Dietary intake and BCAA metabolism regulate pulmonary fibrosis through KDM4A-mediated epigenetic remodeling in male mice

Abstract

Idiopathic pulmonary fibrosis is a progressive and fatal disorder characterized by abnormal activation of alveolar fibroblasts. However, the metabolic reprogramming of alveolar fibroblasts during lung injury remains unclear. Here we show that uptake of branched-chain amino acids is increased, whereas their catabolism is significantly impaired in fibrotic lung fibroblasts and mouse lung tissues. Branched-chain amino acids promote lung fibroblast activation and bleomycin-induced lung fibrosis. Genetic inactivation of branched-chain amino acid transaminase 2 exacerbates fibrosis, whereas inhibition of the corresponding transporter SLC7A5 or enhancement of catabolism attenuates pulmonary fibrosis in male mice. Mechanistically, ATF4 and PPARγ regulate the expression of SLC7A5 and BCAA catabolic genes, respectively. We identify KDM4A as a key mediator of the epigenetic regulation of fibrotic genes. Notably, dysregulated BCAA metabolism is associated with disease severity in patients, suggesting that targeting BCAA metabolism may serve as a promising therapeutic strategy for idiopathic pulmonary fibrosis.

Idiopathic pulmonary fibrosis is a fatal lung disease driven by aberrant fibroblast activation, but the underlying metabolic changes remain unclear. Here, the authors show that dysregulated branched chain amino acid metabolism promotes lung fibrosis and can be therapeutically targeted.

Introduction

Idiopathic pulmonary fibrosis (IPF) is a devastating, progressive interstitial lung disease characterized by relentless scarring of the lung parenchyma, leading to irreversible decline of lung function and a poor prognosis¹. The pathogenesis of IPF involves recurrent epithelial injuries and aberrant activation and differentiation of alveolar fibroblasts, which cause the excessive deposition of extracellular matrix (ECM) proteins, and eventually destroy normal alveolar architecture and impair gas exchange^2,3. Currently, nintedanib and pirfenidone are the only FDA-approved drugs for the treatment of pulmonary fibrosis⁴. However, these drugs only slow the progression of IPF and have limited efficacy in improving patient survival. Therefore, there is an urgent need to identify better therapeutic targets that can halt or even reverse disease progression in IPF.

Metabolic dysfunction is increasingly recognized as a pathogenic process in fibrosis across various organs, including the lung, liver, and kidneys⁵. Alveolar fibroblasts are the primary effector cell type in pulmonary fibrosis. Metabolic reprogramming associated with the activation and differentiation of fibroblasts has emerged as key, targetable mechanisms driving organ fibrosis⁶. Glycolysis, fatty acid oxidation, and fatty acid synthesis are well-established metabolic pathways implicated in fibrogenesis, and targeting these pathways has shown promising results in preclinical studies⁷. However, the role of amino acid metabolism in the pathogenesis of pulmonary fibrosis remains poorly understood.

Emerging evidence highlights the pivotal role of cellular metabolism in regulating cellular function and disease progression^8,9. Among the metabolic pathways implicated in organ fibrosis, the role of branched-chain amino acids (BCAAs) has recently garnered attention¹⁰. BCAAs, including leucine, isoleucine, and valine, are essential amino acids pivotal for energy production and metabolic regulation¹¹. SLC7A5 is the primary amino acid transporter that facilitates cellular uptake. Once enter the cell, BCAAs undergo catabolism initiated by branched-chain aminotransferases (BCAT1 and BCAT2), which convert them into branched-chain α-keto acids (BCKAs). The rate-limiting branched-chain α-keto acid dehydrogenase (BCKDH) complex further irreversibly catabolizes these BCKAs, resulting in the production of acetyl-CoA and succinyl-CoA that enter the TCA cycle for energy generation. Dysregulation of BCAA metabolism has been linked to fibrosis in organs like the liver and kidney, but exhibiting contrasting roles in each organ^12–14. In the case of IPF, previous studies have suggested that BCAA content in the breath of patients is elevated¹⁵; however, the underlying mechanisms driving BCAA metabolic remodeling and their contribution to the development and progression of IPF have yet to be fully elucidated.

In this study, we employed a combination of transcriptomic analysis, metabolic profiling, and metabolite efflux assays using ex vivo lung fibroblasts, in vivo rodent models of IPF, and clinical samples from IPF patients. Our results reveal significant BCAA metabolic reprogramming during pulmonary fibrosis. Using genetic, dietary, and pharmacological approaches, we demonstrate that BCAA accumulation promotes myofibroblast activation and lung fibrosis, while enhancing BCAA catabolism or depriving BCAAs alleviates bleomycin-induced fibrosis. Mechanistically, we show that ATF4 and PPARγ regulate the expression of SLC7A5 and BCAA catabolic genes, respectively, which contribute to BCAA accumulation and drive profound epigenetic changes, particularly at the H3K36me3 mark, in lung fibroblasts. Importantly, we identify KDM4A as a key epigenetic regulator of fibrotic gene expression in response to BCAA metabolism. Clinical analysis reveals that the expression of BCAA metabolic genes and plasma BCAA concentrations correlate strongly with the extent of pulmonary fibrosis and lung function in IPF patients. These findings highlight BCAA metabolism as a central mediator of IPF pathogenesis and suggest that targeting BCAA metabolic pathways could offer a promising therapeutic strategy to combat pulmonary fibrosis.

Results

BCAA metabolic remodeling in fibrotic lungs and TGFβ-treated alveolar fibroblasts

To elucidate the metabolic reprogramming during lung fibrosis, we generated multi-omics data, including transcriptomics and metabolomics, from TGFβ-induced lung fibroblast activation and bleomycin-induced lung fibrosis models. Transcriptomic analysis revealed that the extracellular matrix (ECM) organization pathway was significantly enriched in TGFβ-induced genes in primary mouse lung fibroblasts (MLFs) (Supplementary Fig. 1a, b). Similarly, we observed a substantial increase in fibrosis-related and inflammatory genes in bleomycin-induced fibrotic lungs (Supplementary Fig. 1c, d). These data collectively support the establishment of faithful cellular and animal models for IPF.

We then examined the dysregulations in metabolomics from the same alveolar fibroblasts and mouse lungs using targeted metabolomics approach. Metabolomics analysis revealed 62 metabolites that were differentially expressed between TGFβ-treated and control lung fibroblasts (Supplementary Fig. 1e, f). These metabolites were associated with pathways such as purine metabolism, pyrimidine metabolism, and BCAA biosynthesis and degradation. In addition, 52 metabolites showed significant changes between bleomycin-induced and control lungs, with strong associations to pathways including purine metabolism, BCAA biosynthesis and degradation, and the TCA cycle (Supplementary Fig. 1g, h). Notably, the majority of the dysregulated metabolites in both MLFs and lung tissues overlapped, with the BCAA metabolic pathway being most significantly enriched (Fig. 1a, b). We then specifically examined the expression levels of individual BCAAs and found that leucine, isoleucine, and valine were all highly upregulated in TGFβ-treated MLFs and bleomycin-induced lungs (Fig. 1c, d). These findings were consistent with the reanalysis of a previous metabolic profiling study of bleomycin-induced mouse lungs (Supplementary Fig. 1i)¹⁶. Moreover, several BCAA pathway-associated metabolites, including 2-oxoisovalerate, oxaloacetate (OAA), succinate, and malate, were also dysregulated in MLFs and fibrotic lungs (Supplementary Fig. 1j, k).

Fig. 1. Dramatic BCAA metabolic reprogramming during pulmonary fibrosis progression.

a Venn diagram showing the overlap among differentially expressed metabolites identified by LC–MS between TGF-β1-stimulated MLFs (left) and lung tissues from BLM-treated mice (right). b Top enriched KEGG pathways of shared dysregulated metabolites. P-values were calculated by one-sided hypergeometric test. c, d Relative quantification of BCAA levels in TGF-β1-treated MLFs (c, n = 3) and bleomycin-treated mouse lung tissues (d, n = 7), as determined by targeted metabolomics. Center line indicates the median; box limits denote the 25th and 75th percentiles; whiskers extend to the most extreme values within 1.5 × IQR. e Heatmap showing the expression profile of BCAA metabolic genes and profibrotic markers in TGF-β1 or PBS-treated MLFs (n = 3). f, g RT-PCR validation of selected BCAA metabolic genes in TGF-β1-treated MLFs (f, n = 3) and BLM-treated lung tissues (g, n = 5 for PBS group; n = 8 for BLM group). h, i Immunoblot analysis of BCAA transporters and catabolic genes in MLFs treated with TGF-β1 (h, n = 3) and lung tissues from fibrotic mice (i, n = 5). j, k Quantification of BCAA concentrations in culture supernatants from MLFs treated with TGF-β1 (j, n = 4) and in serum from BLM-treated mice (n = 5). l, m Schematic (l) and quantitative results (m) from [¹³C₆] leucine isotope-tracing experiments in MLFs cultured in BCAA-free medium supplemented with 0.80 mM labeled leucine under PBS or TGF-β1 conditions (n = 4). Box plots show median (center), IQR (box), and minimum to maximum (whiskers). n Schematic representation of BCAA metabolic reprogramming during pulmonary fibrosis by integrated metabolic profiling and transcriptional analysis. Red denotes upregulated genes or metabolites. Blue denotes downregulated genes or metabolites. BCKA, branched-chain α-keto acid. The value of n indicates biologically independent samples. Data are presented as mean ± SEM (g, k) or mean ± SD (f, j). Statistical significance was assessed by two-tailed unpaired Student’s t test (c, d, f, g, j, k, and m). Source data are provided as Source Data file.

In addition to the changes in metabolite levels, we also observed significant dysregulation in the expression of key BCAA metabolic genes (Fig. 1e). Specifically, the BCAA transporter gene Slc7a5 was upregulated, whereas the majority of BCAA catabolic genes, including Bcat2, Bckdha, Dbt, and Dld, were dramatically downregulated in both TGFβ-treated MLFs and bleomycin-induced mouse lungs (Fig. 1f, g). Moreover, BCAA levels in the culture medium of TGFβ-treated MLFs and serum of bleomycin-induced mice were significantly reduced, likely attributable to the upregulation of the BCAA transporter Slc7a5 (Fig. 1j, k), although direct measurement of BCAA uptake remains to be explored. Interestingly, we observed an increase in Bcat1 expression during lung fibrosis, which may be explained by previous reports identifying Bcat1 as a downstream target of TGFβ signaling¹⁷. Similar alterations were also detected at the protein level (Fig. 1h, i and Supplementary Fig. 1l, m). We further confirmed the changes of BCAA metabolic genes in butylated hydroxytoluene (BHT)-induced lung injury model (Supplementary Fig. 1n)¹⁸. There was significant downregulation of BCAA catabolic genes when fibrosis was most severe at 7 days post-BHT treatment (Supplementary Fig. 1o).

To further explore the dynamic remodeling of BCAA metabolism during lung fibrosis, we conducted a metabolic flux assay using ¹³C-labeled leucine in MLFs cultured under BCAA-free conditions (Fig. 1l). The ¹³C-leucine tracing analysis revealed impaired BCAA catabolism during lung fibroblast activation, resulting in reduced levels of L-malic acid, fumaric acid, and 2-oxoglutaric acid (α-KG), alongside the accumulation of L-leucine (Fig. 1m). These multi-omics analyses suggest enhanced cellular uptake but diminished catabolic flux of BCAAs during lung fibrosis (Fig. 1n). Collectively, these findings demonstrate that BCAA metabolic profiles are substantially altered in fibrotic lung tissue and identify BCAA metabolic reprogramming in alveolar fibroblasts as a key cellular contributor to the global BCAA metabolic dysregulation observed in fibrotic lungs.

Supplementation of BCAA enhances myofibroblast activation and accelerates lung fibrosis progression

Given the dramatic BCAA metabolic reprogramming observed during lung fibrosis, we next investigated whether BCAA metabolism plays a role in fibroblast activation and the progression of lung fibrosis. We found that TGFβ treatment induced fibroblast activation and ECM production, as evidenced by increased expression of COL1A1, ACTA2, and FN1 in MLFs and human lung fibroblasts (MRC-5) under standard culture conditions (Supplementary Fig. 2a–d). However, lung fibroblasts cultured in BCAA-free medium showed a diminished response to TGFβ treatment. Interestingly, when BCAAs were reintroduced into the medium (Supplementary Fig. 2e–g), MLFs and MRC-5 regained their ability to respond to TGFβ, resulting in fibroblast activation and ECM production (Fig. 2a, b and Supplementary Fig. 2h–k). To determine which specific BCAA is most critical for fibroblast activation, we examined the individual effects of each BCAA (Supplementary Fig. 2e–g). The results suggested that leucine alone was sufficient to induce fibroblast activation, similar to the effect observed with the full BCAA mixture (Fig. 2c, d and Supplementary Fig. 2l). Additionally, BCAAs and their corresponding branched-chain keto acids (BCKAs) are interconvertible through reversible enzymatic catalysis by BCAT1 and BCAT2¹⁹, and supplementation with BCKAs in MLFs significantly rescued TGFβ-induced fibroblast activation (Fig. 2e, f and Supplementary Fig. 2m). Immunofluorescence analysis further supported these findings, showing reduced fibroblast activation under BCAA-free conditions, and increased activation upon reintroduction of BCAAs or BCKAs (Fig. 2g, h). Moreover, we also found that BCAA restriction impaired TGFβ-induced fibroblast proliferation and migration, while supplementation with BCAAs rescued these processes (Supplementary Fig. 3a, b). We also identified a large set of genes (C2) responsive to TGFβ that were blunted under BCAA-free conditions, but were restored to normal levels by BCAA or BCKA supplementation (Fig. 2i). We refer to the genes in the C2 cluster as BCAA-dependent genes. These genes were highly enriched in pathways such as collagen fibril organization and actin filament organization (Fig. 2j). Indeed, the TGFβ-induced ECM genes, including Col1a1, Col5a1, Col5a3, and Col7a1, showed reduced responsiveness to TGFβ under BCAA-free conditions, but exhibited high responsiveness to TGFβ upon BCAA or BCKA supplementation (Fig. 2k and Supplementary Fig. 3c).

Fig. 2. BCAA is crucial for myofibroblast activation and lung fibrosis development.

a–f Representative immunoblot analysis of COL1A1, ACTA2, and FN1 in mouse lung fibroblasts (MLFs) treated with TGF-β1 under the indicated conditions. Quantification of COL1A1 was performed using ImageJ (b, d, f; n = 4 for b; n = 3 for d and f). g, h Representative immunofluorescence images (g) and quantification analysis (h, n = 3) of COL1A1 (red) and ACTA2 (green) expression in MLFs cultured in the indicated conditions. Scale bars: 100 μm. i Heatmap of differentially expressed genes (DEGs) from RNA-seq analysis of MLFs cultured in the indicated conditions (n = 3). DEGs were identified using DESeq2 with q-value < 0.01 and fold change ≥2. j Gene ontology (GO) biological process enrichment analysis of Cluster 2 genes identified in (i). P-values were calculated by one-sided hypergeometric test with Benjamini–Hochberg FDR correction. k Quantitative RT-PCR analysis of collagen-related genes in MLFs treated as in (a) (n = 3). l Schematic diagram of animal study. m, n Quantification of BCAA concentrations in serum (m, n = 6) and lung tissues (n, n = 5) of mice from indicated groups. o, p Representative axial (top) and coronal (bottom) micro-CT images (o) and quantitative assessment of lung fibrosis severity (p) via micro-CT analysis of mouse lungs from indicated groups (n = 5). q Representative images of Masson staining of mouse lungs from indicated groups. Lower left panels (scale bar: 100 μm) are magnified views of upper left panels (scale bar: 1 mm). Quantifications of collagen deposition in lung tissues based on Masson staining were shown on right panel (n = 8). r Hydroxyproline content of lung tissues (n = 6 each group). s Quantitative RT-PCR analysis of collagen-related genes in lung tissues from indicated groups (n = 8 in PBS group; n = 12 in BLM group; n = 7 in BLM + BCAA group). t Immunoblots of COL1A1, COL1A2 and FN1 in lung tissues from indicated groups. Quantifications were shown on right (n = 4). The value of n indicates biologically independent samples. Data are presented as mean ± SEM (m–t) or mean ± SD (b, d, f, h, k). Statistical significance was assessed by two-way ANOVA with correction for multiple comparisons (b, d, f, h, and k), by one-way ANOVA with correction for multiple comparisons (p, q–t), or two-tailed unpaired Student’s t test (m and n). Source data are provided as Source Data file.

To assess whether BCAA treatment accelerates experimental lung fibrosis, we administered BCAA in the drinking water to mice one week prior to bleomycin injection, which induces lung injury and fibrosis (Fig. 2l). After 3 weeks of BCAA treatment, we confirmed a significant increase of BCAA levels in lung and blood (Fig. 2m, n). Notably, BCAA-treated mice were more susceptible to developing bleomycin-induced lung fibrosis, as evidenced by pronounced fibrotic changes observed on micro-computed tomography (microCT) scans compared to vehicle-treated, bleomycin-challenged mice (Fig. 2o, p). H&E staining confirmed increased lung fibrosis in the BCAA-treated mice, and the Ashcroft score was significantly higher in this group (Supplementary Fig. 3d). Collagen accumulation was markedly elevated in the lungs of BCAA-treated mice, as demonstrated by Masson’s trichrome staining and hydroxyproline content (Fig. 2q, r). Transcriptomic analysis revealed that BCAA treatment induced a large set of genes (C1) closely associated with ECM organization and immune response (Supplementary Fig. 3e, f). Furthermore, we confirmed that the mRNA and protein levels of several fibrotic and collagen-related genes were significantly elevated in the BCAA-treated mice (Fig. 2s, t). These data suggest that BCAA supplementation enhances myofibroblast activation and accelerates the progression of lung fibrosis both in vitro and in vivo.

BCAT2 deficiency exacerbates pulmonary fibrosis pathogenesis

BCAT1 and BCAT2 are enzymes that catalyze the reversible transamination of BCAAs, transferring amino groups from BCAAs to α-KG to produce BCKAs and glutamate²⁰. We further confirmed that BCAT1 was increased, while BCAT2 was reduced in TGFβ-treated mouse and human lung fibroblasts (Fig. 1h and Supplementary Fig. 4a). Knockdown of either BCAT1 or BCAT2 significantly promoted COL1A1 expression in response to TGFβ treatment (Supplementary Fig. 4b, c, f). In contrast, overexpression of BCAT1 substantially reduced COL1A1 expression (Supplementary Fig. 4d), which was further supported by immunofluorescence analysis showing weaker COL1A1 staining in BCAT1-overexpressing cells compared to BCAT1-negative controls (Supplementary Fig. 4e).

Given that BCAT1 is primarily expressed in the brain, whereas BCAT2 is abundantly expressed in various tissues, including the lung²¹, we specifically examined the role of BCAT2 in lung fibrosis (Fig. 3a). BCAT2 knockout (KO) mice were generated (Supplementary Fig. 4g), and primary lung fibroblasts derived from these mice exhibited elevated intracellular BCAA levels and enhanced fibroblast activation (Supplementary Fig. 4h–k). Since homozygous BCAT2 mutation (BCAT2−/−) leads to severe phenotypes, such as reduced body weight, under a regular diet²², heterozygous BCAT2+/− (BCAT2 Het) mice were used in this study. BCAT2 Het mice exhibited elevated BCAA accumulation in the blood and lungs compared to controls following bleomycin (BLM) treatment, indicative of impaired BCAA catabolism in the absence of full BCAT2 activity (Fig. 3b–d). Notably, these mice showed enhanced BLM-induced lung fibrosis, as illustrated by microCT imaging (Fig. 3e, f). H&E staining revealed more pronounced fibrosis in BCAT2 Het mice (Supplementary Fig. 4l), which was further confirmed by Masson’s trichrome staining and hydroxyproline content (Fig. 3g–i). Consistently, fibrotic gene Col1a1 was also significantly upregulated (Fig. 3j). These findings suggest that depletion of BCAT2 promotes the progression of lung fibrosis.

Fig. 3. Deficiency of BCAA catabolic gene BCAT2 accelerates pulmonary fibrosis.

a, b Schematics of BCAA catabolic reaction catalyzed by BCAT2 (a, Created in BioRender. Hu, W. (2026) https://BioRender.com/4agqzs2) and animal study (b). c, d Quantification of BCAA concentrations in serum (c, n = 6) and lung tissues (d, n = 4 in WT group; n = 6 in Het group). e, f Representative axial (top) and coronal (bottom) micro-CT images (e) and quantitative assessment of lung fibrosis severity (f) via micro-CT analysis of mouse lungs from indicated groups (n = 6). g, h Representative images of Masson staining (g) of mouse lungs from indicated groups. Lower left panels (scale bar: 100 μm) are magnified views of upper left panels (scale bar: 1 mm). Quantifications of collagen deposition were shown in right panel (h, n = 5). i Hydroxyproline content of lung tissues from indicated groups (n = 5). j Quantitative RT-PCR analysis of Col1a1 mRNA expression in lung tissues from indicated groups (n = 4, 5, 7, and 7 for each group, respectively). k Uniform manifold approximation and projection (UMAP) plot visualizing cell populations across three experiment conditions. l Bar plot showing the fraction of each cell type across experimental conditions. m, n Fibrosis (m) and injury (n) scores across conditions based on gene signature analysis (n = 8051, 5745, and 11781 for each group, respectively). Center line indicates the median; box limits denote quartiles; whiskers extend to 1.5 × IQR. P value was calculated by two-sided Wilcoxon test. o Heatmap showing expression levels of collagen genes in fibroblast subsets across experimental conditions. p, q UMAP plot (p) and relative abundance (q) of fibroblast subtypes across groups. r qRT-PCR analysis of Pparγ mRNA expressions in MLFs treated with TGF-β1 (n = 3). s The average signals of PPARγ enrichment on BCAA catabolic genes (n = 2). t Integrative Genomics Viewer (IGV) browser tracks showing PPARγ binding peaks. u Immunoblots of COL1A1 and BCAA catabolic proteins in MLFs treated with TGF-β1 in the presence or absence of Rosiglitazone (n = 3). v Schematic model illustrating PPARγ-mediated transcriptional regulation of BCAA catabolic genes in lung fibroblasts. Created in BioRender. Hu, W. (2026) https://BioRender.com/9gn0t4z. The value of n indicates biologically independent samples. Data are presented as mean ± SEM (c, d, f, h–j) or mean ± SD (r). Statistical significance was assessed by two-way ANOVA with correction for multiple comparisons (f, h–j), by one-way ANOVA with correction for multiple comparisons (r), or two-tailed unpaired Student’s t test (c and d). Source data are provided as Source Data file.

To gain a comprehensive understanding of the cellular diversity, states, and molecular pathways in BCAT2 heterozygous lungs in response to BLM, we performed scRNA-seq on cell suspensions from three groups of mice (WT PBS, WT BLM, and BCAT2 Het BLM). Unsupervised clustering analysis identified 12 distinct clusters, representing a wide range of epithelial, mesenchymal, endothelial, and immune cell populations (Fig. 3k and Supplementary Fig. 4m). Notably, we observed a reduced number of epithelial cells in the BCAT2 Het mice (Fig. 3l and Supplementary Fig. 4n). In contrast, the fibroblast and macrophage populations were significantly increased in the BCAT2 Het mice compared to WT mice after bleomycin-induced lung injury, indicating an enhanced inflammatory response and fibrosis in the BCAT2 Het mice. Consistently, both the fibrosis and injury scores were significantly higher in BCAT2 Het mice than those in WT mice (Fig. 3m, n). To better understand the cellular dynamics driving lung fibrosis in BCAT2 Het mice, we focused on fibroblast populations, which are pivotal in fibrotic progression^3,23,24. Pseudo-bulk RNA sequencing revealed a significant increase in collagen gene expression in BCAT2 Het mice (Fig. 3o and Supplementary Fig. 4o), reinforcing the link between BCAA metabolism and ECM remodeling. Gene Set Enrichment Analysis (GSEA) further corroborated these findings, demonstrating that the ECM pathway was notably enriched in the upregulated genes of BCAT2 Het mice compared to controls (Supplementary Fig. 4p). Moreover, we identified six distinct fibroblast populations and observed a significant increase in the Cthrc1-positive pathological fibroblast subtype in BCAT2 Het mice (Fig. 3p, q and Supplementary Fig. 4q, r)²⁵. Conversely, one subset of alveolar fibroblasts (Alv. Fib-1), marked by Col13a1 expression, were markedly reduced in these mice, suggesting a shift toward more activated fibroblast states^26,27.

To examine how BCAT2 and other catabolic genes respond to injury signals and reprogram BCAA metabolism in the context of lung fibrosis, we investigated the potential involvement of peroxisome proliferator-activated receptor gamma (PPARγ), a transcription factor previously shown to regulate BCAA catabolic genes in adipocytes²⁸ and implicated in the conversion of lung lipofibroblasts to myofibroblasts²⁷. We demonstrated that PPARγ expression was significantly downregulated in TGFβ-treated MLFs (Fig. 3r). Further analysis using Cut&Tag sequencing revealed that PPARγ binding to BCAA metabolic genes was markedly reduced upon TGFβ treatment compared to controls (Fig. 3s, t). Remarkably, activation of PPARγ using rosiglitazone rescued the expression of key BCAA catabolic genes and inhibited fibroblast activation (Fig. 3u and Supplementary Fig. 5a). Moreover, rosiglitazone treatment attenuated pulmonary fibrosis in bleomycin-induced mouse model, consistent with its ability to enhance BCAA catabolic gene expression and restrain fibroblast activation (Supplementary Fig. 5b–f). These findings highlight the importance of the PPARγ-BCAA catabolic genes axis in regulating BCAA metabolism and lung fibrosis (Fig. 3v).

BCAA regulates lung fibrosis through KDM4A-mediated epigenetic remodeling of H3K36me3

To elucidate the underlying mechanisms by which BCAA metabolism drives fibroblast activation and lung fibrosis, we initially investigated classical pathways known to be regulated by BCAA metabolism²⁹. These include pathways related to mTORC1 activation, TGFβ signaling, AKT activation, the HIF1α pathway, and phosphorylation of ERK and GSK3β, all of which are typically responsive to changes in BCAA availability. However, our analysis revealed that none of these classical pathways explained the observed fibroblast activation and ECM remodeling associated with BCAA metabolism in the context of lung fibrosis (Supplementary Fig. 6a). This led us to hypothesize that BCAA metabolism may exert its effects through an alternative, epigenetic mechanism^30–32. To explore this possibility, we performed ATAC-seq on MLFs treated with TGFβ, both with and without BCAA supplementation. By comparing the chromatin landscapes of fibroblasts under these different conditions, we identified 3251 open chromatin regions (C1 cluster) that were induced by TGFβ but remained unaffected under BCAA-free conditions (Fig. 4a and Supplementary Data 1 and 2). These peaks, referred to as BCAA-dependent peaks, reverted to an open chromatin state following TGFβ treatment upon re-supplementation with BCAAs (Supplementary Fig. 6b). Notably, the genes located near these BCAA-dependent peaks were enriched in pathways related to actin filament organization and ECM-receptor interaction (Supplementary Fig. 6c, d). Further analysis revealed that many of the BCAA-dependent genes were located in proximity to these BCAA-dependent peaks, particularly genes involved in ECM remodeling (Fig. 4b, c). For example, TGFβ treatment in MLFs induced stronger chromatin accessibility near collagen genes such as Col1a1, Col7a1, and Tgfβ2 (Fig. 4d). However, this induction was significantly blunted under BCAA-free conditions. In contrast, BCAA supplementation restored the chromatin accessibility at these regions, making them responsive to TGFβ once again. These results suggest that BCAA metabolism plays a crucial role in modulating the chromatin state of key fibrotic genes, thereby driving the transcriptional response that underlies fibroblast activation and lung fibrosis.

Fig. 4. BCAAs regulate lung fibrosis through KDM4A-mediated epigenetic remodeling.

a Heatmap showing chromatin accessibility profiles of MLFs cultured under different conditions (n = 2). Differentially accessible peaks were identified with FDR < 0.05 and fold change ≥1.5. b Venn diagram illustrating the overlap between BCAA-dependent genes and BCAA-dependent ATAC peaks. c GO analysis of overlapped BCAA-dependent genes. P-values were calculated by one-sided hypergeometric test with Benjamini–Hochberg FDR correction. d IGV tracks depicting chromatin accessibility at representative loci. e Average enrichment profiles of H3K36me3, H3K9me3, H3K27me3, H3K4me3, and H3K27ac on BCAA-dependent genes (n = 2). f IGV tracks showing the histone modification signals at fibrotic gene loci. g Heatmap showing the expression profile of H3K36me3-associated epigenetic regulators. q value < 0.01 and fold change ≥2 were used to mark the significant differentially expressed genes. h Representative immunoblot analysis of KDM4A and COL1A1 expression in MLFs cultured under various conditions. i Immunoblot analysis of COL1A1 expression in MLFs overexpressing KDM4A following TGF-β1 stimulation. j, k Representative immunofluorescence images of MLFs stained for COL1A1 (j, red) or H3K36me3 (k, red) together with FLAG-KDM4A (green) under the indicated conditions. Scale bars, 20 μm. n = 20 (TGF-β1⁻ FLAG⁺) and n = 19 (other groups) in (j); n = 23 (TGF-β1⁻ FLAG⁻), 14 (TGF-β1⁻ FLAG⁺), 19 (TGF-β1⁺ FLAG⁻), and 11 (TGF-β1⁺ FLAG⁺) in (k). l Representative immunoblot analysis was performed to assess the expression levels of COL1A1, KDM4A, and H3K36me3 in MLFs cultured in the indicated conditions. m Average H3K36me3 enrichment on BCAA-dependent genes in the indicated conditions (n = 2). n IGV tracks showing the modification signals of H3K36me3 at fibrotic gene loci. o Representative immunoblot analysis of COL1A1 expression in MLFs following pharmacological inhibition of KDM4A with ML324 under TGF-β1 stimulation. p Schematic model illustrating KDM4A-mediated epigenetic regulation of collagen genes. Created in BioRender. Hu, W. (2026) https://BioRender.com/syxqqb0. The value of n indicates biologically independent samples. Data are presented as mean ± SD (j and k). Statistical significance was assessed by two-way ANOVA with correction for multiple comparisons (j and k). Source data are provided as Source Data file.

Given the dramatic epigenetic remodeling mediated by BCAA metabolism during fibrosis, we next sought to perform a more comprehensive epigenetic analysis to further investigate the histone modifications involved in the transcriptional reprogramming of fibroblasts during fibrosis. We focused on a range of histone modifications known to be crucial for transcriptional regulation, including H3K4me3, H3K27ac, H3K27me3, H3K36me3, and H3K9me3³³. By performing Cut&Tag on MLFs treated with TGFβ in the presence or absence of BCAAs, we revealed the mild to moderate histone modifications reprogramming following BCAA deprivation (Supplementary Fig. 6e and Supplementary Data 3). Notably, the genes near the downregulated histone modifications such as H3K36me3, H3K4me3, and H3K27ac, which are associated with active transcription, were enriched for fibrotic gene programs (Supplementary Fig. 6f). In contrast, we identified that cell-cell adhesion pathways were enriched for upregulated peaks of the repressive histone modifications, such as H3K9me3 and H3K27me3 (Supplementary Fig. 6f), which are typically associated with transcriptional silencing³⁴. It is noteworthy that the H3K36me3 modification exhibited the most dramatic remodeling near the BCAA-dependent genes (Fig. 4e). A substantial overlap was observed between the BCAA-dependent genes and the BCAA-regulated H3K36me3 peaks, with the overlapping genes being highly enriched in ECM-related pathways (Supplementary Fig. 6g, h). For example, we observed a significant reduction in H3K36me3 marks at the genomic loci of Col1a1 and Col5a1 (Fig. 4f). The alteration in H3K36me3 patterns suggests that this histone mark plays a plausible role in regulating the transcriptional response of fibroblasts to BCAA metabolism during lung fibrosis.

The H3K36me3 modification is tightly regulated by a balance between histone methyltransferases and demethylases³⁵. To investigate how BCAA metabolism influences the loss of H3K36me3 marks, we examined the expression levels of histone methylases and demethylases targeting H3K36me3 in TGFβ-treated MLFs, both with and without BCAA supplementation. Among the candidate enzymes, we observed a significant upregulation of lysine-specific demethylase 4 A (KDM4A), a key enzyme responsible for the removal of H3K36me3 marks³⁶, in response to BCAA depletion under TGFβ treatment (Fig. 4g). Notably, the increased expression of KDM4A was reversed upon reintroduction of BCAA (Fig. 4g, h and Supplementary Fig. 6i). Moreover, we found that KDM4A expression was significantly reduced in TGFβ-treated MLFs and MRC-5 cells, both at the mRNA and protein levels (Supplementary Fig. 6j–l). Indeed, overexpression of KDM4A inhibited the expression of COL1A1 (Fig. 4i, j and Supplementary Fig. 7a), while knockdown of KDM4A in MLFs or MRC-5 cells enhanced TGFβ-induced collagen production (Fig. 4l and Supplementary Fig. 7b, c). Consistently, KDM4A overexpression led to a global reduction in H3K36me3 levels (Fig. 4k), while KDM4A knockdown increased overall H3K36me3 abundance and its enrichment near BCAA-dependent genes, including Col5a3 and Col7a1 (Fig. 4l–n). Moreover, increasing BCAA concentrations reduced KDM4A expression and promoted fibroblast activation in a dose-dependent manner (Supplementary Fig. 7d, e), establishing a direct regulatory link between BCAA availability and KDM4A-mediated fibroblast activation. Pharmacological inhibition of KDM4A demethylase activity using ML324 phenocopied the effects of KDM4A loss (Fig. 4o and Supplementary Fig. 7f). These data suggest the involvement of the BCAA-KDM4A-H3K36me3 axis in regulating fibroblast activation and the progression of lung fibrosis (Fig. 4p).

Pulmonary fibrosis is alleviated by dietary BCAA restriction

Building on our previous findings that BCAA accumulation contributes to lung fibrosis, we investigated whether decreasing BCAA bioavailability through dietary intervention could attenuate lung fibrosis³⁷. Mice were placed on a BCAA-free diet one week after bleomycin-induced fibrosis to assess its therapeutic potential (Fig. 5a). The BCAA-free diet resulted in a significant reduction of BCAA concentrations in blood and lung tissues, as measured three weeks after bleomycin treatment (Fig. 5b, c). Mice on the BCAA-free diet showed significantly reduced fibrotic changes, as evidenced by microCT scans (Fig. 5d, e), with less lung fibrosis and collagen deposition, as shown by H&E, Masson’s trichrome staining, and hydroxyproline content measurement (Fig. 5f, g and Supplementary Fig. 8a). At the molecular level, transcriptomic analysis revealed a marked attenuation of the fibrotic gene program in BCAA-free diet-treated mice (Supplementary Fig. 8b, c). Notably, we found that BCAA restriction inhibited the expression of several key collagen genes (Fig. 5h and Supplementary Fig. 8d). To further test whether earlier dietary intervention could prevent the development of lung fibrosis, we treated these mice using BCAA-free diet one week before the bleomycin-induced fibrosis (Fig. 5i). This intervention markedly reduced BCAA concentrations in both blood and lung tissues (Fig. 5j, k). Interestingly, we found that initiating BCAA restriction before the induction of fibrosis significantly prevented the onset of fibrotic changes. Mice on the BCAA-free diet starting one week prior to bleomycin treatment exhibited substantially less lung fibrosis compared to control animals, as assessed by microCT (Fig. 5l), H&E (Supplementary Fig. 8e), Masson’s trichrome staining (Fig. 5m), and hydroxyproline quantification (Fig. 5n). This early intervention also resulted in lower collagen deposition (Supplementary Fig. 8f, g). These findings suggest that BCAA restriction not only slows the progression of established fibrosis but may also have preventive effects when applied before fibrosis onset.

Fig. 5. Pulmonary fibrosis is alleviated by dietary BCAA restriction or inhibition of BCAA uptake.

a Schematic overview of the dietary BCAA restriction study in BLM-treated mice. b, c Quantification of BCAA concentrations in serum (b, n = 6) and lung tissues (c, n = 5) of mice from indicated groups. (d–f, l, m, and s, t) Representative lung micro-CT images and quantification of fibrosis scores, and Masson’s trichrome staining of mouse lungs (scale bars, 1 mm and 100 μm). n = 9, 10, and 10 per group for (d–f); n = 3 (fibrosis scores) and 5 (collagen quantification) for (l, m); and n = 5 (fibrosis scores) and 6, 9, and 8 per group (collagen quantification) for (s, t). g Hydroxyproline content of lung tissues in various groups (n = 6, 10, and 9 per group). h Quantitative RT-PCR analysis of collagen-related genes in lung tissues from indicated groups (n = 9, 10, and 10 per group). i Schematic overview of the early dietary BCAA restriction study in BLM-treated mice. j, k Quantification of BCAA concentrations in serum (j, n = 4) and lung tissues (k, n = 5) of mice from indicated groups. n Hydroxyproline content of lung tissues in various groups (n = 6). o, p Schematic overview of the experimental design for BCAA transporter inhibition using BCH in vivo. Created in BioRender. Hu, W. (2026) https://BioRender.com/h5j01ha. q, r Quantification of BCAA concentrations in serum (q, n = 4) and lung tissues (r, n = 6) of mice from indicated groups. u Immunoblot analysis of ATF4 protein expression in MLFs treated with increasing concentrations of TGF-β1 (n = 3). Results are representative of two independent experiments. v IGV tracks showing ATF4 binding around the Slc7a5 locus. w Quantitative RT-PCR analysis of Slc7a5, Atf4, Col1a1, and Col1a2 in MLFs following ATF4 knockdown (n = 3). x Schematic summary of ATF4-mediated transcriptional regulation of Slc7a5 in fibrotic MLFs. Created in BioRender. Hu, W. (2026) https://BioRender.com/xxs3cgn. The value of n indicates biologically independent samples. Data are presented as mean ± SEM (b, c, e–h, j–n, q–t) or mean ± SD (w). Statistical significance was assessed by one-way ANOVA with correction for multiple comparisons (e–h, l–n, s, t, and w), or two-tailed unpaired Student’s t test (b, c, j, k, q, and r). Source data are provided as Source Data file.

Given that complete deprivation of essential amino acids may non-specifically suppress protein synthesis, we further included dietary groups containing 25, 50, or 100% of normal BCAA levels (Supplementary Fig. 8h). Notably, the extent of fibrosis correlated positively with dietary BCAA content, as assessed by microCT (Supplementary Fig. 8i, j), H&E staining (Supplementary Fig. 8k, l), Masson’s trichrome staining (Supplementary Fig. 8k, m), and molecular profiling (Supplementary Fig. 8n), revealing a clear dose-dependent relationship between BCAA intake and fibrotic severity. Thus, partial BCAA restriction effectively ameliorated fibrotic outcomes without causing overt health deficits, underscoring its potential as a safe and feasible dietary strategy for preventing fibrosis.

We next evaluated whether BCAA restriction could exert protective effects in an established, non-resolving model of lung fibrosis induced by repeated bleomycin administration³⁸. In this setting, mice received the BCAA-free diet after four dose of bleomycin challenge, when fibrotic lesions were already apparent (Supplementary Fig. 9a, b). Remarkably, dietary BCAA deprivation markedly attenuated further fibrotic progression, as demonstrated by microCT imaging (Supplementary Fig. 9b), H&E and Masson’s trichrome staining (Supplementary Fig. 9c, d). Consistently, expression of profibrotic markers, including Col1a1, Col1a2, were significantly decreased in the lungs of BCAA-restricted mice (Supplementary Fig. 9e). These findings indicate that BCAA restriction not only prevents fibrosis onset but also mitigates the progression of established fibrotic disease.

Inactivation of SLC7A5 disrupts BCAA uptake and mitigates lung fibrosis

To further investigate the role of BCAA uptake in lung fibrosis, we focused on SLC7A5, the key transporter responsible for BCAA uptake³⁹. SLC7A5 facilitates the cellular uptake of BCAAs, and its expression is upregulated in IPFs (Supplementary Fig. 9f)⁴⁰. Consistently, we observed that SLC7A5 was significantly upregulated in both TGFβ-treated IPF lung fibroblasts and MRC-5 cell lines (Supplementary Fig. 9g, h), suggesting that enhanced BCAA transport plays a pivotal role in fibroblast activation and the pathogenesis of lung fibrosis. Indeed, knockdown or inhibition of SLC7A5 using BCH (2-amino-2-norbornanecarboxylic acid), a selective inhibitor of the transporter⁴¹, significantly reduced the expression of COL1A1 (Fig. 5o and Supplementary Fig. 9i, j). We then treated mice with BCH during the progression of bleomycin-induced fibrosis to explore the therapeutic potential of SLC7A5 inhibition (Fig. 5p). BCH treatment resulted in lower BCAA concentration in blood and lung tissues (Fig. 5q, r), and significant attenuation of fibrotic changes, as evidenced by micro-CT scans (Fig. 5s), H&E (Supplementary Fig. 9k), and Masson’s trichrome staining (Fig. 5t). Additionally, a significant downregulation of the key fibrotic genes COL1A1 was observed in BCH-treated mice (Supplementary Fig. 9l). These findings suggest that blocking BCAA uptake through SLC7A5 inhibition can suppress fibroblast activation and ECM production in the context of lung fibrosis.

To explore the upstream regulation of SLC7A5 in lung fibrosis, we identified ATF4, a transcription factor known for its role in amino acid metabolism⁴², as a potential regulator. ATF4 is upregulated in IPF and has been implicated in the activation of myofibroblasts⁴³. Consistent with this, our study demonstrates that ATF4 expression was significantly increased in MLFs following TGFβ treatment (Fig. 5u and Supplementary Fig. 9m). Through high-fidelity ATF4 Cut&Tag sequencing in MLFs (Supplementary Fig. 9n, q), we revealed that ATF4 directly bound to the genomic region of Slc7a5 (Fig. 5v). Knockdown of ATF4 significantly reduced Slc7a5 expression, leading to a corresponding reduction in TGFβ-induced fibrotic gene expression, including key genes associated with ECM deposition (Fig. 5w). These findings highlight the critical role of the ATF4-SLC7A5 axis in regulating fibroblast activation and ECM production during lung fibrosis (Fig. 5x).

Enhancing BCAA catabolism by BT2 ameliorates pulmonary fibrosis

In addition to restricting BCAA uptake, we also explored whether enhancing BCAA catabolism could similarly affect the development and progression of lung fibrosis. BT2, a potent inhibitor of BCKDK, promotes the activity of the BCKDH complex, which enhances BCAA catabolism (Fig. 6a)⁴⁴. We examined the effect of BT2 on TGFβ-induced fibroblast activation. Our results demonstrated that BT2 inhibited fibroblast activation in a dose-dependent manner (Supplementary Fig. 10a–d). To investigate the in vivo effects of BT2, we established an acute high-dose bleomycin-induced fibrosis model and treated the mice with BT2 immediately following bleomycin administration (Supplementary Fig. 10e). We observed that BT2 treatment significantly reduced the severity of fibrosis, as assessed by micro-CT imaging (Supplementary Fig. 10f), H&E staining (Supplementary Fig. 10g), and Masson’s trichrome staining (Supplementary Fig. 10h). Mice treated with BT2 exhibited reduced collagen deposition compared to the vehicle-treated group (Supplementary Fig. 10i, j). Additionally, transcriptomic analysis revealed that BT2 treatment led to a marked reduction in the expression of fibrotic genes associated with ECM remodeling (Supplementary Fig. 10k–m).

Fig. 6. Enhancing BCAA catabolism mitigates pulmonary fibrosis.

a, b Schematic illustration of the BCAA catabolic pathway and the mechanism of action of the BCKDK inhibitor BT2 (a) and the BT2 treatment regimen in the BLM-induced fibrosis model (b). c, d Quantification of BCAA concentrations in serum (c, n = 4) and lung tissues (d, n = 5) of mice from indicated groups. e, l, and m Representative lung micro-CT images (e, l) and quantification of fibrosis scores (e, m; n = 5 for e and n = 3 for m). f, g, n, and o Representative H&E (f, n) and Masson’s trichrome (g, o) staining of mouse lungs (Scale bars, 1 mm and 100 μm). Quantification of fibrosis severity (H&E; f, n) and collagen deposition (Masson; g, o) is shown (n = 5). h Quantitative RT-PCR analysis of Col1a1, Col5a3, and Col7a1 in lung tissues from indicated groups (n = 7, 8, and 10 per group). i Schematic overview of animal experiment using BT2 and Nintedanib (NTB) intervention. j, k Quantification of BCAA concentrations in serum (j, n = 3) and lung tissues (k, n = 5) of mice from indicated groups. p Quantitative RT-PCR analysis of Col1a1, Col1a2, and Col5a1 mRNA expressions from lung tissues in various groups (n = 6, 7, 7, 6, and 6 per group). The value of n indicates biologically independent samples. Data are presented as mean ± SEM (c–h, j, k, and m–p). Statistical significance was assessed by one-way ANOVA with correction for multiple comparisons (e–h, j, k, and m–p), or two-tailed unpaired Student’s t test (c and d). Source data are provided as Source Data file.

We next examined the therapeutic potential of BT2 in the progressive bleomycin-induced lung fibrosis model. BT2 was administered one week after bleomycin injection when fibrosis was well-established (Fig. 6b). BT2 treatment enhanced BCAA catabolic flux, leading to reduced BCAA concentrations in both serum and lung tissues (Fig. 6c, d). Imaging, histological assessments, and molecular analyses all supported the conclusion that BT2 treatment could prevent the progression of lung fibrosis. Micro-CT imaging showed less pronounced fibrotic changes (Fig. 6e), while H&E (Fig. 6f) and Masson’s trichrome staining (Fig. 6g) revealed reduced collagen deposition in the lungs of BT2-treated mice, which was confirmed by decreased expression of collagen genes in BT2-treated mice (Fig. 6h). Moreover, to further evaluate the therapeutic potential of BT2, we compared its effects to those of the FDA-approved drug, Nintedanib (NTB), a well-established treatment for IPF (Fig. 6i–k). Our results revealed that both BT2 and NTB effectively reduced the severity of lung fibrosis through multimodal analysis, including micro-CT imaging (Fig. 6l, m), H&E staining (Fig. 6n), Masson’s trichrome staining (Fig. 6o), and transcriptional profiling (Fig. 6p). Both treatments demonstrated similar reductions in fibrotic markers and ECM deposition, indicating their efficacy in modulating the fibrotic process. Although the combination of BT2 and NTB did not produce synergistic effects in our model, these data suggest that BT2, as an adjunct to existing therapies like NTB, could offer a potential strategy for managing lung fibrosis.

Clinical relevance of BCAA metabolic dysregulation in IPF patients

To assess the clinical relevance of BCAA metabolic remodeling in IPF, we reanalyzed previously published transcriptomic data from a larger lung dataset, which included 123 IPF patients and 91 healthy controls⁴⁵. Consistent with our findings in the mouse model, we observed that most of BCAA catabolic genes were significantly downregulated in IPF lungs, whereas SLC7A5 and BCAT1 were upregulated in the lungs of IPF patients (Supplementary Fig. 11a). Reanalysis of another study focusing on invasive IPF fibroblasts further confirmed these results²¹, showing elevated expression of SLC7A5 and BCAT1, alongside reduced expression of BCAA catabolic genes in invasive lung fibroblasts compared to non-invasive lung fibroblasts (Supplementary Fig. 11b). Furthermore, we collected lung tissues from both IPF patients and healthy controls and performed RT-qPCR and Western blot analyses. These findings corroborated the transcriptomic data, revealing impaired expression of BCAA catabolic genes and increased levels of BCAA uptake genes in the IPF lungs (Fig. 7a, b and Supplementary Fig. 11c). Immunohistochemical staining of an independent IPF cohort further substantiated these results, showing increased SLC7A5 protein expression and diminished BCAT2 levels in fibrotic lesions (Fig. 7c).

Fig. 7. Impaired BCAA catabolism and increased BCAA uptake are associated with lung fibrosis in IPF patients.

a–c Quantitative RT-PCR (a) and immunoblot (b) analyses of BCAA metabolic and fibrosis-related genes in lung tissues from patients with IPF (n = 8) and control subjects (n = 10 for a; n = 4 for b), and representative images and quantification of SLC7A5 and BCAT2 staining in human lung tissues (c; n = 9 (IPF) and 7 (control)). d Partial Spearman correlation of BCAA metabolic genes with fibrosis-associated genes, based on publicly available GEO dataset GSE47460. Dot sizes are proportional to −log10 (adjusted P values). Partial Spearman correlations and corresponding P values were calculated using two-sided tests, with P values adjusted for FDR separately for each gene. e Semipartial Spearman correlation of BCAA metabolic genes with lung function, based on publicly available GEO dataset GSE47460. Correlation coefficients and P values were calculated using a two-sided semi-partial Spearman correlation test. f Linear relationship and association between lung function, measured as DLCO % predicted and log of age- and sex-adjusted expression of BCAA metabolic genes (n = 211), based on publicly available GEO dataset GSE47460. Semi-partial correlations were performed using the pcor.test function from the RVAideMemoire R package. A robust linear regression model was used to estimate the association. P-values were calculated using two-sided tests. g Quantification of serum BCAA concentration in patients with IPF (n = 33). h, i Pearson correlation analyses (Two-tailed) between serum leucine levels and lung function parameters: FVC (n = 27), FVC% predicted (n = 27), DLCO (n = 23), and DLCO% predicted (n = 23). Some clinical parameters were unavailable for certain patients due to technical limitations. j, k Pearson correlation analysis (Two-tailed) between serum leucine levels and clinical severity scores: GAP index (j) and modified Medical Research Council (mMRC) scale (k) (n = 33). l–o Immunoblot analysis and quantification of COL1A1 (l, n; n = 3) and COL1A1 mRNA expression (m, o; n = 3) in lung fibroblasts from patients with IPF treated with BCH or BT2. The value of n indicates biologically independent samples. Data are presented as mean ± SEM (a, c, and g) or mean ± SD (l–o). Statistical significance was assessed by one-way ANOVA with correction for multiple comparisons (g and l–o), or two-tailed unpaired Student’s t test (a and c). Source data are provided as Source Data file.

Since physiological lung parameters were available alongside the transcriptomic data, we constructed linear regression models to examine the correlation between the RNA expression of BCAA metabolic genes and functional lung parameters, including lung function and fibrosis markers^45,46. We found that the RNA levels of SLC7A5 and BCAT1 were positively correlated with fibrosis marker genes, while the majority of BCAA catabolic genes showed negative correlations with fibrosis marker genes (Fig. 7d). Importantly, there was a strong negative correlation between the expression levels of SLC7A5 and BCAT1 and key lung function parameters, including DLCO (diffusing capacity of the lung for carbon monoxide), FEV1 (forced expiratory volume in 1 s), and FVC (forced vital capacity) in individuals (Fig. 7e, f and Supplementary Fig. 11d). These findings highlight the potential clinical relevance of BCAA metabolism in modulating fibrosis and lung function in IPF, underscoring the value of BCAA metabolic pathways as biomarkers or therapeutic targets for lung fibrosis.

To further explore the relationship between circulating BCAA levels and disease severity, we performed targeted metabolomic profiling of leucine, isoleucine, and valine in a small cohort of IPF patients (Supplementary Table 1). Patients were stratified into three groups based on the prognostic gender-age-physiology (GAP) index⁴⁷. Notably, serum levels of all three BCAAs declined significantly with increasing disease severity (Fig. 7g), mirroring the reduction observed in the culture medium of TGFβ-treated MLFs, and suggesting enhanced BCAA uptake by lung fibroblasts and fibrotic lung tissue (Fig. 1j, k). This reduction is likely attributable to enhanced BCAA uptake in fibrotic lungs, consistent with the upregulation of BCAA transport observed in IPF fibroblasts. Correlation analysis revealed a strong positive association between circulating BCAA levels and pulmonary function metrics. Higher serum leucine levels were significantly associated with improved FVC, FVC% predicted, DLCO, and DLCO% predicted (Fig. 7h, i). Similar trends were observed for isoleucine and valine, reinforcing the link between BCAA availability and preserved pulmonary function in IPF (Supplementary Fig. 11e–h). In parallel, we assessed whether serum BCAA levels were associated with clinical indices of disease progression. Using GAP staging and modified Medical Research Council (mMRC) scores⁴⁸, we found that higher circulating BCAA levels were associated with milder disease stages, suggesting a potential role for serum BCAAs as protective metabolic markers in the context of IPF severity (Fig. 7j, k and Supplementary Fig. 11i, j). To investigate the therapeutic potential of targeting BCAA metabolism in IPF, we treated primary lung fibroblasts derived from IPF patients with BT2 and BCH drugs. Remarkably, treatment with either compound led to a robust reduction in the expression of profibrotic genes (Fig. 7l–o), suggesting that modulation of BCAA metabolic pathways may hold clinical promise as a therapeutic strategy for pulmonary fibrosis.

Discussion

In this study, we demonstrated that pulmonary fibrosis is associated with dramatic reprogramming of BCAA metabolism, characterized by impaired BCAA catabolism and increased BCAA uptake, although the latter remains challenging to quantify directly in the current study. BCAA metabolism acts as a crucial metabolic checkpoint essential for myofibroblast activation during lung fibrosis. Supplementation of BCAAs promoted myofibroblast activation and fibrosis progression, whereas inhibition of BCAA uptake via targeting SLC7A5 or enhancing BCAA catabolism through the BCKDK inhibitor BT2 effectively alleviated pulmonary fibrosis. Through integrative analysis of transcriptomic, metabolomics, and epigenomic data, we showed that ATF4 and PPARγ regulated the expression of SLC7A5 and BCAA catabolic genes, respectively. The resultant BCAA accumulation induced profound epigenetic remodeling, notably H3K36me modification of fibrotic genes, mediated through KDM4A. These findings underscore the critical role of this metabolic-epigenetic signaling axis in regulating lung fibroblast function and suggest potential therapeutic targets for pulmonary fibrosis.

Our findings underscore the critical role of BCAA metabolism in myofibroblast activation and lung fibrosis, demonstrating that BCAA dysregulation is both required and sufficient for fibrosis development, with lowering BCAA levels offering a protective effect in halting lung fibrosis progression. Although complete deprivation of essential amino acids can non-specifically suppress protein synthesis, our results show that partial BCAA restriction markedly ameliorates fibrotic outcomes without inducing overt health abnormalities, supporting its potential as a safe and practical dietary strategy for fibrosis prevention. While previous studies have implicated BCAAs in fibrosis across multiple organs, such as the liver, heart, and kidneys, their roles appear to be context-dependent, varying significantly among tissues. For example, increased BCAA concentrations have been observed in the livers of cirrhotic patients compared to healthy donor tissues, with this increase more pronounced in those with severe fibrosis⁴⁹. However, circulating BCAA levels are decreased in patients with liver cirrhosis⁵⁰. In non-alcoholic steatohepatitis (NASH), BCAAs have been shown to prevent hepatic fibrosis in mouse models, suggesting a potentially beneficial role of BCAA in liver fibrosis⁵¹. The function of BCAAs in kidney fibrosis is also controversial. One study found that downregulation of BCAA catabolic genes through Klf6 and subsequent BCAA accumulation alleviated kidney fibrosis¹², while another study showed that exogenous BCAA administration attenuated renal fibrosis¹³. In the heart, BCAA catabolic genes are suppressed in failing murine and human hearts, and recent studies suggest that extra-cardiac BCAA catabolism lowers blood pressure and protects against heart failure⁴⁴. Moreover, in cardiomyopathy patients with impaired BCAA catabolism, cardiac fibrosis formation was increased^52,53. These contrasting findings highlight the complex, organ-specific roles of BCAAs in fibrotic diseases, potentially driven by different mechanisms. Understanding the specific mechanisms that contribute to these organ-specific phenotypes will help elucidate the context-dependent roles of BCAA metabolism in regulating fibrogenesis and provide insights into potential therapeutic strategies.

The fibroblast population in the lungs exhibits remarkable heterogeneity, with distinct subsets originating from different sources and playing unique roles in the pathogenesis of fibrosis^3,23,24. These fibroblast subsets include resident alveolar fibroblasts, adventitial fibroblasts, lipofibroblasts, myofibroblasts, and other lineage-committed progenitors. Each subset contributes differentially to the fibrotic microenvironment through processes such as activation, differentiation, and extracellular matrix deposition. Resident alveolar fibroblasts, under normal physiological conditions, maintain alveolar homeostasis. However, recent studies have identified them as the predominant source of multiple emergent fibroblast subsets, a process that is driven by inflammatory and pro-fibrotic signals following lung injury³. In our scRNA-seq analysis, we observed a significant increase in the proportion of myofibroblasts in BCAT2-deficient mice, suggesting a potential role for BCAA metabolism in regulating fibroblast differentiation. However, the precise mechanisms by which BCAA metabolism influences each fibroblast subset and their cell fate decisions remain poorly understood. Future studies should focus on fibroblast subtype-specific perturbations of BCAA metabolic pathways, such as fibroblast-specific BCAT2 knockout models, to elucidate the distinct contributions of BCAA metabolism to the activation and differentiation of these subsets in pulmonary fibrosis. Additionally, we found a decrease in AT2 cell numbers in bleomycin-induced BCAT2-deficient mice. The underlying mechanism, however, remains unclear. Notably, dysregulation of the BCAA metabolic pathway was also evident in transitional KRT8⁺ AT2 cells. Whether BCAA directly regulates AT2 cell self-renewal and differentiation, or whether it operates through cell-cell communication pathways, requires further investigation. Given the complex cellular composition and the involvement of multiple cell types in pulmonary fibrosis, including macrophages and endothelial cells⁵⁴, elucidating the cell-type-specific roles of BCAA in lung fibrosis will significantly advance our understanding of the disease pathogenesis.

We observed that most BCAA catabolic genes were downregulated in response to fibrotic stimuli, with the exception of BCAT1, which was significantly upregulated, as well as BCAA transporter SLC7A5. We hypothesize that BCAT1 acts as a downstream target of TGFβ-SMAD signaling in pulmonary fibroblasts. Upon TGFβ-induced upregulation, BCAT1 exerts an inhibitory effect on the TGFβ pathway, thereby establishing a negative feedback loop. Previous studies have shown that BCAT1 catalyzes the reverse transamination of BCKAs to BCAAs to fuel cancer cell growth⁵⁵. BCAT1 and BCAT2 likely exert distinct catalytic and functional roles in lung fibroblasts. BCAT1 predominantly mediates the amination of branched-chain α-keto acids (BCKAs) to synthesize BCAAs, whereas BCAT2 catalyzes BCAA deamination to generate BCKAs. This functional divergence may arise from their differential subcellular localizations (cytoplasmic BCAT1 versus mitochondrial BCAT2) and/or context-dependent substrate availability, which warrants further investigation. Importantly, the observation that BCAT1 knockdown promotes fibroblast activation despite reducing intracellular BCAA levels suggests that BCAT1 also exerts BCAA-independent functions. Previous studies have implicated BCAT1 in regulating processes such as reactive oxygen species (ROS) production⁵⁶ and DNA damage responses⁵⁷, both of which may contribute to fibroblast activation independently of BCAA abundance.

Moreover, the expression of BCAA metabolic genes is regulated by various mediators in different cellular contexts^12,58,59. In lung fibroblasts, we demonstrated that PPARγ mediates the downregulation of BCAA catabolic genes during lung fibrosis²⁸. The PPARγ signaling pathway, which promotes fatty acid oxidation and directly facilitates ECM degradation through CD36, is often downregulated in fibroblasts across various fibrotic diseases^60,61. Given the extensive genomic binding sites of PPARγ on these BCAA catabolic genes, the reduction in PPARγ expression likely contributes to impaired BCAA metabolism in fibrosis. Thus, activation of the PPARγ pathway may serve as a potential therapeutic strategy to prevent lung fibrosis, not only through its classical role in lipid metabolism but also via its interplay with BCAA metabolism. ATF4, a basic leucine zipper transcription factor traditionally associated with stress responses, is induced in lung fibroblasts early after TGFβ exposure. Knockdown of ATF4 prevents collagen accumulation downstream of TGFβ by inhibiting the serine-glycine synthesis enzymes⁶². Recent studies have also shown that ATF4 and mTOR regulate glycolytic and TCA cycle metabolites in TGFβ-treated lung fibroblasts⁶³. In our study, we demonstrated that ATF4 directly binds the enhancer region of SLC7A5 and regulates both BCAA uptake and myofibroblast activation.

To explore the underlying mechanisms by which BCAA influences IPF progression, we investigated the classical pathway targeted by BCAA and involved in myofibroblast activation. Interestingly, depletion of BCAA did not inhibit lung fibrosis through the mTOR pathway, AKT signaling, or other well-known fibrosis-related pathways such as HIF1α signaling, GSK3β, TGFβ signaling, and ERK pathway. This led us to propose an alternative mechanism wherein BCAA metabolism intersects with epigenetic regulation to control fibrotic gene expression. Our analysis revealed significant chromatin accessibility remodeling in BCAA-depleted conditions in response to fibrotic stimuli. A detailed examination of histone modifications revealed a widespread redistribution of the H3K36me3 mark. Notably, genes associated with ECM production and fibrosis showed a reduction in H3K36me modification in the absence of BCAA. By systematically examining the histone methylases and demethylases involved in BCAA-induced fibrosis, we identified that KDM4A is upregulated in BCAA-depleted conditions, but its expression returns to baseline levels upon BCAA repletion in the context of fibrosis. However, the precise regulatory mechanism underlying this regulation remains unclear. Previous studies suggest that BCAA-induced ROS signaling may suppress KDM4A expression^64,65. Our data suggest the presence of a BCAA-KDM4A-H3K36me axis in regulating lung fibrosis. Consistent with this, previous studies have highlighted the role of BCAA catabolic metabolites in modulating protein acetylation and histone methylation across various physiological and pathological conditions^19,66. Future research into the interactions between BCAA metabolic reprogramming and epigenetic remodeling will provide deeper insights into the intricate regulatory network governing lung fibroblast homeostasis.

There is a critical need to understand the causes of BCAA metabolic dysfunction in IPF patients. Our findings reveal that transcriptional remodeling of BCAA metabolic genes and BCAA accumulation in the lung are strongly correlated with pulmonary fibrosis and deteriorating lung function in IPF patients⁶⁷. Interestingly, circulating BCAA levels in IPF patients show a positive correlation with lung function parameters, suggesting that BCAAs may serve as diagnostic or prognostic markers for the development and progression of IPF. Future studies involving larger, multi-center cohorts of IPF patients are crucial for assessing the diagnostic potential of serum and pulmonary BCAA profiles in distinguishing IPF patients from healthy controls. In addition to advancing non-invasive methods for identifying IPF patients, our study provides a roadmap for developing more precise and effective pharmacological treatments. We demonstrated that promoting BCAA oxidation alleviates pulmonary fibrosis. Administration of BT2 in mice, which accelerates BCAA oxidation and reduces plasma BCAA levels, effectively prevented lung fibrosis. Although BT2 is not suitable for human use, sodium phenylbutyrate (NaPB), an FDA-approved drug for urea cycle disorders, also targets BCKDK, reduces plasma BCAA levels, and could be a viable therapeutic option⁶⁸. Future studies should evaluate the clinical potential of NaPB in IPF patients. Our findings also highlight an emerging opportunity for dietary modulation, specifically the limitation of BCAA intake, as a strategy for managing IPF.

Notably, the serum concentrations of BCAAs measured in our study were a bit lower than those reported in previous studies^69,70. This difference likely reflects variations in analytical methods, sample processing, and the use of serum versus plasma, rather than biological variation, and should be considered when comparing amino acid levels across studies.

In summary, our study sheds light on the dramatic metabolic remodeling of BCAA during lung fibrosis and highlights the essential roles of BCAA in driving the progression of IPF. We reveal an unrecognized metabolism-epigenetics mechanism, coordinated by KDM4A, which regulates lung fibroblast activation and fibrosis. Furthermore, we demonstrate that targeting BCAA uptake and catabolism through pharmacological and dietary strategies offers a promising therapeutic approach to halt the progression of lung fibrosis.

Methods

Cell culture

Primary mouse lung fibroblasts (MLFs) were isolated from lungs of 6–8-week-old C57BL/6J mice using enzymatic digestion with collagenase I followed by differential adhesion⁷¹. Human lung fibroblasts (HLFs) were derived from lung tissue samples of patients with idiopathic pulmonary fibrosis (IPF) following established protocols^72,73. Minced lung tissues were digested in 2 mg/mL collagenase type IV (GIBCO, USA) at 37 °C for 30–60 min with agitation. Dissociated cells were passed through a 70 μm Nylon mesh (Falcon) to remove masses and then were washed four times in cold F12 media containing a cocktail of antibiotics, including Gentamycin (Life Technologies, Cat# 15710-064), Fungizone (Life Technologies, Cat# 15290018), and Penicillin-streptomycin (Invitrogen, Cat# 15140122). Primary Human Lung Fibroblasts were enriched by Miltenyi magnetic separation beads to remove immune (CD45+), epithelial (Epcam+), and Endothelial (CD31+) cells. Human fetal lung fibroblasts (MRC-5 cells, CL-0161) and HEK-293T cells (CL-0005) were obtained from Procell.

MLFs were maintained in DMEM/F12 medium (Gibco, 11330032) supplemented with 10% fetal bovine serum (FBS) (LONSERA, SH05713) and 1% Penicillin-Streptomycin (P/S) (Gibco, 15140122), while HLFs were cultured in DMEM (Gibco, 11965092) supplemented with 10% FBS and 1% P/S. MRC-5 cells were cultured in Minimum Essential Medium (MEM) (Gibco, 11090081) supplemented with 10% FBS, 1% GlutaMAX (Gibco, 35050061), 1% non-essential amino acids (Gibco, 11140050), 1% sodium pyruvate (Gibco, 11360070), and 1% P/S.

All primary fibroblasts were used within seven passages. Cells were cultured at 37 °C in a humidified atmosphere containing 5% CO₂. For in vitro lung fibrosis model, cells were treated with recombinant human TGF- β1 (PeproTech, 100-21) at a final concentration of 20 ng/mL for 48 h unless otherwise indicated. BCAA-free and BCAA-complete culture media (DMEM/F12) were obtained from Beijing Liweining Biotechnology (Lvn1004-BCAA and Lvn1004, respectively). The BCAA-free medium was reconstituted by supplementing with BCAAs at a final concentration of 0.45 mM L-leucine, 0.42 mM L-isoleucine, and 0.45 mM L-valine, matching the BCAA concentration in the complete formulation.

Animal studies

Wild-type C57BL/6J mice (10–12 weeks old) were purchased from Charles River Laboratories (Beijing, China). BCAT2 heterozygous mutants (BCAT2^+/−; C57BL/6 background) were obtained from Gempharmatech (Nanjing, China) and maintained under specific pathogen-free (SPF) conditions (22–26 °C, 40–60% humidity, 12 h light/dark cycle). All animal procedures were approved by the Institutional Animal Care and Use Committee of Guangzhou National Laboratory (GZLAB-AUCP-2023-02-A02).

For the bleomycin (BLM)-induced pulmonary fibrosis model, male C57BL/6J mice (10–12 weeks old) were administered a single intratracheal dose of BLM (1.25 mg/kg; Selleck, S1214). Lungs were harvested and analyzed 21 days post-administration. For the butylated hydroxytoluene (BHT)-induced lung injury model, mice received a single intraperitoneal injection of BHT (225 mg/kg body weight, Selleck, S6202) and lung tissues were collected 7 days post-injection. To evaluate the impact of BCAA supplementation on lung fibrosis, mice were provided drinking water supplemented with 30 mM L-leucine (Macklin, L812333), 30 mM L-isoleucine (Macklin, L811667), and 30 mM L-valine (Macklin, L820396) for 7 days prior to BLM exposure (1.25 mg/kg, intratracheal). With an average daily water intake of ~4.05 mL per mouse, the daily BCAA consumption was 15.9 mg L-leucine, 15.9 mg L-isoleucine, and 14.2 mg L-valine per mouse during the experimental period. Lungs were collected and analyzed 14 days after BLM challenge. To assess the role of BCAT2 in pulmonary fibrosis, wild-type (BCAT2 +/+) and heterozygous (BCAT2 +/−) mice were given intratracheal BLM (1.25 mg/kg), and lung tissues were collected at day 14.

Separate cohorts of mice were used for prevention and treatment studies. In the prevention model, mice were pre-fed either a standard control diet or a BCAA-free diet for 7 days before BLM challenge (1.25 mg/kg, intratracheal), and the respective diets were maintained throughout the experimental period. Lungs were harvested 21 days post-BLM administration. In the therapeutic model, mice received BLM (1.25 mg/kg, intratracheal), and starting from day 7 post-injury, were treated with either diets containing 0% (Dyets, Cat# D240219), 25% (50% BCAA diets and 0% diets mixed 1:1 (w/w)), 50% (Dyets, Cat# D240218), or 100% of normal BCAA levels (Dyets, Cat# D510090) or daily intraperitoneal injection of BCH (100 mg/kg, MCE, HY-108540) or BT2 (20 mg/kg, TargetMol, T14834). Oral administration of nintedanib (60 mg/kg/day, MCE, HY-50904) served as a positive control. Lungs were collected 21 days after BLM exposure. To further validate the therapeutic efficacy of BT2, an additional cohort of mice received a higher BLM dose (3.5 mg/kg, intratracheal), followed by daily intraperitoneal BT2 (20 mg/kg) starting on day 3 post-injury. Lung tissues were collected on day 14. To further determine the therapeutic effect of BCAA-restricted diet on IPF, a repetitive, non-remodeling BLM-induced fibrosis model was established³⁸. Mice received BLM (1.0 mg/kg, intratracheal) once weekly for 4 weeks, followed by a one-week rest period. Subsequently, the mice were fed either a standard control diet or a BCAA-free diet for 14 days before lung tissue collection

Human studies

All experiments with human tissue and blood samples were performed under protocols approved by the Institutional Review Boards at Guangzhou National Laboratory and Ethics Committee of Guangzhou Medical University. IPF lung tissues were collected during lung transplantation procedures. Control lung tissues were sourced from two groups: (1) histologically normal regions of lungs resected from lung cancer patients, and (2) donor lungs rejected for transplantation due to non-IPF factors. IPF diagnoses were rigorously confirmed via multidisciplinary evaluation, integrating clinical examination and high-resolution computed tomography (HRCT) imaging, in adherence to the 2018 ATS/ERS international diagnostic guidelines. Informed written consent was obtained from all participants. All human studies were performed in accordance with relevant ethical regulations. Human IPF tissue and plasma samples were obtained from the First Affiliated Hospital of Guangzhou Medical University. Tissue samples were destroyed during analysis. Blood samples are stored at Guangzhou National Laboratory, but cannot be shared due to regulatory and IRB restrictions, however, collaboration requests can be directed to hu_wenxiang@gzlab.ac.cn.

Metabolomics profiling by LC–MS

Sample preparation

Cellular and tissue samples were processed using optimized extraction protocols: cell lysates were treated with pre-cooled (−20 °C) acetonitrile/methanol/water (4:4:2 v/v) containing 2 mM alkaline buffer salt, sonicated (4 °C, 10 min, 30 s ON/OFF cycles), and centrifuged (13,000 × g, 4 °C, 10 min). The supernatant was transferred into a new polypropylene vial and run with 3 μL injection volume by LC–MS/MS system. 15 μL of each sample was taken to mix into quality control (QC). Tissue homogenates from 20 mg mouse lung (homogenized with three 2 mm steel balls in 200 μL ddH₂O at 4 °C, 60 Hz, 30 s ON, 15 s OFF using KZ-III-FP grinder) underwent methanol/acetonitrile (1:1 v/v, 800 μL, 4 °C, 10 min, 30 s ON/OFF cycles) extraction, protein precipitation (−20 °C, 1.5 h), centrifugal separation (13,000 × g, 4 °C, 15 min) for supernatant collection and centrifugal evaporation (CV600 concentrator) prior to reconstitution in 50 μL acetonitrile/water (1:1 v/v). The reconstitution solution was vortexed for 30 s and sonicated 10 min at 4 °C. After being centrifuged at 13,000 × g for 15 min at 4 °C, the supernatant was transferred into a new polypropylene vial and stored at 4 °C until being tested. Quality control (QC) samples were prepared by mixing 4 μL of each sample.

Data acquisition and processing

A 3 μL injection volume was used for all analyses. Chromatographic separation was performed using an Agilent 1290II ultra-high-pressure liquid chromatography (UHPLC) system equipped with 6546 quadrupole time-of-flight (QTOF) mass spectrometry. A Waters ACQUITY UPLC BEH Amide column (2.1 × 100 mm × 1.7 μm) and guard column (2.1 × 5 mm × 1.7 μm) at 35 °C was used to separate metabolites with mobile phase A: 100% aqueous containing 15 mM ammonium acetate and 0.3% ammonium hydroxy and mobile phase B: 90% acetonitrile (v/v) aqueous containing 15 mM ammonium acetate and 0.3% ammonium hydroxy. The linear gradient was set as follows: 10% A (0.0–8.0 min), 50% A (8.0–10.0 min), 50% A (10.0–11.0 min), and 10% A (11.0–20.0 min). The total run time was 20 min and flow rate was 0.3 mL/min. The mass spectrometer was equipped with Agilent Jet-stream source operating in negative and positive ion mode with source parameters set as follow: Nebulizer gas, 45 psi; Sheath gas temperature, 325 °C; Sheath gas flow, 10 L/min; Dry gas temperature, 280 °C; Dry gas flow, 8 L/min; Capillary voltage, 3500 v for two ion modes and nozzle voltage, 500 v for positive and 1000 v for negative mode. The QTOF scan parameters were set as follows: Scan speed, 1.5 scan/s; scan range, 50–1700 m/z and ion fragmentor voltage, 140 v. The acquired data quality was monitored by Amino Acid standards (Merck, Sigma-Aldrich Production GmbH, Switzerland), mixed-samples QC and blanks. Peak integration and metabolite identification were accomplished using Profinder 10.0 (Agilent). Statistical analysis and graphing were fulfilled using Mass Profiler Professional 15.1(Agilent).

Metabolites flux analysis

Metabolites flux analysis was conducted according to previously published protocols^74,75. One million MLFs were pre-cultured in BCAA-free DMEM/F12 medium supplemented with 0.80 mM ¹³C_6-Leucine (Cambridge lsotope Laboratories, Tewksbury, MA) for 24 h and treated with TGF-β1 (20 ng/mL) for an additional 48 h. Cells were then rapidly collected for metabolites examination. Untargeted metabolic flux analysis was performed at LipidALL Technologies. Polar metabolites were extracted with ice-cold methanol supplemented with phenylhydrazine, followed by vortex-mixing (30 min, 4 °C) and subsequent α-keto acid derivatization (−20 °C, 1 h). After phase separation by centrifugation (12,000 × g, 15 min, 4 °C), the supernatant was lyophilized using a SpeedVac concentrator (H₂O mode). Residual protein content in the pellet was quantified via Pierce® BCA Protein Assay Kit per manufacturer specifications. For LC–MS analysis (Agilent 1290 II UPLC-Sciex 5600 + Q-TOF), reconstituted extracts (5% acetonitrile) were chromatographed on a Waters ACQUITY HSS-T3 column (3.0 × 100 mm, 1.8 μm) with the following MS parameters: ESI voltage −4.5 kV, vaporizer 500 °C, N2 pressures (drying/nebulizer/curtain) 50/50/35 psi. Full-scan spectra (m/z 60–700) were acquired in information-dependent acquisition mode with collision energy (−)35 ± 15 eV. Data processing utilized Analyst® TF 1.7.1 for acquisition, with MarkerView 1.3 generating m/z-retention time matrices. Metabolite annotation combined PeakView 2.2 spectral matching against SCIEX Metabolites Database, HMDB, and authentic standards. L-Leucine-d10 served as internal standard for inter-sample peak area normalization.

RNA extraction and qRT-PCR

Total RNA was extracted from cultured cells using the EASYspin RNA Mini Kit (Aidlab, RN07) and from mouse lung tissues using TRIzol reagent (Vazyme, R401), following the manufacturers’ protocols. 1 μg of total RNA was reverse transcribed using the HiScript III 1st Strand cDNA Synthesis Kit (Vazyme, R211-02). Quantitative real-time PCR (qRT-PCR) was performed using ChamQ SYBR qPCR Master Mix (Vazyme, Q711-03) on Applied Biosystems QuantStudio 5, with gene-specific primers. The relative expression levels were normalized against the internal control. Primers used were listed in Supplementary Table 2.

RNA-seq and data processing

Total RNA samples were prepared with EASYspin RNA Mini Kit or TRIzol reagent according to manufacturer’s instructions. 2 μg RNA from biological replicates were sent to Anoroad for library preparation (Illumina) and sequencing on the Illumina NovaSeq X Plus platform (150 bp paired-end reads). Raw sequencing data were subjected to quality control using FastQC. Adapter sequences and low-quality bases were trimmed using Trim Galore. Clean reads were aligned to the mouse reference genome (GRCm39/mm39) using HISAT2 with default parameters. Gene-level quantification was performed using featureCounts from the Subread package. Differential gene expression analysis was conducted using DESeq2. Significantly differentially expressed genes (DEGs) were subjected to Gene Ontology (GO) and KEGG pathway enrichment analysis using the clusterProfiler R package. For visualization, TPM was scaled and plotted as heatmaps.

Immunoblot analysis

Whole-cell lysates were prepared in SDS loading buffer (0.5 M Tris-HCl [pH 6.8], 2% SDS, 10% glycerol, 0.1% bromophenol blue) and denatured at 95 °C for 15 min. Proteins were resolved by 10% SDS-PAGE (120 V, 90 min) and electrophoretically transferred onto PVDF membranes (100 V, 90 min). Membranes were blocked with 5% non-fat dried milk in TBST (Tris-buffered saline with 0.1% Tween-20) for 1 h at room temperature (RT), then probed with primary antibodies overnight at 4 °C. After three TBST washes (10 min each), membranes were incubated with HRP-conjugated secondary antibodies for 1 h at RT, followed by three TBST washes. Protein signals were detected using the Pierce ECL chemiluminescent detection system (Thermo Fisher Scientific, Waltham, MA, USA) and imaged with a ChemiDoc MP imaging system (Bio-Rad Laboratories, Hercules, CA, USA). Densitometric quantification was performed using ImageJ software. The primary antibody information is as follows: rabbit anti-SLC7A5 (1:1000, Proteintech, Cat# 28670-1-AP), rabbit anti-BCAT1 (1:1000, ABclonal, Cat# A16351), rabbit anti-BCAT2 (1:1000, ABclonal, Cat# A7426), rabbit anti-BCKDHA (1:1000, ABclonal, Cat# A21588), rabbit anti-BCKDHB (1:1000, ABclonal, Cat# A23916), rabbit anti-DBT (1:1000, ABclonal, Cat# A20381), rabbit anti-ACADS (1:1000, ABclonal, Cat# A0945), rabbit anti-FN1 (1:1000, ABclonal, Cat# A16678), rabbit anti-COL1A1 (1:1000, Cell Signaling Technology, Cat# 72026S), mouse anti-ACTA2 (1:10,000, Abcam, Cat# ab7817), rabbit anti-COL1A2 (1:1000, ABclonal, Cat# A21059), rabbit anti-Flag-tag (1:1000, Cell Signaling Technology, Cat# 14793S), rabbit anti-KDM4A (1:1000, ABclonal, Cat# A9267), rabbit anti-H3K36me3 (1:2000, Abcam, Cat# ab9050), rabbit anti-H3 (1:10,000, ABclonal, Cat# A2348), rabbit anti-ATF4 (1:1000, Proteintech, Cat# 10835-1-AP), rabbit anti-Phospho-SMAD2 (Ser465/467) (1:1000, Cell Signaling Technology, Cat# 3108), rabbit anti-Phospho-SMAD3 (Ser423/425) (1:1000, Cell Signaling Technology, Cat# 9520), rabbit anti-mTOR ((1:1000, Cell Signaling Technology, Cat# 2983), rabbit anti-Phospho-mTOR (Ser2448) (1:1000, Cell Signaling Technology, Cat# 5536), rabbit anti-Phospho-AKT1 (Ser473) (1:1000, ABclonal, Cat# AP0098), rabbit anti-HIF-1α (1:1000, abclonal, Cat# A26889), rabbit anti-Phospho-Erk1/2 (1:1000, Thr202/Tyr204) (Sino Biological, Cat# 110441-R0072), Phospho-GSK-3β (Ser9) (1:1000, Sino Biological, Cat# 110455-R0016), HRP-conjugated GAPDH (1:10,000, Proteintech, Cat# HRP-60004). The secondary antibody information is as follows: Anti-rabbit IgG, HRP-linked Antibody (1:5000, Cell Signaling Technology, Cat# 7074); Anti-mouse IgG, HRP-linked Antibody (1:5000, Cell Signaling Technology, Cat# 7076). All unprocessed scans of the blots are provided in the Source Data or Supplementary Information.

Measurement of BCAA content

Cell culture supernatants and mouse blood

BCAA concentrations in cell culture supernatants and mouse blood were measured using a commercial Branched Chain Amino Acid (BCAA) Assay Kit (Abcam, ab83374), following the manufacturer’s instructions. Values were normalized to the input cell number to account for variations in cell density.

Cell, plasma, and lung tissue samples

BCAA levels in cells, mouse plasma (Figs. 2m and 5b) and human plasma, as well as mouse lung tissues, were measured by enzymatic assay or LC–MS as described below.

Sample preparation

Serum: extraction solution (Acetonitrile/Methanol/H₂O = 4:4:2) was prepared and placed in −20 refrigerator overnight to pre-cool. 10 μL mouse serum and 10 μL extraction solution were added into 200 μL centrifuge tube and centrifuged using a handheld centrifuge after vortexing 2 s. Then sample was incubated at −4 °C for 0.5 h to precipitate the protein and then centrifuged at 16,000 × g for 10 min at 4 °C. The supernatant was transferred to a new polypropylene vial and run with 1 μL injection volume by LC–MS/MS system.

Tissue: The extraction steps are the same as metabolite measurement by LC–MS.

Data acquisition and processing

A 1 μL injection volume was used for all analyses. Chromatographic separation was performed using an Agilent 1290II ultra-high-pressure liquid chromatography (UHPLC) system equipped with 6495B triple quadrupole (QQQ) mass spectrometry. A Waters ACQUITY UPLC BEH Amide column (2.1 × 100 mm × 1.7 μm) and guard column (2.1 × 5 mm × 1.7 μm) at 40 °C was used to separate metabolites with mobile phase A: 100% aqueous containing 15 mM ammonium acetate and 0.3% ammonium hydroxy and mobile phase B: 90% acetonitrile (v/v) aqueous containing 15 mM ammonium acetate and 0.3% ammonium hydroxy. The linear gradient was set as follows: 5% A (0.0–5.0 min), 50% A (5.0–7.0 min), 95% A (7.0–17.0 min). The total run time was 17 min and flow rate were 0.2 mL/min. 0.0–0.5 min and 6.4–17.0 min were switched to waste and 0.5–6.4 min was into MS. Delta EMV in positive mode was 400. The mass spectrometer was equipped with Agilent Jet-stream source with source parameters set as follow: Nebulizer gas, 45 psi; Sheath gas temperature, 325 °C; Sheath gas flow, 10 L/min; Dry gas temperature, 280 °C; Dry gas flow, 8 L/min; Capillary voltage, 3500 v for two ion modes and nozzle voltage, 500 v for positive and 1000 v for negative mode. Leucine, isoleucine, and valine were quantified by multiple reaction monitoring (MRM) in positive ESI mode. Calibrations of leucine, isoleucine, and valine with certain concentration gradients (0.01, 0.025, 0.05, 0.1, 0.5, 1, 5, 10, 25, 50, and 100 mg/L) were prepared with ddH2O and detected with UHPLC -QQQ. The MRM transition for leucine and isoleucine (m/z 132.1 > 86.1, 132.1 > 44.1) and valine (m/z 118.1 > 72.1, 118.1 > 55.1) were monitored. MS peak data from UHPLC -QQQ analyses were subjected to Agilent MassHunter Quantitative Analysis 10.0 for peak detection and integration. Linear regression equations were derived from standard calibration curves. The concentration of leucine, isoleucine, and valine in mouse serum and lung tissue was calculated according to the standard calibration curves. BCAA measurements in lung tissues were normalized to lung weight.

Immunofluorescence staining

Immunofluorescence staining was conducted according to a previously published protocol⁷⁶. Cells were seeded onto eight-chamber glass-bottom dishes (Cellvis, C8-1.5H-N) and fixed in 4% paraformaldehyde (Biosharp, BL539A) for 10 min at room temperature following two washes with PBS. After three additional PBS washes, cells were permeabilized for 10 min using PBS containing 0.1% Triton ×-100 (Beyotime, ST1723) and 1% bovine serum albumin (BSA) (BBI, A600332-0100). Following four PBS washes, cells were incubated with primary antibodies overnight at 4 °C. The next day, cells were washed and incubated with fluorescent secondary antibodies for 1 h at room temperature. Nuclei were counterstained with DAPI (0.5 μg/mL; Beyotime, P0131) for 10 min. After final washes, cells were visualized using NIKON A1. Fluorescence intensity was quantified using ImageJ across randomly selected fields (n = 3) from three independent biological replicates. The following antibodies were used: rabbit anti-COL1A1 (1:100, Cell Signaling Technology, Cat# 72026S), mouse anti-ACTA2 (1:200, Abcam, Cat# ab7817), mouse anti-Flag (1:100, proteintech, Cat# 66008-4-Ig), rabbit anti-H3K36me3 (1:100, Abcam, Cat# ab9050), Donkey anti-Mouse IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor™ 488 (1:100, Thermo Fisher Scientific, Cat# A21202), Donkey anti-Rabbit IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor™ 594 (1:100, Thermo Fisher Scientific, Cat# A21207).

Micro-CT analysis

Mice were anesthetized and positioned in the scanning chamber of the Quantum GX2 microCT Imaging System (PerkinElmer). High-resolution scans of the thoracic cavity were acquired following the manufacturer’s protocols to visualize lung architecture. A validated semi-quantitative fibrosis scoring system (0–6) was applied to assess the severity of pulmonary fibrosis based on micro-CT images⁷⁷. In this system, a score of 0 corresponds to normal lung parenchyma with no detectable fibrosis, while a score of 6 denotes severe fibrotic remodeling involving the majority of lung fields.

Tissue histopathology analysis

Left lung lobes were fixed in 4% neutral-buffered formalin for 24 h at room temperature, followed by dehydration, paraffin embedding, and sectioning at 5 μm thickness. For histopathological analysis, paraffin-embedded lung sections were subjected to hematoxylin and eosin (H&E) and Masson’s trichrome staining using standard protocols to assess tissue architecture and collagen deposition, respectively. H&E and Masson’s staining images were captured by Olympus SLIDEVIEW VS200 digital slide scanner. The quantification analysis of H&E and Masson’s staining were performed using ImageJ software.

Hydroxyproline content

Hydroxyproline levels in the upper lobe of the right lungs were quantified to assess collagen deposition. Lung tissues were subjected to alkaline hydrolysis, and hydroxyproline content was measured using a commercial colorimetric assay kit (Solarbio, BC0255), following the manufacturer’s instructions. Results were normalized to tissue weight.

Cell viability assay

MLFs were seeded in 96-well plates at approximately 5000 cells per well and pre-cultured in three types of media: normal medium (Complete), BCAA-free medium (Free), and BCAA-free medium added with BCAA (Free + BCAA). After incubation for 24 h, TGF-β1 (20 ng/mL) was added to the cells for an additional 48 h, and then, 10% CCK-8 reagent (APExBIO, K1018) was added for 2–3 h at 37 °C. Absorbance was measured at 450 nm by SpectraMax Paradigm.

Scratch-wound assay

For the scratch-wound assay, culture inserts (Ibidi, 80209) were placed in 24-well plates. 1 × 10⁶ cells/mL suspensions in three types of media: normal medium (Complete), BCAA-free medium (Free), and BCAA-free medium were prepared. 70 μL suspensions were placed in culture inserts for 24 h, culture inserts were then gently removed to expose the defined cell-free gap. Cells were washed with PBS three times to remove cell debris and cultured with corresponding medium without FBS. After 48 h incubation at 37 °C, the migration distance of the cells was observed under Olympus IX73 inverted microscope. The areas between the edges of the wounds were measured and analyzed using ImageJ software.

scRNA-seq and data processing

DNBelab C Series High-throughput Single-Cell RNA Library (940-001818-00) was utilized for scRNA-seq library preparation. In brief, Briefly, the Single-cell RNA libraries were prepared using the DNBelab C Series kit (940-001818-00). Cells were diluted to 1000 cells/mL and loaded into a microfluidic chip alongside barcoded beads and droplet oil. Droplets were generated with a DNBelab C4/TaiM4 system, followed by mRNA capture, reverse transcription (RT), and PCR amplification. cDNA was purified (Qubit dsDNA kit) and processed into 3’-end transcript libraries via fragmentation, size selection, end repair, adapter ligation, and indexing. Final libraries were quantified (Qubit ssDNA kit, Qsep100) and sequenced on the DNBelab C4 /DNBelab TaiM4 Series Single-Cell Library Prep Set (MGI).

The raw sequencing reads were aligned to the Mus musculus reference genome (GRCm39 assembly) using dnbc4tools. Following alignment, the mapped reads were processed using Seurat (v4.2.0)⁷⁸. Individual Seurat objects were created for each sample, followed by doublet removal using DoubletFinder (v2.0.4)⁷⁹ and ambient RNA contamination filtering with decontX from celda package (v1.10.0), applying a stringent contamination threshold of <0.3 for each sample. Only high-quality cells meeting stringent QC thresholds were retained for downstream analysis, including: (1) nUMI ≥ 500; (2) nGene ≥ 200; (3) log10GenesPerUMI > 0.80; (4) mitoRatio <0.50. We also filtered genes with low counts or expressed in less than 10 cells to improve the reliability of variance analysis. After quality control, all samples were integrated by CCA-based integration in Seurat to correct batch effect. Lowly expressed genes (detected in fewer than 10 cells) were filtered out to ensure robust differential expression analysis. Following quality control, we performed canonical correlation analysis (CCA)-based batch correction by Seurat to mitigate technical variability across samples. In detail, SelectIntegrationFeatures select the most 3000 variable features, PrepSCTIntegration prepare the SCT list, FindIntegrationAnchors find the best anchors between samples with normalization.method = “SCT” and IntegrateData integrate all conditions. After integration of samples, we used RunPCA and RunUMAP to perform PCA analysis and reduction using dims = 1:50. Then, determination of k-nearest neighbor and cell clusters were implemented by FindNeighbors and FindClusters under resolution 0.2. Celltypist were used to define cell types⁸⁰.

Fibrosis and injury score analysis

We utilized the AddModuleScore function from Seurat to calculate the fibrotic activity levels and injury scores across cell populations under distinct samples, based on a set of 20 fibrosis genes and GO term: 0050679 “positive regulation of epithelial cell proliferation” in mouse⁸¹.

Gene clusters in fibroblast cells

To delineate alterations in gene expression within fibroblast cells between different conditions, we categorized the marker genes of fibroblast into several clusters. In detail, we firstly acquired the marker genes of fibroblast by FindMarkers with ident.1 set to “Fibroblast”, and only genes with avg_log2FC > 0.5 and p_val_adj < 0.01 were marker genes. Then, we calculated the gene level under each condition by AverageExpression and classified the genes to 6 clusters according to gene expression level by Heatmap from ComplexHeatmap v 2.14.0⁸².

CUT&Tag and data analysis

CUT&Tag assays were performed using the CUT&Tag High-Sensitivity Kit (Novoprotein, N259-YH01) following manufacturer’s protocols. The following antibodies were used: rabbit anti-ATF4 (1:50, Proteintech, Cat# 10835-1-AP), rabbit anti-PPARγ (1:50, Abcam, Cat# ab45036), rabbit anti-H3K36me3 (1:100, Abcam, Cat# ab9050), rabbit anti-H3K9me3 (1:100, Abcam, Cat# ab8898), rabbit anti-H3K27me3 (1:50, Cell Signaling Technology, Cat# 9733 T), rabbit anti-H3K4me3 (1:100, Millipore, Cat# 04-745), rabbit anti-H3K27ac (1:100, Active Motif, Cat# 39135), Goat Anti-Rabbit IgG H&L (1:200, Novoprotein, Cat# N269). In MLFs, ATF4 and PPARγ chromatin occupancy was profiled under basal conditions or TGF-β1 treatment conditions. Histone modification landscapes (H3K4me3, H3K27ac, H3K27me3, H3K36me3, H3K9me3) were mapped in MLFs cultured in complete medium or BCAA-free medium prior to TGF-β1 treatment. The sequencing libraries were sequenced on the Illumina NovaSeq X Plus platform (150 bp paired-end reads). We performed quality control using FastQC on all samples. The raw reads were trimmed using Trim-galore to remove adapters. The clean reads were mapped to mouse reference genome (mm39) using Bowtie2 with the “–end-to-end –very-sensitive -I 10 -X 700 –no-mixed” option. Low-quality mapped reads and those mapped to mitochondrial reads were filtered using Samtools. The duplicated reads were removed using Sambamba. The Counts Per Million mapped reads (CPM) normalization method was used to transform alignment BAM files into read coverage files in bigWig format on deepTools. Peak calling was performed using MACS2. For H3K4me3 and H3K27ac, the narrow peak model was applied with default parameters. For H3K27me3, H3K36me3, and H3K9me3, the broad peak model was used with the “-p 0.01 –broad-cutoff 0.01” parameters. Differentially enrichment peaks were identified using the DiffBind with a p-value < 0.05 and absolute log2(fold change)> 1. The nearest genes around the peak were annotated using ChipSeeker. Motif enrichment analysis was performed using HOMER.

ATAC-seq and data analysis

MLFs were cultured in three media conditions: complete medium (Complete), BCAA-free medium (Free), or BCAA-free medium supplemented with BCAAs (Free + BCAA) for 24 h prior to TGF-β1 (20 ng/mL) stimulation for 48 h. ATAC-seq analyses were conducted on treated MLFs by using the ATAC-Seq Library Prep Kit (Vazyme, TD711-02) following manufacturer’s protocols. The sequencing libraries were sequenced on the Illumina NovaSeq X Plus platform (150 bp paired-end reads). We performed quality control using FastQC on all samples. The raw reads were trimmed using Trim-Galore to remove adapters. The clean reads were mapped to mouse reference genome (mm39) using Bowtie2 with the “–very-sensitive -X 1000” option. Low-quality mapped reads and those mapped to mitochondrial reads were filtered using Samtools. The duplicated reads were removed using Sambamba. The Counts Per Million mapped reads (CPM) normalization method was used to transform alignment BAM files into read coverage files in bigWig format on deepTools. Peak calling was performed using MACS2 with the parameter “–SPMR –nomodel –shift −100 –extsize 200”. Differentially accessible peaks were identified using the DiffBind with an adjusted p-value < 0.05 and absolute log2(fold change) > 0.585. The genes around the peak 20 kb range were annotated using ChipSeeker. Motif enrichment analysis was performed using HOMER.

RNA interference and gene transfection

siRNA and cDNA transfection

For gene knockdown or overexpression in MRC-5 cells, Lipofectamine 3000 (Thermo Fisher Scientific, L3000-015) was used according to the manufacturer’s protocol. For siRNA transfection in MLFs, INTERFERin reagent (Polyplus, 409-01) was used following the manufacturer’s instructions. SiRNAs against human BCAT1 (Gene ID: 586), BCAT2 (Gene ID: 587), SLC7A5 (Gene ID: 8140), KDM4A (Gene ID: 9682), and mouse Kdm4a (Gene ID: 230674) were synthesized by Sangon Biotech. The sequences were listed in Supplementary Table 2.

Lentiviral transduction

For some sgRNA- or cDNA-based gene manipulation in MLFs, lentiviral expression constructs encoding Flag-tagged BCAT1 or Flag-tagged KDM4A were generated by cloning the full-length mouse Bcat1 (Gene ID: 12035) or Kdm4a (Gene ID: 230674) cDNA into the pLVX-3Flag-MCS-IRES-Puro vector. For ATF4 stable knockdown cell lines, the gRNA targeting mouse Atf4 (Gene ID: 11911) were designed by website (http://crispr.mit.edu). The sequences were listed in Supplementary Table 2. Lentiviral particles were generated in HEK293T cells. Plasmids encoding sgRNA or cDNA constructs were co-transfected with psPAX2 and pMD2.G packaging vectors using PEG300 (Selleck, S6704). After 48 h, the viral supernatant was collected, filtered through a 0.45 μm filter, and mixed with 8 μg/mL polybrene (Sigma-Aldrich, H9268) to enhance transduction efficiency. 48 h post infection, transduced MLFs were selected with 5 μg/mL puromycin (Sigma-Aldrich, P9620) for 3 days to ensure stable gene knockdown or overexpression.

Quantification and statistical analysis

GraphPad Prism 9.5.0 software was used for statistical analyses. Data are presented as mean ± SEM for in vivo experiments and mean ± SD for in vitro assays. Statistical significance was assessed using Student’s t-tests or ANOVA, where appropriate.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

W.H. conceptualized the study, interpreted data, and wrote the manuscript, which was revised and approved by all authors. J.Y. and S.F. performed most experiments. M.L. and N.H. performed bioinformatics analysis. W.P., Z.O., C.H.W., C.Z., S.W., J.L., and W.Z. help with animal experiments. Z.O., L.Y., and M.F. performed in vitro experiments. B.G., Y.P., W.R., P.Y., H.H., and Q.H. collected patient samples and clinical data. C.Q.W. and Z.R.interpreted data and revised the manuscript.

Peer review

Peer review information

Nature Communications thanks Ross S Summer, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The RNA-seq data generated in this study have been deposited in the Gene Expression Omnibus (GEO) under accession number GSE295426. The ATF4 and PPARγ CUT&Tag datasets are available under accession number GSE295424. The H3K4me3, H3K27ac, H3K27me3, H3K36me3, and H3K9me3 CUT&Tag datasets are available under accession number GSE295425 and GSE325209. The ATAC-seq datasets are available under accession number GSE295423. The scRNA-seq datasets are available under accession number GSE295427. Metabolomic data generated in this study have been deposited in the OMIX database and are available at https://ngdc.cncb.ac.cn/omix/release/OMIX009983. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-72273-3.