INTRODUCTION

Globally, the prevalence of non-alcoholic fatty liver disease (NAFLD) is approximately 24% [1] and it is estimated that non-alcoholic steatohepatitis (NASH) prevalence will rise by 63% over the next 10–15 years [2]. Nowadays, NAFLD remains a challenge to researchers and clinicians due to its increased prevalence, complex pathophysiology, and lack of diagnostic and therapeutic approaches [3]. Among various diagnostic approaches, the procedure used for diagnosis and staging of NASH is liver biopsy, which is a so-called “gold standard”. However, it is an expensive, invasive, and painful procedure which may sometimes cause risk to the patients [4]. There are some non-invasive steatosis biomarkers, such as hepatic steatosis index (HSI), SteatoTest, and fatty liver index, whose diagnostic performance is accepted, but the ability to distinguish between the grades is not up to the required range [5,6,7]. The most common non-invasive fibrosis markers are NAFLD fibrosis score (NFS), AST-to-platelet ratio index (APRI), and fibrosis-4-index (FIB-4) [8,9,10], however, they are clinically not much utilized for diagnosis. To overcome these situations, there is an urgent need to develop circulatory non-invasive biomarkers that could reliably discriminate NASH from NAFL. Additionally, biomarkers can be applied for early prediction of the disease, which helps to prevent disease progression, identify the mechanisms involved in disease pathophysiology, develop novel therapeutic strategies and identify the novel therapeutic targets for improving personalised medicine.

NAFLD is characterized by intra-hepatic lipid accumulation (simple steatosis) which could further develop into non-alcoholic steatohepatitis (NASH) characterized by simple steatosis, hepatocyte ballooning, and lobular inflammation, and finally leads to fibrosis, cirrhosis, and carcinoma. The molecular mechanisms that drive simple steatosis, inflammation, fibrosis, and other progressive stages of NAFLD are insufficiently understood. Additionally, there is emerging evidence that NAFLD is associated with the alteration of immune responses and chronic low-grade inflammation [11, 12]. Inflammatory cytokines and chemokines upregulate Kupffer cells, activate hepatic stellate cells, and infiltrate immune cells [13]. Several studies have reported that inflammatory mediators such as cytokines, chemokines, proteins, oxygen free radicals, and metabolites are involved in the pathophysiology of NAFLD disease [14,15,16,17,18]. Generally, pro-inflammatory cytokines promote insulin resistance, production of reactive oxygen species, hepatic inflammation, and hepatic fibrosis [19, 20]. On the other hand, anti-inflammatory cytokines are responsible for protecting the liver from damage or injury by regulating these processes [21]. As hepatic inflammation is a hallmark of NAFLD progression, the identification of a specific panel of inflammatory markers to characterize NAFL and NASH is very important. However, currently, no serum inflammatory biomarker is available that could diagnose NASH as a distinct entity from NAFL. Therefore, the present study was designed to investigate an array of non-invasive disease-specific markers in NAFL and NASH patients.

METHODS

Study Design and Ethics Statement

The present study is an observational, cross-sectional study that is being conducted in accordance with the principles of the Declaration of Helsinki. This study was approved by the Institutional Ethics Committee of the Downtown Hospital, Guwahati [IEC/dth/2022/MS/001]. Clinical samples were collected from Downtown Hospital, Guwahati and the Institute of Liver and Biliary Sciences (ILBS) biobank, New Delhi. The study participants have shown their willingness to participate in the study by signing the written informed consent.

Study Participants

Male and female adults aged between 20 and 65 years were included in the study. Among the study participants, healthy individuals with no signs of fatty liver and normal levels of biochemical parameters were grouped under the control group (n = 64); patients with signs of fatty liver (simple steatosis) of different grades confirmed by ultrasound imaging were grouped under NAFL group (n = 109); and patients with simple steatosis and inflammation with or without fibrosis confirmed by liver-biopsy or fibroscan were grouped under NASH group (n = 60). NAFLD patients with or without metabolic diseases, with no history of alcohol consumption or having a limit of 30 g/day in males and 20 g/day in females were included in the study. The participants with other liver diseases, chronic diseases/complications, pregnancy, and history of abdominal surgery were excluded from the study.

Clinical and Laboratory Assessment

After enrolling the participants in the study, anthropometric data, including height (cms), weight (kg), body mass index (BMI), and waist circumference (WC) were measured. Socio-demographic data including age (yrs), gender, history of alcohol consumption and smoking were collected. The blood samples were collected from the study participants and processed for serum within two hours. The laboratory biochemical parameters such as aspartate aminotransferase (AST), alanine aminotransferase (ALT), alkaline phosphatase (ALP), blood glucose, total cholesterol (TC), low-density lipoprotein (LDL), high-density lipoprotein (HDL), triglycerides (TG’s), urea, creatinine, uric acid, and blood urea nitrogen (BUN) were assessed using an automated biochemistry analyzer (Randox, RX Daytona +). Additionally, non-invasive biomarkers such as APRI [22], AST/ALT ratio, HSI [6], FIB-4 score [8], and NFS [23] were calculated using anthropometric data, socio-demographic data, and biochemical parameters. The specific formulae to calculate non-invasive biomarkers were given in the Supplementary file.

Measurement of Serum Cytokines and Chemokines

Prior to the experiment, the frozen serum samples were thawed and centrifuged at 1,000 × g for 15 min at 4 °C to remove particulates from all the samples. The serum samples were diluted with sample diluent in the ratio of 1:4. Bio-Plex Pro Human Cytokine 27-Plex Panel (#M500KCAF0Y) was used which includes FGF (fibroblast growth factor) basic, Eotaxin, G-CSF (Granulocyte colony-stimulating factor), GM-CSF (Granulocyte–macrophage colony-stimulating factor), IFN-γ (Interferon-gamma), IL (Interleukin)-1β, IL-1ra, IL-2, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IL-10, IL-12, IL-13, IL-15, IL-17A, IP (Interferon-gamma inducible protein)-10, MCP-1 (Monocyte chemoattractant protein-1), MIP-1a (Macrophage inflammatory protein-1 alpha), MIP-1b, PDGF-BB (Platelet-Derived Growth Factor BB), RANTES (regulated upon activation, normal T cell), TNF-α (Tumor necrosis factor alpha), and VEGF (Vascular endothelial growth factor) [To refer all these analytes, we used the global term “cytokines” in the entire manuscript]. The measurement of cytokines was done in serum samples of all study participants using a Bio-Plex® 200 System (Bio-Rad). The whole experiment was performed according to the manufacturer’s instructions. The standards and blank samples were run in duplicate, whereas, the serum samples were run in single. Out of 27 cytokines, 10 cytokines (FGF, IFN-γ, IL-5, IL-6, IL-7, IL-10, IL-12, IL-15, IL-17A, and VEGF) were excluded as they did not show the quantification in 90% of the samples.

Statistical Analysis

Initially, the Kolmogorov–Smirnov (KS) test was used to check the data for normality. Descriptive statistics of normally distributed data were represented as mean ± SD, whereas, non-normally distributed data were represented as median (interquartile range). In order to compare the variables among study groups, ordinary one-way ANOVA followed by Tukey’s multiple comparison tests was applied for normally distributed data, whereas Kruskal-Walli’s test, followed by Dunn’s post-hoc test for multiple comparisons, was applied for non-normally distributed data.

The principal coordinate analysis (PCoA) was performed on the dissimilarity matrix and then the standard deviation of distances between the samples was calculated; a threshold was set to identify and remove the outliers (s.class function to visualize PCoA results). Next, the correlation was performed between the values of the X-axis and Y-axis coordinates of the PCoA plot and each cytokine profile. Volcano plots are generated to visualize the significant cytokine markers. Additionally, rank-scaling for each study group was done and the values of all the cytokines were ranked 0–1. The heatmap is generated to visualize the rank-scaled data of the markers across three different groups (heatmap.2 function). In order to obtain a fold change, the cytokine value of each patient was calculated as a log-ratio value and then divided with the control median which was taken as a standard for the corresponding cytokine marker. To understand how each biochemical parameter influences cytokine levels, we performed a linear regression model by adjusting age and BMI. p-value < 0.05 was considered statistically significant, whereas Benjamini–Hochberg correction was applied to adjust p-values that account for the increased risk of false positives (q-value < 0.15). To perform correlation analysis, spearman correlation and false discovery rate (FDR) correction methods were used. A binary matrix was generated by thresholding q-values using a significance threshold of 0.05 and then considering those variables for network analysis. Network analysis was conducted using Cytoscape software to elucidate the relationships among clinical and biochemical characteristics, cytokine markers, and non-invasive markers. Each marker was represented as a node, and significant correlations were depicted as edges connecting the nodes.

The Random Forest approach was then utilized for computing the overall accuracy for discriminating between the three groups based on the cytokine profiles. Two different investigations were performed using the random forest model. In the first investigation, we performed 50 iterations to identify the most diagnostic cytokine markers facilitating the three-group classification. In each iteration, for each group (control, NAFL, and NASH), the data was split randomly into training (70%) and testing (30%) subsets. A random forest classification model was developed by training on 70% of the data and testing on the rest 30% of the data. A confusion matrix was created to check whether these predictions were correct or not. The feature importance was calculated by determining each variable contribution for differentiating between the three groups. The mean feature importance scores of all variables across the 50 iterations were then computed for ranking the cytokines based on their power/performance in discriminating across the three groups. In the second investigation, we adopted a similar approach, but this time, we considered each pair of groups separately. Based on these feature’s importance, a venn diagram was created to illustrate the top cytokine markers extracted from each pairwise group comparison (https://csbg.cnb.csic.es/BioinfoGP/venny.html). Network analysis was performed using the STRING database, Version 12.0. We considered the disease-specific networks based on a high confidence threshold (0.700) and < 1% of the FDR threshold.

The packages used for this analysis are randomForest (for building the Random Forest classification model), ggplot2 (for creating visualizations), Base R (for data manipulation and confusion matrix generation), gplot (for generating heatmaps), gridExtra (for arranging multiple plots into a grid), psych (to compute the correlation matrix).

RESULTS

Clinical and Biochemical Characteristics of Study Participants

A total of 233 participants were included in the study based on inclusion and exclusion criteria, of which 64 were in the control group, 109 in the NAFL group, and 60 in the NASH group. There were statistically significant differences in age, BMI, and biochemical characteristics, except for gender, alcohol consumption and smoking habits among the study groups. The levels of biochemical parameters such as glucose, AST, ALT, and ALP are increased in NASH patients, whereas, TC, TGs, and LDL are increased in NAFL patients when compared to the corresponding two groups. The detailed information on the clinical and biochemical characteristics of study participants was represented in Table 1.

Table 1 Clinical and Biochemical Characteristics of Control, NAFL, and NASH Subjects
Full size table

Measurement of Non-Invasive Markers

Non-invasive markers such as APRI, AST/ALT, HSI, FIB-4, and NFS were found to be statistically significant among the study groups (Supplementary file). It was found that APRI, HSI, FIB-4, and NFS scores were higher in patients who were included in the NASH group when compared to control and NAFL groups.

Distinct Cytokine Profiles Between the Study Groups

We first used the PCoA method to investigate whether the control, NAFL, and NASH study groups had distinct profiles based on the serum concentration of cytokines. These results demonstrated statistically significant differences in cytokines/chemokines profiles between each of the study groups (PERMANOVA p = 0.001; R2 = 0.102) (Fig. 1a). The separation between the groups was significantly pronounced along both PCoA axes (PC1 or X-Axis and PC2 or Y-Axis). For the PC1, we observed a progressive significant increase following pattern: Control < NAFL < NASH (Control v/s NAFL: p = 0.009; Control v/s NASH: p = 0.00001; NAFL v/s NASH: p = 0.003) (Fig. 1b). To identify the markers associated with this gradual progression along this disease axis, we then performed a correlation analysis of each cytokine with PC1. The results shown as a volcano plot (Fig. 1c) indicate that all the cytokine markers have significant positive associations, indicating that all the cytokines gradually increase across the different disease stages. The only exceptions were eotaxin and IP-1. In contrast, we observed the following pattern of progressive decrease of PC2: Control > NAFL > NASH (Control v/s NAFL: p = 0.005; Control v/s NASH: p = 0.00001; NAFL v/s NASH: p = 0.001) (Fig. 1d). IL-2, IL-8, IL-13, IL-1b, TNF-α, MIP-1a, MIP-1b, G-CSF, RANTES, and eotaxin were observed to show significant negative correlations with PC2, indicating their progressive increase with different disease stages (in line with our findings in Fig. 1c) (Fig. 1e).

Fig. 1
figure 1

a Principal component analysis (PCoA) plot representing the variation of cytokines profiles among study groups [p = 0.001; R2 = 0.102]. b and d Box-plot of cytokines profile variations within and between the study groups based on PC1 and PC2 coordinates, respectively. Dunn’s test was applied to check the statistical significance between the study groups. **p ≤ 0.009 and ***p < 0.0009 indicates statistical significance. c and e Volcano plots of cytokines profile. Red dotted cytokines above the red line are significant.

Full size image

Serum Cytokines Concentration in Study Participants

We then compared the concentration of serum cytokines in control, NAFL, and NASH study participants Table 2. It was found that all the cytokines, except IL-4, were significantly different among the study groups. We observed significantly higher levels of TNF-α, IL-1β, IL-1ra, G-CSF, PDGF-BB, MCP-1, MIP-1a, MIP-1b, RANTES, eotaxin, IL-8 and IP-10 in NASH subjects when compared to control subjects. On the other hand, the levels of IL-9 and IL-13 were found to be significantly decreased in NASH patients. Moreover, IL-2 levels were significantly decreased in NAFL subjects when compared to control subjects. There were no statistical differences observed in IL-1b, IL-8, MCP-1, MIP-1a, MIP-1b, and PDGF-BB concentrations between NAFL and NASH groups.

Table 2 Serum Concentration of Cytokines in Control, NAFL, and NASH Subjects
Full size table

Figure 2 shows the ranked measured levels of the 16 serum cytokines across all the individuals of control, NAFL, and NASH study groups, as well as the median abundances of the same in each of the three groups. Our data showed that the median fold change values of serum IL-2, IL-8, IL-13, IL-1b, TNF-α, MIP-1a, MIP-1b, G-CSF, RANTES, and eotaxin are significant among NAFL and NASH groups when compared to the control group. However, there were no statistically significant differences in median fold change values of IL-4, IL-9, IP-10, IL-1ra, PDGF-BB, and MCP-1.

Fig. 2
figure 2

Heatmap representing the fold change of various cytokines among the study groups. The median fold change of cytokine levels in NAFL and NASH groups with respect to control was also shown.

Full size image

Feature Importance of Biochemical Parameters and their Association with Distinct Cytokines

Random forest models were first utilized in an iterative manner (See Methods) to identify the diagnostic cytokine markers facilitating the three-group classification. Herein, the study group was considered as the dependent variable (response) and the cytokines/chemokines were considered as predictors. The confusion matrix was created to check the robustness of the model’s performance in distinguishing control, NAFL, and NASH classes. Our results indicated that the model correctly classified 9 positive class data points as control, 13 instances as NAFL, and 11 instances as NASH for class 1 (control), class 2 (NAFL), and class 3 (NASH) respectively (Fig. 3a). Our investigation further revealed ALT as the most important diagnostic feature followed by AST, TP, ALP, creatinine, TGs, glucose, urea, LDL, albumin, HDL, BUN, TC, and uric acid for the three-group classification (Fig. 3b).

Fig. 3
figure 3

a Confusion matrix was created to evaluate the model’s performance. The rows indicate the “actual” class and the columns indicate the “predicted” class. All the values in the diagonal represent “true positives”. b Feature importance plot of biochemical characteristics based on variable importance score. c Linear regression analysis was performed to show the association of biochemical characteristics with cytokine markers.

Full size image

To further probe this aspect, we also applied a linear regression model after adjusting for age and BMI to investigate the strength of association and direction of relationships between biochemical parameters and markers profile (Refer to the heatmap in Fig. 3c). In case of liver function parameters, a positive significant association (FDR corrected q-value ≤ 0.15) was observed between AST and RANTES, IP-10, MCP-1, IL-1ra, G-CSF; ALT and RANTES, IP-10, MCP-1, IL-1ra, G-CSF; ALP and IP-10, G-CSF. A similar association was also observed between IL-1ra and glucose, creatinine, BUN, urea; total protein and IL-9, RANTES; albumin and IL-9. Furthermore, a significant negative association (FDR corrected q-value ≤ 0.15) was observed between ALP and IL-13; albumin and IP-10, G-CSF; LDL, creatinine and G-CSF.

Correlation Network Analysis to Understand the Relationship Between the Cytokine Markers

We also investigated the pairwise relationship between the cytokine markers using correlation analysis. The investigation of the pairwise associations revealed significant positive correlations between G-CSF and most of the markers (TNF-α, RANTES, MIP-1a, MIP-1b, MCP-1, IP-10, IL-2, IL-4, IL-8, IL-1ra, IL-1b). A significant negative correlation was observed between IL-13 and G-CSF; PDGF-BB and eotaxin (Fig. 4a). Next, we extended upon our earlier linear regression-based analysis to perform correlation analysis between the inflammatory cytokines and study parameters such as clinical, biochemical characteristics and non-invasive markers. We observed that G-CSF and IP-10 have shown a significant positive correlation with AST, ALT, and ALP, whereas IL-1ra, MCP-1, and RANTES also show a significant positive correlation with AST and ALT. A significant negative correlation was observed between IL-13 and ALP. G-CSF and IL-1ra have shown a significant positive correlation with non-invasive markers such as APRI, HSI, Fib-4, and NFS. The only cytokine, i.e., IL-9, has shown a significant negative correlation with HSI and NFS (Fig. 4b). Finally, correlation network analysis was performed for all the study parameters, non-invasive markers, and cytokine markers. A significant positive and negative correlations between them are depicted in Fig. 4c.

Fig. 4
figure 4

Spearman correlation analysis was performed a among cytokine markers and b between cytokine markers and study parameters. Orange, red, and blue colour indicate a positive correlation, no correlation, and negative correlation respectively. *Indicates statistically significant. c Correlation network analysis was plotted between cytokine markers, non-invasive markers and clinical and biochemical characteristics. Green-coloured lines indicate a significant positive association, whereas red-coloured lines indicate a negative association.

Full size image

Feature Importance of Cytokine Markers Across Different Group Comparisons

Markers facilitating pairwise discrimination between groups were identified using a second RF-based investigation (See Methods). The feature importance of the cytokine markers was based on the variable importance score across different group comparisons and is represented in Fig. 5a. The order of importance of cytokines for control v/s NAFL group classification was found to be RANTES, IL-1ra, MIP-1b, IL-2, G-CSF, IL-9, TNF-α, IL-13, eotaxin, MCP-1, IL-1b, PDGF-BB, MIP-1a, IP-10, IL-8, and IL-4. For control v/s NASH group classification, the order of importance of cytokines was found to be G-CSF, IL-1ra, TNF- α, RANTES, IL-9, MIP-1b, IL-13, MIP-1a, IL-1b, IL-2, IL-8, IP-10, PDGF-BB, eotaxin, MCP-1, and IL-4. Similarly, for NAFL v/s NASH group classification, the order of importance of cytokines was found to be G-CSF, IL-9, IL-13, eotaxin, TNF- α, IP-10, IL-2, IL-8, MIP-1a, IL-1b, RANTES, MIP-1b, IL-1ra, PDGF-BB, MCP-1, and IL-4. We found that the feature importance marker profiles are stable throughout the three disease state combinations (Supplementary figure).

Fig. 5
figure 5

a Feature importance plot of cytokines based on variable importance score for i) Control v/s NAFL, ii) Control v/s NASH, and iii) NAFL v/s NASH groups. b Venn diagram representing the top predictive cytokine panel for Control v/s NAFL, Control v/s NASH, and NAFL v/s NASH groups. c Network analysis revealed the involvement of various pathways for i) Control v/s NAFL, ii) Control v/s NASH, and iii) NAFL v/s NASH groups. Double-rounded cytokines are the predictive cytokines that can distinguish the study groups from each other.

Full size image

Potential Predictive Cytokines Panel for NAFLD

According to VIP scores, the top 30% of the cytokines are considered to predict each pairwise group comparison. Among them, RANTES, IL-1ra, MIP-1b, IL-2, and G-CSF could be considered for predicting the control v/s NAFL group and the protein–protein network analysis revealed that these markers are associated with cytokine-cytokine receptor interaction, Th1 and Th2 cell differentiation, Th17 cell differentiation, TNF, chemokine, JAK/STAT, IL-17, and P13K/Akt signaling pathways. Furthermore, G-CSF, IL-1ra, TNF-α, RANTES, and IL-9 could be considered for predicting the control v/s NASH group; G-CSF, IL-9, IL-13, eotaxin, and TNF- α for NAFL v/s NASH (Fig. 5b). The protein–protein network analysis revealed that cytokine-cytokine receptor interaction, MAPK, TNF, chemokine, TLR, NOD-like receptor, NF-kB, IL-17, T-cell receptor, and adipocytokine signaling pathways are commonly involved mechanisms for predicting control v/s NASH and NAFL v/s NASH groups. Additionally, the JAK/STAT signaling pathway is also involved in predicting the NAFL_NASH group (Fig. 5c). Among the three distinct sets of markers, the highest predictive value with a sensitivity of 88.20% and specificity of 93.76% was observed for G-CSF, IL-1ra, TNF-α, RANTES, IL-9 panel that could distinguish NASH patients from the control group Table 3.

Table 3 Receiver Operating Characteristic Curve (ROC) Analysis
Full size table

DISCUSSION

Numerous pathophysiological mechanisms such as lipid accumulation, oxidative stress, mitochondrial dysfunction, and endoplasmic reticulum stress could lead to the release of inflammatory cytokines as well as damage to the hepatocytes [24]. In the present study, we evaluated cytokines/chemokines in a clinical setting to demonstrate their use as a disease-specific panel of markers in NAFL and NASH patients. This would also be helpful in identifying promising therapeutic targets for future research.

The present study showed that serum AST, ALT, and ALP levels were significantly increased in NASH patients. It was well-documented that increased AST [25] and ALP [26] levels are independent predictors for diagnosing hepatic fibrosis in NASH patients, whereas increased ALT levels are indicators of hepatic damage [27]. Furthermore, AST and ALT levels were considered as a part of NAFLD non-invasive diagnostic panels such as APRI, HSI, FIB-4, and NFS. Our study has shown that the levels of these non-invasive markers were increased in NASH patients. Sometimes, the levels of hepatic enzymes fluctuate over disease progression, which indicates that there is no clear relationship between aminotransferase levels and histological features of NAFLD [28]. In the case of lipid profile, serum TGs, TC, and LDL levels were significantly higher in NAFL subjects when compared to control and NASH subjects. However, previous study findings revealed that serum TC and LDL levels are directly proportional to the increasing grades of NAFLD [29]. Moreover, it was found that serum HDL levels are significantly decreased with an increase in grades of NAFLD. Alterations of lipid levels in various stages of NAFLD might be due to abnormalities of the lipid metabolism pathway in the hepatocytes. It was observed that the accumulation of more lipids in the hepatocytes is a hallmark of NAFLD development. As a result, cells undergo necrosis and apoptosis, which may further accelerate inflammation by activating various immune cells [30]. Several inflammatory cytokines and chemokines (TNF-α, IL-6, IL-12, IL-23, IL-1β, CCL2, and CCL5) will be released into the liver, promoting disease progression [31]. Therefore, lipid metabolism and activation of inflammatory pathways are linked together and responsible for disease development and progression.

During NASH, hepatic recruitment and activation of leucocytes together with neutrophils and macrophages promote local inflammation and the derived inflammatory signals activate the NF-kB pathway [32]. It has been reported that activation of NF-kB induces the production of TNF-α which is the first crucial step for the progression of the disease, triggering the release of other cytokines and finally destroying hepatocytes [33]. Several evidences confirmed that TNF-α, IL-1, IL-6, IL-8, and IL-18 play a role in the pathogenesis of NAFLD [34, 35]. In line with the previous study results [36,37,38], we observed increased serum TNF-α levels in NASH patients when compared to NAFL and control subjects. Furthermore, the results of the previous study have shown a positive association between TNF-α and NAFLD [39], concluding that TNF-α can increase the risk of NAFLD. IL-1β, a member of the IL-1 cytokine family, is induced through Toll-like receptors (TLR) in Kupffer cells [40]. A study has shown that IL-1 levels are remarkably higher in NAFLD conditions than in other liver diseases [41]. IL-1ra, an anti-inflammatory cytokine, antagonizes the functions of IL-1 and protects the liver by modulating the inflammatory responses. Experimental models have shown that IL-1β contributes to hepatic steatosis and fibrosis [42, 43]. This could be possible by activating Kupffer cells, promoting HSCs conversion to myofibroblasts [44]. Excessive IL-1, IL-1β, and TGF- β levels were observed in IL-1ra-/- mice liver. Additionally, IL-1ra-/- mice have shown a higher degree of steatosis and steatohepatitis when fed with an atherogenic diet [45]. Our study has shown increased levels of IL-1ra and IL-1β in NASH patients. In contrast, recent studies have shown decreased levels of serum IL-1β with increasing steatosis [46] and NASH [47]. The role of these cytokines in the pathophysiology of NAFLD was not clearly determined. In normal conditions, G-CSF is expressed in very low concentration in the hepatocytes, whereas, during NAFLD, G-CSF has shown a five-fold increase in its concentration [48]. Zhang Y et al. have also reported that serum G-CSF levels were significantly higher in the NAFLD mouse model [49]. In line with these animal models of NAFLD, the same has been observed in our clinical study findings. Moreover, a study has shown that deficiency of G-CSF alleviates insulin resistance and hepatic steatosis through the GCSFR-SOCS3-JAK-STAT3 negative feedback pathway, indicating the therapeutic potential in NAFLD [49]. Sometimes, the protective effects of cytokines depend on organ-specific and disease conditions, mediating via different pathways. G-CSF protects the liver from NAFLD via the PI3K/Akt pathway [48], whereas it protects the myocardium via the JAK/STAT3 pathway [50]. We found that G-CSF is a common biomarker that could distinguish NAFL and NASH patients from control subjects. These results indicate that G-CSF plays a potential role in both NAFLD development and progression. However, in order to better understand the role of G-CSF in NAFL and NASH groups, more studies should be conducted in clinical settings. PDGF-BB is another cytokine that promotes the production of collagen and is responsible for the development and progression of hepatic fibrosis. It is involved in HSC proliferation and differentiation [51]. A study has shown that overexpression of PDGF-BB in transgenic mice might be due to HSC activation, thus resulting in hepatic fibrosis [52]. The levels of PDGF-BB increased with respect to the progression of alcoholic liver disease and correlated with hepatic fibrosis [53]. In contrast to our study findings, another study has reported the levels of PDGF-BB were decreased in NASH patients, with the lowest levels detected in cirrhotic NAFLD patients when compared to healthy controls [54]. As the fibrosis progressed, the levels of PDGF-BB declined to a larger extent, proposing this growth factor as a biomarker for the prediction of fibrosis [55]. However, the role of PDGF-BB in the context of chronic liver diseases warrants further investigation.

Our study has shown that serum IL-9 and IL-13 levels were comparatively lower than control subjects in NAFL and NASH patients. IL-9 is produced by Th1, Th17, T-regulatory, and natural killer T cells [56]. It was believed that IL-9 ameliorates inflammation by reducing the production of pro-inflammatory cytokines. However, one study has shown increased IL-9 levels in T2DM patients [57], and another study has shown decreased IL-9 levels in T2DM patients [58]. Studies have not specifically highlighted the IL-9 role in NAFLD and NASH groups. IL-4 and IL-13 are Th2-specific cytokines that counteract the inflammatory processes driven by Th1 cells. Although there is an increased serum levels of IL-4 in NASH patients than controls, it was not significant. Contrary to our study, previous study results have shown increased serum IL-13 levels [59] and hepatic IL-13 mRNA expression in NASH patients [60]. Inhibition of IL-13 might worsen the conditions responsible for the NAFLD pathogenesis such as insulin resistance, metabolic dysfunction, and inflammation in NAFLD disease models. IL-13 suppresses hepatic gluconeogenesis and lowers the production of glucose in the hepatocytes by activating the STAT3 pathway [61]. As far as IL-2 is concerned, serum IL-2 levels were significantly decreased in NAFL patients but not in NASH patients when compared to controls. Consistent with our results, a recent study has also shown a non-significant association of IL-2 with NAFLD [62].

In our cohort, we observed the highest levels of chemokines CCL2/MCP-1, CCL3/MIP-1a, CCL4/MIP-1b, and CCL5/RANTES in sera from patients with NASH. In line with our results, a recent study focused on CCL2/MCP-1 has shown that the serum levels of CCL2 were considerably increased in NAFL patients, and raised to the highest levels in NASH patients. The same study has proven its stronger expression in the liver of NASH patients than in NAFL patients, suggesting its role in the transition from steatosis to NASH progression by increasing the leukocyte infiltration into the liver [63]. These results were confirmed by the other study reporting reduced hepatic steatosis and inflammation in the absence of CCR2, a receptor for CCL2 [64]. During this phase, RANTES/CCR5 take part in leucocyte infiltration into the liver. Recent in vitro data has shown that hepatic RANTES expression has been observed when there was a lipid accumulation in the hepatocytes but not due to activated HSCs or infiltrating inflammatory cells. Nevertheless, along with the other cytokines/chemokines, RANTES may also be involved as a part of low-grade inflammation and lead to the progression of NAFLD [65]. In addition to inflammatory effects, RANTES intervene in hepatic fibrotic effects. A recent study reported higher levels of RANTES when compared to healthy controls in patients with NAFLD-associated cirrhosis [66]. Inconsistent with our study findings, another recent study reported lower circulatory levels of RANTES in NAFL patients than in healthy controls [54]. These inconsistent findings between the studies warrant further investigation with a large cohort. It has been reported that CCL3/MIP-1a was induced by other pro-inflammatory cytokines such as TNF-α, IL-1β, and IFN-γ [67]. Recent studies have demonstrated that MIP-1a levels are increasingly higher at circulatory and molecular levels according to the histological phenotypes of NAFLD [68,69,70]. Animal study results have also shown the hepatic expression of CCL3 in mice fed with a high-cholesterol and high-fat diet, confirming that CCL3 recruits macrophages into the liver and thus contributes to hepatic inflammation [70]. A previous study has shown that MIP-1a/CCL3, MIP-1b/CCL4, IL-8/CXCL8, and IP-10/CXCL10 levels were found to be higher in NASH patients [69], the same has been observed in our study findings.

The CXC chemokines participate through various pathways in the pathogenesis of NAFLD and are responsible for producing inflammatory and fibrotic responses. Serum IL-8/CXCL8 and IP-10/CXCL10 levels were found at higher concentrations in NAFL and NASH patients when compared to control subjects. In addition, serum IL-8 levels and hepatic mRNA expression of IL-8 were found to be higher in NAFL patients [69]. The same trend has been observed in a recent study, with the highest levels in NAFLD-associated cirrhosis patients [54]. It has been reported that increased IL-8 levels are associated with hepatocyte ballooning and significant fibrosis [71], thus leading to a worse prognosis. Furthermore, IL-8 activates the expression of smooth muscle actin in hepatic stellate cells, further promoting liver injury [72]. During the development of NASH, IL-8 plays a role in recruiting neutrophils by activating the AKT/mTOR/STAT3 pathway [69] and IP-10 in the MLK3 signaling pathway [73]. IP-10 is secreted by hepatocytes in case of lobular inflammation and is associated with its severity in NASH patients. In addition, previous data indicate that IP-10 promote hepatic steatosis by stimulating lipogenesis and activating macrophages and induces oxidative stress, inflammation, and fibrosis by activating the NF-κB pathway, hence leading to NASH and NASH-associated fibrosis [69, 74]. Both these chemokines have the potential and could be considered therapeutic targets for NAFLD. Furthermore, we observed that RANTES, IP-10, MCP-1, IL-1ra, and G-CSF have shown significant positive associations with liver function parameters. Besides, IL-2, IL-4, IL-8, MIP-1a, MIP-1b, and TNF-α have not shown significant associations with either of the biochemical characteristics. Overall, the association between lipid metabolism and inflammation in the hepatocytes was influenced by cytokines/chemokines. Our study results demonstrates that the abnormal adipocytokine production and disrupted lipid metabolism together might activate proinflammatory signalling pathways that contribute to chronic inflammation in NAFLD progression. Our distinct panel of cytokine/chemokine markers may serve as diagnostic markers for predicting the early and late stages of NAFLD and are also useful for risk stratification of patients with NAFLD. Additionally, these markers involved in signalling pathways may provide new insights into the pathogenesis of NAFLD and also help in developing novel therapeutic strategies. Together, these results may suggest that inflammatory markers act as mediators directly or indirectly in promoting and modulating the crucial processes of NAFLD progression.

STRENGTHS AND LIMITATIONS

Our study has some strengths and limitations which need to be considered for future studies. First, we have included a significant number of participants in NAFL and NASH groups and evaluated multiple cytokines. Some of our results are well supported by previous studies. We provided FDR-adjusted q-values along with p-values in correlation analysis and linear regression analysis for readers to interpret. Furthermore, we analysed our data using different approaches such as PCoA and random forest classification approach. Using this study data, we revealed that these biomarkers are associated with some signaling pathways which might play a role in the pathogenesis of NAFLD. The major limitation is that we cannot draw causal inferences from the study as it is a cross-sectional study. Due to the presence of inconsistent findings between the studies, large-scale population-based studies are needed in the future to confirm the specific biomarkers of NAFLD.

CONCLUSION

Overall, these results indicate that inflammatory cytokines play a crucial role in the pathophysiology of NAFLD and that there exists a relationship between cytokines and NAFLD. Our study findings revealed a set of distinct cytokine markers for each disease stage, and they might be useful in predicting NAFLD progression. However, further validation is necessary in a separate cohort with a large sample size. Future studies are recommended focusing on the various molecular pathways that are identified in our study for therapeutic decision making.