Supplementary Table 3: Basic Clinical Information of Patients from External Test Cohort 2

Figure S1 Quality control and fundamental baseline analysis

Figure S7 Dynamic metabolite model construction and performance evaluation

The peritumoural adipose tissue (PAT) is a key contributor to cancer therapy resistance, yet its role in regulating ferroptosis remains unclear. Here we demonstrate that PAT confers ferroptosis resistance to cancer cells by upregulating ferritin (FTH1/FTL) and sequestering intracellular iron. PAT-derived kynurenine (KYN) was identified as the principal mediator. KYN is taken up by cancer cells and metabolized to 3-hydroxykynurenine, which directly binds to nuclear receptor coactivator 4 (NCOA4). This interaction inhibits NCOA4-mediated ferritinophagy, preventing ferritin degradation and limiting the free iron pool required for ferroptosis. In murine models, pharmacological inhibition of the KYN pathway synergized with PD-1 blockade to overcome ferroptosis resistance and suppress tumour progression. These findings reveal a PAT-KYN-ferritinophagy axis that promotes ferroptosis resistance, highlighting the potential of targeting adipose-tumour cross-talk to enhance immunotherapy in PAT-associated tumours.

Figure S9 Characterization and communication of NK cell subtypes in tumor and PBMC

Supplementary Table 4: Basic Clinical Information of Patients from External Test Cohort 3

Figure S8 Additional information of tumor response after SAC and I3C administration and safety assessment

Supplementary Table 1: Basic Clinical Information of Patients collected from the ESCORT-1st trial

Figure S10 Characterization CD8+ T cell subtypes in tumor and PBMC

Figure S5 Identification of optimal long PFS cutoff to distinguish long and short term responders and key metabolite differences and predictive performance of on-treatment model

Methionine metabolism generates the substrate S-adenosylmethionine (SAM), which regulates epigenetic modifications crucial for various cellular processes, particularly tumorigenesis. However, whether methionine metabolism involves epigenetic mechanisms independent of SAM and what roles such mechanisms play in tumorigenesis remain unclear. We show here that the adenosylhomocysteinase (AHCY)–adenosine complex increases mRNA m6A levels in a non-global manner, promoting fatty acid synthesis and tumorigenesis. Adenosine increases mRNA m6A levels by binding to the methionine metabolism enzyme AHCY to form a complex, rather than depending on adenosine receptors. The AHCY–adenosine complex facilitates AHCY dimerization, with adenosine being crucial for dimer stability. AHCY dimers hinder the binding of fat mass and obesity-associated protein (FTO) at the Q86 site to RNA containing the VWDRACH motif, increasing m6A levels and upregulating lipogenesis genes, especially ACACA and SCD1, thus leading to reprogramming of lipid metabolism. Conversely, AHCY mutants that have lost dimerization or FTO-binding ability but retain hydrolase activity suppress lipogenesis and tumor growth without significantly affecting methionine catabolism mediated by AHCY. Loss of AHCY in mice and disruption of AHCY dimerization in tumor cells and patient-derived xenograft models restricted tumor growth. Our findings demonstrate a key SAM-independent link between methionine metabolism and mRNA m6A modification that affects demethylase substrate specificity. This novel link between the methionine cycle and lipid metabolism suggests new strategies for anticancer therapy.

Supplementary Table 2: Basic Clinical Information of Patients from External Test Cohort 1

Figure S6 Time-dependent metabolic risk analysis and lipid subtype characterization

Esophageal squamous cell carcinoma (ESCC) exhibits heterogeneous responses to chemoimmunotherapy, with only a minority achieving durable benefit, necessitating dynamic precision monitoring. Through longitudinal plasma metabolomics of 541 serial samples from 252 ESCORT-1st trial patients receiving chemoimmunotherapy plus 3 independent cohorts of 288 samples, we established an integrated risk assessment framework spanning the entire therapeutic continuum: (i) a baseline predictor for initial responders based on metabolite signatures; (ii) an on-treatment predictor in prognosticating long-term responders among initial ones based on treatment-induced metabolic shift patterns; and (iii) a real-time model based on dual alteration of sphingolipid and glycerophospholipid dynamically stratifying progression risk. Meanwhile, 2 dietary metabolites, garlic-derived S-allyl-L-cysteine and cruciferous vegetable-derived indole-3-carbinol, were confirmed to improve outcomes by promoting NK-cell infiltration and reversing CD8+ T-cell exhaustion. In conclusion, we provide the first metabolomic roadmap for precision chemoimmunotherapy in ESCC, unifying baseline prediction, longitudinal surveillance, and dietary modulation into a clinically actionable paradigm. SIGNIFICANCE:We established a large-scale plasma metabolomic database from patients with ESCC undergoing chemoimmunotherapy, defining the first noninvasive, comprehensive, and precise monitoring framework for treatment response prediction and risk assessment. Our findings reveal clinically actionable dietary metabolites that may serve as readily accessible adjuvants to enhance chemoimmunotherapy efficacy.

Figure S4 TG pattern shifts from baseline to on-treatment

Figure S3 Baseline metabolite model performance validation and clinical prognostic analysis

Supplementary Table 6: Prediction Result of On-Treatment Metabolite Model for Long-Term Responder

Figure S2 Metabolic block construction and validation analysis

Supplementary Table 5: Prediction Result of Baseline Metabolite Model of Each Cohort