Lipase production and growth modeling of a novel thermophilic bacterium: Aneurinibacillus thermoaerophilus strain AFNA

Afshin Ebrahimpour¹ · Raja Noor Zaliha Raja Abd. Rahman*¹ · Nor Hafizah Ahmad Kamarudin¹ · Mahiran Basri² · Abu Bakar Salleh¹

¹Faculty of Biotechnology and Biomolecular Sciences, Universiti Putra Malaysia, Serdang, Selangor, Malaysia
²Faculty of Science, Universiti Putra Malaysia, Serdang, Selangor, Malaysia

*Corresponding author: rnzaliha@biotech.upm.edu.my

Keywords: artificial neural network, modified response surface methodology, thermostable, organic solvent tolerant.

Financial support: The financial support by Universiti Putra Malaysia is gratefully acknowledged.

Aneurinibacillus thermoaerophilus strain AFNA as a novel isolated extracellular thermostable organic solvent tolerant lipase producing bacterium was employed in the present study. The lipase production of strain AFNA and its correlation with bacterial growth was studied via a modeling assessment by response surface methodology (RSM) and artificial neural network (ANN) techniques. The best achieved models were multilayer full feed forward incremental back propagation network and modified cubic response surface model (mRSM) using backward elimination. The highest lipase specific activity (13.1 Umg^-1) and bacterial growth (OD₆₀₀ = 3.0) were obtained at technically similar: growth temperature (53 and 53ºC), inoculum size (2.6 and 3.0%), agitation rate (118 and 115 rpm) and initial pH (7.0 and 7.2) but different medium volume (139 and 87 ml) and incubation period (48 and 38 hrs), respectively. In addition, the importance of effective parameters on the bacterial growth and lipase production was studied where pH and inoculum size were the most and the least effective factors, respectively. Significant correlation between lipase production and bacterial growth was observed when Bivariate correlation was employed to analyse the data. As a conclusion, lipase production was the result of a synergistic combination of effective parameters interactions and these parameters were in equilibrium.

Nowadays, thermostable organic solvents tolerant bacterial lipases (EC 3.1.1.3) gain wide industrial and biotechnological interest due to their novel, multifold applications and resistance to harsh industrially conditions. They are capable of catalyzing an abroad range of novel and important reactions in both aqueous and nonaqueous media. Thereby these lipases present an attractive field for future research and create interest to isolate and study of novel thermostable organic solvents tolerant lipase producing bacteria and the factors affecting their lipase production (Sharma et al. 2001; Svendsen, 2004; Hasan et al. 2006).

Predictive models have been accepted as informative tools for rapid and cost effective study of microbial growth, their products development, risk assessment and scientific purposes (Ross, 1999). Response surface methodology (RSM) is a mathematical modelling system, which assesses the relationships between the response(s) and the independent variables (Manohar and Divakar, 2004), and defines the importance of the independent variables, alone or in combination, in the model. Although RSM is the most frequently used techniques for modelling the biological processes, it is not applicable to all modelling studies (Baş and Boyaci, 2007b). Artificial neural networks (ANNs) occupy a place of prominence, in the past decade, among modeling systems in biological studies (Dutta et al. 2004). Indeed, an artificial neural network is an adaptive data processing paradigm that is inspired by the way biological nervous systems process information. ANNs learn by examples via a parallel processing style, improve their performance and, therefore, produce reasonable response(s).

In the present investigation, the lipase production behaviour of a newly isolated thermophilic bacterium; Aneurinibacillus thermoaerophilus strain AFNA and its correlation with bacterial growth was studied via a modelling assessment by RSM and ANN techniques. In addition, the importance of effective parameters on the bacterial growth and lipase production was studied.

Bacterial strain

The bacterial strain used in this study was a novel thermostable, organic solvent tolerant lipase producer, isolated from cooking oil contaminated soil from Selangor, Malaysia. It was identified as Aneurinibacillus thermoaerophilus strain AFNA via 16S rDNA analysis and deposited in DSMZ, Germany (DSM 21497) and NCIMB, UK (NCIMB 41584). The 16S rDNA sequence of this bacteriumhas been assigned GenBank accession numberGenBank: EF032876. This strain was preserved in sterile 16% (v/v) glycerol in Tryptic Soy Broth (TSB) at -80ºC.

Inoculum preparation

The inoculum was prepared by transferring a loopfull of bacterial cells from a fresh pure culture into 10 ml TSB and incubated at 55ºC, overnight, under 150 rpm agitation. The cells were harvested by centrifugation at 12,000 x g and 4ºC for 10 min. The bacterial pellet was resuspended in sterile normal saline solution (8.5 g/l NaCl) to give an absorbance of 0.5 at 600 nm.

Lipase production medium

In order to select the best lipase production medium, eight different media were tested. The compositions of the media were (% w/v): M1: peptone (3), yeast extract (1), NaCl (0.5), olive oil (1% v/v) (Hun et al. 2003); A1 (modified M1): M1 + CaCl₂.2H₂O (0.05) + gum arabic (1) (Ebrahimpour et al. 2008); A2 (modified M1): A1 + MgSO₄.7H₂O (0.01), FeCl₃.6H₂O (0.004) (Ebrahimpour et al. 2008); GYP: glucose (2), yeast extract (1), peptone (1), CH₃COONa·3H₂O (1), MgSO₄·7H₂O (0.03), MnSO₄ (0.01), KCl (0.05), olive oil (2% v/v) (Hun et al. 2003); M3: nutrient broth (0.325), gum arabic (1), CaCl₂·2H₂O (0.05), Tween 80 (1% v/v), olive oil (1% v/v) (Hun et al. 2003); M5: nutrient broth (0.8), triolein (1% v/v) (Hun et al. 2003); TYEM: tryptone (0.6), yeast extract (0.2), CaCl₂.2H₂O (0.02), MgSO₄.7H₂O (0.01), FeCl₃.6H₂O (0.04), olive oil (1.5% v/v) (Lee et al. 1999); MTYEM (modified TYEM): TYEM + gum arabic (1) (Ebrahimpour et al. 2008).

The media were sterilized for 15 min at 121ºC after pH adjustment to 7.0. Bacterial inoculum (2%) was then inoculated into 50 ml production medium and incubated by agitation under 150 rpm, for 48 hrs at 60ºC. The cell free supernatant was obtained by centrifugation at 12,000 x g, 4ºC for 15 min prior to lipase assay.

Data collection; lipase activity, protein content and bacterial growth

Lipase activity was assayed according to Kwon and Rhee (1986) method using olive oil as substrate. The reaction mixture, consisting of 1 ml crude enzyme (culture filtrate), 2.5 ml olive oil emulsion (properly mixed of an equal volume olive oil with sodium phosphate buffer, 50 mM, pH 7.0), and 0.02 ml of 20 mM CaCl₂, was incubated in a water bath shaker for 30 min at 50ºC under 200 rpm agitation. The enzyme reaction in the emulsion system was stopped by adding HCl (1 ml, 6 M) and isooctane (5 ml), followed by mixing for 1 min. The upper isooctane layer (4 ml) containing the free fatty acid was transferred to a test tube and properly mixed with 1 ml copper reagent. The reagent was prepared by adjusting the solution of 5% (w/v) copper (II) acetate-1-hydrate to pH 6.1 with pyridine. The free fatty acid dissolved in isooctane was determined by measuring the absorbance of the upper layer at 715 nm after mixture settlement. Lipase activity was determined by measuring the amount of free fatty acid released based on the standard curve of oleic acid in isooctane. The within-run (n = 20) coefficient of variation was < 8.0%. One unit of lipase activity was defined as 1.0 μmol of free fatty acid liberated min^-1 and reported as Uml^-1.

Protein concentration was determined according to the Bradford method using the Bio-Rad assay reagent (catalogue number 500-0006) and bovine serum albumin as standard, according to the manufacturer's instructions. In order to determine the biomass, 10 ml sample of culture was filtered (cellulose acetate filter, pore size of 0.22 µm, Sartorius) and washed properly with acetone followed by distilled water. The filter was dried at 80ºC to a constant weight. Cell growth was monitored by turbidity measurement at 600 nm (Garcia-Gimeno et al. 2005). Each sample (5 ml) was centrifuged (12,000 x g and 4ºC for 10 min) and the precipitate was washed twice by resuspending in distilled water after removing the supernatant and repeating the centrifugation process. Optical density was measured at 600 nm after resuspending the precipitate.

Experimental design

Response surface method (RSM) offers a large amount of information from a small number of experiments because of using special designs those help the appropriate model be fitted to the response(s). We chose a central composite rotatable design (CCRD), as the most popular RSM design, for the experiments due to its obvious advantages of rotability and the ability to analyse the interaction effects. A central composite design (CCD) includes three groups of design points. Factorial points that consists of all possible combinations of the +1 and -1 levels of the factors; star or axial points that have all of the factors set to 0, the midpoint, except one factor, which has the value +/- Alpha; and center points, which are points with all levels set to coded level 0, the midpoint of each factor range. Typically, the center point of the design is repeated, often four or more times to get a good estimate of experimental error (pure error variance). This gives an adequate estimate of the variation of the response and provides the number of degrees of freedom needed for an adequate statistical test of the model. To summarize, central composite rotatable designs require 5 levels of each factor: -Alpha, -1, 0, 1, and +Alpha. Rotatable designs provide the desirable property of constant prediction variance at all points that are equidistant from the design center, thus improving the quality of the prediction (Montgomery, 2004; Baş and Boyaci, 2007a; Hill and Lewicki, 2007).

Although, CCRD is appropriate for calibration the full quadratic models, a complete description of the process behaviour might require a cubic or higher order model. Design-Expert allows creating the higher order models (up to the fifth order) by adding the appropriate terms. On the other hand, rarely all of the terms of the full models are needed in an application. Model reduction (modification) consists of eliminating those terms that are not desired (Montgomery, 2004; Baş and Boyaci, 2007a; Hill and Lewicki, 2007).

A five-levels-six-factors central composite rotary design (CCRD) was employed throughout the study, requiring 33 experiments (Cochran and Cox, 1992). The fractional factorial design consisted of 16 factorial points, 12 axial points and 5 center points. The following variables and levels were incorporated: growth temperature (45-65ºC); medium volume (50-200 ml); inoculum size (1-5%); agitation rate (0-200 rpm); incubation period (24-72 hrs) and initial pH (5-9) (Table 1). The experimental data [40 points include CCRD design (Table 1) and optimization data (Table 2)] were divided into three sets: training set, testing set and validating set. Each of different combinations was tested in triplicate.

RSM modelling

The CCRD design experimental data were analyzed using Design Expert version 6.06 (Stat Ease Inc. Minneapolis, USA) and then interpreted. The predicted values obtained from RSM models were compared with actual values for testing the models (Table 3) and predicted optimal conditions (Table 2) were used as validating sets.

ANN modelling

A commercial ANN software, NeuralPower version 2.5 (CPC-X Software) was used throughout the study. Multilayer normal feedforward and multilayer full feedforward neural networks were used to predict the responses. The networks were trained by different learning algorithms (incremental back propagation, IBP; batch back propagation, BBP; quickprob, QP; genetic algorithm, GA; and Levenberg-Marquardt algorithm, LM). In order to determine the optimal network topology, only one hidden layer was used and the number of neurons in this layer and the transfer functions of hidden and output layers (sigmoid, hyperbolic tangent function, Gaussian, linear, threshold linear and bipolar linear) were iteratively determined by developing different networks. Each network was trained until the network root of mean square error (RMSE) was lower than 0.0001, and average correlation coefficient (R) and average determination coefficient (DC) were equal to 1. At the start of the training, weights were initialized with random values and adjusted through a training process in order to minimize network error.

Since the replicates at center point do not improve the prediction capability of the network (Baş and Boyaci, 2007a), the models were improved by using mean of center points. The experimental data (Table 1 and Table 3) were divided into two sets: Training set and testing set, and experimental values of predicted optimal conditions (Table 2) were used as validating set.

Verification of estimated data and correlation study

In order to test the estimation capabilities of the techniques, the predicted responses obtained from RSM and ANNs were compared with the actual responses. The coefficient of determination (R²) and absolute average deviation (AAD) were determined for finding the best models. The AAD and R² were calculated by Equation 1 and Equation 2, respectively.

    [Equation 1]

where y_i,exp and y_i,cal are the experimental and calculated responses, respectively, and p is the number of the experimental run.

    [Equation 2]

where n is the number of experimental data.

R² is a measure of the amount of the reduction in the variability of response obtained by using the repressor variables in the model. Because R² alone is not a measure of the model’s accuracy, it is necessary to use absolute average deviation (AAD) analysis, which is a direct method for describing the deviations. Evaluation of R² and AAD values together would be better to check the accuracy of the model. R² must be close to 1.0 and the AAD between the predicted and observed data must be as small as possible. The acceptable values of R² and AAD values mean that the model equation defines the true behavior of the system and it can be used for interpolation in the experimental domain (Baş and Boyaci, 2007a).

Bivariate correlation was used to test the correlation between lipase production and bacterial growth using SPSS software (Version 18).

Effect of media composition on lipase production

The production of lipases is mostly inducer-dependent (Lotti et al. 1998). Different media have different stimulation effects on lipase production (He and Tan, 2006) based on the physiological and biochemical pathways of the bacterium. In order to select the best lipase production medium, the ability of bacterium to produce lipase was tested in eight different liquid media (Figure 1) those were appropriate for the bacterium growth. These media could be categorized in four groups.

The first group included M1, A1 and A2 media. The media of this group were established based on the M1 basal medium, which was composed of peptone and yeast extract as organic nitrogen sources, olive oil as oil carbon source and Na⁺ as metal ion. Generally, microorganisms provide high yields of lipase when organic nitrogen sources are used, such as peptone and yeast extract, which have been used for lipase production by various thermophilic Bacillus sp. (Ghanem et al. 2000; Sharma et al. 2002). Yeast extract is one of the most important nitrogen sources for high level lipase production by different microorganisms (Bora and Kalita, 2007). In addition, yeast extract supplies vitamins and trace elements for the growth of bacteria and increases their lipase production (Gupta et al. 2007). A1 was a modified M1 by adding Ca²⁺ as second metal ion and gum arabic as emulsifier those increased the lipase production by 2.43 times compared to M1. A2 was a modified medium of A1 with additional metal ions, Mg²⁺ and Fe³⁺. Although the lipase production in this medium was higher than the basal medium (M1) most probably due to existence of Ca²⁺ and gum arabic, additional metal ions mentioned seems to suppress the lipase production compared to A1. Different microorganisms have different requirements for metal ions. Calcium ions play essential roles for many microbial species. They are important in maintaining cell wall rigidity, stabilizing oligomeric proteins and covalently bounding protein peptidoglycan complexes in the outer membrane (Macció et al. 2002). Lipase production by various Bacillus sp. was stimulated in the presence of Ca²⁺ alone (Sharma et al. 2002; Alkan et al. 2007) or in combination with other ions such as Mg²⁺, and Fe²⁺ (Janssen et al. 1994). The lipase production induction by this group was the highest among different groups, where A1 was the best lipase production medium.

The second group includes TYEM and its modified form, MTYEM. This group, similar to first group was composed of organic nitrogen sources, tryptone and yeast extract, and oil carbon source, olive oil. These media had Ca⁺², Mg⁺² and Fe⁺³ as metal ions. MTYEM with extra emulsifier, gum arabic, showed 2.34 times higher lipase production stimulation compared to TYEM. It has been reported by different researchers that highly branched, helically configurated, non-metabolizable polysaccharides such as gum arabic are able to enhance the lipase production. This might probably be due to the emulsification of culture media containing oil to increase the lipid surface (interfacial area between oil and water) for lipase action, detachment of lipase from the oil surface, and from binding sites at the outer membrane of Gram-negative bacteria (Winkler and Stuckman, 1979; Lee et al. 1999).

The bacterium in the last 3 tested media produced very low amount of the lipase. M3 and M5 media as the third group composed of peptone, meat extract and yeast extract as organic nitrogen sources, olive oil (triolein) as oil carbon source and Na⁺ as metal ion. In addition, M3 had Tween 80 as surfactant and extra carbon source, gum arabic as emulsifier and Ca⁺² as additional metal ion. The roles of Ca⁺² and gum arabic as lipase production stimulators, beside Tween 80, could be seen as previously was mentioned for A1, A2 and MTYEM. M3 showed 2.6 times higher lipase production compared to M5. The last medium, GYP, was made of peptone and yeast extract as organic nitrogen sources, olive oil as oil carbon source, sodium acetate and glucose as simple non oil carbon sources, and Na⁺, K⁺, Mg²⁺ and Mn²⁺ as metal ions. The lipase production in this medium almost was zero. Difference between this medium and M1 as one of the effective media on the lipase production, in spite of some additional metal ions that could act as the lipase production suppressors, was sodium acetate and glucose as simple carbon sources those can be easily used by bacteria for growth. High levels of lipase production were reported from various thermophilic Bacillus sp. in the presence of olive oil as carbon source in the culture medium (Lee et al. 1999; Eltaweel et al. 2005; Bora and Kalita, 2007). Most published experimental data have shown that lipid carbon sources (especially natural oils) stimulate lipase production (He and Tan, 2006; Kaushik et al. 2006), while carbon sources that are easily broken down and used by bacteria play an inhibitory role (Lee et al. 1999). It seems when bacteria cannot easily find the suitable carbon sources, try to produce extracellular hydrolytic enzymes such as lipases, proteases and amylases and use of other sources. Then oils not only are important for higher lipase production via increase the number of lipase producers as carbon sources but also act as lipase production inducers.

As a result, A1 production medium was chosen as the medium to be used in the further study.

Predictive models

In this research we tried to analyse, model and interpret the experimental data using two completely different processing views; response surface methodology (RSM) as a mathematical modeling system, and artificial neural network (ANN) as an adaptative data processing method based on learning and interpretation.

RSM is a collection of statistical and mathematical techniques and shows several advantages over the ANNs (Manohar and Divakar, 2004). This method offers a large amount of information from a small number of experiments and provides mathematical model and equation for system studied. In addition, RSM has been furnished by ANOVA analysis, which can help to statistically analyse the whole model produced, every single parameters involved and their interactions. Although, RSM is based on the use of a second order equation as its major drawback, the biggest mistake is fitting the all-experimental data to the quadratic model. In fact, it is possible to predict a model equation with a higher degree than the second order and modified it by removing non significant terms. We employed the ability of Design Expert (Stat Ease Inc. Minneapolis, USA) mathematically packed program throughout the study to find the best modified fitted models for the experimental data analysis (Montgomery, 2004; Baş and Boyaci, 2007a).

On the other hand, in most cases an ANN is an adaptive system that changes its structure based on external or internal information that flows through the network during the learning phase. ANNs gather their knowledge by detecting the patterns and relationships in data and learn through experience, not from programming. An ANN is formed from hundreds of single units, artificial neurons or processing elements, connected with coefficients (weights), which constitute the neural structure and are organised in layers. The power of neural computations comes from connecting neurons in a network. The ability of ANNs to learn the process characteristics with little prior knowledge is desirable and eases their implementation and heightens their modelling potential. This property makes ANNs powerful and flexible tools that are well suited to model complex relationships between inputs and outputs or to find patterns in data via non-linear, distributed, parallel and local processing and adaptation (Agatonovic-Kustrin and Beresford, 2000; Baş and Boyaci, 2007b).

Indeed ANN is a superior and more accurate modeling technique when compared to the RSM as it represents the nonlinearities in much better way (Dutta et al. 2004). On the other hand, neural networks also have the disadvantage of requiring large amounts of training data in comparison with RSM that offers a large amount of information from a small number of experiments. This advantage of RSM is because of its experimental design (Montgomery, 2004). To overcome this ANN problem, in present study, we used the RSM idea, and then a statistical experimental design, CCRD, was employed to reduce the number of experiments.

RSM models

The best fitting models were determined through multiple linear regressions with backward elimination. Finally, the modified cubic polynomial model with very small "model P-values" (< 0.0001) and large "lack of fit P-values" (0.2593 for growth and 0.3623 for specific activity) from the analysis of ANOVA (Table 4 and Table 5) and a suitable coefficient of determination (R² = 0.99) and adjusted coefficient of determination (R²_adjusted = 0.98), was highly significant to represent the actual relationships between the responses and the significant variables (Equation 3 and Equation 4).

Lipase specific activity (Umg^-1) = -18.9 + 0.6 T + 0.3 V + 4.3 IS - 2.5 Ag - 4.1 t + 25.8 pH - 0.05 T2 -8.9 E-4 V2 - 0.8 IS2 - 2.7 E-4 Ag2 - 0.01 t2 - 2.4 pH2 + 0.05 T.Ag + 0.1 T.t + 0.06 T.pH - 0.03 V.IS + 9.3 E-4 V.t + 5.0 E-3 V.pH - 6.8 E-3 IS.Ag + 0.8 IS.pH + 0.05 Ag.t + 0.01 Ag.pH - 9.9 E-4 T.Ag.t      [Equation 3]

Bacterial growth (OD₆₀₀) = -76.77 + 1.74 T - 0.02 V + 4.08 IS + 0.12 Ag - 0.15 t + 7.29 pH - 0.02 T² - 0.2 IS² - 1.8 E-4 Ag² - 2.5 E-3 t² - 0.38 pH² - 0.02 T.IS - 1.3 E-3 T.Ag + 8.8 E-3 T.t - 0.02 T.pH - 4 E-5 V.Ag + 3.8 E-3 V.pH + 1.8 E-3 IS.Ag - 0.01 IS.t - 0.2 IS.pH - 2.9 E-4 Ag.t - 8.1 E-3 t.pH      [Equation 4]

Where T, V, IS, Ag and t are symbols for growth temperature, medium volume, inoculum size, agitation rate and incubation time, respectively.

ANN models

The best produced ANN models in the present study were multilayer full feed forward incremental back propagation networks with Gaussian transfer function and one hidden layer consisted of 16 neurons. The optimized values of networks for learning rate and momentum were 0.15 and 0.8, respectively. In the case of training data set, the coefficient of determination (R²) and absolute average deviation (AAD) were 1 and 0.20%, respectively, whereas for the testing data set, R² was 1.0 and AAD was 0.08% (Table 3) and for validating data sets R² and AAD were 0.98 and 1.21%, respectively (Table 2). Comparison of predicted and experimental values in training, testing and validating data sets, not only revealed capability of ANN in prediction of known data responses (the data that have been used for training) but also showed the ability of generalization for unknown data (the data that have not been used for training) and implying that empirical models derived from ANN can be used to adequately describe the relationship between the input factors and responses.

Main effects and interactions between parameters

The interaction of the parameters and optimal value of each variable is clearly represented in the three dimensional response surface plots. Figure 2a i and Figure 2a ii, represent the three dimensional plots as function of growth temperature and inoculum size on lipase specific activity and bacterial growth, respectively, when other parameters were kept at optimum point. Maximum lipase specific activity of 13.1 Umg^-1 was obtained at the 53ºC and 2.6% inoculum size that was technically similar to the condition for maximum bacterium growth, 53ºC and 3.0% inoculum size. Further increase or decrease in these parameters led to the decrease in the responses. For extracellular enzymes, temperature influences their secretion, possibly by changing the physical properties of the cell membrane (Rahman et al. 2005). On the other hand, though higher temperature causes higher reaction rates and higher solubility of substrate and products, yet according to Henry's law (Atkins and De Paula, 2001) oxygen solubility is decreased.

Suitable inoculum size provides sufficient nutrient and oxygen levels for enough growth of bacteria and therefore, enhance the lipase production. If the inoculum size is too small, insufficient number of bacteria will lead to reduced amount of secreted lipase. High inoculum size can result in the lack of oxygen and nutrient depletion in the culture media (Rahman et al. 2005; Shafee et al. 2005).

Figures 2b i and ii, depict the interaction between medium volume and agitation rate. The maximum lipase specific activity and bacterial growth were obtained at different culture volume of 139 and 87 ml, respectively but technically similar agitation rate (118 and 115 rpm). Generally, suitable agitation lead to sufficient supply of dissolved oxygen in the culture medium (Kumar and Takagi, 1999). Agitation also promotes a reduction in nutrient particle size, favouring the nutrient homogenization in the culture medium, providing additionally a rise in mass transfer rates and nutrient uptake by bacteria, which favouring microbial growth (Beg et al. 2003).

Though medium volume may have a great effect on the enzyme production, yet Figure 2b i reveals that medium volume had no significant effect on bacterial growth at optimum condition within the chosen range (50-200 ml). Although a larger medium volume initially contains more oxygen, nutrients and space for growth of bacteria, the void in the container and subsequently oxygenation of the medium will be decreased. On the other hand, it seems that ratio of surface area to volume (A/V) is important for lipase production where higher ratio cause higher oxygenation and lipase production (Woolley and Petersen, 1994).

The combined effect of growth medium initial pH and incubation time on lipase production and bacterial growth is shown in Figure 2c i and Figure 2c ii, respectively. According to the plot, a neutral initial pH (7.0 and 7.2) caused maximum lipase production and bacterial growth. Maximum bacterial growth was achieved after 38 hrs incubation period, while the lipase production needed longer incubation time (48 hrs).

Lipases are produced throughout bacterial growth, with peak production being obtained by late exponential growth phase (Gupta et al. 2004). Figure 2c i reveals that after 30 hrs incubation (near to end of the bacterial growth phase), lipase production was stimulated and drastically increased to reach a maximum amount at 48 hrs following by decrease of lipase specific activity due to reduction of active cells, and enzymes most probably because of proteolysis by present proteases as well as physical damages.

To obtain the maximum lipase production, achievement of maximum possible number of bacteria (lipase producers) and enough lipase production stimulation is necessary. Significant correlation between lipase production and bacterial growth was observed when Bivariate correlation was used to analyse the data. Pearson correlation coefficient of 0.566 was statistically significant at the 0.01 level (2-tailed). It means that there was 32% correlation between lipase production and bacterial growth.

On the other hand, Figure 3 shows the importance percentage of effective parameters on the bacterial growth and lipase production based on the ratio of optimized ANN weights. An ANN is formed from hundreds of artificial neurons, connected with coefficients (weights). The weights are the adjustable parameters and, in that sense, a neural network is a parameterized system. The weighed sum of the inputs constitutes the activation of the neuron. The activation signal is passed through transfer function to produce a single output of the neuron. During training, the weights are optimized until the error in predictions is minimized and the network reaches the specified level of accuracy. Once the network is trained and tested it can be given new input information to predict the output. In this stage, the ratio of optimized weights shows the incorporation percentage of each input parameter in final output that can be computed and presented as an importance value. pH with 18.2% and 21% of importance on bacterial growth and lipase production, respectively was the most effective factor, while inoculum size was the least effective factor with 13.5% and 13.7%, respectively. Incubation period (18.5%), Growth temperature (16.8%), agitation rate (15.2%) and medium volume (15%) were subsequent degrees of importance for bacterial growth. This sequence was different for lipase production as follows: Growth temperature (17.7%), medium volume (17.4%), incubation period (16.9%) and agitation rate (16%).

Optimization

The optimal conditions for lipase production and bacterial growth are presented in Table 2. Among the various optimum conditions, highest lipase specific activity (13.1 Umg^-1) and bacterial growth (OD₆₀₀ = 3.0) were obtained at technically similar: growth temperature (53 and 53ºC), inoculum size (2.6 and 3.0%), agitation rate (118 and 115 rpm) and initial pH (7.0 and 7.2) but different medium volume (139 and 87 ml) and incubation period (48 and 38 hrs), respectively. Since medium volume was not an effective factor on bacterial growth (Figure 1b ii), duration of incubation, when other parameters were kept at optimum values, was critical to achieve firstly a maximum bacterial growth, then maximum lipase induction.

In the current study, ANN and modified RSM were carried out and compared to study the bacterial growth and lipase production of a newly isolated thermophilic lipolytic bacterium, Aneurinibacillus thermoaerophilus strain AFNA. The artificial neural network (ANN) method was improved by using the statistical experimental design to overcome the ANN disadvantage of requiring large amounts of training data and reduce the number of experiments. On the other hand, although as an adaptative system, ANN represents the nonlinearities in much better way than other modelling methods and, therefore it is capable to analyse more complex nonlinear data sets, the improved RSM by using equation modification (modified RSM) was comparable to ANN to provide high quality modelling for bacterial growth and lipase production. Finally, although each of the methods could be separately used to study the system behaviour, a combination of their information provided a powerful and flexible tool for modelling, interpretation and optimization of bacterial growth and enzyme production.

On the other hand, this study was a good example to support the following inference. Lipase production is the result of a synergistic combination of effective parameters interactions and these parameters are in equilibrium. Achievement of maximum possible number of bacteria as well as enough lipase production stimulation is necessary to obtain the maximum lipase production. In addition, some nutritional factors can act as "inducer" for production of extracellular hydrolytic enzymes. In the case of lipases, most published experimental data have shown that lipid carbon sources (especially natural oils) stimulate lipase production, while carbon sources that are easily broken down and used by bacteria play an inhibitory role.

AGATONOVIC-KUSTRIN, S. and BERESFORD, R. (2000). Basic concepts of artificial neural network (ANN) modeling and its application in pharmaceutical research. Journal of Pharmaceutical and Biomedical Analysis, vol. 22, no. 5, p. 717-727. [CrossRef]

ALKAN, H.; BAYSAL, Z.; UYAR, F. and DOGRU, M. (2007). Production of lipase by a newly isolated Bacillus coagulans under solid-state fermentation using melon wastes. Applied Biochemistry and Biotechnology, vol. 136, no. 2, p. 183-192. [CrossRef]

ATKINS, P. and DE PAULA, J. (2001). Physical Chemistry. 7^th edition, Freeman, Oxford, Oxford University Press. 1149 p. ISBN 0716735393.

BAŞ, D. and BOYACI, I.H. (2007a). Modeling and optimization I: Usability of response surface methodology. Journal of Food Engineering, vol. 78, no. 3, p. 836-845. [CrossRef]

BAŞ, D. and BOYACI, I.H. (2007b). Modeling and optimization II: Comparison of estimation capabilities of response surface methodology with artificial neural networks in a biochemical reaction. Journal of Food Engineering, vol. 78, no. 3, p. 846-854. [CrossRef]

BEG, Q.K.; SAHAI, V. and GUPTA, R. (2003). Statistical media optimization and alkaline protease production from Bacillus mojavensis in a bioreactor. Process Biochemistry, vol. 39, no. 2, p. 203-209. [CrossRef]

BORA, L. and KALITA, M.C. (2007). Production and optimization of thermostable lipase from a thermophilic Bacillus sp. LBN 4. The Internet Journal of Microbiology, vol. 4, no. 1.

COCHRAN, W.G. and COX, G.M. (1992). Experimental Design. New York, Wiley. 640 p. ISBN 9780471545675.

DUTTA, J.R.; DUTTA, P.K. and BANERJEE, R. (2004). Optimization of culture parameters for extracellular protease production from a newly isolated Pseudomonas sp. using response surface and artificial neural network models. Process Biochemistry, vol. 39, no. 12, p. 2193-2198. [CrossRef]

EBRAHIMPOUR, A.; RAHMAN, R.N.Z.R.A.; CH'NG, D.H.E.; BASRI, M.; and SALLEH, A.B. (2008). A modeling study by response surface methodology and artificial neural network on culture parameters optimization for thermostable lipase production from a newly isolated thermophilic Geobacillus sp. strain ARM. BMC Biotechnology, vol. 8, no. 1, p. 96. [CrossRef]

ELTAWEEL, M.A.; RAHMAN, R.N.Z.R.A.; SALLEH, A.B. and BASRI, M. (2005). An organic solvent-stable lipase from Bacillus sp. strain 42. Annals Microbiology, vol. 55, no. 3, p. 187-192.

GARCIA-GIMENO, R.M.; HERVAS-MARTINEZ, C.; RODRIGUEZ-PEREZ, R. and ZURERA-COSANO (2005). Modeling the growth of Leuconostoc mesenteroides by Artificial Neural Networks. International Journal of Food Microbiology, vol. 105, no. 3, p. 317-332. [CrossRef]

GHANEM, E.H.; AL-SAYED, H.A. and SALEH, K.M. (2000). An alkalophilic thermostable lipase produced by a new isolate of Bacillus alcalophilus. World Journal of Microbiology and Biotechnology, vol. 16, no. 3, p. 459-464. [CrossRef]

GUPTA, R.; GUPTA, N. and RATHI, P. (2004). Bacterial lipases: An overview of production, purification and biochemical properties. Applied Microbiology and Biotechnology, vol. 64, no. 6, p. 763-781. [CrossRef]

GUPTA, N.; SAHAI, V. and GUPTA, R. (2007). Alkaline lipase from a novel strain Burkholderia multivorans: statistical medium optimization and production in a bioreactor. Process Biochemistry, vol. 42, no. 4, p. 518-526. [CrossRef]

HASAN, F.; SHAH, A.A. and HAMEED, A. (2006). Industrial applications of microbial lipases. Enzyme and Microbial Technology, vol. 39, no. 2, p. 235-251. [CrossRef]

HE, Y.Q. and TAN, T.W.(2006). Use of response surface methodology to optimize culture medium for production of lipase with Candida sp. 99-125. Journal of Molecular Catalysis B: Enzymatic, vol. 43, no. 1-4, p. 9-14. [CrossRef]

HILL, T. and LEWICKI, P. (2007). STATISTICS Methods and Applications. StatSoft, Tulsa, OK.

HUN, C.J.; RAHMAN, R.N.Z.A.; SALLEH, A.B. and BASRI, M. (2003). A newly isolated organic solvent tolerant Bacillus sphaericus 205y producing organic solvent-stable lipase. Biochemical Engineering Journal,vol. 15, no. 2, p. 147-151. [CrossRef]

JANSSEN, P.H.; MONK, C.R. and MORGAN, H.W. (1994). A thermophilic, lipolytic Bacillus sp., and continuous assay of its p-nitrophenyl-palmitate esterase activity. FEMS Microbiology Letters, vol. 120, no. 1-2, p. 195-200. [CrossRef]

KAUSHIK, R.; SARAN, S.; ISAR, J. and SAXENA, R.K. (2006). Statistical optimization of medium components and growth conditions by response surface methodology to enhance lipase production by Aspergillus carneus. Journal of Molecular Catalysis B: Enzymatic, vol. 40, no. 3-4, p. 121-126. [CrossRef]

KUMAR, C.G. and TAKAGI, H. (1999). Microbial alkaline proteases: from a bioindustrial viewpoint. Biotechnology Advances, vol. 17, no. 7, p. 561-594. [CrossRef]

KWON, D.Y. and RHEE, J.S. (1986). A simple and rapid colorimetric method for determination of free fatty acids for lipase assay. Journal of the American Oil Chemists' Society, vol. 63, no. 1, p. 89-92. [CrossRef]

LEE, D.W.; KOH, Y.S.; KIM, K.J.; KIM, B.C.; CHOI, H.J.; KIM, D.S.; SUHARTONO, M.T. and PYUN, Y.R. (1999). Isolation and characterization of a thermophilic lipase from Bacillus thermoleovorans ID-1. FEMS Microbiology Letters, vol. 179, no. 2, p. 393-400. [CrossRef]

LOTTI, M.; MONTICELLI, S.; MONTESINOS, J.L.; BROCCA, S.; VALERO, F. and LAFUENTE, J. (1998). Physiological control on the expression and secretion of Candida rugosa lipase. Chemistry and Physics of Lipids, vol. 93, no. 1-2, p. 143-148. [CrossRef]

MACCIÓ, D.; FABRA, A. and CASTRO, S. (2002). Acidity and calcium interaction affect the growth of Bradyrhizobium sp. and attachment to peanut roots. Soil Biology and Biochemistry, vol. 34, no. 2, p. 201-208. [CrossRef]

MANOHAR, B. and DIVAKAR, S. (2004). Applications of surface plots and statistical design to selected lipase-catalyzed esterification reactions. Process Biochemistry, vol. 39, no. 7, p. 847-853. [CrossRef]

MONTGOMERY, D.C. (2004). Design and Analysis of Experiments. 6^th edition. John Wiley & Sons, Inc. 647 p. ISBN 047148735X.

RAHMAN, R.N.Z.A.; LEE, P.G.; BASRI, M. and SALLEH, A.B. (2005). Physical factors affecting the production of organic solvent-tolerant protease by Pseudomonas aeruginosa strain K. Bioresource Technology, vol. 96, no. 4, p. 429-436. [CrossRef]

ROSS, T. (1999). Predictive food microbiology models in the meat industry. Meat and livestock Australia, Sydney, Australia, 196 p. ISBN 0958582513.

SHAFEE, N.; ARIS, S.N.; RAHMAN, R.N.Z.A.; BASRI, M. and SALLEH, A.B. (2005). Optimization of environmental and nutritional conditions for the production of alkaline protease by a newly isolated bacterium Bacillus cereus strain 146. Journal of Applied Sciences Research, vol. 1, no. 1, p. 1-8.

SHARMA, R.; CHISTI, Y. and BANERJEE, U.C. (2001). Production, purification, characterization, and applications of lipases. Biotechnology Advances, vol. 19, no. 8, p. 627-662. [CrossRef]

SHARMA, R.; SONI, S.K.; VOHRA, R.M.; JOLLY, R.S.; GUPTA, L.K. and GUPTA, J.K. (2002). Production of extracellular alkaline lipase from a Bacillus sp. RSJ1 and its application in ester hydrolysis. Indian Journal of Microbiology, vol. 42, no. 1, p. 49-54.

SVENDSEN, A. (2004). Enzyme functionality, design, engineering, and screening. New York, Marcel Dekker, Inc., 712 p. ISBN 0824747097.

WINKLER, U.K. and STUCKMAN, M. (1979). Glycogen, hyaluronate, and some other polysaccharides greatly enhance the formation of exolipase by Serratia marcescens. Journal of Bacteriology, vol. 138, no. 3, p. 663-679.

WOOLLEY, P. and PETERSEN, S.B. (1994). Lipases: their structure, biochemistry and application. Cambridge University Press, UK, 363 p. ISBN 0521445469.