Serpine1 as a potential therapeutic target in pyrrolidine alkaloids-induced hepatic sinusoidal obstruction syndrome

Abstract

Background & Aims

Hepatic sinusoidal obstruction syndrome (HSOS) induced by pyrrolizidine alkaloids (PAs) is a life-threatening liver injury for which standardized chronic models are lacking. This study aimed to establish a clinically relevant chronic PA-HSOS murine model and to identify potential therapeutic targets.

Methods

A 28-day model was developed using key alkaloids from Gynura segetum (seneciphylline, senecionine, and N-oxide derivatives), mirroring human exposure. Disease characteristics were evaluated via hematological profiling, CT imaging, histopathology, and transcriptomics. A hepatocyte–liver sinusoidal endothelial cell (LSEC) co-culture system was used to validate molecular mechanisms. TM5441, a Serpine1 inhibitor, was evaluated for therapeutic efficacy.

Results

The model faithfully recapitulated human HSOS, characterized by elevated liver enzymes (ALT, 1.6-fold, p <0.001; AST, 5.7-fold, p <0.0001, n = 6), coagulation dysfunction (prothrombin time, 1.5-fold, p <0.001; activated partial thromboplastin time, 1.1-fold, p <0.05; D-dimer, 8.4-fold, p <0.0001, n = 6), increased liver weight (1.2-fold, p <0.0001, n = 6), reduced VE-cadherin expression (0.28-fold, p <0.001, n = 6), and sinusoidal fibrosis. RNA sequencing identified Serpine1 as the most significantly upregulated gene (|log₂fold-change| ≥2, p <0.05) associated with p53 signaling. Serum Serpine1 was elevated in patients (5.0-fold, p <0.01, n = 16,) and mice (3.7-fold, p <0.001, n = 6) with PA-HSOS. Co-culture confirmed that Serpine1-driven p53 activation (1.8-fold, p <0.01, n = 3) promoted endothelial senescence. TM5441 reduced Serpine1-p53 activity (0.5-fold, p <0.001, n = 6), improved liver function (ALT and AST reduced 30-40%, p <0.001, n = 6), normalized coagulation, restored LSEC integrity (VE-cadherin, 3.2-fold, p <0.001, n = 6), and attenuated inflammation and fibrosis.

Conclusions

This first chronic PA-HSOS murine model closely mirrors human disease and identifies Serpine1 as a critical therapeutic target. Targeting the Serpine1–p53 axis with TM5441 represents a promising strategy to mitigate endothelial injury in PA-HSOS.

Impact and implications

This study establishes a chronic PA-HSOS (pyrrolizidine alkaloid-induced hepatic sinusoidal obstruction syndrome) model that more accurately replicates human disease progression, thereby overcoming the limitations of previous acute models for mechanistic and therapeutic research. The researchers identify Serpine1 as a key regulatory target that could lead to new therapies for PA-HSOS by reducing endothelial senescence and inflammation. These findings provide clinicians and researchers with improved tools for studying drug-induced liver injury and a pathway for translating discoveries into clinical trials. The therapeutic compound TM5441 shows promise for reversing liver dysfunction and vascular damage, while the model itself may aid in the development of better diagnostic criteria and prevention strategies.

Clinical trial number

Not applicable.

Graphical abstract

Pyrrolizidine alkaloids (PAs) are metabolized in hepatocytes to toxic dehydropyrrolizidine alkaloids (DHPAs), which injure liver sinusoidal endothelial cells (LSECs). PA exposure induces Serpine1 upregulation in LSECs, leading to inhibition of MDM2-mediated p53 ubiquitination and proteasomal degradation. Stabilized p53 accumulates in the nucleus and drives transcriptional programs associated with endothelial senescence, apoptosis, and inflammatory responses, ultimately contributing to PA-HSOS pathogenesis. Pharmacological inhibition of Serpine1 with TM5441 restores p53 turnover and alleviates endothelial dysfunction.

Highlights

• •A 28-day oral PA murine model reproduces key clinical and pathological features of human PA-HSOS.
• •Bulk and single-cell RNA-seq identify Serpine1 as the most upregulated gene, linked to p53 activation.
• •PAs activate the Serpine1–p53 pathway, driving LSEC senescence, apoptosis, and barrier dysfunction.
• •TM5441 alleviates PA-HSOS by improving liver and coagulation function and restoring LSEC integrity.
• •The model meets clinical PA-HSOS criteria, with elevated serum Serpine1 supporting diagnostic potential.

Introduction

Hepatic sinusoidal obstruction syndrome (HSOS), also known as hepatic veno-occlusive disease, is a rare yet potentially life-threatening condition characterized by damage to liver sinusoidal endothelial cells (LSECs) and subsequent circulatory obstructions in the liver.¹ While HSOS is linked to hematopoietic stem cell transplantation in Western countries, it is predominantly caused by pyrrolidine alkaloids (PAs) from Gynura segetum in China² – accounting for 50-89% of domestic HSOS cases (2214 diagnosed between 1980-2019).² Globally, over 15,000 PA poisoning cases (including PA-HSOS) have been reported across >10 regions, linked to contaminated plants, herbal products, or foodstuffs.^3,4 With the globalization of Asian herbal remedies, Western incidence is rising, but underdiagnosis (45% misdiagnosis rate^5,6) due to overlapping symptoms and unreported herbal intake likely leads to underestimation of the true prevalence, underscoring the need to elucidate disease mechanisms to enable the development of effective therapies.

PA-HSOS pathogenesis centers on hepatic metabolic activation: hepatocyte cytochrome P450 enzymes (CYP3A4, CYP2B)^7,8 convert PAs to toxic dehydropyrrolizidine alkaloids,⁹ which form pyrrole-protein adducts that target LSECs.¹⁰ While prior studies identified dysregulated proteins (e.g. carbamoyl phosphate synthetase 1), MMP-9/2 activation, and coagulation pathway involvement,[11], [12], [13], [14] the precise signaling cascades initiating disease remain unclear.

Diagnostic advances (e.g. Nanjing criteria¹) have improved clinical recognition, but PA-HSOS research is hindered by inadequate models. Most animal models use monocrotaline (MCT) to induce acute HSOS (1-3 days),[12], [13], [14] failing to recapitulate clinical chronicity or the full toxicity profile of Gynura segetum-derived PAs (e.g. seneciphylline⁶). In vitro studies often bypass hepatocyte metabolism by directly exposing LSECs to PAs,¹⁵ overlooking critical hepatocyte-LSEC crosstalk. Additionally, Gynura segetum’s chemical heterogeneity complicates standardized patient-derived models.¹⁶ Herein, we establish a controllable chronic PA-HSOS animal model using Gynura segetum components and demonstrate that inhibiting Serpine1 signaling protects against disease, offering a promising therapeutic strategy.

Materials and methods

Patient selection and preparation of biological samples

Blood samples and liver tissues were obtained from patients with PA-HSOS after receiving approval from the Ethics Committee of Zhongshan Hospital of Fudan University. Informed consent was obtained from all participants before sample collection. All patients were initially diagnosed according to the Nanjing criteria,¹ with subsequent confirmation at our hospital. For comparison, blood samples were collected from healthy individuals matched for age and sex. Demographic data are summarized (Table S1). Liver tissues were obtained from the normal tissue adjacent to hepatic hemangiomas in patients undergoing resection.

Preparation and qualitative and quantitative analysis of lyophilized powder extracted from Gynura segetum

Given the challenges in differentiating between San Qi and Gynura segetum, mass spectrometry analysis was employed to accurately detect the components of Gynura segetum, including PAs, as per previously reported methods.¹⁷ Lyophilized powder was extracted from the crude Gynura segetum consumed by patients diagnosed with PA-HSOS to remove excess impurities, enhance component preservation and detection, and facilitate storage. The lyophilized powder was stored in dark conditions (Fig. S1A). Quantitative and qualitative analyses of the lyophilized powder were performed using our previously established high-performance liquid chromatography coupled with UV detection mass spectrometry (HPLC-UV-MS) method.⁶ Four reference compounds, including seneciophylline, senecionine, seneciphylline N-oxide, and senecionine N-oxide, were analyzed. The HPLC-UV-MS analysis utilized a gradient elution of 0.1% formic acid and acetonitrile, with MS detection in positive/negative modes at a resolution of 70,000. Reference solutions at 2 μg/ml were prepared in methanol. Test samples were processed from the lyophilized powder, extracted, and centrifuged for analysis.

PA-HSOS animal model establishment

All animal experiments were approved by the Animal Experiment Ethics Committee of Zhongshan Hospital of Fudan University. Six-week-old C57BL/6 mice were housed in a specific pathogen-free facility under controlled conditions. Mice were fed a standard laboratory diet with access to filtered water. The four high-purity alkaloids, including seneciophylline (Cat# TN1078, TargetMol), senecionine (Cat# TQ0192, TargetMol), seneciphylline N-oxide (Cat# TN2204, TargetMol), and senecionine N-oxide (Cat# TN2203, TargetMol), were formulated to mimic the proportions found in Gynura segetum associated with diagnosed cases. The alkaloids were suspended in 0.5% carboxymethylcellulose sodium (CMC-Na) to ensure effective dissolution during oral administration.

We selected a 28-day exposure period because clinical cases of PA-induced hepatotoxicity typically develop symptoms within 1 month.¹ Also, exploring the optimal accurate dose selection is challenging as patients can rarely specify their Gynura segetum intake, often consumed via alcohol infusion. Hence, we conducted a 28-day preliminary experiment with dose gradients and daily gastric administration to ascertain the optimal concentration of PAs. C57BL/6 mice were randomly divided into four groups (n = 6/group) as follows: (1) Control (Group C): 0.5% CMC-Na; (2) Low-dose (Group L): 25 mg/kg/d PAs; (3) Medium-dose (Group M): 50 mg/kg/d PAs; and (4) High-dose (Group H): 100 mg/kg/d PAs. Livers were collected for analysis after spontaneous death or euthanasia at day 28. Based on the dose optimization results, C57BL/6 mice were randomly divided into two groups (n = 6/group) for the 28-day study as follows: (1) Control group: 0.5% CMC-Na and (2) PA-HSOS group: 50 mg/kg/d PAs. On day 28, peripheral blood was collected, followed by imaging examinations and subsequent analyses.

Serpine1 inhibitor treatment

Following the 28-day establishment of the PA-HSOS model, C57BL/6 mice underwent treatment with the Serpine1 inhibitor TM5441 (Cat# HY-101761, MCE). The treatment was administered 20 mg/kg daily¹⁸ for 1 week. After the treatment period, mice were sacrificed, and tissues were collected for further analysis to evaluate the effect of TM5441 on the PA-HSOS model. Concurrently, in our in vitro cell model, TM5441 was applied at a concentration of 20 μM to assess its impact on cellular processes.

Statistical analyses

Statistical analysis was conducted with GraphPad Prism, employing the Shapiro-Wilk test to assess data normality. Data are presented as SEM or median with IQR for continuous variables, and as frequencies and percentages for categorical variables. Group comparisons were made using the Student's t test or the Mann-Whitney U test for continuous variables, and the Chi-square test or Fisher's exact test for categorical variables. Correlations were evaluated with Pearson's or Spearman's coefficients. A p <0.05 indicated statistical significance.

Additional information regarding the materials and methods used can be found in the supplementary materials.

Results

Serum biochemical profiles in the PA-HSOS mouse model resemble human HSOS

To simulate PA-HSOS in mice, we first conducted a comprehensive component analysis of Gynura segetum, using samples derived from patients diagnosed with PA-HSOS in Anhui Province, China. The samples were processed using freeze-drying to retain the active components, resulting in an extraction yield of 82.9%. We focused specifically on qualifying and quantifying four proven key toxic components, including seneciphylline, senecionine, seneciphylline N-oxide, and senecionine N-oxide, which represent the bioactivated forms responsible for PA-mediated hepatotoxicity (Fig. S1B; Table S2). The chromatographic profiles and mass spectral characteristics of these four components are detailed (Fig. S1C–F). The average concentrations of these four key components were calculated (Table S3). Our findings revealed that seneciphylline was the most abundant of the four toxic components, with an average concentration of approximately 11 mg/g. Senecionine N-oxide exhibited the lowest concentration at approximately 3 mg/g. Seneciphylline N-oxide and senecionine were not significantly different from each other, both measuring approximately 6 mg/g. Meanwhile, HPLC-UV-MS analysis was employed to profile all constituents of Gynura segetum (Table S4).

After a 28-day daily oral administration of the four aforementioned PAs to establish a PA-HSOS model in mice, we initially conducted a preliminary experiment to determine the optimal dosage, which was found to be 50 mg/kg/day (Fig. S2A–C). Using this dosage, we proceeded with the formal experiment, with the operational timeline illustrated (Fig. 1A). Compared to controls, the PA-HSOS group showed significantly increased body weight gain (>5% from baseline) at the end of modeling, indicating systemic toxicity (Fig. 1B). In this model, to directly assess the toxic impact of polyphthalamide on protein structures, plasma pyrrole-protein adduct levels, serving as a biomarker for PA metabolite-induced hepatotoxicity, were measured and found to be significantly increased in both patients and mice with PA-HSOS (Fig. 1C). Concurrently, liver function parameters, including total bilirubin, alanine aminotransferase (ALT), aspartate aminotransferase (AST), and alkaline phosphatase (ALP), were markedly elevated (Fig. 1D). Additionally, markers of hepatic synthetic function showed reduced serum albumin and elevated direct bilirubin and total bile acid levels in both the mouse model and human patients with HSOS, mirroring liver dysfunction observed clinically (Fig. S3A). Further assessments revealed impaired coagulation function, as indicated by prolonged prothrombin time (PT), activated partial thromboplastin time (aPTT), and increased D-dimer levels (Fig. 1E), suggesting disseminated intravascular coagulation or compromised hemostasis. Moreover, we observed elevated pro-inflammatory markers IL-1β and TNF-α, indicative of a systemic inflammatory response characteristic of the liver.[19], [20], [21] These markers followed the same elevated trend as clinical measures (Fig. 1F). Hematological evaluations in the PA-HSOS mice also showed a pattern of abnormalities, including reduced platelet and red blood cell (RBC) counts with variable white blood cell responses (Fig. 1G), indicating bone marrow dysfunction and peripheral blood alterations associated with PA-HSOS pathophysiology.

Fig. 1.

Hematological and biochemical alterations in mice and humans with PA-HSOS.

(A) Schematic representation of PA-HSOS mouse model establishment through Gynura segetum dosing. (B) Progressive body weight increase in mice following PA exposure indicates systemic toxicity (n = 6/group). (C) Plasma PPA levels are substantially elevated in both mice and humans with PA-HSOS compared to controls. (D) Liver function parameters, including T-BIL, ALT, AST, and ALP, are significantly elevated in PA-HSOS. (E) Coagulation dysfunction, evidenced by prolonged PT, aPTT, and elevated D-dimer levels, demonstrates impaired synthetic function and a hypercoagulable state in mice and humans with PA-HSOS. (F) Increased pro-inflammatory cytokines, including IL-1β and TNF-α, in mice and humans with PA-HSOS compared to controls. (G) Hematological abnormalities, including reduced PLT and RBC counts with variable WBC responses, indicate bone marrow dysfunction and peripheral blood alterations associated with PA-HSOS pathophysiology. Data points represent individual measurements from mice (n = 6/group) and humans (n = 16/group). All data represent means ± SEM. ∗p <0.05, ∗∗p <0.01, ∗∗∗p <0.001, ∗∗∗∗p <0.0001 by Mann-Whitney U test. ALP, alkaline phosphatase; ALT, alanine aminotransferase; aPTT, activated partial thromboplastin time; AST, aspartate aminotransferase; PA, pyrrolizidine alkaloid; PA-HSOS, pyrrolizidine alkaloid–induced hepatic sinusoidal obstruction syndrome; PLT, platelet count; PPA, pyrrole–protein adducts; PT, prothrombin time; RBC, red blood cell; T-BIL, total bilirubin; WBC, white blood cell.

While minor parameter disparities exist, the PA-HSOS mouse model consistently mirrors key hematological features of human HSOS. Hematological dysfunction concurrent with liver injury – including alterations in RBC/platelet indices and white blood cell subpopulations – further validates the model’s clinical relevance (Fig. S3B). Furthermore, a decline in renal function was observed in both the mouse model and human patients, as evidenced by elevated blood creatinine and blood urea nitrogen levels (Fig. S3C). Together, these findings collectively establish a systemic disease pattern characterized by primary hepatic failure with secondary effects on hematological and renal function, underscoring the multisystem nature of PA-HSOS pathophysiology.

Murine liver pathology and hemodynamic dysfunction closely recapitulate human PA-HSOS

To verify that the mouse model accurately recapitulates human PA-HSOS, we performed comparative analyses at the imaging, histopathological, and hemodynamic levels. CT scans demonstrated consistent hepatic abnormalities between PA-HSOS mice and human patients. Healthy controls exhibited normal liver morphology with smooth surfaces, symmetrical lobes, and uniform parenchymal density on non-contrast CT. At the same time, contrast-enhanced images showed evenly distributed vasculature without stenosis. In contrast, both mice and patients with PA-HSOS exhibited diffuse hepatomegaly with rounded contours, heterogeneous parenchymal hypodensity, and significant ascites on non-contrast CT scans. Contrast-enhanced arterial phase imaging further revealed a distinctive "irregular" or "mottled" pattern of enhancement, characterized by thickening of the hepatic artery and its branches (Fig. 2A). These imaging features confirm the fidelity of the murine model in recapitulating human HSOS pathology, reflecting the heterogeneous nature of vascular injury across the hepatic parenchyma.

Fig. 2.

Hepatic imaging, histology, and hemodynamics in mice and humans with PA-HSOS.

(A) Representative non-contrast and contrast-enhanced CT scans reveal hepatic pathology in control and PA-HSOS. Red arrows denote ascites, while red circles highlight characteristic "map-like" and "mottled" heterogeneous enhancement patterns of the liver. (B) Gross liver morphology reveals macroscopic differences between control and PA-HSOS mice. scale bar, 1 cm. (C) Representative sections show H&E, Masson trichrome, and IL-1β immunohistochemical staining, along with CD31 and VE-cadherin immunofluorescence in control and PA-HSOS tissues from both mice and humans. Black arrows indicate RBC accumulation within hepatic sinusoids, a hallmark of sinusoidal obstruction. Scale bars, 200 μm (or 50 μm for enlarged regions). (D) Hemodynamic parameters including PP, MAP, PBF, and HVPG reveal systemic and portal hemodynamic dysfunction in PA-HSOS mice and humans compared to controls (mice: n = 6/group; humans: n = 16/group). All data represent means ± SEM. ∗p <0.05, ∗∗ p <0.01, ∗∗∗ p <0.001, ∗∗∗∗p <0.0001 by Mann-Whitney U test. HVPG, hepatic venous pressure gradient; MAP, mean arterial pressure; PA-HSOS, pyrrolizidine alkaloid–induced hepatic sinusoidal obstruction syndrome; PBF, portal blood flow; PP, portal pressure; RBC, red blood cell.

Morphological and immunohistochemical analyses further corroborated the model's similarity to human HSOS. On day 28, PA-HSOS mice exhibited significantly more hepatic congestion and enlargement compared to the control group (Fig. 2B). Moreover, hepatomegaly is a hallmark finding in PA-HSOS, with both absolute liver weight (Fig. S4A) and liver-to-body weight ratio (Fig. S4B) significantly increased compared to controls, reflecting substantial hepatic hypertrophy and congestion. Meanhwile, ascites quantification demonstrated fluid accumulation in PA-HSOS mice, consistent with portal hypertension and hepatic decompensation (Fig. S4C).

Comprehensive quantitative analyses revealed striking pathological similarities between murine and human PA-HSOS (Fig. 2C). In H&E staining, both murine and human PA-HSOS groups exhibited characteristic sinusoidal congestion (black arrows) and hepatocyte necrosis, with quantitative analysis showing significantly increased RBCs (p <0.001, Fig. S4D). Masson's trichrome staining revealed marked collagen deposition (blue fibers), with murine cases (p <0.001) exhibiting more severe fibrosis than human cases (p <0.01, Fig. S4E). IL-1β immunostaining demonstrated inflammatory cell infiltration, quantified as 4.4-fold (mice, p <0.001) and 2.5-fold (human, p <0.01) increases in positive cells/mm² (Fig. S4F). Immunofluorescence showed that VE-cadherin expression in CD31⁺ LSECs was reduced both in mice (p <0.001) and humans (p <0.01, Fig. S4G). These findings validate the murine model's ability to recapitulate human PA-HSOS pathology across all examined parameters. Moreover, H&E staining revealed that PA-HSOS affected the lungs (RBC leakage) and stomach (tissue swelling), while the kidneys, spleen, and heart appeared normal (Fig. S4H).

Hemodynamic parameters including portal pressure (PP), mean arterial pressure (MAP), pulmonary blood flow (PBF), and hepatic venous pressure gradient (HVPG) revealed parallel systemic and portal hemodynamic dysfunction in PA-HSOS compared to controls (Fig. 2D). In mice, PA-HSOS induced a significant increase in portal hypertension (1.74-fold, p <0.0001), systemic hypoperfusion (MAP reduction to 115.2 ± 2.1 mmHg, p <0.05), and portal flow impairment (PBF reduction of 37.8%, p <0.01). Human patients exhibited similar severe changes, with HVPG reaching 21.87 ± 2.89 mmHg (vs. normal ≤5 mmHg,²² p <0.0001), MAP dropping to 81.23 ± 3.12 mmHg (p <0.05), and PBF slowing to 15.32 ± 3.78 cm/s (p <0.0001). These abnormalities, combined with the CT findings and histopathological evidence, establish a multiscale correlation between the pathophysiology of murine and human PA-HSOS.

Transcriptomic profiling reveals Serpine1 as a key mediator in PA-HSOS pathogenesis

Transcriptomic analysis of liver tissues from PA-HSOS and control mice identified 1,267 differentially expressed genes (DEGs) (|log₂fold-change| ≥2, p <0.05), with Serpine1 (Pai-1), Fosl1, Lcn2, Il1r2, and Mt2 being the top five upregulated genes and Prss2, Cpa1, 2210010C04RIK, Amy2a3, and Try4 the top five downregulated genes (Fig. 3A). Functional enrichment analyses, including DEGs (Fig. S5A) and Gene Ontology enrichment analysis (Fig. S5B), were performed to identify hepatic transcriptomic alterations in PA-HSOS mice. KEGG pathway enrichment analysis identified dysregulated biological pathways in PA-HSOS livers, highlighting mechanisms underlying sinusoidal obstruction and hepatic injury. It revealed the p53 signaling pathway as the most significantly upregulated molecular pathway (Fig. 3B). Notably, Serpine1 emerged as a key player within this pathway, exhibiting marked upregulation, which suggests its potential role in mediating PA–HSOS–related hepatic injury (Fig. 3C).

Fig. 3.

Transcriptomic signatures in PA-HSOS mouse livers.

(A) Volcano plot of differential gene expression reveals dysregulation of genes critical to PA-HSOS pathogenesis, with log₂ fold-change and statistical significance plotted to identify transcriptional alterations between PA-HSOS and control livers (n = 6/group). (B) KEGG pathway enrichment analysis identifies dysregulated biological pathways in PA-HSOS livers (n = 6/group). (C) Heatmap of key differentially expressed genes demonstrates distinct transcriptomic signatures distinguishing PA-HSOS from control samples (n = 6/group). (D) Single-cell RNA-seq UMAP clustering delineates distinct hepatic cell populations and their transcriptional heterogeneity in PA-HSOS vs. control livers (n = 3/group). (E) Violin plots of Serpine1 expression in LSECs (n = 3/group). (F) Ridge plot visualization further characterizes the dynamic range of Serpine1 expression in LSECs (n = 3/group). All data represent means ± SEM. LSEC, liver sinusoidal endothelial cell; PA-HSOS, pyrrolizidine alkaloid–induced hepatic sinusoidal obstruction syndrome; RNA-seq, RNA sequencing; UMAP, uniform manifold approximation and projection.

To further investigate the cellular origin and distribution of Serpine1, we performed single-cell RNA sequencing on liver samples from mice. Uniform manifold approximation and projection analysis revealed distinct changes in endothelial cell populations between the control and PA-HSOS groups (Fig. 3D). In controls, endothelial cells formed a well-defined, concentrated cluster accounting for 18.5% of total cells; while in PA-HSOS, this population showed the most significant proportional decline (to 9.1%) with fragmented clustering.

Violin plots of Serpine1 expression within LSECs (exclusively focusing on this cell type) demonstrated marked upregulation in PA-HSOS, implicating this profibrotic and prothrombotic mediator in disease pathogenesis (Fig. 3E). In control mice, Serpine1 was rarely detected in LSECs (0.48% positive cells), with the vast majority (99.52%) showing no expression. In contrast, PA-HSOS induced profound Serpine1 activation, with 55.0% of LSECs becoming Serpine1-positive and only 44.9% remaining negative. Ridge plot visualization further highlighted the dynamic range of Serpine1 expression in LSECs (Fig. 3F). These findings, consistent with clinical observations, confirm LSECs as the primary target and most severely affected cell type in PA-HSOS,¹ collectively implicating Serpine1 as a central molecular effector linking p53 pathway activation to endothelial dysfunction and disease progression.

PAs activate the Serpine1-p53-p16/p21 pathway to induce endothelial senescence and dysfunction in vivo

To elucidate the mechanism of PA-HSOS, we explored the role of the Serpine1-p53 pathway in LSECs using the PA-HSOS murine model. First, mRNA analysis of LSECs isolated from mice showed that PAs significantly upregulated Serpine1 and p53 expression, indicating transcriptional activation of the stress-response pathway (Fig. 4A). Western blot analysis of LSECs at the animal level further verified that protein levels of Serpine1, p53, p21, and p16 were all elevated in PA-HSOS LSECs compared to controls (Fig. 4B), confirming activation of the p53-mediated cell cycle checkpoint cascade.

Fig. 4.

The Serpine1/p53 pathway activation and associated pathologies in vivo.

(A) qPCR analysis demonstrates upregulation of Serpine1 and p53 mRNA in mouse LSECs from PA-HSOS mice (n = 6/group). (B) Western blot analysis of mouse LSECs confirms elevated Serpine1 protein alongside increased p53, p21, and p16 expression. (C) Flow cytometry-based cell cycle analysis reveals G1/S phase arrest in PA-HSOS LSECs (n = 6/group). (D) qPCR quantification of senescence-associated molecules in mouse LSECs shows increased expression of key senescence markers (n = 6/group). (E) Immunofluorescence intensity of p16 in ERG-positive cells in mouse livers localizes senescent endothelial cells within hepatic sinusoids (n = 6/group). Scale bars, 50 μm. (F) TUNEL staining reveals enhanced endothelial cell apoptosis in PA-HSOS livers compared to controls (n = 6/group). Scale bars, 200 μm (or 50 μm for enlarged regions). (G) Flow cytometry of mouse LSECs quantifies apoptotic populations (n = 6/group). (H) Western blot analysis of Bax and cleaved caspase-3 in mouse LSECs confirms activation of the intrinsic apoptotic pathway. (I) In vivo endothelial permeability assay using tail-vein injection of 40 kDa FITC-dextran demonstrates compromised sinusoidal barrier function in PA-HSOS mice, with quantified fluorescence intensity reflecting enhanced vascular leakage into hepatic tissue (n = 6/group). Scale bars, 100 μm (or 50 μm for enlarged regions). (J) Western blot of eNOS and VE-cadherin in mouse LSECs reveals reduced expression of these vascular integrity markers. (K) Serum Serpine1 levels are significantly elevated in PA-HSOS mice and humans compared to respective controls (mice: n = 6/group; humans: n = 16/group). All data represent means ± SEM. ∗∗p <0.01, ∗∗∗p <0.001, ∗∗∗∗p <0.0001 by Mann-Whitney U test.

Cell cycle analysis revealed that PA-HSOS induces a G1/S-phase arrest in LSECs, as evidenced by an increased proportion of cells in the G1/S phase and a concomitant reduction in the G2 phase (Fig. 4C), demonstrating a mitotic blockade characteristic of cellular senescence. mRNA analysis of senescence-related factors in LSECs (Fig. 4D) displayed significant upregulation, further supporting enhanced cellular senescence in PA-HSOS LSECs. Immunofluorescence staining showed that the average fluorescence intensity of p16 in ETS-related gene (ERG)-positive cells – where ERG, a nuclear marker specific to ECs, localizes to the nucleus²³ – was markedly higher in the PA-HSOS group (Fig. 4E), directly indicating endothelial senescence.

TUNEL staining revealed enhanced endothelial cell apoptosis in the PA-HSOS group (Fig. 4F), indicating programmed cell death within vascular compartments. Flow cytometry analysis of isolated LSEC apoptosis (Fig. 4G), along with western blot detection of apoptosis-associated proteins (Bax and cleaved caspase-3)²⁴ in LSECs (Fig. 4H), consistently confirmed elevated apoptosis in PA-HSOS LSECs. For endothelial function, the FITC-Dextran (40 kDa) permeability assay in mouse liver (Fig. 4I) showed that the PA-HSOS group had significantly increased endothelial permeability, indicating compromised sinusoidal barrier function. Western blot analysis of endothelial function-related proteins (eNOS and VE-cadherin) showed decreased expression in the PA-HSOS group (Fig. 4J), indicating endothelial dysfunction and loss of adherens junction stability. Finally, the ELISA result (Fig. 4K) indicated that serum Serpine1 levels were elevated in both patients and murine PA-HSOS models compared to respective controls, establishing Serpine1 as a circulating biomarker of disease severity and reflecting systemic endothelial dysregulation. Collectively, these in vivo data demonstrate that PAs activate the Serpine1-p53 pathway in LSECs, triggering a cascade of events that includes endothelial senescence, heightened apoptosis, and disrupted barrier function, all of which are pivotal in the pathogenesis of PA-HSOS.

PAs activate the Serpine1-p53 pathway and promote endothelial senescence in vitro experiments

To further validate the role of the Serpine1-p53 pathway in PA-induced endothelial senescence, we established in vitro models. As highlighted in the expert consensus,¹ the hepatotoxicity of PAs is primarily mediated by metabolic activation in hepatocytes, and LSECs are the primary target cells of PA damage, prompting us to use a Transwell co-culture system (hepatocytes in the upper chamber and LSECs in the lower chamber; Fig. 5A) to mimic the microenvironment where hepatocyte-metabolized PAs act on LSECs, enabling paracrine signaling and physiologically relevant PA metabolite exposure. The co-culture system was treated with the PA mixture, following the same component ratios as in our animal model. Through time-course experiments, 24 h was identified as the optimal treatment duration (Fig. 5B).

Fig. 5.

PA-induced endothelial dysfunction and senescence in a hepatocyte-LSEC co-culture system.

(A) Schematic representation of the hepatocyte-LSEC co-culture system, with hepatocytes cultured in the upper Transwell compartment and LSECs in the lower chamber, enabling paracrine signaling and physiologically relevant PA metabolite exposure. (B) PA treatment impairs LSEC morphology and viability in co-culture, as demonstrated by phase-contrast microscopy and quantified by CCK-8 assay (n = 3 independent experiments). Scale bars, 100 μm. (C) qPCR analysis reveals transcriptional upregulation of Serpine1 and p53 alongside their downstream effectors p21 and p16 in PA-exposed LSECs (n = 3 independent experiments). (D) Western blot analysis demonstrates corresponding increases in Serpine1, p53, p21, and p16 protein levels in LSECs following PA treatment. (E) SA-β-Gal staining quantifies senescent LSECs following PA exposure, providing morphological evidence of endothelial senescence induction and confirming functional engagement of the senescence program (n = 3 independent experiments). (F) qPCR profiling of inflammatory cytokines and SASP genes demonstrates coordinated upregulation of pro-inflammatory mediators and senescence markers in PA-treated LSECs (n = 3 independent experiments). (G) Western blot analysis of Serpine1, p53, p21, and p16 in PA-exposed LSECs treated with si-Serpine1 demonstrates that Serpine1 knockdown attenuates p53-pathway activation and senescence markers. All data represent means ± SEM. ∗p <0.05, ∗∗p <0.01, ∗∗∗p <0.001, ∗∗∗∗p <0.0001 by Mann-Whitney U test. n = 3/group (B-G). LSEC, liver sinusoidal endothelial cell; PA, pyrrolizidine alkaloid; qPCR, quantitative PCR; SA-β-Gal, senescence-associated β-galactosidase; SASP, senescence-associated secretory phenotype; si, small-interfering.

Additionally, building on our previous whole-liver RNA-seq results, we focused specifically on metabolic pathways and identified fatty acid metabolism as the most significantly downregulated among 33 altered pathways (Fig. S6A and B). Subsequent targeted metabolomic sequencing of 37 fatty acids in hepatocytes confirmed these findings, showing decreased saturated and increased monounsaturated fatty acids (Table S5), suggesting that PAs may alter fatty acid synthase and stearoyl-CoA desaturase activities.

Subsequent analysis of LSECs revealed that, at the mRNA level, the expression of Serpine1, p53, p21, and p16 was significantly increased (Fig. 5C), confirming p53-pathway activation at the mRNA level. In contrast, the mRNA levels of Fosl1, Lcn2, Il1r2, and Mt2 showed no significant elevation or lacked statistical significance (Fig. S7A). At the protein level, Western blot analysis also demonstrated increased expression of Serpine1, p53, p21, and p16 in LSECs (Fig. 5D), validating transcriptional findings and indicating sustained pathway activation. These results confirmed the establishment of a valid in vitro cell model and the activation of the Serpine1-p53-p16/p21 pathway by PAs.

To further dissect the role of Serpine1 in this pathway, we performed Serpine1 knockdown using small-interfering RNA targeting Serpine1. As shown in Fig. 5E, senescence-associated β-galactosidase staining revealed that PA-induced senescence in LSECs was significantly alleviated by Serpine1 knockdown, providing morphological evidence of endothelial senescence induction and confirming functional engagement of the senescence program. mRNA expression analysis of senescence-associated secretory phenotype genes (including p21 and p16), inflammatory cytokines (CCL2, CD14, P-selection, iNOS, IL-17, IL-1β, ICAM1, and VACM1) showed that, in PA-treated LSECs, the upregulation of p21, p16, CCL2, CD14, SELP, iNOS, IL-17, IL-1β, ICAM1, and VACM1 was notably reversed by Serpine1 knockdown (Fig. 5F), revealing the senescence-associated inflammatory state. Western blot analysis confirmed that the protein levels of Serpine1, p53, p21, and p16, which were elevated in PA-treated LSECs, were reduced after Serpine1 knockdown (Fig. 5G), establishing Serpine1 as a critical upstream driver of PA-induced endothelial senescence and dysfunction. As shown in Fig. S7B, Serpine1 overexpression further increased its levels (especially under PA stimulation). Collectively, these in vitro data demonstrate that PAs activate the Serpine1-p53 pathway to promote endothelial senescence, and Serpine1 knockdown can mitigate this process, highlighting the central role of Serpine1 in PA-induced LSEC senescence.

Silencing Serpine1 reverses PA-induced endothelial dysfunction in LSECs in vitro

To investigate whether Serpine1 mediates PA-induced endothelial dysfunction by activating the p53-p16/p21 pathway, we examined the effects of Serpine1 knockdown on key functional parameters of PA-treated LSECs. Consistent with prior observations, PA treatment caused significant G1 phase arrest in LSECs, as shown by flow cytometry and quantitative analysis. Notably, si-Serpine1 reversed this arrest, alleviating the abnormal distribution of cells across G1 and S phases, indicating restored cell cycle progression (Fig. 6A) and establishing Serpine1 as a critical mediator of PA-induced cell cycle checkpoint activation.

Fig. 6.

Serpine1 mediates LSEC dysfunction through cell cycle arrest, fenestration loss, and apoptosis in PA-HSOS.

(A) Representative flow cytometry histograms and quantification of propidium iodide-stained LSECs reveal G1/S arrest in PA-exposed LSECs, which are reversed by Serpine1 knockdown. (B) Scanning electron microscopy visualizes LSEC fenestrations in control and PA-treated conditions, with yellow arrows highlighting characteristic pores. Fenestration density (number per unit area) is significantly reduced in PA-HSOS LSECs but restored upon Serpine1 inhibition. Scale bar, 5 μm. (C) Immunofluorescence shows disrupted VE-cadherin organization in PA-treated LSECs, with reduced VE-cadherin expression partially recovered by Serpine1 knockdown. Scale bar, 100 μm. (D) In vitro permeability assay using 40 kDa FITC-dextran leakage quantifies barrier dysfunction in PA-exposed LSECs, demonstrating enhanced paracellular flux that reflects compromised endothelial integrity and is attenuated by Serpine1 knockdown. (E) Western blot analysis of endothelial function-related proteins (including eNOS and VE-cadherin) and apoptotic markers, including pro-apoptotic factors (cleaved caspase-3 and Bax) reveals downregulated endothelial function and activation of the apoptotic cascade in PA-treated LSECs, which is suppressed by Serpine1 inhibition. (F) Representative flow cytometry analysis delineates that PA-induced LSEC apoptosis is significantly reduced by Serpine1 knockdown. Data represent means ± SEM (n = 3/group). ∗p <0.05, ∗∗p <0.01, ∗∗∗p <0.001, ∗∗∗∗p <0.0001 by Mann-Whitney U test. n = 3/group independent experiments. LSEC, liver sinusoidal endothelial cell; PA, pyrrolizidine alkaloid; PA-HSOS, pyrrolizidine alkaloid–induced hepatic sinusoidal obstruction syndrome.

Structural integrity, a hallmark of endothelial function, was assessed via scanning electron microscopy. PA treatment induced LSEC defenestration (yellow arrows, Fig. 6B), which si-Serpine1 reversed by restoring fenestration area, demonstrating Serpine1-mediated sinusoidal endothelial dysfunction. Barrier function, critical for endothelial homeostasis, was evaluated through VE-cadherin expression and permeability assays. PA treatment decreased VE-cadherin fluorescence intensity (Fig. 6C) and increased dextran leakage (Fig. 6D), demonstrating enhanced paracellular flux that reflects compromised endothelial integrity. Silencing Serpine1 restored VE-cadherin levels and reduced dextran permeability, demonstrating recovered barrier function.

Molecular validation via western blot analysis showed that PA treatment downregulated endothelial function and upregulated apoptosis (Fig. 6E). These changes were reversed by si-Serpine1, confirming functional recovery at the protein level. Finally, apoptosis analysis via flow cytometry and quantification revealed that PA-induced LSEC apoptosis was significantly reduced by Serpine1 knockdown (Fig. 6F). Together, these findings demonstrate that silencing Serpine1 reverses PA-induced endothelial dysfunction in LSECs across multiple interconnected endpoints, establishing Serpine1 as a critical mediator of PA-induced endothelial cell death and confirming its central role in driving LSEC dysfunction across multiple pathways, including cell cycle, fenestration, and survival.

PAs maintain the protein stability of p53 by facilitating the binding between p53 and Serpine1

To clarify the regulatory relationship between Serpine1 and p53 in PA-treated LSECs, we performed a series of experiments to determine their upstream-downstream hierarchy and interaction mechanism. First, we investigated the protein level changes of p53 and Serpine1 after reciprocal knockdown. Western blot analysis showed that knocking down Serpine1 in PA-treated LSECs led to a decrease in p53 protein levels, while knocking down p53 had no significant effect on Serpine1 protein levels (Fig. 7A). At the mRNA level, knocking down p53 did not affect the expression of Serpine1, and knocking down Serpine1 also did not change the mRNA level of p53 (Fig. 7B). These results indicate that Serpine1 acts upstream of p53, and PA-induced Serpine1 upregulation is independent of p53. Additionally, Serpine1 regulates p53 at the protein level rather than the transcriptional level.

Fig. 7.

Molecular interaction between Serpine1 and p53 in PA-HSOS pathogenesis.

(A) Western blot analysis of p53 and Serpine1 in LSECs treated with si-Serpine1 or si-p53, showing reciprocal regulation of p53 and Serpine1 protein levels under PA exposure. (B) qPCR analysis showing that Serpine1 or p53 knockdown does not significantly alter p53 or Serpine1 mRNA levels in PA-treated LSECs. (C) Cycloheximide (CHX) chase assay combined with autophagy inhibitor (3-MA) or proteasome inhibitor (MG-132) demonstrating that Serpine1 stabilizes p53 protein by inhibiting proteasomal degradation. (D) Co-immunoprecipitation analysis of human liver samples showing enhanced interaction between Serpine1 and p53 in PA-HSOS compared with control tissue. (E) Immunofluorescence staining demonstrating subcellular colocalization of Serpine1 and p53 in LSECs under PA-HSOS conditions. Scale bar, 5 µm. (F) Co-immunoprecipitation analysis showing that Serpine1 interferes with the interaction between p53 and MDM2. (G) Ubiquitination assay demonstrating increased p53 polyubiquitination following Serpine1 knockdown, indicating enhanced p53 degradation. (H) Western blot analysis showing that Serpine1 knockdown promotes MDM2-dependent p53 degradation, whereas intact Serpine1 expression stabilizes p53 protein levels. Data represent means ± SEM (n = 3/group). LSEC, liver sinusoidal endothelial cell; PA-HSOS, pyrrolizidine alkaloid–induced hepatic sinusoidal obstruction syndrome; qPCR, quantitative PCR; si, small-interfering.

We next investigated the mechanism by which Serpine1 stabilizes p53 protein. Using cycloheximide (which blocks new protein synthesis)²⁵ combined with 3-MA (an autophagy inhibitor)²⁶ or MG-132 (a proteasome inhibitor),²⁷ we found that Serpine1 knockdown accelerated p53 degradation, and MG-132 but not 3-MA reversed this effect (Fig. 7C). This indicates that Serpine1 inhibits p53 degradation via the ubiquitin-proteasome pathway. Given that MDM2 is the most well-known E3 ubiquitin ligase of p53,²⁸ we explored whether Serpine1 modulates the p53-MDM2 interaction. Co-immunoprecipitation (Fig. 7D) and confocal microscopy (Fig. 7E) in PA-HSOS samples showed direct colocalization and binding between Serpine1 and p53, suggesting that Serpine1 may competitively inhibit p53-MDM2 interaction. Indeed, Serpine1 knockdown significantly enhanced p53-MDM2 binding (Fig. 7F) and increased p53 ubiquitination (Fig. 7G). Additionally, in Serpine1-knockdown LSECs, p53 levels decreased in the absence of MDM2 but partially recovered with its presence (Fig. 7H). These findings collectively demonstrate that Serpine1 maintains p53 protein stability by directly binding to it, thereby blocking MDM2-mediated ubiquitination and proteasomal degradation.

Validating Serpine1 as a PA-HSOS therapeutic target using TM5441

To translate the in vitro findings and validate Serpine1 as a therapeutic target for PA-HSOS, we employed TM5441, a Serpine1 inhibitor.¹⁸ We first established the chronic PA-HSOS model using the main components of Gynura segetum feeding from days 0-28, then treated mice with TM5441 from days 29-35 (Fig. 8A). Monitoring general health, TM5441-treated PA-HSOS mice exhibited body weight recovery (Fig. 8B), a reduction in ascites volume (Fig. S8A) and liver weight (Fig. S8B), with the liver/body weight ratio remaining unchanged (Fig. S8C), indicating improvement in systemic health status.

Fig. 8.

Therapeutic efficacy of TM5441 in ameliorating PA-HSOS.

(A) Schematic overview of the experimental design for PA-HSOS mouse model treatment with TM5441, a selective p53 activator. (B) Body weight trajectories demonstrate that TM5441 treatment significantly attenuates PA-induced weight loss. (C) Hemodynamic analysis reveals that TM5441 restores abnormal hemodynamic parameters, including PP, MAP, and PBF. (D) Comprehensive biochemical assessment demonstrates that TM5441 treatment ameliorates liver dysfunction (ALT, AST, ALP), coagulation abnormalities (PT, aPTT, D-dimer), systemic inflammation (IL-1β, TNF-α), and hematological dysregulation (platelets, RBC, WBC). (E) Histopathological analysis including H&E, Masson trichrome, IL-1β immunohistochemistry, and DAPI/CD31/VE-cadherin immunofluorescence demonstrates that TM5441 reduces sinusoidal RBC accumulation, hepatic fibrosis, inflammatory infiltration, and endothelial dysfunction. Black arrows indicate RBC accumulation within hepatic sinusoids. Scale bars, 200 μm (or 50 μm for enlarged regions). (F) FITC-dextran permeability assay with immunofluorescence visualization demonstrates that TM5441 restores sinusoidal barrier integrity and reduces vascular leakage. Scale bars, 100 μm (or 50 μm for enlarged regions). (G) qPCR profiling reveals that TM5441 suppresses inflammatory cytokine expression and reduces senescence-associated gene upregulation in liver tissue. (H) Western blot analysis of LSEC function and apoptotic markers shows that TM5441 reduces pro-apoptotic protein levels and restores anti-apoptotic factors. (I) Flow cytometry demonstrates that TM5441 significantly reduces apoptosis rates in hepatic endothelial cells. (J) Western blot of the Serpine1-p53 pathway reveals that TM5441 downregulates Serpine1 and p53 protein levels alongside reduced expression of downstream senescence effectors, including p21 and p16. (K) Immunofluorescence of ERG and p16 in hepatic tissue reveals reduced coexpression in TM5441-treated mice. Scale bar, 50 μm. (L) Flow cytometry-based cell cycle analysis demonstrates that TM5441 reverses PA-induced G1/S phase arrest in hepatic endothelial cells. Data represent means ± SEM (n = 6/group). ∗p <0.05, ∗∗p <0.01, ∗∗∗p <0.001, ∗∗∗∗p <0.0001 by Mann-Whitney U test. ALP, alkaline phosphatase; ALT, alanine aminotransferase; aPTT, activated partial thromboplastin time; AST, aspartate aminotransferase; LSEC, liver sinusoidal endothelial cell; MAP, mean arterial pressure; PA-HSOS, pyrrolizidine alkaloid–induced hepatic sinusoidal obstruction syndrome; PBF, portal blood flow; PP, portal pressure; PT, prothrombin time; qPCR, quantitative PCR; RBC, red blood cell; WBC, white blood cell.

Assessing physiological and biochemical parameters, TM5441 treatment reversed hemodynamic defects, including decreased portal pressure, along with increased MAP and PBF (Fig. 8C). Biochemical analysis of blood further confirmed the reversal of alterations in liver dysfunction (ALT, AST, and ALP), coagulation abnormalities (PT, aPTT, and D-dimer), systemic inflammation (IL-1β and TNF-α), and hematological dysregulation (platelets, RBCs, and white blood cells), indicating broad therapeutic benefit across multiple organ systems (Fig. 8D and S8D). Additionally, serum ELISA revealed that TM5441 treatment also reduced the concentration of Serpine1 (Fig. S8E).

Histological analysis via H&E staining revealed that TM5441 alleviated liver tissue damage in PA-HSOS mice, reducing inflammatory cell infiltration and improving tissue architecture. Immunohistochemistry for IL-1β, Masson, and VE-cadherin revealed reduced inflammation and restored collagen deposition, indicating improved endothelial barriers (Fig. 8E). TM5441 also restored endothelial permeability in PA-HSOS mice, as evidenced by reduced FITC-Dextran extravasation (Fig. 8F).

qPCR further showed that TM5441 treatment downregulated the mRNA levels of Serpine1 (Fig. S8F), along with senescence-related genes (including p21 and p16) and pro-inflammatory genes (including CCL2, CD14, SELP, iNOS, IL-17, IL-1β, ICAM1, and VCAM1), consistent with reduced senescence and inflammation (Fig. 8G). Molecular and cellular function assays reinforced these findings. Western blot showed TM5441 restored endothelial function and reduced pro-apoptotic markers (Fig. 8H). Flow cytometry (Fig. 8I) and TUNEL staining (Fig. S8G) revealed decreased apoptosis in TM5441-treated PA-HSOS mice. Additionally, Western blot analysis of key pathway proteins (Fig. 8J) confirmed that TM5441 downregulated Serpine1 and its downstream p53-mediated senescence targets (including p53, p21, and p16), validating on-target Serpine1 inhibition. Consistent with p16 downregulation in the p53 pathway, TM5441-treated PA-HSOS mice had fewer p16-ERG⁺ cells, confirming endothelial senescence reversal (Fig. 8K and S8H). While cell cycle analysis showed a normalized distribution, indicating improved cell survival and proliferation (Fig. 8L). Additionally, co-culture experiments showed that TM5441 also reversed PA-induced Serpine1/p53 upregulation (Fig. S8I and J). These results indicate that TM5441-mediated Serpine1 inhibition alleviates PA-HSOS by restoring mice’s physical and hepatic function, repairing endothelial barriers, and regulating the Serpine1-p53 pathway to reduce senescence, inflammation, and apoptosis.

Discussion

The lack of standardized animal models that accurately replicate the chronic progression of PA-HSOS has long hindered mechanistic research and therapeutic development for this life-threatening condition. This gap is particularly critical given the substantial clinical burden of PA-HSOS – predominantly linked to Gynura segetum ingestion in China² – and its growing global incidence associated with herbal medicine use, which is likely underestimated due to non-specific clinical presentation and unreported herbal intake[3], [4], [5], [6]. Our study addresses this critical gap by establishing a clinically relevant 28-day murine model using the primary toxic components of Gynura segetum, including seneciphylline, senecionine, and their N-oxide derivatives. This model faithfully recapitulates hallmark features of human PA-HSOS, including elevated liver enzymes, coagulation dysfunction, hepatomegaly, sinusoidal fibrosis, and LSEC dysfunction, thereby providing a robust platform for investigating disease pathogenesis and testing potential therapies.

Notably, our PA-HSOS model overcomes key limitations of the well-established MCT-induced model, which has been widely used but carries inherent drawbacks. MCT, a PA primarily associated with acute pulmonary vascular toxicity,[29], [30], [31] fails to replicate the hepatic sinusoidal specificity and chronicity that define human PA-HSOS. In contrast, by incorporating the PA mixture from the leading etiology of PA-HSOS in China, our model better reflects the metabolic activation and chronic liver damage observed in clinical settings. In vivo, our 28-day exposure protocol aligns with the delayed onset of human PA-HSOS (symptoms typically emerging after ≥1 month of toxic plant consumption), as opposed to MCT models that induce acute injury within 1–3 days,[12], [13], [14] a time course that poorly reflects the gradual disease progression observed clinically. Additionally, our model prioritizes liver-centric pathology, which aligns with the hepatic focus of human PA-HSOS. In contrast, MCT is widely reported to cause dominant pulmonary toxicity,^29,30,32 an off-target effect that often confounds the interpretation of hepatic injury in MCT-based studies. Complementing these advantages, our in vitro Transwell co-culture system further recapitulates human PA metabolism: the Gynura segetum-derived PAs in this system require hepatocyte metabolism (via CYP enzymes) to form toxic dehydropyrrolizidine alkaloids that then act on LSECs, mirroring the human bioactivation pathway – unlike MCT, which directly injures LSECs without relying on hepatocyte-mediated conversion.[12], [13], [14] Together, this alignment with clinical metabolic processes, chronic time course, and hepatic pathological focus makes our model a more clinically relevant tool for dissecting PA-HSOS pathogenesis and evaluating targeted therapies.

Histologically, PA exposure in our model led to lung (RBC leakage) and stomach (tissue swelling) involvement. These changes align with prior reports of PA-induced systemic vascular effects³³, but their clinical relevance requires further validation, given that pulmonary CT and gastroscopy are not routine for PA-HSOS, limiting clinical data on extrahepatic manifestations. Stomach changes in the model may also reflect gavage-related irritation, and direct correlation with human gastric pathology awaits future investigation. Further enhancing translational relevance, the model aligns with the Nanjing criteria for PA-HSOS diagnosis and recapitulates core pathophysiological features (e.g. weight gain, bilirubin elevation, ascites) consistent with international HSOS standards (Seattle/Baltimore criteria)^34,35, demonstrating its capacity to capture core HSOS pathophysiology across etiologies.

Transcriptomic profiling of liver tissues from our PA-HSOS model identified Serpine1 as the most significantly upregulated gene. This finding is particularly compelling given Serpine1’s known roles in fibrinolysis regulation,³⁶ endothelial dysfunction, and fibrosis,³⁷ and extends its relevance to PA-HSOS pathogenesis. Our data demonstrate that Serpine1 upregulation is tightly linked to the activation of the p53-p21/p16 pathway, a critical axis in senescence and apoptosis.^38,39 Co-immunoprecipiation and confocal microscopy confirm a physical interaction between Serpine1 and p53 in LSECs. Notably, while commonalities with previously reported acute models using MCT were observed, such as the upregulation of MMP9 and MMP2¹⁴ and downregulation of the ADAMTS13,⁴⁰ these genes did not rank within the top 20 DEGs in our induced chronic model. This discrepancy highlights the distinct molecular landscape of chronic PA-HSOS compared to acute MCT-induced models.

Mechanistically, we show that Serpine1 acts upstream of p53, maintaining p53 protein stability by directly binding to it and blocking MDM2-mediated ubiquitination and proteasomal degradation. This interaction drives LSEC senescence, as evidenced by elevated p21 and p16 levels, and exacerbates endothelial dysfunction through increased permeability, reduced LSEC function, and heightened apoptosis. These effects collectively contribute to sinusoidal obstruction, a hallmark of PA-HSOS. The Serpine1-p53 axis may also promote microthrombosis by disrupting fibrinolytic balance, given Serpine1’s role as an inhibitor of plasminogen activators,³⁶ further amplifying liver damage.

The therapeutic potential of targeting Serpine1 was validated using TM5441, a small-molecule inhibitor of Serpine1.^41,42 In our PA-HSOS model, TM5441 treatment significantly improved liver function (reduced ALT, AST, and bilirubin levels), normalized coagulation parameters (PT, APTT, and D-dimer), and mitigated systemic inflammation (decreased TNF-α and IL-1β). Histopathologically, TM5441 reduced sinusoidal congestion, collagen deposition, and inflammatory cell infiltration, while restoring the integrity of LSECs (increased VE-cadherin expression) and their barrier function (reduced endothelial permeability). These improvements were associated with downregulation of the Serpine1-p53-p21/p16 cascade, confirming on-target inhibition and validating Serpine1 as a therapeutic target. These results align with prior studies showing TM5441’s efficacy in ameliorating vascular senescence and thrombosis in non-hepatic models.^41,43,44 The inhibitor’s ability to reverse hemodynamic abnormalities (portal hypertension and reduced portal blood flow) further supports its potential as a multifaceted therapeutic agent for PA-HSOS.

Despite these advancements, our study has limitations. First, while our targeted LC-MS of 37 fatty acids in hepatocytes identified alterations in the profiles of saturated and monounsaturated fatty acids, suggesting that PAs may affect fatty acid synthase and stearoyl-CoA desaturase activities, species-specific differences in PA metabolism remain a consideration. In particular, cytochrome P450 enzymes (e.g. human CYP3A4 vs. murine CYP3A isoforms⁴⁵) exhibit differences in substrate specificity, which could alter the profile of toxic metabolites generated and limit direct translational relevance to human PA-HSOS. Second, our co-culture system, while innovative in mimicking hepatocyte-LSEC interactions, simplifies the liver’s multicellular complexity by omitting contributions from Kupffer cells and hepatic stellate cells, which play critical roles in fibrosis and inflammation.⁴⁶ Finally, the current model is tailored for PA-induced HSOS, and its applicability to SOS from other etiologies (e.g. chemotherapeutic agents) remains to be explored; future studies could test whether the Serpine1-p53 pathway is conserved across SOS subtypes to expand translational relevance.

In conclusion, our study establishes a clinically relevant chronic PA-HSOS model and identifies Serpine1 as a pivotal therapeutic target. Targeting the Serpine1-p53 pathway with TM5441 presents a promising strategy for mitigating endothelial injury and fibrosis in PA-HSOS, with implications for enhancing patient outcomes. These findings underscore the importance of chronic models in studying drug-induced liver injury and provide a foundation for future translational research, including the need for human studies to address species-specific differences and validate Serpine1 inhibition as a clinical strategy.

Abbreviations

ALP, alkaline phosphatase; ALT, alanine aminotransferase; aPTT, activated partial thromboplastin time; AST, aspartate aminotransferase; CMC-Na, carboxymethylcellulose sodium; DEGs, differentially expressed genes; ERG, ETS-related gene; HSOS, hepatic sinusoidal obstruction syndrome; HVPG, hepatic venous pressure gradient; LSEC, liver sinusoidal endothelial cell; MAP, mean arterial pressure; MCT, monocrotaline; PA-HSOS, pyrrolizidine alkaloid-induced hepatic sinusoidal obstruction syndrome; PAs, pyrrolizidine alkaloids; PBF, portal blood flow; PT, prothrombin time; RBC, red blood cell.

Authors' contributions

JMZ and HG conceived the project and designed the experiments. YFM, ZRQ, and ZYG performed most of the experimental work. JYY and MYJ contributed to part of the molecular and cell experiments. SNC contributed to part of the animal experiments. YFM wrote the first version of the manuscript. JMZ and HG are the guarantors of this work and, as such, had full access to all the data in the study and took responsibility for the data’s integrity and the accuracy of the data analysis. All authors agreed to be accountable for the work and ensured the integrity of the research.

Data availability

Data are available upon request.

Financial support

This study was funded by Zhongshan Hospital, Fudan University (2020ZSLC24), and Shanghai Science and Technology (21Y11921800).

Conflict of interest

The authors declare no conflicts of interest.

Please refer to the accompanying ICMJE disclosure forms for further details.

Supplementary data

The following are the Supplementary data to this article: