We conducted comprehensive searches of the following electronic bibliographic databases: PubMed, Web of Science, Embase, Cochrane Library, China National Knowledge Infrastructure (CNKI), Wanfang Database, VIP Database, and the Chinese Biomedical Database. The search encompassed animal experimental studies investigating emodin treatment for PF from January 2000 through July 2025. The search strategy incorporated the terms “pulmonary fibrosis,” “idiopathic fibrosis,” and “emodin,” combined using Boolean operators “AND” or “OR.” Literature collection was supplemented by manual screening of eligible citations from relevant articles to ensure comprehensive and accurate retrieval. Detailed search strategies are provided in Supplementary material. Reference lists of included studies were screened, and literature management was performed using Zotero software. Two independent reviewers screened eligible literature, eliminated duplicates and irrelevant articles, and extracted data. Discrepancies were resolved through consensus discussion.

Eligibility criteria were formulated according to the PICOS framework: rats or mice with PF (Population, P); emodin (Intervention, I); animal models with PF receiving no treatment or placebo/vehicle control (Comparison, C); assessment of PF severity, alveolitis for pathological lung injury, collagen deposition, inflammatory response, oxidative stress levels, and EMT (Outcomes, O); preclinical controlled studies (Study design, S).

Studies were included if they met the following criteria: (1) preclinical animal model studies (including rats and mice); (2) studies employing emodin as the intervention, without restrictions on route of administration, formulation, or treatment duration; (3) studies containing at least one PF control group, without restrictions on modeling method or modeling duration, with control animals receiving equal volumes of non-therapeutic vehicle or no intervention; (4) studies investigating PF as the target disease; (5) studies reporting at least one of the following outcome measures: pulmonary pathological injury indicators (assessment of PF severity and alveolitis), collagen deposition indicators (hydroxyproline content for quantitative meta-analysis; other collagen-related measures were analyzed qualitatively), EMT indicators (TGF-β content and expression levels), inflammation-related indicators [interleukin-6 (IL-6), interleukin-1β (IL-1β), and tumor necrosis factor-α (TNF-α) levels], and oxidative stress-related indicators [superoxide dismutase (SOD) and malondialdehyde (MDA) levels]. Subgroup analyses were performed for studies employing multiple dosages or different species. When measurement units were inconsistent across studies, appropriate conversions were performed for standardization.

Studies were excluded if they met any of the following criteria: (1) duplicate data or studies unrelated to the research topic; (2) studies that did not report relevant results or failed to include specified outcome measures; (3) studies involving disease models combined with other pathological conditions.

During the data extraction phase, two reviewers independently screened the titles and abstracts of the identified studies to select those meeting the inclusion criteria and subsequently reviewed the full texts. Duplicate publications were identified by comparing author names, institutions, sample sizes, intervention details, and outcome data. When the same study was published in multiple languages (e.g., both Chinese and English versions), these were treated as a single study to avoid duplicate data inclusion. After discussion, duplicates or unrelated studies were excluded. The reviewers then extracted the data independently, compiling the information into an Excel data extraction table. This table included the following: (1) baseline information of the included studies (including first author and publication year); (2) species characteristics (including strain, gender, weight, and number of rats or mice); (3) drugs used to induce the PF model; (4) dosages and administration routes of emodin extract; (5) anesthesia methods; (6) outcome measures. In cases where units of the outcome measures were inconsistent across studies, a third reviewer was involved to standardize the units.

If experimental results were not reported in a study, the authors were contacted to retrieve the necessary data. When raw numerical data were unavailable from the text or tables, data were extracted from figures using GetData Graph Digitizer software (version 2.26), where data points were manually extracted by generating coordinates on the graph. To ensure reliability, data extraction from figures was performed independently by two reviewers, and any discrepancies were resolved through re-extraction and consensus discussion. Data presented as mean ± standard error of the mean (SEM) in the articles were converted to mean ± standard deviation (SD) for consistency. In studies with multiple intervention groups, the analysis focused solely on the PF control group and the emodin intervention group. Throughout these stages, we strictly adhered to the relevant guidelines in the Cochrane Handbook for Systematic Reviews of Interventions (Edition 6.5, 2024) (24).

Two reviewers independently assessed the quality and risk of bias of the included studies using the SYRCLE animal risk of bias tool (25). The evaluation addressed six types of bias: selection bias, performance bias, detection bias, attrition bias, reporting bias, and other sources of bias, comprising a total of 10 questions. Each study was categorized into low risk, high risk, or unclear risk for each bias domain. Disagreements between reviewers were resolved through discussion, and if consensus could not be reached, a third reviewer was consulted. Inter-rater reliability was assessed using Cohen’s kappa statistic. “Other bias” was defined as any potential source of bias not covered by the other five domains, including industry funding, baseline imbalances, or inappropriate statistical methods. All included studies were assessed using Review Manager 5.4 software.

Meta-analysis results were visualized using forest plots generated by Review Manager 5.4 and Stata 17 software. The meta-analysis employed either a random-effects model (I² > 50%) or a fixed-effects model (I² < 50%) based on heterogeneity. Outcome measures were analyzed using continuous data. Effect sizes were calculated as standardized mean differences (SMD) using Hedges’ g (bias-corrected SMD) with 95% confidence intervals (95% CI), and heterogeneity was quantified using the Higgins index (I²) and chi-square statistics. A result with I² > 50% or p < 0.01 was considered indicative of significant heterogeneity, and sensitivity analyses were performed by sequentially removing individual studies to assess the impact of heterogeneity. If a meta-analysis could not be performed, qualitative analysis was used to report the data. Additionally, subgroup analyses based on predefined criteria were conducted to evaluate the impact of different doses (≥40 mg/kg vs. <40 mg/kg) and species (rats vs. mice) on outcome heterogeneity. Sensitivity analyses for outcomes were performed using Stata 17 software, with publication bias assessed using funnel plots and Egger’s test, where p < 0.05 indicated the presence of bias. It should be noted that Egger’s test has limited reliability when fewer than 10 studies are included per outcome, and results should be interpreted with caution in such cases. The threshold for statistical significance was set at p < 0.05 for all outcomes, which aligns with conventional standards in meta-analysis. The heterogeneity threshold (p < 0.01) was maintained as a more conservative criterion for assessing between-study variation.

A total of 395 relevant studies were identified from eight databases, including 104 English-language studies and 291 Chinese-language studies. Specifically, PubMed contained 19 studies, Embase included 55, Web of Science had 30, Cochrane Library had no relevant studies, CNKI contained 37, Wanfang included 114, CQVIP contained 112, and CBM included 28. After excluding 109 duplicate studies, 286 studies remained. Further screening of titles and abstracts led to the exclusion of 6 studies that did not involve rats or mice, 242 studies irrelevant to the research topic, and 15 review or meta-analysis studies, leaving 23 studies for full-text review. Upon full-text assessment, 4 studies that did not report relevant outcomes, 1 study with duplicate data published in a different language, 1 study that combined other diseases in the model, and 1 study that did not include the specified outcome measures were excluded. Ultimately, 16 studies were included in the systematic review and meta-analysis (Figure 1).

We included 16 studies involving 425 animals, of which two studies utilized both female and male rats (26, 27), while one study did not report mouse sex Chen et al. (21). Eleven studies used rats, comprising 228 Sprague–Dawley rats (≈ 54%) and 52 Wistar rats (≈ 12%), Sprague–Dawley rats were the most commonly used species. Five studies utilized C57Bl/6 mice (145 mice, 34%), with only one study using the C57Bl/6 J strain. The above statistics include the number of animals used across different dose groups. One study did not report mouse weight (28). The methods used to induce PF included intratracheal instillation of BLM in 10 studies, intratracheal instillation of silica suspension in 3 studies, intragastric administration of BLM in 1 study, intraperitoneal injection of LPS in 1 study, and paraquat solution administered by gavage in 1 study. The most common method for inducing PF was intratracheal instillation of BLM. The anesthesia methods varied widely, with 4 studies not specifying the details.

Among the 16 studies, emodin dosages ranged from 2 mg/kg to 80 mg/kg. Two studies used modified emodin nanoparticle formulations (Emo-NCs solution at 2 mg/kg and EDn solution at 5 mg/kg). Ten studies included 2–3 dose groups. The administration routes of emodin included oral gavage, intraperitoneal injection (28, 29), intratracheal aerosolization using a MicroSprayer® Aerosolizer (30), and intravenous injection (27). Details are provided in Table 1.

Two reviewers independently assessed the risk of bias for each study using the SYRCLE risk of bias tool. One study (29) employed “randomization via random number tables” and was classified as low risk for sequence generation. However, while the remaining 15 studies mentioned randomization, the specific methods were not provided, resulting in an unclear risk rating. Two studies did not report initial body weight (30) and sex (21), leading to an unclear risk rating for baseline comparability. All 16 studies failed to provide sufficient information on whether appropriate allocation concealment procedures were implemented, resulting in an unclear risk rating for this domain. Ten studies employed random allocation of animals, while the remaining six did not specify this method and were rated as high risk. For blinding of the intervention implementer and random outcome assessment, all studies were evaluated as high risk and unclear risk, respectively. Additionally, the blinding of outcome assessors was considered a high risk due to unclear descriptions regarding the researchers’ knowledge of animal treatment. No studies reported incomplete outcome data or selective reporting, and as such, all were classified as low risk in these domains. For other sources of bias, all studies were rated as unclear risk. Notably, while some studies demonstrated low risk in certain areas, others showed unclear or even high risk in specific domains (Figure 2). Overall, despite some studies failing to report critical details such as animal age, sex, or allocation concealment methods, the majority provided comprehensive and satisfactory reporting in other areas.

Prior to initiating this study, we found no comprehensive evaluation of emodin in treating PF; therefore, we undertook this work. We included 16 studies using preclinical animal models, involving 425 animals with PF. Of these, 15 studies were conducted in China between 2007 and 2025, and one study was completed in India in 2024. The overall reporting quality demonstrated comprehensiveness and standardization, with methodological assessment revealing potential risks of bias. Our findings indicate that emodin treatment ameliorated pulmonary histopathological damage related to the degree of PF and alveolitis. Additionally, emodin alleviated collagen deposition by reducing HYP content in lung tissue and decreased both TGF-β content and mRNA expression levels in lung tissue, reflecting emodin’s capacity to block the PF process and inhibit EMT. Regarding anti-inflammatory effects, emodin application significantly reduced levels of relevant inflammatory factors (IL-6, IL-1β, and TNF-α), effectively mitigating the inflammatory response. Emodin also demonstrated notable potential in antioxidant activity, as evidenced by elevated SOD and reduced MDA in lung tissue. While our meta-analysis focused on these measurable endpoints, individual studies within our review also explored underlying molecular mechanisms.

Our evaluation of this study confirmed our findings that emodin exerts therapeutic effects on PF through modulation of diverse mechanisms, including anti-inflammatory, antioxidant, anti-apoptotic pathways, and inhibition of EMT (Figure 10). Beyond the outcomes directly synthesized in our meta-analysis, individual studies within our review investigated specific molecular mechanisms that may contribute to emodin’s anti-fibrotic effects. These findings were not quantitatively meta-analyzed due to limited replication across studies or heterogeneity in measurement methods, and therefore represent hypotheses requiring validation through future research rather than established conclusions from our systematic review.

Persistent inflammation and oxidative stress constitute the fundamental driving forces underlying the onset and progression of PF (35). The inflammatory cascade is initiated by transcription factors such as NF-κB, which rapidly respond and induce massive release of inflammatory mediators. Emodin significantly reduced the levels and mRNA expression of pro-inflammatory cytokines (including IL-6, TNF-α, IL-1β, and IL-4) in BALF and lung tissue (27, 34, 36). Mechanistically, emodin suppressed NF-κB p65 expression and IκBα phosphorylation, thereby blocking nuclear translocation of inflammatory signals (37). Individual studies within our analysis attributed this to inhibition of NF-κB p65/p-IκBα signaling, though this mechanism itself was not quantitatively synthesized. The consistent reduction in inflammatory cytokines across studies provides robust evidence for emodin’s anti-inflammatory capacity. Furthermore, as a key transcription factor in antioxidant defense, emodin activated the Nrf2 signaling pathway to promote nuclear translocation and induced expression of downstream antioxidant and detoxifying enzymes, including HO-1, SOD, GSH, GSH-Px, and CAT (37, 38). Enhanced activity of these enzymes effectively scavenged reactive oxygen species (ROS), thereby substantially decreasing MDA content, a lipid peroxidation product, and reducing tissue damage (27). Notably, negative crosstalk exists between Nrf2 and the NF-κB pathway, where Nrf2 activation can indirectly suppress NF-κB activity (39), establishing the mechanistic foundation for emodin’s synergistic anti-inflammatory and antioxidant effects.

Transforming growth factor-β (TGF-β) is recognized as the “master regulator” in the pathogenesis of organ fibrosis, particularly PF (40). Through receptor binding, it activates the downstream Smad signaling pathway, promoting phosphorylation of Smad2 and Smad3. This complex subsequently translocates to the nucleus, driving transcription of mesenchymal genes and ECM genes such as COL1A1 and FN (41, 42). Studies demonstrate that emodin significantly decreased protein expression levels of TGF-β1, Smad2, Smad3, and Smad7 in lung tissue of rats with PF, directly blocking transmission of this classical pro-fibrotic signal (28, 43, 44). Emodin reduced mRNA and protein levels of Col1 and Col3, as well as serum content of Col1 and Col3 precursors P1CP and P3NP, by inhibiting the TGF-β1/Smad and EMT axis (45). Ultimately, HYP content in lung tissue was significantly reduced, directly confirming alleviation of collagen deposition (21, 26, 46). This effect likely results from the demonstrated TGF-β suppression observed in our meta-analysis. Concurrently, emodin upregulated ADAMTS-1 expression, an ECM-degrading enzyme that facilitates breakdown of excess ECM. Conversely, emodin decreased TIMP-1 mRNA levels, relieving inhibition of ECM degradation, with this dual mechanism promoting ECM remodeling and clearance (47). This mechanism was examined in one study and, while consistent with the observed reduction in HYP content across our meta-analysis, requires replication to establish causality.

Beyond classical inhibition of Smad phosphorylation, emodin enhances anti-fibrotic effects through a non-canonical regulatory mechanism: activation of Sirt1 signaling. Sirt1 is an NAD + -dependent deacetylase whose function relates to cellular metabolism and anti-aging (48). Research indicates that emodin promotes Sirt1 expression, which subsequently deacetylates Smad3. Since acetylated Smad3 possesses higher pro-fibrotic activity, Sirt1-mediated deacetylation effectively attenuates the pro-fibrotic capacity of Smad3, providing additional protection even in the pathological environment of elevated TGF-β1 (28). This suggests that emodin’s inhibition of TGF-β1/Smad signaling operates through multiple layers and dimensions. However, this mechanism was identified in a single study within our review and represents a promising non-canonical pathway that warrants investigation in additional experimental models.

Epithelial-mesenchymal transition (EMT) represents one of the critical events initiating PF, where epithelial cells lose polarity and transform into myofibroblasts capable of migration and ECM secretion, characterized by downregulation of the epithelial marker E-cadherin and upregulation of mesenchymal markers α-SMA and vimentin (49). Research confirms that emodin effectively inhibits EMT progression, specifically manifested by promoting E-cadherin expression while reducing expression of α-SMA, vimentin, and FSP-1 (a marker of activated fibroblasts) (47). More detailed mechanistic investigation revealed an important molecular axis through which emodin regulates EMT: the c-MYC/miR-182-5p/ZEB2 axis. ZEB2 is a core transcriptional repressor in the EMT process that drives the mesenchymal phenotype by suppressing transcription of genes such as E-cadherin. Emodin inhibits c-MYC expression, thereby upregulating miR-182-5p. Because miR-182-5p functions as a negative regulator of ZEB2, its elevation leads to significant reduction in ZEB2 protein expression, thus relieving repression of E-cadherin and decreasing transcription of mesenchymal genes like vimentin (42). Discovery of this regulatory axis provides a precise molecular mechanism for emodin’s inhibition of EMT and suggests potential crosstalk with TGF-β1’s regulation of ZEB2. While these findings are mechanistically plausible and align with our meta-analyzed reductions in fibrosis markers, the specific molecular pathways were investigated in isolated studies and could not be quantitatively synthesized. Future research should examine these mechanisms across multiple models to enable meta-analytic confirmation.

Additionally, damage and apoptosis of alveolar epithelial cells (AECs) are considered the initial signals triggering EMT. Emodin demonstrates protective anti-apoptotic effects through regulation of Bcl-2 family proteins (50). Emodin elevated expression of the anti-apoptotic protein Bcl-2 while reducing expression of the pro-apoptotic protein Bax and the apoptosis executor caspase-3, thereby protecting alveolar epithelial cells and indirectly inhibiting EMT initiation (29).

An important limitation of our analysis is that while we quantitatively synthesized key pathological and biochemical outcomes, many underlying molecular mechanisms (e.g., Sirt1, c-MYC/miR-182-5p/ZEB2 axis, ADAMTS-1/TIMP-1) were investigated in only one or few studies and could not be meta-analyzed. Therefore, our discussion of these mechanisms relies on narrative synthesis rather than pooled effect estimates. Future preclinical studies should employ standardized protocols to measure these mechanistic endpoints, enabling comprehensive meta-analytic evaluation. Additionally, the mechanistic network (Figure 10) represents a theoretical framework synthesizing individual study findings rather than pathways directly validated through our meta-analysis. Although this meta-analysis preliminarily reveals the therapeutic potential of emodin against PF, we must recognize several inherent limitations. These limitations primarily involve geographic distribution of included literature, methodological quality, and heterogeneity in statistical analysis.

First, the literature included in this study exhibits marked geographic bias, with the majority of studies originating from China. This high degree of geographic concentration may limit the external validity of results, failing to adequately represent treatment effects across different ethnicities, environments, or experimental conditions. We strongly recommend that future research should pursue broader international collaboration, incorporating literature from diverse countries and regions to enhance the generalizability of study conclusions. Additionally, some early included studies may present certain methodological and technical limitations. With the rapid advancement of molecular biology techniques, contemporary pharmacological research should adhere strictly to standardized, unified experimental protocols, particularly regarding animal model establishment, intervention dosage selection, and outcome assessment procedures. This is essential for ensuring reproducibility and reliability of experimental data. At present, we advocate active utilization of multi-omics integration technologies (such as genomics, transcriptomics, and metabolomics) alongside other advanced molecular biology approaches to conduct more comprehensive mechanistic investigations of emodin, thereby providing more robust scientific evidence for clinical treatment of PF and effectively improving the efficiency of drug clinical translation.

Second, the included literature revealed deficiencies in risk of bias and methodological quality assessment. Specifically, many studies exhibited high risk regarding blinding of outcome assessors and uncertain risk concerning allocation concealment. These biases represent major factors affecting the accuracy of systematic review results. Of particular concern is that quantitative assessment of pathology forms the foundation for evaluating emodin efficacy, yet existing literature lacks a unified gold standard for grading PF severity and pathological damage (such as the use of Ashcroft and Szapiel scoring systems). To address this shortcoming, future preclinical studies should develop more rigorous and standardized experimental protocols, unify pathological assessment criteria, and strictly adhere to randomization and blinding principles to minimize resulting research bias and error.

Finally, we must candidly acknowledge inherent statistical limitations of this systematic review and meta-analysis. According to the PICO principle, meta-analysis results for certain key outcome indicators exhibited significant heterogeneity. This heterogeneity relates closely to variations in animal model selection, intervention dosage settings, outcome assessment methods, and the number of included studies among investigators. Different experimental conditions and assessment approaches exerted substantial influence on statistical results, thereby limiting our evaluation of the accuracy and reliability of meta-analysis findings.