Huihui Xiang ^1,2 Rika Kasajima ^1,3 Koichi Azuma ⁴ Tomoyuki Tagami ⁵ Asami Hagiwara ⁵ Yoshiro Nakahara ^6,7 Haruhiro Saito ⁶ Yuka Igarashi ^8,9 Feifei Wei ^8,9 Tatsuma Ban ¹⁰ Mitsuyo Yoshihara ^1,11 Yoshiyasu Nakamura ^1,11 Shinya Sato ^1,11,2 Shiro Koizume ^12,2 Tomohiko Tamura ¹⁰ Tetsuro Sasada ^8,9 Yohei Miyagi ^12,2*

• ¹ Molecular Pathology & Genetics Division, Kanagawa Cancer Center Research Institute, Yokohama, Japan, Yokohama, Japan
• ² Department of Pathology, Kanagawa Cancer Center, Yokohama, Japan, Yokohama, Kanagawa, Japan
• ³ Center for Cancer Genome Medicine, Kanagawa Cancer Center, Yokohama, Japan, Yokohama, Japan
• ⁴ Department of Internal Medicine, Kurume University Medical School, Kurume, Fukuoka, Japan
• ⁵ Research Institute for Bioscience Products and Fine Chemicals, Ajinomoto Co., Inc., Kawasaki, Japan
• ⁶ Department of Thoracic Oncology, Kanagawa Cancer Center, Yokohama, Japan
• ⁷ Department of Respiratory Medicine, School of Medicine, Kitasato University, Sagamihara, Kanagawa, Japan
• ⁸ Division of Cancer Immunotherapy, Kanagawa Cancer Center Research Institute, Yokohama, Kanagawa, Japan
• ⁹ Cancer Vaccine and Immunotherapy Center, Kanagawa Cancer Center, Yokohama, Japan
• ¹⁰ Department of Immunology, Yokohama City University Graduate School of Medicine, Yokohama, Japan
• ¹¹ Morphological Analysis Laboratory, Kanagawa Cancer Center Research Institute, Yokohama, Japan
• ¹² Molecular Pathology & Genetics Division, Kanagawa Cancer Center Research Institute, Yokohama, Japan

Background: Studies have shown that tumor cell amino acid metabolism is closely associated with lung adenocarcinoma (LUAD) development and progression. However, the comprehensive multiomics features and clinical impact of the expression of genes associated with amino acid metabolism in the LUAD tumor microenvironment (TME) are yet to be fully understood. Methods: LUAD patients from The Cancer Genome Atlas (TCGA) database were enrolled in the training cohort. Using least absolute shrinkage and selection operator Cox regression analysis, we developed PTAAMG-Sig, a signature based on the expression of tumor-specific amino acid metabolism genes associated with overall survival (OS) prognosis. We evaluated its predictive performance for OS and thoroughly explored the effects of the PTAAMG-Sig risk score on the TME. The risk score was validated in two Gene Expression Omnibus (GEO) cohorts and further investigated against an original cohort of chemotherapy combined with immune checkpoint inhibitors (ICIs). Somatic mutation, chemotherapy response, immunotherapy response, gene set variation, gene set enrichment, immune infiltration, and plasma-free amino acids (PFAAs) profile analyses were performed to identify the underlying multi-omics features. Results: TCGA datasets based PTAAMG-Sig model consisting of nine genes, KYNU, PSPH, PPAT, MIF, GCLC, ACAD8, TYRP1, ALDH2, and HDC, could effectively stratify the OS in LUAD patients. The two other GEO-independent datasets validated the robust predictive power of PTAAMG-Sig. Our differential analysis of somatic mutations in the high-and low-risk groups in TCGA cohort showed that the TP53 mutation rate was significantly higher in the high-risk group and negatively correlated with OS. Prediction from transcriptome data raised the possibility that PTAAMG-Sig could predict the response to chemotherapy and ICIs therapy. Our immunotherapy cohort confirmed the predictive ability of PTAAMG-Sig in the clinical response to ICIs therapy, which correlated with the infiltration of immune cells (e.g., T lymphocytes and nature killer cells). Corresponding to the concentrations of PFAAs, we discovered that the high PTAAMG-Sig risk score patients showed a significantly lower concentration of plasma-free α-aminobutyric acid. Conclusions: In patients with LUAD, the PTAAMG-Sig effectively predicted OS, drug sensitivity, and immunotherapy outcomes. These findings are expected to provide new targets and strategies for personalized treatment of LUAD patients.