International Journal of Biological Macromolecules 64 (2014) 443–452

Contents lists available at ScienceDirect

International Journal of Biological Macromolecules

journal homepage: www.elsevier.com/locate/ijbiomac

Statistical based media optimization and production of naringinase

using Aspergillus brasiliensis 1344
M. Shanmugaprakash ∗ , J. Kirthika, J. Ragupathy, K. Nilanee, A. Manickam
Downstream Processing Laboratory, Department of Biotechnology, Kumaraguru College of Technology, Coimbatore 641049, India

a r t i c l e i n f o a b s t r a c t

Article history: Statistics based optimization, Plackett–Burman design (PBD) and response surface methodology (RSM)
Received 18 October 2013 were employed to screen and optimize the media components for the production of naringinase from
Received in revised form 3 December 2013 Aspergillus brasiliensis MTCC 1344, using solid state fermentation. Cassava waste (CW) was used as both
Accepted 19 December 2013
the solid support and carbon source for the growth of A. brasiliensis. Based on the positive influence of
Available online 28 December 2013
the Pareto chart obtained from PBD on naringinase activity, three media components – maltose, peptone
and calcium chloride were screened. Box–Behnken design (BBD) was employed using these three factors
Keywords:
at three levels, for further optimization, and the second order polynomial equation was derived, based
Aspergillus brasiliensis
Cassava waste
on the experimental data. Derringer’s desired function methodology showed that the concentrations of
Derringer’s desired function maltose (7.74 g/L), peptone (4.19 g/L) and calcium chloride (7.63 mM) were the optimal levels for maximal
Naringinase naringinase activity (889.91 U/mg) which were validated through experiments.
Plackett–Burman design © 2013 Elsevier B.V. All rights reserved.
Response surface methodology

1. Introduction for the production of industrially important enzymes [6]. Agricul-

tural by-products were largely used as substrates for solid substrate
Naringinase is an extracellular enzyme system, consisting of fermentation, which also largely help in solid waste management
both ␣-l-rhamnosidase (EC 3.2.1.40) and ␤-d-glucosidase (EC [7]. The substrate serves as an inert or insoluble substrate, as
3.2.1.21). It specifically hydrolyzes the naringin (4,5,7-trihydroxy well as carbon and other source. In addition, the nutrient solu-
flavonone 7-rhamnoglucoside) to l-rhamnose and prunin; the lat- tion added to the substrate provides a good support for the growth
ter is further hydrolyzed to yield naringenin (4,5,7-trihydroxy and metabolism of microorganisms. The SSF is simple to perform
flavonone) (Fig. 1) and d-glucose [1,2]. This enzyme system is also and requires less processing energy for the production of differ-
found in different species of microorganisms and plants [3]. Naring- ent microbial products. It is mostly used when a fermentation
inase production is low in bacterial while but widely distributed in process makes use of fungal sources, while seldom for bacterial
fungi [4]. This enzyme system has great potential in both pharma- sources [7,8]. SSF has many advantages over submerged fermen-
ceutical and food industries for the biotransformation of steroids, tation including higher productivity, simple technique, low capital
antibiotics and particularly for glycosides hydrolysis. It is also used investment, low energy requirement and less water output, thus
in the debittering of naringin, limonin and neohesperidin, which lesser contamination, lack of foam build up, but improved prod-
impart bitterness in the citrus juice. The hydrolytic by-product of uct recovery. Thus, several advantages have made SSF a method
naringin is naringenin, which is bitterless and tasteless. This unique of choice for the production of industrially important compounds
property of naringinase can be used for processing, and to help sta- [9–11].
bilize citrus juice and improve its property [3,5]. Such applications Different solid substrates were used for naringinase production
have attracted several researchers to search for cheap, economical under the SSF, including soybean [1], wheat bran, washed sug-
and newer methods of producing this enzyme from various agro arcane bagasse [12], and rice bran [13]. Cassava waste, a fibrous
industrial wastes as substrates. Solid state fermentation (SSF) has material, is the by-product in the processing industry [6]. It has
gained importance for the production of microbial enzymes due to greater starch content [14] and is used as protein enrichment for
the economical advantages over the conventional, submerged fer- the enzyme production [15]. Not much work has been done to our
mentation. SSF has mostly been used for the bioconversion process knowledge, using cassava starch as solid support for the production
of debittering enzymes.
Most of the earlier work on the production of enzymes involves
∗ Corresponding author. Tel.: +91 0422 2669401; fax: +91 0422 2669406. one variable at a time (OVAT) on the experimental output while
E-mail address: [email protected] (M. Shanmugaprakash). other parameters are maintained constant; this procedure is not

0141-8130/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.ijbiomac.2013.12.033
444 M. Shanmugaprakash et al. / International Journal of Biological Macromolecules 64 (2014) 443–452

only time consuming, but expensive as well, terms of the con- India) was maintained on Czapeck Dox Agar (CDA) medium slant.
sumption of reagents and materials, and also it does not give Subculturing was carried out every three-week period, and the cul-
any information about the interactive effects among the various ture was maintained at 4 ◦ C.
parameters [16]. In order to overcome the limitations in OVAT, var-
ious statistical experiment designs, response surface methodology
(RSM), Graeco-Latin square design and Plackett–Burman design 2.2. Solid substrate
(PBD) have been tried for various engineering applications [17].
These statistical methods are very useful for examining the simulta- Cassava waste (CW) was used as a solid substrate. It was pro-
neous, systematic and efficient variations of all the components, the cured from a local sago industry, pulverized and sieved (0.85 mm)
cumulative interactive effect present on the system, and to predict before use.
the optimal conditions for a given operating system. RSM is useful
for a small number of variables (up to five), but impractical for a
large number of variables, due to the higher number of experimen- 2.3. Inoculum
tal runs required. Therefore, for screening more than five factors,
the Plackett–Burman design is recommended [18,19]. Optimiza- The inoculum was prepared by harvesting the spores from
tion of the different parameters was attempted by several workers, 6-day-old cultures grown in Czapeck broth containing 0.1%
using the Box–Behnken design under the response surface method- (w/v) Tween-80, and diluted to the desired spore concentration
ology for the maximal production of enzymes such as ␣-amylase (1 × 109 spores/g CW). The pour plate technique was followed to
[19,20], extracellular ribonuclease [21], thermo-tolerant xylanase calculate the number of spores in the inoculum.
[22], extracellular protease [23], pectinase [18] and ␣-galactosidase
[24]. Puri [25] used RSM as a statistical tool for naringinase produc-
tion. In the present study, the various parameters that influence the 2.4. Preparations of the substrate and solid state fermentation
production of naringinase were screened using Plackett–Burman
design, and the screened variables were optimized using the BBD Exactly 5 g of CW was accurately weighed in a Petri-plate,
combined with Derringer’s desired function methodology. incubated in a tray chamber (50 cm × 50 cm × 70 cm) under auto-
matically controlled temperature (27 ◦ C) and humidity. The
substrate was moistened with mineral solution containing (g/L):
2. Materials and methods NaNO3 – 3.0, K2 HPO4 – 1.0, MgSO4 – 0.5, KCl – 0.5, FeSO4 – 0.01,
with naringin (0.01%, w/v) as the inducer (pH 5.0). The contents
2.1. Microorganism were thoroughly mixed and autoclaved at 121 ◦ C. The inoculum
(10%, w/v) [26], of 5-day-old culture [13] was added to the above
A pure culture of Aspergillus brasiliensis MTCC 1344 procured substrate and incubated for 5 days. Experiments were conducted
from the Microbial Type Culture Collection (MTCC, Chandigarh, according to the statistical design (Table 1).

Fig. 1. Hydrolysis of naringin into prunin, rhamnose, naringenin, and glucose by naringinase expressing ␣-l-rhamnosidase and ␤-d-glucosidase activities.
 M. Shanmugaprakash et al. / International Journal of Biological Macromolecules 64 (2014) 443–452 445

Table 1
Experimental design using Plackett–Burman method for screening of nutrients supplemented to Cassava waste.

Medium code A B C D E F G H I J K L Specific activity (U/mg

dry CW) at 5th day

1 + + + − − + + − + + − − 246.30 ± 1.61
2 + + − − − − + − + − + + 410.23 ± 2.02
3 − + − + − + + + + − − + 562.35 ± 1.36
4 + − − + + − + + − − − − 412.85 ± 1.64
5 − − + + − + + − − − − + 374.69 ± 2.55
6 − − − − − − − − − − − − 297.59 ± 1.07
7 + + − − + + − + + − − − 502.35 ± 1.93
8 − + + + + − − + + − + + 297.45 ± 1.15
9 + − + − + + + + − − + + 398.54 ± 1.14
10 − − − + − + − + + + + − 231.56 ± 2.02
11 + − − − − + − + − + + + 206.58 ± 1.01
12 + − + + − − − − + − + − 365.21 ± 1.67
13 + + + + − − + + − + + − 154.87 ± 2.18
14 − + + − − − − + − + − + 75.54 ± 1.50
1+ − − + − + − + + + + − − 131.56 ± 1.84
16 + − + + + + − − + + − + 108.42 ± 1.56
17 − + + − + + − − − − + − 194.62 ± 2.12
18 − + − + + + + − − + + − 19.54 ± 0.99
19 + + − + + − − − − + − + 12.36 ± 1.72
20 − − − − + − + − + + + + 23.25 ± 2.48

b 61.0 −7.5 −33.1 5.3 −82.4 66.4 44.2 92.1 73.2 −260.6 −42.2 −8.2

Each data point represents the mean of triplicate determinations ± SD.

2.5. Enzyme extraction (sodium nitrate, peptone, urea) and various trace metals (cal-
cium chloride, magnesium sulphate, zinc sulphate, ferric chloride
At the end of SSF, the crude enzyme was extracted from the and manganese sulphate) present in the medium. The experimen-
fermented material. A weighed quantity (g) of fermented CW was tal setup for the Plackett–Burman design were carried out using
mixed thoroughly with 0.2 M acetate buffer (pH 4.0) at room tem- Minitab software (Trial version 16, USA).
perature on a rotary shaker for 1 h at 120 rpm (Scigenics, India).
Then, the suspension was centrifuged at 8000 rpm for 10 min (Kub-
ota 3700, Japan) and the clear, brown supernatant was used for 2.8. Response surface methodology
subsequent analysis.
Response surface methodology [30] is a collection of mathemat-
2.6. Naringinase assay ical and statistical techniques, used for optimizing the significant
parameters, that influence the production of an enzyme. The
The modified Thammawat method [27] was followed to assay Box–Behnken [31] method is an independent, rotatable quadratic
the naringinase activity. 0.1 mL of enzyme extract was added to design with no embedded factorial or fractional factorial points,
0.65 mL of 0.1% (w/v) naringin in acetate buffer (pH 4.0). After incu- where the variables combination is at the midpoints of the edges
bation at 60 ◦ C for 15 min, 0.2 mL was withdrawn. To this solution, of the variable space and at the centre [32]. This method optimizes
4.0 mL of diethylene glycol (90%, v/v) and 0.2 mL of (4.0 N) NaOH the concentrations of three factors at two different levels and a
were added, and incubated for 10 min at room temperature. The centre point. Also, it is useful for building the second-order surface
intensity of the yellow colour produced was measured at 420 nm. models.
One unit of naringinase activity is defined as the amount of enzyme The application of the statistical experimental design tech-
that is required to hydrolyse 1 ␮mol of naringin per ml per minute, niques in bioprocess development and optimization result in the
under the assay conditions. The protein concentration was esti- maximum yield. Based on one variable at a time, the independent
mated by Lowry method [28]. variables and their ranges were chosen. Experiments were con-
ducted, based on a BBD with three factors, at three levels; each
2.7. Plackett–Burman design independent variable was coded at three levels −1, 0 and 1, and
these values are – maltose (5.0–10.0 g/L), peptone (3.0–6.0 g/L) and
The Plackett–Burman experimental design was acquired to calcium chloride (5.0–10.0 mM). The coding variables are expressed
screen ‘N’ variables that influence the naringinase production by by the following equation:
‘N + 1 experiments [29]. Each of the variables is represented at
(Xi − Xz )
two levels – high (+) and low (−) levels. The effect of the variables xi = (2)
was calculated and considered as significantly important when the Xi
confidence level is >85%.
where Xi , is the dimensionless value of an independent variable;
The first order model for the Plackett–Burman design is:
 Xi is the real value of an independent variable; Xz is the real value
Y = ˇ0 + ˇ xi (1) of an independent variable at the centre point; and Xi is the step
i change of the real value of the variable i, corresponding to the vari-
where Y is the estimated target regression function and ˇ is the ation of a unit for the dimensionless value of the variable i. Based
regression coefficient. on the results obtained from the Plackett–Burman experimental
Twenty experiments were performed for 12 variables with design, the effect of maltose (X1 ), peptone (X2 ) and calcium chloride
naringinase activity as the response under each experiment (X3 ) were further optimized by the BBD investigation. A non-linear
(Table 1). The variables include carbon sources (maltose, rham- regression method was used to fit the second order polynomial
nose, glucose, and sucrose), inorganic and organic nitrogen sources equation that relates the predicted response (Y) in terms of the
446 M. Shanmugaprakash et al. / International Journal of Biological Macromolecules 64 (2014) 443–452

independent variables (X1 , X2 and X3 ): Table 2

Nutrients supplements to CW for screening using Plackett–Burman design.
Y = b0 + b1 X1 + b2 X2 + b3 X3 + b11 X12 + b22 X22 + b33 X32 + b12 X1 X2
Nutrient code Components High level Low level
+ b23 X2 X3 + b13 X1 X3 (3) A Maltose (g/L) 5 1
B Rhamnose (g/L) 5 1
where b0 is the intercept term, b1 , b2 , b3 the linear coefficients, C Glucose (g/L) 5 1
b11 , b22 , b33 the squared coefficients, and b12 , b23 , b13 the inter- D Sucrose (g/L) 5 1
action coefficients. Combinations of factors (such as X1 X2 , X2 X3 , E Sodium nitrate (g/L) 3 1
F Peptone (g/L) 3 1
X1 X3 ) represent an interaction between the individual factors in
G Urea (g/L) 3 1
that term. H Calcium chloride (mM) 5 1
I Magnesium sulphate (mM) 5 1
2.9. Statistical analysis J Zinc sulphate (mM) 5 1
K Ferric chloride (mM) 5 1
L Manganese sulphate (mM) 5 1
The multiple regression analysis and the analysis of variance
(ANOVA) were performed for fitting the mathematical model,
using Design Expert software (Trial version 8.0.7.1, Stat-Ease Inc., where di is the desirability of the response and ‘n’ is the num-
USA). The mathematical modelling was developed with a quadratic ber of responses studied in the optimization process. If any of the
model, including the linear, squared, and interaction terms. The sig- responses are beyond the desirability, then the overall function
nificant terms in the model for each response were found by the would be zero. It can be extended to
ANOVA, and the significance was judged by the F-statistic calcu- 1/n 1/n
lated from the data. The experimental data were evaluated with G = (d1˛1 × d2˛2 × d3˛3 × . . . × dn ) 0 ≤ ˛i ≤ 1(i = 1, 2, 3, . . .n),
various descriptive statistical parameters such as p-value, F-value, ˛1 + ˛2 + . . . + ˛n = 1 (8)
degrees of freedom (DF), sum of squares (SS), mean sum of squares
(MSS), coefficient variation (CV), determination coefficient (R2 ), where di indicates the desirability of the different responses Yi (i = 1,
adjusted determination of coefficient (Ra2 ), correlation coefficient 2, 3,. . .,n) and i represents the importance of the responses. Thus,
(R), Mallow’s (Cp ) statistic, Durbin–Watson (DW) statistic and chi- the maximal overall desirability function G depends solely on the
square (2 ) test, in order to show the statistical significance of the ˛i value [35]. For the optimization of any response, the goal must
developed quadratic mathematical model. After fitting them to the have a low and high value. In this study, the goals were assigned
developed model, the data generated were used for plotting the between the low and high values. The following desirability func-
response surface and contour plots. tion criterion was used to predict the naringinase production.
di = 0 if response < low value.
2.10. Percentage contributions of the process variables 0 ≤ di ≤ 1 as response varies from low to high.
di = 0 if response > high value.
Based on the sum of squares obtained from the ANOVA, the per- Weight factors of unity and default importance of 3 were cho-
centage contribution (PC) of each individual process variable was sen for the individual desirability of the response, which defines
calculated by the following equations, as described by [33]: the shape and represents the goals to be equally important for the
n response.
SSi
TPi = n n i=1
× 100 (4)
i=1 i=1
SSi + SSi + SSij 2.12. Verification of the predicted optimized conditions

n In order to determine the validity of the developed mathe-

SSii
TPCii = n n i=1
× 100 (5) matical model equation, experiments in triplicate were performed
i=1 j=1
SSi + SSii + SSij under the optimal conditions as predicted by the model. The aver-
age values of the experiments were compared with the predicted
n n values of the developed model in order to find out the accuracy and
SSij
TPCij = n ni=1 i=1 × 100 (6) the suitability of the model.
i=1 j=1
SSi + SSii + SSij
3. Results and discussion
where TPCi , TPCij and TPCii are the total percentage contributions
(TPC) of the linear, interactive and quadratic terms; and (SSi ), (SSij )
3.1. Screening of the nutrient supplements using a
and (SSii ) are the computed sum of squares for the linear, interactive
Plackett–Burman design
and quadratic terms, respectively.

Based on our earlier studies, 12 components were investigated

2.11. Optimization by Derringer’s desired function methodology
for their effect on naringinase activity, when supplemented with
Cassava waste (CW), using the Plackett–Burman Design. In the
After the results were obtained, Derringer’s desired function
experimental design, each row represents an experiment and each
methodology [34] was performed to evaluate the optimal operating
column represents an independent variable. The positive and neg-
conditions in order to obtain the maximum naringinase produc-
ative signs represent high and low levels of the given independent
tion. The general approach of the desirability function is to first
variable (Table 1). Codes A to L were used to represent each of
transform the response into a dimensionless individual desirability
the components (Table 2). The experimental design using different
function (di ) that varies from the lowest to the highest desirability
media showed a wide range (12.36–562.35 U/mg) of naringinase
(0 to 1). From the geometric means of different individual (di ) val-
activity. The magnitude of the coefficient ‘b’ of each term repre-
ues the overall desirability function (G) was worked out, by adding
sents the significant effect on the naringinase activity, and clearly
the individual desirability values:
suggests that the enzyme activity is strongly influenced by the com-
1/n ponent(s) present in the designed medium. We used the Pareto
G = (d1 × d2 × d3 × . . . × dn ) (7)
chart in order to determine the magnitude and importance of an
 M. Shanmugaprakash et al. / International Journal of Biological Macromolecules 64 (2014) 443–452 447

response. Any effect that extends past this reference line is poten-
tially significant for that particular process [36]. The fact that the
bars for sucrose and rhamnose remained inside the reference line
as shown in Fig. 2, indicate that these terms contributed to the least
prediction of the naringinase activity. The negative coefficients
of the model components (ZnSO4 , MgSO4 , MnSO4 and glucose)
showed an antagonistic effect while the positive coefficients (CaCl2 ,
peptone, FeCl3 , NaNO3 and maltose) showed a synergistic effect on
the enzyme activity [33]. Based on these observations, most signif-
icant factors (maltose, peptone, and calcium chloride) were used as
supplements to CW for the production of naringinase, and used for
further optimization by a Box–Behnken design.

3.2. Optimization of screened nutrient supplements using RSM

The three significant factors screened from PBD – maltose, pep-

Fig. 2. Pareto chart of an independent variable and the interaction on dependent tone and calcium chloride were supplemented along with CW as
variable. solid substrates, and further optimized using the Box–Behnken
design (Table 3). Totally, 15 designed runs of experimental condi-
independent variable, and its interaction with the dependent vari- tions, including three centre points were obtained from the BBD,
able (Fig. 2). The chart displays the absolute value of the effects and the experiments were carried out thrice. The results of the
and draws a reference line on the chart. The length of each bar Box–Behnken experiments are shown with the mean observed and
in the chart indicates the standardized effect of that factor on the predicted responses from the experimental model in Table 3.

Table 3
ANOVA of the response surface model for the naringinase activity.

Source Coefficient estimate Sum of squares Degree of freedom Mean square F value p-Value Prob > F Remarks

Model 859.25 962,333.92 9 106,925.99 852.27 <0.0001 Significant

A – Maltose 0.34 12,724.11 1 12,724.11 101.42 0.0002
B – Peptone −82.99 36,944.42 1 36,944.42 294.47 <0.0001
C – Calcium chloride −18.41 8.90 1 8.90 0.07 0.8006
AB −1.15 34,661.13 1 34,661.13 276.27 <0.0001
AC 55.80 6134.02 1 6134.02 48.89 0.0009
BC −146.52 87,995.29 1 87,995.29 701.38 <0.0001
A2 −301.19 373,975.09 1 373,975.09 2980.81 <0.0001
B2 −181.80 152,648.28 1 152,648.28 1216.70 <0.0001
C2 −345.71 369,832.69 1 369,832.69 2947.79 <0.0001
Residual error 627.30 5 125.46

Lack of fit 223.60 3 74.53 0.37 0.7872 Not significant

Pure error 403.70 2 201.85
Cor total 962,961.22 14
Mean 433.96
CV% 2.58
Adequate precision 4485.97

Table 4
Sequential model fitting for the naringinase activity.

Source Sum of squares Degree of freedom Mean square F value Prob > F Remarks

Sequential model sum of squares

Mean 2,824,879.98 1 2,824,879.98
Linear 49,677.43 3 16,559.14 0.20 0.8946
2FI 128,790.44 3 42,930.15 0.44 0.7321
Quadratic 783,866.04 3 261,288.68 2082.63 <0.0001 Suggested
Cubic 223.60 3 74.53 0.37 0.7872 Aliased
Residual 403.70 2 201.85
Total 3,787,841.20 15 252,522.75

Lack of fit tests
Linear 912,880.09 9 101,431.12 502.51 0.0020
2FI 784,089.64 6 130,681.61 647.42 0.0015
Quadratic 223.60 3 74.53 0.37 0.7872 Suggested
Cubic 0.00 0 Aliased
Pure error 403.70 2 201.85

Source Std. Dev. R2 Adjusted R2 Predicted R2 PRESS Remarks

Model summary statistics

Linear 288.14 0.0516 −0.2071 −0.4123 1,359,950.27
2FI 313.15 0.1853 −0.4257 −0.7367 1,672,342.74
Quadratic 11.20 0.9993 0.9982 0.9953 4485.97 Suggested
Cubic 14.21 0.9996 0.9971 + Aliased
448 M. Shanmugaprakash et al. / International Journal of Biological Macromolecules 64 (2014) 443–452

Table 5
Box–Behnken experimental design with response- and predicted values along with residuals.

Run Maltose (g/L) Peptone (g/L) Calcium Naringinase activity (U/mg) Rounded residuals (Ei − Ep ) Error (%) Absolute
chloride (mM) rounded
residuals

Experimental value (Ei ) Predicted value (Ep )

1 5 3 7.5 300.38 ± 2.55 294.34 6.04 2.01 2.01

2 10 3 7.5 564.62 ± 2.45 560.28 4.34 0.77 0.77
3 5 6 7.5 340.26 ± 3.31 344.60 −4.34 −1.28 1.28
4 10 6 7.5 232.15 ± 1.66 238.19 −6.04 −2.60 2.60
5 5 4.5 5 245.15 ± 3.15 246.53 −1.38 −0.56 0.56
6 10 4.5 5 248.29 ± 1.76 247.97 0.32 0.13 0.13
7 5 4.5 10 165.78 ± 1.08 166.10 −0.32 −0.19 0.19
8 10 4.5 10 325.56 ± 1.49 324.18 1.38 0.42 0.42
9 7.5 3 5 277.15 ± 1.20 281.81 −4.66 −1.68 1.68
10 7.5 6 5 448.26 ± 1.00 442.54 5.72 1.28 1.28
11 7.5 3 10 570.62 ± 0.96 576.34 −5.72 −1.00 1.00
12 7.5 6 10 148.45 ± 1.41 143.79 4.66 3.14 3.14
13 7.5 4.5 7.5 897.25 ± 1.37 880.93 16.32 1.82 1.82
14 7.5 4.5 7.5 874.25 ± 1.36 880.93 −6.68 −0.76 0.76
15 7.5 4.5 7.5 871.30 ± 2.15 880.93 −9.63 −1.11 1.11

Each data point represents the mean of triplicate determinations ± SD.

In order to determine the model adequacy among various mod- that the linear and interactive (2FI) models resulted in lower R2
els, three different tests via the sequential model sum of squares, values, adjusted R2 , predicted R2 and higher p-values when com-
lack of fit tests and model summary statistics were performed. The pared with the quadratic model. Furthermore, a cubic model was
maximum naringinase activity is given in Table 4. It was found found to be aliased. The quadratic model was found to have the

Table 6
Detailed descriptive statistics of the regression analysis for the derived quadratic model for the naringinase activity.

Descriptive statistics Equation Regression results


n

Sum of residuals (Yo − Yp ) 0

i=1

n

1
Average residual n
(Yo − Yp ) 0
i=1

n
2
Residual or error sum of squares (absolute) SSE = (Yo − Yp ) 403.70
i=1

n
2
Residual or error sum of squares (relative) (SSE )R = [(Yo − Yp ) /i2 ] 403.70
i=1

n
2
Error variance of the estimate (MSSE ) (∂2 = Se2 = (Yo − Yp ) /(n − p) = SSE /(n − p) 201.85

i=1n
 
2
Standard error of the estimate (Se ) ∂ = Se = (Yo − Yp ) /(n − p) = SSE /(n − p) 14.207

 i=1
  n

 (Yp −Yp )2

R2 =  n
i=1 SSreg
Determination coefficient (R2 ) =
  n SSreg +SSE
0.99
 (Yp −Yp )2 + (Yo −Yp )2

 i=1 n i=1
 

 (Yp −Yp )2

R=  i=1
Correlation coefficient (R)
n
 n
0.99
 (Yp −Yp )2 + (Yo −Yp )2

i=1SS dfT

i=1  n−1


Adjusted determination coefficient (Ra2 ) Ra2 = 1 − E
= 1 − (1 − R2 ) 0.99
 SST dfE n−k−1

Coefficient of variation (CV) CV = ( MSSE /o × 100) 2.58


n

n

Durbin–Watson statistic DW = (et − ei−1 )/ et2 1.86

i=2 i=1
Mallow’s Cp statistic Cp = (SSE /MSSE ) + 2p − n 7.0

n
2
2
Chi-square ( ) test 2cal = (Oi − Ei ) /Ei 1.20
i=1
 M. Shanmugaprakash et al. / International Journal of Biological Macromolecules 64 (2014) 443–452 449

Fig. 3. Diagnostic plots for the model adequacy.

maximum ‘Adjusted R-squared’ and the ‘Predicted R-squared’ val- second-order polynomial equation for the experimental values is
ues. Hence, the quadratic model, incorporating linear, interactive given in Table 5.
and quadratic effects, was applied to explain the effects of the given
process variables on the naringinase activity. The adequacy of the
quadratic model was determined by ANOVA (Table 5). 3.4. Statistical analysis

Statistical testing of the model was done following Fisher’s sta-

3.3. Fitting of the second order polynomial equation tistical test for ANOVA (Table 6). The analysis of variance of the
quadratic regression model demonstrates that it is highly signifi-
The regression equation obtained after the analysis of variance, cant; the F-value (Fmodel = 852.27) is much greater than the tabular
is presented as a function of maltose, peptone and calcium chloride. F(10,9) value (19.38) at 5% level. Since Fcal > Ftab (852.27 > 19.38), the
The model obtained in terms of the coded factor is: Fisher’s F-test indicated with 95% certainty, that the derived regres-
sion model could explain a significant amount of the variation in
Y = 859.25 + 0.34X1 − 82.99X2 − 18.41X3 − 1.15X1 X2 the dependent variables. The High F-value and non-significant lack
+ 55.80X1 X3 − 146.52X2 X3 − 301.1921 − 181.80X22 of fit indicated that the quadratic model was highly significant,
as reported [33,38]. The closer the value of R (multiple correla-
+ 345.71X32 (9) tion coefficient) to 1, the greater the correlation noticed between
the observed and the predicted values. R2 was 0.99 for naringi-
where Y = naringinase activity (U/mg), X1 , X2 and X3 are the coded nase activity. A coefficient of variance (CV) value of 2.58 reveals
values of maltose, peptone and calcium chloride, respectively. The the high precision and reliability of the experimental data, while
above equation is used to find out the optimal values of the given a non-significant lack of fit value (>0.05) indicates the validity of
process variables, and shown as the surface and contour plots. Hii the quadratic model. In addition, the value of the adjusted deter-
[37] have reported that the ANOVA is very important to test the mination coefficient (Ra2 = 0.99) was also very high, which reflects
significance of the predicted model. The significance of fit of the the high significance of the given model, as reported [38]. Adequate

Table 7
Predicted and experimental values of the naringinase activity under optimum conditions.

Optimum condition Specific activity (U/mg) % Error Residual error

Maltose (g/L) Peptone (g/L) Calcium chloride (mM) Predicted value Experimental value

7.74 4.19 7.63 889.91 890.72 ± 3.10 0.09 0.81

Each data point represents the mean of triplicate determinations ± SD.

450 M. Shanmugaprakash et al. / International Journal of Biological Macromolecules 64 (2014) 443–452

precision greater than 4.0 is desirable, and the ratio was found to close to the experimental values, and the data of all predicted and
be 80.6 (>4.0), which indicates that the derived model is significant experimental response values lie in a straight line (Fig. 3a) indi-
for the naringinase activity. The predicted residual sum of squares cating that the derived model was able to successfully predict the
(PRESS) value was 201.90, which showed how the derived model correlation between the process variables on the response. Fig. 3b
fits with each point in the given design [32]. shows that the normal % probability plot of residuals for response
The Durbin–Watson (DW) statistic method was used, to eval- was normally distributed, as they lie reasonably close in a straight
uate whether the autocorrelation or correlation between errors is line and show no deviation of the variance.
present in the given model. The range of the DW statistic value The internally studentized residuals plot was used to analyze
between 0 and 4 is useful for checking the linear association the goodness fit for the model. Fig. 3c shows that all the data points
between adjacent residuals [39]. The DW value below 2 indicates lay within the limits. Since all the leverage values are lesser lower
a positive correlation, while that over 2 indicates a negative corre- than 1 (Fig. 3d), there are no outliers or unexpected errors in the
lation [40]. In the present study, the DW value of 1.86 is very close model. However, the difference in the beta values plot (Fig. 3e)
to 2, indicating the best fit of the model (Table 6). Mallow’s (Cp ) is
another statistical tool used to determine how many terms can be
omitted from the response surface model [41]. For a response sur-
face model including all terms, (Cp ) = P, where P is the number of
parameters or variables in the regression model that includes the
intercept term. For response surface models with omitted terms,
Cp ∼ P indicates a good model with little bias, while (Cp ) ≤ P indi-
cates a very good prediction model. The ultimate goal is to remove
the terms from the response surface model until a minimum Cp
value closer to p is obtained. If (Cp ) > P, then it indicates that too
many terms have been removed or some remaining terms are not
necessary [41]. In our study, Mallow’s Cp statistic (Cp = 7.0) showed
the third condition (Cp ≤ P and P = 10 including b0 ,b1 ,b2 ,. . .,b3 2 )
indicating a very good prediction model [33]. The chi-square (2 )
test was performed to check whether there was a significant dif-
ference between the expected responses and the observed data.
The calculated chi-square value (2cal = 1.20) was found lower than
the tabulated chi-square value (2˛,(n−1) = 20.05,14 = 2tab = 23.69),
indicating that there was no statistically significant difference
between the observed and expected data.

3.5. Adequacy of the model

Model adequacy checking was used, in order to find whether

the derived model would give sufficient approximation values to
the actual values [32]. Fig. 3(a–e) depicts the residual and the influ-
ence plots for the experimental data. Raw residuals are elements
of variation for the given data, which cannot be explained by the
model, representing the deviations between the experimental and
predicted values. The predicted values from the model were quite

Fig. 5. Response surface plots representing the effects of process variables on pro-
Fig. 4. Diagram showing the percentage contribution of given process variables. duction of naringinase.
 M. Shanmugaprakash et al. / International Journal of Biological Macromolecules 64 (2014) 443–452 451

Fig. 6. Desirability ramp for numerical optimization for four selected goals.

showed no undue influence of any observation on any of the regres- and activity, that has generally been preferred for bioconversion
sion coefficients. Since the external studentized residuals are in the processes utilizing Cassava waste [6].
determined range (Fig. 3f), there is no strong evidence of influential
observations in the experimental data. The results suggest that the 3.9. Effect of calcium chloride
model used in the study is able to predict the optimum conditions
for the production of naringinase. Fig. 5b shows the interaction between calcium chloride and pep-
tone and it depicts that the contour plots are parallel to both the
3.6. Percentage contribution of the process variables axes suggesting that the two variables are quite independent of
each other. CaCl2 as a stabilizing effect on enzyme production has
Using the ANOVA, the sum of squares for each individual model been reported by several researchers [20,44,45].
components, and the percentage of contribution (PC) for each indi-
vidual term were calculated (Fig. 4). The quadratic terms showed
3.10. Effect of peptone
the highest level of contribution (83.40%) by the TPC compared
with the other terms, and this is followed by the interactive terms
Fig. 5c shows the combined influence of calcium chloride and
(11.98%). Among the calculated TPC values, the linear terms showed
peptone on the naringinase activity. It is seen that the enzyme
the lowest level of contribution (4.62%) and did not greatly influ-
activity increased steadily, reached the maximal level at 4.50 g/L,
ence the prediction of naringinase activity. Based on the TPC values,
and tended to decline as the concentration of peptone increased.
it is obvious that the quadratic independent variables have a direct
Under optimal concentration, peptone was used as the nitrogen
impact on the dependent variable.
source by A. brasiliensis. Puri [2] reported that peptone could be
used as an organic component for naringinase by A. niger MTCC
3.7. Effect of the process variables on the naringinase activity
1344. Similarly, Rahman [46] reported that peptone was used as the
nitrogen source for the production of cyclodextrin glucanotransfer-
In the present study, three factors at three levels of BBD were
ease (CGTase) by Bacillus strearothermophilus HR1. Peptone type I,
applied, to investigate the influence of the process variables such as
a complex mixture of peptides, with a small amount of free amino
maltose, peptone and calcium chloride on the naringinase activity.
acids, could play an important role in cell metabolism by supplying
The interaction between the independent variables and depen-
essential amino acids and metabolic energy [47].
dent variable was plotted in the form of three dimensional (3-D)
response surface and contour plots (Fig. 5a–c). These plots illus-
trate the relative effects of any two variables by maintaining the 3.11. Optimization of the process parameters and validation
third variable constant; and furthermore, it is used to predict the
optimum conditions for the given process variables. According to the Box–Behnken design, combined with the
numerical optimization technique, the optimal condition to obtain
3.8. Effect of maltose as supplemented to CW on naringinase maximal naringinase production using A. brasiliensis MTCC 1344
activity was calculated by Derringer’s desired function methodology as
follows: maltose 7.74 g/L, peptone 4.19 g/L and calcium chloride
Three-dimensional response surfaces were plotted on the basis 7.63 mM. Based on the optimal values, a desirability ramp was
of the model equation in order to investigate the interaction among developed using numerical optimization techniques (Fig. 6). The
the variables and to determine the optimum concentration of each experiments were performed at optimal values in triplicates, and
factor for maximal naringinase activity of A. brasiliensis. The effect the results were compared with the predicted values from model
of varying the concentrations of peptone and maltose as supple- equation (Table 7). The maximum naringinase activity under the
mented with CW (Fig. 5a) indicates that both variables facilitate optimum condition was found at 889.91 U/mg. The mean values of
higher enzyme activity at both lower- and higher levels. A strong the naringinase activity were compared along with the predicted
degree of curvature of the 3-D surface is noticed. A similar result ones obtained from Derringer’s desirability function method. The
was reported by Bocchini [42] for the utilization of maltose as the good correlation between the predicted values and the experi-
carbon source for the production of xylanase by Bacillus circulus. mental values indicates that the derived Box–Behnken combined
Omemu [43] observed a similar behaviour for cellulase produc- desirability method could be effectively applied to optimize the
tion. Aspergillus niger utilizes starch as the substrate for growth process variables for the naringinase activity.
452 M. Shanmugaprakash et al. / International Journal of Biological Macromolecules 64 (2014) 443–452

4. Conclusion [14] L.G. Lacerda, R.R. Almeida, I.M. Demiate, M.A.l.S. Carvalho Filho, E.C. Vasconce-
los, A.L. Woiciechowski, G. Bannach, E. Schnitzler, C.R. Soccol, Brazilian Archives
of Biology and Technology 52 (2009) 143–150.
The Plackett–Burman design was successfully employed to [15] O.O. Ezekiel, O.C. Ogugua, H.P. Blaschek, T.C. Ezeji, African Journal of Biotech-
screen various process variables and response surface methodol- nology 9 (2010) 187–194.
ogy was applied to optimize the screened variables involved in the [16] D. Gangadharan, S. Sivaramakrishnan, K.M. Nampoothiri, R.K. Sukumaran, A.
Pandey, Bioresource Technology 99 (2008) 4597–4602.
production of naringinase by A. brasiliensis MTCC 1344 under solid [17] G.E.P. Box, J.S. Hunter, Annals of Mathematical Statistics 28 (1957) 195–241.
state fermentation. The results indicated that the factors selected in [18] D.C. Sharma, T. Satyanarayana, Bioresource Technology 97 (2006) 727–733.
this study had a significant effect on the enzyme activity, and acted [19] G. Dey, A. Mitra, R. Banerjee, B.R. Maiti, Biochemical Engineering Journal 7
(2001) 227–231.
as supplement effects along with CW. Model summary statistics
[20] F. Francis, A. Sabu, K.M. Nampoothiri, S. Ramachandran, S. Ghosh, G. Szakacs,
showed that the predicted model was in close agreement with the A. Pandey, Biochemical Engineering Journal 15 (2003) 107–115.
experimental data. From the ANOVA results, the second-order poly- [21] Y.H. Xiong, J.Z. Liu, H.Y. Song, L.N. Ji, Biochemical Engineering Journal 21 (2004)
27–32.
nomial regression model developed has a high correlation R2 value.
[22] J.l.X. Heck, S.H. Flores, P.F. Hertz, M.A.n.Z.c. Ayub, Bioresource Technology 97
The best conditions were found to be, maltose 7.74 g/L, peptone (2006) 1902–1906.
4.19 g/L and calcium chloride 7.63 mM and the predicted maxi- [23] Y.K. Rao, S.-C. Lu, B.-L. Liu, Y.-M. Tzeng, Biochemical Engineering Journal 28
mum naringinase activity was 889.91 U/mg. Under these optimized (2006) 57–66.
[24] C.Q. Liu, Q.H. Chen, B. Tang, H. Ruan, G.Q. He, Letters in Applied Microbiology
conditions, the experimental values of enzyme production closely 45 (2007) 206–212.
agreed with the predicted values. [25] M. Puri, A. Kaur, R.S. Singh, A. Singh, Applied Biochemistry and Biotechnology
162 (2010) 181–191.
[26] M. Puri, A. Banerjee, U.C. Banerjee, Process Biochemistry 40 (2005) 195–201.
Acknowledgements [27] K. Thammawat, P. Pongtanya, V. Juntharasri, P. Wongvithoonyaporn, Kaestsart
Journal: Nature Science 42 (2008) 61–72.
The authors are indebted to the Department of Science and Tech- [28] O.H. Lowry, N.J. Rosebrough, A.L. Farr, R.J. Randall, Journal of Biological Chem-
istry 193 (1951) 265–275.
nology (DST), New Delhi, for the financial support (Project Sanction [29] R.L. Plackett, J.P. Burman, Biometrika 33 (1946) 305–325.
No. SR/FT/LS-85/2012) to carry out this work. The authors are also [30] G.E.P. Box, N.R. Draper, Response Surfaces, Mixtures, and Ridge Analyses, John
thankful to the management of Kumaraguru College of Technology, Wiley & Sons Inc., Hoboken, New Jersey, 2007.
[31] G.E.P. Box, D.W. Behnken, Technometrics 2 (1960) 455–475.
Coimbatore, India, for providing the research facilities. [32] J.P. Maran, S. Manikandan, K. Thirugnanasambandham, C.V. Nivetha, R. Dinesh,
Carbohydrate Polymers 92 (2013) 604–611.
References [33] K. Yetilmezsoy, S. Demirel, R.J. Vanderbei, Journal of Hazardous Materials 171
(2009) 551–562.
[34] G. Derringer, Journal of Quality Technology 12 (1980) 214–219.
[1] M. Puri, U.C. Banerjee, Biotechnology Advances 18 (2000) 207–217.
[35] M. Mourabet, A. El Rhilassi, H. El Boujaady, M. Bennani-Ziatni, R. El Hamri, A.
[2] M. Puri, S. Kalra, World Journal of Microbiology and Biotechnology 21 (2005)
Taitai, Applied Surface Science 258 (2012) 4402–4410.
753–758.
[36] A.B. Solanki, J.R. Parikh, R.H. Parikh, AAPS PharmSciTech 8 (2007) 43–49.
[3] M.H. Ribeiro, Applied Microbiology and Biotechnology 90 (2011) 1883–1895.
[37] L.S. Hii, M. Rosfarizan, T.C. Ling, A.B. Ariff, Food and Bioprocess Technology 5
[4] M. Puri, Applied Microbiology and Biotechnology 93 (2012) 49–60.
(2012) 729–737.
[5] H. Ni, F. Chen, H. Xiao, A. You, L.Q. Yunzhen, Journal of Food Science 77 (2012)
[38] H.L. Liu, Y.W. Lan, Y.C. Cheng, Process Biochemistry 39 (2004) 1953–1961.
1–7.
[39] G.J. Hewings, S. Changnon, C. Dridi, Paper Regional Economics Applications
[6] A. Pandey, C.R. Soccol, P. Nigam, V.T. Soccol, L.P.S. Vandenberghe, R. Mohan,
Laboratory, University of Illinois Urbana, 2000.
Bioresource Technology 74 (2000) 81–87.
[40] M. Khajeh, Food Chemistry 129 (2011) 1832–1838.
[7] A. Pandey, Biochemical Engineering Journal 13 (2003) 81–84.
[41] P.L. Dawson, R. Martinez-Dawson, Pollution Science 77 (1998) 468–474.
[8] S. Bhargav, B.P. Panda, M. Ali, S. Javed, Chemical Biochemical Engineering Quar-
[42] D.A. Bocchini, E. Gomes, R.D. Silva, Applied Biochemistry and Biotechnology
terly 22 (2008) 49–70.
146 (2008) 29–37.
[9] U. Holker, M. Hofer, J. Lenz, Applied Microbiology and Biotechnology 64 (2004)
[43] A.M. Omemu, I. Akpan, M.O. Bankole, O.D. Teniola, African Journal of Biotech-
175–186.
nology 4 (2005).
[10] M. Stredansky, E. Conti, L. Navarini, C. Bertocchi, Process Biochemistry 34 (1999)
[44] J.P. Chessa, G. Feller, C. Gerday, Canadian Journal of Microbiology 45 (1999)
11–16.
452–457.
[11] G.V. Reddy, P. Ravindra Babu, P. Komaraiah, K. Roy, I.L. Kothari, Process Bio-
[45] P. Viswanathan, N.R. Surlikar, Journal of Basic Microbiology 41 (2001) 57–64.
chemistry 38 (2003) 1457–1462.
[46] R.A. Rahman, R.M. Illias, M.G.M. Nawawi, A.F. Ismail, O. Hassan, K. Kamaruddin,
[12] S. Elinbaum, H. Ferreyra, G. Ellenrieder, C. Cuevas, Letters in Applied Microbi-
Process Biochemistry 39 (2004) 2053–2060.
ology 34 (2002) 67–71.
[47] M.R. Amezaga, I.R. Booth, Applied and Environmental Microbiology 65 (1999)
[13] M. Shanmugaprakash, V. Vinoth Kumar, M. Hemalatha, V. Melbia, P. Karthik,
5272–5278.
Engineering in Life Sciences 11 (2012) 322–325.