Untargeted LC-HRMS approaches combined with feature-based molecular networking to annotate reaction markers in processed foods

The rising consumption of ultra-processed foods (UPFs), linked to multiple health risks, underscores the need to characterise the chemical profile of formulated and processed goods to improve their quality and safety. This study aims to develop untargeted approaches based on LC-HRMS and LC-HRMS/MS, coupled with feature-based molecular networking (FBMN), to explore, for the first time, the chemical reactivity within a well-characterised UPF-like food matrix (sponge cake). Three controlled baking conditions were applied to the formulated cake to induce thermal reactivity and generate diverse chemical profiles. Principal component analysis and heatmap clustering of untargeted LC-HRMS data from cake extracts were able to effectively discriminate samples based on the thermal process intensity. Approximately 75% of the detected features were indeed involved in the reactivity. FBMN revealed different groups of compounds (including precursors and advanced products) associated with Maillard and caramelisation reactions, and helped annotation of several reaction markers. This is the first application of FBMN on widely consumed processed foods such as baked products, opening new perspectives for generating high-throughput untargeted data to annotate reaction markers from complex food matrices.