Protection Against Insulin Resistance by Apolipoprotein M/Sphingosine-1-Phosphate

Insulin resistance is one of the main pathogenetic mechanisms of type 2 diabetes. Subjects with insulin resistance often exhibit low serum levels of HDL cholesterol, which could explain, in part, the proatherosclerotic effect of insulin resistance (1). In addition, subjects with low serum HDL cholesterol levels are also reportedly susceptible to the development of diabetes (2–4). Although the anti-inflammatory actions (5) or roles in the reverse cholesterol transfer system (6) of HDL might partly explain these epidemiological associations, the precise mechanisms have yet to be elucidated.

Sphingosine-1-phosphate (S1P) is carried on apolipoprotein M (apoM) (7), a minor apolipoprotein on HDL (8), and has been proposed as being responsible for many of the pleiotropic effects of HDL, such as antiapoptotic (9), anti-inflammatory (10), and vasoprotective effects (11). In regard to the clinical association between apoM and diabetes, single nucleotide polymorphisms of apoM have been shown to be associated with type 1 and type 2 diabetes (12,13), and serum apoM levels have been demonstrated to be lower in subjects with MODY3 diabetes (14). As for the mechanisms underlying the association between apoM/S1P and diabetes, we have reported that apoM promotes insulin secretion through activation of S1P receptor 1 (S1P1) and/or S1P receptor 3 (S1P3) signaling (15). One recent study reported that apoM suppresses the activity of brown adipose tissue and that deficiency of apoM protects mice against diet-induced obesity and, to some extent, against glucose intolerance (16). However, the association between apoM/S1P and insulin resistance has not yet been clearly elucidated, especially in terms of the disturbed glucose metabolism in the liver and skeletal muscles, which is involved in the pathogenesis of insulin resistance, as well as in the adipose tissue.

In the current study, considering these background data, we hypothesized that apoM/S1P might explain the protective effects of HDL against the development of diabetes and investigated the roles of apoM/S1P in protecting against glucose intolerance, especially insulin resistance.

The commercially available materials and reagents used in the current study are listed in Supplementary Tables 1–4 and were used according to the manufacturers’ protocols.

We collected serum samples from subjects who had fasted overnight prior to undergoing a medical examination at Aizu Fujitsu Semiconductor Ltd. (n = 179) and from outpatients with type 2 diabetes who had fasted overnight (n = 46) at the Aizu Medical Center, Fukushima Medical University, after applying the following exclusion criteria: persons receiving insulin therapy, persons receiving antidyslipidemia reagents, and persons with severe renal complications (albuminuria >300 mg/g creatinine or estimated glomerular filtration rate <30 mL/min/1.73 m²) (17). The clinical characteristics of the subjects are shown in Table 1. The current study was conducted with the approval of the Institutional Research Ethics Committee of Fukushima Medical University and Aizu Fujitsu Semiconductor Ltd., and written informed consent was obtained from each of the subjects (2041, UMIN Clinical Trials Registry number UMIN000015010).

For investigating the modulation of apoM and S1P by insulin resistance, wild-type (WT) C57BL6 mice obtained from CLEA Japan, Inc. (Tokyo, Japan) were fed chow, a 60% high-fat diet (HFD) (HFD-60; Oriental Yeast Co. Ltd., Tokyo, Japan), a choline methionine–deficient diet (F2MCD; Oriental Yeast Co. Ltd.), or a 60% HFD supplemented with 1% cholesterol (n = 6) for 12 weeks from 8 weeks old.

For the experiments with apoM knockout (KO) mice, apoM-KO mice were generated using the CRISPR-Cas9 system, as described previously (18). ApoM-KO mice and WT mice, their littermates, were fed chow or an HFD from 8 weeks old for 12 weeks or 18 months.

For investigating the effects of apoM overexpression, 20-week-old WT mice reared on an HFD for 12 weeks were injected with an adenovirus vector encoding human apoM (huApoM mice) or a control blank adenovirus vector (WT mice) via the tail vein, at a dose of 2.5 × 10⁸ plaque-forming units/g body weight (BW) (19). To investigate the effects of VPC23019 (VPC), an antagonist against S1P1/3, or JTE013 (JTE), an antagonist against S1P2, the mice were intraperitoneally injected with vehicle alone, VPC (0.75 mg/kg BW), or JTE (0.75 mg/kg BW) twice a day from the second day after administration of the viral vector until 1 h before the mice were sacrificed. We used VPC and JTE, as they are commonly used antagonists for the S1P receptors.

We performed a glucose tolerance test and an insulin tolerance test (ITT) as follows: the mice were denied access to food for 6 h and challenged intraperitoneally with glucose (2 g/kg BW) or Humulin R (0.5 units/kg or 0.375 units/kg BW).

All of the animal experiments were conducted in accordance with the guidelines for Animal Care and approved by the animal committee of The University of Tokyo (P11–074 and P16–044).

The plasma S1P levels were measured by a previously validated high-performance liquid chromatography method, as previously described (20). Hepatic lipid was extracted with methanol and chloroform, and the cholesterol and triglyceride levels were measured by enzymatic methods and adjusted with the hepatic protein levels. Blood glucose was measured using a Glutest sensor. Metabolome analyses were performed with a gas chromatography–mass spectrometry system (QP2020; Shimadzu Corporation) as described previously (21). The ELISA kits used in the current study are listed in Supplementary Table 1. The human apoM levels were measured by an ELISA method developed and validated by us, as described previously (15).

HepG2 cells (ATCC, Manassas, VA), 3T3L1 fibroblasts (Japanese Collection of Research Bioresources Cell Bank, Osaka, Japan), and C2C12 myocytes (European Collection of Authenticated Cell Cultures, Salisbury, U.K.) were cultured in DMEM supplemented with 10% FBS and 1% penicillin/streptomycin. The 3T3L1 fibroblasts and C2C12 cells were used after induction of adipogenic and myogenic differentiation with 0.5 mmol/L 3-isobutyl-1-methylxanthine, 1 μmol/L dexamethasone, 10 μg/mL insulin, and 2% horse serum, respectively.

To investigate the effects of overexpression or knockdown of apoM, HepG2 cells were infected with an adenovirus vector encoding apoM or GFP at a multiplicity of infection of 25 or treated with siRNA against apoM or control siRNA using Lipofectamine RNAi MAX. After 48 h, the medium was replaced with FBS-free medium, and then 24 h later, the cellular contents were collected and subjected to the analyses.

We performed the experiments with conditioned media of apoM-overexpressing HepG2 cells (CM-apoM), GFP-overexpressing HepG2 cells (CM-GFP), or HepG2 cells infected with blank adenovirus (CM-Null) as previously described (22). In several experiments, we treated HepG2 cells with siRNA against SIRT1 or control siRNA, siRNA against S1P receptor 1 (S1P1), S1P receptor 2 (S1P2), or S1P receptor 3 (S1P3), or control siRNA. After 48 h, the medium was replaced by serum-free medium containing CM-apoM, CM-GFP, or CM-Null. For the experiments conducted using the 3T3L1 adipocytes and C2C12 myocytes, after the induction of cell differentiation, the media were replaced with FBS-free medium containing CM-apoM, CM-GFP, or CM-Null with or without VPC or JTE. The experiments were performed after a further 24 h of incubation.

To investigate the degradation of SIRT1, the cells were treated with cycloheximide at 100 μg/mL in FBS-free medium. Then the proteins were collected at 0 h, 8 h, and 24 h to determine the SIRT1 levels by Western blot analysis.

Western blotting was performed using 2–10 μg of the tissue or cellular proteins or a volume corresponding to 0.01–0.02 μL of plasma according to a standard method. The antibodies used in the current study were anti-human apoM antiserum, which was developed previously (19), and commercially available antibodies listed in Supplementary Table 2. The intensities of the bands were measured using ImageJ (from the National Institutes of Health), and bands from different gels were compared after normalization using the same sample run on both gels as the internal standard.

The mitochondrial membrane potentials of the cells in the 96-well plate were assessed using the JC-1 MitoMP Detection Kit, and the oxygen consumption rates of the cells in the 96-well plate or the murine liver at the protein concentration of 8 mg/mL, isolated with the MitoCheck Mitochondrial (Tissue) Isolation Kit from WT, apoM-KO mice, or huApoM mice, were evaluated by the MitoXpress Xtra Oxygen Consumption Assay (23). For these experiments, we used CM-Null instead of CM-GFP to avoid the interference with GFP.

The total RNA extracted from the murine tissues or cells using the GenElute Mammalian Total RNA Miniprep kit was subjected to reverse transcription with ReverTra Ace qPCR RT Master Mix. Quantitative PCR was performed using an ABI 7300 Real-Time PCR System (Applied Biosystems), the primers for the S1P receptor constructed in a previous article (24), and commercially available primers listed in Supplementary Table 3.

Insulin staining was performed using the pancreas sections according to a standard method with anti-insulin rabbit antibody, biotin-conjugated anti-rabbit Ig, peroxidase-conjugated streptavidin, and diaminobenzidine.

All of the data were statistically analyzed using SPSS (SPSS Inc., Chicago, IL). Some results are expressed as the means ± SD. In the clinical studies, comparison between any two groups was performed by the Mann-Whitney U test and correlations between two parameters were determined by the Spearman rank correlation test. For basic studies, differences between two groups were evaluated by Student t test, and differences among more than two groups were assessed by one-way ANOVA, followed by post hoc analysis using the Tukey-Kramer test. The fasting and random blood glucose levels in the WT, apoM-KO, and huApoM mice were evaluated by two-way ANOVA. A value of P < 0.05 was regarded as denoting statistical significance in all of the analyses.

All relevant data are within the article, and the data sets generated or analyzed during the current study are available on reasonable request.

First, we investigated the correlations between the serum apoM levels and diabetes-related parameters. The serum apoM level was lower in the subject group with type 2 diabetes, whereas no significant difference was found in the serum apoA-I level (Fig. 1A and Supplementary Fig. 1A). The apoM/apoA-I ratio was significantly lower in the group with diabetes (Fig. 1B). In regard to the correlations with diabetes-related parameters, the serum apoM level showed negative correlations with BMI, HOMA-IR, HbA_1c, and HOMA of β-cell function and a weak positive correlation with the serum adiponectin level, while no significant correlations were observed with the leptin levels (Fig. 1C and D and Supplementary Fig. 1B–E). These correlations with diabetes-related parameters were also observed for apoA-I (Supplementary Fig. 1F–I). Moreover, the serum apoM levels showed a moderate correlation with the serum apoA-I levels and weak correlation with the serum apoB levels (Fig. 1E and F).

We next investigated the modulation of the serum apoM levels in mice with diet-induced obesity. Plasma apoM levels, determined by an ELISA or Western blot analysis, were higher in the mice reared on HFD or an HFD supplemented with cholesterol (Fig. 1G and Supplementary Fig. 3A). The plasma S1P levels were also higher in the mice with diet-induced obesity (Fig. 1H). Hepatic levels of LDL receptor (LDLr) were lower and the hepatic expression of apoM was higher in the mice with diet-induced obesity (Fig. 1I and J and Supplementary Fig. 3B). In the mice with diet-induced obesity, the serum leptin and insulin levels, which reportedly regulate the serum apoM levels, were modulated in a way similar to that in obese subjects (Supplementary Figs. 2 and 3C–F); however, the serum apoA-I level was negatively correlated with the BMI in human subjects, while they were elevated in the mice with diet-induced obesity (Supplementary Fig. 1F and G).

To elucidate the roles of apoM in the pathogenesis of diabetes, especially in the development of insulin resistance, we investigated the modulation of glucose metabolic parameters in apoM-KO mice reared on an HFD (Fig. 2) or chow (Supplementary Fig. 5) for 12 weeks. Plasma S1P levels were lower in the apoM-KO mice (Fig. 2A). Investigation of the time course of the parameters in the apoM-KO mice reared on an HFD revealed that random blood glucose levels were higher in 18- and 20-week-old apoM-KO mice (Fig. 2B and D). While there were no significant differences in the BW/organ weights or food intake, the ratio of the liver weight to the BW was lower and that of the epididymal (EP) fat weight to the BW was higher in the apoM-KO mice (Fig. 2B, E, and F and Supplementary Fig. 4). The hepatic triglyceride content was also lower in the apoM-KO mice (Supplementary Fig. 4C and D). Although the fasting plasma insulin levels were not significantly different, the apoM-KO mice showed higher blood glucose levels in ITT, suggesting that the apoM-KO mice were less tolerant to HFD-induced obesity (Fig. 2G and H). In addition to the progression of insulin resistance, we also observed that the induction of insulin secretion in the intraperitoneal glucose tolerance test was attenuated in the apoM-KO mice (Fig. 2I and J), although no obvious modulation of the pancreatic β-cells by apoM was observed (Supplementary Fig. 6).

Consistent with the finding that insulin resistance was augmented in the apoM-KO mice, phosphorylation of AKT in the liver and skeletal muscle, as also that of AMPK, which is regulated by insulin signaling, was attenuated in the apoM-KO mice (Fig. 3A–D). Moreover, the expression levels of Pepck, G6pc, Glut2, and glycogen synthase 2 (Glysn), which are involved in hepatic glucose metabolism and are negatively regulated by insulin resistance, were higher in the liver of the apoM-KO mice (Fig. 3E). In contrast, the hepatic expression level of Ucp2, which is involved in the metabolism of fatty acids in the mitochondria (25), tended to be lower in the apoM-KO mice. Concerning the expression levels of the proteins involved in glucose metabolism in the adipose tissue, the expression level of Ucp2 was significantly lower in the EP fat of the apoM-KO mice, although there were no changes in the expression levels of Glut4 or adiponectin (Fig. 3F).

In the apoM-KO mice reared on chow, decrease in insulin secretion during a glucose tolerance test, slight attenuation of the insulin sensitivity during an ITT, and phosphorylation of AKT in the liver, as compared with the observations in the WT mice, were observed (Supplementary Fig. 5).

We next investigated whether overexpression of apoM might modulate insulin resistance in the mice with diet-induced obesity. When we overexpressed apoM using an adenovirus gene transfer (Fig. 4A), the plasma S1P levels increased, and the blood glucose levels decreased (Fig. 4B and C). Both the results of the ITT and the fasting plasma insulin levels suggested that the insulin resistance improved in huApoM mice (Fig. 4D and E). Following treatment of the huApoM mice with VPC or JTE (S1P1/3 and S1P2 antagonists, respectively), while the fasting plasma insulin levels remained unchanged (Supplementary Fig. 7A), treatment with VPC, but not JTE, partially reversed the beneficial effects of apoM on the fasting blood glucose levels and results of the ITT (Fig. 4F and G). Consistent with these results, phosphorylation of AKT and AMPK in the liver and that of AKT in the skeletal muscle were enhanced in the huApoM mice, while treatment with VPC partially inhibited these protective effects of apoM overexpression against insulin resistance (Fig. 4H–K) and also partially decreased the suppression by apoM overexpression of the hepatic proteins involved in glucose metabolism (Fig. 4L). The increased expression of Ucp2 in the liver and EP fat by apoM overexpression was also partially reversed by treatment with VPC (Fig. 4M and N).

In addition to modulating the activity of the insulin signaling pathways, including AKT and AMPK, apoM also positively influenced the expression of Ucp2. Considering that UCP2 is involved in the functioning of the mitochondria (25), we next investigated the modulation of the mitochondria functions by apoM.

Although there were no significant changes in the shape of the hepatic mitochondria on electron-microscopic observation (Supplementary Fig. 8), the hepatic protein levels of SIRT1, PGC1a, and TFAM, which have been established as being positively associated with the mitochondrial functions (26,27), the mRNA levels of two cytochrome b mitochondrial DNA subunits, Cyba and Cybb (28,29), and of Pgc1a were significantly downregulated in the livers of the apoM-KO mice (Fig. 5A–D). The mRNA levels of Pgc1a, Cyba, and Cybb were also lower in the EP fat (Fig. 5E), and the protein levels of PGC1a were lower in the skeletal muscle of the apoM-KO mice (Fig. 5F and G). The oxygen consumption rates of the mitochondria isolated from the livers of the apoM-KO mice were lower than those of the mitochondria isolated from the livers of the WT mice (Fig. 5H). Concordantly, metabolome analyses revealed that the levels of several metabolites of the tricarboxylic acid (TCA) cycle, such as citrate, were lower in the liver and EP fat of the apoM-KO mice (Fig. 5I and J).

Furthermore, we investigated the modulation of SIRT1, PGC1a, TFAM, Cyba, and Cybb in the huApoM mice treated with or without VPC/JTE. As shown in Fig. 6, we found that apoM overexpression was associated with increases of the hepatic protein levels of SIRT1, PGC1a, and TFAM and of the mRNA levels of Pgc1a, Tfam, Cyba, and Cybb in the liver (Fig. 6A–D), the mRNA levels of Pgc1a, Cyba, Cybb, and Tfam in the EP fat (Fig. 6E), and the protein levels of PGC1a in the skeletal muscle tissues (Fig. 6F and G). Treatment with VPC inhibited the modulations of the aforementioned by apoM overexpression. The oxygen consumption rates of the mitochondria isolated from the livers of the huApoM mice were also higher than those of the mitochondria isolated from the livers of the WT mice (Fig. 6H).

Because SIRT1 reportedly regulates mitochondrial functions and is also involved in age-induced metabolic dysfunction (30), we investigated whether KO of apoM might deteriorate glucose metabolism in 18-month-old senile mice reared on chow. Contrary to the findings in the 4-month-old mice (Supplementary Figs. 5 and 9), deletion of apoM was associated with overt deterioration of insulin resistance (Supplementary Fig. 10) and downregulation of SIRT1 and other genes involved in mitochondrial functions in the liver, EP fat, and skeletal muscle tissues (Supplementary Fig. 11). These results suggest that the modulation of SIRT1 by apoM might have an important role in the association between apoM and insulin resistance, which could deteriorate with aging.

Lastly, we investigated the mechanisms of regulation of SIRT1 by apoM in vitro. We observed similar modulation, as described above, of SIRT1 by apoM in in vitro experiments performed using HepG2 cells (Fig. 7A, B, and D). Furthermore, we treated HepG2 cells with CM-apoM, CM-GFP, or CM-Null. The S1P and apoM levels in CM-apoM and CM-GFP were 121.6 ± 1.33 nmol/L and 0.04 ± 0.005 nmol/L and 10.82 ± 0.20 mg/L and 0.58 ± 0.01 mg/L (n = 3), respectively, and the protein levels of SIRT1 were increased in the cells treated with CM-apoM (Fig. 7C), suggesting the possibility that S1P bound to apoM might modulate SIRT1 expression from outside the cells. When we suppressed the expression of SIRT1 with siRNA and treated the cells with CM-apoM, the enhanced expressions of Pgc1a, Cyba, and Tfam by CM-apoM were suppressed (Fig. 7E and F), suggesting that CM-apoM enhanced the mitochondrial functions, at least in part, by modulating the SIRT1 protein levels. When we investigated the mitochondrial membrane potentials and oxygen consumption rates in the HepG2 cells, we found that overexpression of apoM was associated with an increase in both of these parameters, while knockdown of apoM was associated with a decrease in both of these parameters (Fig. 7G–J).

We further investigated which S1P receptors might be involved in the upregulation of SIRT1 by apoM using siRNAs against the S1P receptors. The efficacies of knockdown of the S1P receptors are shown in Supplementary Fig. 12. Treatment with CM-apoM did not increase the SIRT1 protein levels in the HepG2 cells with S1P1 knockdown (Fig. 8A and B). Consistent with this finding, the induction of Pgc1a, Cyba, and Tfam, the augmented mitochondrial potentials, and enhanced oxygen consumption rates by CM-apoM were inhibited in the HepG2 cells with SIP1 knockdown (Fig. 8C and D and Supplementary Fig. 13). These results suggest that the increase in SIRT1 protein expression levels induced by apoM might be mediated by S1P1 signaling.

When we investigated the modulation of SIRT1 in the 3T3L1 adipocytes and C2C12 myocytes, we observed that CM-apoM increased the SIRT1 protein levels, expression levels of Pgc1a and Tfam, the mitochondrial potentials, and the oxygen consumption rates, all of which were reversed by treatment with VPC, but not by treatment with JTE (Supplementary Figs. 14–16).

We also investigated how the SIRT1 protein levels were increased by apoM. As shown in Figs. 5D and E and 6C and Supplementary Fig. 11C and D, the mRNA levels of Sirt1 were not positively regulated by apoM, suggesting that regulation at the transcriptional level might not be involved in the positive association between apoM and SIRT1 protein expression. SIRT1 protein expression is reportedly regulated by both degradation and transcription. As shown in Fig. 8E–H, the degradation of SIRT1 was delayed in the apoM-overexpressing HepG2 cells, while it was accelerated in the HepG2 cells with apoM knockdown. These results suggest that apoM might increase the protein levels of SIRT1 by delaying the degradation of SIRT1.

In this study, we demonstrated that apoM/S1P protects against the development of insulin resistance in the liver, adipose tissue, and skeletal muscle by activating AKT and AMPK signaling, which are the main pathways of insulin signaling, through S1P1 and/or S1P3 and by enhancing mitochondrial functions, perhaps through upregulation of the SIRT1 protein levels (Supplementary Fig. 17). This is the first study to demonstrate the potential protective effect of apoM/S1P against the development of insulin resistance through the aforementioned mechanisms.

In regard to the activation of AKT and AMPK signaling, it is reasonable to assume that the S1P carried by apoM might activate these pathways, because these pathways have been established with near certainty to be regulated by S1P1 and S1P3 (31). Contrary to these results, several previous studies have reported that S1P might promote insulin resistance in the hepatocytes, pancreatic β-cells, and adipose tissue (32–34). However, in these studies, the S1P levels were modulated by delivering S1P using albumin as the vehicle or by modulating the level of an enzyme involved in the production of S1P. Recently, several articles have demonstrated that S1P bound to apoM/HDL might possess different properties from S1P bound to albumin; S1P bound to apoM/HDL might be a biased agonist for S1P1 and/or S1P3 (10,15,35–37). Considering that several previous studies have reported that insulin resistance is induced by the activation of S1P2 (32), the results of the current study were not contrary to these reports. Actually, the activation of AKT and AMPK signaling was inhibited by treatment with VPC, an antagonist of S1P1 and S1P3, but not by JTE, an antagonist of S1P2, although the inhibitory effect of VPC was not complete (Fig. 4H–N). Because S1P has been proposed to have several physiological roles inside the cells, we cannot deny the possibility that the protective effect of apoM against the development of insulin resistance might also be attributed to intracellular S1P and/or apoM.

Another possible mechanism underlying the improvement of insulin resistance by apoM is that apoM/S1P affects the amount and functions of mitochondria in the liver and adipose tissue. In this study, apoM positively influenced Ucp2, the mRNA expression levels of mitochondrial proteins (Figs. 4M and N and 5D and E), the metabolites of the TCA cycle (Fig. 5I and J), and the oxygen consumption rate of isolated mitochondria (Figs. 5H and 6H). The results of the current study suggest that these effects of apoM might be exerted, at least in part, via maintenance of the SIRT1 protein levels by apoM. As for the association between apoM/S1P and mitochondrial functions, a recent elegant study reported that apoM/S1P suppresses the activity of brown adipose tissue and deletion of apoM protects against diet-induced obesity in mice (16). The results in this aforementioned article are partly contrary to the results of the current study. Especially, the weight of apoM-KO mice used in the previous report was much lower than that of the WT mice, while in the current study, there was no obvious difference in the weight between the apoM-KO mice and the WT mice. Although the reasons for these differences remain unclear, we generated the apoM-KO mice using the CRISPR-Cas9 system (18), while Christoffersen et al. (38) produced them by inserting a neomycin resistance-encoding cassette in the APOM locus. In regard to the association between apoM/S1P and SIRT1, ApoM and S1P are reportedly positively regulated by SIRT1 activity (39), which, together with the results of the current study, suggests that SIRT1 and apoM/S1P form a positive-feedback cycle. Considering the beneficial effects of apoM/S1P on human pathological conditions (18,37,40–42), this pathway might contribute, at least in part, to the longevity properties of SIRT1. In this study, we also proposed a potential mechanism of the regulation of SIRT1 by apoM/S1P; apoM/S1P delayed the degradation of SIRT1, perhaps through S1P1. Several previous researchers have proposed modulation of the rate of degradation as the major regulatory mechanism of SIRT1 (43). Further studies are necessary to investigate the roles and mechanisms of this regulatory pathway in the pathogenesis of diabetes as well as in the pathogenesis of other disorders than those of glucose metabolism, because SIRT1 has been reported to be involved in the pathogenesis of many human diseases.

In the clinical study, we observed that the serum apoM levels were significantly negatively correlated with the insulin resistance (Fig. 1B and C and Supplementary Fig. 1). Previous reports have demonstrated that plasma apoM levels are lower in subjects with obesity (44,45), which are consistent with the findings of the current study, although some studies have reported that the association between apoM and insulin resistance might differ according to the country of birth (46), and others entirely denied the existence of any association with insulin levels (44). In contrast to the negative association between apoM and obesity observed in the human study, we observed that plasma apoM levels were higher in the mice with diet-induced obesity. This discrepancy could be explained, at least in part, by the different modulation of HDL by obesity between mice and humans; in human subjects, the serum apoA-I levels were negatively correlated with the BMI, while in mice with diet-induced obesity, the serum apoA-I levels were elevated, perhaps because HDL is the main lipoprotein in mice, which do not possess CETP. In mice with diet-induced obesity, decreased hepatic levels of the LDLr, which is involved in the clearance of apoM-containing lipoproteins (22), and/or increased expression of apoM in the liver (Fig. 1I and J and Supplementary Fig. 3B) could also contribute to the increased plasma levels of apoM. Considering that apoM has a protective effect against insulin resistance, modulation of apoM by obesity might be one of the causes of insulin resistance in human subjects, while the modulation of apoM may serve a compensatory role in mice.

In conclusion, apoM/S1P has a protective effect against the development of insulin resistance through activating insulin signaling and enhancing the mitochondrial functions, in addition to its role in promoting insulin secretion through S1P1 and/or S1P3, as reported by us previously (15,47). These results suggest novel molecular mechanisms underlying the association between HDL and diabetes.

Funding. This work was supported by Japan Society for the Promotion of Science KAKENHI grant 16H06236 (to M.K.), the Research Fund of the Mitsukoshi Health and Welfare Foundation 2017 (to M.K.), Japan Heart Foundation and Astellas Grant for Research on Atherosclerosis Update (to M.K.), the MSD Life Science Foundation, Public Interest Incorporated Foundation (to M.K.), the Japan Agency for Medical Research and Development Leading Advanced Projects for Medical Innovation (LEAP) program, the Japan Agency for Medical Research and Development Practical Research for Innovative Cancer Control program (to Y.Y.), and Japan Society for the Promotion of Science Grant-in-Aid for Scientific Research on Innovative Areas 15H05906 (to Y.Y.).

Duality of Interest. T.S. is an employee of Sekisui Medical Co., Ltd. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. M.K., K.T., and Y.Y. designed research. M.K. and M.H. analyzed data. M.K., T.S., H.K., K.N., A.A., and M.H performed research. M.K., K.T., M.H., and Y.Y. wrote the article. All authors read and approved the final manuscript. M.K. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.