A comprehensive literature search was performed to identify all relevant preclinical animal studies investigating the efficacy of resveratrol in preventing LF. The following electronic databases were searched: Web of Science, Embase, PubMed, China Biology Medicine (CBM), China National Knowledge Infrastructure (CNKI), Wanfang Database (WF), and China Science Journal Database (VIP). The search was limited to studies published before December 31, 2024, to ensure the inclusion of the most recent evidence. Additionally, manual searches of reference lists from included studies and relevant reviews were conducted to identify potentially eligible studies that might have been missed in the electronic database searches. The search strategy employed a combination of Medical Subject Headings (MeSH) terms and free-text words to maximize the sensitivity and specificity of the search (Supplementary Table 1).

The inclusion criteria for this study were defined based on the PICO framework to ensure the selection of relevant and high-quality preclinical animal studies. Animal: Studies utilizing animal models of LF were included, regardless of the species, sex, age, or weight of the animals. Both induced (e.g., chemically induced, diet-induced) and genetic models of LF were considered eligible. Intervention: Studies in which animals were treated with resveratrol, regardless of the dose, duration, frequency, or route of administration, were included. Comparison: Studies must have included a control group receiving either an equivalent vehicle, physiological saline, or no treatment. Outcomes: Studies reporting outcomes related to the protective effects of resveratrol on LF were included. Primary outcomes of interest included histopathological changes in LF (e.g., collagen deposition, degree of tissue fibrosis), LF progression markers [e.g., hydroxyproline (HYP), α-smooth muscle actin (α-SMA)], TGF-β, and inflammatory cytokine levels [e.g., interleukin-6 (IL-6), tumor necrosis factor-α (TNF-α)]. Secondary outcomes such as liver function biomarkers (e.g., ALT, AST) and oxidative stress markers [e.g., malondialdehyde (MDA), superoxide dismutase (SOD), and glutathione (GSH)] were also considered.

Exclude the following studies: (1) Study Type: Clinical studies, in vitro experiments, and computer simulation studies were excluded. (2) Intervention: Studies without a control group or those in which the treatment group received a combination of resveratrol and other therapeutic interventions were excluded. (3) Duplicate Publications: In cases of duplicate publications, the most recent or comprehensive study was retained, and earlier versions were excluded.

Data extraction was conducted independently by two researchers (QH and JK) to ensure accuracy and minimize bias. The process involved the following steps: (1) Initial Screening: Titles and abstracts of all retrieved studies were screened to exclude irrelevant publications. Studies that did not meet the inclusion criteria were removed at this stage. (2) Full-Text Review: The remaining studies were subjected to a full-text review to assess their eligibility based on the predefined inclusion and exclusion criteria. (3) Data Collection: For studies meeting the inclusion criteria, the following information was extracted: Publication Details: Authors and year of publication. Animal Characteristics: Species, sex, age, weight, and sample size. LF Model: Method used to induce LF (e.g., chemical induction, diet-induced, genetic models). Intervention Details: Resveratrol administration parameters, including dose, duration, frequency, route of administration, and control group treatment. Outcome Measures: Data on histopathological changes, LF progression markers, inflammatory cytokine levels, liver function biomarkers, and oxidative stress markers. If outcome data were presented only in graphical form, attempts were made to contact the corresponding authors to obtain raw data. If raw data were unavailable, graphical data were digitized using WebPlotDigitizer 4.5,¹ a validated tool for extracting numerical data from graphs. For studies reporting multiple data points due to varying doses or time points, data from the group receiving the maximum effective dose or the latest effective time point were extracted for meta-analysis. Any disagreements between the two researchers during data extraction were resolved through discussion and, if necessary, consultation with a third researcher (SX) to reach a consensus.

The methodological quality and risk of bias of the included studies were independently assessed by two reviewers (ZS and QX) using the Systematic Review Center for Laboratory Animal Experimentation (SYRCLE) risk of bias tool (23). This tool is specifically designed to evaluate the risk of bias in animal studies and includes the following domains: sequence generation, baseline characteristics, allocation concealment, random housing, blinding of experimentalists, random outcome assessment, blinding of outcome assessors, incomplete outcome data, selective outcome reporting, other sources of bias. Each domain was assessed and categorized as “yes” (low risk of bias), “no” (high risk of bias), or “unclear” (uncertain risk of bias) based on the information provided in the studies. Discrepancies between the two reviewers were resolved through discussion, and if consensus could not be reached, a third reviewer (SX) was consulted to make the final decision.

Statistical analyses were performed using STATA software (version 17.0). For continuous outcome measures, the overall effect size was expressed as the standardized mean difference (SMD) with 95% confidence intervals (CIs). A p-value of < 0.05 was considered statistically significant. Heterogeneity among studies was assessed using the I² statistic, which quantifies the proportion of total variability in effect estimates attributable to heterogeneity rather than chance. The following thresholds were used to interpret the I² values: I² ≤ 50%: Low to moderate heterogeneity, indicating that a fixed-effects model was appropriate for meta-analysis. I² > 50%: Substantial heterogeneity, prompting further investigation through sensitivity analysis and subgroup analysis to identify potential sources of heterogeneity. If significant heterogeneity persisted and could not be resolved, a random-effects model was applied to account for between-study variability. Sensitivity analysis was conducted by sequentially excluding individual studies to evaluate their impact on the overall effect size and heterogeneity. Subgroup analyses were performed based on predefined factors, such as animal species, LF induction method, resveratrol dosage, and treatment duration, to explore potential sources of heterogeneity. Publication bias was assessed using Egger’s linear regression test and Begg’s rank correlation test. If evidence of publication bias was detected (p < 0.05), the trim-and-fill method was employed to adjust for potential bias and estimate the corrected effect size.

A total of 732 potentially relevant articles were retrieved from seven online databases, including PubMed (24), Embase (295), Web of Science (233), CNKI (25), CBM (26), Wanfang (27), and VIP (19). After removing duplicates, 511 articles remained. Subsequently, 440 articles were excluded based on the screening of titles and abstracts. Following a full-text review, an additional 25 articles were excluded, resulting in the final inclusion of 46 studies. The flow diagram of the study selection process is presented in Figure 2.

The 46 included studies involved a total of 751 animals, with 375 in the treatment groups and 376 in the control groups. Among these, 11 studies used 172 Sprague–Dawley (SD) rats (26, 28–37), 8 studies used 124 C57BL/6 J mice (25, 38–44), 11 studies used 229 Wistar rats (45–55), 5 studies used 46 Balb/c mice (27, 56–59), 3 study used 40 Kunming mice (60–62), 2 studies used 32 Swiss mice (24, 63), 1 study used CD-1 mice (64), and 1 study used 20 Japanese quails (65). Additionally, 4 studies involved 86 rats of unspecified strains (66–69). Regarding the sex of the animals, 39 studies used male animals, 1 study used female animals, 4 studies used both male and female animals, and 2 studies did not specify the sex. The age of the animals was reported in 26 studies, and the weight was described in 36 studies. In terms of resveratrol administration, 34 studies utilized oral gavage or intragastric administration, 11 studies employed intraperitoneal injection, and 1 study used tail vein injection. For LF assessment, 24 studies reported histopathological changes in LF, 18 studies measured TGF-β levels, 21 studies evaluated α-SMA expression, 19 studies assessed type I collagen, 18 studies measured HYP, 6 studies reported hyaluronic acid (HA), 5 studies evaluated type III procollagen N-terminal propeptide (PIINP), 3 studies measured collagen type IV (COL-IV), and 6 studies reported laminin (LN). Regarding liver function, 37 studies reported ALT levels, 31 studies reported AST levels, 11 studies measured albumin (ALB), and 12 studies evaluated alkaline phosphatase (ALP). For oxidative stress markers, 21 studies reported MDA levels, 10 studies measured GSH, and 15 studies evaluated SOD. In terms of inflammatory cytokine levels, 13 studies reported TNF-α, 6 studies measured IL-6, 5 studies assessed IL-1β, and 6 studies evaluated NF-κB. Detailed characteristics of the included studies are presented in Table 1.

Quality assessment of the 46 included studies using the SYRCLE risk-of-bias tool revealed the following score distribution: 4 studies scored 4, 7, and 8 points each, with 14 and 17 studies attaining 5 and 6 points, respectively. Methodological quality analysis (Figure 3) demonstrated that baseline characteristics of experimental subjects were documented in 45 studies, among which 31 explicitly implemented animal randomization, 23 reported randomized housing, and 18 utilized randomized outcome assessment. Notably, while 8 studies declared blinded outcome evaluation, none provided specific details regarding allocation concealment or operator blinding. All included studies met low-risk criteria in three critical domains: data completeness, avoidance of selective reporting, and control of additional biases. Comprehensive evaluation data are compiled in Supplementary Table 2.

Sensitivity analysis excluded through sequential research indicates that despite high heterogeneity, the estimated effects of all outcome measures were stable. The exclusion of Chávez (51) and Hung et al. (64) yielded marginal variations in collagen deposition effect sizes (minimum: −3.04, 95% CI: −3.43 to −2.65; maximum: −3.64, 95% CI: −4.07 to −3.21). Similarly, removing Hessin et al. (47) and Lv (2005) revealed comparable HYP effect size ranges (minimum: −2.93, 95% CI: −3.36 to −2.51; maximum: −3.61, 95% CI: −4.07 to −3.16). The results of secondary outcomes still have robustness (Supplementary Table 4).

To address substantial between-study heterogeneity, stratified subgroup analyses of collagen deposition and HYP levels were performed across four covariates: disease induction methods, animal species, administration routes, and dosing regimens. The analysis revealed disease modeling approaches, animal species selection, and treatment duration as key contributors to collagen deposition heterogeneity. For HYP variability, primary sources included disease induction protocols, administration methods, and dosage parameters. Comprehensive stratification data are presented in the attached Supplementary Table 3.

In the presence of sufficient data, Egger and Begg were used to assess publication bias, and the results showed that all outcome measures included in the analysis were at risk of publication bias (Supplementary Figure 1). To address potential missing research effects, sensitivity analysis conducted through the pruning and filling methods showed that unpublished data had no significant impact on the combined effect estimation of degree of liver fibrosis, HYP, α-SMA, ALT, AST, and SOD, and the research results were robust (Table 2; Figure 10). However, unpublished data had a significant impact on the combined effect estimation of TGF -β, Col1α1, ALB, ALP, MDA, GSH, and TNF-α, and the results obtained from the study were not robust.

This meta-analysis synthesizing 46 preclinical studies demonstrates resveratrol’s therapeutic potential against hepatic fibrosis by attenuating collagen deposition and HYP accumulation, suppressing fibrogenic markers (TGF-β, α-SMA, Col1α1), and modulating ECM components (HA, LN, PIIINP, COL-IV). Hepatoprotective effects were evidenced by reduced ALT/AST/ALP levels and elevated albumin expression. Notwithstanding substantial heterogeneity observed in primary outcomes (collagen deposition: I² = 87.8%; HYP: I² = 81.8%), sensitivity analyses confirmed result stability. Stratified subgroup analyses identified disease induction methods, animal species, administration routes, and dosage regimens as key heterogeneity contributors. Potential publication bias was identified in all results, the robustness of secondary outcome measures decreased after adjusting for pruning and padding, indicating that the bias may have been exaggerated.

The activation of HSC constitutes the central pathological mechanism underlying LF (71, 72). Activated HSCs generate excessive reactive oxygen species (ROS) (73), inducing persistent oxidative stress that compromises cellular membrane integrity and organelle architecture, ultimately driving hepatocyte injury, necrosis, and apoptosis (74). This cyclical process perpetuates HSC activation and fibrotic progression. Mechanistically, MDA, a terminal lipid peroxidation byproduct, serves as a dual biomarker of oxidative damage severity and hepatocyte injury (75, 76). SOD, the primary oxygen radical scavenger, becomes depleted under oxidative assault, while its restoration inhibits MDA-mediated free radical generation (77). GSH depletion exacerbates mitochondrial ROS leakage, triggering apoptotic cascades and profibrotic factor release (78). Resveratrol shows anti-fibrotic potential in preclinical studies via MDA reduction and enhanced SOD/GSH defenses, indicating oxidative stress mitigation.

Fibrogenesis critically depends on inflammatory dysregulation, where early repair-to-pathology transition follows pro−/anti-inflammatory imbalance (79). Chronic inflammation, whether driven by steatosis-induced lipotoxicity or ROS-mediated oxidative stress (80), activates proapoptotic pathways and inflammatory cascades, perpetuating HSC activation via phagocytosis of cellular debris (81–83). Interleukin is a key link in immune regulation. Among them, IL-6 plays a role in promoting fibrosis, and IL-22, IL-24 exerts anti-fibrotic effects (10). TNF-α can regulate the expression of matrix metalloproteinase (MMP) -9, thereby promoting LF (84). Oxidative stress can also activate NF-κB, promote the release of inflammatory factors such as TNF-α and IL-6, exacerbate inflammatory reactions, and lead to fibrosis (85). In addition, inflammation can induce epithelial-mesenchymal transition in liver cells which was associated with reduced TGF-β expression. TGF-β is the initial signal for activation and transformation of quiescent HSC and plays an important role in the occurrence and development of LF (86–88). Studies have shown that inhibiting the TGF-β/SMAD signaling pathway can alleviate the progression of LF (89). Resveratrol can downregulate the expression of TNF-α, IL-6, and TGF-β. Therefore, we speculate that resveratrol may delay the progression of LF by inhibiting inflammatory responses and the TGF-β/SMAD signaling pathway.

The substantial heterogeneity observed in this study primarily stems from differences in experimental design and limitations of animal models. Chemically induced models (e.g., CCl₄/TAA) induce rapid fibrosis through acute liver injury, whereas metabolic models (high-fat diet/choline deficiency) replicate the progressive mechanism of NAFLD. Consequently, resveratrol’s inhibitory effect on TGF-β is markedly weaker in metabolic models compared to chemical models. Meanwhile, the essential differences between the CYP1B1 metabolism of resveratrol in rodents and the human CYP1A2 pathway, as well as the lack of core comorbidity features of human liver fibrosis in existing models (such as insulin resistance and gut microbiota disorders), further weaken the clinical extrapolation of the results.

The 46 preclinical studies included in this meta-analysis lacked dose-ranging toxicity evaluations and standardized documentation of organ-specific adverse reactions (e.g., weight loss, multi-organ injury, or abnormal mortality). Although doses varied significantly (10–500 mg/kg), only 17.4% (8/46) monitored biochemical parameters beyond baseline liver function (ALT/AST), and none provided histopathological assessment of extrahepatic organs (e.g., kidneys, heart). This precludes assessment of whether resveratrol’s known pharmacological risks (e.g., CYP450 enzyme inhibition or estrogen receptor modulation) manifest at anti-fibrotic doses. Furthermore, the longest-duration study (36 weeks) evaluated efficacy endpoints exclusively, neglecting chronic exposure cumulative toxicity assessment. Consequently, neither the No Observed Adverse Effect Level (NOAEL) nor the safety margin required for regulatory dose translation could be established.

Future research must mandate tiered dosing designs (encompassing 20–50 mg/kg human-equivalent and 2–5 therapeutic doses), pathological screening of core organs (liver/kidney/heart), and dynamic CYP450 activity monitoring to address critical gaps in systematic safety reporting. In the preclinical phase, humanized liver models (e.g., FRG mice) or organoid co-culture systems should integrate metabolic-inflammatory interactions, alongside establishing multi-etiological sequential injury models (e.g., HCV infection combined with high-fat diet). In addition, future research on the mechanism of action can focus on microRNAs to comprehensively elucidate the role of resveratrol in preventing and treating LF. Early clinical trials should prioritize liposomal formulations like ResVida®, conducting Phase I maximum tolerated dose studies focused on CYP450/hormonal disturbances while quantifying target engagement (e.g., p-Smad2/3 inhibition rates) via Phase IIa liver biopsies. Ultimately, a precision treatment framework should be developed using validated biomarkers (plasma miR-29a, CK-18) for cohort stratification, coupled with antioxidant-synergistic combination regimens such as obeticholic acid or selonsertib.