Unrevealing the Halyomorpha halys Damage Fingerprint on Hazelnut Metabolome by Multiomic Platforms and AI-Aided Strategies

The brown marmorated stink bug (Halyomorpha halys) poses a significant threat to hazelnut crops by affecting kernel development and causing quality defects, reducing the market value. While previous studies have identified bitter-tasting compounds in affected kernels, the impact of stink bug feeding on the hazelnut metabolome, particularly concerning aroma precursors, remains underexplored. This study aims to map the nonvolatile metabolome and volatilome of hazelnut samples obtained by caging H. halys on different cultivars in two locations to identify markers for diagnosing stink bug damage. Using a multiomic approach involving headspace solid-phase microextraction (HS-SPME), comprehensive two-dimensional gas chromatography-time-of-flight mass spectrometry (GC × GC-TOF MS), and liquid chromatography-high-resolution mass spectrometry (LC-HRMS), both raw and roasted hazelnuts are analyzed, with artificial intelligence (AI) and machine learning tools employed to explore data correlations. The study finds that the hazelnut metabolome and volatilome exhibit high chemical complexity with significant classes of compounds such as aldehydes, ketones, alcohols, and terpenes identified in both raw and roasted hazelnuts. Multivariate analysis indicates that the orchard location significantly impacts the metabolome, followed by damage type, with cultivar differences being less pronounced. Partial least-squares discriminant analysis (PLS-DA) models achieve high predictive accuracy for orchard location (99%) and damage type (≈80%), with the roasted volatilome showing the highest predictive accuracy. Correlation matrices reveal significant relationships between raw hazelnut metabolites and aroma compounds in roasted samples, suggesting potential markers for stink bug damage that could guide the quality assessment and mitigation strategies. Data fusion techniques further enhance classification performance, particularly in predicting damage type, underscoring the potential of integrating multiple data sets for comprehensive quality assessment.