Trajectories of Liver Fibrosis and Gene Expression Profiles in Nonalcoholic Fatty Liver Disease Associated With Diabetes

Nonalcoholic fatty liver disease (NAFLD) ranges from simple nonalcoholic fatty liver to nonalcoholic steatohepatitis (NASH) with inflammation and fibrosis. Among the histological features of NASH, liver fibrosis is associated with long-term outcomes, such as liver cirrhosis, liver failure, hepatocellular carcinoma, and all-cause mortality (1,2). However, despite intensive pharmaceutical efforts in drug development and clinical trials, no promising therapy against liver fibrosis has been established (3). It is well recognized that obesity is a leading cause of NAFLD. Still, it remains unclear what clinical features besides obesity are responsible for the pathological progression of liver fibrosis, especially in Japanese patients, who generally are not severely obese, in contrast to Western patients.

In the natural course of posttransfusion chronic hepatitis C, type 2 diabetes accelerates liver fibrosis and subsequent development of hepatocellular carcinoma and mortality (4), suggesting a causal role of hyperglycemia in liver fibrosis. Consistently with this observation, our previous comprehensive analysis of the hepatic gene expression profile revealed that in the livers of subjects with diabetes, expression of the genes in the transforming growth factor-β1family, together with angiogenic factors such as platelet-derived growth factor, are cooperatively upregulated (5). These cytokines activate stellate cells to myofibroblasts that produce collagen. Indeed, genes for collagens are upregulated in the liver of patients with type 2 diabetes, even those without NASH (6). These findings suggest that hyperglycemia independently accelerates liver fibrosis in various liver diseases.

In a small-scale serial liver biopsy analysis of 39 Japanese patients with NAFLD, we previously found that reduction in HbA_1c and insulin treatment independently contribute to alleviating liver fibrosis (7). However, the small number of subjects and short observation periods prevented us from confirming the significance of the findings in the analyses of subgroups with and without diabetes. In addition, gene expression signatures underlying pathological progression of human NAFLD/NASH remain underinvestigated.

To identify the clinical parameters that predict and determine the development of liver fibrosis, we extended the previous study by increasing the number of subjects with serial liver biopsies and observation periods and analyzed the trajectories of liver fibrosis by the presence or absence of diabetes using trend tests and a generalized linear mixed model. Furthermore, we examined serial global hepatic gene expression profiles associated with hepatic fibrosis and glycemic control to address the molecular basis underlying hepatic fibrosis progression.

In this prospective cohort study, subjects who had undergone two or more serial liver biopsies for NAFLD since 1998, with a maximum of 15 years between biopsies, were observed. Subjects’ eligibility and exclusion criteria are described in the Supplementary Appendix.

As shown in Supplementary Fig. 1, we initially recruited 350 subjects with histologically confirmed NAFLD who had undergone liver biopsy from 1998 to 2014 at Kanazawa University Hospital. Of these subjects, 121 had undergone serial liver biopsies. Histology of the serial liver biopsy samples from these 121 subjects was reanalyzed by a pathologist who was not allowed access to the subjects’ clinical information. Three subjects were excluded after histological reexamination (one with hepatic sarcoidosis and two without significant liver steatosis). The accepted subjects were followed until December 2017. Finally, 118 subjects diagnosed histologically with NAFLD were enrolled (Supplementary Fig. 1).

The study was conducted with the approval of the ethics committee of Kanazawa University Hospital in accordance with the Declaration of Helsinki. All subjects provided written informed consent.

Clinical information was obtained for each subject, including age, sex, body measurements, alcohol consumption, medical history, medication use, and laboratory tests. Details are provided in the Supplementary Material.

Biopsies were obtained after a thorough clinical evaluation and the receipt of signed informed consent from each subject and were performed during hospitalization. Liver biopsy slides were stained with hematoxylin-eosin, Azan Mallory, and silver reticulin impregnation. A single pathologist (K.H.) blind to both clinical information and the order in which the biopsies had been obtained analyzed all biopsies. The NAFLD activity score was calculated as the unweighted sum of the scores for steatosis (0–3), lobular inflammation (0–3), and ballooning (0–2) as reported by Kleiner et al. (8). NASH was histologically diagnosed based on the presence of ballooned hepatocytes with lobular hepatitis along with >5% steatosis. Fibrosis was scored as 0 (normal connective tissue), 1 (perisinusoidal fibrosis in zone 3), 2 (perisinusoidal fibrosis in zone 3 with portal fibrosis), 3 (bridging fibrosis), or 4 (cirrhosis) according to the standard criteria developed by Brunt et al. (9).

We examined serial global hepatic gene expression profiles associated with hepatic fibrosis and glycemic control to address the molecular basis underlying hepatic fibrosis progression. Hepatic gene expression profiles were examined in 33 patients using serial liver biopsy samples, as previously reported (10). Details are provided in the Supplementary Appendix.

Among 33 subjects, the histological stage of liver fibrosis progressed in 6, improved in 7, and did not change in others. Regarding HbA_1c levels, values increased in 9 subjects and decreased in 16 (Supplementary Tables 3B and C).

First, we used trend tests to identify clinical parameters that predict liver fibrosis development. Next, we used a generalized linear mixed model for multilevel analyses of continuous liver biopsy to determine the pathological conditions and parameters contributing to liver fibrosis development (details provided in the Supplementary Appendix). SPSS version 27.0 software (IBM Corporation, Armonk, NY) was used for all statistical analyses. P < 0.05 was considered to indicate statistical significance.

All data are included in the article and supplementary materials. The data sets and resources generated and analyzed during the study are available from the corresponding author upon reasonable request.

This study included 118 subjects (67 males and 51 females) diagnosed histologically as having NAFLD (Supplementary Fig. 1). These subjects had a mean duration of follow-up of 3.8 ± 3.45 years (maximum 15 years). A total of 342 liver biopsy samples were systematically rescored according to the classifications. The baseline clinical, laboratory, and histological characteristics of the 118 subjects are described in Supplementary Table 1. The mean (SD) age, BMI, and HbA_1c scores of these subjects with NAFLD were 50.7 (14.1) years, 28.8 (5.6) kg/m², and 7.6% (2.0%), respectively. At baseline, 86 (72.9%) of the 118 subjects had type 2 diabetes. At the initial liver biopsy, 26 subjects had nonalcoholic fatty liver, and 92 had NASH.

First, a trend test was performed to identify baseline clinical factors associated with liver histological fibrosis scores at the end of the follow-up (Table 1). Hepatic fibrosis at follow-up was positively correlated with age (P = 0.001), AST (P = 0.001), AST/ALT ratio (P < 0.001), fibrosis-4 (FIB-4) index (P < 0.001), treatment with ursodeoxycholic acid (P = 0.002), and having a history of hepatocellular carcinoma (P = 0.003) at baseline. It was negatively correlated with total cholesterol (P = 0.011) and platelet count (P < 0.001). On the other hand, HbA_1c and BMI at baseline were not associated with hepatic fibrosis at follow-up (Table 1).

Next, we performed multivariable analyses to evaluate associations between changes in clinical parameters and the progression of liver pathology over time using generalized linear mixed models, adjusting for covariates. In the generalized linear mixed model analysis, the number of liver biopsies and age did not independently affect liver fibrosis progression (Table 2). AST (P = 0.022), γ-glutamyl transferase (GGT) (P = 0.038), HbA_1c (P = 0.038), and treatment with ursodeoxycholic acid (P = 0.043) were positively correlated and platelet count (P < 0.001) negatively correlated with the progression of liver fibrosis in all subjects (Table 2). To discriminate the interaction of these factors with diabetes, we performed multivariable analyses separately in subjects with and without diabetes. In the group with diabetes, HbA_1c (P = 0.038) and GGT (P = 0.025) were positively correlated and platelet count (P = 0.001) negatively correlated with the progression of hepatic fibrosis. On the other hand, in the group without diabetes, AST (P = 0.011) was positively correlated and ALT (P = 0.044) negatively correlated with the progression of hepatic fibrosis (Table 2). Administration or discontinuation of ursodeoxycholic acids, but not insulin, metformin, and sulfonylureas, was significantly associated with changes in liver fibrosis scores.

We calculated the coefficients of interaction by type 2 diabetes and HbA_1c using a generalized linear mixed model with adjustments for sex, age, and BMI. As shown in Supplementary Table 2, diabetes tended to interact with HbA_1c and fibrosis progression (P = 0.054).

To address the molecular basis underlying liver fibrosis progression, we examined serial hepatic gene expression profiles associated with liver fibrosis and glycemic control using RNA sequencing (RNA-seq) in 33 subjects. Baseline clinical characteristics in these subjects are described in Supplementary Table 3A. Among 33 subjects, the histological stage of liver fibrosis progressed in 6, improved in 7, and did not change in others. Regarding HbA_1c levels, values increased in 9 subjects and decreased in 16 (Supplementary Tables 3B and C).

Pathway analyses identified genes coordinately altered in the liver of subjects with hepatic fibrosis progression (Supplementary Table 5) or HbA_1c elevation (Supplementary Table 6) during the course. To further address which components of resident cells participate in alleviating liver fibrosis and glycemic control, we performed gene set enrichment analyses using gene sets associated with resident cells in the human liver defined by single-cell RNA-seq analyses (11). Tables 3 and 4 show the coordinately altered pathways associated with the progression or regression of fibrosis and the elevation or reduction in HbA_1c levels, respectively.

This study is the first to use a trend test and a generalized linear mixed model to reveal the factors contributing to the long-term real-world histological course of the liver in subjects with NAFLD. Trend analyses revealed that the baseline FIB-4 Index and its components (age, platelet count, AST, and ALT) predicted the future progression of liver fibrosis. The FIB-4 Index reflects the real-time status of hepatic fibrosis in subjects with NAFLD in the current study (data not shown), as previously reported (12,13). In a study evaluating the prevalence of NAFLD in type 2 diabetes, the estimated fibrosis score by FIB-4 Index was consistent with the actual score obtained by liver biopsy (14). In addition, of the 4,899 patients with type 1 diabetes, advanced liver fibrosis was present in 6.7% based on a FIB-4 Index of ≥2.67 (15).

Following the trend analysis, we created a generalized linear mixed model to identify the therapeutic target against liver pathology progression in NAFLD. In subjects with NAFLD and diabetes, an increase in HbA_1c was significant, and an increase in BMI tended to be associated with the progression of liver fibrosis. Unlike in the subjects with NAFLD and diabetes, neither changes in HbA_1c nor BMI were associated with fibrosis progression in subjects without diabetes.

In the current study, baseline levels of and changes in BMI, often associated with insulin resistance, were not significantly associated with liver fibrosis development (Tables 1 and 2). In our recent intervention study, reduction in HbA_1c, but not BMI, was significantly associated with a reduction in liver fibrosis in patients with type 2 diabetes and NAFLD treated with the sodium–glucose cotransporter 2 inhibitor tofogliflozin (10). On the basis of these findings, we speculate that insulin resistance exerts limited effects on liver fibrosis progression, at least in relatively slender Japanese patients with NAFLD.

The following two studies support our conclusion. In a serial liver biopsy study in 103 subjects with NAFLD, multivariable analysis showed that diabetes was the cause of the progression of liver fibrosis (16). In addition, it was reported that diabetes incidence increased as liver fibrosis progressed in a serial liver biopsy study of 59 subjects with NAFLD (17). However, these studies did not use a generalized mixed model or identify biomarkers that predict and are associated with fibrosis progression in subjects with type 2 diabetes.

In the current study, surrogate liver enzymes were also distinct between subjects with NAFLD and with and without diabetes. GGT, like HbA_1c, was significantly associated with the progression of liver pathology in the group with NAFLD and diabetes. In contrast, changes in ALT and AST were negatively and positively associated with fibrosis progression, respectively, in the group with NAFLD but without diabetes. These findings are consistent with a previous report that elevated GGT is independently associated with undiagnosed advanced hepatic fibrosis in patients with type 2 diabetes (18) and suggest that a unique pathophysiology may underlie the progression of liver fibrosis between NAFLD with and without diabetes.

On the basis of these findings, we propose a disease entity, “diabetic steatohepatitis,” in which hyperglycemia exacerbates liver fibrosis independent of weight gain. GGT, rather than ALT and AST, may be the best surrogate liver enzyme for diabetic steatohepatitis. Itoh et al. (19) in Japan first reported cirrhotic NASH in five postmenopausal women with obesity and type 2 diabetes. They named its pathology “diabetic cirrhosis” in 1979, prior to the proposal of the NASH pathology by Ludwig et al. (20) in 1980.

To address which components of resident cells participate in alleviating liver fibrosis and glycemic control, we performed gene set enrichment analyses using gene sets associated with resident cells in the human liver defined by single-cell RNA-seq analyses (11). Broad ranges of hepatocytes seemed to be functionally impaired because genes involved in zone 1, zone 2, and zone 3 hepatocytes, including genes encoding coagulation factor F5 and complement C6, were coordinately downregulated in association with liver fibrosis progression (Table 3), suggesting extensive hepatocyte damage during fibrosis progression.

Liver sinusoidal endothelial cells (LSECs) act as a platform for the adhesion of liver-resident noninflammatory macrophages, Kupffer cells, and stellate cells (21). In combination with Kupffer cells, LSECs themselves have a vital scavenger function. Conversely, Kupffer cells protect LSECs from Fas-dependent apoptosis in sepsis (22). In the current study, genes involved in LSECs in zone 3 (central LSEC) and noninflammatory macrophages were coordinately downregulated with the progression of liver fibrosis (Table 3). Inadequately controlled diabetes may injure LSECs in zone 3 and Kupffer cells, which mediate crosstalk between inflammatory cells and stellate cells in the space of Disse, leading to liver fibrosis.

Gene sets of stellate cells and cholangiocytes were coordinately upregulated with the progression of liver fibrosis. Cholangiocytes release autocrine or paracrine factors that regulate proliferative responses and activate hepatic stellate cells into activated myofibroblasts (23), suggesting cholangiocyte proliferation as a potential therapeutic target against liver fibrosis in NASH.

Genes involved in the inflammatory macrophages were coordinately downregulated in association with fibrosis regression. However, unexpectedly, genes involved in immune cells, such as natural killer–like cells, γδ-T cells, and αβ-T cells, were coordinately downregulated in association with fibrosis progression. We interpret these profiles as that some clusters of inflammatory cells may be downregulated in association with so-called burnout NASH during fibrosis progression.

In association with the HbA_1c elevation, genes involved in response to hypoxia and oxidative stress in zone 2 and 3 hepatocytes were coordinately upregulated (Table 4). The findings potentially may reflect the impaired oxygen-binding capacity of HbA_1c and central LSEC injury in the relatively hypoxic zone 3 area. Acceleration of hypoxia in zone 3 would worsen hepatocyte damage and stellate cell activation that might trigger inflammation and fibrosis in zone 3. Indeed, genes involved in stellate cells were coordinately downregulated in association with HbA_1c reduction, suggesting that this pathway may be a common target of diabetes and liver fibrosis.

In conclusion, our generalized linear mixed model analysis indicates that an increase in HbA_1c is significantly associated with hepatic fibrosis progression in subjects with NAFLD and diabetes. Therefore, in subjects with diabetes, hyperglycemia, together with obesity, may be the essential therapeutic target against the progression of NAFLD to cirrhosis and hepatocellular carcinoma. Hepatic gene expression profiles suggest that diabetes-induced hypoxia and oxidative stress injure LSECs in zone 3, which may mediate inflammation and stellate cell activation, leading to liver fibrosis.

Acknowledgments. The authors thank the subjects and medical staff who participated in this study.

Funding. This study was funded by The Junior Scientist Development Grant supported by Novo Nordisk Pharma Ltd. and The Japan Diabetes Society (to S.S.) and supported by Japan Society for the Promotion of Science grant-in-aid 21K19503 (to T.T.) and Japan Agency for Medical Research and Development grant JP20fk0210073 (to T.T. and M.H.).

The funders had no role in the design, data collection, analysis, interpretation, decision to publish, or manuscript writing.

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

Author Contributions. S.S., Y.T., H.A., H.Ts., H.N., M.H., and T.T. participated in statistical analyses and interpreted the data. S.S., Y.T., H.Ta., H.G., Y.N., T.Y., K.A., S.K., and T.T. were responsible for subject recruitment and data collection. S.S. and T.T. wrote and edited the manuscript with input from all authors. K.H. contributed to histological evaluation for all liver biopsy samples. S.S., M.H., and T.T. analyzed and interpreted hepatic gene expression profiles. All authors participated in data interpretation and manuscript review. T.T. designed the study. T.T. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.