Metabolomic Analysis of Yunnan Cigar Tobacco Leaves: Impact of Geography and Climate on Flavor Characteristics and Machine Learning-Based Origin Traceability

Guijuan Zhao¹Yuping Wu²Yi Li¹Guifeng Li¹Wenyuan Wang³Lei Yang³Zhonglong Lin²Heng Yao²Guanghui Kong²Fangchan Jiao²Gaokun Zhao²Guanghai Zhang²Meiwei Zhao⁴Tao Zhang³Jin Wang^3*

• ¹Yunnan Minzu University, Kunming, China
• ²Yunnan Academy of Tobacco Agricultural Science, kunming, China
• ³China Tobacco Yunnan Industrial Corporation, Kunming, China
• ⁴Kunming University, Kunming, China

In order to explore how the geographical and climatic characteristics of Yunnan shape the unique metabolic profile of its cigar tobacco leaves (CTLs) and develop a reliable method for tracing the origin of CTLs using machine learning algorithms. 71 samples were collected from country scale Dominica, Indonesia and Yunnan, and prefecture scale Lincang, Puer, Yuxi in Yunnan. A non-targeted metabolomics approach was applied to investigate the metabolic differences, and 778 highly reliable metabolites were identified. Associated with high altitude, large day-night temperature differences, intense ultraviolet radiation, and relative dryness (drought), Yunnan CTLs exhibit distinct metabolic profiles. Specifically, pathways such as flavone and flavonol biosynthesis and betalain biosynthesis are significantly enriched. With higher contents of polyphenols, indoles, jasmonates, carotenoids and so on, contribute to Yunnan CTLs' distinctive woody, roasted, and astringent flavor profile. 12 key biomarkers were selected using Multivariable methods with unbiased variable selection in R (MUVR). Machine learning algorithms—including LDA, LR, GMM, KNN, and SVM—leveraged these biomarkers to achieve exceptional origin traceability across national (Yunnan vs. Dominica/Indonesia) and regional (Lincang, Pu'er, Yuxi) scales. Validation demonstrated with a median of 100 false classification rates about 0.1 and the AUC close to 1, underscoring the model's high accuracy and robustness.