A pig model of human radiation-induced veno-occlusive liver disease reveals ferroptosis as a therapeutic target Authors: Zhen Sun, Xinjie Zhang, Qiang Zheng, Baihua Wu, Gonglin Fan, Guozhi Shi, Junjie Nan, Enjie Zhou, Yiheng Lin, Weijing He, Sihan Zhao, Jingqi Wu, Shengyao Guan, Songyi Li, Mingyu Chen, Fanzheng Meng, Lianxin Liu, Lijian Hui, Xiujun Cai, Chandan Guha, Yifan Wang, Ludi Zhang Published: 2026 ABSTRACT SummaryRadiation-induced liver disease (RILD) poses a major clinical challenge in radiotherapy, transplantation preconditioning, or radiation accidents, yet its pathogenesis is poorly understood due to limited animal models. Here, we establish a translational pig model recapitulating human RILD pathology within 4 weeks post-40 Gy irradiation, featuring veno-occlusive disease (VOD) and centrilobular necrosis. Single-cell atlas analyses identify ferroptosis as a key driver of hepatocyte death during RILD initiation. Ferroptosis inhibition with liproxstatin-1 (Lip-1) not only prevents RILD progression but also reverses histological damage and restores liver function. Mechanistically, Lip-1 treatment restores dysregulated gene expression profiles, particularly associated with hepatocyte ferroptosis, while stimulating hepatic regeneration via coordinated proliferation of hepatocytes and endothelial cells. Our findings establish ferroptosis inhibition as a therapeutic strategy for RILD, demonstrating its dual role in cytoprotection and regeneration. This large animal model provides a robust platform to optimize radiotherapy regimens, improve transplant conditioning, and develop targeted radioprotectants.Keywords: radiation-induced liver disease, veno-occlusive disease, miniature pigs, ferroptosis, translational model Introduction Radiation therapy (RT) is a cornerstone of cancer treatment and transplantation medicine, yet its efficacy is limited by its toxicity to radiosensitive organs.1,2,3 Due to its radiosensitive nature and substantial organ volume, the liver is at high risk of radiation-induced liver disease (RILD), posing a significant challenge in RT.4,5 The liver is additionally vulnerable to radiation during preconditioning for hematopoietic stem cell and hepatocyte transplantation, as well as in nuclear accidents.6,7,8 RILD occurs in 44% of patients who receive >35 Gy whole-liver irradiation, with a mortality rate of 10%–76%.4,9 Unfortunately, no pharmacological therapies exist for RILD, largely because of the lack of a molecular understanding of its pathogenesis. To date, studies on RILD have relied on clinical observations and the histopathology of patients with terminal RILD.10 RILD typically occurs 4–8 weeks after RT, and patients present with fatigue, ascites, and increased alkaline phosphatase (ALP).4 In addition, patients with underlying liver diseases develop severe hepatic dysfunction, presenting with jaundice and markedly elevated serum transaminase levels.4 The histopathological hallmark of RILD is hepatic veno-occlusive disease (VOD), also known as sinusoidal obstruction syndrome, which is characterized by the obliteration of the central vein by erythrocytes trapped in a mesh of reticulin and collagen fibers.11 This results in centrilobular necrosis and hepatic stellate cell (HSC) activation, which typically progress toward liver fibrosis or even deteriorate into liver failure.11 However, elucidating the pathogenesis of RILD using patient samples is not feasible, primarily because of the scarcity of liver samples from the early stages of the disease. Thus, developing animal models for RILD is essential for studying its pathogenesis, evaluating drug efficacy, and optimizing RT regimens.10 Significant differences in liver radiosensitivity have been observed across species. Currently available animal models, including rodents, rabbits, dogs, and rhesus monkeys, do not fully recapitulate many key pathological characteristics of human RILD,10,12,13,14,15,16 especially VOD. In contrast, cynomolgus monkeys receiving 40 Gy of whole-liver irradiation have been reported to exhibit marked veno-occlusive changes in the central veins 10 weeks after treatment.17 Nonetheless, liver injury in these animals was moderate, as evidenced by mild liver fibrosis and the absence of centrilobular necrosis, indicating a failure to model the severe form of RILD. In addition, the cost and accessibility of monkey models hinder their broader application. New insights into the molecular mechanisms underlying RILD would facilitate the development of effective treatment strategies. Previous studies have focused on understanding the mechanisms of radiation-induced liver fibrosis, the primary histological change observed in irradiated rodents. TGF-β and Hedgehog (Hh) pathways are known to be involved in hepatic fibrogenesis after irradiation by promoting the activation of HSCs and collagen deposition.14,18 However, these rodent models did not develop VOD, and the mechanisms underlying the development of RILD, and more specifically VOD, remain elusive. Pigs are commonly used as large animal models owing to their physiological, metabolic, and anatomical similarities to humans.19,20 Specifically, they share notable parallels with humans in terms of liver weight, hepatic segmentation, liver enzyme profiles, serum albumin levels, and other key physiological parameters.21 Furthermore, pigs with disruptions in the CFTR (cystic fibrosis transmembrane conductance regulator) and FAH (fumarylacetoacetate hydrolase) genes develop human-like pathological phenotypes, such as progressive liver injury accompanied by severe fibrosis, that are not recapitulated in mouse models,20,22 likely due to species-specific differences in liver metabolism.23 In addition, their size enables the use of clinically relevant interventions and procedures.24,25 These advantages, combined with the model’s cost-effectiveness and ethical acceptability, led us to select the pig for our study. In this study, we performed a dose-escalation study of whole-liver irradiation in miniature pigs, establishing a pig model replicating the clinical and histological features of severe RILD, particularly VOD and hepatic necrosis. Furthermore, our single-cell atlas analysis revealed significant ferroptosis, a unique type of programmed cell death driven by iron-dependent lipid peroxidation26 that occurs in hepatocytes early in RILD. Notably, treatment with the ferroptosis inhibitor liproxstatin-1 (Lip-1) significantly prevented the development of RILD and improved liver regeneration. Overall, our pig model is crucial for understanding the mechanisms underlying RILD and for evaluating therapeutic agents aimed at reducing radiation-induced hepatic toxicity and improving patient survival. Results Hepatic irradiation induces the characteristic manifestations of RILD in pigs Given that 30 Gy is regarded as the clinical maximum safe dose,4 a dose-escalation study (20, 30, and 40 Gy) was designed to determine liver radiosensitivity in miniature pigs (Figure 1A). The onset of human RILD is typically observed as early as 28 days after irradiation. Thus, we monitored the irradiated pigs for 35 days to assess hepatic toxicity by conducting weekly evaluations of blood biochemistry, body weight, and appetite, as well as histological analysis of biopsy samples (Figure 1A). To mimic the clinical scenario, the total dose was precisely delivered using hypofractionated stereotactic body RT (SBRT) in five consecutive daily fractions without causing lethal adverse effects on the surrounding organs (Figures 1A–1C and S1A–S1E). HIR induces the characteristic manifestations of RILD in pigs (A) Schematic outline for monitoring the irradiated pigs for 35 days to evaluate hepatic toxicity. (B) Model diagram of applying SBRT to pigs. This plot is created with BioRender.com. (C) Radiation dose of volume histogram. (D) Survival curve of pigs receiving different radiation doses (n = 3 per group). (E) Dynamic changes in body weight of pigs after irradiation (n = 3 per group). (F) Dynamic changes in appetite of pigs after irradiation (n = 3 per group). (G–N) Dynamic changes of blood index of pigs after irradiation. Serum ALP levels (G), serum AST (H), TBIL (I), TBA (J), ALB (K), ALT (L), INR (M), and ammonia (N) are detected (n = 3 per group). Scale bars: 100 μm. Data are presented as mean ± SEM; The p values represent an unpaired, two-tailed t test comparing day 28 to day 1 (baseline) within the 40 Gy group. Pigs receiving 20 Gy (n = 3) survived for 35 days and maintained their body weight (Figures 1D and 1E). Their appetite decreased transiently on day 21 and recovered quickly by day 28 (Figure 1F). Compared with the baseline, ALP and aspartate aminotransferase (AST) levels increased significantly by 2.2- and 5.4-fold, respectively, on day 35 (Figures 1G and 1H). Other liver function indices, including total bilirubin (TBIL), total bile acids (TBAs), serum albumin (ALB), alanine aminotransferase (ALT), international normalized ratio (INR), and ammonia, were comparable to baseline levels on day 35 (Figures 1I–1N). Necropsy showed that the pigs did not develop ascites (Figure S1F). Pigs receiving 30 Gy (n = 3) exhibited phenotypes similar to those receiving 20 Gy (Figures 1D–1N and S1F), suggesting that both 20 and 30 Gy induced moderate liver injury that was tolerated by the pigs. When treated with 40 Gy (n = 3), pigs exhibited appetite loss starting on day 21, followed by body weight loss and the development of fatigue and jaundice on day 28 (Figures 1E and 1F). One of the three pigs died on day 33, and the other two were moribund on day 35 (Figure 1D). Consistent with these symptoms, ALP, AST, TBIL, and TBA levels drastically increased on day 28 (4.5-, 33.0-, 29.6-, and 9.5-fold higher than the baseline, respectively) and remained elevated on day 35, suggesting significant liver injury (Figures 1G–1J). The mean increase in TBIL was marked, though it exhibited high inter-individual variability. Moreover, ALB levels decreased by 37.1% on day 35, indicating massive hepatocyte loss (Figure 1K). The INR and ammonia levels remained within the normal range, while ALT levels were slightly elevated in the irradiated pigs (Figures 1L–1N). Furthermore, necropsy showed ascites development (Figure S1F). Collectively, these symptoms develop as early as 28 days and resemble the clinical features observed in patients with severe RILD. Taken together, these results suggest that hepatic irradiation (HIR)-induced damage to pig livers is dose dependent, with high-dose HIR causing lethal liver dysfunction in pigs, resembling a more severe form of RILD. HIR induces VOD in pigs VOD is a hallmark of RILD and has not been fully recapitulated in animal models.10 To investigate whether HIR induces VOD in pigs, we performed hematoxylin and eosin (H&E) and Masson’s trichrome staining of endpoint liver samples. In the 20 and 30 Gy groups, we observed significant histological changes in the central vein, including prominent endothelium, subendothelial inflammation, and moderate fibrosis extending to the surrounding sinusoids, while their lumens remained unoccluded (Figure 2A). Notably, pigs receiving 40 Gy showed pronounced veno-occlusive changes in the central vein, characterized by subintimal edema, red blood cell extravasation, and extensive collagen deposition, which completely obstructed the central vein (Figure 2A). These results suggest that 40 Gy can induce VOD in pigs; therefore, we focused on this dose to analyze VOD-related pathological characteristics. HIR induces veno-occlusive liver disease in pigs (A) Representative images of H&E and Masson’s trichrome staining in irradiated livers 35 days post-irradiation. (B–G) Representative images and quantification of immunofluorescence staining of LYVE1 (B and C), VE-cadherin (D and E), and CD31 (F and G) in irradiated livers under 40 and 0 Gy 35 days post-irradiation (n = 3 per group). (H) GSEA analysis showing the enrichment of genes associated with abnormal bleeding and hypoxia in 40 versus 0 Gy. Left: genes higher in 0 Gy; right: genes higher in 40 Gy. (I and J) Representative images of H&E staining of irradiated livers and quantification of necrosis area under 40 and 0 Gy 35 days post-irradiation (n = 3 per group). (K–N) Representative images and quantification of immunofluorescence staining of GS (K and L) and MRP2 (M and N) in irradiated livers under 40 and 0 Gy 35 days post-irradiation (n = 3 per group). (O) GSEA analysis showing the enrichment of genes associated with cell death and CYP metabolism in 40 versus 0 Gy. Left: genes higher in 0 Gy; right: genes higher in 40 Gy. (P–S) Representative images and quantification of immunofluorescence staining of α-SMA (P and Q) and collagen I (R and S) in irradiated livers under 40 and 0 Gy 35 days post-irradiation (n = 3 per group). (T and U) Representative images and quantification of Masson’s trichrome staining in irradiated livers under 40 and 0 Gy 35 days post-irradiation (n = 3 per group). Liver lobules in normal pigs are delineated by collagen-rich connective tissue. Necrosis is not detected in the 0 Gy controls. (V) GSEA analysis showing the enrichment of genes associated with ECM and fibrosis in 40 versus 0 Gy. Left: genes higher in 0 Gy; right: genes higher in 40 Gy. Scale bars: 100 μm. Data are presented as mean ± SEM; p values were assessed using unpaired two-tailed Student’s t test. Radiation-induced damage to liver sinusoidal endothelial cells (LSECs) and the central vein endothelium is considered the primary cause and major pathological feature of VOD.10,27 Immunofluorescence staining for the sinusoidal endothelial markers, including lymphatic vessel endothelial hyaluronan receptor 1 (LYVE1) and vascular endothelial cadherin (VECAD), revealed a 2.4-fold decrease in hepatic sinusoidal density and sinusoidal dilatation in the 40 Gy group compared with the non-irradiated controls (Figures 2B–2E). These results were further confirmed by CD31 staining (Figures 2F and 2G). Moreover, co-staining for the endothelial marker CD31 and the pericentral hepatocyte marker GS showed that the endothelium of the central vein was nearly absent (Figure S2A). To further identify genes and signaling pathways exhibiting altered expression related to VOD, we performed bulk tissue RNA sequencing (RNA-seq) analysis on irradiated livers from the 40 Gy group and compared them with non-irradiated livers. Gene set enrichment analysis (GSEA) showed enrichment of the abnormal bleeding (e.g., F2) and hypoxia (e.g., HIF1A) pathways in the 40 Gy group (Figures 2H, S2B, and S2C), consistent with previous findings that radiation-induced endothelial damage, followed by hemorrhage into the enlarged subendothelial space, is the main cause of outflow obstruction and induces a hypoxic environment.11 Centrilobular hepatocyte death triggered by a hypoxic environment is another significant pathological feature of VOD that has not been observed in other animal models of RILD.10 In pigs treated with 40 Gy, we observed foci of yellow necrosis in the central regions of the liver lobules, covering 10.2% of the parenchyma, as evidenced by H&E staining (Figures 2I and 2J). Significant neutrophil infiltration was detected by immunofluorescence staining for MPO (Figures S2D and S2E). More importantly, a 16.4-fold decrease in the area of GS+ hepatocytes compared with non-irradiated livers further demonstrated a significant loss of centrilobular hepatocytes (Figures 2K and 2L). These findings are consistent with the dramatically elevated AST levels and decreased ALB levels observed in the 40 Gy group. Given the significantly elevated levels of ALP and TBAs, we performed MRP2 staining to investigate radiation-induced bile canaliculus injury in hepatocytes. Unlike the contiguous and radial bile canaliculi in non-irradiated livers, the MRP2+ bile canaliculi were disorganized, and their density was reduced by 3.1-fold in the 40 Gy group, suggesting severe damage to the bile canaliculus of hepatocytes (Figures 2M and 2N). Consistent with the staining data, GSEA revealed significant enrichment of the necrosis pathway (e.g., CASP3 and XIAP) in the 40 Gy group (Figures 2O and S2F). By contrast, the CYP pathway genes (e.g., CYP8B1 and CYP27A1), which are predominantly expressed in pericentral hepatocytes,28 were significantly downregulated (Figures 2O and S2G), further supporting the presence of centrilobular necrosis. In addition to endothelial cell and hepatocyte damage, patients with VOD show HSC activation and progressive liver fibrosis, which are associated with a poor prognosis. To determine whether 40 Gy induced liver fibrosis, we performed immunofluorescence staining of α-SMA, a marker of activated HSCs, and found a significant increase in the number of activated HSCs in irradiated livers (Figures 2P and 2Q). Moreover, excessive extracellular matrix (ECM) accumulation was demonstrated by immunofluorescence staining of collagen I and fibronectin I (Figures 2R, 2S, S2H, and S2I). Additionally, Masson’s trichrome staining revealed that the collagen volume fraction in the 40 Gy group was 4.2-fold higher than that in the normal liver (22.2% vs. 5.3%) (Figures 2T and 2U), indicating marked collagen deposition and fibrosis after 40 Gy irradiation. Consistent with the staining data, GSEA revealed significant enrichment of the ECM (e.g., COL1A1, COL5A1, MMP2, and MMP7) and fibrosis pathways (e.g., FN1, ACTA2, ITGA3, ITGA4, LAMA3, and TIMP1) in the 40 Gy group (Figures 2V and S2J). Taken together, these results indicate that pigs exposed to 40 Gy HIR develop pathological features resembling those of human VOD, accompanied by endothelial cell damage, massive centrilobular necrosis, and liver fibrosis. High-dose HIR impairs liver regeneration capacity in pigs The liver has a remarkable regenerative capacity upon injury. However, a substantial proportion of patients with RILD will die of liver failure, possibly due to impaired liver regeneration.4,29 However, a single low dose (14 Gy) of HIR reportedly induces hepatocellular proliferation in pigs.24 In the present study, we observed that a dose of 40 Gy induced lethal RILD, whereas 30 Gy, which is considered the clinically maximum safe dose, caused moderate liver injury but was tolerable. Thus, we assessed differences in liver regeneration in pigs receiving doses of 20, 30, and 40 Gy. Immunostaining for phosphorylated retinoblastoma protein (pRb), a marker of the G1-S-phase transition, showed that 22.7% and 32.4% of pRb-positive cells were observed in pigs receiving 20 and 30 Gy, respectively (Figures 3A and 3B). However, only 5.3% of pRb-positive cells were detected in pigs receiving 40 Gy (Figures 3A and 3B). Moreover, Ki67 and HNF4A co-staining demonstrated that 2.9% and 6% of hepatocytes proliferated in response to liver injury induced by 20 and 30 Gy, respectively (Figures 3C and 3D). Compared with the 20 and 30 Gy groups, only 1.2% of Ki67-positive hepatocytes were observed in pigs receiving 40 Gy (Figures 3C and 3D). In addition, expression profile analysis and GSEA revealed the downregulation of cell proliferation-associated genes in the 40 Gy group compared with the 20 Gy group (Figures 3E, S3A, and S3B). Collectively, these results indicated that 40 Gy HIR impairs hepatocyte proliferation in response to irradiation injury. High-dose HIR impairs liver regeneration capacity in pigs (A–D) Representative images and quantification of immunofluorescence staining of pRb (A and B), HNF4A, and Ki67 (C and D) in irradiated livers 35 days post-irradiation (n = 3 per group). (E) The heatmap of expression of genes involved in cell proliferation and senescence-associated secretory phenotype (SASP) genes after irradiation. (F–I) Representative images and quantification of immunofluorescence staining of γH2A.X (F and G), HNF4A, and P21 (H and I) in irradiated livers 35 days post-irradiation (n = 3 per group). (J–M) Representative images and quantification of co-staining of HNF4A and P21 (J and K) and staining of SA-β-Gal (L and M) in irradiated livers under 20 and 40 Gy within 28 days post-irradiation (n = 3 per group). Scale bars: 100 μm. Data are presented as mean ± SEM; p values were assessed using unpaired two-tailed Student’s t test. We next sought to investigate the mechanism by which high-dose HIR impairs liver regeneration. Radiation can induce numerous DNA lesions. Notably, 62.5% of cells showed activation of DNA damage response 35 days after receiving 40 Gy HIR, 8.6-fold higher than in the 20 Gy group, evidenced by immunostaining of γH2A.X, a DNA damage marker (Figures 3F and 3G). In line with γH2A.X staining, GSEA showed enrichment of the DNA damage pathway in the 40 Gy group (Figures S3C and S3D), whereas the DNA repair pathway was enriched in the 20 Gy group (Figures S3E and S3F). Cellular senescence occurs when DNA damage is not successfully repaired.30 Remarkably, 21.6% of hepatocytes were in cell-cycle arrest or senescence in pigs receiving 40 Gy, as shown by co-immunostaining for P21 and HNF4A (Figures 3H and 3I). Moreover, senescence-associated secretory phenotype-associated genes, such as IL6, IL10, IL18, CCL26, TNFRSF1B, and CXCL2, were significantly upregulated under the 40 Gy condition (Figures 3E and S3G). Meanwhile, GSEA revealed enriched DNA damage-induced senescence pathways (Figure S3H). Taken together, these results suggest that high-dose irradiation impairs liver regeneration through severe DNA damage and extensive hepatocyte senescence. Potential treatment window for RILD Current treatments for RILD primarily offer supportive care and do not alter natural disease progression.4,11 Due to the potentially fatal nature of RILD, early intervention is crucial. We hypothesized that the significant impairment in liver regeneration could be a critical point for effective treatment. To this end, we analyzed cellular senescence in liver biopsy samples from the 40 Gy group on days 1, 7, 14, 21, and 28. The 20 Gy group was included as a control because of the occurrence of liver regeneration and limited cellular senescence. P21 and HNF4A co-staining revealed hepatocyte cell-cycle arrest or senescence as early as day 1, which became more pronounced by day 14 in the 40 Gy group compared to the 20 Gy group (Figures 3J and 3K). Similarly, SA-β-Gal staining demonstrated significant cellular senescence from day 14 onward in the 40 Gy group (Figures 3L and 3M). These findings suggest that day 14 is a critical time point for studying molecular mechanisms and that the optimal treatment window for RILD may be within 14 days post-irradiation. To delineate the kinetics of cellular dysfunction, we performed an analysis of serial liver biopsy samples. A significant elevation of P21, a marker of senescence, was observed in both hepatocytes as early as day 1 (Figures 3J and 3K) and LSECs by day 7 post-irradiation (Figures S4A and S4B) in the 40 Gy group compared to the 20 Gy group. In contrast, activation of HSCs, reflected by an increasing yet statistically non-significant trend in α-SMA-positive cells, was observed from day 7 onward (Figures S4C and S4D). These data suggest that hepatocytes and LSECs may be the primary early responders to radiation. snRNA-seq analyses implicate hepatocyte ferroptosis in pigs with RILD To comprehensively understand the pathogenesis and molecular mechanisms underlying RILD in the early stages, we performed single-nucleus RNA sequencing (snRNA-seq) on liver biopsy samples from the 40 Gy group on day 14 and non-irradiated controls (Figure 4A). A total of 9,432 and 15,217 cells were captured from 0 to 40 Gy samples, respectively, for library construction and paired-end sequencing, and 87.6% and 91.6%, respectively, passed quality control (STAR Methods). Uniform manifold approximation and projection dimensionality reduction and sub-clustering of these cellular transcriptomes showed that our approach captured the main cell types that comprise the porcine liver (Figures 4B and 4C). The 15 cell clusters were identified as 11 cell types (Figures S5A and S5B), including hepatocytes, cholangiocytes, LSECs, HSCs, natural killer (NK) cells, Kupffer cells (KCs), dendritic cells (DCs), T cells, B cells, fibroblasts (FBs), and proliferating cells (Figures S5C and S5D), according to previously reported markers. These results suggest that our single-cell atlas effectively captures the predominant cell types in high-dose radiation liver biopsy samples as well as their transcriptomic diversity during the early stage of RILD pathology. snRNA-seq analyses implicate hepatocyte ferroptosis in pigs with RILD (A) Schematic illustration of the experimental workflow. This plot is created with BioRender.com. (B and C) Dimension-reduction presentation of the combined single-cell transcriptome data from 0 to 40 Gy samples 14 days post-irradiation (n = 2 per group). Uniform manifold approximation and projection (UMAP) plot showing major cells. (D–I) Heatmap of marker genes and GO term enrichment in the endothelial cell cluster (D–F) and the hepatocyte cluster (G–I). (J) The bubble map showing that ferroptosis-related genes were enriched in 0 and 40 Gy samples. (K–N) Representative images and quantification of staining of 4-HNE (K and L) and Perls (M and N) in irradiated livers under 0 and 40 Gy 35 days post-irradiation (n = 3 per group). 4-HNE+ cells and Perls+ cells are not detected in the 0 Gy controls. Scale bars: 100 μm. (O–R) Representative images and quantification of immunofluorescence staining of SLC7A11 (O and P) and GPX4 (Q and R) in irradiated livers under 0 and 40 Gy 35 days post-irradiation (n = 3 per group). Scale bars: 100 μm. (S and T) Representative images of staining of 4-HNE and Perls in normal human liver and irradiated human liver. Normal human liver (sample-1) and irradiated human liver (sample-7) are also shown in Figures S8B and S8C. Sample-1, scale bars: 500 μm. Sample-7, scale bars: 1,000 μm. Data are presented as mean ± SEM; p values were assessed using unpaired two-tailed Student’s t test. Using this dataset, we compared the transcriptome changes in each cell type between 0 and 40 Gy samples. Differentially expressed genes (DEGs) were obtained by comparing gene expression between 0 and 40 Gy with log2FC >0.25 and p value <0.05 (STAR Methods). We first performed Gene Ontology (GO) enrichment analysis of the DEGs in LSECs, HSCs, and KCs because these cell types exhibit pronounced responses during HIR.27,31 As the second most abundant cell type in the liver, LSECs were highly radiosensitive, showing upregulated apoptosis, senescence, and autophagy following radiation damage (Figures 4D–4F). In addition, pathways related to fibrinolysis, blood vessel morphogenesis, oxidative damage response, and angiogenesis were significantly downregulated (Figures 4D–4F), indicating potential vascular repair and coagulation dysfunctions that could contribute to disrupted microcirculatory blood flow, leading to hepatocyte damage, hepatic congestion, and liver dysfunction. In HSCs, various signaling pathways, including response to TGF-β and Wnt, were significantly upregulated in the 40 Gy samples (Figures S6A–S6C). These are well known for their critical role in HSC activation.14,32 In addition, the fibrinolysis and response to wounding pathways were significantly downregulated (Figures S6A–S6C), potentially accelerating the fibrosis progression. Gene expression analyses of KCs in the 40 Gy samples showed the upregulation of genes involved in macrophage cytokine production and endocytosis (Figures S6D–S6F), indicating an activated adaptive immune system. Moreover, KCs in the 40 Gy samples exhibited significant downregulation of pathways related to the response to bacteria, hypoxia, and inflammation compared with the 0 Gy samples (Figures S6D–S6F), suggesting impaired KC function following HIR, which may adversely affect liver repair and regeneration. Interestingly, in contrast to LSECs and hepatocytes, HSCs and KCs showed no significant activation of either apoptosis or necroptosis pathways, suggesting the involvement of an alternative mechanism underlying their radiation resistance (Figures S6C and S6F). Indeed, we observed the upregulation of senescence markers in KCs following radiation (Figure S7A). This aligns with the established phenomenon wherein fetal-derived KCs adopt a senescence phenotype as a mechanism to resist radiation-induced cell death.33 Previous studies have shown that DNA sensing and interferon signaling activation in non-parenchymal cells (NPCs) contribute to the progression of radiation-induced liver injury.34 In line with this, we found significant activation of the type I interferon (IFN) signaling pathway specifically in LSECs and KCs but not in other NPC populations (Figure S7B). This suggests that LSECs and KCs may be identified as a major source of IFN-α/β secretion in irradiated pig livers. Next, we focused on hepatocytes, as they are the dominant cell type in the liver (Figure 4G). GO analyses confirmed that signaling pathways, including responses to irradiation and DNA damage, were significantly upregulated in hepatocytes treated with 40 Gy compared with non-irradiated controls, indicating a robust hepatocyte response to radiation (Figures 4H and 4I). Moreover, this process was accompanied by the activation of apoptosis, along with the downregulation of pathways that negatively regulate cell death (Figures 4H, 4I, S6G, and S6H). These findings are consistent with the centrilobular necrosis observed in the histological analysis. In addition, we found that compared with the non-irradiated controls, 40 Gy HIR downregulated the mitotic process and negative regulators of cell senescence (Figures 4H, 4I, S6I, and S6J). This suggested that the cells shifted toward a state of cell-cycle arrest, resulting in failed regeneration. Notably, we observed the activation of ferroptosis signals and noted a decrease in iron ion transport signals in the 40 Gy samples (Figures 4H and 4I). In connection with ferroptosis, we observed significant enrichment of fatty acid oxidation and P53 signaling pathways in the 40 Gy samples (Figures 4H and 4I), both of which have been previously implicated in ferroptosis.35,36 Specifically, the gene expression of ferroptosis suppressors (SOD2, CD44, GCH1, ATF4, and DHODH) was significantly decreased, whereas that of ferroptosis drivers (ACSL1, TP53, ACADSB, and TFR2) was significantly increased in the 40 Gy samples (Figure 4J). These data indicated that early in the pathogenesis of RILD, hepatocytes enter a state of cell-cycle arrest accompanied by various forms of cell death, including ferroptosis. The identification of ferroptosis during early RILD pathogenesis was unexpected. To determine whether ferroptosis persisted in the late stages of RILD, we analyzed liver samples 35 days after 40 Gy irradiation. Key indicators of ferroptosis, including lipid peroxidation and Fe2+ accumulation, were also assessed.26 Given that lipid peroxidation is the key driver of ferroptosis, we performed immunofluorescence staining for 4-hydroxynonenal (4-HNE), a well-established marker of lipid peroxidation. The results showed an increased number of 4-HNE-positive cells in irradiated livers compared with those in non-irradiated controls (Figures 4K and 4L). Additionally, we measured serum lipid peroxidation by detecting malondialdehyde (MDA) levels at multiple time points over 28 days. Notably, lipid peroxidation was significantly elevated at day 28, suggesting that serum MDA could serve as a valuable marker for tracking RILD progression (Figure S6K). Hepatocyte accumulation of Fe2+ in the irradiated livers was demonstrated by Perls staining (Figures 4M and 4N). Moreover, immunofluorescence staining for the ferroptosis suppressors SLC7A11 and GPX4 showed a significant decrease in the number of positive cells in irradiated livers (Figures 4O–4R). Taken together, these data suggest that hepatocytes undergo significant ferroptosis in both the early and late stages of RILD. To validate the clinical relevance of ferroptosis in RILD, we analyzed available liver samples from patients who had undergone RT, including HIR and radioembolization. Notably, all 8 samples from 5 independent patients exhibited markedly elevated levels of lipid peroxidation and Fe2+ accumulation compared to normal liver tissue (Figures 4S, 4T, and S8). This finding underscores the role of ferroptosis in the pathophysiology of liver damage following irradiation and has potential implications for therapeutic strategies targeting this form of cell death. Reversal of RILD by the ferroptosis inhibitor Lip-1 To test the hypothesis that inhibiting ferroptosis can reverse RILD, we treated 40 Gy-irradiated pigs with the ferroptosis inhibitor Lip-1, which inhibits lipid peroxidation.37 Given that 14 days post-irradiation is a critical period for intervention, we administered Lip-1 at 2 mg/kg/day to the animals during this window. The dose was selected based on previous pharmacokinetic and efficacy studies in mouse models,37,38 guided by body surface area scaling for translation to larger species.39 Six adult Bama miniature pigs were randomly divided into two groups following 40 Gy irradiation. The pigs assigned to group 1 received vehicle control (HIR [40 Gy+veh] group, n = 3). Those assigned to group 2 were treated with Lip-1 as early as on day 0 (E-Lip-1 group, n = 3) (Figure 5A). To better mimic the clinical scenario where patients often present with established injury, we also included a third group in which Lip-1 treatment was initiated later, at 21 days post-irradiation (L-Lip-1 group, n = 3) (Figure 5A). Non-irradiated (0 Gy+vehicle) livers were also included as a reference control for baseline histological and immunostaining readouts. Reversal of RILD by ferroptosis inhibitor-Lip-1 (A) Schematic outline for monitoring the irradiated pig treatment with HIR (40 Gy+veh), early-Lip-1 (E-Lip-1), or late-Lip-1 (L-Lip-1) to evaluate therapeutic effect. (B) Miniature pigs treated by HIR (40 Gy+veh) or E-Lip-1. (C) Survival curve of pigs treated by HIR (40 Gy+veh), E-Lip-1, or L-Lip-1 within 35 days post-irradiation (n = 3 per group). (D–H) Dynamic changes of blood index of pigs treated by HIR (40 Gy+veh), E-Lip-1, or L-Lip-1 within 28 days post-irradiation (n = 3 per group). The annotated p values represent an unpaired, two-tailed t test comparing day 28 to day 1 (baseline) in the HIR (40 Gy+veh) group. (I–L) Representative images and quantification of immunofluorescence staining of SLC7A11 (I and J) and GPX4 (K and L) in normal livers treated by vehicle (0 Gy+veh) and irradiated livers treated by HIR (40 Gy+veh), E-Lip-1, or L-Lip-1 35 days post-irradiation (n = 3 per group). (M–P) Representative images and quantification of H&E (M and N) and Masson’s trichrome (O and P) staining in normal livers treated by vehicle (0 Gy+veh) and irradiated livers treated by HIR (40 Gy+veh), E-Lip-1, or L-Lip-1 35 days post-irradiation (n = 3 per group). Necrosis is not detected in the E-Lip-1or L-Lip-1. Images shown in (M) and (O) were performed on adjacent serial sections. Scale bars: 100 μm. Data are presented as mean ± SEM; p values were assessed using unpaired two-tailed Student’s t test. We first evaluated whether E-Lip-1 treatment could prevent the development of RILD. One of the three pigs in the HIR group died 30 days post-irradiation, and the remaining two pigs showed lethargy and an unsteady gait within the same period (Figures 5B and 5C). Notably, all three pigs in the E-Lip-1 group survived for 35 days (Figures 5B and 5C). On day 1 post-irradiation, we collected serum from the pigs, which was used to determine baseline levels. Pigs in the HIR group exhibited a significant increase in serum levels of ALP and TBIL, along with a notable elevation in serum AST levels, 28 days following exposure to 40 Gy irradiation (Figures 5D–5G). Additionally, serum ALB levels progressively decreased (Figure 5H), suggesting liver dysfunction. Consistent with our earlier observations, ALT did not show an obvious change (Figure 5F). In contrast, pigs treated with Lip-1 displayed no significant changes in serum biochemical parameters compared with the baseline (Figures 5D–5H). These results suggested that Lip-1 treatment at early stage effectively prevented the development of RILD symptoms in pigs. To further evaluate whether Lip-1 treatment inhibits ferroptosis and VOD-related pathological characteristics, we analyzed liver samples 35 days post-irradiation. Lipid peroxidation levels following Lip-1 treatment were significantly reduced compared to the HIR group, as evidenced by 4-HNE staining (Figures S9A and S9B). Perls staining revealed a significant decrease in Fe2+ accumulation in the Lip-1 group (Figures S9C and S9D). Additionally, SLC7A11 and GPX4 immunofluorescence staining revealed a substantial increase in positive cells in the E-Lip-1 group, reaching a level comparable with that in non-irradiated livers (Figures 5I–5L), indicating the effective inhibition of ferroptotic stress. H&E staining demonstrated minimal hepatocyte necrosis in the Lip-1 group, in contrast to massive necrosis in the control group (Figures 5M and 5N). Furthermore, Masson’s trichrome, collagen I, and α-SMA staining revealed a marked reduction in liver fibrosis in the E-Lip-1 group, reaching levels comparable to those in non-irradiated livers (Figures 5O, 5P, S9E, and S9H), suggesting that Lip-1 therapy effectively reversed VOD-related pathological characteristics. We next assessed the therapeutic potential of Lip-1 when treatment was started after injury was established. Following a 2-week course of this delayed treatment, all pigs survived to day 35 and exhibited nearly normal biochemical profiles (Figures 5C–5H). Histologically, the severe centrilobular necrosis observed in the HIR group was resolved in the L-Lip-1 group (Figures 5M and 5N). Furthermore, liver fibrosis was significantly attenuated, although the fibrotic level remained slightly higher than that in the non-irradiated liver (Figures 5O and 5P). Collectively, these findings underscore the remarkable therapeutic effects of Lip-1 in RILD treatment. Lip-1 treatment inhibits hepatocyte ferroptosis and promotes liver regeneration To further investigate the therapeutic mechanism of Lip-1 in preventing RILD, we performed snRNA-seq analysis on liver biopsy samples from pigs in the Lip-1 treatment group (E-Lip-1) 14 days post-irradiation (Figure 6A). Due to the highly consistent pathological and transcriptomic profiles among 40 Gy-irradiated animals (Figures 1, 5, and S10A), the same 40 Gy biopsy samples served as the control group. Using the aforementioned transcriptome-based characterization of all cell types of pig liver biopsies, we compared the transcriptome changes for each cell type between Lip-1 and the control (Figures 6B, 6C, S10B, and S10C). To determine whether Lip-1 treatment was cell type specific, we first quantified the number of gene alterations in each cell type from Lip-1-treated biopsies compared with control samples. Notably, hepatocytes displayed the highest number of DEGs, with 749 genes showing significant upregulation or downregulation following Lip-1 treatment (Figure S10D). Moreover, GO analyses showed that the ferroptosis pathway was significantly enriched in the DEGs of hepatocytes compared to other cell types, including KCs, LSECs, and HSCs (Figure 6D). Gene expression bubble analysis also revealed that 68% of the ferroptosis driver genes were significantly downregulated in hepatocytes, whereas 65% of the ferroptosis suppressor genes were significantly upregulated (Figures 6E and 6F). These results suggest that Lip-1 therapy exhibits cell-type specificity, primarily affecting hepatocytes. Lip-1 treatment inhibits hepatocyte ferroptosis and promotes liver regeneration (A) Schematic illustration of the experimental workflow. This plot is created with BioRender.com. (B and C) Dimension-reduction presentation of the combined single-cell transcriptome data from 40 Gy and E-Lip-1 samples (n = 2 per group); UMAP plot showing the major cells. (D) The binding significance of reverted genes in different cell types on ferroptosis-associated genes. (E and F) Bubble map showed that ferroptosis-related genes were enriched in 40 Gy and E-Lip-1 samples. (G–J) Heatmap of marker genes and GO term enrichment analysis in the hepatocyte cluster (G and H) and endothelial cell cluster (I and J) in 40 Gy and E-Lip-1 samples. (K and L) Representative images and quantification of immunofluorescence staining of HNF4A, CD31, Ki67, and P21 in irradiated livers treated by HIR (40 Gy+veh) and E-Lip-1 35 days post-irradiation (n = 3 per group). Ki67+ CD31+ cells are not detected in the HIR (40 Gy+veh) group. Scale bars: 100 μm. Data are presented as mean ± SEM; p values were assessed using unpaired two-tailed Student’s t test. In conjunction with the inhibition of ferroptosis, genes associated with lipid peroxidation were significantly downregulated, whereas those related to ATP synthesis and response to hypoxia were markedly upregulated (Figures 6G and 6H), which enhanced hepatocyte resistance to oxidative damage. In addition, liver function-related pathways, including coagulation and xenobiotic drug metabolism (Figures 6G and 6H), were activated, indicating improved liver function. Notably, Lip-1-upregulated DEGs were significantly enriched in pathways related to liver regeneration and negative regulation of cell death (Figures 6G and 6H), suggesting increased hepatocyte proliferation and survival. Additionally, other cell types, such as LSECs, exhibited increased expression of genes involved in cell proliferation and angiogenesis, whereas pathways associated with hemostasis were downregulated (Figures 6I and 6J), indicating vascular function and circulatory system homeostasis recovery. These data suggest that ferroptosis inhibition may directly stimulate regeneration in hepatocytes and LSECs. For HSCs, we observed enhanced expression of genes associated with fibrinolysis and response to wounding (Figures S10E and S10F), suggesting potential fibrosis inhibition. Regarding KCs, we observed the activation of pathways related to bacterium response and hypoxia (Figures S10G and S10H), indicating the restoration of their function. Consistent with these findings, Ki67, HNF4A, and CD31 co-staining provided further evidence that Lip-1 treatment significantly improved the proliferation of both hepatocytes and endothelial cells (Figures 6K and 6L). The number of γH2A.X-positive cells was significantly reduced after Lip-1 treatment (Figures S9I and S9J), which is consistent with previous studies showing that ferroptosis inhibition can mitigate DNA damage and revert senescence.40 Additionally, co-staining for P21 and HNF4A revealed that the proportion of senescent hepatocytes in the control group was 60.5-fold higher than in the Lip-1-treated liver (13.3% vs. 0.2%), indicating that Lip-1 therapy markedly inhibits hepatocyte senescence and cell-cycle arrest (Figures 6K and 6L). Taken together, these findings demonstrate that Lip-1 treatment effectively restores dysregulated DEGs in RILD, particularly those associated with hepatocyte ferroptosis, and improves liver regeneration. Discussion RILD is a limiting factor in RT, particularly in the treatment of liver cancer, and is associated with a high mortality rate in patients with underlying liver diseases. Approximately six decades after the initial histological characterization of RILD, which resembles VOD, our understanding of its pathogenesis remains limited, and the disease remains incurable. One major barrier is the lack of a readily available animal model capable of reproducing the hallmarks of RILD. Herein, we developed a pig model of RILD using clinically relevant SBRT to replicate the clinical symptoms and pathological characteristics of human RILD, including VOD and centrilobular necrosis. Notably, we revealed that hepatocyte ferroptosis plays a crucial role in the pathogenesis of RILD and that treatment with a ferroptosis inhibitor significantly reverses its manifestations, offering promising potential for clinical translation. To investigate the pathogenesis and molecular mechanisms of RILD, the effects of HIR have been studied in various animal models ranging from rodents to nonhuman primates. In rodents, high-dose HIR (30–60 Gy) induces elevated levels of transaminase and hepatocyte apoptosis, which can lead to mild liver fibrosis after 2–6 months.12,13,14 Several large animals, such as rabbits, dogs, and rhesus monkeys, reportedly tolerate HIR and show no clinical or biochemical signs of liver dysfunction.15,16 Among these, mild liver fibrosis has been found in irradiated dogs over a period of years.15 Importantly, none of the animals developed VOD after HIR. Yannam et al. observed elevated ALP and decreased ALB levels along with marked veno-occlusive changes and mild liver fibrosis in cynomolgus monkeys after hypofractionated radiotherapy.17 However, this requires a dose exceeding 40 Gy and typically manifests 10 weeks post-HIR. In contrast, our miniature pig model showed a VOD phenotype within 5 weeks of exposure to 40 Gy. Notably, it phenocopied a more severe form of RILD than that observed in cynomolgus monkeys. HIR (40 Gy) caused extensive centrilobular necrosis and bile canaliculus injury in pigs, correlating with significantly elevated AST and ALP levels. Severe liver fibrosis and compromised liver regeneration were also observed. These results suggest that pigs are more susceptible to the effects of HIR than cynomolgus monkeys, showing more acute and severe liver damage within a shorter time frame. This miniature pig model addresses the shortage of animal models for RILD research and has significant implications for studying RILD pathogenesis. Hepatocytes are more radioresistant than non-parenchymal cells, and radiation-induced hepatocyte injury is often underappreciated in the pathogenesis of RILD.17 In addition to liver fibrosis, both VOD and centrilobular necrosis have been observed in pig models of RILD. Our snRNA-seq confirmed the activation of TGF-β pathways in HSCs. Notably, we found that the ferroptosis pathway was specifically enriched in irradiated hepatocytes. Radiation-induced lipid peroxidation41,42,43 and downregulation of ferroptosis suppressors such as SLC7A11 and GPX4 may contribute to hepatocyte ferroptosis. Existing evidence suggests that lipid peroxidation susceptibility of hepatocyte membranes in pigs are more similar to those in humans than to mice,44 which may underlie the pig’s superior ability to mimic human RILD pathology. More importantly, inhibiting hepatocyte ferroptosis with Lip-1 for 2 weeks post-irradiation not only prevented VOD-related centrilobular necrosis and liver fibrosis but also promoted hepatocyte and endothelial cell proliferation, highlighting the pivotal role of hepatocyte injury in RILD pathogenesis. These results warrant further genetic manipulation of ferroptosis in pig hepatocytes to demonstrate hepatocyte ferroptosis as a key driver of RILD. While ferrostatin-1 (Fer-1) and Lip-1 are broad-spectrum inhibitors of lipid peroxidation, their impact on specific gene expression can be complex and context dependent. Existing studies have shown that Fer-1 or Lip-1 can downregulate ferroptosis driver genes (e.g., ACSL1, ACSL4, and ALOX15) and upregulate suppressor genes like GPX4 in liver injury models.40,45,46 Consistent with this, our study demonstrated that Lip-1 treatment downregulates ACSL1,47 ACADSB (a known negative regulator of GPX4),48 and P53 (a transcriptional repressor of SLC7A11)36 in the RILD pig model. Lip-1 treatment broadly downregulates ferroptosis driver genes and upregulates suppressor genes, likely by reducing lipid peroxidation and oxidative stress and thereby indirectly normalizing the activity of upstream transcriptional regulators such as P53. The upregulation of transferrin (TF), a key marker of hepatocyte function, likely reflects an overall functional improvement in irradiated livers following Lip-1 treatment. In contrast, the upregulation of DPP4 was more unexpected, as it has been reported to promote ferroptosis in specific contexts, such as P53-deficient cancer cells, and is typically regulated at the enzymatic level rather than via mRNA expression.49 Thus, its role in ferroptosis inhibition in the liver warrants further investigation. Radiation-based preconditioning has shown promise in a range of applications, including hematopoietic stem cell, hepatocyte, mesenchymal cell, and nerve cell transplantation.7,14,50 In hepatocyte transplantation, HIR disrupts the endothelial barrier and inactivates KCs, thereby enhancing the engraftment of transplanted hepatocytes.27 At high doses, HIR also inhibits mitosis of host hepatocytes, providing transplanted cells with a proliferative advantage.27 Thus, transplantation efficiency improves in a dose-dependent manner. However, it is important to note that a single dose of 50 Gy, typically delivered to the entire liver or 30% of the liver volume in rodents to achieve therapeutic liver repopulation, exceeds the maximum safe dose for humans.51 Such high doses are strongly associated with the risk of RILD and can lead to significant morbidity. Additionally, a clinical study showed that hepatocyte transplantation within 24 h of administering 10 Gy HIR resulted in a lower anticipated engraftment rate.7 This highlights the need to optimize the timing and dose of irradiation to improve the success of hepatocyte transplantation. In the current study, pigs developed RILD that closely resembled human conditions in terms of dose, timing, and pathology. This makes them an ideal model for evaluating the efficacy and toxicity of HIR for hepatocyte transplantation and for optimizing HIR preconditioning protocols. The insights gained from these studies can be effectively translated into clinical settings, leading to safer and more effective preconditioning strategies for human patients. In summary, we used SBRT to develop a miniature pig model that closely mimics the human RILD phenotype. This model allowed us to characterize the pathological events and multicellular responses in the early stages of the disease and to uncover ferroptosis as a therapeutic target. These findings are crucial for advancing the clinical treatment and prevention of RILD, enhancing our understanding of its pathogenesis, and preparing effective HIR regimens for hepatocyte transplantation. Limitations of the study Our study highlights the therapeutic promise of ferroptosis inhibitors for treating RILD; nonetheless, several questions remain. Although Lip-1 has been demonstrated as one of the superior radical-trapping antioxidants (RTAs) in terms of activity, solubility, and stability in various mouse models,38 its inhibition of CYP2D enzymatic activity poses concerns for clinical safety.37 Therefore, conducting a comprehensive pharmacokinetic study of Lip-1 in large animals and developing a new generation of RTAs with improved drug properties are crucial for future clinical applications.52 Furthermore, iron chelators such as FerroTerminator1, deferoxamine, and deferasirox have been reported to attenuate disease in metabolic dysfunction-associated steatohepatitis (MASH) mouse models by inhibiting ferroptosis.45,46,53,54 It would be worthwhile to test the efficacy of these iron chelators in RILD pigs. Finally, in addition to ferroptosis, we observed apoptosis activation in the hepatocytes of irradiated pigs. Future studies employing apoptosis inhibitors (e.g., Z-VAD-FMK) in these large animal models could help elucidate the role of apoptosis in RILD pathogenesis. Resource availability Lead contact Further information and requests for resource and reagents should be directed to and will be fulfilled by the lead contact, Ludi Zhang (zhangludi@sibcb.ac.cn). Materials availability This study did not generate new unique reagents. Data and code availability •Single-cell and bulk tissue RNA-seq data are publicly available at the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/) under the GSE numbers GEO: GSE200123 and GEO: GSE286326. Single-cell and bulk tissue RNA-seq data are publicly available at the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/) under the GSE numbers GEO: GSE200123 and GEO: GSE286326. •This paper does not report original code. The bioinformatics for the single-cell data was performed using open-source tools, as listed in the key resources table. This paper does not report original code. The bioinformatics for the single-cell data was performed using open-source tools, as listed in the key resources table. •Any additional information required to reanalyze data reported in this work is available from the lead contact upon request. Any additional information required to reanalyze data reported in this work is available from the lead contact upon request. Acknowledgments We thank the core facilities for molecular biology, cell biology, and animal care at Center for Excellence in Molecular Cell Science. This project is supported by the Strategic Priority Research Program of the Chinese Academy of Sciences (XDB1150000 and XDB1030000), the National Science Foundation of China (NSFC) (U25A600024, 32570880, and 32400606), Major Program of Natural Science Foundation of Zhejiang Province, China (no. LHDMD23H030001), Xinjiang Regional Collaborative Innovation Program (2024E02040), Postdoctoral Fellowship Program of Chinese Postdoctoral Science Foundation (GZB20230755), Shanghai Super Postdoctoral Incentive Program (2023635), 75th General Program of Chinese Postdoctoral Science Foundation (2024M753224), Peak Disciplines (Type IV) of Institutions of Higher Learning in Shanghai, Youth Innovation Promotion Association CAS, and the SA-SIBS Scholarship Program. Author contributions X.C., L.H., C.G., Y.W., and L.Z. conceived and supervised the study. Z.S. and L.Z. designed and initiated the study project. Z.S., Q.Z., and X.Z. performed most of the experiments. B.W., Z.S., and L.Z. analyzed the data of RNA-seq. G.F. and G.S. assisted with the ultrasound guidance and irradiation experiments. J.N., E.Z., Y.L., S.L., and M.C. assisted with the animal experiments. W.H., S.Z., S.G., and J.W. assisted with the TSA staining assays. F.M. and L.L. assisted in obtaining human liver samples. Z.S., L.Z., X.C., L.H., C.G., and Y.W. analyzed the data. Z.S. and L.Z. wrote the manuscript, and L.H. and C.G. were involved in manuscript preparation. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesLYVE1 (1:2000, IHC)AbcamCat# ab219556, RRID:AB_2884014VE-cadherin (1:1000, IHC)AbcamCat# ab205336, RRID:AB_2891001pRb (Ser795) (1:200, IHC)Cell Signaling TechnologyCat# 9301, RRID:AB_330013MRP2 (1:200, IHC)GeneTexCat# GTX23373, RRID:AB_374356GS (1:500, IHC)BD BiosciencesCat# 610517, RRID:AB_397879E-cadherin (1:500, IHC)Cell Signaling TechnologyCat# 3195, RRID:AB_2291471FN-1 (1:200, IHC)SolarbioCat# K000214M, RRID:N/AMPO (1:1000, IHC)AbcamCat# ab208670, RRID:AB_2864724Ki67 (1:1000, IHC)AbcamCat# ab15580, RRID:AB_443209α-SMA (1:400, IHC)BosterCat# BM0002, RRID:AB_2811044γH2A.X (pS139) (1:500, IHC)AbcamCat# ab26350, RRID:AB_470861Collagen I (1:200, IHC)AbcamCat# ab34710, RRID:AB_731684CD31 (1:100, IHC)AbcamCat# ab28364, RRID:AB_726362SLC7A11/xCT (1:200, IHC)ArigoCat# ARG57998, RRID:AB_2815029GPX4 (1:200, IHC)ArigoCat# ARG41400, RRID:N/A4-Hydroxynonenal (1: 100, IHC)AbcamCat# ab48506, RRID:AB_867452HNF4A (1:1000, IHC)AbcamCat# ab181604, RRID:AB_2890918P21 (1:1000, IHC)AbcamCat# ab188224, RRID:AB_2734729Critical commercial assaysOpal4-Color Manual IHC KitPerkinelmerCat#NEL810001KTLipid Peroxidation Assay (MDA) (Colorimetric) color reaction kitAbcamCat#ab118970Perls staining kitSolarbioG1424DAB-enhanced Perls staining kitSolarbioG1428Deposited dataRaw and analyzed bulk RNA-seq dataThis paperGEO: GSE200123snRNA-seq datasetThis paperGEO: GSE286326Software and algorithmsGraphPad Prism 8GraphPad SoftwareVersion 8.0.2ImageJNIHhttps://imagej.nih.gov/ij/Cell Ranger (v8.0.1)10X Genomicshttps://www.10xgenomics.com/support/software/cell-ranger/latestSeurat R package (v4.1.0)Stuart et al.55https://satijalab.org/seurat/Metascape (v3.5)Zhou et al.56https://metascape.org/gp/hisat2-2.10Kim et al.57https://daehwankimlab.github.io/hisat2/Cufflinks v.2.2.1Trapnell et al.58https://cole-trapnell-lab.github.io/cufflinks/htseq-countAnders et al.59https://htseq.readthedocs.io/en/master/htseqcount.htmlDESeq2Love et al.60https://bioconductor.org/packages//release/bioc/html/DESeq2.html Experimental model and study participant details Animals and radiation treatment All animal procedures were performed according to institutional and national guidelines and approved by the Animal Care Ethics Committee of Sir Run Run Shaw Hospital of Zhejiang University Medical School (SRRSH20190401). Anesthetized, male Bama miniature pigs (Purchased from Wujiang Tianyu Biological Technology Co., Ltd.;15–18 kg) received total doses of 20, 30, and 40 Gy in five consecutive daily fractions, using 6 MV energy photon beams from a PRIMUS linear accelerator (Siemens Oncology Systems, Erlangen, Germany), according to plans to whole liver. Appetite assessment Appetite was assessed using a simple semi-quantitative visual scoring system based on feeding behavior. Pigs were fed under a consistent daily feeding schedule, and appetite was evaluated by direct observation at each scheduled monitoring time point (days 1, 7, 14, 21, 28, and 35 post-irradiation), consistent with the timeline in Figure 1. Each animal was observed during the feeding period, and an appetite score was assigned as follows: 3, normal feeding behavior (promptly approached feed and consumed a usual amount); 2, reduced appetite (reduced interest in feed and/or noticeably decreased intake compared with the animal’s usual behavior); 1, anorexia (refused feed or did not eat). Scoring was performed by trained staff using the same criteria throughout the study. Biopsy procedure Liver biopsies were performed by an interventional radiologist with more experience in percutaneous biopsies. All procedures were performed under ultrasound guidance using a 3-6MHz convex transducer (DC-6T; Mindray Diagnostic Ultrasound System, Guangdong, China). The procedure was performed under aseptic conditions. Lobe selection depended on operator preference. An 18-gauge core needle (BardMaxCore, America) was advanced through the liver capsule until the tip was positioned in the subcapsular parenchyma and directed away from intrahepatic vascular structures that were visible on ultrasound. The liver capsule was only penetrated once. The cutting needle with automated spring-fire mechanism (BardMaxCore) was co-axially advanced through the core needle and fired to acquire tissue samples. After each pass, the radiologist visually inspects the specimen for adequacy to determine if more passes are required. The retrieved biopsy specimens were placed in 4% Paraformaldehyde (Fdbio Science, Hangzhou, Zhejiang, China) prior to transporting to the pathology department. Human subject Human liver tissue specimens were obtained from the First Affiliated Hospital of the University of Science and Technology of China (USTC). The study protocol was approved by the Hospital’s Medical Research Ethics Committee (Approval No. 2025KY Lun Shen 449). Written informed consent was obtained from all participants (or their legally authorized representatives) when applicable, and all procedures were conducted in accordance with institutional guidelines and the Declaration of Helsinki. A total of 10 liver tissue specimens were included. Two specimens (Sample-1 and Sample-2) were normal control liver tissues collected as excess tissue from transplant donors at the time of orthotopic liver transplantation (OLT). Three specimens (Sample-3 to Sample-5) were collected from three independent patients who had received internal radiation therapy (Y-90 radioembolization) and were classified as post-radiotherapy liver tissues. Five specimens (Sample-6 to Sample-10) were collected from two patients who had received external radiotherapy (SBRT) and were also classified as post-radiotherapy liver tissues; specifically, Samples 6 and 7 were obtained from the same SBRT-treated patient, and Samples 8–10 were obtained from another SBRT-treated patient. Sample identifiers, radiation modality, and specimen IDs are provided in Figures S8A. Method details Histology, immunohistochemistry and immunofluorescence For paraffin sections, liver samples were fixed overnight in 4% PFA (4°C) and embedded in paraffin blocks. For making frozen sections, livers were fixed 30min in 4% PFA (4°C), followed by cryopreserved in 30% sucrose overnight (4°C) prior to freezing in OCT tissue blocks. Immunohistochemistry staining and hematoxylin and eosin staining were performed as previously described.61 For immunofluorescence, after antigen retrieved (AKOYA, NEL810001KT), paraffin sections (2 μm) were washed in TBST (0.1% Tween-80 in TBS) for 3 min, blocked in antibody diluent (AKOYA, NEL810001KT) for 1 h at room temperature, and stained with primary antibody at 4°C overnight. Primary antibodies included LYVE1 (Abcam, ab219556, 1:2000), VE-cadherin (Abcam, ab205336, 1:1000), pRb (CST, 9301, 1:200), MRP2 (GeneTex, GTX23373, 1:200), GS (BD, 610517, 1:500), E-cadherin (CST, 3195, 1:500), FN-1 (Solarbio, K000214M, 1:200), MPO (Abcam, ab208670, 1:1000), Ki67 (Abcam, ab15580, 1:1000), α-SMA (Boster, BM0002, 1:400), γH2A.X (pS139) (Abcam, ab26350, 1:500), Collagen I (Abcam, ab34710, 1:200), CD31 (Abcam, ab28364, 1:100), SLC7A11/xCT (Arigo, ARG57998, 1:200), GPX4 (Arigo, ARG41400, 1:200), 4-Hydroxynonenal (Abcam, ab48506, 1:100), HNF4A (Abcam, ab181604, 1:1000), P21 (Abcam, ab188224, 1:1000). Primary antibodies were detected using OpalTM 4-Color Manual IHC Kit (AKOYA, NEL810001KT) on paraffin sections according to the manufacturer’s protocol. Sections were stained with DAPI (4′,6-Diamidino-2-Phenylindole) and mounted with fluorescence mounting medium (Dako). Images were taken using an Olympus BX53 microscope or a confocal microscopy Leica TCS SP8 WLL. Images in different channels were merged by ImageJ software. Perls staining Perls staining was performed on liver sections as follows: porcine samples were stained using a kit with eosin counterstain (Solarbio, G1424), and human samples were stained using a DAB-enhanced kit for ferric iron (Solarbio, G1428), in accordance with the manufacturer’s instructions. In vivo experiment The rescue compound (Lip-1, HY-12726, MCE) or vehicle control was administered starting on either day 1 (early stage) or 21 days (late stage) after 40 Gy irradiation. Lip-1 was dissolved in a vehicle containing 5% dimethyl sulfoxide (DMSO), 40% PEG300/400, 5% Tween-80, and 50% saline, and administered intraperitoneally at a dose of 2 mg kg−1 daily for two weeks. The vehicle control group (40 Gy + Veh) received an equal volume of the vehicle solution on the same schedule. Both solutions were odorless and colorless to prevent bias. Drug administration and daily animal health assessments were performed in a blinded manner. Animals were euthanized upon exhibiting terminal endpoints. Liver function tests Blood samples were collected for subsequent analyses. Albumin (ALB), total bile acid (TBA), total bilirubin (T-BIL), aspartate aminotransferase (AST), alanine aminotransferase (ALT), alkaline phosphatase (ALP) and ammonia were estimated using an AU 5800 automated analyzer (Beckmann) at the clinical laboratory of Sir Run Run Shaw Hospital. Plasma malondialdehyde (MDA) measurement Plasma was isolated by centrifugation. MDA was measured using a Lipid Peroxidation Assay (MDA) (Colorimetric) color reaction kit (ab118970, Abcam) following the manufacturer’s protocol. RNA-seq Total RNA was extracted from pig livers using Trizol (Invitrogen) according to the manufacturer’s instructions. Sequencing library was prepared from one microgram of total RNA using Illumina TruSeq RNA Sample Prep Kit. Paired-end 150 bp read length sequencing was performed on Illumina Hiseq 4000 sequencer. RNA-seq analysis RNA-seq analysis was performed as previously described.25 All sequencing reads from RNA-seq were mapped to the porcine reference genome sequence (Sscrofa11.1) using hisat2-2.10. FPKM (fragments per kilobase of exon per million fragments mapped) values were calculated by Cufflinks v.2.2.1 using default parameters for gene expression levels. We used htseq-count to count reads on genes. Differential expression analysis was performed using DESeq2. Genes were considered differentially expressed if FPKM>1 in all sample and | log2(fold changes) | ≥ 1, q-value<=0.05. Cell suspension preparation The liver tissue samples were subjected to nuclear isolation, sequencing, and library preparation following the 10X Genomics protocol. Approximately 500 mg of hepatic tissue was dissociated into a singular nuclear suspension via tissue homogenization in a chilled lysis buffer (0.25 M sucrose, 5 mM CaCl2, 3 mM MgAc2, 10 mM Tris-HCl pH 8.0, 1 mM DTT, 0.1 mM EDTA, 1 × Protease Inhibitor and 1U/μL RiboLock RNase Inhibitor (Thermo Scientifc, cat no. O0381) with pestle strokes. The resulting homogenate was subsequently filtered through a 70 μm cell strainer, yielding a nuclear fraction collected in a 15mL centrifuge tube, with a 1 mL volume. This nuclear fraction was mixed with an equal volume of 50% iodixanol solution (0.16 M sucrose, 10 mM NaCl, 3 mM MgCl2, 10 mM Tris-HCl pH 7.4, 1 U/μL RiboLock RNase Inhibitor, 1 mM DTT and 0.1 mM PMSF Protease Inhibito (Thermo Scientifc, cat no. 36978), yielding a final concentration of 25%, and supplemented with 1 mL of 33% iodixanol solution at the tube’s base and 30% iodixanol solution at the top. The solution centrifugation at 500 × g for 8 min at 4°C, subsequent to myelin layer removal from the gradient’s apex. Nuclei were harvested from the 30% iodixanol interface, resuspended in nuclear wash and resuspension buffers, and centrifuge (0.04% bovine serum albumin, 0.2 U/μL RiboLock RNase inhibitor, 500 mM mannitol and 0.1 mM PMSF protease inhibitor in PBS) at 500 × g and 4°C for 5 min. Filtration through a 40 μm cell filter eliminated cell debris and sizable aggregates. The nuclei’s total count, concentration, and integrity ratio were ascertained via hemocytometer-assisted microscopic examination of trypan-stained samples. Ultimately, the nuclear concentration was regulated to 700-1,200 nuclei/μL, with nuclei inspected using the 10X Chromium platform. snRNA-seq library preparation and sequencing Cellular suspensions were loaded on a 10X Genomics GemCode Single-cell instrument that generates single-cell Gel Bead-In-EMlusion (GEMs). Using Chromium Next GEM Single Cell 3′Reagent Kit v3.1 for Library Generation and Sequencing of cDNA. Upon dissolution of the Gel Bead in a GEM, primers containing (i) an Illumina R1 sequence (read 1 sequencing primer), (ii) a 16 nt 10 × Barcode, (iii) a 10 nt Unique Molecular Identifier (UMI), and (iv) a poly-dT primer sequence were released and mixed with cell lysate and Master Mix.62 Silane magnetic beads were used to remove leftover biochemical reagents and primers from the post GEM reaction mixture. Full-length, barcoded cDNAs were then amplified by PCR to generate sufficient mass for library construction. R1 (read 1 primer sequence) were added to the molecules during GEM incubation. P5, P7, a sample index, and R2 (read 2 primer sequence) were added during library construction via End Repair, A-tailing, Adaptor Ligation, and PCR. The final libraries contained the P5 and P7 primers used in Illumina bridge amplification. A Single Cell 5′ Library comprised standard Illumina paired-end constructs which begin and end with P5 and P7. The Single Cell 5′ 16 bp 10X Barcode and 10 bp UMI were encoded in Read 1, while Read 2 was used to sequence the cDNA fragment. Sample index sequences were incorporated as the i7 index read. Read 1 and Read 2 were standard Illumina sequencing primer sites used in paired-end sequencing.63 snRNA-seq data processing FASTQ files were processed with 10X Genomics Cell Ranger (v8.0.1) using the default parameters and aligned to the Ensembl_release104 Sscrofa 11.1 reference genomes. The gene expression matrix was processed and analyzed by Seurat package (v4.1.0). To filter out low-quality cells, the investigation removed reads from cells in which less than 240 and over 3,600 genes were detected, cells with more than 16,000 UMIs, and cells with more than 10% share of mitochondrial gene expression. Cell clustering by Seurat Multiple criteria were employed to filter cells, eliminating multi-cellular entities and cells with suboptimal states, while retaining high-quality cells. The gene expression levels were normalized using a log transformation method. The resulting high-quality cell data was subjected to data integration and batch effect correction using CCA Integration. To mitigate noise interference from individual gene expression levels in single-cell transcriptomics data, the vst algorithm was applied for high-variable gene selection. Principal Component Analysis (PCA) was performed on the integrated data to reduce the dimensionality to 30 principal components for capturing the main information. The Louvain algorithm was used for clustering the dimension-reduced data with a resolution parameter set to 30, partitioning all cells in the single-cell transcriptome into distinct cell subpopulations, facilitating subsequent analyses.64 Differentially expressed genes (DEGs) and pathway enrichment analysis Expression value of each gene in given cluster was compared against the rest of cells using Wilcoxon rank-sum test.55 Significant up-regulated (log2Foldchange larger than 0.25, padj less than 0.05) genes were identified. Gene ontology analysis was performed using Metascape v3.5.56 Quantification and statistical analysis All experimental data are presented as mean ± SEM. “n” represents the number of animals and is indicated in the figures and figure legends. For quantification of liver sections, three to five random fields of each liver sample were imaged and then quantified using ImageJ. Animals were randomly used in this study. For statistical evaluation, an unpaired two-sided Student’s t test was performed using GraphPad Prism 8 software and is indicated in the figure legends. A p value <0.05 was considered statistically significant. Published: April 16, 2026 Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.xcrm.2026.102745. Contributor Information Xiujun Cai, Email: srrsh_cxj@zju.edu.cn. Chandan Guha, Email: cguhamd@gmail.com. Yifan Wang, Email: anwyf@zju.edu.cn. Ludi Zhang, Email: zhangludi@sibcb.ac.cn. Supplemental information Document S1. Figures S1–S10 Document S2. Article plus supplemental information References Associated Data Supplementary Materials Document S1. Figures S1–S10 Document S2. Article plus supplemental information Data Availability Statement •Single-cell and bulk tissue RNA-seq data are publicly available at the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/) under the GSE numbers GEO: GSE200123 and GEO: GSE286326. Single-cell and bulk tissue RNA-seq data are publicly available at the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/) under the GSE numbers GEO: GSE200123 and GEO: GSE286326. •This paper does not report original code. The bioinformatics for the single-cell data was performed using open-source tools, as listed in the key resources table. This paper does not report original code. The bioinformatics for the single-cell data was performed using open-source tools, as listed in the key resources table. •Any additional information required to reanalyze data reported in this work is available from the lead contact upon request. Any additional information required to reanalyze data reported in this work is available from the lead contact upon request.