To the best of our knowledge, this is the first study based on MRI-PDFF and liver biopsy to investigate the clinical, laboratory and genetic features of the MAFLD population. Due to the inadequate diagnosis of NASH, insufficient knowledge of MAFLD and greater genetic predisposition to MAFLD, these conditions need to be further assessed.[31] It has been reported that steatosis is associated with significantly increased overall mortality, and this risk progressively increases with worsening MAFLD histology.[8] MRI-PDFF can accurately assess the steatosis degree of the liver, which was also demonstrated in our study; thus, it is necessary to evaluate the status of NAFLD patients by MRI-PDFF. However, this technique has not been commonly applied in clinical services in China. Here, this cross-sectional study investigated the clinical characteristics and noninvasive serum biomarkers, including potential risk SNPs, of MAFLD patients based on MRI-PDFF and liver biopsy.
In this study, we found that the incidences of MS and DM were higher in the MAFLD group and specifically were highest in the moderate/severe MAFLD group; in addition, most of the metabolic-related factors (insulin, HOMA, TG, ferritin, and FGF21) were significantly higher in MAFLD. Even between the mild MAFLD and the healthy groups, whose distinction was smaller (healthy was defined as LFC ~ 5.0%, and mild MAFLD was defined as LFC 5.1%-14.1%), correlation analysis showed that insulin, TG and FGF21 had strong positive correlations with LFC, suggesting that the glucose and lipid metabolic pathways might be the focus of disease aggravation and emphasizing the need for renaming NAFLD as MAFLD. Moreover, CK18, which indicates the degree of hepatocyte apoptosis,[32] has the strongest correlation with LFC among the serum biomarkers (R = 0.6, P < 0.05), suggesting the need to evaluate CK18 in long-term liver status monitoring.
However, we did not find a significant difference in FGF19 among the three groups; even between the moderate/severe group and the healthy group, FGF19 remained similar between the NASH and non-NASH groups. As a marker related to bile acid metabolism,[33] FGF19 was found to be obviously lower in the NASH group in another study,[34] whereas in our study, FGF19 only showed a significant difference between non-obese and obese MAFLD. Adiponectin and leptin also exhibited similar results. Thus, these indicators similar between different degree of liver steatosis may not be an efficacy parameter to reflect treatment effects in NASH clinical trials in Chinese population. Inflammatory markers have been demonstrated to play a pivotal role in the pathogenesis of experimental steatohepatitis[35] and can also reflect the disease status. IP-10 and IL-6 levels showed differences between different MAFLD statuses, and a great difference was also found between the mild MAFLD and healthy groups, indicating that chronic low-grade inflammation exists in patients with mild MAFLD.
To verify the diagnostic value of the MAFLD molecular prediction model (FLI, HSI, APRI and FAST), MRI-PDFF serves as the main diagnostic standard in testing prediction ability. The predictive ability of FAST, which is calculated based on AST, CAP, and LSM, has not been identified in China based on MRI-PDFF since it has only been recently reported.[36] After testing in test subsets, FAST performed better in predicting steatosis among the four reported models. Machine learning-based approaches were used to address the issue of a high-dimensional small dataset.[29, 30, 37] When predictive abilities of the reported and developed models were tested using the same test subset, their performances were similar and better than that of the previous model. These models (LASSO, elastic net, and logistic) were capable of accurately establishing the relationships between our analysed features and MAFLD, but they still need further investigation in other MAFLD cohorts.
Decision tree-based models were used to identify predictors of MAFLD, not only to differentiate healthy (LFC < 5%) and MAFLD (LFC > 5%) but also to differentiate mild MAFLD (LFC ≤ 14%) and moderate/severe MAFLD (LFC > 14%). An increased LFC (MRI-PDFF ≥ 15%) is associated with increased odds of fibrosis progression in patients with NAFLD at an early stage of fibrosis.[10] In addition, regression tree could reveal strong predictors of LFC. Compared with generalized linear models, the decision tree model is easier to understand because the results exported in the decision tree model resemble clinical decision-making processes and the tree structures are more flexible for distributing the response variable without preassumptions.[38] Our tree models provided strong evidence that ALT, AST, TC, CAP, CK18, and insulin were important predictors of MAFLD.
It has been demonstrated that approximately 44.3% of MAFLD patients are nonobese in China, which is similar to our result (49%).[39] Wong et al.[40] found similar clinical characteristics between nonobese and obese MAFLD; likewise, in our study, LFCs were comparable between obese and nonobese patients. In addition, more severe tissue inflammation was found in obese Chinese patients, as indicated by higher IP-10 and IL-6 levels. A study using liver biopsies or FibroScan also showed that obese NAFLD patients are more likely to develop advanced liver complications than are nonobese patients in China (AF 31.6% versus 6.3% and HCC 0.9% versus 0%).[7] Further follow-up studies could concentrate on comparing the different outcomes of nonobese and obese patients.
Genetic factors may be important in the development of MAFLD in either obese or nonobese patients.[41] The SNP rs738409 of PNPLA3 has been most widely reported as a hereditary risk factor associated with MAFLD. Other SNPs also contribute to the development of NAFLD (such as HSD17B13 and MBOAT7).[42] However, as the gold standard for MAFLD, histology is limited as a quantitative analysis of liver steatosis. MRI-PDFF, which can quantify liver steatosis, uncovers minor changes in liver steatosis.
In this study, based on MRI-PDFF, a significantly higher LFC was found in patients with the PNPLA3 G allele; furthermore, subjects with the PNPLA3 G allele showed higher concentrations of biomarkers related to liver injury and tissue inflammation. Among nonobese subjects, people with the MBOAT7 T allele had a significantly higher LFC, and thus, the MBOAT7 T allele was identified as a risk factor in nonobese subjects. The HSD17B13 rs72613567 variant allele was associated with moderate/severe MAFLD in both the nonobese and whole cohorts.
UQCC1 plays a role in mitochondrial respiratory chain complex III protein expression and is structurally similar to the mouse Bfzb controlling mouse brown fat.[43, 44] Previous associations at the UQCC1 rs4911494 locus included arm fat reduction,[45] reduced height, reduced body weight, and increased WHRadjBMI.[46] In our study, UQCC1 rs878639 (A > G) was identified as having a protective role against MAFLD, and people with the variant allele showed lower LFC, CAP and UA, which has not been found in other studies, while apoptosis and inflammation markers were not found to be different. Higher GRSs showed more severe disease status, suggesting the role of genetic factors in MAFLD.
4.1 Strengths and limitations
The data were collected using standardized protocols, which add validity to our study’s results. To the best of our knowledge, this is the first study to investigate the clinical characteristics and serum biomarkers (including gene polymorphisms) of MAFLD patients in a Chinese population based on MRI-PDFF. As a foundation, this study first reported a MAFLD diagnostic model that uses a machine learning method based on MRI-PDFF. The small sample size is a main limitation in the study. The cross-sectional design of the study may represent a limitation. In cross-sectional analyses, reverse causation cannot be excluded. Further investigation needs to be conducted in follow-up studies.