Mutations of NRG4 Contribute to the Pathogenesis of Nonalcoholic Fatty Liver Disease and Related Metabolic Disorders

Nonalcoholic fatty liver disease (NAFLD) is the most common chronic liver disease, affecting 25.24% of the population worldwide (1). Growing evidence shows that NAFLD is a multisystem disease strongly associated with obesity, type 2 diabetes, and insulin resistance (2). In the pathogenesis of NAFLD, adipose tissue inflammation exacerbates hepatic steatosis, which consequently activates innate inflammatory responses in the liver (3,4). Adipose tissue is a highly active endocrine organ that secretes hormones and cytokines, such as leptin and adiponectin, to regulate metabolic homeostasis through a comprehensive communication network with other organs, such as the liver (5–7). Several endocrine factors, such as fibroblast growth factor 21, interleukin-6, fatty acid–binding protein 4, and neuregulin 4 (Nrg4), have been linked to adipose tissue and the liver (6,8) and play crucial roles in the physiological function of the liver.

Nrg4 is a novel adipokine and a member of the epidermal growth factor (EGF) family, with the highest expression levels in brown adipose tissue (9). Nrg4 contains a highly conserved N-terminal EGF-like (EGFL) domain that is released after proteolytic cleavage at Ser53 (9,10). Heterotopic expression and binding assays indicate that Nrg4 signals via ErbB3 and ErbB4, which belong to the ErbB/HER family of protein-tyrosine kinase receptors to regulate several biological processes (9,11,12). Nrg4 knockout mice showed weight gain and liver fat accumulation accompanied by exacerbated glucose intolerance and insulin resistance upon high-fat diet (HFD) feeding (9). On the contrary, Nrg4 transgenic mice showed higher basal metabolic rates and energy expenditure (13). Nrg4-deficient livers also showed significantly higher expression of several genes involved in de novo lipogenesis, which may account for the protective effects of Nrg4 (9).

Although the beneficial role of Nrg4 in metabolic disorders in mouse models is evident from several studies, its pathophysiological implications in human NAFLD have not been explored fully (14). Furthermore, its regulation in humans, especially the function of its variants, is unclear. It is well established that studies in human genetics can enrich the understanding of disease mechanisms and help to develop new therapeutics. To evaluate the physiological roles of NRG4 in humans, we performed genotyping in a community-based population of 11,022 subjects, whole-exome sequencing (WES) in 247 severely obese subjects, and exome arrays in 2,388 subjects from the Shanghai Obesity Study (SHOS) to screen for NRG4 mutations. Therefore, in this study, we aimed to evaluate the physiological roles of NRG4 in humans, particularly to understand how the NRG4 variants regulate the pathogenesis of NAFLD and related metabolic disorders.

Ethical approval was granted by the institutional review board of Shanghai Jiao Tong University Affiliated Sixth People’s Hospital according to the principles of the Declaration of Helsinki. A total of 11,022 community-based subjects from the Shanghai Nicheng Cohort Study were recruited for genome-wide arrays to screen for metabolism-related hot spot chromosomal regions (15). The flowchart describing the inclusion and exclusion criteria for subject selection is shown in Supplementary Fig. 1. Additionally, 2,388 individuals from SHOS were recruited for exome genotyping (16–18), and 247 participants with BMI >35 kg/m² were recruited for WES screening (19). Metabolic trait assessment was performed as we have previously described (15). Moreover, to screen for NRG4 mutations in a larger sample size, 1,848 subjects from another cross-sectional population (Shanghai Diabetes Study [SHDS]) (20,21) were used.

WES was performed following the method described in a previous study (19). Infinium Asian Screening Array version 1.0 (Illumina, San Diego, CA) and Infinium Exome BeadChip version 1.0 (Illumina) were used for genotyping loci across the genome. Before analysis, genotype data were confirmed by quality control procedures as we previously described (22). Imputation by Minimac3 software was conducted in 504 East Asian individuals in 1000 Genomes phase 3 (23). Estimated r² ≥0.4 was selected for further analysis. Data were mapped to National Center for Biotechnology Information build 37. To screen the mutation of NRG4, a larger sample size of 5,577 subjects of the Shanghai Nicheng Cohort Study and 1,848 subjects of the SHDS were genotyped (20,21) by matrix-assisted laser desorption ionization/time-of-flight mass spectroscopy (Agena Bioscience, San Diego, CA).

Six-week-old male C57BL/6J mice were purchased from the Nanjing Biomedical Research Institute of Nanjing University (Nanjing, China). Mice were fed an HFD (D12492; Research Diets Inc.) starting at 8 weeks of age. Tail vein injections of adeno-associated virus (AAV) were performed after 8 weeks of HFD feeding. Mice were placed individually in a chamber for 96 h, and a Comprehensive Laboratory Animal Monitoring System (Columbus Instruments) was used to monitor physical activity and metabolism. The first 24 h served as an adaptive phase, and in the following 72 h, oxygen consumption, food intake, and total movement were recorded for analysis. Body fat content was assessed by nuclear magnetic resonance (echo MRI). Metabolic measurements and liver lipid analysis were performed following the method described earlier (24).

AAV9-green fluorescent protein (GFP) plasmids were generated for the overexpression of Nrg4. The vector of pAAV-thyroxine-binding globulin (TBGp)-enhanced GFP (EGFP)-3Flag-simian virus 40 (SV40) PolyA with BamHI/BamHI restriction enzyme sites was used to insert the mouse wild-type (WT) Nrg4 cDNA (purchased from GeneChem, Shanghai, China). Two missense point mutations were introduced into WT Nrg4 cDNA by a PCR-based method to obtain the AAV-TBG mutant Nrg4 construct.

Intraperitoneal glucose tolerance and intraperitoneal insulin tolerance tests were performed in mice at 12 weeks and 14 weeks after AAV delivery, respectively, following procedures described previously (25).

Primary hepatocytes were isolated and maintained as previously described (26). An adenovirus encoding mouse ErbB4 (pDC315-3Flag-SV40-EGFP) was purchased from GeneChem. Adenovirus infection experiments were conducted on the same day of isolation as previously described (9). After 24 h, cells were treated with Nrg4 (20 μg/mL) for 20 min for use in further experiments.

Immunoblotting experiments were performed using specific antibodies against SREBP-1C (sc-13551; Santa Cruz Biotechnology). The following antibodies were purchased from Cell Signaling Technology: lamin A/C (#2032), phospho-ErbB4 (Tyr1284, #4757), ErbB4 (#4795), phospho-ErbB3 (Tyr1289, #4791), ErbB3 (#12708), phospho-Stat5 (Tyr694, #9359), Stat5 (#94205), phospho-Akt (Ser473; #4060), Akt (#9272), β-actin (#4970), and HSP90 (#4877). Anti-Nrg4 antibody was purchased from Invitrogen (PA5-102641).

Surface plasmon resonance (SPR) measurements were performed on a Biacore T200 instrument equipped with a CM5 sensor chip. All the proteins used in SPR analysis were diluted with Biacore buffer consisting of 10 mmol/L HEPES (pH 7.5), 150 mmol/L NaCl, and 0.005% (v/v) Tween 20 via gel filtration. ErbB4 protein from R&D Systems (4387-ER) was immobilized on a chip. The WT and mutant Nrg4 proteins were diluted serially. The analytes were then used to flow over the chip surface with the measured response units. The binding kinetics were analyzed using Biacore T200 Insight Evaluation Software, and the equilibrium binding and disassociation constants were calculated.

The primary hepatocytes were infected with GFP-ErbB4 adenovirus for 36 h and then incubated with glutathione S-transferase (GST)-Nrg4 (500 μg/mL) for 1 h. Alexa Fluor 647–conjugated GST monoclonal antibodies purchased from Cell Signaling Technology (#3445) diluted 1:50 were incubated with cells for 1 h. Samples were run on an ImageStreamX MKII using INSPIRE data acquisition software (Amnis EMD-Millipore) at a concentration of ∼1 × 10⁶ cells in 50 μL of PBS. Data analyses were performed using IDEAS version 6.0 software (Amnis EMD-Millipore). HepG2 cells were fixed with 4% formaldehyde for 15 min at 20–25°C. The remaining steps were the same as in the primary hepatocytes.

Statistical differences were determined by ANOVA and Student-Newman-Keuls (SNK) multiple pairwise comparison test using SAS 9.0 software. P < 0.05 was considered significant.

The data generated in this study are available from the corresponding author upon reasonable request.

First, we examined the association between NRG4 locus with obesity and related metabolic perturbations in humans. In the cohort study of 11,022 subjects (15) (Supplementary Table 1 and Supplementary Fig. 1), we found that the loci associated with serum triglyceride levels, glycosylated hemoglobin (HbA_1c), γ-glutamyl transferase, fasting plasma insulin, and HOMA of insulin resistance (HOMA-IR) were enriched in the NRG4 gene region on chromosome 15q24.2 (Fig. 1A), suggesting the possible role of NRG4 in metabolic disorders. However, all loci in the NRG4 gene region were defined in introns and regulatory regions, which could be due to the limitations of gene arrays (Supplementary Figs. 2 and 3). To further understand the pathogenic mutations, we next took advantage of exon arrays in 2,388 subjects from SHOS and WES in 247 subjects with severe obesity to explore the target region (Fig. 1B and C). One novel heterozygous mutation in NRG4, R44H (p. Arg44his), was identified in two siblings (Fig. 1B and Supplementary Tables 2 and 3). Interestingly, one of the siblings showed dyslipidemia and liver dysfunction.

Using WES of the 247 subjects with severe obesity (19), we identified a carrier of the NRG4 E47Q heterozygous variant, which was further validated by Sanger sequencing (Fig. 1C). The NRG4 E47Q carrier was a 22-year-old woman who exhibited severe obesity with a BMI of 39.78 kg/m² (Table 1). Abdominal fat distribution is usually correlated with metabolic disorders such as NAFLD, type 2 diabetes, and atherosclerosis (27–29). In Chinese women with obesity, the average visceral and subcutaneous fat areas are 104.2 and 166.0 cm², respectively, with a mean BMI of 24.6 kg/m² (15). In our severe obesity population with similar BMI, E47Q carriers had less visceral fat area (64.56 vs. 142.44 cm²) and waist-to-hip ratio (WHR) (0.76 vs. 0.95) (Table 1), suggesting that fat accumulation was mainly subcutaneous. The glucose and lipid metabolism of the carriers were normal.

In addition, we examined the receptor ERBB4 locus (chromosome 2: 212212021–213433166) for genetic variants and their effects on obesity and associated metabolic perturbations. The results are shown in Supplementary Fig. 4 and Supplementary Tables 4 and 5. Further, we genotyped NRG4 R44H and E47Q in 7,425 subjects (5,577 from the Shanghai Nicheng Cohort Study and 1,848 from the SHDS) to validate the results (Supplementary Table 6), but we did not observe any of the two variants.

In terms of protein structure, NRG4 has a highly conserved EGFL domain to exert biological functions. However, in the p.R44H variant, the 44th amino acid was altered from arginine (pI 10.76) to histidine (pI 7.59), and in the p.E47Q variant, the 47th amino acid was altered from glutamic acid (pI 3.22) to glutamine (pI 5.65) (Fig. 1D). As both arginine and glutamic acid are charged between the third disulfide bond, their relative positions are highly conserved within the neuregulin family. Therefore, it is speculated that mutations at these sites might alter metabolism through aberrant NRG4 function (Supplementary Fig. 5).

To further dissect the functional involvement of Nrg4 in metabolism, we generated mouse models through AAV9-mediated delivery (Fig. 2A). For better comparison, mice were fed an HFD for 8 weeks and then divided into four groups: WT Nrg4 overexpression, reconstituted mutant Nrg4 R44H overexpression, reconstituted mutant Nrg4 E47Q overexpression, and AAV-GFP. After 1 month of intravenous tail injections, expression of Nrg4 was significantly increased in the liver in Nrg4 overexpression mice (Fig. 2B and Supplementary Fig. 6). We next analyzed the metabolic parameters of Nrg4 and GFP overexpression mice. Upon HFD feeding for 22 weeks, WT Nrg4 mice gained less weight accompanied by significantly reduced percentages of fat and lean mass compared with that of GFP mice (Fig. 2C and D). R44H mice had more body weight gain than WT mice but less than GFP mice. E47Q mice showed a similar growth curve and percentages of fat and lean mass to WT mice. There were significant differences in liver and white adipose tissue (WAT) weight among the groups (Supplementary Fig. 7A–C). Wang et al. (9) demonstrated that Nrg4 could attenuate hepatic steatosis by the activation of STAT5 signaling in diet-induced obesity (DIO), but the activation status of AKT pathways appeared unchanged. Consistently, our experiments showed a similar attenuation effect on the hepatic lipogenic process of Nrg4 (Supplementary Fig. 7D). Compared with the WT mouse livers, the E47Q mouse livers exhibited a notably strong activation status of the STAT5 pathways, whereas the R44H mouse livers displayed moderately weakened activation. Rectal body temperature was nearly indistinguishable among the HFD groups (Supplementary Fig. 7E). However, no significant difference was found among groups fed a chow diet (Supplementary Fig. 8). Furthermore, to establish how Nrg4 influences diet-induced weight gain, we performed metabolic cage studies in HFD-fed overexpression mice using a Comprehensive Laboratory Animal Monitoring System to assess energy balance. Oxygen consumption rate and energy expenditure were significantly elevated in WT and E47Q mice compared with GFP mice, while they were lower in R44H mice than in WT mice (Fig. 2E). In contrast, food intake appeared comparable in the four groups during both dark and light phases. WT and E47Q mice also exhibited increased locomotor activity during both dark and light phases, whereas the total activity was similar in GFP and R44H mice (Fig. 2E). These results strongly suggest that Nrg4 elevation protects mice from DIO by stimulating fuel oxidation and increasing energy expenditure, while R44H mutation loses this effect partially.

We next determined how Nrg4 regulates whole-body glucose metabolism after 12 or 14 weeks of AAV9-mediated delivery. Consistent with decreased adipose mass, WT mice showed significant improvement in glucose intolerance, insulin resistance, and decreased fasting plasma insulin levels following DIO, and E47Q mice showed enhanced protective effects on glucose homeostasis (Fig. 3A–C). WAT histological analyses (Fig. 3D) showed that WT and E47Q mice had suppressed HFD-induced macrophage infiltration. Reportedly, the activation of Nrg4 could ameliorate adipose tissue hypoxia under obese conditions by promoting angiogenesis in WAT (30). We also performed CD31 staining in the epididymal adipose tissue (EAT) of the four mouse groups to observe the blood vessel density (Fig. 3E and F). The capillary area of the WT Nrg4 group was higher than the GFP group, whereas that of the R44H and GFP groups were similar. However, there was no significant difference between the E47Q and WT groups. E47Q mouse livers developed the least fat accumulation accompanied by decreased liver triacylglycerol (TAG) (Fig. 3D and G and Supplementary Fig. 7), whereas R44H mice showed similar adipose inflammatory status and hepatic steatosis as the GFP control. These results provide direct evidence that WT Nrg4 blocks hepatic lipid accumulation, whereas E47Q enhances this process to some extent. Moreover, the impairment in insulin resistance in the liver was proportional to hepatic and visceral fat contents (Fig. 3G). Our results showed that E47Q mice had similar ALT and AST values to WT mice, while R44H mice had a higher ALT value than WT mice (Fig. 3H).

Because R44H and E47Q mice exhibited partially different systemic metabolism and hepatic steatosis, we hypothesized that E47Q could be a gain-of-function mutation, whereas R44H could be a loss-of-function mutation. We next performed studies to confirm the regulatory effects of the three types of Nrg4 in liver lipogenesis in cultured primary mouse hepatocytes. Hepatocytes transduced with GFP or ErbB4 adenovirus were treated with palmitate and reconstituted Nrg4 protein for 24 h. Similar to the in vivo results, intracellular triglyceride accumulation was blocked significantly in the WT Nrg4–treated group, whereas E47Q mutation enhanced the inhibition efficacy of WT Nrg4 (Fig. 4A). The combination of oleate and palmitate treatment strongly induced the mRNA expression of lipogenic genes, such as Fasn, Lpin1, and Scd1 (Fig. 4B), and proinflammatory cytokines, such as tumor necrosis factor-α and Ccl2. As expected, treatment of WT and E47Q Nrg4 significantly lowered the stimulatory effects of long-chain fatty acids on these genes. However, compared with WT Nrg4, R44H showed a significant decrease in the inhibition of lipid synthesis (Fig. 4B).

To investigate the mechanism underlying the alteration of function observed in R44H and E47Q Nrg4, we examined the ErbB receptor–activated signaling pathway. Our results indicated that phosphorylation of ErbB4 decreased upon R44H treatment and increased slightly upon E47Q treatment without affecting the total ErbB4 levels (Fig. 4C). The transcription factor STAT5 was notably involved in NAFLD (31). Our results also showed that the phosphorylation of ErbB3 and STAT5 accorded with the trend of ErbB4 in the case of the two mutant proteins. SREBP-1C has been identified as a key regulator of the de novo lipogenesis pathway in the liver (32,33). In our study, the cleaved and transcriptionally active isoform of Srebp-1 in the nucleus (nSREBP-1) was significantly suppressed in WT and E47Q Nrg4 treatment groups. Protein levels of precursor Srebp-1 (pSREBP-1) from total lysates were similar in the three groups (Fig. 4D). Together, our studies revealed that Nrg4 attenuated the progress of hepatic lipogenesis and that R44H Nrg4 weakened its protective function.

To explore the interaction among the three versions of purified Nrg4 (WT, R44H, and E47Q) and ErbB4 protein, we performed SPR experiments to test the binding affinity of the Nrg4 proteins to ErbB4 protein using a Biacore T200 instrument, which identified ErbB4 as the direct receptor of Nrg4. The active binding affinity of WT Nrg4 to ErbB4 was 2.20 μmol/L (Fig. 5A). As expected, R44H Nrg4 did not bind to ErbB4, while E47Q Nrg4 showed a higher binding affinity of 1.66 μmol/L (Fig. 5B and C). Subsequently, we performed imaging flow cytometry to measure protein-protein interactions (Fig. 5D). To facilitate detection, we generated fusion proteins between GST and the extracellular fragment of WT or mutant Nrg4. Using these constructs, we performed binding assays in primary hepatocytes that were transfected with GFP-ErbB4 for 36 h. While the transfection efficiency of GFP-ErbB4 adenovirus was similar in the three groups, the proportions of ErbB4⁺Nrg4⁺ hepatocytes in ErbB4⁺ hepatocytes were different. Compared with WT Nrg4, increasing combining capacity of E47Q Nrg4 was evident; however, R44H Nrg4 almost lost its combining capacity (18.12% for WT vs. 0.85% for R44H vs. 27.21% for E47Q) (Fig. 5E). Similar results were obtained in HepG2 cell lines (Supplementary Fig. 9). These data indicate that the mutations either enhance or attenuate the physiological function of Nrg4 by altering the binding affinity between Nrg4 and ErbB4, which could be due to the alteration in the highly conserved protein structure.

Nrg4, a fat-derived endocrine factor, plays an important role in systemic glucose, lipid, and energy homeostasis in the adipose-hepatic endocrine axis in mouse models (14,30,34–36). In this study, we explored human genetic variants of NRG4 and identified R44H and E47Q as novel mutations that affect Nrg4 function. We validated that Nrg4 inhibits hepatic lipogenesis through the ErbB4-STAT5-SREBP-1C pathway. We then demonstrated that mutations altered the binding affinity of Nrg4 to ErbB4 and further mediated the signal pathway by up/downregulating the phosphorylation of ErbB4 to influence hepatic lipid metabolism.

Multiple large-scale genome-wide association studies (GWAS) have been conducted to construct the genetic basis of NAFLD (37,38). However, most variants account for only a modest proportion of the estimated heritability of most complex traits (39,40). One of the reasons is that common genetic variants confirmed by GWAS are always located in intron regions, which neglects the rare variants located in exons (39–41). WES provides a supplementary approach to molecular diagnosis, including detection of rare genetic variants and new mutations contributing to disease state (42).

Given that GWAS have been prominent in identifying novel variant-trait associations, we first performed genotyping and identified the NRG4 gene region as a candidate because of its enrichment in metabolism-related variants. We then carried out exome arrays in SHOS and WES in severely obese subjects to screen NRG4 for pathogenic variants. Interestingly, two novel and rare variants, p.R44H and p.E47Q, were discovered in NRG4.

Although genetic studies based on humans implied functional involvement and a potential mechanism of variants, these results lack experimental evidence that can provide direct causality. Thus, it is critical to conduct further animal studies to verify the function of the target protein. We successfully constructed a mouse model by AAV injection, and three kinds of Nrg4 were overexpressed >10 times. However, because of the current limitations of technology, it is difficult to detect serum Nrg4 in rodents, even in the overexpression group. Therefore, we instead performed quantitative PCR analysis to evaluate the overexpression efficiency and found that overexpression was mainly concentrated in liver and adipose tissue. Despite the overexpressed levels of Nrg4 being much higher than regular physiological expression, the overexpression efficiencies of WT Nrg4 and mutant Nrg4 were similar; therefore, the effects of the three kinds of Nrg4 could be compared. Of note, WT Nrg4 significantly alleviated hepatic steatosis by promoting ErbB signaling in HFD-fed mice. The different hepatic phenotype in the four groups could be partly due to the enhanced or inhibitory effect of Nrg4/ErbB4 signaling on hepatic lipogenesis. Compared with the WT Nrg4 mouse livers, the E47Q mouse livers exhibited a notably strengthened activation status of the STAT5 pathways, but the R44H mouse livers displayed a moderately weakened activation. However, we also found that the E47Q Nrg4 overexpression mice showed a similar weight gain to that of the WT mice, which could be because the Nrg4 concentration in the overexpression mice might have maximized its beneficial effect on weight and fat mass; therefore, E47Q, even as a gain-of-function mutation, was unlikely to be phenotypically different from the WT in vivo.

Neuregulins (Nrg1–4) are a family of proteins containing an EGFL motif that activates the ErbB/HER receptor tyrosine kinase family, which is involved in the regulation of diverse biological processes. SPR is a method of detecting both weak and strong protein-protein interactions ranging from the millimolar to the nanomolar range in vitro (43). In this study, we confirmed through SPR experiments that Nrg4 indeed binds to ErbB4 directly with low binding affinity (equilibrium dissociation constant [KD] = 2.20 μmol/L). Several groups have reported that the KD varies from 127 to 179 nmol/L, indicating moderate affinity between NRG1 and ErbB4 and is >10² μmol/L between NRG1β and ErbB1 using Biacore (44–46). Thus, the binding affinity of the NRGs to the ErbB family seems to be in the hundreds of nanomolar to micromolar affinity range. In our experiments, the active binding affinity of WT Nrg4 to ErbB4 was 2.20 μmol/L, weaker than the affinity between Nrg1 and ErbB4. To the best of our knowledge, there are no other SPR experiments that have measured the binding affinity of Nrg4 and ErbB4 directly. Although our extracellular amino acid sequence has been confirmed to be biologically active, the influence of protein refolding on the protein-protein interaction cannot be excluded. It is possible that it is partially biologically active because of partial refolding of the protein. However, we paid more attention to the change value of the binding affinity between WT and R44H or E47Q rather than its absolute KD value.

In summary, our results suggest that mutations in the NRG4 gene from humans translate to loss-of-function and gain-of-function proteins and participate in metabolism. We demonstrated the possibility that NRG4 deficiency increases susceptibility to the pathological consequences of NAFLD. These findings also provide strong human genetic evidence to inform the development of NRG4 agonists for weight loss and the treatment of obesity-associated metabolic disease. Clinically, the translation of these gene mutations to affect diagnosis and treatment remains a significant and growing challenge. Together, our study suggests that these two mutations from clinical data in the NRG4 gene locus translate to loss-of-function and gain-of-function proteins, and it advances our understanding of how to effectively target specific molecules for the treatment of common complex diseases, such as obesity and NAFLD.

Acknowledgments. The authors are grateful to all study participants, without whom this study would not have been possible. The authors thank the Department of Endocrinology and Metabolism, Shanghai Jiao Tong University Affiliated Sixth People’s Hospital, for data collection. They also thank Prof. Jiandie Lin (University of Michigan Medical Center) for constructive comments on the studies.

Funding. This study was supported by the National Key Research and Development Project of China (2016YFC1304902 and 2018YFA0800402), Outstanding Academic Leaders of Shanghai Health System Program (2017BR008), National Science Foundation of China (81800702), Yangtze River Scholar Program, National Natural Science Foundation of China (81974118), Shanghai Outstanding Academic Leaders (20XD1433300), Shanghai Sailing Program (18YF1418900), and Shanghai Municipal Education Commission Gaofeng Clinical Medicine Grant Support (20152527).

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

Author Contributions. Y. Li, L.J., Q.Y., and Yi. Z. contributed to the animal experiments. F.J., J. Yan, and R. Z. contributed to the data acquisition. Y. Li and L.J. contributed to the statistical analysis. H.Z. conducted the bioinformatics analysis. H.Y., Yu. Z., and Z.H. collected the clinical samples and data. Y. Lu, J. Yang, and C.H. provided supervision and mentorship. C.H. contributed to the research idea and study design. Each author contributed important intellectual content during manuscript drafting or revision and accepts accountability for the overall work by ensuring that questions pertaining to the accuracy or integrity of any portion of the work are appropriately investigated and resolved. C.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.

Prior Presentation. Parts of this article were presented in poster form at the 80th Scientific Sessions of the American Diabetes Association, 12–16 June 2020.