figure a

Introduction

Non-alcoholic fatty liver disease (NAFLD) is the most common cause of chronic liver disease worldwide and affects approximately 25% of the population [1]. In persons with NAFLD, the liver disease- and cardiovascular disease (CVD)-specific mortality rates are 0.77 and 4.79, respectively, per 1000 person-years [1]. NAFLD is closely associated with obesity, the metabolic syndrome, dyslipidaemia and type 2 diabetes [2]. However, a proportion of individuals develop NAFLD in the absence of obesity [3] and susceptibility to NAFLD clearly varies in part due to genetic polymorphisms [4, 5].

The rs738409 C>G variant of the gene encoding patatin-like phospholipase domain containing 3 (PNPLA3) protein is the first and strongest common variant that modifies genetic susceptibility to NAFLD [6]. The rs738409 C>G variant has been fully demonstrated to be a strong determinant of liver fat deposition independent of age [7], ethnicity [8], features of the metabolic syndrome and other risk factors for steatosis [6]. In several meta-analyses, this gene variant was shown to increase the risk of non-alcoholic steatohepatitis (NASH), fibrosis, cirrhosis and hepatocellular carcinoma [9]. However, the effect of PNPLA3 gene variants on diabetes is still uncertain. Most studies have indicated that PNPLA3 C>G variant carriers with NAFLD are not predisposed to insulin resistance [10] and type 2 diabetes [11]. An inverse correlation between the PNPLA3 G variant and the risk of diabetes was found in the NASH Clinical Research Network cohort [11]. In one large-scale cohort study in Germany, the PNPLA3 rs738409 G variant was found to be associated with increased liver disease-associated mortality but reduced coronary heart disease-associated mortality [12]. Adiposity can amplify the effect of the PNPLA3 rs738409 C>G variant on liver fat accumulation according to recent studies [13,14,15]. Weight gain is the most important risk factor for both NAFLD [16] and type 2 diabetes [17]. Weight loss is the mainstay for NAFLD treatment [18] and is an approved effective method to prevent diabetes [19]. Thus, we can obtain a better understanding of the role of the PNPLA3 C>G variant in the natural progression and remission of NAFLD and diabetes by investigating changes in the liver fat content (LFC) and metabolic variables in relation to changes in body weight in different PNPLA3 genotype carriers.

In this study, we prospectively monitored the natural course of changes in body weight and investigated their impact on the LFC and glucose and lipid metabolic variables in 2189 middle-aged and elderly individuals with different PNPLA3 genotypes from the Shanghai Changfeng Study [20].

Methods

Participants

The Shanghai Changfeng Community Study recruited 4300 Chinese community residents (1652 men, 2648 women) aged >45 years from May 2010 to December 2012. Participants with medical records or self-reports of previous viral hepatitis, excessive alcohol consumption and chronic liver diseases other than NAFLD were excluded [14]. In the current study, we report the first-round follow-up results of the Changfeng large-scale community study. Among the 4300 participants at baseline, a total of 156 participants died before November 2014 according to registration data from the Shanghai Center for Disease Control. The remaining 4144 participants were invited to attend the first-round follow-up appointment by phone call, letter or email from November 2014 to March 2017. Finally, 3262 (78.7%) participants responded to the invitation. After excluding 685 participants being treated with lipid-lowering medication, 232 with hypoglycaemic medication, 75 with a combination of hypoglycaemic and lipid-lowering medication and 81 with hepatoprotectants (e.g. polyene phosphatidylcholine, silymarin, adenosylmethionine, reduced glutathione), 2189 participants (889 men and 1300 women) with an average follow-up period of 4.2 years were included in the analysis (see electronic supplementary material [ESM] Fig. 1). The baseline characteristics of the enrolled 2189 participants were similar to those of the 4300 overall participants and could well represent the whole community population (ESM Table 1). The study was approved by the Research Ethics Committees of the Shanghai Health Bureau, China, and each participant provided written informed consent.

Anthropometric and serum biochemical measurements

A uniform questionnaire about the past history of diabetes, medications, smoking status and alcohol consumption for each participant was completed in a face-to-face interview. Body height and weight were measured with the participants wearing no shoes or outer clothing. The BMI was calculated by dividing the weight (kg) by the square of the height (m2). Waist circumference was measured using a soft tape at the midpoint between the lowest rib and the iliac crest in a standing position. For blood pressure, the mean of three resting measurements was used for the analysis. After at least a 12 h overnight fast, a venous blood sample was collected for the biochemical examinations. The serum total cholesterol, HDL-cholesterol and triacylglycerol levels were measured by an oxidase method and alanine aminotransferase (ALT) and aspartate aminotransferase (AST) were measured by the ultraviolet lactate and malate dehydrogenase methods, respectively, on a model 7600 automated bio-analyzer (Hitachi, Tokyo, Japan). LDL-cholesterol was calculated using the Friedewald equation [21]. All participants underwent a 75 g OGTT. The fasting and OGTT 2 h post-load blood glucose (PBG) concentrations were measured using the glucose oxidase method. An electrochemiluminescence immunoassay was used to measure the serum insulin concentrations. HOMA-IR was calculated by multiplying the fasting blood glucose (FBG) (mmol/l) by fasting insulin (pmol/l) and dividing by 156.3.

PNPLA3 genotype

The PNPLA3 rs738409 C/G variants were genotyped at the baseline examination using primer extension of multiplex products with detection by matrix-assisted laser desorption/ionisation time-of-flight mass spectrometry using the MassARRAY platform (MassARRAY Compact Analyzer; Sequenom, San Diego, CA, USA).

Determination of LFC

LFC was determined using a quantitative ultrasound method. Trained ultrasonographers who were unaware of the clinical data performed the ultrasound examinations. Ultrasound images were captured using the GE Logiq P5 scanner (GE Healthcare, Milwaukee, WI, USA), analysed using imaging software (ImageJ 1.41o; National Institutes of Health, Bethesda, MD, USA) and standardised using a tissue-mimicking phantom (Model 057; Computerized Imaging Reference Systems, Norfolk, VA, USA). The LFC was calculated using the following equation: LFC (%) = (62.592 × standardised US hepatic:renal ratio) + (168.076 × standardised US hepatic attenuation rate) − 27.863 [22].

Diagnosis of diabetes

For both the baseline and follow-up evaluations, diabetes was defined according to the 1999 WHO criteria [23] as follows: (1) FBG ≥7.0 mmol/l or (2) OGTT 2 h PBG ≥11.1 mmol/l.

Definition of metabolically healthy and unhealthy status

The waist circumference measurement and HOMA-IR have been used to define metabolic health status in previous epidemiological studies [24]. For the subgroup analyses in this study, we divided the participants into metabolically healthy and unhealthy groups according to a combination of the final HOMA-IR and waist circumference as follows: metabolically healthy group, HOMA-IR <2.5 and waist circumference <90 cm in men and <80 cm in women; metabolically unhealthy group, HOMA-IR ≥2.5 or waist circumference ≥90 cm in men and ≥80 cm in women.

Statistical analysis

All statistical analyses were performed using SPSS software (SPSS 18.0 software, SPSS Inc., Armonk, NY, USA; URL https://www.ibm.com/products/spss-statistics). The data are presented as the mean ± SD, except for skewed variables, which are presented as the median with the interquartile range (25–75%) reported in parentheses. Four-year changes in body weight were calculated by dividing the changes in body weight (in kg) by the follow-up times in 4 year periods. Participants were divided into quintiles based on the distribution of the 4 year changes in body weight during the follow-up: quintile 1, 4 year weight loss >2.6 kg; quintile 2, 4 year weight loss 0.9–2.6 kg; quintile 3, 4 year weight loss <0.9 kg or weight gain <0.5 kg; quintile 4, 4 year weight gain 0.5–2.1 kg; quintile 5, 4 year weight gain >2.1 kg. General linear models were used for comparisons of baseline continuous data among groups, whereas the linear-by-linear association χ2 test was used for comparisons of categorical variables. The distributions of the baseline LFC, triacylglycerol, insulin and HOMA-IR were highly skewed to the right and were log-transformed to approximate normality before entering into the general linear models. Our primary outcomes of interest were the hepatic outcomes (including changes in LFC and serum ALT) and glucose outcomes (including FBG and OGTT 2 h PBG); secondary outcomes included changes in other NAFLD-related metabolic variables (waist circumference, triacylglycerol, HDL-cholesterol and HOMA-IR). We examined the associations of changes in the LFC, ALT and metabolic variables with quintiles of the 4 year changes in body weight and different PNPLA3 genotypes using general linear models. Potential confounders considered in the multivariable models were age, sex, alcohol consumption, cigarette smoking, baseline body weight and baseline value of the investigated metabolic variable. Changes in LFC were further adjusted in the multivariable models on the associations between PNPLA3 genotypes and changes in glucose and lipid metabolic variables. Furthermore, we evaluated the interactions between changes in body weight (as an ordered categorical variable) and the PNPLA3 genotype and their effects on changes in the LFC, plasma glucose and lipid variables by inclusion of interaction terms into the models. Given the fact that the primary outcomes in the study included both hepatic and glucose outcomes, four tests (changes in LFC, ALT, FBG and OGTT 2 h PBG) were controlled in the primary interaction analyses. The interaction of PNPLA3 genotype and body weight change on other metabolic outcomes was evaluated as exploratory tests. False discovery rate (FDR) was used to adjust for multiple testing of the primary interaction analyses and the associations of PNPLA3 genotypes and quintiles of body weight change with changes in LFC and all metabolic variables [25]. A correlation coefficient heatmap was used to depict the correlations between changes in the LFC and changes in various metabolic variables in participants with different PNPLA3 genotypes. In the secondary analyses, logistic regression models were used to estimate ORs for incident diabetes in 1741 non-diabetic participants with different PNPLA3 genotypes. Age, sex, alcohol consumption, cigarette smoking, baseline body weight and LFC and changes in body weight and LFC were adjusted in the multivariable logistic regression models. Since our previous cross-sectional study indicated that PNPLA3 gene variants increased the risk of NAFLD in a metabolic factor-dependent manner [14], subgroup analyses were performed in participants with different metabolic health statuses. Values of p < 0.05 were considered statistically significant for all analyses.

Results

Baseline characteristics and changes in body weight and metabolic variables

A total of 889 men and 1300 women were included in the study, with average age of 62.2 years, BMI 24.1 kg/m2 and LFC 7.8%. There were 841 (38.4%) PNPLA3 CC homozygotes, 1036 (47.3%) CG heterozygotes and 312 (14.3%) GG homozygotes. The baseline characteristics of the study participants across categories of PNPLA3 genotypes and changes in body weight are shown in ESM Table 2 and Table 1, respectively. The PNPLA3 G variant carriers showed significantly higher LFCs and serum AST levels but lower serum triacylglycerol levels at baseline. Compared with participants in the lower quintiles of body weight change, those in the higher quintiles had lower baseline body weights, LFCs, FBG, OGTT 2-h PBG, HOMA-IR, and serum ALT, AST and triacylglycerol levels and higher serum HDL-cholesterol levels (all p < 0.05). The frequencies of the PNPLA3 CC, CG, and GG genotypes were similar among the groups with different grades of body weight changes (p = 0.254, Table 1). Despite their initially better metabolic status when compared with participants in the lower quintiles of body weight change, those in the higher quintiles showed greater increases in the LFC, OGTT 2 h PBG, HOMA-IR and serum levels of ALT, triacylglycerol and LDL-cholesterol and a reduction in the HDL-cholesterol levels (ESM Table 3). Except for a lack of association between changes in body weight and changes in PBG and LDL-cholesterol in the PNPLA3 GG genotype carriers, an association between change in body weight and changes in the LFC and metabolic variables was found in all participants regardless of their PNPLA3 genotype, even after adjustment for age, sex, alcohol consumption, cigarette smoking, baseline body weight and baseline LFC or the investigated metabolic variable (ESM Table 3).

Table 1 Baseline characteristics of the study participants according to 4 year changes in body weight

Effect of the interaction between the PNPLA3 GG genotype and changes in body weight on NAFLD and liver enzymes

As shown in Fig. 1a, participants with the PNPLA3 GG genotype had significantly greater increases in the LFC than PNPLA3 CC homozygotes in quintiles 1, 2, 4 and 5 (all FDR-adjusted p < 0.05). For participants with decreases in body weight during the follow-up period, the LFC was significantly decreased in the PNPLA3 CC homozygotes but was still increased in the PNPLA3 GG homozygotes. The average change in the LFC of the CG heterozygotes was between that of the homozygote groups. An interaction between the changes in body weight and PNPLA3 polymorphisms was also observed (FDR-adjusted pinteraction = 0.044) (Fig. 1a).

Fig. 1
figure 1

Changes in LFC (a) and serum ALT (b) according to changes in body weight and PNPLA3 genotype. Data are presented as the mean ± SEM; *p < 0.05 (FDR adjusted) compared with PNPLA3 CC genotype carriers; p < 0.05 (FDR adjusted) compared with the PNPLA3 CG genotype carriers. PNPLA3 GG genotype carriers had significantly greater increases in the LFC in the first, second, fourth and fifth quintiles of body weight changes (a) and greater increases or smaller reductions in ALT than the PNPLA3 wild-type (CC) carriers in all quintiles (b). An interaction was found between changes in body weight and PNPLA3 polymorphisms that affected changes in the LFC (FDR-adjusted pinteraction = 0.044) (a) and serum ALT (FDR-adjusted pinteraction = 0.044) (b). Quintiles of change in body weight over 4 years: quintile 1 (Q1), weight loss >2.6 kg; quintile 2 (Q2), weight loss 0.9–2.6 kg; quintile 3 (Q3), weight loss <0.9 kg or weight gain <0.5 kg; quintile 4 (Q4), weight gain 0.5–2.1 kg; quintile 5 (Q5), weight gain >2.1 kg

Liver enzyme markers, especially ALT, are the most commonly used variables for evaluating liver inflammation. Participants with the PNPLA3 GG genotype had a greater increase or lesser reduction in their ALT levels than the PNPLA3 CC carriers for all participants with body weight changes (all FDR-adjusted p < 0.05) (Fig. 1b). The effect of the interaction between changes in body weight and the PNPLA3 genotype on serum ALT was statistically significant (FDR-adjusted pinteraction = 0.044) (Fig. 1b).

Effect of the interaction between the PNPLA3 GG genotype and changes in body weight on glucose metabolic variables

We also assessed the possible interactions between PNPLA3 genotypes and changes in body weight on other metabolic characteristics related to NAFLD. In participants with a body weight gain of 0.5–2.1 kg and >2.1 kg, the PNPLA3 GG genotype carriers showed a significantly lower increase in the OGTT 2 h PBG (Fig. 2b) than the PNPLA3 CC genotype carriers. The PNPLA3 CG genotype carriers with a body-weight gain >2.1 kg also showed a lower increase in the OGTT 2 h PBG than PNPLA3 CC genotype carriers. However, no significant differences were found in the changes in waist circumference, HOMA-IR, FBG, serum triacylglycerol or HDL-cholesterol among the PNPLA3 CC, CG, and GG genotype carriers with the same degree of body weight changes (Fig. 2a and ESM Fig. 2a–d). No interactions between changes in body weight and the PNPLA3 genotype were identified for changes in the FBG, OGTT 2 h PBG or other lipid metabolic traits.

Fig. 2
figure 2

Changes in FBG (a) and OGTT 2 h PBG (b) according to changes in body weight and the PNPLA3 genotype. Data are presented as the mean ± SEM; *p < 0.05 (FDR adjusted) compared with PNPLA3 CC genotype carriers. PNPLA3 GG genotype carriers in the fourth and fifth quintiles and CG genoytpe carriers in the fifth quintile had significantly smaller increases in OGTT 2 h PBG compared with PNPLA3 wild-type (CC) carriers (b). There were no interactions between changes in body weight and PNPLA3 genotypes that affected blood glucose. Quintiles of change in body weight over 4 years: quintile 1 (Q1), weight loss >2.6 kg; quintile 2 (Q2), weight loss 0.9–2.6 kg; quintile 3 (Q3), weight loss <0.9 kg or weight gain <0.5 kg; quintile 4 (Q4), weight gain 0.5–2.1 kg; quintile 5 (Q5), weight gain >2.1 kg

Among the 1741 participants without diabetes at baseline, the rate of incident diabetes diagnosed by both FBG and the OGTT 2 h PBG was 8.9%, 7.1% and 6.7% for the PNPLA3 CC, CG and GG genotype carriers, respectively (Table 2). The incident diabetes rate diagnosed by the OGTT 2 h PBG alone was 3.8% in the PNPLA3 GG genotype carriers, only half that in the PNPLA3 CC genotype carriers. The multivariable logistic regression analyses showed that the multivariate-adjusted ORs were 0.509 (0.260, 0.998) for incident diabetes diagnosed by FBG and OGTT 2 h PBG and 0.335 (0.149, 0.751) for incident diabetes diagnosed by OGTT 2 h PBG alone in the PNPLA3 GG genotype carriers (Table 2).

Table 2 ORs (95% CIs) for incident diabetes in participants with different PNPLA3 genotypes

The PNPLA3 GG genotype dissociates changes in the LFC and the OGTT 2 h PBG

Changes in the LFC were positively associated with changes in body weight, waist circumference, HOMA-IR and serum ALT, AST and FBG in all participants regardless of their PNPLA3 genotype. However, the changes in LFC were significantly correlated with changes in the OGTT 2 h PBG and triacylglycerol only in participants with the PNPLA3 CC and CG genotype, and not in those with the PNPLA3 GG genotypes (Fig. 3).

Fig. 3
figure 3

Correlation coefficient heatmap for the correlations between changes in the LFC and changes in metabolic variables among participants with different PNPLA3 genotypes. Changes in the LFC were significantly correlated with changes in the OGTT 2 h PBG and triacylglycerol in participants with the PNPLA3 CC and CG genotypes but not those with the PNPLA3 GG genotype. *p < 0.05, **p < 0.01 and ***p < 0.001. HDL-c, HDL-cholesterol; LDL-c, LDL-cholesterol; TC, total cholesterol; WC, waist circumference

Subgroup analyses of participants with differing metabolic health status

We divided all participants into categories according to metabolic health (healthy vs unhealthy) based on their final HOMA-IR and final waist circumference. In the metabolically healthy participants with a final HOMA-IR <2.5 and no abdominal obesity, no significant differences were found in the changes in the LFC and OGTT 2 h PBG among the different PNPLA3 genotype groups displaying the same degree of changes in body weight (Fig. 4a, b). However, in the metabolically unhealthy participants with a final HOMA-IR ≥2.5 or abdominal obesity, PNPLA3 GG genotype carriers showed a significantly greater increase in the LFC in the first, second, fourth and fifth quintiles of body weight changes (all FDR-adjusted p < 0.05) but smaller increase in the OGTT 2 h PBG with 4 year weight gain over 0.5 kg when compared with the PNPLA3 CC genotype carriers (Fig. 4a, b).

Fig. 4
figure 4

Changes in the LFC (a) and OGTT 2 h PBG (b) according to changes in body weight and the PNPLA3 genotype in subgroups with a metabolically healthy (final HOMA-IR <2.5 and waist circumference <90 cm in men and <80 cm in women) and unhealthy (final HOMA-IR ≥2.5 or waist circumference ≥90 cm in men and ≥80 cm in women) status. Data are presented as the mean ± SEM; *p < 0.05 (FDR adjusted) compared with the PNPLA3 CC genotype carriers. For metabolically healthy participants, the PNPLA3 GG genotype had no effect on changes in the LFC or PBG, whereas for metabolically unhealthy participants, the PNPLA3 GG genotype was associated with a greater increase in LFC in the first, second, fourth and fifth quintiles of weight change (a), but a smaller increase in the OGTT 2 h PBG in the fourth and fifth quintiles of weight change (b). Quintiles of change in body weight over 4 years: quintile 1 (Q1), weight loss >2.6 kg; quintile 2 (Q2), weight loss 0.9–2.6 kg; quintile 3 (Q3), weight loss <0.9 kg or weight gain <0.5 kg; quintile 4 (Q4), weight gain 0.5–2.1 kg; quintile 5 (Q5), weight gain >2.1 kg

Discussion

An interaction between the PNPLA3 genotype and obesity has been previously reported to affect liver steatosis and increase the serum ALT level. The major finding of this study is that the LFC and OGTT 2 h PBG in participants with different PNPLA3 genotypes respond differently to long-term spontaneous changes in body weight. In a cohort of middle-aged and elderly Chinese individuals from a community population from the Shanghai Changfeng Study, the PNPLA3 homozygous GG genotype dissociated the changes in the LFC and OGTT 2 h PBG. This variant was also associated with seemingly paradoxically greater increases in the LFC but less elevation of the OGTT 2 h PBG and diabetes incidence when compared with individuals having the PNPLA3 CC genotype and the same degree of weight gain. Furthermore, an interaction between change in body weight and PNPLA3 genotype on the change in LFC and serum ALT was discovered. Subgroup analyses indicated that the effect of the PNPLA3 GG genotype on changes in the LFC and PBG was significant in metabolically unhealthy participants but not in those who were metabolically healthy. Taken together, these results indicate that the PNPLA3 genotype plays an important role in influencing individual hepatic and glucose metabolic outcomes related to the natural course of body weight change.

Several recent studies support a close association between body weight changes and the risk of NAFLD [26], type 2 diabetes and CVD [27]. Our previous study in the same cohort also showed that changes in body weight could predict risk of impaired glucose regulation and diabetes [28]. In the current study, we further found that the PNPLA3 rs738409 C>G variant interacted with changes in body weight and influenced the clinical outcomes of NAFLD and diabetes. In fact, several previous studies also reported that individuals with NAFLD and the PNPLA3 GG genotype responded differently to weight-loss interventions when compared with PNPLA3 CC genotype carriers. Several previous studies have indicated that PNPLA3 GG genotype carriers benefit most from body weight reduction resulting from an intensive short-term low-energy low-carbohydrate diet [29], 12 month lifestyle intervention programme with >10% body weight reduction [30], or bariatric surgery with an average weight loss of 40 kg [31]. Other intervention studies showed that PNPLA3 GG genotype variant carriers displayed changes in the LFC similar to those seen in PNPLA3 CC homozygotes, with an average of 3 kg of weight loss over 6 months in a Finnish population [32] and a 5% reduction in body weight over 6 months in children with severe obesity [33]. The results from the Wessex Evaluation of Fatty Liver and Cardiovascular Markers in NAFLD with Omacor Therapy (WELCOME) trial even showed that the LFC was increased after docosahexaenoic acid and eicosapentaenoic acid treatment and weight loss in PNPLA3 GG homozygotes, in contrast to the changes seen in PNPLA3 CC and CG genotype carriers [34]. In our subgroup analysis, we found that the effect of the PNPLA3 G variant on the natural course of changes in the LFC was dependent on the individual’s final metabolic health status. In participants with central obesity or insulin resistance, the PNPLA3 GG genotype interacted with changes in body weight and aggregated liver steatosis, whereas in metabolically healthy participants, presence of the PNPLA3 GG genotype had no effect on the changes in the LFC.

The reasons for the discrepancies in the effect of PNPLA3 polymorphisms on the progression and remission of NAFLD under different metabolic conditions are not known. Possibly, the expression of harmful proteins resulting from PNPLA3 mutation is regulated by insulin and the nutritional status. PNPLA3 has been demonstrated to be directly regulated by the insulin-regulated transcription factor sterol regulatory element binding protein-1c (SREBP-1c), and pathogenic PNPLA3 C>G mutant products accumulate under conditions of insulin resistance or central obesity, thereby exacerbating liver steatosis and inflammation [35]. In accordance with our findings in the current study, previous laboratory studies found that the PNPLA3 G variant alone was insufficient to cause liver steatosis in chow-fed mice but elicited a two- to threefold increase in the risk of NAFLD in sucrose-fed mice [36]. The regulation of PNPLA3 expression by metabolic status also provides a theoretical foundation for the interaction between PNPLA3 gene variant and changes in body weight.

We found that participants with the PNPLA3 GG genotype who gained body weight displayed smaller increase in the OGTT 2 h PBG than the PNPLA3 CC genotype carriers despite baseline OGTT 2 h PBG and changes in FBG and HOMA-IR being similar among the different PNPLA3 genotype carriers. Several cross-sectional studies have shown that the PNPLA3 G variant is correlated with better insulin sensitivity [37, 38]. However, the beneficial effect of the PNPLA3 G variant on glucose metabolism was not able to be detected by the baseline FBG, OGTT 2 h PBG, HOMA-IR or even euglycaemic–hyperinsulinaemic clamp in all participants. Only in participants with a high percentage of body fat could a better insulin sensitivity status precisely measured by OGTT be found in PNPLA3 G variant carriers [37]. An interaction of PNPLA3 G variant with obesity on insulin sensitivity was reported previously [39]. This might explain our finding of a lower blood glucose level and diabetes incidence in PNPLA3 GG genotype carriers with a metabolically unhealthy status, when compared with the PNPLA3 CC homozygotes with the same degree of body weight change. Since PNPLA3 expression levels are extremely low in a fasting status [35] and an increased PBG is better correlated with insulin resistance than the FBG [40], it is understandable that the PNPLA3 G variant has more influence on the OGTT 2 h PBG than the FBG. The mechanism underlying the beneficial effect of PNPLA3 gene variants on glucose metabolism may be related to changes in the liver lipid composition from saturated triacylglycerol to polyunsaturated triacylglycerol and a marked reduction in insulin-resistance-inducing ceramides [41]. One recent study indicated that PNPLA3 functions as a very-long-chain polyunsaturated fatty acid-specific triacylglycerol hydrolase, which promotes transfer of polyunsaturated fatty acids from triacylglycerol to phosphatidylcholine and that the PNPLA3 C>G variant causes an 80% reduction in PNPLA3 activity and a reduction in the ratio of saturated to polyunsaturated triacylglycerol in the liver [42].

To the best of our knowledge, our current study might be the first large-scale community population-based cohort study to evaluate the influence of PNPLA3 genotype on the association between natural changes in body weight and the progression or remission of NAFLD and metabolic complications. Several limitations are associated with our current study. First, information regarding diet and exercise during the follow-up period was not recorded. Therefore, our results did not permit evaluation of the different hepatic responses to changes in weight for specific reasons among different PNPLA3 genotype carriers. Second, this study was performed in a Chinese cohort aged >45 years and the results need to be confirmed in participants with different ethnicities and in different age groups. Third, the LFC was quantified using a quantitative ultrasound method, which is not as accurate as a liver biopsy or proton magnetic resonance spectroscopy (1H-MRS). However, invasive liver biopsy and 1H-MRS cannot be carried out routinely in large-scale prospective population studies, and previous studies have indicated that the LFC measured by our quantitative ultrasound method is suitable for large-scale human studies and agrees well with LFC measurement determined by 1H-MRS (r = 0.85, p < 0.001) and the histological liver steatosis grades (r = 0.79, p < 0.001) [43].

In conclusion, the PNPLA3 GG genotype and its interaction with body weight change aggravates liver steatosis but protects against increased risk of incident diabetes. The interaction between the PNPLA3 GG genotype and changes in body weight on NAFLD is highly dependent on an individual’s metabolic status. Since no formal recommendations exist for the treatment of NAFLD associated with PNPLA3 gene variant at present, our data indicate the need for personalised treatment of NAFLD in those with the PNPLA3 rs738409 C>G variant.