CHAPTER FIVE

Sequential Optimization Methods

for Augmentation of Marine
Enzymes Production in Solid-State
Fermentation: L-Glutaminase
Production a Case Study
T. Sathish*,1,2, K.B. Uppuluri, P. Veera Bramha Chari{, D. Kezia

*Bioengineering and Environmental Centre, Indian Institute of Chemical Technology, Hyderabad, India

Bioprospecting Laboratory, School of Chemical and Biotechnology, SASTRA University, Thanjavur, India
{
Krishna University, Machilipatnam, India

Center for Biotechnology, Andhra University, Visakhapatnam, India

2
Corresponding author: e-mail addresses: sathish@niot.res.in; satish.tadikamalla@gmail.com

Contents
1. Introduction
2. L-Glutaminase Production a Case Study
2.1 Experimentation
2.2 Observations
3. Conclusions
References

96
97
97
102
113
113

Abstract
There is an increased L-glutaminase market worldwide due to its relevant industrial
applications. Salt tolerance L-glutaminases play a vital role in the increase of flavor of
different types of foods like soya sauce and tofu. This chapter is presenting the economically viable L-glutaminases production in solid-state fermentation (SSF) by Aspergillus
flavus MTCC 9972 as a case study. The enzyme production was improved following a
three step optimization process. Initially mixture design (MD) (augmented simplex lattice design) was employed to optimize the solid substrate mixture. Such solid substrate
mixture consisted of 59:41 of wheat bran and Bengal gram husk has given higher
amounts of L-glutaminase. Glucose and L-glutamine were screened as a finest additional
carbon and nitrogen sources for L-glutaminase production with help of Plackett
Burman Design (PBD). L-Glutamine also acting as a nitrogen source as well as inducer
for secretion of L-glutaminase from A. flavus MTCC 9972. In the final step of optimization
various environmental and nutritive parameters such as pH, temperature, moisture
1

Current address: Andaman and Nicobar Centre for Ocean Science and Technology, National Institute
of Ocean Technology, Port Blair 744103, India.

Advances in Food and Nutrition Research, Volume 78

ISSN 1043-4526
http://dx.doi.org/10.1016/bs.afnr.2016.06.003

2016 Elsevier Inc.

All rights reserved.

95

T. Sathish et al.

96

content, inoculum concentration, glucose, and L-glutamine levels were optimized

through the use of hybrid feed forward neural networks (FFNNs) and genetic algorithm
(GA). Through sequential optimization methods MDPBDFFNNGA, the L-glutaminase
production in SSF could be improved by 2.7-fold (4531690 U/g).

1. INTRODUCTION
The environmental conditions of marine microorganisms are harsh and
diverse; they can be hot or cold, acidic or basic, pressurized, saline, or mineral
rich as found in deep ocean hydrothermal vents, polar oceans, and extremely
saline water body. The extremophiles that survive in these diverse harsh conditions may produce enzymes that help in their survival (Chandrasekaran &
Kumar, 2010). The marine enzymes are relatively stable and active than
the corresponding enzymes from the terrestrial sources (Zhang & Kim,
2010). The major characteristics of marine microbial enzymes are salt tolerance, hyperthermostability, barophilicity, and cold adaptivity.
L-Glutaminase (L-glutamine amidohydrolase, EC 3.5.1.2) is the enzyme
deamidating L-glutamine. L-Glutaminase has many therapeutic and industrial
applications. It is a potent anticancer agent and it acts as a flavor-enhancing
agent in fermented foods by increasing the glutamic acid content and thereby
imparting a palatable and enhanced taste (Kashyap, Sabu, Pandey, Szakacs, &
Soccol, 2002; Sabu, Keerthi, Kumar, & Chandrasekaran, 2000). The enzyme
can also be used as biosensor for monitoring glutamine levels in mammalian
and hybridoma cell cultures (Katikala, Bobbarala, Tadimalla, & Guntuku,
2009). Salt tolerant and heat stable glutaminase from marine microbes are
favored in food industry. Extracellular L-glutaminase is reported to be produced from marine yeasts (eg, Zygosaccharomyces rouxii) (Kashyap et al.,
2002), marine fungi (eg, Beauveria sp.) (Sabu et al., 2000), and marine bacterium
(eg, Pseudomonas sp.) (Kumar & Chandrasekaran, 2003).
Production of marine microbial enzymes can be done using batch, continuous, or fed-batch fermentation. Batch process can be done by submerged fermentation (Estrada-Badillo & Marquez-Rocha, 2003) and
solid-state fermentation (SSF) (Kashyap et al., 2002). Production of L-glutaminase in SSF is an economically viable process (Sathish, Laxmi, Rao,
Brahmaiah, & Prakasham, 2008).
The selected solid substrate and cultural conditions play a vital role in the
enzymes production in SSF. Optimization of various cultural conditions
such as pH, temperature, moisture content, inoculum size and age of

Optimization Methods for Augmentation of Marine Enzymes

97

inoculum, and incubation time is vital for the production of microbial

enzymes in SSF. Interactive effects of these factors have significant effect
on the production of microbial products.
The traditionally method of optimization is one factor at a time
method, where only one parameter is optimized and all other parameters
are kept constant. Though this method is simple, it lacks the interactive
effects. To account the interactive effects of selected fermentation conditions, statistical methods such as factorial designs, orthogonal designs
(Taguchi method), and response surface methodology are evolved
(Sathish & Prakasham, 2010). Along with optimization methods, screening
designs such as two-factorial designs and PlackettBurman designs (PBDs)
were also evolved, to select the best suitable component from a huge variety
of choices with less number of experiments. Due to the availability of various
softwares and literature on such statistical methods, many researchers are
employing these designs in their work. Each method has some particular
limitations as well as advantages. The assumption of mathematical models
from linear to quadratic polynomials is the major limiting factor of these
designs. To overcome these problems, artificial neural networks (ANNs),
genetic algorithms (GAs), and particle swarm optimization are used to solve
the nonlinear problems (Sathish & Prakasham, 2010). In this study, the production of L-glutaminase by Aspergillus flavus MTCC 9972 was improved by
optimizing the various conditions by using statistical methods (MD, PBD)
and artificial intelligence methods (ANN, GA). This case study represents a
good example on how the application of combined statistical methods can
help to optimize the enzyme production and obtain higher enzyme yields.

2. L-GLUTAMINASE PRODUCTION A CASE STUDY

2.1 Experimentation
Marine fungus A. flavus MTCC 9972 isolated from coastal waters of Bay of
Bengal at Visakhapatnam, A.P., India was used in this study. The spore solution prepared from 72 h culture was used as an inoculum.
Different agroindustrial materials such as wheat bran (WB), rice husk
(RH), rice bran (RB), corn cobs (CCs), sugar cane bagasse (SCB), red gram
husk (RgH), green gram husk (GGH), Bengal gram husk (BeH), black gram
husk (BgH), ground nut oil cake (GC), and coconut oil cake (CoC) were procured from the local market. The pineapple waste (PW) was obtained from
the local fruit market and juice shops. The spent coffee (SC) and spent tea
(ST) waste were collected from the canteen and home. In a different

T. Sathish et al.

98

250 mL flask 10 g of processed substrate was taken individually, moisturized

with 10 mL of distilled water and sterilized. The sterile substrates were initially
received 1 mL of spore suspension (1
108 spores per mL) as inoculum. The
content of flasks was thoroughly mixed and incubated at 28C in a slanting
position to get maximum surface area. Fermentation was carried up to 48 h.
The crude enzyme was extracted by simple contact method as described
by Sathish et al. (2008). L-Glutaminase activity was determined using L-glutamine as substrate (Imada, Igarasi, Nakahama, & Isono, 1973). Enzyme
yield was expressed as units per gram of dry substrate (U/g).
Mixture design (MD) was used to study the combined substrates impact
on L-glutaminase production and optimized their concentrations to attain
higher amounts of enzyme yield. Three best substrates suitable for fungal
growth as well as enzyme production were chosen. In MD, the total proportions of the different factors must be 100% that is 10 g using three solid
substrates. A 10 run augmented simplex lattice design was used in the present
study. Table 1 shows the MD along with obtained L-glutaminase yield
corresponding to the substrate mixture.
Table 1 Mixture Design Along with Observed and Predicted L-Glutaminase Yield (U/g)
L-Glutaminase Activity (U/g)
Wheat
Green Gram Bengal Gram
S. No. Bran (g) Husk (g)
Husk (g)
Observed Predicted Error

1 (10)

0 (0)

0 (0)

453

454.57

1.57

0 (0)

1 (10)

0 (0)

388

385.66

2.34

0 (0)

0 (0)

1 (10)

402

401.94

0.06

0.5 (5)

0.5 (5)

0 (0)

436

435.24

0.76

0.5 (5)

0 (0)

0.5 (5)

502

503.52

1.52

0 (0)

0.5 (5)

0.5 (5)

406

403.61

2.39

0.6667
(6.7)

0.1667
(1.65)

0.1667 (1.65)

496

492.07

3.93

0.1667
(1.65)

0.6667 (6.7)

0.1667 (1.65)

428

435.79

7.79

0.1667
(1.65)

0.1667
(1.65)

0.6667 (6.7)

463

463.98

0.98

10

0.3333
(3.35)

0.3333
(3.35)

0.3333 (3.30)

494

491.57

2.43

The real values are presented in braces. Predicted values were calculated by special cubic model (Eq. 7).

Optimization Methods for Augmentation of Marine Enzymes

99

The analysis of MD varies from the response surfaces models because of

the constraint xi 1. Linear to cubic models were used to determine the
best blend of components for optimal medium.
Y

p
X

i xi Linear model

(1)

i1

p
X

i xi +

p
XX

p
X

i xi +

p
XX

ij xi xj +

i<j

i1

ij xi xj Quadratic model

(2)

i<j

i1

p
XX X

ijk xi xj xk Special cubic

(3)

i<j<k

p
p
p
X
XX
XX


i xi +
ij xi xj +
ij xi xj xi  xj
i<j

i1

p
XX X

i<j

(4)

ijk xi xj xk Full cubic

i<j<k

where Y is a response, i is a linear, ij is a quadratic, and ijk is a cubic coefficients, ij is a parameter of the model. The ixi represents linear blending
portion, and the parameters ij represent either synergic or antagonistic
blending.
A PBD was employed to screen the supplementary carbon and nitrogen
sources for increasing the L-glutaminase production by isolated marine fungus. Table 2 shows the selected nine variables and their levels along with
experimental design. PBD analysis was performed based on the first-order
model using the following Eq. (5):
X
i xi i 1,2,3k
(5)
Y 0 +
where Y is productivity (response; L-glutaminase yield), 0 is model intercept, and i is coefficient of variables. Variables whose coefficient value is
higher indicate that those variables have major influence on the enzyme production. Similarly, the coefficients values close to zero suggest a little or no
impact on L-glutaminase production.
The effect of each selected variable on L-glutaminase production was
determined using the following Eq. (6):
X

2
Yi +  Yi
E xi
(6)
N

Table 2 The PlackettBurman Experimental Design Matrix for Screening of Carbon and Nitrogen Sources for L-Glutaminase
Production by A. flavus MTCC 9972
2 Level (g) 100
100
100
100
100
50
50
50
50
+ Level (g) 200
S. No.

200

200

200

200

150

Yeast
Sucrose cellulose Starch Fructose Glucose Extract

150

150

Malt
Peptone Extract

150

L-Glutaminase Activity (U/g)

L-Glutamine

Observed Predicted Error

1

1

1

1

696

701.5

5.5

1

1

1

1

677

671.5

5.5

1

1

1

1

1

892

900.5

8.5

1

1

1

1

1

403

411.5

8.5

1

1

1

1

592

597.5

5.5

1

1

1

725

716.5

8.5

1

1

1

647

638.5

8.5

1

1

1

1

725

719.5

5.5

1

1

1

1

828

833.5

5.5

10

1

1

1

1

747

741.5

5.5

11

1

1

1

1

1

563

571.5

8.5

12

1

1

1

1

1

1

1

1

1

537

528.5

8.5

Optimization Methods for Augmentation of Marine Enzymes

101

where E(xi) is concentration effect of the tested variable, Yi + and Yi are
obtained L-glutaminase yield at high and low concentrations of variable
(xi), and N is the number of trials. Sign of the variables effect indicates that
the level at which it is considered for further improvement. The variables
having above 90% confidence levels were considered as significantly influence on L-glutaminase production by A. flavus MTCC 9972.
A hybrid of feed forward neural network (FFNN) and GA was used to
optimize the various process and nutritive conditions of L-glutaminase production by A. flavus MTCC 9972. The hybrid FFNN helps to nonlinear
modeling of the experimental data and GA plays a role to determine the
optimum condition for enzyme production based on FFNN outputs.
The details of FFNNGA hybrids were given elsewhere (Prakasham,
Sathish, & Brahmaiah, 2011). The architecture of FFNN consists of three
layers of neuron, an input layer, hidden layer, and an output layer. An additional node presents in the input and hidden layers known as bias. Each neuron present in layer is connected to subsequent layer neurons. The
connection between neurons is termed as weights. These weights play an
important role in modeling of data.
Each selected parameter acts as an input neuron and the experimental
conditions were chosen as inputs for the network while output is L-glutaminase activity (U/g). After entering the input layers, the signals were sent
to the hidden layer neurons where they were processed and sent to the output neuron. The neuron in hidden and output layer acts as a summing junction, which combines and modifies the inputs from the previous layer.
Sigmoid transfer function was used for hidden layer, and the linear transfer
function was used for output layer to obtain the best fitted values. A sigmoid
transfer function generally normalizes the data within a range 1 to +1, to
avoid any numerical overflow prior to training.
The training algorithm plays a pivotal role in the functionality of created
FFNN. Various training algorithms such as conjugate gradient backpropagation with PowellBeale restarts (TRAINCGB) (Powell, 1977), Bayesian regularization back-propagation (TRAINBR) (Foresee & Hagan, 1997;
MacKay, 1992), MarquardtLevenberg method (TRAINLM) (Mizytani,
1997), and scaled conjugate gradient back-propagation (TRAINSCG)
(Moller, 1993) were tested to select the best training algorithm to train the
network. The number of neurons in the hidden layer were fixed by increasing
step by step till the best correlation was achieved (care was taken to avoid
the over learning by limiting the no of neurons was 18 at hidden layer as a
maximum). Total 30 experiments were conducted by altering the various

102

T. Sathish et al.

concentrations of six parameters. The training consisted of 24 runs (80%) and

the testing of six runs (20%). Initially weight and bias values were taken randomly. However, in subsequent training steps, the weights and biases, in hidden and output layers, were adjusted in accordance with a convergence
criterion to get the similarity in training and testing experimental values.
The experimental design along with obtained L-glutaminase yield is shown
in Table 3.
The FFNN predictability was determined by the coefficient R2, mean
square error (MSE), and mean absolute percentage error (MAPE). The error
was calculated based on the differences between the experimental and
predicted values.
GA optimization was performed using FFNN weights and bias values.
Various parameters of GA optimization such as chromosome length (Lchr)
as 60, population size (Npop) as 60, crossover probability as 0.7, mutation
probability (Pmut) as 0.01 were assumed based on previous studies. GA


was evaluated up to 300 generations Ngmax 300 . Best fermentation conditions in the given range of input parameters were selected from the 300
generations. Neural networks and GA toolboxes of MATLAB 7.0 (The
Mathworks, USA) were used in modeling studies.

2.2 Observations
Cost and availability of solid substrate are the major controlling factors in SSF
(Chiranjeevi, Pandian, & Sathish, 2014). Among selected solid substrates
WB supported the highest enzyme production (453 U/g), followed by
BeH (402 U/g) and GGH (387 U/g) by A. flavus MTCC 9972 (Fig. 1).
CoC and ST yielded the lowest L-glutaminase production (128 U/g).
The dissimilarity of the production is attributed to variation in the composition of the solid materials (Sathish et al., 2008). The obtained results are in
accordance with El-Sayed (2009), who noticed that WB was the best solid
substrate for the L-glutaminase production from Trichoderma koningii. Sathish
et al. (2008) reported that BeH was the finest support for L-glutaminase production from Bacillus sp. RSPGLU followed by WB. Kashyap et al. (2002)
noticed that sesame oil cake yielded more L-glutaminase than WB as substrate material with Z. rouxii. L-Glutaminase production was also carried
out by using sufi (soya bean cheese) and polystyrene as substrate materials
(Han, Mab, Romboutsa, & Nout, 2003; Prabhu & Chandrasekaran,
1997; Sabu et al., 2000).

Table 3 Experimental Design and L-Glutaminase Production by A. flavus MTCC 9972 on Selected Parameters at Various Levels in SSF
L-Glutaminase Activity (U/g)
Moisture Inoculum
Concentration
Glucose
Content
L-Glutamine
(mL)
Concentration (g) Concentration (g) Experimental Predicted Error
S. No. pH Temperature (C) (%, w/w)

29

90

1.75

0.75

0.5

452

459.99

7.99

29

90

1.25

1.25

0.5

631

630.99

0.01

29

70

1.75

0.75

1.5

376

376.00

0.00

27

90

1.25

1.25

1.5

693

692.99

0.01

29

70

1.75

1.25

1.5

768

768.00

0.00

27

90

1.75

0.75

1.5

856

855.99

0.01

29

90

1.25

0.75

0.5

244

245.99

1.99

29

70

1.25

1.25

1.5

431

431.00

0.00

27

70

1.75

0.75

0.5

316

315.99

0.01

10a

27

90

1.25

0.75

1.5

863

862.99

0.01

11

29

70

1.25

0.75

1.5

828

827.99

0.01

12

27

70

1.25

1.25

0.5

693

693.00

0.00

13a

27

70

1.75

1.25

0.5

830

830.00

0.00

14

27

90

1.75

1.25

1.5

888

888.00

0.00

15

29

90

1.75

1.25

0.5

917

915.47

1.53

3
4

Continued

Table 3 Experimental Design and L-Glutaminase Production by A. flavus MTCC 9972 on Selected Parameters at Various Levels in SSFcont'd
L-Glutaminase Activity (U/g)
Moisture Inoculum
L-Glutamine
Content
Concentration
Glucose
S. No. pH Temperature (C) (%, w/w)
(mL)
Concentration (g) Concentration (g) Experimental Predicted Error

16a

27

70

1.25

0.75

0.5

1021

1020.43

0.57

17

28

80

1.5

852

849.43

2.57

18

28

80

1.5

609

609.00

0.00

19

26

80

1.5

756

756.00

0.00

30

80

1.5

807

806.99

0.01

28

60

1.5

926

925.77

0.23

22

28

100

1.5

743

742.94

0.06

23

28

80

557

556.99

0.01

24

28

80

933

932.99

0.01

25

28

80

1.5

0.5

796

795.99

0.01

26

28

80

1.5

1.5

865

865.00

0.00

27

28

80

1.5

630

630.91

0.91

28

28

80

1.5

748

747.96

0.04

28

80

1.5

1008

1007.81

0.19

28

80

1.5

996

995.99

0.01

20
21

29
30
a

Runs used for testing.

Optimization Methods for Augmentation of Marine Enzymes

105

Fig. 1 Role of various agroindustrial waste materials on L-glutaminase production by

A. flavus MTCC 9972 in SSF. WB, wheat bran; RH, rice husk; RB, rice bran, CC, corn cobs;
SCB, sugar cane bagasse; RgH, red gram husk; GGH, green gram husk; BeH, Bengal gram
husk; BgH, black gram husk, GC, ground nut oil cake; CoC, coconut oil cake; PW, pineapple waste; SC, spent coffee; ST, spent tea.

MD was employed to optimize the solid substrate composition. An augmented simplex lattice MD consists of 10 experiments with various combinations of WB, GGH, and BeH. L-Glutaminase production was varied from
388 to 502 U/gds (Table 1) at various runs of MD. This was attributed due
to alteration in the overall sugars and nitrogenous compounds of the solid
substrate. The MD data were analyzed by employing a multiple linear
regression analysis by taking the L-glutaminase production as a dependent
variable. Based on regression coefficient (R2) and F-test, an appropriate type
of model was selected from a linear to full cubic models. The analysis of variance (ANOVA) results signifies quadratic and special cubic models are significant and other models such as linear and cubic are insignificant (Table 4).
 2

0:9990 than the
Though cubic models have higher R2 value Rcubic


2

0:9931
, the model was rejected because
special cubic model Rspecial
cubic
of smaller F-value (Fcubic 2.8589) and higher p-value (Pcubic 0.3858)
(Rispoli & Shah, 2007; Sathish & Prakasham, 2013). Based on ANOVA,
special cubic model was used for the further studies.

T. Sathish et al.

106

Table 4 ANOVA of Different Models and Special Cubic Model Regression Coefficients
R2
Adjusted R2 SS
df MS
F
p-Value

Linear

0.4642 0.3111

7239.00 2

3619.500

Quadratic

0.9442 0.8745

7486.53 3

2495.508 11.4726 0.0196

Special cubic

0.9936 0.9808

770.46 1

770.459 23.2028 0.0170

Cubic

0.9990 0.9914

84.79 2

42.394

15595.60 9

1732.844

Total adjusted

3.0319 0.1126

2.8590 0.3858

Special Cubic Model Terms

X1 * X2
WB (X1) GGH (X2) BeH (X3) (ns)

X1 * X3

X2 * X3
(ns)

X1 * X2 *
X3

Coefficient 454.5789 385.6698 401.9425 60.4973 301.0428 39.2246 890.4706

p-Value

0.000004 0.000007 0.000006 0.119859 0.001728 0.256317 0.017044

t-Value

81.6050 69.2346 72.1559 2.1575

10.7360 1.3988

4.8169

SS, sum of squires; MS, mean square; df, degree of freedom; ns, not significant.

The empirical relationship between selected solid substrates and L-glutaminase production obtained by the application of special cubic model is
given by Eq. (7):
L  Glutaminase activity U=gds 454:5789* X1
+ 385:6698 *X2 + 401:9425* X3
+ 60:4973* X1* X2 + 301:0428 *X1 *X3
+ 39:2246* X2* X3
+ 890:4706 *X1 *X2 *X3
(7)

where X1X3 are the coded values of the test variables. Based on Students
t-test and p-values significance of each coefficient was determined. Except
interaction terms of GGH with WB and BeH remaining all other terms are
significant. Fig. 2 is the triangular graph, which illustrates the L-glutaminase
yield as a function of Eq. (7). The triaxial graph illustrates that the highest
enzyme production is observed near to the WB and BeH axial line.
The regression Eq. (7) was solved based on numerical method given by
Myers and Montgomery (1995). The results showed that 59% WB and 41%
BeH blend are the best support for fungi growth as well as enzyme production. At this combination the predicted L-glutaminase production was

Optimization Methods for Augmentation of Marine Enzymes

107

Fig. 2 Triaxial surface (A) and triaxial contour (B) plots of selected three agroindustrial
waste materials mixture effect of on L-glutaminase production by A. flavus MTCC 9972.

505.82 U/g, by conducting the validation experiments 516 U/g enzyme

was obtained. With the help of MDs, 14% enzyme yield was improved
by optimizing the composition of substrate mixture alone. Sangeetha,
Ramesh, and Prapulla (2004) noticed an enhanced FTase production with
addition of RB to bagasse and tippi. Sathish et al. (2008) optimized the substrate composition using MD for L-glutaminase production by Bacillus sp.
RSPGLU reporting that two-level combination BeH and WB in a ratio
of 66:34 was the best substrate. Sathish and Prakasham (2013) also employed
MD to optimize the solid substrate for FTase production by Aspergillus
awamori GHRTS. Current experimental results and literature reports signify
the bi- or trimixtures of solid substrates could enhance the enzyme production in SSF.
In the process of L-glutaminase production in SSF, the second step is the
selection of suitable carbon and nitrogen sources. With the help of PBD,
each component could be studied at two different concentrations (levels).
In current study, nine components (five carbon and four nitrogen) were
studied for L-glutaminase production by isolated marine fungus. Table 2
shows the experimental design along with components at high and low
values.
The obtained L-glutaminase yield at various runs in PBD denotes the
selected nine compounds and their levels have significant effect on enzyme
production by A. flavus MTCC 9972. The highest enzyme activity 892 U/g

108

T. Sathish et al.

was observed at third run and 403 U/g as a lowest enzyme activity was
noticed at fourth run.
The effect of all selected components was plotted as a Pareto chart of
effects (Fig. 3). This chart shows the factors main effect estimates on the horizontal axis (Hymavathi, Sathish, Brahmaiah, & Prakasham, 2010). Fig. 3
denotes that glucose and L-glutamine have the highest effect on L-glutaminase production by A. flavus MTCC 9972. It indicates that the isolated fungus
utilizes L-glutamine as a nitrogen source as well as inducer for enzyme
secretion.
By taking L-glutaminase activity as a response, regression coefficients and
effects (nine variables and a constant term) were calculated and results are
presented in Table 5 along with ANOVA results. Table 5 indicated that glucose has the highest effect (137.66) and the lowest p-value (0.00372)
followed by L-glutamine. Both compounds have positive effects, indicating
that more amount of these compounds is needed to obtain higher amounts
of L-glutaminase by A. flavus MTCC 9972. Based on Pareto chart of effects
and ANOVA, glucose and L-glutamine were selected as supplementary carbon and nitrogen for L-glutaminase production by A. flavus MTCC 9972.

Fig. 3 Pareto chart for selected carbon and nitrogen source on L-glutaminase production by A. flavus MTCC 9972.

Table 5 Effect and ANOVA for PBD Analysis Data

S. No
Variable
Effect
Coefficients

SS

df

p-Value

133.3259

0.000056

7271.8

7271.76

34.1116

5.8405

0.028086

11.1633

1495.4

1495.44

7.0151

2.6486

0.117872

19.8333

9.9167

1180.1

1180.08

5.5357

2.3528

0.142913

Fructose

40.2500

20.1250

4860.2

4860.19

22.7990

4.7748

0.041172

Glucose

137.6667

68.8333

56856.3

56856.33

266.7113

16.3313

0.003728

Yeast extract

28.6533

14.3267

2463.0

2463.04

11.5540

3.3991

0.076722

Peptone

12.7967

6.3983

491.3

491.26

2.3045

1.5181

0.268310

Casein

55.2900

27.6450

9171.0

9170.95

43.0207

6.5590

0.022464

10

L-Glutamine

131.5700

65.7850

51932.0

51931.99

243.6114

15.6081

0.004080

Error

426.4

213.18

Total

136147.4

11

Mean/intercept

561.9433

561.9433

Maltose

49.2333

24.6167

Lactose

22.3267

Starch

t-Value

MS

110

T. Sathish et al.

Further these compounds concentrations were optimized along with other

parameters by hybrid ANNGA.
Based on preliminary studies (data not shown) and PBD, temperature,
pH, moisture content, inoculum concentration, glucose, and L-glutamine
concentrations were identified as major influencing parameters on L-glutaminase production by A. flavus MTCC 9972 and those parameters were
optimized by FFNN and GA. Table 3 depicts the experimental design along
with the obtained enzyme production values. The L-glutaminase production
was varied from 244 to 1021 U/g. The observed very low and high enzyme
production (approximately fourfold) signifies the selected parameters and
their levels on L-glutaminase production by A. flavus MTCC 9972. Further
the obtained data was analyzed with FFNN. To obtain the best network
topology different training algorithms with different number of neurons
were tested.
The scaled conjugate gradient back-propagation training algorithm
(TRAINSCG) with 12 neurons at hidden layer was given the best fitted
values nearest to the experimental values. As a final point of FFNN construction, the topology for ANN was fixed as 6-12-1 neurons in input, hidden
and output layers (Fig. 4). The goodness of FFNN predictions was analyzed
by calculating the regression coefficient (R2). The R2 of whole data

Fig. 4 FFNN architecture used for L-glutaminase production data modeling. WH, weights
between input and hidden neurons; WO, weights between hidden and output neurons;
bI, input bias; bH, hidden layer bias.

Optimization Methods for Augmentation of Marine Enzymes

111

(together with training and testing data) was found to be 0.9999 representing
that the constructed network was significant and network predicted data is
more accurate. The overall MAPE and the MSE was noted to be 0.0009
and 3.234 for training and 0.0001 and 0.063 for testing, respectively. The
obtained low MAPE and MSE values indicate that the constructed FFNN
was best suits for modeling of obtained L-glutaminase data.
From the 300 set of conditions given by the GA trails only the best five
conditions were selected and validated. By conducting the validation experiments 1690 U/g of L-glutaminase was obtained at pH 8.5, temperature
27.5C, moisture content 94% (w/w), inoculum concentration 1.9 mL,
glucose 1.1 g, and L-glutamine 1.4 g. The maximum L-glutaminase yield
was increased up to 1690 U/gds by hybrid ANNGA optimization. With
the help of FFNNGA optimization 65% of improvement in enzyme production was achieved in SSF.
3D surface plots were drawn using FFNNGA predicted data (Fig. 5).
The surface plots were used to determine the interaction of selected factors
on L-glutaminase production. The interaction of L-glutamine and glucose
on L-glutaminase yield was represented in Fig. 5A. The surface plot indicated that both compounds above 1 g give the higher amount of enzyme
production. Contradicting reports on role of L-glutamine are available in
the literature. According to Iyer and Singhal (2007) L-glutamine acted as
inducer for L-glutaminase production by Z. rouxii; however Kashyap
et al. (2002) noticed the retarding effect of L-glutamine on L-glutaminase
production by Z. rouxii in SSF. In Bacillus licheniformis L-glutamine acted
as a sole carbon and nitrogen source for L-glutaminase production (Cook,
Hoffman, & Bernlohr, 1981). Sathish and Prakasham (2010) also noticed
that L-glutamine acted as an inducer as well as sole nitrogen and carbon
source in Bacillus sp. RSPGLU. In the present study, L-glutamine acted
as an inducer as well as nitrogen source. The inducer effect of L-glutamine
is concentration dependent in this marine fungus, at higher L-glutamine
concentrations, the enzyme production was reduced. Fig. 5C represents
interaction of temperature and moisture content which is important in
SSF. At higher temperatures higher amount of moisture content was necessary for attain the higher amount of L-glutaminase production by A. flavus
MTCC 9972. Overall with the help of FFNNGA optimization 65% of
enzyme production was improved in SSF. This study reveals that L-glutaminase production in SSF by A. flavus MTCC 9972 is controlled by various
physical and nutritive parameters and their concentration.

Fig. 5 Interaction influence of selected fermentation factors on L-glutaminase production by A. flavus MTCC 9972 (A) glucose concentration
vs L-glutamine concentration, (B) pH vs glucose concentration, (C) temperature vs moisture content, and (D) moisture content vs inoculum
concentration.

Optimization Methods for Augmentation of Marine Enzymes

113

3. CONCLUSIONS
Cost of fermentation and downstream is the major limiting parameters
to fix the final price of enzymes in the market. Present study reveals that a
sequential optimizations procedures could increase the enzyme production
in the SSF. In the first step solid substrate was optimized by MD. With the
help of MD the L-glutaminase production was improved from 453 to
516 U/g (14% improvement). Later, by selecting the suitable carbon and
nitrogen sources by PBD the enzyme production was increased from 516
to 892 U/g (72% enhancement). Further by optimizing the various parameters and their levels by FFNNGA the L-glutaminase production was
enhanced from 892 to 1690 U/g (89% increase). By using MDPBD
FFNNGA in a progressive way 2.7-fold increases in L-glutaminase production by A. flavus MTCC 9972 was achieved in SSF. Overall, this study shows
that optimizing the various parameters such as solid substrate composition,
environmental, and nutritive parameters in SSF, a significant enzyme production could be achieved without any alteration of genetic material of
microorganisms.

REFERENCES
Chandrasekaran, M., & Kumar, S. R. (2010). Marine microbial enzymes. Biotechnology, 9,
115.
Chiranjeevi, P. V., Pandian, M. R., & Sathish, T. (2014). Integration of artificial neural network modeling and genetic algorithm approach for enrichment of laccase production in
solid state fermentation by Pleurotus ostreatus. Bioresource Technology, 9, 24592470.
Cook, W. R., Hoffman, J. H., & Bernlohr, R. W. (1981). Occurrence of an inducible glutaminase in Bacillus licheniformis. Journal of Bacteriology, 148, 365367.
El-Sayed, A. S. (2009). L-Glutaminase production by Trichoderma koningii under solid-state
fermentation. Indian Journal of Microbiology, 49, 243250.
Estrada-Badillo, C., & Marquez-Rocha, F. J. (2003). Effect of agitation rate on biomass and
protease production by a marine bacterium Vibrio harveyi cultured in a fermentor. World
Journal of Microbiology and Biotechnology, 19(2), 129133.
Foresee, F. D., & Hagan, M. T. (1997). GaussNewton approximation to Bayesian regularization. In Proceedings of the 1997 international joint conference on neural networks.
Han, B., Mab, Y., Romboutsa, F. M., & Nout, M. J. R. (2003). Effects of temperature and
relative humidity on growth and enzyme production by Actinomucor elegans and Rhizopus
oligosporus during sufu pehtze preparation. Food Chemistry, 81, 2734.
Hymavathi, M., Sathish, T., Brahmaiah, P., & Prakasham, R. S. (2010). Impact of carbon and
nitrogen sources on L-asparaginase production by isolated Bacillus circulans (MTCC
8574): Application of saturated PlackettBurman design. Chemical Biochemical Engineering
Quarterly, 24, 473480.
Imada, A., Igarasi, S., Nakahama, K., & Isono, M. (1973). Asparaginase and glutaminase
activities of microorganisms. Journal of General Microbiology, 76, 8599.

114

T. Sathish et al.

Iyer, P., & Singhal, R. S. (2007). Production of glutaminase (E.C.3.2.1.5) from

Zygosaccharomyces rouxii: Statistical optimization using response surface methodology.
Bioresource Technology, 99, 43004307.
Kashyap, P., Sabu, A., Pandey, A., Szakacs, G., & Soccol, C. R. (2002). Extra-cellular L-glutaminase production by Zygosaccharomyces rouxii under solid-state fermentation. Process
Biochemistry, 38, 307312.
Katikala, P. K., Bobbarala, V., Tadimalla, P., & Guntuku, G. S. (2009). Screening of L-glutaminase producing marine bacterial cultures for extracellular production of
L-glutaminase. International Journal of ChemTech Research, 1, 12321235.
Kumar, S. R., & Chandrasekaran, M. (2003). Continuous production of L-glutaminase by an
immobilized marine Pseudomonas sp. BTMS-51 in a packed bed reactor. Process Biochemistry, 38(10), 14311436.
MacKay, D. J. C. (1992). Bayesian interpolation. Neural Computation, 4, 415447.
Mizytani, J. R. S. (1997). Neuro-fuzzy and soft computing. New Jersey: Prentice Hall.
Moller, M. F. (1993). A scaled conjugate gradient algorithm for fast supervised learning. Neural Networks, 6, 525533.
Myers, R. H., & Montgomery, D. C. (1995). Response surface methodology: Process and product
optimization using designed experiments (1st ed.). USA: Wiley-Interscience.
Powell, M. J. D. (1977). Restart procedures for the conjugate gradient method. Mathematical
Programming, 12, 241254.
Prabhu, G. N., & Chandrasekaran, M. (1997). Impact of process parameters on L-glutaminase
production by marine Vibrio costicola in solid state fermentation using polystyrene as an
inert support. Process Biochemistry, 32, 285289.
Prakasham, R. S., Sathish, T., & Brahmaiah, P. (2011). Imperative role of neural networks
coupled genetic algorithm on optimization of biohydrogen yield. International Journal of
Hydrogen Energy, 36, 43324339.
Rispoli, F. J., & Shah, V. (2007). Mixture design as a first step for optimization of fermentation medium for cutinase production from Colletotrichum lindemuthianum. Journal of
Industrial Microbiology and Biotechnology, 34, 349355.
Sabu, A., Keerthi, T. R., Kumar, S. R., & Chandrasekaran, M. (2000). L-Glutaminase production by marine Beauveria sp. under solid state fermentation. Process Biochemistry, 35,
705710.
Sangeetha, P. T., Ramesh, M. N., & Prapulla, S. G. (2004). Production of fructosyl transferase by Aspergillus oryzae CFR 202 in solid-state fermentation using agricultural
by-products. Applied Microbiology and Biotechnology, 65, 530537.
Sathish, T., Laxmi, G. S., Rao, Ch. S., Brahmaiah, P., & Prakasham, R. S. (2008). Mixture
design as first step for improved glutaminase production in solid-state fermentation by
isolated Bacillus. Letters in Applied Microbiology, 47, 256262.
Sathish, T., & Prakasham, R. S. (2010). Enrichment of glutaminase production by Bacillus
subtilis RSP-GLU in submerged cultivation based on neural networkGenetic algorithm approach. Journal of Chemical Technology and Biotechnology, 85, 5058.
Sathish, T., & Prakasham, R. S. (2013). Intensification of fructosyltransferases and fructooligosaccharides production in solid state fermentation by Aspergillus awamori GHRTS.
Indian Journal of Microbiology, 53, 337342.
Zhang, C., & Kim, S.-K. (2010). Research and application of marine microbial enzymes:
Status and prospects. Marine Drugs, 8(6), 19201934.