外文文献

Medium optimization for glutathione production by Saccharomyces cerevisiaeChi-Hsien Liu,Chin-Fa Hwang,Chii-Cherng LiaoFood Industry Research and Development Institute. P.O. Box 246, Hsinchu, Taiwan, ROC Received 28 April 1997. Revised 25 March 1998. Accepted 28 March 1998. Available online 19 August 1999./10.1016/S0032-9592(98)00055-7, How to Cite or Link Using DOI Permissions & ReprintsAbstractGlucose, peptone and magnesium sulphate were found to be suitable components for the cell growth and glutathione (GSH) production in the yeast strain. Saccharomyces cerevisiae CCRC 21727. The Box–Behnken design and response surface methodology were employed to derive a statistical model to investigate the effects of glucose, peptone and magnesium sulphate concentrations on GSH production. Neural networks were compared with a second-order-polynomial model in predicting the effects of component concentrations on the production of GSH and dry cell weight (DCW). Neural network models can predict cell growth and GSH production more precisely than second-order-response-surface models.Keywordsmedium optimization;glutathione;neural network prediction;response-surface model1. IntroductionGlutathione (GSH) is a tripeptide consisting of l-glutamate, l-cysteine and glycine. In 1921, Hopkins isolated GSH from yeasts, animal livers and muscle [1]. GSH widely affects oxidation and reduction in vivo. It is used as a medicine for liver and as a scavenger of toxigen [2]. High levels of GSH content exist in some strains of yeasts which are the microorganisms commonly used for commercial GSH production [3] and [4].Neural networks (NNs), computer model systems, match the functionality of the brain in a fundamental manner. NNs have recently been used in many areas of fermentation such as brewing [5], [6] and [7] and process state variable estimate [8], [9] and [10]. Kennedy et al. compared the use of NNs to factorial design as tools for medium design and reported that NNs could he trained on less data and predicted the experiments with accuracy for medium design [11].Response surface methodology (RSM), an experimental strategy for seeking the optimum conditions for a multivariable system, is a much more efficient tool for optimization than the one-factor-at-a-time approach [12]. RSM has been successfully applied for optimizing media gradients in shake-flask systems [13], [14], [15] and [16]. The steepest ascent method (SAM) was used to evaluate the influence of component concentrations on the product via the regression model [12], [13] and [17]. In this study, a GSH-producing strain of Saccharomyces cerevisiae, was used. The authors adopted Box–Behnken design [12] and RSM to determine the optimal medium for GSH production and cell growth. Neural-network models were compared withsecond-order-response-surface models to predict GSH production and dry cell weight (DCW) along the steepest ascent paths.2. Materials and methods2.1. Organism and cultivation conditionsSaccharomyces cerevisiae CCRC 21727 (ATCC 7754) from the Culture Collection and Research Center of Food Industry Research and Development Institute in Taiwan was maintaincd on YM Agar (Difco Laboratories, MI, USA) and grown in YM Broth (Difco Laboratories) at 30°C and 150 rpm for 18 h as the inoculum. For all experiments, a 4% inoculum was used in a 50 mlitre Hinton-flask containing 10 mlitre culture medium and incubated at 30°C and 150 rpm for 24 h. Each trial was conducted in triplicate.2.2. Experimental design and statistical analysisVarious carbon, nitrogen and salt sources were tested to determine the best components for GSH production and cell growth. The most important components were then chosen for optimization of concentrations. The three-variable Box–Behnken design with 13 combinations and three replications of the centre point was used [18]. Each trial was conducted in triplicate. The average data of GSH production and DCW were analysed by SAS software [19] to construct the second-order models and used to train NNs software [20]. A set of experiments designed by SAM was used to evaluate the accuracy of the second-order models and the trained NNs.The artificial NN has been inspired by biological models and the building blocks of artificial NN are processing elements (PEs). PEs are combined into layers, the input is received and weighted according to weighting factors and the resulting quantities are summed. PEs then execute a transfer function on the weighted sum, and pass the value onto the next PE. The transfer function maps the possibly infinite summation of PE input to a predefined range, the output. The learning or training phase of a NN requires paired input-output data. The input is fed into the network, transferred through the network layers. Ultimately, the network calculates a predicted output. This predicted output is subsequently compared with the actual output, and the connection weights between the PEs are modified to minimize the deviation between the predicted output and the actual one. The learning rule specifies how connection weights are changed during the learning process. The procedure continues until a defined accuracy has been reached. This is the concept of back-propagation [24].In this study, two back-propagation NNs consisting, of four layers, one input layer, two hidden layers and one output layer, were used. The numbers of PEs in these two hidden layers were six and three, respectively. The hyperbolic tangent function and delta rule were chosen for the transfer function and learning rule.2.3. AssaysAfter fermentation the broth was boiled for 5 min, ice-cooled and then centrifuged at 12 000 rpm for 3 min, the GSH content in the supernatant was determined according to the published methods [21] and [22] by measuring the absorbance of reaction solutions at 412 nm using a spectrophotometer (Beckman Inc., DU 7400 Spectrophotometer, CA, USA). Cell growth was determined by measuring the DCW. DCW was determined after yeast cells were centrifuged at 12 000 rpm for 3 min and washed twice with deionized water before drying at 70°C for 24 h.3. Results and discussionThe effects of various carbon sources on GSH production and cell growth in S. cerevisiae culture are illustrated in Table 1. Glucose was found to be the best carbon source for GSH productionwhile fructose and sucrose were the two most suitable carbon sources for growth of S. cerevisiae. Shimizu and coworkers reported that when glucose rather than ethanol was used as sole carbon source, both specific growth rate and GSH specific production rate were higher [3]. Cell growth was similar but GSH production was 92% higher when glucose was the carbon source as compared to ethanol. With glycerol as carbon source, the ratio of GSH to DCW was the highest among all the carbon sources tested but GSH and DCW productions was 56·9 and 25·8% of those when glucose was used as carbon source. When lactose or mannitol was used, GSH and DCW production was poor. Glucose was used as the carbon source for further experiments.Table 1. Effects of various carbon sources on relative GSH production and relative DCW in S. cerevisiae cultureaCarbon source Relative GSH production(%)Relative DCW(%)Ratio of GSH to DCW(%)Glucose 100·0 100·0 0·48Lactose 22·8 21·6 0·51Fructose 77·1 117·3 0·32Sucrose 87·9 115·3 0·37Glycerol 56·9 25·8 1·06Mannitol 24·0 28·9 0·40Ethanol 51·7 100·0 0·25aThe medium contained 1% yeast extract, and the carbon sources were added at a concentration of 1% (w/v). GSH production and DCW which equate to 100% are 28·26 mg litre−1 and 5·88 g litre−1, respectively.Full-size tableThe effects of nitrogen sources on GSH production and cell growth are shown in Table 2. The highest GSH content per unit of DCW was obtained when tryptone was used as nitrogen source, although the amounts of GSH and DCW were 61·4 and 57·6% of those when peptone was used. Peptone and yeast extract were the two best nitrogen sources for GSH production by S. cerevisiae. There was a two-fold increase in growth when yeast extract was used instead of peptone as nitrogen source, resulting in lower GSH content per unit of DCW. Peptone was thus chosen as the nitrogen source for further experiments.Table 2. Effects of various nitrogen sources on relative GSH production and relative DCW in S. cerevisiae cultureaNitrogen source Relative GSH production(%)Relative DCW(%)Ratio GSH to DCW(%)Peptone 100·0 100·0 0·97Tryptone 61·4 57·6 1·04Meat extract 80·7 112·8 0·70Yeast extract 100·0 202·8 0·48Malt extract 10·7 42·8 0·24aThe medium contained 1% glucose, and the nitrogen sources were added at a concentration of 1% (w/v). GSH production and DCW which equate to 100% are 28·26 mg litre−1 and 2·90 g litre−1,respectively.Full-size tableMetal ions such as K+, Mg+2 and Ca+2 directly influence fermentative metabolism in yeasts [25] and (NH4)2SO4 and NH4Cl stimulate ethanol production by S. cerevisiae[27]. Khan et al. reported that (NH4)2SO4 was a suitable nitrogen source for the production of yeast biomass [26]. The effects of various salt sources on GSH production and cell growth are shown in Table 3. All salts except MgSO4 showed inhibition on both GSH production and cell growth. There was no previous report on the effect of MgSO4 on GSH production in yeast. Mills et al. reported that magnesium deficient diet inhibits biosynthesis of blood GSH in rat [23]. As MgSO4 could stimulate GSH production and cell growth, it was chosen for further study.Table 3. Effects of various salts on relative GSH production and relative DCW in S. cerevisiae cultureaSalt source Relative GSH production (%) Relative DCW (%) Ratio of GSH to DCW (%) Controlb 100·0 100·0 0·97CaCl2 39·0 87·2 0·44KH2PO4 46·2 75·2 0·60(NH4)2SO4 43·7 73·4 0·58NH4CL 41·4 43·1 0·94MgSO4 125·9 126·9 0·97K2SO4 47·4 55·2 0·84KHO3 53·0 54·5 0·95aThe medium contained 1% glucose and 1% peptone, and the salts were added at a concentration of 0·25% (w/v). The GSH production and DCW which equate to 100% are 28·26 mg litre−1 and 2·90 g litre−1, respectively.bThe control contained 1% glucose and 1% peptone only.Full-size tableThe effects of glucose concentrations on DCW and GSH production are shown in Fig. 1. Biomass increased in parallel with the increase of glucose concentration. GSH production was highest when the glucose concentration was 1·5%. The effects of peptone concentrations on DCW and GSH production are shown in Fig. 2. The cell concentration increased with the elevation of peptone concentration. GSH production was the highest when the concentration of peptone was 1·0%, beyond which it decreased. The influence of MgSO4 concentrations on GSH production and DCW were then studied (Fig. 3). MgSO4 at a concentration of 0·15% was sufficient for S. cerevisiae to grow and produce GSH. A further increase in MgSO4 concentration had no effect on cell growth and GSH production. From these one-factor-at-a-time experiments, the best medium composition for GSH production in S. cerevisiae was 1·5% glucose, 1·0% peptone and 0·125% MgSO4. The GSH concentration reached 66·64 mg litre−1 using this medium.Fig. 1. Influence of glucose concentration on GSH production and DCW of S. cerevisiae, where the concentrations of peptone and MgSO4 were 1·0 and 0·25%, respectively.View thumbnail imagesFig. 2. Influence of peptone concentration on GSH production and DCW of S. cerevisiae, where the concentrations of glucose and MgSO4 were 1.0 and 0·25%, respectively.View thumbnail imagesFig. 3. Influence of MgSO4 concentration on GSH production and DCW of S. cerevisiae, where the concentrations of glucose and peptone were 1·0 and 1·0, respectively.View thumbnail imagesThe range of concentration tested in the optimization procedure was 0–4% glucose, 0–4% peptone and 0–1% MgSO4 according to the preliminary experiments. The coded levels for each component are shown in Table 4. Box–Behnken design was employed to fit a second-order RSM. The total number of trials needed for this design was 15, fewer than those of the central composite design or the 33 factorial design. The Box–Behnken design and its experimental results based on the three-factor-three-level response surface analysis are shown in Table 5. The regression coefficients, t-values and determination coefficients (R2) for the second-order RSMs of GSH and DCW production are shown in Table 6.Table 4. Assigned concentrations of variables at different levels in Box–Behnken design Independent variables Coded levels−1 0 1X1 Glucose (%) 0 2 4X2 Peptone (%) 0 2 4X3 MgSO4 (%) 0 0·5 1Full-size tableTable 5. Experimental design and results of Box–Behnken design for GSH and DCW production Glucose Peptone MgSO4 GSH (mg litre−1) DCW (g litre−1)1 1 0 106·36 9·431 −1 0 7·39 3·30−1 1 0 17·15 1·73−1 −1 0 6·07 0·201 0 1 101·95 6·501 0 −1 18·92 3·23−1 0 1 9·01 1·07−1 0 −1 9·36 1·650 1 1 89·72 7·800 1 −1 22·32 2·700 −1 1 6·70 1·650 −1 −1 6·48 0·770 0 0 89·72 6·300 0 0 94·88 6·100 0 0 83·88 6·00Full-size tableTable 6. Regression coefficients, t-values and determination coefficients (R2) of a full second-order polynomial model for GSH and DCW productionaTerm GSH DCWCoefficient t-Value Coefficient t-Valueβ089·49 31·32b 6·13 13·24bβ124·13 13·79b 2·23 7·85bβ226·11 14·92b 1·97 6·94bβ318·79 10·74b 1·08 3·82cβ11−25·89 −10·04b −1·29 −3·10β1221·97 8·88b 1·15 2·87β22−29·38 −11·40b −1·18 −2·82β1320·84 8·42b 0·96 2·40β2316·80 6·79b 1·06 2·63β33−28·81 −11·18b −1·73 −4·14cR2 0·9952 0·9723ay=β0+β1X1+β2X2+β3X3+β11X21+β22X22+β33X23+β12X1X2+β13X1X3+β23X2X3bSignificant at 0·1%.cSignificant at 1%.Full-size tableThe Student’s t-test, based on the hypothesis that the true parameter is zero, was employed in the second-order models to elucidate the significance of the factors. All factors including linear, quadratic and interaction terms of glucose, peptone and MgSO4 presented significant effects on GSH production. For cell growth, all linear terms of these three components and the quadratic term of MgSO4 showed significant effects, and no significant interaction existed between these three factors. The regression models of GSH production and cell growth were highly significantwith satisfactory R2 values of 0·995 and 0·972, respectively, which means 99·5 and 97·2% variation could be explained by these two models. The result indicates that these regression models are suitable to represent GSH production and cell growth. Judging by the regression coefficients and t values, it may be concluded that all the carbon source, nitrogen sources and salt sources had the same positive effects on GSH production and cell growth. The analysis of variance (ANOV A) was used to test the adequacy of models (Table 7). The results of ANOV A were in good agreement with the Student’s t-test (Table 6). The lack of fit and pure errors did not influence the accuracy of models.Table 7. Analysis of variance for regression models of GSH and DCW production obtained from the response-surface experimentsSource Degree of freedom Sum of squaresGSH DCWModel (regression 9 25 300·00a 112·92aLinear 3 12 937·00a 80·01aQuadratic 3 7565·69a 19·43bCross-product 3 4797·51a 13·44Residual (error) 5 122·49 3·22Lack of fit 3 61·91 3·17Pure error 2 60·58 0·05Total 14 25 422·59 116·14F ratio (model) 114·70 19·50P-value (model) <0·0001 0·0022aSignificant at 0·1%.bSignificant at 1%.Full-size tableThe canonical analysis was a mathematical procedure used to simplify a second order polynomial model and to decide the stationary point to be a maximum, minimum or saddle point. The results from the canonical analysis of response surfaces are shown in Table 8. The stationary points determined by canonical analysis for GSH and DCW production were maximum points, and the optimal medium for GSH production was 5·20% glucose, 4·84% peptone and 1·16% MgSO4. The optimal medium for cell growth was determined to be 8·45% glucose, 8·86% peptone and 1·63% MgSO4. A set of experiments designed by SAM was performed to check the paths to the maximum points of GSH and DCW production predicted by the RSM model. NNs were trained by the data of the Box–Behnken design and used to predict GSH and DCW production along the steepest ascent paths. The concentrations of components, experimental results of the steepest ascent paths, and prediction by RSM and NN are shown in Table 9 and Table 10. When component concentrations exceeded those of the maximum points (trial 7 of Table 7 and Table 8) predicted by RSM models, the GSH and DCW production remained almost constant. ANOV A and R2 values indicate that these models are suitable to represent GSH production and cell growth, but experimental results were at variance with the data predicted by the RSM model in the designed concentration ranges. The data predicted by NN models fit better to the experimental data thanthose predicted by RSM. When the component concentrations were outside the experimental range, the NN model prediction would remain almost constant. The NN model may give erroneous predictions outside the experimental range when the production or growth was inhibited in high substrate concentration ranges. Such errors could be eliminated by the second-order model for its parabolic curve characteristic. Besides, GSH production in the optimum medium composition predicted by RSM was 115·28 mg li tre−1, which was 73% more than that in the best medium composition designed by one-factor-at-a-time experiments.Table 8. The canonical analysis of response surfaces for GSH and DCW production Response variable Critical value of process factor (%) Predicted value Stationary pointGlucose Peptone MgSO4GSH 5·20 4·84 1·16 139·74 mg litre−1 MaximumDCW 8·45 8·86 1·63 14·32 g litre−1MaximumFull-size tableTable 9. Concentrations of medium components and GSH production along the steepest ascent pathTrials Component concentration (%) GSH (mg litre−1)Glucose Peptone MgSO4 Experimental Predicted by RSM Predicted by NN1 2·00 2·00 0·50 89·15 89·49 89·542 2·53 2·47 0·61 92·77 104·85 93·273 3·07 2·95 0·72 98·00 117·41 98·144 3·60 3·42 0·83 94·78 127·18 102·765 4·13 3·89 0·94 99·20 134·16 102·606 4·67 4·37 1·05 108·45 138·34 102·487 5·20 4·84 1·16 115·28 139·74 102·508 5·74 5·31 1·27 124·13 138·34 102·559 6·27 5·79 1·38 124·93 134·15 102·61Full-size tableTable 10. Concentrations of medium components and DCW production along the steepest ascent pathTrials Component concentration (%) DCW (g litre−1)Glucose Peptone MgSO4 Experimental Predicted by RSM Predicted by NN1 2·00 2·00 0·50 4·83 6·13 6·122 3·08 3·14 0·69 5·90 8·64 8·903 4·15 4·29 0·88 6·97 10·68 9·254 5·23 5·43 1·06 7·95 12·28 9·985 6·30 6·57 1·25 8·68 13·41 9·296 7·38 7·72 1·44 8·90 14·10 9·307 8·45 8·86 1·63 9·45 14·32 9·308 9·53 10·00 1·82 9·53 14·10 9·309 10·60 11·15 2·01 9·50 13·42 9·30Full-size tableAcknowledgementsThis work was supported by grants from the Ministry of Economic Affairs of the Republic of China.。