Optimization of Fermentation Medium for Producing α-Hydroxyphenylacetic Acid by Using Plackett-Burman Design and Response Surface Methodology *

Plackett-Burman design and response surface methodology were applied in order to optimize the fermentation medium of (R)-α-hydroxyphenylacetic acid ((R)-HPA) producing Bacillus sp. HZG-19. The factors playing important roles in the production of (R)-HPA were selected based on Plackett-Burman design. The path of steepest ascent was undertaken to optimize said fermentation medium. Finally, the optimal levels of the factors with the greatest change in regard to product yield were further optimized using Box-Behnken and response surface analysis. The optimal conditions were found to be as follows: casein peptone 30.49 (g × L), glycerol 14.09 (g × L), KH2PO4 0.1345 (g × L), K2HPO4 0.01 (g × L), CaCl2 0.1 (g × L), MnSO4 0.01 (g × L). Under the optimal conditions described above, the yield of (R)-HPA reached 63.30%, which indicated an increase of 14.9%, as compared to the yield obtained before optimization.


Introduction
α-Hydroxyphenylacetic acid (mandelic acid) (HPA) and its derivatives are key intermediates for the production of various pharmaceuticals, such as semi-synthetic penicillins and cephalosporins [1][2][3].It is also used as a chiral resolving agent and chiral synthon for the synthesis of anti-tumor and anti-obesity agents [4].Many methods have been reported for the preparation of enantiomerically pure (S)-or (R)-α-Hydroxyphenylacetic acid [5][6][7][8][9].For example, the synthesis of enantiopure α-Hydroxyphenylacetic acid has been investigated through diastereomeric crystallization [10].It has also been prepared by the chemo-enzymatic routes [11], from methyl mandelate by lipase-catalyzed hydrolysis [12], or from kinetic resolution [13].However, these approaches are limited in their industrial application due to the expensive catalysts, limited efficiency and low yields.
Recently we have reported a new bacterial strain, Bacillus sp.HZG-19, which is capable of degrading phenylglyoxylic acid (PGA) and affording (R)-HPA with high optical purity.Numerous variables will have an effect on the production of HPA, hence, factors playing important roles in the production of pure enantimomeric (R)-HPA are crucial for large scale production.Plackett-Burman, Box-Behnken design and response surface methodology are inexpensive and accurate methods for further optimization of the fermentation medium.

Strain and Chemicals
The strain (Bacillus sp.HZG-19) was preserved in the lab of Chemical Engineering Department, Fuzhou University.(R)-HPA and (S)-HPA were purchased from Sigma (St. Louis, MO USA).Phenylglyoxylic acid (PGA, >98%) was supplied by Pharmaceutical & Chemical Co., Ltd of Taizhou, China.Methyl alcohol of HPLC grade was purchased from Merck, Germany.Hydroxypropyl-β-cyclodextrin (HP-β-CD) was supplied by Fluka (Neu Ulm, Germany).All other chemicals were obtained from local suppliers and of reagent grade.
Medium for slant culture (SM): same as seed medium with 2% (w/v) of agar.

Culture Conditions
The strain, Bacillus sp.HZG-19, was inoculated into 20 ml seed medium in 50 ml Erlenmeyer flasks and cultured aerobically at 32˚C on a shaker (180 rpm).When the cells reached logarithmic phase, cultures were then inoculated (1%, v/v) into 50 ml flasks containing 20 ml of FM medium at the same fermentation conditions.PGA solution, with a final concentration of 15 mM, was added directly into the fermentation broth under aseptic conditions.After 24 h of incubation, the supernatant that was obtained after centrifugation from the medium and was subjected to HPLC analysis and HPCE analysis to determine the concentration and purity of the HPA generated.

Analytical Methods
The concentrations of substrate and product were determined by HPLC using a reverse phase column (Agilent HC-C 18, Ø 4.6 mm × 250 mm, 5 μm).The mobile phase is composed of methanol and phosphate buffer (25 mmol/L) (15:85, v/v) containing 6 mmol/L of tetrabutylammonium bromide (pH 6.8) at the rate of 1.0 ml/min.A UV detector (at 220 nm) was employed for quantification.
The yield of HPA is expressed as: where CHPA and CPGA represent the concentration of the HPA generated and initial concentration of PGA, respectively.The concentration of (R)-HPA and (S)-HPA were determined by HPCE.Detection was made at 214 nm using a buffer solution of Tris-phosphoric acid (100 mmol/L, pH 7.6) containing 150 g/L of hydroxypropylβ-cyclodextrin.A voltage of 20 KV was applied at a temperature of 20˚C.
where [R] and [S] are the concentration of (R)-HPA and

Results and Discussion
is extremely useful in ation of experim Table 1.Design matrix and experimental results of Plack- (X 4 ) X 5 (X 6 ) X 7 (X 8 ) X 9 Yield/%

Plackett-Burman Design
The Plackett-Burman (PB) design screening and selecting for the most vital factors within a large candidate pool [14].The experiments were carried out according to the design matrix shown in  2. Generally, effects with a confidence level of greater than 95% are considered to be actual effects, whereas a confidence level of less than 95% indicates that the effect may have resulted due to chance.As shown in Table 2, the confidence level of factors casein peptone (X 1 ), glycerol (X 2 ) and KH 2 PO 4 (X 3 ) are shown to be above 95% and considered to be significant.The rest of the factors, K 2 HPO 4 (X 5 ), CaCl 2 (X 7 ) and MnSO 4 (X 9 ) had a confidence of below 95% in HPA production and hence, were considered insignificant.The high level of X 5 , X 7 and the low level of X 9 were selected during further optimization studies.

Steepest Ascent Design to Approach the Optimal Region
The steepest ascent method was applied in order to investigate optimal substrate concentration, by slowly increasing substrates along the path of steepest ascent until no further increase in response is observed [15].Based on the experimental results of PB, casein peptone, glycerol and KH 2 PO 4 should be increased due to their positive effect (positive value for t).Table 3 represents the design and results of the steepest ascent search experiment.As shown in Table 3, optimal fermentation medium should be around run 2, so the level of run 2 is considered as center point in the follow-up response surface experiment.

Response Surface Analysis (RSM) and Establishment of Optimum Fermentation Medium
RSM is a statistical method by which one can find the best conditions in a multi-factor system [16].Among RSM, Box-Behnken design (BBD) method and central composite design are more frequently used than the others.When the number of the experimental factors does not exceed 5, Box-Behnken method is more economic.Casein peptone, glycerol and KH 2 PO 4 were determined to be the main factors as shown through the PB experiment.X 1 , X 2 , X 3 represent casein peptone, glycerol, KH 2 PO 4 respectively and the yield of HPA shows the response value.Factor levels were determined by steepest ascent experiment.In order to investigate the optimum levels of those variables and study their interactions, a three-factor three-level BBD was applied.
A total of 15 experiments were performed in triplicates, 12 of which used different factors while 3 trials were used as controls.Controls were used to estimate experimental error.Table 4 illustrates the coded and non-coded values of the experimental variables, with results of the BBD experiments given in Table 5. ased on the experimental results of BBD (Table 5) an 2 12 B d regression analysis, a quadratic polynomial equation was established to identify the relationship between yield and variables.The model of coded units can be expressed as: where Y 1 is the yield of HPA, X 1, X 2 and X 3 represent casein peptone, glycerol and KH 2 PO 4 , respectively.
eterm was reliable with an R 2 value of 0.9913.It suggested that model was unable to explain only 0.87% of the total variations.A low value of coefficient of the variation (C.V.) (1.0298) clearly indicated a very high degree of precision and a good deal of reliability of the experimental values.
Analysis of variance (ANOVA) is important in d ining the adequacy and significance of the quadratic model.ANOVA summary was shown in Table 6.The Fvalue of 63.1957 implies that the model was significant.A p value of <0.0001 indicated that there was only a 0.01% chance that model's large F-value could occur due to noise.Values of "prob > F" less than 0.0500 indicated the significant model terms.The mathematical model     and 2, there was a maximum response at the optimum level of each variable and exist the interaction among the three variables, so it was not simple linear relationship for the effect of response value.The ridge analysis indicated that the regression equation has no singularity.Simulta st response surface conditions and the predictive value of yield from the regression equation.The best theoretical levels in experiment were: X 1 = 0.816, X 2 = 0.592 and X 3 = −0.045,that is, casein peptone 30.49g•L −1 , glycerol 14.09 g•L −1 and KH 2 PO 4 0.1345 g•L −1 .Predicted value of yield given these conditions was 63.35%.Validation under the optimized conditions was performed in a 50 ml Erlenmeyer flask containing 20 ml reaction medium.The experiments were conducted in triplicate.Under optimized conditions, HPA yield achieved in the verification experiment was 63.30%, which was very close to the value predicted by model based on BBD (63.35%).
Selective reduction of PGA with Bacillus sp.HZG-19 is a very promising techno

Conclusion
Results of this stu fective method for the optimization of fermentation medium.Optimum casein peptone, glycerol and KH 2 PO 4 were found to be 30.49,14.09 and 0.1345 g•L −1 , respectively.

Figures 1 -
Figures 1-2 show the response and contour curves for casein peptone, glycerol and KH 2 PO 4 .Contour curves

Figure 1 .
Figure 1.Response surface plot and contour plot for HPA yield (%) as a function of casein peptone and KH 2 PO 4 (X 1 and X 3 represent casein peptone and KH 2 PO 4 , respectively).

Figure 2 .
Figure 2. Response surface plot and contour plot for HPA yield (%) as a function of glycerol and KH 2 PO 4 (X 2 and X 3 repr sent glycerol and KH 2 PO 4 , respectively).

Table 1 ,
with each row representing one trial while each column represents a single variable.The six factors listed are: casein peptone, glycerol, KH 2 PO 4 , K 2 HPO 4 , CaCl 2 , MnSO 4 , respectively.Three variables (X 4 , X 6 , X 8 ) are dummy variables employed to evaluate the standard errors of the experiment.Elements (1) and (−1) are representative of the relative amounts each variable factor, either high or low, within each experimental trial.Yield of HPA is considered as response value.Levels selection and significance evalu ental variables are summarized in Table

Levels selection and significance evaluation of ex- perimental variables.
2.

Estimated value of regression equation partial re- ression coefficient and analysis of square g deviation.
).