Dedifferentiation of insulin-secreting β cells in the islets of Langerhans has been proposed to be a major mechanism of β-cell dysfunction. Whether dedifferentiated β cells can be targeted by pharmacological intervention for diabetes remission, and ways in which this could be accomplished, are unknown as yet. Here we report the use of streptozotocin-induced diabetes to study β-cell dedifferentiation in mice. Single-cell RNA sequencing (scRNA-seq) of islets identified markers and pathways associated with β-cell dedifferentiation and dysfunction. Single and combinatorial pharmacology further show that insulin treatment triggers insulin receptor pathway activation in β cells and restores maturation and function for diabetes remission. Additional β-cell selective delivery of oestrogen by Glucagon-like peptide-1 (GLP-1–oestrogen conjugate) decreases daily insulin requirements by 60%, triggers oestrogen-specific activation of the endoplasmic-reticulum-associated protein degradation system, and further increases β-cell survival and regeneration. GLP-1–oestrogen also protects human β cells against cytokine-induced dysfunction. This study not only describes mechanisms of β-cell dedifferentiation and regeneration, but also reveals pharmacological entry points to target dedifferentiated β cells for diabetes remission.
The progressive loss or dysfunction of insulin-producing β-cell mass ultimately leads to type 1 diabetes (T1D) or type 2 diabetes (T2D), respectively1. Current pharmacological treatments do not stop the decline of β-cell function and number that leads to glucose excursions and, eventually, to devastating microvascular and macrovascular complications. Hence, the ideal treatment should be initiated when the first diabetic symptoms appear, to protect or regenerate glucose-sensing and insulin-secreting β cells for optimal blood glucose regulation and to prevent secondary complications. Recently, T1D progression has been halted by anti-CD3 immunotherapy for 2 years (ref. 2), but it will be important to test whether additional β-cell regenerative therapy can further or permanently delay the onset of diabetes. Intensive insulin therapy at disease onset has been shown to partially restore β-cell function, which slows the disease progression in patients with T1D and T2D3,4,5. Despite both therapies providing similar glycaemic control, early intensive insulin treatment, compared with oral anti-diabetic drugs, in patients with T2D better preserves β-cell function, which suggests additional glucose-independent beneficial effects of insulin therapy6,7. Thus, an understanding of the mechanisms of β-cell dysfunction and pharmacological replenishment is urgently required to stop or reverse diabetes progression and to improve the therapeutic options for patients. Dedifferentiation of β cells has been observed in genetic mouse models of T1D and T2D as well as in patients with diabetes, and is characterised by the loss of the expression of key maturation marker genes (for example, Slc2a2 (also known as Glut2) and Ucn3) and by impaired insulin secretion, and thereby contributes to β-cell dysfunction and hyperglycaemia8,9,10. To investigate whether dysfunctional β cells under hyperglycaemic conditions can be targeted pharmacologically to restore β-cell function, we explored the multiple-low-dose model of streptozotocin-induced diabetes (mSTZ) in mice. STZ specifically ablates β cells, but when STZ is injected in multiple low doses, some residual β cells can survive11. Furthermore, the absence of genetic lesions and autoimmunity in the mSTZ model permits the investigation of the fate of those remaining β cells and the effect of pharmacological treatment on β-cell protection and regeneration.
Insulin restores β-cell function
Ten days after the last mSTZ injection, mice were severely hyperglycaemic (Extended Data Fig. 1a) and showed an impaired islet architecture (Extended Data Fig. 1b) with markedly decreased β-cell numbers in comparison to those in control mice (Extended Data Fig. 1b,c). Proliferation of remaining β cells was unchanged (Extended Data Fig. 1d), whereas β-cell apoptosis was significantly increased in mSTZ-treated mice compared to control mice (Extended Data Fig. 1e) and was accompanied by a loss of identity and function (Extended Data Fig. 1f–h). Hence, at this time point (diabetes onset), when a fraction of dysfunctional β cells were still remaining, we initiated a permanent drug treatment over 100 d (Fig. 1a). Vehicle-treated mSTZ mice remained diabetic over the length of the study, which suggests that the residual β-cell functionality or endogenous β-cell regeneration12,13 is insufficient to maintain or restore sufficient glucose homeostasis (Fig. 1b–f). To correct the insulin deficiency in mSTZ mice, we treated diabetic mice with a long-acting pegylated insulin analogue (PEG–insulin, once daily), which improved glycaemia (Fig. 1b), increased C-peptide levels (Fig. 1c), improved islet structure (Fig. 1d) and increased the number of insulin-positive cells (Fig. 1e,f). This shows functional β-cell recovery upon glycaemia normalisation, which extends findings from insulin treatment in genetic mouse models14,15. However, the risk of hypoglycaemia and unwanted weight gain are undesirable hallmarks of insulin therapy, and thus alternative pharmacological approaches are required to mitigate these possible effects.
GLP-1–oestrogen and insulin polypharmacotherapy
Oestrogen and glucagon-like peptide 1 (GLP-1) have been repeatedly implicated in the treatment of diabetes due to insulinotropic and β-cell protective effects in preclinical studies16,17. Chemically optimised GLP-1 analogues profoundly improve glucose and body weight management in obese people and individuals with T2D18. However, severe gynaecological, oncogenic and mitogenic side effects preclude chronic oestrogen use as a drug for diabetes and, as yet, GLP-1 analogues have failed to preserve β-cell function and mass in obese and diabetic humans17.
To circumvent the gynaecological, oncogenic and mitogenic actions of oestrogen, we recently designed and evaluated a stable GLP-1–oestrogen conjugate, which reversed the metabolic syndrome in diet-induced obese male and female mice19. Here, we used the GLP-1–oestrogen conjugate to test whether the specific delivery of oestrogen into the GLP-1 receptor protein (GLP-1R)-expressing β cells could restore β-cell functionality. GLP-1–oestrogen treatment for 100 d was more efficacious in decreasing fasting glucose (Fig. 1b) and increasing fasting C-peptide (Fig. 1c) and insulin levels (Extended Data Fig. 2a) than either of the monoagonists (oestrogen or GLP-1 alone). Moreover, only GLP-1–oestrogen treatment improved pancreatic islet architecture (Fig. 1d) and increased β-cell number, as compared with those in mSTZ-diabetic mice treated with vehicle (Fig. 1e,f). These effects were independent of body weight loss (Extended Data Fig. 2b).
Polypharmacotherapy holds the potential to simultaneously activate redundant or additive pathways to enhance efficacy, and to enable reduced dosing of the individual components and consequently reduce the risk of unwanted side effects20. We tested the combination of insulin and GLP-1–oestrogen to investigate a triple pharmacological approach to enhance the efficacy of both compounds, and particularly to lessen the amount of insulin required. The combination therapy normalised glycaemia (Fig. 1b) and increased C-peptide levels (Fig. 1c), as compared with those in mSTZ-diabetic mice treated with vehicle. Furthermore, the combination therapy displayed a superior effect compared to treatment with insulin alone to limit weight gain (Extended Data Fig. 2b), normalise glucose tolerance (Extended Data Fig. 2c), increase pancreatic insulin content (Extended Data Fig. 2d) and increase β-cell number (Fig. 1e,f). Importantly, we were able to reduce the insulin dose by 60% (10 nmol kg−1) compared with that in the insulin monotherapy (25 nmol kg−1) and still achieve superior therapeutic outcomes, which reduces the risk of hypoglycaemia and unintended weight gain.
To test whether treatment-induced improvements on glucose and islet homoeostasis are maintained, we switched a group of mice after 12 weeks of GLP-1–oestrogen treatment to two additional weeks of vehicle injections. Positive effects of GLP-1–oestrogen treatment to reduce fasting glycaemia (Fig. 2a), increase fasting C-peptide levels (Fig. 2b), and enhance the β-cell maturation state (Fig. 2c,d) were sustained after these two weeks, which supports the notion of preserved islet cell function even after treatment cessation.
GLP-1–oestrogen targets β cells
We next wanted to confirm the absence of systemic toxicity that was related to the oestrogen component of the GLP-1–oestrogen conjugate, a pre-requisite for clinical use. To that end, we investigated whether GLP-1–oestrogen (doses up to 10× higher than are generally used in mouse experiments, and at least 1,000× the plasma oestradiol exposure as compared with that in women on hormone replacement therapy (see Methods)) increased uterus weight in ovariectomised (OVX) rats after two weeks of treatment (Extended Data Fig. 3a and Methods). In contrast to treatment with oestrogen alone, no treatment-related effect was observed with the GLP-1–oestrogen conjugate (Extended Data Fig. 3b), which is consistent with previously reported results in OVX mice19.
To confirm β-cell-specific targeting of the GLP-1–oestrogen conjugate, we used a double knock-in fluorescent reporter mouse model (Foxa2-Venus Fusion (FVF) × Pdx1-BFP (blue fluorescent protein) fusion (PBF); FVFPBFDHom)21, which allows α- and β-cell sorting (Extended Data Fig. 3c,d). Male FVFPBFDHom mice develop maturity onset diabetes of the young owing to reduced Pdx1 levels in islets, accompanied by hyperglycaemia, reduced β-cell number and impaired islet architecture at weaning age21. In this genetic diabetes model, none of the therapies used in this study improved glycaemia after four weeks of treatment (Extended Data Fig. 3e,f). These results suggest that β-cell function cannot be restored pharmacologically in the presence of the genetically induced β-cell lesions in the FVFPBFDHom mice. However, the GLP-1–oestrogen conjugate, but not the monoagonists, specifically increased β-cell granularity of FVFPBFDHom mice in comparison to control mice, which shows that the conjugate selectively targets β cells (Extended Data Fig. 3g).
GLP-1–oestrogen improves human β-cell function
To provide human relevance to the findings, we tested GLP-1–oestrogen and the monoagonists in human micro-islets in the absence or presence of β-cell stressors (cytokine cocktail, see Methods). After acute compound exposure, GLP-1–oestrogen was more potent than either of the individual components in increasing glucose-stimulated insulin secretion (GSIS) from human micro-islets in comparison to that in untreated and monoagonist-treated micro-islets (Fig. 3a). We next exposed the human micro-islets to cytokines to determine whether the beneficial effects of GLP-1–oestrogen protect against stress-induced impairment of β-cell functionality (Fig. 3b). Seven-day treatment with GLP-1–oestrogen enhanced GSIS, and exceeded the effects that were seen with either of the individual components (Fig. 3b). Moreover, only GLP-1–oestrogen treatment increased the total insulin content of cytokine-exposed human micro-islets (Fig. 3c). This was independent of changes in the total ATP content (Extended Data Fig. 4a) and in caspase luciferase activity (Extended Data Fig. 4b), which suggests that the compound treatment improved functionality, but did not improve cell survival of human micro-islets. These results show that GLP-1–oestrogen is superior to both of the monoagonists in improving β-cell function in homoeostasis, and in acting upon cytokine stress in both mice and humans.
β-cell heterogeneity in homoeostatic and healthy mice
To elucidate the molecular mechanisms that underlie β-cell failure in mSTZ diabetes, and β-cell recovery after the different therapeutic approaches, we performed scRNA-seq of isolated islets from mice that responded to treatment (Extended Data Fig. 5).
In normal islet homoeostasis, we identified the four main endocrine cell subtypes, α cells, β cells, δ cells and pancreatic polypeptide (PP) cells, by unbiased graph-based clustering. Clusters were annotated on the basis of predominant endocrine hormone expression of glucagon (Gcg), insulin (Ins), somatostatin (Sst) and PP (Extended Data Fig. 6a,b and Methods). For each of the four endocrine subtypes we identified a specific marker gene signature (Supplementary Table 1 and Methods). Refined clustering of insulin-positive cells revealed the presence of two main β-cell subpopulations, β1 and β2 (Extended Data Fig. 6c and Methods). Single-cell trajectory inference suggests a continuum of transcriptional states, rather than discrete phenotypes within β cells and a transition between β1 and β2 subpopulations (Extended Data Fig. 6d and Methods). We found a progressive increase in expression of β-cell maturation marker genes (for example Ins1, Ins2 and Ucn3 (ref. 22)), and genes of the secretion machinery (G6pc2, Sytl4 and Slc2a2), as well as a concomitant decrease in expression of the β-cell immaturity gene Mafb23, and pan-endocrine lineage marker genes (Chga and Chgb) along the pseudotime trajectory from β2- to β1-cells (Extended Data Fig. 6d and Supplementary Table 2). The expression of transcription factor genes that are associated with β-cell identity (for example Pdx1, Nkx6.1 and NeuroD1) was unchanged (Extended Data Fig. 6d). The β2-cell cluster was characterised by the downregulation of genes that are involved in insulin secretion, oxidative phosphorylation and cell-cycle inhibition, as well as by the upregulation of genes that are involved in cAMP and WNT signalling (Extended Data Fig. 6e), both of which observations are suggestive of a more immature and/or proliferative state for β2 cells24. We consistently observed upregulation of cell-cycle-associated genes, such as Ki67 and Cdk1 in the immature β2 subpopulation and, in accordance with this result, 16 out of 403 of the β2-cells, but only 2 out of 5,319 of the mature β1-cells, were classified as being involved in cell cycling (Extended Data Fig. 6f (see Methods)). Taken together, these results confirm the co-existence of mature (β1) and immature and/or proliferative β cells (β2) in healthy mouse islets24,25. In addition, we found subpopulations of polyhormonal cells that could be distinguished from doublets and ambient RNA, both of which are common problems with the scRNA-seq technology (Extended Data Fig. 6g,h and Methods).
β-cell dedifferentiation at the single-cell level
To obtain a deeper understanding of the cell autonomous and non-cell autonomous effects that underlie chemical β-cell ablation in the islet cell niche, we performed a scRNA-seq survey of diabetic islets after 100 d of persistent hyperglycaemia. Unsupervised clustering and embedding of the scRNA-seq data revealed altered endocrine subtype composition and cell-intrinsic gene expression profiles that were indicated by a shifted cell cluster location in the Uniform Manifold Approximation and Projection (UMAP) space of mSTZ-diabetic compared to healthy mice (Fig. 4a and Supplementary Table 3). In particular, there was a threefold decrease in the proportion of β cells from mSTZ-treated mice (β-mSTZ), and β-mSTZ formed a cluster that was clearly distinct from healthy β cells (Fig. 4a). In contrast to the results for β cells, few transcriptional changes were detected for α cells, δ cells and PP cells (Extended Data Fig. 7a-f and Supplementary Table 3). We next sought to describe the progression from healthy to dysfunctional β cells and the associated gene expression changes, using single-cell trajectory inference. Cells were ordered on the basis of a cell-to-cell distance metric that was calculated using the concept of diffusion pseudotime (see Methods). We identified a cellular trajectory in which cells transitioned from mature to immature to β-mSTZ cells (Fig. 4b). Remaining β-mSTZ cells expressed low Ins1 and/or Ins2 messenger RNA and also showed sustained low expression of β-cell identity transcription factor genes, such as Pdx1, Nkx2.2, Nkx6.1, Pax6, Isl1 and NeuroD1 (refs. 26,27,28,29,30,31) (Fig. 4b). Pathways that are associated with β-cell maturity and functionality included genes with downregulated expression, whereas genes in ER stress and oxidative phosphorylation pathways were upregulated in the remaining β-mSTZ cells; together these observations indicate an ER stress response and β-cell dysfunction (Fig. 4c and Supplementary Table 3). Along this trajectory, expression of key markers of β-cell maturity and functionality gradually decreased concomitantly with an increase in the very few known markers of β-cell immaturity and dedifferentiation (for example Aldh1a3 (ref. 32) and gastrin33) (Fig. 4b,d). Strikingly, our single-cell analysis uncovered a large number of upregulated genes and pathways in β-mSTZ cells that are not expressed at all or are expressed only subtly in mature, functional murine β cells (Fig. 4e). We confirmed the increased expression of, for example, Cck and Slc5a10 proteins by immunohistochemistry in mSTZ diabetic mice (Fig. 4f). These identified targets may serve as biomarkers for dysfunctional β cells, and may have the potential to be part of druggable pathways to restore β-cell function. Some of these were also identified recently in β cells and pancreata of T1D and T2D human specimens34,35 (Extended Data Fig. 7g,h). There is some debate as to whether β-cell dedifferentiation resembles reversal to a pluripotent (Oct3/4, Nanog, Sox2) or endocrine progenitor state (Neurogenin 3 or Neurog 3, hereafter called Ngn3), whether it is part of normal phenotypic variation described as β-cell heterogeneity, or whether it resembles a glucotoxic-induced reversible state36. Although dedifferentiated β cells had been characterised previously by upregulation of pluripotency or endocrine transcription factors8,9,15, in our study expression levels of Sox9, Pou5f1 (Oct3/4), Myc and Ngn3 genes were unaltered in mSTZ-treated β cells (Extended Data Fig. 7i).
To further characterise the maturation state of mSTZ-derived β cells, we compared our data set to β-cell expression profiles during embryonic (E17.5) to postnatal development (P60)37. We assessed transcriptional similarity of β-cell subpopulations using Partition-based graph abstraction (PAGA) after data integration and computation of a common embedding. The enforcement of integration using only genes associated to β-cell maturation in the reference data allowed us to match maturation states independent of differences in other biological processes between the two data sets (see Methods). PAGA uses a statistical model to measure the relatedness of groups of single cells (see Methods). We found that dedifferentiated β-mSTZ cells were more strongly connected to early time points of the maturation data set, whereas β2 cells clustered to intermediate time points and β1 cells to late time points (Fig. 5a). Remarkably, the inferred cellular trajectory from β-mSTZ to β2 to β1 cells aligned with the trajectory of β-cell maturation from embryonic (E17.5) to mature β cells (P60) of the reference data set (Fig. 5b). An increase in known β-cell maturation and a decrease of immaturity markers along this trajectory further indicated that the cells follow a similar differentiation or dedifferentiation programme (Fig. 5c,d). To validate these findings, we scored each β cell using the gene sets that are characteristic for the start (E17.5/P0) and end (P60) point of the developmental trajectory (Supplementary Table 4 and Methods). Healthy mature β cells (β1) scored high for maturity genes, whereas healthy immature cells (β2) and dedifferentiated cells (β-mSTZ) scored higher for the embryonic immaturity gene set (Fig. 5e). Taken together, these results imply that during the transition from healthy β1 to β2 to dedifferentiated β-mSTZ, β cells revert, at least in part, back to a more immature and, further, to an embryonic-like state.
To separate the altered maturation state from other processes induced in dedifferentiated β cells, we compared differentially regulated gene ontologies and pathways between embryonic (E17.5/P0) and mature (P60) β cells and between β-mSTZ cells and healthy control β cells (see Methods). According to the trajectory and PAGA analysis, these β-cell states of the reference data set correspond best to dedifferentiated β-mSTZ cells and mature, healthy β cells, respectively. Embryonic and mSTZ-diabetic β cells shared downregulation of molecular processes connected to β-cell function and maturity (for example, insulin secretion and FoxO signalling) compared to healthy mature β cells (Extended Data Fig. 8a and Supplementary Table 4), whereas genes involved in oxidative phosphorylation and gene and protein transcription were upregulated (Extended Data Fig. 8b and Supplementary Table 4). Specific to embryonic β cells was a downregulation of lipid and carbohydrate metabolism and an upregulation of WNT signalling (Extended Data Fig. 8a,b). This corresponds to known mechanisms of β-cell maturation during embryogenesis37. Interestingly, in mSTZ-diabetic β cells, but not the embryonic β cells, we found an upregulation of pathways and ontologies associated with ER stress and a decreased expression of genes involved in insulin and MAPK signalling in comparison to healthy mature β cells (Extended Data Fig. 8a,b).
Thus, β-cell dedifferentiation involves partial reversal to an embryonic or immature β-cell programme and upregulation of an ER stress response and altered signalling state. These results suggest that surviving β cells are dedifferentiated, which shows that mSTZ-induced diabetes is a good model in which to study mechanisms of β-cell dedifferentiation and redifferentiation in the absence of genetic lesions.
Mechanisms of β-cell redifferentiation
In line with the pharmacological data, single-cell analysis of the different treatments revealed that β cells of mice treated with vehicle (Extended Data Fig. 9a), oestrogen (Extended Data Fig. 9b) and GLP-1 (Extended Data Fig. 9c) remained dedifferentiated. By contrast, we observed an increased fraction of immature β2-cells from GLP-1–oestrogen-treated mice (Extended Data Fig. 9d). In PEG–insulin-treated mice (Extended Data Fig. 9e) and GLP-1–oestrogen plus PEG–insulin-cotreated mice (Extended Data Fig. 9f), almost no dedifferentiated β cells remained and most cells clustered with immature β2-cells. To further assess the transcriptional state of β cells from the treated mice, we calculated a cell-to-cell distance so that cells could be ordered along the cellular trajectory from dedifferentiated to healthy β cells (see Methods). On this one-dimensional axis, β cells of mice treated with PEG–insulin or with the combination of PEG–insulin and GLP-1–oestrogen were located closest to β cells from healthy mice (Fig. 6a–c). This transcriptional similarity to healthy β cells was further supported by PAGA (see Methods). In the PAGA graph, β cells of PEG–insulin-treated and GLP-1–oestrogen plus PEG–insulin-cotreated mice showed the strongest connection to healthy β cells (Fig. 6d). The observed overall re-establishment of the healthy β-cell expression profiles was substantiated by an increased expression of β-cell maturity markers and decreased expression of immaturity and dedifferentiation markers along the inferred trajectory (Fig. 6e–g). Moreover, Ucn3 expression recovered during the pharmacological treatment (Extended Data Fig. 10a). This shows that the maturation state before treatment was different from that achieved after treatment. Hence, upon PEG–insulin or PEG–insulin plus GLP-1–oestrogen treatment, β cells adopt a molecular immature yet functional phenotype that is sufficient for blood glucose normalisation and diabetes remission.
β cells from mice treated with PEG–insulin or the combination of PEG–insulin and GLP-1–oestrogen were grouped in distinct β-cell subpopulations, albeit at a similar maturation state (Fig. 6b). This implies the existence of a compound-specific mechanism-of-action (MOA) that underlies the recovery of β-cell function. To investigate the distinct MOAs of the different treatments, we identified the β-cell-specific transcriptional signature of treated and mSTZ-derived β cells (Supplementary Table 5). An increased expression of genes involved in β-cell functionality and maturation (Fig. 6h) was common to both treatments and may result from improved blood glucose levels and/or stimulation of shared pathways of insulin and GLP-1–oestrogen signalling. GLP-1, oestrogen and insulin receptor activation regulate MAPK and FoxO signalling38,39,40, both of which were increased after PEG–insulin, GLP-1–oestrogen and PEG–insulin and GLP-1–oestrogen co-therapy (Fig. 6h). Although we cannot dissect the signalling contribution of each individual receptor, we believe that the polypharmacological approach might potentiate the simultaneous activation of commonly regulated pathways.
Unexpectedly, treatment with PEG–insulin elicited a β-cell-specific stimulation of the insulin signalling cascade as well as stimulation of the recently characterised RNA polymerase II mediated pathway41 (Fig. 6h). Hence, our data suggest that direct effects of insulin on β cells contribute to the improvement of β-cell function and recovery, as was proposed for T2D42,43.
Our goal was to use GLP-1–oestrogen to selectively deliver oestrogen into β cells. We consistently found β-cell-specific induction of the ER-associated degradation (ERAD) pathway and induction of transfer RNA signalling in PEG–insulin and GLP-1–oestrogen cotreated mice (Fig. 6h). ERAD mitigates ER stress, which, when unresolved, contributes to functional β-cell mass loss in T1D and T2D44. Chemical and genetic disturbances of ERAD impair β-cell function44. We found an increased proinsulin-to-C-peptide ratio in mSTZ-diabetic mice, which is used as an ER-stress surrogate in diabetes45 (Extended Data Fig. 10b). GLP-1–oestrogen, PEG–insulin and GLP-1–oestrogen and PEG–insulin co-therapy normalised this ratio (Extended Data Fig. 10b). Recently, it has been reported that oestrogen, via nuclear oestrogen receptor alpha signalling, stabilizes the ERAD proteins Sel1l and Hrd1in β cells, an effect associated with diabetes amelioration in Akita mice46. We observed increased co-staining for insulin and Sel1l in GLP-1–oestrogen and PEG–insulin cotreated islets only 25 d after treatment initiation (Extended Data Fig. 10c). In addition to upregulation of Sel1l and Hrd1 (also known as Syvn1), ERAD-associated gene expression (for example, expression of Sdf2l1, Herpud1, Dnajb11, Dnajb9, Derl3 and Hspa5 genes) was specifically increased in β cells of GLP-1–oestrogen and PEG–insulin cotreated mice (Extended Data Fig. 10d). The Sdf2l1, Herpud1 and Hspa5 genes encode ERAD-proteins that have beneficial effects on β cells that allow correct insulin folding and/or function47,48,49. Dnajb9 and Dnajb11 are chaperone proteins that might aid correct insulin protein folding. Derl3 is required for ER-associated degradation50. The specific role of Derl3 in β cells is unknown, but interestingly, Derl3 expression was shown to protect cardiomyocytes against ER-stress-induced death by enhancing ERAD50. These results support a role for ERAD activation by GLP-1–oestrogen and PEG–insulin cotreatment that induces a treatment-specific molecular profile for the protection and regeneration of β cells (Fig. 6b).
In a similar manner, tRNA signalling is known to be an intracellular target of oestrogen, and an increased abundance of tRNA has been associated with proliferating cells51,52. Indeed, from the single-cell data, we observed the highest fraction of proliferative β cells in the PEG–insulin and GLP-1–oestrogen conjugate cotreated mice (Fig. 7a). Moreover, we found increased β-cell proliferation in GLP-1–oestrogen and PEG–insulin cotreated mice that was already evident after 25 d of treatment, and that, importantly, was not evident in the single-treatment groups (Fig. 7b). Stressed β cells, such as those under chronic hyperglycaemic conditions, lack an adequate response to GLP-1 therapy probably as a result of decreased expression of GLP-1R53,54,55. Here, we reasoned that the restoration of glycaemia in mSTZ mice, notably through additional chronic insulin therapy, may increase the expression of GLP-1R. Indeed, we found progressively increased levels of GLP-1R on the surfaces of β cells of mice with improved glucose levels (Fig. 7c). This may have facilitated the enhanced delivery, uptake and action of oestrogen and GLP-1 in β cells, especially for those cells cotreated with GLP-1–oestrogen and insulin.
To examine whether other endocrine cells have contributed to the regeneration of functional β cells, we explored cluster relations and possible cellular transitions using PAGA and RNA velocity estimation (see Methods). For this we included β-mSTZ cells as the origin (starting point) of treated cells and we investigated where cells moved from that baseline (Fig. 8a). We found no direct connection or cell movement from other (non-β) cell populations towards redifferentiated β cells. We next examined the RNA velocity and potential fate of treated endocrine cells (Fig. 8b). The velocity of some of the immature β cells observed after GLP-1–estrogen, PEG–insulin and combined treatment pointed towards mature β cells of healthy mice, thus further substantiating β-cell redifferentiation. Moreover, the scRNA-seq data suggested that neogenesis was not increased after 100 d of treatment, as the expression levels of Ngn3 mRNA remained unchanged in endocrine cell subtypes (Fig. 8c). We also found no indication, from Ngn3 immunostaining in tissue sections of earlier time points, that overt neogenesis contributed to β-cell regeneration (Fig. 8d). Together, these results suggest that redifferentiation of β cells along dedifferentiation and redifferentiation trajectories is the main mechanism that underlies the re-establishment of functional β cells in the mSTZ model by treatment with GLP-1–oestrogen or PEG–insulin, or by their cotreatment. By using a combination of low dose insulin with GLP-1–oestrogen treatment, we were able to trigger a β-cell-specific transcriptional response that was characterised by increased β-cell proliferation and enhanced functionality.
Herein we have established the mSTZ model of diabetes as a model to study β-cell dysfunction and dedifferentiation. Single-cell profiling of remaining β cells identified many markers of β-cell dedifferentiation that were undescribed previously and that code for surface molecules, receptors and secreted proteins. These may be used as biomarkers or may allow the detection, isolation and characterisation of dedifferentiated β cells. This could reveal pathomechanisms of T1D and T2D and may have the potential to identify unique diagnostic markers and therapeutic targets. Using scRNA-seq we were able to delineate a β-cell fate trajectory in which cells transitioned from mature to immature to dedifferentiated β cells, which implies that β cells can be characterised by a continuum of transcriptional states that reflect discrete phenotypes. Inference of cell transitions using the RNA velocity concept further suggested that there was no ongoing transdifferentiation from other non-β and non-endocrine cells towards dedifferentiated β cells. Upregulation of the endocrine master regulator Ngn3 might depend on the severity of hyperglycaemia; high glucose levels ( >33 mM) have been shown to induce Ngn3 expression9,15, whereas lower levels ( <25 mM) had no effect14,56. We show that β-cell dedifferentiation in mSTZ-diabetic mice is independent of induction of Ngn3+ endocrine progenitors and that the transcriptional state of dedifferentiated β cells is more similar to late embryonic or early postnatal β cells.
Recently, the Kushner laboratory has provided evidence that some of the remaining insulin in the blood stream of patients with long-term T1D57 originates from dedifferentiated β cells and/or from polyhormonal non-β cells that function as ‘insulin microsecretors’58. Similarly, the Korsgren laboratory found histological evidence for β-cell dedifferentiation at T1D onset59. Therefore, triggering the redifferentiation of dedifferentiated β cells seems to be an intuitive approach for the treatment of diabetes that does not involve β-cell proliferation or neogenesis per se60. Preclinical as well as clinical findings from patients with type 1 and 2 diabetes suggest that a transient recovery of β-cell dysfunction occurs upon glycaemia normalisation by intensive insulin treatment by either β-cell rest or redifferentiation3,15. By using scRNA-seq we can dissect endocrine subtype-specific treatment responses and show that insulin treatment triggers transcriptional changes in β cells that are connected to insulin and/or IRS signalling. This supports the idea that in addition to lowering the glucotoxic stress on β cells, direct insulin- or IGF-signalling improves β-cell health and performance and can redifferentiate β-cell mass in diabetic models43. Importantly, the redifferentiated β cells that were induced by insulin therapy were functional and responded to physiological stimuli, as indicated by increased fasting plasma C-peptide levels.
Moreover, we show that targeted delivery of oestrogen using GLP-1 as a peptide carrier and intensive insulin co-therapy by a distinct MOA alleviates hyperglycaemia, increases fasting C-peptide levels and redifferentiates β cells, whereas it reduces daily insulin requirements by 60% and limits weight gain in mice. The enhanced restoration of GLP-1R expression in dedifferentiated β cells by the GLP-1–oestrogen and insulin cotreatment renders them susceptible to targeted delivery of oestrogen. As has been proposed previously in the Akita mouse model46, we observed that stimulating the ERAD pathway by GLP-1–oestrogen beneficially influences β-cell physiology in rodent models of diabetes. In future studies, it might be of specific interest to test GLP-1–oestrogen with and without PEG–insulin co-therapy in genetically perturbed mouse models of ERAD. Finan et al. showed previously that peptide-based targeting prevented adverse side effects of oestrogen, such as uterus and tumour growth19. There was also no measurable oestrogen-induced increase in bone content owing to the effect of the absence or limited expression of GLP-1R on off-target tissues and cells19. Here, we extended the safety profile of GLP-1–oestrogen and demonstrated that GLP-1–oestrogen did not stimulate uterine tissue growth in OVX rats. This study further verified that there is insufficient free, systemic oestrogen to drive toxicity as well as the selectivity and specificity of GLP-1–mediated oestrogen targeting. The strategy to use GLP-1 as a carrier may be adopted to selectively target any other small molecule or biologic to β cells. The prerequisite for the transport of the molecule of interest into the target cell (that is, stressed and dedifferentiated β cells) is adequate GLP-1R expression. Under hyperglycaemic conditions, adjunctive treatments that reduce the glycaemic burden, such as chronic insulin therapy as demonstrated here, can facilitate the restoration of GLP-1R expression in stressed β cells. Notably, chronic PEG–insulin treatment also increased functional β-cell number and the scRNA-seq data suggest a direct effect of insulin on β cells. Thus, combinatorial pharmacological treatments that include insulin might have additional beneficial effects on β-cell survival, protection, proliferation and function. In summary, our work has identified mechanisms and pathways of β-cell dedifferentiation and opens avenues for pharmacological targeting of these dedifferentiated cells for diabetes remission.
mSTZ (Sigma-Aldrich cat. no. S0130) was injected intraperitoneally in 8-week-old male C57BLJ/6 mice (n = 125) at 50 mg kg−1 for five consecutive days following the mSTZ model to induce diabetes. A subset of age-matched male mice was injected with ice-cold citrate buffer (pH 4.5) as control animals (n = 20). C57BLJ/6 mice were obtained from Janvier Labs. Ten days after the last mSTZ injection, fasting blood glucose was taken, as well as fasting plasma, to determine fasting insulin and C-peptide levels. We included hyperglycaemic mice with fasting blood glucose levels of >190 mg dl−1 (n = 116). We estimated β-cell function and mass of mSTZ-treated mice by combining fasting blood glucose levels, the homeostatic model assessment (HOMA)-β-score and the ratio of fasting C-peptide to blood glucose levels. Among mSTZ-treated mice, animals with fasting glucose levels of >25th percentile, a HOMA-β-score of <25th percentile and a C-peptide to blood glucose ratio of <25th percentile were excluded from the study (n = 9).
Pharmacological study in mSTZ mice
mSTZ diabetic mice were randomised and evenly distributed to different treatments according to fasting blood glucose levels. Ten days after the last STZ injection, mice were allocated to different treatments of daily subcutaneous injection with vehicle (PBS; n = 17, not mSTZ-treated), vehicle (PBS; n = 17, mSTZ-treated), a GLP-1 analogue (n = 16), oestrogen (n = 14), GLP-1–oestrogen (n = 28, of which n = 11 mice were switched to vehicle (PBS) treatment after 12 weeks of GLP-1–oestrogen treatment), PEG–insulin (n = 13), or GLP-1–oestrogen and PEG–insulin (n = 16) at the doses indicated in Fig. 1a for 100 d. Mice were housed up to four per cage on a 12:12-h light-dark cycle at 22 °C and given free access to a normal chow diet (Altromin 1314) and water. Compounds were administered in a vehicle of PBS (Gibco) and were given by daily subcutaneous injections at the indicated doses at a volume of 5 μl g−1 body weight for the indicated durations as indicated in the figure legends. The investigators were not blinded to group allocation during the in vivo experiments or to the assessment of longitudinal endpoints. All rodent studies were approved by and performed according to the guidelines of the Animal Use and Care Committee of Bavaria, Germany.
Study in FVFPBFDHom mice
Male 8-week old FVFPBFDHom mice with fasting blood glucose > 250 mg dl−1 were randomised to vehicle (n = 7), oestrogen (n = 5), GLP-1 (n = 9) or GLP-1–oestrogen (n = 11) treatment according to their fasting blood glucose levels. FVFPBFDHom mice were obtained from breeding. Mice were treated daily for four weeks with subcutaneous injections. Fasting blood glucose was measured after a 6-h fast. We single- or group-housed the mice on a 12-h light, 12-h dark cycle at 22 °C and provided free access to food and water. This study was approved by and performed according to the guidelines of the Animal Use and Care Committee of Bavaria, Germany.
Uterotrophic assessment in OVX rats
The study was designed in accordance with the Endocrine Disruptor Screening Programme Test Guidelines OPPTS 890.1600: Uterotrophic Assay, a standardised in vivo screening test intended to evaluate the ability of a chemical to elicit biological activities consistent with agonists of natural oestrogens (for example, 17ß-oestradiol). The test measures the increase in uterine weight or uterotrophic response in comparison with non-treated controls61. The study was further designed according to accepted pharmacological principles and followed Good Laboratory Practice (conducted by Envigo). A total of 44 OVX female Sprague-Dawley rats (Charles River) were supplied for the study, of which 40 animals were allocated to treatment groups (randomised by body weight to ensure equal group mean starting body weight) and the remaining 4 animals were allocated as spares. On the day of dosing (following 14–22 d of acclimatisation), rats were approximately 9–12 weeks of age, and weighed 217 g to 348 g. Four groups of 8 rats each were treated for 14 consecutive days with once-daily subcutaneous administration of GLP-1–oestrogen at doses of 400 μg kg−1, 2,200 μg kg−1 and 4,000 μg kg−1 per day or with volume-matched vehicle (PBS). An additional group of animals received once-daily subcutaneous injections of vehicle for the first 10 d followed by once-daily subcutaneous injection of 17 alpha-ethynyl oestradiol (positive control) at 300 μg kg−1 d−1 on days 11 to 14. Animals were weighed daily from day 1 (prior to dosing) until the day of necropsy, and food consumption was recorded on days −4, 1, 4, 8, 11 and 15 (day of necropsy). Standard toxicological organ weight measurement was carried out at necropsy, including the weighing of wet and dry (blotted) uterine tissue. The uterus was sampled and weighed according to OPPTS 890.1600.
Administration of EdU
To investigate cell proliferation, we used the modified uracil analogue 5´ethynyl-2´-desoxyuridine (EdU). EdU was injected intraperitoneal (i.p.) at 50 µg kg body weight −1 24 h prior to sacrifice.
Compound formulation and peptide synthesis and cleavage
GLP-1 and GLP-1–oestrogen
Peptides were synthesised by solid-phase peptide synthesis methods using in situ neutralisation for fluorenylmethoxycarbonyl (Fmoc)-based chemistries. For Fmoc-based neutralization peptide synthesis, we used 0.1 mmol Rink MBHA resin (0.5–0.7 mmol g–1, 100–200 mesh, Novabiochem) on a Symphony peptide synthesiser by standard Fmoc methods using diisopropylcarbodiimide/1-hydroxybenzotriazole chloride (DIC/HOBt-Cl, both from AAPPTec) in dichloromethane (DMF, VWR Chemicals) for coupling and 20% piperidine in DMF for deprotection of N-terminal amines. Completed peptidyl resins were treated with triflouracetic acid (TFA, VWR Chemicals), triisopropylsilane (TIS, Sigma-Aldrich) and water (9.0:0.5:0.5 (v/v/v)) for 2 h with agitation. We confirmed the peptide molecular weights and character by liquid chromatography–mass spectrometry on an Agilent 1260 Infinity/6120 Quadrupole instrument with a Kinetex C8 column and a gradient of 10%–80% eluent B. Eluent A is water with 0.05% TFA and eluent B is 10% water, 90% MeCN, 0.05% TFA. Preparative HPLC purifications were performed on a Waters instrument with a Kinetex 5 μ C8.
To synthesise the derivatized oestrogen for construction of the conjugate, we reacted oestradiol 17-acetate (Sigma) and a tenfold excess of ethyl 2-bromoacetate in dioxane in the presence of K2CO3 under reflux conditions and agitation. After the removal of dioxane in vacuo, we resuspended the intermediate product in dioxane with 1 N NaOH with heat, followed by a re-acidification by HCl in dichloromethane. After the removal of DCM in vacuo, the crude product was resuspended in aqueous solvent containing at least 20% MeOH, 20% acetronitrile (ACN, VWR Chemicals) and 1% AcOH. The crude extract was purified by reversed-phase HPLC.
The derivatized oestrogen was added covalently to the side-chain amine on the C-terminal lysine amide residue. We used a C-terminal N-methyltrityl-l-lysine (Lys(Mtt)-OH) residue, whose side chain was orthogonally deprotected with four 10-min treatments with 1.5% TFA, 2% TIS and 1% anisole in DMF. The oestrogen was attached through this side-chain amine of the C-terminal lysine after treatment with a threefold excess of the purified derivatized oestrogen and DIC/HOBt-Cl in N-methyl- 2-pyrrolidone. After TFA cleavage as described above, the crude extract was resuspended in aqueous buffer containing 20% ACN and 0.1 M NH4HCO3. The conjugate was purified as described above.
Pegylated insulin (PEG–insulin) was prepared by insulin N-terminal amine reductive amination with 20 K methoxy PEG propionaldehyde. In brief, human insulin was dissolved in 50 mM sodium acetate buffer (pH 5.0) and 50% ACN. A 30-fold excess of sodium cyanoborohydride and a 1.5-fold excess of methoxy PEG propionaldehyde (M-ALD-20K, JenKem Technology) was added to the buffer containing insulin for 3 h at room temperature with stirring. Purification by reverse phase chromatography on a C-8 column in 0.1% TFA acetonitrile solvents yielded the final product at greater than 95% purity.
Oestrogen (17β-oestradiol, Sigma-Aldrich) was dissolved in 100% ethanol (Sigma-Aldrich) at a concentration of 1 mg ml−1 and was diluted with PBS to the required concentration.
Blood was collected from tail veins after a 4-h fast, using EDTA-coated microvette tubes (Sarstedt). Blood was immediately chilled on ice. Plasma was separated by centrifugation at 5,000g at 4 °C for 10 min using a micro centrifuge and was stored at −20 °C until further usage. Plasma insulin and C-peptide (Crystal Chem) and Proinsulin (Alpco) were quantified by ELISA (enzyme-linked immunosorbent assay) following the manufacturer’s instructions. Four-hour fasting blood glucose levels were determined using a handheld glucometer (FreeStyle).
Adult pancreata were dissected and fixed in 4% PFA in PBS for 24 h at 4 °C. The tissues were cryoprotected in a sequential gradient of 7.5%, 15% and 30% sucrose-PBS solutions at room temperature (2 h incubation for each solution). Next, the pancreata were incubated in 30% sucrose and tissue embedding medium (Leica) (1:1) at 4 °C overnight. Afterwards, they were embedded in a cryoblock using tissue-freezing medium (Leica), frozen in dry ice and stored at −80 °C. Sections of 20 μm thickness were cut from each sample, mounted on a glass slide (Thermo Fisher Scientific) and dried for 10 min at room temperature before use or storage at −20 °C.
Immunostaining of sections
The cryosections were rehydrated by three washes with 1× PBS, and were permeabilised with 0.2–0.15% Triton X-100 in H2O for 30 min. Permeabilisation was not performed for staining with GLP-1R. Then, the samples were blocked in blocking solution (PBS, 0.1% Tween-20, 1% donkey serum, 5% FCS) for 1 h. The following primary antibodies were used: guinea pig polyclonal anti-insulin (1:300, Thermo Fisher Scientific), goat polyclonal anti-Glut2 (1:500, Abcam), goat polyclonal anti-Nkx6.1 (1:200, R&D Systems), goat polyclonal anti-somatostatin (1:500, Santa Cruz), rat monoclonal anti-somatostatin (1:300, Invitrogen), rabbit polyclonal anti-urocortin 3 (1:300, Phoenix Pharmaceuticals), rabbit monoclonal anti-insulin (1:300, Cell Signaling), guinea pig polyclonal anti-glucagon (1:500, Takara), guinea pig polyclonal anti-insulin (1:300, ABD Serotec), rabbit polyclonal cleaved caspase-3 (Asp 175) (1:300, Cell Signaling), rabbit monoclonal anti-ki67 (1:300, Abcam), rabbit polyclonal anti-Aldh1a3 (1:300, Abcam), rabbit monoclonal anti-GLP-1R (1 µg ml−1, Novo Nordisk), rabbit polyclonal anti-gastrin (1:100, Abcam), rabbit polyclonal anti-cholecystokinin (1:100, ENZO), goat polyclonal anti-Sel1l (1:300, Novus Biologicals) and rabbit anti-Ngn3 (1:800, donated by H. Edlund). Dilutions were prepared in blocking solution and sections were incubated overnight at 4 °C. Thereafter, sections were rinsed 3× and washed 3× with 1× PBS. All secondary antibodies were used at a 1:800 dilution prepared in blocking buffer. We used the following secondary antibodies: donkey anti-goat IgG (H + L) secondary antibody (Alexa Fluor 633, Invitrogen A-2108), donkey anti-rabbit IgG (H + L) secondary antibody (Alexa Fluor 555, Invitrogen A-31572); donkey anti-rabbit IgG (H + L) secondary antibody (Alexa Fluor 488, Invitrogen A-21206); donkey anti-guinea pig IgG (H + L) secondary antibody (DyLight 649, Dianova, 706–495–148); donkey anti-rat IgG (H + L) secondary antibody (DyLight 647 Dianova 711–605–152); donkey anti-rat IgG (H + L) secondary antibody (Cy3, Dianova 712–165–153); donkey anti-guinea pig (H + L) secondary antibody (Alexa Fluor 488, Dianova 706–545–148). After 4–5 h of incubation, pancreatic sections were stained with DAPI (1:500 in 1× PBS) for 30 min, rinsed and washed 3× with 1× PBS, and subsequently mounted. All images were obtained with a Leica microscope of the type DMI 6000 using LAS AF software (Leica). Images were analysed using the LAS AF and/or ImageJ software programme.
Automatic tissue analysis
Stained tissue sections were scanned with an AxioScan.Z1 digital slide scanner (Zeiss) equipped with a ×20 magnification objective. We scanned three sections per animal. Images were evaluated using the commercially available image analysis software Definiens Developer XD 2 (Definiens). First, regions of interest were annotated manually to select islets of Langerhans for analysis. A specific rule set was then defined to detect and quantify the cells within each defined region on the basis of the fluorescence intensity of DAPI, morphology, size and neighbourhood. The Ins-, Gcg- or Sst-expressing cells were classified automatically using the fluorescence intensity of each hormone.
EdU detection protocol
EdU staining was carried out according to the EdU imaging kit manual (Life Technologies) after staining with the secondary antibody. DAPI staining and mounting was performed as described above.
Pancreatic insulin content
Pancreatic insulin content was determined by an acid ethanol extraction. The pancreas was dissected, washed in 1× PBS and homogenised in an acid-ethanol solution (5 ml 1.5% HCl in 70% ethanol) followed by incubation at −20 °C for 24 h. After 2 rounds of acid-ethanol precipitation, the tissue was centrifuged (2000 r.p.m., 15 min, 4 °C) and the supernatant was neutralised with 1 M Tris pH 7.5. Insulin was measured using a mouse insulin ELISA (Crystal Chem) and was normalised over the protein concentration that was determined by BCA protein assay.
Islet isolation was performed by collagenase P (Roche) digestion of the adult pancreas. In brief, 3 ml of collagenase P (1 mg ml−1) was injected into the bile duct and the perfused pancreas was consequently dissected and placed into another 3 ml of collagenase P for 15 min at 37 °C. Then, 10 ml of G-solution (HBSS (Lonza) + 1% BSA (Sigma-Aldrich)) was added to the samples, and this was followed by centrifugation at 1,600 r.p.m. at 4 °C. After another washing step with G-solution, the pellets were re-suspended in 5.5 ml of gradient preparation (5 ml 10% RPMI (Lonza) + 3 ml of 40% Optiprep (Sigma-Aldrich) per sample), and placed on top of 2.5 ml of the same solution. To form a 3-layer gradient, 6 ml of G-solution was added on the top. Samples were then incubated for 10 min at room temperature before centrifugation at 1,700 r.p.m. Finally, the interphase between the upper and the middle layers of the gradient was harvested and was filtered through a 70 μm Nylon filter then washed with G-solution. Islets were hand picked under the microscope.
To achieve a single-cell suspension of islets, islets were hand picked into a 1.5 ml Eppendorf tube, pelleted (800 r.p.m., 1 min) washed with PBS (minus Mg or Ca, Gibco) and digested with 0.25% trypsin with EDTA (Gibco) at 37 °C for 8 min. Mechanical disaggregation every 2–3 min was required. The digestive reaction was then stopped and cells were pelleted (1200 r.p.m., 5 min).
Single-cell libraries were generated using the Chromium Single-cell 3′ library and gel bead kit v2 (PN 120237) from 10x Genomics. In brief, to reach a target cell number of 10,000 cells per sample 16,000 cells per sample were loaded onto a channel of the 10x chip to produce Gel Bead-in-Emulsions (GEMs). This underwent reverse transcription to barcode RNA before cleanup and cDNA amplification followed by enzymatic fragmentation and 5′ adaptor and sample index attachment. Libraries were sequenced on the HiSeq4000 (Illumina) with 150 bp paired-end sequencing of read2.
FACS sorting of endocrine cells was performed using the FACS-Aria III (BD Biosciences). Single cells were gated according to their FSC-A (front scatter area) and SSC-A (side scatter area). Singlets were gated dependent on the FSC-W (front scatter width) and FSC-H (front scatter height) and dead cells were excluded using the marker 7AAD (eBioscience). The FVF endocrine populations were discriminated according to their Venus fluorescence emission at 488 nm and the β- and α-lineages were discriminated according to their BFP emission at 405 nm (positive and negative respectively). To enrich for β cells the distinct SSC-A high populations were gated. FACS sorted cells were sorted directly into Qiazol (Qiagen) for RNA isolation.
RNA isolation and cDNA preparation
The mRNA isolation was performed using the miRNeasy Micro Kit (Qiagen) according to the manufacturer’s instructions. On-column DNase I treatment was applied to degrade DNA. For cDNA preparation the SuperScript Vilo cDNA Synthesis Kit (Thermo Fisher Scientific) and the GoScript Reverse Transcript System Kit (Promega) were used. The cDNA synthesis was carried out according to both manufacturers’ instructions.
Quantitative PCR (qPCR) was carried out using the Viia7 Real Time PCR System (Thermo Fisher Scientific) and the following TaqMan probes (Life Technologies): Ins 1 (Mm01950294_s1), Ins 2 (Mm00731595_Gh), Gcg (Mm01269055_m1), Sst (Mm00436671_m1), Ppy (Mm01250509_g1), Ghrelin (Mm00445450_m1), Pecam1 (Mm01242584_m1), Gapdh (Mm99999915_g1) and 18S (Mm03928990_g1). Each reaction contained 25 ng of cDNA. For analysis, the Ct-values were transformed to the linear expression values and were normalised to the reference genes (Gapdh and 18S) and to the control samples.
Reaggregated human micro-islets
All primary human islets were obtained through Prodo Laboratories with no information on the identity of the donor for ethical and privacy reasons (donor 1: male, BMI 32.38, age 48, HbA1c 5.6%; donor 2: male, BMI 33.2, age 46, HbA1c 5.4%; donor 3: male, BMI 28.65, age 34, HbA1c 5.2%). For all donors, consent was obtained from next of kin. For each preparation of InSphero 3D InSight human islet microtissues, 10,000–20,000 islet equivalents were dispersed in dissociation solution (1× TrypLE Express solution (Thermo Fisher Scientific 12604013) with 40 µg ml−1 DNase I (Sigma-Aldrich 10104159001)) by gentle pipetting at 37 °C. Remaining cell clumps were removed by filtering the cell suspension through a cell strainer (70 μm pore size). Islet microtissues were produced by hanging-drop-based scaffold-free reaggregation of 2,500 cells in each well of the InSphero 96-well Hanging Drop System for 5 d. The primary aggregates were then transferred to the Akura 96 well-plate to further mature for at least another 8 d before the start of the experiments. All experiments were performed within 30 d after the start of the aggregation. Islet microtissues were maintained in 3D InSight Human Islet Maintenance Medium (InSphero).
Compound, cytokine treatments and GSIS with human micro-islets
Dilution series of compounds were performed in 3D InSight Human Islet Maintenance Medium. Each of the assessed compounds was added to the culture medium at the concentration indicated below, one day prior to the start of the cytokine treatment. The cytokine cocktail contained: tumour necrosis factor alpha (TNFα, 10 ng ml−1, Thermo Fisher Scientific PHC3016), interferon gamma (IFNγ, 10 ng ml−1, Sigma-Aldrich I3265) and interleukin-1beta (IL-1β, 2 ng ml−1, Sigma-Aldrich I17001), was prepared in PBS containing 0.1% BSA (Sigma-Aldrich A7888). The same concentration of PBS–BSA solution was maintained in each experimental condition. Regular redosings with cytokines and compounds were performed every 2–3 d. Prior to GSIS, culture medium was removed and islet microtissues were washed twice with Krebs Ringer Hepes Buffer (KRHB; 131 mM NaCl, 4.8 mM KCl, 1.3 mM CaCl2, 25 mM Hepes, 1.2 mMm KH2PO4, 1.2 mM MgSO4, 0.5% BSA) containing 2.8 mM glucose and equilibrated for 1 h in the same solution. GSIS was performed in KRHB containing indicated glucose concentrations for 2 h. Following GSIS, the tissues were lysed using the CellTiter-Glo Luminescent Cell Viability Assay (Promega G9241) with protease inhibitor cocktail (Promega G6521) and the luminescence was recorded with a microplate reader (Infinite M1000, TECAN) for the analysis of total ATP content. The lysates were then used for assessment of total insulin content. After the correct dilutions in KRHB were performed, total and secreted insulin was quantified using the Stellux Chemi Human Insulin ELISA (Alpco, 80-INSHU-CH10). The Caspase-Glo 3/7 Assay (Promega, G8090) was used to assess caspase-3 and caspase-7 activity in the islet microtissues following compound treatment.
Statistical analysis not including scRNA-seq data
Preliminary data processing and calculations during ongoing studies were carried out using Microsoft Excel 2016. All further statistical analyses were performed using GraphPad Prism 8. We used the one-way analysis of variances (ANOVA) followed by Tukey’s post hoc analysis to determine significance among the different treatment groups. In the case of only two groups, the unpaired Student two-tailed t-test was used to detect significant differences. The human micro-islets were derived from three different donors, each of whom naturally varied in their GSIS. To compare the treatment effects among all donors, we used one-way ANOVA with the different donors as random effect followed by Tukey’s post hoc analysis. This analysis was performed in R. Grubbs’ test (α < 0.05) was used to detect significant outliers, which were then excluded from subsequent statistical analysis and figure drawing. P < 0.05 was considered to be statistically significant. All results are mean ± s.e.m. unless otherwise indicated.
Preprocessing of droplet-based scRNA-seq data
Demultiplexing of binary base call (BCL) files, alignment, read filtering, barcode and unique molecular identifier (UMI) counting were performed using the CellRanger analysis pipeline (v2.0.0) provided by 10x Genomics. High quality barcodes were selected on the basis of the overall distribution of total UMI counts per cell using the standard CellRanger cell detection algorithm. All further analyses were run with python3 using the scanpy package62 (v1.0.4 + 92.g9a754bb, https://github.com/theislab/scanpy) except stated otherwise. Genes with expression in less than ten cells were excluded. Furthermore, as is also applied as standard preprocessing steps in scanpy tutorials, low quality or outlier cells were removed if they: (1) had a high fraction of counts from mitochondrial genes (40% or more); (2) expressed more than 7,000 genes; or (3) had more than 100,000 UMI counts. Cell-by-gene count matrices of all samples were then concatenated to a single matrix. To account for differences in sequencing depth, UMI counts of each cell were normalised by total counts of that cell (pp.normalize_per_cell with mean = TRUE) and values were log-transformed. Highly variable genes (n = 1,625) were selected on the basis of normalised dispersion using the setting the lower cutoffs for the mean to 0.0125 and for the dispersion to 0.5. This matrix was used as input for all further analyses unless otherwise indicated.
Embedding, clustering and cell type annotation
Clustering was performed on the full data set to reduce systematic biases such as batch effects as was recommended in ref. 63. A single-cell neighbourhood graph (kNN-graph) was computed on the 50 first principal components using 15 neighbours. To minimise condition effects and to facilitate clustering we recomputed the kNN-graph using the first 15 diffusion components of the PCA-based graph as suggested in ref. 64. For clustering and cell type annotation, Louvain-based clustering65 was used as implemented in louvain-igraph (v0.6.1 https://github.com/vtraag/louvain-igraph) and adopted by scanpy (tl.louvain). The resolution parameter was varied in different parts of the data manifold to account for strong changes in resolution (for details, see Data availability). Clusters were annotated on the basis of the mRNA expression of the four main hormone genes Ins1 and Ins2, Gcg, Sst and Ppy (endocrine cells) and other known marker genes (non-endocrine cells) and were merged if they reflected heterogeneity only in a cell type outside the focus of this study.
Ductal cells (that express Krt19), acinar cells (that express Prss2), endothelial cells (that express Plvap), stellate cells (that express Col1a2) and small clusters of potential doublet-like cells that co-express endocrine and non-endocrine markers were removed from further analysis. Immune cells (that express Cd74) that infiltrate the islets were finely subclustered into macrophages (that express Cd86, Adgre1 and Cd14), dendritic cells (that express Cd86, Itgax and Iftim3), B cells (that express Cd79a and Cd79b) and T cells (that express Cd8a and Cd3d).
The hormone genes Ins1 and Ins2, Gcg, Sst and Ppy were expressed at very high levels and showed background level expression in all other endocrine subtypes and non-endocrine cell types. Such background expression is a common phenomenon in droplet-based scRNA-seq data. It is commonly said to be due to free-floating mRNA in the single-cell solution that comes from lysed cells and that is incorporated into all droplets. For annotation, only hormone expression that was well above the background level in non-endocrine cells, such as ductal, immune and endothelial cells, was considered.
For the identification of β-cell substates a new kNN-graph on the first 50 principal components was calculated and put into Louvain-based clustering of both Ins monohormonal and the connected Ins-PP cells. Similarly, Ins-Sst cells were subclustered from Ins-Sst-PP cells after recalculating the kNN-graph on the first 50 principal components. Ins-Gcg-Sst cells were assigned using a manual threshold for all three hormones that was well above ambient levels (threshold = 6 for normalised data).
For visualisation, UMAP was run as recommended in ref. 66. For each UMAP-plot the UMAP was newly calculated by recomputing the kNN-graph on the represented cell subset using the first 50 principal components.
Identification of polyhormonal singlets and doublet-like endocrine cell clusters
Polyhormonal cells have previously been reported to exist in pancreatic islets67,68,69. However, the expression of multiple hormones in the same droplet can also be an indication of a doublet. It can therefore be difficult to distinguish polyhormonal singlets from doublets. A doublet rate of ~8–10% was measured in experiments with the same concentration of cells using the 10x technology70. This rate includes doublets with contributions from two different cell types (here, polyhormonal doublets) and from the same cell type (here, monohormonal doublets). The monohormonal doublets resemble monohormonal singlets and do not affect subsequent analyses71. We calculated the expected doublet frequency of polyhormonal doublets for a doublet rate of 10% using the frequency of monohormonal cell types that contribute to the doublet (doublet contributors) and assuming that doublets are generated by sampling singlet cells uniformly at random71. In every sample the proportion of observed polyhormonal cells clearly exceeded the expected polyhormonal doublet frequency. Therefore, in our data set it is unlikely that all detected polyhormonal cells are doublets. Application of doublet cell detection tools Scrublet71 (v0.1, https://github.com/AllonKleinLab/scrublet) and DoubletDetection (https://github.com/JonathanShor/DoubletDetection) failed to resolve which clusters represent doublets and which clusters represent polyhormonal singlets. Predictions of the tools disagreed with each other and the doublet rate was consistently overestimated. We therefore used the following criteria to evaluate polyhormonal cell clusters and to distinguish between singlets and doublets:
(1) Doublets should not express unique genes. All genes should also be expressed in at least one doublet contributor.
(2) Doublets should express marker genes or lineage-determining transcription factors of the doublet contributors. Downregulation of these genes indicates singlet populations.
(3) Previous reporting of polyhormonal singlet cells in the literature.
(4) Clusters of polyhormonal cells with a higher frequency than expected by our doublet simulation indicate polyhormonal singlet clusters (Extended Data Fig. 6g).
(5) Clusters with Scrublet doublet score distributions that are comparable to monohormonal singlet clusters indicate polyhormonal singlets (Extended Data Fig. 6h).
On the basis of these criteria we found sufficient evidence for Ins-PP, Ins-Sst-PP, Gcg-PP (low) and Gcg-PP (high) cells to be polyhormonal singlets, but we excluded Ins-PP-Gcg, Ins-Gcg, Ins-Gcg-Sst, Gcg-Sst-PP and Sst-PP(high) cells as probable doublets.
To classify cells into cycling and non-cycling cells, first, a score was assigned to each cell for a set of S and G2/M phase genes72 as proposed73, and second, all cells with an S-score or a G2/M-score > 0.25 were classified as cycling. The threshold was chosen on the basis of the score distribution. The score for a given gene set was computed as described74 and implemented in scanpy (tl.score_genes_cell_cycle).
Marker genes of the main endocrine cell types
For the characterisation of the four endocrine cell types, specific marker genes were identified by comparing the gene expression profile of each cell type against all cells of the other three cell types using a test with overestimated variance as implemented in scanpy (tl.rank_genes_groups). All genes that ranked within the top 300 genes, had a test score of >8 and were unique markers for one cell type were considered as marker genes.
Differential expression testing to describe subpopulations and treatment responses
Differential expression testing between treatments and for the characterisation of immature β cells and polyhormonal subpopulations was performed on quantile-normalised (quantile threshold = 0.95) and log-transformed data to account for extremely highly expressed genes (for example, the main hormones in endocrine cells) that may incorrectly alleviate the expression of other genes in a cell when total count normalisation is applied. By quantile normalisation, each cell is normalised by the total UMI count in the cell, of genes that account for less than 5% of the total UMI counts across all cells. Thus, very highly expressed genes are not considered for normalisation. For differential expression testing we used limma-trend75,76, as implemented in the Bioconductor package limma (v3.28.10), through an rpy2 (v2.9.1) interface as recommended in ref. 77. In each test only genes expressed in >1% of cells in any of the two subsets tested were considered. Gene set enrichment was performed using EnrichR78 through its web interface. Genes with a false discovery rate (FDR) of <0.01 and an estimated log(fold change) (output from limma model, not the actual log(fold change), as log-transformed data were the input) of >0.25 were used as input. Notably, the hormone genes Ins1 and Ins2, Gcg, Sst and Ppy, as well as other known cell type marker genes Pyy, Iapp, Ttr, Gpx3, Ctrb1 and Try5 were also differentially expressed in other cell-types in which they are expressed only at background levels (free-floating mRNA, see Embedding, clustering and annotation). These genes are indicated in the Supplementary Tables and were excluded from plotting.
Identification of specific β-cell dedifferentiation markers
Genes specific for dedifferentiated β cells (β-mSTZ) were extracted from the list of all significantly upregulated genes (FDR < 0.05, estimated log(fold change) > 0.25) in β cells from mSTZ-treated mice compared to β cells of healthy mice by two filtering steps. First, non-specific genes that were also differentially expressed in any of the other monohormonal endocrine cell types (α, δ and PP) were excluded. Second, only genes that were expressed in at least 25% of β cells from mSTZ-treated mice and in less than 5% of β cells from healthy mice were considered. Location was extracted from the GeneCards database (https://www.genecards.org).
Inference of β-cell maturation, dedifferentiation and regeneration trajectories
Pseudotime of β-cell maturation in healthy islets and dedifferentiation upon mSTZ treatment was calculated using diffusion pseudotime (dpt)79,80 as implemented in scanpy (tl.dpt), selecting a random root cell within the starting population. The choice of root cell did not affect the inferred pseudotemporal ordering strongly. Similarly, the dpt approach was used to model β-cell regeneration and to estimate the location of treated β cells along the path from dedifferentiated to mature β cells. Here, dpt was used as a cell-to-cell distance metric across samples. Cycling cells as well as a small subpopulation of β cells were excluded from visualisation as they were not part of the linear trajectory and they showed very high pseudotime values.
Comparison of β-cell dedifferentiation trajectory to embryonic and postnatal maturation
To compare the dedifferentiation trajectory to embryonic and postnatal maturation we used a publicly available single-cell RNA-seq data set as a reference that contained cells that were sorted from Gcg-Venus and Ins-GFP reporter mice at 6 different time points (E17.5, P0, P3, P9, P18 and P60)37. The filtered and annotated raw count matrix was downloaded from the Gene Expression Omnibus (GEO) (GEO accession number: GSE87375). The analysis was run using the updated scanpy package v1.4.4, as only this version includes the necessary data integration methods. ERCC RNA spike-in and genes with expression in fewer than three cells were excluded. The data were normalised to total counts per cell using the pp.normalize_total function in scanpy with default parameters and excluding highly expressed genes, and were log-transformed (pp.log1p). For the scope of this manuscript we used a subset of the data that contained only β cells (Ins1-positive cells). Therefore, we computed a kNN-graph on the 50 first principal components using 15 neighbours and performed a first round of Louvain-based clustering (tl.louvain). As input, data were subset to the 3,000 top ranking highly variable genes (pp.highly_variable_genes). We excluded all clusters that showed high expression of Gcg, which is indicative of α cells, or that showed high expression of Mki67, which is indicative of proliferative cells. In addition, we filtered cells that showed high expression of the δ cell markers Sst or Hhex.
We integrated this reference data set with the data from our study, which was subset to β cells from Ctrl and mSTZ-Vehicle treatment, using BBKNN81 from the scanpy external package (sce), which allowed us to compute an embedding and trajectory that included both data sets. To reduce noise, we excluded genes that were expressed in fewer than 15 cells in each data set. In addition, cycling cells (Mki67 > 1, 33 reference cells and 30 of the cells from this study) were excluded, as their expression profile is dominated by the expression of cell-cycle genes and these cells therefore formed a separate cluster which was not part of the linear maturation trajectory. To ensure that β-cell maturation dominates the gene expression variation we first considered only the 2,000 top-ranked highly variable genes (pp.highly_variable_genes) of the reference data set that were also expressed in our data, and, second, we reduced the contribution of heterogeneity within the β1 and β-mSTZ cluster to gene expression variation by randomly subsetting both clusters to 500 cells. We then scaled and zero centred each data set separately (pp.scale) and concatenated the two data sets, which resulted in a 1,788 cells by 1,654 genes count matrix. We computed a common kNN-graph on the first ten principal components using the sce.pp.bbknn function with default parameters and k = 5 within batch neighbours. As the data sets did not show a strong batch effect even without integration, assessed by visual inspection of the first principal components and diffusion components, and thus transitions between cells from the two data sets also showed a high probability, we were able to use diffusion pseudotime (tl.dpt) to infer the maturation trajectory with the common kNN-graph as input. The trajectory was calculated by selecting a random root cell from the embryonic cells that were sampled at E17.5. The choice of root cell did not affect the inferred ordering strongly. The ordering of the cells from each data set was largely consistent with the ordering obtained prior to integration, assessed by the distribution along the trajectory of the time points or β-cell subgroups. To quantify the cluster similarity, PAGA was applied to the common kNN-graph (tl.paga).
To compute an embryonic or immaturity cell score and a maturity cell score we extracted the gene signatures from the reference data set. The reference data was subset to the 2,000 top ranked highly variable genes and Louvain-based clustering was performed on the kNN-graph that was computed on the first 50 principal components with 15 neighbours. The cluster that consisted of cells that were sampled at E17.5 and P0 was annotated as ‘embryonic or neonatal’, whereas the cluster consisting mainly of cells that were sampled at P60 was annotated as ‘mature’. Differentially expressed genes between these two clusters were used for scoring. Genes that were upregulated in the mature cluster or in the embryonic or neonatal cluster were used as a gene set for maturity or for embryonic or immaturity score, respectively. For differential expression testing the t-test with overestimated variance implemented in the tl.rank_genes_groups function of scanpy was used. The top 500 ranked genes with a log(foldchange) > 0.25 and an adjusted P value < 0.01 were considered. Cell scores were computed using the tl.score_genes function in scanpy.
Inference of cluster-to-cluster distances, lineage relations and cell movement
PAGA64 was performed to infer cluster and lineage relations using the tl.paga function of scanpy with a threshold on edge significance of 0.05. In a PAGA graph, paths represent cluster connections or relations that indicate potential routes of cell transitions. Edge weights represent the confidence of a connection calculated on the basis of a measure of cluster connectivity.
To infer the direction of possible transitions82 and cell movements we estimated RNA velocity using a stochastic version implemented in the scVelo python package (v0.1.16.dev13 + c1a6dad, https://github.com/theislab/scvelo with scanpy v1.3.2). Splicing information of reads was extracted using the velocyto pipeline (v0.17.7, http://velocyto.org). We then followed the recommended steps described in scVelo to estimate RNA velocities and RNA velocity force fields. First, data were preprocessed by filtering genes with less than 30 spliced or 30 unspliced counts and both unspliced and spliced counts were normalised by total counts. Then the first- and second-order moments for each cell were computed across its 15 nearest neighbours of the kNN graph in PC space (50 PCs). Next, RNA velocities were estimated using a stochastic model of transcriptional dynamics. To obtain a more conservative estimate a 95% quantile fit was used. Finally, to project the velocity vector of each cell into the low-dimensional UMAP embedding for visualisation and interpretation, the expected mean direction given all potential cell transitions on the kNN graph was computed. Each potential cell transition is assigned a probability corresponding to the correlation to the predicted transition by the velocity vector (velocity graph). For example, a high probability corresponds to a high correlation with the velocity vector. The projection results in a low dimensional map of RNA velocity which indicates the predicted cell state transitions. For computation of the velocity graph and embedding only genes with an r2 > 0.1 of the velocity fit were considered.
The velocities of each gene were calculated over all treatments except for healthy β cells, for which only healthy cells were used. A treatment can here be considered as a process by which cells move from the diseased cells potentially towards healthy cells, as for the pseudotime inference described above. During this process genes are induced and/or repressed, as approximated by RNA velocity. Therefore, to also take into account these intermediate gene states, all treatments were included for model fitting and velocity estimation. Both PAGA and the RNA velocity graph and projection were instead only computed on the represented cell subset. For this, the kNN-graph was recalculated for the cell subset using the first 50 principal components and the highly variable genes as initially defined.
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Custom python scripts written for performing scRNA-seq data analysis are available in a github repository (https://github.com/theislab/pancreas-targeted_pharmacology). Versions of packages that might influence numerical results are indicated in the scripts. Raw data and gene expression matrices of scRNA-seq are deposited in GEO under the accession number GSE128565. Source data for Figs. 1–3 and 7 and Extended Data Figs. 1–5 and 10 are provided with the paper.
Matveyenko, A. V. & Butler, P. C. Relationship between β-cell mass and diabetes onset. Diabetes Obes. Metab. 10, 23–31 (2008).
Herold, K. C. et al. An anti-CD3 antibody, teplizumab, in relatives at risk for type 1 diabetes. N. Engl. J. Med. 381, 603–613 (2019).
Harrison, L. B., Adams-Huet, B., Raskin, P. & Lingvay, I. β-cell function preservation after 3.5 years of intensive diabetes therapy. Diabetes Care 35, 1406–1412 (2012).
Chen, H.-S. et al. Beneficial effects of insulin on glycemic control and beta-cell function in newly diagnosed type 2 diabetes with severe hyperglycemia after short-term intensive insulin therapy. Diabetes Care 31, 1927–1932 (2008).
The Diabetes Control and Complications Trial Research Group. Effect of intensive therapy on residual β-cell function in patients with type 1 diabetes in the diabetes control and complications trial: a randomized, controlled trial. Ann. Intern. Med. 128, 517–523.
Weng, J. et al. Effect of intensive insulin therapy on β-cell function and glycaemic control in patients with newly diagnosed type 2 diabetes: a multicentre randomised parallel-group trial. The Lancet 371, 1753–1760 (2008).
Alvarsson, M. et al. Beneficial effects of insulin versus sulphonylurea on insulin secretion and metabolic control in recently diagnosed type 2 diabetic patients. Diabetes Care 26, 2231–2237 (2003).
Rui, J. et al. β cells that resist immunological attack develop during progression of autoimmune diabetes in NOD mice. Cell Metab. 25, 727–738 (2017).
Talchai, C., Xuan, S., Lin, H. V., Sussel, L. & Accili, D. Pancreatic β cell dedifferentiation as a mechanism of diabetic β cell failure. Cell 150, 1223–1234 (2012).
Cinti, F. et al. Evidence of β-cell dedifferentiation in human type 2 diabetes. J. Clin. Endocrinol. Metab. 101, 1044–1054 (2016).
Like, A. A. & Rossini, A. A. Streptozotocin-induced pancreatic insulitis: new model of diabetes mellitus. Science 193, 415–417 (1976).
Thorel, F. et al. Conversion of adult pancreatic α-cells to β-cells after extreme β-cell loss. Nature 464, 1149–1154 (2010).
Chera, S. et al. Diabetes recovery by age-dependent conversion of pancreatic δ-cells into insulin producers. Nature 514, 503–507 (2014).
Brereton, M. F. et al. Reversible changes in pancreatic islet structure and function produced by elevated blood glucose. Nat. Commun. 5, 4639 (2014).
Wang, Z., York, N. W., Nichols, C. G. & Remedi, M. S. Pancreatic β cell dedifferentiation in diabetes and redifferentiation following insulin therapy. Cell Metab. 19, 872–882 (2014).
Tiano, J. P. & Mauvais-Jarvis, F. Importance of oestrogen receptors to preserve functional β-cell mass in diabetes. Nat. Rev. Endocrinol. 8, 342–351 (2012).
Chon, S. & Gautier, J.-F. An update on the effect of incretin-based therapies on β-cell function and mass. Diabetes Metab. J. 40, 99–114 (2016).
Marso, S. P. et al. Semaglutide and cardiovascular outcomes in patients with type 2 diabetes. N. Engl. J. Med. 375, 1834–1844 (2016).
Finan, B. et al. Targeted estrogen delivery reverses the metabolic syndrome. Nat. Med. 18, 1847–1856 (2012).
Clemmensen, C. et al. Emerging hormonal-based combination pharmacotherapies for the treatment of metabolic diseases. Nat. Rev. Endocrinol. 15, 90–104 (2019).
Bastidas-Ponce, A. et al. Foxa2 and Pdx1 cooperatively regulate postnatal maturation of pancreatic β-cells. Mol. Metab. 6, 524–534 (2017).
Blum, B. et al. Functional beta-cell maturation is marked by an increased glucose threshold and by expression of urocortin 3. Nat. Biotechnol. 30, 261–264 (2012).
Nishimura, W. et al. A switch from MafB to MafA expression accompanies differentiation to pancreatic beta-cells. Dev. Biol. 293, 526–539 (2006).
Bader, E. et al. Identification of proliferative and mature β-cells in the islets of Langerhans. Nature 535, 430–434 (2016).
Roscioni, S. S., Migliorini, A., Gegg, M. & Lickert, H. Impact of islet architecture on β-cell heterogeneity, plasticity and function. Nat. Rev. Endocrinol. 12, 695–709 (2016).
Ediger, B. N. et al. Islet-1 is essential for pancreatic β-cell function. Diabetes 63, 4206–4217 (2014).
Gao, T. et al. Pdx1 maintains β cell identity and function by repressing an α cell program. Cell Metab. 19, 259–271 (2014).
Gu, C. et al. Pancreatic β cells require NeuroD to achieve and maintain functional maturity. Cell Metab. 11, 298–310 (2010).
Gutiérrez, G. D. et al. Pancreatic β cell identity requires continual repression of non–β cell programs. J. Clin. Invest. 127, 244–259 (2016).
Swisa, A. et al. PAX6 maintains β cell identity by repressing genes of alternative islet cell types. J. Clin. Invest. 127, 230–243 (2016).
Taylor, B. L., Liu, F.-F. & Sander, M. Nkx6.1 is essential for maintaining the functional state of pancreatic beta cells. Cell Rep. 4, 1262–1275 (2013).
Kim-Muller, J. Y. et al. Aldehyde dehydrogenase 1a3 defines a subset of failing pancreatic β cells in diabetic mice. Nat. Commun. 7, 12631 (2016).
Dahan, T. et al. Pancreatic β-cells express the fetal islet hormone gastrin in rodent and human diabetes. Diabetes 66, 426–436 (2017).
Solimena, M. et al. Systems biology of the IMIDIA biobank from organ donors and pancreatectomised patients defines a novel transcriptomic signature of islets from individuals with type 2 diabetes. Diabetologia 61, 641–657 (2018).
Camunas-Soler, J.et al. Pancreas patch-seq links physiologic dysfunction in diabetes to single-cell transcriptomic phenotypes. Preprint at bioRxiv https://doi.org/10.1101/555110 (2019).
Weir, G. C. & Bonner-Weir, S. Islet β cell mass in diabetes and how it relates to function, birth, and death: islet β cell mass in diabetes. Ann. N. Y. Acad. Sci. 1281, 92–105 (2013).
Qiu, W.-L. et al. Deciphering pancreatic islet β cell and α cell maturation pathways and characteristic features at the single-cell level. Cell Metab. 25, 1194–1205.e4 (2017).
Taniguchi, C. M., Emanuelli, B. & Kahn, C. R. Critical nodes in signalling pathways: insights into insulin action. Nat. Rev. Mol. Cell Biol. 7, 85–96 (2006).
Rowlands, J., Heng, J., Newsholme, P. & Carlessi, R. Pleiotropic effects of GLP-1 and analogs on cell signaling, metabolism, and function. Front. Endocrinol. 9, 672 (2018).
Segars, J. H. & Driggers, P. H. Estrogen action and cytoplasmic signaling cascades. Part I: membrane-associated signaling complexes. Trends Endocrinol. Metab. 13, 349–354 (2002).
Hancock, M. L. et al. Insulin receptor associates with promoters genome-wide and regulates gene expression. Cell 177, 722–736.e22 (2019).
Kulkarni, R. N. et al. Altered function of insulin receptor substrate-1–deficient mouse islets and cultured β-cell lines. J. Clin. Invest. 104, R69–R75 (1999).
Ueki, K. et al. Total insulin and IGF-I resistance in pancreatic β cells causes overt diabetes. Nat. Genet. 38, 583–588 (2006).
Fonseca, S. G., Gromada, J. & Urano, F. Endoplasmic reticulum stress and pancreatic β-cell death. Trends Endocrinol. Metab. 22, 266–274 (2011).
Sims, E. K. et al. Elevations in the fasting serum proinsulin-to-C-peptide ratio precede the onset of type 1 diabetes. Diabetes Care 39, 1519–1526 (2016).
Xu, B. et al. Estrogens promote misfolded proinsulin degradation to protect insulin production and delay diabetes. Cell Rep. 24, 181–196 (2018).
Tiwari, A. et al. SDF2L1 interacts with the ER-associated degradation machinery and retards the degradation of mutant proinsulin in pancreatic β-cells. J. Cell Sci. 126, 1962–1968 (2013).
Ho, D. V. & Chan, J. Y. Induction of Herpud1 expression by ER stress is regulated by Nrf1. FEBS Lett. 589, 615–620 (2015).
Wong, N., Morahan, G., Stathopoulos, M., Proietto, J. & Andrikopoulos, S. A novel mechanism regulating insulin secretion involving Herpud1 in mice. Diabetologia 56, 1569–1576 (2013).
Belmont, P. J. et al. Roles for endoplasmic reticulum-associated degradation and the novel endoplasmic reticulum stress response gene Derlin-3 in the ischemic heart. Circ. Res. 106, 307–316 (2010).
Zhu, D. et al. Single-cell transcriptome analysis reveals estrogen signaling coordinately augments one-carbon, polyamine, and purine synthesis in breast cancer. Cell Rep. 25, 2285–2298.e4 (2018).
Torrent, M., Chalancon, G., de Groot, N. S., Wuster, A. & Madan Babu, M. Cells alter their tRNA abundance to selectively regulate protein synthesis during stress conditions. Sci. Signal. 11, eaat6409 (2018).
Xu, G. et al. Downregulation of GLP-1 and GIP receptor expression by hyperglycemia: possible contribution to impaired incretin effects in diabetes. Diabetes 56, 1551–1558 (2007).
Fritsche, A., Stefan, N., Hardt, E., Häring, H. & Stumvoll, M. Characterisation of beta-cell dysfunction of impaired glucose tolerance: evidence for impairment of incretin-induced insulin secretion. Diabetologia 43, 852–858 (2000).
Kjems, L. L., Holst, J. J., Vølund, A. & Madsbad, S. The influence of GLP-1 on glucose-stimulated insulin secretion: effects on beta-cell sensitivity in type 2 and nondiabetic subjects. Diabetes 52, 380–386 (2003).
Jonas, J. C. et al. Chronic hyperglycemia triggers loss of pancreatic beta cell differentiation in an animal model of diabetes. J. Biol. Chem. 274, 14112–14121 (1999).
Keenan, H. A. et al. Residual insulin production and pancreatic β-cell turnover after 50 years of diabetes: Joslin Medalist Study. Diabetes 59, 2846–2853 (2010).
Lam, C. J., ChatterjeeA., ShenE., Cox, A. R. & Kushner, J. A. Low-level insulin content within abundant non-á islet endocrine cells in long-standing type 1 diabetes. Diabetes 68, 598–608 (2019).
Seiron, P. et al. Characterisation of the endocrine pancreas in type 1 diabetes: islet size is maintained but islet number is markedly reduced. J. Pathol. Clin. Res. 5, 248–255 (2019).
Zhou, Q. & Melton, D. A. Pancreas regeneration. Nature 557, 351–358 (2018).
Waaseth, M. et al. Hormone replacement therapy use and plasma levels of sex hormones in the Norwegian Women and Cancer postgenome cohort: a cross-sectional analysis. BMC Womens Health 8, 1 (2008).
Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).
Büttner, M., Miao, Z., Wolf, F. A., Teichmann, S. A. & Theis, F. J. A test metric for assessing single-cell RNA-seq batch correction. Nat. Methods 16, 43–49 (2019).
Wolf, F. A. et al. PAGA: graph abstraction reconciles clustering with trajectory inference through a topology preserving map of single cells. Genome Biol. 20, 59 (2019).
Blondel, V. D., Guillaume, J.-L., Lambiotte, R. & Lefebvre, E. Fast unfolding of communities in large networks. J. Stat. Mech. Theory Exp. 2008, P10008 (2008).
Becht, E. et al. Dimensionality reduction for visualizing single-cell data using UMAP. Nat. Biotechnol. 37, 38–44 (2018).
Chiang, M.-K. & Melton, D. A. Single-cell transcript analysis of pancreas development. Dev. Cell 4, 383–393 (2003).
Katsuta, H. et al. Single pancreatic beta cells co-express multiple islet hormone genes in mice. Diabetologia 53, 128–138 (2010).
Alpert, S., Hanahan, D. & Teitelman, G. Hybrid insulin genes reveal a developmental lineage for pancreatic endocrine cells and imply a relationship with neurons. Cell 53, 295–308 (1988).
Zheng, G. X. Y. et al. Massively parallel digital transcriptional profiling of single cells. Nat. Commun. 8, 14049 (2017).
Wolock, S. L., Lopez, R. & Klein, A. M. Scrublet: computational identification of cell doublets in single-cell transcriptomic data. Cell Syst. 8, 281–291.e9 (2019).
Kowalczyk, M. S. et al. Single-cell RNA-seq reveals changes in cell cycle and differentiation programs upon aging of hematopoietic stem cells. Genome Res. 25, 1860–1872 (2015).
Tirosh, I. et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 352, 189–196 (2016).
Satija, R., Farrell, J. A., Gennert, D., Schier, A. F. & Regev, A. Spatial reconstruction of single-cell gene expression data. Nat. Biotechnol. 33, 495–502 (2015).
Law, C. W., Chen, Y., Shi, W. & Smyth, G. K. voom: precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biol. 15, R29 (2014).
Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 43, e47–e47 (2015).
Soneson, C. & Robinson, M. D. Bias, robustness and scalability in single-cell differential expression analysis. Nat. Methods 15, 255–261 (2018).
Kuleshov, M. V. et al. Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res. 44, W90–W97 (2016).
Haghverdi, L., Büttner, M., Wolf, F. A., Buettner, F. & Theis, F. J. Diffusion pseudotime robustly reconstructs lineage branching. Nat. Methods 13, 845–848 (2016).
Tritschler, S. et al. Concepts and limitations for learning developmental trajectories from single cell genomics. Development 146, dev170506 (2019).
Polański, K. et al. BBKNN: fast batch alignment of single cell transcriptomes. Bioinformatics 36, 964–965 (2019).
La Manno, G. et al. RNA velocity of single cells. Nature 560, 494–498 (2018).
We thank L. Müller, L. Sehrer, E. Malogajski and M. Kilian from the Helmholtz Diabetes Center in Munich for excellent assistance with in vivo mouse experiments. We thank J. Jaki, C. Salinno, F. Volta, J. Beckenbauer, A. Savoca and R. Fimmen for excellent assistance with in vitro experiments. We thank C. Pyke and P. Gottrup Mortensen for providing the GLP-1R antibody. We thank V. Bergen, M. Lücken, L. Simon and D. Fischer for fruitful discussions on the computational analysis. This work was supported in part by funding to M.H.T. from the Alexander von Humboldt Foundation and the Initiative and Networking Fund of the Helmholtz Association and funding by the European Research Council (AdG HypoFlam grant no. 695054). In addition, this work was supported by funds from the Helmholtz future topic ‘Aging and Metabolic programming’, the Helmholtz Alliance ICEMED, the Helmholtz Initiative on Personalised Medicine, iMed, by the Helmholtz Association, and the Helmholtz cross-programme topic ‘Metabolic Dysfunction’. M.A.S.-G. was funded by a Marie Sklodowska-Curie individual Fellowship (grant no. 706965, GCG-T3 Dyslipidemia). F.J.T. acknowledges support by the BMBF (grant no. 01IS18036A and grant no. 01IS18053A), by the German Research Foundation (DFG) within the Collaborative Research Centre 1243, Subproject A17, by the Helmholtz Association (Incubator grant sparse2big, grant no. ZT-I-0007) and by the Chan Zuckerberg Initiative DAF (advised fund of Silicon Valley Community Foundation, 182835).
C.B.J., M.C., B.Y, B.F., and R.D.D. are current employees of Novo Nordisk. Novo Nordisk has licensed from Indiana University intellectual property pertaining to this report. M.H.T. serves as a scientific advisory board member of ERX Pharmaceuticals, Inc., Cambridge, MA. F.J.T. reports receiving consulting fees from Roche Diagnostics GmbH and Cellarity Inc., and ownership interest in Cellarity, Inc. and Dermagnostix. S.T. reports receiving consulting fees from Cellarity, Inc. The Institute for Diabetes and Obesity receives research support from Novo Nordisk. All other authors declare no conflict of interest.
Peer review information Primary Handling Editor: Elena Bellafante.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Effects of either vehicle or the mSTZ treatment on a, fasting blood glucose (No STZ: n = 20, mSTZ: n = 107; unpaired two-sided t-test; t = 14.64, df = 125), b, pancreatic islets histology (No STZ: 179 islets of n = 3 mice, mSTZ: 182, n = 3; unpaired two-sided t-test; β: t = 11.44, df = 358; α: t = 10.98, df = 356; δ: t = 4.27, df = 338; images are representative from no STZ n = 3 and mSTZ n = 3 mice), c, the insulin positive area within pancreatic sections (No STZ: 27 sections of n = 3 mice; STZ: 27, n = 3; unpaired two-sided t- test; t = 3.646, df = 52), d, the proliferation (No STZ: 58 islets of n = 3 mice; STZ: 69, n = 3) and e, apoptosis rate in β cells (No STZ: 46 islets of n = 3 mice; STZ: 42, n = 3; unpaired two-sided t-test, t = 3.955, df = 86), f, the expression of β-cell functional marker Ucn3 and Glut2 (images are representative of dataset plotted in b from no STZ n = 3 and mSTZ n = 3 mice), g, the homoeostatic model assessment of β-cell function (HOMA-β) (No STZ: n = 20, STZ: n = 107; unpaired two-sided t-test; t = 20.65, df = 124) and h, the ratio of fasting C-peptide to fasting blood glucose (No STZ: n = 20, STZ: n = 106; unpaired two-sided t-test; t = 14.03, df = 122). Boxplots covering all data points are depicted. Line indicates the median. Scale bar, 50 μm. Scale bar zoom-in, 20 µm. Source data
Effect of treatment with indicated compounds and doses on a, fasting plasma insulin levels at week 12 of treatment (mSTZ-vehicle, n = 12; oestrogen, n = 10; GLP-1, n = 11; GLP-1/oestrogen, n = 11; one-way ANOVA with Tukey post-hoc; F (3, 39) = 10.66) and b, body weight in the end of the study (no STZ-vehicle, n = 12; mSTZ-vehicle, n = 13; GLP-1, n = 11; oestrogen, n = 11; GLP-1/oestrogen, n = 11; PEG-insulin, n = 9; GLP-1/oestrogen and PEG-insulin, n = 10; unpaired two-sided t-test; t = 2.436, df = 17). (c, d) Comparison of PEG-insulin and GLP-1/oestrogen plus PEG-insulin co-treated mice. c, Blood glucose after intraperitoneal glucose (0.5 g/kg) at week 12 (no STZ-vehicle, n = 12; PEG-insulin, n = 9; GLP-1/oestrogen and PEG-insulin, n = 10; one-way ANOVA with Tukey post-hoc (F (2, 27) = 24.71)). d, Pancreatic insulin content in the end of the study (no STZ-vehicle, n = 4; PEG-insulin, n = 4; GLP-1/oestrogen and PEG-insulin, n = 4; unpaired two-sided t-test; t = 4.534, df = 6). All data are mean ± SEM. Source data
(a, b) Treatment of female OVX Sprague-Dawley rats. a, Study scheme. b, Dry uterus weight. Data are mean ± SEM. N = 8 female rats per group. One-way ANOVA with Tukey post-hoc (F (4, 34) = 44.89). (c-g) Treatment of male FVFPBFDHom mice. (c) FACS gating strategy of dispersed endocrine cells based on granularity (Side Scatter Cell (SSC)) and PBF (405 nm) intensity. d, qPCR analysis confirmed sorting strategy of endocrine cells. Data are values of sorted cells from n = 2 mice. e, Study scheme; FVFPBFDHom male mice were treated with vehicle (n = 7), oestrogen (n = 5), GLP-1 (n = 9), or GLP-1/oestrogen (n = 11) at the indicated doses for four weeks. f, Fasting blood glucose. Data are mean ± SEM. g, Sorted endocrine cell populations after treatment (vehicle (n = 4, cells of n = 2 mice each were pooled), oestrogen (n = 4, cells of n = 2 mice each were pooled), GLP-1 (n = 5, cells of n = 2 and n = 3 mice were pooled), or GLP-1/oestrogen (n = 6, islets of n = 3 mice each were pooled)). Data are mean ± SD. Source data
Measurement of human micro-islet viability and cell death with and without cytokine exposure and in the present of different compounds at the indicated doses. (a) ATP content and (b) Caspase 3/7 activity of human micro-islets. a, N = 6 micro-islets of n = 3 human donors for each condition. Boxplot of all data points. Line indicates the median. #P indicates P-value to vehicle no stress condition. *P indicates P-value to vehicle cytokine exposure. No stress versus cytokines stress by unpaired two-sided t-test (t = 1.756, df = 33). Otherwise one-way ANOVA with donor as random effect followed by Tukey post-hoc (Flow dose(4, 81) = 6.68; Fmedium dose(4, 82) = 10.68; Fhigh dose(4, 81) = 6.30). b, N = 3–4 micro-islets of n = 3 human donors for each condition. Boxplot of all data points. Line indicates the median. #P indicates P-value to vehicle no stress condition. *P indicates P-value to vehicle cytokine exposure. No stress versus cytokines stress by unpaired two-sided t-test (t = 2.567, df = 22). Otherwise one-way ANOVA with donor as random effect followed by Tukey post-hoc (Flow dose(4, 52) = 5.58; Fmedium dose(4, 52) = 4.44; Fhigh dose(4,53) = 4.23). Source data
Representative mice (n = 3) of each treatment were used for scRNA-seq. a, Fasting glucose levels. One-way ANOVA with Tukey post-hoc test among mSTZ, oestrogen, GLP-1 and GLP- 1/oestrogen treated mice (F (3, 8) = 21.23). One-way ANOVA with Tukey post-hoc among no STZ, GLP-1/oestrogen, PEG-insulin, and co-treated mice (F (3, 8) = 94.06). b, Fasting C-peptide levels. One-way ANOVA with Tukey post-hoc test among STZ, oestrogen, GLP-1 and GLP- 1/oestrogen treated mice (F (3, 8) = 9.073). All data are mean ± SEM. Source data
a, Endocrine cell annotation is based on the hormone expression of insulin (Ins), glucagon (Gcg), somatostatin (Sst), and pancreatic polypeptide (PP). b, UMAP plot showing all endocrine cells (7578 cells in total) from healthy mice. The cell number and proportion of each endocrine cluster is indicated. c, Redefined clustering of the Ins+ β cells revealed two main β-cell subpopulations. d, Expression changes of genes from selected pathways along a pseudotime trajectory from β2- to β1 cells. β1-cells were downsampled to 1000 cells for better visualisation. e, GO term and KEGG pathway enrichment analysis of up- (log(fold change) > 0.25) and downregulated (log(fold change) < -0.25) genes in β2- (278 cells) compared to β1 cells (5380 cells). Cells were pooled from n = 3 mice. Representative terms from Supplementary table 2 are depicted. We used limma-trend to find differentially expressed genes (M&M). Gene enrichment was done with EnrichR using Fisher’s exact test to identify regulated ontologies/pathways (M&M). f, Violin plots showing the distribution of the expression of proliferation and β-cell maturation genes suggesting an immature phenotype of cycling β-cells. Accordingly, 16/403 of the β2 cells, whereas only 2/5319 of the mature β1 cells were classified as cycling (M&M). Cells were pooled from n = 3 mice. Violin shows the distribution as a kernel density estimate fit. Points in violin interior show individual data points. Boxplot in violin interior shows median, quartile and whisker values. g, Measured proportion and expected doublet frequency of polyhormonal cell clusters. Expected doublet frequency is calculated given a doublet rate of 10% (M&M). h, Boxplot displaying the doublet score distribution of mono- and polyhormonal cell clusters. A high score indicates a high doublet probability. Cells were pooled from n = 3 mice. Boxplot shows the quartile values and extreme values. Whiskers extend to 1.5 IQRs of the lower and upper quartile. Outliers are displayed individually.
(a–d), Volcano plots showing differential expression and its significance (-log10(adjusted p-Value), limma-trend) for each gene in (a) β-, (b) α-, (c) PP-, and (d) δ-cells from mSTZ treated versus healthy mice. Red line indicates thresholds used on significance level and gene expression change. Significantly regulated genes are highlighted in black. Genes significantly regulated in only one cell type but not the others are highlighted in blue. p-values were correct for multiple testing using BH. Cells were pooled from no STZ (n = 3) and mSTZ-vehicle (n = 3) treated mice. (e, f) GO term and KEGG pathway enrichment analysis of up- (log(fold change) > 0.25) and downregulated (log(fold change) < -0.25) genes in (e) α- and (f) δ cells in mSTZ treated versus healthy mice. We used limma-trend to find differentially expressed genes (M&M). Gene enrichment was done with EnrichR using Fisher’s exact test to identify regulated ontologies/pathways (M&M). Cells were pooled from no STZ (n = 3) and mSTZ-vehicle (n = 3) treated mice. (g, h) Comparison between dysregulated genes in mSTZ-β cells in mice with (g) data from RNA-seq of human T2D pancreata and (h) from scRNA-seq of human T1D β cells. Gene names of overlapping genes and identified dedifferentiation markers in Fig. 4e) are listed. i, Violin plots showing the distribution of the expression of endocrine developmental genes in beta cells of mSTZ treated and healthy mice. Violin shows the distribution as a kernel density estimate fit. Points in violin interior show individual data points. Boxplot in violin interior shows median, quartile and whisker values. Cells were pooled from no STZ (n = 3) and mSTZ-vehicle (n = 3) treated mice.
Gene ontologies (Pvalue < 0.0001) and KEGG pathways (Pvalue < 0.05) that are commonly and specifically (a) down- and (b) upregulated in embryonic and mSTZ-derived β-cells. Representative terms from Supplementary Table 4 are depicted. Gene enrichment was done with EnrichR using Fisher’s exact test to identify regulated ontologies/pathways (M&M). Cells were pooled from mSTZ-vehicle (n = 3) treated mice.
UMAP plot of all endocrine cells after 100 days of treatment showing endocrine cell distribution in each individual treatment. Total cell number for (a) mSTZ diabetic mice 5001, for (b) oestrogen treated mice 4889, for (c) GLP-1 treated mice 3874, for (d) GLP-1/oestrogen treated mice 5201, for (e) PEG-insulin treated mice 3217, and for (f) GLP-1/oestrogen (GLP-1/E) and PEG-insulin (PEG-ins) co-treated mice 3276. Values indicate the proportions of each cell cluster. Cells of n = 3 mice for each treatment were pooled.
a, Immunohistochemical analysis of Ucn3 expression during the course of the study. Scale bar, 50 μm. Day 0: Images are representative of no STZ-vehicle (n = 3) and mSTZ-vehicle (n = 3) treated mice. Day 25: Images are representative of no STZ-vehicle (n = 3), mSTZ-vehicle (n = 3), GLP-1/oestrogen (n = 3), PEG-insulin (n = 3), and GLP-1/oestrogen and PEG-insulin (n = 3) co-treated mice. Day 100: Images are representative of no STZ-vehicle (n = 3), mSTZ-vehicle (n = 3), GLP-1/oestrogen (n = 2), PEG-insulin (n = 2), and GLP-1/oestrogen and PEG-insulin (n = 3) co-treated mice. b, Plasma proinsulin/C-peptide ration in the end of the study (no STZ-vehicle, n = 8; mSTZ-vehicle, n = 8; oestrogen, n = 6; GLP-1, n = 5; GLP-1/oestrogen, n = 6; PEG-insulin, n = 6, GLP-1/oestrogen and PEG-insulin, n = 6; one-way ANOVA with Tukey post-hoc: F(6, 36) = 8.12). Data are mean ± SEM. c, Representative staining for insulin and Sel1l after 25 days of treatment. Arrow indicates Sel1l + insulin + -cells, which were especially found in GLP-1/oestrogen and PEG-insulin co-treated mice. Sel1l + insulin–cells (arrow head) were more common in mSTZ-diabetic and PEG-insulin treated mice. Images are representative of no STZ-vehicle (n = 3), mSTZ-vehicle (n = 3), PEG-insulin (n = 3), and GLP-1/oestrogen + PEG-insulin (n = 3) co-treated mice. Scale bar, 20μm. d, Expression of selected ER stress and ERAD-associated genes by scRNA-seq at study end. Source data
Statistical Source Data for each panel
Statistical Source Data for each panel
Statistical Source Data for each panel
Statistical Source Data for each panel
Statistical Source Data for each panel
Statistical Source Data for each panel
Statistical Source Data for each panel
Statistical Source Data for each panel
Statistical Source Data for each panel
Statistical Source Data for each panel
About this article
Cite this article
Sachs, S., Bastidas-Ponce, A., Tritschler, S. et al. Targeted pharmacological therapy restores β-cell function for diabetes remission. Nat Metab 2, 192–209 (2020). https://doi.org/10.1038/s42255-020-0171-3