Intratumoral Fusobacterium nucleatum promotes ferroptosis resistance and a disulfide-stress-associated vulnerability under glucose restriction

Abstract

Fusobacterium nucleatum (Fn) is an intracellular oncogenic bacterium implicated in multiple cancers. Although Fn has been linked to resistance to anticancer therapies, its effects on therapies targeting metabolic cell-death programs remain unclear. Here, we show that intratumoral Fn abundance is positively associated with SLC7A11 expression and that Fn infection attenuates Era- and RSL3-induced ferroptosis in tumor cells. Live-versus-killed, donor-based, and pharmacological evidence suggests that H₂S generated during Fn infection contributes to SLC7A11 induction and engagement of the downstream GSH/GPX4 axis. Fn infection also reduces lipid-peroxide and Fe²⁺ accumulation and is accompanied by ACSL4 downregulation and FTH1 upregulation. Fn-infected cells display increased glucose uptake and a heightened dependence on glucose. Under glucose-replete conditions, Fn-infected cells are relatively resistant to ferroptosis, whereas glucose deprivation exposes a conditional vulnerability accompanied by disulfide stress. RSL3 further sensitized Fn-infected cells to glucose restriction, and combined GPX4 inhibition with the GLUT1 inhibitor BAY876 enhanced antitumor activity in Fn-infected solid-tumor models. These findings identify a nutrient-dependent metabolic liability associated with Fn infection and support further evaluation of combined glucose-transport and GPX4 inhibition in appropriately controlled models.

Graphical abstract

Introduction

Recent studies have demonstrated the presence and functional relevance of intratumoral bacteria across multiple cancer types. ^{1 , 2} Many of these microorganisms promote cancer progression by remodeling the tumor microenvironment (TME). ² Fusobacterium nucleatum (Fn), an obligate anaerobe that commonly colonizes the oral cavity, has been described as a symbiont, opportunistic pathogen, and oncobacterium. ³ As a facultative intracellular pathogen, ^{4 , 5} Fn promotes tumor progression through virulence factors including FadA, ⁶ Fap2, ⁶ and Fn-Dps. ⁵ Fn also stimulates host cells to produce inflammatory cytokines ⁷ and pro-metastatic exosomes, ⁸ thereby fostering an immunosuppressive TME. ⁹ Recent studies have further linked Fn to altered responses to immune checkpoint inhibitors, ^{10 , 11} chemotherapy, ¹² and radiotherapy, ¹³ underscoring the therapeutic challenges posed by Fn-positive tumors.

Ferroptosis is an iron-dependent, nonapoptotic form of regulated cell death characterized by lipid-peroxide accumulation. ¹⁴ Pharmacological induction of ferroptosis may help overcome therapy resistance in cancer. ¹⁵ Solute carrier family 7 member 11 (SLC7A11), a cystine/glutamate antiporter frequently overexpressed in human cancers, mediates the transport activity of system x_c ^- at the plasma membrane. ¹⁶ Inhibition of SLC7A11 blocks cystine uptake, depletes glutathione (GSH), inactivates glutathione peroxidase 4 (GPX4), and promotes ferroptosis. ^{17 , 18} Conversely, under glucose deprivation, high SLC7A11 expression promotes intracellular cystine accumulation, NADPH depletion, aberrant disulfide bonding of actin cytoskeletal proteins, and disulfidptosis. ¹⁹ Thus, SLC7A11 has opposing, nutrient-dependent effects on ferroptosis and disulfidptosis, complicating strategies intended to trigger both cell-death programs.

Fn produces H₂S through cysteine catabolism, and this gasotransmitter can quench reactive oxygen species (ROS) and facilitate intracellular bacterial survival. ¹¹ Whether H₂S generated during Fn infection influences host-cell ferroptosis or glucose-restriction-associated disulfide stress through changes in cystine transport and metabolism remains unclear. A recent study reported that Fn suppresses ferroptosis by increasing GPX4 expression and that GPX4 inhibition can overcome oxaliplatin resistance in colorectal cancer (CRC). ²⁰ However, the relationships among Fn infection, the SLC7A11/GPX4 axis, ferroptosis, and disulfide-stress vulnerability remain incompletely defined.

In this study, we investigated the association between intratumoral Fn abundance and SLC7A11 expression and examined whether H₂S generated during Fn infection contributes to engagement of the SLC7A11/GSH/GPX4 axis in tumor cells. We further examined whether the Fn-associated SLC7A11^high state is linked to nutrient-dependent disulfide and GPX4-directed lipid-peroxide stress and evaluated glucose-transport/GPX4 cotargeting within Fn-infected tumor models.

Results

Intratumoral Fn abundance is positively correlated with SLC7A11 expression in human tumor tissues

To identify conserved transcriptional alterations induced by Fn infection with high reliability, we performed RNA-seq in two classic esophageal squamous cell carcinoma (ESCC) cell lines with distinct differentiation statuses: kyse30 (well-differentiated) and kyse150 (poorly differentiated). The use of two heterogeneous ESCC cell lines allowed us to minimize cell-line-specific transcriptional changes and focus on conserved gene expression events driven by Fn infection. Venn diagrams revealed the overlap of 14 significantly upregulated genes common to both cell lines after 48 h of infection with Fn (Figure 1a). Among these, 4 genes (SLC7A11, GDF15, CHAC1 and NUPR1) are implicated in ferroptosis regulation, while the others are implicated in tumor proliferation and progression (Figure 1b).

Figure 1.

SLC7A11 expression is positively correlated with Fn abundance in human tumor tissues. a, b. Venn diagram (a) and heat map (b) showing genes upregulated in both kyse30 and kyse150 cells after Fn infection (MOI = 10:1, 48 h). c. SLC7A11 mRNA expression in normal-adjacent tissue (NAT) and tumor tissue from TCGA-BRCA, TCGA-ESCA, and TCGA-PAAD. d. Kaplan–Meier analyzes of overall survival (OS) and disease-free interval (DFI) according to SLC7A11 expression in the TCGA-BRCA, TCGA-ESCA, and TCGA-PAAD cohorts. For each cohort and survival endpoint, patients with available SLC7A11 expression and corresponding survival data were dichotomized into high- and low-expression groups using an optimal cutoff determined by maximally selected rank statistics. Hazard ratios are presented for the high-expression group relative to the low-expression group, together with their 95% confidence intervals. e. Representative immunofluorescence images of SLC7A11 (green) and Fn (red) in human ESCC and BRCA tissues; nuclei were stained with DAPI (blue). Scale bars, 100 μm. f. Spearman correlation between genus-level Fusobacterium abundance and SLC7A11 mRNA expression in paired TCMA/TCGA ESCA samples (n = 25). g. Linear-regression analysis of Fn abundance [ΔCt = -(Ct (Fn) - Ct(18S))] and SLC7A11 IHC H-score in ESCC (n = 13) and BRCA (n = 9). Two-tailed Mann-Whitney U tests were used in c; log-rank tests in d; Spearman’s rank-correlation test in f; and simple linear regression with a two-sided test of the slope in g. No inferential statistical test was applied to the representative images in e. *P < 0.05, **P < 0.01, and ***P < 0.001.

Given that SLC7A11 is a key regulator of ferroptosis and disulfidptosis, we analyzed its expression in three cancer types (BRCA: Breast invasive carcinoma; ESCA: Esophageal carcinoma; PAAD: Pancreatic adenocarcinoma) using the TCGA database. As shown in Figure 1c, SLC7A11 mRNA levels were significantly upregulated in these tumor tissues compared to normal adjacent tissues (NAT). Kaplan-Meier analyzes indicated that high SLC7A11 expression was associated with poorer survival outcomes in the statistically significant comparisons across the BRCA, ESCA, and PAAD cohorts. However, the ESCA disease-free interval analysis showed only a nonsignificant trend toward shorter disease-free interval (P = 0.056; Figure 1d). Using immunofluorescence staining, we also detected Fn in SLC7A11-positive ESCC and BRCA tissues (Figure 1e).

Because microbiome resources differ across cancer types, we explicitly separated database-derived from literature-supported evidence. In the TCMA ESCA cohort, Fusobacterium was detected in 57.3% of samples and ranked fourth among the displayed genera by prevalence (Fig. S1). Published pan-cancer and disease-focused studies support the presence of intratumoral Fn in subsets of breast cancers and in pancreatic tumors or precursor lesions, ^{1 , 21-23} but paired microbiome-transcriptome data were not available for a reliable PAAD correlation analysis. In paired TCMA/TCGA ESCA data, Fusobacterium abundance positively correlated with SLC7A11 mRNA (Spearman r = 0.4469, P = 0.0251; Figure 1f).

We collected 31 paraffin-embedded tumor specimens from patients with ESCC (n = 17) or BRCA (n = 14). After exclusion of specimens with unsuccessful DNA extraction or undetectable Fn, 13 ESCC and 9 BRCA specimens were included in the correlation analysis. SLC7A11 expression was evaluated by immunohistochemistry (IHC), with a higher H-score indicating greater SLC7A11 staining. Fn abundance was quantified by qPCR using ΔCt = −[Ct(Fn) − Ct(18S)]; under this definition, a higher ΔCt value indicates greater Fn abundance (Table S1). Linear-regression analysis revealed significant positive associations between Fn abundance and SLC7A11 expression in both ESCC (R² = 0.3780, P = 0.0253) and BRCA tissues (R² = 0.5608, P = 0.0202) (Figure 1g).

H₂S is associated with SLC7A11/GSH/GPX4 engagement during Fn infection

We next examined the effects of Fn infection on the SLC7A11/GSH/GPX4 axis in four tumor cell lines: two human ESCC cell lines (Eca-109, also termed E109; kyse30), the human breast cancer cell line MDA-MB-231 (hereafter, 231), and the human pancreatic cancer cell line PANC-1. Fn infection (multiplicity of infection [MOI] = 10:1) significantly increased SLC7A11 mRNA and protein expression in all four cell lines (Figure 2a,b).

Figure 2.

H₂S is associated with SLC7A11/GSH/GPX4 engagement during Fn infection. a. SLC7A11 mRNA expression in E109, 231, kyse30, and PANC-1 cells after Fn infection (MOI = 10:1, 48 h). b. SLC7A11 immunoblots and densitometric quantification after Fn infection. c,d. WSP-5 flow-cytometric profiles in E109 and 231 cells (c) and kyse30 and PANC-1 cells (d) after treatment with live Fn, heat-killed Fn (K-Fn), or NaHS (100 μM) for 48 h; d also shows representative WSP-5 confocal images in kyse30 cells. Scale bar, 30 μm. e. Viability after treatment with the indicated NaHS concentrations for 48 h. f. SLC7A11 immunoblots and densitometric quantification after NaHS treatment (100 μM, 48 h). g. WSP-5 flow cytometry and quantification in Fn-infected E109 and 231 cells with or without Ga (2 mM). h. SLC7A11 immunoblots and densitometric quantification after Fn and/or Ga treatment. i. KEGG enrichment of Fn-responsive genes in kyse30 and kyse150 cells. j-l. Metabolomic profiling of E109 cells after Fn infection (24 h) followed by Era treatment (24 h): positive-ion principal-component analysis (j), pathway-topology analysis (k), and heat map of glutathione-metabolism-related metabolites (l). m. Cystine-analog uptake in E109 and 231 cells after Fn infection (24 h) followed by Era treatment (35 min). n. Intracellular GSH after Fn infection (24 h) followed by Era treatment (24 h). o, p. GSH after SLC7A11 silencing and Fn infection (o) or NaHS treatment (p). q, r. GPX4 immunoblots and densitometric quantification after Fn infection (q) or NaHS treatment (r). Data are mean ± SD from three biological replicates. Two-tailed unpaired t tests were used in a, b, f, q, and r; two-way ANOVA in e; and one-way ANOVA in g, h, and m–p. *P < 0.05, **P < 0.01, and ***P < 0.001. Era concentrations are listed in Table S4.

To assess whether H₂S generated during Fn infection contributes to this response, we used the fluorescent H₂S-sensing probe WSP-5 to monitor intracellular H₂S. Flow cytometry and confocal imaging showed a marked increase in the WSP-5 signal in cells infected with live Fn, but not in cells exposed to heat-killed Fn (K-Fn) (Figure 2c,d). Treatment with the H₂S donor NaHS increased normalized cell viability over the tested concentration range (Figure 2e) and markedly increased SLC7A11 protein expression after 48 h at 100 μM (Figure 2f).

Gallic acid (Ga), an inhibitor of the Fn H₂S-generating enzyme Fn1419, ²⁴ had no measurable effect on Fn growth or tumor-cell viability over the tested concentration range (Fig. S2a, b). Ga (2 mM) reduced the Fn-induced WSP-5 signal (Figure 2g and Fig. S2c) and attenuated SLC7A11 induction, whereas Ga alone did not alter basal SLC7A11 expression (Figure 2h and Fig. S2d, e). Together, the live-versus-killed comparison, donor phenocopy, and pharmacological inhibition support a contributory role for H₂S generated during Fn infection in SLC7A11 induction, without establishing a direct molecular target.

KEGG analysis of the RNA-sequencing data identified amino-acid biosynthesis among the pathways altered by Fn infection (Figure 2i). We therefore profiled E109-cell metabolism after Fn infection and erastin (Era) treatment. Principal-component and pathway-topology analyzes separated the treatment groups and identified glutathione metabolism as a prominent altered pathway (Figure 2j,k; Tables S5 and S6). Era depleted reduced- and oxidized-glutathione-related metabolites (Figure 2l). Direct measurements showed that Era reduced cystine uptake and intracellular GSH. Relative to Era alone, Fn partially preserved cystine uptake in E109, 231, and PANC-1 cells and preserved GSH in a cell-line-dependent manner; some comparisons in kyse30 and PANC-1 cells were not statistically significant (Figure 2m,n and Fig. S3). Fn alone also increased basal cystine uptake and/or GSH in the tested models.

To determine whether SLC7A11 is required for this response, we screened three SLC7A11-targeting siRNAs and used the most effective construct (Fig. S4a). SLC7A11 silencing prevented the Fn- or NaHS-dependent increase in SLC7A11 protein (Fig. S4b, c), lowered basal GSH, and markedly blunted GSH maintenance after Fn or NaHS treatment (Figure 2o,p). Fn and NaHS increased GPX4 expression (Figure 2q,r and Fig. S5), whereas SLC7A11 silencing attenuated GPX4 induction (Fig. S6). These findings support an SLC7A11-dependent GSH/GPX4 response associated with H₂S signaling during Fn infection, although the relevant transcriptional or post-translational H₂S sensor remains to be identified.

Fn infection attenuates Era- and RSL3-induced ferroptosis in vitro and in vivo

We next tested whether Fn infection alters tumor-cell sensitivity to the ferroptosis inducers Era and RSL3. LDH-release assays, expressed as cell viability, showed that cells infected with Fn for 24 h retained higher viability after Era or RSL3 treatment than the corresponding inducer-only groups in all four cell lines. Ferrostatin-1 (Fer-1) served as a pharmacological control for ferroptosis inhibition (Figure 3a,b and Fig. S7a, b).

Figure 3.

Fn infection partially attenuates Era- and RSL3-induced ferroptosis in E109 and 231 cells. a, b. Viability of E109 and 231 cells after Era (a) or RSL3 (b) treatment with the indicated Fn and Fer-1 conditions. c. Representative TEM images of kyse30 cells after Era treatment with or without Fn. Scale bars, 2 μm. d. Three-dimensional confocal visualization of α-tubulin (green) and Fn (red) in E109 and 231 cells during Era exposure; nuclei were stained with DAPI. e, f. Total viable cell and bacterial counts (e) and viable intracellular Fn per cell (f) in gentamicin-protection assays. g, h. DCFH-DA-associated fluorescence (g) and C11-BODIPY oxidation (h) in E109 and 231 cells. i. FerroOrange imaging and quantification of intracellular Fe²⁺. Scale bar, 30 μm. j. ACSL4 immunoblots and densitometric quantification after Fn infection. k. C11-BODIPY analysis after SLC7A11 silencing and/or ACSL4 overexpression with or without Fn. l. Viability after FTH1 silencing and/or ACSL4 overexpression with or without Fn under the indicated ferroptotic challenge. m. FTH1 immunoblots and densitometric quantification after Fn infection. n. FTH1 immunoblots after FTH1 silencing with or without Fn. Cells were infected with Fn (MOI = 10:1) for 24 h before Era or RSL3 treatment for 24 h; Fer-1 (1 μM) was added 4 h before the inducer. Data are mean ± SD from three biological replicates. Two-tailed unpaired t tests were used in e, f, j, and m; one-way ANOVA in a, b, g–i, k, l, and n. No inferential statistical test was applied to the representative images in c and d. *P < 0.05, **P < 0.01, and ***P < 0.001; ns, not significant. Drug concentrations are listed in Table S4.

Transmission electron microscopy of kyse30 and PANC-1 cells showed shrunken mitochondria with increased membrane density after Era treatment; these changes were attenuated by Fn preinfection (Figures 3c and S7c). Confocal imaging and three-dimensional reconstruction of z-stacks confirmed intracellular Fn during Era exposure (Figs. 3 d and S7d). In gentamicin-protection assays, viable intracellular Fn remained recoverable 48 h after the initial infection (24 h after Era addition) (Figures 3e,f and S7e, f).

Flow-cytometric analysis using DCFH-DA and C11-BODIPY showed that Era increased the intracellular DCFH-DA signal and lipid oxidation, respectively, in all four tumor cell lines, whereas Fer-1 reduced these responses (Figure 3g,h and Fig. S7g, h). In Fn-infected cells exposed to Era, the DCFH-DA signal increased, whereas C11-BODIPY oxidation was lower than in the Era-only group. Fn infection also reduced Era-induced Fe²⁺ accumulation (Figure 3i and Fig. S7i). These changes were accompanied by lower ACSL4 and higher FTH1 expression after Fn infection (Figure 3j,m and Fig. S7j, k). ACSL4 overexpression was validated in E109 and 231 cells (Fig. S8), and FTH1-targeting siRNAs were screened to identify an effective knockdown reagent (Fig. S9). ACSL4 overexpression and/or FTH1 silencing shifted lipid-oxidation and viability readouts toward greater ferroptotic stress in Fn-infected cells (Figure 3k,l), whereas FTH1 silencing blunted Fn-dependent FTH1 induction (Figure 3n). Because these interventions did not uniformly eliminate Fn-associated protection and intracellular Fe²⁺ was not reassessed after FTH1 silencing, the data do not establish a unified ACSL4/FTH1 axis. Instead, ACSL4 downregulation and FTH1 upregulation are complementary changes that each partially contribute to Fn-associated ferroptosis resistance alongside the SLC7A11/GSH/GPX4 response.

We then established two animal tumor models: an E109 ESCC subcutaneous xenograft model and a 231 BRCA orthotopic mammary fat-pad model. Fn-associated fluorescence was visualized in E109 xenograft-bearing mice after intravenous administration of DiD-labeled Fn. Fourteen days after a single injection, fluorescence remained detectable predominantly in the E109 xenograft, liver, and spleen (Figure 4a). In a separate experiment, male mice received two intravenous injections of Fn (10⁶ CFU) before subcutaneous implantation of E109 cells and subsequent Era treatment (10 mg/kg; Figure 4b). Era significantly reduced tumor volume and weight, but this effect was attenuated by Fn preinfection (Figure 4c,d). qPCR confirmed Fn DNA in tumors from Fn-infected mice (Figure 4e). Consistent with the in vitro findings, SLC7A11 and GPX4 expression increased in tumors from Fn-infected mice (Figure 4f,g and Fig. S10).

Figure 4.

Fn infection attenuates Era- and RSL3-induced antitumor effects in vivo. a. Ex vivo fluorescence images of the E109 xenograft and excised organs 14 days after intravenous administration of DiD-labeled Fn; tissue identities are indicated in the panel. b-g. E109 ESCC subcutaneous xenografts (n = 5 per group) in mice infected with Fn before E109 implantation and treated with Era: treatment schedule (b), tumor volume (c), tumor weight (d), tumor Fn DNA measured by qPCR (e), SLC7A11/Fn immunofluorescence (f; scale bar, 30 μm), and SLC7A11/GPX4 immunoblotting (g). h, i. Tumor volume (h) and weight (i) in E109 xenografts infected with Fn after implantation and treated with RSL3. j-l. 231 orthotopic mammary fat-pad tumors (1 × 10⁶ cells; n = 5 per group) treated with Era: schedule (j), tumor volume (k), and tumor weight (l). m, n. Tumor volume (m) and weight (n) in 231 orthotopic tumors treated with RSL3. Tumor data are mean ± SEM. A two-tailed Mann–Whitney U test was used in e; two-way ANOVA in c, h, k, and m; and one-way ANOVA in d, i, l, and n. No inferential statistical test was applied to the ex vivo images in a, treatment schedules in b and j, immunofluorescence images in f, or representative immunoblots in g. *P < 0.05, **P < 0.01, and ***P < 0.001; ns, not significant.

Next, we varied the timing of infection to repeat this experiment. Mice were infected with Fn after tumor cell inoculation and subsequently treated with another ferroptosis inducer, RSL3 (Fig. S11a). As expected, Fn infection after tumor cell implantation similarly attenuated the anti-tumor efficacy of RSL3 in mice (Figure 4h,i). Finally, consistent results were obtained in orthotopic mouse models (Figure 4j–l). Fn infection was found to impair the anti-tumor effects of Era, though no significant difference in tumor weight was observed between the Fn+Era and Era groups, likely due to the small size of tumors. The results of Fn infection and RSL3 treatment in orthotopic mice models were similar to those of E109 subcutaneous xenograft models (Fig. 4m-n and S11b). Collectively, these in vivo results demonstrate that Fn infection significantly attenuates the anti-tumor effects of ferroptosis inducers.

GPX4 inhibition increases the sensitivity of Fn-infected tumor cells to glucose restriction

Under glucose deprivation, aberrant intracellular disulfide accumulation in SLC7A11^high cells has been linked to canonical disulfidptosis. ¹⁹ We therefore examined whether Fn-associated SLC7A11 upregulation altered the response of tumor cells to glucose restriction. In glucose-free medium, Fn infection and NaHS treatment increased SLC7A11 and GPX4 expression in E109, 231, kyse30, and PANC-1 cells (Figure 5a and Fig. S12a-d), indicating that Fn-infected cells maintain an SLC7A11^high, glucose-dependent metabolic state.

Figure 5.

RSL3 increases the vulnerability of Fn-infected tumor cells to glucose restriction. a. SLC7A11 and GPX4 immunoblots in E109 and 231 cells after Fn infection or NaHS treatment in glucose-free medium. b. 2-NBDG uptake after Fn infection followed by Era or RSL3 treatment. c. Viability during glucose withdrawal after Fn infection and Era or RSL3 treatment. d. Viability after BAY876 treatment with Era or RSL3 in Fn-infected cells. e. F-actin architecture in glucose-free medium after the indicated treatments; nuclei were stained with DAPI (blue), and F-actin is shown in red. Scale bars, 20 μm. f. Cystine-analog uptake during glucose withdrawal. g. NADP⁺/NADPH ratios during glucose withdrawal. h. Working model for the nutrient-dependent effects of Era and RSL3 in Fn-infected SLC7A11^high cells. Data are mean ± SD from three biological replicates. One-way ANOVA was used in b–d, f, and g. No inferential statistical test was applied to the representative immunoblots in a, representative images in e, or schematic model in h. *P < 0.05, **P < 0.01, and ***P < 0.001; ns, not significant. Fn infection preceded Era/RSL3 treatment by 24 h; BAY876 (5 μM) preceded Era/RSL3 treatment by 3 h. Drug concentrations are listed in Table S4.

Using the fluorescent glucose analog 2-NBDG, we found that Fn infection increased glucose uptake in glucose-containing medium, including in cells treated with Era or RSL3 (Figure 5b). Fn infection increased cell viability when glucose was available (Figure 3a) but reduced viability when glucose was withdrawn (Figure 5c). Era and RSL3 had contrasting effects during glucose deprivation: RSL3 further reduced the viability of Fn-infected cells, whereas Era partially preserved viability (Figure 5c). Pharmacological inhibition of glucose transport with BAY876 produced a similar pattern, with Fn-infected cells showing greater sensitivity to BAY876 plus RSL3 than to BAY876 plus Era (Figure 5d).

Fn infection increased cystine uptake and the NADP⁺/NADPH ratio during glucose deprivation and was accompanied by cell shrinkage and F-actin collapse (Figure 5e–g). Era reduced cystine uptake and partially preserved F-actin architecture, whereas RSL3 did not suppress cystine uptake and further increased redox imbalance and cytoskeletal collapse. These findings are consistent with the compounds' distinct sites of action: Era inhibits SLC7A11-mediated cystine transport upstream, thereby relieving the cystine-dependent NADPH burden, whereas RSL3 inhibits GPX4 downstream without interrupting cystine influx. To assess whether the Fn+BAY876+RSL3 phenotype could be explained by canonical ferroptosis, apoptosis, or necroptosis, we performed rescue experiments using Fer-1, Z-VAD-fmk, and necrostatin-1 (Nec-1), respectively. None of these inhibitors significantly restored cell viability or uniformly preserved F-actin architecture in Fn-infected E109 or 231 cells treated with BAY876 plus RSL3 (Fig. S13). Moreover, high-molecular-weight β-actin-containing species were preferentially detected under nonreducing conditions and diminished after reduction (Fig. S14), consistent with disulfide-linked cytoskeletal protein complexes. Together, these findings indicate that the response cannot be explained by a single canonical cell-death pathway and support a composite phenotype with a prominent disulfide-stress component superimposed on GPX4-directed lipid-peroxide stress (Figure 5h).

We next evaluated glucose-transport and ferroptosis-inducer combinations in Fn-infected tumor models. In Fn-infected E109 subcutaneous xenografts, BAY876 alone significantly inhibited tumor growth, and BAY876 plus RSL3 further reduced tumor volume and weight relative to BAY876 alone; BAY876 plus Era did not produce a significant additional effect (Figure 6a–c). A comparable treatment pattern was observed in Fn-infected 231 orthotopic mammary fat-pad tumors (Figure 6d–f). Representative IHC images from tumors treated with Fn+BAY876+RSL3 were consistent with higher SLC7A11 and lower GPX4 staining than those treated with Fn+BAY876+Era (Figure 6g,h).

Figure 6.

BAY876 plus RSL3 enhances antitumor activity within the tested Fn-infected tumor models. a-c. Fn-infected E109 subcutaneous xenografts (n = 5 per group) treated with BAY876 plus Era or RSL3: treatment schedule (a), tumor volume (b), and tumor weight (c). d-f. Fn-infected 231 orthotopic mammary fat-pad tumors (1 × 10⁶ cells; n = 5 per group) treated with the same regimens: treatment schedule (d), tumor volume (e), and tumor weight (f). BAY876 (3 mg/kg), Era (10 mg/kg), and RSL3 (5 mg/kg) were administered intraperitoneally every 2 days. g, h. Representative tumor IHC images for SLC7A11 (g) and GPX4 (h). Scale bars, 50 μm. Tumor data are mean ± SEM. Two-way ANOVA was used in b and e; and one-way ANOVA in c and f. No inferential statistical test was applied to the treatment schedules in a and d or the representative IHC images in g and h. *P < 0.05, **P < 0.01, and ***P < 0.001.

We additionally evaluated 2-DG-based combinations in Fn-infected tail-vein metastasis models using survival as the endpoint. In the E109 model, 2-DG+Era prolonged survival relative to Fn alone, whereas the difference between 2-DG+RSL3 and Fn alone did not reach statistical significance. In the 231 model, both 2-DG+Era and 2-DG+RSL3 prolonged survival relative to Fn alone. However, 2-DG+RSL3 did not significantly outperform 2-DG+Era in either model (Fig. S15). Endpoint complete blood counts showed no consistent combination-specific changes across the measured leukocyte, erythrocyte, and platelet parameters (Table S7). Representative H&E-stained sections of the liver, spleen, lung, and kidney from the corresponding E109 and 231 cohorts showed no overt treatment-associated histopathological abnormalities (Fig. S16). These endpoint observations provide only a preliminary tolerability assessment. Serum liver- and kidney-function markers and long-term or repeated-dose toxicity were not evaluated. Collectively, the treatment data demonstrate activity of glucose-targeting/ferroptosis-inducer combinations within the tested Fn-infected models.

Discussion

Although bacterial modulation of ferroptosis has recently gained attention, how intracellular pathogens reshape ferroptosis together with nutrient-dependent disulfide stress remains unclear. Here, we found that intratumoral Fn abundance was positively associated with SLC7A11 expression in human tumor tissues. SLC7A11 supports ferroptosis resistance through cystine uptake, GSH synthesis, and GPX4 activity, whereas under glucose limitation continued cystine import can promote NADPH depletion and cytoskeletal disulfide stress in SLC7A11^high cells. Fn-associated SLC7A11 upregulation was therefore linked to a double-edged phenotype: ferroptosis resistance under nutrient-replete conditions and increased sensitivity to glucose restriction accompanied by disulfide and GPX4-directed lipid-peroxide stress. Recent studies have identified divergent effects of specific gut bacteria on ferroptosis and tumor progression. For example, Peptostreptococcus anaerobius produces a tryptophan metabolite that suppresses ferroptosis and promotes tumor development, ²⁵ whereas Clostridium butyricum or its metabolite butyrate can increase ferroptosis sensitivity in mouse models of pancreatic ductal adenocarcinoma. ²⁶ These findings highlight the species-specific effects of tumor-associated or gut-derived bacteria on metabolic cell death.

Fn can produce substantial amounts of H₂S, ²⁷ and our previous work showed elevated intracellular H₂S signals in phagocytes after Fn infection. ¹¹ In the present study, Fn infection and NaHS treatment were each associated with increased SLC7A11 expression and a downstream GSH/GPX4 response. H₂S signaling has also been linked to ferroptosis suppression through the xCT/GSH/GPX4 pathway, ²⁸ and the H₂S donor GYY4137 can attenuate ferroptosis by increasing SLC7A11 and GPX4. ²⁹ Bacterially derived H₂S may also contribute to oxidative-stress defense. ³⁰ Together, these observations are consistent with H₂S contributing to host-cell redox remodeling during Fn infection. However, because bacterial H₂S production was not genetically disrupted and complemented, they do not establish a specific causal relationship. The present data do not establish that SLC7A11 inhibition will be ineffective in Fn-positive tumors; rather, they suggest that incomplete suppression of the Fn-associated SLC7A11/GSH/GPX4 response may limit ferroptosis induction under the tested conditions.

Our donor-based and pharmacological data are consistent with a contributory role for H₂S in the Fn-associated SLC7A11/GSH/GPX4 response, but they do not establish a specific causal relationship or a direct H₂S-SLC7A11 molecular mechanism. Fn infection increased SLC7A11 mRNA and protein expression, and NaHS reproduced SLC7A11 protein induction, suggesting that transcriptional or post-transcriptional regulation may contribute. However, the present experiments cannot distinguish promoter activation from altered mRNA stability, protein stabilization, or persulfidation of SLC7A11 or an upstream regulator. Quantitative persulfidomics, targeted tag-switch assays, genetic disruption and complementation of bacterial H₂S production, and site-directed mutational validation will be required to determine whether H₂S directly modifies SLC7A11 or another pathway component. ^{31 , 32}

Fn infection was also accompanied by ACSL4 downregulation, and ACSL4 overexpression shifted the tested readouts toward greater ferroptotic stress, supporting a partial contributory role for this change. Separately, Fn infection reduced Era-induced Fe²⁺ accumulation and increased host-cell FTH1 expression. FTH1 oxidizes and stores iron and can limit the labile iron pool. ^{33 , 34} Fn also expresses the bacterial ferritin-like protein Fn-Dps, which supports intracellular bacterial survival. ⁵ Our functional data indicate that FTH1 contributes to Fn-associated ferroptosis resistance, but Fe²⁺ was not reassessed after FTH1 silencing, and the contribution of Fn-Dps was not tested. A cooperative effect of host FTH1 and bacterial Fn-Dps therefore remains a hypothesis requiring direct validation.

These findings define a nutrient-dependent, double-edged consequence of Fn-associated SLC7A11 upregulation. When glucose-dependent reducing capacity is sufficient, increased cystine uptake can support GSH synthesis and GPX4-dependent lipid-peroxide detoxification. In contrast, SLC7A11^high tumors are highly dependent on glucose and sensitive to glucose restriction. ³⁵ Under glucose deprivation, continued cystine import consumes NADPH and can promote disulfide accumulation and F-actin collapse, features implicated in canonical disulfidptosis. ^{19 , 36} In the present study, we therefore interpret the glucose-restriction phenotype as disulfide-stress-associated rather than as definitive evidence of canonical disulfidptosis. One plausible spatial model is that these opposing effects vary among tumor regions with different nutrient availability; however, intratumoral glucose, Fn, and SLC7A11 were not spatially mapped in this study. This interpretation should therefore be considered a working model rather than a directly demonstrated spatial mechanism.

Fn-targeted treatment has been proposed as an anticancer strategy. ^{6 , 37} Although several antibiotics, including metronidazole, β-lactams, and doxycycline, can reduce or eradicate Fn, antibiotic resistance and microbiota disruption warrant caution. ⁶ Combined modulation of ferroptosis and disulfide stress has also emerged as a potential antitumor approach, and GPX4 inhibitors have shown therapeutic promise in several cancers. ^{38 , 39} A recent study reported that RSL3 overcomes oxaliplatin resistance in Fn-infected CRC, whereas our data indicate that Fn infection weakens RSL3-induced ferroptosis under nutrient-replete conditions. These context-dependent findings underscore the need to evaluate glucose-restriction/GPX4 combinations using matched Fn-positive and Fn-negative models. Because CRC is the most extensively studied Fn-associated cancer, determining whether the same SLC7A11-associated metabolic state and treatment response occur in Fn-positive CRC is an important direction for future work.

Rapid tumor growth requires an abundant glucose supply, motivating interest in glucose-restriction strategies. The opposing effects of Era and RSL3 in glucose-restricted, Fn-infected cells can be understood from their distinct positions within the SLC7A11/GSH/GPX4 pathway. Although both compounds are widely used as ferroptosis inducers, Era inhibits SLC7A11-mediated cystine transport upstream. ¹⁶ Under glucose-replete conditions, this limits cysteine availability for GSH synthesis and promotes ferroptosis. Under glucose restriction, however, limiting cystine entry also reduces the amount of imported cystine that must be reduced to cysteine at the expense of NADPH. Era can therefore partially relieve the NADPH burden and disulfide stress implicated in canonical disulfidptosis in SLC7A11^high cells. ^{19 , 35} This context-dependent model provides a plausible explanation for why Era partially preserved viability and F-actin architecture rather than cooperating with glucose restriction in Fn-infected cells. Consistent with this interpretation, a previous study found that Era reduced cell viability in high-glucose medium but not under glucose-free conditions, whereas glucose deprivation did not markedly alter RSL3-induced ferroptosis. ⁴⁰

RSL3 acts downstream by inhibiting GPX4 and does not interrupt SLC7A11-mediated cystine uptake. ¹⁸ In this model, Fn-infected SLC7A11high cells continue to import cystine and consume NADPH when glucose-dependent NADPH regeneration is impaired, while GPX4 inhibition simultaneously removes a major defense against phospholipid-peroxide stress. Rather than relieving the upstream disulfide burden, RSL3 may allow cystine-driven NADPH depletion and cytoskeletal disulfide stress to occur together with GPX4-directed lipid-peroxide stress. ^{19 , 35} The increased NADP⁺/NADPH ratio, F-actin collapse, reduction-sensitive β-actin-containing complexes, and failure of Fer-1, Z-VAD-fmk, or Nec-1 to restore viability support a composite phenotype with a prominent disulfide-stress component rather than canonical ferroptosis alone. Because the SLC7A11 dependence of this glucose-restriction phenotype was not directly tested and the full defining criteria for canonical disulfidptosis were not established, we interpret it as a disulfide-stress-associated component rather than definitive disulfidptosis. This distinction offers a hypothesis for the treatment pattern observed with BAY876 plus RSL3 relative to BAY876 plus Era within the tested Fn-infected models; it does not establish Fn-selective efficacy or a synergistic interaction.

The different in vivo outcomes obtained with BAY876 and 2-DG should nevertheless be interpreted cautiously. BAY876 inhibits GLUT1-dependent glucose transport, ⁴¹ whereas 2-DG is a glucose analog whose activity depends on cellular uptake, intracellular phosphorylation, and competition with circulating glucose. ⁴² Their effects may therefore differ according to systemic exposure, dosing interval, toxicity constraints, transporter expression, and compensatory use of alternative substrates. ⁴³ More importantly, the experiments were not designed as a head-to-head comparison: BAY876 was administered every 2 days in established subcutaneous or orthotopic tumors, with tumor growth and weight as endpoints, whereas 2-DG was administered once weekly in tail-vein metastasis models, with survival as the endpoint. The 2-DG experiment also lacked a 2-DG-only group. Accordingly, these data cannot establish that BAY876 is intrinsically superior to 2-DG or that the absence of a relative advantage for 2-DG+RSL3 contradicts glucose dependence. Rather, they indicate that therapeutic efficacy is sensitive to the mechanism, duration, and anatomical context of glucose restriction. A definitive comparison will require BAY876 and 2-DG to be evaluated within the same tumor model and dosing framework, together with pharmacokinetic and pharmacodynamic measurements of intratumoral glucose uptake, NADP⁺/NADPH balance, cystine accumulation, cytoskeletal disulfide formation, and lipid peroxidation.

Overall, our findings indicate that Fn infection promotes ferroptosis resistance in tumor cells and that live-versus-killed, donor-based, and pharmacological evidence implicates H₂S in the associated SLC7A11/GSH/GPX4 response. Fn infection is also associated with reduced lipid-peroxide and Fe²⁺ accumulation, and functional perturbation supports partial and complementary contributions of ACSL4 downregulation and FTH1 upregulation. However, the Fn-associated SLC7A11^high state is linked to a nutrient-dependent liability because sustained cystine uptake becomes cytotoxic when glucose-dependent NADPH regeneration is restricted. Within the tested Fn-infected models, combined GLUT1 and GPX4 inhibition was associated with enhanced antitumor activity and a composite redox cell-death phenotype with a prominent disulfide-stress component. These findings provide a rationale for targeting the metabolic state associated with intratumoral Fn, while further studies are required to determine whether and to what extent canonical disulfidptosis contributes to the composite phenotype, to define the relative contributions of disulfide and lipid-peroxide stress, and to identify the most effective glucose-targeting regimens across Fn-positive tumor contexts.

Materials and methods

Clinical samples

A total of 17 esophageal squamous cell carcinoma (ESCC) and 14 breast cancer (BRCA) tissue samples were obtained from Sun Yat-sen University Cancer Center (SYSUCC). Nine samples-four ESCC and five BRCA samples-were excluded from subsequent analyzes because of unsuccessful DNA extraction or undetectable Fn levels. The specimens were residual tissue samples collected following routine clinical examination at SYSUCC between November 2019 and March 2024. Patients who had received prolonged antibiotic treatment before sample collection were excluded. Written informed consent for participation in the study and the research use of residual tissue specimens was obtained from all participants. All participants were adults; therefore, parental or guardian consent and minor assent were not applicable. The study protocol, including the collection and analysis of residual clinical tissue specimens, was reviewed and approved by the Ethics Committee of Sun Yat-sen University Cancer Center, Sun Yat-sen University (approval No. SL-B2024-738-01). The study was conducted in accordance with the approved protocol and the Declaration of Helsinki. Detailed patient characteristics are provided in Table S1.

Strains and growth conditions

Fusobacterium nucleatum (Fn; strain ATCC 25586; American Type Culture Collection, Manassas, VA, USA) was cultured on blood agar plates (Huankai Microbial, Guangzhou, China; Cat. No. CP0160) for 48-72 h at 37 °C under anaerobic conditions generated using an AnaeroPack system (Mitsubishi Gas Chemical Co., Tokyo, Japan; product type AnaeroPack-Anaero 2.5 L; Cat. No. C-01). For cell-infection experiments, Fn was added at a multiplicity of infection (MOI) of 10:1 and incubated with tumor cells for 24 or 48 h, as indicated for each experiment. Heat-killed Fn (K-Fn) was prepared by incubating live bacteria at 100 °C for 30 min.

Cell culture and compounds

The human ESCC cell line E109 (CCTCC, Cat. No. GDC0207), kyse30 (a gift from Professor Musheng Zheng, SYSUCC, Guangzhou, China), the human breast cancer cell line 231 (ATCC, Cat. No. HTB-26), and the human pancreatic cancer cell line PANC-1 (ATCC, Cat. No. CRL-1469) were maintained at 37 °C in a humidified atmosphere containing 5% CO₂. E109 and kyse30 cells were cultured in RPMI 1640 medium, whereas 231 and PANC-1 cells were cultured in DMEM medium. RPMI 1640 medium (Gibco; Cat. No. A1049101) and DMEM (Gibco; Cat. No. C11995500BT) were supplemented with 10% fetal bovine serum (Gibco; Cat. No. 10099141 C).

Erastin (Era; Selleck Chemicals, Cat. No. S7242), RSL3 (Selleck Chemicals, Cat. No. S8155), Fer-1 (Selleck Chemicals, Cat. No. S7243), and BAY876(Selleck Chemicals, Cat. No. S8452) were obtained from Selleck Chemicals. 2-Deoxy-D-glucose (2-DG; MedChemExpress, Cat. No. HY-13966), Z-VAD-fmk (MedChemExpress, Cat. No. HY-16658B), and necrostatin-1 (Nec-1; MedChemExpress, Cat. No. HY-15760) were obtained from MedChemExpress. Gallic acid (Ga; Solarbio, Cat. No. G8720) and sodium hydrosulfide (NaHS; Adamas, Cat. No. 40878 A; CAS No. 16721-80-5) were obtained from the indicated suppliers.

Stock solutions were prepared in DMSO for Era, RSL3, Fer-1, BAY876, Z-VAD-fmk, Nec-1; ddH₂O for 2-DG, Ga, NaHS, stored at -20 °C, and diluted in culture medium immediately before use. The final concentration of the solvent was maintained at 0.1% in all groups, including vehicle controls.

For experiments conducted in glucose-containing medium, the final Era and RSL3 concentrations were, respectively, 30 and 10 μM for E109 cells, 20 and 0.05 μM for kyse30 cells, 10 and 2 μM for 231 cells, and 10 and 0.05 μM for PANC-1 cells. For glucose-free or BAY876 combination experiments, the corresponding Era and RSL3 concentrations were 10 and 5 μM for E109 cells, 10 and 0.05 μM for kyse30 cells, 10 and 0.2 μM for 231 cells, and 10 and 0.05 μM for PANC-1 cells.

Unless otherwise specified, BAY876was used at 5 μM and added 3 h before Era or RSL3. Fer-1 was used at 1 μM and added 4 h before the ferroptosis inducer. Z-VAD-fmk and Nec-1 were each used at 20 μM and added 4 h before BAY876and RSL3 treatment. Ga was used at 2 mM, and NaHS was used at 100 μM for 48 h, except in dose-response experiments.

qPCR analysis

Total RNA was extracted from 1 × 10⁶ tumor cells using 1 mL AG RNAex Pro Reagent (Accurate Biology, Hunan, China; Cat. No. AG21102) per sample. cDNA was synthesized from 1 μg total RNA in a 20 μL reaction using the Evo M-MLV Reverse Transcription Kit (Accurate Biology; Cat. No. AG11711). Genomic DNA was extracted from 5 μm-thick paraffin-embedded ESCC and BRCA tissue sections using the TIANamp FFPE DNA Kit (TIANGEN, Beijing, China; Cat. No. DP331). qPCR was performed using the SYBR Green Pro Taq HS qPCR Kit (Accurate Biology; Cat. No. AG11701) on a CFX96 Real-Time PCR Detection System (Bio-Rad). Each 20 μL reaction contained 10 μL 2 × qPCR master mix, 2 μL template DNA or cDNA, and 0.2 μM each forward and reverse primer. Primer sequences are provided in Table S2.

siRNA transfection and plasmid overexpression

Small interfering RNAs targeting human SLC7A11 and FTH1 and the corresponding nontargeting control were custom synthesized by Sangon Biotech; therefore, commercial catalog numbers were not applicable. The sequences and, where available, synthesis order numbers are provided in Table S3.

The ACSL4 overexpression construct was generated by cloning the full-length human ACSL4 coding sequence into pcDNA3.1 (Invitrogen; Cat. No. V79020). Cells cultured in 6-well plates were transfected with siRNA at a final concentration of 50 nM or with 2 μg plasmid DNA per well using 4 μL Lipofectamine 3000 (Invitrogen; Cat. No. L3000015) per well. Knockdown or overexpression was evaluated by western blotting 48 h after transfection, and the indicated experimental treatments were initiated 24 h after transfection.

Cell viability and pharmacological rescue assays

Cells were seeded in 96-well plates at 5 × 10³ cells in 100 μL medium per well and subjected to the indicated treatments. Cytotoxicity was measured using an LDH Cytotoxicity Assay Kit (Beyotime Biotechnology, Shanghai, China; Cat. No. C0017). 50 μL culture supernatant and 50 μL reaction reagent were used for each measurement according to the manufacturer’s protocol. Absorbance was measured at 490 nm using an iMark microplate reader (Bio-Rad) after correction for background and maximum-LDH-release controls. Cell viability was calculated as 100% minus the cytotoxicity percentage.

For pharmacological rescue experiments, cells were pretreated with Fer-1 (1 μM), Z-VAD-fmk (20 μM), or Nec-1 (20 μM) for 4 h before treatment with BAY876and RSL3 under glucose-free conditions. Cell viability was measured after an additional 24 h.

RNA-sequencing analysis

RNA sequencing and bioinformatic analysis were performed by Biomarker Technologies Corporation (Beijing, China), as mentioned previously. ¹⁰ Briefly, kyse30 and kyse150 cells were infected with Fn (MOI = 10:1) for 48 h. Total RNA (1 μg per sample) was used to prepare sequencing libraries with the NEBNext Ultra RNA Library Prep Kit for Illumina (New England Biolabs; Cat. No. E7530L) according to the manufacturer's recommendations. Differential expression was analyzed using DESeq2. Genes with a fold change ≥ 1.5 and an adjusted P < 0.01 were considered significantly differentially expressed.

Data sources and bioinformatic analysis

Transcriptomic and clinical data for the TCGA-BRCA, TCGA-ESCA, and TCGA-PAAD cohorts were downloaded from UCSC Xena (http://xena.ucsc.edu/). Genus-level Fusobacterium relative-abundance data for ESCA were obtained from the Tumor Microbiome Cancer Atlas (TCMA; https://tcma.pratt.duke.edu/). Data processing and statistical analyzes were performed using R version 4.3.1. Differences in SLC7A11 expression between tumor and normal-adjacent tissues were assessed using the Wilcoxon test. For each TCGA cohort and survival endpoint, patients with available expression and corresponding survival data were dichotomized into SLC7A11-high and SLC7A11-low groups using an optimal cutoff determined by maximally selected rank statistics implemented in the surv_cutpoint function of the survminer R package. Kaplan-Meier curves were generated, and survival differences were evaluated using two-sided log-rank tests. Hazard ratios and corresponding 95% confidence intervals for high versus low SLC7A11 expression were estimated using univariable Cox proportional-hazards regression implemented in the survival R package (version 3.5–5).

Metabolomic analysis

At least 1 × 10⁶ cells per sample were lysed by three freeze–thaw cycles in 300 μL methanol/acetonitrile/water solution (2:2:1, v/v/v). LC–MS-grade methanol (Fisher Chemical; Cat. No. A456-4), acetonitrile (Fisher Chemical; Cat. No. A998-4), and water (Fisher Chemical; Cat. No. W6-4) were used. After centrifugation at 12000 × g, 15 min, 4 °C, the supernatant was vacuum-dried for 4 h using a refrigerated concentrator. The dried residue was reconstituted in 100 μL acetonitrile/water (1:1, v/v).

Metabolomic profiling was performed using ultra-high-performance liquid chromatography coupled to a TSQ Quantum Access MAX triple-quadrupole mass spectrometer (Thermo Fisher Scientific). Data processing was conducted by One-Map Biotech (Dalian, China), and features were retained according to the 80% rule.

Immunofluorescence

Double immunofluorescence (IF) staining was performed as follows. Cells were fixed with 4% paraformaldehyde (Beyotime Biotechnology Cat. No. P0099) for 30 min. Subsequently, cells or tissue sections were permeabilized with 0.3% Triton X-100 (Biofroxx; Cat. No. 1139ML500) for 30 min at room temperature and were blocked with ready-to-use goat serum (Boster Biological Technology; Cat. No. AR0009) for 45 min at 37 °C. Samples were incubated overnight at 4 °C with anti-SLC7A11/xCT antibody (1:1,000; ABclonal, Cat. No. A13685), anti-α-tubulin antibody (1:200; Ray Antibody, Cat. No. RM2007), or an in-house-generated rabbit or mouse anti-Fn antibody (1:1,000; no commercial catalog number), prepared as previously described. Species-appropriate secondary antibodies (1:200; Fudebio, Cat. Nos. FD0150, FD0129, and FD0136) were applied for 2 h at room temperature. Nuclei were stained with DAPI at 1 × working concentration (Fudebio, Cat. No. FD7229). Images were acquired using an FV3000 laser-scanning confocal microscope (FV3000, Olympus, Japan).

Immunohistochemistry

Antigen retrieval was performed by heating tissue sections in citrate antigen-retrieval solution (Boster; Cat. No. AR0024) at 98 °C for 20 min. Endogenous peroxidase activity was blocked with 3% hydrogen peroxide (Boster Biological Technology; Cat. No. AR1108) for 10 min. Sections were incubated overnight at 4 °C with anti-SLC7A11 antibody (1:1,000; Abcam, Cat. No. ab307601) or anti-GPX4 antibody (1:500; Servicebio, Cat. No. GB120010). HRP-conjugated secondary antibodies (1:200; Servicebio, Cat. Nos. GB23303 and GB23301) were applied for 2 h at room temperature. Signals were developed for 3-5 min using a DAB substrate kit (CWBIO; Cat. No. CW0125S).

Western blotting

Total cell proteins were extracted using RIPA lysis buffer (Beyotime, Cat. No. P0013B). Protein concentrations were determined using BCA Protein Assay Kit (Beyotime, Cat. No. P0012) and 30 μg protein per lane was separated by 10% SDS-PAGE. Proteins were transferred to 0.22 μm PVDF membranes (Bio-Rad; Cat. No. 1620177). Membranes were blocked with 5% skim milk (Yili) for 1 h at room temperature.

Membranes were incubated overnight at 4 °C with antibodies against SLC7A11/xCT (1:1,000; ABclonal, Cat. No. A13685), GPX4 (1:1,000; Abcam, Cat. No. ab125066), ACSL4 (1:10,000; Abcam, Cat. No. ab155282), FTH1 (1:500; ABclonal, Cat. No. A1144), or GAPDH (1:5,000; Bioworld, Cat. No. AP0063). HRP-conjugated goat anti-rabbit IgG (1:5,000; Fdbio Science, Cat. No. FDR007) was applied for 1 h at room temperature. Protein bands were visualized using an ECL chemiluminescence reagent (NCM Biotech; Cat. No. P10100) and quantified using ImageJ (Java 1.6.0_20, NIH, USA).

Reducing and nonreducing western blotting

Cell lysates were prepared in 200 μL RIPA buffer (Beyotime; Cat. No. P0013B) without DTT for nonreducing conditions or containing 100 mM DTT (Solarbio; Cat. No. D8220) for reducing conditions. Samples were mixed with 5 × SDS loading buffer (Beyotime; Cat. No. P0015L) at a sample-to-buffer ratio of 4:1. Reducing samples were boiled for 10 min, whereas nonreducing samples were incubated at 37 °C for 30 min. 30 μg protein per lane was separated by 8% SDS-PAGE and immunoblotted using an anti-β-actin antibody (1:5,000; Bioworld, Cat. No. AP0060).

Fluorescent staining

F-actin was stained using the TraKine F-actin Staining Kit (Abbkine Scientific, Wuhan, China; Cat. No. KTC4009). Cells were fixed with 4% paraformaldehyde for 30 min, permeabilized with 0.1% Triton X-100 for 10 min, and incubated with 1 × phalloidin staining solution for 30 min at room temperature in the dark.

Intracellular H₂S and Fe²⁺ were detected using WSP-5 (Maokang Biotechnology, Shanghai, China; Cat. No. MX5302) and the FerroOrange Assay Kit (Dojindo Laboratories; Cat. No. F374), respectively. Live cells were incubated with 50 μM WSP-5 or 1 μM FerroOrange for 30 min at 37 °C in the dark.The cells were visualized using a laser scanning confocal microscope (FV3000, Olympus, Japan).

Flow-cytometric analysis

Intracellular reactive oxygen species and lipid peroxidation were detected using DCFH-DA (Beyotime Biotechnology; Cat. No. S0033M) at 10 μM and C11-BODIPY 581/591 (Invitrogen; Cat. No. D3861) at 2 μM, respectively. Cells were incubated with the probes for 1 h at 37 °C in the dark.

Intracellular H₂S was measured by incubating cells with 50 μM WSP-5 for 30 min at 37 °C. Glucose uptake was assessed using 50 μM 2-NBDG (APExBIO; Cat. No. B6035) in glucose-free medium for 30 min at 37 °C. Samples were analyzed using a CytoFLEX S flow cytometer (Beckman Colter), and at least 10,000 cellular events were recorded per sample. Data were analyzed using FlowJo version 10.0.

Biochemical assays

Total NADP (NADPH+NADP⁺) and NADPH were quantified using an NADP⁺/NADPH Assay Kit with WST-8 (Beyotime Biotechnology; Cat. No. S0179) using 1 × 10⁶ cells per sample. Absorbance was measured at 450 nm, and NADP⁺ was calculated by subtracting NADPH from total NADP.

Total glutathione was measured using the Total Glutathione Assay Kit (Beyotime Biotechnology; Cat. No. S0052) using 1 × 10⁶ cells per sample. After the NADPH solution was added, absorbance at 405 nm was measured every 5 min for 25 min.

Cystine uptake was measured using the Cystine Uptake Assay Kit (Dojindo Laboratories; Cat. No. UP05) using 5 × 10⁴ cells per well. Fluorescence was measured at excitation/emission wavelengths of 490/535 nm using a Synergy H1 microplate reader (BioTek Instruments).

Intracellular survival assays

Cells were infected with Fn at an MOI of 10:1 for 24 h and then treated with Era at the cell-specific concentrations described (Table S4) above for an additional 24 h. Cells were washed with PBS (Beyotime; Cat. No. P1020) and incubated for 2 h in RPMI 1640 or DMEM containing 50 μg/mL gentamicin (Solarbio; Cat. No. G8170) to eliminate extracellular bacteria. Cells were lysed in 500 μL PBS containing 0.1% Triton X-100 (Biofroxx; Cat. No. 1139ML500) and 0.01% SDS (Sigma-Aldrich; Cat. No. L3771). Serial dilutions were spread on blood agar plates, and intracellular bacterial loads were determined from colony counts obtained from three replicate measurements.

Transmission electron microscopy

Mitochondrial morphology was examined by transmission electron microscopy (TEM). Briefly, kyse30 and PANC-1 cells subjected to the indicated treatments were initially fixed by adding 2.5% glutaraldehyde solution (Servicebio; Cat. No. G1102) for 5 min in the dark. After low-speed centrifugation, the cell pellets were collected and further fixed in fresh glutaraldehyde solution for 30 min at room temperature in the dark. The fixed samples were then stored at 4 °C. TEM imaging was performed using a Hitachi HT7800 TEM system (Tokyo, Japan).

Animal experiments

BALB/c nude mice (4-5 weeks old) were obtained from the Nanjing Biomedical Research Institute of Nanjing University (Nanjing, China). Male mice were exclusively used for E109 esophageal squamous cell carcinoma subcutaneous xenograft models, while female mice were used for 231 orthotopic mammary fat pad tumor models. Mice were housed under specific pathogen-free (SPF) conditions in individually ventilated cages (IVC), with 4 mice per cage, maintained at 22 ± 2 °C and 50 ± 10% relative humidity under a strict 12-h light/12-h dark diurnal cycle. Sterilized standard rodent chow and autoclaved purified water were provided ad libitum. All mice underwent a 7-day acclimatization period after delivery before initiation of any experimental manipulations. Environmental enrichment consisted of sterile cotton nesting material and plastic gnawing sticks inside each cage. Trained, certified animal care staff performed daily visual health and welfare monitoring of all mice throughout the entire experimental period.

The individual mouse was defined as the experimental unit for all in vivo assays. The Fn in vivo organ localization experiment contained 5 mice per group; subcutaneous E109 and orthotopic 231 tumor growth therapeutic experiments were assigned 5 mice per group; tail vein intravenous metastasis survival cohorts utilized 8 mice per group. All sample sizes were predetermined based on previously published tumor microbiology and ferroptosis in vivo studies combined with preliminary pilot xenograft data, with power analysis to ensure ≥0.8 statistical power for detecting medium-to-large treatment effect sizes, balancing experimental variability and the minimal number of animals required to generate statistically interpretable tumor growth and survival outcomes. Prespecified animal exclusion criteria included complete failure of tumor engraftment, severe procedural complications following intravenous/orthotopic injection, and sustained >20% body weight loss within 72 h of manipulation.

To visualize systemic and intratumoral localization of Fn in vivo, male BALB/c nude mice were subcutaneously inoculated into the right flank with 5 × 10⁶ E109 cells suspended in 100 μL ice-cold sterile PBS. Once subcutaneous tumors reached a target volume of approximately 60 mm³, 1 × 10⁶ CFU DiD-labeled Fn (Beyotime Biotechnology, Cat. No. C1995S) was resuspended in 100 μL sterile PBS and administered via tail vein intravenous injection. Fourteen days after Fn injection, all mice were humanely euthanized via CO₂ inhalation followed by secondary cervical dislocation for confirmation of death; primary xenograft tumors and major visceral organs (liver, spleen, lung, kidney, heart) were harvested for ex vivo whole-organ fluorescence imaging using an IVIS Spectrum In Vivo Imaging System (PerkinElmer, Waltham, MA, USA).

For tumor growth and therapeutic intervention experiments, male BALB/c nude mice received subcutaneous flank inoculation of 5 × 10⁶ E109 cells in 100 μL sterile PBS; female BALB/c nude mice underwent orthotopic mammary fat pad injection of 1 × 10⁶ 231 cells suspended in 50 μL sterile PBS. Fn (1 × 10⁶ CFU/mouse in 100 μL PBS) was delivered via tail vein injection using two distinct administration schedules: two injections prior to tumor cell implantation, or three sequential injections after visible tumor establishment, as illustrated in schematic treatment timelines (Figures 4b,j, 6a,d and S11a-b).

When the average tumor volume across all cages reached 60-100 mm³, mice were allocated to experimental and vehicle control groups using a computer-generated random number sequence stratified by baseline tumor volume and body weight to guarantee balanced group baseline characteristics, with n = 5 mice per experimental arm. Erastin (Era, 10 mg/kg), RSL3 (5 mg/kg), and GLUT1 inhibitor BAY876 (3 mg/kg) were prepared according to the ratio of 5% DMSO+40% PEG300 + 5% Tween-80 + 50% ddH₂O and administered by intraperitoneal injection every 48 h; vehicle-only control mice received identical volumes of the matching solvent on the identical treatment schedule. Digital caliper measurements of tumor length (L) and width (W), alongside whole-body mouse weight, were recorded every 3 days throughout treatment. Tumor volume was calculated using the standard ellipsoid formula:(L × W²)/2. Blinding was not implemented during tumor caliper measurement and sample tissue collection due to distinct visual differences in tumor burden between treatment cohorts; however, all downstream quantitative tumor volume, weight and IHC data analyzes were completed by an independent researcher fully blinded to group identity.

For systemic metastasis survival analysis, male or female BALB/c nude mice received tail vein intravenous injection of 1 × 10⁶ E109 or 231 tumor cells (100 μL PBS vehicle), respectively. Mice were randomly assigned to treatment groups (n = 8 per group) using stratified computer randomization based on baseline body weight. Starting 14 days post-tumor cell inoculation, mice received weekly tail vein Fn injections (1 × 10⁶ CFU in 100 μL PBS) alongside weekly intraperitoneal therapy with 2-DG(1 g/kg), erastin (10 mg/kg), or RSL3 (5 mg/kg) as specified per experimental arm.

All animals were inspected twice daily (morning and evening) to monitor general physical condition, spontaneous mobility, grooming behavior, food/water intake, respiratory function, and tumor-associated adverse events. Predefined institutional humane endpoint criteria were strictly enforced: maximum single tumor volume exceeding 1200 mm³, sustained ≥20% loss of initial body weight, visible tumor ulceration/necrosis, impaired ambulation, persistent hunched posture, or severe respiratory distress. Any mouse meeting one or more humane endpoint criteria was immediately euthanized by CO₂ inhalation followed by cervical dislocation, and the event recorded as a survival endpoint. Survival duration was defined as the interval from tumor cell intravenous inoculation until natural death or mandatory euthanasia per humane endpoint rules.

All animal handling, injection, treatment and tissue harvest procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of Sun Yat-sen University (Guangzhou, China), under approved study protocols (Nos. SYSU-IACUC-2022-001227, 2023-001289, 2024-000385, 2024-001331, 2025-000525, 2026-002726).

Endpoint hematologic and histopathologic assessment

At the experimental endpoint, peripheral blood was collected from the retro-orbital venous plexus and analyzed using a BC-5000Vet automated veterinary hematology analyzer (Mindray). The liver, spleen, lungs, and kidneys were fixed in 4% paraformaldehyde (Beyotime Biotechnology; Cat. No. P0099) for 24 h, embedded in paraffin (Servicebio; Cat. No. G6004), sectioned at 4 μm, and stained using hematoxylin (Servicebio; Cat. No. G1004) and eosin (Servicebio; Cat. No. G1001).

Statistical and reproducibility

All statistical analyzes were conducted with GraphPad Prism 10 (GraphPad Software, CA, USA). Data from at least three independent in vitro experiments are expressed as the mean ± SD. For in vivo studies, tumor volume and tumor weight data are presented as the mean ± SEM. The normality of data distribution was assessed using the Shapiro-Wilk test. For comparisons between two groups of normally distributed data, an independent-samples t-test was used. For comparisons across multiple groups, one-way ANOVA was used. Non-normally distributed data were analyzed using the Mann-Whitney U test. A p-value of less than 0.05 was considered statistically significant.

Supplementary Material

Supplemental material_R2.pdf

Supplemental material_R2.pdf

Funding Statement

This work was supported by the National Natural Science Foundation of China (grant No. 82172890; 32300761) and the China Postdoctoral Science Foundation (grant No. 2025M772659).

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.

Data availability statement

The RNA-sequencing data have been deposited in the Gene Expression Omnibus under accession number GSE222917.

Supplementary material

Supplemental data for this article can be accessed at https://doi.org/10.1080/19490976.2026.2748782.