Soft-Sensing-Guided Glucose/l-Phenylalanine Cofeeding Strategy and Transcriptomic Analysis of the Optimal l-Phenylalanine Supply Range for Efficient β-Phenylethanol Biosynthesis in Saccharomyces cerevisiae.

β-Phenylethanol (2-PE) is a natural aroma compound used in food and cosmetic applications, and its efficient microbial production requires the coordinated control of glucose and precursor availability. Here, Raman and near-infrared (NIR) spectroscopy were implemented as soft sensors to guide glucose/l-phenylalanine (l-Phe) cofeeding during 2-PE biosynthesis by Saccharomyces cerevisiae. Raman-Artificial Neural Network and NIR-Partial Least Squares models showed the best performance (external validation R2 up to 0.99). A soft-sensing-guided glucose-feeding strategy maintained glucose below 20 g/L and triggered refeeding after ethanol depletion. Under these conditions, the optimal l-Phe supply range was found to be 5-15 g/L, yielding 16.72 g/L 2-PE and 0.66 g/g l-Phe-to-2-PE conversion yield. Transcriptomics at low, intermediate, and high l-Phe levels indicated that intermediate l-Phe promoted SPS-module/Ehrlich-pathway responses, whereas low or excessive l-Phe impaired nitrogen- and redox-related metabolism. These results support the process and strain optimization for 2-PE biomanufacturing.