Cotton (Gossypium hirsutum) is an economically important crop and is widely cultivated around the globe. However, the major problem of cotton is its high vulnerability to biotic and abiotic stresses. It has been around three decades since the cotton plant was genetically engineered with genes encoding insecticidal proteins (mainly Cry proteins) with an aim to protect it against insect attack. Several studies have been reported on the impact of these genes on cotton production and fiber quality. However, the metabolites responsible for conferring resistance in genetically modified cotton need to be explored. The current work aims to unveil the key metabolites responsible for insect resistance in Bt cotton and also compare the conventional multivariate analysis methods with deep learning approaches to perform clustering analysis. We aim to unveil the marker compounds which are responsible for inducing insect resistance in cotton plants. For this purpose, we employed 1H-NMR spectroscopy to perform metabolite profiling of Bt and non-Bt cotton varieties, and a total of 42 different metabolites were identified in cotton plants. In cluster analysis, deep learning approaches (linear discriminant analysis (LDA) and neural networks) showed better separation among cotton varieties compared to conventional methods (principal component analysis (PCA) and orthogonal partial least square discriminant analysis (OPLSDA)). The key metabolites responsible for inter-class separation were terpinolene, α-ketoglutaric acid, aspartic acid, stigmasterol, fructose, maltose, arabinose, xylulose, cinnamic acid, malic acid, valine, nonanoic acid, citrulline, and shikimic acid. The metabolites which regulated differently with the level of significance p < 0.001 amongst different cotton varieties belonged to the tricarboxylic acid cycle (TCA), Shikimic acid, and phenylpropanoid pathways. Our analyses underscore a biosignature of metabolites that might involve in inducing insect resistance in Bt cotton. Moreover, novel evidence from our study could be used in the metabolic engineering of these biological pathways to improve the resilience of Bt cotton against insect/pest attacks. Lastly, our findings are also in complete support of employing deep machine learning algorithms as a useful tool in metabolomics studies.
Keywords: NMR; cotton; defense related metabolites; metabolomics; multivariate data analysis.