Beta-cell specific insulin resistance promotes glucose-stimulated insulin hypersecretion

Insulin receptor (Insr) protein can be found at higher levels in pancreatic β-cells than in most other cell types, but the consequences of β-cell insulin resistance remain enigmatic. Ins1cre allele was used to delete Insr specifically in β-cells of both female and male mice which were compared to Ins1cre-containing littermate controls at multiple ages and on multiple diets. RNA-seq of recombined β-cells revealed significant differences in multiple pathways previously implicated in insulin secretion and cellular fate, including rewired Ras and NFκB signaling. Male, but not female, βInsrKO mice had reduced oxygen consumption rate, while action potential and calcium oscillation frequencies were increased in Insr knockout β-cells from female, but not male mice. Female βInsrKO and βInsrHET mice exhibited elevated insulin release in perifusion experiments, during hyperglycemic clamps, and following i.p. glucose challenge. Deletion of Insr did not reduce β-cell mass up to 9 months of age, nor did it impair hyperglycemia-induced proliferation. Based on our data, we adapted a mathematical model to include β-cell insulin resistance, which predicted that β-cell Insr knockout would improve glucose tolerance depending on the degree of whole-body insulin resistance. Indeed, glucose tolerance was significantly improved in female βInsrKO and βInsrHET mice when compared to controls at 9, 21 and 39 weeks. We did not observe improved glucose tolerance in adult male mice or in high fat diet-fed mice, corroborating the prediction that global insulin resistance obscures the effects of β-cell specific insulin resistance. We further validated our in vivo findings using the Ins1-CreERT transgenic line and found improved glucose tolerance 4 weeks after tamoxifen-mediated Insr deletion. Collectively, our data show that loss of β-cell Insr alone is sufficient to drive glucose-induced hyperinsulinemia, thereby improving glucose homeostasis in otherwise insulin sensitive dietary and age contexts.


Introduction
Type 2 diabetes is a multifactorial disease wherein several cell types, most prominently pancreatic β-cells, are dysfunctional prior to and after diagnosis (Prentki and Nolan, 2006). Hyperinsulinemia, insulin resistance, impaired fasting glucose, and impaired glucose tolerance can all be observed prior to the onset of frank diabetes (Tabak et al., 2012), but the causal relationships between these factors remain incompletely understood (Page and Johnson, 2018). Impaired insulin receptor signaling is associated with obesity and often precedes the onset of overt type 2 diabetes, but it has been studied primarily in skeletal muscle, fat, and liver where it manifests differently (Boucher et al., 2014). Recent work in mice has established that β-cell specific insulin resistance can be observed early in the progression towards type 2 diabetes, when hyperinsulinemia is prominent, and independently of insulin resistance in other tissues (Paschen et al., 2019). The physiological consequences of reduced insulin receptor (Insr)-mediated signaling in β-cells remain controversial.
It remains unresolved whether physiological insulin action on β-cells manifests as positive feedback to stimulate further insulin secretion, or negative feedback to inhibit its own release (Leibiger et al., 2008). Human studies provide evidence for both possibilities. In vivo hyperinsulinemic-euglycemic clamps have been shown to reduce circulating C-peptide, a marker of endogenous insulin secretion (Cavallo-Perin et al., 1993;Elahi et al., 1982). In some studies, this inhibition was impaired in the obese state suggesting that systemic insulin resistance also extends to the β-cells (Cavallo-Perin et al., 1993). Bouche and colleagues replicated this result at basal glucose, but also found evidence that insulin can potentiate glucose-stimulated insulin secretion under specific conditions (Bouche et al., 2010). Others have shown that insulin can either stimulate or inhibit its own secretion depending on the metabolic context (Mari et al., 2011). Administration of insulin to single β-cells in vitro increases intracellular Ca 2+ (Johnson and Misler, 2002a) and, in some studies, stimulates exocytosis (Aspinwall et al., 1999). However, Ca 2+ release from intracellular stores is not always sufficient to evoke insulin exocytosis. We previously did not detect robust exocytosis or C-peptide release from human β-cells in response to exogenous insulin despite observed changes in Ca 2+ release (Luciani and Johnson, 2005).
Whether chronic changes in autocrine insulin signaling affect β-cell development, survival and adaptation conditions is also controversial. Mice with chronically reduced insulin production have impaired β-cell expansion in the context of a high fat diet (Mehran et al., 2012). In vitro, physiologically relevant concentrations of insulin support the survival of both human and mouse β-cells (Johnson et al., 2006a). We also reported insulin is sufficient to increase proliferation of cultured primary mouse βcells and that blocking insulin secretion with somatostatin blunts proliferation induced by hyperglycemia in vitro (Beith et al., 2008) and that the majority of glucose-dependent changes in gene expression in MIN6 cells are Insr-dependent (Ohsugi et al., 2005). However, it has also been proposed hyperglycemia-induced β-cell proliferation bypasses the Insr (Jhala et al., 2003;Stamateris et al., 2016). Thus, whether these signals from insulin and/or glucose are transmitted through Insr, Igf1r, or both receptors, remains unresolved.
To address the short-and long-term consequences of eliminating Insr signaling in vivo, Kulkarni and colleagues generated mice with a floxed Insr allele and an Ins2 promoter driven Cre transgene (Kulkarni et al., 1999). Using this and related models, they reported that mice lacking β-cell Insr have profound glucose intolerance, and frank diabetes in some cases, due to impaired glucose stimulated insulin secretion, Glut2 loss, and insufficient β-cell mass (Kulkarni et al., 1999;Okada et al., 2007;Otani et al., 2004). Mice lacking Insr in β-cells (and other off-target cell types) failed to exhibit the compensatory increase in β-cell mass that accompanies a high fat diet in mice (Okada et al., 2007). Doubt was cast on some of these results when these Cre lines were subsequently shown to have offtarget tissue effects owing to endogenous Ins2 expression in the brain and thymus (Fan et al., 2009;Johnson, 2014;Mehran et al., 2012;Wicksteed et al., 2010a). More recently, Wang and colleagues studied the roles of β-cell Insr in utero and in adult mice using an inducible Ins1-CreER transgenic mouse model (Oakie et al., 2020;Wang et al., 2018), but these studies are confounded by the presence of the human growth hormone (hGH) minigene (Oropeza et al., 2015), which necessitates the use of Cre-containing controls exclusively.
In the present study, we used the constitutive Ins1 Cre knock-in strain with robust and specific recombination in β-cells (Thorens et al., 2015) to precisely reduce insulin receptor signaling and define the impacts of β-cell specific insulin resistance on glucose homeostasis. Using this approach, we found that insulin receptor signaling plays a suppressive role on insulin secretion by modulating β-cell electrical excitability and that this effect is absent in conditions of global insulin resistance.

Insr abundance in human islets and β-cells
We initiated our studies by conducting an unbiased analysis of insulin receptor abundance across tissues using publicly accessible data. Pancreatic islets had the 2 nd highest protein abundance of both isoforms of the insulin receptor across a panel of 24 human tissues, as quantified by massspectrometry (Fig. 1A). These results, which are not complicated by the limitations associated with anti-insr antibodies, show that islets have more INSR protein than 'classical' insulin target tissues, including the liver and adipose. This also supports our previous observations suggesting that insulin receptors are more abundant in β-cells relative to other cells in the pancreas . Compilation of open source public single-cell RNA sequencing data from human islets demonstrated at least one count of insulin receptor mRNA in 62.4% β-cells, alongside other islet cell types (Fig.  1A,B).

β-cell specific Insr deletion with Ins1 Cre mice
We next sought to examine the function of the Insr, and the consequences of β-cell-specific insulin resistance/sensitivity, using an in vivo β-cell specific knockout mouse model. To limit recombination of the floxed Insr allele to pancreatic β-cells, we used a Cre allele knocked into the endogenous Ins1 locus which, unlike Ins2 promoters, drives specific expression in β-cells (Thorens et al., 2015). Experimental Insr f/f ;Ins1 cre/wt ;nTnG (βInsr KO ) and Insr f/wt ;Ins1 cre/wt ;nTnG (βInsr HET ) mice and littermate control Insr wt/wt ;Ins1 cre/wt ;nTnG mice were generated using a breeding scheme to ensure consistency of the Cre donor parent (Fig. 1C). Cre-recombinase efficiency was assessed using the nuclear TdTomato-to-nuclear EGFP (nTnG) lineage trace reporter (Prigge et al., 2013) and found to be robust on the Insr f/f ;Ins1 cre/wt ;nTnG genetic background (Fig 1D). We confirmed by Western blotting that Insr protein was almost completely absent from βInsr KO islets and partially reduced from βInsr HET (Fig. 1E). qPCR showed that Insr mRNA was not decreased in any of the 18 tissues examined, including the whole brain or hypothalamus specifically, hence we did not perform Western blots for other tissues 5 (Fig. 1F). Together with other published data on Ins1 cre mice, these data strongly suggest that Insr deletion with Ins1 cre is robust and sufficiently specific to pancreatic β-cells.

Loss of β -cell Insr alters gene expression in purified β-cells
To establish a baseline gene expression profile of our β-cell specific Insr knockout model we performed an unbiased global analysis of β-cell gene expression in FACS purified GFP-positive β-cells labelled with the nTnG reporter (Fig. S1A). RNA sequencing revealed significant differences in mRNA expression between βInsr KO β-cells and wild type β-cells ( Fig. 2; Fig. S1B-E). Overall, differentially expressed genes were enriched in Reactome pathways related to NFκB signaling, Ras signaling, and mRNA splicing (Fig. S1B). In many cases, Insr gene deletion showed a gene-dose-dependent effect on regulated gene expression meaning that heterozygous βInsr HET β-cells exhibited intermediate expression of target genes between the full Insr knockout and wildtype β-cells (Fig. S1C).
Individual genes that were significantly decreased included regulators of β-cell function, including the regulation of K ATP channels, and β-cell fate (Brsk2, Rab3a, Bmpr2, Camk2b, Mapk8ip3), a negative regulator of Glut2 protein abundance (Prnp), a regulator of endocytosis (Vps8), a mediator of autophagy and cellular cholesterol traffic (Gramd1a), an anti-apoptosis gene (Rel), a transducer of ER stress (Atf6b), and a histone methyltransferase and STAT target gene (Mettl8) ( ). Several of these downregulated genes showed strong evidence for genome-wide association with glycemic traits, including Ttll12, C2, Atf6b, Sympk, and Camk2b (Fig. S1C). Interestingly, Ddx31 is a genetic regulator of circulating proinsulin levels in women but not men (Strawbridge et al., 2011). Hnf4g belongs to a network of transcription factors that play essential roles in β-cell function and fate but its specific role is unclear (Hara et al., 2000). Together, analysis of upregulated and downregulated genes points to the possibility that β-cells lacking Insr may exhibit increased glucose-stimulated insulin secretion but increased susceptibility to stress.
We also used unbiased protein-protein interaction network analysis to examine the relationships between the differentially expressed genes, as well as known components of the insulin signaling pathways. We identified key network hubs downstream of β-cell Insr (Fig. S1D). We also predicted the transcriptional regulators of Insr-regulated differentially expressed genes. Of these transcription factors Rel, Pygo2, Polr2a, Nanos1 were significantly decreased and Jak2, Thap3 significantly increased in Insr-deficient β-cells. These are likely to be the more proximal mediators of the Insrdependent transcriptional program in β-cells (Fig. S1E).

Loss of β-cell Insr increases
β -cell excitability The RNA-seq analysis of Insr-regulated genes implicated multiple factors known to promote insulin secretion, including regulators of Glut2, mitochondria, K ATP channels, and granule trafficking. We did not observe significant differences in the intensity or gross localization of Glut2 protein using immunofluorescence imaging (Fig. 3A,B). We examined mitochondrial function using the Seahorse bioanalyzer and found no significant differences between groups in females. In males, we observed a significant reduced oxygen consumption rate in dispersed islets from both βInsr KO and βInsr HET in comparison to controls during basal respiration and after antimycin A/rotenone injections, suggesting reduced ATP production (Fig. 3C). Insulin administration has been reported to open β-cell K ATP channels, mediating negative feedback on insulin secretion (Khan et al., 2001). Lack of this endogenous insulin action through Insr would therefore be expected to lead to β-cell hyper-excitability in our mouse model in the context of high glucose. Indeed, electrophysiological analysis of single β-6 cells from female βInsr KO mice confirmed a significant increase in action potential firing frequency during glucose stimulation, when compared to control β-cells, with no differences in resting potential, firing threshold, or action potential height (Fig. 3D,E). The reversal potential was right-shifted in βInsr KO β-cells, further suggesting reduced K + conductance (Fig. 3G). Interestingly, the hyper-excitability was observed in Insr knockout β-cells from female, but not male mice (Fig. 3E,F). Perhaps the reduced mitochondrial efficiency in males prevented β-cell hyperexcitability. It is also possible that male control β-cells were already maximally insulin resistant when studied. No significant differences have been identified in Insr mRNA expression between sorted β-cells from female and male islets (Stancill et al., 2019). We did not observe a statistically significant difference in depolarization induced exocytosis in single cells from either sex (Fig. 3H), suggesting that the late stages of insulin granule exocytosis are not altered under these conditions. Next, we analyzed Ca 2+ responses to 15mM glucose in thousands of Fura-2-loaded dispersed islet cells and analyzed the data with an adaptation of our TraceCluster algorithm (Wills et al., 2016). In agreement with the electrophysiological data, Insr knockout β-cells from female mice exhibited a significantly greater number of oscillation peaks within the glucose stimulation period compared to control cells (Fig. 3I,J, S2). A similar increase in excitability was observed in βInsr HET β-cells.
Collectively, these experiments demonstrate that β-cells lacking Insr have increased electrical activity, suggesting that insulin has a negative feedback influence on excitability in the context of hyperglycemia.
Loss of β -cell Insr causes insulin hypersecretion in the context of stimulatory glucose Insulin secretion is driven by electrical excitability, so we next carefully examined the effects of partial and full β-cell Insr deletion on secretory function employing multiple orthogonal in vitro and in vivo assays. We used islet perifusion to examine the dynamics of insulin secretion ex vivo at rest (3 mM glucose) and in response to 20mM glucose or 10mM glucose, as well as direct depolarization with 30mM KCl. Islets from female 16 week-old βInsr KO and βInsr HET mice secreted more insulin in response to 20 mM glucose and 30 mM KCl compared to islets from control mice (Fig. 4A). No significant differences were observed at 3mM or 10mM glucose (Fig. 4A). Consistent with our electrophysiology data, we did not observe differences in islets from males of the same age and on the same diet (Fig. 4B). It seems possible that reduced ATP generation in male mice (Fig. 4C) offsets any enhanced excitability/Ca 2+ responsiveness, to result in absence of significant changes in insulin secretion.
This potentiation of high glucose-stimulated insulin secretion ex vivo in the complete and partial Insr knockout β-cells, led us to examine how insulin levels were affected by glucose stimulation in vivo using the hyperglycemic clamp technique in awake mice. For this cohort of mice, there were no significant differences in body mass (control 20.5 +/-0.5g n=8 vs βInsr KO 20.8 +/-0.4g n=10), lean mass (control 17.5 +/-0.4g vs βInsr KO 17.3 +/-0.2g), or fat mass (control 1.8 +/-0.3g vs βInsr KO 2.2 +/-0.2g). Glucose infusion rates were adjusted in order to reach hyperglycemic levels (~19 mM) in βInsr KO and wild type control mice (Fig. 4C, D, G, H). Interestingly, slightly higher glucose infusion rates were necessary in female βInsr KO mice in comparison to control mice in order to reach similar hyperglycemic levels (Fig. 4C). In accordance with our ex vivo insulin secretion data, glucosestimulated insulin secretion was higher in female, but not in male βInsr KO mice compared with control mice (Fig. 4E, I). We further tested whether in vivo insulin secretion would be potentiated after a single bolus of glucose in mice with reduced β-cell Insr. At 11 weeks of age, plasma insulin response, relative to baseline, was significantly elevated 30 min after i.p. injection of 2g/kg body mass glucose in female, but not in male, βInsr KO mice compared to controls (Fig. 4K, L). In accordance with the electrophysiology, Ca 2+ oscillation, and islet perifusion data, we detected no statistical difference between female genotypes in fasting plasma insulin in vivo at multiple ages (Fig. S3). Together, these experiments suggest that β-cell Insr can play suppressive role in glucose-stimulated insulin secretion without much impact on basal insulin secretion.
Effects of β -cell Insr loss on insulin production, storage, processing and clearance Insulin and insulin signaling can modulate protein synthesis in many cell types and we have previously provided evidence that soluble cellular insulin protein transiently increases in human islet cell cultures treated with exogenous insulin (Luciani and Johnson, 2005). To assess the quantitative contribution of Insr signaling to insulin production and mRNA translation rates we measured total islet 7 insulin content after acid-ethanol extraction and examined total protein synthesis rate using S 35methoioine/cysteine pulse labelling in isolated islets. Total insulin content and protein synthesis in isolated islets were unaffected by Insr deletion under these basal glucose conditions ( Fig. 5M-P).
To investigate the role of Insr signaling on β-cell stress and insulin clearance, we conducted analysis of plasma proinsulin to C-peptide ratios, and C-peptide to insulin ratios in the fasting state (4h) across multiple ages in both male and female mice, and on multiple diets (Fig. S4). While many of these parameters changed with age, no statistical differences between genotypes was seen in any of these parameters of mice fed either a low-fat diet (LFD) or a high fat diet (HFD). A trend towards lower insulin clearance was observed in LFD-fed female βInsr KO mice in comparison to wild type control mice at 7 weeks. Collectively, these experiments show that β -cell insulin receptor signaling has only a small, if any, effect on insulin processing and clearance at baseline glucose conditions.

Beta-cell area and hyperglycemia-induced proliferation in mice lacking β-cell Insr
As our RNA sequencing data revealed pathways downstream of Insr signaling that may affect β-cell proliferation capacity (Fig. 2), we examined baseline β-cell area and proliferation reserve capacity. Islet architecture and β-cell-to-α-cell ratio were not obviously perturbed (Fig. 5A). We did not detect significant differences associated with genotype in β-cell area in either female or male mice, at 13, 42 weeks (LFD) or 54 weeks (HFD) of age (Fig. 5B). In comparison to control mice, HFD fed female βInsr KO mice had a tendency toward a smaller β-cell area at 54 weeks of age that were consistent with tendencies towards lower plasma insulin (Fig. S3C), proinsulin and C-peptide levels (Fig. S4). These data suggest that Insr may help support age-dependent β-cell expansion under some conditions.
Prolonged hyperglycemia can stimulate β-cell proliferation in adult mouse β-cells (Sharma et al., 2015), but whether this requires intact insulin receptor signaling remains controversial. To examine the role of Insr-mediated signaling on hyperglycemia-induced β-cell proliferation, we performed 4-day hyperglycemic infusions in βInsr KO and wild type control mice. In female mice, hyperglycemia (>10 mM; Fig. 5F) resulted in mildly elevated insulin secretion in βInsr KO relative to control mice for the initial 48 h, which was not sustained for the duration of the experiment (Fig. 5G,H), while glucagon levels declined similarly in both genotypes (Fig. 5I,J). There was no effect of Insr deletion on hyperglycemiainduced proliferation of either β-cells or α-cells in females (Fig. 5K,L). In male mice, 96 h of hyperglycemia resulted in sustained hyperinsulinemia in βInsr KO mice ( Fig. 5M-O), with no differences in circulating glucagon (Fig. 5P,Q). In male mice lacking β-cell Insr, this manipulation was associated with more β-cell proliferation (Fig. 5R,S). The fact that we did not observe a suppression of glucoseinduced proliferation of β-cells lacking Insr prompted us to determine the degree to which the broadly defined insulin signaling pathway was inhibited in our model. Indeed, glucose-induced Akt phosphorylation, shown by western blot of whole islet lysate, was statistically unaffected, and glucoseinduced Erk phosphorylation was only reduced ~50% in Insr knockout β-cells (Fig. 5T). It is likely that the Igf1r or a receptor tyrosine-kinase-independent mechanism initiates parts of intracellular postreceptor 'insulin signaling' in the absence of Insr. Testing this hypothesis in the future will require truly β-cell specific double deletion of Insr and Igf1r.
Modelling contributions of peripheral and β -cell specific insulin sensitivity to glucose homeostasis The continuum between insulin sensitivity and resistance impacts multiple tissues, including pancreatic β -cells. We observed that β -cell-specific insulin resistance resulted in insulin hypersecretion in the context of unchanged β -cell mass. We next used mathematical modelling to generate quantitative predictions of the dependence of glucose tolerance on both β -cell and wholebody insulin resistance, both independently and synchronously. As described in the methods section, we modified the Topp model (Topp et al., 2000) to include insulin receptor-mediated negative feedback on insulin secretion, as indicated by our experimental data, with S β Insr KO mice) corresponded with elevated peak and plateau insulin secretion (Fig. 6B). As expected, this was predicted to drive more rapid clearance of blood glucose (Fig 6C). Analysis of areas under the curve for glucose and insulin resulting from in silico glucose tolerance tests while varying S P and S β indicated that β -cell insulin resistance should have a marked effect on insulin secretion and glucose tolerance, most dramatically in conditions of low peripheral insulin sensitivity. Next, we compared the predictions of this model with experimental results. We used the in silico AUC Glucose values as a function of both S P and S β combined with averaged experimental AUGC values to estimate S P as a function of age for the low-fat diet conditions (Fig. S5C,D). We found that male values of S P for wildtype and mutant were indistinguishable from each other while females showed significant differences from each other and from the male values at all ages. As expected, HFD led to reduced peripheral insulin sensitivity (Fig. S5E). Collectively, these simulations show how β-cell insulin sensitivity and peripheral insulin sensitivity may combine to the regulation of glucose tolerance.
Condition-dependent improved glucose tolerance with reduced β -cell Insr signaling To test our theoretical model experimentally, we examined glucose tolerance in female and male βInsr KO , βInsr HET , and control littermates at multiple ages between 4 and 52 weeks in the context of two diets. Significant improvements in glucose tolerance were observed in female mice with reduced Insr signaling at multiple ages, and in young males (Fig. 6F). Consistent with our mathematical modelling that suggested a diminished contribution of β-cell insulin resistance to glucose homeostasis in the context of greater whole-body insulin resistance, we did not observe significant effects of genotype in older male mice, or mice of either sex fed an insulin-resistance-inducing HFD (Fig. 6F). Thus, Insr deletion has little consequence on glucose tolerance in mice with already impaired pantissue insulin resistance, which we and others have shown increases with age and is more pronounced in male mice (Fig. S5).
The Ins1 cre allele results in pre-natal gene deletion (Thorens et al., 2015). To assess the effects of Insr deletion in adult β-cells, and to determine the role of Insr on a different genetic background and under different housing conditions, we also phenotyped multiple cohorts of mice in which the Insr f/f allele was recombined by the Ins1 promoter-driven CreERT transgenic allele (commonly known as MIP-Cre) after injection with tamoxifen. In agreement with our observations in mice with constitutive loss of β-cell Insr, we found that glucose tolerance was significantly improved 4 weeks after β-cellspecific Insr deletion in male mice (Fig. 6G). These differences were not maintained as the mice aged and presumably became more insulin resistant. In these mice, there were no significant differences observed in fasting glucose (control 4.8 +/-0.3mM n=7 vs βInsr KO 4.4 +/-0.2mM n=10), β-cell mass (control 1.4 +/-0.3% n=3 vs βInsr KO 2.1 +/-0.2% n=5), or body mass (control 26.1 +/-1.1g n=7 vs βInsr KO 23.1 +/-0.7g n=17. Collectively, these observations using an independent model and independent housing conditions lend support to our conclusion that the initial consequence of β-cell specific Insr deletion is improved glucose tolerance. . Thus, we asked how the context-dependent glucosestimulated insulin hyper-secretion induced by targeted β-cell specific insulin resistance may affect insulin sensitivity, adiposity, and body mass over time. Insulin sensitivity was assessed at multiple ages in the same mice. Interestingly, insulin sensitivity was significantly improved in 10-week-old female βInsr HET mice compared to littermate controls without Insr deletion (Fig. S6). On a high fat diet, male βInsr KO and βInsr HET mice had significantly improved insulin sensitivity compared to controls at 22 weeks of age. Longitudinal tracking of 4-h fasting glucose identified relative hypoglycemia in young LFD female βInsr KO and βInsr HET mice, older LDF male βInsr KO and βInsr HET mice, and across the tested ages in HDF female mice ( Fig. 7A-D). Longitudinal tracking of body weight revealed that female mice with reduced β-cell Insr consistently weighed more than controls when fed a HFD ( Fig. 7E-H), consistent with the known role of hyperinsulinemia in diet-induced obesity (Mehran et al., 2012;Templeman et al., 2015). We also examined the mass of several tissues at 13 weeks of age. Interestingly, liver mass was lower in both female and male mice lacking β-cell Insr (Fig. 7IJ). Pilot experiments showed that liver Insr protein abundance appeared reduced in mice with partially or completely reduced β-cell Insr, in the context of the LFD but not the HFD (Fig. 7K). These data are consistent with the concept that modest hyperinsulinemia can drive down insulin receptor levels and the concept that insulin signaling is a trophic signal for liver. Together, these data demonstrate that specifically preventing autocrine insulin feedback can have systemic effects on insulin sensitivity, body mass, and the size of some tissues. These changes may affect the eventual susceptibility to type 2 diabetes.

Discussion
The goal of the present study was to establish the role of β-cell specific insulin resistance on β-cell function and glucose homeostasis using specific genetic loss-of-function tools. We found that in vivo Insr deletion potentiated glucose-stimulated insulin secretion by increasing action potential firing and Ca 2+ oscillation frequency, leading to improved glucose tolerance in insulin sensitive animals. Our data therefore suggest a model in which insulin inhibits its own secretion in a context-dependent manner and that this local negative feedback loop has physiological consequences for glucose tolerance.
Autocrine signaling in endocrine cells is generally a negative feedback (Norman and Henry, 2014), with a few exceptions in specific conditions . Given the abundance of Insr protein in islets and the physiological modulation of both insulin and Insr signaling in health and disease, autocrine and paracrine insulin signaling have been topics of interest and controversy for decades (Braun et  Consistent with a narrow range of responsiveness, our results also show that the contribution of autocrine insulin feedback to glucose homeostasis depends on whether mice are on a diet or at an age where insulin resistance is high, and potentially saturated in β-cells. Background genetics, diet, housing conditions, microbiome, or glucose concentrations could contribute to differences in observed phenotypes between our β-cell specific Insr knockout models and the frank diabetes reported for transgenic models that use fragments of the Ins2 or Pdx1 promoters to drive Cre-mediated β-cell insr deletion ( . We expect that our βInsr KO mouse line is the most tissue specific model used to date for the study of autocrine insulin signaling and β-cell insulin resistance. Using genetic and genomic tools, our work complements previous studies in humans and animal models. Ex vivo studies of perfused canine pancreata found an inhibitory autocrine effect of insulin (Iversen and Miles, 1971). Similarly, exogenous insulin perfusion of canine pancreas in situ was shown to lead to reduced endogenous insulin production (Rappaport et al., 1972). In vivo insulin infusion rapidly suppressed C-peptide levels in healthy men, but not those with obesity and presumably global insulin resistance (Cavallo-Perin et al., 1993). In isolated human islets, perifusion studies showed that treatment with physiological doses of insulin had no effect on C-peptide release Beyond the potentiation of glucose-stimulated insulin secretion, transcriptomic analysis of fully and partially Insr-deficient β-cells revealed a broad re-wiring of multiple signaling pathways and gene networks that are expected to change how these cells respond to stresses. In particular, Insr loss in our model biased the combined insulin/Igf signaling pathway towards Ras/MAP-kinase signaling and increased in the capacity for cytokine signaling that could modulate cell fate. The conditions under which these gene expression changes could manifest in β-cell proliferation or increased β-cell survival remain incompletely explored. Otani et al reported modestly reduced β-cell mass in non-diabetic Ins2-Cre transgenic Insr knockouts, which was exacerbated by diabetes, but they did not employ Cre controls . Okada et al reported impaired compensatory β-cell proliferation in the context of high fat diet or liver Insr knockout in the same experimental groups (Okada et al., 2007). These previous studies comparing β-cell knockout mice of Insr versus Igf1r, suggested a more important role for the former in β-cell proliferation and survival (Okada et al., 2007). Insulin receptor over-expression experiments also support the idea that β-cells are key insulin targets (Okamoto et al., 2004). In the present study, we were unable to identify conditions which would result in statistically significant differences in β-cell mass in mice with Insr deficiency, but variability was high and relatively few animals were studied at older ages. We have previously shown that the increase in β-cell mass resulting from high fat diet requires local insulin production and is independent of hyperglycemia (Mehran et al., 2012), consistent with the known direct anti-apoptotic and mitogenic effects of insulin in vitro (Beith et al., 2008;Johnson et al., 2006a;Muller et al., 2006). These findings can be reconciled by proposing that high insulin concentrations within the islet are sufficient to activate remaining Igf1 receptors linked to biased post-receptor signaling. We observed that sustained hyperglycemia and hyperinsulinemia over 4 days was associated with increased proliferation, but whether these are effects are mediated through Igf1r function compensation will require double receptor knockout experiments. It is also possible that hyperglycemia itself is a major driver of β-cell proliferation under these conditions, through Irs2-Creb signaling that may bypass Insr/Igf1r (Jhala et al., 2003). It has also been demonstrated that ~80% of the gene expression changes attributed to glucose in MIN6 cells require full insulin receptor expression (Ohsugi et al., 2005). We have previously found that glucose cannot stimulate primary mouse β-cell proliferation when autocrine insulin signaling is blocked by somatostatin and that 'glucose-induced' Erk phosphorylation requires full insulin secretion Our studies illustrate the power of using both females and males to study integrated physiology. Indeed, β-cell Insr loss led to increased β-cell action potentials, calcium oscillations, and glucosestimulated insulin secretion in 16-week-old female mice, but not in age-matched males. No intrinsic sex differences of Insr mRNA levels were in a transcriptomic analysis of sorted β-cells (Stancill et al, 2019). Given the abundance of data showing more pronounced insulin resistance in males (Parks et al., 2015;Yki-Jarvinen, 1984), the most likely reason for these female-specific phenotypic responses to β-cell Insr deletion was the development of insulin resistance in control males but not in control females. To this point, glucose tolerance is improved in 4-week-old βInsr KO males, an age at which control males remain insulin-sensitive, and in male mice with acute loss of Insr. More work will be needed to confirm this possibility, and to determine factors in addition to sex hormones and sex chromosomes that impact these sex differences in insulin sensitivity and glucose homeostasis (Mauvais-Jarvis, 2015). The sex-specific nature of phenotypes arising from our genetic manipulation of Insr in β-cells highlights the importance of including both sexes to accurately interpret data and to draw conclusions that will apply to both sexes.
While our study is comprehensive and employs the best genetic tools available today, this work has limitations. Ins1 Cre is the most β cell-specific Cre deletion strain available today (Thorens et al., 2015), but this fact does not preclude off-tissue effects that have yet to be discovered. Cre recombinase itself is not totally benign. These facts and our detailed comparison of Ins1 wt/wt mice with Ins1 Cre/wt mice (Figs. S7-10), highlight the importance of the Cre control group we employed throughout our studies. Recently, it was reported that some colonies of Ins1 cre mice exhibited some silencing via DNA hypermethylation at the Ins1 locus and this was suggested as an explanation for discordance between the phenotypes compared to gene deletions using Pdx1-Cre and Ins2-Cre transgenic lines (Mosleh et al., 2020). In our study, we observed virtually complete recombination in β-cells and no evidence for off-tissue Insr deletion. We believe a major source of discrepancy with previously reported phenotypes stems from the propensity of previous promoter transgenic strains to recombine in the brain and robust expression of Insr throughout the brain. Another caveat of our experiments using βInsr KO mice is that Insr is expected to be deleted from β-cells starting in late fetal development (Thorens et al., 2015). In our hands, tamoxifen-inducible Ins1 CreERT mice have insufficient β-cell specific recombination for in vivo physiological studies. Our validation experiments using the tamoxifen-inducible Ins1-CreERT address this limitation and confirm that β-cell insulin resistance improves glucose tolerance, at least under the initial insulin sensitive conditions. It should also be noted that, because we show that long-term deletion of Insr results in profound re-wiring of the β-cell transcriptome, the physiological changes can be the result of either direct or indirect action of Insr signaling. It should also be emphasized that, while we have deleted Insr, insulin can signal through Igf1r and Igf2r, especially at the higher concentrations predicted in the pancreas. A further caveat is that the molecular mechanisms involved in insulin secretion may be different in mouse and human β-cells In conclusion, our work demonstrates a key modulatory role for autocrine insulin negative feedback and the lack thereof (i.e. β-cell resistance) in insulin secretion, glucose homeostasis and body mass, which depend on the physiological context studied. We hope our studies resolved longstanding and controversial questions about the local effect of insulin on β-cells, and lead to experimental and theoretical studies that incorporate Insr-mediated signaling in other islet cell types.

Bioinformatics
Human tissue-level proteome and transcriptome were downloaded from the ProteomicsDB resource (https://www.proteomicsdb.org) (Schmidt et al., 2018) in 2019 and sorted by relative protein abundance in Microsoft Excel. Publicly available human islet scRNAseq data were acquired from the panc8.SeuratData package and the SCTransform pipeline was followed to integrate the studies (Hafemeister and Satija, 2019). Expression data were normalized using the Seurat::NormalizeData function with default parameters and visualized using the Seurat::RidgePlot and Seurat::UMAPPlot functions, all from the Seurat package in R (Stuart et al., 2019).

Mouse model and husbandry
All animal protocols were approved by the University of British Columbia Animal Care Committee (Protocol A16-0022), the Institutional Animal Care and Use Committee of the University of Massachusetts Medical School (A-1991-17) and the University of Michigan Institutional Animal Care and use Committee, in accordance with national and international guidelines. Mice were housed at room temperature on a 12/12 light dark cycle at the UBC Modified Barrier Facility, unless otherwise indicated.
MIP cre mice were generously obtained from Dr Dempsey (Wicksteed et al., 2010b). The CAG-YFP reporter transgenic animals were from Jax (#011107). All animals were males that were intraperitoneally injected at 8 weeks of age for 5 consecutive days with tamoxifen (Sigma, T5648) freshly dissolved in corn oil (Sigma, C8267) with 3 injections at 200 mg/kg over a 1-week period.

Comparison of control genotypes
Before conducting our main study, we did a pilot experiment to determine whether the Ins1 cre knock-in mice had any phenotype on their own under both low fat and high fat diets, and we tracked both 'control' genotypes for the majority of our studies. Although Insr wt/wt ;Ins1 cre/wt ;nTnG and Insr f/f ;Ins1 wt/wt :nTnG control mice exhibited generally similar phenotypes, we observed key differences that reinforced the rationale for using controls containing Cre and lacking 1 allele of Ins1, matching the experimental genotypes. For example, male HFD-fed Insr f/f ;Ins1 wt/wt ;nTnG mice showed significantly higher levels of plasma proinsulin in comparison to Insr wt/wt ;Ins1 cre/wt ;nTnG mice at 16 and 28 weeks of age (Fig. S7). At several ages, both LFD and HFD fed female Insr f/f ;Ins1 wt/wt ;nTnG /mice exhibited trends toward slightly improved glucose tolerance (Fig. S8), most likely due to one extra allele of insulin, in comparison to Insr wt/wt ;Ins1 cre/wt ;nTnG mice. Insulin sensitivity was generally similar, although not identical (Fig. S9). Longitudinal tracking of body weight revealed a consistent tendency for mice with a full complement of insulin gene alleles to be heavier than mice in which 1 allele of Ins1 had been replaced with Cre. With the statistical power we had available, female HFD-fed Insr f/f ;Ins1 wt/wt ;nTnG mice had significantly increased body mass at 11 and 16 weeks of age in comparison to Insr wt/wt ;Ins1 cre/wt ;nTnG mice ( Fig. S10). Once we had established the effects of the Ins1 cre allele on its own, we used a breeding strategy ensuring that all pups were born with 3 insulin alleles to control for any effects of reduced insulin gene dosage (See Fig. 1C). This strategy gave us cohorts of: Insr wt/wt ;Ins1 cre/wt ;nTnG /-(βInsr KO ) control mice, Insr f/wt ;Ins1 cre/wt ;nTnG /-(βInsr KO ) β-cell specific Insr heterozygous knockout mice, and Insr f/f ;Ins1 cre/wt ;nTnG /-(βInsr KO ) β-cell specific Insr complete knockout mice. In some studies, the nTnG allele was not present (see Figure legends).

Targeted gene expression analysis
Tissue samples were kept frozen during grinding using mortals and pestles. cDNA was synthesized using qScript TM cDNA synthesis kits (QuantaBio, #95047-500) following RNA was isolated from 50-100mg of sample using RNeasy mini kits (Qiagen, #74106) according to manufacturer's recommendations. qPCR was performed in 15ul reaction volumes using a CFX384 Touch Real-Time

RNA sequencing
To generate transcriptomic data from β-cells lacking Insr and littermate controls, groups of 50 islets were dispersed using mild trypsin digest according to our standard protocol (Alejandro et al. Protein-protein interaction network, sub-network and TF-gene network was generated from differentially expressed genes between KO and WT control group using Network Analyst 3.0 platform (http://www.networkanalyst.ca/) (Zhou et al., 2019). PPI network data were based on STRING interactome with high (900-1000) confidence score. PPI sub-network enriching KEGG 'Insulin signaling pathway' was generated using the Module Extraction function. TF-gene network data were based on ENCODE ChIP-seq data.

Islet metabolism and oxygen consumption analysis
The Seahorse XF Cell Mito Stress Test kit (cat#103015-100) was used to measure oxygen consumption rate (OCR) in dispersed mouse islets using an Agilent Seahorse XF96 Analyzer XF96 (Seahorse Bioscience, North Billerica, MA). Dispersed islet cells were seeded at a density of 40,000 cells/well. 48h post seeding, additionally, 180μL of Seahorse XF Calibrant Solution (cat#100840-000) was added to each well of the Seahorse XF Sensor Cartridge to hydrate the XF Utility Plate (cat#102416-100). The hydrated cartridge was kept in a non-CO 2 incubator at 37C o for 24h hereby removing the CO 2 from media. To allow the assay media to pre-equilibrate, 180 μ L of assay (eagle's minimal essential medium (DMEM, no glucose, Gibco, (cat#103334-100) supplemented with 25 × 10 −3 M glucose and 25 × 10 −3 M 4-(2-hydroxyethyl) -1-piperazinemethanesulfonic acid(HEPES)) was added to each well 1 h prior to the run and used for extracellular flux measurements ofXF. The plate was kept at 37 C o in a non-CO 2 incubator. Mitochondrial respiration was analyzed by sequential injections of modulators including oligomycin (1M) used to block ATP synthase, carbonyl-cyanide-4-(trifluoromethoxy) phenylhydrazone (FCCP 2.5M) to activate uncoupling of inner mitochondrial membrane allowing maximum electron flux through the electron transport chain, and a mix of rotenone (1M) and antimycin A (1M) were used together to inhibit complexes I and III, respectively. The modulators were diluted in XF Assay Medium and loaded into the injection ports of the hydrated sensor cartridge corresponding to the order of injection 1 h prior to the run.

Patch-clamp electrophysiology
Islets from 16-week-old chow-fed male and female mice were isolated at UBC. 100-300 islets from each mouse shipped in a blinded manner overnight to the University of Alberta in RPMI (Invitrogen, 11875) with 10% FBS (Invitrogen #12483020), and 1% penicillin-streptomycin (Thermo Fisher, #15070063). Islets were dissociated into single cells using StemPro accutase (Thermo Fisher Scientific, Cat# A11105-01) one day after receiving the islets. Dispersed cells were cultured in RPMI-1640 containing 11.1 mM glucose with 10% FBS and 100 U/ml penicillin/streptomycin for up to 2 days.
Membrane potential and current measurements were collected using a HEKA EPC10 amplifier and PatchMaster Software (HEKA Instruments Inc, Lambrecht/Pfalz, Germany) in either the current-or voltage-clamp mode in the perforated patch-clamp configuration. All the measures were done in a heated chamber (32-35°C). Membrane potential measurement was performed with patch pipettes pulled from thick-walled borosilicate glass tubes (Sutter Instrument), with resistances of 8-10 MΩ when filled with 76 mM K 2 SO 4 , 10 mM KCl, 10 mM NaCl, 1 mM MgCl 2 and 5 mM Hepes (pH 7.25 with KOH), and back-filled with 0.24 mg/ml amphotericin B (Sigma, cat# a9528). The extracellular solution consisted of 140 mM NaCl, 3.6 mM KCl, 1.5 mM CaCl 2 , 0.5 mM MgSO 4 , 10 mM Hepes, 0.5 mM NaH 2 PO 4 , 5 mM NaHCO 3 , 5 mM glucose (pH 7.3 with NaOH). Membrane potential was measured with 5 mM G starting from the beginning, for 5 min, then changed to 1 mM G for 4-5 min, then changed to 10 mM G for 8-10 min, finally changed back to 5 mM G. K ATP currents and reversal potential were recorded during and after membrane potential measurement on each cell. β-cells were distinguished by characteristic differences in the voltage-dependent inactivation of Na + channel currents (Gopel et al., 2000).

Calcium imaging and analysis
Two days following cell seeding on glass coverslips, adherent islet cells were loaded with 5μM of the acetoxymethyl (AM) ester form of the calcium indicator Fura-2 (Thermo Fisher Scientific #F1221) for 30min. Islet cells were perifused at 1mL/min for 45min prior to experimental procedure to ensure washout of excess FURA2. During experiments, cells were mounted on a temperature-controlled stage and held at 37°C on a Zeiss Axiovert 200M inverted microscope equipped with a FLUAR 20× objective (Carl Zeiss,Thornwood, NY), while perifused with Krebs-Ringer (KRB) solution (144mM NaCl, 5.5mM KCL, 1mM MgCl 2 , 2mM CaCl 2 , 20mM HEPES) of various glucose concentrations as indicated in figures. Ca 2+ traces were analyzed automatically, as follows. Taking a similar approach to that described previously (Wills et al., 2016), 8 features were extracted from the Traces (Fig S2). Peaks during each phase were identified as local maxima reaching a value with a percent difference above the median baseline level greater than 20%. P-values for calcium analysis were generated using ANOVA with correction for multiple comparisons performed using Tukey's method. Figures were generated using the ggplot2 package in R (Wickham2016). Code used to analyze data and generate figures are available upon request.

Analysis of total protein synthesis rate
For the purpose of pulse labeling of newly translated proteins, 50 isolated islets were incubated in complete RPMI media without cysteine and methionine (MP Biomedicals, #SKU 091646454) for 1hr.
Subsequently media was supplemented with 250 μCi of [ 35 S]-cysteine/methionine mixture (PerkinElmer, NEG772002MC) and islets were incubated under normal conditions for 30 min. Islets were then lysed and proteins separated by SDS-gel electrophoresis as described above. Gels were fixed for 30 min in 50% (v/v) ethanol in water with 10% (v/v) acetic acid, dried in gel dryer (Bio-Rad model 583) and then exposed to storage phosphor screen (GE Healthcare) overnight. Screens were imaged and digitised using Typhoon FLA 9000 biomolecular imager (GE Healthcare). Protein bands intensity was quantified with Adobe Photoshop software.

Dynamic insulin secretion perifusion analysis
For perifusion experiments, islets from 16week old chow-fed male and female mice were isolated using collagenase, filtration and hand-picking as previously described (Johnson et al., 2006b). Our standard approach compared the insulin response to both 20mM and 10mM glucose stimulation as well as direct depolarization with 30 mM KCl. More specifically, 150 hand-picked islets per column were perifused (0.4 mL/min) with 3 mM glucose KRB solution containing (in mM) 129 NaCl, 4.8KCL, 1.2 MgSO4•7H2O, 1.2 KH 2 PO 4 , 2.5 CaCl 2 , NaHCO 3 , 10 HEPES, as well as 0.5% BSA (Sigma # A7030) for 60 min to equilibrate the islets to the KRB and flow rate, and then treated as indicated. Samples were analyzed using a rat insulin radioimmunoassay that has 100% cross-reactivity for mouse insulin (Millipore-Sigma #ri-13k). Insulin content was measured after acid-ethanol extraction using an insulin ELISA (Stellux Rodent Insulin ELISA, Alpco #80-INSMR-CH10).

Hyperglycemic clamps
In vivo hyperglycemic clamp experiments were performed at the National Mouse Metabolic Phenotyping Center (MMPC) at UMass Medical School. Body composition analysis was conducted by noninvasively measuring whole body fat mass and lean mass using 1 H-MRS (Echo Medical Systems, Houston, TX). A survival surgery was performed at 5-6 days before hyperglycemic clamp experiments to establish an indwelling catheter in the jugular vein. On the day of experiment, mice were fasted overnight (~15h), and a 2-h hyperglycemic clamp was conducted in awake mice by intravenously infusing 20% dextrose to rapidly raise and maintain plasma glucose levels at ~19 mM (Zhang et al., 2012).

Intravenous 4-day glucose infusion
Mice were bred and genotyped at University of British Columbia and shipped at 5 weeks of age to the Division of Diabetes, Department of Medicine, University of Massachusetts Medical School, USA. In a blinded manner, glucose infusions were performed as described (Alonso et al., 2007). Jugular vein catheters were placed in 9-12-week-old male and female mice with blinded genotypes. From postoperative recovery through euthanasia mice were unrestrained and were maintained on a 12-h light/dark cycle, with access to 2.2 g diet (to ensure isocaloric intake across all mice) and water. After 2 days of recovery, mice received continuous 4-day intravenous infusions of 50% dextrose (Baxter) containing 500ug/mL BrdU. Tail blood was sampled for plasma insulin, glucagon and blood glucose at Day 0, 1, 2 and 4. Blood glucose was measured using ReliOn glucometer (Walmart), glucagon was measured using mouse Glucagon ELISA (Mercodia 10-1281-01), and plasma insulin was measured using mouse Insulin ELISA kit (Mercodia 10-1247-01). Mice were euthanized at the end of the experiment and pancreas and duodenum were harvested for histology. Tissues were fixed for 5 h in 10% formalin and then stored in 1X PBS until processing, paraffin embedding and sectioning. Images were acquired using a NIKON fully motorized for Phase and Fluorescence Ti-E microscope. Images were taken of at least 10 randomly selected islets, all four channels at the same time. To generate RGB images, channels were inserted to show Insulin-BrdU-DAPI, Glucagon-BrdU-DAPI or Insulin-Glucagon-DAPI. To generate yellow-magenta-white images to accommodate colorblind viewers, new files were generated in Adobe Photoshop in which original channel data were displayed in multiple channels using the merge function (e.g. to change green to yellow, green channel data were added to both green and red channels; for more detailed information please contact LCA). Cells were counted using Cell profiler automated counting software from Broad Institute (Cambridge, MA); all counts were manually checked.

β -Cell Mass and immunohistochemistry
Pancreata were perfused, then fixed for 24h with 4% paraformaldehyde, and then washed twice with 70% ethanol prior to paraffin embedding and sectioning (5 μm) to obtain 5 different regions of the pancreas (100 μm apart) by WAXit Inc. (Vancouver, Canada). Paraffin was removed by 5 min xylene incubation steps. Sections were rehydrated in decreasing concentrations of ethanol and rinsed with water and PBS. Epitope retrieval was done either by immersing samples 10 mM citrate buffer, pH 6.0 for 15min at 95 o C, or by transferring sections to prewarmed 1N HCl for 25 min at 37 o C. Samples were washed with PBS twice and were either blocked for 10 min at room temperature (Dako protein block #X0909), or with goat block (GB) with Triton X-100 (10% BSA + 5% Goat Serum with 0.5% Triton X-

Blood collection and in vivo analysis of glucose homeostasis and insulin secretion
Tail blood was collected for blood glucose measurements using a glucometer (OneTouch Ultra 2 meter, Lifescan, Canada) for single time points as well as during glucose and insulin tolerance tests. Mice were fasted for 4h or 16h during single timepoints and for 6h during glucose and insulin tolerance tests, as well as glucose stimulated insulin secretion tests. The i.p. glucose dose was 2g/kg unless otherwise specified. The i.p. Humalog (Eli Lilly and Co) insulin dose was 0.75U unless otherwise indicated. 1-2 days prior to femoral blood collection the experimental mice were tube handled and shaved. Femoral blood was collected for single timepoints, as well as for measurements of in vivo glucose-stimulated insulin secretion after i.p. injection of 2g/kg glucose. Blood samples were kept on ice during collection, centrifuged at 2000rpm for 10min at 4 o C and stored as plasma at -20 o C. Plasma samples were analysed for insulin (Stellux Chemi Rodent Insulin ELISA, Alpco #80-INSMR-CH10), proinsulin (Alpco #80-PINMS-E01), C-peptide (Alpco #80-CPTMS-E01), glucagon (Mercodia, #10-1281-01). Measurements were performed on a Spark plate reader (TECAN), and analysed using (GraphPad Software Inc., San Diego, CA).

Mathematical modeling
We used a modified version of the Topp model (Topp et al., 2000) to simulate glucose and insulin dynamics (see equations in Fig. 6A). Parameter values for the glucose and insulin equations were modified by hand so that the model matched a typical wildtype female low-fat-diet glucose tolerance test time series and predicted the observed beta cell mass in these mice. We included an inhibition factor, 1/(1 + S β I), multiplying the insulin secretion term that accounts for insulin receptor-mediated negative feedback on insulin secretion, parametrized by S  2 , where I wt * is the steady state insulin level for wild type control mice, and I m * is the equivalent for the βInsrKO mice. We assumed that, at each time point after the first one in the glucose-clamp insulin measurements, the blood insulin levels were hovering around steady state and all other parameters and variables were constant (Fig. S5A). We used those steady state data points to estimate S β for the male and female wild type mice using this equation. Glucose tolerance tests were simulated for various S β and S P values by calculating the IG system steady state values for each S β and S P , then using the steady state value as the initial condition for insulin, and steady state plus 20 mM as the initial condition for glucose. Fig. 6A shows sample in silico glucose tolerance tests for various values of S β , with S P fixed at 70. The glucose tolerance area under the glucose curve (AUGC) was then calculated for each set of S β and S P , generating a numerical map from S β and S P to AUGC. From the glucose tolerance tests in wildtype and mutant male mice and wildtype and mutant female mice, we calculated the experimental AUGC values at 4, 9, 21 and 39 weeks. Using the computed AUGC map and the S β values calculated from the glucose-clamp data, we estimated the S P value that would give the experimental AUGC (Fig. S5C)

Statistics
Data are presented as mean +/-SEM, with individual data points, unless otherwise indicated. Ttests were performed when only 2 groups were compared at a single timepoint. One-way ANOVA was applied when three groups were compared at a single timepoint. Mixed effects model statistics were used for statistical analysis of GTT, ITT, GSIS experiments. Specifically, a fitted mixed model (Prism 8, Graph Pad), which allows for missing measurements and un-even group sizes, was applied when 2 or more groups were compared at multiple time points (e.g. 4h fasted blood glucose, insulin, proinsulin, c-peptide and body weight).                 Blood Glucose (mM)        Fig. S4. Fasting proinsulin and C-peptide. (A-D) Circulating plasma proinsulin and C-peptide levels after a 4-hour fast, and their associated ratios in control, bInsr HET , and bInsr KO mice at multiple ages. Statistical analyses were done with a mixed model.