DNA Methylation–Dependent Restriction of Tyrosine Hydroxylase Contributes to Pancreatic β-Cell Heterogeneity

Pancreatic islets comprise several cell types, including the insulin-producing β-cells, which collectively regulate glucose homeostasis. β-Cells display molecular and functional heterogeneity, with unique β-cell subpopulations characterizing specific developmental and disease contexts to subserve robust systemic adaptation to changing physiological demands (1,2). All pancreatic endocrine cell types, including the different β-cell subpopulations, originate from a common progenitor. Spatiotemporal cues during development modify and restrict progenitor gene expression profiles to drive differentiation and generate cellular heterogeneity. However, we have a limited understanding of the mechanisms that orchestrate differential gene expression underlying β-cell heterogeneity.

Epigenetic mechanisms, such as DNA methylation, mediate context-specific changes in gene expression to direct the progressive refinement of cell fates during development (3). DNA methylation is the best-studied epigenetic module, with the de novo DNA methyltransferases Dnmt3a and Dnmt3b establishing new methylation patterns, whereas Dnmt1 maintains and propagates existing patterns through cell division (4). Modulation of DNA methylation patterns is known to dictate cell fate choices. In pancreas, maintenance methylation by Dnmt1 is essential for the survival of pancreatic progenitors during embryonic development (5) and for maintaining β-cell identity in postnatal life (6). Restriction of Dnmt1 expression also contributes to lineage commitment bias within the endocrine progenitor pool (7). Much less is known about the contribution of new methylation patterns in endocrine specification and lineage restriction. Dnmt3a is a part of the Nkx2.2 repressor complex that directs β-cell specification and is also essential for β-cell maturation in postnatal life (8,9). This suggests that de novo DNA methylation is important in refining cell fate choices during differentiation.

Here, we focus on a unique β-cell subpopulation as a paradigm to illustrate the contribution of DNA methylation in guiding endocrine lineage choices and heterogeneity. Pancreatic islets harbor a small number of β-cells that express the enzyme tyrosine hydroxylase (TH), which is a signature of sympathetic neurons. TH catalyzes the rate-limiting step in the synthesis of catecholamines that inhibit insulin secretion (10,11). Restriction of the TH⁺ β-cells within islets is essential for appropriate insulin secretion in mice, such that mouse strains with a larger number of TH⁺ β-cells have blunted insulin secretion (12). However, to our knowledge, there have been no systematic studies defining the molecular characteristics of these cells, nor do we know the mechanisms that restrict the TH⁺ β-cells in islets. We demonstrate the essential requirement of DNA methylation in establishing endocrine cell heterogeneity during differentiation, using TH⁺ β-cells as a model. We establish that the TH⁺ islet cells co-expressing insulin represent a bona fide replication-competent β-cell subtype with distinct molecular and metabolic characteristics and display species-specific developmental profiles in mice and humans. Using developmental stage–specific ablation of DNA methyltransferases, we demonstrate that DNA methylation establishes the restriction of TH expression in β-cells during the transition from endocrine progenitors to β-cells and maintains it subsequently. We show that loss of Th promoter methylation leads to increased TH expression and impaired β-cell identity in response to chronic overnutrition. Our data establish an essential role for DNA methylation patterning in endocrine lineage determination toward generation of β-cell heterogeneity.

Pancreatic sections from human fetal (n = 1), neonatal (n = 3), and adult nondiabetic donors (n = 5) were procured from brain-dead organ donors by the Network for Pancreatic Organ Donors with Diabetes, University of Florida at Gainesville. Human fetal pancreas (n = 1; 16 weeks postconception) was obtained through the Terminal Tissue Bank at the University of Southern California. The investigators were given 4-μm pancreatic sections from each donor (deidentified). Sample characteristics are described in Supplementary Table 1.

All animal experiments were performed according to protocols approved by the Institutional Animal Care and Use Committee at City of Hope, University of Southern California, and University of California, Los Angeles. Mice were maintained on a C57BL/6J background. Dnmt3a^fl/fl and Dnmt1^fl/fl mice have been described (13,14). We used Pdx1-Cre (15) and Ngn3-Cre (16) mice to ablate Dnmts in the pancreatic and endocrine progenitors, respectively, and Ins1-Cre^Thor (17) for β-cell–specific deletion. The Cre lines harbored Rosa26R-YFP or Rosa26R-mTmG lineage reporters (18). We used Pdx1-Cre:Rosa26R-YFP and heterozygous Ngn3-EGFP embryos (19) to sort pancreatic and endocrine progenitors, respectively. Transgenic mice expressing mouse Ins1 promoter (MIP-GFP) (20) were used to sort β-cells. Mice were fed ad libitum and kept under a 12-h light/dark cycle. Male and female mice were used for all studies except the high-fat diet (HFD) studies. For the HFD studies, 1.5-month-old male C57BL/6J mice (n = 8/group) were fed a control diet (CD; 10% calories from fat; Research Diets D12450B), or a HFD (55% calories from fat; Envigo TD.93075) for a short term (8 weeks) or long term (16 weeks).

A standard immunofluorescence protocol was used to detect various proteins in pancreatic sections, as previously described (21,22). Primary antibodies were diluted in the blocking solution, as noted in Supplementary Table 2. Donkey- and goat-derived secondary fluorescent antibodies (Jackson ImmunoResearch) were diluted 1:200. Antifade mounting medium with DAPI was used to label nuclei (Vector Laboratories). Slides were viewed using a Leica DM6000 or DM6 B microscope (Leica Microsystems) and imaged using OpenLab (Improvision) or Leica Application Suite X (Leica). Confocal imaging was done on a Zeiss LSM 800 with AiryScan using ZEN software. To quantify TH⁺ cells, all islets were imaged and the total numbers of insulin⁺, TH⁺, and TH⁺–insulin⁺ cells were manually counted. The data were expressed either as percentage of total TH⁺ cells, or β-cells and islet cells. Senescence associated β-galactosidase (SA-β-gal) staining was performed at pH 6.0 using the Cell Signaling kit (catalog 9860), followed by immunofluorescent staining for TH and insulin. Quantification of replication (and senescence) in different subpopulations was done as described by Rodnoi et al. (22). We counted a minimum of 1,500 cells for adults and 700 cells for neonates. β-Cell mass was measured per our published protocol (21). Briefly, pancreatic sections spanning every 80 μm were stained for insulin, counterstained with hematoxylin-eosin, and scanned using a Leica DM6 B microscope. Images were processed using ImageJ (National Institutes of Health), and β-cell mass was calculated as a ratio of insulin⁺ area to total pancreas area, multiplied by dry pancreas weight. Fluorescence Lifetime Imaging Microscopy (FLIM) was performed on an SP8 DIVE FALCON multiphoton microscope (Leica Microsystems) using a ×40/1.10 numerical aperture water-immersion objective (see Supplementary Methods).

Islets were isolated from mouse pancreas using perfusion with the Liberase enzyme blend (Sigma Aldrich), as described (21). For cell sorting, pancreatic buds from Pdx1-Cre:Rosa26R-YFP embryos, heterozygous Ngn3-EGFP embryos, or the MIP-GFP transgenic pups were harvested and digested into single cells using TrypLE Express (ThermoFisher). Cells were sorted for green fluorescent protein imaging using a BD FACS Aria II to an average percentage purity of 85–95%, with wild-type cells being the negative control for FACS gating (9).

Intraperitoneal glucose tolerance tests were performed as previously described (21,22). Briefly, mice were fasted overnight and were injected intraperitoneally with 2 mg/kg d-glucose in sterile 1× PBS. Blood glucose levels were measured immediately before injection (0 min) and then at 15, 30, 60, 90, and 120 min after injection. Glucose-stimulated insulin secretion (GSIS) was assessed using a static incubation assay, per our previous studies (9). For TH inhibition, freshly isolated islets were cultured for 24 h in the presence of 10 μmol/L α-methyl-para-tyrosine (AMPT; Sigma-Aldrich) or vehicle (1× PBS) and used for GSIS. Ins in supernatants and islet extracts was measured by ELISA (Mercodia).

Chromatin immunoprecipitation (ChIP) was carried out on sorted cells using a low-cell-number protocol (6,9), using 2.5 μg of antibody. The antibodies and primers used for ChIP are listed in Supplementary Tables 3 and 4. Bisulfite conversion of DNA from sorted cells and islets and sequencing of the converted DNA were performed according to our published methods (6,9) using primers described in Supplementary Table 5. We analyzed 20–25 independent clones per group; each experiment was repeated in biological triplicates. RNA extraction and quantitative real-time RT-PCR was performed as published (22) using primers listed in Supplementary Table 6.

Single-cell RNA-sequencing (scRNA-seq) analysis was performed on publicly available scRNA-seq data sets (n = 3) from islets of mice fed a low-fat diet (accession no. GSE12512; ref. 23), as described in Supplementary Methods.

All data are expressed as mean ± SE. Mean and SEM values were calculated from at least triplicates (biological) of a representative experiment. The statistical significance of differences was measured by an unpaired Student t test for experiments with two groups and a continuous outcome; a one-way ANOVA with Šídák, Bonferroni, or Fisher least significant difference post hoc tests was used for experiments with repeat measures. P < 0.05 indicated statistical significance. In figures, asterisks indicate P values as follows: *P < 0.05, **P < 0.01, ***P < 0.005, and ****P < 0.001.

All data are available in the article or supplementary materials. Data sets and resources are available from the corresponding author upon reasonable request.

Islet cells that co-express the enzyme TH have been described before (24), though their precise molecular identity remains obscure. TH expressing islet cells were readily detectable in adult wild-type mice, the majority of which co-expressed insulin (Fig. 1A–C). These β-cells had variable TH levels and frequently appeared as clusters (Fig. 1B). There was no overlap between TH and glucagon in adult islets, whereas TH was present in few adult δ-cells (Supplementary Fig. 1A–C). TH expression in β-cells was first observed at embryonic day (E) 15.5, prior to which TH was restricted to α-cells (Supplementary Fig 1D). We confirmed the endocrine and β-cell lineage identities of the islet-resident TH-expressing cells by using endocrine-specific (Ngn3-Cre:R26R-mTmG) and β-cell–specific (Ins1-Cre:R26R-YFP) lineage reporters and by colocalization of TH with the endocrine marker chromogranin A (Fig. 1D, Supplementary Fig. 1E and F). TH⁺ β-cells expressed Glut2, maturity marker Ucn3, and β-cell transcription factors Pdx1, Nkx6.1, MafA, and NeuroD, similar to other β-cells (Fig. 1E–G, Supplementary Fig. 1G). Because TH also marks sympathetic neurons that are intimately linked to islet vasculature (25), we examined the interaction of TH⁺ β-cells with sympathetic neurons and vasculature. These cells had greater contact with the Tuj1⁺ sympathetic fibers and PECAM1⁺ capillaries, compared with TH⁻ β-cells (Fig. 1H–J, Supplementary Fig. 1H–L, Supplementary Movies 1 and 2). Prior work showed an inverse correlation between TH⁺ β-cell number and islet function (12). Accordingly, TH inhibition in mouse islets using AMPT led to improved GSIS (Supplementary Fig. 1M).

To determine if TH⁺ β-cells exist in humans, we mined two publicly available human islet scRNA-seq data analyses (26,27) and found that no more than 1% of adult human β-cells express TH at very low levels (Fig. 2A, Supplementary Fig. 2A–E). A histological survey of adult human pancreases did not show any TH⁺ β-cells in the samples analyzed (0% of 2,000 β-cells per sample, n = 3) (Fig. 2B and D), regardless of the islet size, confirming their rarity in adult humans (Fig. 2B). Next, we examined TH expression in fetal and neonatal human pancreases. Although most islet structures lacked TH, small endocrine-cell clusters frequently harbored brightly stained TH⁺ β-cells (Fig. 2C and D, Supplementary Fig. 2F–I). To confirm the developmental regulation of TH, we queried the β-Cell Hub and Broad Institute Single Cell Portal databases for single-cell transcriptomic comparison of the immature and mature human β-cell states modeled by stem-cell derived β-like (sc-β-like; early/immature and late/mature) cells and human islet origin β-cells (28,29). Our queries showed higher TH expression in the sc-β-like cells compared with islet-derived β-cells (Fig. 2E and F) and revealed that Rho-associated protein kinase (ROCK) inhibitor treatment, which promotes the maturation of sc-β-like cells (30), led to a fivefold reduction in TH mRNA (Fig. 2F). These data suggest that reduced TH levels mark the human β-cell maturation process.

To define the unique molecular features of TH⁺ β-cells, we meta-analyzed publicly available islet scRNA-seq data from adult C57BL/6J mice (23). All samples were integrated into one data set, and various cell types were identified by their standard gene signatures (Fig. 3A, Supplementary Fig. 3A and B). β-Cells were grouped into Th⁺ and Th⁻ categories; three of four β-cell clusters (clusters 0, 2, and 3) were enriched in Th, with ∼5% cells expressing high levels (Fig. 3B). Next, we surveyed for differentially expressed genes (DEGs) between the Th⁺ and Th⁻ β-cells and identified 130 such genes (Fig. 3C). The top enriched genes in the Th⁺ group were Mt1, Ftl1, Selenow, Mt2, and mt-Nd3, whereas Fkbp11, Manf, Pcsk1, Txnip, Pdia6, and Ero1lb represented top downregulated genes. Gene-set enrichment analysis of these DEGs showed enrichment of pathways associated with protein synthesis, ribosomes, energy production, and ATP metabolism and downregulation of endoplasmic reticulum (ER) protein transport and ER stress response pathways (Fig. 3D, Supplementary Fig. 3C), hallmarks of ER stress and high oxidative phosphorylation (OxPhos).

To validate these findings, we first determined the metabolic characteristic of the TH⁺ β-cells. We leveraged a FLIM method we developed to measure NADH levels in fixed tissue as a proxy of cellular metabolic status (31). FLIM measures the lifetime of excited NADH; unbound NADH exhibits short lifetimes and is a byproduct of glycolysis, whereas enzyme-bound NADH exhibits an ∼10× longer lifetime and is a substrate of OxPhos (32). We used immunofluorescence to identify TH⁺ and TH⁻ β-cells and collected their fluorescence intensity and NADH autofluorescent lifetimes. Comparison of average lifetimes for TH⁺ and TH⁻ β-cells (n = 8 islets per pancreas; n = 4 mice) showed a higher lifetime for TH⁺ β-cells, indicating more oxidative metabolism (Fig. 3E, Supplementary Fig. 3D). We plotted the NADH intensity of each pixel of a TH⁺ and TH⁻ β-cell in an islet onto a phasor plot and determined the coordinates of the mode of NADH intensity for individual islets. We then plotted the average of those modes of all islets per animal on a G versus S axis (i.e., horizontal and vertical components, respectively, of lifetime measurements in arbitrary units) (Fig. 3F). The TH⁺ β-cells clustered together to the left, implying more OxPhos (Fig. 3F). Next, we tested islet levels of Ero1lb-encoded endoplasmic reticulum oxidoreductase 1 beta (Ero1b), which catalyzes disulfide bond formation in the ER, is critical for insulin biogenesis, and protects from ER stress (33,34). Immunostaining confirmed scRNA-seq findings that, indeed, Ero1b was less abundant in TH⁺ β-cells (Fig. 3G).

It was previously proposed that TH⁺ β-cells represent a senescent population that does not replicate (10). Replication is the primary mechanism for the growth and maintenance of β-cell mass in postnatal life, with rapid replication in neonatal life followed by gradual, age-dependent decline (21,35). Therefore, we examined the growth and replication of TH⁺ β-cells in postnatal life. To establish their growth profile, we quantified TH⁺ β-cells in wild-type murine pancreas at various ages. The percentage of TH⁺ β-cells increased steadily during the growth phase (postnatal day 2 [P2]–2.5 months) and then stabilized (Fig. 4A). Immunostaining for replication markers Ki67 and pHH3 revealed that the TH⁺ β-cells do replicate in neonatal life (Fig. 4B–E, Supplementary Fig. 4A and B). The percentage of replicating TH⁺ β-cells was highly variable, likely due to their small number in the islets. In adult mice in which β-cells are typically quiescent but can replicate in response to increased insulin demand (36), several TH⁺ β-cells were marked by the licensing factor Mcm2, which indicates replication competence (Supplementary Fig. 4C and D). The ability of TH⁺ β-cells to replicate suggested these cells may not be senescent. Quantification of key senescence markers such as the presence of p21, p16, γ-H2AX, and SA-β-gal, and loss of LaminB1 (37,38) revealed that TH⁺ β-cells are nonsenescent (Fig. 4F–M, Supplementary Fig. 4E–R). Moreover, none of the key senescence pathway genes were upregulated in the TH⁺ β-cells in our scRNA-seq meta-analysis (Supplementary Fig. 4S). These data show that TH⁺ β-cells can replicate and are not senescent under homeostatic conditions.

DNA methylation patterning is a key mechanism that directs the establishment of pancreatic endocrine identity (6,8). We hypothesized that promoter methylation restricts TH expression to a select subset of β-cells, and we examined Th promoter methylation at different stages of endocrine differentiation. We performed bisulfite sequencing for the Th promoter in pancreatic progenitors (Pdx1⁺), endocrine progenitors (Neurogenin3 [Ngn3⁺]), and β-cells (insulin⁺) isolated from pancreatic tissue using lineage reporter mice at E11.5 (Pdx1-Cre:R26R-YFP), E13.5 (Ngn3-Cre:R26R-YFP), and E18.5 (Ins1-Cre^Thor:R26R-YFP), respectively. We found stage-specific, differential DNA methylation at the −2K region of the Th promoter, with the promoter being hypomethylated in the pancreatic and endocrine progenitors and hypermethylated in β-cells (Fig. 5A). No differences in DNA methylation were found in the other CG-rich regions in the promoter (Supplementary Fig. 5A and B).

This differential methylation was not due to stage-specific expression of Dnmt3a (the only de novo Dnmt present in endocrine lineage) (9), because Dnmt3a was expressed in the progenitors as well as β-cells (Fig. 5B–D). ChIP for Dnmt3a at the −2K region of the Th promoter revealed higher enrichment of Dnmt3a in the endocrine progenitors compared with pancreatic progenitors and β-cells (Fig. 5E). Expression analysis revealed very little overlap of Pdx1 and TH at E12.5 except for a rare few TH⁺ cells marked by very low Pdx1, whereas none of the Ngn3⁺ endocrine progenitors expressed TH at E13.5 (Supplementary Fig. 5C and D). Because TH was not expressed in the pancreatic or endocrine progenitors, the lack of promoter DNA methylation at these stages led us to determine the mechanism of Th repression at these stages. Histone H3–lysine 9 trimethylation (H3K9me3) creates a reversible repressive chromatin state prior to the establishment of the more stable DNA methylation (39). ChIP for H3K9me3 confirmed its enrichment at the −2K region of Th promoter in sorted pancreatic and endocrine progenitors (Supplementary Fig. 5E). These data show the stepwise establishment of repressive marks at the Th promoter through β-cell differentiation with DNA methylation as the final step and suggest a role for DNA methylation in restricting TH expression in differentiated β-cells.

To establish a direct regulatory role of DNA methylation in restricting Th expression in β-cells during differentiation, we generated mice harboring loss of Dnmt3a in pancreatic, endocrine, and β-cell lineages and examined TH expression in the pancreas. Mice with Dnmt3a ablation in the pancreatic and endocrine progenitor lineages (Pdx1-Cre:Dnmt3a^fl/fl [3aPancKO] and Ngn3-Cre:Dnmt3a^fl/fl [3aEndoKO], respectively) showed an ∼16-fold increase in the number of TH⁺ β-cells. Loss of Dnmt3a in the β-cell lineage (Ins1-Cre^Thor:Dnmt3a^fl/fl [3aBetaKO]) did not alter the number of TH⁺ β-cells (Fig. 6A–F, Supplementary Fig. 6A–C). We confirmed the efficiency of recombination and Dnmt3a ablation by Cre-driven fluorescent protein expression and Dnmt3a immunostaining (Supplementary Fig. 6A–H). We chose neonatal stages to confirm Dnmt3a ablation because its levels decline dramatically after weaning (9). The changes in β-cell identity upon Dnmt3a ablation did not involve any changes in β-cell mass but were accompanied by blunting of GSIS (Supplementary Fig. 6I and J). Additionally, Dnmt3a ablation in the progenitors led to loss of TH restriction in δ- but not α-cells (Supplementary Fig. 6K–N). In agreement with the changes in TH expression, the −2K region of the Th promoter was hypomethylated in islets from 3aPancKO and 3aEndoKO mice but not in 3aBetaKO islets (Fig. 6D–I). Thus, de novo methylation by Dnmt3a is required to restrict Th expression in differentiated β-cells.

Dnmt3a dependent de novo DNA methylation is essential for β-cell maturation (9). Thus, we tested if the DNA methylation–dependent restriction of Th in β-cells is linked to functional maturation. We compared TH⁺ β-cells in 3aPancKO and control pancreas specimens at P0 and P5 before functional maturation. The increase in the number of TH⁺ β-cells in the 3aPancKO was very pronounced at P5 and was evident even immediately after birth (Fig. 7A and E, Supplementary Fig. 7A and B). Dnmt3a ablation in the endocrine lineage showed a similar de-repression of TH in β-cells at P0 (Fig. 7B and F). To determine the developmental timeline of TH de-repression in the absence of Dnmt3a, we examined 3aPancKO, 3aEndoKO, and control tissue during late endocrine differentiation (E16.5) and observed an increase in TH⁺ β-cells in the two KO models, even at this stage before functional maturation (Supplementary Fig. 7C–F).

Maintenance of the existing DNA methylation patterns through replication is essential for the preservation of β- versus α-cell identity in postnatal life (6). To determine if maintenance of promoter methylation is required for sustained Th restriction in β-cells, we generated mice lacking the maintenance methyltransferase Dnmt1 in endocrine and β-cell lineages (Ngn3-Cre:Dnmt1^fl/fl [1EndoKO], and RIP-Cre^Herr:Dnmt1^fl/fl [1RCBetaKO]). Because Dnmt1 depletion in β-cells results in their conversion to α-cells in late adult life (6), we focused our analysis on neonates and young adults (P5 to 1 month) before the onset of trans-differentiation to α-cells. We observed an increased number of TH⁺ β-cells in the 1EndoKO and 1RCBetaKO mice at P5 (Fig. 7C and D), the increase being much higher in 1EndoKO, congruent with the timeline of Th promoter methylation (Fig. 7G and H). This increase was accompanied by loss of Th promoter methylation in β-cells harvested from 1-month-old 1RCBetaKO mice (Fig. 7I–K). These data show that propagation of Th promoter methylation is required to maintain β-cell heterogeneity.

Given the essential requirement of maintenance DNA methylation in restricting TH in β-cells, we asked if conditions that warrant β-cell expansion alter the proportion of TH⁺ β-cells. A prior study had reported increased TH⁺ β-cell numbers in genetically obese ob/ob mice that undergo β-cell expansion in young age. However, the increased number of TH⁺ β-cells only occurred in the older ob/ob mice (10). Because there is no age-dependent increase in TH⁺ β-cells in wild-type mice (Fig. 4A), we hypothesized that chronic increase in insulin demand because of insulin resistance was an underlying cause. To test this, young adult, wild-type mice (7 weeks old) were either exposed to a short-term (8 weeks) or long-term (16 weeks) HFD regimen or maintained on a CD. The two regimens are designed to model successful β-cell adaptation to increased demand for insulin (short-term HFD) or the subsequent decompensation as β-cell de-differentiation and failure ensue (40). As expected, both short- and long-term HFD led to increased body weight and impaired glucose tolerance (Supplementary Fig. 8A–D). Although end point analysis showed increased islet size and β-cell mass in mice fed the short- and long-term HFDs, only the long-term HFD regimen resulted in an increase in TH⁺ β-cells (Fig. 8A–F) concomitant with impaired islet function (Supplementary Fig. 8E) that hallmarks β-cell decompensation in long-term HFD (41). This finding suggests that chronic β-cell workload due to persistently high insulin demand could disrupt the epigenetic restriction of TH in β-cells. Thus, we performed bisulfite sequencing analysis of Th promoter in islets from mice fed a short- and long-term HFD versus appropriate controls. We observed loss of promoter DNA methylation corresponding to increased Th mRNA expression in islets from mice fed the long-term but not short-term HFD (Fig. 8G, Supplementary Fig. 8F and G), indicating a failure to maintain DNA methylation upon chronically high insulin demand. This suggests that maintenance of epigenetic patterns is essential to preserve β-cell identity during adaptation and fails during β-cell decompensation.

Pancreatic β-cells share several features with neurons, including shared developmental transcriptional programs (42). Islet cells contain neurotransmitters that regulate insulin secretion and enzymes such as TH that regulate their synthesis (43,44). We observed TH expression in a small subpopulation of adult β-cells and rare δ-cells, suggesting that neuron-like features underlie heterogeneity of many endocrine cell types. Early endocrine precursors express specific neuronal markers whose expression gets restricted as endocrine identities are refined. Accordingly, it has been postulated that the TH⁺ β-cells may indicate a developmentally immature precursor (45), which appears to be the case in humans. Presence of TH in fetal and neonatal human pancreases and sc-β-cells, but not adult human pancreases, highlights a major developmental difference in TH regulation compared with mice. Recent studies have identified β-cells harboring serotonergic markers in human sc-β-cell differentiation (46). This finding suggests that expression of neuron-like features characterize sc-β-cells, revealing a bottleneck in generating mature human sc-β-cells.

We show that DNA methylation is essential for regulating endocrine identity and heterogeneity in homeostatic and stress conditions. The specification and maintenance of cell fates require the precise temporal orchestration of epigenetic patterns to regulate stage-specific transcriptional programs (3). De novo DNA methylation is essential for β-cell lineage specification and functional maturation in fetal and early postnatal life. Maintenance of these methylation patterns by Dnmt1 allows the continued restriction of α-cell fate to preserve β-cell identity in postnatal life (6,8,9). We establish that the de novo DNA methylation–dependent restriction of TH expression in β-cells occurs during lineage specification prior to their functional maturation. Heterogeneity of Dnmt expression within the endocrine progenitor pool dictates their lineage commitment bias (7); similar mechanisms might be involved in limiting TH expression to a select few β-cells. The ablation of Dnmt1 in endocrine progenitors and β-cells results in the ectopic TH expression in β-cells prior to their trans-differentiation to α-cells. These temporal changes in β-cell identity upon the loss of maintenance methylation suggest differences in the stringency of restriction between cell fates with different degrees of relatedness. It is likely that the epigenetic barriers between different β-cell subtypes are more plastic than those between β- and α-cells. DNA methylation plays distinct regulatory roles in different endocrine lineages, whereas loss of Dnmt1 in β-cells results in their conversion to α-cells; Dnmt1 ablation in α- or δ-cells does not convert them to β-cells (6,47,48). The dysregulation of Th in β- and δ-cells, but not α-cells, upon Dnmt ablation further highlights the lineage specificity of DNA methylation regulation.

It has been suggested that TH⁺ islet cells may be senescent (10). We show that TH⁺ β-cells do replicate during the neonatal growth phase and do not harbor senescence markers under homeostatic conditions. Instead, these cells bear signatures of overactive protein synthesis, poor ER stress response, and higher OxPhos, hallmarks of β-cells predisposed to ER and oxidative stress (49). We also show that conditions of β-cell decompensation can disrupt the epigenetic repression of the Th promoter and alter β-cell identity, marked by an increase in TH⁺ β-cell numbers. These data explain prior observations of increased TH⁺ β-cell numbers in older, genetically obese ob/ob mice without any accompanying replication (10). The older ob/ob mice represent a β-cell decompensation state in response to chronically elevated insulin demand, where β-cell replication does not occur (40). Accumulation of these cells during β-cell decompensation, along with the inverse correlation between TH⁺ β-cell abundance and islet function (12), suggests the possibility of β-cells de-differentiating into a TH⁺ state and acquiring a dysfunctional phenotype. The large-scale changes in human islet methylome associated with β-cell identity and function genes in type 2 diabetes strengthen this idea (50). Our data suggest that embryonic DNA-methylation patterning may affect postnatal β-cell identity, which is noteworthy, given that over- or undernutrition during development can predispose to risk of β-cell failure (4). Future work will identify the specific aspects of postnatal β-cell phenotype that rely upon embryonic epigenetic patterning and will elucidate the significance of such a paradigm.

Acknowledgments. The authors thank Alexander Ham, Giovanni Hernandez-De La Peña, and Pope Rodnoi (S.D. Laboratory) for technical support; Dr. Rachel Steward and Family Planning Associates for the (fetal) tissue donation (Children’s Hospital Los Angeles tissue repository); and Drs. Jason Junge and Peiyu Wang (University of Southern California) for expertise and consultation in FLIM studies. The authors thank Dr. David Ron (University of Cambridge) and Dr. Mark Huising (University of California, Davis) for the gifts of Ero1b and Ucn3 antibodies, respectively.

Funding. This work was supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK; grants R01DK120523 to S.D., R01DK119590 to H.-P.S., and 1R01DK133504 to S.K.G.), the Human Islet Research Network (a New Investigator Award to S.D. via NIDDK grant UC4DK104162), the Homer & Gloria Harvey Family Foundation, the Paul Lester Foundation, and the Saban Research Institute (to S.K.G.). Work in the S.D. and H.-P.S. laboratories is also supported by City of Hope (start-up support to S.D. and H.-P.S.), the Wanek Family Foundation to Cure Type 1 Diabetes (S.D. and H.-P.S.), and theLions Club Initiative (S.D.).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. S.K.G. and S.D. conceived and planned the study. N.P., J.K.W., J.C., M.S.R., S.K.G., and S.D. performed the experiments. N.P., J.K.W., S.B., J.C., M.S.R., A.E.B., X.W., S.K.G., and S.D. performed the analyses. N.P., J.K.W., S.B., J.C., A.E.B., X.W., H.-P.S., S.K.G., and S.D. interpreted data. N.P., S.B., X.W., H.-P.S., S.K.G., and S.D. wrote the manuscript. H.-P.S., S.K.G., and S.D. acquired funding. S.D. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Part of this work was presented at the American Diabetes Association 79th Scientific Sessions, San Francisco, CA, 7–11 June 2019, and the American Diabetes Association 82nd Scientific Sessions, New Orleans, LA, 3–7 June 2022.