The IR spectrum of fengycin in KBr reveals bands appearing at 3400 cm⁻¹ for amino- and hydroxyl groups of amino acids. The bands appearing at 2860 cm⁻¹ and 2930 cm⁻¹ reflect the aliphatic side chains and at 2060 cm⁻¹, the phenolic ring of tyrosine. At 1650 and 1520 cm⁻¹ strong bands appeared due to the peptide bonds. The shoulder peak appearing at 1760 cm⁻¹ could be attributed to an ester linkage (Figure S1(a)). The IR spectrum of fengycin from B. subtilis F29-3 was also consistent with the literature (Figure S1(b)) [5].

UV absorption maxima of the fengycin complex at 278 nm in methanol and at 293 nm in alkaline methanolic solution are indicative of tyrosyl peptides (data not shown).

The ¹³C NMR spectrum exhibits carbonyl resonances between 173 and 177 ppm, both of which are carbon signals of various amino acids known from amino acid analyses. The resonances of the various fatty acid chains are found mainly between 10 and 40 ppm (Figures S2(a) and S2(c)), most of which could be assigned by a comparison with published data (Figures S2(b) and S2(d)) [14]. Some of the unsaturated carbon atoms showing resonances at 122.4 and 131.5 can be attributed to olefinic fatty acid residues.

For various homologues of fengycin, the signals responsible for fengycin in MALDI-TOF/MASS spectra ranged from 1435–1529 m/z (Table 1). During HPLC analysis, samples were collected from two to 16 minutes of elution time at one minute intervals and the collected fractions were then subjected to MALDI-TOF/MASS analysis. Table 1 summarizes the mass number of fengycin lipopeptide families observed in the MALDI-TOF mass spectra (data not shown). The mass peak appearing at m/z 1475.8 could be attributed to a fengycin isoform containing a β-hydroxy fatty acid with a chain length of 17 carbon atoms containing one double bond. The compounds with mass numbers of m/z 1497.8 and m/z 1505.8 were identified as fengycins with β-hydroxy fatty acid components possessing the chain lengths of 17 carbon atoms. The first species (m/z 1497.8) is sodium adduct of a C17 isoform with an alanine at position 6. The other compound (m/z 1505.8) is a protonated form of a C17 isoform with a valine instead of an alanine at position 6 (Table 1).

Exactly how seven variables affect fengycin production by B. subtilis F29-3 was analyzed based on fractional factorial design. Table 2 summarizes the regression analysis results of the fractional factorial. The model had a coefficient of determination (R²) of 0.9109, suggesting that the sample variation exceeding 91.09% was attributed to the variables, while the model could not explain only 8.91% of the total variance. The F-value of 11.69 suggested that the model was significant. Moreover, four of the several variables examined, i.e., mannitol, soybean meal, NaNO₃ and MnSO₄·4H₂O, significantly affected fengycin production according to the ‘Prob > F’ value (Table 3) (considering ‘Prob > F’ values of less than 0.05 as significant). Thus, concentrations of mannitol, soybean meal, NaNO₃ and MnSO₄·4H₂O were selected as independent variables to perform response surface analysis. According to the fractional factorial design, the preferable medium composition (g/L) consisted of the following: mannitol, 27.1; soybean meal, 20.8; NaNO₃, 2.5; FeCl₂·4H₂O, 0.55; MgSO₄·7H₂O, 3.0; MnSO₄·4H₂O, 0.1; Na₂MoO₄, 0.055.

Although a highly effective means of screening variables, fractional factorial can neither estimate the optimum levels of the variables, nor determine the appropriate range of the selected variables for response surface method design. Therefore, the steepest ascent method was applied to increase fengycin production. The path of the steepest ascent was determined based on Table 4 to identify the proper direction of changing variables in order to increase fengycin production. According to this table, fengycin production was increased by elevating the concentrations of mannitol and soybean meal as well as by decreasing the concentrations of NaNO₃ and MnSO₄·4H₂O. This table also revealed the yield plateau reached during the third step. Therefore, these variables were selected for further optimization via RSM design.

Based on the results of fractional factorial design and the steepest ascent method, the optimal medium composition was determined based on four variables, i.e., mannitol, soybean meal, NaNO₃ and MnSO₄·4H₂O, which significantly influenced fengycin production, leading to optimization of fengycin production. The optimal levels of the four factors, and exactly how interactions between the four factors affect fengycin production, were determined based on central composite design (CCD) of RSM. The CCD results were analyzed by standard analysis of variance (ANOVA). Table 5 lists the mean predicted and observed responses. Thirty experiments with various combinations of mannitol (X₁), soybean meal (X₂), NaNO₃ (X₃) and MnSO₄·4H₂O (X₄) were performed (Tables 5 and 6). A second order regression equation (Equation 1) describes the levels of fengycin production as a function of initial values of mannitol, soybean meal, NaNO₃ and MnSO₄·4H₂O. Based on the simulation results, the response surface can be estimated by the following equation (Equation 1): (1) Y = 3371.8333 + 18.958333 X 1 + 145.125 X 2 − 229.3021 X 1 2 − 100.1875 X 2 X 1 − 136.5521 X 2 2 − 169.5417 X 3 − 150.625 X 4 − 139.0521 X 3 2 − 40.3125 X 3 X 4 − 194.6771 X 4 2 + 79.0625 X 3 X 1 + 79.8125 X 3 X 2 + 48.81 X 4 X 1 − 20.6875 X 4 2where Y refers to fengycin production, and X₁, X₂, X₃ and X₄ refers to the coded value of mannitol, soybean meal, NaNO₃ and MnSO₄·4H₂O concentration, respectively. Model terms with values of ‘Prob > F’ less than 0.05 are considered significant, whereas those exceeding 0.10 are insignificant. According to the proposed model, three (X₂, X₃ and X₄) out of the four linear terms and all of the squared model terms X₁², X₂², X₃², and X₄² were significant for fengycin production (Table 6). Coefficient of determination (R²) for fengycin production was estimated as 0.9043 (a value of R² > 0.75 indicated the aptness accuracy of the model, which can explain up to 90.43% variability of the response. Next, the optimum level of each variable and exactly how their interactions affect fengycin production were studied by plotting three dimensional response surface curves against any two independent variables, while maintaining other variables at their respective ‘0’ levels. Figures 1(a) to 1(f) display the three dimensional curves of the estimated responses from the interaction between mannitol and soybean, mannitol and NaNO₃, mannitol and MnSO₄·4H₂O, soybean meal and NaNO₃, soybean meal and MnSO₄·4H₂O, and NaNO₃ and MnSO₄·4H₂O, respectively. Estimated results of the response surface model equation indicated that a combination of adjusting the mannitol concentration to 26.2 g/L, increasing the soybean meal concentration to 21.9 g/L, decreasing the NaNO₃ concentration to 3.1 g/L and adjusting the MnSO₄·4H₂O concentration to 0.15 g/L, would maximize fengycin production, yielding a fengycin production of 3.5 g/L. This value is significantly higher than the control value (1.45 g/L) obtained from the SMN medium, indicating that the RSM design strategy markedly improved fengycin production. Confirmation experiments based on optimal medium composition also indicated a fengycin yield of 3.55 g/L, which is consistent with the model estimates.

The purified fengycin powder was analyzed using FAB-MS (Model JMS-700, JEOL, Tokyo, Japan). FAB-MS spectra were collected over a range of 0 to 2,000 m/z.

Once identified on a KRS-5 cell, the purified fengycin powder was analyzed using Fourier transform-infrared spectrum (FT-IR) (PerkinElmer, Paragon 500, U.S.). FTIR spectra were collected between 400 and 4,000 wave numbers (per centimeter).

The purified fengycin powder was analyzed by UV spectrophotometry (Spectronic 601, Milton Roy, U.S.). UV spectra were collected between a range of 200 to 900 nm using methanol as the solvent and between a range of 240 to 900 nm using methanol/0.5 N NaOH (Sigma) as the solvent.

The sample was dissolved in CD₃OD and analyzed by NMR (Bruker, Rheinstetler, Germany).

The purified fengycin powder and all of the fractions collected from HPLC elutant were analyzed using MALDI-TOF/MASS (Bruker, Daltonic, Germany).

The most significant parameters affecting fengycin production by B. subtlis F29-3 were screened based on fractional factorial design. Seven variables, i.e., mannitol, soybean meal, NaNO₃, FeCl₂·4H₂O, MgSO₄·7H₂O, MnSO₄·4H₂O and Na₂MoO₄, were studied in 16 experiments (Table 2). Each variable was represented at two levels, i.e., high and low, as denoted by (+) and (−) signs, respectively. Concentration ranges for the variables were determined based on an extensive literature survey [19]. Finally, experiments were performed following the instruction of a design matrix (Table 3).

The central point and ranges of the variables that significantly influenced fengycin production used for response surface methodology (RSM) experimental design, were determined based on a single steepest ascent experiment (Table 4) [19].

Following selection of the ranges of appropriate variables, the optimum concentration of these variables was determined using RSM to increase fengycin production. Based on a central composite design (CCD) the concentration of the variables, i.e., medium constitutes, was optimized along with their interactions studied. Each variable was then presented at five levels, denoted by (−2), (−1), (0), (+1), (+2), respectively (data not shown). A 2⁴ factorial design was used with eight axial points and six replicates at the center point with a total of 30 experiments (Table 5). Fengycin production was taken as the response (Y), and multiple regression analysis of the data was performed to derive an empirical model that relates the response measured to the independent variables. The system behavior was described according to the following quadratic equation (Equation 2): (2) Y = β 0 + β 1 X 1 + β 2 X 2 + β 3 X 3 + β 4 X 4 + β 11 X 1 2 + β 22 X 2 2 + β 33 X 3 2 + β 44 X 4 2 + β 1 β 2 X 1 X 2 + β 1 β 3 X 1 X 3 + β 1 β 4 X 1 X 4 + β 2 β 3 X 2 X 3 + β 2 β 4 X 2 X 4 + β 3 β 4 X 3 X 4where, Y refers to the predicted response, β₀ refers to the intercept, β_1, β_2, β_3, β₄ refer to the linear coefficients, β_1,1, β_2,2, β_3,3, β_4,4 refer to the squared coefficients, β_1,2, β_1,3, β_1,4, β_2,3, β_2,4, β_3,4 refer to the interaction coefficients, and X₁, X₂, X₃, X₄ refer to the independent variables [20–23].