Transcriptional and functional cellular specialization has been described for insulin-secreting β-cells of the endocrine pancreas. However, it is not clear whether β-cell heterogeneity is stable or reflects dynamic cellular states. We investigated the temporal kinetics of endogenous insulin gene activity using live cell imaging, with complementary experiments using FACS and single-cell RNA sequencing, in β-cells from Ins2GFP knockin mice. In vivo staining and FACS analysis of islets from Ins2GFP mice confirmed that at a given moment, ∼25% of β-cells exhibited significantly higher activity at the evolutionarily conserved insulin gene, Ins2. Live cell imaging over days captured Ins2 gene activity dynamics in single β-cells. Autocorrelation analysis revealed a subset of oscillating cells, with mean oscillation periods of 17 h. Increased glucose concentrations stimulated more cells to oscillate and resulted in higher average Ins2 gene activity per cell. Single-cell RNA sequencing showed that Ins2(GFP)HIGH β-cells were enriched for markers of β-cell maturity. Ins2(GFP)HIGH β-cells were also significantly less viable at all glucose concentrations and in the context of endoplasmic reticulum stress. Collectively, our results demonstrate that the heterogeneity of insulin production, observed in mouse and human β-cells, can be accounted for by dynamic states of insulin gene activity.

Pancreatic β-cells in the islets of Langerhans are the only source of circulating insulin, a conserved and essential hormone required for nutrient homeostasis and life (1). Insulin production is demanding, as insulin mRNA can account for one-half of all β-cell mRNA. Insulin’s synthesis, folding, and processing require semispecialized transcription factors, enzymes, and cellular conditions (2,3). However, not all β-cells appear to be the same. Indeed, functional β-cell heterogeneity is well established (4), including cellular specialization for islet cell synchronization, insulin secretion, insulin production, and marker gene expression (511). The in vivo existence of extreme β-cells, defined as having more than twofold Ins2 mRNA than the median expression, was revealed with single-molecule fluorescence in situ hybridization (4,12). Single-cell RNA sequencing has shown that human β-cells also express INS over a similarly wide range (8). However, it remains unclear whether this variation is the hallmark of distinct stable populations of β-cells or is indicative of transitions between more labile β-cell states.

To date, the majority of islet cell subpopulations have been defined by single time-point snapshots, making it impossible to know to what extent observed β-cell heterogeneity represents distinct cell fates or variations in cell states. This information is essential for interpreting existing and future data. Using a dual Ins1 and Pdx1 promoter reporter construct and live cell imaging, we previously demonstrated that mouse and human β-cells can transition between less and more differentiated states over a timescale of ∼24 h (1315). However, the artificial promoter constructs in these early studies may not reflect endogenous gene activity, leaving open the question of whether endogenous insulin gene activity is similarly dynamic.

In this study, we measured endogenous insulin gene activity using an Ins2GFP knockin/knockout mouse line in which the coding sequencing of the evolutionarily conserved Ins2 gene has been replaced with green fluorescent protein (GFP) (16). Live cell imaging of dispersed cells from these mice revealed that GFP fluorescence changed over time in a subset of cells, suggesting that variation in Ins2 levels can result from dynamic transcriptional activity at the Ins2 gene locus rather than stable heterogeneity. Single-cell RNA sequencing was used to characterize the Ins2(GFP)HIGH cellular state in an unbiased way, revealing increased markers of β-cell maturity as well as alterations in protein synthesis machinery and cellular stress response networks. Pancreatic β-cells in the Ins2(GFP)HIGH cellular state were also more fragile across a range of stress conditions. To our knowledge, our observations are the first to define the temporal kinetics of endogenous insulin gene activity, which represents a previously uncharacterized form of β-cell plasticity. Understanding the dynamics of insulin production has relevance for understanding the pathobiology of diabetes and for regenerative therapy research (17).

Animals and In Vivo Physiology

Animals were housed and studied in a modified barrier facility using protocols approved by the University of British Columbia animal care and use committee in accordance with international guidelines. Ins2WT/GFP knockin mice were obtained from Shouhong Xuan (16). These mice were crossed with transgenic mice where the Ins1 promoter drives an mCherry:H2B fluorescent fusion protein (18). Glucose tolerance and insulin secretion were assessed in both male and female mice (aged 12–14 weeks) injected intraperitoneally with 2 g/kg (20%) glucose after a 5-h fast. Insulin from in vivo samples was measured using ELISA kits (Alpco, Salem, NH). Insulin tolerance was assessed after injection of 0.75 units insulin/kg body weight after a 5-h fast.

Immunostaining and RT-PCR

Pancreata from PBS-perfused mice were harvested and fixed in 4% paraformaldehyde for 24 h before being washed and stored in 70% ethanol and paraffin embedding. Pancreatic sections (5 μm) were taken from at least three different regions of the pancreas 100 μm apart. Sections were deparaffinized, hydrated with decreasing concentrations of ethanol, and rinsed with PBS. Sections were subjected to 15 min of heat-induced epitope retrieval at 95°C using a 10 mmol/L citrate buffer (pH 6.0). Sections were blocked and then incubated with primary antibodies overnight in a humid chamber at 4°C. A list of primary antibodies can be found in Table 1. Primary antibodies were visualized after incubation with secondary antibodies conjugated to Alexa Fluor 488, 555, 594, or 647 as required (1:1,000; Invitrogen). Counterstaining was done using VECTASHEILD mounting media with DAPI (H-1200). Images were taken on a Zeiss Axiovert 200M microscope using a 20× air (numeric aperture [NA] 0.75), 40× oil (NA 1.3), and/or 100× oil (NA 1.45) objective and analyzed using SlideBook software (Intelligent Imaging Innovations, Denver, CO). For quantification of immunofluorescence, we used the segment masking function of SlideBook, generating a GFP high mask and a GFP low mask and comparing the two (Fig. 8C). For human pancreas sections stained for C-peptide and PDX-1, we obtained images from the Human Pancreas Analysis Program database (19) and performed the analysis using custom scripts in CellProfiler. Real-time RT-PCR was conducted as described previously (1). A list of primers used can be found in Supplementary Table 2.

Islet Isolation, Dissociation, and Culture

Pancreatic islets were isolated using collagenase, filtered, and handpicked as previously described (15). Islets were cultured overnight (37°C in 5% CO2) in RPMI medium (Thermo Fisher Scientific) with 11 mmol/L glucose (Sigma-Aldrich, St. Louis, MO), 100 units/mL penicillin, 100 μg/mL streptomycin (Thermo Fisher Scientific), and 10% v/v FBS (Thermo Fisher Scientific). For islet dissociation in preparation for imaging experiments, islets were washed four times with minimum essential medium (Corning) and immersed for 5 min in 0.05% trypsin at 37°C. Cells were then resuspended in RPMI medium and given 24 h of rest (37°C in 5% CO2) before imaging. For experiments involving various culture conditions, cells were incubated overnight in serum-free media for serum-free and serum-starved conditions and media with 10% FBS for serum conditions. Before imaging, cells were given a media change and stained with Hoechst cell nuclei marker and propidium iodide cell death marker for 2 h. Serum-free condition cells were given media with no serum, while serum-starved (rescue) and serum conditions were given media with 10% FBS. Cells were then treated with various glucose concentrations immediately before imaging.

FACS

Pancreatic islets were dispersed using 0.05% trypsin and resuspended in 1× PBS with 0.005% FBS. Dispersed islets were then filtered into 5 mL polypropylene tubes. FACS was conducted on a Cytopeia inFlux cell sorter (Becton Dickinson, Franklin Lakes, NJ) at the Life Sciences Institute core facility. Cells were excited with a 488-nm laser (530/40 emission) and a 561-nm laser (610/20 emission).

Live Cell Imaging

To define the incidence and kinetics of the transitions in Ins2 gene activity, we cultured dispersed islet cells on 96-well glass bottom plates and imaged them every 5 or 30 min for up to 96 h through a 40× air objective using an ImageXpress Micro environmentally controlled, robotic imaging system (Molecular Devices) and a 300-W xenon lamp (20). Cells were exposed to 359-nm light for 110 ms, 491-nm light for 15 ms, 561-nm light for 75 ms.

Our methods for live cell imaging of islet cell survival have been published (20). Briefly, pancreatic islet cells were dissociated and cultured in RPMI media with 10% FBS (Gibco, Thermo Fisher Scientific, Gaithersburg, MD), and penicillin/streptomycin on 96-well glass bottom plates for 48 h. Dispersed islet cells were then exposed to different glucose concentrations as well as to thapsigargin (Sigma-Aldrich). Islet cell death was measured by propidium iodide incorporation. Images were taken every 30 min for up to 48 h, as described earlier. Propidium iodide incorporation was traced throughout the time period, and the area under the curve (AUC) was measured.

Live Cell Imaging Analysis

Analysis of Ins2GFP:Ins1-mCherry cells was done using ImageXpress software and custom R scripts (21). Hierarchical clustering with eight cellular behavior traits as features was done, and a heat map was constructed using the pheatmap function in R. Traits were calculated using R functions and formulas, including mean (average lifetime fluorescent intensity), sharp (calculates number of sharp peaks in a cell’s lifetime), AUC, full width at half maximum (FWHM), and oscillation (mean average deviation). Principal component analysis (PCA) was done using the same eight cellular behavior traits as variables. We performed model-based clustering using the mclust function in R and projected the cells onto the first two principal components (PCs).

Analysis of Ins2GFP cells was done using MetaXpress software and custom R scripts. Hierarchical clustering was done with various cellular behavior traits as features, and a heat map was constructed using the pheatmap function in R. Features include autocorrelation function (ACF) time (calculates when the ACF equals 0 in a time series), FWHM, difference (calculates the range of fluorescent intensity change), deviation (mean average deviation), mean (average lifetime relative fluorescence intensity), and sharp (a function that detects sharp peaks). PCA was done using the same cellular behavior traits as variables. We performed k-means clustering using the k-means function in R and projected the cells onto the first two PCs. Categorization of cells into high- and low-GFP–expressing states in live cell imaging and single-cell RNA sequencing data was done using model-based clustering (22).

We noticed in our imaging experiments a slow decrease in GFP fluorescence over the course of 48 h (Supplementary Fig. 4A). An experiment with 2-h imaging intervals and fewer wells showed that decreasing the total exposure time over 48 h of imaging did not remove this trend (Supplementary Fig. 4B). Detrending and normalizing to control for this trend were done by comparing various treatments to the basal condition (10 mmol/L glucose) and normalizing to the average fluorescence intensity of the first 2 h of imaging to correct for a possible linear artifact and variations in the initial hours of imaging, using the following equation:
formula

Autocorrelation analysis of single-mutant Ins2WT/GFP was done using the acf function in R (23). To calculate the oscillation periods of GFP, we first selected for cells with at least one oscillation. We calculated the ACF of all cells for all time points and filtered for cells with at least two significantly positive peaks (ACF >0.05) using the findpeaks function in R (21), indicating an oscillation. We calculated oscillation periods by finding the distance of the two positive peaks to get a full wavelength period. Oscillation periods of oscillating cells with more than one oscillation were then binned in 4-h intervals and plotted as a bar chart using R scripts.

Single-Cell Transcriptomics

Islets were isolated, handpicked, dispersed, and subjected to FACS as described above into Ins2(GFP)HIGH and Ins2(GFP)LOW cells. Single-cell suspension was loaded onto a Chromium Controller (10× Genomics, Pleasanton, CA) for capture in droplet emulsion. The 60-week-old mouse islet cell libraries were prepared using the Chromium Single Cell 3′ Reagent v2 Chemistry kit (10× Genomics), and the standard protocol was followed for all steps. The 8-week-old mouse islet cell libraries were prepared using the Chromium Single Cell 3′ Reagent v3 Chemistry kit (10× Genomics). Libraries were then sequenced on a NextSeq 500 (Illumina). Cell Ranger 3.0 (10× Genomics) was used to process raw FASTQ data. The results were visualized using Loupe Browser software (10× Genomics). Cells with high mitochondria gene ratios (>0.1), low numbers of genes (<500), and low unique molecular identifier counts (<1,000) were excluded before analysis. This was done for both 60-week-old and 8-week-old mouse data.

Analyses were performed using Seurat package version 4.0.4 in R (24). Data were normalized, scaled, and analyzed with PCA. We used the generated PCs to build a uniform manifold approximation and projection (UMAP) to identify populations of cells using known cellular markers (Ins1 for β-cells, glucagon [Gcg] for α-cells, somatostatin [Sst] for δ-cells, pancreatic polypeptide [Ppy] for Ppy/γ-cells, Ghrl for ε-cells, and Pcam1 for endothelial cells). Cells expressing Ins1 were selected, and the rest were used as a base expression level for differential expression analysis using the FindMarkers function in Seurat. Differential expression analysis between the subsets of cells was performed using the Wilcoxon rank sum test option in Seurat. Differential expression analysis over GFP gradient was performed by first calculating correlation functions for each gene with GFP, then selecting for the genes that had the highest values. Top hits were used for pathway enrichment analyses and heat maps. Pathway enrichment analysis was performed using custom R scripts with data from the Gene Ontology Resource database. RNA velocity analysis was performed by mapping reads to exons and introns with velocyto in Python, while RNA velocity vectors were inferred and overlaid onto UMAPs using the dynamical mode in scVelo.

Statistical Analysis

Data are shown as mean ± SEM unless otherwise indicated. Differences between two groups were evaluated using Student t test and among more than two groups using ANOVA and Kruskal-Wallis one-way ANOVA. Data were analyzed using custom R scripts or GraphPad Prism software (GraphPad Software, San Diego, CA).

Data Resource and Availability

The data sets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request. Key resources are listed in Table 1. Additional information can be found in Supplementary Table 1.

In Vivo Heterogeneity of Insulin Content in Human β-Cell and Insulin Gene Activity in Ins2GFP Mice

Heterogeneity of insulin production is an established phenomenon. We started our study by investigating human pancreata from an online database (19) that were stained with antibodies to C-peptide and PDX-1, a key transcription factor for β-cell survival and function (25,26). As expected, based on single-cell RNA sequencing data (8) and previous single-cell imaging of human β-cells (27), we identified β-cells with both high and low C-peptide and PDX-1 protein levels in human pancreas (Fig. 1A–C). We have previously shown limited correlations between insulin content and PDX-1 immunofluorescence or nuclear localization in human β-cells (27). Analysis of single-cell RNA sequencing data previously compiled by our laboratory (2) showed that there was a bimodal distribution in insulin gene expression in β-cells from donors with and without type 2 diabetes, with a left shift in high insulin–expressing cells in patients with type 2 diabetes (Fig. 1D).

To study insulin gene activity in living cells, we examined islets from mice in which the coding sequencing of the evolutionarily conserved Ins2 gene has been replaced with GFP (16) (Fig. 1E). Mice lacking one or two wild-type (WT) functional Ins2 alleles had normal glucose homeostasis (Fig. 1F), consistent with our previous studies of Ins2 knockout mice (28) and the ability of Ins1 to compensate (29). Immunofluorescence staining of pancreata from Ins2GFP mice revealed a bimodal distribution of endogenous insulin production in vivo, similar to our single-cell analyses (Fig. 2A). Hand counting 1,879 cells across 11 randomly selected islets showed that 38.7% of β-cells had substantially higher GFP immunofluorescence above an arbitrary, but consistently applied, threshold (Fig. 2A). The percentage of cells with high GFP did not appear to vary as a function of islet size. We have observed similar heterogeneity when examining β-galactosidase knockin into the Ins2 locus (28), meaning that this observation is not unique to this knockin line or to GFP as a reporter.

FACS confirmed this bimodal distribution and that less than one-half of all β-cells engage in high Ins2 gene transcription at a given time (Fig. 2B). FACS analysis also validated that gfp mRNA, Ins2 mRNA, and pre-mRNA were significantly increased in high-GFP–expressing cells compared with cells expressing less GFP (Supplementary Fig. 1A), strongly suggesting that GFP protein levels accurately reflect Ins2 mRNA in this system. We did not expect there to be a perfect correlation due to the different predicted transcription-to-protein time courses for GFP and insulin (Fig. 2C and Supplementary Material). Nevertheless, these data demonstrate that GFP production, reflecting the activity of the endogenous Ins2 gene locus, is bimodal in vivo and ex vivo, with 35% of β-cells showing significantly higher activity. Hereafter, we refer to cells with high GFP abundance as Ins2(GFP)HIGH.

Live Cell Imaging of Insulin Gene Activity in Ins2WT/GFP Mice

We next generated a mouse model to examine endogenous Ins2 activity dynamics relative to a stable background β-cell marker. We crossed the Ins2GFP knockin line with transgenic mice with an allele wherein histone-fused mCherry is driven by the less complex Ins1 promoter that is known to have relatively stable red fluorescence in virtually all β-cells (18), allowing us to track Ins2 gene activity in real time (GFP) while observing all β-cells (mCherry). As expected, immunofluorescence of intact pancreatic sections and FACS analysis of dispersed islets from Ins2WT/GFP:Ins1-mCherry mice showed that mCherry labeled virtually all β-cells, while GFP was robustly expressed in a clearly separated subset of β-cells we deemed Ins2(GFP)HIGH (Fig. 3A–C). Quantitative PCR of these FACS-purified cells confirmed the expected elevated expression of GFP, Ins2, pre-Ins2, Ins1, and pre-Ins1 mRNA in Ins2(GFP)HIGH cells (Fig. 3D and Supplementary Fig. 1B). Ins2 pre-mRNA would be expected to precede GFP fluorescence by at least 1.3 h, as per estimations in Fig. 2C and calculations in the Supplementary Material, which provides a possible explanation for the elevated pre-Ins2 in the mCherry-positive, but GFP-low β-cells. Both intact islets and dispersed islet cells isolated from Ins2WT/GFP:Ins1-mCherry mice showed a similar proportion of Ins2(GFP)HIGH and Ins2(GFP)LOW cells to that we observed in vivo, demonstrating that this heterogeneity was not altered by isolation or dispersion/culture.

A high-throughput, live cell imaging system with environmental control was used to study dispersed islet cells from Ins2WT/GFP:Ins1-mCherry mice. Remarkably, live cell imaging over ∼3 days identified a subset of β-cells that transitioned in and out of Ins2(GFP)HIGH activity states over the course of 36-h-long recordings (Fig. 3E and Supplementary Fig. 2A and B). Analysis of 547 cells with various cellular behavior metrics as variables identified three distinct clusters of Ins2 gene activity (Supplementary Fig. 2B and C). While testing for viability of the Ins2WT/GFP β-cells, we noticed GFP fluorescence in Ins2WT/GFP cells sometimes rapidly decreased just before cell death. Therefore, in a larger study, we used propidium iodide, a dye that permits the real-time detection of the terminal stage of cell death, to identify changes in GFP fluorescence that are not associated with cell death (Fig. 4A–B). As propidium iodide emits red fluorescence at 636 nm, we used Ins2WT/GFP mice without Ins1-mCherry for these experiments. We chose the heterozygous Ins2WT/GFP genotype to keep the experiment as close as possible to WT insulin levels. Dispersed islets from Ins2WT/GFP mice were studied for up to 4 days, imaging cells at 5-min intervals. Similar to our pilot study, we identified β-cells that transitioned between high and low fluorescence states over the course of recordings (Fig. 4A and B). We tracked and quantified fluorescence intensities of Hoechst, GFP, and propidium iodide in 2,115 individual cells, confirming the bimodal distribution of mean GFP fluorescence in the cellular population (Fig. 4D and Supplementary Fig. 3A) we found in our FACS analysis. Quantification of other GFP wavelength parameters, including ACF and FWHM, revealed groups of cells with fluctuating Ins2 gene activity (Fig. 4E). We combined k-means clustering with PCA to reveal four clusters displaying distinct cellular behaviors (Supplementary Fig. 3B and C). Collectively, our long-term live cell imaging showed significant dynamic fluctuations in the activity of the endogenous Ins2 locus in primary β-cells, demonstrating that the apparent heterogeneity observed in vivo is not necessarily stable.

Autocorrelation Analysis of Fluctuations in Ins2 Gene Dynamics

The observation of fluctuations in Ins2 gene activity in some β-cells within our 4-day time window prompted us to determine the most common oscillation frequencies in Ins2WT/GFP β-cells. We performed autocorrelation analysis on GFP fluorescence intensity over time in Ins2WT/GFP β-cells that had at least one oscillation during the imaging period (Fig. 5A) and found that 357 (17%) of 2,115 cells had at least one oscillation, of which 226 had oscillation periods of 8–20 h. The most common frequency fell within the 12–16-h bin, with an average of ∼17 h across all oscillating cells (Fig. 5B), suggesting that this oscillatory behavior may be diurnal in nature. Interestingly, we also found that cells that oscillated had significantly higher GFP fluorescence intensity compared with nonoscillating cells (Fig. 5B). Together, our long-term live cell imaging data suggest that circadian-influenced oscillations of Ins2 gene activity persist in culture.

Ins2 Gene Dynamics Under Various Glucose and Serum Conditions

Glucose is a primary driver of insulin production, including insulin gene transcription (3,30). Therefore, we asked whether altering glucose concentrations would affect endogenous Ins2 gene activity in our knockin model. We treated isolated islet cells from Ins2WT/GFP mice with multiple culture conditions, including without serum, with serum, starvation from serum for 9 h before imaging, and an array of glucose concentrations (0, 5, 10, 15, 20 mmol/L). The time required for mature Ins2 mRNA production to respond to elevated glucose is still under debate, with various studies describing an increase in mature Ins2 mRNA being detected at a range of several hours after high-glucose treatments up to ∼48 h (31,32). We found significantly higher Ins2 gene activity (inferred by GFP fluorescence) in higher glucose concentrations under serum-containing conditions at the 48-h mark (Fig. 6A and B). We did not find significantly higher Ins2 gene activity in serum-starved conditions, but there was still a trend toward higher gene activity in higher glucose concentrations compared with 5 and 10 mmol/L glucose. In serum-free and all 0 mmol/L glucose conditions, cells died rapidly, confounding analysis of Ins2 gene activity (data not shown). After detrending our data as described in the Research Design and Methods section, a rise in Ins2 gene activity was observed ∼12 h into the experiment in cells cultured in higher glucose concentrations in both serum and serum-starved conditions (Fig. 6A and Supplementary Fig. 4A). Interestingly, we found more cells transitioning from the Ins2(GFP)LOW state to the Ins2(GFP)HIGH state, as well as fewer cells transitioning from the Ins2(GFP)HIGH state to the Ins2(GFP)LOW state in higher glucose concentrations, which may explain the increase in overall Ins2 gene activity in those conditions (Fig. 6C). We also investigated the range in which GFP fluorescence can fluctuate and found that cells in higher glucose conditions tended to have a greater increase in GFP fluorescence than in lower glucose conditions (Supplementary Fig. 6A). We performed hierarchical clustering, extracted individual variables, and compared the treatments (Supplementary Fig. 5AI). We found a significant decrease in higher versus lower glucose concentrations in the ACF time variable (Supplementary Fig. 5H). PCA with k-means clustering analyses of all measurements did not reveal any distinct clusters (Supplementary Fig. 6B and C).

Autocorrelation analyses showed a considerable difference in populations of oscillating versus nonoscillating cells in the various treatments. Of particular interest was the higher ratio of oscillating cells in higher glucose concentrations, suggesting that glucose may also influence Ins2 gene expression volatility (Fig. 5B). Combined with data in our initial Ins2WT/GFP imaging study in basal 10 mmol/L glucose conditions, this finding suggests that oscillating cells may be a contributor to the increase in Ins2 gene activity in response to high glucose (Figs. 3 and 4). Oscillation periods between treatments were not significantly different, but there was a trend toward higher oscillation periods in higher glucose concentrations (Supplementary Fig. 5C). We also looked at the amplitude of the oscillations by comparing the range of GFP fluorescence in oscillating cells, but we did not find any differences (Supplementary Fig. 5B). Collectively, these results show that various glucose concentrations under serum conditions may influence Ins2 gene activity and behavior.

Profiling β-Cell States With Single-Cell RNA Sequencing

To characterize the Ins2(GFP)HIGH state in a comprehensive and unbiased way, we performed single-cell RNA sequencing on FACS-purified Ins2(GFP)HIGH and Ins2(GFP)LOW cells from islets pooled from multiple mice (Fig. 7A–E and Supplementary Fig. 10AD). We focused on β-cells by including only those cells expressing Ins1 for downstream analysis (Fig. 7A and B and Supplementary Fig. 9A). Examination of the distribution of Ins2 gene expression in these Ins1-expressing cells further confirmed our previous quantitative PCR data (Fig. 3A) that gfp can accurately represent Ins2 expression levels (Fig. 7C and Supplementary Fig. 9B). We also found that gfp mRNA was expressed in a bimodal distribution, consistent with our FACS and live cell imaging results (Fig. 7C, Supplementary Fig. 9B). We then examined differential gene expression as a function of gfp mRNA (Fig. 7D and E and Supplementary Fig. 9C and D). We also considered Ins2(GFP)HIGH and Ins2(GFP)LOW as binary categories and examined differential gene expression (Supplementary Fig. 8A and B).

In 8-week-old Ins2WT/GFP mice, pathway enrichment analysis of cells expressing high gfp showed that the Ins2(GFP)HIGH cells had significant alterations in genes involved in hormone regulation, hormone secretion, hormone transport, protein synthesis, and protein cleavage (Supplementary Fig. 9C). Clusters of genes ordered by correlation with gfp in heterozygous Ins2WT/GFP β-cells showed that Sema6a, and the antioxidant metallothionein genes Mt1 and Mt2, were most closely correlated to gfp mRNA at the single-cell level. Classical genes related to β-cell function and maturity that were positively correlated with gfp included Ins2, Ero1lb, chromogranin A (Chga), the Glut2 glucose transporter Slc2a2, Iapp, Nkx6.1, and Pdx1 (18,33,34) (Supplementary Fig. 9D). The gfp-high state was also characterized by increased mRNA expression of genes encoding signal recognition particle 9 (Srp9), G-protein subunit γ 12 (Gng12), nucleotide pyrophosphatase/phosphodiesterase 2 (Enpp2), aldehyde dehydrogenase 5 family member A1 (Aldh5a1), and tetraspanin-28 (Cd81). Stress response genes, such as Sec61b, Sec61g, and Sec23b, were also highly expressed (35,36) (Supplementary Fig. 9D). The gfp-high state was associated with decreased expression of genes including soluble factors, such as peptide YY (Pyy), Sst, Gcg, and Ppy, suggesting in aggregate a less polyhormonal and, therefore, more mature gene expression profile (17,37) (Supplementary Fig. 9D). In agreement, gfp-high β-cells had lower expression of pappalysin 2 (Pappa2), an α-cell–selective regulator of IGF bioavailability and the ε-cell marker (Etv1) (38) (Supplementary Fig. 9D). The gfp-high cells also had reduced expression of several genes linked to insulin production and secretion, such as multiple subunits of the 40S and 60S ribosomes, eukaryotic translation initiation factor 3 subunit C (Eif3c), eukaryotic translation initiation factor 2 subunit 2 (Eif2s2), and heat shock 70-kDa protein 9 (Hspa9/mortalin) (Supplementary Fig. 9D). Interestingly, a decrease in the expression of certain endoplasmic reticulum (ER) stress genes was found, including Prkcsh, Nucb2, and Fam129a (3941) (Supplementary Fig. 9D). Thus, in young mice, the Ins2(GFP)HIGH cell state is associated with a mature single-cell gene expression profile, optimal insulin secretion, and a reorganization of the protein synthesis machinery.

Age is known to significantly alter the properties of pancreatic β-cells, including their function and ability to enter the cell cycle (42). Thus, we conducted an additional study in three 60-week-old mice. In this experiment, we studied cells from both Ins2WT/GFP (Fig. 7A–E) and Ins2GFP/GFP islets (Supplementary Fig. 10AC). Pathway enrichment analysis of cells expressing high gfp showed that the Ins2(GFP)HIGH cells had significant alterations in genes involved in vasculature development, metabolism, and ion homeostasis (Fig. 7D). Cluster analysis of genes ordered by gfp gradient in old heterozygous Ins2WT/GFP β-cells revealed that Ins2 expression most closely matches gfp mRNA (Fig. 7E). Similar to the younger mice, genes that increased with gfp expression included those related to optimal insulin secretion, such as Chga and Slc2a2, as well as to key β-cell transcription factors and maturity markers Nkx6.1 and Pdx1 (Fig. 7E). Other genes related to insulin production and secretion that were upregulated include Ins1, secretogranin 3 (Scg3), proprotein convertase subtilisin/kexin type 1 (Pcsk1), ubiquinol-cytochrome c reductase complex assembly factor 2 (Uqcc2), and synaptotagmin-like 4 (Sytl4) (25,43,44) (Fig. 7E). There was also an enrichment in key β-cell transcription factors and maturity markers Ucn3 and Neurod1 (45,46) (Fig. 7E). Interestingly, many of the mRNAs that were increased are known to be Pdx1 target genes in islets (47). Other notable genes that were upregulated in gfp-high cells were Gng12, Enpp2, Ppp1r1a, and metabolism-regulating genes, such as glucose-6-phosphatase catalytic subunit 2 (G6pc2) and hydroxyacyl-CoA dehydrogenase (Hadh) (Fig. 7E). ER stress–related gene Neat1 was also increased (48). Also in agreement with the analysis of young islets, gfp-high β-cells from old homozygous mice had decreased expression of Pappa2, Etv1, and Sst, as well as other markers of β-cell immaturity (Gcg, Ppy, Pyy). Ldha, another dedifferentiation marker, was also negatively correlated with gfp (Fig. 7E). Several ER stress genes were also anticorrelated with gfp, including Eef2 and Spp1 (49,50) (Fig. 7E). Thus, in older mice, the Ins2(GFP)HIGH cell state is similar overall to younger mice, with a more mature single-cell expression profile and association with genes related to metabolism.

Separately, we analyzed genes ordered by gfp mRNA gradient in old homozygous Ins2GFP/GFP β-cells (Supplementary Fig. 10). Many of the same genes that we observed in the heterozygous samples had similar expression patterns in the GFP homozygous samples, including maturity markers Chga, Neurod1, Pdx1, Ucn3, Slc2a2, Gng12, and Ppp1r1a and dedifferentiation markers Sst, Pyy, Ppy, and Gcg. Nuclear protein 1, transcriptional regulator (Nupr1), a stress adaptation gene, was reduced in homozygous Ins2GFP/GFP β-cells with high gfp expression. When we combined data from old Ins2WT/GFP and Ins2GFP/GFP cells, pathway enrichment analysis of high-gfp–expressing cells in older mice showed that the Ins2(GFP)HIGH cells had significant alterations in genes involved in protein translation, RNA splicing, and mRNA processing (Supplementary Fig. 10B). With this combined analysis, gfp correlation with many of the same genes found by analyzing Ins2WT/GFP and Ins2GFP/GFP cells separately were also identified (Supplementary Fig. 11). Pathway enrichment analysis of differential expression in all Ins1-positive cells revealed upregulation in genes related to the translation machinery in younger mice, while in older mice, there was upregulation in genes related to cellular respiration (Supplementary Fig. 12A and B).

The single-cell RNA sequencing studies identified enrichment in genes that control β-cell stress responses and markers of β-cell maturity. We followed this up by performing RNA velocity analysis on older Ins2WT/GFP and Ins2GFP/GFP mice. We found that the direction of the velocity vectors, which point toward clusters of β-cells with higher gfp expression, indicates that β-cells of older Ins2WT/GFP and Ins2GFP/GFP with higher levels of gfp mRNA were more mature than β-cells with lower levels of gfp mRNA (Fig. 8A and Supplementary Fig. 12A). Because of high dropout rates of gfp mRNA in the younger mice, we were unable to perform RNA velocity analysis on those mice.

Our single-cell analyses also showed correlation of gfp with various gene products related to optimal insulin secretion and processing. Immunofluorescence staining for insulin and proinsulin indicated increased abundance of both proteins in cells with high GFP, consistent with a possible difference in functional maturity between high- and low-GFP cell states (Fig. 8D and E). We found that PDX1 levels were significantly elevated in cells with high GFP staining compared with those in the same islet with lower GFP (Fig. 8F). However, we did not find a significant difference in CHGA, another putative maturity marker (Supplementary Fig. 12C). We next selected individual gene products from a list of significantly altered mRNAs in high-gfp–expressing cells (Supplementary Fig. 11). We have previously shown that insulin production itself is a significant stress under basal conditions in β-cells (1), and we therefore predicted that cells with increased Ins2 gene activity and GFP production would be more sensitive to stress. Interestingly, Hspa5 (GRP78), a major ER stress regulator, was downregulated at the mRNA level in high-gfp–expressing cells (Fig. 8G) but increased at the protein level in cells with higher GFP staining in pancreatic sections obtained from Ins2WT/GFP mice. The autocorrelation analyses revealed possible connections to circadian cycles. Indeed, high-gfp–expressing cells had increased Clock gene expression (Fig. 8B). Immunofluorescence staining showed that cells with higher GFP abundance had higher levels of CLOCK protein than cells with lower GFP, confirming our single-cell analyses (Fig. 8H). Thus, the states marked by high gene activity at the endogenous Ins2 locus likely possess critical functional differences. Taking the immunofluorescence data together with the single-cell RNA sequencing analysis, we conclude that β-cells with high gfp expression have a more mature β-cell profile.

Finally, we asked whether high insulin production was associated with increased β-cell fragility. Indeed, using live cell imaging, we found that Ins2(GFP)HIGH cells were >10-fold more sensitive to apoptosis at multiple glucose concentrations and in the presence of the model ER stressor thapsigargin compared with Ins2(GFP)LOW cells in the same cultures (Fig. 8I). Collectively, we have shown that insulin gene activity marks a range of β-cell maturity states, with consequences for function and survival.

The goal of the current study was to determine the nature of Ins2 gene expression heterogeneity. Analysis of pancreatic tissue sections from Ins2GFP knockin mice showed that at any given time, ∼40% of all β-cells exist in a GFP-high state, suggesting that not all β-cells have simultaneously high levels of active transcription at the Ins2 locus in vivo. Over the course of multiday in vitro imaging experiments, we observed clear transitions between Ins2(GFP)HIGH and Ins2(GFP)LOW states in single β-cells, possibly linked to the circadian clock. However, Ins2 gene activity was stable for the duration of these studies in the majority of cells. We used single-cell RNA sequencing to characterize the Ins2(GFP)HIGH cellular state and found that Ins2(GFP)HIGH cells were significantly more fragile under all stress conditions examined. Together with previous live cell imaging data, the results of the current study demonstrate that a substantial component of β-cell heterogeneity in relation to insulin gene expression is dynamic in time (Fig. 8J).

Live cell imaging of dispersed islet cells from Ins2GFP mice provided an unprecedented look at insulin gene activity in populations of single β-cells. Many of the Ins2(GFP)LOW and Ins2(GFP)HIGH β-cell states were maintained over at least 24 h. Autocorrelation analysis showed that of the cells that displayed at least one oscillation, a significant proportion have oscillation periods clustered at 8–12 h. There were significantly more GFPHIGH cells with oscillations than GFPLOW cells. We also found a significant increase of Clock gene and protein levels in cells with high GFP staining. Circadian control of islet function is essential for maintaining normal glucose homeostasis, and circadian-based variations on the transcriptional level have been described in diverse cell types and shown to be critical for optimization of cellular function (51). Circadian influences are known to persist in cell culture in β-cells, and insulin is both a target of circadian regulation and a regulator of the circadian machinery in many cell types (5254). Diurnal rhythms of β-cell function and metabolism have long been recognized (55,56), and our study extends this to activity at the insulin gene locus.

The study of pancreatic islet cell heterogeneity is currently experiencing resurgence, in part due to the application of single-cell sequencing and optogenetic technologies. There are many examples of β-cell heterogeneity, and these were reviewed recently (4). Insulin gene expression is a cardinal feature of pancreatic β-cells, but cell-by-cell variability in insulin production has remained underappreciated despite published evidence (11). For example, two states of chromatin accessibility were identified at the INS gene using single nucleus assay for transposase-accessible chromatin sequencing analyses (snATAC-seq) in human islets (57). There are reports of significant β-cell heterogeneity in transgenic mice expressing GFP under the Ins2 promoter (58). Similarly, we have shown variation in fluorescent protein expression under the control of Ins1 promoters in vivo and in vitro (1315). However, a limitation in these studies is that artificial promoter constructs may not recapitulate the more complex and long-range regulation available at the endogenous gene locus. Notwithstanding, cell-by-cell analysis of insulin mRNA, either by single-molecule fluorescence in situ hybridization or single-cell RNA sequencing, also showed a 2- to 10-fold range in native gene expression from endogenous insulin gene loci (8,12). The Ins2(GFP)HIGH β-cells we identified in our study are possibly a temporal manifestation of the extreme β-cells reported by Farack et al. (12). For instance, their extreme β-cells had significantly elevated Chga, Pdx1, Slc2a2, and Ucn3 mRNA expression (12), which is consistent with our single-cell RNA sequencing data. It is possible that the dynamic nature of insulin transcription is an adaptation to the extremely high demands of producing and maintaining adequate stores of insulin in β-cells. INS expression levels between β-cell subpopulations identified by Dorrell et al. (7) did not show significant variation, but certain subpopulations had differential expression in genes related to optimal insulin production and secretion (Neurod1, Ppp1r1a, Slc2a2), similar to the Ins2(GFP)HIGH cells. Assuming that there are homeostatic mechanisms to maintain stable insulin protein stores, it is unclear how the need for a burst of insulin gene activity would be sensed by individual β-cells, and future studies should attempt to define these mechanisms. It will be interesting to perform fluorescence tracking experiments on stem cell lines with INS-GFP knockins or maturity markers undergoing directed differentiation toward a β-cell fate to see whether dynamics in fluorescence can be observed as cells mature (58); it is likely that in vitro differentiation protocols will need to be optimized to produce a full range of dynamic β-cell characteristics.

The relationship between β-cell states demarked by insulin expression and disease pathogenesis remains unclear. Type 1 diabetes risk is associated with a paradoxical increase in INS mRNA in whole human fetal pancreas (59) and single adult β-cells (2). Isolated islets and single human β-cells from people with type 2 diabetes were reported to have reduced INS expression on average (38,60), although our recent meta-analysis of single-cell RNA sequencing data sets did not find a statistically significant reduction in INS mRNA levels in β-cells from donors with type 2 diabetes (2). Extreme β-cells were significantly more common in diabetic db/db mice (12). We found that the Ins2(GFP)HIGH state was associated with significant vulnerability to cellular stress, so having an excess number of Ins2(GFP)HIGH β-cells at a given time may negatively affect islet health and robustness. These results are consistent with our previous data defining the interrelationships among maximal insulin production, ER stress, and β-cell proliferation (1). We also note that in more hyperglycemic conditions, more cells may start to transfer toward the Ins2(GFP)HIGH state, but because of the small time frame that the imaging allows, we were unable to consistently observe cell transitions from Ins2(GFP)LOW to the higher end of Ins2(GFP)HIGH states. Rest may be important for β-cells, as dysfunctional cells that were removed from diabetic mice were shown to recover functionality when cultured at normal glucose levels (3). Thus, the ability to transition back to Ins2(GFP)LOW states may also be critical, as it may represent a form of rest for the β-cells, with failure in doing so possibly resulting in chronic stress and dysfunction. It is also likely that stress may modulate the frequency of β-cell state transitions, although we did not test this directly.

The presence of a proportion of β-cells in the Ins2(GFP)LOW state may be essential for islet function, with recent studies showing that loss of less mature β-cells affects overall islet function (61). Specifically, so-called “hub β-cells” that help to synchronize islets were reported to have lower insulin content compared with typical β-cells (5), and we have speculated that this represents a trade-off needed for their synchronizing function (6). Rodent and human β-cells are long-lived (62,63), and perhaps β-cells cycle through multiple states during their existence, including taking turns supporting the oscillatory coupling of the islet. Interestingly, many of the genes that are differentially expressed in Ins2(GFP)HIGH β-cells are known to play roles in type 2 diabetes susceptibility, including common alleles of the maturity-onset diabetes of the young/neonatal diabetes genes Pdx1, Neurod1, Nkx6.1, Abcc8, and Slc2a2, as well as Slc30a8 and Pam previously identified by genome-wide association (64). It will also be interesting to examine the frequency of β-cell states in the context of type 1 diabetes, given that proinsulin, Pdx1, Iapp, and Chga are autoantigens (6567). Indeed, β-cells undergoing proliferation or with lower insulin/maturity are protected in the NOD mouse model of type 1 diabetes (68,69). Collectively, these observations suggest that modulation of β-cell states could be a therapeutic and/or prevention target for both type 1 and type 2 diabetes.

Temporal transcriptional plasticity and gene expression bursting on a similar timescale as to what we have observed have been documented in bacteria, yeast, and other mammalian cell types (7072). For example, bursting gene expression patterns have been observed in pituitary cells (7375). Interestingly, luteinizing hormone subunit β transcription in pituitary gonadotrophs is directly linked to proteasome activity (74), suggesting a possible mechanism for coupling protein loads and transcription in secretory cell types. Many cell-extrinsic and cell-intrinsic factors have been implicated in the modulation of transcriptional burst frequency, including histone modifications and chromatin topology (7680). Future studies will be required to determine the molecular mechanisms mediating transcriptional bursting at the insulin gene locus in β-cells. Future studies should also seek to directly measure Ins2 mRNA transcription, perhaps using new clustered regularly interspaced short palindromic repeat–based probes or other mRNA tagging systems (8183).

In this study, we identified a subpopulation of cells that had relatively rapid transitions between Ins2(GFP)LOW and Ins2(GFP)HIGH β-cell states, suggesting dynamic transcriptional activity at the Ins2 gene locus rather than stable heterogeneity. However, our study has several limitations. For example, our measurements of GFP fluorescence originating from the Ins2 locus–mediated transcription cannot distinguish the relative contribution in changes in GFP mRNA transcription/stability or GFP protein translation/degradation. The half-life of unmodified GFP is ∼26 h (84), and the changes in GFP fluorescence in some cells suggest coordinated increases of protein synthesis and coupled protein degradation, which may also explain the slow decrease observed in our over time analyses. In addition, oscillations detected in the in vitro live cell imaging experiments cannot be used to determine the ratio of oscillating cells in vivo because of the possible loss of intercellular signals and additional stress in imaging-dispersed cells. While some intercellular signals could be missing in our analyses of dispersed islet cells (85,86), our in vivo analysis revealed a similar proportion of Ins2(GFP)LOW and Ins2(GFP)HIGH, giving us confidence that the two-state phenomenon we measured in vitro is generalizable to in vivo conditions. Future experiments should involve more powerful live cell imaging techniques, including tissue slides and intravitreal imaging, ideally with techniques that allow long-term cell tracking without excess phototoxicity or photobleaching (87,88).

In conclusion, our data demonstrate that single β-cells can switch between states marked by high and low activity of the phylogenetically conserved, endogenous insulin gene locus. This newly discovered phenomenon may account for much of the observed heterogeneity in β-cell insulin gene expression measured at single time points and needs to be comprehensively studied and leveraged in efforts to protect β-cells in the context of diabetes and to generate new β-cells from stem cells (17,89).

This article contains supplementary material online at https://doi.org/10.2337/figshare.21171376.

C.M.J.C. and H.M. contributed equally to this work.

Acknowledgments. The authors thank many colleagues for helpful discussions. This manuscript used data acquired from the Human Pancreas Analysis Program database (RRID:SCR_016202) (https://hpap.pmacs.upenn.edu) and Human Islet Research Network Consortium (RRID:SCR_014393) (UC4-DK-112217, U01-DK-123594, UC4-DK-112232, and U01-DK-123716).

Funding. This study was supported by a Canadian Institutes of Health Research operating grant (PJT-152999) to J.D.J. and the JDRF Centre of Excellence at the University of British Columbia (3-COE-2022-1103-M-B).

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

Author Contributions. C.M.J.C. designed studies, performed experiments, analyzed/interpreted data, and revised the manuscript. H.M. designed studies, performed experiments, analyzed/interpreted data, and wrote the original version of the manuscript. C.E., N.A.J.K., Y.B.Z., and Y.H.X. analyzed/interpreted data. S.S., H.C., N.N., and D.A.D. performed experiments and analyzed/interpreted data. E.P. supervised work and analyzed/interpreted data. X.H. performed experiments. S.X. provided a key reagent. M.O.H. provided a key reagent, designed studies, and analyzed/interpreted data. T.J.K. supervised work. F.C.L. supervised work, analyzed/interpreted data, and edited the manuscript. J.D.J. conceived the project, designed studies, interpreted data, and edited the manuscript. J.D.J. 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. Parts of this study were presented in poster form virtually at the Diabetes Canada Annual General Meeting, 27–30 October 2020.

1.
Szabat
M
,
Page
MM
,
Panzhinskiy
E
, et al
.
Reduced insulin production relieves endoplasmic reticulum stress and induces β cell proliferation
.
Cell Metab
2016
;
23
:
179
193
2.
Wang
S
,
Flibotte
S
,
Camunas-Soler
J
,
MacDonald
PE
,
Johnson
JD
.
A new hypothesis for type 1 diabetes risk: the at-risk allele at rs3842753 associates with increased beta-cell INS messenger RNA in a meta-analysis of single-cell RNA-sequencing data
.
Can J Diabetes
2021
;
45
:
775
784.e2
3.
Boland
BB
,
Rhodes
CJ
,
Grimsby
JS
.
The dynamic plasticity of insulin production in β-cells
.
Mol Metab
2017
;
6
:
958
973
4.
Benninger
RKP
,
Hodson
DJ
.
New understanding of β-cell heterogeneity and in situ islet function
.
Diabetes
2018
;
67
:
537
547
5.
Johnston
NR
,
Mitchell
RK
,
Haythorne
E
, et al
.
Beta cell hubs dictate pancreatic islet responses to glucose
.
Cell Metab
2016
;
24
:
389
401
6.
Kolic
J
,
Johnson
JD
.
Specialized hub beta cells trade maximal insulin production for perfect timing
.
Cell Metab
2016
;
24
:
371
373
7.
Dorrell
C
,
Schug
J
,
Canaday
PS
, et al
.
Human islets contain four distinct subtypes of β cells
.
Nat Commun
2016
;
7
:
11756
8.
Xin
Y
,
Dominguez Gutierrez
G
,
Okamoto
H
, et al
.
Pseudotime ordering of single human β-cells reveals states of insulin production and unfolded protein response
.
Diabetes
2018
;
67
:
1783
1794
9.
Wills
QF
,
Boothe
T
,
Asadi
A
, et al
.
Statistical approaches and software for clustering islet cell functional heterogeneity
.
Islets
2016
;
8
:
48
56
10.
Pipeleers
DG
.
Heterogeneity in pancreatic beta-cell population
.
Diabetes
1992
;
41
:
777
781
11.
Kiekens
R
,
In ’t Veld
P
,
Mahler
T
,
Schuit
F
,
Van De Winkel
M
,
Pipeleers
D
.
Differences in glucose recognition by individual rat pancreatic B cells are associated with intercellular differences in glucose-induced biosynthetic activity
.
J Clin Invest
1992
;
89
:
117
125
12.
Farack
L
,
Golan
M
,
Egozi
A
, et al
.
Transcriptional heterogeneity of beta cells in the intact pancreas
.
Dev Cell
2019
;
48
:
115
125.e4
13.
Szabat
M
,
Luciani
DS
,
Piret
JM
,
Johnson
JD
.
Maturation of adult beta-cells revealed using a Pdx1/insulin dual-reporter lentivirus
.
Endocrinology
2009
;
150
:
1627
1635
14.
Szabat
M
,
Pourghaderi
P
,
Soukhatcheva
G
, et al
.
Kinetics and genomic profiling of adult human and mouse β-cell maturation
.
Islets
2011
;
3
:
175
187
15.
Szabat
M
,
Johnson
JD
,
Piret
JM
.
Reciprocal modulation of adult beta cell maturity by activin A and follistatin
.
Diabetologia
2010
;
53
:
1680
1689
16.
Wakae-Takada
N
,
Xuan
S
,
Watanabe
K
,
Meda
P
,
Leibel
RL
.
Molecular basis for the regulation of islet beta cell mass in mice: the role of E-cadherin
.
Diabetologia
2013
;
56
:
856
866
17.
Johnson
JD
.
The quest to make fully functional human pancreatic beta cells from embryonic stem cells: climbing a mountain in the clouds
.
Diabetologia
2016
;
59
:
2047
2057
18.
Benner
C
,
van der Meulen
T
,
Cacéres
E
,
Tigyi
K
,
Donaldson
CJ
,
Huising
MO
.
The transcriptional landscape of mouse beta cells compared to human beta cells reveals notable species differences in long non-coding RNA and protein-coding gene expression
.
BMC Genomics
2014
;
15
:
620
19.
Kaestner
KH
,
Powers
AC
,
Naji
A
,
Atkinson
MA
;
HPAP Consortium
.
NIH initiative to improve understanding of the pancreas, islet, and autoimmunity in type 1 diabetes: the Human Pancreas Analysis Program (HPAP)
.
Diabetes
2019
;
68
:
1394
1402
20.
Yang
YH
,
Johnson
JD
.
Multi-parameter single-cell kinetic analysis reveals multiple modes of cell death in primary pancreatic β-cells
.
J Cell Sci
2013
;
126
:
4286
4295
21.
Matthews
DR
,
Lang
DA
,
Burnett
MA
,
Turner
RC
.
Control of pulsatile insulin secretion in man
.
Diabetologia
1983
;
24
:
231
237
22.
Scrucca
L
,
Fop
M
,
Murphy
TB
,
Raftery
AE
.
mclust 5: clustering, classification and density estimation using gaussian finite mixture models
.
R J
2016
;
8
:
289
317
23.
Chen
S
,
Thakor
NV
,
Mower
MM
.
Ventricular fibrillation detection by a regression test on the autocorrelation function
.
Med Biol Eng Comput
1987
;
25
:
241
249
24.
Butler
A
,
Hoffman
P
,
Smibert
P
,
Papalexi
E
,
Satija
R
.
Integrating single-cell transcriptomic data across different conditions, technologies, and species
.
Nat Biotechnol
2018
;
36
:
411
420
25.
Szabat
M
,
Lynn
FC
,
Hoffman
BG
,
Kieffer
TJ
,
Allan
DW
,
Johnson
JD
.
Maintenance of β-cell maturity and plasticity in the adult pancreas: developmental biology concepts in adult physiology
.
Diabetes
2012
;
61
:
1365
1371
26.
Johnson
JD
,
Ahmed
NT
,
Luciani
DS
, et al
.
Increased islet apoptosis in Pdx1+/− mice
.
J Clin Invest
2003
;
111
:
1147
1160
27.
Johnson
JD
,
Bernal-Mizrachi
E
,
Alejandro
EU
, et al
.
Insulin protects islets from apoptosis via Pdx1 and specific changes in the human islet proteome
.
Proc Natl Acad Sci U S A
2006
;
103
:
19575
19580
28.
Mehran
AE
,
Templeman
NM
,
Brigidi
GS
, et al
.
Hyperinsulinemia drives diet-induced obesity independently of brain insulin production
.
Cell Metab
2012
;
16
:
723
737
29.
Leroux
L
,
Desbois
P
,
Lamotte
L
, et al
.
Compensatory responses in mice carrying a null mutation for Ins1 or Ins2
.
Diabetes
2001
;
50
(
Suppl. 1
):
S150
S153
30.
Knoch
KP
,
Nath-Sain
S
,
Petzold
A
, et al
.
PTBP1 is required for glucose-stimulated cap-independent translation of insulin granule proteins and coxsackieviruses in beta cells
.
Mol Metab
2014
;
3
:
518
530
31.
Evans-Molina
C
,
Garmey
JC
,
Ketchum
R
,
Brayman
KL
,
Deng
S
,
Mirmira
RG
.
Glucose regulation of insulin gene transcription and pre-mRNA processing in human islets
.
Diabetes
2007
;
56
:
827
835
32.
Tillmar
L
,
Carlsson
C
,
Welsh
N
.
Control of insulin mRNA stability in rat pancreatic islets. Regulatory role of a 3′-untranslated region pyrimidine-rich sequence
.
J Biol Chem
2002
;
277
:
1099
1106
33.
Zito
E
,
Chin
KT
,
Blais
J
,
Harding
HP
,
Ron
D
.
ERO1-beta, a pancreas-specific disulfide oxidase, promotes insulin biogenesis and glucose homeostasis
.
J Cell Biol
2010
;
188
:
821
832
34.
Augsornworawat
P
,
Maxwell
KG
,
Velazco-Cruz
L
,
Millman
JR
.
Single-cell transcriptome profiling reveals β cell maturation in stem cell-derived islets after transplantation
.
Cell Rep
2021
;
34
:
108850
35.
Sundaram
A
,
Plumb
R
,
Appathurai
S
,
Mariappan
M
.
The Sec61 translocon limits IRE1α signaling during the unfolded protein response
.
eLife
2017
;
6
:
e27187
36.
Yehia
L
,
Liu
D
,
Fu
S
,
Iyer
P
,
Eng
C
.
Non-canonical role of wild-type SEC23B in the cellular stress response pathway
.
Cell Death Dis
2021
;
12
:
304
37.
Basford
CL
,
Prentice
KJ
,
Hardy
AB
, et al
.
The functional and molecular characterisation of human embryonic stem cell-derived insulin-positive cells compared with adult pancreatic beta cells
.
Diabetologia
2012
;
55
:
358
371
38.
Segerstolpe
Å
,
Palasantza
A
,
Eliasson
P
, et al
.
Single-cell transcriptome profiling of human pancreatic islets in health and type 2 diabetes
.
Cell Metab
2016
;
24
:
593
607
39.
Shin
GC
,
Moon
SU
,
Kang
HS
,
Choi
HS
,
Han
HD
,
Kim
KH
.
PRKCSH contributes to tumorigenesis by selective boosting of IRE1 signaling pathway
.
Nat Commun
2019
;
10
:
3185
40.
Zhang
D
,
Lin
J
,
Chao
Y
, et al
.
Regulation of the adaptation to ER stress by KLF4 facilitates melanoma cell metastasis via upregulating NUCB2 expression
.
J Exp Clin Cancer Res
2018
;
37
:
176
41.
Evstafieva
AG
,
Kovaleva
IE
,
Shoshinova
MS
,
Budanov
AV
,
Chumakov
PM
.
Implication of KRT16, FAM129A and HKDC1 genes as ATF4 regulated components of the integrated stress response
.
PLoS One
2018
;
13
:
e0191107
42.
Rankin
MM
,
Kushner
JA
.
Adaptive beta-cell proliferation is severely restricted with advanced age
.
Diabetes
2009
;
58
:
1365
1372
43.
Cambier
L
,
Rassam
P
,
Chabi
B
, et al
.
M19 modulates skeletal muscle differentiation and insulin secretion in pancreatic β-cells through modulation of respiratory chain activity
.
PLoS One
2012
;
7
:
e31815
44.
Lin
CC
,
Cheng
KP
,
Hung
HC
, et al
.
Serum secretogranin III concentrations were increased in subjects with metabolic syndrome and independently associated with fasting plasma glucose levels
.
J Clin Med
2019
;
8
:
1436
45.
Fu
Z
,
Gilbert
ER
,
Liu
D
.
Regulation of insulin synthesis and secretion and pancreatic beta-cell dysfunction in diabetes
.
Curr Diabetes Rev
2013
;
9
:
25
53
46.
Blum
B
,
Hrvatin
S
,
Schuetz
C
,
Bonal
C
,
Rezania
A
,
Melton
DA
.
Functional beta-cell maturation is marked by an increased glucose threshold and by expression of urocortin 3
.
Nat Biotechnol
2012
;
30
:
261
264
47.
Sachdeva
MM
,
Claiborn
KC
,
Khoo
C
, et al
.
Pdx1 (MODY4) regulates pancreatic beta cell susceptibility to ER stress
.
Proc Natl Acad Sci U S A
2009
;
106
:
19090
19095
48.
Adriaens
C
,
Standaert
L
,
Barra
J
, et al
.
p53 induces formation of NEAT1 lncRNA-containing paraspeckles that modulate replication stress response and chemosensitivity
.
Nat Med
2016
;
22
:
861
868
49.
Lépine
S
,
Allegood
JC
,
Park
M
,
Dent
P
,
Milstien
S
,
Spiegel
S
.
Sphingosine-1-phosphate phosphohydrolase-1 regulates ER stress-induced autophagy
.
Cell Death Differ
2011
;
18
:
350
361
50.
Boyce
M
,
Py
BF
,
Ryazanov
AG
, et al
.
A pharmacoproteomic approach implicates eukaryotic elongation factor 2 kinase in ER stress-induced cell death
.
Cell Death Differ
2008
;
15
:
589
599
51.
Kalsbeek
A
,
la Fleur
S
,
Fliers
E
.
Circadian control of glucose metabolism
.
Mol Metab
2014
;
3
:
372
383
52.
Perelis
M
,
Marcheva
B
,
Ramsey
KM
, et al
.
Pancreatic β cell enhancers regulate rhythmic transcription of genes controlling insulin secretion
.
Science
2015
;
350
:
aac4250
53.
Allaman-Pillet
N
,
Roduit
R
,
Oberson
A
, et al
.
Circadian regulation of islet genes involved in insulin production and secretion
.
Mol Cell Endocrinol
2004
;
226
:
59
66
54.
Petrenko
V
,
Gosmain
Y
,
Dibner
C
.
High-resolution recording of the circadian oscillator in primary mouse α- and β-cell culture
.
Front Endocrinol (Lausanne)
2017
;
8
:
68
55.
Groop
L
,
Harno
K
.
Diurnal pattern of plasma insulin and blood glucose during glibenclamide and glipizide therapy in elderly diabetics
.
Acta Endocrinol Suppl (Copenh)
1980
;
239
:
44
52
56.
Polonsky
KS
,
Given
BD
,
Van Cauter
E
.
Twenty-four-hour profiles and pulsatile patterns of insulin secretion in normal and obese subjects
.
J Clin Invest
1988
;
81
:
442
448
57.
Chiou
J
,
Zeng
C
,
Cheng
Z
, et al
.
Single-cell chromatin accessibility identifies pancreatic islet cell type- and state-specific regulatory programs of diabetes risk
.
Nat Genet
2021
;
53
:
455
466
58.
Katsuta
H
,
Aguayo-Mazzucato
C
,
Katsuta
R
, et al
.
Subpopulations of GFP-marked mouse pancreatic β-cells differ in size, granularity, and insulin secretion
.
Endocrinology
2012
;
153
:
5180
5187
59.
Veres
A
,
Faust
AL
,
Bushnell
HL
, et al
.
Charting cellular identity during human in vitro β-cell differentiation
.
Nature
2019
;
569
:
368
373
60.
Vafiadis
P
,
Bennett
ST
,
Colle
E
,
Grabs
R
,
Goodyer
CG
,
Polychronakos
C
.
Imprinted and genotype-specific expression of genes at the IDDM2 locus in pancreas and leucocytes
.
J Autoimmun
1996
;
9
:
397
403
61.
Yang
BT
,
Dayeh
TA
,
Kirkpatrick
CL
, et al
.
Insulin promoter DNA methylation correlates negatively with insulin gene expression and positively with HbA(1c) levels in human pancreatic islets
.
Diabetologia
2011
;
54
:
360
367
62.
Nasteska
D
,
Fine
NHF
,
Ashford
FB
, et al
.
PDX1LOW MAFALOW β-cells contribute to islet function and insulin release
.
Nat Commun
2021
;
12
:
674
63.
Perl
S
,
Kushner
JA
,
Buchholz
BA
, 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
2010
;
95
:
E234
E239
64.
Teta
M
,
Long
SY
,
Wartschow
LM
,
Rankin
MM
,
Kushner
JA
.
Very slow turnover of beta-cells in aged adult mice
.
Diabetes
2005
;
54
:
2557
2567
65.
Thomsen
SK
,
Raimondo
A
,
Hastoy
B
, et al
.
Type 2 diabetes risk alleles in PAM impact insulin release from human pancreatic β-cells
.
Nat Genet
2018
;
50
:
1122
1131
66.
Nakayama
M
,
Abiru
N
,
Moriyama
H
, et al
.
Prime role for an insulin epitope in the development of type 1 diabetes in NOD mice
.
Nature
2005
;
435
:
220
223
67.
Stadinski
BD
,
Delong
T
,
Reisdorph
N
, et al
.
Chromogranin A is an autoantigen in type 1 diabetes
.
Nat Immunol
2010
;
11
:
225
231
68.
Han
S
,
Donelan
W
,
Wang
H
,
Reeves
W
,
Yang
LJ
.
Novel autoantigens in type 1 diabetes
.
Am J Transl Res
2013
;
5
:
379
392
69.
Rui
J
,
Deng
S
,
Arazi
A
,
Perdigoto
AL
,
Liu
Z
,
Herold
KC
.
β cells that resist immunological attack develop during progression of autoimmune diabetes in NOD mice
.
Cell Metab
2017
;
25
:
727
738
70.
Dirice
E
,
Kahraman
S
,
De Jesus
DF
, et al
.
Increased β-cell proliferation before immune cell invasion prevents progression of type 1 diabetes
.
Nat Metab
2019
;
1
:
509
518
71.
Raj
A
,
Peskin
CS
,
Tranchina
D
,
Vargas
DY
,
Tyagi
S
.
Stochastic mRNA synthesis in mammalian cells
.
PLoS Biol
2006
;
4
:
e309
72.
Vera
M
,
Biswas
J
,
Senecal
A
,
Singer
RH
,
Park
HY
.
Single-cell and single-molecule analysis of gene expression regulation
.
Annu Rev Genet
2016
;
50
:
267
291
73.
Suter
DM
,
Molina
N
,
Gatfield
D
,
Schneider
K
,
Schibler
U
,
Naef
F
.
Mammalian genes are transcribed with widely different bursting kinetics
.
Science
2011
;
332
:
472
474
74.
Norris
AJ
,
Stirland
JA
,
McFerran
DW
, et al
.
Dynamic patterns of growth hormone gene transcription in individual living pituitary cells
.
Mol Endocrinol
2003
;
17
:
193
202
75.
Walsh
HE
,
Shupnik
MA
.
Proteasome regulation of dynamic transcription factor occupancy on the GnRH-stimulated luteinizing hormone beta-subunit promoter
.
Mol Endocrinol
2009
;
23
:
237
250
76.
Harper
CV
,
Featherstone
K
,
Semprini
S
, et al
.
Dynamic organisation of prolactin gene expression in living pituitary tissue
.
J Cell Sci
2010
;
123
:
424
430
77.
Phillips
NE
,
Mandic
A
,
Omidi
S
,
Naef
F
,
Suter
DM
.
Memory and relatedness of transcriptional activity in mammalian cell lineages
.
Nat Commun
2019
;
10
:
1208
78.
Nicolas
D
,
Zoller
B
,
Suter
DM
,
Naef
F
.
Modulation of transcriptional burst frequency by histone acetylation
.
Proc Natl Acad Sci U S A
2018
;
115
:
7153
7158
79.
Nicolas
D
,
Phillips
NE
,
Naef
F
.
What shapes eukaryotic transcriptional bursting?
Mol Biosyst
2017
;
13
:
1280
1290
80.
Atger
F
,
Gobet
C
,
Marquis
J
, et al
.
Circadian and feeding rhythms differentially affect rhythmic mRNA transcription and translation in mouse liver
.
Proc Natl Acad Sci U S A
2015
;
112
:
E6579
E6588
81.
Molina
N
,
Suter
DM
,
Cannavo
R
,
Zoller
B
,
Gotic
I
,
Naef
F
.
Stimulus-induced modulation of transcriptional bursting in a single mammalian gene
.
Proc Natl Acad Sci U S A
2013
;
110
:
20563
20568
82.
Pichon
X
,
Lagha
M
,
Mueller
F
,
Bertrand
E
.
A growing toolbox to image gene expression in single cells: sensitive approaches for demanding challenges
.
Mol Cell
2018
;
71
:
468
480
83.
Pichon
X
,
Robert
MC
,
Bertrand
E
,
Singer
RH
,
Tutucci
E
.
New generations of MS2 variants and MCP fusions to detect single mRNAs in living eukaryotic cells
.
Methods Mol Biol
2020
;
2166
:
121
144
84.
Vera
M
,
Tutucci
E
,
Singer
RH
.
Imaging single mRNA molecules in mammalian cells using an optimized MS2-MCP system
.
Methods Mol Biol
2019
;
2038
:
3
20
85.
Corish
P
,
Tyler-Smith
C
.
Attenuation of green fluorescent protein half-life in mammalian cells
.
Protein Eng
1999
;
12
:
1035
1040
86.
Chowdhury
A
,
Dyachok
O
,
Tengholm
A
,
Sandler
S
,
Bergsten
P
.
Functional differences between aggregated and dispersed insulin-producing cells
.
Diabetologia
2013
;
56
:
1557
1568
87.
Scarl
RT
,
Corbin
KL
,
Vann
NW
, et al
.
Intact pancreatic islets and dispersed beta-cells both generate intracellular calcium oscillations but differ in their responsiveness to glucose
.
Cell Calcium
2019
;
83
:
102081
88.
Kiepas
A
,
Voorand
E
,
Mubaid
F
,
Siegel
PM
,
Brown
CM
.
Optimizing live-cell fluorescence imaging conditions to minimize phototoxicity
.
J Cell Sci
2020
;
133
:
jcs242834
89.
Laissue
PP
,
Alghamdi
RA
,
Tomancak
P
,
Reynaud
EG
,
Shroff
H
.
Assessing phototoxicity in live fluorescence imaging
.
Nat Methods
2017
;
14
:
657
661
90.
Alvarez-Dominguez
JR
,
Donaghey
J
,
Rasouli
N
, et al
.
Circadian Entrainment Triggers Maturation of Human In Vitro Islets
.
Cell Stem Cell
2020
;
26
:
108
122.e10
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.