SOX4-STAT6-MTHFD2 axis drives hepatocellular carcinoma progression and treatment resistance.

Introduction
Liver cancer is the sixth diagnosed malignancy and ranks as the third leading cause of cancer-related deaths globally [1]. Hepatocellular carcinoma (HCC) comprises approximately 90% of primary liver cancers and remains a significant global public health concern. Curative options for early-stage HCC include surgery, local ablation, or transplantation, but recurrence leads to poor 5-year survival rates [2, 3]. Advanced HCC is managed with embolization and targeted therapies like immune checkpoint inhibitors (ICIs), tyrosine kinase inhibitors (TKIs), or anti-angiogenic agents [2, 4, 5]. Despite therapeutic advancements, disease progression remains a concern, highlighting the need to identify critical targets to improve treatment efficacy.
SOX4, a member of the SOX family, contains the SRY-related high-mobility-group box, which binds to [(A/T)(T/A)CAA(A/T)G] DNA consensus sequences in promoters to regulate genes involved in neural development, sex determination, differentiation, epithelial–mesenchymal transition (EMT), and angiogenesis in cancers [6]. Its overexpression correlates with poor prognosis in multiple cancers, including HCC [7–16]. Recent studies suggest SOX4 as a potential biomarker for poor immunotherapy outcomes [17, 18]. Despite its role in tumor progression, SOX4 remains undruggable. Functionally, SOX4 acts as a hub to interact with proteins like EZH2, P53, Smad3, β-catenin, and TCF/LEF in response to various signaling [6]. Notably, the SOX4–EZH2 axis has been shown to drive metastasis via epigenetic reprogramming, including modulation of microRNAs and transcriptional/metabolic changes [8, 10]. In addition, SOX4 has also been reported to be associated with SMARCA4- the catalytic ATPase subunit of the SWI/SNF chromatin remodeling complex, to regulate transcriptional targets such as TGFBR2 [19]. Collectively, these findings highlight a critical role for SOX4 in epigenetic regulation. Continued investigation into its regulation and interactome may uncover its role in carcinogenesis and therapeutic potential.
STAT6 is essential for M2 macrophage polarization, driven by IL-4/IL-13, which fosters tumor progression by creating an immunosuppressive microenvironment [20]. In the canonical IL-4/IL-13 pathway, these cytokines bind to interleukin-4 receptor alpha, activating Janus kinase 1, which phosphorylates STAT6 at tyrosine residue 641 (p-STAT6 Y641), prompting its dimerization and nuclear translocation to regulate gene expression [21]. Beyond phosphorylation, the STAT6 promoter is modulated by epigenetic modifications [22–24]. Recent therapeutic strategies target STAT6 in immune disorders and cancers [21, 25–27]. Besides, STAT6 overexpressed in various malignancies, including breast, pancreatic, prostate, and colorectal cancers, promotes macrophage activation and tumor metastasis [28–33]. Knockdown of STAT6 reduced HepG2 and Hep3B cell proliferation by suppressing nuclear factor kappa-B ligand (RANKL) expression [34]. However, its role in HCC carcinogenesis remains unclear due to limited studies.
Methylenetetrahydrofolate dehydrogenase 2 (MTHFD2) is a mitochondrial enzyme in one-carbon folate pathway, catalyzing the NAD + /NADP + - dependent conversion of 5,10-methylenetetrahydrofolate (CH2-THF) to 10-formyltetrahydrofolate (10CHO-THF) [35]. This reaction supports redox balance via the glutathione system and provides formate for purine biosynthesis [35]. CH2-THF also fuels thymidylate synthase (TYMS) in converting deoxyuridine monophosphate (dUMP) to deoxythymidine monophosphate (dTMP), essential for pyrimidine synthesis [35]. Besides, the interconnection between the folate pathway and the methionine cycle—where S-adenosylmethionine (SAM), a critical intermediate, serves as a universal methyl donor for DNA, RNA, and protein methylation [35]. Additionally, MTHFD2 promotes hexosamine biosynthesis pathway (HBP) to activate UDP-N-acetylglucosamine (UDP-GlcNAc) production, stabilizing c-Myc protein and enhancing PD-L1 expression, facilitating immune evasion in pancreatic cancer [36]. Besides, given its role in nucleotide metabolism, MTHFD2 drives cancer cell growth, stemness, drug resistance, and immune evasion, making it a promising therapeutic target for various kinds of cancers [37–39].
In our previous study, two SOX4 knockout cell lines (Hep3B SOX4−/− #1 and #2) were previously generated using the clustered regularly interspaced short palindromic repeats and Cas9 (CRISPR/Cas9) system to investigate SOX4-regulated cellular pathways [11]. In this study, STAT6 was identified as a downstream regulatory protein of SOX4 in HCC cells through liquid chromatography–mass spectrometry (LC-MS) analysis. Therefore, we aimed to evaluate how SOX4 regulated STAT6, as well as their co-regulated genes and pathways, including the folate metabolism pathway and its key enzyme, MTHFD2, in HCC cells. The impact of the SOX4/STAT6/MTHFD2 axis on cellular metabolites, and prognosis of patients with HCC were also discussed in this study.

Results

SOX4 epigenetically regulated STAT6 expression in HCC cells
To explore SOX4-regulated cellular pathways in HCC, its endogenous expression was assessed by western blotting across various liver cancer cell lines. SOX4 was highly expressed in Hep3B cells, moderately in Huh7 and SNU475, weakly in PLC5 and SNU398, and undetectable in HepG2 and two immortalized hepatocyte lines (Supplementary. Fig. 1). Based on this, two SOX4-knockout Hep3B cell lines (Hep3B SOX4⁻/⁻ #1 and #2), generated via CRISPR/Cas9 previously [11], were used to analyze SOX4-dependent protein changes using LC-MS. Among the differentially expressed proteins, STAT6 showed the most significant downregulation in SOX4 knockout cells (Fig. 1A). Results of western blotting and quantitative reverse transcription- real-time PCR (qRT-PCR) confirmed that repression of SOX4 reduced STAT6 at both protein (Fig. 1B, lane 2, 3 and lane 5) and mRNA levels in HCC cells (Fig. 1C), suggesting transcriptional regulation. Data from The Cancer Genome Atlas-Liver Hepatocellular Carcinoma Collection (TCGA-LIHC) supported a positive correlation between SOX4 and STAT6 transcripts (Fig. 1D). Although no SOX4 consensus binding sites were found in the STAT6 promoter, multiple CpG sites were located from the 5’-UTR to the first exon of STAT6 gene (Fig. 1E), suggesting potential epigenetic regulation. Given previous evidences implicating SOX4 in epigenetic regulation and reports of STAT6 promoter methylation [8, 10, 19, 22, 23], we next analyzed DNA methylation profiles in wild-type and SOX4-knockout Hep3B cells using bisulfite DNA sequencing. Bisulfite DNA sequencing showed increased methylation in the STAT6 5′-UTR to exon 1 in SOX4−/− cells compared to wild-type (Fig. 1E). Treatment with the DNA methylation inhibitor 5-Aza-2′-deoxycytidine (5’-AZA) restored STAT6 mRNA (Fig. 1F) and protein levels (Fig. 1G, H) in SOX4-knockout/or knockdown cells, suggesting that SOX4 maintains STAT6 expression by preventing promoter methylation. These results indicate an epigenetic mechanism by which SOX4 regulates STAT6 in HCC cells.

SOX4 complexed with STAT6 protein to modulate IL-4-mediated phosphorylation of STAT6 in HCC cells
Since both SOX4 and STAT6 are transcription factors, their interaction was also examined via co-immunoprecipitation in HCC cells. SOX4 formed a complex with both total STAT6 and p-STAT6 Y641 (Fig. 2A), while IgG served as a negative control. Mapping of interaction domains using deletion constructs in HEK293 cells revealed that removal of the STAT6 DNA-binding domain (DBD) abolished its binding to SOX4 (Fig. 2B, lane 4), and deletion of the SOX4 serine-rich region (SRR) repressed its interaction with STAT6 (Fig. 2C, lanes 4–5). Confocal microscopy further confirmed nuclear colocalization of SOX4 and STAT6 in Hep3B cells (Fig. 2D) and HCC tumoroids (Fig. 2E).
To investigate the functional of the SOX4–STAT6 complex, Hep3B, Hep3B SOX4−/−, and SNU475 cells with scramble or SOX4 si transfection were treated with or without interleukin-4 (IL-4). IL-4 induced p-STAT6 Y641 in both HCC cell lines (Fig. 2F, lanes 2, 6), but this induction was reduced in SOX4-deficient cells (Fig. 2F, lanes 4, 8). The total STAT6 and p-STAT6/STAT6 ratio was reduced in Hep3B SOX4−/− cells or SOX4 si-transfected SNU-475 cells, relative to their corresponding controls (Hep3B and scramble-transfected SNU-475, respectively. (Fig. 2F right panel). Furthermore, IL-4-induced expression of RANKL, a downstream STAT6 target, was diminished in SOX4-knockout Hep3B cells, as shown by qRT-PCR (Fig. 2G). These results indicate that SOX4 could modulate IL-4–mediated phosphorylation of STAT6 Y641 and its downstream expression of RANKL in Hep3B cells. Together, these results demonstrate that SOX4 physically interacts with STAT6 and colocalizes with STAT6 in the nucleus of HCC cells and tumoroids. Functionally, SOX4 is required for efficient IL-4–mediated phosphorylation of STAT6 at Y641 and for the subsequent induction of STAT6 target genes such as RANKL in HCC.

SOX4 and STAT6 co-Regulate folate pathway in HCC Cells
To identify genes co-regulated by SOX4 and STAT6, chromatin immunoprecipitation (ChIP) assays were performed in Hep3B cells using SOX4 and STAT6 antibodies, with IgG as a negative control. Both SOX4 and STAT6 antibodies showed enriched binding signals near peak centers compared to IgG (Fig. 3A), with 5.5% and 7.6% of peaks located in promoter regions; respectively (Fig. 3B). The identified consensus binding motifs matched known SOX4 and STAT6 sequences (Fig. 3C). Integration of ChIP-seq and RNA-seq data from Hep3B vs. SOX4−/− cells and Hep3B scramble vs. STAT6 si knockdown cells identified 666 SOX4-regulated and 95 STAT6-regulated genes (Supplementary Fig. 2), with 86 overlapping targets (Supplementary Table 1). Furthermore, Gene Ontology analysis of the 86 shared targets showed enrichment in xenobiotic response and drug absorption, distribution, metabolism, and excretion (ADME) pathways, as well as folate-related metabolism (Fig. 3D). Notably, genes involved in folate transport and metabolism were enriched in SOX4 and STAT6 ChIP & RNA-seq. data (Fig. 3E), suggesting that SOX4 and STAT6 co-regulate pathways critical for drug metabolism and folate homeostasis in HCC cells. These results indicate that SOX4 and STAT6 co-occupy promoter regions and share a subset of transcriptional targets, with enrichment in folate-related metabolic pathways. This cooperative regulation highlights their role in modulating drug metabolism and folate homeostasis in HCC, we focused on their regulation of MTHFD2, a key enzyme in folate metabolism in this study.

MTHFD2 was regulated by SOX4 and STAT6 in HCC cells
Overexpression of MTHFD2 was associated with poor prognosis and drug resistance in various cancers [36–41]. ChIP-seq results confirmed SOX4 and STAT6 binding to the MTHFD2 proximal promoter (Fig. 4A), and RNA-seq data showed reduced MTHFD2 expression in SOX4−/− (red trace) as compared to Hep3B cells (blue trace) (Fig. 4A-panel I), or STAT6 si-knockdown Hep3B (red trace) as compared with scramble-transfected cells (green trace) (Fig. 4A-panel II). ChIP-qPCR further validated SOX4 and STAT6 binding to the MTHFD2 proximal promoter, which was diminished in SOX4−/− cells (Fig. 4B). MTHFD2 transcripts were decreased in Hep3B and SNU475 cells upon SOX4 or STAT6 knockdown as revealed by qRT-PCR (Fig. 4C, lane 2, 3 and lane 5, 6). Consistently, MTHFD2 protein levels were reduced in SOX4−/− (lane 2, 3), SOX4 si- (lane 5) or STAT6 si-knockdown (lane 7, 9) HCC cells (Fig. 4D). These results suggest that SOX4 and STAT6 regulate MTHFD2 expressions at the transcriptional level.
Analysis of the MTHFD2 promoter revealed putative SOX4 (−181 to −174) and STAT6 (−152 to −140) binding sites within the proximal promoter (Fig. 4E). A series of promoter deletion constructs were cloned upstream of a TATA-less luciferase reporter to assess transcriptional activity. Mutation of either binding site reduced promoter activity, while dual mutations abolished it to basal levels (Fig. 4E). TCGA-LIHC data showed significant positive correlations between STAT6 and MTHFD2 (p = 6.22e−5, r = 0.206) and between SOX4 and MTHFD2 (p = 4.22e−28, r = 0.509) (Fig. 4F). Moreover, high MTHFD2 expression was associated with poorer overall survival in HCC patients (Fig. 4G). Collectively, these findings demonstrate that MTHFD2 is a transcriptional target of the SOX4–STAT6 complex, with both factors required for its promoter activity. Elevated MTHFD2 expression correlates with poor prognosis in HCC using TCGA-LIHC data.

Role of SOX4/STAT6/MTHFD2 regulatory axis in NADH/NADPH production, purine synthesis in HCC cells
MTHFD2 is critical for NADH/NADPH production, nucleotide synthesis [35] and HBP pathway [36] (Fig. 5A). Since our results showed that SOX4 and STAT6 regulate MTHFD2, the related metabolites including NADH/NADPH, purine nucleotides, and HBP metabolites such as UDP, UDP-glucose, and UDP-GlcNAc, were analyzed using LC-MS in Hep3B versus Hep3B SOX4⁻/⁻ cells, as well as in scramble- versus STAT6 si-transfected Hep3B cells. In SOX4⁻/⁻ or STAT6 knockdown cells, levels of NADH, IMP, AMP, ADP, GMP, GDP, UDP, UDP-glucose, and UDP-GlcNAc were significantly reduced compared to Hep3B cells or scramble transfected cells, while NAD levels remained unchanged (Figs. 5B, C). Due to LC-MS sensitivity limits for NADPH, ELISA assays were used, revealing significantly decreased NADPH levels in SOX4−/− (Fig. 5D) and STAT6-knockdown cells (Fig. 5E), with a corresponding increase in NADP/NADPH ratio. These results suggest that SOX4 and STAT6, through regulation of MTHFD2, promote NADH/NADPH production, de novo purine synthesis, and HBP metabolite generation in HCC cells.

Expression of SOX4/STAT6/MTHFD2 in HCC specimen and their association with metabolite profiles and drug resistance
To assess the clinical relevance of the SOX4/STAT6/MTHFD2 axis, we analyzed their protein expression in 62 HCC tumor and adjacent normal tissue pairs, with representative results from 12 pairs shown in Fig. 6A and the remaining 50 in Supplementary Fig. 3. The demographic data of the enrolled 62 cases was listed in Supplementary Table 2. Western blotting revealed significantly elevated levels of SOX4, STAT6, p-STAT6 (Y641), and MTHFD2 in tumor tissues compared to adjacent normal (Fig. 6B). Protein expression levels were positively correlated across the cohort (Fig. 6C). Immunohistochemistry further confirmed co-expression and spatial colocalization of SOX4, STAT6, and MTHFD2 in tumor lesions (Supplementary Fig. 4)
Based on protein expression, samples were stratified into SOX4/STAT6/MTHFD2high (n = 12) and SOX4/STAT6/MTHFD2low (n = 9) groups for metabolomic profiling via LC-MS. Partial Least Squares-Discriminant Analysis (PLS-DA) revealed clear separation between the two groups (Supplementary Fig. 5A) and volcano plot analysis identified significantly altered metabolites (Supplementary Fig. 5B). Heatmap analysis showed increased levels of MTHFD2-associated metabolites—including GSH, GSSG, AMP, GMP, NADH, NADPH, UDP, and UDP-glucose—in the SOX4/STAT6/MTHFD2high group (Fig. 6D, metabolites 38–58).
We next examined the relationship between SOX4, STAT6, and MTHFD2 expression and treatment response in HCC patients. Among the 62 patients, 18 received (TKIs), immunotherapy, or combination therapies (Table 1). Protein levels of SOX4, STAT6, and MTHFD2 were quantified as relative folds (expression level in tumor/normal) and correlated with clinical response (Fig. 6E). Two patients achieved complete response (CR). Patient #6, who had multiple intrahepatic recurrences, showed complete tumor regression after nine cycles of atezolizumab and bevacizumab, maintaining CR for 18 months post-treatment (Table 1, Fig. 6F). Patient #12, treated with adjuvant sorafenib followed by surgical resection of peritoneal metastases, remained in CR for six months (Table 1). Both CR cases had low SOX4, STAT6, and MTHFD2 expressions compared to patients with progressive disease (PD) (Fig. 6E). Two patients (#54, #28) with stable disease (SD) also showed relatively low expression levels (Fig. 6E). In contrast, patient #8, who experienced disease progression following sorafenib and TACE, had high expression of all three proteins (Table 1, Fig. 6F). Analysis of transcriptomic data from GSE109211 further supported these findings, as elevated SOX4, STAT6, and MTHFD2 were observed in nearly half of sorafenib-nonresponsive lesions (22/46), but rarely in sorafenib-responsive tumors (Supplementary Fig. 6).
Taken together, these findings indicate that tumor tissues exhibited higher expression of SOX4, STAT6, and MTHFD2 associated with increased nucleotide synthesis, potentially contributing to drug resistance of patients with HCC.

Targeting MTHFD2 or STAT6 inhibits cell growth and reduces tumor growth in a HCC PDx model
As SOX4 is not directly druggable, targeting its downstream effectors may offer therapeutic potential in SOX4-positive tumors. We assessed the effects of STAT6 (AS1517499) and MTHFD2 (DS18561882) inhibitors on cell viability in SOX4⁺ HCC cell lines. Given that previous studies reported enhanced sensitivity to TKIs or immunotherapy upon MTHFD2 inhibition [39], therefore combination of sorafenib with AS1517499 and DS18561882 were applied for examining cell viability in HCC cells as revealed by CCK8 assay. Treatment with AS1517499, DS18561882, or sorafenib alone significantly reduced cell proliferation in both Hep3B and SNU-475 cells (Fig. 7A–D). Notably, sorafenib combination with DS18561882 showed a synergistic effect in Hep3B cells (ZIP synergy score: 12.119) and an additive effect in SNU-475 cells (ZIP synergy score: 9.402) (Fig. 7A, C; Supplementary Fig. 7A, C). In contrast, sorafenib combined with AS1517499 did not produce synergy in either cell line (Fig. 7B, D; Supplementary Fig. 7B, D).
Consistent findings were observed in sphere formation assays using Hep3B cells. Spheres were treated with AS1517499, DS18561882, sorafenib, or their combinations (Fig. 7E), and stemness was assessed by qRT-PCR analysis of SOX2 and Nanog. As expected, SOX2 and Nanog expression was elevated when Hep3B cells transitioned from 2D to 3D culture (Fig. 7F). Importantly, DS18561882 combined with sorafenib suppressed both sphere growth and stemness marker expression more effectively than monotherapies, whereas AS1517499 did not exert a synergistic effect (Fig. 7F).
To evaluate in vivo efficacy, we employed an HBV⁺ HCC patient-derived xenograft (PDx) model from a 50-year-old male with advanced HCC. Histopathology confirmed a trabecular HCC architecture (Supplementary Fig. 8A). The patient’s tumor progressed rapidly despite sorafenib, and the PDx model (n = 3) demonstrated resistance to sorafenib, regorafenib, and ribavirin (Supplementary Fig. 8B). RNA-seq analysis showed elevated SOX4, MTHFD2 expression relative to GEPIA median levels; however, expression of STAT6 was slightly higher than median levels (Supplementary Fig. 8C). Treatment with AS1517499 or DS18561882 began on day 14 following PDx implantation in NPG mice (n = 4), both inhibitors significantly reduced tumor volume and weight over a 4-week treatment period (Figs. 7G, H).
These findings suggest that targeting STAT6 or MTHFD2 can inhibit cell proliferation in vitro and suppress tumor growth in a drug-resistant HCC PDx model, offering a promising strategy for SOX4⁺ HCC.

Discussion
This study is the first to reveal a mechanistic link between SOX4, STAT6, and MTHFD2 in cancer. We identified a novel SOX4–STAT6–MTHFD2 regulatory axis that plays a critical role in HCC cells, influencing NADH/NADPH production, de novo nucleotide biosynthesis, cancer stemness, and patient prognosis (Fig. 7I). SOX4, previously recognized as a key transcription factor in maintaining cellular stemness, was found to cooperate with STAT6 to transcriptionally regulate MTHFD2, an essential metabolic enzyme required for nucleotide synthesis and cell proliferation in this study. Notably, we demonstrated that SOX4 modulates STAT6 expression through both epigenetic mechanisms—by preventing promoter hypermethylation—and direct protein–protein interactions. Based on our data the SOX4/STAT6/MTHFD2 axis plays a role in modulating folate transport and methionine metabolism in HCC cells (Fig. 3), potential linking SOX4 loss to impaired one-carbon metabolism and possible secondary STAT6 promoter hypermethylation. However, the specific mechanism of demethylation is uncertain. SOX4 has been previously implicated in cancer cell proliferation and angiogenesis [6, 11, 12], its newly established connection to MTHFD2 further underscores its pivotal role in promoting HCC progression and therapeutic resistance.
Among the components of this regulatory axis, STAT6 has been well-documented in macrophage activation [21], but its involvement in HCC tumorigenesis has been less explored [34]. Notably, therapeutic targeting of STAT6 expression by its specific anti-sense oligonucleotide in macrophages has been investigated in clinical trials for advanced HCC, showing promising anti-tumor effects [42]. Although the therapy mentioned above was targeted to macrophages instead of cancer cells, the treatment effect on cancer cells itself should be paid attention to since the expression of STAT6 protein was also overexpressed in tumor lesions of HCC (Fig. 6A). Our findings indicate that inhibition of p-STAT6 using AS1517499 reduced HCC cell growth in vitro (Fig. 7B-D). However, the effect on HCC PDx tumor growth was less pronounced compared to MTHFD2 inhibition (Figs. 7G, H). One possible explanation is that STAT6 transcript levels were not as much elevated in the HBV + HCC PDx model as compared with the median STAT6 transcript level from GEPIA (Supplementary Fig. 8C). Furthermore, while AS1517499 reduced Hep3B sphere size, it did not significantly downregulate the stemness markers SOX2 and Nanog, suggesting that inhibition of p-STAT6 may exert its effects through alternative, yet unidentified, pathways that require further investigation.
Several inhibitors have been developed to target MTHFD2 for cancer treatment based on its structure [43]. A study has shown that treatment with MTHFD2 inhibitors in acute myeloid leukemia slows the replication fork, increasing replication stress and inducing cellular apoptosis in vitro [37]. In HCC, MTHFD2 repression has been shown to inhibit cell growth and sensitize cells to TKIs [39]. Additionally, TP53 deficiency has been reported to trigger the transcriptional activation of MTHFD2, promoting cell proliferation [44]. TCGA-LIHC data indicate that TP53 mutations occur in approximately 32% of HCC specimens, though site-specific mutant effects [45, 46] on activation of MTHFD2 remain unclear. Our study’s limitation is that it did not assess the status of TP53 mutations in clinical samples. Besides, in folate pathway, 5-Fluorouracil (5-FU) is a chemotherapy drug that targets TYMS to reduce pyrimidine synthesis, which is used to treat various kinds of cancers [47], but HCC is considered as chemotherapy-resistant tumor, therefore 5-FU is not routinely administered [48]. However, dihydropyrimidine dehydrogenase (DPD) activity may predict 5-FU efficacy [49]. To assess whether simultaneous inhibition of TYMS and MTHFD2 could enhance therapeutic effects, we evaluated the combination of 5-FU with MTHFD2 or STAT6 inhibitors. In Hep3B cells, neither combination produced synergy (Supplementary Fig. 9A, B). By contrast, 5-FU combined MTHFD2 inhibition showed a synergistic effect in SNU-475 cells (Supplementary Fig. 9C), whereas no synergy was observed with 5-FU and STAT6 inhibition (supplementary Fig. 9D). Since DPD activity was not examined in these cells, the observed differences in responsiveness may reflect cell line–specific metabolic features and merit further investigation.
In conclusion, SOX4 epigenetically regulates STAT6 transcription and forms a complex with STAT6 to co-regulate MTHFD2 expression in HCC cells. This regulation enhances NADPH production, purine synthesis, and activation of the HBP pathway in tumors (Fig. 7I). Tumors with high SOX4/STAT6/MTHFD2 expression are more resistant to TKIs and immunotherapy, though our sample size was limited. However, analysis of transcriptomic data from GSE109211 supported our findings, as elevated SOX4, STAT6, and MTHFD2 were observed in nearly half of sorafenib-nonresponsive lesions, but rarely in sorafenib-responsive tumors. Besides, targeting STAT6 or MTHFD2 significantly reduces cell proliferation in vitro and attenuates tumor growth in sorafenib-resistant PDx models (Fig. 7I). Therefore, disrupting this regulatory axis may help restore therapeutic sensitivity in HCC patients, providing a potential strategy for improving treatment outcomes.

Patients and methods
This study was conducted in accordance with the Declaration of Helsinki and all relevant institutional guidelines and regulations. The study protocol was approved by the Institutional Review Board of Chang Gung Memorial Foundation, Taiwan (approval numbers: 202100994B0, 202200985B0, and 202202182B0). A total of 62 patients (Supplementary Table 2) who underwent curative-intent surgical resection for hepatocellular carcinoma at Chang Gung Memorial Hospital (CGMH), Linkou or New Taipei Municipal Tucheng Hospital were retrospectively enrolled. Written informed consent was obtained from all participants prior to tissue collection and data analysis. Patients aged 20–85 years were included, and individuals with a history of other malignancies were excluded.
Paired tumor and adjacent non-tumorous liver tissues (n = 62) were obtained and processed by the CGMH Research Specimen Processing Laboratory. Clinical and pathological data were collected retrospectively and analyzed in an anonymized manner. All downstream protein expression analyses and molecular studies were performed using de-identified samples. The clinical images presented in Fig. 6F are non-identifiable and contain no personal or identifying information. Therefore, additional consent for publication of images was not required. However, the written informed consent for publication of these images was obtained from patient #6; for patient #8, who passed away prior to manuscript preparation, publication consent was provided by his legally authorized representative.

Cell lines and three-dimensional (3D) culture
Cell lines used in this study are listed in Supplementary Table 3. All were maintained per provider instructions. Hep3B SOX4−/− cells [11] were cultured in complete medium with 4 ng/ml puromycin (Thermo Scientific). For 3D culture, 5000 cells were seeded in six-well plates coated with polypeptide polyelectrolyte multilayer films (Acrocyte Therapeutics, Taiwan) [50] The complete culture medium was supplemented with 1× B27 (Thermo Fisher Scientific), 20 ng/mL EGF (Thermo Fisher Scientific), and 20 ng/mL basic FGF (Sigma-Aldrich) for sphere culture.

HCC tumoroid culture
Liver cancer tumoroids were isolated and cultured using a previously described method [51]. About 1 cm³ of liver cancer tissue was minced, digested at 37 °C for 1–2 h, filtered through a 70 µm strainer, and centrifuged at 300–400 g for 5 min. After PBS washing, cells were cultured in a 24-well organoid culture plate (Acrocyte Therapeutics, Taiwan).

Genome DNA methylation sequence
Genomic DNA (500 ng) was fragmented (150–200 bp) (Covaris S2, Covaris Inc., USA), processed with the TruSeq® Methyl Capture EPIC Kit (Illumina), and enriched for 3.3 M CpG sites. Bisulfite conversion and PCR amplification were performed with Kapa HiFi Uracil+ polymerase. Libraries were assessed by Qubit and Bioanalyzer and sequenced on a NovaSeq 6000. Data processing followed [52] and was deposited on Gene Expression Omnibus (GEO) with accession number GSE286327.

RNA sequencing
RNA-seq was conducted at the CGMH Genomic Medicine Research Core [53]. Data from Hep3B vs. Hep3B SOX4⁻/⁻ (#1, #2) and Hep3B/Scramble vs. Hep3B/STAT6 siRNA were deposited in GEO (GSE277540).

Immunofluorescence staining
The immunofluorescence staining was as previously described [54] and the antibodies used in this study were listed in Supplementary Table 4. The slides were examined under a Leica TCS SP2 confocal laser scanning microscope (Leica Microsystems GmbH, Wetzlar, Germany).

Immunohistochemistry (IHC)
The procedure of IHC staining was as previously described [11] and the antibodies used for IHC are listed on listed in Supplementary Table 4.

Western blotting and co-immunoprecipitation
The western blotting and co-immuno-precipitation procedures were as previously described [55] and the primary antibodies used in this study were listed in Supplementary Table 4.

Quantitative reverse-transcription polymerase chain reaction (qRT-PCR)
RNA was extracted with TRIzol (Thermo Fisher Scientific, 15596026), reverse transcribed, and analyzed by qPCR as described [11] using primers in Supplementary Table 5.

Chromatin immunoprecipitation (ChIP) sequence
CUT&RUN (Cell Signaling, 86652) was performed on 1 × 10⁵ Hep3B or Hep3B SOX4−/− cells with anti-SOX4 (Abnova, PAB14092). SOX4–DNA complexes were digested by pAG-MNase, purified, followed by library construction with a DNA library prep kit (Cell Signaling, #56795 and #29580) and sequenced (150 bp paired-end reads; NovaSeq 6000). Data were processed in Partek Flow and were deposited in GEO (GSE277728).

Luciferase assay
Cells were transfected with luciferase and Renilla (1/20 of luciferase) reporters (Lipofectamine 3000, Thermo Fisher, L3000015). Activity was measured at 24 h using Dual-Glo® (Promega, E2920) and normalized to Renilla.

NADPH/NADP assay
Total NADP and NADPH levels in HCC cells were measured using the NADP/NADPH Assay Kit (Abcam, ab65349) per the manufacturer’s instructions. Values were determined from a standard curve and expressed as pmol/well. The NADP total (NADPt)/NADPH ratio was calculated as (NADPt-NADPH)/NADPH.

Cell proliferation assay
Hep3B cells (1 × 10⁴/per 96 well) were treated with AS1517499 (HY-100614, MedChemExpress, USA), DS18561882 (HY-130251, MedChemExpress, USA), sorafenib (HY-10201, MedChemExpress, USA), or 5-FU (Nang-Kuang pharmaceutical Co.). DMSO served as control. Viability was measured with CCK-8 (Biotools, Taiwan). Synergy was assessed by the ZIP model (SynergyFinder 3.0) [56].

HCC Patient-derived xenograft (PDx) models
HCC PDx models were established as described [57]. Tumor specimens from metastatic HCC patients were cut into 3–5 mm pieces and implanted subcutaneously into anesthetized 5-week-old male NPG mice (NOD.Cg-Prkdcscid Il2rgtm1Vst/Vst, BioLASCO Taiwan), which were maintained under specific pathogen-free conditions under an approved protocol (2015121808) at the CGMH Laboratory Animal Center, Taiwan. The study comprised three groups (Vehicle, AS1517499, and DS18561882), with four mice per group. Mice were randomly assigned to the Vehicle, AS1517499, or DS18561882 treatment groups using a simple randomization procedure. All animals were processed in parallel under the same technician, housing, and handling conditions to minimize bias. AS1517499 (20 mg/mL) and DS18561882 (40 mg/mL) were injected intraperitoneally (0.1 mL) twice weekly from day 14 for 4 weeks. The investigators were not blinded to group allocation during treatment or outcome assessment. The tumor size measurements using calipers were objective and automated, minimizing investigator bias.

Protein identification by tandem mass spectrometry
Protein identification was performed at the Clinical Proteomics Core Laboratory of CGMH, Linkou. Peptide spectra analysis followed established procedures [55].

Cellular metabolites analyzed via LC/MS
Soluble metabolites in HCC cells were analyzed at the Core Laboratory of the Healthy Aging Research Center, Chang-Gung University, following a previously described protocol [58]. Data analysis was conducted using MetaboAnalyst 6.0 [59].

Statistical Analysis
Analyses were performed in GraphPad Prism 9; Student’s t test for quantitative data. Overall survival was analyzed by Kaplan–Meier in SPSS 21. P < 0.05 was considered significant. Transcripts of SOX4, STAT6, MTHFD2 was download from GSE109211 dataset. The transcript data of SOX4, STAT6 and MTHFD2 in tumor lesions was normalized as percentile gene expression for statistical analysis.

Supplementary information