Discovery of Thermophilic Enzyme Properties Through Structural Modeling, Phylogenetic Analysis, and Machine Learning Prediction

doi:10.21203/rs.3.rs-7756125/v1

The rapid growth of the global population and the urgent need for sustainable resource utilization have intensified research into renewable natural resources, particularly lignocellulosic biomass. Among the various enzymes involved in lignocellulose degradation, thermophilic glycoside hydrolases (GH) family 5 cellulases have gained significant attention due to their ability to sustain enzymatic activity at elevated temperatures exceeding 60°C. These high temperatures not only accelerate enzymatic reactions, improving reaction rates and process efficiency, but also enhance substrate solubility and reduce the risk of microbial contamination, making them highly valuable for paper, food, feed, pharmaceutical, and biofuel industries. In this work, we performed structural modeling of GH5 cellulases and investigated their thermophilicity, integrating data from experimental structures and computational simulations. We analyze structural characteristics such as compact protein folds, conserved active site residues, increased hydrophobic interactions, hydrogen bonds, and other interatomic contacts. Understanding these features is crucial for protein engineering and optimizing these enzymes for various high-temperature biotechnological applications.

Computational Biology

Structural Biology

Bioinformatics

cellulases

glycoside hydrolases

thermophilicity

structural analysis

computational simulations

The exponential growth of the global population, projected to reach 11.2 billion by 2100, presents significant challenges to achieving the Sustainable Development Goals established by the United Nations in 2021.¹ This population increase exacerbates environmental concerns and accelerates anthropogenic activities that contribute to climate change, resource depletion, and waste generation.^2,3 In response to these pressing issues, research efforts are increasingly focused on leveraging renewable natural resources, such as lignocellulosic biomass, while also seeking ways to reduce food waste.^4,5 It is estimated that about 1.3 billion tons of food is wasted globally each year, with a significant amount coming from agricultural and agro-industrial sources.^6–8 This highlights the critical need for sustainable solutions in managing agricultural residues and reducing overall waste generation.^9,10

Lignocellulose, the most abundant renewable biomass on Earth, is primarily composed of cellulose, hemicellulose, and lignin.¹¹ The enzymatic degradation of this complex structure, often preceded by pretreatment processes to increase accessibility, is crucial for converting biomass into biofuels and other valuable bioproducts.^12–14 Glycoside hydrolase family 5 (GH5) enzymes are key players in this process, as they break down cellulose into fermentable sugars.⁵ Thermophilic GH5 cellulases, which sustain enzymatic activity at temperatures exceeding 60°C, have garnered significant interest for their applications in high-temperature industrial processes, including biorefineries, the pulp and paper industry, textile manufacturing, and food processing.^15–17 Structural studies of these enzymes have revealed molecular features that contribute to their thermophilicity, including compact protein folds, conserved active site residues, and increased hydrophobic interactions, hydrogen bonds, and other interatomic contacts.^18,19 These characteristics are crucial for maintaining enzymatic activity at elevated temperatures, making them valuable in industrial settings where efficiency and robustness are essential.

Structural modeling of GH5 cellulases has provided invaluable insights into their molecular architecture, catalytic mechanisms, and features contributing to thermophilicity. Crystal structures of thermophile GH5 cellulases from Bacillus subtilis (PDB code: 3PZT),²⁰ Thermotoga maritima (PDB code: 3MMU),²¹ Clostridium thermocellum (PDB code: 1CEN)²² and Pyrococcus horikoshii (PDB code: 2ZUN)²³ have revealed insights into protein folds, active site architectures, and domain arrangements. In addition to structural studies, metagenomics has emerged as a powerful tool for exploring the vast genetic diversity of non-culturable microorganisms, unlocking a rich reservoir of potential novel enzymes.^9,24 However, identifying, selecting and characterizing these enzymes requires analysis of large and complex datasets. In this context, machine learning techniques could be used in the accurate prediction and classification of enzymes from metagenomic sequences with high precision. In a previous study conducted by our research group, we analyzed the gut microbiome of two native termites and identified lignocellulose-degrading enzymes.⁵ From this data set, two GH5 cellulases (GH5CelA and GH5CelB) from non-culturable bacterium have been characterized.²⁵ Both enzymes exhibited moderate thermostability, functioning effectively at temperatures ranging from 35 to 50°C.²⁶

In the present study, we employed structural modeling and machine learning to prospect novel thermophilic GH5 cellulases from termite gut metagenomes. We focused on analyzing the structural characteristics that confer thermophilicity to these enzymes, including compact protein folds and hydrophobic interactions, among other interatomic contacts such as hydrogen bonds and polar interactions. The insights gained from this study provide a foundation for selection of GH5-encoding genes from metagenomic datasets and for engineering and optimizing GH5 cellulases for high-temperature industrial applications, potentially improving their performance in biofuel production and other biotechnological processes.

Selection of GH5-encoding Genes from a Termite Gut Microbiome

In our prior studies, several contigs encoding lignocellulose degradation were identified from shotgun metagenomic sequencing of two termite gut microbiomes (Cortaritermes fulviceps and Nasutitermes aquilinus). In that study GH5 was the most predominant family involved in cellulose degradation.⁵ The contigs had been annotated using Prokka²⁷ and further scanned for GH putative enzyme families associated with plant biomass degradation with the dbCAN3 web server²⁸, and we selected all predicted GH5-encoding genes. Only candidate genes with an e-value < to 1e^-12 were selected.

Thermophilicity Prediction of Cellulases

To assess thermodynamic stability, several specialized prediction tools were used. The Metagenome Cellulase Identification and Characterization (MCIC) tool²⁹ was used for screening cellulases based on thermal and pH dependencies. This tool classifies cellulases as mesophilic (T_opt < 50°C), thermophilic (50°C < T_opt < 75°C), and hyper-thermophilic (T_opt >75°C). Additionally, we integrated ThermoPred³⁰ and iThermo³¹ predictors tools into our prediction pipeline. ThermoPred is a thermophilic enzymes predictor based on a Support Vector Machines (SVM) with a radial basis function kernel. iThermo, is a sequence-based model for identifying thermophilic proteins, it uses a multi-feature fusion strategy and a Multilayer Perceptron (MLP) model to discriminate thermophilic from non-thermophilic proteins. This tool classifies proteins as thermophilic (T_opt >60°C). These tools collectively provided insights into cellulase thermodynamic stability and identified thermophilic proteins, enhancing the comprehensiveness of our study.

Sequence Alignment and Phylogenetic Analysis

A comprehensive analysis was done by downloading 41,708 amino acid sequences of GH5 family from the CAZy database (htt://www.cazy.org/). A custom database was constructed using the CAZy database and BLASTP tool to identify subfamilies.³² The taxonomic classification of each GH5 sequence analyzed in this study was performed using the Kaiju web server ³³. The first top five matched sequences for each analysis with an e-value equal or lower than 1e^-11 were used for multiple sequence alignment using the super5 algorithm of the MUSCLE v5 aligner.³⁴ Subsequently, a maximum likelihood phylogenetic tree was obtained using the FastTreeMP tool³⁵ employing the Le-Gascuel (LG) amino acid substitution matrix with gamma correction for variability in evolutionary rates among different positions in the sequence. In addition, the BIONJ algorithm was utilized for the reconstruction of the likelihood tree based on the distance matrix obtained. The bootstrap analysis comprised 1,000 replicates. Tree visualization and manipulation were performed using the iTol v6 tool.³⁶

Physiochemical Properties and Amino Acid Usage of the GH5 Enzymes

The Isoelectric points (pI/Mw) and the grand averages of hydropathy were calculated using the ProtParam web server.³⁷ The amino acid composition of each enzyme was also analyzed using the Dayhoff metric, calculated by EMBOSS Pepstats³⁸ to provide detailed insights into the relative abundance of amino acids compared to mesophilic and thermophilic enzymes as well as to GH5CelA, previously characterized by our research group.

Additionally, the IPC2.0³⁹ web server was used to predict the isoelectric point of each enzyme. To predict the secondary structures of the enzymes, the Garnier (EMBOSS) tool was employed.⁴⁰ Furthermore, SignalP ⁴¹ was used to identify potential signal peptides on the N-terminal end of the enzymes, providing valuable information about cellular localization or secretion mechanisms.

The data mining analysis for the abundance of specific amino acids was implemented in two topological areas of structure/function relationship of relevance on thermotolerance: the core and the catalytic pocket of the enzymes. Starting from the molecular structure of three set of enzymes (thermophiles, mesophiles and candidate), the residues with higher hydrophobic character Ile, Phe, Tyr, Trp (HYD group) was mined inside the less exposed amino acids core and the full enzyme amino acids as well (Chimera software script: select #0: Ile, Phe, Tyr, Trp & :/areaSAS < = 10). The more reactive residues Lys, His, Glu (CAT group) was counted for the full enzyme amino acids but also along the catalytic pocket (Chimera software script: select #0: Glu, His, Lys & sel zr < 10).

Molecular Structure Prediction and Ligand Binding Site Prediction

Cellulase structures were predicted using advanced computational tools. Molecular models of the cellulases structure were developed with the I-TASSER server⁴² and the AlphaFold software, using its online ColabFold.⁴³ AlphaFold improves the accuracy of structure predictions by incorporating novel neural network architectures and training procedures based on the evolutionary, physical and geometric constraints of protein structures. For the model refining stage, we used FoldX ⁴⁴ software routines (script RepairPDB; http://foldx.crg.es/). FoldX identifies residues with unfavorable torsion angles or Van der Waals clashes, repairs them, and explores various rotamer combinations to achieve a new energy minimum.⁴⁵

The quality of the models was evaluated at the SWISS-MODEL Workspace (Structure Assessment tool) using QMEAN ⁴⁶ and MolProbity.⁴⁷ Model manipulation and imaging were developed using the Chimera software.⁴⁸ These structures were used as reference structures for calculation of the root mean square deviation (RMSD).⁴⁹ Interatomic interaction analysis in protein structures was conducted using the Arpeggio web server calculations.⁵⁰ The most refined structural model was analyzed using the COACH algorithm, available through the COACH metaserver.⁵¹ This method compares the submitted models with proteins in the BioLiP2 database to predict potential ligand-binding sites.⁵²

Protein Sequence Accession Numbers

The nucleotide sequences of T1Cel5 (KBCPBGKF_14422), T2Cel5 (KBCPBGKF_22703), T3Cel5 (AFHCADON_14335), T4Cel5 (KBCPBGKF_00253), and T5Cel5 (KBCPBGKF_33952) were deposited in the GenBank database as PV80055, PV800558, PV800559, PV800560 and PV800560, respectively.

Selection of Candidate Genes Based on Thermophilicity Prediction Properties

From the metagenomic data of the intestines of two termites, we identified 115 protein sequences corresponding to the GH5 family, the most predominant family involved in cellulose degradation.⁴ Using three tools employing different machine learning algorithms (MCIC, iThermo and ThermoPred), we classified these enzymes as potentially mesophilic or thermophilic (Table 1). From the 115 sequences, we selected five enzymes predicted to be thermophilic by two or more different tools, which also possessed a signal peptide and an e-value ≤ 1e^-12. The five selected GH5-encoding genes were KBCPBGKF_14422, KBCPBGKF_22703, AFHCADON_14335, KBCPBGKF_00253, and KBCPBGKF_33952, henceforth referred to as T1Cel5, T2Cel5, T3Cel5, T4Cel5, and T5Cel5, respectively (Table 1). According to BLASTP analysis (Table S1), T1Cel5 showed 61.38% identity and 99% query coverage with a GH5 protein from Treponema sp. (MDR0301551.1). Similarly, T2Cel5 had 47.48% identity and 99% coverage with a GH5 protein from Treponema sp. (MDR2575803.1), while T3Cel5 exhibited 68.73% identity and 99% coverage with another GH5 protein from Treponema sp. (MDR1839590.1). T4Cel5 had 74.56% identity and 98% coverage with a GH5 protein from Treponema sp. (MDR2953146.1). In addition, T5Cel5 displayed 79.07% identity and 100% coverage with a cellulase from Chitinispirillales bacterium (MDR0303540.1). The amino acid sequences with the highest similarity were obtained from a termite gut metagenome analysis conducted by Salgado et al.⁵³

Phylogenetic Analysis

A phylogenetic analysis of GH5 was conducted to establish the evolutionary relationship of the studied enzymes. We generated a maximum likelihood tree using FastTreeMP (Fig. 1). A subfamily analysis using the CAZy database and BLASTP grouped the five GH5 enzymes into four GH5 subfamilies (GH5_2, GH25, GH5_39, and GH5_40). T1Cel5 grouped within the GH5_40 subgroup, an uncharacterized subfamily primarily represented by organisms from the Terrabacteria group. T2Cel5 and T4Cel5 were found in the same subgroup (GH5_39), composed of enzymes that are distributed across a wide variety of organisms, predominantly within the Terrabacteria group. This includes species from the families Herpetosiphonaceae and Synechococcaceae, such as Synechococcus sp. and Herpetosiphon aurantiacus, respectively. T3Cel5 grouped in the GH5_2 subgroup predominantly represented by organisms from the Terrabacteria group, including species from the phylum Firmicutes (e.g., Bacillus sp.) and Spirochaetes (e.g., Treponema sp.). T5Cel5 groups within the GH5_25 subgroup, predominantly composed of enzymes from the Actinobacteria and Bacteroidetes phyla, including thermophilic species such as Dictyoglomus thermophilum and Caldithrix abyssi. GH5_39 is characterized as a monospecific cellulase subfamily, exhibiting a single activity among its characterized enzyme members, while GH5_2 and GH5_25 are polyspecific. Additionally, this subfamily includes sequences from uncultured bacteria and members of the Spirosomaceae family.

Physicochemical Characterization and Secondary Structure Prediction

To gain insight into the physicochemical properties of the enzymes, we assessed features such as molecular weight (Mw), isoelectric point (pI), general hydrophobicity (GRAVY), aliphatic index, and instability index to predict protein characteristics (Table 2). The differences in Mw among the thermophilic cellulase candidates can be explained by both their domain architecture and the physicochemical properties data. T1Cel5 and T3Cel5, with molecular weights of 41.97 kDa and 38.89 kDa, respectively, consist solely of the cellulase domain, as shown in Fig. S1. In contrast, T2Cel5, T4Cel5, and T5Cel5, which have significantly higher molecular weights (66.6 kDa, 65.96 kDa, and 63.14 kDa, respectively), possess an additional galactose-binding-like domain, contributing to their increased Mw (Fig. S1). The GRAVY value, which evaluates hydrophobicity and hydrophilicity, serves as an indicator of protein solubility and its interactions with water. A positive GRAVY value indicates that a protein is generally hydrophobic, while a negative value suggests that it is more hydrophilic. In this study, the GRAVY values observed ranged from -0.34 to -0.45, which could indicate the hydrophilic potential of the five proteins evaluated in this study (Table 2). The aliphatic index of a protein is a measure of the relative volume occupied by the aliphatic side chain of the following amino acids: alanine, valine, leucine and isoleucine. The sequence-based analysis of the aliphatic index among five GH5 revealed a mean value of 81.33 ± 4.10. In addition, the instability index, the measure of the in vivo half-life of a protein, of the five GH5 enzymes tested revealed a mean value of 33.29 ± 6.81 (Table 2). Furthermore, the number of negatively charged amino acid residues (R-: aspartic acid and glutamic acid) is higher than the number of positive residues (R+: arginine and lysine) in the two candidates (T1Cel and T2Cel), except for T3Cel, which showed the same number of positive and negative residues. For the T4Cel and T5Cel5 enzymes, the number of positively charged residues was higher than the number of negatively charged residues (Table 2). The secondary structure analysis determines the spatial arrangement of amino acids, classifying them into helices, strands, or coils. In the identified enzymes, the secondary structure composition averaged 28.66% ± 6.50 α-helix, 25.5% ± 3.04 β-sheet, 25.44% ± 3.74 turn, and 24.38% ± 3.03 random coil. This composition typically shows a well-balanced distribution of four characteristics (Table 2).

In addition, the in silico Mw and pI for C. fulviceps and N. aquilinus were evaluated (Fig. S2A). The in silico Mw ranged from 21 to 71 kDa (C. fulviceps) and from 21 to 83 kDa (N. aquilinus). Whereas the average theoretical pI ranged from 4.5 to 7.4 for C. fulviceps and from 4.6 to 8.6 for N. aquilinus (Fig. S2A). The average GRAVY values observed were -0.42 ± 0.129 for C. fulviceps and -0.36 ± 0.129 for N. aquilinus (Fig. S2B).

Amino Acid Composition

Subsequently, the candidate cellulases were analyzed for their amino acid compositions. Regarding amino acid composition, using the Dayhoff metric, we observed that enzymes predicted as thermophilic show a higher relative occurrence of the amino acids Phe, Glu, Asp, Cys, Tyr, Trp, Pro, Asn, Leu, Lys, Ile and His compared to mesophilic enzymes (Fig. 2A). In Fig. 2B, T1Cel5 showed an enrichment in Glu, Cys, Tyr, Trp, Asn, Leu, Lys, Ile, and His compared to mesophilic enzymes, which have higher levels of Gly, Val, Thr, and Ser, and GH5CelA, which is richer in Phe, Asp, Ala, Arg, Gln, Pro, and Met. In Figure 2C, T2Cel5 contains a higher amount of Glu, Cys, Tyr, Trp, Leu, Lys, Ile, and His compared to GH5CelA, while mesophilic enzymes display a greater proportion of Gly, Val, Thr, Ser, and Asn. For Figures 2D to 2F, T3Cel5, T4Cel5, and T5Cel5 are all enriched in Glu, Cys, Tyr, Trp, Leu, Lys, Ile, and His in comparison to GH5CelA, which is richer in Phe, Asp, Ala, Arg, Gln, Pro, and Met, whereas Mesophilic shows higher levels of Gly, Val, Thr, Ser, and Asn.

In Figure 3, the thermophilic and mesophilic enzymes are further categorized based on their amino acid properties. Although mesophilic enzymes exhibited a dominant composition of small and tiny amino acids compared to thermophilic enzymes, thermophilic enzymes show a higher presence of aromatic, acidic, basic, polar and charged amino acids in their molecular composition (Fig. 3A). In Figures 3B to 3F, the categorized amino acid properties of the individual thermophilic cellulases T1Cel5, T2Cel5, T3Cel5, T4Cel5, and T5Cel5 are presented alongside GH5CelA and mesophilic enzymes. In Figures 3B to 3F, tiny and small amino acids are more enriched in mesophilic proteins and GH5CelA than in thermophilic candidates. Aliphatic amino acids are primarily enriched in GH5CelA, with the exception of T3Cel5 (Fig. 3D) and T5Cel5 (Fig. 3F). Aromatic amino acids show enrichment in most thermophilic candidates, except for T3Cel5 and T4Cel5 (Fig. 3E). GH5CelA displays the highest levels of non-polar amino acids. With respect to polar amino acids, thermophilic candidates exhibit greater enrichment, with the exception of T1Cel5 (Fig. 3B). Charged amino acids are more abundant in thermophilic candidates. Basic amino acids are predominantly present in GH5CelA, with T3Cel5 and T4Cel5 as exceptions. Lastly, acidic amino acids are enriched in all thermophilic candidates.

Tertiary Structure Prediction

We analyzed the tertiary structures of different cellulase sequences using the I-TASSER prediction server (Table 3). We selected the best model based on the C-score and TM-score. The C-scores for the selected models ranged from -1.73 to 0, indicating varying confidence levels in their predictions. The TM-scores, which measure the structural similarity to native proteins, ranged from 0.50 to 0.75, with T3Cel5 exhibiting the highest TM-score of 0.75. Afterwards, these models were refined using AlphaFold2_mmseqs2 and ColabFold.⁴⁴ AlphaFold2 not only predicts 3D structures but also evaluates the quality of the predicted models using two key metrics: the pLDDT and pTM scores. The pLDDT score, ranging from 0 to 100, signifies the confidence of the model prediction for each amino acid residue relative to the α carbon atoms. The refined models demonstrated optimal pLDDT confidence values greater than 90, with T3Cel5 achieving the highest pLDDT score of 97.2, indicating strong predictive accuracy. Structural modeling revealed that the enzymes exhibit the typical TIM barrel structure, with eight α-helices and eight β-sheets (Fig. 4A). Furthermore, the catalytic residues for each enzyme were identified: two glutamic acid residues, Glu169 and Glu279, and the histidine His250 were identified as probably involved in the catalytic site of T1Cel5; Glu149, Glu293, and His207 (T2Cel5); Glu150, Glu239 and His211 (T3Cel5); Glu156, Glu298, and His214 were identified for T4Cel5 and Glu128, Glu245 and His187 for T5Cel5 (Fig. 4C, T2Cel5). Regarding the structure and sequence alignment, the GH5 candidates contain the eight key residues Arg46, His90, Asn139, Glu140, His198, Tyr200, Glu280, and Trp313, referencing T1Cel5, which are conserved in the functional motif of the GH5 family (Fig. 5).

All five refined models were associated with specific substrates identified from the BioLip2 database, with cellobiose assigned to T1Cel5 and T4Cel5, and cellotriose associated with T2Cel5, T3Cel5 and T5Cel5.

The structural similarity among the thermophilic cellulase candidates was evaluated through RMSD values obtained from pairwise alignments of their three-dimensional structures. Typically, RMSD values below 3Å indicate strong structural similarity, allowing for reliable comparisons between proteins.⁵⁴ As shown in Table 4, the candidates exhibited RMSD values predominantly below this threshold, with T3Cel5 displaying the lowest RMSD of 0.57Å in comparison to its closest structural homolog, 3pzt. This structural proximity among the candidates potentially contributes to their classification as thermophilic proteins. In addition, the TM-scores further corroborate these findings, with values above 0.76, indicating a high degree of structural alignment with known thermophilic proteins in the PDB.

To assess the interatomic features of thermophilic and mesophilic enzymes, we conducted an analysis using the Arpeggio tool. Our findings showed a distinct pattern: thermophilic enzymes exhibited a significantly higher tendency towards hydrophobic contacts compared to mesophilic enzymes (Table 5). Specifically, the hydrophobic contacts per amino acid for thermophilic enzymes ranged from 3.05 to 3.37, while mesophilic enzymes exhibited lower values, indicating a robust hydrophobic character among the thermophilic candidates. This observation underscores potential structural adaptations that contribute to the stability of thermophilic enzymes in high-temperature environments.

Finally, we performed a data mining analysis by querying the molecular structure of a set of reference enzymes and our top five enzymes to evaluate the abundance of certain amino acids, in two topological scenarios of structure/function relationship, that influence on thermotolerance.

Specifically, we extracted from Figure 2A the residues overrepresented in the set of thermophilic enzymes over mesophilic enzymes and, based on what is described in the literature, we assigned the amino acids Ile, Phe, Tyr, Trp to the HYD group and the amino acids Lys, His, Glu to the CAT group.^55-57 Then, the residues of higher hydrophobic character of the HYD group were counted and relativized according to the ratio with the total amino acids of the enzymes (Fig. S3, C and D) or according to the ratio with the total HYD amino acids per enzyme. In this sense, we found a higher ratio of presence of HYD-type amino acids in the core residues of our five enzymes compared to a set of mesophilic enzymes. Similarly, for these novel enzymes we found a higher ratio of catalytically influenced residues in the catalytic pocket environment compared to the mesophilic enzyme set (Fig. S3, A and B).

The results of this study not only underscores the relevance of predictive methods in enzymology and metagenomics but also emphasizes the importance of integrating biotechnology into the sustainable production of bioproducts from biomass. In silico tools have proven to be highly promising for characterizing enzymes with industrial potential, facilitating their selection for appropriate applications.^58–61 However, the structural basis of thermophilicity, a trait paramount biotechnological importance, has so far remained elusive.

Thermophilic enzymes can accelerate transformation rates, enhance substrate hydrolysis, improve product solubility, and reduce microbial contamination.⁶² To evaluate their thermophilicity, we employed a variety of specialized prediction tools, acknowledging that no single biochemical, biophysical, or biological factor can fully explain account for the thermophilicity of enzymes or proteins in general.⁶³

Specifically, we employed the MCIC tool to identify cellulases based on their optimum temperature and pH⁶⁴ and integrated ThermoPred and iThermo into our analysis. Older models, such as ThermoPred, employed a support vector machine (SVM) classifier trained on a dataset that included both non-thermophilic and thermophilic proteins.³⁰ The iThermo model, a multi-layer perceptron (MLP), was developed using an expanded dataset featuring a greater diversity of thermophilic and non-thermophilic proteins.³¹ This model utilizes a multi-feature fusion approach to accurately identify proteins with thermophilic characteristics. In combination, these tools provided valuable insights into the thermodynamic stability of cellulases and facilitated the identification of thermophilic proteins, thereby enhancing the comprehensiveness of our study. In this analysis, we defined thermophiles as organisms with growth temperatures exceeding 60°C, while mesophiles were characterized by growth temperatures below this threshold. Our findings show that the selected GH5 enzymes, T1Cel5 through T5Cel5, possess characteristics indicative of thermophilicity, as predicted by at least two out of the three employed tools. The reliance on multiple prediction algorithms not only reinforces the robustness of our selections but also highlights the complexities inherent in accurately predicting protein thermophilicity based on sequence data alone. From a biotechnological point of view, the correlation between the predictions and the presence of signal peptides suggests that these enzymes are likely to be functionally relevant in thermophilic conditions. Additionally, the stringent selection criteria employed, including the e-value threshold, enhance the reliability of our candidate selection.

The five selected GH5 cellulases were comprehensively studied to compare their phylogenetic relationships, physicochemical characteristics, and interactomic structure. To further contextualize our findings, a phylogenetic study using accessions from the CAZy database and BLASTP comparisons showed that GH5 cellulases are closely related to specific bacterial groups, with a clear separation of subfamilies among the enzymes T1Cel5, T2Cel5, T3Cel5, T4Cel5, and T5Cel5. These enzymes cluster into distinct groups, indicating varying relationships with different bacterial taxa. Specifically, T1Cel5 groups with species of the genus Treponema, while T2Cel5 and T4Cel5 are associated with clades related to bacteria from the Terrabacteria group. In contrast, T3Cel5 is linked to bacteria from the phylum Spirochaetota, and T5Cel5 shows a close relationship with species from the family Bacillaceae. These findings broaden our understanding of GH5 enzyme distribution across bacteria and reinforce the idea of their diverse microbial origins. The phylogenetic relationships identified provide valuable insights for engineering enzymes tailored to specific biotechnological applications.

The physicochemical characterization and secondary structure predictions provided key insights into the properties of the GH5 enzymes from C. fulviceps and N. aquilinus gut metagenomes, contributing to a deeper understanding of their potential functionality and stability. The isoelectric point is the pH value at which the mobility of the protein becomes zero, resulting in a more compact and stable conformation.⁶⁰ For N. aquilinus, the pI values were found to be widely spread, with a higher abundance of enzymes having pI values less than 7. Similarly, the pI analysis for C. fulviceps GH5 indicated a predominance of acidic enzymes, suggesting they contain more positively charged residues.

The negative GRAVY values for all cellulase sequences indicate a hydrophilic nature, suggesting good solubility. Lower GRAVY values, which measure hydrophilicity, are associated with enhanced aqueous interactions.⁶⁵ This supports the hypothesis that these enzymes are well-suited for biotransformation processes in aqueous environments. The aliphatic index, a key indicator of protein thermostability, strongly suggests that these cellulases possess robust structural integrity across a wide temperature range. Defined as the relative volume occupied by aliphatic side chains, this index is considered a positive factor in enhancing the thermal stability of globular proteins.^66,67 The consistently high aliphatic index values in all candidate sequences further reinforce their potential stability under diverse temperature conditions. Additionally, the instability index, which measures the in vivo half-life of a protein consistently averaged around 33.29 ± 6.81 across the five GH5 enzymes presented here.⁶⁸ This value, being below the threshold of 40, indicates that the analyzed cellulases are likely to maintain stability, with a predicted half-life of over 16 hours.⁶⁹ These characteristics collectively highlight the suitability of these enzymes for industrial applications requiring stable and thermostable proteins. Additionally, this higher abundance of negatively charged residues has been noted to play significant roles in the three-dimensional structures and can significantly influence the orientation of proteins, especially in transmembrane proteins.⁷⁰ The secondary structure analysis indicates a well-balanced composition of α-helix, β-sheet, turn, and random coil across the enzymes. In N. aquilinus, α-helices dominate, suggesting greater stability, while in C. fulviceps, the prevalence of extended strands and random coils may indicate higher flexibility. These features likely contribute to the enzymes' stability and adaptability in varying conditions, supporting their potential for industrial applications.

Previous studies have highlighted the stabilizing role of an increased number of hydrophobic interactions in thermophilic enzymes. In agreement with these findings,^18,19,71 our comparative analysis of interatomic contacts reveals that thermophilic enzymes exhibit a higher frequency of hydrophobic interactions relative to their mesophilic homologs. This pattern was consistently observed across all five GH5 enzymes examined in this study.

To further explore structural determinants of thermotolerance, we analyzed the distribution of hydrophobic (HYD) and catalytically influential (CAT) amino acids in core and pocket regions. Our data mining revealed that the candidate enzymes (T1Cel5–T5Cel5) exhibit a significantly higher HYD-amino acids ratio in their hydrophobic cores (Fig. S3C-D) and an elevated CAT-amino acids ratio in catalytic pockets compared to mesophilic counterparts (Fig. S3A-B). These trends—consistent across all five candidates—support the hypothesis that increased hydrophobicity in structural cores and catalytic residue enrichment in active sites are hallmarks of thermo-adaptation.^55–57 Such features likely contribute to the compactness and rigidity observed in our structural models, reinforcing the thermophilic potential of these enzymes. The structural characteristics of thermophilic GH5 enzymes can be summarized as follows: they typically possess a compact and rigid architecture that allows them to withstand the extreme temperatures of their natural habitats. This robustness is achieved through various features, including reduced surface loops, increased hydrophobic interactions, and enhanced secondary structural elements, all of which minimize protein flexibility and contribute to thermostability.¹⁸ Additionally, these enzymes often exhibit a tightly packed hydrophobic core, which stabilizes the protein structure by limiting the exposure of hydrophobic residues to the solvent, thereby reducing the likelihood of denaturation at elevated temperatures. Furthermore, thermophilic GH5 enzymes may possess an increased number of α-helices, which are more stable secondary structural elements than β-sheets or random coils.¹⁸ This higher α-helix content further enhances the overall rigidity and stability of the protein structure enabling it to endure the extreme conditions found in hyper-thermophilic environments.⁶⁰ Additionally, alignment analysis revealed that the five selected protein GH5 share similar folding and secondary structural attributes and exhibit high homology with thermostable enzymes characterized both in silico and biochemically, as reported in the PDB (Table 4). This further reinforces the validity of this approach regarding their potential as thermophilic enzymes. The information produced in this study strongly suggest that the five GH5 enzymes analyzed are thermophilic candidate and encouraging moving to the next, more costly and labor-intensive stages of biochemical and biotechnological validation.

In this study, we identified new thermophilic lignocellulolytic enzymes. The application of machine learning facilitated the discovery of promising candidates, enhancing characterization while reducing costs. The predicted substrates suggest a potential role in biomass conversion. Furthermore, these enzymes exhibit high homology with hyper-thermophilic enzymes, highlighting their potential for applications in the biotechnology industry and contributing to a more sustainable economy.

Author Contributions

Pablo Daniel Farace and Rubén Marrero Diaz de Villegas: conceptualization, investigation, validation, visualization, formal analysis, data curation, methodology, writing original draft, writing and reviewing and editing the final version of the Manuscript. María Laura Mon: investigation, methodology, formal analysis, writing – review and editing. Marcelo A. Soria: investigation, validation, visualization, formal analysis, data curation, methodology, writing – review and editing. Paola M. Talia: conceptualization, supervision, methodology, writing – original draft, formal analysis, conceptualization funding acquisition, project administration.

Acknowledgments

MLM and PMT are members of the National Scientific and Technical Research Council of Argentina (CONICET). PDF acknowledges the CONICET fellowship. RMDV and PMT belong to the National Institute of Agricultural Technology of Argentina (INTA).

Conflict of Interest

The authors declare no conflicts of interest.

Funding

This study was supported by grants from the Instituto Nacional de Tecnología Agropecuaria (INTA) (PI 085, 089, 122 and 159), the Agencia Nacional de Promoción Científica y Tecnológica (ANPCyT) Proyectos de Investigación Científica y Tecnológica (PICT) 2018-#4149, 2020-#3570, Fundación Williams #014 and CONICET Project # PIP-2021-2561, Argentina.

Data Availability Statement

The data that supports the findings of this study are available from the corresponding author upon reasonable request.

Global Population Growth and Sustainable Development. 2021. https://www.un.org/development/desa/pd/sites/www.un.org.development.desa.pd/files/undesa_pd_2022_global_population_growth.pdf.
Kumar L, Chhogyel N, Gopalakrishnan T, Hasan MK, Jayasinghe SL, Kariyawasam CS, Kogo BK, Ratnayake S. Chapter 4 - Climate change and future of agri-food production. In: Bhat R, ed. Future Foods. Academic Press; 2022:49-79.
Garg R, Sabouni R, Ahmadipour M. From waste to fuel: Challenging aspects in sustainable biodiesel production from lignocellulosic biomass feedstocks and role of metal organic framework as innovative heterogeneous catalysts. Industrial Crops and Products. 2023;206:117554.
Ebaid R, Wang H, Sha C, Abomohra AE-F, Shao W. Recent trends in hyperthermophilic enzymes production and future perspectives for biofuel industry: A critical review. Journal of Cleaner Production. 2019;238:117925.
Romero Victorica M, Soria MA, Batista-García RA, Ceja-Navarro JA, Vikram S, Ortiz M, Ontañon O, Ghio S, Martínez-Ávila L, Quintero García OJ, Etcheverry C, Campos E, Cowan D, Arneodo J, Talia PM. Neotropical termite microbiomes as sources of novel plant cell wall degrading enzymes. Sci Rep. 2020;10(1):3864.
Kour R, Singh S, Sharma HB, Naik TSSK, Shehata N, N P, Ali W, Kapoor D, Dhanjal DS, Singh J, Khan AH, Khan NA, Yousefi M, Ramamurthy PC. Persistence and remote sensing of agri-food wastes in the environment: Current state and perspectives. Chemosphere. 2023;317:137822.
Sadh PK, Duhan S, Duhan JS. Agro-industrial wastes and their utilization using solid state fermentation: a review. Bioresources and Bioprocessing. 2018;5(1):1.
Prado-Acebo I, Cubero-Cardoso J, Lu-Chau TA, Eibes G. Integral multi-valorization of agro-industrial wastes: A review. Waste Management. 2024;183:42-52.
Batista-García RA, del Rayo Sánchez-Carbente M, Talia P, Jackson SA, O'Leary ND, Dobson ADW, Folch-Mallol JL. From lignocellulosic metagenomes to lignocellulolytic genes: trends, challenges and future prospects. Biofuels, Bioproducts and Biorefining. 2016;10(6):864-882.
Adegboye MF, Ojuederie OB, Talia PM, Babalola OO. Bioprospecting of microbial strains for biofuel production: metabolic engineering, applications, and challenges. Biotechnology for biofuels. 2021;14(1):5-5.
Ilić N, Milić M, Beluhan S, Dimitrijević-Branković S. Cellulases: From Lignocellulosic Biomass to Improved Production. Energies. 2023;16(8).
Mosier N, Wyman C, Dale B, Elander R, Lee YY, Holtzapple M, Ladisch M. Features of promising technologies for pretreatment of lignocellulosic biomass. Bioresource Technology. 2005;96(6):673-686.
Ashokkumar V, Venkatkarthick R, Jayashree S, Chuetor S, Dharmaraj S, Kumar G, Chen WH, Ngamcharussrivichai C. Recent advances in lignocellulosic biomass for biofuels and value-added bioproducts - A critical review. Bioresour Technol. 2022;344(Pt B):126195.
Yiin CL, Yap KL, Ku AZE, Chin BLF, Lock SSM, Cheah KW, Loy ACM, Chan YH. Recent advances in green solvents for lignocellulosic biomass pretreatment: Potential of choline chloride (ChCl) based solvents. Bioresource Technology. 2021;333:125195.
Gomes E, de Souza AR, Orjuela GL, Da Silva R, de Oliveira TB, Rodrigues A. Applications and Benefits of Thermophilic Microorganisms and Their Enzymes for Industrial Biotechnology. In: Schmoll M, Dattenböck C, eds. Gene Expression Systems in Fungi: Advancements and Applications. Cham: Springer International Publishing; 2016:459-492.
Turner P, Mamo G, Karlsson EN. Potential and utilization of thermophiles and thermostable enzymes in biorefining. Microb Cell Fact. 2007;6:9.
Singhal G, Bhagyawant SS, Srivastava N. Chapter 3 - Cellulases through thermophilic microorganisms: Production, characterization, and applications. In: Tuli DK, Kuila A, eds. Current Status and Future Scope of Microbial Cellulases. Elsevier; 2021:39-57.
Nezhad NG, Rahman RNZRA, Normi YM, Oslan SN, Shariff FM, Leow TC. Thermostability engineering of industrial enzymes through structure modification. Applied Microbiology and Biotechnology. 2022;106(13):4845-4866.
Pucci F, Rooman M. Physical and molecular bases of protein thermal stability and cold adaptation. Curr Opin Struct Biol. 2017;42:117-128.
Santos CR, Paiva JH, Sforça ML, Neves JL, Navarro RZ, Cota J, Akao PK, Hoffmam ZB, Meza AN, Smetana JH, Nogueira ML, Polikarpov I, Xavier-Neto J, Squina FM, Ward RJ, Ruller R, Zeri AC, Murakami MT. Dissecting structure-function-stability relationships of a thermostable GH5-CBM3 cellulase from Bacillus subtilis 168. Biochem J. 2012;441(1):95-104.
Pereira JH, Chen Z, McAndrew RP, Sapra R, Chhabra SR, Sale KL, Simmons BA, Adams PD. Biochemical characterization and crystal structure of endoglucanase Cel5A from the hyperthermophilic Thermotoga maritima. J Struct Biol. 2010;172(3):372-379.
Domínguez R, Souchon H, Lascombe M, Alzari PM. The crystal structure of a family 5 endoglucanase mutant in complexed and uncomplexed forms reveals an induced fit activation mechanism. J Mol Biol. 1996;257(5):1042-1051.
Kim HW, Ishikawa K. Functional analysis of hyperthermophilic endocellulase from Pyrococcus horikoshii by crystallographic snapshots. Biochem J. 2011;437(2):223-230.
Handelsman J. Metagenomics: application of genomics to uncultured microorganisms. Microbiol Mol Biol Rev. 2004;68(4):669-685.
Ben Guerrero E, Marrero de Villegas R, Soria M, Santangelo Mde L, Campos E, Talia P. Characterization of two GH5 endoglucanases from termite microbiome using synthetic metagenomics. Applied Microbiology and Biotechnology. 2020;104(19):8351-8366.
Sakon J, Adney WS, Himmel ME, Thomas SR, Karplus PA. Crystal Structure of Thermostable Family 5 Endocellulase E1 from Acidothermus cellulolyticus in Complex with Cellotetraose. Biochemistry. 1996;35(33):10648-10660.
Seemann T. Prokka: rapid prokaryotic genome annotation. Bioinformatics. 2014;30(14):2068-2069.
Zheng J, Ge Q, Yan Y, Zhang X, Huang L, Yin Y. dbCAN3: automated carbohydrate-active enzyme and substrate annotation. Nucleic Acids Research. 2023;51(W1):W115-W121.
Foroozandeh Shahraki M, Farhadyar K, Kavousi K, Azarabad MH, Boroomand A, Ariaeenejad S, Hosseini Salekdeh G. A generalized machine-learning aided method for targeted identification of industrial enzymes from metagenome: A xylanase temperature dependence case study. Biotechnology and Bioengineering. 2021;118(2):759-769.
Lin H, Chen W. Prediction of thermophilic proteins using feature selection technique. Journal of Microbiological Methods. 2011;84(1):67-70.
Ahmed Z, Zulfiqar H, Khan AA, Gul I, Dao F-Y, Zhang Z-Y, Yu X-L, Tang L. iThermo: A Sequence-Based Model for Identifying Thermophilic Proteins Using a Multi-Feature Fusion Strategy. Frontiers in Microbiology. 2022;Volume 13 - 2022.
Camacho C, Coulouris G, Avagyan V, Ma N, Papadopoulos J, Bealer K, Madden TL. BLAST+: architecture and applications. BMC Bioinformatics. 2009;10:421.
Menzel P, Ng KL, Krogh A. Fast and sensitive taxonomic classification for metagenomics with Kaiju. Nat Commun. 2016;7:11257.
Edgar RC. Muscle5: High-accuracy alignment ensembles enable unbiased assessments of sequence homology and phylogeny. Nat Commun. 2022;13(1):6968.
Price MN, Dehal PS, Arkin AP. FastTree 2--approximately maximum-likelihood trees for large alignments. PLoS One. 2010;5(3):e9490.
Letunic I, Bork P. Interactive Tree Of Life (iTOL) v5: an online tool for phylogenetic tree display and annotation. Nucleic Acids Research. 2021;49(W1):W293-W296.
Gasteiger E, Gattiker A, Hoogland C, Ivanyi I, Appel RD, Bairoch A. ExPASy: The proteomics server for in-depth protein knowledge and analysis. Nucleic Acids Res. 2003;31(13):3784-3788.
Madeira F, Madhusoodanan N, Lee J, Eusebi A, Niewielska A, Tivey ARN, Lopez R, Butcher S. The EMBL-EBI Job Dispatcher sequence analysis tools framework in 2024. Nucleic Acids Res. 2024;52(W1):W521-w525.
Kozlowski LP. IPC 2.0: prediction of isoelectric point and pKa dissociation constants. Nucleic Acids Res. 2021;49(W1):W285-w292.
Garnier J, Osguthorpe DJ, Robson B. Analysis of the accuracy and implications of simple methods for predicting the secondary structure of globular proteins. J Mol Biol. 1978;120(1):97-120.
Teufel F, Almagro Armenteros JJ, Johansen AR, Gíslason MH, Pihl SI, Tsirigos KD, Winther O, Brunak S, von Heijne G, Nielsen H. SignalP 6.0 predicts all five types of signal peptides using protein language models. Nat Biotechnol. 2022;40(7):1023-1025.
Yang J, Zhang Y. I-TASSER server: new development for protein structure and function predictions. Nucleic Acids Res. 2015;43(W1):W174-181.
Mirdita M, Schütze K, Moriwaki Y, Heo L, Ovchinnikov S, Steinegger M. ColabFold: making protein folding accessible to all. Nat Methods. 2022;19(6):679-682.
Delgado J, Radusky LG, Cianferoni D, Serrano L. FoldX 5.0: working with RNA, small molecules and a new graphical interface. Bioinformatics. 2019;35(20):4168-4169.
Schymkowitz J, Borg J, Stricher F, Nys R, Rousseau F, Serrano L. The FoldX web server: an online force field. Nucleic Acids Res. 2005;33(Web Server issue):W382-388.
Benkert P, Biasini M, Schwede T. Toward the estimation of the absolute quality of individual protein structure models. Bioinformatics. 2010;27(3):343-350.
Chen VB, Arendall WB, 3rd, Headd JJ, Keedy DA, Immormino RM, Kapral GJ, Murray LW, Richardson JS, Richardson DC. MolProbity: all-atom structure validation for macromolecular crystallography. Acta Crystallogr D Biol Crystallogr. 2010;66(Pt 1):12-21.
Pettersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt DM, Meng EC, Ferrin TE. UCSF Chimera—A visualization system for exploratory research and analysis. Journal of Computational Chemistry. 2004;25(13):1605-1612.
Sargsyan K, Grauffel C, Lim C. How Molecular Size Impacts RMSD Applications in Molecular Dynamics Simulations. Journal of Chemical Theory and Computation. 2017;13(4):1518-1524.
Jubb HC, Higueruelo AP, Ochoa-Montaño B, Pitt WR, Ascher DB, Blundell TL. Arpeggio: A Web Server for Calculating and Visualising Interatomic Interactions in Protein Structures. J Mol Biol. 2017;429(3):365-371.
Yang J, Roy A, Zhang Y. Protein–ligand binding site recognition using complementary binding-specific substructure comparison and sequence profile alignment. Bioinformatics. 2013;29(20):2588-2595.
Zhang J, Liu Y, Hu J, Leng G, Liu X, Cui Z, Wang W, Ma Y, Sha S. Cellulase Promotes Mycobacterial Biofilm Dispersal in Response to a Decrease in the Bacterial Metabolite Gamma-Aminobutyric Acid. Int J Mol Sci. 2024;25(2).
Salgado JFM, Hervé V, Vera MAG, Tokuda G, Brune A. Unveiling lignocellulolytic potential: a genomic exploration of bacterial lineages within the termite gut. Microbiome. 2024;12(1):201.
Carugo O. How root-mean-square distance (r.m.s.d.) values depend on the resolution of protein structures that are compared. Journal of Applied Crystallography. 2003;36(1):125-128.
Nezhad NG, Rahman R, Normi YM, Oslan SN, Shariff FM, Leow TC. Recent advances in simultaneous thermostability-activity improvement of industrial enzymes through structure modification. Int J Biol Macromol. 2023;232:123440.
Torktaz I, Karkhane AA, Hemmat J. Rational engineering of Cel5E from Clostridium thermocellum to improve its thermal stability and catalytic activity. Appl Microbiol Biotechnol. 2018;102(19):8389-8402.
Xu Z, Cen YK, Zou SP, Xue YP, Zheng YG. Recent advances in the improvement of enzyme thermostability by structure modification. Crit Rev Biotechnol. 2020;40(1):83-98.
Lugani Y, Sooch B. In silico Characterization of Cellulases from Genus Bacillus. 2017.
Dodda SR, Hossain M, Kapoor BS, Dasgupta S, B VPR, Aikat K, Mukhopadhyay SS. Computational approach for identification, characterization, three-dimensional structure modelling and machine learning-based thermostability prediction of xylanases from the genome of Aspergillus fumigatus. Comput Biol Chem. 2021;91:107451.
Sanjaya RE, Putri KDA, Kurniati A, Rohman A, Puspaningsih NNT. In silico characterization of the GH5-cellulase family from uncultured microorganisms: physicochemical and structural studies. J Genet Eng Biotechnol. 2021;19(1):143.
Bakli M, Bouras N, Pașcalău R, Smuleac L. In silico Structural and Functional Characterization of an Endoglucanase from Actinoalloteichus hoggarensis. Advanced Research in Life Sciences. 2023;7:135-141.
Vieille C, Zeikus GJ. Hyperthermophilic enzymes: sources, uses, and molecular mechanisms for thermostability. Microbiol Mol Biol Rev. 2001;65(1):1-43.
Hildén K, Hakala TK, Lundell T. Thermotolerant and thermostable laccases. Biotechnol Lett. 2009;31(8):1117-1128.
Foroozandeh Shahraki M, Ariaeenejad S, Fallah Atanaki F, Zolfaghari B, Koshiba T, Kavousi K, Salekdeh GH. MCIC: Automated Identification of Cellulases From Metagenomic Data and Characterization Based on Temperature and pH Dependence. Frontiers in Microbiology. 2020;11.
Kyte J, Doolittle RF. A simple method for displaying the hydropathic character of a protein. Journal of Molecular Biology. 1982;157(1):105-132.
Ikai A. Thermostability and aliphatic index of globular proteins. J Biochem. 1980;88(6):1895-1898.
Delavari A, Zare S, Ghaemi MR, Kashfi R, Ebrahimi M, Tahmasebi A, Ebrahimi M, Ebrahimie E. Determining the structural amino acid attributes which are important in both protein thermostability and alkalophilicity: a case study on xylanase. BioTechnologia. 2014;95(2):161-173.
Guruprasad K, Reddy BV, Pandit MW. Correlation between stability of a protein and its dipeptide composition: a novel approach for predicting in vivo stability of a protein from its primary sequence. Protein Eng. 1990;4(2):155-161.
Rogers S, Wells R, Rechsteiner M. Amino acid sequences common to rapidly degraded proteins: the PEST hypothesis. Science. 1986;234(4774):364-368.
Rutz C, Rosenthal W, Schülein R. A Single Negatively Charged Residue Affects the Orientation of a Membrane Protein in the Inner Membrane of Escherichia coliOnly When It Is Located Adjacent to a Transmembrane Domain*. Journal of Biological Chemistry. 1999;274(47):33757-33763.
Camilloni C, Bonetti D, Morrone A, Giri R, Dobson CM, Brunori M, Gianni S, Vendruscolo M. Towards a structural biology of the hydrophobic effect in protein folding. Sci Rep. 2016;6:28285.

Table 1. Thermostability prediction of GH5 cellulases candidates.

ID	Locus tag	e_value (dbCAN)	Subfamily	Amino acid	MCIC	iThermo	ThermPred
T1Cel5	KBCPBGKF_14422	6.10E-42	GH5_40	331	hyper-thermophilic	thermophilic (score= 0.9620)	thermophilic (score= 0.9950)
T2Cel5	KBCPBGKF_22703	8.10E-45	GH5_39	554	hyper-thermophilic	thermophilic (score= 0.8441)	thermophilic (score= 0.7632)
T3Cel5	AFHCADON_14335	1.90E-43	GH5_2	320	hyper-thermophilic	mesophilic (score= 0.3729)	thermophilic (score= 0.7231)
T4Cel5	KBCPBGKF_00253	1.30E-47	GH5_39	554	hyper-thermophilic	mesophilic (score= 0.4295)	thermophilic (score= 0.5845)
T5Cel5	KBCPBGKF_33952	8.40E-38	GH5_25	541	mesophilic	thermophilic (score= 0.5433)	thermophilic (score= 0.5215)

Table 2. Physicochemical properties of the thermophilic cellulases candidates

ID	IPC 2.0		ProtParam					Garnier (EMBOSS)
	Theoretical pI	Weight (kDa)	GRAVY	Aliphatic index	Inestability index	R+*	R- **	α-helix (%)	β-sheet (%)	Turn (%)	Random-coil (%)
T1Cel5	5.14	41.97	-0.403	84.34	29.51	47	32	22.8	28.7	29.1	25.1
T2Cel5	5.69	66.6	-0.428	82.20	34.02	68	61	32.3	22.7	26.8	21.2
T3Cel5	7.02	38.89	-0.404	84.42	44.15	40	40	23.3	28.2	24.6	29.2
T4Cel5	5.58	65.96	-0.456	74.40	32.73	65	75	26.8	25.8	27.3	23.0
T5Cel5	5.08	63.14	-0.342	81.29	26.03	57	72	38.1	22.1	19.4	23.4

* Positive residues.

** Negative residues.

Table 3. Summary of metrics data of selected protein structural models

	I-TASSER			Alphafold2
ID	C-score	TM-score	Substrate (BioLip2 database)	pLDDT	TM-score
T1Cel5	0.00	0.71±0.1	cellobiose	94.8	0.92
T2Cel5	-1.73	0.50±0.1	cellotriose	93.7	0.93
T3Cel5	0.25	0.75±0.1	cellotriose	97.2	0.95
T4Cel5	-1.77	0.50±0.15	cellobiose	93.8	0.90
T5Cel5	-1.58	0.52±0.1	cellotriose	91.1	0.80

Table 4. Thermophilic proteins are structurally close to the target in the PDB.

ID	PDB Hit	TM-score	RMSD
T1Cel5	4hty	0.85	1.03
	2zun	0.76	3.19
	5fip	0.74	2.92
	5ecu	0.74	2.97
T2Cel5	4yzp	0.84	1.87
	5e0c	0.79	3.02
	4v2x	0.78	3.16
T3Cel5	5ihs	0.94	0.91
	5ecu	0.92	1.78
	5fip	0.92	1.52
	3pzt	0.91	0.57
T4Cel5	4yzp	0.82	2.27
	5e0c	0.80	2.97
	4v2x	0.78	3.11
T5Cel5	4yzp	0.88	1.52
	5e0c	0.83	2.34
	4v2x	0.83	2.48

Table 5. Characterization of interactomic contacts in the structural modeling of candidate GH5 cellulases. Contacts were normalized by calculating contacts per amino acid.

ID		Amino acid	Hidrofobic contacts	Hidrofobic contacts /aa*	Polar contacts	Polar contacts/aa*	Hydrogen bond	Hydrogen bonds / aa*
T1Cel		331	1010	3.05	568	1.72	394	1.19
T2Cel		554	1818	3.28	901	1.63	601	1.08
T3Cel		320	1012	3.16	560	1.75	374	1.17
T4Cel5		554	1867	3.37	884	1.60	569	1.03
T5Cel5		541	1683	3.11	769	1.42	513	0.95
Termophilic	1cen	293	1007	3.44	575	1.96	389	1.33
	2zun	379	1288	3.4	633	1.67	407	1.07
	3nco	310	1073	3.46	554	1.79	355	1.15
	3mmu	309	1066	3.45	556	1.8	360	1.17
	3pzt	293	859	2.93	539	1.84	350	1.19
	5ecu	350	1044	2.98	590	1.69	369	1.05
	5fip	331	971	2.93	554	1.67	357	1.08
Mesophilic	1egz	291	813	2.79	541	1.86	348	1.2
	1edg	380	1098	2.89	711	1.87	458	1.21
	1g01	357	1020	2.86	634	1.78	391	1.1
	3l55	345	963	2.79	676	1.96	427	1.24

The authors declare no competing interests.

supplementary.docx

Discovery of Thermophilic Enzyme Properties Through Structural Modeling, Phylogenetic Analysis, and Machine Learning Prediction

Status:

Version 1

Abstract

Figures

Introduction

Materials and Methods

Selection of GH5-encoding Genes from a Termite Gut Microbiome

Thermophilicity Prediction of Cellulases

Sequence Alignment and Phylogenetic Analysis

Physiochemical Properties and Amino Acid Usage of the GH5 Enzymes

Molecular Structure Prediction and Ligand Binding Site Prediction

Protein Sequence Accession Numbers

Results

Discussion

Conclusion

Declarations

References

Tables

Additional Declarations

Supplementary Files

Status:

Version 1