Redefining the coenzyme A transferase superfamily with a large set of manually annotated proteins

Abstract

The coenzyme A (CoA) transferases are a superfamily of proteins central to the metabolism of acetyl‐CoA and other CoA thioesters. They are diverse group, catalyzing over a 100 biochemical reactions and spanning all three domains of life. A deeply rooted idea, proposed two decades ago, is these enzymes fall into three families (I, II, and III). Here we find they fall into different families, which we achieve by analyzing all CoA transferases characterized to date. We manually annotated 94 CoA transferases with functional information (including rates of catalysis for 208 reactions) from 97 publications. This represents all enzymes we could find in the primary literature, and it is double the number annotated in four protein databases (BRENDA, KEGG, MetaCyc, UniProt). We found family I transferases are not closely related to each other in terms of sequence, structure, and reactions catalyzed. This family is not even monophyletic. These problems are solved by regrouping the three families into six, including one family with many non‐CoA transferases. The problem (and solution) became apparent only by analyzing our large set of manually annotated proteins. It would have been missed if we had used the small number of proteins annotated in UniProt and other databases. Our work is important to understanding the biology of CoA transferases. It also warns investigators doing phylogenetic analyses of proteins to go beyond information in databases.

1. INTRODUCTION

Acetyl coenzyme A (CoA) and other acyl‐CoA compounds are central in metabolism. Reactions that form or consume these compounds are thus of great interest in biochemistry. The CoA transferases (EC 2.8.3._) are a superfamily that catalyze one such set of reactions. Specifically, they transfer of a CoA group from an acyl‐CoA donor to a carboxylate acceptor. ¹ In total, they catalyze over 100 such reactions, each involving a different donor/acceptor pair.

The diversity of these enzymes and the reactions they catalyze has led to several ways of classifying them. ^{1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9} One of the first and most enduring ways is now two decades old. In 2001, Heider ¹ named three families of transferases (I, II, and III). Members of family I are classical CoA transferases and use an acyl‐CoA as a substrate. Members of family II are distinct by being able to use acyl‐[acyl‐carrier protein], also. Members of family III have amino acid sequences that differ from other families. Prior to Heider, ¹ it was thought different families may exist, ^{3 , 10} but Heider ¹ was crucial to naming the differences. Heider ¹ used only 16 named enzymes, which reflected the amount of evidence then available.

Since that time, investigators have proposed changes to the three families, but changes have been minor. For example, investigators have examined family I transferases and proposed splitting them into different subfamilies. ^{4 , 8 , 9} One problem is that rather than analyze all CoA transferases, investigators have focused on only a few enzymes, usually those closely related to an enzyme just discovered. ^{4 , 6 , 7 , 8 , 9} There has been no attempt to go back, find all CoA transferases, and determine if the original families are still appropriate.

The situation with CoA transferases represents a common problem in analyses of protein families. It is laborious to go back to the primary literature and find all members of a family or superfamily, and so analyses often rely on a few well‐known proteins and sequence homologs. Another approach is to find proteins through UniProt ¹¹ or other databases. ^{12 , 13 , 14} Databases are popular because they have large numbers of proteins, and some proteins are already annotated with functional information from the literature. Indeed, several analyses of protein families have used databases as a major or sole source of functional information. ^{15 , 16 , 17 , 18 , 19} However, it is not clear how complete is information in databases and if it can substitute for a search of the primary literature.

Here we find, annotate, and analyze nearly 100 experimentally characterized CoA transferases from the primary literature. This analysis shows the three traditional families of CoA transferases are not appropriate, and they need to be regrouped into six to reflect evolutionary relationships. Importantly, the six families were readily apparent when only analyzing our large number of manually annotated enzymes. If we used the small numbers of proteins annotated in UniProt and three other databases, the families were not as apparent. Our study of CoA transferases serves as a warning and shows phylogenetic analyses need to go beyond the small numbers of proteins in databases.

2. RESULTS

2.1. Our analysis involves nearly 100 experimentally characterized enzymes

The CoA transferases have been divided into three families, but this classification is based on analyzing few enzymes. ¹ We aimed to do an analysis with all experimentally characterized enzymes reported in the literature. Accordingly, we looked for all enzymes with (a) experimental evidence for catalyzing at least one reaction and (b) an amino acid sequence. Enzymes with catalytic activity inferred by homology, but with no experimental evidence, were not considered (except where noted).

In total, we found 94 enzymes described in 97 publications (Table 1 and Table S1). These publications measured rates of 208 CoA transferase reactions (Table S2). Of these, 105 reactions occurred at rates we considered biochemically significant and were included subsequently in our analysis (see Table S2 and Materials and Methods). In addition to bona fide CoA transferases, we included n = 5 enzymes that do not catalyze CoA transferase reactions but are closely related (indistinguishable from sequence alone).

TABLE 1.

Coenzyme A transferases that have been experimentally characterized in the primary literature ^a

Enzyme	Organism	References	Enzyme	Organism	References
AarC (Aac)	Acetobacter aceti	[20, 21]	FldA (Csp)	Clostridium sporogenes	[22]
AbfT (Cam)	Clostridium aminobutyricum	[23, 24, 25]	Frc (Lac)	Lactobacillus acidophilus	[26]
ACH1 (Sce)	Saccharomyces cerevisiae	[27]	Frc (Ofo)	Oxalobacter formigenes	[29, 30, 31]
Act (Fsp)	Firmicutes sp.	[32]	GctAB (Afe)	Acidaminococcus fermentans	[33, 34, 35, 36]
Act (Vpa)	Variovorax paradoxus	[37]	HadA (Cdi)	Clostridium difficile	[38]
ActA (Cgl)	Corynebacterium glutamicum	[39]	IaaL (Aar)	Aromatoleum aromaticum	[7]
AibAB (Mxa)	Myxococcus xanthus	[40]	IctA (Ate)	Aspergillus terreus	[41]
Asct (Fhe)	Fasciola hepatica	[42]	IpdAB (Mtu)	Mycobacterium tuberculosis	[43]
AtoDA (Eco)	Escherichia coli	[44, 45]	IpdAB (Rjo)	Rhodococcus jostii	[43]
BaiF (Csc)	Clostridium scindens	[46]	MadA (Mru)	Malonomonas rubra	[3, 28]
BaiK (Csc)	Clostridium scindens	[46]	Mcr (Mtu)	Mycobacterium tuberculosis	[47]
BbsEF (Tar)	Thauera aromatica	[48, 49]	Mct (Cau)	Chloroflexus aurantiacus	[50]
Bct (Gme)	Geobacter metallireducens	[51]	Mct (Hhi)	Haloarcula hispanica	[52]
CaiB (Atu)	Agrobacterium tumefaciens	[53]	MdcA (Aca)	Acinetobacter calcoaceticus	[54]
CaiB (Eco)	Escherichia coli	[55, 56, 57]	MdcA (Kpn)	Klebsiella pneumoniae	[58, 59]
CaiB (Psp)	Proteus sp.	[60]	MdcA (Ppu)	Pseudomonas putida	[61]
CarA (Awo)	Acetobacterium woodii	[62]	OXCT1 (Hsa)	Homo sapiens	[63, 64]
Cat1 (Aba)	Acinetobacter baumannii	[65]	Oxct1 (Mmu)	Mus musculus	[66]
Cat1 (Aca)	Anaerostipes caccae	[67]	Oxct1 (Rno)	Rattus norvegicus	[68]
Cat1 (Ace)	Acetobacter cerevisiae	[65]	OXCT1 (Ssc)	Sus scrofa	[69, 70, 71, 72]
Cat1 (Asp)	Anaerostipes sp.	[30]	p49 (Afr)	Artemia franciscana	[73]
Cat1 (Bfr)	Bacteroides fragilis	[65]	PcaIJ (Aba)	Acinetobacter baylyi	[74, 75]
Cat1 (Bsp)	Butyricicoccus sp.	[76]	PcaIJ (Atu)	Agrobacterium tumefaciens	[77]
Cat1 (Cdi)	Corynebacterium diphtheriae	[65]	PcaIJ (Ppu)	Pseudomonas putida	[78]
Cat1 (Ckl)	Clostridium kluyveri	[79]	PcaIJ (Sme)	Sinorhizobium meliloti	[80]
Cat1 (Ibu)	Intesimonas butyriciproducens	[76]	Pct (Cpr)	Clostridium propionicum	[8, 81]
Cat1 (Mca)	Moraxella catarrhalis	[65]	Pct (Mel)	Megasphaera elsdenii	[32]
Cat1 (Mel)	Megasphaera elsdenii	[76]	Pct (Reu)	Ralstonia eutropha	[4]
Cat1 (Pgi)	Porphyromonas gingivalis	[82]	RipA (Ype)	Yersinia pestis	[83, 84, 85]
Cat1 (Psp)	Peptoniphilus sp.	[76]	SCACT (Cgr)	Cutibacterium granulosum	[86]
Cat1 (Rho)	Roseburia hominis	[67]	ScoAB (Bsu)	Bacillus subtilis	[87]
Cat1 (Rsp)	Roseburia sp.	[76]	ScoAB (Hpy)	Helicobacter pylori	[88, 89]
Cat1 (Sal)	Snodgrassella alvi	[65]	Scot (Ala)	Anaerotignum lactatifermentans	[32]
Cat2 (Aca)	Anaerostipes caccae	[67]	SCOT (Dme)	Drosophila melanogaster	[90]
Cat2 (Asp)	Anaerostipes sp.	[76]	Scot (Msp)	Megasphaera sp.	[32]
Cat2 (Pgi)	Porphyromonas gingivalis	[82]	ScpC (Eco)	Escherichia coli	[91]
Cat2 (Rsp)	Roseburia sp.	[76]	Sct (Cau)	Chloroflexus aurantiacus	[92]
Cat3 (Pgi)	Porphyromonas gingivalis	[82]	SmtAB (Cau)	Chloroflexus aurantiacus	[93]
CatIJ (Pkn)	Pseudomonas knackmussii	[94]	SptAB (Aar)	Aromatoleum aromaticum	[95]
CitF (eco)	Escherichia coli	[96, 97]	SUGCT (Hsa)	Homo sapiens	[98]
CitF (Kpn)	Klebsiella pneumoniae	[10, 99]	TbASCT (Tbr)	Trypanosoma brucei	[100]
CoAT (Aac)	Acidipropioni. acidipropionici	[101]	TvASCT (Tva)	Trichomonas vaginalis	[102]
CoaT (Ani)	Aspergillus nidulans	[103]	UctB (Aac)	Acetobacter aceti	[104]
CoaT (Cty)	Clostridium tyrobutyricum	[105]	UctC (Aac)	Acetobacter aceti	[6]
CoAT (Pfr)	Propionibacterium freudenreichii	[106]	YdiF (Eco)	Escherichia coli	[107]
CoAT (Rsp)	Ruminococcaceae sp.	[105]	YfdE (Eco)	Escherichia coli	[6]
CtfAB (Cac)	Clostridium acetobutylicum	[108, 109]	YfdW (Eco)	Escherichia coli	[30, 110]

^a See Table S1 for more details.

The number of enzymes and reactions we found is double that in UniProt and three other protein databases (see below). Our search of the literature was thus exhaustive.

2.2. Phylogeny of CoA transferases reveals six, not three, families

With the large number of experimentally characterized enzymes in hand, we built a phylogenetic of their protein sequences. We constructed these trees using W‐IQ‐TREE ¹¹¹ after aligning sequences with Clustal Omega. ¹¹² We also constructed a sequence similarity network, which is another way to explore how sequences are related. ¹¹³ We constructed this network using Cytoscape ¹¹⁴ after aligning sequences using the Needleman–Wunsch algorithm. ¹¹⁵

The phylogenetic tree revealed problems with dividing the CoA transferases into the three traditional families (I, II, and III) (Figure 1a). The sequences of family I were not closely related, as was apparent in the phylogenetic tree. Indeed, some members of family I were more closely related to members of other families than each other. Worst of all, family I (or family II) is not even monophyletic. These problems were most apparent after including all family II enzymes. Previous analyses either omitted family II enzymes entirely, ^{4 , 8 , 9} or they did not report a phylogenetic tree of their sequences. ^{1 , 2 , 5}

FIGURE 1.

Phylogenetic tree of CoA transferase sequences shows the superfamily should be divided into six, not three families. (a) The three traditional CoA transferase families. Subfamilies of family I also shown. (b) Our six proposed families. The proposed families are consistent with other properties noted (see Figure 3 for names of enzymes and branch support values). Sequences in panel (a) were assigned to families based on descriptions in Reference [1] and to subfamilies based on Figure S1

Some analyses have proposed splitting family I into different subfamilies. ^{4 , 8 , 9} The phylogenetic tree (Figure 1A) shows one proposed split (according to Reference [9]), but it does not solve all problems. First, not all sequences from family I fall into a defined subfamily (IA, IB, and IC) (see also Figure S1). Second, no matter how it is split into subfamilies, family I is not monophyletic.

The tree suggests dividing the CoA transferases into six, not three, families (Figure 1b). These six families are monophyletic (Figure 1b), and members within the same family are more closely related to each other than to members of other families. This division is also consistent with other properties, such as reactions catalyzed, type of catalysis, and crystal structure (see below).

The sequence similarity network also showed that CoA transferases fall into six or more families (Figure 2). We constructed the network to show clusters of sequences with ≥25% identity (the minimum value observed for most homologous sequences). ^{116 , 117} Sequences fell into six clusters, with one additional sequence not falling into any cluster. This supports the idea that CoA transferases form more than three and at least six families.

FIGURE 2.

Sequence similarity network of CoA transferases show they cluster into six groups, which correspond to the six families we propose (see Table S1 for enzyme IDs)

2.3. The families of CoA transferases differ in reactions catalyzed

We explored if our six proposed families of enzymes would differ in the reactions they catalyzed. We created a heat map of the n = 94 enzymes and n = 105 reactions we identified earlier, and it revealed members within a given family generally catalyze similar reactions (Figure 3).

FIGURE 3.

Heat map of CoA transferases and reactions catalyzed, affirming division into six families (see Table S1 for information on these enzymes and Table S2 for their reactions)

The Cat1 family catalyzes reactions involving small acyl‐CoA (Figure 3). Acetyl‐CoA/butyrate was the most common pair of substrates and used by 51% of enzymes. Other common substrates were acetyl‐CoA/propionate (49% enzymes), acetyl‐CoA/succinate (46% enzymes), and propionyl‐CoA/succinate (14% enzymes).

The OXCT1 family is different in using oxo and hydroxy acyl‐CoA (Figure 3). The most common substrates were acetyl‐CoA/lactate (29% of enzymes), acetoacetyl‐CoA/succinate (29% of enzymes), acetyl‐CoA/acetoacetate (19% of enzymes), and β‐ketoadipyl‐CoA/succinate (14% of enzymes).

Half of all members of the Gct family catalyze only nontransferase reactions (Figure 3). For example, two members cleave the ring of a cholesterol‐CoA derivative. ⁴³ All reactions still involve acyl‐CoA as a substrate.

All members of CitF and MdcA families can catalyze CoA transferase reactions (Figure 3). Specifically, CitF uses acetyl‐CoA/citrate, and MdcA uses acetyl‐CoA/malonate. However, they also catalyze transferase reactions involving acyl‐ACP. ^{10 , 118 , 119} Acyl‐ACP is the likely substrate in cells, as CitF and MdcA are part of larger enzymes systems containing ACP. ^{54 , 97 , 118 , 119 , 120}

The reactions of the Frc family differ from those of other families, but they otherwise have little in common (Figure 3). Formyl‐CoA/oxalate is used by 21% of enzymes, but all other pairs of substrates are used by two or fewer enzymes.

In sum, there are several differences in the reactions catalyzed by the CoA transferases. These differences provide further support for dividing the transferases into six families.

2.4. The families of CoA transferases have different amino acid residues in the active site

In a number of CoA transferases, the residues of the active site have been identified by mutation or crystallography (see Table S1). We wanted to see if one key residue—that involved in covalent catalysis—and the surrounding region differed across our proposed families. We aligned sequences of the enzymes and highlighted the residue involved in covalent catalysis.

Our analysis showed some similarities across families, but these are punctuated by clear differences (Figure 4). In the Cat1 family, the residue involved in covalent catalysis was a glutamate (E567). Glutamate served this role in families OXCT1 and Gct, also, but the glutamate occupies a completely different position in the alignment (E582). This resulted from a number of indels between the glutamate and a conserved glycine (G584). In the Frc family, the residue involved in covalent catalysis was aspartate (D533). In CitF and MdcA families, no residues for covalent catalysis were identified, as covalent catalysis is not thought to occur. ^{10 , 121} In the Gct family, some members only catalyze nontransferase reactions, and for these, no residues for covalent catalysis were identified. Indeed, these members are not thought to use covalent catalysis. ^{40 , 43}

FIGURE 4.

Sequence alignment of the active site of CoA transferases reveals differences among families (see Table S1 for information on these enzymes, their sequences, and their active site residues)

This analysis shows our proposed families differ in residues of the active site. These differences, such as in the region surrounding the catalytic glutamate, become clear by analyzing a large number of proteins where the catalytic residue was annotated.

2.5. The families of CoA transferases differ in structure

Our analysis showed families differ by sequence, and we wanted to see if crystal structures would also differ. Accordingly, we aligned structures of n = 24 enzymes, and we made a tree of the alignment.

We found our proposed families have different structures (Figure 5). In the tree, structures clustered by family. Differences were also clear by comparing structures visually. All transferases had two domains, each with several layers of α helices and β sheets. However, they differed in the number and arrangement of layers. Domain 1 of Gct, for example, had a large number of layers, with three layers of α helices and two β sheets (Figure 5 and Figure S2).

FIGURE 5.

Phylogenetic tree of CoA transferase structures shows that differences among families extend to the structural level. Structures were aligned together, colored individually, then overlaid within family (see Table S1 for information on these enzymes and PDB accession numbers for structures)

Compared to the phylogenetic tree built with protein sequences, the tree built with structures differed somewhat. For example, families CitF and MdcA were distantly related according to their sequences, but appeared more closely related from their structures. The families were still distinct, showing our proposed families are valid at both sequence and structural levels.

One caveat is there were few structures available for families CitF and MdcA, and those available are for putative enzymes (Table S3). As putative enzymes, they have no experimental evidence of activity, and instead their catalytic activity is inferred by homology. If we exclude them, the position of the remaining families in the tree is not affected (Figure S3). We did not use putative enzymes elsewhere in our analysis. In sum, despite some caveats, this analysis agrees with the sequence analysis and suggests dividing the CoA transferases into six families.

2.6. The structure of CoA transferases differs around the active site

Having found the CoA transferases differ in overall structure, we wanted to see if differences extended to the active site. We were most interested in comparing Cat1, Gct, and OXCT1 families, given glutamate serves as a catalytic residue in all three. The crystal structures showed differences around the catalytic residue (Figure S4). In the Cat1 family, the residue was followed by an alpha helix. In Gct, and OXCT1 families, the helix was replaced by part of a β sheet. The helix and sheet correspond to a region of indels in the sequence alignment (from E567 to G584; see Figure 4). These results point to structural elements gained or lost during the evolution of the CoA transferases. Despite clear similarities, there are also differences around the active site for Cat1, Gct, and OXCT1 families.

2.7. Enzyme databases are missing half of all enzymes and reactions

Our analysis used the primary literature as a source for information on enzymes. Searching for information in this way is laborious, and many investigators search protein databases instead. We wanted to see if these databases had comparable information and would have led us to the same conclusions.

We searched four widely used databases (BRENDA, KEGG, MetaCyc, UniProt) for CoA transferases. ^{11 , 12 , 13 , 14} We searched for enzymes according to their enzyme commission (EC) numbers and three other identifiers (Rhea ID, KEGG REACTION ID, MetaCyc Reaction ID/BioCyc ID) ^{12 , 14 , 122 , 123} (Table S4). We counted enzymes annotated with a reaction, amino acid sequence, and literature reference.

We found information in databases was far from complete. The four databases we searched had information annotated for only n = 48 enzymes (Figure 6a and Table S5). Half (49%) of the enzymes we found in the primary literature were thus missing information. The databases reported the enzymes catalyzed only n = 57 CoA transferase reactions (Figure 6a and Table S6). About half (46%) of the reactions we found were missing.

FIGURE 6.

Protein databases are missing large numbers of CoA transferases in the primary literature. (a) Venn diagram of CoA transferases annotated by databases. (b) Heat map of CoA transferases and reactions annotated by databases (see Table S5 for information on these enzymes and Table S6 for their reactions)

We built a reaction heat map of all enzymes and reactions available in databases (Figure 6b). The result looked different from the heat map built using information from the primary literature (see Figure 2). When using information in databases, all families had fewer members, and the MdcA family was completely missing. Furthermore, reactions catalyzed by families Cat1 and OXCT1 were no longer clearly different. Consequently, it is not obvious that family I proposed by Heider ¹ should be divided into separate families, as it is when using information from the primary literature. Without digging into the primary literature, we would have reached incorrect conclusions about the CoA transferase superfamily.

2.8. Additional phylogenetic trees shed more insight into the CoA transferases

Crystal structures show CoA transferases have two protein domains. We examined how these domains are organized within each enzyme, and then we built a phylogenetic tree to see if the organization differed by family (Figure S5). We found domains were organized in different ways (Figure S5a). In some enzymes, domains were separated (on two polypeptide chains), but in others, they were fused (on one chain). In rare cases, domains were fused and duplicated (two polypetide chains with two domains each). A phylogenetic tree (Figure S5b) showed members of Gct had only separated domains, whereas Cat1, MdcA, and CitF had only fused domains. OXCT1 had both separated and fused subunits. Frc had both fused domains and fused domains that had been duplicated. These differences in organization are small, but help further define the families of CoA transferases.

To this point, we have built phylogenetic trees using one amino acid sequence per enzyme. Alternatively, we could build trees using two separate sequences per enzyme, with one sequence per protein domain. Such a tree would reveal if domains evolved separately from a common ancestor, as proposed for some CoA transferases. ^{2 , 34} We thus built such a tree, and it showed sequences did not always cluster by domain (Figure S6). For example, domain 1 of CitF clustered with domain 2 of Gct. If the domains evolved separately, all sequences for domain 1 should cluster together, and sequences for domain 2 should cluster apart. This suggests that domains did not evolve separately, at least for some families.

We built phylogenetic trees using CoA transferases and very few (n = 5) additional enzymes. We in fact considered many more proteins, but they turned out not to be closely related to CoA transferases. In one phylogenetic tree, we included proteins considered by Pfam, ¹²⁴ SCOP, ¹²⁵ and other authorities ² to be related to CoA transferases (see Table S7). The tree showed these proteins were not as closely related to the CoA transferases as the CoA transferases were to themselves (Figure S7). Furthermore, none are known to use acyl‐CoA as a substrate or perform transferase reactions. It makes sense to set the boundaries of the CoA superfamily around the proteins of our original tree (Figure 1).

3. DISCUSSION

Our analysis is important to the biology of CoA transferases and to phylogenetic analyses of proteins. It is important to the biology of the CoA transferases because it uproots the idea that these enzymes fall into three families. ¹ This idea has stood for two decades, and others have proposed only minor changes as scores of enzymes have been discovered. ^{4 , 5 , 6 , 7 , 8 , 9} The strength of our approach is it uses all experimentally characterized enzymes and examines several properties (e.g., sequences, reactions, and structures). It is by using this comprehensive approach that we can conclude CoA transferases fall into six families.

Our work is important to the biology of CoA transferases, also, by shedding insight into their evolutionary history. Specifically, it suggests that the superfamily evolved (or lost) the ability to catalyze their reactions multiple times. Three families (MdcA, CitF, and Gct) have members that catalyze non‐CoA transferase reactions (either alone or in addition to CoA transferase reactions). The other three families (Frc, Cat1, and OXCT1) catalyze CoA transferase reactions, and few others. Given their complex phylogeny, the CoA transferases must have gained (or lost) the ability to catalyze CoA reactions at least twice. Reflecting this, the three families that catalyze mostly CoA transferase reactions (Frc, Cat1, and OXCT1) show key differences around the active site. This complex history is apparent only from analyzing our large set of manually annotated enzymes and reactions.

Besides its importance to the CoA transferases, our analysis is important to phylogenetic analyses. Namely, it shows the importance of conducting phylogenetic analyses with all proteins reported in the primary literature. It is convenient to focus on only a few proteins, or to use proteins with information reported in UniProt or other databases. Unfortunately, using only databases would lead to an incomplete picture. It is already known that databases rely on few publications for most of their information. ¹²⁶ For gene ontology annotations in UniProt, the top 0.14% publications are used as a source of information for 25% of the proteins. ¹²⁶ Our study shows that databases are missing information outright; half of all enzymes were missing the reaction they catalyze, their amino acid sequence, or a literature reference. Databases can be useful for starting a literature search, but they are not a replacement for one.

In sum, our work is important to the CoA transferase superfamily and to phylogenetic analyses in general. It warns investigators doing these analyses to go beyond proteins annotated in databases. Though our work on CoA transferases is comprehensive, a phylogenetic analysis is never done—it simply awaits discovery of the next protein. Digging deep in the literature for proteins ensures analyses are as accurate as possible until the next discovery.

4. MATERIALS AND METHODS

4.1. Search for CoA transferases in primary literature

We searched for CoA transferases in the primary literature. As explained in Results, we considered only enzymes with (a) experimental evidence for catalyzing at least one reaction and (b) an amino acid sequence. An important resource was Zhang et al. ⁸⁶ which had manually annotated information for n = 51 enzyme from Cat1, OXCT1, and Gct families. We also used databases ^{11 , 12 , 13 , 14} as starting points.

We identified reactions as biochemically significant if they occurred at fast rates. We arbitrarily defined these as reactions that occur at ≥10% of the fastest CoA transferase reaction for a given enzyme. If an enzyme catalyzed only one CoA transferase reaction, the reaction was significant by default. This definition is arbitrary, and we wanted to identify reactions that are statistically different from 0 instead (p < .05). However, many literature studies do not report statistical significance.

Some publications did not give report gene or protein names. We assigned a name in cases where it was missing. Information is otherwise as reported in the publication.

4.2. Construction of phylogenetic tree of sequences

To construct phylogenetic trees of amino acid sequences, we first performed multiple sequence alignment with Clustal Omega. ¹¹² Following References [127, 128], we ran Clustal Omega using the package msa ¹²⁹ of R ¹³⁰ and the default parameter values.

With the aligned sequences, we calculated the tree using maximum‐likelihood with W‐IQ‐TREE ¹¹¹ and default parameter values. The substitution model used was LG + F + G4, which gave the lowest AICc value. Branch support values were calculated using the ultrafast bootstrap analysis ¹³¹ with 1,000 maximum iterations.

We visualized the final tree with the package ggtree ¹³² of R and the Interactive Tree of Life (iTOL). ¹³³ We used ggtree for rectangular trees (Figures 3, 4, and 6 and Figures S1, S5, S6, and S7) and the iTOL for equal‐daylight trees (Figures 1 and 5 and S3).

As mentioned, CoA transferases show different organization of domains (Figure S5), and we had to take this into account before alignment. For enzymes with fused domains, we used the amino acid sequence as is. For enzymes with separated domains, we concatenated the sequences of the two subunits before alignment. For enzymes with fused domains that had been duplicated, we arbitrarily chose the first subunit for alignment and discarded the second. If we chose the second subunit instead, the tree was not affected. This procedure ensured that each domain was represented once and only once in the alignment.

Protein sequences are available in FASTA format on Figshare (https://figshare.com/articles/dataset/Sequences_and_phylogenetic_trees_of_coenzyme_A_transferases/16617892). Phylogenetic trees are available in Newick format at the same resource.

4.3. Construction of sequence similarity networks

To construct sequence similarity networks, we first performed pairwise sequence alignment using the Needleman–Wunsch algorithm. ¹¹⁵ We ran this algorithm using the package Biostrings ¹³⁴ in R. We used BLOSUM62 substitution matrix, gap opening penalty of 10, and gap extension penalty of 0.5. We then calculated pairwise identity of the sequences.

With the pairwise identity of the aligned sequences, we constructed the sequence similarity network in Cytoscape. ¹¹⁴ We filtered out (removed) edges with <25% identity. We visualized the network using the yFiles organic layout.

4.4. Construction of heat map of reactions

We constructed a heat map of CoA transferase reactions using data from the primary literature. We took reactions we found in the literature and ordered them according to their frequency within family. Following this order, we assigned reactions IDs, created a heat map using ggplot2, ¹³⁵ and laid it next to the phylogenetic tree constructed with ggtree.

4.5. Construction of sequence alignment of active site

We constructed a sequence alignment of the active site using the same alignment for the phylogenetic trees. The sequence alignment was visualized using ggplot2.

4.6. Construction of phylogenetic tree of crystal structures

To construct the tree of crystal structures, we first aligned the structures with SALIGN. ¹³⁶ We chose this alignment tool because it can accommodate proteins with multiple subunits. For proteins with two subunits (A and B), we used the command “FIRST:A:LAST:B” to ensure all residues were included. The tool outputted (a) the aligned structures in pdb format and (b) pairwise distances between structures.

With the pairwise distances, we calculated the tree using the minimum evolution method. We used the package ape ¹³⁷ of R and default parameter values. The pairwise distances were from the log file of SALIGN, and they were for the guide tree during the last alignment iteration.

We visualized the aligned structures using PyMOL (v. 2.0, Schrödinger, LLC). We loaded aligned structures (pdb format) into PyMOL, colored them using palettes generated by package colorspace ¹³⁸ of R, and then captured images as ray traces. For best comparison, we captured all ray traces from the same view for a given figure.

4.7. Searching for CoA transferases in databases

To find CoA transferase reactions in databases, we searched for EC numbers for CoA transferase reactions (Table S4). For databases that allowed them, we searched for additional reaction IDs (Table S4). For UniProt, we also searched for Rhea IDs. For MetaCyc, we also searched for Rhea IDs and MetaCyc Reaction IDs/BioCyc IDs. For KEGG, we also searched for KEGG REACTION IDs.

Our analysis also included n = 5 enzymes that catalyze non‐CoA transferase reactions only. For fair comparison, we searched databases for these enzymes, too.

We counted enzymes annotated with a reaction, amino acid sequence, and literature reference. We did not count enzymes with partial or incorrect information (see Table S5). For UniProt, we counted only enzymes with “Experimental evidence at protein level” (see Table S5).

The Venn diagram of reactions by database was generated using package venn ¹³⁹ in R. The phylogenetic tree and heat map were generated as described previously.

4.8. Prediction of sequence regions of protein domains

To build a phylogenetic tree of the two different domains of CoA transferases (Figure S6), we had to predict sequences regions for each domain. First, we identified sequence regions in enzymes where they were known (or obvious). These enzymes included n = 24 with crystal structures and n = 12 where domains were part of separate subunits. Second, we built profile hidden Markov models (pHMMs) of these sequence regions. We used the hmmbuild command of HMMER, ¹⁴⁰ using the known sequence regions as the input. Third, we predicted sequence regions of domains in all remaining CoA transferases. We used the hmmsearch command of HMMER, using the pHMMs and sequences of CoA transferases as an input.

Supporting information

Figure S1. Phylogenetic tree of CoA transferases that shows subfamilies IA, IB, and IC proposed in Tielens et al. (2010). Sequences included are from Fig. 3 of Tielens et al. (2010), plus all sequences from our own analysis. Some sequences from Tielens et al. (2010) belong to putative enzymes (ones with no biochemical evidence of activity), and for this reason were not included in our own analysis.

Figure S2. Crystal structure of a Gct family protein, showing the large number of layers in domain 1 that distinguishes this family. Structure of a Cat1 family protein, with typical three layers, shown for comparison. Proteins shown are GctAB (Afe) and AarC (Aac). Structures are presented in stereo and can be seen in 3D.

Figure S3. Phylogenetic tree of CoA transferase structures after removing putative enzymes, showing positions of other families in tree are unchanged.

Figure S4. Crystal structures show the region around active site of Cat1 family differs from that of Gct and OXCT1 families. Structures were aligned together and then overlaid within family. The indel region refers to E567 to G584 in Figure 4. The catalytic residue refers to that shown in Figure 4. Only structures with an identifiable catalytic residue are shown. See Table S1 for information on these enzymes and PDB accession numbers for structures.

Figure S5. CoA transferases show different organizations of their domains. (A) Types of organization. The N‐terminus and C‐terminus of each polypeptide (subunit) is shown. (B) Enzyme by type of organization. See Table S1 for information on these enzymes.

Figure S6. Phylogenetic tree of CoA transferases with sequences split by domains, showing that sequences do not always cluster by domain. See Table S1 for information on these enzymes.

Figure S7. Phylogenetic tree of CoA transferase superfamily plus proteins from additional families, showing the additional families are not closely related. See Table S1 for information on enzymes of the CoA transferase superfamily (Cat1, OXCT1, Gct, MdcA, Frc, CitF), and see Table S7 for information on proteins from additional/other families.

Table S1. Coenzyme A transferases that have been experimentally characterized in the primary literature

Table S2. Reactions of coenzyme A transferases in the primary literature

Table S3. Coenzyme A transferases that areputative enzymes and included in this analysis only for their structures

Table S4. Accession numbers of reactions in databases

Table S5. Coenzyme A transferases that have been annotated in databases

Table S6. Reactions of coenzyme A transferases in databases

Table S7. Proteins that some authorities consider part of a CoA transferase superfamily (1 representative protein per family)

Table S8. Coenzyme A transferase sequences and proposed subfamilies from Tielens et al. (2010)

Hackmann TJ. Redefining the coenzyme A transferase superfamily with a large set of manually annotated proteins. Protein Science. 2022;31:864–881. 10.1002/pro.4277