Adiponectin, Leptin, and IGF-1 Are Useful Diagnostic and Stratification Biomarkers of NAFLD

Background: Nonalcoholic fatty liver disease (NAFLD) is the most common chronic liver disease where liver biopsy remains the gold standard for diagnosis. Here we aimed to evaluate the role of circulating adiponectin, leptin, and insulin-like growth factor 1 (IGF-1) levels as non-invasive NAFLD biomarkers and assess their correlation with the metabolome. Materials and Methods: Leptin, adiponectin, and IGF-1 serum levels were measured by ELISA in two independent cohorts of biopsy-proven obese NAFLD patients and healthy-liver controls (discovery: 38 NAFLD, 13 controls; validation: 194 NAFLD, 31 controls) and correlated with clinical data, histology, genetic parameters, and serum metabolomics. Results: In both cohorts, leptin increased in NAFLD vs. controls (discovery: AUROC 0.88; validation: AUROC 0.83; p < 0.0001). The leptin levels were similar between obese and non-obese healthy controls, suggesting that obesity is not a confounding factor. In the discovery cohort, adiponectin was lower in non-alcoholic steatohepatitis (NASH) vs. non-alcoholic fatty liver (AUROC 0.87; p < 0.0001). For the validation cohort, significance was attained for homozygous for PNPLA3 allele c.444C (AUROC 0.63; p < 0.05). Combining adiponectin with specific serum lipids improved the assay performance (AUROC 0.80; p < 0.0001). For the validation cohort, IGF-1 was lower with advanced fibrosis (AUROC 0.67, p < 0.05), but combination with international normalized ratio (INR) and ferritin increased the assay performance (AUROC 0.81; p < 0.01). Conclusion: Serum leptin discriminates NAFLD, and adiponectin combined with specific lipids stratifies NASH. IGF-1, INR, and ferritin distinguish advanced fibrosis.


INTRODUCTION
Non-alcoholic fatty liver disease (NAFLD) is one of the most prevalent chronic liver conditions and an important risk factor for liver cirrhosis and hepatocellular carcinoma. The NAFLD spectrum varies from simple fat accumulation to nonalcoholic steatohepatitis (NASH), characterized by hepatocellular injury, inflammation, and, eventually, fibrosis (1). In fact, the increasing prevalence of NAFLD parallels that of obesity and is accompanied by rising liver-related morbidity and mortality and, consequently, increased socio-economic burden (2). Clinical practice guidelines recommend NAFLD screening in individuals with metabolic risk factors, such as obesity, type 2 diabetes, and hypertension (metabolic syndrome) (3,4). Most NAFLD patients are clinically asymptomatic, and when present, the symptoms are unspecific and associated with advanced disease. Furthermore, liver function tests may be within normal range or only slightly increased. Thus, panels composed by anthropometric and blood biochemistry data have been proposed to assess hepatic steatosis (fatty liver index, hepatic steatotic index, NAFLD liver fat score, SteatoTest), although with modest performance compared with imaging biomarkers. In turn, imaging techniques [ultrasonography, magnetic resonance imaging (MRI)] are not fully conclusive at identifying liver inflammation and hepatocyte injury typical of NASH or distinguishing NASH from fatty liver (3,4). Indeed differential diagnosis between non-alcoholic fatty liver (NAFL) and NASH is an important indicator of increased risk of cirrhosis and other hepatic co-morbidities, although the stage of fibrosis is increasingly recognized as the most relevant NASH prognostic marker (5).
Although not yet fully understood, the disease pathogenesis involves disturbances in adipose tissue, insulin resistance, and inflammation (11). Thus, adipokines' and related cytokines' role in NAFLD pathogenesis and severity has been equated (12), and their potential as non-invasive biomarkers of diagnosis and staging deserves further investigation. Moreover, lipid and metabolite signatures in plasma and in liver tissue associated with NAFLD and/or fibrosis (13)(14)(15)) may further support non-invasive NAFLD diagnosis (16).
We hypothesize that serum adipokines-leptin and adiponectin-and liver-produced insulin-like growth factor 1 (IGF-1) are useful circulating biomarkers for NAFLD diagnosis and stratification. We assessed their potential as non-invasive biomarkers in the discovery and validation cohorts of biopsy-proven NAFLD patients and evaluated their impact on the metabolome.

Patient Cohorts
Human serum samples were collected from two independent cohorts of morbidly obese adult patients with biopsy-proven diagnosis of NAFLD, who were undergoing bariatric surgery [body mass index (BMI) ≥35 kg/m 2 ], and healthy-liver individuals (17) (BMI <35 kg/m 2 ) from Santa Maria Hospital (Lisbon, Portugal; discovery cohort) and Donostia University Hospital (San Sebastian, Spain; validation cohort). Up to 20% of the patients were taking statins. The patient inclusion criteria consisted in liver biopsy-proven NAFLD diagnosis and BMI ≥35 kg/m 2 . The exclusion criteria included alcohol consumption ≥20 (females) or ≥30 g/day (males), viral hepatitis B and C, and other causes of chronic liver disease. The liver biopsies were obtained during bariatric surgery, under standard procedures.
Fibrosis was assessed from a distance >0.5 cm from the liver capsule. Blood and tissue samples were collected at a maximum of 1-week interval, while liver imaging was performed within 24 h of surgery. Sample collection was performed after receipt of the patient's informed consent and the Institutional Review Board's approval, in accordance with the Declaration of Helsinki. For histology purposes, formalin-fixed paraffin-embedded tissue sections were routinely stained with hematoxylin-eosin method and with Masson's trichrome for fibrosis evaluation. Histological evaluation was performed by experienced pathologists in a blinded fashion following agreement on the histological features of NAFLD/NASH assessment and NASH diagnosis. NAFLD activity score (NAS 0-8) and fibrosis stage (F0-4) were assigned to each biopsy according to the NASH Clinical Research Network histological scoring system (18). Fibrosis extent was dichotomized as early (F0-2) and advanced (F3-4) fibrosis. Where possible, fibrosis scores APRI (19), , and NAFLD fibrosis score (NFS) (21) were calculated. Individuals were classified as healthy non-obese controls (BMI <35 kg/m 2 ), healthy obese controls [BMI ≥35 kg/m 2 , maximum NAS = 1 (if steatosis = 0) and maximum fibrosis score = 1], NAFL (NAS ≤4), or NASH (NAS ≥5). Liver biopsies were obtained from healthy non-obese individuals that were referred for hepatic surgery without an underlying liver disease (17).

Serum Hormones and Data Analysis
The serum levels of leptin, adiponectin, and IGF-1 were measured in single determination using specific enzyme-linked immunosorbent assay kits according to the manufacturer's instructions (Mediagnost GmbH, Reutlingen, Germany) and correlated with NAFLD activity score, fibrosis stage, biochemical parameters, genotypes, imaging data, and co-morbidities. Data analysis was first performed in the discovery cohort and then expanded to the validation cohort. The normality of values distribution was tested with Kolmogorov-Smirnov (n ≥ 50) or Shapiro-Wilk (n < 50) tests. According to the normality of values distribution, one-way ANOVA or Kruskal-Wallis test followed by Bonferroni or Dunn's multiple-comparison test were used to determine differences between sample groups according to NAS score, steatosis, lobular inflammation and ballooning stages, and genotypes. Differences according to fibrosis dichotomy and presence of co-morbidities were determined with unpaired ttest or Mann-Whitney U-test. To correlate hormone serum levels and biochemical parameters or imaging data, Pearson or Spearman correlation coefficient and linear regression analysis were applied. Uniparameter receiver operating characteristic (ROC) curve analysis tested the hormones' discriminatory power. Cutoff values were retrieved from Youden index. Variables presenting an area under the ROC curve (AUROC) value superior to 0.60 and p <0.05 were selected for binomial logistic regression analysis. Multi-parameter ROC curves were established for relevant parameter combinations, where an AUROC value >0.80 was considered a very good assay performance. Associated sensitivity, specificity, and positive predictive value (PPV) and negative predictive value (NPV) were estimated. A p-value <0.05 was considered significant. All statistical analysis were performed using GraphPad Prism, version 8.0.2 (La Jolla, California, USA) and IBM SPSS Statistics, version 26 (Armonk, New York, USA).

Metabolomic Analysis
Serum samples from 198 patients in the validation cohort were randomly selected for metabolomic analysis by liquid chromatography-time of flight-mass spectrometry. Two separate liquid chromatography-time of flight-mass spectrometry-based platforms were used: platform 1 to analyze methanol lipid extracts [including fatty acids (FA), acyl carnitines, bile acids, lysoglycerophospholipids, Nacyl ethanolamines, and oxidized FA] and platform 2 for methanol/chloroform lipid extracts (including glycerolipids, cholesterol esters, sphingolipids, and glycerophospholipids) (22). Quality control calibration and validation samples were used to assess data quality. The spectra were mass-corrected by reference to leucine enkephalin. System control and data pre-processing were performed using Masslynx 4.1 software and TargetLynx application manager (Waters Corp., Milford, USA). Peak detection, noise reduction, and data normalization were performed as previously described (23). For univariate analyses, the samples were grouped according to the median serum level of each hormone (leptin = 39.61 ng/ml; adiponectin = 4.84 µg/ml; IGF-1 = 103.65 ng/ml). Log 2 fold change for metabolite levels between higher than median vs. lower than median hormone level groups was determined and compared using unpaired Student's t-test (p < 0.05) using R v.3.4.1 (R Development Core Team, 2017; http://cran.r-project.org). A p-value <0.05 was considered significant.

Clinical and Biochemical Features of Patient Cohorts
In both cohorts, the NAFL and NASH patients were similar with respecting to age, gender, BMI, gamma-glutamyl transferase (GGT), serum glucose, triglycerides, and cholesterol [total, highdensity lipoprotein (HDL) or low-density lipoprotein (LDL)] concentrations (Tables 1, 2). While in the discovery cohort the NAFL and NASH patients had similar circulating AST and ALT, in the validation cohort these liver injury markers
Overall, serum leptin predicts NAFLD, correlates with serum lipid changes, and is a potentially valuable tool for NAFLD diagnosis.
Although not significant, higher serum adiponectin positively correlated with serum lipid classes, except for DG, TG, and steroids (Figure 2). Both PE and PC containing arachidonic (20:4) or docosahexaenoic (DHA; 22:6n-3) acids were increased in patients with high circulating adiponectin (p < 0.05 and p < 0.01). Moreover, the serum levels of 147 metabolites were significantly different between NAFL and NASH, of which 35 were highly correlated with serum adiponectin (Supplementary Table 2). Among those, nine TG were already described as capable of distinguishing NAFL from NASH (16).
Sixty-five out of the 147 metabolites presented AUROC >0.60 and were considered for binomial regression analysis. This allowed establishing a panel combining adiponectin and nine serum lipids (mainly TG, PC, and SM), reaching an AUROC value of ∼0.80 and an overall accuracy of 74% in distinguishing NAFL from NASH in all patients, regardless of PNPLA3 genotype. When this panel was applied to homozygous carriers of the PNPLA3 c.444C allele, the AUROC value increased to 0.83 and the accuracy to 80% (Table 4).
Overall, serum adiponectin inversely correlated with liver injury markers and, when combined with nine specific lipids, distinguished NAFL from NASH patients with high accuracy.    (Figures 4A,B). In agreement, in the validation cohort, there were also no significant differences in IGF-1 relating to liver steatosis, lobular inflammation, or hepatocyte ballooning severity (Supplementary Figure 3), yet IGF-1 was significantly lower in NAFLD patients with advanced fibrosis (F3-4; p < 0.05) ( Figure 4C). By establishing a cutoff value of IGF-1 inferior to 98.83 ng/ml for the presence of advanced fibrosis, 70% sensitivity, 61% specificity, and 63% overall accuracy were obtained ( Figure 4C and Table 5). Overall, IGF-1, as well as the pre-established biomarker fibrosis panels APRI, Fib-4, and NFS, displayed a suboptimal performance in F0-2 vs. F3-4 discrimination. However, by combining serum IGF-1, ferritin, and INR, the discriminatory power was significantly improved, resulting in AUROC value of 0.81 and overall accuracy of 93%, considering the ∼10% prevalence of advanced fibrosis in this cohort ( Table 5). Importantly, besides a robust NPV, this panel displayed 71% PPV value, contrasting with the low PPV values from APRI, Fib-4, and NFS. Circulating IGF-1 did not correlate with other parameters or co-morbidities (Supplementary Figure 3). Serum metabolomic analysis showed that higher IGF-1 was associated with an overall increase in main lipid classes, significant for LPE (p < 0.05), PC containing arachidonic acid (20:4; p < 0.05), diacylglycerophosphocholine (p < 0.05), and monoacylglycerophosphoethanolamine (p < 0.05) ( Figure 4D). Moreover, 26 specific serum metabolites were able to distinguish advanced fibrosis, with several bile acids increased in advanced fibrosis, and both cholic acid and PC (17:0/18:2) positively correlated with IGF-1 (Supplementary Table 3).
IGF-1 altogether could identify advanced fibrosis, particularly in combination with INR and ferritin, and correlates with serum lipid changes associated with fibrosis.

DISCUSSION
The strong association between obesity and NAFLD, parallel with the continuous rise in the prevalence of obesity, prompts a more detailed study of the role of adipokines and other hormones as potential non-invasive biomarkers for NAFLD (1). Therefore, we measured serum leptin, adiponectin, and IGF-1 in two independent cohorts, comprising both biopsyproven healthy-liver controls as well as obese NAFL and NASH patients. In the discovery cohort, the serum hormones were correlated with clinical and histological information, including NAS, fibrosis stage, and biochemical parameters. Results were then validated in a second cohort, which further allowed us to explore the role of obesity as a confounding factor by including a control group of obese healthy-liver individuals. Additionally, in this validation cohort, hormone serum levels were correlated with liver imaging, co-morbidities, and NAFLD risk-associated polymorphisms. For a subset of patients, serum metabolomic analysis was also performed. This strategy showed that particularly leptin and adiponectin can be potential diagnosis and stratification biomarkers in NAFLD. The role of these two adipokines in NAFLD pathophysiology, diagnosis, and even treatment has attracted significant attention despite the controversial results.
Leptin is a 16-kDa pro-inflammatory adipokine produced by adipocytes. Serum leptin levels reflect total body mass. Leptin may exert an anti-steatotic action in early NAFLD by promoting FA oxidation and decreased lipogenesis and a pro-inflammatory and pro-fibrotic action at later disease stages by increasing hepatic reactive oxygen species generation, pro-inflammatory cytokine release, and enhanced fibrinogenesis (28,29). Previous reports have generally found increased serum leptin in NAFLD patients (12,29,30). Still the role of obesity as a bias remains poorly explored. Our results in the discovery cohort showed that leptin accurately identified NAFLD patients, where there was no correlation with BMI. Moreover, in the validation cohort, although leptin levels positively correlated with BMI, they were higher in NAFL and NASH compared with both non-obese and obese controls, suggesting that the association between NAFLD and leptin is more robust than the effect of obesity on leptin levels. In this cohort, serum leptin showed again a very good performance as a potential biomarker to distinguish NAFLD patients from healthy individuals. After applying the cutoff value of leptin at >9.33 ng/ml for diagnosing NAFLD in the validation cohort, despite having a lower general accuracy, it still presented sensitivity, specificity, and NPV values far superior than those previously reported (31,32).
In turn, adiponectin, a 30-kDa protein mainly secreted by adipose tissue, is the most well-studied adipokine in the pathogenesis of NAFLD (30,33). It is considered to have antiinflammatory, anti-steatotic, and anti-fibrotic effects. In the liver, adiponectin prevents lipid accumulation by promoting FA oxidation via peroxisome proliferator-activated receptor alpha (33). Adiponectin also downregulates pro-inflammatory cytokines produced by Kupffer cells and hepatic stellate cells, thus inhibiting their transformation into myofibroblasts and consequently decreasing liver fibrosis (30,(34)(35)(36). Several reports have shown a reverse relation between adiponectin circulating levels and body fat mass and its decrease in obesity, type 2 diabetes, insulin resistance, and dyslipidemia (30,34). A meta-analysis showed that adiponectin decreases in healthy controls compared to NAFL, and in the latter compared to NASH (37). However, there is significant heterogeneity among studies, and the importance of including biopsyproven healthy controls is underlined; most differences in adiponectin between individuals with NAFLD and controls were observed when the controls were not subjected to liver biopsy (37). In the discovery cohort, adiponectin presented a very promising performance in distinguishing NAFL vs. NASH. However, when tested in the larger validation cohort, adiponectin only presented a fair discriminatory power in this stratification when homozygous carriers of the PNPLA3 c.444C allele were considered. Importantly, the discriminatory power of adiponectin regarding NAFL vs. NASH patients was improved when combined with nine specific lipids. Previously, others have shown that combining adiponectin with homeostatic model assessment for insulin resistance and type IV collagen 7S increased the assay sensitivity from 68 to 94% in early-stage NASH prediction (38), while combining adiponectin, resistin, and cleaved cytokeratin-18 predicted NASH with an AUROC of 0.90 (39). Similar approaches combining adiponectin serum levels with a panel of serum lipids, mainly PC species, have been shown to discriminate between healthy individuals, NAFL, and NASH with high accuracy (40). Curiously, previous observations showed that NAFLD patients homozygous for the c.444G allele presented decreased adiponectin (41); however, the opposite was observed here. Nonetheless, these results could explain how this subgroup of NAFLD patients has an improved response to treatment and lifestyle interventions, including bariatric surgery,  when compared to CG and CC carriers (42). Additionally, we showed that lower levels of serum adiponectin associated with higher AST and ALT, thus suggesting a role of adiponectin in preventing liver damage. Overall, our observations suggest that adiponectin might be a valuable tool in NAFL vs. NASH stratification; however, its role as a biomarker in NAFLD can be challenged by risk-conferring genetic variants, such as the PNPLA3 c.444G allele, and can benefit from combination panels, such as the one presented comprising serum lipids. In fact, NAFLD has been associated with increased lipids in both liver tissue and plasma. Here we observed that higher circulating leptin associated with increased serum lipid profile and increased hepatic steatosis as assessed by histology and imaging. Similarly, higher circulating adiponectin associated with an increased serum lipid profile. The fact that adiponectin is inversely correlated with liver fat contents, as evaluated by MRI, might suggest a protective action for adiponectin by promoting lipid removal from the liver into circulation. For instance, higher adiponectin associated with higher serum PE containing arachidonic acid (C20:4n-6), while lower liver levels of this acid have been observed in NAFLD due to active conversion into pro-inflammatory prostaglandins (13). Furthermore, increased serum PE containing DHA also followed higher adiponectin. DHA, whose lower levels have been associated with both NAFL and NASH, is believed to have anti-inflammatory and metabolic effects and to be able to lower liver TG (13,14,43,44). This might also explain why higher adiponectin inversely correlates with TG as measured by blood biochemistry and serum metabolomics.
In turn, IGF-1, mainly produced in the liver in response to growth hormone stimulation, is known to regulate insulin sensitivity and decrease hepatic TG accumulation in the liver (45). While a few clinical studies have associated low serum IGF-1 with increased severity of steatosis (46,47) or lobular inflammation (48), our results and those of others (49) failed to show any association. On the other hand, clinical studies are consensual when reporting low serum IGF-1 association with the severity of fibrosis (46,(48)(49)(50). In fact, in vitro and in vivo studies showed that IGF-1 can induce cellular senescence and hepatic stellate cell inactivation through a p53-dependent manner, thus improving experimental fibrosis (51). Accordingly, here we showed that patients with NAFLD and advanced fibrosis presented significantly lower serum IGF-1. More importantly, we found that the 63% accuracy of serum IGF-1 alone in distinguishing low/mild fibrosis from advanced fibrosis is further improved to 93% if IGF-1 is combined with ferritin and INR. These results are very similar to those determined for the Enhanced Liver Fibrosis (ELF) test, recommended by the NICE guidelines for the evaluation of fibrosis in NAFLD (9). A poor PPV value has still been reported for the generality of fibrosis tests, including ELF (52). The panel here proposed displayed high PPV value, being matched by the recent Hepamet fibrosis scoring system where a PPV of 76% was achieved (53). Recent reports suggest that IGF-1 biodisponibility might be decreased during disease progression toward NASH and liver fibrosis and advance IGF-1/intact IGFBP3 ratio as a fibrosis predictor (54). Moreover, IGF-1 might associate with a fibrosis-specific serum metabolite profile, where many bile acids were found increased in advanced fibrosis, namely, cholic acid, which also correlated with serum IGF-1. Increased levels of many bile acids, including cholic acid, have already been associated with liver fibrosis (55).
Inclusion of a patient group comprising lean-NAFLD would allow further inquiries on adipokines as NAFLD biomarkers, particularly for leptin. It is also recognized that the number of healthy non-obese controls is relatively small, mostly due to the difficulty in obtaining biopsies from healthy-liver individuals. Lastly, the performance of the proposed biomarker panels comprising serum adiponectin and lipids for NAFL vs. NASH stratification and serum IGF-1 with INR and ferritin for identification of advanced fibrosis should be prospectively tested in a different cohort.

CONCLUSION
In conclusion, we suggest that serum leptin can identify NAFLD without obesity as a confounding factor, whereas adiponectin combined with specific serum lipids might be advantageous for NAFL vs. NASH stratification. In turn, IGF-1, together with ferritin and INR, could embody a valuable biomarker panel to identify advanced fibrosis, which could be easily implemented in clinics and presents a better PPV than the regularly used algorithms. Serum metabolomics correlate with hormone levels and offer new opportunities for improved noninvasive biomarker panels.

DATA AVAILABILITY STATEMENT
The data that support the current study are available within the article/Supplementary Materials, and from the corresponding author upon reasonable request.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by sample collection was performed after patient informed consent and Institutional Review Board approval, in accordance with the Declaration of Helsinki, by the ethics committees of Hospital de Santa Maria (Lisbon, Portugal) and Donostia University Hospital (San Sebastian, Spain). The patients/participants provided their written informed consent to participate in this study.

AUTHOR CONTRIBUTIONS
VM contributed to the methodology, validation, formal analysis, investigation, writing-original draft, writing-review and editing, and visualization. MA contributed to the supervision, methodology, writing-review and editing, and visualization. NB, MP, RL, AC, EA, and JA conducted the investigation. FD-R contributed to the formal analysis, writing-review, and editing. AS-L contributed to data curation, writing-review, and editing. RJ-A and EE contributed to data curation. LB, DT, and BF contributed to writing-review and editing. MM contributed to data curation. CA, MK, and FL contributed to the investigation and writing-review and editing. HC-P contributed to the resources, data curation, and writing-review and editing. JB contributed to the resources, data curation, supervision, funding acquisition, writing-review, and editing. RC contributed to writing-review and editing. AN contributed to the conceptualization, resources, writing-review, and editing. CR contributed to the supervision, conceptualization, funding acquisition, resources, writing-review, and editing. All the authors read and approved the final manuscript.