The Ruegeria pomeroyi acuI Gene Has a Role in DMSP Catabolism and Resembles yhdH of E. coli and Other Bacteria in Conferring Resistance to Acrylate

The Escherichia coli YhdH polypeptide is in the MDR012 sub-group of medium chain reductase/dehydrogenases, but its biological function was unknown and no phenotypes of YhdH− mutants had been described. We found that an E. coli strain with an insertional mutation in yhdH was hyper-sensitive to inhibitory effects of acrylate, and, to a lesser extent, to those of 3-hydroxypropionate. Close homologues of YhdH occur in many Bacterial taxa and at least two animals. The acrylate sensitivity of YhdH− mutants was corrected by the corresponding, cloned homologues from several bacteria. One such homologue is acuI, which has a role in acrylate degradation in marine bacteria that catabolise dimethylsulfoniopropionate (DMSP) an abundant anti-stress compound made by marine phytoplankton. The acuI genes of such bacteria are often linked to ddd genes that encode enzymes that cleave DMSP into acrylate plus dimethyl sulfide (DMS), even though these are in different polypeptide families, in unrelated bacteria. Furthermore, most strains of Roseobacters, a clade of abundant marine bacteria, cleave DMSP into acrylate plus DMS, and can also demethylate it, using DMSP demethylase. In most Roseobacters, the corresponding gene, dmdA, lies immediately upstream of acuI and in the model Roseobacter strain Ruegeria pomeroyi DSS-3, dmdA-acuI were co-regulated in response to the co-inducer, acrylate. These observations, together with findings by others that AcuI has acryloyl-CoA reductase activity, lead us to suggest that YdhH/AcuI enzymes protect cells against damaging effects of intracellular acryloyl-CoA, formed endogenously, and/or via catabolising exogenous acrylate. To provide “added protection” for bacteria that form acrylate from DMSP, acuI was recruited into clusters of genes involved in this conversion and, in the case of acuI and dmdA in the Roseobacters, their co-expression may underpin an interaction between the two routes of DMSP catabolism, whereby the acrylate product of DMSP lyases is a co-inducer for the demethylation pathway.


Introduction
The compatible solute dimethylsulfoniopropionate (DMSP) is made by many marine phytoplankton, including dinoflagellates, diatoms, and coccolithophores, several marine macroalgal seaweeds and a few land plants [1]. This zwitterion probably acts as an osmolyte, although other anti-stress functions have been suggested [2]. Its high concentrations in the producing organisms (,0.5 M in some cases) coupled to their wide distributions in the oceans, means that DMSP is one of the most abundant (,10 9 tons made per annum) organic, sulfur-containing molecules on Earth.
Furthermore, the DMSP that is released following the death or damage of the producing organisms provides an important nutrient source for many marine microbes, including the prodigiously abundant SAR11 clade, and it is therefore a key component of the global sulfur cycle [3,4].
The catabolic fate of DMSP is complex, and varied. Some of the eukaryotic DMSP producers can themselves cleave it into acrylate plus the volatile dimethyl sulfide (DMS), as can many marine bacteria, and a few fungi (see [5]). Although the enzyme activity for this reaction is generically termed ''DMSP lyase'', biochemical and genetic studies show that widely divergent enzymes, which occur in different microbes, can cleave DMSP into these products. To date, no less than five ''Ddd'' (DMSPdependent DMS) gene products (DddL, DddP, DddQ, DddW and DddY) have been found in different marine bacteria, and although they all catalyse the same cleavage of DMSP into acrylate plus DMS, they are in different polypeptide families [6,7]. A sixth bacterial ''Ddd'' enzyme, termed DddD, also generates DMS by cleaving DMSP, but in this case, the resultant C3 compound is 3hydroxypropionate (3HP) [8].
In addition to these cleavage pathways, DMSP can also be catabolised in a completely different manner, in which the initial step involves demethylation to methylmercaptopropionate (MMPA), which is then further catabolised to methane thiol and acetate [9,10]. The gene responsible for the initial demethylation is termed dmdA, which occurs in two groups of abundant marine a-Proteobacteria, namely a lineage known as the Roseobacters and also in the hugely populous SAR11 clade; thus, demethylation accounts for most of the global DMSP catabolic flux, although it does not liberate any DMS (see [7]).
In several bacteria, the ''primary'' ddd genes, whose lyase products act directly on DMSP, are in clusters, together with other genes that are variously involved in the import of DMSP, in downstream catabolic steps that feed into central metabolism, or in their transcriptional regulation in response to the appropriate coinducer molecule (see Figure 1 for some examples of these).
In addition, several different ddd gene clusters, in a range of bacteria, contain another gene, termed acuI (acrylate utilisation). First described in the a-Proteobacterium, Rhodobacter sphaeroides strain 2.4.1 [11], acuI is the central gene of a three-gene operon whose promoter-proximal acuR encodes a regulator in the TetR family, and whose promoter-distal gene, dddL, encodes a DMSP lyase that cleaves DMSP into acrylate plus DMS ( [12], Figure 1). Although R. sphaeroides does not grow well on acrylate as sole carbon source, it can catabolise 14 C-labelled acrylate substrate, with the concomitant release of labelled 14 CO 2 . An AcuI 2 mutant of R. sphaeroides was less effective in this transformation, and, strikingly, it was hypersensitive to the inhibitory effects of acrylate compared to the wild type [11]. Significantly, acuI is not only linked to other ddd genes that encode different types of DMSP lyases, but is also next to the DMSP demethylation gene dmdA in the model Roseobacter strain, Ruegeria pomeroyi [6,13].
AcuI is in the ''medium chain reductase/dehydrogenase'' (MDR) family, members of which are widespread and occur in all Domains of life. In a detailed sequence analysis, Hedlund et al. [14] delineated 86 different MDR sub-families, with functions ranging from quinone reductases to a f-crystallin in vertebrate lens. According to this scheme, AcuI polypeptides are in the MDR012 sub-family, which is also termed the YhdH group, in recognition of the Escherichia coli gene product of that name, which is 54% identical to AcuI of Rhodobacter. The function of E. coli YhdH is unknown and no mutant phenotypes have been reported. Although the overall fold structure of the YhdH polypeptide resembles that of quinone reductases, it has very little sequence similarity to any known enzymes that act on quinones [15]. Furthermore, the YhdH polypeptide lacks a Zn co-factor, which distinguishes it from the many zinc-containing alcohol dehydrogenases and several other members of the MDR super-family [14,15,16]. Thus, AcuI is not a ''putative Zn-dependent oxidoreductase'', as suggested by its annotation (http://www. ncbi.nlm.nih.gov/protein/ABA77575.1).
Recent biochemical evidence that complements these genetic links between AcuI and acrylate has recently shown that AcuI of Rhodobacter sphaeroides has acryloyl-CoA reductase activity, catalysing the conversion of acryloyl-CoA to propionyl-CoA [13]. Here, we present more insights into the function and distribution of acuIlike genes and their products in a wide range of bacteria, including those that do and those that do not catabolise DMSP. We present a model that indicates that AcuI-like enzymes have an unusual and widespread general role in bacteria, as well as affecting how DMSP can be catabolised. Figure 1. Locations of acuI relative to various ddd and dmdA genes in different bacteria. Locations of the acuI genes relative to those that encode DMSP demethylase (dmdA -yellow fill) or the ddd genes that encode the DddD, DddL, DddP and DddY DMSP lyases (arrows filled with various colors) are shown. Grey-filled arrows signify genes with other, known roles in DMSP catabolism. The fccA and fccB genes in Shewanella sp. MR4 encode a flavocytochrome c and a tetraheme cytochrome c respectively. Gene tags from left to right as shown above are: Halomonas HTNK1: ACV84065 to ACV84073 inclusive Rhodobacter sphaeroides 2.4.1: RSP_1433 to RSP_1435 inclusive Canididatus Puniceispirillum marinum IMCC1322: SAR116_1428, SAR116_1427 Alcaligenes faecalis M3A: ADT64689 to ADT64696 inclusive Arcobacter nitrofigilis DSM 7299: Arnit_0113, Arnit_0112 Shewanella sp. MR4: Shewmr4_2154 to Shewmr4_2151 inclusive Ruegeria pomeroyi DSS-3: SPO1913, SPO1914 Also shown are the dimensions of the Ruegeria pomeroyi DSS-3 dmdA-lacZ and acuI-lacZ fusion plasmids (pBIO2020 and pBIO2021 respectively) in which the reporter lacZ gene in pBIO1878. The cloned R. pomeroyi DNA (shown as blue lines) was cloned into pBIO1878 to form pBIO2022 (acuI+dmdA) and pBIO2024 (dmdA alone) for the complementation tests. doi:10.1371/journal.pone.0035947.g001 Results acuI-like genes are near some, but not all, the ddd and dmdA genes involved in DMSP catabolism It had been noted that there were acuI-like genes near some bacterial ddd and dmdA genes involved in DMSP catabolism. In a more thorough examination of these linkage relationships, we found that the distributions of acuI genes in relation to each of the individual types of dmdA and ddd genes in different bacteria are instructive, as follows.
Of the seven known enzymes that act directly on DMSP, either demethylating it (DmdA) or cleaving it to release DMS (DddD, DddL, DddP, DddQ, DddW and DddY), all but two of the corresponding genes (dddQ and dddW) are closely linked to acuI in at least one bacterial strain. These are considered individually in the following section.
dmdA. The dmdA gene is largely confined to two groups of marine bacteria (see [6,7]). One of these, the Roseobacters, comprises several genera in a sub-group of the Rhodobacterales. These are abundant and widespread in the oceans and coastal waters and most of them can catabolise DMSP -indeed, several individual strains can both demethylate it and cleave it, releasing DMS [17]. This is due to their possession of both the DmdA demethylase plus one or more DMSP lyases (DddD, DddL, DddP, DddQ and/or DddW), each of which generates acrylate or 3HP plus DMS. Individual Roseobacter strains have various portfolios of these different lyases [6,18].
We found that all 37 of the searchable, genome-sequenced Roseobacter strains listed in ''Roseobase'' (http://www.roseobase. org/), contained a single acuI gene, whose products mostly ranged from 44%-57% identity to AcuI of Rhodobacter sphaeroides (bit score .289); that of Citreicella sp. SE45 was more closely related, being 85% identical ( Figure 2). Of these Roseobacter strains, 26 contain the DmdA demethylase and in all but two of these cases, acuI was immediately 39 of, and likely co-transcribed with, dmdA. In the two exceptions (Phaeobacter gallaeciensis BS107 and Rhodobacterales bacterium HTCC2255), acuI was elsewhere in the genome, and in the former of these, it was adjacent to a gene whose product is annotated as a betaine/carnitine/choline transporter (BCCT family), which resembled DddT, a DMSP transporter in other bacteria, such as Halomonas HTNK1 ( [8]; see below). All 11 Roseobacter strains that lack dmdA also contain acuI, and in some cases (e.g. Oceanicola granulosus HTCC2516 and Phaeobacter gallaeciensis 2.10), it was next to a gene that resembled dddT, as in Phaeobacter gallaeciensis BS107 (see above).
The other major group of bacteria that contain DmdA demethylase are species of Pelagibacter, in the SAR 11 clade; unlike the Roseobacters, SAR11 bacteria lack any known DMSP lyases. All three genome-sequenced strains of Pelagibacter have close homologues of AcuI (which are ,40% identical to AcuI of Rhodobacter sphaeroides), but none is closely linked to dmdA; in turn, none of their acuI genes is near any genes known to be involved in DMSP catabolism.
Turning to the linkages of acuI to the ddd genes that encode the various DMSP lyases, these reveal a number of different patterns, as follows.
DddD. The DddD enzyme is, so far, unique, since it generates DMS plus 3HP as the 3C catabolite, in contrast to the other DMSP lyases, which cleave DMSP into DMS plus acrylate. The dddD gene occurs, sporadically, in a range of Proteobacteria. In several of these, including Marinomonas MWYL1, in which it was first identified [19], it is clustered with other ddd genes with ancillary functions involved in DMSP catabolism (transport, regulation, subsequent catabolic steps).
To date, the only dddD-containing cluster that includes acuI is in Halomonas HTNK1 ( Figure 1). Significantly, though, this c-Proteobacterium grows on both DMSP and on acrylate as sole carbon sources [8], unlike, for example, Marinomonas MWYL1, which grows only on the former substrate. This ability is due to conversion of acrylate to 3HP, via the activities of the products of the acuN and acuK genes, which are also clustered with dddD and the other ddd genes in Halomonas HTNK1 ( [8]; Figure 1). To date, all other dddD-containing bacteria lack acuN and acuK, anywhere in their genomes. In light of the postulated function of AcuI in conferring resistance to acrylate-mediated toxicity, the association of acuI with genes involved in the catabolism of this substrate is unlikely to be coincidental (see below).
In several other bacteria that contain dddD, there is no nearby acuI, but, in nearly all such cases, there is a version of acuI elsewhere in their genomes, unlinked to any other known gene involved in DMSP catabolism. These include representatives of the a-(e.g. Hoeflea), b-(Burkholderia) and c-(e.g. Marinomonas) Proteobacteria.
Three Roseobacter strains, namely Sagittula stellata E-37, Citreicella sp. SE45 and Rhodobacterales bacterium HTCC2083, contain DddD homologues with predicted DMSP lyase activities. All three contain close homologues of AcuI and, in strain HTCC2083, the corresponding gene is downstream of dmdA, as in many other Roseobacters (see above). However, Sagittula stellata E-37 and Citreicella sp. SE45 are unusual among the Roseobacters in that they lack dmdA and their versions of acuI are not near any genes that are known to be involved in DMSP catabolism.
One final point concerning dddD-containing bacteria is that those strains in the family Rhizobiaceae that contain this gene, namely Rhizobium leguminosarum WSM2304, Sinorhizobium fredii NGR234 and Agrobacterium tumefaciens 5A, lack any detectable acuI homologue anywhere in their genomes.
DddY. The DMSP lyase product of dddY differs from the other five DMSP lyases in that it is located in the bacterial periplasm, not the cytoplasm. First described in the b-Proteobacterium Alcaligenes faecalis M3A [20,21], dddY also occurs occasionally in other sub-phyla, namely Arcobacter nitrofigilis DSM7299 (e sub-phylum), Desulfovibrio acrylicus W218 (d) and in the c-Proteobacteria Ferrimonas balearica DSM9799 and several different species of Shewanella [20].
The Alcaligenes faecalis M3A acuI gene is immediately 59 of and co-transcribed with dddY, under the control of an acrylateinducible promoter [20]. The cluster that contains these two genes resembles that of Halomonas HTNK1, since it includes regulatory genes (dddZ and dddR), downstream catabolic genes (dddA, dddC) and the acuN and acuK genes, whose products act on the acrylate, either supplied exogenously, or generated by cleavage of DMSP by the DddY lyase ( Figure 1).
The genome of the e-Proteobacterium Arcobacter nitrofigilis has two dddY genes, one of which (locus tag Arnit_0113) lies immediately 59 of an acuI-like gene (Arnit_0112), whose product is in the MDR012 family, though somewhat divergently related to the AcuI sequences of other bacteria ( Figure 2; see below) Of the nine species of Shewanella that contain dddY, eight have a nearby version of acuI ( Figure 1) the exception being S. frigidimarina, whose acuI gene is elsewhere in the genome (see below).
DddL. The DddL DMSP lyase cleaves DMSP into DMS plus acrylate. To date, dddL is largely confined to strains of bacteria in the Rhodobacterales and Rhizobiales families, with one representative (Marinobacter sp. MnI7-9) in the c-Proteobacteria. All these strains contain acuI somewhere in their genomes, but the only known example in which it is closely linked to dddL is in the originally described Rhodobacter sphaeroides strains 2.4.1, Figure 2. Molecular phylogenetic analysis of selected AcuI polypeptides. Strains, clades, and protein identification numbers of a selection of the AcuI polypeptides in are shown. Strains in which MDR012-type polypeptides are encoded by acuI genes that are close to dmdA or to the various ddd genes are highlighted in yellow and the two MDR028 gene products near dddQ in green. Those cases where the cloned acuI genes were shown experimentally to correct the acrylate sensitivity of the E. coli YhdH 2 mutant are underlined. Other examples illustrate the wide taxonomic range of ATCC17029 and WS8N, in which these two genes are cotranscribed ( [11,12]; Figure 1).
DddP. The dddP gene encodes a DMSP lyase in the M24 family of metallo-peptidases and is frequently found in the Roseobacters, with the majority (24 out of 37) of the genomesequenced strains harbouring this gene [6,22]. As described above, the acuI gene of most strains in this clade is 39 of dmdA, and in no case was there tight linkage between acuI and dddP in any of the Roseobacters. There is evidence for horizontal gene transfer of dddP to some c-Proteobacteria (Vibrio orientalis and Oceanimonas doudoroffii, the latter which has two different dddP genes [23]), to Candidatus Puniceispirillum marinum (a member of the abundant SAR116 clade of a-Proteobacteria), and even to some Ascomycete fungi [22]. In strains of Candidatus P. marinum, acuI lies upstream of dddP (Figure 1), the only case in which dddP and acuI are closely linked in any known bacterium.
DddQ. The DMSP lyase encoded by dddQ is confined, so far, to the Roseobacters, in which it occurs in seven different strains [6]. In none of these is dddQ near an acuI-like gene, but we did note that the dddQ of R. pomeroyi DSS-3 and the adjacent dddQ1 and dddQ2 genes in Roseovarius nubinhibens ISM [24] were closely linked to and likely co-transcribed with a gene (SPO1593 in R. pomeroyi DSS-3 and ISM_14095 in Roseovarius nubinhibens) whose product was in the MDR super-family. However, this polypeptide was predicted to be in the MDR028 sub-family and is markedly divergent to the MDR012-type AcuI polypeptide described here ( Figure 2 and see below).
DddW. To date, dddW is only seen in two Roseobacter strains, namely R. pomeroyi DSS-3 and Roseobacter sp. MED193 [25]; in neither of these was it linked to an acuI-like gene.

Widespread distribution of AcuI-like gene products in Bacterial phyla
Apart from those MDR028 versions that are near dddQ, all the acuI-like genes that are tightly linked to the various ddd and dmdA genes encode MDR012 sub-family medium chain reductase/ dehydrogenases, as judged by the sequence similarities of their products to at least one member of this sub-group, listed by Hedlund et al. [14]. However, it is clear that these AcuI proteins do not comprise a particular out-group that is distinct from many other MDR012 polypeptides that occur in other bacteria, in a range of taxa, and which have no known link with DMSP and/or acrylate catabolism. This is illustrated in Figure 2, which shows a maximum likelihood tree of a range of different AcuI-like polypeptides; the gene products that were included in the tree were chosen either because they had been shown, above, to correct the acrylate sensitivity of AcuI 2 and/or YhdH 2 mutants, or had sequences that closely resembled such ratified polypeptides, or because their corresponding genes were closely linked to ddd or dmdA genes or, finally, because they were from a range of different Bacterial phyla. Some of these are described in more detail, below.
With the exception of Citreicella sp. SE45 (see above) the AcuI polypeptides of the Roseobacters are all very similar to each other (,80% identical) whether their acuI gene is downstream of dmdA or not. Their closest matches are in bacteria with no known role in DMSP catabolism, including Polymorphum gilvum SL003B-26A1 (a-Proteobacterium) and Thioalkalivibrio sulfidophilus (c-Proteobacterium), whose homologues are respectively ,80% and ,66% identical to the AcuIs in the Roseobacters. Of the AcuI polypeptides encoded by the loci near the various ddd genes, the Roseobacter AcuIs are more similar (,60% identical) to the product of the acuI that adjoins dddP in Candidatus Puniceispirillum marinum IMCC1322 than to those that are linked to any of the various dddY genes and to dddD of Halomonas. And, as shown in Figure 2, the AcuI of the Roseobacters, exemplified by that of R. pomeroyi, is closely related to those of Candidatus Pelagibacter ubique, and the eponymous YhdH polypeptide of E. coli.
The original AcuI of Rhodobacter sphaeroides strain 2.4.1 more closely resembles that of Citreicella sp. SE45 (86%) and some other, more taxonomically distant strains, such as the b-Proteobacterium Pusillimonas sp. T7-7 and the Verrucomicrobiae bacterium strain DG1235 (,65%) than it does to the polypeptides encoded by the acuI genes near dmdA of the Roseobacters (53%) or dddY of Alcaligenes (60%) or dddD of Halomonas HTNK1 (62%) ( Figure 2). In turn, the respective AcuI products in these last two strains most closely resemble homologues in organisms with no links with DMSP, namely the b-Proteobacterium Dechloromonas aromatica RCB (78% identical) and the c-Proteobacterium Alkalilimnicola ehrlichii MLHE-1 (61%), two species that are unusual in having two separate genes for AcuI-like polypeptides ( Figure 2).
The sequences of the AcuI polypeptides in all but one of the Shewanella strains that harbour dddY form a closely related group, with 75-94% identify to each other, and closely resembling YhdH of E. coli. The exception, Shewanella frigidimarina, is the only one whose acuI gene is not linked to dddY; its AcuI polypeptide is rather different to the others in this genus and is closely related to that of Halomonas HTNK1 ( Figure 2).
Finally, the product of the acuI gene Arnit_0112 that abuts dddY in Arcobacter nitrofigilis [26] is less closely related to the polypeptides encoded by any of the other acuIs that are linked to ddd or dmdA genes, and range from 37% identity (to those in the Roseobacters) to 43% (in Halomonas HTNK1). As with the examples above, the closest homologues to the Arnit_0112 gene product more closely resemble homologues in strains that are not involved in DMSP catabolism, such as Psychrobacter cryohalolentis K5 (ã-Proteobacterium; 69% identical).
These observations show clearly that although the AcuI-like polypeptides encoded by genes near dddD, dddL, dddP, dddY and dmdA of different bacteria are all within the MDR012 sub-family, their relatedness to each other is not necessarily congruent with either the taxonomic status of the organisms, or with the particular class of Ddd DMSP lyase encoded by the neighboring genes. For example, the AcuI's of Rhodobacter sphaeroides (a-Proteobacterium, DddL), Alcaligenes faecalis M3A (b-Proteobacterium, DddY) and Halomonas HTNK1 (c-Proteobacterium, DddD) are more closely related to each other than they are to those encoded by the genes downstream of dmdA in the Roseobacters or next to dddP in Candidatus Puniceispirillum marinum ( Figure 2).
It was also apparent from the examples above that AcuI-type polypeptides in the MDR012 sub-family occur in a very wide range of bacteria. This was further demonstrated when we used bacteria that harbour AcuI homologues, several of which are closely related to those encoded by genes linked to ddd or dmdA, and include some genera (Xanthomonas, Streptomyces, Geobacter) in which only some strains contain AcuI homologues. (C./V. = Chlamydiae/Verrucomicrobia; a-, b-, d-, c-, erefer to the corresponding sub-phylum of Proteobacteria). Three strains have two separate AcuI homologues, as indicated for Dechloromonas aromatica, Alkalilimnicola ehrlichii and Psychrobacter cryohalolentis. The tree with the highest log likelihoood is shown. The percentage of trees in which the associated taxa clustered together is shown next to the branches. Initial tree(s) for the heuristic search were obtained automatically using maximum parsimony method or by the BIONJ method with MCL distance matrix. The tree is drawn to scale, with branch lengths measured in the number of substitutions per site. doi:10.1371/journal.pone.0035947.g002 BLASTP to interrogate the current NCBI listing of genomesequenced microbes, using AcuI of Rhodobacter sphaeroides as the in silico probe. Close homologues were present in many Bacterial phyla, with that of Arcobacter being the least similar (36% amino acid sequence identity; bit score 211) as illustrated in Figure 2. However, MDR102 family members were absent from all Archaeal strains currently available, and from all but two eukaryotes.
Within the Proteobacteria, MDR012 proteins are highly prevalent, though not universal among the c-sub-phylum. Thus, although all known, genome-sequenced Enterobacteriaceae (including Escherichia coli -see below) and Vibrioonales contain an acuI homologue, it is missing from the Pasteurellales, including Haemophilus species. Interestingly, some individual c-Proteobacterial genera include species that contain (e.g. Xanthomonas gardneri) or lack (X. campestris pv. campestris) a polypeptide of the MDR012 subfamily. Similarly, in the a-, b-, dand e-sub-phyla of Proteobacteria, acuI was present in only some species within a family (e.g. the Rhizobiaceae {a}), or even a genus (Burkholderia {b} or Geobacter {e}). Thus, Rhizobium leguminosarum contains acuI, but strains of the closely related Sinorhizobium, including NGR234, which harbours dddD (see above) lack a close homologue. Intra-genus diversity also occurs in other bacterial phyla; the much-studied Streptomyces coelicolor lacks AcuI, but other strains in this Actinomycete genus contain a homologue that closely resembles those described above -the homologue in Streptomyces sp. SA3_actG is 70% identical to that in Halomonas, for example (see Figure 2). It was also apparent that some Bacteria, for example the Chlamydiae, Cyanobacteria and Spirochaetes, have few or no strains with close homologues of AcuI.
Intriguingly, there are also close AcuI homologues (,50% identical to Rhodobacter AcuI) in two marine animals, namely the Tunicate sea-slug Oikopleura dioica and the Cnidarian Clytia hemisphaerica. For both species, these matches were seen to EST sequences, showing that their acuI-like genes are expressed. In the case of O. dioica, whose small (70 Mb) genome has been sequenced [27], it was possible to deduce that the corresponding gene (BACOIKO008_47) has two introns, precluding any possibility that this sequence arose via bacterial contamination and strongly pointing its acquisition by inter-Domain horizontal gene transfer.
The AcuI homologue YhdH in Escherichia coli protects against the toxic effects of acrylate The finding that mutations in acuI of Rhodobacter sphaeroides caused hypersensitivity to the inhibitory effects of acrylate [11] was the first reported phenotypic effect of mutations in any gene that encoded a medium chain reductase/dehydrogenase in the MDR012 sub-family in any bacterium. It therefore was of interest to see if mutations in similar genes in other bacteria conferred a similar phenotype. One such gene (see above) was yhdH of E. coli K-12, whose function was previously unknown.
We therefore obtained a YhdH 2 insertional mutant strain JW3222-1, and compared it with its wild type E. coli K-12 parental strain BW25113 by growing both these strains on M9 minimal agar plates containing glycerol as carbon source, plus varying concentrations of acrylate. As shown in Figure 3 the YhdH 2 mutant was extremely sensitive to acrylate compared to the wild type, being unable to grow at concentrations as low as 50 mM, some 100-fold lower than the concentration that was tolerated by the wild type.
We examined if the yhdH mutation affected growth in the presence of other compounds with structural similarities to acrylate and/or which might be metabolically converted to or from acrylate; namely MMPA, propionic acid, 3HP, methacrylic acid, 3-butenoic acid, 4-pentenoic acid, acrylamide and allyl alcohol (Figure 4). Of these, the only one in which the wild type and the YhdH 2 mutant differed in their responses was 3HP. Although less inhibitory to both strains than acrylate, the wild type tolerated 40 mM 3HP in the medium, but the YhdH 2 mutant failed to form colonies at 5 mM. We also exposed the wild type and YhdH 2 mutant E. coli strains for 24 hours to high levels (10 mM) acrylate in M9 buffer that lacked any other carbon source and which therefore did not support cell growth. Following this treatment, the cells were washed, serially diluted, and plated on LB complete media and incubated, before counting the numbers of colonies. Compared to the control (buffer with no acrylate), exposure to acrylate did not affect the survival of either strain. Thus, the effects of acrylate appear to be bacteriostatic rather than bacteriocidal, and/or this observation may mean that acrylate must be transformed to the genuine inhibitory compound and that this only occurs in actively metabolising cells.
Correction of the acrylate sensitivity of an E. coli YhdH 2 mutant with cloned acuI from other bacteria The finding of such a clear phenotypic difference in acrylate tolerance in the wild type and a YhdH 2 mutant of E. coli offered a facile way to establish if this phenotype could be corrected by the cloned acuI-like genes from other organisms. We therefore amplified and cloned the individual acuI genes from genomic DNA of E. coli itself and from several bacteria whose acuI gene was closely linked to different ddd genes or to dmdA. These were Halomonas HTNK1 (next to dddD), Alcaligenes faecalis M3A (dddY), Rhodobacter sphaeroides (dddL), Arcobacter nitrofigilis (dddY), Burkholderia ambifaria (which has a distantly linked dddD gene [19]) and R. pomeroyi DSS-3 (dmdA). We also examined the effects of the cloned acuI gene of Rhizobium leguminosarum 3841, a strain with no known links to DMSP catabolism, whose pRL120182 gene encodes a polypeptide that is 53% identical to AcuI of Rhodobacter sphaeroides. The resultant recombinant plasmids were each transformed into the YhdH 2 E. coli mutant JW3222-1, and the transformants were tested for acrylate tolerance, in comparison with the wild type E. coli strain BW25113. As shown in Figure 3, all the clones tested could restore acrylate resistance to the E. coli YhdH 2 mutant.
In contrast, a plasmid that contained the Roseovarius nubinhibens ISM gene (ISM_14095) in the dddQ cluster of that strain and which encodes an MDR028-type polypeptide (see above), failed to correct the acrylate sensitivity of the E. coli YhdH 2 mutant.
Repercussions of the close linkage of acuI and dmdA in the model Roseobacter Ruegeria pomeroyi DSS-3 R. pomeroyi DSS-3 was the first Roseobacter to be genomesequenced [28] and has had a relatively long history as a subject for studies on DMSP catabolism. Earlier biochemical studies showed that it can both demethylate DMSP and cleave it to acrylate plus DMS, with the former pathway being relatively more important at lower DMSP concentrations [17]. Genetic analyses then showed that the demethylation pathway was mediated by the DmdA demethylase, with the subsequent, sequential downstream catabolic reactions being catalysed by the products of the dmdB, dmdC and dmdD genes [9,10]. This strain also has no less than three DMSP lyases, DddP, DddQ and DddW, all of which contribute to the cleavage of DMSP to acrylate plus DMS [23].
As stressed by Reisch et al. [7], any ''switch'' that modulates the partitioning of the fluxes through the cleavage and demethylation pathways may be important, not only for the individual strains and species of Roseobacters, but more widely, with significant environmental consequences. This is because the volatile DMS product of the DMSP lyases is a major vehicle to transfer sulfur from the seas to the atmosphere, thence back to land [29]. The finding in many Roseobacter strains that dmdA, a key gene in the demethylation pathway is immediately 59 of acuI, which is implicated in handling acrylate, a major product of the cleavage pathway, suggests that there may be some intimate links between these two catabolic routes in the Roseobacters, which may impinge on this switch.
To further examine these links, we first generated insertional mutations into dmdA and into acuI of R. pomeroyi DSS-3 (see Materials and Methods). In light of the hypersensitivity of the AcuI 2 /YhdH 2 mutants of Rhodobacter sphaeroides and E. coli, it was not unexpected to find that the insertion into acuI of R. pomeroyi conferred a similar phenotype. This mutant failed to grow on minimal medium containing 0.2 mM acrylate, with or without the alternative carbon source succinate, whereas wild type R. pomeroyi was unaffected for growth at acrylate concentrations greater than 5 mM.  Since catabolism of DMSP by one or more of the R. pomeroyi Ddd DMSP lyases generates acrylate as one of the products, we also examined if the mutation in acuI affected the growth of R. pomeroyi when DMSP was in the medium. The wild type tolerated concentrations greater than 20 mM, but growth of the AcuI 2 mutant was strongly inhibited by this concentration of DMSP.
The insertional mutation into dmdA of R. pomeroyi led to similar phenotypes as those in the AcuI 2 mutant. Thus, growth of the DmdA 2 mutant was strongly inhibited on medium that contained acrylate as low as 0.5 mM and on DMSP at 1 mM even in the presence of the ''regular'' carbon source, succinate. There are two likely contributing factors to explain why the dmdA mutation would cause hypersensitivity to acrylate and to DMSP. First, the insertion into dmdA is predicted to be polar on the expression of the downstream acuI, which would abolish or reduce the AcuImediated protective effects against acrylate, either added exogenously, or generated by the cleavage of DMSP. Secondly, the DmdA 2 mutant would be expected to channel more of the DMSP catabolic flux through one or more of the DMSP lyases, since the demethylation pathway was blocked. Indeed, it was directly shown by Howard et al. [9], that a DmdA 2 mutant of this strain produced more acrylate from DMSP than the wild type. Consistent with this, we noted that production of DMS, the other product of the cleavage pathway, was also enhanced, ,5-fold, when our DmdA 2 mutant was assayed for DMSP lyase activity.
Further evidence for these explanations was obtained from a series of complementation tests, using a series of separate plasmid constructions as follows. We cloned the R. pomeroyi acuI and dmdA genes, both individually and together, into the wide host-range cloning vector pBIO1878 [25], and separately introduced the resultant recombinant plasmids into the R. pomeroyi AcuI 2 and DmdA 2 mutants in tri-parental conjugational matings. The acrylate and DMSP sensitivities of both mutants were corrected by the plasmids that contained acuI alone (pBIO2024) or in tandem with dmdA (pBIO2022) but, significantly, the plasmid that contained only dmdA (pBIO2021) did not overcome the sensitivity to either of these compounds. Thus, the sensitivities to both acrylate and to DMSP in the DmdA 2 mutant must be due to the effect of the insertional mutation on the expression of the downstream acuI gene.
We noted a second link between acrylate and the R. pomeroyi dmdA-acuI operon, which concerns its transcriptional regulation. In the course of a microarray experiment, the expression levels of both acuI and dmdA were substantially enhanced (,14-fold) when the cells were grown in MBM minimal media in the presence of 5 mM DMSP. Significantly, when 5 mM acrylate was present in the growth medium, the expression of these two genes was also induced, and by a similar factor (,12-fold). This marked induction of dmdA and of acuI by DMSP was not seen in a different transcriptomic analysis of this strain by Bürgmann et al. [30], most likely because of the much lower concentration (80 mM) of DMSP that was added to the medium in that study.
To examine the expression of the R. pomeroyi DSS-3 dmdA-acuI operon in more detail, we made two lacZ transcriptional fusion plasmids, both based on the wide host-range reporter vector pBIO1878 (See Figure 1). Both plasmids contained the promoter region of the dmdA-acuI operon, but their 39 ends, fused to the lacZ reporter, were either in dmdA (pBIO2020) or acuI (pBIO2021). These two plasmids were each mobilised into R. pomeroyi DSS-3 by conjugation, and cultures of the transconjugants were each grown in minimal medium, containing or lacking either acrylate or DMSP (each at 5 mM) before assaying their b -galactosidase activities. The results tallied with those in the microarrays; the expression of both genes was markedly enhanced by both compounds compared to the levels in the control medium, with a ,10-fold increase with DMSP and ,14-fold when acrylate was in the growth medium. Thus, for the acuI-lacZ fusion, the bgalactosidase activities were 5663 Miller Units in the control medium, 502620 in the +DMSP and 780632 in the +acrylate media. The finding that a breakdown product of the DMSP cleavage pathway can induce the production of the DMSP demethylase DmdA means that there is a regulatory tie-in between the two pathways, with possible physiological consequences (see below).

Discussion
Interest in the acuI gene stemmed from its close linkage to a range of different ddd and dmdA genes that are involved in the initial steps of DMSP catabolism in a wide range of bacteria. A more direct link with DMSP, via one of its catabolites, acrylate, came from observations on the acuR-acuI-dddL operon of Rhodobacter sphaeroides; not only was its expression enhanced by the co-inducer acrylate, but AcuI 2 mutants were less effective in the breakdown of acrylate and were more sensitive to the growthinhibiting effects of this compound [11].
However, the bioinformatically and experimentally based observations described here show that the role of the acuI gene extends far beyond the realm of DMSP catabolism and DMSPcatabolising bacteria. Furthermore, the recent demonstration that AcuI of Rhodobacter sphaeroides has acryloyl-CoA reductase activity, which converts acryloyl-CoA to propionyl-CoA [13], provides significant, and highly relevant biochemical insights. Taken together, these new data prompt us to propose a novel, and potentially widespread, functional role for the AcuI-type gene products, as follows.
In general terms, we suggest that the intracellular presence of a compound that can be formed endogenously, but whose production is elevated in cells that are grown with exogenous acrylate, is extremely inhibitory to growth. The primary role of AcuI is to act as a ''cleansing agent'' to reduce the concentrations of this compound to sub-inhibitory levels. Given the enzymatic activity of AcuI, the most likely suspect for this molecule is acryloyl-CoA itself, whose cellular toxicity was alluded to by Herrmann et al. [31]. Here, we consider the evidence in favour of this model. First, and most obviously, it explains why mutations in acuI of Ruegeria, Rhodobacter and in the equivalent gene, yhdH, in E. coli are hypersensitive to the inhibitory effects of acrylate. Of the other compounds tested, the AcuI 2 /YhdH 2 mutants were also sensitive to 3-OH-propionate, consistent with its conversion to 3-OHpropionyl-CoA, thence to acryloyl-CoA, as suggested by Schneider et al. ( [13]; see Figure 5). The fact that these hypersensitive phenotypes of the E. coli YhdH 2 mutant were corrected by the cloned acuI genes from a range of different bacteria confirmed that these genes, too, are all functionally equivalent.
Acryloyl-CoA is predicted to be a very strong electrophile that would react with other important molecules in the cell [31]. Although it was not demonstrated formally that acryloyl-CoA is indeed the culprit that is responsible for the growth inhibition in acrylate-grown cells (and, less so, in response to 3HP), there is a precedent in which a related molecule, propionyl-CoA, was reported to be a ''suicide substrate'' that inactivated a short-chain acyl-CoA dehydrogenase, due to modification of its flavin adenine dinucleotide (FAD) cofactor [32]. It is unlikely that the inhibitory effects seen with acrylate are due solely to the interaction of acryloyl-CoA with the FAD of bacterial acyl-CoA dehydrogenase itself, since mutations in the E. coli fadE gene that encodes this enzyme are capable of normal growth, unless forced to use a fatty acid such as oleate as a carbon course [33]. However, if the more reactive acryloyl-CoA did react with FAD of other enzymes that contain this cofactor, or its reduced FADH 2 form, then this might form the basis of its inhibitory effects if such enzymes include any that are needed for normal growth.
One implication of this model is that many bacteria (including E. coli K-12) can convert exogenous acrylate to acryloyl-CoA. There have been no reports of such an activity in E. coli, and, to our knowledge this bacterium does not encounter acrylate in its natural environment. The same holds true for many of the other bacteria, in a range of different taxa, that also harbour close homologues of AcuI. It therefore seems unlikely that E. coli has a dedicated acryloyl-CoA ligase activity. However, at least one acetyl-CoA ligase enzyme has broad specificity and can attach CoA to other substrates, including acrylate [34]. Indeed, we have noted that when 1 mM 14 C acrylate was fed to mid-log phase cultures of E. coli BL21, ca. 75% of the counts appeared as 14 CO 2 after 16 hours incubation, with concomitant loss of the original labelled substrate (M.J. Sullivan, unpublished). So, E. coli, and perhaps an unexpectedly wide range of bacteria, may be able to catabolise acrylate in vivo, albeit rather inefficiently.
Concerning the linkage of acuI with several different classes of DMSP catabolic genes in different bacteria, we propose that this is an adaptive response in which the AcuI enzyme may counter the inhibitory effects of the acrylate that is obtained directly from the environment and/or which is made by cleavage of DMSP by Ddd lyase(s). Although the terms ''acrylate'' and ''acrylic'' are in common parlance, this is largely because these are widely used feed-stocks in the petrochemical industry. To our knowledge, the only natural environments in which there are significant amounts of acrylate are those, such as coral reefs, with very high levels of DMSP, much of which is converted to acrylate by microbial action [35,36].
As shown above, though, the acuI gene of many DMSPcatabolising bacteria is not linked to the DMSP catabolic genes.
However, the features and distributions of those ddd/dmdA genes that do or do not have a closely linked acuI gene are informative. Most notably, the only case in which acuI is near the dddD gene is in Halomonas HTNK1, and this is the only known dddD cluster that also includes the acuN and acuK genes. Significantly, DddD is the only known DMSP lyase that does not generate acrylate as its C3 product, so, in most bacteria that harbor dddD, there is no need for any dedicated protection system from the damaging effects of acrylate and its subsequent catabolite(s). However, the possession of the acuN and acuK genes confers on Halomonas HTNK1 the ability to catabolise exogenous acrylate, converting it to 3HP, the same initial catabolite as that generated by the action of DddD on DMSP. Furthermore, acryloyl-CoA is a predicted transient intermediate in the conversion of acrylate to 3HP, since the AcuN polypeptide is in a family of acyl-CoA transferases. Therefore, this unusual, dual-purpose ddd/acu gene cluster of Halomonas may have recruited a closely linked acuI gene for ''added protection'' [8].
Turning to DddY, another enzyme that does release acrylate from DMSP, there are nearby acuI genes in nearly all the diverse bacteria that contain this lyase, the only exceptions, to date, being in Shewanella frigidimarina NCIMB 400 and Ferrimonas balearica DSM9799, both of which have acuI homologues, but these are elsewhere in their genomes. In Alcaligenes faecalis M3A, the intimate relationship of acrylate and its acuI gene is further emphasised by the fact that expression of its adjacent acuI-dddY genes is massively increased when the cells were grown in the presence of acrylate [20].
In contrast, concerning the genes for the four other DMSP lyases, there are no acuI-like genes near dddQ and dddW in any bacteria to date, and only one case each in which dddP (Candidatus Puniceispirillum marinum) or dddL (Rhodobacter sphaeroides) is close to acuI, even though all these genes encode DMSP lyases that cleave DMSP into acrylate plus DMS. Our explanation for this is that with the exceptions of the very few cases of horizontal gene transfer, these four ddd genes are confined to the Roseobacters and, in the majority of genome-sequenced strains of this clade, Figure 5. AcuI-mediated conversion of acryloyl-CoA to propionyl-CoA. The pathway from 3-hydroxypropionate to propionyl-CoA is adapted from Schneider et al. [13]. Also shown is how DMSP lyases can generate acrylate, which is postulated to be converted to acryolyl-CoA by a CoA-ligase, as yet unidentified. The exact identity of the DMSP lyase is strain-dependent; e.g., Rhodobacter sphaeroides has the DddL lyase, and in Ruegeria pomeroyi, the acrylate can be generated from DMSP by DddP, DddQ and/or DddW. doi:10.1371/journal.pone.0035947.g005 their acuI gene is immediately downstream of their dmdA genes. The ability to catabolise DMSP, both by demethylation and by cleavage, is part of the core lifestyle of the Roseobacters [18,37], as witnessed by the fact that the great majority of them contain both dmdA plus at least one ddd gene. However, the particular portfolio of ddd gene varies considerably from strain to strain [18,23]. From an adaptive point of view, it may therefore be most efficient if acuI is linked to dmdA, so that its product can deal with the potentially harmful consequences of the acrylate that is formed by any of the DMSP lyases in the same strain. In the SAR11 clade, the other major group of bacteria with the DmdA DMSP demethylase, their acuI genes are not linked to dmdA, consistent with the fact that these bacteria do not cleave DMSP into acrylate plus DMS, so have no need for a specialised acrylate protection system that connects to DMSP catabolism.
The linkage and co-expression of dmdA and the downstream acuI gene may also affect the ways in which the Roseobacters partition the DMSP catabolic flux between the cleavage and the demethylation routes. The relative importance of these pathways in R. pomeroyi DSS-3 had been shown to be influenced by the concentrations of the DMSP substrate, with a higher relative flux through the cleavage route as the DMSP concentrations increased [17]. Furthermore, DmdA 2 mutants produce proportionally more acrylate and DMS from DMSP than does the wild type ( [9]; see above), due to the blockage of the demethylation pathways, which results in more of the substrate being available for cleavage by the DddP, DddQ and/or DddW DMSP lyases of this strain. However, our new finding that the acrylate that is produced by these lyases is itself a likely co-inducer of dmdA means that there is also a more direct regulatory link between the expression of the cleavage and demethylation pathways. This could have a homeostatic outcome, whereby, as the intracellular levels of acrylate rise due to lyase activities, there is increased expression of the DmdA demethylase. Thus, by increasing the flux through the demethylation pathway, less DMSP substrate is available for the acrylate-generating action of the Ddd lyases and by enhancing the levels of AcuI, there is enhanced protection against the potential damage, most likely inflicted by acryloyl CoA. Then as the acrylate levels drop, the relative inputs of the various lyases rise, with a concomitant increase in intracellular acrylate. The situation may be more complex, though, for at least two reasons.
Firstly, the expression of one of the ddd genes, dddW, is considerably enhanced in cells of R. pomeroyi DSS-3 grown in the presence of DMSP, but acrylate is not a co-inducer for this gene [25]. Secondly, although the demethylation pathway in R. pomeroyi DSS-3 that was described recently [10] does not generate any acrylate, a previously suggested catabolic scheme proposed that MMPA, the initially demethylated product of DMSP, is subject to demethiolation in a step that would yield acrylate plus methane thiol [38,39]. Since this has not been formally precluded as method for DMSP catabolism, albeit a minor one, it is possible that some of the acrylate that acts as a co-inducer of the dmdA-acuI operon may have originated by this route.
The very fact that a catabolic product (in this case, acrylate) enhances the expression of an enzyme that acts on the substrate (DMSP) is, in itself, an unusual phenomenon in bacterial gene regulation. However, this mode of control is a feature of bacterial catabolism of DMSP and was first noted in physiological experiments, in which pre-growth of different bacteria in the presence of DMSP or 3HP enhanced their levels of DMSP lyase activity [5]. Since then, these DMSP catabolites were shown to be co-inducers of ddd catabolic genes in other bacteria, such as Halomonas and Alcaligenes [6]. Indeed, in Rhodobacter sphaeroides, although the substrate DMSP appeared to induce its dddL gene, the DMSP had to be cleaved by the DddL lyase, forming the bona fide co-inducer, acrylate [11]. It is striking that this phenomenon of catabolite-responsive gene induction extends to the dmdA gene involved in the demethylation pathway. Prompted by this observation, we also investigated if MMPA, the demethylated derivative of DMSP generated by the action of the DmdA demethylase was a co-inducer of the dmdA-acuI operon; however, there was no induction of the dmdA-lacZ or acuI-lacZ fusions when strains of Ruegeria carrying the corresponding reporter plasmids were pre-grown in the presence of 5 mM MMPA.
It is noteworthy that many bacteria do not have close homologues of AcuI in their deduced proteomes. These may be entire clades, such as the Chlamydiae Phylum or the Order Pasteurellales within the c-Proteobacteria. Perhaps more strikingly, there are also several cases in which individual genera include some strains that do and some that do not contain acuI. Indeed, we demonstrated directly that the cloned pRL120182 gene of Rhizobium leguminosarum 3841, whose product is a predicted AcuI enzyme, conferred acrylate resistance to the E. coli YhdH 2 mutant. Yet, a very closely related strain, Sinorhizobium fredii NGR234, lacks an AcuI homologue but in our hands (unpublished) was as tolerant of acrylate in the medium as R. leguminosarum 3841.
Do these various strains that lack AcuI have alternative methods to circumvent the toxic effects of acrylate? In that connection, Hetzel et al. [40] purified a heterotrimeric enzyme with acryloyl-CoA reductase activity from Clostridium propionicum, but the Nterminal sequences of none of the polypeptides resembled that of AcuI. Similarly, the conversion of acryloyl-CoA to propionyl-CoA by two other MDR family members has been reported in the Crenarchaeota, Sulfolobus tokodaii [41] and the green non-sulfur bacterium, Chloroflexus aurantiacus [42]. Both strains engage in CO 2 fixation, using a cycle that includes the conversion of acryloyl CoA to propionyl-CoA. In Chloroflexus, this is mediated by two adjacent MDR-type domains within a multi-functional protein, but in the Archaea, this reductive step was performed by a stand-alone MDR polypeptide. However, the sequence similarities of both these enzymes to the MDR012 class of AcuI/YhdH polypeptides studied here is very limited.
Although acuI was first of interest because of its link with acrylate in bacteria that synthesised this molecule via DMSP catabolism, this is just one aspect of a wider and more important role for this gene and its close relatives. This even extends to the allocation of a function to an E. coli gene whose current description is ''Putative quinone oxidoreductase, function unknown'' (''ecogene''; http://ecogene.org/geneinfo.php?eg_id=EG11315). Although the work described here includes the first example of a phenotype that can been ascribed to mutations in the E. coli yhdH gene, the molecular basis of the inhibitory effects of acrylate remain to be formally confirmed as, indeed, does the proposal that acryloyl-CoA is the toxic molecule. It will also be of interest to establish if other enzymes, which do not resemble AcuI in their sequence, but which mimic its role in conferring acrylate resistance, occur in other organisms that do not have a recognisable acuI gene in their genomes.
To assay b-galactosidase, R. pomeroyi cells were grown overnight in either MBM minimal medium with 10 mM succinate as carbon source and, where appropriate, the co-inducers DMSP and acrylate, each at 5 mM, prior to being assayed for b-galactosidase as described in Rossen et al. [46]. The transcriptional fusion plasmid vector was pBIO1878, which is based on pMP220 [47] and includes a selectable Spc R cassette to facilitate selection in Roseobacters [25].

In vitro and in vivo genetic manipulations
General handling and manipulation of DNA were done as in Wexler et al. [48]. Plasmids were conjugated into the Rif R R. pomeroyi strain J470 by triparental mating using helper plasmid pRK2013 [49].

Gene amplification and construction of plasmids and mutants
Fragments of genomic DNA containing the intact yhdH of E. coli and the acuI genes of a selection of other bacteria, were each amplified by PCR from genomic DNA obtained from the E. coli K12 strain BW25113, Ruegeria pomeroyi DSS-3, Rhodobacter sphaeroides 2.4.1 [50], Alcaligenes faecalis M3A [20], Halomonas sp. HTNK1 [8], Rhizobium leguminosarum 3841 [51] and Burkholderia ambifaria AMMD [19] using primers that contained appropriate restriction sites (Table S2). Digested PCR products were then ligated into pET16b, pET21a, pRK415 [52] or pBIO1878 and transformed into E. coli strain 803 [53]. For cloning Roseovarius nubinhibens ISM_14095, a 4 kb PstI fragment from pBIO1880 was sub-cloned into pBluescript [54]. The Arcobacter nitrofigilis acuI gene was identified by screening a genomic library in the cosmid vector pLAFR3 [55] of this strain for any cosmids that corrected the acrylate sensitivity of the YhdH 2 mutant.
To facilitate the cloning of the R. pomeroyi dmdA and acuI genes, together with their promoter, a 2.4 kb PCR fragment containing both these genes was first cloned into pBluescript using XbaI and BamHI sites in the primer sequences, yielding plasmid pBIO2019. To construct the dmdA-lac fusion plasmid, pBIO2019 was digested with XbaI and PstI, releasing a 350 bp fragment that contained the dmdA/acuI promoter and whose 39 end was within dmdA. This fragment was then cloned into pBIO1878 (Spc R /Tet R ), forming pBIO2020. To construct an acuI-lac fusion plasmid, pBIO2019 was digested with XbaI and NsiI, releasing a 1.4 kb fragment that contained the dmdA/acuI promoter and whose 39 end was in acuI. This fragment was cloned into pBIO1878 that had been digested with XbaI plus PstI to form pBIO2021.
For the complementation tests with the AcuI 2 and DmdA 2 mutants of R. pomeroyi DSS-3, the following plasmids were constructed. The 2.4 kb fragment that contains intact dmdA and acuI plus their promoter, which was used to construct pBIO2019 (see above), was released from that plasmid and cloned into pBIO1878 to form pBIO2022. To clone acuI alone, though still under the control of its own promoter, plasmid pBIO2019 was first digested with SphI, then religated in a procedure that removed three SphI fragments internal to dmdA, but which leaves acuI intact. The deletant plasmid pBIO2023 was then digested with XbaI plus BamHI and the released fragment was sub-cloned into pBIO1878, forming pBIO2024. The plasmid that contained only dmdA was pBIO2021, described above. The dimensions and names of the relevant plasmids are shown in Figure 1 and Table S1. All recombinant plasmids were ratified by sequencing of the inserts, performed by Genome Enterprise Ltd, Norwich Research Park, Norwich, UK.
The insertional mutations into acuI and into dmdA of R. pomeroyi were made by a procedure in which fragments internal to each of the two genes were cloned, separately, into the suicide plasmid vector pBIO1879 [24], a Spc R derivative of the suicide vector pK19mob [56]. The internal dmdA fragment was made by amplifying a 800 bp fragment from R. pomeroyi DSS-3 genomic DNA using forward and reverse primers (Table S2) which respectively contain EcoR1 and PstI restriction sites, prior to cloning into pBIO1879, cut with the same enzymes to form pBIO1870. A 960 bp acuI internal DNA fragment was made by digesting pBIO2019 with SalI, and this was then cloned into SalIdigested pBIO1879 to form pBIO2025. The two plasmids pBIO1870 and pBIO2025 were then each conjugated to R. pomeroyi DSS-3 in triparental matings, selecting Rif R /Spc R /Kan R transconjugants, which should arise via a single cross-over event within the corresponding genes. These mutants were confirmed by colony PCR and Southern Blotting and were termed J527 (AcuI 2 ) and J471 (DmdA 2 ).
The E. coli strains BW25113 and its yhdH 2 mutant derivative JW3222-1 were obtained from the Keio collection of E. coli K-12 in-frame, single-gene knockout mutants [57] through the E. coli Genetic Stock Center, Yale University, New Haven, Connecticut.

Growth inhibition experiments
Starter cultures of the various E. coli or R. pomeroyi strains were grown to mid log-phase in complete medium. Cells were adjusted to equivalent OD 600 values, washed in minimal medium. For E. coli, 10 ml aliquots were spotted onto plates of M9 minimal agar medium containing varying levels of acrylate, DMSP or other tester compounds. Growth was scored after incubation at 37uC for 20 hours. For R. pomeroyi, cells were used to inoculate 5 ml MBM minimal media containing 10 mM succinate as carbon source and varying levels (50 mM to 10 mM) of acrylate or DMSP (200 mM -20 mM). Cultures were incubated at 28uC, with shaking and the growth levels recorded after 48 hours.