Integrative analysis of sensory evaluation and non-targeted metabolomics to unravel tobacco leaf metabolites associated with sensory quality of heated tobacco

Abstract

Introduction: Heated tobacco (Nicotiana tabacum L.) products are heating tobacco plug at a temperature of 350°C and produce different emissions in aerosol and sensory perceptions of tobacco leaf compared with combustible tobacco. Previous study assessed different tobacco varieties in heated tobacco for sensory quality and analyzed the links between sensory scores of the final products and certain chemical classes in tobacco leaf. However, contribution of individual metabolites to sensory quality of heated tobacco remains largely open for investigation.

Methods: In present study, five tobacco varieties were evaluated as heated tobacco for sensory quality by an expert panel and the volatile and non-volatile metabolites were analyzed by non-targeted metabolomics profiling.

Results: The five tobacco varieties had distinct sensory qualities and can be classified into higher and lower sensory rating classes. Principle component analysis and hierarchical cluster analysis showed that leaf volatile and non-volatile metabolome annotated were grouped and clustered by sensory ratings of heated tobacco. Orthogonal projections to latent structures discriminant analysis followed by variable importance in projection and fold-change analysis revealed 13 volatiles and 345 non-volatiles able to discriminate the tobacco varieties with higher and lower sensory ratings. Some compounds such as β-damascenone, scopoletin, chlorogenic acids, neochlorogenic acids, and flavonol glycosyl derivatives had strong contribution to the prediction of sensory quality of heated tobacco. Several lyso-phosphatidylcholine and lyso-phosphatidylethanolamine lipid species, and reducing and non-reducing sugar molecules were also positively related to sensory quality.

Discussion: Taken together, these discriminating volatile and non-volatile metabolites support the role of leaf metabolites in affecting the sensory quality of heated tobacco and provide new information on the types of leaf metabolites that can be used to predict applicability of tobacco varieties for heated tobacco products.

Keywords: discriminating metabolites; heated tobacco products; non-targeted metabolomics profiling; orthogonal projections to latent structures discriminant analysis; sensory evaluation.

Figure 1

Metabolite variation among five tobacco varieties. (A) PCA score plot for the 309 annotated volatile metabolites (R²X = 0.923, Q² = 0.783). (B) PCA score plot for the 1168 annotated non-volatile metabolites (R²X = 0.932, Q² = 0.871).

Figure 2

Hierarchical clustering of volatile (A) and non-volatile (B) profiles of tobacco leaf by variety. Volatile and non-volatile metabolite data were normalized within each variety via z-transformation. The resulting z-scores were converted into colors and grouped using hierarchical clustering. The color in each cell represents the z-transformed abundances of the averaged replicates (n = 3) per tobacco variety (red: high, blue: low).

Figure 3

OPLS-DA performed on (A) 309 volatile metabolites and (B) 1168 non-volatile metabolites identified in five tobacco varieties according to the higher and lower sensory ratings. The colors indicate the different levels of sensory ratings (red: higher sensory rating, blue: lower sensory rating).

Figure 4

Analysis of metabolite variation among five tobacco varieties within chemical classes. (A) Heatmap of 13 volatile metabolites including one alcohol, two aldehydes, two esters, five heterocyclic compounds, one ketone, one nitrogen compound and one terpenoid. (B) Heatmap of the top 50 discriminating non-volatile compounds ranked by t-test including three alkaloids, 12 flavonoids, three lignans and coumarins, six lipids, 17 phenolic acids, three tannins, two terpenoids, one quinone, one saccharide and two other non-volatile metabolites. Volatile and non-volatile metabolite data were normalized within each variety via z-transformation. The resulting z-scores were converted into colors and grouped using hierarchical clustering. The color in each cell represents the z-transformed abundances of the averaged replicates (n = 3) per tobacco variety (red: high, blue: low).