Evolution by leaps: gene duplication in bacteria

When Charles Darwin wrote The Origin of Species, no data existed that could inform him of the molecular nature of genetic variation that fuels evolutionary change. Today the existence of sequences of entire genomes and the ability to compare related sequences allows identification and characterization of sources of genetic variation. Evolution at the molecular level is now known to have taken place through both selection and neutral drift acting on genetic variation arising from many avenues: single base changes, horizontal transfer of genes, loss of genes, rearrangements of genomic segments and, discussed here, gene duplication followed by divergence of the copies. The comparative analysis of sequences of related and unrelated bacteria has filled out our understanding of some of these mechanisms of evolution.

Views of the nature of genetic change underlying evolution have changed over the last century. Koonin has summarized the history of these changes up to the present view [1]. In the beginning, Darwin thought that genetic changes were small and evolution was gradual. This view was maintained as plausible after the structure of DNA became known. Successive single nucleotide changes by point mutation would be small, conforming to the view of the gradual nature of the process. Evolutionary change according to this gradualist view was brought about by selection, that is the fixation of beneficial mutations, elimination of the deleterious. Subsequently Kimura [2] and others introduced the neutral theory, stating that selectively neutral mutations dominate and fixation occurs by random drift. At this time, the type of genetic change was still viewed as gradual accumulation of point mutations.

However in 1970, Ohno [3] introduced the idea of gene duplication as an important form of genetic variation, a process that would go beyond gradualism and would permit quantum changes. The process of gene duplication in microbes as the agent of evolution of novel gene functions is being studied by many scientific groups today e.g. [4–7]. Another source of sudden change was the discovery of horizontal transfer of genes from one organism to another not necessarily related organism [8]. Both these mechanisms, gene duplication and lateral transfer, have the capacity to bring about relatively large changes.

With availability of complete genome sequences of many bacteria, studies have used such data to understand the power law behavior of sizes of paralogous groups of genes in many bacterial species [4]. Others have used collections of genomic sequence data to enumerate types of fates of ancestral genes, concluding that there has been a great deal of loss following duplication, that selection for novel functions has played a prominent role and that rates of divergence of paralogous genes depends on selection pressure and functional constraints [6]. Gevers et al. [7] analyzed presence of sequence-related groups from a functional standpoint. They found that in all the genomes, largest families contained transport genes and regulation genes, smaller families were involved in metabolism and energy production. They considered that duplicated genes were retained if adapted to a changing environment.

As distinguished from such studies of sequence-related families in large data sets like collections of whole genome sequences, we planned to examine a few paralogous groups in a limited number of bacteria where the great majority of the functions of the individual proteins in each family is known. We wanted to see what kind of impact expansion of a family by duplication and divergence has on the host cell. Different paths of divergence would be expected to create the differences one sees in the taxa today. As to what kinds of proteins to examine, we chose to look at enzymes even though they form smaller data sets than those for transport and regulation proteins. Our goal was not to reconstruct evolutionary events over time, but to look at the power of duplication to affect the identity of the cell in specific biochemical terms. We ask in qualitative terms if the content of a family of enzymes bears a relationship to the biological characteristics of the organisms in which they reside.

A companion study to this one from our laboratory, used MrBayes methodology to develop unrooted trees of the enzymes of this study [9]. These data show that the enzyme trees do not correspond to trees of the organisms, nor would we expect them to. Protein family trees are different from phylogenetic trees of organisms. The selection factors that operate on enzymes such as availability and concentration of cofactors, energy supply (e.g. ATP, NADH), interactions within metabolic pathways, response to regulatory chains, tolerance to inhibitors, to ion concentrations, the breadth of substrate accommodation, and so on and so on, need not connect quantitatively with the factors that affect phylogeny of the organism as a whole.

There have been few studies confined to enzymes as factors in molecular evolution. Jensen in 1976 pointed out the importance of "recruitment" of new enzymes in evolution by gene duplication followed by changes in the specificity of the new copies so as to take on a related, but new role [10]. Some relationships of enzymes within a pathway could be understood in these terms. Another mechanism is duplication and modification of one copy by addition of another domain. An example of such a relationship is the pair of genes in Escherichia coli for the ribose repressor (RbsR) and the periplasmic protein for ribose transport (RbsB). These proteins share the sequence spanning the periplasmic binding protein (PBP) domain (PF00352) but differ in the acquisition of a DNA-binding domain by RbsR. An alignment of RbsR and RbsB is shown in Figure 1. While both proteins have maintained their ability to bind ribose using the PBP domain, RbsR has gained the ability to bind DNA and regulate transcription while the RbsB has been modified to allow for export to the periplasmic space and for interaction with the membrane components of the ABC type transporter.

Different from the rbs story, there are families of sequence similar enzymes that use the same reaction mechanisms but vary in substrate specificity. An example is the family of aminotransferases Class III. However, perhaps even more interesting, there are other families of sequence-similar enzymes that catalyze related but different reactions. Such mechanistically diverse collections are called superfamilies of enzymes. Several enzyme superfamilies, isolated from many biological sources, have been studied carefully from a structural and biochemical point of view. These include the enolase, Nudix, amidohydrolase, crotonase and haloacid dehalogenase superfamilies (reviewed in [11]). We have focused on identifying the members of a superfamily within one organism, a group of enzymes that could have arisen by duplication and divergence. We ask whether the members of the family are of a kind that would contribute to the metabolic identity of the organism.

One such superfamily is the Short Chain Dehydrogenase-Reductase (SDR) family. Similarities among certain dehydrogenases from Streptomyces spp., Drosophila melanogaster and several mammals, led to the identification of a type of dehydrogenase given the name SDR [12]. All reactions catalyzed by members of this superfamily require the cofactor NAD(P)/H and all members possess the Rossman fold. As more and more members of this superfamily were identified, the family was found also to include epimerases, dehydratases and isomerases [13]. It is variations on a theme of reaction chemistry that ties members of the superfamily together. This is different from earlier ideas on evolution of enzymes where a single enzyme would change by modifying substrate affinities, not by varying the reaction.

In the context of evolution, one can ask what kinds of biochemical properties have been conferred on a single organism by this process. To answer the question we decided to gather the members of the SDR family in E. coli, and then expand the study to other sequence-related enzyme families, not only from E. coli but from other bacteria as well.

To find out how many members of the SDR family are present in E. coli K-12 MG1655, henceforth E. coli, we assembled enzymes identified with an EC number 1.1.1.x. Among these are enzymes with the structural and sequence characteristics of the SDR superfamily. Initially we used the AllAllDb program of the Darwin system [14] (after first separating independent, fused proteins into their components) to collect all sequence related E. coli enzymes from this group. Parameters of the initial pair-wise similarity search were set as requiring a Pam value of at least 200, an alignment of 83 residues and an involvement of at least 50% of the length of the smaller protein of any sequence-similar pair. Related enzymes were assembled by transitive relationship. To extend membership in the groups to include proteins whose sequence may have diverged further, we submitted all members to PSI-BLAST analysis [15].

Examination of the sequence alignments of the E. coli SDR enzymes revealed four regions that aligned for all members of the extended family, the substrate binding site, the NAD(P)/H-binding Rossman fold and two sites of unknown function, likely to be important for folding (Fig. 2). Each of the conserved sequences occurs in approximately the same region within each protein. Small changes in the residues in conserved regions have large effects on the affinity for particular substrates and on the specific reaction that is catalyzed.

Table 1 shows the separation into two types of crotonases and the variety of pathways and resulting phenotypes served by the SDR superfamily. Some pathways are used by many organisms, such as fatty acid synthesis, but many products and processes are characteristic of the enteric organisms only, such as bile acid emulsification, biosynthesis of colanic acid, lipid A, enterobactin and enterobacterial common antigen. It appears that the process of duplication and divergence has contributed to the metabolic characteristics of a unique phylogenetic group of bacteria.

One can ask how broad the phenomenon of families is among E. coli enzymes. Even before the sequence of the E. coli genome was completed, the existence of families of related sequence within its genome was observed [17, 18]. Such sequence-related families are viewed as paralogous families that arose by duplication of genes within the genome of the organism itself or in that of an ancestor, although as previously mentioned some members of these families could have been introduced by lateral gene transfer. After completion of the full genomic sequence of E. coli [19], the complete set of paralogous families in relation to the whole genome could be determined. Pair-wise related sequences from the entire genome were assembled, using the criteria of similarity as having Pam values below 200 and alignments of at least 83 residues. By requiring an alignment of 83 amino acids or more we seek to avoid grouping sequences by small common domains or motifs, such as DNA binding domains, instead we detect protein level duplications. For example in the RbsR/RbsD case, the 45 amino acid DNA-binding domain (PF00356) is present in 14 additional E. coli transcriptional regulators. Since the main components of these proteins, the ligand-binding domains, not are related to RbsR we do not consider them paralogs. Our groups ranged in size from 92 members in the largest group down to the smallest size, simple pairs. Over half of the E. coli proteins resided in these sequence-related groups [20–22].

The existence of families of sequence-similar proteins making up a large fraction of the genomic content supports the proposal that duplication followed by divergence is an important mechanism of molecular evolution. The largest groups in the E. coli genome were those of related transport proteins, regulatory proteins, and redox (i.e. iron-sulfur) subunits of enzyme complexes. Groups of sequence similar enzymes were smaller, had fewer members, than the groups of transporters and regulators. However, we concentrated on the class of enzymes because studying families of enzymes has the advantage of being able to draw on the detailed knowledge in the extensive biochemical literature concerning their properties, prosthetic groups, the mechanisms of the reactions they catalyze and pathways they belong to. One is in a position to link genetic information with biochemical information and thus with phenotypes of the organism. Examining the members of enzyme families of E. coli allowed a view at the molecular level of what kind of creation of function occurred as a consequence of presumed duplication and divergence.

Another superfamily that is structurally and mechanistically related but catalyzes diverse reactions is the crotonase family. The family was originally characterized by similarities in three-dimensional structure of four enzymes derived from different sources. Although structurally related, sequence related and mechanistically related, their biochemistry showed that they catalyzed four different reactions [23]. Subsequent investigation has shown that the crotonase enzymes are related in sequence, though often distantly, and catalyze a broad range of reactions i.e. dehalogenation, hydration/dehydration, decarboxylation, formation/cleavage of carbon-carbon bonds and hydrolysis of thioesters [24].

To look at crotonases in an evolutionary context, one can ask if they could have arisen by duplication and divergence. To approach this question, one could enumerate all crotonases in one organism. Starting with a crotonase in E. coli, encoded in the N-terminal portion of FadB (here designated FadB_1) with demonstrable structural similarity at the active site to the rat liver crotonase, we assembled the group of sequence-similar enzymes in E. coli as before by the Darwin AllAllDb program. Figure 3 presents the alignment of residues at the active site for the E.coli crotonase family. The greatest amino acid conservation is seen for the residues involved in acyl-CoA-binding and the catalytic site. There is a CoA-binding site and an expandable acyl-binding pocket as well as an oxyanion hole for binding the thioester C = O bond, crucial to the reaction catalyzed by members of this superfamily [23, 25]. Variations in residues at critical positions in the active sites dictate which of the related reactions occurs. Again, as for the SDR family, one can visualize that the broad family of crotonases, spanning several kinds of reactions, could have arisen by gene duplication and divergence early in evolutionary time.

By assembling the crotonase family members in a few organisms, one expects that some individual enzymes will be present in all the organisms as they are virtually universal. However other members of the crotonase family are expected to differ from one organism to another. We expect that bacteria in separate lineages would have some enzymes that catalyze different reactions. Differentiation of bacteria as they evolved along different lineages is expected to be partly as a consequence of generating different enzyme family members in the course of the divergence process. Other molecular evolution events are occurring at the same time as the duplication and divergence, such as lateral transfers and gene loss. To focus on gene duplication we decided to look at families of enzymes in a set of both similar and distant bacteria.

We asked whether members of three enzyme families are the same in the bacteria examined or whether there are differences dictated by separate evolutionary histories and separate selective pressures. Three enzyme families were compared in four bacteria. The families chosen for comparison were the crotonases, pyridoxal phosphate-requiring aminotransferases Class III, and thiamin diphosphate-requiring decarboxylases. The four bacteria are E. coli, Salmonella enterica subsp. enterica serovar Typhimurium LT2 (henceforth S. enterica), the distant γ-proteobacterium Pseudomonas aeruginosa PAO1 and the gram positive bacterium Bacillus subtilis subsp. subtilis strain 168 (henceforth "B. subtilis).

We note that some of the enzymes are present in all four bacteria, suggesting they are integral parts of core metabolic functions. This is supported by the pathways they participate in; biotin synthesis and porphyrin synthesis (BioA and HemL), aminobutyrate utilization (GabT), pyruvate oxidation (PoxB/YdaP), and fatty acid oxidation (FadB). One supposes such commonly held important functions are conserved in many bacteria in many taxa.

Other enzymes differ in their distribution (presence or absence) among the four organisms. This is presumably a result of different evolutionary histories in different lineages during the divergence processes, leading to establishment of bacterial taxa with biochemical and metabolic differences. For example the MenD decarboxylase and MenB crotonase used for menaquinone biosynthesis are absent from P. aeruginosa and present in the other three organisms. This distribution is reflective of the Pseudomonads using only ubiquinone, and not both ubiquinone and menaquinone, as electron carriers for respiration. Gcl, tartronate-semialdehyde synthase of glyoxalate utilization, is present in three bacteria, and not in B. subtilis. Degradation of glyxolate in B. subtilis has been shown to occur by a different pathway from the other three organisms. In the two enteric organisms, their particular paths of metabolizing putrescine and carnitine are reflected in the presence of putrescine aminotransferase (PatA) and carnityl-CoA dehydratase (CaiD) in both E. coli and S. enterica.

Several of the aminotransferases are involved in arginine metabolism, and the occurrences of these enzymes also vary among the organisms. E. coli and its close relative S. enterica both have ArgD and AstC for biosynthesis and degradation of arginine, respectively. AruC is used by P. aeruginosa for both arginine synthesis and degradation. While in B. subtilis, ArgD is used for arginine synthesis and RocD, another member of the aminotransferase family, is used to degrade arginine by a different pathway. We observe that the two more closely related enteric organisms have a higher similarity in their aminotransferase content.

Some of the protein family members represent isozymes, sequence similar enzymes that catalyze the same reaction but with definable differences such as substrate breadth, feedback inhibition, binding constants, reaction rates and the like. Based on the common nature of the isozymes, we suppose they have arisen by gene duplication and slight divergence. Examples of isozymes are the trio of acetolactate synthases; IlvB, IlvI and IlvG, found in E. coli and S. enterica. These isozymes function in the isoleucine and valine biosynthesis pathway, each responding to distinct feed back. One copy, IlvG, is mutated and inactive in E. coli, rendering E. coli valine sensitive. This phenotype is used in identification protocols to distinguish E. coli and S. enterica. A second type of acetolactate synthase (AlsS) is also present in B. subtilis, but this enzyme is used exclusively for catabolism and not synthesis of isoleucine and valine.

E. coli and S. enterica have another set of isozymes, FadB and FadJ. Both enzymes are used for fatty acid oxidation, but FadB is used under aerobic conditions and FadJ is used under anaerobic conditions. Other isozymes are GabT and PuuE in E. coli, GsaB and HemL in B. subtilis. Isozymes are often specific to pathways, such as PuuE, which is specific to putrescine utilization. One supposes that simply by small changes in duplicate genes, pathway content and biochemical capability of an organism can expand.

In addition there are protein family members that are unique to only one of the four organisms and absent in the other three. These enzymes often confer metabolic properties unique to their host. An example is oxalyl-CoA decarboxylase (Oxc) that is present E. coli where it is believed to confer oxalate degrading capabilities. As is the case for any of the enzymes present in one organism, not the others, the gene could have been acquired by lateral transmission [26]. However when an enzyme like oxalyl-CoA decarboxylase, is found in many bacteria, it is at least as possible that it arose by gene duplication and divergence. Other organism specific enzymes, in this case B. subtilis, include the IolD for myo-inositol degradation and the crotonases PksH and PksI used for polyketide synthesis. Polyketides are a group of secondary products peculiar to the Bacilli. Other unique B. subtilis enzymes AlsS, GsaB and RocD have been mentioned above. It seems evident that formation of different enzymes by unique divergence events, add up to creation of taxa with different metabolic characteristics.

P. aeruginosa has the largest number of unique, or organism specific, enzymes in our dataset. This is shown for all three enzyme families (Tables 2, 3, 4). These Pseudomonas specific enzymes include synthesis of the siderophore pyoverdine (PvdH), and utilization of mandelate (MdlC), leucine and isovalerate (LiuC) and acyclic terpenes (AtuE). Other predicted family members include two aminotransferases: PA5313, evidently an isozyme for 4-aminobutyrate, and OapT, likely a beta-alanine:pyruvate enzyme. Each of these enzymes contributes to the distinct metabolic character of P. aeruginosa as a pseudomonad. In addition there are 5 aminotransferases, 5 decraboxylases and 14 crotonases whose functions remain unknown in P. aeruginosa. Our phylogenetic analysis [9] suggests that these are unique enzymes representing additional functions yet to be discovered. Combining genes of known and unknown function for the three families, the number of unique P. aeruginosa genes (33) far surpasses that of B. subtilis (12), E. coli (2) and S. enterica (1). The large number of Pseudomonas specific enzymes detected is in agreement with the well-documented metabolic versatility of this group [27, 28].

These examples of differences among enzyme families in four organisms suggest that the distinct events of divergence in genes of protein families over time have generated taxa of bacteria that are distinguished in part by their metabolic differences. Bacteria that are closely related have fewer differences in these families. For all three enzyme families we noted that the two most closely related organisms, E. coli and S. enterica, contain the most similar complement of enzymes. Larger differences in both number of dissimilar enzymes and enzyme functions were seen when comparing either B. subtilis or P. aeruginosa to any of the other three.

Overall, our protein family analysis includes several examples of how the functional and metabolic diversity of today's organisms is reflected in a history of duplicated and diverged gene copies in their genome sequences. In some instances the gene copies are the same in all the bacteria. These are enzymes for universal functions. Some of the gene copies did not undergo much divergence and resulted in isozymes catalyzing the same reactions but with different properties. Such enzymes usually contribute to phenotypic differences, for instance by changes in substrate specificity or regulation. Still other gene copies were not found in other bacteria. These were functions characteristic of the phenotype of the particular organism. We do not suggest that duplication of genes was the only source of diversity in these organisms. In addition there lateral transfer could have introduced a new function and also gene losses would have changed the composition of protein families. Some analyses suggest that lateral gene transfer has played a large role in assembling gene families [29]. However one needs to take into account the lack of congruence between organism trees and gene trees, the latter being affected by different selective pressures on individual enzymes (such as gene family composition, cofactor/substrate availablility) compared to those affecting the organism as a whole. Lawrence and Hendrickson [30] have discussed in a thoughtful way the difficulties in distinguishing horizontal transmission from duplication of existing genes. We have therefore not attempted to identify laterally transferred genes in our enzyme families. While possibly there we do not expect them to predominate. In summary, it is a combination of all these genetic changes (duplications, divergence, loss and acquisitions) in ancestors of contemporary organisms that has generated the characteristic phenotypes of today's organisms.

By assembling selected superfamilies of enzymes of sequence and structural similarity in four different bacteria whose entire genomes have been sequenced, we suggest that members of the families arose in the course of evolution at least in large part, by duplication followed by divergence. We observed that differences in the enzyme families, both in functions and numbers of homologs, were greater as the organisms were less closely related. Functional differences of family members were reflective of metabolic diversity of the host genome. Events such as gene loss and gain also must have made changes to enzyme family rosters over time, but we suggest that the outline of the duplication and divergence process remains visible in the contemporary paralogous groups of sequence-related superfamilies.

Pairwise sequence alignments and scores were generated using the AllAllDb program of Darwin (Data Analysis and Retrieval With Indexed Nucleotide/peptide sequence package), version 2.0, developed at the ETHZ in Zurich [14]. Maximum likelihood alignments are generated with an initial global alignment by dynamic programming followed by dynamic local alignments. A single scoring matrix is used for these steps. After the initial alignment, the scoring matrix is adjusted to fit the approximate distance between each protein pair to produce the minimum Pam value. Pam units are defined as the numbers of point mutations per 100 residues [35, 36].

Referee 1:

Pierre Pontarotti

Directeur de Recherche CNRS

Marseilles, France