Different types of contextual information can be obtained by means of prokaryotic comparative genomics and used to elucidate functionally uncharacterized proteins. First, fusions of uncharacterized domains or genes to functionally characterized domains or genes suggest participation of the former in processes similar to those of the latter. Second, clustering of genes in operons usually implies coordinated gene expression, and conserved prokaryotic gene neighborhoods are a strong indication of functional interaction, especially through physical interactions of the encoded protein products. The power of contextual inference, especially for the less prevalent protein families, has been considerably boosted due to the enormous increase in data from the various microbial genome sequencing projects 4142 and the development of publicly available resources such as WIT2/PUMA2 and STRING/SMART that integrate a variety of contextual information 43444546.

Accordingly, we set up a protocol to identify comprehensively the network of contextual connections centered on the prokaryotic Ub-related proteins detected in the above searches, and used it to infer the functional pathways in which they participate. We first determined the complete domain architectures of all the Ub-like proteins using a combination of case-by-case PSI-BLAST searches and searches against libraries of position specific score matrices (PSSMs) or HMMs of previously characterized protein domains. We then established the gene neighborhoods (see Materials and methods, below) for these Ub-like proteins and found a number of conserved neighborhoods containing genes for specific protein families often co-occurring with the Ub-like proteins. Each of the families belonging to the conserved neighborhoods were used as starting points for further PSI-BLAST searches to identify homologous proteins in prokaryotic genomes. These homologs were then used as foci to identify any conserved gene neighborhoods occurring with them. This way we built up a comprehensive set of conserved gene neighborhoods for the Ub-like proteins as well as their putative functional partners and their homologs, which were identified via contextual analysis. As a result we identified several persistent architectural and gene neighborhood themes associated with the prokaryotic Ub-like proteins. We discuss below the most prominent of these, especially those with relevance to the early evolution of the Ub-signaling related pathways.

Several families of prokaryotic Ub-like proteins, namely ThiS, MoaD, RnfH, TmoB, and a newly detected family typified by Ralstonia solanacearum RSc1661 (gi: 17428677; see below), are characterized by a single standalone Ub-like domain. In several cases the ThiS and MoaD are fused to ThiG and MoaE (Figure 3), which respectively are their functional partners in the transfer of sulfur to the substrates (Figure 1). We also noted that a distinct version of ThiS is fused to the carboxyl-terminus of the sulfite reductase in certain actinobacteria (for example, Nocardiodes and Acidothermus cellulolyticus), whereas MoaD might be fused to aldehyde ferredoxin oxidoreductase (Azoarcus; Figure 3). Another newly characterized family of Ub-domains typified by the protein mlr6139 from Mesorhizobium loti (gi: 14025878) is characterized by three tandem repeats of the Ub-like domain (Figure 3; see below for details).

A family of Ub-like domains, distinct from ThiS, is found fused to the amino-terminus of the adenylating Rossmann fold domain of certain ThiF proteins, such as that from Campylobacter jejuni (gi: 57166736; Figure 3). In the lambda and T1 phage TAPI proteins, the Ub-like domain is fused to another small globular carboxyl-terminal domain via a glycine-rich low complexity linker. In some cases the TAPI protein itself may be fused to the tail-assembly protein J (TAPJ) or K (TAPK), which contain two peptidase domains, namely the JAB domain and NlpC/P60 domain with the papain-like fold (Figure 3) 13.

In the proteins typified by the Thermotoga maritima TM_0779, the amino-terminal Ub-like domain is linked to a carboxyl-terminal Mut7-C RNAse domain and a zinc ribbon domain (Figure 3) 47. Iterative sequence profile searches with the Mut7-C domain as a query recovered the previously characterized PIN (PilT-N) RNAse domains with significant e values (e < 10^-3). The two domains share an identical pattern of conserved catalytic residues, suggesting a similar enzymatic mechanism 48. In the actinobacteria, the YukD-like β-grasp domain is fused to an integral membrane domain with 12 transmembrane helices (Figure 3). The TGS domain, as previously reported, was almost always found in various RNA-binding multidomain proteins; hence it is not discussed here in detail 37. Likewise, the architectures of β-grasp ferredoxins, which are typically found as a part of multidomain oxido-reductases, have previously been considered in depth and are not dwelt upon in detail here 49.

The multistep biosynthetic pathways for the major cofactor thiamine is the experimentally best characterized of the prokaryotic systems involving Ub-like sulfur transfer proteins and associated E1-like enzymes. Furthermore, there has also been a comprehensive comparative genomics analysis of the components of the prokaryotic thiamine biosynthetic pathway 50. In the present report we focus only on associations in these systems that are pertinent to the evolution of the Ub-signaling related pathways and previously unnoticed features of the distribution and gene neighborhoods of the ThiS genes.

The ThiS protein is highly conserved in all of the major bacterial and archaeal lineages, suggesting that it may be traced back to the last universal common ancestor (LUCA). In most bacterial lineages ThiS is encoded within a large operon including several other genes for thiamine biosynthesis. These include genes encoding proteins for both the major branches of the thiamine biosynthetic pathway (for instance, the aminoimidazole ribotide utilizing branch with ThiC and ThiD, and the sulfur transfer and hydroxyl-ethyl-thiazole forming branch with ThiS, ThiG, ThiO, ThiH) and the stem combining the products of branches to form thiamine phosphate (ThiE; Figure 4) 50.

Although the individual genes occurring in this conserved gene neighborhood exhibit some variability across different bacteria, ThiS is most strongly coupled with ThiG (approximately 80%) - its physically interacting functional partner within the operon. The next strongest coupling of ThiS in bacteria is with its other complex forming partner, namely the adenylating enzyme ThiF (approximately 20%). This is not surprising, given that ThiF and ThiG compete for ThiS to catalyze two successive steps in the sulfur incorporation process 2551. Very rarely, ThiS may also be coupled with ThiC (for example, Cytophaga hutchinsonii). The genes for the group of ThiF proteins containing a fused Ub-like domain at their amino-termini (see above) typically co-occur in predicted operons with standalone ThiS genes (Figure 4). This suggests that their fused Ub-like domain plays a role different from the standalone ThiS protein. However, in a single case (Pelobacter propionicus), the Ub-like domain-ThiF fusion proteins do not occur in an operon with other thiamine biosynthesis genes, instead co-occurring with O-acetylhomoserine sulfhydrylase and cysteine synthase (Figure 4). Similar operonic association of ThiS alone, or ThiS and ThiG with genes for cysteine biosynthesis such as cysteine synthase, and sulfite transporter genes are also seen in Pelodictyon and Chlorobium (Figure 4 and Additional data file 1). These represent multiple independent associations of thiamine biosynthetic genes with sulfur assimilation and cysteine biosynthesis genes, which is consistent with the fact that cysteine is the sulfur donor for the ThiS thiocarboxylate.

The genes of the archaeal ThiS orthologs are not found in any conserved gene neighborhoods, and this is consistent with the previously noted absence of ThiF and ThiG orthologs in the archaea, and the presence of an alternative branch for hydroxyl-ethyl-thiazole biosynthesis 50. This observation suggests that the archaeal ThiS genes might even have been recruited for a sulfur transfer process distinct from thiamine biosynthesis.

The MoaD-MoeB system in molybdenum and tungsten cofactor biosynthesis mirrors the ThiS-ThiF system in thiamine biosynthesis. MoaD is also conserved across all major archaeal and bacterial lineages, suggesting that it existed in the LUCA. Unlike ThiS, MoaD is present in Mo/W cofactor biosynthesis operons in both bacteria and archaea (Table 1). This implies that both ThiS and MoaD had probably diverged from each other by the time of the LUCA, but the recruitment of ThiS for a sulfur transfer system in thiamine biosynthesis emerged early in the bacterial lineage, only after it had split from the archaeal lineage. In contrast, the deployment of MoaD in Mo/W cofactor biosynthesis appears to have happened in the LUCA itself. The Mo/W cofactor biosynthesis operons from different bacteria encode a variety of proteins, including those involved in using the GTP precursor (MoaA and MoaC); the MoeB, MoaD and MoaE products, which are downstream of the former and involved in molybdopterin biosynthesis; and MoeE, MogA, MobD, and the MOSC domain proteins, which are involved in formation of MoCo/WCo and its terminal derivatives (Figure 4, Table 1 and Additional data file 1) 525354. Although the predicted operons exhibit variability across prokaryotes in terms of the different genes included in them, the core conserved gene neighborhood in bacteria contains the genes for MoaD and MoaE, which together constitute the molybdopterin (MPT) synthase, which transfers the sulfur from the MoaD thiocarboxylate to the precursor Z (cyclic pyranopterin monophosphate) to form MPT 5255 (Figures 1 and 4). In a few cases MoaD may be adjacent to the gene for MoeA, which acts on the product downstream of the reaction catalyzed by the MPT synthase. MoaD, unlike ThiS, is rarely found immediately adjacent to the gene for its adenylating enzyme, MoeB (Figure 4). This distinction may be related to experimental results, which indicate that MoaD and MoeB do not form a covalently linked persulfide or thioester complex, unlike ThiS and ThiF or the Ub/Ubl and the E1s (Figure 1) 30.

A distinct set of MoaD genes are found strictly adjacent to genes encoding an aldehyde ferredoxin oxidoreductase (AOR) in a sporadic group of phylogenetically distant archaea and bacteria (Table 1), suggesting that they might constitute a mobile gene cluster. Additionally, these gene neighborhoods often include MoeB and occasionally other cofactor biosynthesis genes such as MoaA and MoaE, and a pyridine disulfide oxidoreductase in close vicinity to MoaD and the AOR genes (Figure 4). In some organisms this MoaD containing gene cluster is distinct from the MoCo biosynthesis operon found elsewhere in the genome of the same organism. Experimentally characterized versions of these AORs have been shown to utilize a tungsten-containing variant of the cofactor 56. Taken together, these observations suggest that these AOR linked MoaD genes might specifically participate in the synthesis of molybdopterin for WCo generation for the AORs.

Beyond the above-stated predicted operons, with the bona fide ThiS/MoaD and the ThiF/MoeB enzymes involved in conventional thiamine and MoCo/WCo biosynthesis, we also recovered several other predicted bacterial operons encoding homologous proteins. These gene clusters typically encode a ThiS/MoaD related protein and an E1-like enzyme related to ThiF/MoeB with a carboxyl-terminal rhodanese domain, but they do not contain any genes encoding other components of the two cofactor biosynthesis pathways (Figures 3 and 4, and Table 1). The bacteria that contain these predicted operons also contain independent thiamine or molybdenum operons, highlighting the functional distinctness of the pathways encoded by these gene neighborhoods (Table 1). Interestingly, this class of predicted operons also often contains a gene encoding a standalone version of the JAB metallopeptidase, which forms a monophyletic clade within the tree of all JAB domains (Figures 4 and 5; see Materials and methods, below, for details). There are at least five distinct subtypes of this class of gene neighborhoods, which exhibit a sporadic distribution across phylogenetically diverse bacteria, suggesting possible dispersion through lateral gene transfer (Table 1 rows 4a-4e and Figure 4). One of these subtypes of gene clusters has been shown to encode components of the biosynthetic pathway for the siderophores and secreted protective compounds PDTC (pyridine-2,6-bis[thiocarboxylic acid]) and quinolobactin in Pseudomonas stutzeri/P. putida and P. fluorescens, respectively 5758. Our analysis of gene neighborhoods revealed that related conserved gene neighborhoods are also found in several distantly related proteobacteria, such as Ralstonia solanacearum and Nitrosomonas europaea, suggesting that such compounds might be widely produced (Table 1 row 4a and Figure 4).

There are considerable differences in the genes and corresponding biosynthetic pathways (related to amino acid biosynthetic pathways) producing the basic molecular skeleton of each of these metabolites. For example, in the case of quinolobactin a xanthurenic acid skeleton is used, whereas in the case of PDTC a dipicolinic acid skeleton is used (Figure 1) 5758. However, all of these operons contain a conserved core of genes whose products catalyze the critical sulfurylation step required for the production of all of these compounds 5758. This core group encodes a carboxylate AMP ligase, which adenylates a carboxylate group on the precursor, and proteins for a sulfur transfer system that forms a thiocarboxylate group from the carboxy adenylate produced by the AMP ligase (Figure 1). The proteins of the sulfur transfer system include an E1-like protein with a carboxyl-terminal rhodanese domain, a ThiS/MoaD-like protein, and a protein with a JAB metallopeptidase domain (Figure 4). The first two enzymes are likely to participate in a sulfur transfer pathway similar to those seen in the conventional thiamine and MoCo/WCo pathways, with the rhodanese domain probably abstracting the sulfur from a small molecule donor such as cysteine (as in the case of ThiI), and the E1-like protein adenylating and transferring the sulfur to the ThiS/MoaD-like protein to form a terminal thiocarboxylate (Figure 1).

Most other predicted operon subtypes of this class appear to exhibit different variants of the core sulfur transfer system seen in the above-described siderophore biosynthesis gene clusters (Table 1 and Figure 4). A simple subtype seen in a wide range of bacteria contains just three genes encoding a ThiS/MoaD-like protein, a protein combining an E1-like module and a rhodanese domain, and JAB domain peptidase. Derivatives of this basic subtype might simply contain genes for the JAB domain peptidase and E1 + rhodanese protein (Table 1 row 4b and Figure 4). Another subtype additionally combines the cysteine synthase with the three genes of the basic operon, suggesting that they might couple sulfur transfer to production of the major cellular sulfur donor cysteine (Table 1 row 4c and Figure 4). A variant of the cysteine synthase containing operon subtype, which is particularly prevalent in the actinobacteria, includes ClpS that is involved in degradation of proteins through the Clp system and an uncharacterized helical protein that is almost exclusively encoded in this predicted operon subtype (Table 1 row 4d and Figure 4). Other links to sulfur metabolism are hinted at by another major subtype of this class of gene neighborhoods, where genes for the ThiS/MoaD, JAB, and E1-like proteins are combined with genes coding sulfite/sulfate ABC transporters, PAPS reductase, ATP sulfurylase, sulfite reductase, O-acetylhomoserine sulfhydrylase, and adenylylsulfate kinase. The E1-like protein of these predicted operons always lacks the carboxyl-terminal rhodanese-like domain. However, these gene neighborhoods always contain a SirA (cysteine containing domain 1 [CCD1]) protein, which was predicted to play a role similar to that of rhodanese 59 (Table 1 row 4e and Figure 4). These observations suggest that these gene clusters are principally involved in the assimilation of sulfur from sulfate/sulfite and that this sulfur might be terminally transferred to the ThiS/MoaD-like proteins encoded by them.

The genomes of lambdoid and T1-like phages are known to contain related tail assembly gene complexes 60. In a large number of phages this complex encodes a protein TAPI that contains an Ub-like domain related to ThiS/MoaD (Figure 2). The exact function of this protein tail assembly is unclear, but it is not incorporated into the mature tail. Analysis of the gene neighborhoods revealed that TAPI is most often flanked by the genes encoding the TAPK protein, with JAB and NlpC/P60 peptidase domains, and the TAPJ protein, which is required for host specificity (Table 1 row 5 and Figure 4). The JAB domains found in these gene associations are also a part of the monophyletic clade, including those from the above-described class of gene neighborhoods. Variants of this organization lacking either of the two flanking genes are seen in a few phages/prophages, and in a small group of phages TAPI is flanked by a version of TAPK containing only an NlpC/P60 peptidase domain (Figure 4). It is possible that the latter versions are actually degenerate variants of the former versions and are typical of integrated prophages.

A number of sets of predicted operons, each with a distinctive sporadic distribution across several phylogenetically distant bacteria and encoding proteins with JAB domain and E1-like enzymes, were recovered in our search for conserved gene neighborhoods. E1-like enzymes in these gene neighborhoods never contained a carboxyl-terminal rhodanese domain. However, they were typically fused, either at the amino-terminus or the carboxyl-terminus, to the JAB domain. In the instances in which they were not fused to the JAB domain, there was always a JAB domain protein encoded by the immediately adjacent gene in the predicted operon (Table 1 rows 6a-6e and Figure 4). One group of proteins, typified by an E1-like protein fused to a JAB domain at the carboxyl-terminus, also contained an additional conserved amino-terminal domain, with a conserved histidine and cysteine (for example, Mdeg02000735 from Microbulbifer degradans, gi: 48864353; Table 1 row 6a and Figure 3). Iterative PSI-BLAST searches with the alignment of this domain as a seed recovered eukaryotic E2 (ubiquitin conjugating enzymes [UBC]) enzymes as hits with significant e values (e = 10^-3, iteration 3). The predicted secondary structure of these domains was congruent with that of eukaryotic E2 domains, with a four-strand β-meander and two flanking helices on either side 61. Furthermore, the conserved histidine and cysteine of the bacterial proteins also precisely matched the cognate active site residues of the eukaryotic E2 enzymes, suggesting that the amino-terminal domains of the bacterial domain are homologs of the E2 enzymes and likely to possess similar activity (Figure 6).

In addition, each set of these predicted operons contained a distinct group of genes that almost exclusively co-occurred with a particular operon type. Based on the different groups of co-occurring genes, we were able identify at least five major operon types (Table 1 rows 6a-6e and Figure 4). These groups of co-occurring genes encoded several conserved uncharacterized proteins, whose evolutionary relationships we systematically investigated using sequence profile searches, secondary structure prediction, and matches to libraries of profiles and HMMs for various previously characterized domains.

The first of these operon types exhibited a very simple organization, usually with two genes. One of them encoded the triple module protein, with amino-terminal E2-like and E1-like domains followed by a carboxyl-terminal JAB domain (Figure 3). The second gene in the operon encoded a specialized version of the metallo-β-lactamase domain (Table 1 row 6a and Figure 4). Another operon group typified by a conserved gene neighborhood from the Escherichia coli integrative and conjugative element (ICE) 62 and related mobile elements was found to contain a nucleotidyl transferase of the polymerase β-fold 63, in addition to the genes encoding the E1-like and JAB domain proteins (Table 1 row 6b and Figure 4). Like the E1-like proteins from the first group of conserved gene clusters the E1-like proteins of this group also show a fusion to an E2-related domain with a conserved active site cysteine (Figure 6). Similarly, a conserved operon group prototyped by a gene neighborhood from the megaplasmid NGR234 of Rhizobium sp. contains genes encoding two conserved uncharacterized proteins, one of which is predicted to contain a metal-binding domain based on the conserved pattern of two cysteines, a histidine, and an acidic residue (Table 1 row 6c and Figure 4). We observed that the E1-like proteins encoded by both of these operon types contained an additional amino-terminal domain with a conserved cysteine. Sequence searches with this amino-terminal region recovered the UBC-like E2 domains from a variety of eukaryotes. The best hit to these domains was from a profile of the E2-like proteins and included a match to the conserved cysteine (P < 10^-5 match for this cysteine containing motif in a Gibbs sampling search, with the MACAW program, including a wide range of known E2 domains). Secondary structure prediction for this conserved domain also showed complete congruence with the known structure of the E2 fold, suggesting that these amino-terminal domains fused to the E1-like enzymes are also homologs of the eukaryotic E2 ubiquitin conjugating enzymes (Figure 6).

A fourth operon type found in several diverse bacteria (Table 1 row 6d) typically contained three additional genes in the conserved gene neighborhood, in addition to the genes of the JAB domain and E1-like proteins (Figure 4). Furthermore, the JAB domain has an amino-terminal α + β domain that has a strictly conserved arginine and tryptophan residue (JAB-N; Figure 3). The first of these encodes a small protein with a highly conserved glycine at the carboxyl-terminus. Secondary structure prediction revealed that this small protein has a progression of structural elements identical to that seen in the β-grasp fold (Figure 2). The conservation pattern in this protein also strongly resembles that seen in the known β-grasp domains, and sequence-structure threading using the PHYRE program also recovered β-grasp proteins (for example, ThiS and PDB: 1tyg) as the best hits, suggesting that these are small standalone Ub-like proteins. The second protein encoded by this operon type was found to encode a largely α-helical protein with absolutely conserved charged and polar residues, suggesting that it might be an uncharacterized enzyme. The third conserved protein from these gene neighborhoods contained a conserved cysteine and gave significant hits to the profiles of the E2 Ub-conjugating enzymes, with the alignments spanning the conserved cysteine (Figure 6). This relationship was also supported by their predicted secondary structure and general conservation pattern. Although these proteins did not have the conserved histidine at the position often encountered in most E2 enzymes, they had an absolute conserved histidine further downstream (Figure 6). Mapping of the sequences of representatives of this family of proteins on the structures of E2 enzymes showed that this downstream histidine from the helix would be positioned very close to the active site histidine of the classical E2 enzymes (Figure 6). This would mean that these proteins are likely to effectively contain an active site similar to the classical E2 enzymes.

The fifth operon type is found sporadically in most proteobacterial lineages, cyanobacteria, and certain actinobacteria (Table 1 row 6e). Usually these gene neighborhoods contain two or three genes in addition to the central gene for an E1-like enzyme, which in most cases contains a JAB domain fused to the amino-terminus of the E1-like module. However, in a subset of bacteria the E1-like protein contains a fusion to an uncharacterized amino-terminal domain in place of the JAB domain (Figure 2). The conservation pattern of this domain is unrelated to that of the JAB domain, but it contains several conserved charged residues, making it tempting to speculate that it might perform a function analogous to the JAB domains. The other gene found in all gene neighborhoods of this type encodes a protein containing one to three repeats of an approximately 70-75 amino acid domain. The conservation pattern is similar to that seen in Ubls, and the predicted secondary structure of this domain exhibits a progression completely congruent to other β-grasp fold domains (Figure 2). Consistent with this, sequence-structure threading with the PHYRE program recovered the structures of the ThiS/MoaD proteins as the top hits (for example, PDB: 1tyg). These observations strongly suggest that this group of proteins is comprised of one or more Ub-like domains Table 1.

Furthermore, we noted that these predicted β-grasp domain proteins might also be fused with either of two unrelated carboxyl-terminal domains (Table 1). The first of these domains is a small domain of about 75 residues exhibiting a conservation pattern and secondary structure progression similar to the Ubls (Figure 2). These domains also recovered ThiS/MoaD as their best hits in sequence-structure threading with the PHYRE program, implying that it might form the third Ub-like domain in a subset of these proteins. The second carboxyl-terminal domain found in a mutually exclusive subset of these proteins also occasionally occurs as a standalone protein encoded by a separate gene sandwiched between the genes for the multi-β-grasp domain protein and the JAB + E1 domain proteins (Figure 3). Profile searches with an alignment of this domain recovered hits to the E2 enzymes and the eukaryotic RWD domain 6164, which contains a catalytically inactive version of the E2 fold as the best hits (e about 0.01-0.005). This relationship was also supported by the congruence of the predicted secondary structure of these domains with that of the E2 and RWD domains 61. Like the eukaryotic RWD domains, these bacterial domains also lacked the conserved cysteine residue, implying that they are likely to be catalytically inactive representatives of the E2-like fold (Figure 6). The above operon type was also seen to encode another conserved protein with a C-x(3)-C-x(35-38)-H-x(2)-C signature (Figure 4). The predicted secondary structure of this potential metal-binding signature is consistent with proteins containing a Zn finger domain, perhaps of the treble-clef fold.