Diversity and distribution of transcription factors: their partner domains play an important role in regulatory plasticity in bacteria

The proportion of wHTHs contributes significantly to the total repertoire of TFs in bacteria

In order to gain insights into the commonalities and differences in gene regulation between bacterial species from the perspective of TFs, we compared the repertoires of TFs identified in 428 nonredundant bacterial sequenced genomes by using diverse bioinformatics tools. From this analysis, we found that the wHTH comprises 48 % of the total TFs identified in bacteria, being the most abundant superfamily of DNA-binding structures described so far in this cellular domain (Table 1 and Supplementary Table S1). Indeed, alternative DNA-binding structures have been identified in minor proportions, such as the lambda repressor, homeodomain-like domain and C-terminal effector domain. This result not only correlates with the fact that the DBDs are associated with TFs, and in particular wHTHs are among the most ancient domains, probably derived from a relatively small set of folds (Aravind & Koonin, 1999; Madan Babu & Teichmann, 2003; Pérez-Rueda & Collado-Vides, 2001), but also shows that the wHTHs have been highly successful domains in nature.

Based on this apparent overrepresentation of wHTH proteins in the repertoire of TFs, we evaluated their abundance in the context of genome size. In this regard, it was previously described that TFs follow a quadratic distribution, i.e. large genomes contain a high proportion of TFs and, vice versa, small genomes have a small repertoire of TFs (Pérez-Rueda et al., 2004, 2009). Thus, we asked whether wHTH abundance correlates with the number of open reading frames. From this analysis, we found that wHTHs follow a similar distribution to the total repertoire of TFs, suggesting that this superfamily contributes significantly to this distribution (see Fig. 1). Indeed, the wHTH corresponds to around 50 % of the total of TFs per genome and in some organisms more than 70 % of the total repertoire of TFs corresponds to wHTH proteins, such as in Mycoplasma genitalium, Prochlorococcus marinus NATL1A, Bordetella bronchiseptica RB50 and Thermosynechococcus elongatus BP-1. Based on these results, we consider that the abundance of TFs and wHTHs may be associated with organismal diversity, i.e. organisms with free-living lifestyles and a large genome size contain a major proportion of transcriptional regulators, compared with parasitic or symbiont organisms with their small genome sizes, as described previously (Pérez-Rueda et al., 2009). For instance, diverse free-living bacteria, such as Burkholderia sp. 383, with its large genome size, contain a high proportion of wHTH TFs. Recent studies suggest that organisms with free lifestyles require a large proportion of regulatory proteins as a consequence of an increase in the genome size, which also increases the number of putative regulatory interactions (Croft et al., 2003). Alternatively, organisms with smaller genome sizes, such as Mycoplasma species, contain a small proportion of wHTH TFs, between one and six wHTH TFs. This proportion of TFs correlates with the fact that symbionts and/or parasitic obligate bacteria have substantially reduced genome sizes. Indeed, in some organisms with a substantial reduction in the gene repertoire, we were not able to identify TFs or wHTH TFs, such as in the Buchnera, Rickettsia and Wolbachia genera, suggesting that they exhibit alternative regulatory mechanisms beyond TFs. Although the wHTH contributes significantly to the total TFs, probably following a similar path of duplication events to the rest of the genes in Bacteria (as illustrated in Fig. 1), we identified organisms such as Bordetella pertussis and ‘Candidatus Protochlamydia amoebophila’ in which the wHTH does not represent the most abundant superfamily. In fact, in these organisms the homeodomain-like superfamily is the most abundant DBD associated with TFs, suggesting alternative means for regulation of gene expression.

Another important question that emerges from the distribution of wHTHs is whether the abundance of these kinds of proteins in all the genomes also reflects a large proportion of families. In this sense, we found that the abundance of wHTH proteins is derived preferentially by duplication events more than by the existence of different families. Therefore, we classified the repertoire of wHTH TFs into 39 different families, and their distribution among all bacteria was evaluated (Supplementary Table S2, available with the online version of this paper). From this analysis, we describe in the next two sections the most evident results.

(i) We found that the abundance of TFs in larger genomes does not necessarily involve diversity in the repertoire of families, but it does suggest an increase in the size of the family, i.e. whereas there is a large proportion of TFs as a consequence of genome size, the number of different families per genome is almost constant, from 1 to 26, with an average of 14 families in bacteria. However, in some cases we identified a large number of families, such as for the bacterium Saccharopolyspora erythreaea, a filamentous soil microbe used for industrial-scale production of the antibiotic erythromycin (Oliynyk et al., 2007), for which 26 different wHTH families were identified, the maximum number of families per organism. A probable identification of abundant families might suggest diverse duplication events in Bacteria, whereas small families would suggest gene loss, lateral gene transfer or invention de novo (Supplementary Table S1).

(ii) Seven families include 80 % of the total TFs with wHTHs: ArsR, MarR, LysR, biotin, CAP, GntR and Lrp (Fig. 2). From these families, LysR represents an interesting evolutionary group, because it contains the most abundant family of TFs identified so far as well as a large proportion of proteins with dual activity (repressor and activator proteins). This family is mainly responsible for the regulation of basic and ancient physiological processes, such as amino acid biosynthesis, associated with the last common ancestor of Bacteria, Archaea and Eukarya (Hernández-Montes et al., 2008). In contrast, small families were also identified, such as iron, ArgR and LexA, proposed to be essential under standard growth conditions and in maintaining DNA integrity in E. coli. Based on these data, a family’s abundance not only suggests ancient evolutionary events in Bacteria but also reveals a limit in the number of wHTH families in all bacteria. Thus, it seems that there is a limit of expansion for all families in bacteria, independent of the genome size and an increase in the number of duplication events associated with each family, suggesting that the TF repertoire in bacteria is associated mainly with events of duplication, recombination and lateral gene transfer. An interesting observation is that in cell division in Chlamydiae, during which massive gene loss events have been identified, the HrcA family of TFs was exclusively identified. HrcA is a small family that contains a large proportion of negative regulators of class I heat-shock genes (grpE–dnaK–dnaJ and groELS operons) that prevent heat-shock induction of these operons (Fischer et al., 2002). We suggest that proteins of this family may be associated with alternative functions beyond heat-shock induction, playing an important role in the adaptation of parasitic bacteria to their hosts.

A universal pattern of distribution is observed in bacterial wHTH families

In order to evaluate the distribution of the 39 TF families in bacteria, we analysed their distribution across 24 taxonomical divisions (Fig. 2; see also Supplementary Tables S1 and S2). The relative abundance of families was calculated by phylum, with a value of 0 representing absence and 1 representing presence. Based on this analysis we identified a cluster of eight families widely distributed in all the bacterial divisions: the LexA, LysR, FUR, ArsR, MarR, CAP, HrcA and Rrf2 families. An interesting finding was that the seven most abundant families described previously were included in this cluster, together with families that are less abundant. All these families regulate a plethora of functions, such as amino acid metabolism (LysR), carbon source assimilation (CAP, Körner et al., 2003; Maddocks & Oyston, 2008), resistance to diverse stresses (MarR, Ellison & Miller, 2006; LexA, Shimoni et al., 2009; HrcA, Ellison & Miller, 2006), cysteine biosynthesis and benzoate degradation (Rrf2, Even et al., 2006; Peres & Harwood, 2006) and metal assimilation (FUR, Pennella & Giedroc, 2005) and resistance (ArsR, Busenlehner et al., 2003). It is probable that all these families could have been present in the last common ancestor of Bacteria and as a consequence the bacterial cenancestor would have a high capability to contend with diverse, challenging environments and also be a self-supporting system.

In a second cluster, families with a large distribution pattern, except in small organisms, were identified: GntR, Lrp, biotin and IclR families. These families regulate biotin synthesis, carbon source assimilation and amino acid biosynthesis, among other processes.

A third cluster with a low distribution pattern was identified and included the ArgR, Iron and Rex families. These families regulate genes associated with carbon source assimilation, arginine biosynthesis, iron uptake and responses to changes in the cellular NADH/NAD⁺ redox state, respectively, and include few members per organism.

Finally, a high diversity of families with erratic distribution patterns were included in diverse clusters, such as Rio2, associated with Bacillales, Ets, associated with Actinobacteria, and MetH, exclusively associated with Actinobacteria. In addition, we identified probable families with lateral gene transfer events, including Vacu, which is associated with Bacillales and Actinobacteria. This finding suggests that diverse lineage-specific TFs are involved in specific and important processes, such as sporulation in bacilli, or in some specific amino acid biosynthesis routes. It is interesting that the absence of TFs for several important amino acid biosynthesis routes in B. subtilis and other Firmicutes may have been complemented by the invention of novel regulatory mechanisms, such as transcription attenuation (Gollnick et al., 2005; Merino & Yanofsky, 2005; Rodionov et al., 2004). Indeed, a large diversity of regulatory mechanisms beyond TFs was recently described, including antisigma factors, RNAs and protein–protein interactions, among others (Martínez-Núñez et al., 2010).

Multidomain proteins are highly abundant in bacterial TFs

At the present time, the considerable diversity of sequenced bacterial genomes available to the scientific community offers an invaluable source of information for evaluating the abundance and diversity of the repertoire of regulatory proteins controlling gene expression at the level of transcription initiation. These proteins can be analysed to evaluate their influence on bacterial adaptation and responses to environmental stimuli. Therefore, in order to evaluate the contributions of these domains in the total set of proteins identified as TFs, the domain repertoire beyond the DBD was analysed. From this analysis we identified different groups based on domain architectures, i.e. 57 % of the TFs exhibited more than one structural domain (the DBD and PaDos), whereas 43 % of the total repertoire was associated only with the DBD (Table 2 and Supplementary Table S3, available with the online version of this paper). The monodomain proteins can be further subdivided into two categories: the first one includes proteins for which more than 94 % of the protein is occupied by the DBD and the second category includes proteins for which the DBD covers only 50 % of the sequence. These latter proteins may exhibit additional domains not identified using structural data. Therefore, in Fig. 3 we present the distribution of multidomain TFs in all the bacterial genomes. From this illustration, it is evident that the abundance of multidomain TFs follows a similar distribution to total TFs, i.e. their abundance correlates with the genome size. For instance, organisms with small genomes may contain a lower proportion of multidomain proteins than do larger genomes. This result reinforces the notion that small genomes are associated with stable environments, where a limited number of TFs are necessary to regulate gene expression. In contrast, larger genomes contain a great proportion of multidomain TFs, suggesting that these domains contribute to functional adaptations to environmental changes. Based on this analysis, the diversity and abundance of TF families and their PaDos would contribute significantly to the regulatory diversity.

Therefore, 79 PaDos were identified in the whole collection of wHTH TFs associated with bacteria, and in order to evaluate the distribution of these PaDos in bacteria, 22 divisions were analysed. From this analysis we identified diverse groups. The most representative clusters are shown in Fig. 4 (see also Supplementary Figs S1 and S2, available with the online version of this paper). The first group contains four different PaDos, such as periplasmic-binding protein-like II (PBP II), cAMP-binding domain-like, GAF domain-like and LexA/signal peptidase domains. PBP II and cAMP domains are associated with the LysR and CAP families, whereas GAF is associated with diverse families, such as as IclR, HrcA, Plan, FUR and others. The LexA/signal peptidase domain is associated with the LexA family. It is important to mention that the first three PaDos are highly abundant in all bacteria, representing 70 % of the total PaDos, and they are also related to large TF families. These findings suggest that these PaDos are very successful in all the bacteria and are intimately related to large families of TFs.

The second cluster includes eight different PaDos: dimeric α- and β-barrel (dimeric), PLP-dependent transferases (PLP), NagB/RpiA/CoA transferase-like (NagB), C-terminal domain of transcriptional repressors, C-terminal domain of arginine repressor, class II aaRS and biotin synthetases, iron, and NAD(P)-binding Rossmann fold domains. From these domains, dimeric, PLP and NagB have been identified as highly abundant, representing 15 % of total PaDos. An interesting observation for these domains is that they exhibit a similar distribution pattern to the PaDos included in the first cluster, except they are absent in parasites, symbionts and, in general, in small genomes, suggesting probable gene loss events. The third cluster is integrated by MOP-like, S-adenosyl-l-methionine-dependent methyltransferases and acyl-CoA N-acyltransferases (Nat), which have been mainly identified in the divisions Proteobacteria and Acidobacteria. In the subsequent clusters, we identified diverse PaDos, including the PRTase-like, ribokinase-like and CBS domains and the putative transcriptional regulator TM1602, C-terminal domain, constrained to Firmicutes and Fusobacteria. Finally, we found the rhodanese/cell cycle control phosphatase, fatty acid-responsive transcription factor FadR C-terminal, SIS, phosphorytosine, Bet v1-like, and nucleotidyltransferase domains clustered together. These domains have been identified in only low proportions, mainly in Proteobacteria.

In summary, we identified six PaDos that represent 84 % of the total domains identified in bacterial TFs (PBP II, GAF, dimeric, PLP, NagB and cAMP domains), many of which are universally distributed in bacteria and intimately associated with specific families. It is also interesting that these PaDos are found in abundant families, reinforcing their probable roles in basic physiological proceses, such as PBP II being exclusively associated with the largest family of TFs identified so far, LysR (see Supplementary Fig. S2). This finding suggests that PaDos and wHTH domains probably coevolved, based on their pattern distributions. Alternatively, we identified some PaDos as probably lineage specific, such as the PTS-reg, TM1602, PTS-sys, PRTase-like, CBS, HPr kN-T and Thio/thiol domains. These latter domains were exclusively identified in Firmicutes. We suggest that these PaDos may be the consequence of invention de novo, because of their specific distribution in Firmicutes. In addition, to reinforce these previous observations, we asked if the PaDos were specifically associated with wHTH TFs. To obtain further insights into the specific and general associations of these domains and wHTHs, we evaluated all DNA-binding structures reported so far and classified them as homeodomain-like, lambda repressor or nucleic acid-binding domains (Table 1) in order to identify the distributions of these domains. Based on this analysis, we determined that almost 40 % of the total PaDos associated with wHTHs are exclusive to this class of structural domains, suggesting that these domains have been preferentially recruited by wHTHs. A similar finding has been identified in other superfamilies, for instance, 30 % of the total PaDos are lamba specific, and 33 % of the total PaDos are homeodomain-like specific (see Supplementary Fig. S3, available with the online version of this paper).

The diversity of PaDos defines three main groups of families

In order to evaluate the diversity of TFs in terms of their structural domains, we defined three main groups of families: (i) monodomain families, those families that exhibit only the DBD; (ii) monolithic families, in which most of the protein members exhibit the DBD and a PaDo, usually in the same domain; and (iii) promiscuous families, those families with a large diversity of domains. We next describe each of these categories in more detail.

Conclusions

What are the roles of the diversity and distribution of TFs in the regulatory plasticity of bacteria? The answer to this question is important in order to determine the contribution of these regulatory families in the evolution of these organisms and in the context of their responses to diverse environmental stimuli. Thus, based on the repertoire of DNA-binding TFs associated with the wHTH domains in 668 completely sequenced bacterial genomes, representing adaptative designs for different lifestyles, we have attempted to gain an understanding of the relationships between the DBD and PaDo distribution patterns involved in the modelling of transcriptional regulatory networks.

We have shown that TF families expand or contract in a lineage-specific manner to adapt to the varied environmental needs of the organisms. Similar trends have been observed in previous comparative studies on TF families in plants versus animals and at the level of taxa (Coulson et al., 2001). A more general perspective regarding lineage-specific expansion of protein families and the implications for the diversification of organisms has also been described for eukaryotic species (Lespinet et al., 2002). In this regard, the abundance levels of families and their domain organizations were evaluated in bacteria. From this analysis, we identified that TF families have preferentially suffered diverse expansion events more than invention de novo, i.e. in all organisms evaluated in our study there was a similar distribution of families, but the members were different among the families. Alternatively, PaDo families are present at between 1 and 26 per organism, contributing significantly to the diversity of the regulatory machinery. Finally, three main categories of families were defined on the basis of their domain architecture: monodomain, monolithic and promiscuous. We suggest that the interplay of all these elements, TF abundance, recombination of DBDs, and PaDos, and also duplication events, allows bacteria to adapt to changing environmental conditions and shows that they are models of regulatory networks.