4. Discussion
Traditional fermentation process regulation is based on manual experience. The analysis of relevant parameters can identify connections between sensitive parameters and key indicators (Chen et al., 2013; Zhang et al., 2014). With the development of sensor technology, a variety of advanced on-line sensors have been applied in the fermentation process to achieve multi-dimensional monitoring of cell metabolism, thus making the process control more rational and reliable. These sensors have included process mass spectrometry, near-infrared spectroscopy, raman spectroscopy, and viable cell biosensors (Chen et al., 2021a; Iversen, Berg & Ahring, 2014). In the present study, we developed an on-line monitoring platform for the fermentation of SLs for the first time. This platform allowed the simultaneous real-time detection of substrates, products, and cellular metabolic states, thus creating a significant amount of process data. We found that the production of SLs under different oil feeding strategies during the fed-batch fermentation could be predominantly divided into three stages: the first stage was limited by cell production capacity; this stage was associated with a relatively higher specific productivity of SLs (up to 0.15 g/gDCW/h). Then, the cells entered a stage that was inhibited by high product concentration when the concentration of SLs reached 75 g/L (non-normalized); at this point, the SLs productivity decreased, whereas the RQ value stabilized at 0.70. Finally, when the concentration of SLs exceeded 140 g/L (non-normalized), the cells entered a third stage which was limited by oxygen supply. During this stage, the SLs productivity continued to decline, and the RQ began to rise. Thus, RQ is a key parameter to characterize the difference of substrate metabolism in SLs fermentation and can be adopted as a potential control parameter to guide process optimization.
The method of mining key parameters from a huge amount of process data by manual analysis is both time-consuming and laborious. With recent advancements in data science technology, the regulation of fermentation based on mathematical algorithms is gradually being applied to realize the processing of many data samples, to identify the relationship between multiple variables and build association models (Zhu et al., 2021). Currently, mathematical algorithms predominantly include linear and non-linear algorithms, such as unary linear models, neural network models, and support vector machines. A data model combining a large number of parameters and mathematical algorithms could help us to predict and regulate the fermentation process, thereby significantly reducing labor costs (Lopez et al., 2013). In a previous study, Lu et al. (2016) established a continuous fermentation control system for feedback that adjusted the glucose supplementation rate during the fermentation process of sodium gluconate by correlating the changing trend between OUR and DO, thus significantly improving the fermentation efficiency of sodium gluconate. In another study, Wang et al. (2020b) proposed a non-linear predictive method for the real-time control of product concentration during L-lysine fermentation to improve the efficiency. Data showed that predictive control, as based on Grey-Wolf Optimization, led to better levels of prediction accuracy, adaptability, real-time tracking ability, overall error, and control accuracy. Herein, six common mathematical algorithms, including LR, MLR, SVMs, PLS, RFs and GBRs, have been used to establish a parameter correlation model for the first and second stages of SLs synthesis, as based on the productivity of SLs and using the consumption of glucose and rapeseed oil as dependent variables. The integrated application of multiple mathematical algorithms was able to achieve the real-time and rational control of process feeding based on fermentation process parameters. Furthermore, the model could accomplish self-optimization under continuous data iteration. However, it was not possible for the model to fully understand the physiological state of the cells.
On the other hand, an intracellular metabolic flux model was established for cells to interpret the mechanism underlying the data model. The processes involving the two substrates (glucose and rapeseed oil) included four main pathways: glucose oxidation, β-oxidation of fatty acids, the synthesis of SLs, and the synthesis of extracellular by-products. Through structure and component analyses, the main proportion of the lactone-form of SLs reached 85.0%; of these, the C18:1 lactone-form of SLs accounted for the highest proportion at 58.2%. This may have been because the fatty acid component in rapeseed oil was mainly composed of oleic acid and linoleic acid; the proportion of oleic acid reached 68.5% (Tables S1 and S2). The extracellular organic acids were mainly citric acid (CIT), pyruvic acid (PYR), malic acid (MAL), and succinic acid (SUC); these organic acids were present in only very low concentrations (Fig. S1). The carbon ratio of extracellular organic acids to the added substrates during the synthesis of SLs was only 7.7%.
During the fed-batch fermentation of 24-96 h, we calculated the consumptions of glucose and rapeseed oil, the synthesis of CO2, SLs, and extracellular organic acids; this allowed us to calculate that the carbon balance at this stage was 95.8% (Table S3). Metabolic flux analysis demonstrated that when the RQ was 0.70, the ratio of glucose and rapeseed oil to SLs was fixed. The synthesis of 1 g of SLs required 0.824 g and 0.719 g of glucose and rapeseed oil, respectively. Furthermore, 66.5% of the carbon sources in the substrate were transferred to SLs, 21.7% to CO2, and 7.7% to organic acids (Fig. S2). Therefore, there was a strong relationship between the substrate consumption rate, the productivity of SLs and respiration metabolism (carbon dioxide production rate, oxygen consumption rate) during the synthesis stage of SLs. The difference in respiration metabolism is closely related to the change of culture conditions. The multi-parameter integrated control platform could realize precise and intelligent control according to the changes in the culture environment and cellular metabolism. The mechanism-assisted data model was applied for the rational and precise control of residual oil concentration in semi-continuous SLs fermentation. The oxygen limitation problem encountered during the third stage was effectively alleviated, thereby further increasing the SLs productivity to 2.30 g/L/h with an increase of 40.2%.