2 Materials and methods
2.1 Strain, media, and culture conditions
Candida bombicola ATCC 22214 was obtained from the Guangdong Culture Collection Center (China) and stored at -80℃ in 20% glycerol solution. Seed medium contained 50 g/L of glucose, 1 g/L of KH2PO4, 4 g/L of (NH4)2SO4, 0.5 g/L of MgSO4·7H2O, and 10 g/L of corn steep liquor (CSL). Seed was cultured in a 1 L baffled shake flask with 200 mL working volume at 200 rpm and 25℃ for 48 h.
The initial fermentation medium was placed in a 5 L bioreactor (Shanghai Guoqiang Bioengineering Equipment Co., Ltd., China) and consisted of 100 g/L of glucose, 1 g/L of KH2PO4, 4 g/L of (NH4)2SO4, 0.5 g/L of MgSO4·7H2O, and 10 g/L of CSL. All culture media were sterilized at 115℃ for 30 min. An initial volume of 2.5 L was mixed with 3% of inoculum and cultured at 25℃ for 168 h. Aeration was provided at 0.5 vvm and dissolved oxygen (DO) was maintained above 30% of the saturation concentration by adjusting agitation in a stepwise manner. A pH of 3.5 was maintained during the entire process by the addition of 4 M NaOH solution. The fed-batch fermentation cycle lasted 168 h. During the fed-batch fermentation process, the feeding rate for rapeseed oil was 1 g/L/h in the first 24 h, and was then set to 2.0 (Low), 2.6 (Medium), and 3.2 (High) g/L/h. During the late stage of fermentation, the residual oil concentration of the fermentation broth was controlled so that it did not exceed 30 g/L. During semi-continuous fermentation, the concentration of glucose and rapeseed oil was controlled at 40 g/L and 10 g/L, respectively, and the glucose concentration was maintained below 30 g/L by controlling the feeding rate prior to in-situ separation. In-situseparation was carried out when the concentration of SLs in the fermenter exceeded 140 g/L (non-normalized), as determined by the on-line sensor.
2.2 Determination of fermentation process parameters
During the fermentation process, the concentration of SLs, rapeseed oil, and glucose, were determined by using a real-time on-line detection platform involving a near-infrared spectrum (Chen et al., 2021a). The relationship between off-line data and near-infrared spectrum data was established during the early stages to achieve the real-time detection of substance concentration in the fermentation broth. OUR, CER, and RQ were calculated on-line through detecting the oxygen and carbon dioxide proportion in the inlet air and off-gas by a process mass spectrometer as described previously by Chen et al. (2019). For dry cell weight (DCW) determination, 2 mL of fermentation broth was sampled, washed three times with an equal volume of 70% ethanol (v/v) solution, and then dried in an oven at 80°C. Next, the sample was weighed. SLs structure was determined by LC-MS as described previously by Chen et al. (2020).
Extracellular organic acids were analyzed by the HPLC method. A 2 mL sample of fermentation broth was taken at 24 h, 48 h, 72 h, and 96 h. The samples were then centrifuged to isolate the supernatant. The supernatant was filtered into a liquid phase vial with a 0.22 μm filter and then analyzed by HPLC, including a VARIAN Metacarb-H plus chromatographic column, a RID detector, 0.01 mol/L dilute sulfuric acid as the mobile phase, a flow rate of 0.4 mL/min, an injection volume of 10 μL, a column temperature of 50°C, and a detection temperature of 35°C.
The composition of rapeseed oil was determined by the GC method, as described previously by Gao, Liu, Jin & Wang (2019).
2.3 The influence of SLs concentration on SLs synthesis
First, we added 200 mL of fermentation broth (the concentration of the mixture of glucose and rapeseed oil was 50 g/L) into a 1 L baffled shake flask. Next, we added SLs to the shake flask to prepare the following concentrations: 0, 15, 30, 45, 60, 75, 90, and 105 g/L. Then, we took 50 mL of the fermentation broth and cultured this sample in a 5 L fermenter for 30 h; this was followed by centrifugation to remove the supernatant. After washing, the cells were resuspended in 20 mL of sterile water and inoculated in a shake flask. This was then cultured for 24 h at 200 rpm and 25°C.
2.4 Feedback feeding model
The feedback feeding model achieved stable, real-time, and rational regulation of fermentation process feeding based on on-line parameters and fermentation control. This model was mainly divided into an on-line parameter acquisition module, a parameter analysis module, and a feedback feeding module. First, we established a real-time detection system for fermentation process parameters (independent variables: OUR, CER, RQ, pH, DO, temperature, agitation, aeration, and DCW). This was achieved by a variety of on-line sensors (near infrared spectroscopy, a process mass spectrometer, a dissolved oxygen electrode, temperture electrode and pH electrode ). We also monitored a range of dependent variables, including SLs productivity, glucose consumption rate, and rapeseed oil consumption rate. Then, we used six algorithms to fit these key parameters, including multiple linear regression, partial least squares, a support vector machine, random forest, and gradient boosting regression. The correlations between parameters (R2) were then used to construct a data correlation model. Next, we used the data correlation model to select the optimal feedback value for feed output in accordance with the on-line parameters and fermentation control. Finally, we used the output correlation (R2) for multiple algorithm models to achieve real-time control over the feed module. The model could read a range of detection data (SLs productivity, fermenter volume, DCW, residual oil concentration, residual glucose concentration, OUR, and other parameters) every 1 h. The control of oil and glucose concentrations enabled real-time feedback and regulation of oil and glucose supplementation. During the verification process, fed-batch fermentation was performed to control the residual oil concentration of the fermentation broth to 2 g/L and 10 g/L; the glucose concentration was controlled at 40 g/L. During semi-continuous fermentation, the residual oil concentration was controlled at 10 -15 g/L, and the glucose concentration was controlled at 30-40 g/L. We used gravity sedimentation separation and washing recovery to achieve in-situ separation and fermentation for 234 h. The fermentation control platform and feedback feeding model are shown in Fig. 1.
2.5 Data analysis
Due to the consideration of changes in working volume during feeding, all of the data presented in the figures and tables were normalized to the initial volume (except those marked ‘non-normalized in part’), and represent the real-time detection of SLs concentration in the fermenter. All experiments were performed in triplicate. The data shown in the tables and figures represent the mean ± standard deviation of three experiments. One-way analysis of variance (ANOVA) and the T-test (P < 0.05) were used to identify significant differences between treatments (GraphPad Prism 8.3.0, GraphPad Software Inc., USA).