Enzyme and Microbial Technology 31 (2002) 213–220

Application of statistical experimental methods to optimize production of

poly(␥-glutamic acid) by Bacillus licheniformis CCRC 12826
I.L. Shih a,∗ , Y.T. Van a , Y.N. Chang b
a Department of Environmental Engineering, Da-Yeh University, 112, Shan-Jiau Rd., Da-Tsuen, Chang-Hwa 51505, Taiwan, ROC
b Department of Food Engineering, Da-Yeh University, Chang-Hwa 51505, Taiwan, ROC

Received 18 August 2001; received in revised form 12 February 2002; accepted 13 February 2002

Abstract
Statistical experimental methods (SEM) were employed to study the effects of glutamic acid, citric acid, glycerol and initial medium pH
on the production of poly(␥-glutamic acid) (␥-PGA) by Bacillus licheniformis CCRC 12826 in shaken cultures. The results of first-order
factorial design experiments showed that the liner terms of glutamic acid, citric acid and glycerol had significant positive effects, but the
initial medium pH exhibited insignificant effect on ␥-PGA production. The effects decreased in the order of glycerol, glutamic acid and
citric acid. In addition, the interaction term of glutamic acid–glycerol exhibited a significant positive effect. Based on the results of the
first-order factorial design experiment, the optimum composition was then investigated by using a central composite design (CCD). The
experimental results of CCD were fitted with a second-order polynomial equation by a multiple regression analysis. The coefficient of
determination (R2 ) was 0.9078, the Fisher F-test was significant at upper 5% level and the lack of fit was insignificant at 5% level. All of
these indicated a good adequacy of the second-order polynomial model proposed to explain the data observed.
The optimal ␥-PGA yield (19.80 ± 1.59 g/l) was determined by the CCD experiments and was predicted to be at the regions where
respective concentrations of citric acid, glutamic acid and glycerol were around 24.50, 57.30 and 157.11 g/l, respectively. When the strain
was cultivated using the optimized medium predicted by the model, the yield of ␥-PGA production was 19.62 ± 1.07 g/l (average of three
repeats). The ␥-PGA production by B. licheniformis CCRC 12826 was increased significantly by 372%, from 5.27 to 19.62 g/l when the
strain was cultivated in the optimal medium developed by SEM, as compared to conventional medium E used in the literature. The ␥-PGA
thus produced was shown to be a homogeneous polymer of glutamic acid by thin-layer chromatography and amino acid analysis, and its
molecular weight was over 2 × 106 Da by GPC.
© 2002 Elsevier Science Inc. All rights reserved.
Keywords: Poly(␥-glutamic acid); Statistical experimental methods (SEM); Bacillus licheniformis

1. Introduction biopolymer flocculants [21], heavy metal absorber [22,23],

Poly(␥-glutamic acid) (␥-PGA), is an unusual an-
ionic, naturally occurring, water-soluble polyamide. It is
biodegradable, edible and nontoxic toward humans and the
environment. Therefore, potential applications of ␥-PGA
and its derivatives have been of interest in the past few
years in a broad range of industrial fields such as food,
cosmetics, medicine and water-treatment [1]. ␥-PGA and
its derivatives offer a wide range of unique applications
including being used as thickener [2], humectant [3], bit-
terness relieving agent [4], cryoprotectant [5], drug carrier
[6–9], curable biological adhesive [10–13], biodegradable
fibers [14–17], highly water absorbable hydrogels [18–20],
∗Corresponding author. Tel.: +886-4-8511344; fax: +886-4-8511344.
E-mail address: ils@mail.dyu.edu.tw (I.L. Shih).
0141-0229/02/$ – see front matter © 2002 Elsevier Science Inc. All rights reserved.
PII: S 0 1 4 1 - 0 2 2 9 ( 0 2 ) 0 0 1 0 3 - 5
and animal feed additives [24]. The application of ␥-PGA
is versatile, safe and environmental friendly. Therefore, the
development of this ecomaterial is both economically and
environmentally valuable.
Several bacteria produce ␥-PGA as an extracellular
viscous material [25–28], all of which belong to the
Bacillus genus. In order to enhance the ␥-PGA productivity,
researchers have investigated the nutrient requirements for
␥-PGA production and found that the nutrient requirements
varied according to the strain used. In studies which tried
to clarify the metabolic pathway for PGA synthesis, re-
searchers found that there are different mechanistic systems
for PGA production in different bacteria, indicating that
PGA production is diverse in microorganisms [29]. Accord-
ing to the nutrient requirements, theses bacteria are divided
into two groups: one requires l-glutamic acid as a carbon
214 I.L. Shih et al. / Enzyme and Microbial Technology 31 (2002) 213–220
and nitrogen source for ␥-PGA production and growth, the
other does not require l-glutamic acid for the same pur-
pose. The l-glutamic acid dependent bacteria include B.
subtilis ATCC 9945a [25], B. subtilis IFO3335 [30] and B.
subtilis F-2-01 [28]; the l-glutamic acid independent bac-
teria includes B. subtilis 5E [31], B. subtilis TAM-4 [32]
and Bacillus licheniformis A35 [27]. The l-glutamic acid
independent bacterial usually used citric acid or glucose as
major carbon sources for ␥-PGA production. Although B.
licheniformis 9945a used citric acid and glycerol as carbon
sources, it is however l-glutamic acid dependent. The role
of l-glutamic acid in this example is presumably an acti-
vator for the ␥-PGA enzyme system but not as a carbon
source [33]. Besides carbon sources, factors such as nitro-
gen source, ionic strength, aeration, and medium pH all
affected the productivity and quality of ␥-PGA [34]. For
example, variation of the medium ionic strength by the ad-
dition of up to 4% (w/v) NaCl to cultures of B.licheniformis
9945a led to the formation of ␥-PGA of relatively higher
molecular weight but lower volumetric yield. A recent re-
port showed that pH 6.5 was optimal for attaining increased
␥-PGA yields and maintaining high specific productivity
for B. licheniformis 9945a culture. Besides, citrate utiliza-
tion occurred more rapidly and to great extent at pH 6.5
indicating that the formation of ␥-PGA from citrate at pH
6.5 was of increased importance [34].
Recently, we have found that B. licheniformis CCRC
12826 produced an excellent bioflocculant, which was
later proved to be a homopolymer of glutamic acid [35].
In addition, B. licheniformis CCRC 12826 required the
presence of glutamic acid, citric acid and glycerol in the
culture medium for PGA production. We attempted to
further enhance the production of ␥-PGA by B. licheni-
formis CCRC 12826. Therefore, in this study we applied
statistical experimental methods (SEM) to determine the
effects of three medium components (glutamic acid, cit-
ric acid and glycerol) and initial medium pH on ␥-PGA
production by this microorganism. SEM is a powerful tech-
nique for testing multiple process variables because fewer
experimental trials are needed compared to the study of
one variable at a time. Also, interactions between vari-
ables can be identified and quantitated by such a technique
[36].
2. Materials and methods
2.1. Materials
Reagents for cultivation such as nutrient agar (NA) and
nutrient broth (NB) were purchased from DIFCO Lab-
oratories (Detroit, MI). Glutamic acid, citric acid, glyc-
erol, NH4 Cl, MgSO4 ·7H2 O, FeCl3 ·6H2 O, K2 HPO4 , and
CaCl2 ·2H2 O were obtained from Sigma. All other reagents
used were of the highest grade available unless otherwise
indicated. B. licheniformis CCRC 12826 used in this study
was obtained from the Culture Collection and Research
Center (CCRC) Taiwan.
2.2. Production of γ -PGA in medium E
Medium E (l-glutamic acid (20 g/l), citric acid (12 g/l),
glycerol (80 g/l), NH4 Cl (7 g/l), MgSO4 ·7H2 O (0.5 g/l),
FeCl3 ·6H2 O (0.04 g/l), K2 HPO4 (0.5 g/l), CaCl2 ·2H2 O
(0.15 g/l), and MnSO4 ·H2 O (0.04 g/l)) is a medium of
choice for many researchers investigating production of
␥-PGA by Bacillus species [33,34,37]. For initial experi-
ments on growth and ␥-PGA production, B. licheniformis
CCRC 12826 was cultivated in medium E.
B. licheniformis CCRC 12826, obtained as a lyophilized
powder in a glass ampoule sealed under vacuum, was first
cultured on NA plates to induce spore formation. The NA
containing agar (15 g/l), beef extract (3 g/l), peptone (5 g/l)
was prepared by suspending 23 g of Bacto® NA powder
in 1 l deionized water, boiled to dissolve completely, and
sterilized at 122–124 ◦ C for 15 min. The final pH was kept
at 7.2 ± 0.2 at 25 ◦ C. The bacterial powder was suspended
in 0.2 ml of sterilized water and one loopful (3 mm di-
ameter loop) of bacterial suspension was spread onto the
NA plate and then incubated. After overnight incubation
(37 ◦ C, pH 7.4), highly mucoid colonies that appeared on
the plates were picked up (1 cm squares) and inoculated
into 5 ml of NB (pH 7.4) in a 30 ml test tube, followed by
incubation at 37 ◦ C for 48 h with shaking at 150 rpm. The
NB, composed of beef extract (3 g/l) and peptone (5 g/l),
was prepared by suspending 8 g of NB powder (Difco
Laboratories) in 1 l of deionized water and sterilizing at
122–124 ◦ C for 15 min. The spores in the broth were inoc-
ulated (5%, v/v) into 100 ml of the medium E (l-glutamic
acid (20 g/l), citric acid (12 g/l), glycerol (80 g/l), NH4 Cl
(7 g/l), MgSO4 ·7H2 O (0.5 g/l), FeCl3 ·6H2 O (0.04 g/l),
K2 HPO4 (0.5 g/l), CaCl2 ·2H2 O (0.15 g/l), and MnSO4 ·H2 O
(0.04 g/l)) in a 500 ml flask and incubated at 37 ◦ C, pH
6.5 with shaking at 150 rpm. Periodically, aliquots were
removed from cultures and used to determine cell growth
by measurement of optical density at 660 nm or to mea-
sure the viscosity. In our previous paper, we have shown
that the course of ␥-PGA production and the relative vis-
cosity of the medium increased with the culture time and
reached a maximum after 96 h of incubation [35]. There-
fore, the viscosity of the medium was monitored and used
as an indicator of ␥-PGA production. The concentration
of citric acid and glutamic acid was determined by high
performance liquid chromatography (HPLC, equipped with
a Hitachi 6200 solvent delivery system and a Hitachi 4250
UV detector) on a reverse phase ODS column (Phenomnex,
USA, 250 mm × 4.6 mm, 5 ␮m). The sample was eluted
with a mobile phase consisting of methanol and 0.2%
aqueous phosphate, pH 2.4 (1:9, v/v). The flow rate was
set at 0.2 ml/min, and the eluant was monitored at 220 nm.
After cultivation for 96 h, a viscous culture broth was ob-
tained which was centrifuged at 20,000 × g to separate the
 I.L. Shih et al. / Enzyme and Microbial Technology 31 (2002) 213–220
cells. The viscous materials were further purified by the
procedures described further.
2.3. γ -PGA purification
The viscous culture broth obtained from above was diluted
with the same volume of distilled water and then centrifuged
at 20,000 × g for 15 min. The supernatant was poured into
four volumes of cold ethanol and to precipitate the ␥-PGA.
The resultant precipitate was collected by centrifugation at
25,000×g for 15 min and redissolved in distilled water. After
three such ethanol precipitation steps, the crude ␥-PGA thus
obtained was dialyzed at 4 ◦ C overnight in deionized water
and lyophilized to give pure material.
2.4. Characterization of γ -PGA
2.4.1. Molecular weight determination
The number-average molecular weight (Mn ) of the
␥-PGA was measured by gel permeation chromatography
(GPC) using a Hitachi L6200 system controller equipped
with Shodex KB800 series columns (two KB80M, one
KB802.5), a refractive index (RI) detector (Bischoff, model
8110) and a waters model 730 data module. Dextrans
(polysaccharides) standards obtained form Phenomenex
Inc., USA were used to construct a calibration curve from
which molecular weights of ␥-PGA were calculated with
no further correction. The eluant contained 0.3 M Na2 SO4 ,
0.05% (w/v) NaN3 was brought to a pH of 4.0 using glacial
acetic acid and the flow rate was set at 1.0 ml/min.
2.4.2. Amino acid analysis and thin-layer chromatography
The purified material was hydrolyzed with 6N HCl at
110 ◦ C for 24 h in a sealed and evacuated tube, and the
amino acid compositions were determined with a Beckman
system 6300E analyzer using a column (0.9 cm × 55 cm)
packed with Beckman PA-35 resin. Thin-layer chromatogra-
phy was performed on a cellulose plate (Merck) with solvent
systems of butanol–acetic acid–water (3:1:1, w/w) and 96%
ethanol–water (63:37, w/w). Amino acids were detected by
spraying with 0.2% ninhydrin in acetone [38,39].
2.4.3. Viscosity measurement
Viscosity of cell-free culture fluid was measured with
Brookfield Digital Rheometer (model DV-III, USA)
equipped with a spindle SC4-34 at different shear rates
(0.28–56.0 s−1 ). Silicon oil (99.8 cP at 25 ◦ C, Brookfield
Engineering laboratory Inc., USA) was used as a viscosity
standard.
2.4.4. Sugar content
The total carbohydrate content of the ␥-PGA was de-
termined by the phenol–sulfuric acid method [40] and ex-
pressed as the glucose equivalent. The protein moiety in the
purified material was determined by the Bradford method
[41] with bovine serum albumin as a standard.
215
2.5. SEM experimental design
Application of SEM requires the identification of the
major factors that are suitable to sustain good production
of ␥-PGA. In the previous studies of PGA production by
B. licheniformis CCRC 12826 [35] or B. licheniformis
9945a [25,34], it was indicated that the levels of citric
acid, glutamic acid, glycerol, and pH were the major vari-
ables affecting the performance of the culture in terms of
␥-PGA yields. Therefore, these four factors were chosen
for further optimization through SEM. Initially, a complete
four-factor-two-level factorial design was carried out, which
was followed by the central composite design (CCD) to
optimize the factors for ␥-PGA production.
2.6. Factorial design
In the first experiment of this series, the ranges of the
variables tested were citric acid 6–18 g/l, glutamic acid
10–50 g, glycerol 40–120 g/l, pH 6–7. For the first phase of
the optimization process in which the region close to the
optimum is to be approached, a two-level factorial design
was chosen. In this experimental design the main effects and
interactions of different factors, each at two different lev-
els, can be simultaneously investigated. For a full factorial
design, all possible combinations of the two levels of the
independent variables were investigated. For a 24 factorial
design with 4 factors at 2 levels, 16 experimental runs are
required [36]. Two center points were added to estimate the
experimental error and check the adequacy of the first-order
model. Table 1 shows the four independent variables and
their concentrations at the different coded levels of the fac-
torial design experiments. The matrix corresponding to the
24 factorial designs, together with the observed experimen-
tal data and predicted values from the model equation is
also shown in Table 1. To avoid bias, a total of 18 runs was
performed in a random order (overall randomization). The
statistical analyses of data obtained from the 24 factorial
designs of experiments are shown in Tables 2 and 3.
2.7. Central composite design
Based on the results obtained from the factorial design,
the CCD was conducted in the optimum vicinity to locate
the true optimum concentrations of citric acid, glutamic acid
and glycerol for ␥-PGA production. The pH was excluded
in CCD design because it showed little effect on ␥-PGA
production according to the results of factorial experiments
which will be described further. For the three factors, this
trial was essentially a full 23 factorial design augmented by
six axial points (or called star points) coded ± α and two
replications of center point (all factors at level 0), resulting
in a total number of 16 experiments [36]. The distance of
the star points from the center point is given by α = 2n/4
(for three factors n = 3, α = 1.682). The matrix corre-
sponding to the CCD is shown in Table 4, together with the
216 I.L. Shih et al. / Enzyme and Microbial Technology 31 (2002) 213–220

Table 1
Experimental design and results of the four-factor-two-level factorial design together with the predicted yields from the model equation
Trial number Factors Observed ␥-PGA Predicted ␥-PGA
yield (g/l) yield (g/l)
Citric acid (g/l) Glutamic acid (g/l) Glycerol (g/l) pH

1 6 10 40 6 0.0276 −0.5670
2 18 10 40 6 0.1461 0.8419
3 6 50 40 6 0.4634 0.4135
4 18 50 40 6 1.3252 1.8024
5 6 10 120 6 1.7170 2.3835
6 18 10 120 6 6.2316 5.9924
7 6 50 120 6 7.0059 7.5124
8 18 50 120 6 11.5067 11.1013
9 6 10 40 7 0.0246 0.7498
10 18 10 40 7 0.1912 −0.1066
11 6 50 40 7 0.7537 1.2016
12 18 50 40 7 0.6718 0.3250
13 6 10 120 7 3.0818 2.8132
14 18 10 120 7 3.7870 4.1567
15 6 50 120 7 7.7895 7.4134
16 18 50 120 7 7.9334 8.7368
17a 12 30 80 6.5 4.1352 3.4231
18a 12 30 80 6.5 4.8255 3.4231
a Center point.

Table 2 observed experimental data and predicted values from the

Results of regression analysis of the 24 factorial design experiments
model equation.
Parameter Coefficient Standard error t ratio Probability

Intercept 3.42318∗ 0.22529∗ 15.19456∗ 0.000001∗ 2.8. Shake-flask cultivation for all
Curvature 0.8442 1.75953 0.95954 0.381356 optimization experiments
X1 0.6831∗ 0.44912∗ 2.85866∗ 0.024384∗
X2 1.3902∗ 0.44912∗ 5.81768∗ 0.000652∗
B. licheniformis CCRC 12826 was cultured on NA plates
X3 2.8406∗ 0.44912∗ 11.88747∗ 0.000007∗
X4 −0.2619 0.44912 −1.09604 0.309329 and transferred into NB as described previously. The spores
X 1 X2 −0.0050 0.44912 −0.02100 0.983830 in the broth were inoculated (5%, v/v) into a 500 ml flask
X 1 X3 0.5500 0.44912 2.30115 0.054866 containing 100 ml of medium E with various pH and vari-
X 1 X4 −0.5664 0.44912 −2.37018 0.049593 ous concentrations of citric acid, glutamic acid and glycerol
X 2 X3 1.0371∗ 0.44912∗ 4.34011∗ 0.003396∗
according to the experimental design. The inoculated flasks
X2 X4 −0.1322 0.44912 −0.55321 0.597334
X 3 X4 −0.2218 0.44912 −0.92813 0.384229 were incubated at 37 ◦ C with continuous shaking at 150 rpm
for 96 h. The viscous ␥-PGA was further isolated and puri-
R2 : 0.9646. X1 : citric acid, X2 : glutamic acid, X3 : glycerol, X4 : pH. fied by the procedures described previously and the ␥-PGA
∗ Significant at the 5% level.
yields were calculated.

Table 3 2.9. Software for experimental design and statistical

ANOVA table of 24 factorial design experiments analysis
Factor SS d.f. MS F P
Statistica, Version 5.0 (Statsoft Inc., Tulsa, OK) was used
Model 196.82 10 19.68 21.54∗
for the experimental design and regression analysis of the
Linear 168.57 4 42.15 46.14∗ experimental data obtained. The data of ␥-PGA production
Citric acid 7.46 1 7.46 8.17 0.11
Glutamic acid 30.92 1 30.92 33.84 0.05
obtained from SEM experiments were further analyzed
Glycerol 129.10 1 129.10 141.31 0.02 by the REG or RSREG (Response Surface REGression)
pH 1.09 1 1.09 1.19 0.27 procedure of Statistical Analysis System (SAS) [42]. The
Cross product 28.24 6 4.70 5.14∗ quality of fit of the model equation was expressed by the
coefficient of determination (R2 ), and its statistical signifi-
Residual 6.39 7 0.91
Lack of fit 6.15 6 1.02 4.30 0.35
cance was determined by an F-test. The significance of the
Pure error 0.23 1 0.23 regression coefficients was tested by a t-test. For analysis
of the nature of the fitted response and for prediction of the
Total SS 203.23 17
maximum point, the second-order equation was reduced
∗ Significant at 5% level. to its canonical form [43,44]. Canonical analysis was one
 I.L. Shih et al. / Enzyme and Microbial Technology 31 (2002) 213–220 217

Table 4
Experimental design and results of the CCD together with predicted yields from the model equation
Trial number Factors Observed ␥-PGA Predicted ␥-PGA
yield (g/l) yield (g/l)
X1 (citric acid (g/l)) X2 (glutamic acid (g/l)) X3 (glycerol (g/l))

1 14 45 130 8.9285 9.7936

2 14 45 170 8.0028 6.2971
3 14 65 130 17.2343 17.7647
4 14 65 170 18.5571 18.3661
5 22 45 130 15.770 15.0004
6 22 45 170 15.3543 13.8633
7 22 65 130 15.4728 16.2179
8 22 65 170 21.0043 19.1786
9 11.3 55 150 14.7228 14.5577
10 24.7 55 150 18.0957 19.6193
11 18 38.2 150 5.4871 6.8680
12 18 71.8 150 18.0628 18.0404
13 18 55 116.4 15.5128 14.23449
14 18 55 183.6 11.1471 13.7840
15a 18 55 150 13.3214 13.8106
16a 18 55 150 14.5328 13.8106
a Center point.

part of the RSREG SAS output. The ridge max option
was used to compute the estimated ridge of optimum re-
sponse for increasing radii from the center of the origin
design.
3. Results and discussion
3.1. Production of γ -PGA in medium E
Medium E, consisted of l-glutamic acid (20 g/l),
citric acid (12 g/l), glycerol (80 g/l), NH4 Cl (7 g/l),
MgSO4 ·7H2 O (0.5 g/l), FeCl3 ·6H2 O (0.04 g/l), K2 HPO4
(0.5 g/l), CaCl2 ·2H2 O (0.15 g/l), and MnSO4 ·H2 O (0.04 g/l),
is a medium of choice for many researchers investigating
production of ␥-PGA by Bacillus species. We showed pre-
viously that substitution of glutamic acid, citric acid, glyc-
erol or ammonium chloride in medium E for other carbon
and nitrogen sources resulted in a significant decrease of
␥-PGA production by B. lciheniformis CCRC 12826 [35].
It was concluded that glutamic acid, citric acid, glycerol
and ammonium chloride were still the most suitable carbon
and nitrogen sources for the production of ␥-PGA by B.
licheniformis CCRC 12826, and that the effects of these
three carbon sources on ␥-PGA production were synergistic.
Nevertheless, when B. licheniformis CCRC 12826 was cul-
tivated in medium E, the amount of ␥-PGA produced was
only 5.27 g/l.
The number-average molecular weight (Mn ) of ␥-PGA
thus obtained was determined by GPC and found to be
over 2 × 106 Da. ␥-PGA was further characterized by
amino acid analysis and thin-layer chromatography. The
6N HCl hydrolysate of the purified material was composed
solely of glutamic acid. Thin-layer chromatography of the
hydrolysate performed on a cellulose thin-layer plate and
visualized with 0.2% ninhydrin indicated a single spot with
a Rf value identical to that of authentic glutamic acid. Fur-
thermore, the ninhydrin and Biuret reactions for this viscous
material were negative. The purified ␥-PGA was tested by
the phenol–sulfuric acid method for polysaccharide con-
tent. Reduced to glucose, less than 1% (w/w) of total sugar
was detected indicating that B. licheniformis CCRC 12826
scarcely produced polysaccharides.
It is known that the molecular weight of ␥-PGA varies
from 100,000 to 2,000,000 Da depending upon the species
and the cultivation conditions used for its production
(Kunioka, 1997; [29]). Under the conditions used in this
study the molecular weight of ␥-PGA produced by B.
licheniformis CCRC 12826 was over 2 × 106 Da by GPC.
To our knowledge, ␥-PGA with a molecular weight over
2 million Da was rarely produced by B. licheniformis. Al-
though little is known of the physiology and biochemistry
of ␥-PGA formation, several factors in the culture medium
have been indicated to affect the molecular weight of
␥-PGA. For example, ␥-PGA molecular weight could be
modulated by medium ionic strength (NaCl concentration)
in the B. licheniformis 9945a culture and increased agitation
was found to decrease molecular weight of ␥-PGA pro-
duced by B. licheniformis 9945a. In addition, the addition
of ammonium sulfate to the medium for PGA production
by B. subtilis IFO3335 is important to obtain ␥-PGA with a
high molecular weight. Whether the above factors affect the
molecular weight of ␥-PGA produced by B. licheniformis
CCRC 12826 is yet to be determined.
It was indicated in some instances that ␥-PGA produc-
tion was complicated by the production of an extracel-
lular polysaccharide by-product. For example, B. subtilis
IFO3335 produced ␥-PGA together with a polysaccharide
218 I.L. Shih et al. / Enzyme and Microbial Technology 31 (2002) 213–220
by-product when it was grown in the medium contain-
ing glucose, l-malic acid, succinic acid or fumaric acid.
However, it was generally found that in the glutamic
acid/citric acid medium, polysaccharide by-products were
scarcely produced. Therefore, It is of no surprise that the
little amount of polysaccharide by-product was produced in
association of ␥-PGA production by B. licheniformis CCRC
12826 in glutamic acid/citric acid medium as suggested by
the data.
3.2. Statistical experimental design
Response surface methodology (RSM) is a sequential
procedure with an initial objective to lead the experimenter
rapidly and efficiently along a path of improvement toward
the general vicinity of the optimum. Although two-level
(full or fractional) factorial experiments will only yield data
to fit a limited model (Eq. (1) or (2)), they are the most
common initial experiments in the application of RSM,
because orthogonality of the design minimizes the vari-
ance of the regression coefficients. Besides, any first-order
(two-level) orthogonal design is rotatable [43,44]. Since
the location of the optimum is unknown prior to running
RSM experiment, it is conceivable to use a design with
rotability that ensures equal precision of estimation in all
direction.

k
Y = β0 + βi xi (1)
i=1

k



Y = β0 + βi xi + βij xi xj
i=1 i<j
Y is the predicted response (␥-PGA yield in this study
(g/l)); β 0 , β i , β ij are constant coefficients, and xi , xj are
the coded independent variables or factors.
3.3. Factorial design experiments
Table 1 is a complete four-factor-two-level (24 ) facto-
rial experiment design augmented with two center points
for ␥-PGA production. Repeated observations at the center
were used to estimate the experimental error and to allow
for checking the adequacy of the first-order model. The four
factors are citric acid (X1 ), glutamic acid (X2 ), glycerol
(X3 ), and pH (X4 ), and their upper and lower levels in this
initial design were chosen in reconciliation with the data of
literature working on ␥-PGA production [25,35]. The ex-
perimental results of ␥-PGA productions are also shown in
Table 1. In order to approach the vicinity of the optimum,
a first-order model was fitted to the data obtained from
the factorial design experiment. From the analysis of the
data in Table 1 by the least-squares method, the regression
coefficients and corresponding t-values for the model were
obtained which are shown in Table 2. The fitted model with
coded variables is shown in Eq. (3).
3.3.1. First-order with interaction model
Y(PGA,g/l) = 3.423 + 0.68X1 + 1.39X2 + 2.84X3
−0.26X4 − 0.005X1 X2 + 0.55X1 X3
−0.57X1 X4 + 1.03X2 X3 − 0.13X2 X4
−0.22X3 X4 (3)
The analyses of variances (Table 3) suggested that the
first-order with two-factor interactions model appeared to
be more adequate than the pure first-order model. This re-
sult is in agreement with the fact that interaction effects of
citric acid, glutamic acid, and glycerol during the ␥-PGA
fermentation were obvious as indicated by the previous re-
ports [25,33,35]. According to the test statistics, F-value for
the overall regression is significant at 5% level and the lack
of fit is insignificant indicating that the first-order model is
very adequate in approximating the response surface of the
experimental design. This statement is further supported by
the satisfactory value of the R2 : 0.9646. Judging from the re-
gression coefficients and the corresponding t-values shown
in Table 2, it is concluded that the linear terms of glutamic
acid, citric acid and glycerol had significant positive effect
on ␥-PGA production, and the effects decreased in the order
of glycerol, glutamic acid and citric acid. In contrast, the pH
exhibited insignificant effect on ␥-PGA production; it was
then omitted from the experiments of CCD in the study.
3.4. Central composite design
The results of the four-factor-two-level (24 ) factorial
(2) experiments indicated that the first-order with two-factor in-
teractions model (R2 : 0.9646) appeared to be more adequate
than the pure first-order model (R2 : 0.8416). Therefore, the
linear model was not adequate to represent the system be-
cause a curvature existed despite the fact that the curvature
check [45], the comparison of the measurements for the
dependent variable at the center point with the average for
the rest of the design, performed using Statistica showed
that the curvature is insignificant at 5% level (Table 2). To
explore the subregion of the response surface in greater
detail requires a model of second-order or higher to ap-
proximate the response. The full quadratic model (Eq. (4))
was most frequently used to approximate the high-order
response and was adequate in most cases [43,44]. The CCD
was chosen in this study to fit a full quadratic equation and
its experimental results together with predicted values from
the model Eq. (5) are shown in Table 4.
k

k




Y = β0 + βi xi + βii xi xj + βij xi xj (4)
i=1 i=1 i<j
Y is the predicted response (␥-PGA yield in this study (g/l));
β 0 , β i , β ii , β ij are constant coefficients, and xi , xj are the
coded independent variables or factors.
By applying multiple regression analysis on the experi-
mental data shown in Table 4, the experimental results of
 I.L. Shih et al. / Enzyme and Microbial Technology 31 (2002) 213–220 219

Table 5
Estimated ridge of maximum response for ␥-PGA production
Coded radius Estimated response (g/l) Standard error Citric acid Glutamic acid Glycerol

0 13.8062 1.4201 17.9950 55.0000 149.9950

0.1 14.3958 1.4095 18.2752 56.5287 150.0582
0.2 14.9418 1.3785 18.5681 58.0300 150.5616
0.3 15.4523 1.3306 18.8929 59.4484 151.5830
0.4 15.9382 1.2720 19.3266 60.6334 153.1180
0.5 16.4190 1.2133 20.1441 61.0086 154.7820
0.6 16.9381 1.1709 21.2206 60.3497 155.7006
0.7 17.5294 1.1664 22.1572 59.5443 156.1930
0.8 18.2019 1.2228 22.9905 58.7937 156.5499
0.9 18.9577 1.3559 23.7649 58.0259 156.8481
1.0 19.7981 1.5689 24.5027 57.3054 157.1146

the CCD were fitted with the polynomial Eq. (4), and the
second-order polynomial equation obtained for ␥-PGA pro-
duction is shown in Eq. (5).
3.4.1. Second-order model equation
Y(PGA,g/l) = 13.8105 + 1.5X1 + 3.32X2 − 0.13X3
+1.15X12 − 0.48X22 + 0.07X32 − 1.69X1 X2
+0.59X1 X3 + 1.02X2 X3
This fit of the model was checked by the R2 , which
was calculated to be 0.9078, indicating that 90.78% of
the variability in the response could be explained by the
model. The significant variables can be identified by means
of their Student t-test values. The results revealed that the
first-order term of glutamic acid displayed significant effect
on the ␥-PGA yield at a 5% level. However, all the other
terms presented insignificant effects on the ␥-PGA yield
at a 5% level. Nevertheless, these terms were not omitted
from the model equation since its adequate fit could be
confirmed. The test statistics F-values for the overall re-
gression is significant at the upper 5% level and the lack
of fit is insignificant. These facts further supported that the
second-order model is very adequate in approximating the
response surface of the experimental design.
For the analysis of the fitted surface, Eq. (5) was trans-
formed into its canonical form, which is shown in Eq. (6).
Y − 14.732 = 4.291507Z12 + 1.007Z22 − 3.179Z32
The coefficients of the Eq. (6) are eigenvalues based on
coded data, and Y is the ␥-PGA yield (g/l). The mixed signs
of the coefficients in the above equation indicated that the
predicted response surface of the stationary point of ␥-PGA
yield is shaped like a saddle. The saddle point appeared at
the point where the concentrations of citric acid, glutamic
acid and glycerol were at 18.98, 58.45 and 99.62 g/l, respec-
tively. Because the canonical analysis resulted in a saddle
point, the estimated surface did not exist with a unique opti-
mum. Therefore, this analysis could not be used to identify
the optimum conditions. The ridge analysis method com-
putes the estimated ridge of optimum response for increas-
ing radii from the center of the original design. The ridge
analysis (Table 5) indicated that the maximum ␥-PGA yield
(19.80 ± 1.59 g/l), where concentrations of citric acid, glu-
tamic acid and glycerol were at 24.50, 57.30 and 157.11 g/l,
respectively was found at the distance of the coded radius
1.0. The stationary point (14.73 g/l) derived from canonical
analysis was much lower than the maximum point (19.80 g/l)
(5) from ridge max analysis. Therefore, the ridge max point
is recommended as the optimal condition in this study.
Verification of the calculated optimum was done with a
culture medium representing this optimal point and yielding
19.62±1.07 g/l (average of three repeats). The excellent cor-
relation between predicted and experimental values justifies
the validity of the response model and the existence of an
optimum point.
This study showed that ␥-PGA production by B. licheni-
formis CCRC 12826 was increased significantly by 372%,
from 5.27 to 19.8 g/l when the strain was cultivated in the
optimal medium developed by SEM, as compared to con-
ventional medium E used in the literature [37]. Therefore,
the SEM proved to be a powerful and useful tool for en-
hancing ␥-PGA production. The major goal of the fermen-
tation research is to identity the environmental factors and
the corresponding levels that combinedly produce the opti-
mum or near optimum result. It is our intention to advance
the ␥-PGA production of B. licheniformis CCRC 12826 by
(6) including other crucial factors, such as rate of aeration, ionic
strength, the concentration and types of various trace ele-
ments, etc. in the future experiments.
Acknowledgment
This work was supported partially by Grant NSC90-2313-
B-212-001 from National Science Council of ROC. We
would like to thank Prof. Chwen-Jen Shieh for the enthu-
siasm he showed for our work as well as for performing
statistical analysis for us.
220 I.L. Shih et al. / Enzyme and Microbial Technology 31 (2002) 213–220
References
[1] Shih IL, Van YT. The production of poly(␥-glutamic acid) from
microorganisms and its various applications. Bioresour Technol
2001;79:207–25.
[2] Yamanaka S. New gamma-polyglutamic acid, production therefore
and drinking agent containing the same. JP Patent 3047087 (1991).
[3] Hasebe K, Inagaki M. Preparation composition for external use
containing gamma-polyglutamic acid and vegetable extract in
combination. JP Patent 11240827 (1999).
[4] Sakai K, Sonoda C, Murase K. Bitterness relieving agent. JP Patent
WO0021390 (2000).
[5] Mitsuiki M, Mizuno A, Tanimoto H, Motoki M. Relationship between
the antifreeze activities and the chemical structures of oligo- and
poly(glutamic acid)s. J Agric Food Chem 1998;46:891–5.
[6] Li C, Yu DF, Newman A, Cabral F, Stephens C, Hunter N, et
al. Complete regression of well-established tumors using novel
water-soluble poly(l-glutamic acid)-paclitaxel conjugate. Cancer Res
1998;58:2404–9.
[7] Li C, Price JE, Milas L, Hunter NR, Ke S, Tansey W, et al.
Antitumor activity of poly(l-glutamic acid)-paclitaxel on syngeneic
and xenografted tumors. Clin Cancer Res 1999;5:891–7.
[8] Markland P, Amidon GL, Yang VC. Modified polypeptides
containing ␥-benzyl glutamic acid as drug delivery platforms. Int J
Pharm 1999;178:183–92.
[9] Dekie L, Toncheva V, Dubruel P, Schacht EH, Barrett L, Seymour
LW. Poly(l-glutamic acid derivatives as vectors for gene therapy. J
Control Release 2000;65:187–202.
[10] Otani Y, Tabata Y, Ikada Y. A new biological glue from gelatin and
poly(l-glutamic acid). J Biomed Mater Res 1996;31:157–66.
[11] Otani Y, Tabata Y, Ikada Y. Rapidly curable biological glue composed
of gelatin and poly(l-glutamic acid). Biomaterials 1996;17:
1387–91.
[12] Otani Y, Tabata Y, Ikada Y. Effect of additives on gelation
and tissue adhesion of gelatin-poly(l-glutamic acid). Biomaterials
1998;19:2167–73.
[13] Otani Y, Tabata Y, Ikada Y. Hemostatic capability of rapidly curable
from gelatin-poly(l-glutamic acid), and carbodiimide. Biomaterials
1998;19:2091–8.
[14] Giannos SA, Shah D, Gross RA, Kaplan DL, Mayer JM.
Poly(glutamic acid) produced by bacterial fermentation. In: Dawes
EA, editor. Novel biodegradable microbial polymers. Dordrecht:
Kluwer Academic Publishers, 1990. p. 457–60.
[15] Kubota H, Nambu Y, Endo T. Convenient and quantitative
esterification of poly(␥-glutamic acid)produced by microorganism. J
Polym Sci Part A Polym Chem 1993;31:2877–8.
[16] Borbley M, Nagasaki Y, Borbely J, Fan K, Bhogle A, Sevoian M.
Biosynthesis and chemical modification of poly(␥-glutamic acid).
Polym Bull 1994;32:127–32.
[17] Yahata K, Sadanobu J, Endo T. Preparation of poly-␣-benzyl-
␥-polyglutamate fiber. Polym Prep Jpn 1992;41:1077.
[18] Kunioka M. Properties of hydrogels prepared by ␥-irradiation
in microbial poly(␥-glutamic acid) aqueous solutions. Kobunshi
Ronbunshu 1993;50:755–60.
[19] Choi HJ, Kunioka M. Preparation conditions and swelling equilibria
of hydrogel prepared by ␥-irradiation from microbial poly(␥-glutamic
acid). Radiat Phys Chem 1995;46:175–9.
[20] Choi HJ, Yang R, Kunioka M. Synthesis and characterization of
pH-sensitive and biodegradable hydrogels prepared by ␥-irradiation
using microbial poly(␥-glutamic acid) and polu (ε-lysine). J Appl
Polym Sci 1995;58:807–14.
[21] Yokoi H, Arima T, Hirose J, Hayashi S, Takasaki Y. Flocculation
properties of poly(gamma-glutamic acid) produced by Bacillus
subtilis. J Ferment Bioeng 1996;82:84–7.
[22] McLean RC, Wolf DC, Ferris FG, Beveridge TJ. Metal-binding
characteristics of the gamma-glutamyl capsular polymer of Bacillus
licheniformis ATCC 9945. Appl Environ Microbiol 1990;56:
3671–7.
[23] McLean RC, Beauchemin D, Beveridge TJ. Influence of oxidation
state on iron binding by Bacillus licheniformis capsule. Appl Environ
Microbiol 1992;58:405–8.
[24] Tanimoto H, Sato H, Karasawa M, Iwasaki K, Oshima A, Adachi
S. Feed composition containing poly-␥-glutamic acid. JP Patent
WO9635339 (2000).
[25] Thorne CB, Gomez CG, Noyes HE, Housewright RD. Production of
glutamyl polypeptide by Bacillus subtilis. J Bacteriol 1954;68:307–
15.
[26] Troy FA. Chemistry and biosynthesis of the poly(␥-d-glutamyl)
capsule in Bacillus subtilis. 1. Properties of the membrane-mediated
biosynthetic reaction. J Biol Chem 1973;248:305–16.
[27] Cheng C, Asada Y, Aaida T. Production of ␥-polyglutamic acid by
Bacillus subtilis A35 under denitrifying conditions. Agric Biol Chem
1989;53:2369–75.
[28] Kubota H, Matsunobu T, Uotani K, Takebe H, Satoh A, Tanaka T, et
al. Production of poly(␥-glutamic acid) by Bacillus subtilis F-2-01.
Biosci Biotechnol Biochem 1993;57:1212–3.
[29] Kunioka M, Furusawa K. Poly(␥-glutamic acid) hydrogel prepared
from microbial poly(␥-glutamic acid) and alkane diamine with
water-soluble carbodiimide. J Appl Polym Sci 1997;65:1889–93.
[30] Goto A, Kunioka M. Biosynthesis and hydrolysis of poly(␥-glutamic
acid) from Bacillus subtilis IFO3335. Biosci Biotechnol Biochem
1992;56:1031–5.
[31] Murao S. On the polyglutamic acid fermentation. Kobunshi
1969;16:1204–12.
[32] Ito Y, Tanaka T, Ohmachi T, Asada Y. Glutamic acid independent
production of poly(␥-glutamic acid) by Bacillus subtilis TAM-4.
Biosci Biotechnol Biochem 1996;60:1239–42.
[33] Birrer GA, Cromwick AM, Gross RA. ␥-Poly(glutamic acid)
formation by Bacillus licheniformis: physiological, (␥-poly(glutamic
acid) formation by Bacillus licheniformis: physiological and
biochemical studies. Int J Biol Macromol 1994;16:265–75.
[34] Cromwick AM, Birrer GA, Gross RA. Effects of pH and aeration
on ␥-poly(glutamic acid) formation by Bacillus licheniformis in
controlled batch fermentor cultures. Biotechnol Bioeng 1996;50:222–
7.
[35] Shih IL, Van YT, Yeh LC, Chang YN. Production of a biopolymer
flocculant from Bacillus licheniformis and its flocculation properties.
Bioresour Technol 2001;78:267–72.
[36] Box GEP, Wilson KB. On the experimental attainment of optimum
conditions. J Roy Statistical Soc (Ser B) 1951;13:1–45.
[37] Leonard CG, Housewright RD, Thorne CB. Effects of some metallic
ions on glutamyl polypeptide synthesis by Bacillus subtilis. J
Bacteriol 1958;76:499–503.
[38] Yokoi H, Natsuda O, Hirose J, Hayashi S, Takasaki Y. Characteristics
of a biopolymer flocculant produced by Bacillus sp. PY-90. J Ferment
Bioeng 1995;79:378–80.
[39] Stewart JM, Young JD. Solid phase peptide synthesis. 2nd ed.
Rockford, Illinois: Pierce Chemical Co., 1984. p. 104.
[40] Chaplin MF, Kennedy JF. Carbohydrate analysis. Washington, DC:
IRL Press, 1986. p. 2–5.
[41] Bradford MM. A rapid and sensitive method for the quantitation of
microgram quantities of protein utilizing the principle of protein-dye
binding. Anal Biochem 1976;72:248–54.
[42] SAS Institute. SAS/STAT user’s guide, Version 6, 4th ed. vol. 2.
Gary, NC: SAS Institute Inc., 1990.
[43] Mason RL, Gunst RF, Hess JL. Statistical design and analysis of
experiments with application to engineering and science. New York:
Wiley, 1989.
[44] Montgomery DC. Design and analysis of experiments. 3rd ed. New
York: Wiley, 1991.
[45] Box GEP, Draper NR. Empirical model-building and response
surfaces. New York: Wiley, 1987.