Trends in biological data integration for the selection of enzymes and transcription factors related to cellulose and hemicellulose degradation in fungi

Abstract

Fungi are key players in biotechnological applications. Although several studies focusing on fungal diversity and genetics have been performed, many details of fungal biology remain unknown, including how cellulolytic enzymes are modulated within these organisms to allow changes in main plant cell wall compounds, cellulose and hemicellulose, and subsequent biomass conversion. With the advent and consolidation of DNA/RNA sequencing technology, different types of information can be generated at the genomic, structural and functional levels, including the gene expression profiles and regulatory mechanisms of these organisms, during degradation-induced conditions. This increase in data generation made rapid computational development necessary to deal with the large amounts of data generated. In this context, the origination of bioinformatics, a hybrid science integrating biological data with various techniques for information storage, distribution and analysis, was a fundamental step toward the current state-of-the-art in the postgenomic era. The possibility of integrating biological big data has facilitated exciting discoveries, including identifying novel mechanisms and more efficient enzymes, increasing yields, reducing costs and expanding opportunities in the bioprocess field. In this review, we summarize the current status and trends of the integration of different types of biological data through bioinformatics approaches for biological data analysis and enzyme selection.

Introduction

Fungi are heterotrophic eukaryotic organisms that are widely distributed in the environment (Maharachchikumbura et al. 2021; Liu et al. 2020). For centuries, fungi have been applied in processes of human interest, such as food production and agriculture. Fungi are an important source of enzymes, platform for the discovery of new enzymes, and/or host for the production of industrial enzyme products. Different classes of enzymes and bioactive compounds from fungi, such as hydrolases, peroxidases, lipases, and laccases, are used in second generation biofuel production, the pharmaceutical industry, food processing, pulp and paper industry, textiles and bioremediation (Singh et al. 2021; Huang et al. 2021; Anasonye et al. 2014; Saravanan et al. 2021; Tomer et al. 2021). With the advent of biotechnology, the application of fungi in bioprocesses has become commonplace (Singh and Gehlot 2020; Jouzani et al. 2020). The bioprospecting of enzymes and compounds produced by fungi involves different scientific disciplines, such as genetics, microbiology, biotechnology, molecular biology, and bioinformatics (Wang et al. 2020; Saldarriaga-Hernández et al. 2020; Abrashev et al. 2021).

Bioinformatics is a scientific discipline that arose with the demand to analyze large-scale data produced by sequencing and other high-throughput techniques, encompassing knowledge of biotechnology, molecular biology, computing, and statistics (Pereira et al. 2020; Orlov and Baranova 2020; Jhalia and Swarnkar 2021). Bioinformatics can be defined as a science that links biological data storage, interpretation, and analyses with computational biology, enabling the integration of large-scale data produced through genomics, transcriptomics, and proteomics (Xu et al. 2020; Diniz and Canduri 2017; Saeys et al. 2007). Advances in bioinformatics have allowed the emergence of techniques for dealing with genome sequencing data, demanding a high computational processing capacity. Both fields evolved quickly in recent decades, and following the advances achieved by the Human Genome Project (Venter et al. 2001), many other industrially and clinically relevant fungal species lines have been sequenced (Martinez et al. 2008a; Machida et al. 2005; Van Den Berg et al. 2008; Batista et al. 2020).

The genomic sequencing of fungi is an essential technique for understanding the evolutionary principles of particular groups and searching for targets of biotechnological interest (Gurjar et al. 2019; Pramesh et al. 2020; Alberti et al. 2020). The process of genome assembly, prediction, and annotation involves diverse bioinformatics tools and analyses that are important for revealing the hypothetical function of a given gene or identifying clusters of nearby genes that are related to a particular process (Giani et al. 2020; Faksri et al. 2016; Pop and Salzberg 2008).

The complete genome of a fungus makes it possible to characterize the profile of a certain group of enzymes, such as carbohydrate-active enzymes (CAZymes) (Gujar et al. 2018; Zhao et al. 2013). CAZymes are enzymes whose activity is related to carbohydrates and are rationally grouped in the CAZy (http://www.cazy.org/) database (Huang et al. 2018). These enzymes are produced by all fungi (Glass et al. 2013) and are applied in bioprocesses related to water decontamination and the food, textile, paper, and cellulose industries. Important degradative CAZymes are also needed and are a target of prospecting by the energy sector for the production of biofuels from plant biomass (Bohra et al. 2018) through enzymatic hydrolysis. These enzymes are grouped into five catalytic modules: glycoside hydrolases (GHs), glycosyl transferases (GTs), polysaccharide lyases (PLs), carbohydrate esterases (CEs), and auxiliary activities (AAs) (Cantarel et al. 2009; Levasseur et al. 2013). In addition, some enzymes are associated with binding modules known as carbohydrate-binding modules (CBMs) (Shoseyov et al. 2006). All of these modules are subdivided into families categorized based on level-specific information.

The study of gene expression levels under certain biological conditions is fundamental for understanding the complex expression systems of eukaryotes. RNA-sequencing (RNA-Seq) techniques associated with the refinement of bioinformatics tools reliably identify and quantify gene expression under different conditions and can even be used to study the exon composition in the alternative splicing process (Geniza and Jaiswal 2017; Costa-Silva et al. 2017; Ding et al. 2017; Luecken and Theis 2019). Different approaches are used to calculate the correlations between expression data, enabling researchers to predict the coregulation between genes. An understanding of the correlations between genes can allow researchers to predict the genes involved in a particular metabolic pathway or process at a systematic level (Hurley et al. 2012; Zhang et al. 2016).

Bioinformatics tools are also fundamental for predictions and in silico modeling for target selection during the biochemical and structural characterization of a protein (Martins-Santana et al. 2018; Guzmán-Chávez et al. 2018; Silva et al. 2020). Protein engineering is a unique process that requires in-depth knowledge of the organism being studied; thus, basic information, such as structural and functional information, on the genome of the organism must first be obtained to generate a set of enzymes that are modified and adapted for a particular bioprocess (Yang and Zhang 2018; Milić and Veprintsev 2015). The prospection of new proteins and the improvement of protein efficiency are strategies to achieve the highest efficiency in different industrial processes. In addition, the future development and improvement of bioinformatics and computational tools for predicting and assigning the functions of nonidentified and/or unannotated proteins show great potential.

The advancement of the application of computational knowledge to solve biological problems has generated a revolution in the way we understand complex systems involving the regulation of genes and proteins. The integration of different types of biotechnological data is only possible through the implementation of bioinformatics tools, the development of algorithms and the construction of databases (Fig. 1). In this review, we summarize the main approaches for the integration of different types of biological data from fungi to guide the selection and design of targets for application in bioprocesses.

Fig. 1.

Data integration approach for selecting enzymes for application in bioprocesses

Application of bioinformatics in the prospecting and engineering of enzymes

Genomic studies for mining enzymes of biotechnological interest in fungi

Genomics technology is a basic tool for searching for genes of interest that may have biotechnological applications. Fungi generally have small genomes relative to other organisms (such as plants) (Stajich 2017), which facilitates the execution of draft genome sequencing projects in fungal species. However, high-quality sequenced genomes are available for few species, which highlights the challenges in this type of project (Stajich 2017; Jauhal and Newcomb 2021). The main challenges in fungal genome assembly are related to the sequencing coverage necessary to fill gaps in the assembly with long and short reads and the assembly of repeated elements in the genome, especially in intron regions, which demands sufficient computational capacity for data processing (Amarasinghe et al. 2020; Magi et al. 2018).

Although many draft fungal genomes have been sequenced in recent years (Lange et al. 2021; Lopes et al. 2020), nearly complete genomes are available for only a few model fungi, such as Saccharomyces cerevisiae (Wei et al. 2007; Peter et al. 2018), Trichoderma reesei (Martinez et al. 2008a; Li et al. 2017), Aspergillus fumigatus (Nierman et al. 2005; Fedorova et al. 2008), and Neurospora crassa (Galagan et al. 2003; Wu et al. 2014). Even these genomes still include many genes with unknown functions (Ellison et al. 2014). The biological databases used for the automatic annotation of genes often include redundant information and show low curation, which hinders the process of gene annotation, and many genes/proteins are annotated as hypothetical in a fungal genomic project (Chen et al. 2017, 2020). FungiDB (https://fungidb.org/fungidb/app) is a platform that integrates different genomic and functional information from fungi to facilitate the process of curating genomic annotation (Basenko et al. 2018).

Sequenced fungal genomes can be used to search for genes with biotechnological applications through several different approaches, such as prospecting coregulated gene clusters and secondary metabolites or protein families, including CAZymes and phosphatases; prospecting for transcription factors (TFs) related to the regulation of genes involved in a particular bioprocess; SNP calling related to gene regions; and analyzing motifs in promoter and regulatory regions.

The genomic contents of CAZymes have been characterized in a large variety of fungi (Fig. 2). The identification of the CAZyme genome content is related to the fungal degradation capacity; however, the number of CAZyme families in the genome does not seem to directly correlate with the production levels and enzymatic efficiency exhibited by the fungus (Ferreira Filho et al. 2017; Druzhinina and Kubicek 2017; Martinez et al. 2008b). Nevertheless, the study of this group of enzymes provides insights into the evolution, mode of life, and biotechnological application of fungi. Barrett et al. (2020) used the predictions and annotations of the CAZyme secretomes of 465 Ascomycota and Basidiomycota genomes to examine the evolutionary relationships of the fungal CAZyme secretome. In another study, the CAZyme contents of the proteomes of Aspergillus terreus, T. reesei, Myceliophthora thermophila, N. crassa, and Phanerochaete chrysosporium in the presence of different lignocellulosic substrates were determined, indicating that the expression of these enzymes is directly related to substrate specificity (Arntzen et al. 2020). Additionally, the type of process (liquid-state (SmF) or solid-state fermentation (SSF)) influences fungal metabolism and subsequent enzyme production as a determinant of the enzymatic activity induced (Teigiserova et al. 2021). SSF culture, a fed batch culture, presents faster oxygenation but a slower sugar supply than SmF. The process is static without mechanical energy expenditures. In contrast, SmF cultures work as homogeneous systems requiring large energy expenditures to supply oxygen at sufficiently fast rates to address the large oxygen demand. Automated fed batch supply of substrates is necessary to avoid catabolite repression (Viniegra-González et al. 2003). SSF processes have been explored for protease production due to the higher yield achieved, and novel microorganisms are being tested (López-Gómez and Venus 2021; Benabda et al. 2019; Usman et al. 2021).

Fig. 2.

Genomic annotation of CAZymes in different fungi of biotechnological interest

Fungi produce a wide variety of secondary metabolites that can be used in the pharmaceutical industry in particular, demonstrating the potential of genomic studies for the discovery of new secondary metabolites. Cheng et al. (2020) sequenced the genome of the fungus Calcarisporium arbuscula and found 65 clusters of secondary metabolite synthesis genes, including genes related to mycotoxin synthesis. Wei et al. (2020) identified a new molecule, pyranoviolin A, from the sequencing and analysis of clusters of genes from the Aspergillus violaceofuscus genome.

Software tools and scripts for data processing automatization are applied to improve genomic analyses, generating large amounts of processed data (Huber et al. 2015; Qu et al. 2016). In the stages of assembly, gene prediction, and automatic annotation, several developed software programs have been applied in each step of data processing (Table 1). For the processing of textual data in biological analysis, the most commonly used programming languages are Python and Perl, which have specific modules for bioinformatics analysis. For tabulation and statistical analysis, the main tools used are Excel and R tools.

Table 1.

Bioinformatics tools used for genome analysis

Software	Application	Operating system	Command line or graphic interface	References
SPAdes	De novo assembler	Linux	Command line	Nurk et al. (2013)
Celera assembler	De novo assembler	Unix system and Mac OS-X	Command line	Myers et al. (2000)
Pilon	Draft assembly improvement and variant detection	Unix System	Command line	Walker et al. (2014)
BWA	Error correction by low divergence alignment	Unix System	Command line	Li and Durbin (2009)
FGNESEH	Ab initio gene predictor	Windows and Unix system	Command line and Graphic interface	Solovyev et al. (2006)
AUGUSTUS	Ab initio gene predictor	Windows and Unix system	Command line and Graphic interface	Stanke and Morgenstern (2005)
MAKER	Genome annotator	Unix system and MAC OS-X	Command line	Cantarel et al. (2008)
LTR_FINDER	Repeat and transposable element annotator	Windows	Graphic interface	Xu and Wang (2007)
SignalP5.0	Signal peptide predictor	Windows	Graphic Interface	Armenteros et al. (2019)
BLAST*	alignment tool	Unix system, MAC OS-X and Windows	Command line and Graphic interface	Altschul et al. (1990)
HMMER3	Homology searches and alignment	Windows and Unix system	Command line and Graphic interface	Krogh et al. (1994)
InterProScan	Protein classification and functional analysis	Windows and Unix system	Command line and Graphic interface	Quevillon et al. (2005)
FunGAP	Annotator and evidence-based model predictor	Unix system	Command line	Min et al. (2017)
Artemis	Sequence or genome visualization, manual editing	Windows, Unix system and MAC OS-x	Command line and Graphic interface	Rutherford et al. (2000), Carver et al. (2005)

Table 1 summarizes some bioinformatics tools for sequence assembly, gene prediction, and automatic annotation that can be used in a fungal genomic project. The choice of the tool will vary according to the objective of each project.

Transcriptomics applied in the selection of enzymes in fungi

Transcriptomic analysis is used to elucidate a specific cell response state. It is an important strategy for studying the expression of large gene sets under particular conditions. Fungi quickly adapt their metabolism to different environments, and the adaptation process is driven by coordinated gene expression responses. The set of expressed genes and their expression levels are directly related to the species and growth conditions, such as the carbon source, temperature, luminosity, and humidity. These characteristics make transcription analysis a precise tool for understanding the state of the cell as well as differences regarding transcriptional regulation in different states. It is essential to understand the transcriptome (the set of all transcribed RNAs) to interpret the functional elements of the genome and reveal the molecular constituents of cells and tissues in different stages of development (Bull et al. 2000; Wang et al. 2009).

Thus, transcriptomic analysis methods have evolved in recent decades from the analysis of expressed sequence tags (ESTs) to RNA-Seq, which is a high-throughput approach for deep transcriptome sequencing that can precisely determine all transcripts and their expression levels in a controlled state, while also providing insight into the regulatory mechanisms of gene expression control. From a transcription profile, the sets of fungal gene transcripts conferring different enzymatic activities can be analyzed (Horta et al. 2014; Glass et al. 2013; dos Santos Castro et al. 2014), providing detailed information on the molecular mechanisms of enzyme production. This technology has also facilitated the genetic manipulation of strains to increase their production potential and the discovery of new enzymes or proteins (Borin et al. 2017; Zhao et al. 2018).

In fungi, ESTs provide initial information on the transcribed regions of genes. The first Trichoderma EST libraries were generated from T. reesei QM6a under biomass degradation conditions (Lorito et al. 2010). The successful application of ESTs in T. harzianum provided the first indication of gene expression in mycelium (Liu and Yang 2005) and later allowed the identification of genes with putative roles in mycoparasitism against Fusarium solani (Steindorff et al. 2012). Using ESTs, transcriptional processes have been explored in industrially relevant species; for example, glucose metabolism has been studied in the enzyme producer T. reesei (Chambergo et al. 2002), and genes with putative roles in crop contamination have been identified in Aspergillus oryzae (Akao et al. 2007) and Aspergillus flavus (Yu et al. 2004).

The arrival of next-generation sequencing (NGS) and RNA-Seq approaches has promoted many successful studies involving transcriptome analysis. These technologies produce millions of base pairs of sequencing reads in the form of short- to medium-length reads (30–400 bp reads, depending on the DNA sequencing technology) in a short time with low costs. Next-generation sequencing has accelerated RNA-Seq experiments, allowing studies of even more diverse conditions and species (Lorito et al. 2010; Wang et al. 2009). The transcriptomic field has developed very quickly, and RNA-Seq has become the preferred method for gene expression profiling (McGettigan 2013; Parkhomchuk et al. 2009; Corchete et al. 2020) by cDNA sequencing. The Illumina Genome Analyzer, HiSeq, MiSeq, and NextSeq platforms are some of the technologies for RNA-Seq that have become available since 2006. These NGS platforms are capable of paired-end sequencing, resulting in high coverage, large numbers of reads, and high-quality sequence data relative to single-end sequencing (Ambardar et al. 2016). Many studies have used RNA-Seq to elucidate enzymatic activities, protein structures, and synergistic reactions among enzymes (Horta et al. 2018; Santos et al. 2016; Mhuantong et al. 2021; Liu et al. 2021). On the basis of RNA-Seq data, genes encoding hydrolytic enzymes involved in plant cell wall degradation have been identified in different species of Trichoderma, providing important information for the selection of species and target genes for use in industrial enzyme production technologies (dos Santos Castro et al. 2014). Transcriptomic studies on well-known filamentous fungi such as T. reesei and Aspergillus niger have been employed to improve enzymatic cocktail technologies, leading to the production of new enzymes and more efficient mutants to reduce costs involved in second-generation bioethanol production (de Gouvêa et al. 2018; de Souza et al. 2011; Borin et al. 2017; dos Santos Castro et al. 2014).

Many functionalities are currently available for transcriptomic analysis, ranging from specific programs to entire platforms dedicated solely to genomic analyses, including tools for RNA-Seq assembly, mapping, annotation, expression, and differential expression analyses, etc. Regarding mapping and quantification, various tools can be used, including reference genomes, spliced read aligners, unspliced read aligners, pseudoalignment and quasi-mapping, and various tools for the processing of all possible situations in RNA-Seq analysis can be found for (Costa-Silva et al. 2017). Table 2 lists the currently available tools that are commonly used for RNA-Seq data management.

Table 2.

Tools used for mapping and differential expression analysis in transcriptome studies

Software	Application	Operating system	Command line or graphic interface	References
HISAT2	Mapping reads	Linux	Command line	Kim et al. (2019)
STAR	Mapping reads	Linux and Mac OS-X	Command line	Dobin et al. (2013)
TopHat2	Mapping reads	Linux and Mac OS-X	Command line	Kim et al. (2013)
Salmon	Quantifying the expression of transcripts using RNA-Seq data	Linux	Command line	Patro et al. (2017)
Kallisto	Quantifying the abundance of transcripts	Linux and Mac OS-X	Command line	Bray et al. (2016)
BWA	Mapping reads	Unix system	Command line	Li and Durbin (2009)
BBMap	Short read aligner	Linux	Command line	Bushnell (2014)
Cufflinks	Transcriptome assembly and differential expression analysis for RNA-Seq	Unix system and MAC OS-X	Command line	Trapnell et al. (2012)
DESeq2	Bioconductor package to perform differential gene expression analysis	Windows, Linux and MAC OS-X	Graphic interface and command line	Love et al. (2014)
EdgeR	R package for differential expression analysis	Windows, Linux and MAC OS-X	Graphic interface and command line	Robinson et al. (2010)

Data integration: coregulation networks

Due to the development of molecular biology and high-throughput technologies, the use of systems biology and combinations of data sources from different omic levels to decipher the enormous network of cellular relationships has attracted increasing interest (Kitano 2002; Aderem 2005). Regarding the industrial enzymatic engineering of fungi, several studies encompassing this holistic perspective have been published (Kubicek 2013; Akcapinar and Sezerman 2016); however, the computational modeling of these interacting structures remains a challenge.

In recent decades, the use of complex networks has emerged as a powerful analytical tool for complex systems, with common applications in diverse areas of knowledge due to the common dynamics and architecture of such structures (Wilson 1999; Strogatz 2001). By using principles of graph theory, network science offers methodological resources for elucidating the dynamics and interactions of these complex systems (Barabási 2013), including molecular biology. When using omics data, there are different possible types of representations of cellular interactions through networks, with nodes (vertices) and links (edges) representing the components of the system and their relationships (e.g., proteins and their metabolic interactions, respectively) (Vazquez et al. 2003).

In fungal research, the first network approaches were mainly based on protein interactions for metabolic reconstruction and functional inferences using omic data and biological databases such as the Kyoto Encyclopedia of Genes and Genomes (KEGG) (Kanehisa 2002) and MetaCyc (Caspi et al. 2014). However, because of the high availability of data, other types of networks have now been employed (Koutrouli et al. 2020), such as protein–protein interaction networks, sequence similarity networks, gene regulatory networks, signal transduction networks, metabolic networks, and gene coexpression networks.

Several biological databases and tools assist with the construction of these structures (Mering et al. 2003; Stark et al. 2006; Ulrich and Zhulin 2007; Barrett et al. 2012; Caspi et al. 2014). The main challenge is the method to properly measure unknown relationships among genes, especially considering the potential presence of noise and nonlinear interactions (Koutrouli et al. 2020). A common and currently employed strategy to predict gene functions is to use coexpression and coregulation data, including coexpressed gene clusters (Lawler et al. 2013; Sieber et al. 2014), Bayesian inferences (Ferreira Filho et al. 2020; Sazal et al. 2020) and coexpression networks.

With the development of high-throughput technologies, including microarrays and RNA-Seq, large amounts of gene expression data are being continuously generated. Because of the large volume of transcriptome data, scalable methods are required to decipher novel relationships (Saha et al. 2017). Through gene coexpression networks, gene–gene expression dependencies can be measured and can supply insights into functional associations (Zhao et al. 2010). Genes with a similar expression pattern across multiple samples are grouped and often participate in the same biological process (Langfelder et al. 2013). By using this principle of guilt-by-association (Wolfe et al. 2005), a gene coexpression network analysis allows the prediction of the functions of unknown genes and, accordingly, the prediction of the pathways involving those genes.

Gene coexpression network construction is based on pairwise correlations between each possible pair of genes in a dataset. Some correlation measures that are widely used to compute pairwise correlations are Pearson's correlation and Spearman's correlation (Song et al. 2012; Usadel et al. 2009). Using the correlation matrix as a similarity measure, a network is constructed in which nodes represent genes that are connected to other genes by edges based on coexpression relationships. These correlation values can also be transformed to guarantee network properties, using, for example, weighted gene correlation network analysis (WGCNA) methodology (Langfelder et al. 2013), the highest reciprocal rank (HRR) method (Mutwill et al. 2011), correlation coefficient cutoffs (Burks and Azad 2016), and mutual ranks (Obayashi and Kinoshita 2009).

With this model structure, clustering approaches are generally employed to find groups of coexpressed genes using methods of hierarchical clustering coupled with a branch cutting method (Langfelder and Horvath 2008) for defining network modules and employing additional algorithms, such as MCODE (Xu and Hejzlar 2008), for defining more precise groups. Using the guilt-by-association principle, these modules can be interpreted through functional enrichment analysis to obtain biological insights about the coexpressed genes and their impact on molecular mechanisms. Generally, the Gene Ontology (GO) (Ashburner et al. 2000) and KEGG (Kanehisa and Goto 2000) databases are the most commonly used to characterize molecular functions and metabolic pathways, respectively.

In addition to the identification of genes acting as a module in a network, an important goal is to determine which gene(s) effectively represent the behavior of the module or may be important for module robustness (Koutrouli et al. 2020). For this purpose, network centrality measures are often employed (Barabási 2013), and the most common strategy is to calculate highly connected genes (hubs). These hub genes are the central nodes in a network structure (Koutrouli et al. 2020) and are considered the central players in this structure (Li et al. 2018), but they are not always directly associated with a trait of interest (Langfelder et al. 2013).

Following a scale-free topology criterion, the WGCNA methodology has been extensively used to infer novel functions. Instead of using binary information (connected = 1, unconnected = 0) with correlation cutoffs, WGCNA uses a ‘soft’ threshold to determine the weights of the edges connecting pairs of genes, which has been proven to yield more robust results than unweighted networks (Zhang and Horvath 2005).

Because of the improvement of ‘omics’ approaches used in biotechnological research, more studies are evaluating the molecular relationships among biological systems. For example, gene coexpression network analysis has been used in various biological contexts, such as human diseases (mostly cancer), plant biotechnology, plant-pathogen interactions, etc. This approach has also been applied in the study of hydrolytic microbial enzymes for industrial bioprocesses, including second-generation ethanol technology, as described below.

In a previous study, gene coexpression networks were inferred based on a transcriptome dataset of A. niger grown under two different experimental conditions (van den Berg et al. 2010). This approach allowed the prediction of biologically relevant modules and a search for enriched putative TF binding sites, which improved the understanding of higher-order regulatory structures in the studied fungi.

Other studies have reported the use of gene coexpression network analysis in this context. Using the transcriptomes of T. reesei, T. harzianum, and T. atroviride grown on cellulose and glucose, a coexpression network was inferred for each species (Horta et al. 2018). Based on the resulting data, a set of 80 genes shared among the three Trichoderma species was obtained, which could represent a common cellulose degradation system. Similarly, Almeida et al. (2021) constructed gene coexpression networks based on the transcriptome data of a novel strain of T. harzianum grown on cellulose and glucose. For this fungus, the results indicated that several genes may function in a coordinated manner during cellulose degradation, which revealed the capacity of T. harzianum as an enzyme producer (Almeida et al. 2021).

Additionally, for species of the filamentous genus Trichoderma, a gene coexpression network based on the transcriptome dataset of T. harzianum grown on cellulose and glucose was constructed to explore CLR2 regulator activity during biomass degradation (Ferreira Filho et al. 2020). In the same study, using gene expression data for secreted proteins, a Bayesian network of induced/repressed genes in T. harzianum was inferred under cellulose-based growth conditions. The results revealed potential candidates for the validation of functions during the degradation of plant biomass in further studies.

Through WGCNA based on a transcriptome dataset of T. reesei grown on steam-exploded sugarcane bagasse, 28 highly connected gene modules were identified (Borin et al. 2018). One of these modules was enriched with the most representative core of cellulolytic enzymes, including their regulators and transporters. For several hub genes, DNA binding sites for the main activator of (hemi)cellulases, XYR1, were found in the promoter regions, which suggests a putative role of these hubs in bagasse cell wall breakdown (Borin et al. 2018).

To explore the biotechnological potential of Trichoderma spp., the WGCNA package was used to explore the genetic mechanisms related to the XYR1 and CRE1 TFs during cellulose degradation (Rosolen et al. 2020). For this purpose, the mycoparasite fungus T. atroviride and two mycoparasitic strains with the hydrolytic potential of T. harzianum were selected to model a network for each strain. Based on the transcriptome dataset obtained when the fungi were grown on cellulose and glucose, gene coexpression network analysis revealed that the strains developed different molecular mechanisms to control the regulation and the expression of genes encoding proteins related to cellulose degradation (Rosolen et al. 2020).

Recently, WGCNA was used to explore the molecular mechanisms used by Penicillium oxalicum during biomass degradation (Li et al. 2020a). Based on a transcriptome dataset for the fungus grown on two novel carbon sources, methylcellulose and 2-hydroxyethyl cellulose, 17 highly connected modules were generated. One of these modules included major cellulase- and xylanase-encoding genes of P. oxalicum as well as transcription factor- and transporter-encoding genes (Li et al. 2020a).

WGCNA is one of the most commonly used approaches for constructing and analyzing gene coexpression networks, it has recently been adapted for proteomics studies (Vella et al. 2017). Furthermore, a large-scale proteomics approach for investigating and comparing the enzymatic responses of the filamentous fungi Aspergillus terreus, Trichoderma reesei, Myceliophthora thermophila, Neurospora crassa, and Phanerochaete chrysosporium has been presented (Arntzen et al. 2020). In this study, these fungi were grown on five different substrates: grass (sugarcane bagasse), hardwood (birch), softwood (spruce), cellulose, and glucose. Gene coexpression networks were inferred for each fungus based on proteomics data and the use of the WGCNA package. The results allowed the identification of the adaptation profile of each fungus in response to the different substrates, and the results suggested the specificity of the CAZymes according to the substrate (Arntzen et al. 2020).

The gene coexpression networks of transcriptome and proteome data allow the application of computational simulations for predicting the behavior of the biological system over time or under different conditions. The knowledge acquired through these approaches paves the way for further experimental studies to validate gene function and to improve the use of filamentous fungi as enzyme producers in the biotechnology field.

In silico characterization of transcription factors related to bioprocesses

TFs control transcription by specifically binding to DNA sequences (Todeschini et al. 2014). Thus, studying these proteins can provide important resources for researchers who are interested in studying the regulation of gene expression. In this context, computational studies have been used to provide support for bioinformatics approaches for searching new targets across genomes. The obtained knowledge could be used as a basis for new biotechnological applications in the gene regulation field.

In filamentous fungi, the expression of the genes encoding plant biomass-degrading enzymes is tightly controlled by TFs, which function according to the carbon source available in the environment. In the presence of readily metabolizable sugars, such as glucose, fungal genes responsible for the expression of cellulolytic and hemicellulolytic enzymes are repressed, while in the presence of inductive carbon sources, such as cellulose, the expression of fungal genes encoding CAZymes is activated (de Paula et al. 2018).

Thus, the regulatory network responsible for cell wall deconstruction is regulated by several transcription factors, including the positive regulators XYR1 (Stricker et al. 2006), ACEII (Aro et al. 2001), LAE1 (Seiboth et al. 2012), BglR (Nitta et al. 2012), VEL1 (Karimi Aghcheh et al. 2014), and the HAP 2/3/5 complex (Zeilinger et al. 2001), and the negative regulators CRE1 (Portnoy et al. 2011), ACEI (Aro et al. 2003), and RCE1 (Cao et al. 2017). Additionally, holocellulose genes may also target other TFs, such as Xpp1 (Derntl et al. 2015), SxlR (Liu et al. 2017), and CRZ1 (Martins-Santana et al. 2020) (Fig. 3).

Fig. 3.

TFs related to the expression of hydrolytic enzymes in fungi grown in the presence of glucose or cellulose

Due to the difficulty in the heterologous expression of some recombinant proteins in prokaryotic and eukaryotic hosts, the in silico characterization of TFs related to bioprocesses is useful to guide experiments and to provide new insights into the regulation of gene expression. These bioinformatics approaches have been reported to show great potential in exploring transcriptional regulation in important enzyme biofactories, particularly in industrial filamentous species, which is explored next.

The elucidation of the molecular mechanisms involved in the expression of genes encoding hydrolytic enzymes at the transcriptional level has attracted increasing interest to allow the rational engineering of fungi with increased enzyme production levels. These efforts might enable the identification of many regulatory proteins and signaling pathways responsible for the coordination of cellulase expression under different environmental conditions.

Regarding the main regulators of the expression of cellulase genes in T. reesei, the consensus binding sequences of XYR1 and CRE1 (5′-GGCWWW-3′ and 5′-SYGGRG-3′, respectively) were initially proposed based on comparisons with homologous regulators from other organisms (Kiesenhofer et al. 2018). However, these sequences cannot be used to distinguish between genes regulated by these regulators and those that are not (Furukawa et al. 2009).

The in silico identification of the cis-regulatory elements of XYR1 and CRE1 was subsequently reported. The authors identified potential binding sites for both regulatory proteins in 22 cellulase promoters and found that CRE1 affected xyr1 expression (Silva-Rocha et al. 2014). Following a similar approach, a recent study investigated the role of the TF CRZ1 in T. reesei during biomass degradation. Using bioinformatics methods, CRZ1 binding motifs were identified in the promoter regions of genes encoding proteins related to plant cell wall depolymerization, such as cellulases, hemicellulases, sugar transporters, calcium transporters, and TFs. These findings are crucial for improving engineering attempts to construct new cellulase-responsive promoters and to understand the role of these regulators in T. reesei at the global scale.

Integrated transcriptomic data are also an option for identifying novel TFs related to the control of the gene expression of lignocellulosic hydrolases in filamentous fungi. In a recent study, the authors used high-throughput data and a comparative genomics approach to identify novel potential TFs related to the control of hydrolytic enzyme-encoding genes in T. reesei and A. nidulans (Antonieto et al. 2019). The bioinformatics approach led to the characterization of an AZF1 homolog in T. reesei and the discovery of a novel Cys2His2-type zinc finger regulator involved in plant biomass deconstruction.

Identifying the genes regulated by a TF through in silico analysis is important for understanding not only the function of the regulatory protein but also the gene expression network of the cells. Although not all of these genes are considered direct target genes in the absence of experimental support, the insights provided by the bioinformatics approaches allow us to elucidate the biological reactions occurring at a molecular level, providing targets for further target-validation experiments.

Hypothetical proteins: approach for identification and characterization

Since the genomics era began in the 1990s, the identification of open reading frames (ORFs) whose products are annotated as hypothetical proteins has become increasingly common. In biochemistry, hypothetical proteins (also called “uncharacterized” or “unknown proteins”) are by definition those for which amino acid residue comparisons according to sequence similarity with other gene/protein sequences deposited in a databank (GenBank, https://www.ncbi.nlm.nih.gov/genbank/, PDB, https://www.rcsb.org/) indicate that no accurate functions for such targets can be assigned and for which no evidence of in vivo existence is available (Sivashankari and Shanmughavel 2006). Interestingly, a quick look at genomic survey results reveals that ~ 50–70% of an organism's total genes generally have attributed biological roles, showing that a high percentage of unknown targets remain to be exploited (Bork 2000; Kawaji and Hayashizaki 2008; Armstrong et al. 2019).

Advances in the field of bioinformatics associated with studies using multiomic methodologies have contributed massively to the identification and functional assignment of hypothetical proteins (Subramanian et al. 2020). Because of these tools, it is possible to not only identify the expression profile of a target gene and the required metabolic conditions for its expression (transcriptome studies) but also to reveal whether the hypothetical protein is produced employing proteomic tools. The combination of such methodologies has enabled exciting discoveries in protein science (Pinu et al. 2019; Krassowski et al. 2020). However, the characterization of the function of a protein remains a bottleneck to be overcome in the postgenomic era.

Briefly, it is possible to enumerate up to five different strategies for the characterization of an unknown protein (Sivashankari and Shanmughavel 2006). The most accurate of these approaches is based on the resolution of the three-dimensional structure of a protein by X-ray diffraction studies or methodologies such as nuclear magnetic resonance spectroscopy (NMR) or single-particle cryoelectron microscopy (cryo-EM) (Ronda et al. 2015). There is a direct correlation between protein structure and function, so the high-resolution structure of a hypothetical protein can provide hints about its biochemical functions and reveal its biological role. Other approaches include methodologies based on protein–protein interactions (Marcotte et al. 1999). It is possible to infer the functions of proteins based on their interaction partners since proteins interact with one another to perform complementary functions. Thus, methods based on protein–protein interactions can be used to ensure the assignment of a protein role. At the genomic level, approaches based on comparative genomics and genome structure are also promising for protein function prediction (Pellegrini et al. 1999; Huynen et al. 2000). These methods focus on the relationship between genome structure and the function of a protein in a metabolic pathway, since it is expected that proteins that work together evolve together. The conservation of a gene neighborhood across different species (synteny) can also be exploited to identify protein function. Clustering approaches are useful as well, allowing the grouping of genes with a similar function in a specific cluster (Overbeek et al. 1999). Other modern protein function prediction methods can involve machine learning via high-throughput data analysis, which instead of focusing on specific protein domain prediction, addresses probability and employs accurate mathematical calculations to assign protein functions (Nakano et al. 2019; Jumper et al. 2021). Each of the approaches described here has advantages and disadvantages in the study of hypothetical proteins; however, regardless of the approach used, appropriate biochemical assays are always necessary to confirm a protein’s function.

Interestingly, studies focusing on the identification and characterization of CAZymes routinely report the presence of several targets annotated as hypothetical conserved proteins (Horta et al. 2018). These targets, which are up- or downregulated according to the cultivation conditions of the studied organisms, indicate that the major puzzle of enzymatic biomass degradation is still incompletely understood. It is worth mentioning that the saccharification of biomass is a synergistic process; thus, not all key players in the process have been discovered and characterized, and corresponding biotechnological and industrial interventions have not yet reached their highest efficiency. In this context, there remains great potential for the future development and improvement of bioinformatics and computation tools for predicting and assigning the functions of nonidentified and/or unannotated proteins.

In addition to predicting the characteristics of unknown proteins, bioinformatics tools can be used for purposes such as in silico bioprospecting of genes/proteins. This method consists of discovering genes and metabolic pathways based solely on the large number of sequences found in databases, which may be used for biotechnological purposes (Ferrer et al. 2016). In silico bioprospecting can generally be divided into two stages: the search for databases and the use of bioinformatics tools to select and analyze potential candidates (Kamble et al. 2019).

Such techniques are possible due to the continuous decline in sequencing costs, which has led to the generation of massive amounts of information from highly diverse types of genomes. The combination of several tools and methods has facilitated the discovery of new enzymes with different functions and applications, in addition to assisting in their functional characterization (Gerlt 2017). Bioprospecting can also be performed with a combination of different approaches to obtain new bioproducts, such as newly designed molecular probes that can aid in prospecting enzymes or in building metagenomic libraries (Lee and Lee 2013). An example of the application of this technique is the discovery of new enzymes related to the degradation of lignocellulosic matter.

Several hydrolytic enzymes have been obtained via the application of metagenomics tools to noncultivable microorganisms (Berini et al. 2017; Popovic et al. 2015), as in the work of Salmeán et al. (2018), who identified CAZymes from metagenomics data from marine environments. In silico bioprospecting has been used to discover fungal esterases from the CAZyme database (Dilokpimol et al. 2018), and research groups have sought to develop computational tools for identifying secondary metabolite genes in microorganisms and plants (Medema 2018). This is possible due to the constant advances in bioinformatics, resulting in large amounts of available data on genomes, metabolic pathways, operons, omics information, gene regulatory networks, differential expression, protein characteristics, and structural data deposited in several databases (Fasim et al. 2021).

Application of bioinformatics in enzyme engineering

The integration of different data and bioinformatics approaches is a very useful tool for identifying new protein targets and for directing the study of protein engineering (Florindo et al. 2018; Lenfant et al. 2017; Zhou et al. 2019).

Genetic information from different levels must be integrated to implement a successful fungal engineering approach, ranging from the gene expression profile encoding an enzyme or natural product (NP) of interest to the structural properties of a certain protein. Thus, bioinformatics plays a central role in mining, integrating, and interpreting the data from these studies. Recently, significant advances in the identification, understanding, and engineering of fungal biosynthetic gene clusters (BGCs) have facilitated the biosynthesis of fungal NPs at the global, pathway, and enzyme levels using in vivo and in vitro approaches. It advances the progress in understanding how fungal BGCs are regulated, and the subsequent applications of these novel biosynthetic enzymes as biocatalysts. In 2019, Skellam (2019) described three principal categories for the methods used in fungal engineering of NPs as follows: (1) induction of transcriptional perturbations (e.g., through epigenetic modifications or the overexpression of global transcriptional regulators); (2) manipulating specific biosynthetic pathways either in the native host or in a heterologous host; and (3) the specific engineering of enzymes to synthesize novel NPs in vivo or in vitro. Each strategy varies according to the nature of the target compound and the specificities of the fungal species from which it was produced, leading to the discovery of new enzyme classes/functions, pathways, and heterologous expression systems, low-cost processes, and high-level production.

The biochemical characterization of proteins is essential for understanding the function and activity of a particular enzyme (Manavalan et al. 2017; Uechi et al. 2020; Dilokpimol et al. 2018). One approach that can be used in this type of study is to search RNA-Seq databases to identify key enzymes involved in a given process (Borin et al. 2017; Li et al. 2020b). In T. harzianum IOC-3844, studies on a β-glucosidase from the GH1 family (rThBgl) revealed high expression of the gene according to RNA-Seq data related to plant biomass degradation (Santos et al. 2016). This enzyme was cloned and expressed heterologously in Escherichia coli, and the product was subjected to structural crystallography. In another study, the accessory protein swollenin was identified in a cluster with CAZyme enzymes in BACs of T. harzianum IOC-3844 (Crucello et al. 2015). In this study, high synergism of swollenin with a xylanase from T. viride was found (Santos et al. 2017).

For a protein to be applied in a bioprocess many times, it needs to be adapted, which is known as protein engineering (DiTursi et al. 2006). This process involves genetically modified organisms and random or induced mutations in specific domains of the protein. In silico studies can direct this process according to the comparison of a protein with the desired properties and the target protein to be engineered; through multiple alignments, synthetic analysis and phylogeny, it is possible to design a tailor-made enzyme suitable for a given bioprocess (Pucci et al. 2017; Yang et al. 2019; Dilokpimol et al. 2018). Santos et al. (2019) performed the alignment of a highly glucose tolerant β-glucosidase from Humicola insolens (ideal for application in industrial cellulose degradation) with a low-tolerance enzyme from T. harzianum and found two amino acid residues responsible for this difference. Using the technique of site-directed mutation, it was then possible to obtain a β-glycosidase from T. harzianum with high glucose tolerance.

Conclusions

In this review, we present an overview of the major bioinformatics tools that could be applied for the bioprospection of hydrolytic enzymes and TFs related to cellulose and hemicellulose degradation. As fungi are producers of potential new enzymes, approaches including genomics, transcriptomics, proteomics, and systems biology are used to reveal the degradative mechanism employed by fungi, producing an unprecedented volume of biological data. We propose biological data integration as a methodological strategy that is useful for prospecting and producing enzymes appropriate for biotechnological process. The correlation of different types of data can contribute to a better understanding of how the expression of genes, enzymes, and regulators plays important roles in pathways or reactions of biotechnological interest that are suitable for improvement. The major challenge is to integrate data from different experiments or biological databases, which often have different formats and are not promptly correlated. Hence, we assume that data integration approaches will become increasingly sophisticated and accessible, thus facilitating the understanding and prediction of the actions of complex biological systems.

Author contributions

APS and JAFF conceptualized the manuscript. JAFF, RRR, DAA, PHCA, MLLM, AHA, CAS, MACH and APS wrote the manuscript. JAFF, AHA and MLLM prepared the figures. JAFF and PHCA prepared the tables. CAS, MACH and APS revised the manuscript.

Funding

This work was supported by grants from the Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP 2015/09202-0 and 2018/19660-4), Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES, Biology Program 88882.160095/2013–01), and Conselho Nacional de Desenvolvimento Científico e Tecnológico for a Research Fellowship to APS (CNPq 312777/2018-3) and grants to CAS (CNPq Univeral 430350/2018-0). MACH received fellowship from FAPESP (2020/10536-9). AHA and RRR received PhD fellowships from FAPESP (2019/03232-6 and 2020/13420-1, respectively).

Data availability statement

Not applicable as this is a review article that does not present any new data.

Declarations

Conflicts of interest

The authors declare that they have no conflicts of interest.

Consent for publication

All authors agree to the publication.