Chang-Chen, K. J., Mullur, R. & Bernal-Mizrachi, E. Beta-cell failure as a complication of diabetes. Rev. Endocr. Metab. Disord. 9, 329–343 (2008). 5. Cefalu, W. T. et al. Contribution of visceral fat mass to the insulin resistance of aging. Metabolism 44, 954–959 (1995). 6. Iozzo, P. et al. Independent influence of age on basal insulin secretion in nondiabetic humans. European Group for the Study of Insulin Resistance. J. Clin. Endocrinol. Metab. 84, 863–868 (1999). 7. Arrojo, E. D. R. et al. Age mosaicism across multiple scales in adult tissues. Cell Metab. 30, 343–351.e3 (2019). 8. Cnop, M. et al. The long lifespan and low turnover of human islet beta cells estimated by mathematical modelling of lipofuscin accumulation. Diabetologia 53, 321–330 (2010). 9. Perl, S. et al. Significant human beta-cell turnover is limited to the first three decades of life as determined by in vivo thymidine analog incorporation and radiocarbon dating. J. Clin. Endocrinol. Metab. 95, E234–E239 (2010). 10. Wang, Y. J. et al. Single-cell mass cytometry analysis of the human endocrine pancreas. Cell Metab 24, 616–626 (2016). 11. Castell, A. L. et al.β Cell mass expansion during puberty involves serotonin signaling and determines glucose homeostasis in adulthood. JCI Insight 7, e160854 (2022). 12. Rieck, S. & Kaestner, K. H. Expansion of β-cell mass in response to pregnancy. Trends Endocrinol. Metab. 21, 151–158 (2010). 13. Avrahami, D. et al. Single-cell transcriptomics of human islet ontogeny defines the molecular basis of β-cell dedifferentiation in T2D. Mol.Metab. 42, 101057 (2020). 14. Shrestha, S. et al. Aging compromises human islet beta cell function and identity by decreasing transcription factor activity and inducing ER stress. Sci. Adv. 8, eabo3932 (2022). 15. Bergman, Y. & Cedar, H. DNA methylation dynamics in health and disease. Nat. Struct. Mol. Biol. 20, 274–281 (2013). 16. Horvath, S. & Raj, K. DNA methylation-based biomarkers and the epigenetic clock theory of ageing. Nat. Rev. Genet. 19, 371–384 (2018). 17. Das, P. M. & Singal, R. DNA methylation and cancer. J. Clin. Oncol.22, 4632–4642 (2004). 18. Volkov, P. et al. Whole-genome bisulfite sequencing of human pancreatic islets reveals novel differentially methylated regions in type 2 diabetes pathogenesis. Diabetes 66, 1074–1085 (2017). 19. Dayeh, T. et al. Genome-wide DNA methylation analysis of human pancreatic islets from type 2 diabetic and non-diabetic donors identifies candidate genes that influence insulin secretion. PLoS Genet. 10, e1004160 (2014). 20. Ronn, T. et al. Genes with epigenetic alterations in human pancreatic islets impact mitochondrial function, insulin secretion, and type 2 diabetes. Nat. Commun. 14, 8040 (2023). 21. Henquin, J. C. & Rahier, J. Pancreatic alpha cell mass in European subjects with type 2 diabetes. Diabetologia 54, 1720–1725 (2011). 22. Wu, M. et al. Single-cell analysis of the human pancreas in type 2 diabetes using multi-spectral imaging mass cytometry. Cell Rep.37, 109919 (2021). 23. Yoon, K. H. et al. Selective β-cell loss and α-cell expansion in patients with type 2 diabetes mellitus in Korea. J. Clin. Endocrinol. Metab. 88, 2300–2308 (2003). 24. Loyfer, N. et al. A DNA methylation atlas of normal human cell types. Nature 613, 355–364 (2023). 25. Drawshy, Z. et al. DNA methylation-based assessment of cell composition in human pancreas and islets. Diabetes 73, 554–564 (2024). 26. Kaestner, K. H., Powers, A. C., Naji, A., Consortium, H. & Atkinson, M. A. NIH initiative to improve understanding of the pancreas, islet, and autoimmunity in type 1 diabetes: The Human Pancreas Analysis Program (HPAP). Diabetes 68, 1394–1402 (2019). 27. Shapira, S. N., Naji, A., Atkinson, M. A., Powers, A. C. & Kaestner, K. H. Understanding islet dysfunction in type 2 diabetes through multidimensional pancreatic phenotyping: The Human Pancreas Analysis Program. Cell Metab. 34, 1906–1913 (2022). 28. Hon, G. C. et al. 5mC oxidation by Tet2 modulates enhancer activity and timing of transcriptome reprogramming during differentiation. Mol Cell. 56, 286–297 (2014). 29. Stefansson, O. A. et al. The correlation between CpG methylation and gene expression is driven by sequence variants. Nat. Genet.56, 1624–1631 (2024). 30. Li, E., Bestor, T. H. & Jaenisch, R. Targeted mutation of the DNA methyltransferase gene results in embryonic lethality. Cell 69, 915–926 (1992). 31. Lei, H. et al. De novo DNA cytosine methyltransferase activities in mouse embryonic stem cells. Development 122, 3195–3205 (1996). 32. Pasquali, L. et al. Pancreatic islet enhancer clusters enriched in type 2 diabetes risk-associated variants. Nat. Genet. 46, 136–143 (2014). 33. Neiman, D. et al. Islet cells share promoter hypomethylation independently of expression, but exhibit cell-type-specific methylation in enhancers. Proc. Natl Acad. Sci. USA 114, 13525–13530 (2017). ### Page 15 Nature Metabolism | Volume 8 | April 2026 | 941–956 955Resource https://doi.org/10.1038/s42255-026-01495-y34. Hamilton, A. et al. Adrenaline stimulates glucagon secretion by Tpc2-dependent Ca2+ mobilization from acidic stores in pancreatic α-cells. Diabetes 67, 1128–1139 (2018). 35. Wortham, M. & Sander, M. Mechanisms of β-cell functional adaptation to changes in workload. Diabetes Obes. Metab. 18, 78–86 (2016). 36. Enge, M. et al. Single-cell analysis of human pancreas reveals transcriptional signatures of aging and somatic mutation patterns. Cell 171, 321–330.e14 (2017). 37. Tserel, L. et al. Age-related profiling of DNA methylation in CD8+ T cells reveals changes in immune response and transcriptional regulator genes. Sci. Rep. 5, 13107 (2015). 38. Tong, H. et al. Cell-type specific epigenetic clocks to quantify biological age at cell-type resolution. Aging (Albany NY) 16, 13452–13504 (2024). 39. Gontier, G. et al. Tet2 rescues age-related regenerative decline and enhances cognitive function in the adult mouse brain.Cell Rep. 22, 1974–1981 (2018). 40. Avrahami, D. et al. Aging-dependent demethylation of regulatory elements correlates with chromatin state and improved β cell function. Cell Metab. 22, 619–632 (2015). 41. Dorrell, C. et al. Human islets contain four distinct subtypes of β cells. Nat. Commun. 7, 11756 (2016). 42. Dror, E. et al. Epigenetic dosage identifies two major and functionally distinct β cell subtypes. Cell Metab. 35, 821–836.e7 (2023). 43. Brooks, E. P. et al. NKX2.2 and KLF4 cooperate to regulate α-cell identity. Genes Dev. 39, 242–260 (2025). 44. Gao, N. et al. Foxa1 and Foxa2 maintain the metabolic and secretory features of the mature β-cell. Mol. Endocrinol. 24, 1594–1604 (2010). 45. Gao, N. et al. Foxa2 controls vesicle docking and insulin secretion in mature β cells. Cell Metab. 6, 267–279 (2007). 46. Lee, C. S., Sund, N. J., Behr, R., Herrera, P. L. & Kaestner, K. H.Foxa2is required for the differentiation of pancreatic α-cells. Dev. Biol. 278, 484–495 (2005). 47. Pullen, T. J. & Rutter, G. A. When less is more: the forbidden fruits of gene repression in the adult β-cell. Diabetes Obes. Metab. 15, 503–512 (2013). 48. Fang, Z. et al. Single-cell heterogeneity analysis and CRISPR screen identify key β-cell-specific disease genes. Cell Rep. 26, 3132–3144.e7 (2019). 49. Segerstolpe, A. et al. Single-cell transcriptome profiling of human pancreatic islets in health and type 2 diabetes. Cell Metab. 24, 593–607 (2016). 50. Vujkovic, M. et al. Discovery of 318 new risk loci for type 2 diabetes and related vascular outcomes among 1.4 million participants in a multi-ancestry meta-analysis. Nat. Genet. 52, 680–691 (2020). 51. Ofori, J. K. et al. Cell-specific DNA methylation in human α- and β-cells regulates gene expression in type 2 diabetes. Nat. Metab.https://doi.org/10.1038/s42255-026-01498-9 (2026). 52. Toperoff, G. et al. Premature aging of leukocyte DNA methylation is associated with type 2 diabetes prevalence. Clin. Epigenetics 7, 35 (2015). 53. Feng, J. et al. Dnmt1 and Dnmt3a maintain DNA methylation and regulate synaptic function in adult forebrain neurons. Nat. Neurosci. 13, 423–430 (2010). 54. Jeong, S. et al. Selective anchoring of DNA methyltransferases 3A and 3B to nucleosomes containing methylated DNA. Mol. Cell. Biol. 29, 5366–5376 (2009). 55. Xu, Z. et al. Regulation of de novo and maintenance DNA methylation by DNA methyltransferases in postimplantation embryos. J. Biol. Chem. 301, 107990 (2025). 56. Bachman, M. et al. 5-Hydroxymethylcytosine is a predominantly stable DNA modification. Nat. Chem. 6, 1049–1055 (2014). 57. Schubeler, D. Function and information content of DNA methylation. Nature 517, 321–326 (2015). 58. Charlton, J. et al. TETs compete with DNMT3 activity in pluripotent cells at thousands of methylated somatic enhancers. Nat. Genet.52, 819–827 (2020). 59. Chen, C. W. et al. Adaptation to chronic ER stress enforces pancreatic β-cell plasticity. Nat. Commun. 13, 4621 (2022). 60. Martin, M. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet J. 17, 10–12 (2011). 61. Pedersen, B. S., Eyring, K., De, S., Yang, I. V. & Schwartz, D. A. Fast and accurate alignment of long bisulfite-seq reads. Preprint at https://arxiv.org/abs/1401.1129 (2014). 62. Danecek, P. et al. Twelve years of SAMtools and BCFtools.Gigascience 10, giab008 (2021). 63. Tarasov, A., Vilella, A. J., Cuppen, E., Nijman, I. J. & Prins, P. Sambamba: fast processing of NGS alignment formats.Bioinformatics 31, 2032–2034 (2015). 64. Quinlan, A. R. & Hall, I. M. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26, 841–842 (2010). 65. Park, Y., Figueroa, M. E., Rozek, L. S. & Sartor, M. A. MethylSig: a whole genome DNA methylation analysis pipeline. Bioinformatics30, 2414–2422 (2014). 66. Gu, Z., Eils, R., Schlesner, M. & Ishaque, N. EnrichedHeatmap: an R/Bioconductor package for comprehensive visualization of genomic signal associations. BMC Genomics 19, 234 (2018). 67. Bramswig, N. C. et al. Epigenomic plasticity enables human pancreatic α to β cell reprogramming. J. Clin. Invest. 123, 1275–1284 (2013). 68. Hinrichs, A. S. et al. The UCSC Genome Browser Database: update 2006. Nucleic Acids Res. 34, D590–D598 (2006). 69. Wickham, H. ggplot2: Elegant Graphics for Data Analysis(Springer, 2016). 70. Zheng, G. X. et al. Massively parallel digital transcriptional profiling of single cells. Nat. Commun. 8, 14049 (2017). 71. Hao, Y. et al. Integrated analysis of multimodal single-cell data.Cell 184, 3573–3587.e29 (2021). 72. Young, M. D. & Behjati, S. SoupX removes ambient RNA contamination from droplet-based single-cell RNA sequencing data. Gigascience 9, giaa151 (2020). 73. Germain, P. L., Lun, A., Garcia Meixide, C., Macnair, W. & Robinson, M. D. Doublet identification in single-cell sequencing data using scDblFinder. F1000Res 10, 979 (2021). 74. Guo, H. & Li, J. scSorter: assigning cells to known cell types according to marker genes. Genome Biol. 22, 69 (2021). 75. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014). 76. Stuart, T., Srivastava, A., Madad, S., Lareau, C. A. & Satija, R. Single-cell chromatin state analysis with Signac. Nat. Methods 18, 1333–1341 (2021). 77. Pongor, L. S. et al. BAMscale: quantification of next-generation sequencing peaks and generation of scaled coverage tracks.Epigenetics Chromatin 13, 21 (2020). 78. Kent, W. J., Zweig, A. S., Barber, G., Hinrichs, A. S. & Karolchik, D. BigWig and BigBed: enabling browsing of large distributed datasets. Bioinformatics 26, 2204–2207 (2010). 79. Subramanian, A. et al. Gene set enrichment analysis: a knowledge- based approach for interpreting genome-wide expression profiles.Proc. Natl Acad. Sci. USA 102, 15545–15550 (2005). 80. Segal, E. et al. Module networks: identifying regulatory modules and their condition-specific regulators from gene expression data. Nat. Genet. 34, 166–176 (2003). 81. Mootha, V. K. et al. PGC-1α-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat. Genet. 34, 267–273 (2003). ### Page 16 Nature Metabolism | Volume 8 | April 2026 | 941–956 956Resource https://doi.org/10.1038/s42255-026-01495-y82. Schmidt, E. M. et al. GREGOR: evaluating global enrichment of trait-associated variants in epigenomic features using a systematic, data-driven approach. Bioinformatics 31, 2601–2606 (2015).AcknowledgementsWe thank the families of the organ donors and the entire HPAP team for making this study possible. We used data acquired from the HPAP (HPAP-RRID:SCR_016202) database (https://hpap.pmacs.upenn.edu), a Human Islet Research Network (RRID:SCR_014393) consortium supported by NIDDK grants UC4-DK-112217, U01-DK-123594, UC4-DK-112232 and U01-DK-123716. Further support was provided by NIDDK grant U01-DK-134995 to D.A., B.G. and K.H.K. B.G. and D.A. are supported by grants from the Israel Science Foundation and JDRF (1782/18 and 2982/20), the US–Israel Binational Science Foundation (2019314) and the EU’s Horizon 2020 research and innovation program no. 874710. D.A. is supported by the Israel Science Foundation (2166/24). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Finally, we thank the editor and reviewers for their constructive comments and suggestions.Author contributionsD.A. and K.H.K. conceived and supervised this work. B.G. provided conceptual input. E.M., D.A. and K.H.K. wrote the manuscript. E.M. pre-processed the data, performed most computational analyses and uploaded the data to PancDB and dbGaP. H.C.D. performed human islet sorting, prepared and sequenced the EM-seq, scRNA-seq and snATAC-seq libraries, performed PC analyses and contributed to data interpretation and manuscript outline. D.A. performed gene set analyses and GSEA. J.L. prepared the WGBS libraries. J.S. oversaw the sequencing, provided feedback and helped with data uploads to PancDB and dbGaP. T.D., D.L., D.B. and H.E.-M. contributed to human islet sorting and generated scRNA-seq and snATAC-seq libraries. M.L. performed the snATAC-seq analyses. C.L. and A.N. performed islet isolation from donor tissues. E.F performed additional biological analyses. All authors reviewed and approved the manuscript.Competing interestsThe authors declare no competing interests.Additional informationExtended data is available for this paper at https://doi.org/10.1038/s42255-026-01495-y.Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s42255-026-01495-y.Correspondence and requests for materials should be addressed to Klaus H. Kaestner or Dana Avrahami.Peer review information Nature Metabolism thanks Guoqiang Gu and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Revati Dewal, in collaboration with the Nature Metabolism team.Reprints and permissions information is available at www.nature.com/reprints.Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/. © The Author(s) 2026 ### Page 17 Nature MetabolismResource https://doi.org/10.1038/s42255-026-01495-yExtended Data Fig. 1 | QC clustering. Quality control clustering of all 154 HPAP methylome samples. Rows represent the top 1% most variable segmentation regions (18,972) on chromosomes other than X and Y. Columns correspond to samples. Rows and columns are clustered using Manhattan distance as similarity metrics. Horizontal bars at the top indicate values of several covariates for each sample. uxm=UXM estimated purity score, ND.AAB= non-diabetic auto-antibody positive. ### Page 18 Nature MetabolismResource https://doi.org/10.1038/s42255-026-01495-yA B C20 25 30 35 40 45 50 55 60 0 500 1000 1500 2000 Age (years) Normalized expression NEUROD120 25 30 35 40 45 50 55 60 0 100 200 300 400 Age (years) Normalized expression GCK20 25 30 35 40 45 50 55 60 0 1000 2000 3000 4000 5000 MAFBAge (years) Normalized expression r=-0.5135, p<0.05 20 25 30 35 40 45 50 55 60 0 500 1000 1500 FOXA2Age (years) Normalized expression 20 25 30 35 40 45 50 55 60 0 200 400 600 800 1000 Normalized expression PDX120 25 30 35 40 45 50 55 60 0 1000 2000 3000 4000 Age (years) Normalized expression ABCC820 25 30 35 40 45 50 55 60 1000 10000 100000 1000000 Age (years) Normalized expression IAPP20 25 30 35 40 45 50 55 60 0 100 200 300 400 Age (years) Normalized expression SLC2A2Extended Data Fig. 2 | Beta cell scRNA-Seq data across age. MSigDB Hallmark gene sets identified by GSEA as (a) up-regulated with age and (b) down-regulated with age in beta cells from CTL donors. (c) DESeq2 normalized expression (from pseudo-bulk counts) of selected genes across age. Unadjusted two-sided Pearson correlation p-values: GCK (p = 0.05), NEUROD1 (p = 0.36), MAFB (p = 0.01), FOXA2(p = 0.40), PDX1 (p = 0.19), ISL1 (p = 0.02), ABCC8 (0.11), IAPP (p = 0.06), andSLC2S2 (p = 0.38). ### Page 19 Nature MetabolismResource https://doi.org/10.1038/s42255-026-01495-yExtended Data Fig. 3 | Heatmaps of beta cell age DMRs. Heatmaps of average methylation of the age-hypomethylated DMRs in beta cells of CTL donors, ordered by (a) age, (b) BMI, and (c) GSIS respectively. GSIS was measured by the insulin secretion ratio (insSecrB2A) described in Methods. The top bars indicate the age, BMI, and insSecrB2A values for each donor. ### Page 20 Nature MetabolismResource https://doi.org/10.1038/s42255-026-01495-yA BExtended Data Fig. 4 | See next page for caption. ### Page 21 Nature MetabolismResource https://doi.org/10.1038/s42255-026-01495-yExtended Data Fig. 4 | Enrichment heatmaps for alpha cells DMRs. Enrichment plot of signals from (a) islet histone modifications ChIP-Seq and from (b) snATAC- Seq in alpha cells and islet TF ChIP-Seq experiments for the 1,466 age-associated DMRs with FDR < 0.01 and |delta | >5% in alpha cells from CTL donors. Each row shows the signal plotted over the 10 kb region centered at one DMR. Top: regions hypermethylated in younger alpha cells; bottom: regions hypermethylated in older alpha cells. ### Page 22 Nature MetabolismResource https://doi.org/10.1038/s42255-026-01495-yExtended Data Fig. 5 | NKX2.2 binding sites methylation heatmaps. Heatmaps of average methylation for NKX2.2 peaks 32 for (a) alpha and (b) beta cells. Shown are the 500 most variable peaks (separately, for each cell type) across 13 age-matched donors. ### Page 23 Nature MetabolismResource https://doi.org/10.1038/s42255-026-01495-yExtended Data Fig. 6 | Linear fit slopes and Pearson correlations for NKX2.2binding sites. (a) Violin plots of the slopes of linear fits to methylation versus age corresponding to the regions from Extended Data Fig. 5 in the two cell types. Boxplots of (b) absolute Pearson correlation coefficients and (c) Pearson correlation p-values (for positive association in alpha cells and negative association in beta cells) of methylation versus age at the 306 regions with slope > 0 in alpha cells and the 408 regions with slope<0 in beta cells from Extended Data Fig. 5. Boxplots description: middle line=median; box upper (resp. lower) bound=3rd (resp. 1st) quartile; upper/lower whisker extends to the largest/smallest value within 1.5×IQR; individual points=outliers. The asterisks indicate a two-sided Wilcoxon p-value < 0.001 (p < 2.22e-16 for (a), p = 2.6e-08 for (b) and (c)). ### Page 24 Nature MetabolismResource https://doi.org/10.1038/s42255-026-01495-yExtended Data Fig. 7 | Beta cell scRNA-Seq data in T2D and CTL. (a) DESeq2 normalized expression (from pseudo-bulk counts) of key beta cell identity genes in CTL (n = 13) and T2D (n = 15) donors of similar age from HPAP scRNA-Seq data, showing stable expression in T2D beta cells. Data are presented as mean values +/− SD. (b) MSigDB Hallmark gene sets identified by GSEA as down-regulated in beta cells from T2D donors compared with CTLs. (c) DESeq2 normalized expression (from pseudo-bulk counts) of selected genes from broader functional pathways for the same donors as (a). Data are presented as mean values +/− SD. ### Page 25 Nature MetabolismResource https://doi.org/10.1038/s42255-026-01495-yExtended Data Fig. 8 | Beta cell methylation versus age in binding sites. Median methylation values across the (a) FOXA2 and (b) NKX2.2 binding sites in the fifth quintile of the median methylation levels for 7 CTL (blue) and 7 T2D (red) age-matched donors. ### Page 26 Nature MetabolismResource https://doi.org/10.1038/s42255-026-01495-yExtended Data Fig. 9 | Pre-processing workflow. Workflow of the HPAP methylation data pre-processing. ### Page 27 ### Page 28 ### Page 29