Weight Cycling Impairs Pancreatic Insulin Secretion but Does Not Perturb Whole-Body Insulin Action in Mice With Diet-Induced Obesity

Blood glucose is controlled within a narrow window, minute-to-minute, through the integration of multiple organ systems and tissues. The balance between endogenous glucose production and blood glucose clearance defines glycemic control—each of which relies on the peripheral actions of pancreatic insulin. Chronic obesity is characterized by impaired glucose tolerance and hyperinsulinemia (1,2). The combination of elevated blood glucose and hyperinsulinemia promotes the expansion of pancreatic β-cell mass (3), which is considered a compensatory adaptation to worsening glycemia. Although hyperinsulinemia is typically considered a consequence of peripheral insulin resistance (4), a growing body of evidence suggests it may be a driver of—or at least a significant contributor toward—worsening insulin resistance that accompanies obesity (1–3,5–8). Nonetheless, prolonged hyperglycemia/hyperinsulinemia causes β-cell fatigue and ultimately β-cell failure, defined clinically as type 2 diabetes (9).

Caloric restriction is known to improve obesity-evoked insulin resistance in humans and rodents, whereas one or more bouts of weight regain (e.g., weight cycling) is linked with increased risk of a cluster of cardiometabolic diseases as compared with obesity alone (10–12). Our group developed a murine model of weight cycling whereby mice undergo periods of diet-induced weight gain, weight loss, and weight regain, which accelerates glycemic dysfunction, compared with weight-matched obese controls (13,14). Though our prior work established that weight cycling augments glucose intolerance, it is unclear whether this is principally attributable to decreased insulin action (i.e., defined in this context as decreased suppression of endogenous glucose production and decreased insulin-mediated glucose clearance), impaired pancreatic islet function, or a combination of both. Uncovering the mechanisms for impaired glucose regulation caused by weight cycling may have clinical implications given that most obese individuals regain lost weight within 1–2 years (15–18).

In this study, we aimed to systematically pinpoint the major control points and mechanisms by which glycemic deterioration manifests in weight-cycled mice, using rigorous in vivo and ex vivo functional tests. We hypothesized that weight cycling worsens peripheral insulin action and glucose clearance. Instead, we report that weight cycling–induced glucose intolerance is not explained by worsened whole-body insulin action but, rather, attributable to inadequate insulin secretion that is directly linked to loss of β-cell identity and adaptive plasticity. Taken together, our results suggest that repeated bouts of weight loss and weight regain impair pancreatic compensation for oscillating body weight and by consequence insulin secretion in mice.

All procedures were approved in advance and carried out in compliance with the Vanderbilt University Institutional Animal Care and Use Committee. Vanderbilt University is accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International.

Antibodies are listed in Supplementary Table 1.

Male C57BL/6J mice were purchased from The Jackson Laboratory (no. 000664) at 7 weeks of age. At 8 or 9 weeks of age, mice were placed on 9 week cycles of high-fat diet (HFD) (60% fat, 5.21 kcal/g food, no. D12492; Research Diets) or low-fat diet (LFD) (10% fat, 3.82 kcal/g, D12450B; Research Diets) for a total of 27 weeks. The feeding paradigm generated two experimental groups, “obese” mice and “weight-cycled” mice (Fig. 1A). In several figures, a lean age-matched reference group (ad libitum LFD, from our published work [13]) is used as a comparator to HFD-fed animals. Two primary cohorts of mice were used to comprehensively determine how weight cycling worsens glucoregulation. Cohort 1 was subjected to hyperinsulinemic-euglycemic clamp procedures to determine glucose flux and whole-body insulin action. Cohort 2 underwent hyperglycemic clamps for assessment of in vivo glucose-stimulated insulin secretion (GSIS). In addition, isolated islets from a subset of obese and weight-cycled mice were used for islet perifusion studies, transcriptomics, and/or microscopy.

After a 5 h fast, basal blood glucose levels were measured (−10 min) via tail cut followed by injection of 2.0 g dextrose/kg fat-free mass i.p. Postinjection blood glucose was sampled at 15, 30, 45, 60, 90,120, and 150 min with a handheld glucometer (Bayer Contour NEXT EZ meter). Glucose area under curve from baseline was calculated using the trapezoidal rule.

After a 5 h fast, blood was collected from a carotid artery indwelling catheter before oral gavage administration of a liquid meal (Ensure, 15 kcal/kg fat-free mass or ∼2.5% of total daily energy intake). The percentage of daily energy intake measurement was based on average daily energy intake over 3 weeks preceding the meal challenge. Arterial blood was collected at 15, 30, 45, and 60 min postgavage. Immediately following the last blood draw, mice were euthanized via lethal dose of isoflurane for tissue harvest. Tissues were either fixed (4% paraformaldehyde [PFA] or 2% glutaraldehyde, 2% PFA, 0.1 mol/L sodium cacodylate, 2 mmol/L CaCl₂) or flash frozen in liquid nitrogen and stored at −80°C.

Mouse body fat and fat free mass were measured by a nuclear magnetic resonance whole-body composition analyzer (Bruker Minispec).

Catheters were surgically placed in the carotid artery and jugular vein for sampling and infusions, respectively, 1 week before clamps were performed. Mice were fasted for 5 h before clamp procedures. Mice were neither restrained nor handled during clamp experiments.

[3-³H]glucose was primed and continuously infused from t = −90 min to t = 0 min (0.04 µCi/min). The insulin clamp was initiated at t = 0 min with a continuous insulin infusion (4 mU/kg/min) and variable glucose infusion rate (GIR), both maintained until t = 155 min. The glucose infusate contained [3-³H]glucose (0.06 µCi/µL) to minimize changes in plasma [3-³H]glucose specific activity. Arterial glucose was monitored every 10 min for provision of feedback for adjustment of the GIR to maintain euglycemia. Erythrocytes were infused to compensate for blood withdrawal. [3-³H]glucose kinetics were determined at −15 min and −5 min for the basal period and every 10 min between 80 and 120 min for the clamp period for assessment of whole-body glucose appearance (R_a), glucose disappearance (R_d), and endogenous glucose production (endoR_a). A 13-µCi intravenous bolus of 2-[¹⁴C]deoxyglucose ([¹⁴C]2DG) was administered at 120 min to determine the glucose metabolic index (R_g), an index of tissue-specific glucose uptake. Blood samples were collected at 122, 125, 130, 135, and 145 min to measure [¹⁴C]2DG disappearance from plasma. At 145 min, mice were anesthetized and euthanized and tissues immediately harvested and freeze clamped.

Arterial glucose was clamped at ∼250 mg/dL with a variable GIR. GIR is an index of glucose tolerance during the hyperglycemic clamp. Plasma insulin, C-peptide, and nonesterified fatty acids (NEFA) were determined during clamps. Full step-by-step descriptions of the surgery, isotope clamp method, and calculations are available from the Vanderbilt Mouse Metabolic Phenotyping Center (VMMPC) website (www.vmmpc.org).

Plasma NEFA (HR Series NEFA-HR; FUJIFILM Wako Diagnostics U.S.A.) were determined via colorimetric assay according to the manufacturer’s instructions. Plasma insulin and C-peptide were determined via radioimmunoassay (Vanderbilt Hormone Assay & Analytical Services Core, Vanderbilt University Medical Center [VUMC], Nashville, TN). Insulin was determined by radioimmunoassay (cat. no. PI-13K; Millipore). The assay uses ¹²⁵I-labled insulin and a double antibody technique to determine serum, plasma, or tissue culture media insulin levels. The assay was modified by the Vanderbilt Hormone Assay & Analytical Services Core to improve the sensitivity to 0.01 ng/mL for the sensitive mouse assay and 0.1 ng/mL for the regular mouse assay. The cross-reactivity between human and mouse insulin is 100% in the radioimmunoassay used. Exogenous insulin was determined with use of a human-specific insulin antibody (no. 80-INSHU-E01.1; Alpco). All fasting blood samples were collected after a 5 h fast.

Mouse islets were isolated as previously described (19) and cultured in RPMI medium containing 10% FBS and 5 mmol/L glucose at 37°C overnight. GSIS was assessed by perifusion with use of size-matched islets and normalized to islet equivalents. Three secretagogues, glucose (16.7 mmol/L), 3-isobutyl-1-methylxanthine (IBMX) (100 μmol/L), and KCl (20 mmol/L), were used during perifusion. Insulin in the culture medium was determined by ELISA. Islet size was assessed with MetaMorph, version 7.7 (Universal Imaging) (20). Islet collection and perifusion studies were conducted by the VUMC Islet Procurement & Analysis Core.

Pancreata were fixed in 4% PFA at room temperature for 2 h and then paraffin embedded and serial sectioned at 5 μm for immunolabeling. Slides were imaged with either a ×20 or ×40 objective on a Leica DMI8 widefield microscope and captured with a Leica DFC9000 GT camera. Image tiles were taken across the entirety of each section and stitched with use of the LAS X software suite. Images shown were taken at ×40 magnification (332.79 μm²; 6.1539 pixels/μm). For islet area quantification, every 10th slide was used (three slides/mouse) to ensure no islet was measured twice (minimum distance between sections of 150 μm). ImageJ was used to threshold for DAPI and insulin, and the area of each threshold was outlined manually for total pancreas area and islet area, respectively. Islets containing fewer than five total nuclei were excluded. For quantification of α-cell–to–β-cell ratio, QuPath was used to identify islets and select intraislet cells based on DAPI⁺ nuclei. Cells were designated as α, β, or other based on cytoplasmic pixel intensity of glucagon or insulin. Further annotation of Ki-67⁺ cells was performed with QuPath based on positive nuclear staining.

Immunoblotting was performed on inguinal white adipose tissue (iWAT) and gastrocnemius muscle. Triton X-100 tissue lysates were prepared 1:1 in Laemmli buffer. Prepared protein samples (5–15 µg/lane) were separated via Criterion TGX (Tris-Glycine eXtended)-PAGE precast gels (Bio-Rad Laboratories). The same amount of protein was loaded within each respective tissue across gels. Proteins were transferred onto polyvinylidene difluoride membranes and blocked overnight at 4°C with 5% nonfat dry milk (Tris-buffered saline 0.05% Tween 20 [TBS-T]). Membranes were washed with TBS-T and probed with primary antibodies overnight at 4°C. After primary antibody incubation, membranes were washed and probed with species-appropriate horseradish peroxidase–conjugated secondary antibodies in 5% nonfat dry milk for 2 h at room temperature. Individual bands were detected via chemiluminescence. Antibody information is listed in Supplementary Table 1. Intensity of individual protein bands was quantified with Image Lab (version 6.0.0; Bio-Rad Laboratories) and expressed as a ratio to β-actin. For phosphorylated protein quantification, band intensities were expressed as a ratio to the parent protein (e.g., phosphorylated [p]AKT relative to total AKT). Values are expressed as fold difference with the lean control group set to 1.

Pancreata were immersion fixed in 2% PFA and 2% glutaraldehyde in 0.1 mol/L cacodylate buffer for 1 h at room temperature, followed by 24 h at 4°C. Tissue was immersed in 1% tannic acid, followed by 1% uranyl acetate, and en bloc stained with 1% uranyl acetate. The samples were dehydrated in a graded ethanol series and infiltrated with epon-812 using propylene oxide as the transition solvent; the resin was polymerized at 60°C for 48 h. Samples were sectioned on a Leica EM UC7 at a nominal thickness of 70 nm, collected onto 300 mesh nickel grids, and poststained with 2% uranyl acetate and lead citrate. Imaging was performed on a Tecnai T12 operating at 100 keV with an AMT Imaging CMOS camera. Single images were acquired with the AMT Imaging software. Images for tiling were acquired and stitched with use of SerialEM and IMOD/Etomo, respectively. Segmentation of individual insulin granules was performed using semisupervised machine learning segmentation tools and thresholding. Briefly, mature insulin granules in random quadrants (between 256 and 512 pixels in XY) of large islet electron microscope (EM) micrographs were manually segmented with use of Labkit to create binary masks to serve as ground truth images. At least 5× random ground truth–EM image pairs per animal were used to train a two-dimensional U-Net model with Aivia software (Leica Microsystems). The trained model was applied to the large EM micrographs to generate probability maps that were thresholded to create a binary 8-bit mask with the top 10% highest probability pixels. Each resulting 8-bit object was detected with use of the “analyze particles” function with a threshold of >2,500 pixels². Individual granules were approximated by applying a binary watershed, and the “measurement” function was applied to determine insulin granule area. The area of the nuclei, vessels, and non-β-cells was subtracted from the total tissue area to determine the β-cell cytoplasmic area in each image. To determine the relative density of empty secretory granules, six unique regions within β-cells of each EM image per mouse were randomly selected and the number of mature insulin granules and empty vesicles was manually quantified. We performed all of the above analyses while blinded to image and treatment group identity.

Liver glycogen was determined according to the methodology of Chan and Exton (21), with minor modifications. Briefly, ∼50 mg liver was weighed and bead homogenized in ∼0.5 mL HCl (0.03 mol/L). Extracts were heated at 80°C and subsequently blotted on chromatography paper along with oyster glycogen controls and incubated. Samples were washed in 70% ethanol 3 × 40 min with agitation. Following last wash, samples were acetone rinsed and allowed to dry overnight. Samples were transferred to a 15-mL tube and incubated in a sodium acetate (40 mmol/L)/amyloglucosidase (0.04 mg/mL) mixture for 3 h at 55°C with agitation. The mixture was allowed to cool at room temperature overnight. Glycogen was estimated via glucose plate assay using a glucose standard (no. G6918; Millipore-Sigma) and reaction mix (200 mmol/L Tris HCl, 10 mmol/L MgCl₂, 3 mg/mL ATP, 2 mg/mL NADPH, 1.8 units/mL hexokinase, 0.9 units/mL glucose-6-phosphate dehydrogenase). Absorbance was measured at 340 nm. Data are presented as milligrams of glycogen per gram of liver.

Biochemical liver triglyceride content was determined as previously described (22). Triglyceride content was measured with a commercially available assay (L-Type Triglyceride; Wako Life Sciences). Liver triglyceride concentrations are presented as milligrams per gram of liver wet weight.

RNA was isolated from purified mouse islets with the RNeasy Plus Mini Kit (cat. no. 74134; QIAGEN). Poly-A selection for enrichment of mRNA was performed with use of poly-A magnetic beads (cat. no. E7490L; New England Biolabs) and cDNA library preparation was performed with the NEBNext Ultra kit (cat. no. 7760L; New England Biolabs). Paired-end sequencing (150 bp) was performed on an Illumina NovaSeq 6000 System targeting 50 million reads per sample. Demultiplexed FASTQ files were processed with fastp, version 0.23.1 (23), to trim adapters and remove poor quality sequences. Quantification of sequences was performed with Salmon, version 1.5.2 (24), by aligning to the C57Bl/6J GRCm39 reference assembly. Data were analyzed with EdgeR, version 3.36.0 (25), in R, version 4.1. Differential expression between groups was assessed via the exactTest function (EdgeR).

Student t tests were run for between group comparisons. In experiments that contained more than two groups, one-way ANOVA or two-way ANOVA models were conducted with pairwise comparisons using Tukey or Sidak correction. Brown-Forsythe correction was applied to groups with unequal variance. Data are presented as means ± SE. An adjusted P value of <0.05 was used to determine significance. RNA sequencing (RNAseq) data were analyzed in R, version 4.1, whereas the remainder of statistical analyses was performed with GraphPad Prism (version 9.2.0) (https://www.graphpad.com/).

RNAseq data discussed in this publication are deposited in National Center for Biotechnology Information’s Gene Expression Omnibus (GEO) and accessible through GEO series accession no. GSE202697 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE202697).

Our group previously established that weight cycling worsens glucose tolerance in comparison with body weight–matched controls (13,14). To confirm these findings, we fed mice LFD or HFD according to the paradigm shown in Fig. 1A. Weekly body weight of LFD-fed controls is included as a reference from a published cohort of mice (13) for demonstration that during the weight loss phase (9–18 weeks on diets), body weights converge between LFD and weight-cycled mice (Fig. 1B). Body mass, cumulative energy intake, fat mass, and fat-free mass were not different at the end of the feeding paradigm between obese and weight-cycled animals (Fig. 1B–F). Despite identical cumulative energy intake over the 27 week study, cumulative energy intake within each 9 week block of the study was different between groups (Fig. 1D). Fasting NEFA and insulin concentrations were not different between weight-cycled and obese mice (Fig. 1G). However, consistent with previous findings (13,14), after an intraperitoneal glucose bolus, glucose clearance was delayed in weight-cycled versus obese controls (Fig. 1H).

For determination of whether whole-body insulin action and glucose flux are worsened by weight cycling, hyperinsulinemic-euglycemic clamps were conducted (Fig. 2A). Body mass was not different between obese and weight-cycled mice (Fig. 2B). Arterial glucose was clamped at 120 mg/dL via variable GIR. Neither clamped glucose nor GIR was different between groups (Fig. 2C and D). In addition, EndoR_a and R_d were not different between groups during basal or clamp conditions (Fig. 2E and F), respectively. In comparison with obese mice, clamped insulin (endogenous mouse + exogenous human), but not fasting insulin, concentrations were significantly lower in weight-cycled mice (Fig. 2G). Importantly, exogenously infused insulin (human) was not different between weight-cycled and obese mice (Supplementary Fig. 1A), suggesting that the differences are due to endogenous insulin secretion or insulin processing. Consistent with a secretion phenotype, clamp C-peptide levels were lower in weight-cycled animals (Fig. 2H). For demonstration of the relationship between glucose flux and insulin, arterial insulin was plotted as a function of glucose turnover and EndoR_a (Fig. 2I and J). In this representation, a shift of the curve to the right with a flattened slope is indicative of decreased insulin action. Thus, it appears that weight-cycled mice do not have decreased insulin action compared with obese controls. During hyperinsulinemia, a bolus of [¹⁴C]2DG was administered to determine the glucose metabolic index (i.e., estimate of tissue R_g). Muscle (gastrocnemius, vastus lateralis, and soleus) and epididymal white adipose tissue R_g were deceased in weight-cycled mice (Fig. 1K). However, whole-body glycolysis and glucose storage, which are computed from the detritiation of [3-³H]glucose and glucose turnover, were not different between groups. Similarly, there were no detectible differences in liver glycogen between groups (Supplementary Fig. 1B), whereas liver triglycerides were lower in weight-cycled versus obese mice (Supplementary Fig. 1C). The latter may be related to the effect of chronic hyperinsulinemia on liver steatosis (26,27).

Given that insulin concentrations were lower in weight-cycled mice during the hyperinsulinemic-euglycemic clamp, we performed ex vivo insulin stimulation and assessed phosphorylation of AKT (Ser473) to determine insulin responsiveness in an isolated system. To this end, iWAT and gastrocnemius explants isolated from lean controls and obese and weight-cycled mice were incubated in the absence or presence of insulin (low, 10 nmol/L; high, 100 nmol/L) and subsequently immunoblotted for total AKT and pAKT (Ser473) (Supplementary Fig. 1D). In comparison with lean animals, pAKT (Ser473) was decreased in obese and weight-cycled groups in iWAT with no differences between obese and weight-cycled conditions (Supplementary Fig. 1E). In gastrocnemius muscle, there were also no differences in pAKT (Ser473) between weight-cycled and obese mice (Supplementary Fig. 1F). However, a trend for greater pAKT (Ser473) in weight-cycled versus obese muscle was observed after high-dose insulin. These data suggest that when insulin concentrations are matched in an isolated system, weight-cycled animals do not have worsened proximal insulin signaling compared with obese controls. Taken together, data from hyperinsulinemic-euglycemic clamp and ex vivo insulin responsiveness experiments indicate that lower endogenous insulin rather than impaired insulin action contributes to delayed glucose clearance in weight-cycled compared with obese mice.

To uncover whether weight cycling impairs in vivo pancreatic function, we conducted hyperglycemic clamps on a second cohort of animals. In addition to our standard diet model of obese controls and weight-cycled mice, we included a third group of mice that underwent weight cycling while maintained on HFD throughout the entire experimental period (Fig. 3A). These mice were pair fed (PF) to the weight-cycled group during the weight loss phase for matching of body weight loss. Our reasoning for including this group was to control for the possibility that diet composition during weight loss alters pancreatic response to subsequent weight regain. This PF group is termed “weight-cycled-PF.” Body weight curves, energy intake, and body composition between weight-cycled and weight-cycled-PF groups were not different (Supplementary Fig. 2B–F). In addition, there were no differences in fasting glucose, insulin, or NEFA concentrations following the weight loss period or the weight regain period between weight-cycled and weight-cycled-PF animals, respectively (Supplementary Fig. 2G–I). These data indicate that each weight-cycled group was well matched but also reveal that consuming diets with different macronutrient (high fat vs. low fat) composition during the weight loss phase did not differentially alter fasting indices of glucoregulation after 9 weeks of weight loss or 9 weeks of weight regain.

Hyperglycemic clamp design and procedure are shown in Fig. 3B. The three groups of mice undergoing hyperglycemic clamps did not differ in body weight (Fig. 3C). Arterial glucose was clamped at ∼250 mg/dL via exogenous glucose, which was not different between groups (Fig. 3D). A ∼25% reduction in GIR area under the curve (AUC) (Fig. 3E) (ANOVA, P = 0.1) manifested in both weight-cycled groups versus obese mice, suggesting lower glucose clearance (i.e., GIR during a hyperglycemic clamp is a proxy for glucose tolerance). In comparisons with obese animals, arterial insulin AUC was reduced in weight-cycled groups during hyperglycemia (Fig. 3F). For approximation of early-phase insulin secretion, the insulin change from basal to peak glucose concentration, which occurred within 10 min of glucose infusion, was computed. Percent increase from basal to peak glucose was not statistically different between groups (obese 95 ± 10%, weight-cycled 98 ± 11%, weight-cycled-PF 75 ± 8%); however, percent increase in insulin was approximately threefold higher in obese animals than in either weight-cycled group (Fig. 3G). Basal but not clamped NEFA concentrations were higher in weight-cycled groups relative to obese mice, whereas percent NEFA suppression during hyperglycemia was not different between groups (Fig. 3H and I). Together these data reveal that during clamped hyperglycemia, weight cycling—independent of preceding diet composition—decreases insulin secretion and tends to decrease glucose clearance.

To complement the hyperglycemic clamp studies, we challenged a subset of arterial catheterized mice with an oral mixed meal to determine whether perturbed glucose metabolism in weight-cycled mice manifests in response to oral delivery of mixed nutrients (Fig. 4A). Following oral gavage, a trend for decreased mixed-meal glucose clearance was noted in both weight-cycled groups relative to obese controls (Fig. 4B). Insulin excursions showed a time-dependent response, whereby arterial insulin progressively increased in obese animals from t = 0 min to t = 60 min, while weight-cycled groups displayed a plateau in insulin concentrations between 30 min and 60 min postgavage (Fig. 4C). Compared with obese mice, there was a trend for increased NEFA concentrations in animals that underwent either model of weight cycling (Fig. 4D) (ANOVA, P = 0.06). For weight-cycled-PF mice liver and iWAT mass were elevated compared with obese mice but epididymal white adipose tissue and pancreas mass were not different (Fig. 4E). In sum, these data imply that weight-cycled animals have decreased glucose clearance in response to an oral mixed meal.

Next, we determined whether the reduction in in vivo insulin secretion in weight-cycled mice manifested in isolated pancreatic islets. Isolated islets were perifused with insulin secretagogues (e.g., glucose, IBMX, KCl) ex vivo for determination of dynamic insulin secretion (Fig. 5A). No differences in insulin secretion were detected during low (5.6 mmol/L) or high (16.7 mmol/L) glucose perfusion between obese and weight-cycled islets (Fig. 5B). However, compared with obese controls, weight-cycled islets showed a diminished insulin response to nonspecific phosphodiesterase inhibition with use of IBMX (Fig. 5B), suggesting that cAMP-supported insulin secretion is impaired or less sensitive to stimulation in weight-cycled islets. Membrane depolarization induced insulin secretion using KCl was similar between groups. Insulin content per islet equivalent (i.e., islet of 150 µm diameter) was ∼40% lower in weight-cycled vs. obese mice (Fig. 5C). Immunofluorescence imaging revealed a modest reduction in insulin immunolabeling pixel intensity in contrast-matched images from the weight-cycled islets compared with lean and obese islets, respectively (Fig. 5D). Islet area (as a percent of total pancreas area) was increased in obese mice compared with lean mice (Fig. 5E), whereas weight-cycled islets failed to compensate for hyperglycemia by increasing β-cell area. Furthermore, obese islets were larger than lean and weight-cycled islets (Fig. 5F). Obese islets contained approximately twice as many β-cells as weight-cycled or lean islets (Fig. 5G). However, the ratio of α-cells–to–β-cells was not different between obese and weight-cycled mice (Fig. 5H).

To identify the mechanism behind impaired stimulated insulin secretion caused by weight cycling, we collected islets from obese and weight-cycled mice for bulk RNAseq. Obese and weight-cycled islet transcriptomes were segregated by unbiased linear dimensional reduction using multidimensional scaling (Fig. 6A). Differential expression comparing obese and weight-cycled samples detected 361 significantly different genes with 114 upregulated genes and 247 downregulated genes (Fig. 6B). Gene ontology of differentially expressed genes identified upregulation of cell cycle–related pathways and downregulation of pathways associated with organization and protein localization in weight-cycled islets compared with obese islets (Fig. 6C). Total normalized counts for genes included in cell-cycle pathways were low but significantly different between obese and weight-cycled islets (Fig. 6D). To investigate potential differences in β-cell proliferation, we performed immunofluorescence imaging and quantification for the nuclear marker Ki-67 (Supplementary Fig. 3). Ki-67⁺insulin⁺ cells were sparse in islets and did not differ between groups (lean 0.46%, obese 0.23%, and weight cycled 0.16% of intraislet cells). Multiple drivers of β-cell identity and function, Nkx6.1, Mafa, Pdx1, Ucn3, and Slc2a2, were also downregulated in weight-cycled islets compared with obese islets (Fig. 7A). Conversely, transcripts for insulin (Ins1 and Ins2) and G6pc2 were elevated in weight-cycled islets. Therefore, we determined whether posttranscriptional changes occurred between groups using immunofluorescence imaging. Immunolabeling for NKX6.1 revealed a qualitative reduction in NKX6.1 expression of weight-cycled islets (Fig. 7B). Likewise, nuclear MAFA was also reduced in islets from weight-cycled mice (Fig. 7C). Because markers of β-cell identity were reduced, we also immunolabeled for GLUT2, a β-cell–specific GLUT. GLUT2 localized to the plasma membrane of insulin⁺ cells but was less uniform and had reduced signal intensity in weight-cycled mice (Fig. 7D). Unmerged images are shown in Supplementary Figs. 4–6.

For further identification of whether changes in drivers of β-cell function impaired insulin granule formation in weight-cycled animals, transmission electron microscopy was used (Fig. 8A). The percentage of cytoplasmic insulin-loaded granules was 22% lower (P < 0.05) in β-cells from weight-cycled compared with obese mice (Fig. 8B), despite no differences in the average number of total vesicles per field of view (FOV) (Fig. 8C). The percentage of empty vesicles was ∼30% higher (P = 0.053) in weight-cycled compared with obese islets (Fig. 8D). Taken together, the islet studies support that reduced insulin secretion observed in weight-cycled mice, compared with those from obese mice, is linked with both an insufficient transcriptional/posttranscriptional response to weight regain and impaired insulin granule loading.

Weight cycling has been identified as a risk factor for type 2 diabetes in humans (10), yet the mechanism(s) by which weight cycling alters nutrient flux and/or insulin resistance (compared with obesity alone) is incomplete. Along these lines, it is unclear whether repeated fluctuations in body weight accelerate β-cell fatigue and associated hyperglycemia. Peripheral insulin resistance, defined as liver, muscle, and adipose tissue resistance to insulin action, precedes diabetes development. Similarly, β-cell insulin resistance can also precede type 2 diabetes (28). Long-term weight loss has repeatedly been shown to improve glucose homeostasis by increasing peripheral insulin sensitivity and by maintaining tight pancreatic insulin-to-glucose balance (29–35). Despite these findings, the effects of weight regain on glucose fluxes and insulin action are not clear.

Herein, we used a mouse model of diet-induced weight cycling to systematically interrogate the metabolic phenotype of glucose-intolerant weight-cycled mice with two primary objectives: 1) pinpoint the organ/tissue site of glycemic dysfunction caused by weight cycling and 2) uncover the mechanisms evoking impaired glucoregulation in weight-cycled mice. We hypothesized that weight cycling would worsen peripheral insulin action and glucose clearance. This hypothesis was founded in part by prior evidence that weight-cycled mice have attenuated insulin signaling in adipose tissue and liver compared with obese controls following intraperitoneal insulin administration (14). In contrast to our hypothesis, weight cycling does not worsen whole-body insulin action compared with obesity alone but, rather, decreases pancreatic insulin secretion. These findings suggest that inadequate pancreatic adaptation to fluctuations in body weight is a central player in the deterioration of glycemic control caused by weight cycling in mice.

It is well established that pancreatic β-cells adapt to chronic overnutrition by increasing β-cell mass and hyperinsulinemia (4,36). Yet, the extent to which islets are adaptable to multiple bouts of weight gain and weight loss is unclear. The pattern of nutrient loading (weight gain), unloading (weight loss), and reloading (weight regain) may invoke a state of pancreatic inflexibility. This stimulus of nutrient loading/unloading is important and relevant to the human condition given that many individuals transition through multiple bouts of caloric excess and caloric restriction during their life span—the latter of which is challenging to maintain for reasons previously discussed (37). Nonetheless, uncovering the molecular mechanisms contributing to pancreatic flexibility or loss thereof is of clinical relevance and importance.

We find that during in vivo hyperglycemia (clamped), weight-cycled mice secrete less insulin and require lower GIR (generally indicative of lower glucose clearance) to maintain clamped hyperglycemia. The relative reduction in GIR may be affected by possible differences in the suppression of endogenous glucose production. However, we are not able to make conclusions regarding EndoR_a, given that glucose was not isotopically labeled during the hyperglycemic clamp. On the other hand, whole-body insulin action (determined from hyperinsulinemic-euglycemic clamp) does not appear to be different between obese and weight-cycled animals. Indeed, glucose fluxes (R_a and R_d) assessed during euglycemia—and clamped hyperinsulinemia—did not differ between weight-cycled and obese animals. However, arterial insulin levels were lower in weight-cycled mice, an effect that likely explains the decrease in tissue R_g in weight-cycled mice, which disappears when R_g is normalized to its respective insulin concentration. Yet, it should be noted that the relationships between insulin concentrations and tissue R_g are likely nonlinear. Nonetheless, ex vivo insulin responsiveness is not different between obese and weight-cycled animals when equal amounts of insulin are present.

Although the current study was not designed to interrogate whether exogenous insulin regulates endogenous insulin secretion, it is prudent to discuss this possibility given different arterial insulin concentrations during clamp experiments between obese and weight-cycled animals. In humans, exogenous insulin reportedly potentiates (38–41), inhibits (42–46), or has no effect (47,48) on endogenous insulin secretion during euglycemia. Studies that show insulin-stimulated insulin secretion appear to be dependent on baseline insulin sensitivity. During euglycemia, a group of investigators performed multidose insulin infusions while sampling portal and peripheral blood for the effects of insulin infusion on endogenous insulin secretion (estimated via C-peptide) and hepatic clearance (44). No change in hepatic insulin clearance was noted, and C-peptide concentrations decreased in response to increasing exogenous insulin dose (i.e., negative feedback inhibition). Some data show that hyperinsulinemia accelerates the metabolic clearance of C-peptide (40), which may lead to an underestimation of endogenous insulin secretion based on C-peptide deconvolution modeling. Conversely, in more recent studies where discriminatory insulin measurement techniques were used (biologically equivalent insulin analog “B28-Asp-insulin”) investigators found that pre-exposure to exogenous insulin during euglycemia enhanced insulin secretion in response to graded increases in exogenous glucose in healthy individuals but not individuals with impaired glucose tolerance or patients with type 2 diabetes (39). The same group also reports that insulin pre-exposure augments arginine-induced insulin secretion in humans (49). In isolated human islets, exogenous insulin increases de novo insulin synthesis (50) and enhances GSIS (51). Along these lines, glucose-stimulated insulin synthesis requires autocrine insulin action (52). Treatment of isolated islets from C57BL/6J mice with an insulin mimetic (L-783,281) augmented insulin release in perifused islets, but only under hyperglycemic conditions (11 vs. 3 mmol/L glucose) (53). In sum, the precise regulation by exogenous insulin on endogenous insulin production appears to be linked with baseline insulin sensitivity, but conclusive evidence is still lacking in both animal models and humans.

With regard to isolated islets, GSIS alone is not dampened by weight cycling. In contrast, cAMP-driven insulin secretion during exposure to a nonspecific phosphodiesterase inhibitor (i.e., IBMX) and high glucose was reduced in weight-cycled compared with obese mice. This impairment in islet functional capacity is linked with reductions in insulin content (per islet equivalent), less insulin⁺ immunostaining, and reduced percentage of insulin-loaded granules in weight-cycled mice. Numerous factors can drive altered insulin secretion in β-cells, such as signaling by inflammatory cytokines, endoplasmic reticulum stress, apoptosis, loss of specific endocrine identity, and cellular metabolism (reviewed in 54). Our transcriptomics approach revealed reductions in key transcription factors from lean, obese, and weight-cycled mice. Mafa, Nkx6.1, and Pdx1 are involved in regulating β-cell proliferation and expansion in response to nutrients and are decreased in models of type 2 diabetes (55). Pathway analysis from RNAseq suggested increased proliferation. However, few cells positively stained for Ki-67 within islets, suggesting that β-cell proliferation is low in obese and weight-cycled mice after 27 weeks of diet. Genes associated with proliferation are typically also associated with DNA-damage repair. We did not identify differences in expression for additional genes associated with DNA repair (such as Xrcc2, Xrcc3, Rad52) in the RNAseq data set between obese and weight-cycled mice, but our data do not rule out DNA damage as a mechanism for impaired insulin synthesis or secretion in weight cycling.

β-Cell–specific deletion of one or more of the “identity” transcription factors (Mafa, Nxk6.1, or Pdx1, among others) causes overt hyperglycemia due in part to decreased insulin synthesis/secretion (56). Glycemic control can be improved by sustained expression of Mafa in the hyperglycemic leptin receptor–deficient (db/db) mouse model (57). Haploinsufficiency for Pdx1 is linked with increased β-cell apoptosis (58,59), and Pdx1-insufficient mice have impaired glucose tolerance despite having relatively normal GSIS in perifused islets (58). Our model of weight cycling closely mimics the metabolic phenotype and β-cell characteristics of Pdx1-haploinsufficient mice, which display decreased expression of Slc2a2, islet area, and adaptation to diet-induced obesity. Collectively, these findings are in line with the idea that β-cell proliferation and expansion are stalled in previously obese mice such that when challenged with a second bout of nutrient excess they are unable to mount a sufficient response (i.e., β-cell fatigue).

Weight cycling decreases several β-cell transcription factors critical to maintaining pancreatic flexibility, yet the molecular mechanisms causing the downregulation of these β-cell identity markers is not inherently clear. Prior research reveals that increased oxidative stress decreases expression of Mafa, Nkx6.1, and Pdx1 in human islets (55). Similarly, db/db mice exhibit a reduction in these markers that is restored via transgenic expression of the glutathione peroxidase-1 antioxidant enzyme (55). Transcriptionally, we did not find clear evidence for increased oxidative stress between weight-cycled and obese islets; however, it is possible that functional differences could manifest. In addition, evidence demonstrates that intermediary metabolism of glucose and other substrates contributes to insulin secretion and proliferative activity in islets. Thus, rapid changes in nutrient load during diet transitions may pose a significant burden on β-cells, which are highly susceptible to increased oxidative stress (60–62). Interestingly, the expression of Ins1 and Ins2 was increased in weight-cycled mice; however, this was accompanied by lower insulin content and decreased secretion in vivo and ex vivo. Moreover, the percentage of insulin-loaded granules was lower in weight-cycled β-cells, together suggesting that posttranscriptional insulin gene regulation is implicated in the functional loss of insulin secretion. The question of whether insulin granule formation/trafficking or membrane docking is impaired in weight-cycled islets awaits further testing. While the present findings do not pinpoint the precise molecular signal or mechanism(s) responsible for insufficient β-cell responsiveness, we unveil a model in which β-cells in mice that weight cycle fail to effectively compensate for increased insulin demand during chronic hyperglycemia. This in turn results in reduced peripheral insulin availability and impaired glucose tolerance. Additional studies are needed to determine the precise regulation of changes in β-cell fate and regulation of β-cell mass in response to body weight fluctuations.

In summary, we report that weight cycling–induced glucose intolerance is unlikely attributable to worsened peripheral insulin resistance in mice; rather, it is linked with inadequate insulin secretion. Our findings support the notion that repeated bouts of weight gain challenge pancreatic flexibility. Specifically, reductions at both the transcript and protein levels of the β-cell transcription factors NKX6.1 and MAFA in weight-cycled mice suggest a loss of adequate β-cell compensation normally observed with obesity alone. Additional mechanistic studies are needed to confirm whether restoration of these β-cell transcription factors causally resolves worsened glucose homeostasis evoked by weight cycling.

Acknowledgments. We thank Alec Rodriguez, Vitrag Patel, and Caroline Hegemann of Vanderbuilt University (VU) for mouse husbandry and technical assistance. We thank Dr. David Wasserman, Dr. Owen McGuinness, and Dr. Louise Lantier of Molecular Physiology & Biophysics, VU, VMMPC, for their intellectual contributions regarding glucose fluxes. We also thank Dr. Roland Stein, Dr. Xin Tong of Molecular Physiology & Biophysics, VU, and Dr. Marcela Brissova of Division of Diabetes, Endocrinology and Metabolism, VUMC, for kindly providing primary antibodies and intellectual support for pancreas studies.

Funding. This project was funded by a Veterans Affairs Merit Award 5I01BX002195 and an American Heart Association (AHA) Innovation Award (19IPLOI4760376) to A.H.H. M.A.C. is supported by NIDDK (1F31DK123881) and was supported by METP (T32 DK007563-31) during data curation. N.C.W. was supported by an American Physiological Society Postdoctoral Fellowship and the Molecular Endocrinology Training Program (METP) (T32 DK007563-31) during data curation and analysis and is currently supported by the AHA (21POST834990). M.B. was supported by the Gastroenterology Training Grant (DK007673) and the AHA Strategically Focused Research Network award (17SFRN33520017) during data curation. H.L.C. was supported by METP (T32 DK007563-31) during data curation and is currently supported by AHA (20POST35120547). J.N.G. is supported by the METP (T32 DK007563-31). R.A.D. is supported by Human Islet Research Network (U01DK120447) and dkNET New Investigator Pilot Program in Bioinformatics (5U24DK097771). A.H.H. is supported by a Career Scientist Award from the Veterans Affairs (IK6 BX005649). We also acknowledge the following Vanderbilt University (VU) and Vanderbilt University Medical Center (VUMC) core facilities: VUMC Hormone Assay & Analytical Services Core (NIH DK059637 and DK020593), VU Metabolic Mouse Phenotyping Center [VMMPC (NIH DK059637; http://www.vmmpc.org)], Translational Pathology Shared Resource (NCI/NIH Cancer Center Support Grant 5P30 CA68485-19), VUMC Islet Procurement & Analysis Core (NIH DK020593), and the Cell Imaging Shared Resource (CISR) Core (NIH DK020593).

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

Author Contributions. N.C.W. and M.A.C. conceived and designed research, performed experiments, interpreted results, and drafted the manuscript. M.B. performed experiments, interpreted results, and reviewed and edited the manuscript. H.L.C. performed experiments, interpreted results, and reviewed and edited the manuscript. J.N.G. performed experiments, interpreted results, and reviewed and edited the manuscript. R.A.e.D. interpreted results and reviewed and edited the manuscript. A.H.H. conceived and designed research, interpreted results, and reviewed and edited the manuscript. All authors approved the final version of the manuscript. A.H.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.