Optimization of epothilone B production by Sorangium cellulosum using multiple steps of the response surface methodolog y

The anticancer compound epothilone B is biosynthesized by the myxobacterium Sorangium cellulosum; however, the fermentation characteristics for epothilone production in Sorangium cellulosum have not yet been reported. In this study, medium components for the production of epothilone B in the So0157-2 strain of S. cellulosum were statistically screened and optimized. First, the nutrients in the fermentation medium were optimized in one-factor-at-a-time, Plackett–Burman design, and Box–Behnken design experiments. Afterwards, three nutritional parameters were selected for the optimization of epothilone B production in shaking flask cultures using a central composite response surface methodology design; a polynomial equation model that related the medium components and epothilone B yield was established. The data were further analyzed using response surface plots and canonical mathematical model analyses with the SAS 8.0 software. After optimized, the yield of epothilone B increased to 82.0 ± 3 mg/l, 7.2-fold higher than the initial yield (11.3 ± 0.4 mg/l).

However, the production efficiency of epothilones in Sorangium is poor.Several reports have attempted to improve the production of epothilones.For example, the gene cluster that encodes the epothilone production pathway has been genetically engineered into a Myxococcus xanthus strain (Julien and Shah, 2002); the *Corresponding author.E-mail: lilab@sdu.edu.cn.Tel/Fax: +86 531 8856 4288.production medium for epothilone D was thereafter optimized for the transgenic strain, and 85 mg/l in a 22day semi-continuous fermentation was obtained (Frykman et al., 2005).Previously, we also made mutations in epothilone-producing Sorangium strains to improve production (Gong et al., 2007).However, the effects of culture medium components on the production of epothilones in natural Sorangium producers have not been reported, most likely due to the difficulties in manipulating S. cellulosum strains (Reichenbach and Dworkin, 1992).Myxobacteria are unique among bacteria because they have complex social living patterns; they feed in groups, move in swarms, germinate myxospores in a cell density-dependent manner, and develop multicellular fruiting bodies (Shimkets, 1990).In this study, we investigated the relationships between the production of epothilone B in S. cellulosum strain So0157-2 and the composition of the culture medium and further optimized the concentration for this compound.
In a conventional process optimization, factors that may be involved in production are initially screened and are then optimized using different techniques (Plackett and Burman, 1946;Box and Hunter, 1957;Lewis et al., 1999;Wang and Wan, 2009).The optimization techniques include non-statistical techniques, such as one-factor-ata-time, and statistical techniques, such as the Plackett-Burman (PB) design, Box-Behnken (BB) design, fractional factorial design (FFD), and central composite design (CCD).The response surface methodology (RSM) is a statistical method that determines the relationships between at least one measured dependent response and a number of input factors.Compared to other optimization techniques, RSM has several advantages because it requires fewer experimental trials, is suitable for multiple factor experiments, searches for relationships between factors, will identify the most suitable conditions, and is able to predict responses; thus, RSM has been employed widely for the optimization of medium components (Kipan et al., 1999;Adinarayana and Ellaiah, 2002;Wang and Lu, 2004;Liu et al., 2005;Claucia et al., 2006;Sharma et al., 2007;Bajaj et al., 2009;Ghanem et al., 2009;Palaniyappan et al., 2009).
As the impact of nutrients for epothilone synthesis had not been investigated exhaustively, we assumed that the biosynthesis of epothilone by S. cellulosum So0157-2 would highly improve when we determine the best nutritional conditions.Furthermore, significant differences of antitumor activities between epothilone A and B, and their corresponding derivatives drive multidisciplinary studies to increase B ratio from the mixture (Rowinsky and Calvo, 2006), therefore, our objective was to optimize medium components for increasing the production of epothilone B in S. cellulosum So0157-2 in this study.The optimization steps were based on statistical design methods and included the following stages: (a) screening of multiple nutrient components in one-factor-at-a-time experiments; (b) elucidating the relationships between medium components and epothilone B production using a PB design; (c) optimizing the significant components using a BB design to produce a new medium; (d) elucidating the effects of the new medium on epothilone B production using a FFD; and (e) optimizing three significant nutritional parameters for the production of epothilone B using a CCD.

Strain and growth conditions
Sorangium cellulosum So0157-2 is an epothilone-producing strain and is routinely cultured on M26 agar at 30°C (Nguimbi et al., 2003).For convenience, cells of this strain were cultured in liquid M26 medium for 3 days and then frozen in aliquots (Gong et al., 2007).Each cryo-vial contained 1 × 10 9 cells in a volume of 1.0 ml.The fermentation medium (EPM) (Gong et al., 2007) was used for the production of epothilone B. The EPM medium contained 2.0 g of potato starch, 2.0 g of glucose, 2.0 g of soy powder, 1.0 g of slim milk powder, 1.0 g of MgSO4 .7H2O, 1.0 g of CaCl2, 1.0 ml of EDTA-Fe 3+ solution (0.01 g/l), 1.0 ml of trace element solution (Reichenbach and Dworkin, 1992), 0.5 mg of VB12, 1000 ml of distilled water, and 2% (v/v) Amberlite XAD-16 resin at pH 7.2.One cryo-vial was used to inoculate 50 ml of M26 medium in a 250-ml Erlenmeyer flask, which was then agitated at 30°C for 3 days.This culture was used as the seed for fermentation.The composition of the production medium varied according to the experimental design.The pH of the medium was adjusted to 7.2 using a 20% KOH solution before autoclaving at 121°C for 20 min.A one-ml aliquot of the inoculum was transferred into each 250-ml flask containing 50 ml of production medium.

Determination of epothilone B production capacity
After shaking at 200 rpm at 30°C for 10 days, the Amberlite XAD-16 resin was harvested from the culture, washed with water, air-dried, and extracted with 50 ml of methanol.Next, the extract was concentrated under a vacuum at 40°C and then re-dissolved in 10 ml of methanol for HPLC analysis (Gong et al., 2007).Each experiment was performed three times.

One-factor-at-a-time experiment
The production of epothilone B by Sorangium cellulosum So0157-2 in EPM medium was 11.3 mg/l.To increase the yield of epothilone B, some nutrient sources (Table 1) were evaluated using the onefactor-at-a-time design based on EPM medium.The substitutes for the complex carbon source potato starch were dextrin, soluble starch and cottonseed powder.Sucrose, maltose, lactose, glycerol and cellubiose were selected as potential simple carbon substitutes for glucose.Complex nitrogen substitutes for soy powder included fish flour peptone, peptone, yeast powder, instant dry yeast, cottonseed powder, soy peptone, casitone and tryptone.In addition, acetate, propionate, cysteine, serine and threonine were used as the substitutes for VB12 at concentrations of 40 mol/l each.All results were obtained from triplicate experiments.

Plackett-Burman design
The PB design, a two-level experimental design method, is a popular method for process improvement in which the relevant factors are selected from a long list of multitudinous factors (Kennedy and Krouse, 1999).In this study, 17 factors (Table 2) were tested on the basis of single factor experiment results and the components of EPM medium for further culture medium optimization.Each factor was tested at high (+1) and low (-1) concentrations (Table 2).The experimental protocol was designed using Statistical Analysis System, version 8.0 (SAS 8.0).The yield of epothilone B was listed as the response variable.All experiments were performed in triplicate.

Box-Behnken design
From the aforementioned PB design experiments, three components-soy powder (X3), glucose (X4) and sucrose (X5) (for each, p < 0.05)-were selected for further optimization by BB design.The other factors were adjusted according to the t-test and p-values of the PB experiment, that is, 3.0 g of dextrin, 1.0 g of slim milk powder, 1.0 g of MgSO4 .7H2O, 1.0 g of CaCl2, 2 ml of EDTA-Fe 3+ solution, 0.5 ml of trace element solution, 1000 ml of distilled water, and 2% (v/v) Amberlite XAD-16 resin at pH 7.2.The BB design in SAS 8.0 was used to optimize the concentration of the three factors selected from the PB experiment.Each factor was tested at three levels: sucrose, 0.04, 0.06, and 0.08% ( X5 = 0.02%); glucose, 0.04, 0.06, and 0.08% ( X4 = 0.02%); and soy powder, 0.15, 0.2, and 0.25% ( X3 = 0.05%).All experiments were performed in triplicate.This part of the study included 15 experiments (Table 3).The response variable, epothilone B yield, was analyzed using SAS 8.0.

Fractional factorial design
Based on the aforementioned experiments, we obtained a preliminarily optimized medium (GSM): 3.0 g of dextrin, 0.5 g of sucrose, 0.8 g of glucose, 1.7 g of soy powder, 1.0 g of slim milk 7H2O, 1.0 g of CaCl2, 2 ml of EDTA-Fe 3+ solution, 0.5 ml of trace element solution, 1000 ml of distilled water, and 2% (v/v) Amberlite XAD-16 resin at pH 7.2.Furthermore, the concentrations of 10 factors (Table 4) in this medium were optimized using the FFD in SAS 8.0.Each factor was tested at high (+1) and low (-1) concentrations, which were a 1.25-fold increase or decrease, respectively (Table 4).All experiments were performed in triplicate.The epothilone B yield was used as the response variable.Three components, dextrin (X2), slim milk powder (X6) and MgSO4 .7H2O (X8), each with p-values < 0.05, were selected for further optimization using the CCD.

Central composite design
A five-variable CCD was used to optimize the important variables selected by the FFD.Table 5 lists the design matrix of the experiment according to the 2 3 full factorial design.The central points of these three components were 0.24% dextrin (X2), 0.12% slim milk powder (X6) and 0.08% MgSO4 .7H2O (X8), and the rest of the factors were as follows: 0.5 g of sucrose, 0.8 g of glucose, 1.7 g of soy powder, 1.0 g of CaCl2, 2 ml of EDTA-Fe 3+ solution, 0.5 ml of trace element solution, 1000 ml of distilled water, and 2% (v/v) Amberlite XAD-16 resin at pH 7.2.The epothilone B yield was used as the response variable in the different cycles of runs and was analyzed using SAS 8.0.Twenty experiments were performed in triplicate, and each of the central points was repeated six times.

Modifying nutrient components based on EPM medium
Table 1 lists the effects of different nutrient components on epothilone B production in the one-factor-at-a-time experiments.Dextrin and sucrose were the best carbon sources for epothilone B production.These two carbon sources, as well as potato powder and glucose, were used in further PB design optimization experiments.Suitable nitrogen sources for the production of epothilone B included soy peptone, soy powder, instant dry yeast and cottonseed powder; these were used as nitrogen sources in further optimization experiments.Accordingly, 17 factors (Table 2), seven from the above one-factor-ata-time experiments (dextrin, sucrose, peptone, instant dry yeast, cottonseed powder, serine, and threonine) and 10 from the components of the EPM culture medium, were used in further PB design optimizations.

Determining elements of the GSM medium by the Box-Behnken(BB) design
To detect combination effects, BB experiments were performed (Table 3) according to the above PB design.The production of epothilone B (the response variable) Where, Y is the epothilone B production; X 3 is the soy power; X 4 is the glucose and X 5 is the sucrose.The high F-value and the very low probability (Table S1 (Appendix)) (p < 0.05) indicates that the experimental model agrees well with the experimental results.Furthermore, the response surface regression showed that the linear coefficient of the polynomial model is highly significant (p < 0.05), but the quadratic coefficient has little significance (p > 0.05); the cross coefficient of the medium components was not significantly associated with production (p >> 0.05).The determination coefficient (R 2 = 0.92) in the experimental model suggested that experimental predictions agreed well with results.The precision and reliability of the experiments were confirmed by the relatively low coefficient of variation (CV = 9.21%).
The significance of each coefficient in the experimental model was determined by the t-value and the p-value using SAS 8.0 (Table S2 (Appendix)).High t-values and low p-values indicate that glucose and soy powder had highly significant effects on epothilone B production; however, the effect of sucrose was not significant.Student's t-test of each coefficient of the model showed that two linear and one quadratic coefficients had significant effects (p < 0.05), but the interactive effect of the three factors was not significant.Comparison of the coefficients in the experimental model also revealed descending levels of significance for the three factors: soy powder > glucose > sucrose.
The effect of these three medium components on epothilone B production was further analyzed using 3D response surface plots, which are graphical representations of the regression model.By simulating the experimental results using the empirical model, these plots efficiently identified the optimal values for the variables.From these plots, it is straightforward to determine the interactions between any two factors and to locate their optimum levels (Figure 2).When epothilone B production was observed as a response variable for the interaction of glucose and sucrose as variables and soy powder as the central point, there is an enhancement in epothilone B production at glucose and sucrose concentrations between the central and maximal levels (Figure 2a).Because epothilone B production decreased beyond this range, the maximal epothilone B production could be obtained at the optimal values of glucose and sucrose.The same procedure was followed for other culture medium components to determine the optimal values of each component (Figure 2b and 2c).Therefore, the experimental model has a stationary point, and the predictive yield of epothilone B is the maximal  /l)  The predicted maximal epothilone B yield and the coded value for each factor were obtained by a canonical analysis of the response surface using SAS.The coded values for the three factors sucrose, glucose and soy powder were -0.581, 1.060 and -0.591, respectively and the predicted epothilone B production was 48.0 mg/l.After translating these coded values, the concentrations of sucrose, glucose and soy powder were calculated as 0.5, 0.8 and 1.7 g/l, respectively; validation experiments were performed in triplicate in shaking flasks, and the resulting yield, 46.5 ± 1.2 mg/l epothilone B, indicated that the experimental model could be employed to predict epothilone B production.

Screening for key factors of GSM medium by fractional factorial design
To search for the optimal GSM medium components for epothilone B production, experiments were designed using the FFD, in which epothilone B yield was used as the response variable (Table 7).The positive influence factors on epothilone B production (p < 0.05) included dextrin (X2), slim milk powder (X6) and MgSO 4 .7H 2 O (X8), which were selected for further optimization in CCD (Table 4).In addition, the results of the response surface curvature analysis showed that all variables had significant effects on the yield of epothilone B and that the value of each variable was in the maximum response region.Therefore, we determined that the central point of the CCD was 0.24% dextrin (X2), 0.12% slim milk powder (X6) and 0.08% MgSO 4 .7H 2 O (X8) , instead of the path with the steepest ascent.Because the remaining factors (sucrose, glucose, soy powder, CaCl 2 , EDTA-Fe 3+ , trace element solution and pH) did not significantly influence epothilone B production (p > 0.05), the concentrations of these factors were not changed in the next experiments.

Determining the optimal medium by central composite design
To observe the effects of combinations of components, experiments were designed using a CCD (Table 5), according to the results obtained via FFD.The results from the variance analysis were obtained (Table S3 (Appendix)).The regression equation showed that the epothilone B yield is an empirical function of the test variables in coded units:  Where; Y is the epothilone B production; X 2 is the dextrin; X 6 is the slim milk powder and X 8 is the MgSO 4 .7H 2 O.The high F-value and the very low probability (p < 0.05) indicated that the experimental model agrees well with the experimental results.The response surface regression showed that the linear coefficient of the polynomial model is highly significant (p < 0.05), while the cross product and quadratic coefficient are less significant (p > 0.05).This suggests that there are some subtle interactions among the three factors.The coefficient of determination (R 2 = 0.92) in the experimental model also suggests that experimental predictions agreed well with results.The precision and reliability of the experiments were confirmed by the relatively low coefficient of variation (CV = 13.84%).
The significance of each coefficient in the experimental model was determined by the tvalues and the probabilities (p-value) using SAS 8.0 (Table S4 (Appendix)).Student's t-test of each coefficient of the model showed that one linear, one quadratic and one cross product coefficient had significant effects on epothilone B production (p < 0.05).Dextrin had a significant effect on the production of epothilone B. There is some interaction between dextrin and slim milk powder for epothilone B production.Comparing the value of each coefficient in the experimental model also revealed descending significances of the three factors, from dextrin to slim milk powder to MgSO 4 . 7H 2 O. Accordingly, 3D graphs were generated for the pairwise combination of the three factors by keeping the third factor as the central point (Figure 3).By simulating the experimental results using the empirical model, these plots efficiently identified the optimal values for the variables.When epothilone B production was observed as a response variable with the interaction between dextrin and slim milk powder as variables with MgSO 4 .7H 2 O at the central point, there was an enhancement in epothilone B production at dextrin and slim milk powder concentrations between the central and the maximal levels (Figure 3a).The maximal epothilone B production could be obtained at the optimal values of dextrin and slim milk powder.The same procedure (Figure 3b and 3c) was used to determine the optimal values for each medium component.Therefore, the experimental model had a stationary point, and the predicted yield of epothilone B was maximal at the stationary point.
Table S3.The results of the regression analysis of central composite design.and the coded value for each factor were obtained by a canonical analysis of the response surface using SAS.

Source
The coded values of the three factors, dextrin, slim milk powder and MgSO 4 .
7H 2 O, were 3.240, 2.367 and 6.029, respectively, and the predicted epothilone B production was 82.8 mg/l.After translating these coded values, the concentrations of dextrin, slim milk powder and MgSO 4 . 7H 2 O were calculated as 6.3, 2.6 and 3.2 g/l, respectively.The calculated optimum conditions for epothilone B yield were verified by culturing So0157-2 in the optimized conditions in triplicate shaking flasks.Under these conditions, So0157-2 produced an epothilone B yield of 82.0 ± 3 mg/l, which agreed with the predicted value of 82.8 mg/l.These data suggest that the model is valid for optimizing epothilone B yield.

DISCUSSION
Generally, each organism has its own nutritional requirement for compound production (Elibol, 2004).The yield of epothilones in Sorangium strains is low (Gerth et al., 1996).Probably because of manipulation limitations (Frykman et al., 2002;Julien and Shah, 2002;Park et al., 2008), there is no report regarding to the optimization of media to improve the epothilone production in the natural Sorangium producers.In the past decade, researchers attempted to bypass the limitation by heterologously expressing the epothilone biosynthetic genes in other fermentation-friendly hosts, such as Streptomyces coelicolor (Tang et al., 2000), Escherichia coli.(Mutka et al., 2006) or Myxococcus xanthus (Julien and Shah,

Figure 1 .
Figure 1.Epothilone and its structural analogues.Shadows show the modified groups based on the epothilone structure.

Table 1 .
Designs and results of one-factor-at-one-time.

Table 2 .
Code value of factors; estimate; standard error; t-value and p-value for the Plackett-Burman design.

Table 4 .
Code value of factors; estimate; standard error; t-value and p-value for fractional factorial experiment.
production (t < 0).The epothilone B yield was reduced when these two factors were included in the culture medium.Factors that positively influenced the production of epothilone B with a high probability (p < 0.05) included soy powder, glucose and sucrose; these were selected for further optimization by RSM.According to the t-tests

Table 5 .
Central composite design and results.

Table S1 .
The results of the regression analysis of Box-Behnken design.

Table 7 .
Fractional factorial design and results.