Toll-Like Receptor 3 Influences Glucose Homeostasis and β-Cell Insulin Secretion

The prevalence of obesity and type 2 diabetes is radically rising worldwide, causing serious socioeconomic problems. Type 2 diabetes is a heterogeneous disorder with a complex pathogenesis characterized by the progressive development of hyperglycemia, dyslipidemia, and impaired insulin secretion from pancreatic β-cells (1). Experimental and clinical studies have demonstrated that inflammation mediates its pathophysiology, with Toll-like receptors (TLRs) playing critical functions (2).

In addition to the recognition of pathogen-derived structures, TLRs can activate innate immunity by recognition of endogenous molecules, such as lipids, fatty acids (FAs), and other mediators, that are elevated during tissue stress and cell death in chronic inflammatory diseases (3). Certain TLRs have been implicated in glucose metabolism and type 2 diabetes—TLR2- and TLR4-deficient mice are protected against diet-induced obesity and insulin resistance (4–9), and polymorphisms in human TLR4 have been linked to a lower risk of type 2 diabetes (10,11).

TLRs are expressed predominantly in immune cells, but their expression in nonhematopoietic cells suggests that they have “nonimmune” functions. Notably, TLR3, which is activated by viral double-stranded RNA (dsRNA) and mRNA from dying cells, is highly expressed in pancreatic β-cells (12,13). Further, infectious agents, such as dsRNA, that bind to TLR3 accelerate β-cell dysfunction and apoptosis, causing insulitis and autoimmunity (14). Although TLR3 has been examined in type 1 diabetes, its role in autoimmune diabetes remains unclear (15–18). Whether TLR3 has a role in obesity and the development of type 2 diabetes is unknown. In this study, we examined TLR3 function in glucose metabolism and type 2 diabetes–related phenotypes in animals and humans.

All mice were housed and used as specified by the Stockholm North Committee for Experimental Animal Ethics and the Swedish National Board for Laboratory Animals. The hyperglycemic clamp technique was performed at the National Mouse Metabolic Phenotyping Center, University of Massachusetts Medical School, with approval by the local Institutional Animal Care and Use Committee (Worcester, MA). Male C57BL/6 (Charles River Laboratories, Sulzfeld, Germany) and Tlr3^−/− mice fully backcrossed (N14) onto C57BL/6 (19,20) were weaned at 4 weeks and fed a high-fat diet (HFD) (34.9% fat; Altromin, Lage, Germany) or normal chow diet from age 5 weeks for 20 weeks, unless otherwise stated.

Intraperitoneal glucose tolerance test (ipGTT), intraperitoneal insulin tolerance test (ipITT), and intraperitoneal pyruvate tolerance test were performed in overnight-starved mice (unless otherwise stated) by injecting glucose (1 g/kg body weight; B. Braun Melsungen, Melsungen, Germany), insulin (0.8 units/kg body weight; Actrapid; Novo Nordisk, Bagsværd, Denmark), or pyruvate (1.5 g/kg body weight; Sigma-Aldrich, St. Louis, MO), respectively, and monitoring glucose concentrations (Abbott Scandinavia AB, Solna, Sweden) in tail-vein blood. HOMA-insulin resistance (IR) was calculated as fasting glucose (mg/dL) × fasting insulin (mU/L)/405.

After the HFD for 26 weeks, a survival surgery was performed at 4–5 days before clamp experiments to establish an indwelling catheter in the jugular vein. Mice were fasted overnight before the start of the experiment. A hyperglycemic clamp was conducted in conscious mice, starting with an infusion of 20% dextrose to quickly reach a target hyperglycemia (∼300 mg/dL glucose level) and maintain hyperglycemia by adjusting glucose infusion rates. Plasma samples were collected before the start of infusion (baseline) and at indicated time points to measure glucose, insulin, and C-peptide levels. Insulin clearance was estimated by the ratio of fasted C-peptide to insulin. At the end of the clamps, mice were killed.

Mice were killed with CO₂ after overnight starvation, and blood was collected for plasma analysis. Organs were dissected after vascular perfusion with sterile RNase-free PBS, unless otherwise indicated. Tissues were snap-frozen for RNA and protein analysis or processed for immunohistochemistry.

Whole blood was analyzed by a scil Vet abc hemocounter. Total plasma triglycerides (TGs) and cholesterol were measured using a colorimetric kit per the manufacturer's instructions (Randox Laboratories Ltd., Crumlin, U.K.). Insulin (Crystal Chem Inc., Downers Grove, IL), adiponectin (R&D Systems, Minneapolis, MN), leptin (PeproTech, Rocky Hill, NJ), and serum amyloid A (SAA) (Life Technologies, Carlsbad, CA) were measured by ELISA per the manufacturers' instructions. Plasma TG and cholesterol lipoprotein profiles were examined as previously reported (21).

Mice were fed the HFD for 34 weeks and fasted 8 h before the experiment. VLDL synthesis was assessed after irreversibly blocking VLDL catabolism using 10% (w/v) tyloxapol (500 mg/kg i.v.; Sigma-Aldrich). Mice were bled from the tail into EDTA-coated tubes at the indicated times. Plasma TGs were measured as described above, normalized to baseline TGs, and expressed as fold increase.

Insulin signaling in the liver was assessed by Western blot analysis of basal phosphorylated protein kinase B (PKB)/AKT in TLR3-deficient and control mice. In a separate experiment, 10 min before freezing the tissue in liquid nitrogen, mice were injected with insulin (2 units/kg body weight; Actrapid; Novo Nordisk). Number of mice and duration of HFD are indicated in the figure legends.

Pancreatic tissue was dissected from the surrounding tissues, and specimens from the corpus/cauda regions were fixed in formalin and embedded in paraffin. Sections (4–10 µm) were stained with hematoxylin and eosin, insulin, glucagon, and somatostatin by immunohistochemistry, as described (22). The average islet area was calculated per total pancreatic area in three sections per sample. The detection of apoptotic cells was performed by TUNEL staining technique using an In Situ Cell Death Detection Kit (Roche, Mannheim, Germany) per the manufacturer's instructions.

Islets of Langerhans were isolated from Tlr3^−/− and control mice by collagenase digestion in Hanks’ balanced salt solution, followed by sedimentation, and then cultured for 20–24 h as described (23). After culture, equal-sized islets were preincubated as previously reported (23), followed by 60-min batch incubation (3 islets per tube in triplicate, 2.8 or 16.7 mmol/L glucose), perifusion, RNA extraction, or Western blot. Islets from the batch incubations were treated with acid-ethanol to extract cellular insulin (23).

Secreted insulin and islet insulin content were analyzed by RIA using ¹²⁵I-labeled insulin and anti-porcine insulin (Endocrinology and Diabetes Unit, Karolinska University Hospital Solna, Stockholm, Sweden (24)). Perifusion experiments were performed as described (23). Perfusate samples were collected every minute, and secreted insulin was analyzed by radioimmunoassay.

Polyinosinic-polycytidylic acid (polyI:C) was administered three times per week simultaneously with the HFD in young C57BL/6 mice (age 6 weeks). Mice received 12 intraperitoneal injections of polyI:C (100 µg/mouse, InvivoGen, Toulouse, France) or NaCl (Fresenius Kabi, Uppsala, Sweden). Chow-fed C57BL/6 mice (age 18 weeks) were treated 17 times with polyI:C (100 µg/mouse) or NaCl. Tlr3^−/− mice received 11 intraperitoneal injections of polyI:C or NaCl during HFD. To assess insulin secretion, C57BL/6 mice (age 9 weeks) received 17 intraperitoneal injections of polyI:C or NaCl during HFD.

Total RNA was isolated using the RNeasy Lipid Tissue Mini Kit (Qiagen, Hilden, Germany) or RNeasy Micro Kit (Qiagen) and reverse-transcribed with Superscript II, random hexamers (pdN6), and RNasin (Life Technologies). RNA concentration was measured by spectrophotometry (Thermo Scientific), and RNA quality was assessed on a Bioanalyzer (Agilent Technologies, Waldbronn, Germany). cDNA was amplified on an ABI 7900HT (Applied Biosystems, Foster City, CA) by real-time PCR using primers and probes (Applied Biosystems) for the selected genes. Data were calculated as 2^-ΔΔC_T, where ΔΔC_T = ΔC_T (sample) – ΔC_T (calibrator = average C_T values of all samples in each group) and ΔC_T is the target gene C_T minus the C_T of Hprt.

Pooled islets from two to three mice per sample were used for Western blot as described (25).

For insulin-signaling analysis, a piece of liver was lysed and homogenized in ice-cold buffer (137 mmol/L NaCl, 2.7 mmol/L KCL, 1 mmol/L MgCl₂, 1% Triton X-100, 10% glycerol, 20 mmol/L Tris [pH 7.8], 10 mmol/L NaF, 1 mmol/L EDTA, 5 mmol/L sodium pyrophosphate, 0.5 mmol/L Na₃VO₄, 1 μg/mL leupeptin, 0.2 mmol/L phenylmethyl sulfonyl fluoride, 1 μg/mL aprotinin, and 1 μmol/L microcystin). Homogenates rotated 30 min at 4°C before centrifugation at 12,000g for 15 min at 4°C.

Primary antibodies were used at the following dilutions: mouse anti–VAMP-2, 1:10,000 (Synaptic Systems GmbH, Goettingen, Germany); rabbit anti-glucokinase (Gck), 1:500 (Santa Cruz Biotechnology, Dallas, TX); mouse anti–β-actin, 1:20,000 (Sigma-Aldrich); and rabbit anti–caspase-3 and rabbit anti-cleaved caspase-3, both 1:1,000 (Cell Signaling Technology, Danvers, MA). After incubation with horseradish peroxidase–conjugated anti-mouse or anti-rabbit, bands were visualized by chemiluminescence (Pierce, Rockford, IL).

Male participants from the Polca (n = 625) and Olivia (n = 306) cohorts were included. Polca comprises healthy 50-year-old individuals who were free of coronary heart disease and recruited at random using a population registry (26), and Olivia includes healthy individuals aged 33–80 years, recruited for the Precocious Coronary Artery Disease (PROCARDIS) study (27). Subjects with type 2 diabetes (defined as diagnosis, use of antidiabetic medication, or fasting glucose ≥7 mmol/L) were excluded. All participants were genotyped using the Illumina Infinium 1M or 610K platforms at the SNP&SEQ Technology Platform, Uppsala University, Uppsala, Sweden, or Centre National de Génotypage, Paris, France (Supplementary Table 1 reports the clinical characteristics and genotyping methods). SNPper (28) was used to identify single nucleotide polymorphisms (SNPs) in the TLR3, of which 24 were available in the Polca and Olivia data sets. Because rs3775291 was the only SNP in the coding region, we focused on this. The association between rs3775291 and metabolic phenotypes was examined.

Results from the animal studies are expressed as mean ± SEM. The Mann-Whitney U test was used for group comparisons, and the paired t test of log-transformed values was used for pairwise observations. Associations between rs3775291 and metabolic phenotypes in the human cohorts were analyzed by linear regression. Skewed variables were natural log-transformed before analysis. An additive genetic model was assumed, and adjustments were made for age and BMI. Fixed-effects inverse variance meta-analyses were performed using METAL software (29). A P value of <0.05 indicated a significant association.

After being fed the HFD or chow diet for 20 weeks, Tlr3^−/− and control mice experienced increased body weight, which did not differ between mice with either diet (Fig. 1A). To examine the effects of TLR3 deletion on glucose homeostasis, we performed ipGTT. TLR3-deficient mice had better responses to glucose versus control mice, irrespective of diet (Fig. 1B). Baseline glucose did not differ between control and Tlr3^−/− mice fed chow or the HFD (Fig. 1C).

Tlr3^−/− mice fed the HFD had elevated circulating insulin under fasting conditions versus control mice (Fig. 1D and E), which rose further after the glucose challenge (Fig. 1D). Similarly, incremental insulin was increased in HFD-fed Tlr3^−/− mice compared with control mice (Fig. 1F). Insulin responses did not differ between chow-fed Tlr3^−/− and control mice (Fig. 1D and E). HOMA-IR increased in HFD-fed Tlr3^−/− mice versus control mice (Fig. 1G). Circulating levels of adiponectin and leptin (Supplementary Fig. 1A and B, respectively) were similar between groups.

To determine the effects of TLR3 deficiency on glucose metabolism, the hyperglycemic clamp technique was performed in HFD-fed mice (Fig. 1H–J). Body weight was similar in HFD-fed control and Tlr3^−/− mice, and plasma glucose rose quickly and plateaued at ∼300 mg/dL during infusion in both genotypes (Fig. 1H). Glucose infusion rates were significantly higher in HFD-fed Tlr3^−/− versus control mice (Fig. 1I). Insulin secretion in Tlr3^−/− mice climbed, as evidenced by elevated plasma insulin compared with control mice (Fig. 1J). Further, plasma C-peptide levels increased slightly during clamping in Tlr3^−/− mice (Fig. 1K), reflecting greater insulin production. The C-peptide-to-insulin ratio did not differ between the genotypes, indicating that insulin clearance was similar between the groups (Fig. 1L).

By ipITT, insulin responses did not differ between Tlr3^−/− and control mice fed chow (Fig. 2A) or the HFD (Fig. 2B) for 20 weeks. To rule out the influence of starvation, ipITT was repeated in freely HFD-fed mice, yielding the same result (Fig. 2C). Aged mice that were fed the HFD for 12 months still responded better to the glucose challenge (Fig. 2D) and had increased insulin levels upon glucose stimulation (Fig. 2E), without becoming insulin resistant (Fig. 2F). Despite the rise in the HOMA-IR index, the insulin responses did not differ between Tlr3^−/− and control mice by ipITT, indicating that the increased insulin in Tlr3^−/− mice did not affect whole-body insulin resistance, irrespective of age or metabolic conditions. Liver-specific insulin sensitivity was assessed by analyzing insulin signaling (Fig. 2G and H) and gluconeogenesis (Fig. 2I and J). Basal phosphorylated AKT was reduced in TLR3-deficient mice compared with control mice fed the HFD (Fig. 2G). Insulin-stimulated AKT activation was similar between the genotypes (Fig. 2H). Reduced phosphoenolpyruvate carboxykinase (PEPCK) mRNA in TLR3-deficient mice (Fig. 2I) was accompanied by a trend toward reduced glucose output from the liver compared with control mice (Fig. 2J), demonstrating the ability of insulin to suppress glucose production in TLR3-deficient mice.

Systemic inflammation was assessed by analyzing peripheral blood cells and levels of SAA protein. Whereas total blood lymphocyte, granulocyte, and monocyte numbers did not differ between the genotypes (Fig. 3A), TLR3-deficient mice fed the HFD displayed increased levels of serum SAA (Fig. 3B), indicating augmented systemic inflammation despite the lack of TLR3.

Histopathological analysis of the pancreas showed no difference in the basic structure of the Langerhans cells between control and Tlr3^−/− mice (Fig. 5C). Thus, the size, number, and shape of islets were in the normal range. No bleeding, necrosis, or infiltration of granulocytes could be detected.

Dyslipidemia is a common comorbidity in patients with obesity and type 2 diabetes, characterized by increased FA influx into the liver and higher plasma TG. Despite no difference in weight between control and Tlr3^−/− mice, the latter had decreased circulating TG (Fig. 4A). This was attributed to reduced TG concentrations in the VLDL fraction (Fig. 4B). Similarly, total cholesterol was lower overall and in the VLDL fraction in Tlr3^−/− mice (Fig. 4C and D). Moreover, Tlr3^−/− mice had decreased circulating free FAs (FFAs), the products of VLDL-TG hydrolysis, versus control mice (Fig. 4E). No differences in LDL, HDL, TG, and cholesterol were observed (Fig. 4B and D).

In examining the factors influencing lipoprotein levels, we found that mRNA expression of 3-hydroxy-3-methyl-glutaryl-CoA reductase (HMGCR), the rate-limiting enzyme in cholesterol biosynthesis, was significantly reduced in the liver of Tlr3^−/− mice (Fig. 4F). Further, sortilin-1, which has been implicated in the regulation of VLDL catabolism, rose in the liver of Tlr3^−/− mice (Supplementary Fig. 2A). By inhibiting lipoprotein lipase–dependent TG hydrolysis with tyloxapol, we show that Tlr3^−/− mice had significantly lower TG synthesis (Fig. 4G). Interestingly, circulating TG was raised significantly upon TLR3 stimulation with the TLR3 ligand polyI:C (Supplementary Fig. 2B).

After glucose challenge, islets of Langerhans from Tlr3^−/− mice secreted significantly more insulin than control mice (Fig. 5A), but islet insulin content was unchanged (Fig. 5B). Pancreatic islet structure was similar between TLR3-deficient and control mice, wherein insulin-producing cells were central and glucagon- and somatostatin-expressing cells lay in the periphery (Fig. 5C). Further, islet morphometry showed the islet area was similar between groups (Fig. 5D).

Because TLR3 stimulation with synthetic dsRNA might increase cell death, we performed the TUNEL assay (Fig. 5E) and analyzed cleaved caspase-3 protein (Fig. 5F) but could not detect increased islet apoptosis.

Glucose is the principal signal for insulin release from β-cells. To determine the mechanism of increased insulin secretion in Tlr3^−/− mice, we measured Gck protein expression, the rate-limiting component of glucose metabolism in β-cells that phosphorylates glucose to glucose-6-phosphate (Fig. 6A). Islets in Tlr3^−/− mice expressed more Gck compared with control mice fed the HFD, indicating improved glucose metabolism in the islets, which could affect greater insulin secretion. Furthermore, we measured mRNA levels of GLUTs in pancreatic islets and peripheral tissues (Fig. 6B and Supplementary Fig. 3A–C). GLUT2 did not differ significantly between Tlr3^−/− and control mice in the islets or liver, and GLUT4 expression was similar in adipose tissue and muscle.

Next, we examined glucose-stimulated insulin secretion from islets in vitro by perifusion of islets from Tlr3^−/− and control mice. Stimulation with high glucose (16.7 mmol/L) induced biphasic insulin secretion from chow-fed, under physiological conditions, (Fig. 7A) and HFD-fed (Fig. 7E) control and TLR3-deficient mice. The enhancement in Tlr3^−/− mice involved the first and second phases of insulin secretion in response to high glucose in chow-fed (Fig. 7A–C) and HFD-fed mice (Fig. 7E–G). In addition, depolarization with K⁺ increased insulin secretion in Tlr3^−/− islets versus control mice (Fig. 7A, D, E, and H), indicating that the rise in insulin was not restricted to stimulation by glucose.

The fusion of insulin-containing vesicles with the β-cell membrane and subsequent release of insulin constitute the last crucial step of glucose-stimulated insulin secretion. By Western blot, the exocytotic protein VAMP-2 was upregulated in the pancreatic islets of Tlr3^−/− mice (Fig. 7I), which we speculate facilitates enhanced insulin secretion in Tlr3^−/− islets.

To examine TLR3 function in our model, we treated C57BL/6 chow-fed (Fig. 7J and K) and HFD-fed mice (Fig. 7L–N) with the TLR3 ligand polyI:C and analyzed its effects on glucose control and VAMP-2 expression. PolyI:C reversed the glucose response in C57BL/6 mice under physiological conditions (Fig. 7J) and on the HFD (Fig. 7L). This impairment was accompanied by downregulation of VAMP-2 in pancreatic islets (Fig. 7K and M) and a trend toward reduced insulin secretion in polyI:C-treated C57BL/6 mice fed the HFD (Fig. 7N). Treatment of TLR3-deficient mice with polyI:C had no effect on glucose response (Supplementary Fig. 4A), baseline glucose (Supplementary Fig. 4B), or body weight (Supplementary Fig. 4C).

We next studied the function of lipotoxicity in TLR3-deficient mice, determining whether the improved β-cell function in Tlr3^−/− mice was due to enhanced insulin sensitivity and plasma lipids by reducing β-cell lipotoxicity. Upon stimulation of islets in vitro with palmitate, glucose-induced insulin secretion increased in Tlr3^−/− versus control mice (data not shown). Previous treatment with palmitate inhibited insulin secretion and decreased islet insulin content (Supplementary Fig. 5A–C) in the control and knockout mice. Insulin secretion was inhibited by 60% in Tlr3^−/− animals versus 26% (as a percent of no previous exposure to palmitate) in the control mice (Supplementary Fig. 5A and B). Islet insulin content in the knockout fell to a similar extent as in the control on culture in palmitate (66% vs 62%), likely due to inhibition of proinsulin biosynthesis.

To determine whether TLR3 is linked to metabolic phenotypes in humans, we investigated the effect of genetic variants in TLR3 on insulin-related traits. We searched for all SNPs in TLR3 that could be identified by SNPper (28) and had been previously genotyped in the population studies (Fig. 8). We found 24 SNPs (Fig. 8); however, only 1, rs3775291, was found within the coding region. rs3775291 is predicted to damage TLR3 function (30) and has been studied primarily with regard to TLR3 function in viral infection (31–33). We measured the effects of rs3775291 on insulin-related traits.

A meta-analysis of rs3775291 in two healthy Swedish cohorts was performed, and its minor allele (T) was associated with elevated fasting insulin levels (β = 0.048, SE = 0.023, P = 0.0340) but not blood glucose levels (β = −0.005, SE = 0.005, P = 0.3365), consistent with our findings in Tlr3^−/− mice.

Although TLRs have been implicated in type 2 diabetes, there is little evidence of their involvement, other than TLR2 and TLR4. In this study, we examined the function of TLR3 in the progression of diet-induced obesity in a mouse model and its effects on insulin and type 2 diabetes–related phenotypes in humans.

Our data show that the absence of functional TLR3 protects against metabolic disturbances due to fat intake. TLR3-deficient mice have enhanced glycemic control on glucose challenge, which might be attributed to increased circulating insulin, caused by amplified insulin secretion from β-cells in Tlr3^−/− mice, accelerating the response to glucose. Insulin content in β-cells and islet architecture and area were similar between Tlr3^−/− and control mice.

Enhanced insulin secretion in Tlr3^−/− mice was registered during the first and second phases of insulin secretion in response to high glucose by perifusion, a phenomenon that was glucose dependent and induced by depolarization with high K⁺.

VAMP-2 is a member of the v-SNARE complex of the exocytic apparatus and mediates insulin secretion from β-cells (34). Gck (proximal event) and VAMP-2 (distal event) were upregulated in the pancreatic islets of Tlr3^−/− mice. Notably, VAMP-2 is downregulated in islets from patients with diabetes with attenuated glucose responses (35), contrary to our murine model in which increased VAMP-2 protein expression parallels the ameliorated response to glucose. Further, stimulation of TLR3 signaling by polyI:C impairs this response and decreases VAMP-2 expression, accompanied by decreased insulin secretion in mice fed the HFD and under physiological, chow-fed conditions. However, polyI:C treatment did not affect glucose response in TLR3-deficient mice, suggesting that TLR3 modulates the processes that effect increased insulin secretion in islets.

Our perifusion data demonstrate that insulin release is enhanced in the Tlr3^−/− mice in response not only to elevated glucose but also to elevated potassium levels. This kind of potassium increase functions as an artificial secretagogue that bypasses the regulation of secretion through K⁺-ATP channels, which is the normal mechanism for control of glucose signaling. Because TLR3 deficiency enhanced insulin secretion in response also to potassium, TLR3 modulation of insulin secretion may operate on distal steps in the pathway that are not directly dependent on glucose. A glucose-specific effect is not excluded, however, because glucose is known to influence insulin secretion also at steps distal to that of K⁺-ATP channels (36).

Dysfunctional lipid metabolism causes insulin resistance, reduced insulin secretion (lipotoxicity), and in rare cases, enhanced insulin secretion accompanied by hypoglycemia (37), which primed us to analyze the effect of Tlr3^−/− deficiency on islet lipid metabolism. We tested the effects of palmitate on insulin secretion to determine whether β-cells in Tlr3^−/− mice are resistant to lipotoxicity, thereby mitigating the damage of an HFD. Our results did not support this hypothesis—the inhibitory effects of palmitate were more extensive in Tlr3^−/− mice versus control islets.

Tlr3^−/− mice fed the HFD had elevated basal fasting insulin levels and HOMA-IR values, which could indicate greater insulin resistance. However, the increased insulin response to intraperitoneally administered glucose was accompanied by better glucose tolerance versus controls—a typical primary effect on insulin secretion, not a secondary effect in response to insulin resistance. The primary effect of TLR3 deficiency on insulin secretion was confirmed by the hyperglycemic clamp technique, which demonstrated enhanced insulin secretion at a fixed level of glycemia, a conclusion that was supported by the islet experiments. Notably, the in vitro islet perifusion experiments were performed after identical culture times for control and Tlr3^−/− islets, minimizing residual secondary effects of the in vivo environment. Collectively, the ipITT results indicate that the Tlr3^−/− mice do not become whole-body insulin resistant. The only indication was the HOMA-IR value, a crude measure of resistance, the validity of which has been debated (38). To rule out tissue-specific insulin resistance, we assessed hepatic insulin sensitivity by analyzing insulin signaling and gluconeogenesis. Insulin-stimulated AKT activation was similar between the genotypes, indicating that these effects are unlikely to be caused by impaired hepatic insulin sensitivity. Furthermore, we observed a trend toward reduced glucose output from the liver of TLR3-deficient mice. This is most likely facilitated by the increased insulin, leading to suppression of PEPCK mRNA expression in the liver of TLR3-deficient mice.

A further dissection of tissue-specific insulin resistance (e.g., in muscle and adipose tissue) might be advisable in future studies to rule out a local effect of insulin in other peripheral tissues. Nevertheless, the systemic effects of increased insulin under several conditions (physiological chow-fed, HFD-fed, freely HFD-fed, and aged HFD-fed mice), and the results obtained from the hyperglycemic clamp in addition to our data on hepatic insulin sensitivity and gluconeogenesis make it unlikely that the phenotype of Tlr3^−/− mice is caused by insulin resistance. Instead, the data suggest that Tlr3^−/− mice have a robust metabolic system with adaptive measures that enables them to cope with the increased insulin secretion over a prolonged time.

Lower circulating TG, cholesterol, and FFA concentrations indicated that lipid metabolism in Tlr3^−/− mice was affected. The decline in TG content occurred in the VLDL fraction and was due to reduced VLDL secretion, as revealed by inhibiting lipoprotein lipase-dependent lipoprotein catabolism. In line with this, polyI:C decreased TG levels. Tlr3^−/− mice had higher mRNA levels for sortilin-1, which regulates VLDL secretion; this may account for at least some of the effects on VLDL (39,40).

Consistent with this model, increased plasma insulin concentrations inhibit hepatic VLDL-TG production (39), and insulin-resistant obese patients experience greater basal VLDL secretion (41). The lower circulating TG levels in Tlr3^−/− mice were accompanied by decreased FFAs, which are associated with reduced insulin sensitivity and insulin-stimulated glucose uptake (42). In addition, prolonged elevation of FFAs impairs β-cell secretion.

TLR3 stimulation with a viral mimic accelerates the development of type 1 diabetes in rats (43,44) and induces β-cell apoptosis through TLR3 (14) and Fas-associated protein with death domain recruitment, leading to activation of caspase-8. A dose-dependent effect governs the outcome of TLR3 stimulation—higher doses of exogenous TLR3 ligand induce insulitis and diabetes, and lower doses prevent autoimmune diabetes (15,17,45). Thus, TLR3 likely mediates viral recognition and protects against virally induced type 1 diabetes. The endogenous TLR3 ligand that mediates the effects in our study, in the absence of viral infection or exogenous stimulation, is unknown. TLR3 is activated by mRNA from damaged cells, but we could not detect islet cell death. Various studies have suggested that inflammation is a driving force in the development of metabolic disturbances. The surprising finding that systemic inflammation was increased in the TLR3-deficient mice is in line with findings from a study on atherosclerosis where infiltration of macrophages in atherosclerotic lesions of Tlr3^−/− mice was observed (20). However, we could not detect any signs of local inflammation in the islets.

Notably, pancreatic tissue contains and secretes copious RNases; thus, a control system that involves TLR3, an innate RNA receptor, to detect and respond to abnormalities in this system is logical. Patients with diabetes have increased concentrations of circulating nucleic acids, likely due to inhibition of plasma RNase activity (46). Circulating nucleases might prevent ligand-induced stimulation of inflammatory responses; thus, we speculate that this impairment in patients with diabetes generates endogenous ligands of TLR3.

Our finding that TLR3 mediates glucose tolerance is supported by Wu et al. (47), who demonstrated that TLR3 deficiency in mice improves glucose control. In our study, this improvement was mediated by elevated insulin secretion from Tlr3^−/− islets, at least partly due to increased Gck and VAMP-2 expression in pancreatic β-cells.

Last, we examined whether a well-characterized polymorphism in human TLR3 (rs3775291) was associated with glucometabolic alterations, supporting our findings in mice. In a meta-analysis of two healthy male populations, rs3775291 was associated with increased fasting insulin levels but not fasting glucose. rs3775291 is a nonsynonymous SNP in the coding region of TLR3 that is predicted to impair protein function (30). How rs3775291 affects TLR3 function is unknown, but a recent report indicates that it reduces TLR3 ligand binding and impairs TLR3-mediated cell activation (48). Our results show that its relationship with fasting insulin levels is consistent with the predicted impairment in human TLR3 and the Tlr3^−/− mouse model findings. Unfortunately, the human studies do not allow assessment of insulin kinetics; thus, we are not able to determine whether the increased insulin levels were due to increased production or reduced clearance.

Although TLRs have been shown to be involved in diabetes (17,49), the role of innate immunity in glucose homeostasis and insulin secretion is largely unknown. Interestingly, Krus et al. (50) demonstrated that the complement regulatory protein, CD59, modulates exocytosis machinery by directly interacting with syntaxin-1 and VAMP-2, thereby regulating insulin secretion. Interestingly, TLR3-deficient mice displayed increased Gck and VAMP-2 protein expression. Whether TLR3 directly interacts with exocytosis proteins or whether it modulates glucose sensing via Gck remains to be determined. TLR3 may also act distally in other metabolic tissues and initiate processes that lead to changes in the islets. This notion is supported by the observation that TLR3 increases hepatic production of VLDL and is also in line with published data indicating a role for TLR3 in liver regeneration and hepatocyte proliferation (47,51).

In summary, TLR3 deficiency affects glucose homeostasis, insulin secretion, and lipid metabolism, likely by modulating proximal (improved glycolysis via Gck) and distal events (increased VAMP-2 expression). Further, the rs3775291 polymorphism in TLR3 influences glucose homeostasis in humans.

Acknowledgments. The authors thank Richard Flavell, Yale University, New Haven, CT, and Dr. Claudia Monaco, Imperial College, London, for generously providing the mice, and Anneli Olsson, Ingrid Törnberg, Linda Haglund, Karolinska Institutet, Stockholm, Sweden, and Linda Johansson, Ryhov Hospital, Jönköping, Sweden, for technical assistance. The authors thank Jason K. Kim and Dr. Dae Young Jung, University of Massachusetts, Worcester, MA, for performing the hyperglycemic clamp. The authors thank Valerie Romer, University of Massachusetts, Worcester, MA, for coordinating hyperglycemic clamp experiments. The authors thank Anna Krook and Dr. Mari Björnholm, Karolinska Institutet, Stockholm, Sweden, and Dr. Myriam Aouadi, Karolinska Institutet, Huddinge, Sweden, for helpful discussions.

Funding. This study was supported by the Swedish Research Council, project grants 4203 and 8691, and the Center of Excellence for Research on Inflammation and Cardiovascular disease Linnaeus support (8703), the Swedish Heart-Lung Foundation, Stockholm County Council, the Nanna Svartz Foundation, the Fredrik and Ingrid Thuring Foundation, the Magnus Bergvall Foundation, Karolinska Institutet Cardiovascular Program Career Development Grant, the European Commission (AtheroRemo-201668, Athero-Flux-602222, and LSHM-CT-2007-037273), the Knut and Alice Wallenberg Foundation, the Strategic Cardiovascular and Diabetes Programs of Karolinska Institutet and Stockholm County Council, the Foundation for Strategic Research, the Stockholm County Council and the Research Foundation of Ryhov County Hospital.

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

Author Contributions. D.S. and A.M.L. wrote the manuscript, performed experiments, and analyzed data. Z.M., T.W., D.F.J.K., D.E., and S.F. performed experiments and analyzed data. R.J.S. and A.B. analyzed data. R.C., A.H., G.K.H., and A.B. contributed to discussion and critically edited the manuscript. All authors reviewed and approved the manuscript. D.S. 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.

Previous Presentation. Parts of this work were presented as a poster at the Keystone Symposia on Molecular and Cellular Biology Conference (D6: The Crossroads of Lipid Metabolism and Diabetes), Copenhagen, Denmark, 19–24 April 2015.