Liver Sinusoidal Endothelial Cells Link Hyperinsulinemia to Hepatic Insulin Resistance

We have described vascular EC-specific triple FoxO knockout (VECKO) mice (20). We used only male mice for experiments. We infused insulin (or PBS as control) for 72 h using an Alzet 1003D osmotic pump. We measured body composition by NMR (Bruker Optics), blood glucose with glucometer (One Touch Ultra; Lifescan), plasma insulin (Mercodia), C-peptide (Millipore) by ELISA, and triglyceride (Cayman chemical), cholesterol, and nonesterified fatty acids by colorimetric assays (Cholesterol E and NEFA C; Wako Pure Chemicals). We have previously described the procedure for intraperitoneal glucose (2 g ⋅ kg⁻¹), pyruvate (2 g ⋅ kg⁻¹), and insulin (0.75 units ⋅ kg⁻¹) tolerance tests (21). Mice are maintained with standard or high-fat (60% of calories from fat; Research Diets) diet for indicated periods. The Columbia University Animal Care and Utilization Committee approved all procedures.

For liver sinusoidal ECs (LSECs), we minced livers, which were then incubated for 1 h at 37°C in Dulbecco’s modified Eagle’s medium (DMEM) containing 0.2% collagenase type 1 (Worthington), DNaseI (Sigma), and 0.25% BSA. We resuspended digested tissue with DMEM containing 5% horse serum and centrifuged at 50g for 3 min. Supernatant was centrifuged at 400g for 5 min. The pellets were resuspended in 0.3 mL magnetic-activated cell sorting buffer, and CD146 microbeads (Miltenyi Biotec) were added, mixed, and incubated for 30 min at 4°C. LSEC purified by magnetic-activated cell sorting column were plated and cultured with DMEM with 5% horse serum, nonessential amino acids, 0.2 mg/mL heparin, 0.1 mg/mL endothelial mitogen (Biomedical Technologies), 10 ng/mL vascular endothelial growth factor, 10 ng/mL epidermal growth factor, 100 units/mL penicillin, and 0.1 mg/mL streptomycin. Cells were used after serum starvation for 18 h. Primary mouse hepatocytes were isolated from 8-week-old male mice and cultured with DMEM containing 0.25% BSA for 18 h before experiments as previously described (22). For coculture, we plated isolated LSEC onto cell culture inserts (BD) at a density of 1.0 × 10⁵ cells/cm² and cultured them with DMEM containing 5% horse serum for 4 h. After washing cells with PBS six times, we transferred cell culture inserts above primary hepatocytes in multiwell plates and cultured with DMEM containing 0.25% BSA for an additional 18 h.

We performed FoxO1 immunocytochemistry as previously described (23) with fluorescein isothiocyanate–conjugated anti-rabbit IgG (Jackson ImmunoResearch) as secondary antibody. We fixed liver specimens in 10% buffered formalin for histological analysis and stained using hematoxylin-eosin or nitrotyrosine.

We loaded LSEC cultures with diaminofluorescein-2 diacetate (DAF-2 DA) (0.5 μmol/L; Invitrogen) for 30 min at 37°C, washed three times with DMEM, and incubated them in the dark. We used fluorescent microscopy to score cells, followed by ImageJ analysis. We assayed nitrite levels using Greiss (Enzo).

Frozen liver tissue (100–200 mg) was precipitated in 5% trichloroacetic acid at 4°C and centrifuged at 1,000g for 10 min. The supernatant was extracted with ether saturated with water. cGMP concentration was measured by enzyme immunoassay (Cayman chemical). The amount of cGMP was normalized to tissue weight.

We treated primary mouse hepatocytes with a combination of dexamethasone (500 nmol/L) and cAMP (100 nmol/L) for 4 h and then incubated them for 6 h in glucose production buffer (glucose-free DMEM, pH 7.4, containing 30 mmol/L sodium lactate and 3 mmol/L sodium pyruvate without phenol red) with dexamethasone/cAMP in the presence or absence of insulin (10 nmol/L) and/or DETA-NONOate (2 mmol/L). At the end of the incubation, we aspirated 0.1 mL medium to measure glucose using a glucose oxidase assay (Sigma). Total protein concentration was measured to correct for cell count.

We extracted RNA using TRIzol (Invitrogen), synthesized cDNA with a High Capacity cDNA Reverse Transcription kit (Applied Biosystems), and performed quantitative RT-PCR using a GoTaq SYBR Green qPCR kit (Promega) in a Chromo4 Real-Time PCR Detection system (Bio-Rad). Primer sequences are available on request.

Tissues or cells were lysed by buffer containing 2% SDS, 50 mmol/L Tris-HCl, and 5 mmol/L EDTA. For immunoprecipitation, we lysed liver or cells in radioimmunoprecipitation assay buffer with protease/phosphatase inhibitors. The lysates were subjected to immunoprecipitation with 5 μg anti–InsR-β antibody. Immunoblotting was performed followed by enhanced chemiluminescence detection (GE Lifescience).

We performed comparisons using paired or unpaired t test with appropriate Bonferroni post hoc corrections. We present results as means ± SEM or 95% CI as appropriate. We used the customary threshold P < 0.05 to declare statistical significance.

When VECKO mice are studied in the background of Ldlr^−/− mice and fed a Western-type diet, they show the same insulin sensitivity as Ldlr^−/− controls (20). However, this might be due to the combined effects of Ldlr mutation and diet trumping the effect of the Foxo mutations. To address this question, we studied VECKO mice fed a standard diet (SD). Surprisingly, we observed impaired glucose tolerance and decreased insulin sensitivity in VECKO mice, independent of body weight and composition (Fig. 1A–E). Pyruvate tolerance tests were consistent with increased hepatic glucose output (Fig. 1F and G). VECKO mice also showed higher fed glucose, insulin, C-peptide, and cholesterol, as well as fasting cholesterol and triglycerides, than wild-type (WT) mice (Fig. 1H–N). To assess insulin sensitivity, we analyzed insulin signaling in liver, fat, and skeletal muscle after portal administration of an insulin bolus. In liver of VECKO mice, insulin-induced InsR and Irs1 phosphorylation was attenuated (Fig. 2A and Supplementary Fig. 1A), as was Akt phosphorylation on S473 and T308. Consequently, insulin-induced glycogen synthase kinase (Gsk)3β, extracellular signal–related kinase (Erk), and FoxO1/3a phosphorylation were decreased. In contrast, insulin-induced Akt and Gsk3β phosphorylation were comparable in adipose and muscle (Supplementary Fig. 1B and C).

We also analyzed changes in liver insulin signaling in response to fasting and refeeding. Phosphorylation of InsR, Pdk1, and Akt (T308) was decreased in refed VECKO mice compared with control littermates (Fig. 2B and Supplementary Fig. 1D) and was associated with increased protein and mRNA levels of glucose-6-phosphatase (Fig. 2B and C and Supplementary Fig. 1D). Refeeding-induced changes of Irs2, Pck1, Igfbp1, and Gck were also attenuated (Fig. 2D–G), and glycogen levels were reduced (Fig. 2H and I). In contrast, there were no differences in hepatic triglyceride content between fasted control and VECKO mice (data not shown). These data indicate that VECKO mice have hepatic insulin resistance. SD-fed VECKO mice did not show liver, pancreas, or kidney dysfunction by serum chemistry (data not shown).

The liver is composed of ∼80% hepatocytes and ∼20% other cell types that include LSECs, Kupffer, stellate, ductal, and smooth muscle cells. We wished to determine whether the observed decrease in insulin signaling originated in hepatocytes versus LSEC or other cell types. To this end, we isolated hepatocytes, LSECs, and other cell types using immunoabsorption with anti-CD146 antibodies. Eluted cells demonstrated typical morphology and acLDL uptake (Supplementary Fig. 2A and B), as well as a near-complete absence of FoxO1 and 3a protein and mRNA (20) (Supplementary Fig. 2C and D).

Immunoblotting of eluate and flow through from CD146 immunoabsorption fractions revealed that proteins required for insulin signaling—InsR, Irs and Akt—are enriched ∼20- and 50-fold in hepatocytes compared with LSECs and other cells, respectively (Fig. 3A). In contrast, Erk and endothelial nitric oxide (NO) synthase (eNos) are enriched ∼50-fold in LSECs compared with other cell types. Given that insulin signaling is decreased by ∼40% in VECKO mice, the impairment must largely occur in hepatocytes. In primary hepatocytes, InsR and Akt phosphorylation were comparable between control and VECKO mice, suggesting that a non–hepatocyte-autonomous mechanism induced insulin resistance in VECKO mice liver (Fig. 3B).

In primary LSEC cultures from VECKO mice, we observed a stark increase of eNos protein levels and phosphorylation at the Akt-dependent site, Ser1176 (24). In fact, unlike control LSEC—in which phosphorylated (p)Ser1176 was only detected after insulin stimulation—VECKO LSECs showed increased basal levels of pSer1176 eNos that could not be further stimulated by insulin (Fig. 3C and D and Supplementary Fig. 2E–G). These data are consistent with prior work indicating that FoxOs suppress eNos expression (20,25,26). Consistent with these findings, Nos3 mRNA levels were markedly increased, whereas—unlike aortic ECs from VECKO mice (20)—Nos2 levels were normal (Fig. 3C). These data indicate that eNos is persistently activated in LSECs from VECKO mice, irrespective of insulin stimulation, and confirm the robustness of VECKO mice as a model of constitutive endothelial insulin signaling.

We further determined concentrations of NO metabolites (nitrate/nitrite) in conditioned media of LSEC and in liver. We found that NO metabolites increased in both experimental systems and were normalized by pretreatment with the eNos inhibitor, L-NG-nitroarginine methyl ester (L-NAME) (Fig. 3E and F). Liver cGMP was increased in VECKO mice and was normalized by L-NAME treatment (Fig. 3G). Likewise, basal and insulin-stimulated NO production were enhanced in LSECs from VECKO mice and blunted by L-NAME pretreatment (Fig. 3H). These data indicate that there is increased eNos-dependent NO production in VECKO LSECs.

The findings of 1) impaired hepatic insulin signaling in spite of 2) preserved signaling in isolated primary hepatocytes and 3) increased NO production in LSECs suggest the working hypothesis that NO released by FoxO-deficient LSECs impairs insulin signaling in vivo, contributing to the pathogenesis of hepatic insulin resistance. To test this hypothesis, we first conducted reconstitution experiments in which we cocultured primary hepatocytes with LSEC from VECKO mice, using transwell chambers with gas-permeable bottoms. Coculturing impaired insulin-induced phosphorylation of InsR and Akt (Ser473) (Fig. 4A). The impairment was partially reversed by pretreatment with the eNos inhibitor, L-NAME, consistent with its being mediated by LSEC-released NO. We obtained similar data using cocultures of hepatocytes with WT LSECs (Supplementary Fig. 3A–D).

Next, we compared the effect of coculturing primary hepatocytes with LSECs derived either from WT or from VECKO mice on insulin regulation of gene expression. Addition of dexamethasone and cAMP increased expression of G6pc, Pck1, Pgc1α, and Igfbp1 to varying degrees. Coculture with WT LSECs partly prevented insulin inhibition of dexamethasone/cAMP-induced G6pc and Pck1 expression, with a more limited effect on Igfbp1 (Fig. 4B–E). We saw no effect on Pgc1α expression (Fig. 4D). Coculture with LSECs from VECKO mice had a greater inhibitory effect on the ability of insulin to regulate G6pc and Pck1 than WT LSEC (Fig. 4B and C). These data indicate that LSECs physiologically counteract insulin signaling and that FoxO ablation in LSECs—possibly acting through increased NO production—further impairs insulin action in hepatocytes.

To test the hypothesis that excessive NO production affected hepatic insulin sensitivity in vivo, we performed glucose and insulin tolerance tests in C57Bl6 mice after administering the NO donor, DETA-NONOate. DETA-NONOate increased glucose levels, caused glucose intolerance and insulin resistance (Fig. 4F and G), and increased G6pc protein and mRNA (Fig. 4H and I). This compound also increased expression of gluconeogenic genes and G6pc protein in primary hepatocytes. The effects were preempted by pretreatment with the guanylate cyclase inhibitor, ODQ (Supplementary Fig. 4A and B). The cGMP analog, 8-Br-cGMP, also increased gluconeogenic genes in primary hepatocytes, suggesting that NO induces gluconeogenic genes in a guanylate cyclase/cGMP–dependent manner. Moreover, the ability of insulin to suppress dexamethasone/cAMP induction of gluconeogenic genes (G6pc and Pck1) as well as glucose production in primary hepatocytes was significantly blunted by DETA-NONOate treatment (Fig. 4J and K). These data indicate that exogenous NO blunts the ability of insulin to suppress glucose production via regulating gluconeogenic gene expression.

How does NO produced in LSECs affect insulin signaling in hepatocytes? When NO production exceeds its degradation in the presence of highly reactive oxygen, it gives rise to peroxynitrite that can promote nitration of tyrosine residues on proteins (27). Indeed, treatment of primary hepatocytes with DETA-NONOate caused tyrosine nitration of InsR (Supplementary Fig. 5). Tyrosine nitration of InsR impairs their insulin-induced tyrosine phosphorylation and results in insulin resistance (28). Immunoblotting demonstrated a generalized increase of protein tyrosine nitration in livers of VECKO mice, whereas there was no difference in protein S-nitrosylation (Fig. 5A–C). In addition, immunoblotting with nitrotyrosine antibodies of InsR immunoprecipitates showed an increased tyrosine nitration of InsRs in VECKO liver (Fig. 5D). Immunohistochemistry of liver sections with anti-nitrotyrosine antibodies also showed increased immunoreactivity in VECKO liver (Fig. 5E). These data indicate that tyrosine nitration of InsR, possibly caused by excessive NO from LSECs, is associated with hepatic insulin resistance in VECKO mice.

The findings show that constitutive insulin signaling—as brought about by FoxO ablation—can promote NO-dependent hepatic insulin resistance. But do common forms of hyperinsulinemia result in constitutive endothelial insulin signaling? To address this question, first we examined the effect of chronic insulin treatment using LSEC cultures preincubated with insulin for 24 h. This treatment reduced InsR expression and Akt phosphorylation without impairing FoxO1/3a phosphorylation (Fig. 6A). Accordingly, chronic insulin treatment increased FoxO1 nuclear exclusion, NO production (Fig. 6B–D), eNos, and Nos3 expression (Fig. 6C) in a phosphatidylinositol 3-kinase–dependent manner (Fig. 6B–E). These data are consistent with previous findings in 3T3-L1 cells (29) and demonstrate that FoxO remains sensitive to insulin inhibition even under conditions that impair Akt phosphorylation. Next, we administered insulin to C57BL6 mice using an osmotic minipump for 72 h and compared insulin signaling in liver with PBS-infused mice. Insulin treatment elevated serum insulin levels by approximately threefold (Fig. 6F) and, similar to LSECs, it decreased Akt phosphorylation but not FoxO phosphorylation (Fig. 6G–J). These data indicate that chronic hyperinsulinemia uncouples Akt phosphorylation from FoxO inhibition and are consistent with the possibility that hyperinsulinemia leads to eNos-dependent NO production in LSECs.

High-fat diet (HFD) causes insulin resistance with hyperinsulinemia (30). Therefore, we sought further confirmation of our working model by treating HFD-fed C57BL6 mice with L-NAME. Body weight increased after 5 days on HFD and was not affected by L-NAME treatment (Fig. 7A). Since liver inducible NO synthase (iNos) levels did not change under these conditions (data not shown), we propose that L-NAME treatment worked as a bona fide eNOS inhibitor. Glucose tolerance tests showed that L-NAME treatment normalized HFD-induced glucose intolerance (Fig. 7B) and fasting hyperglycemia (Fig. 7C). Fasting insulin levels tended to improve in L-NAME–treated HFD mice (Fig. 7D). L-NAME treatment also decreased fasting serum cholesterol (Supplementary Fig. 6A). As shown in the experiments with chronic insulin administration, 5-day HFD decreased Akt phosphorylation but was associated with increased FoxO phosphorylation (Fig. 7E and Supplementary Fig. 7A–E). In addition, liver p- and total-eNOS expression increased in HFD mice (Fig. 7E and Supplementary Fig. 7). L-NAME treatment reversed these findings, increasing Akt phosphorylation and decreasing FoxO phosphorylation; as a result, L-NAME treatment also restored G6pc and Igfbp1 levels to normal (Fig. 7F). In contrast, iNOS inhibitor 1400 W did not reverse glucose intolerance in 5-day HFD mice, suggesting that eNOS, rather than iNOS, is the main contributor to the development of glucose intolerance (data not shown). It should be observed that L-NAME treatment was only effective after a short course of HFD (5 days) but not in chronically (15–18 weeks) HFD-fed control (Supplementary Fig. 8A–C) or VECKO mice (Supplementary Fig. 8D–K).

Finally, we determined whether eNos-derived NO overproduction was responsible for the development of hepatic insulin resistance in VECKO mice. To this end, we examined hepatic insulin signaling of VECKO mice after L-NAME treatment. L-NAME improved glucose tolerance and pyruvate tolerance in VECKO mice (Fig. 7G and H). It also improved fasting hyperglycemia (Fig. 7I) and hypercholesterolemia (Supplementary Fig. 6D) in VECKO mice. Fasting insulin levels were not affected by L-NAME treatment (Fig. 7J). Finally, liver protein nitration decreased and insulin signaling improved in VECKO mice treated with L-NAME (Fig. 7K and L). These data are consistent with a causative role of eNos-derived NO in the pathogenesis of hepatic insulin resistance in VECKO mice.

The current study demonstrates that constitutive insulin signaling in EC—as might be expected to ensue from hyperinsulinemia—affects hepatic insulin action in an indirect manner through NO overproduction in LSECs that leads to tyrosine nitration of InsRs. This, in turn, impairs insulin’s ability to restrain expression of glucose production genes, leading to hepatic insulin resistance. Consistent with our findings, a recent report also demonstrated that EC-specific eNOS overexpression in mice was associated with glucose intolerance, although the authors did not investigate the liver phenotype in this model (31). We use mice lacking the three FoxOs in ECs (20) to demonstrate a causative role of the eNos pathway in this process. This turn of events can be reversed by administering L-NAME, providing proof of principle that eNos inhibition in liver has therapeutic benefits in the early stages of insulin resistance. The model arising from these data is that hyperinsulinemia-induced NO overproduction plays an early role in the development of hepatic insulin resistance.

While mechanistically attractive, the hypothesis that paracrine communication between parenchymal and nonparenchymal cells can cause hepatic insulin resistance finds only sparse support in the literature. Aside from reports that prostaglandins in conditioned medium of cultured nonparenchymal cells increase glycogenolysis in rat hepatocytes (32), there is no prior evidence for a physiological role of LSECs in hepatic glucose metabolism. It also was not obvious, prior to this study, that increased insulin signaling in LSECs, as opposed to decreased insulin signaling, adversely affected hepatic insulin sensitivity.

Our findings highlight the dual nature of NO generated in response to insulin in vascular ECs. In the aorta, constitutive activation of eNos expression and Ser1176 phosphorylation is associated with improved endothelial health and prevents atherosclerosis in Western diet-fed Ldlr^−/− mice (20). Conversely, we now show that increased hepatic NO generation predisposes to insulin resistance. This tissue-specific role of eNos-derived NO is not unprecedented. Caveolin1 inhibits NO production by binding to eNos (33). As in Ldlr^−/−;VECKO mice, genetic ablation of Cav1 on the Apoe^−/− background reduces atherosclerosis and increases eNos-dependent NO in ECs (34). But in the lung, Cav1 knockout increases eNos and induces pulmonary hypertension through tyrosine nitration–dependent impairment of protein kinase G activity (35). These observations suggest that the pathophysiological relevance of persistent eNos-dependent NO production varies from tissue to tissue.

There are no precedents, however, for this type of paracrine regulation of insulin signaling by endothelial NO in the context of an otherwise insulin-sensitive mouse on a normal diet. Consistent with the notion that the LSEC-dependent mechanisms play an initiating role in the metabolic syndrome, we find that treatment of established insulin resistance, e.g., in mice fed an HFD for 18 weeks, is no longer reversible by NO inhibition, suggesting that other events (lipotoxicity, inflammation, and metabolic inflexibility) account for the progression of insulin resistance. Likewise, it is interesting to note that VECKO mice only exhibit insulin resistance on a regular diet but not on the more metabolically challenging HFD or Western-type diet on the Ldlr^−/− background (20), suggesting that other factors trump the role of LSECs in the transition from early to advanced insulin resistance. For instance, iNos produced in response to inflammatory cytokines in the obese state can generate significantly more NO than eNos (36–38).

Tyrosine nitration is a biomarker of nitrosative stress in pathological conditions, including insulin resistance (39–41). In insulin signaling, it has been reported that tyrosine residues of InsR, Irs1 and -2, and Akt can be nitrated (28,42), altering their function by preventing phosphorylation (43,44). In contrast, we show that insulin sensitivity in adipose tissue and skeletal muscle was unaffected, possibly due to differences in tissue content of ECs and eNOS-derived NO or tissue sensitivity to NO. We suggest that the eNos-dependent NO overproduction promotes protein nitration, impairing hepatic insulin signaling.

We focused our studies on the role of eNos-derived NO in insulin InsR signaling. But there are likely to be additional mechanisms that mediate endothelial/hepatocyte interactions, potentially affecting insulin sensitivity. In line with our observations, previous studies have suggested that NO can inhibit glycogen synthesis and glucose uptake in liver, possibly independent of insulin signaling. Thus, the NO donor, S-nitroso-N-acetylpenicillamine, can impair glycogen synthase, decreasing glycogen synthesis (45). NO infusion increases glucose output in the perfused rat liver by stimulating glycogenolysis (46,47). Consistent with our results, portal delivery of NO enhances hepatic glucose output (47,48)—an effect that is blocked by inhibition of guanylate cyclase (48). Our findings suggest that local—as opposed to systemic—NO production can explain these effects.

Based on the present data, we propose the following model for the progression of insulin resistance: In response to a decrease in insulin-dependent glucose uptake in skeletal muscle, insulin levels increase, resulting in hyperinsulinemia. Sustained hyperinsulinemia inactivates insulin signaling in LSECs, causes FoxO nuclear exclusion, increases NO production in the liver, and impairs insulin’s ability to suppress glucose production. Hence, we propose that lowering hyperinsulinemia should be a focus of treatment in the early phases of the metabolic syndrome. NO inhibition in liver could have beneficial effects in preventing insulin resistance.

K.T. is the recipient of a postdoctoral fellowship for research abroad from the Japan Society for the Promotion of Science. This work was supported by National Institutes of Health grants HL87123, DK58282, and DK63608 (Columbia University Diabetes Research Center).

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

K.T. and D.A. designed and performed experiments, analyzed data, and wrote the manuscript. D.A. is the guarantor of this work and, as such, had full access to all the data in the study and, as such, takes responsibility for the integrity of the data and the accuracy of the data analysis.

The authors thank members of the Accili laboratory for constructive advice and discussion of data.