Introduction

Expansion of the human population with high growth rate results in an ever-increasing demand for food and energy (Lincoln 2013). Cellulose, being present in the cell wall structure of the green plants is almost ubiquitous and constitutes the major component of forest wastes and agricultural residues (Foyle et al. 2007). Since a large fraction of this polysaccharide remains unutilized, it is not fully exploited for the production of biofuels, food, and other value-added products (Mielenz 2001). On the other hand, the unutilized cellulosic raw material poses a real problem as uncared green garbage (Kumar et al. 2020). As the wastes containing celluloses are predominantly liberated by industrial, urban, sewage sludge and agricultural activities, the absence of a proper waste management system will give rise to enormous problems (Lohri et al. 2017). Hence judicious utilization of cellulosic raw materials is the need of the hour both for economic and ecological reasons. The major concern is the development of sustainable processes or technologies that can be both efficient and economical. The rise of environmental pollution is the biggest global concern as it has a direct effect on public health (Manisalidis et al. 2020).

Although cellulose can be treated with chemicals, bioprocessing with enzyme seems to be a much better way being non-toxic and environment-friendly approach (Rodhe et al. 2011). A large number of bacterial and fungal strains possess the ability to degrade cellulosic wastes with the help of a multi-enzyme system (Johnson et al. 1982). The cellulolytic microorganisms are mostly mesophilic or thermophilic, and aerobic or anaerobic (Kubicek 1993).

The enzyme, cellulase is responsible for the hydrolysis of β-1,4-glycosidic bond present in cellulose and complete enzymatic hydrolysis of the cellulose-containing three different types of the enzymes (Henrissat et al. 1998); namely endoglucanase, (1,4-d-glucan-4-glucanohydrolase; EC 3.2.1.4), exo-cellobiohydrolase (1,4-d-glucan glucohydrolase; EC 3.2.1.74) and β-glucosidase (d-glucoside glucohydrolase; EC 3.2.1.21). Optimization of media for the production of the enzyme from microbial sources is an important step in microbiological research and involves various types of challenges. Classical researches mainly implemented the "one variable-at-a-time" approach but it is very time-consuming as it requires the conduction of various experiments, lot of time, and is laborious (Panda et al. 2007). The classical mechanism of optimizing the production can be achieved when there are no interactions between the variables (Geiger et al. 1996). In recent times the advent of modern mechanisms of implying statistical or mathematical techniques for optimizing the media has been an effective, efficient, vibrant, and robust technology (Goukanapalle et al. 2020). The optimization of various parameters like temperature, pH, speed of agitation and various components of the media using the recent technological advancements has made productivity faster with better yield (Wang et al. 2005). Figure 1 showcases the schematic representation of the optimization techniques.

Fig. 1
figure 1

Schematic representation of optimization techniques

Full size image

In this review, we focussed on the various optimization techniques for the production of the enzyme, their applicability, and feasibility to enhance yield by the involvement of fermentative processes. The discussion was also based on how techniques like response surface methodology (RSM), artificial neural network (ANN), machine learning, and other mathematical models have made the experimentation less laborious with reduced consumption of time. The present overview will also focus on the past, present, and future directions for optimization of process parameters, especially with the help of statistical modelling and computational biology for the production of microbial cellulases.

Need of optimizing media for the production of cellulase

Optimization of media is carried out in various industries to enhance the yield and desired product. Optimization of media essentially focuses on nutrient requirements for the growth of the microbial species that play an essential role in the formation of desired products. The major lookout of industries is to maximize the yield of the product with minimum cost or investment (Singh et al. 2017). Various strategies are involved in the process of optimization of media that are used frequently for better production of desired products. Figure 2 shows the schematic representations of various optimization techniques comprising both classical and modern.

Fig. 2
figure 2

Techniques for the optimization of media

Full size image

Optimization of cellulase activity by one factor at a time

The mechanism of optimization of media by the technique of "one-factor-at-a-time" (OFAT) involves the use of one factor or variable in a particular experiment keeping other variables constant. This is followed by the change in the concentration of the components being present within the media up to the desired range. The dexterity and convenience of OFAT made it a choice for the researchers for designing suitable conditions for optimizing the production of enzymes (Gonzalez et al. 1995). Production of cellulase involves many steps and is followed by optimization of various factors. The most primitive method for optimizing a system conditioned by various factors is by OFAT approach. Numerous studies have been done on production optimization of microbial cellulase in fed-batch culture. Reviews on fermentation of agro-wastes or industrial effluent by microbes for cellulase production by Hussain et al. (2019) and Saranraj et al. (2012), clearly depicted various conditions optimized for the production of cellulase. Cultivation temperature, pH, substrate concentration, presence of an additive, and incubation time are known to be optimized for the production of cellulase (Karmakar and Ray 2010; Mukherjee et al. 2011) by the conventional method. A recent study by Goukanapalle et al. 2020 showed the optimization of cellulase from the endophytic fungi by OFAT (Goukanapalle et al. 2020). The negative aspects of this process are that estimation obtained from the factors are prone to experimental errors and time consuming. 

Optimization of cellulase by RSM

As the conventional OFAT method becomes ineffective and time-consuming (Jeya et al. 2009), researchers are searching for a more advanced method to perform the optimization process within a short period. The use of statistical tools for designing the experiments is the latest trend in bioprocess development and its optimization. Techniques employing statistical tools can be used to check the correctness of the response and the importance of selected factors chosen for the experiment. The failure of the classical method of one-variable-at-a-time (OVAT) bioprocess design in analysis of the collective effect of all selected factors in the fermentation process shifts the focus towards statistical techniques (Selvendran 1985). One such statistical method is RSM which employs various mathematical and statistical techniques for planning experiments, developing models, exploring optimum conditions of independent variables for favorable responses, and assessing the comparative influences of various factors in the presence of complex interactions (Bezerra et al. 2008). It involves a collection of optimal production conditions and controllable variables for yielding a polynomial equation for the fermentation process that helps in the determination of interaction between observed results and governable variables (Zambare and Christopher 2011). It involves identifying the cumulative effect of the variables on the responses (Pandey et al. 1999; Mandal and Ghosh 2018). RSM helps in finding the optimal values for the time of fermentation, substrate amount, and range of temperature than a specific single value, enabling more flexibility during process advancement. A goodness of fit of the statistical models with the experimental data indicates that the empirical models derived from RSM can be used efficiently to describe the association between the dependent and independent variables.

RSM is an efficient and robust mathematical approach that comprises various statistical experiments and the utility of various regression analyses for assessing the optimal conditions for maximum yield (Franco-Lara et al. 2006). RSM requires an accurate model design  on which depends that how correctly it will predict the relationship to the experimentally obtained data (Aanchal et al. 2016).

The RSM is an estimation of the response function (Y), taking various independent variables of cellulase like—pH temperature, concentration, etc., and taking the help of mathematics and statistical tools, which try to fit the empirical models for the experimentally obtained data (Aanchal et al. 2016). Various order designs are used for optimization using RSM. However, the optimization of the microorganism-produced cellulase is commonly done by first and second-order designs. The total ideal fit to the absolute function is important to be calculated in both domains, where two or more responses are present (Oehlert 2000). The various RSM designs are Plackett–Burman design (PBD), central composite design (CCD) (Somayajula et al. 2012; Ozer et al. 2009), Box-Behnken design (BBD) (Box and Hunter 1957), Doehlert design (Doehlert and Klee 1972), Taguchi design (Velazco 1991) that play a significant role in the mechanism of optimization for the maximum yield of the product. Some of these designs along with their applications have been discussed in Table 1.

Table 1 RSM associated optimization of enzyme production
Full size table

Generally, in experiments of optimization of cellulase production, some factors like enzyme concentration, pH, temperature, substrate concentration are selected based on the original center points (Jeya et al. 2009; Jabasingh and Valli 2012). Several trial runs include those for the factorial design, axial points, and replication of the central points. The experiment is repeated at least three times and the average value of the triplicate is the required response function. In an experiment by Jeya et al. (2009), statistical optimization of hydrolysis condition in the saccharification of rice straw using cellulase from T. hirsuta, by RSM, a CCRD (central composite rotary design) for four factors—enzyme concentration, pH, temperature, substrate concentration was selected. A total of thirty experimental trials were performed. The experiments were performed in triplicate and the average value of the data was the response value that gives the enhanced saccharification percentage (Jeya et al. 2009). An optimal substrate concentration, pH, and temperature were tested for optimizing the saccharification process by RSM, based on the initial results. The CCRD design responses were put in a second-order polynomial equation and the statistical significance of the equation was assessed by the F-test for variance analysis. This proved the regression to be significant statistically. The value that indicated the significance of the model and showed the confidence interval is the ‘Prob > F’. The coefficient determination indicated that the model was suitable for representing the real relationship between the selected reaction variables, quite adequately. Three-dimensional RSM plots were obtained by plotting any two variables, keeping the conversion percentage (i.e., the response) on the Z-axis while other variables were kept at their zero levels. The effects of substrate concentration and enzyme on the process of saccharification and its yield, along with other important results can be obtained from surface plots and iso-response contours.

RSM is used for the optimization of cellulase production from various wastes like water hyacinth (Karmakar and Ray 2011a, b, c), pineapple waste (Saravanan et al. 2013) rice bran, corn bran, sugarcane bagasse, rice straw and sawdust (Hareesh et al. 2016); rice bran and coconut water (Gozan et al. 2018) (Table 1). RSM has been the most commonly used experimental design to optimize the production of various products. As this process helps in the evaluation of multiple factors and their interactions on one or more response variables, the method shows its efficacy. The implementation of RSM has been shown to enhance the production of cellulase by diverse types of microbial species by deciphering the clues for selections that influence the factors affecting the activity of enzymes. Among various types of statistical packages, the most commonly used software is Design-Expert, which helps in easier optimization of variously used dependent and independent variables (Aanchal et al. 2016).

Optimization by molecular docking

The extent of interaction between the enzymes and the target substrate regulates the degradation (by the enzymes) and can be presumed through in-silico analysis by molecular docking methods using various tools (Table 2). During molecular docking analysis, the molecules interact with one another in a favorable configuration resulting in the development of a perfect-fit interactive complex (Kim et al. 2006). Docking interaction between cellulose and related enzymes, with respective substrates can be studied using various molecular docking softwares like Schrodinger, AutoDock Vina, Haddock, etc. Computational tools like PyMol, RasMol, Avogadro can be used as molecular visualization tool to understand the chemistry and structures of the enzyme macromolecules. The ligand and substrate binding active sites can be determined by the analysis of the region in proximity to the bound ligands or macromolecules. The value of energy minimization provides the value of the molecular stability and flexibilities of both the enzyme and substrate that play an important role in the docking interaction. The variation in the interactions provides different values for Gibb's energy and thus helps in the optimization of the molecular interaction. A large number of docking studies have been performed between β-glucosidase and cellobiose, cellotetriose, and cellotetraose to reveal the key residues interacting with the substrates through the formation of hydrogen bonds (Mazlan and Khairudin 2010; Kar et al. 2017; Wickramasinghe et al. 2017). In a study conducted by Selvam et al. (2017), the substrate specificity of cellulase from Acinetobacter sp. was determined by 'flexible ligand, rigid receptor' docking studies, where the binding affinities of the enzyme were tested with the substrates laminaribiose cellotetraose, cellobiose, and cellotetriose. In the study, cellulase showed highest affinity to cellotetraose with the binding energy of − 7.87 kJ/mol. Similar docking analyses were performed by Paul et al. (2020) involving five microbial cellulases with β-d glucose to understand binding affinity and the enzyme–substrate interaction that is associated with cellulose hydrolysis. Their study revealed that the endoglucanase-1 enzyme from Streptomyces sp. had the highest affinity of − 5.61 kcal/mol for cellulose.

Table 2 Bioinformatics tools for optimizing the interactions of cellulases
Full size table

These in-silico dynamics and molecular docking approaches can be very useful in real-time screening of cellulases having high catalytic activity and stability, as compared to the laboratory methods, as they save time and effort and are comparatively economical. Molecular docking analysis is often a convenient and handy method of assessing whether mutant enzyme variants serve better catalytic efficiencies as compared to wild-type enzymes without the need for an experimental design involving conventional colorimetric sugar assays, and has been extensively used in research in recent times (Table 2).

Optimization of cellulase production by machine learning

The activity of an enzyme can be governed by several factors such as growth and stress associated environmental conditions that control the ultimate yield of the enzyme.

Even though RSM has proven significant for the optimization of cellulase production, limited research has been done to date on the usage of artificial intelligence (AI), an inner subset of machine learning (ML) tools. In the ML approach, scientists choose specific environmental factors based on which an orthogonal experimental design is formulated. Process variables can be analyzed and shortlisted based on their response to enzyme production by proper parameter design and subsequent logical steps (Cheng et al. 2011). There are various recursive cycles of logical step algorithms and statistical models designed for such purpose of screening process variables and optimizing the parameters; one such widely used is the backpropagation algorithm by ANN. ML uses training datasets to make predictions and decisions by which a foreign dataset can be screened and parameters can be optimized. Directed evolution of enzymes and wet laboratory experimental analyses produce huge data with the potential to be used as training datasets (Li et al. 2019). Therefore, ML techniques are often favored in protein engineering and statistical optimization of enzymes such as cellulases. ML techniques were applied to study the effect of different process conditions on the fermentation of sugarcane bagasse by Saccharomyces cerevisiae (Fischer et al. 2017). The ML-algorithm (MLA) was used to analyze the nature of the cellulase enzyme (Sambasivarao et al. 2014). Recent studies have shown that the use of the ML-based approach for quantitative structural property and activity relation has shown its efficiency and accuracy. The dual-use of ML-based prediction and experimental library generation has markedly enhanced the production of enzymes. To develop the sequence-function relationships of enzyme activity, ML needs a substantial amount of information both in the terms of quality and quantity and thus this results in the uncertainty of the model thereby brings about its impact on the success of the function. Moreover, the data obtained on the enzyme descriptors should possess a correlation with the observed fitness of the mathematical function that helps in the efficient prediction at sufficient levels. Thus, a good amount of information is needed for ML-based predictive tools for optimizing enzyme production and activity (Siedhoff et al. 2020). ML was applied for the identification of unknown cellulolytic microbes and the prediction of their properties by metagenomic analysis (Foroozandeh Shahraki et al. 2020)

In biological studies, ANN methods were often used in the mid-1980s, with high applicability in mathematical methods (Bhat et al. 2000) as MLA. With the use of a computing system for the improvement of the research, ANN has shown a potential application in the procedure of standardization of regulating parameters for enzyme production (Himmel et al. 1999). Despite high demand in industries, the production of cellulase on a large scale is hampered due to inadequate optimization of the enzyme production (Wu et al. 2005). To enhance the yield of cellulase and to make it available at a nominal cost, attention was given to the production parameter optimization process (Filos et al. 2006). But sometimes it faces severe setbacks due to various production constraints. However, this problem can be solved by the integration of an ANN in it. This technique is more accurate and gives the best possible results (Li et al. 2007). Many studies also confirm that this technique is more fruitful than RSM because the representation of non-linearity is done in a much better way in the case of ANN technology (Desai et al. 2008). As a very popular and widely used tool, ANN is applied in the construction of many new models of some bioprocesses like in the functional analyzing of different proteomic or genomic sequences (Liao et al. 2008). This method is generally taken as fitness function and is used for the process of construction of the genetic algorithms (GA) (Wu et al. 2005). It uses the natural selection method which follows the evolution and by this method, the results are well obtained. Recently, ANN was jointly used with GA for the process of optimization. This has become a very friendly way to resolve the problems of optimizations in different bioprocess progressions. These can be mechanistic or theoretical or both. This is eventually used as a new coupling condition method, as a technology based on the ANN, which is more useful compared to RSM processes, and is used for the improvement of the yield of the cellulase by the process of immobilization (Himmel et al. 1999). The advantage of ANN-based prediction is that it allows the mechanism of rational optimization of various types of variables and also possesses the ability to predict the optimization conditions when a sufficient amount of information or data is not available (Bhotmange and Shastri 2011). The process of ANN can be used to stimulate the process of hydrolysis of enzymes for a range of various input variables and help in determining the optimization conditions and complex interactions of variables within the same set of time (Bhotmange and Shastri 2011). Thus, ANN provides a potential complex system that helps in better prediction of optimization parameters (Gama et al. 2017a, b). Improved cellulase activity was found to be achieved using ANN (Fig. 3). In various studies conducted over the years, the ANN architecture is comprised of three neurons namely, the pH, carbodimide concentration, and coupling time which are there in the input layer, the neurons which are hidden in the interior layer and one neuron, which is the activity yield (Roy et al. 2003). Now, the new scaled data are taken for training in an ANN model by using a back-propagation algorithm. The tangent sigmoid and pure linear functions are used to transfer functions in the hidden and output layers of the ANN (Rajoka et al. 2007). So, the mean square error which is taking place between the observations for the output neurons and the actual outputs is usually calculated and propagated backward by the network itself. This algorithm is then adjusted through the weight of each of the other neurons (Taniguchi et al. 1989). After getting the mean square error, we can calculate the training time, and eventually, then we can end it and build the corresponding ANN (Gaur and Lata Khare 2005) (Figs. 3, 4).

Fig. 3
figure 3

Enhanced cellulase activity using optimization by ANN

Full size image

Chang et al. (2011) and Gautam et al. (2011) optimized the cellulase production by solid-state fermentation of Trichoderma viride using wheat straw and wheat bran, using AI-based techniques, where a backpropagation network was designed with Levenberg-Marquardt training algorithm. The pure linear functions and tangent sigmoid were used as the transfer functions in the output layers and hidden layers of ANN respectively. The GA coupled with ANN was used to optimize the process parameters, to detect the concentration of various suitable factors. Saravanan et al. (2012a, b) used BBD statistical tool with GA to optimize the parameters for cellulase production by Trichoderma reesei using pineapple waste as a substrate, where they found the BBD model better than GA. For minimizing the number of experiments but with the accurate result, an ANN was used to model the enzymatic hydrolysis process involved during enzymatic degradation of apple pomace by various fungal cellulases (Gama et al. 2017a, b) Such result effectively could curtail the time and cost of experimentation. Caramihai and Severin (2018) optimized cellulase production in fed-batch culture of Aspergillus niger using an agricultural residue as the substrate with the application of ANN (Caramihai and Severin 2018). ANN was also applied to optimize cellulose and xylanase production by T. ressei utilizing bagasse as a sole carbon source by Singh et al. (2008). They found ANN as a good prediction tool for complex fermentation process involving two enzymes. Figure 4 shows optimization by ANN for the production of cellulase by solid-state fermentation (SSF).

Fig. 4
figure 4

Optimization by the process of ANN for production of cellulase by solid state fermentation (SSF) and observation of its production

Full size image

RSM was applied for optimization of various parameters in a submerged fermentation medium for cellulase production by Trichoderma harzanium ATCC 20846 followed by a computational fluid dynamic (CFD) simulation for additional factor (Kumar et al. 2020)

Application of genetic algorithm for optimization of cellulase production

GA is a type of evolutionary algorithm (EA) that works based on Darwin's theory of natural selection. This algorithm helps in expressing the genes as a combination of digits '1' and '0'. Chromosomes comprising a set of genes are referred to as strings. GA helps in matching and mixing independent variables to develop superior offspring resulting in the enhancement of flexibility, efficacy, and efficiency of the optimized model.

The process parameters of cellulase production by Ochrobactrum haematophilum, a cellulolytic bacterium was developed by simulated annealing (SA) algorithm to increase the production, where after comparison it was found that the SA-based RSM model can give better predictions for cellulase production than the GA-based RSM models or sole RSM models (Parkhey et al. 2017).

Exoglucanase activity from Penicillium roqueforti by SSF was achieved by the use of simplex-centroid design along with the coupling of ANN and GA. It was found that such optimization predicted an increase of production by up to 1263% when a mixture of agro-wastes namely, corn cob, green coconut shell, and sugarcane bagasse was used in comparison to the usage of the individual substrate alone (da Silva et al. 2020).

Sirohi et al. (2018) opined that GA may be used as a tool to determine the optimized conditions for the production of enzymes from wastes after its successful application using mathematical and genetic optimizers for standardizing the parameters for cellulase production by T. reesei using pea hull as carbon source.

Application of LabVIEW-based intelligent system for monitoring of bioprocesses for optimization of cellulase production

LabVIEW is a program that helps to automate the implementation of various software-associated sensors resulting in the development of a graphical programming environment. It possesses a library of functions and subroutines for various types of programming tasks. It also helps in providing library-specific data acquisition, data processing, control of instruments, analysis, and storage (Alford 2006).

Cellulase production by SSF of Aspergillus niger was evaluated by statistical design methodology. The measurements were stored in real-time by using the graphical interface programming package LabVIEW 8.2 (Farinas et al. 2011). On the other hand, in another study Myceliophthora sp. I-1D3b was used by SSF, at pilot-scale, taking sugar cane bagasse (SCB) and wheat bran (WB) as substrates and the bioreactor was forcefully aerated packed bed bioreactor (PBB) connected to LabVIEW 8.5 routine (Casciatori et al. 2013).

Conclusion and future prospects

The solid wastes, generated from agricultural activities or forests pose serious environmental problems. Major part os these wastes are biodegradable cellulosic and lignocellulose substances. Bioconversion of cellulosic wastes is usually left unattended as garbage and needs intensive research, as production of value-added materials from these wastes can be the most important way of waste management. Researches on optimization procedure before enzyme production from agricultural wastes reveal that application of a combined procedure of RSM, ANN or GA can lead to the improvement in product yield. Increased utilization of cellulosic waste not only makes it available at a nominal cost but also will show a new vista for a sustainable way of waste utilization. All the aforementioned optimization methods are found useful and more effective than OFAT method due to many reasons as discussed above. But no particular method can be said to be most preferable, as it is case-specific. The RSM is found to be advantageous over OFAT where several process parameters are involved, which are difficult to handle at a time by OFAT. On the other hand, ANN is proved to be useful, when there is insufficient detailed knowledge of the underlying process, and formulation of a reaction mechanism is not possible as it can simulate an arbitrary bioprocess. ML methods have been increasingly applied to find patterns in data that help predict enzyme protein structures, improve enzyme stability, solubility, and function, predict substrate specificity and guide rational protein design of efficient biocatalysts. But ML requires good information content in terms of the quality and quantity of the enzyme protein. Although molecular docking is not directly applied for optimization of process parameters, it can indicate the activation energy and substrate affinity of the enzyme. Similarly, the involvement of optimization for enzyme production based on genetics lacks a mathematical foundation, but it greatly improves the prediction capability of computer models. Prediction bioprocess thus can remove major bottlenecks in improving the scale-up technology for the production of microbial cellulase from left-overs. Advanced and modern technologies and high-throughput techniques of optimization are of immense importance for the production of novel microbial cellulases with greater potentiality.