Expanding the viewpoint: leveraging sequence information in enzymology

Abstract

The use of protein sequence to inform enzymology in terms of structure, mechanism, and function has burgeoned over the past two decades. Referred to as genomic enzymology, the utilization of bioinformatic tools such as sequence similarity networks and phylogenetic analyses has allowed the identification of new substrates and metabolites, novel pathways, and unexpected reaction mechanisms. The holistic examination of superfamilies can yield insight into the origins and paths of evolution of enzymes and the range of their substrates and mechanisms. Herein, we highlight advances in the use of genomic enzymology to address problems which the in-depth analyses of a single enzyme alone could not enable.

Graphical Abstract

Introduction

Over the past decade, the definitive study of individual enzymes in terms of specificity and catalytic and chemical mechanism has been leveraged using bioinformatic tools to extrapolate that knowledge to numerous homologues. Subsequently, deviations from the original classification are employed to define those enzymes with different structure-function and mechanism for further study. These innovations stemmed from the initial insight in 1996, when Babbitt et al. demonstrated how identifying common ancestry of mechanistically diverse superfamilies can allow for function and sometimes, specificity predictions [1]. The study focused on annotation of members of the enolase superfamily. Extensive mechanistic characterization and key structural knowledge of only a few family members allowed the authors to make the structure-function linkage as related to mechanistically diverse enzymes within the superfamily [1, 2]. The biochemical insight gained through this study gave rise to the concept of genomic enzymology, wherein the sequence-function space of a protein family can be analyzed within its genomic context to allow for the prediction of enzymatic activity and metabolic pathways [3]. Genomic enzymology also can be greatly enhanced by the inclusion of structural information, though the sequence-structure-function linkage is still being explored. The implementation of genomic enzymology, supported by biochemical and biophysical experimentation, has enabled the identification of novel pathways and functions (Figure 1).

Figure 1.

Overview of applications of genomic enzymology in recent years: determining pathways (PDB ID: 6P78), defining a superfamily or family (SSN generated from sequences from Knox et al. and visualized in Cytoscape [4]), exploring substrate specificity (PDB ID: 5V1V), understanding enzyme evolution, and predicting mechanism (mechanism adapted from Braffman et al. [5])

Bioinformatic tools have been developed that facilitate the sequence- and genomic- based study of enzymes [6–12]. A key development has been the ability to visualize the sequence-function space in sequence similarity networks (SSNs) paired with genome neighborhood networks and/or diagrams (GNNs or GNDs) [3, 6, 8, 11]. The SSN shows how homologous sequences relate through grouping into clusters defined by pairwise sequence alignments, whereas a GNN/GND shows the occurrence of genes found in common in the operons encoding homologues of the gene of interest [6, 8, 9]. The colocalization of genes in a GNN/GND is often taken as evidence of participation of the genes in a common pathway (guilt by association) and/or possible protein-protein interactions [13]. In mechanistically diverse superfamilies with overall low sequence identity, the evolutionary drift and resulting noise make it difficult to differentiate signatures for distinct families and subfamilies [14]. To test and quantify relationships within an SSN, Hidden Markov Models (HMMs) can be useful in probing whether subgroupings of enzymes are likely to be isofunctional [15]. HMMs are statistical models that can be used to identify hidden patterns or relationships within protein sequences. Likewise, the convergence ratios (CRs) when used within a superfamily can yield information about the similarity between sequences and aid in distinguishing paralogues from orthologues [16]. CRs measure sequence similarity at a specific alignment score by calculating the ratio of the number of edges of a cluster to the maximum possible number of edges [16]. Taxonomic information can be integrated into SSNs by attributing known properties to the nodes. This information together with phylogenetic analyses, can be powerful tools in making conclusions about relatedness and ancestral origins. In this review, we highlight examples of the conjoined use of primary and tertiary structure information together with genome context, which through these and other tools, reveal mechanisms of substrate specificity, find new chemistries and new pathways, and reveal evolutionary paths.

Benefits of superfamily analysis

As of May 2022, the UniProtKB/Swiss-Prot database contains more than 500,000 unique and annotated sequences and is still expanding [17, 18]. With this wealth of information, comparative analyses of superfamilies can be utilized to aid in the annotation of uncharacterized enzymes as well as determine mechanisms, assign functions, and correct misannotations. A recent application of this “comparative anatomy” approach is the work by Tararina and Allen, which deconvoluted the flavin-dependent amine oxidase (FAO) superfamily, organizing it into eight subgroups based on substrate type (Figure 2) [19]. A notable feature of the FAO superfamily is the flexibility in the multi-domain scaffold, which allows for facile evolution of new reactivity through insertions or deletions to the substrate-binding domains with retention of the critical flavin-binding domain, creating new functions amongst the subfamilies [19]. The evolvability of the superfamily, through use of modular binding domains, was further revealed by the reinvention of functions amongst the eight subfamilies [19].

Figure 2.

Abbreviated SSN of the FAO superfamily adapted from Tararina and Allen [16] and visualized in Cytoscape. The eight enzyme functional subgroups are labeled with representative structures for illustrative purposes. PDB IDs are 2IID (L-amino acid oxidase), 3NKS (protoporphyrinogen IX oxidase), 5KRQ (renalase), 6C71 (nicotine oxidoreductase), 1OJ9 (monoamine oxidase), 2UXN (lysine-specific demethylase), 5MOG (phytoene desaturase), and 5MBX (N1-acetylpolyamine oxidase).

Combining structural information with bioinformatic analysis can aid greatly in surveying a family or superfamily [20]. In Knox et al., the authors structurally characterized an enzyme within the cobalamin-dependent radical S-adenosylmethionine (SAM) methylase class, the largest and most diverse radical SAM methylase class [4]. Radical SAM methylases are able to append methyl groups to inert sp²- and sp³-hybridized carbonas well as phosphinate phosphorus. The low sequence conservation and mechanistic divergence amongst these enzymes make their study challenging. The authors discuss the diversity of this class of enzymes through structural comparisons of three different cobalamin-dependent radical SAM enzymes [4]. Although the three enzymes have the same three structural domains (a Rossmann fold that binds cobalamin, a shortened TIM barrel that ligates the Fe/S cluster, and a structurally variable C-terminal domain), the mechanistic determinants are dictated largely by the cofactor interactions and the secondary structural information [4]_. In a second example, Sun et al. enlisted a SSN and phylogenetic tree to draw links between evolutionary and functional diversification of a phosphodiesterase subfamily, allowing for prediction of the enzyme metallocofactor and reactivity [21]. The study focused on metal-dependent phosphodiesterases involved in regulating cyclic dinucleotides [21]. Within this study, Sun et al. were able to functionally categorize the metal-dependent phosphodiesterases and surmise evolutionary paths by incorporating information on metallocofactors and coordination spheres, secondary structural differences, and specific activities [21]. Generally, the identification of sequence patterns and divergences from those patterns have been very successful in flagging possible new substrates and reactivities [22–26]. The survey of a superfamily, labeled with known reactivities and reaction types, can guide selection of new targets to cover unexplored areas or find additional examples of known reactions, increasing project success.

Sequence analysis is a boon to target selection [27–30]. For instance, identifying homologs can support the study of intractable enzymes. In 2019, Guo et al. completed the experimental mechanistic and structural characterization of an imine reductase (Bsp5) which, unlike its homologues (Ind5 and Pel5), does not require an additional enzyme (Ind4 and Pel4 respectively) for stability and activity [31]. Bsp5 is capable of stand-alone activity and crystallized readily. Moreover, its simplification allowed for a better understanding of the structural constraints for activity [31]. The selection of this more divergent member (Bsp5 has 50% sequence identity to Pel5) also allowed the authors to observe evidence that Bsp5 can accept alternative substrates, which suggests the expansion of the scope of its ability to synthesize D-amino acids. Describing the scope of a superfamily through sequence analysis can also aid in target selection for applications in structural biology [32–34]. Where X-ray crystallographic analysis may be facilitated by using the minimal core catalytic fold, cryo-electron microscopy necessitates a larger macromolecule. The divergence of the folds within a superfamily can be defined through SSNs to identify the minimal, common catalytic fold versus the structurally elaborated folds [35].

In 2021, Taujale et al. designed a convolution neural network with an attention-based deep learning model to predict folds within the glycosyltransferase superfamily by combining secondary structure representations with primary sequence analysis [36]. The authors further delineated the individual folds into distinct clusters using a probabilistic mathematical model to segregate the secondary structures of evolutionarily divergent families [36]. Their approach worked best for the GT-A fold type glycosyltransferases but is also able to predict several unknown fold types [36]. The same approach worked well for the kinase superfamily, leading the authors to suggest that this method is adaptable for exploring fold diversity within large protein superfamilies [36]. The ability to accurately predict structures with no homologous protein fold is a powerful innovation, showing the strength of using deep learning to complement experimental biochemical approaches [37]. As such, machine learning can greatly enhance the studies in which structural information is lacking.

Defining structure-function links within a superfamily

A recent insight is that SSNs can inform not only the relationship, but the evolutionary path between mechanistically divergent enzymes. Bioinformatic studies of the tautomerase superfamily identify linker proteins- enzymes connecting two different clusters within the SSN (see example in Figure 3)- shedding light on the structural changes in the active sites yielding diversification of activity [38, 39]. Structural and kinetic characterization of two linker proteins revealed amino-acid substitutions that allow the switch in mechanism between tautomerase and dehalogenase activities [39]. The identification of linker proteins can uncover sequence variation driving diversity and suggest the evolutionary trajectory within a superfamily, although evolutionary relationships require further support from phylogenetic tree reconstruction [15, 40].

Figure 3.

An example of linker proteins found within a SSN, shown as purple boxes. The enzymes act as a “link” between the two clusters, illustrated in blue and red. SSN generated from sequences in O’Toole et al. [41] and visualized in Cytoscape.

Phylogenetic evaluation can be used to assess evolutionary origins, as highlighted in the recent study of DNA ligases by Pan et al. [42]. Using genomic information gathered in a bioinformatics survey combined with the mechanistic and crystallographic data, the authors identify a probable entry point for horizontal acquisition of ATP-dependent DNA ligases by bacteria [42]. In 2021, O’Toole et al. performed bioinformatic analyses of the monotopic phosphoglycosyl transferase (monoPGT) superfamily, identifying numerous fusion proteins within the SSN [41]. The fusions were grouped into sugar-modifying, glycosyltransferase, and regulatory domains. Intriguingly, a number of fusions with an isofunctional, but distinct superfamily, the polytopic PGTs (polyPGT), were also identified, suggesting an evolutionary link through the shared substrate [41]. Phylogenetic reconstruction revealed a “radial burst of functionalization, with a minority of members comprising only the minimal phosphoglycosyl transferase catalytic domain” [41].

Discovery of new pathways

Genomic enzymology has enabled the identification of novel or unique metabolites and pathways [43–47]. Macdonald et al. surveyed the glycoside phosphorylase family, specifically CAZy family GH94 [48]. The Carbohydrate-Active enZYmes (CAZy) database includes a number of sequences of glycoside phosphorylases, but the classification and characterization of this class are limited [49]_. The phylogenetic exploration by the authors allowed for the identification and then characterization of a previously unclassified glycoside phosphorylase and its substrate, a new biopolymer referred to as acholetin [48]. In Yuan et al., the authors utilized a comparative genomic approach to identify two queuosine salvage pathways in pathogenic, commensal bacteria [50]. Many bacteria are capable of de novo queuosine biosynthesis, but the identification of the salvage pathway in bacteria suggests that there may be direct competition for the queuine-base precursor [50]. The study also uncovered the first example of a radical SAM enzyme catalyzing this queuine-lyase reaction mechanism, which itself had never been previously described in the literature [50].

The catabolic pathway of 1-deoxy-D-sorbitol has remained unknown, but Li et al. in 2022 were able to identify an operon that contained genes involved in catabolism of D-sorbitol and 1-deoxy-D-sorbitol [51]. A phylogenetic distribution analyses suggested the conservation of the gene cluster in the Firmicutes phylum (recently suggested to be renamed Bacillota) [51]. The authors propose that their integrated strategy will be applicable to the discovery of other novel metabolites [51]. Genomic enzymology can also aid in organizing complex families, enabling the dissection of sequence markers for each subfamily or chemical activity. In 2021, Beal et al. performed bioinformatic analysis of aminotransferases, dehydrogenases, and aminopolyol dehydrogenases (a three gene cluster of these is typically indicative of bacterial azasugar biosynthesis), which allowed for determination of a consensus sequence for aminotransferases involved in azasugar biosynthesis. This, in turn, identified more than 400 microorganisms that putatively produce azasugar [52]. The utilization of a pan-taxonomic survey allowed for a broader perspective of the genomic context of azasugar biosynthesis [52].

Often, examining the crystal structures within a bioinformatic study reveals information about the substrate specificity or mechanism [53–55]. Li et al. identified a unique ArsR-SmtB transcriptional regulator that was proposed to use disulfide-bond formation to upregulate the expression of thioredoxin as a way to maintain cellular redox homeostasis [56]. From the structural work, the authors were able to identify an entrance channel and key catalytic residues involved in peroxide activation [56]. Sequence analysis found a conserved Cys residue signature which allowed tracking of this divergent function through the superfamily. Their findings suggest that the vicinal disulfide redox switch is a feature of cyanobacteria within the Nostocales order, identifying a distinct cluster within the ArsR-SmtB regulator family [56]. This provides an excellent example of how evolution of novel pathways often depends on the recruitment and tailoring of existing enzymes to adopt a new substrate [57].

Exploring substrate specificity

Bioinformatic study can help inform the factors determining substrate specificity [58–60]. In the study of ribosomally-synthesized and post-translationally modified peptides (RiPPs), the recent focus is on identifying features that control substrate recognition. RiPPs comprise a large number of natural products, many with favorable pharmacokinetic properties, but the specificity of the enzymes that biosynthesize these peptides is still unclear. Generally, enzymes in RiPP biosynthetic pathways are believed to utilize sequence-specific recognition of the leader peptide sequence, which is removed after post-translation modifications on the core peptide (Figure 4). Typically, there is some conservation of the leader peptide sequence within a RiPP class, which suggests strict leader-sequence specificity [61]. However, enzymatic study guided by bioinformatics has found flexibility in the recognition of the leader peptide [61, 62]. Chekan et al. found in their enzymatic study of the lasso peptide class using alternative leader peptides, that the minimal binding pocket allows for recognition and binding of sequence-divergent leader peptides. The authors conclude that the binding of the leader peptide is controlled by hydrophobic packing rather than a specific amino-acid sequence [61]. Moreover, by curating the extensive SSN for the occurrence of the signature sequence, the authors conclude that it is likely that in many other RiPP classes leader peptide recognition does not dictate activity [61]. Likewise, in a recent study of class II lanthipeptide synthetases, Zhang et al. found that the enzyme’s recognition of the leader peptide is promiscuous whereas the enzyme catalyzes the same reaction on the core peptide, despite significant diversity of the leader peptide [62]. From detailed biochemical analysis and extensive sequence alignment, the authors suggest that the precursor leader peptides coevolve with the cognate enzyme [62]. There is much interest in bioengineering RiPPs and one such point of interest is in the usage of chimeric leader peptides, which can be recognized by enzymes in unrelated RiPP pathways, to allow for biosynthesis of novel RiPPs [63].

Figure 4.

Schematic example of RiPP processing. Amino acids and modified residues are represented by colored spheres.

In 2021, Lachowicz et al. explored the range of substrates accepted by viperins [64]. The radical SAM enzyme viperin (virus inhibitory protein, endoplasmic reticulum associated, interferon inducible) is able to restrict replication of several different RNA and DNA viruses, making it of interest for use in combatting pathogens including HIV, West Nile virus, and Dengue virus [64]. Human viperin catalyzes biosynthesis of 3’-deoxy-3’,4’-didehydro-CTP (ddhCTP) from CTP (making viperin a ddh-synthase), which functions as a chain terminator for RNA polymerases from a variety of flaviviruses [64]. From the structural and kinetic study of viperin homologues, Lachowicz et al. were able to decipher the catalytic mechanism of viperin as well as define the physical and chemical determinants of ddh-synthase activity and substrate selectivity [64]. The SSN revealed discrete clusters delineated by substrate selectivity and this information was used to identify residues that control the specificity [64].

Discovery of novel reaction mechanisms

Enzyme genomic context can aid in identifying unique reactions, mechanisms, or enzymatic properties [65–68]. In 2021, Braffman et al. characterized a cyanobacterial enzyme that performs two unusual alkylation reactions between aromatic rings and secondary alkyl halide substrates in the biosynthesis of cylindrocyclophane natural products [5]. Their bioinformatic study of homologues established that the key catalytic residues likely associate with divergent reactivity and altered secondary metabolism [5]. Phylogenetic study indicated that this specific family of enzymes likely catalyzes other reactions that do not involve alkyl halides or resorcinols [5]. Uncovering the molecular mechanism of this unprecedented biosynthetic reaction allows for a better understanding of the activation of alkyl halides. Similarly, Chekan et al. were able to utilize bioinformatics to aid in identifying key catalytic residues that dictate divergent functionalities in their study of enzymatic dehalogenation [69]. From their crystallographic and computational study of Bmp8, an enzyme that replaces a halogen with a hydrogen in the biosynthesis of the natural product and antibiotic pentabromopseudilin, the authors propose a mechanism similar to that of human deiodinases [69]. To further demonstrate the debromination of Bmp8, the authors characterized a homologue with the same activity found in a distinct genomic context [69]. From this observation, the authors conclude that there are conserved enzymes with the capacity to dehalogenate tetrabromopyrroles outside of the typical biosynthetic pathway [69].

The usage of comparative genomic approaches is a powerful way to discover novel enzymatic reaction mechanisms. In 2019, Rizzolo et al. identified a previously undiscovered class of diheme enzymes that exhibits mechanistically distinct peroxidase activity through stabilization of a bis-Fe(IV) state [70]. Together, the crystallographic, spectroscopic, and bioinformatic study identified essential catalytic residues and distinguished this enzyme from other superfamily members [70]. Another example is found in metalloenzymes, where the metal coordination sphere plays a key role in controlling reactivity. In 2022, Cheng et al. studied a non-heme iron enzyme with two distinct activities- acting as a thiol oxygenase and a sulfoxide synthase [71]. The authors identified homologs that enabled further study of the variation in the coordination sphere of the Fe(II) ion and ultimately concluded that both the substrates and the residues comprising the secondary coordination shell of the Fe(II) cofactor control the dual activity of the enzyme [71].

Enzyme engineering has long been of interest in the biochemical community, especially towards designing catalysts with altered substrate specificity, increased thermostability, and enhanced catalytic activity [72]. Zetzsche et al. explored biocatalytic cross-coupling with cytochrome P450 enzymes [73]. The authors developed a strategy for engineering these biocatalysts for improved activity and selectivity using the structural information available and bioinformatic approaches [73]. The goal of the research was to explore the cytochrome P450 sequence space for biocatalysts with complementary reactivity [73]. The authors identified candidates with the ability to catalyze oxidative cross-coupling reactions with naphthols and were able to use these candidates to construct fusion enzymes with cytochrome P450 to perform complementary reactions [73]. In addition to providing a framework for bioengineering, genomic enzymology provides a starting point for directed evolution or rational design. The assessment of structure-function space can empower the choice of target, pushing design towards the optimal evolutionary path.

Conclusion

With the wealth of genomic information available, structural and kinetic characterization of superfamilies is vital for probing the paths to evolve different families and the factors that distinguish their cognate activities. Looking forward, four main activities will significantly enhance the outcomes from the methodologies discussed above. First, continued enzyme functional assignment and mechanistic work is needed to provide the prototypes for each isofunctional cluster within SSNs. Second, the greater reliability of structure prediction (as implemented in AlphaFold [37]) will allow for integration of structure as additional data in these analyses. Third, the increased use of machine learning approaches to extract information from the evolutionary drift present in the sequence data of divergent superfamilies will facilitate the connection of primary and tertiary structure to new functionalities. Finally, despite the abundance of sequences, the addition of newly sequenced genomes will continue to accelerate study. As sequence information is added, enhancing the understanding of the more divergent members, better links can be made within and between families from both a functional and an evolutionary perspective.

Abbreviations:

ddhCTP

3’-deoxy-3’,4’-didehydro-CTP

CAZy

Carbohydrate-Active enZYmes

CRs

convergence ratios

FAO

flavin-dependent amine oxidase

GNDs

genome neighborhood diagrams

GNN

genome neighborhood networks

HMMs

Hidden Markov Models

monoPGT

monotopic phosphoglycosyl transferase

polyPGT

polytopic phosphoglycosyl transferase

RiPPs

ribosomally-synthesized and post-translationally modified peptides

SAM

S-adenosylmethionine

SSNs

sequence similarity networks

virus inhibitory protein, endoplasmic reticulum associated, interferon inducible

viperin