Screening of Metagenomic and Genomic Libraries Reveals Three Classes of Bacterial Enzymes That Overcome the Toxicity of Acrylate

Acrylate is produced in significant quantities through the microbial cleavage of the highly abundant marine osmoprotectant dimethylsulfoniopropionate, an important process in the marine sulfur cycle. Acrylate can inhibit bacterial growth, likely through its conversion to the highly toxic molecule acrylyl-CoA. Previous work identified an acrylyl-CoA reductase, encoded by the gene acuI, as being important for conferring on bacteria the ability to grow in the presence of acrylate. However, some bacteria lack acuI, and, conversely, many bacteria that may not encounter acrylate in their regular environments do contain this gene. We therefore sought to identify new genes that might confer tolerance to acrylate. To do this, we used functional screening of metagenomic and genomic libraries to identify novel genes that corrected an E. coli mutant that was defective in acuI, and was therefore hyper-sensitive to acrylate. The metagenomic libraries yielded two types of genes that overcame this toxicity. The majority encoded enzymes resembling AcuI, but with significant sequence divergence among each other and previously ratified AcuI enzymes. One other metagenomic gene, arkA, had very close relatives in Bacillus and related bacteria, and is predicted to encode an enoyl-acyl carrier protein reductase, in the same family as FabK, which catalyses the final step in fatty-acid biosynthesis in some pathogenic Firmicute bacteria. A genomic library of Novosphingobium, a metabolically versatile alphaproteobacterium that lacks both acuI and arkA, yielded vutD and vutE, two genes that, together, conferred acrylate resistance. These encode sequential steps in the oxidative catabolism of valine in a pathway in which, significantly, methacrylyl-CoA is a toxic intermediate. These findings expand the range of bacteria for which the acuI gene encodes a functional acrylyl-CoA reductase, and also identify novel enzymes that can similarly function in conferring acrylate resistance, likely, again, through the removal of the toxic product acrylyl-CoA.


Introduction
Acrylate is a well-known compound, largely due to its use as a major chemical feedstock for acrylyl polymers in paints and other products of the petrochemical industries. In natural environments, though, it only occurs in significant amounts in few, specific niches. This is because it is the product of a particular catabolic reaction, namely the cleavage of dimethylsulfoniopropionate (DMSP), a highly abundant (,10 9 tons produced per annum) osmoprotectant and anti-stress compatible solute that is made by many marine photosynthetic phytoplankton, macroalgal seaweeds and a few angiosperms [1][2][3][4]. When stressed, some of these algae can cleave their DMSP into the volatile dimethyl sulfide (DMS) plus acrylate, which may provide protection against further predation by zooplankton [5]. Also, the DMSP that is released when plankton die or are grazed can be catabolised by marine bacteria and some fungi, which produce enzymes that are generically termed 'DMSP lyases' which also yield acrylate plus DMS [1]. Of these two DMSP cleavage products, DMS has received more attention, because of its important environmental effects in the global sulfur cycle and, perhaps, on climate through its effects on cloud formation [6].
Nevertheless, acrylate itself is also important, at least locally. It is abundant in (e.g.) corals, which are massive sources of DMSP, made by the dinoflagellete Symbiodinium [7,8] and/or their hosts [9]. Although at high concentrations and low pH, acrylate may harm some marine microbes [10], some bacteria thrive in the high concentrations around cells of the phytoplankton Phaeocytsis [11], and some can use it as a sole carbon source for growth [1,12,13]. Not surprisingly, therefore, the study of acrylate catabolism has been intimately tied up with the ability of bacteria to use DMSP as a substrate.
Recent genetic studies (reviewed in [3,4]) have revealed striking diversity in the ability to catabolise DMSP. At least seven different ''Ddd'' enzymes that cleave DMSP, and which generate DMS as an initial product, have been described in different marine bacteria. Five of these (DddL, DddP, DddQ, DddW and DddY) yield acrylate as the C3 catabolite, but despite these similarities in their substrate and in the resultant products, these enzymes are in at least four very different polypeptide families. In many different bacteria, these ''primary'' ddd genes are clustered with other ddd genes that are variously involved in DMSP import, in downstream catabolic steps and/or in gene regulation.
DMSP is also subject to a different catabolic fate, in which it is demethylated to methylmercaptopropionate (MMPA) via a DmdA demethylase. This pathway occurs in two abundant marine alphaproteobacterial groups, the SAR11 clade and the Roseobacters [14], and yields neither DMS nor acrylate. However, methylthioacrylyl-CoA is a downstream product [15,16], and it had been suggested that demethiolation of MMPA might also generate acrylate [17].
During the course of analysing the ddd gene clusters of several bacteria, we noted that most of these included a gene that we termed acuI (acrylate utilisation). This was first identified in Rhodobacter sphaeroides 2.4.1 [18], in which acuI is the central gene of an operon whose promoter-distal gene, dddL, encodes a lyase that cleaves DMSP into DMS plus acrylate [19]. The promoterproximal gene, acuR, encodes a transcriptional regulator that represses the entire operon, unless relieved by the presence of acrylate, which is the co-inducer molecule. AcuI 2 mutants were defective in converting 14 C-labelled acrylate to 14 CO 2 , and, strikingly, were hyper-sensitive to the inhibitory effects on growth of exogenous acrylate [18]. They were also defective for growth on 3HP as a carbon source [20,21].
Genes that closely resembled acuI were found not only in the ddd clusters of different bacteria (including those for dddD in Halomonas, dddY in Alcaligenes, and dddP in Candidatus Puniceispirillum marinum) but we noted an acuI-like gene immediately 39 of dmdA in nearly all Roseobacter strains, and that the expression of this dmdA-acuI operon is much-enhanced by growth of cells with acrylate [22].
The AcuI enzyme is in the MDR012 subgroup of the very large medium chain dehydrogenase/reductase (MDR) superfamily [23]. It was recently shown to be an acrylyl-CoA reductase, converting acrylyl-CoA to propionyl-CoA [20] with high specificity for acrylyl-CoA, and a very low K m (,3 mM) for this reaction [21]. This could explain the phenotypes of AcuI 2 mutants, including their increased sensitivity to acrylate, since these would accumulate acrylyl-CoA, a very active cytotoxic electrophile that attacks sulfhydryl groups [24]. One enzyme in particular, pyruvate formate lyase, is hyper-sensitive to acrylyl-CoA [25], perhaps explaining why acrylate was more toxic to E. coli growing in anaerobic than in aerobic conditions [26].
Taken together, these observations suggest that the acuI gene in these various ddd (and dmdA) clusters may have an adaptive, protective role, via the transformation of the very reactive acrylyl-CoA into the less harmful propionyl-CoA [22]. However, other observations have recently shown that AcuI-type enzymes must have more complex and wide-ranging functions than just this.
We noted [22] that the taxonomic distribution of bacteria with acuI homologues was widespread, but sporadic. Thus, there are several bacterial phyla (e.g. the Chlamydia and the Spirochetes to name but two) in which none of the genome-sequenced strains contain an acuI-like gene and others in which only some contain it. For example, only one strain (Gallibacterium anatis UMN179) in the Pasteurellales Order of gammaproteobacteria has an AcuI-like polypeptide; other species, including the important pathogens Haemophilus, Pasteurella and Mannheimia, do not. This diverse, sporadic distribution implies that acuI has been subject to several rounds of horizontal gene transfer. Furthermore, AcuI homologues occur in many bacteria (including E. coli) that do not, as far as we know, normally encounter acrylate or DMSP in their natural environments [22]. Thus, mutations in the E. coli acuI gene (previously termed yhdH) cause marked increased in acrylate sensitivity, a phenotype that could be corrected by cloned acuI from a range of bacteria, not only those with known connections with DMSP or acrylate [22]. Further, the in vitro properties of the E. coli acuI gene product closely resembled those of the Rhodobacter version (see above) in terms of its affinity and specificity for its acrylyl-CoA substrate [21]. Very recently, the enzymatic mechanism of two MDR012 members (including E. coli YhdH) was determined, with a wholly novel and unexpected intermediate being identified, comprising a covalent ene adduct between the NADPH coenzyme and the substrate [27].
In addition to AcuI, three other types of acrylyl-CoA reductase have been described. The first of these is an NADPH-dependent acrylyl-CoA reductase, forming part of the autotrophic CO 2 fixation cycle [28] in some crenarchaea. Like AcuI, this enzyme is in the MDR superfamily, but it has no significant sequence similarity with characterised AcuI proteins. Secondly, the Cterminal domain of a propionyl-CoA synthase from the anoxygenic photosynthetic bacterium Chloroflexus aurantiacus is an NADPH-dependent enoyl-CoA reductase that reduces acrylyl-CoA to propionyl-CoA, also in the 3HP cycle of autotrophic CO 2 fixation [29]. Finally, a different, NADH-dependent acrylyl-CoA reductase in Clostridium propionicum was reported to reduce acrylyl-CoA to propionyl-CoA [30].
The genomes of many other bacteria lack genes for any of these acrylyl-CoA reductases, so we set out to determine if we could isolate other genes, which when cloned, might confer acrylate resistance. To do this, we took a functional metagenomic approach [31][32][33], which sampled a wide portfolio of genomes, and which did not depend on homologies with known polypeptides, or on culture-dependent enrichments. By such means, we obtained various cosmids and plasmids that contained cloned DNA from different metagenomes and bacterial genomes which corrected the extreme sensitivity of the AcuI 2 (YhdH 2 ) E. coli mutant.

Results
Screening Metagenomic Libraries for Genes that Restore Acrylate Resistance to an E. coli AcuI 2 mutant To identify microbial genes that conferred acrylate resistance, we exploited the marked acrylate sensitivity of the E. coli AcuI 2 (YhdH 2 ) mutant, which was used as the recipient strain with donor cultures of E. coli that harboured plasmids or cosmids that comprised individual metagenomic libraries. Selection was made for acrylate-resistant (Acr R ) transconjugants on LB medium containing 2 mM acrylate, a concentration on which wild type E. coli grows well but which completely inhibits the acuI mutant [22] (Figure 1).
Two, pre-existing groups of metagenomic libraries were used (see Table S1 for details). In one set, large-insert (.25 kb) DNA fragments, isolated directly from bacteria in a waste-water treatment plant, had been cloned into pLAFR3, a wide hostrange cosmid vector [34]; these cosmids were transferred, en masse, by triparental conjugational mating into the Rif R E. coli acuI mutant J523. The other libraries comprised smaller inserts, obtained from several different environments (Table S1), cloned in the high copy number plasmid pCR-XL-TOPO; these were electroporated into the Spc R E. coli acuI mutant J557. The entire inserts of the pCR-XL-TOPO-based plasmids were then sequenced; with the cosmid-based clones, the relevant region was first localised by identifying smaller fragments which, when subcloned into pBluescript, conferred an Acr R phenotype.
This procedure yielded a total of 8 different metagenomic cosmids or plasmids that conferred an Acr R phenotype. The introduction of each of these conferred tolerance to acrylate concentrations that were at least 10-fold higher than the 1 mM that completely blocked growth of the E. coli acuI mutant ( Figure 1). Subsequent analyses of these cosmids and plasmids revealed two classes of genes that conferred an Acr R phenotype; one resembled the previously identified acuI, but the other was wholly different.

Metagenomic AcuI-like Medium Chain Dehydrogenase/ reductases
Two cosmids (pBIO2079 and pBIO2081), and five pCR-XL-TOPO-based plasmids (pBIO2151, pBIO2152, pBIO2153, pBIO2154 and pBIO2155), each contained a gene whose product resembled AcuI. The locations of these acuI-like genes, in the context of flanking genes in the cloned metagenomic DNA, and the similarities of their gene products to polypeptides in known bacteria are shown in Tables S2, S3, S4, S5, S6, S7 and S8.
As illustrated in a neighbor-joining tree (Figure 2), the sequences of these seven AcuI-like metagenomic polypeptides were compared with each other and with ratified, AcuI polypeptides, and also with other AcuI homologues from a range of different bacterial taxa. Several of these newly obtained versions of AcuI closely resembled those in known, taxonomically diverse bacteria, most of which had no known links with acrylate and/or DMSP. This further demonstrates the diversity of the AcuI polypeptides and the lack of congruity between the amino acid sequences of the various AcuI gene products and the taxonomic status of the bacteria that harbour them. For example, the AcuI-like polypeptide of the Firmicute Bacillus sp. SG-1 more closely resembles that of the Actinomycete Mycobacterium fortuitum than that of another Bacillus species, B. cereus Rock3-44. Therefore, acuI has likely been subject to wide-range, repeated rounds of horizontal gene transfer. Plates were incubated at 37uC overnight before recording growth, shown in this composite. The strains used were as follows: Lane 1, wild type E. coli J521; Lanes 2 and 6, AcuI 2 mutants J523 and J557 respectively; Lane 3, J523 corrected by E. coli acuI cloned in pET21a (pBIO2011); Lanes 4 and 5, J523 corrected by metagenomic acuI-like genes cloned in pLAFR3-based cosmids pBIO2079 and pBIO2081 respectively; Lanes 7, 8, 9, 10, 11, J557 corrected by metagenomic acuI-like gene cloned in pCR-XL-TOPO-based plasmids pBIO2151, pBIO2152, pBIO2153, pBIO2154 and pBIO2155 respectively; Lane 12, J557 corrected by metagenomic arkA-like gene cloned in pCR-XL-TOPO-based plasmid pBIO2160. doi:10.1371/journal.pone.0097660.g001 Figure 2. Phylogenetic tree of metagenomic AcuI-like proteins and other MDR012 subfamily proteins. In this neighbour-joining tree, the metagenomically identified AcuI enzymes are represented by the plasmid or cosmid from which they originate and are shown boxed, with those that also contain an adjacent CoA ligase marked with a black triangle. Other members of the MDR012 subfamily of the MDR superfamily in other bacteria are also included, with those that have been ratified as conferring an Acr R being marked with black squares. The MDR039 subfamily polypeptide of Burkholderia mallei ATCC 23344 (BMAA0163, accession YP_104995) was used as the outlier sequence. doi: 10 Overall, these metagenomic screens considerably extend the diversity of functional AcuI-type enzymes that confer an Acr R phenotype, so this may be a feature of all MDR012-type polypeptides, in diverse bacteria.

Flanking Genes Next to acuI in Metagenomic Clones
For most of the individual metagenomic clones, all the sequenced genes within the cloned DNA are likely from the same general type of bacterium (Tables S2, S3, S4, S5, S6, S7 and S8), for example, Firmicutes (in pBIO2079), Sphingomonadales (pBIO2153) or betaproteobacteria (pBIO2154). Mostly, the predicted functions of the flanking genes differed in the various metagenomic cosmids and have no known link with acrylate. However, in pBIO2081, pBIO2153 and pBIO2154, acuI was next to a gene whose product was in the Pfam family PF00501 of AMPbinding long-chain fatty acid-CoA ligases; conceivably, these may add CoA to acrylate (or similar molecule) that, in turn is the substrate for the corresponding AcuI acrylyl-CoA reductase.

A Metagenomically Derived Enoyl-acyl Carrier Protein (ACP) Reductase Confers an Acr R Phenotype
The metagenomic screens also yielded pBIO2160, a pCR-XL-TOPO-based plasmid that conferred acrylate resistance ( Figure 1), but in a very different way. The 5901 bps of insert DNA in pBIO2160, obtained from a cast-water biofilm, contains nine genes, none of which resembles acuI (Table S9). By sub-cloning individual genes, we identified one, termed arkA (cloned in pBIO2167), that was responsible for the Acr R phenotype.
ArkA is a strongly predicted (e 286 ) Class II enoyl-[ACP] reductase C (Pfam family PF03060) with .70% identity to a gene product in many (though not all) genome-sequenced species of the genus Bacillus and its close relatives (Table S9). Furthermore, the products of the other eight genes in pBIO2160 also resembled those in these Firmicutes. We noted, though, that arkA is absent from some well-known species of Bacillus (B. cereus, B. thuringiensis, B. anthracis), some of which contain an acuI-like gene (see above).
To show that the arkA-like gene of a known species of Bacillus can confer an Acr R phenotype, we amplified BMD_3924 of B. megaterium DSM 319, whose product is 61% identical to the metagenomic ArkA gene product of pBIO2160. When cloned into pET21a to form pBIO2195, BMD_3924 enhanced the acrylate resistance of the E. coli AcuI 2 mutant J522 allowing growth at concentrations .1 mM.
The sequences of these Bacillaceae ArkA polypeptides placed them into a distinct cluster in Pfam family PF03060, which has two other groups, with only limited sequence identity (,40%) to each other, and to the ArkA enzymes described here. One of these other groups is exemplified by a fungal 2-nitropropane dioxygenase, which removes the nitro group from various anionic nitroalkanes [35,36]. Members of the other sub-group have a very different function; these correspond to FabK, first found in Streptococcus and the closely related Enterococcus [37]. In these bacteria, FabK catalyses the final step in fatty acid biosynthesis, reducing trans-2enoyl-ACP to generate acyl-ACP ( Figure 3). FabK was discovered because these two pathogens were completely resistant to triclosan [38], a widely used antimicrobial additive in various household products. Triclosan inhibits FabI [39], which occurs in many bacteria, including E. coli, and in eukaryotes, and which has the same biochemical function in fatty acid biosynthesis as FabK [40,41], despite a lack of sequence similarity between these polypeptides ( Figure 3).
Introduction of the cloned fabK gene of Streptococcus into E. coli confers high-level tolerance to triclosan, showing that FabK can functionally replace the triclosan-sensitive FabI of E. coli [37]. In contrast, when we introduced plasmids containing the metagenomic and B. megaterium arkA genes (in pBIO2167 and pBIO2195 respectively) into E. coli, there was no such effect; the two derivatives grew in 20 mM triclosan, the same maximal level as for E. coli with the empty vector.
Taken together, these observations show that most, but not all, strains of the Bacillaceae contain the enzyme ArkA. Although similar in sequence to FabK, the ArkA polypeptide is not involved in fatty acid biosynthesis, but may have some other, as yet unknown role. Nevertheless, the similarity of the ArkA and FabK sequences underpinned our naming of arkA (acrylate resistance FabK-type). In light of the reaction that is catalysed by bona fide FabK, it seemed likely that ArkA reduces acrylyl-CoA to propionyl-CoA when exogenous acrylate is present, the same bioconversion as that effected by AcuI (Figure 3).
Novosphingobium Genes that Restore Acrylate Resistance to the E. coli AcuI 2 mutant Given the diversity of the types of genes that confer an Acr R phenotype and the taxonomic and ecological range of bacteria that contain them, we took another approach to see if there were yet other ways in which bacteria can detoxify acrylate. This stemmed from our observation that the deduced proteomes of several genome-sequenced bacterial strains and genera lacked any of the known enzymes that act on acrylyl-CoA, namely AcuI and ArkA, and those in the Crenarchaea, Chloroflexus aurantiacus or Clostridium propionicum (see Introduction). One such genus is Novosphingobium, an alphaproteobacterium that is characterised by its ability to catabolise a wide range of organic aromatic compounds [42].
To identify any Novosphingobium gene(s) that conferred an Acr R phenotype, we exploited a pre-existing genomic library of Novosphingobium tardaugens ARI-1, which comprised ,23,000 cosmids, each with inserts ,25 kb in size, cloned in pLAFR3. These plasmids were mobilised, en masse, into the E. coli AcuI 2 mutant J523, selecting for transconjugants that grew on medium supplemented with 5 mM acrylate. One such colony was obtained and the cosmid, termed pBIO2170, that was responsible for conferring the Acr R phenotype was studied in more detail.
Sequencing the insert in pBIO2170 confirmed that the sequences and the synteny of the cloned genes corresponded closely to those in all the other five genome-sequenced Novosphingobium species, including the much-studied N. aromaticivorans DSM 12444 [43]. One predicted seven-gene operon, equivalent to Saro_0855-Saro_0861 in strain DSM 12444, (Figure 4 ) was of immediate interest, since it included five genes (bauC, vutD, vutE, vutF and vutG -see below) whose products are predicted to catalyse consecutive steps in the oxidative catabolism of valine. Significantly, an intermediate in this pathway is methacrylyl-CoA (MAC-CoA), which closely resembles acrylyl-CoA ( Figure 5). The other two genes in the operon, corresponding to Saro_0856 and Saro_0857 in DSM 12444, encode short polypeptides of no known function, and with no homologues apart from in other Novosphingobium strains.
To confirm that gene(s) in this operon conferred an Acr R phenotype, pBIO2170 was mutagenized with Tn5lacZ (see Materials and Methods) and the mutagenized cosmids were examined for any that no longer corrected the acrylate sensitivity of the E. coli AcuI 2 mutant. All the insertions in these mutant cosmids were in the vutD or vutE genes, so-named because their products catalyse the corresponding steps in the valine utilisation pathway of Pseudomonas putida [44,45].
Given the known toxicity of MAC-CoA [46] and its structural similarity to acrylyl-CoA, it was reasonable that the cloned vutD and vutE genes confer acrylate resistance by effecting the conversion of acrylyl-CoA to 3HP, via the intermediate 3HP-CoA. Figure 5 illustrates the bona fide valine catabolic pathway, as well as the likely alternative conversions that would occur when exogenous acrylate is supplied.
To test this and to study the roles of the bacterial vutD and vutE genes directly, we chose to work on another alphaproteobacterium, Sinorhizobium fredii NGR234, which is more readily grown and genetically amenable than Novosphingobium. Like Novosphingobium, S. fredii NGR234 lacks acuI, and we noted that it contains all the relevant bau/vut valine oxidation genes, although, unlike Novosphingobium, they are not contiguous. Thus, bauC (NGR_b20870), vutE (NGR_b20860) and vutF (NGR_b20850) are adjacent to each other on the large plasmid pNGR234b, but are unlinked to vutD (NGR_c37150) and vutG (NGR_c03390), which are both located (but separately from each other) on the S. fredii chromosome (Figure 4).

The VutD and VutE Enzymes of S. fredii NGR234 Confer Acrylate Resistance
To examine directly if vutD (NGR_c37150) and vutE (NGR_b20860) of S. fredii NGR234 confer acrylate resistance, each of them was individually cloned into different, compatible vectors -pBluescript for vutD (forming pBIO2173) and pRK415 for vutE (forming pBIO2174) -and each of these two plasmids was used to transform the E. coli AcuI 2 mutant J522. Whereas the cloned vutE gene had no effect on the mutant's acrylate sensitivity, pBIO2173 (containing vutD) conferred partial resistance (tolerant to 5 mM acrylate). However, when a derivative of the E. coli mutant that contained both pBIO2173 and pBIO2174 was made, this was fully resistant, up to wild-type levels. Thus, it seems that the conversion of acrylyl-CoA to 3HP-CoA by VutD affords some protection, but that further resistance is conferred by VutE, via the subsequent conversion of 3HP-CoA to the less toxic 3HP [47].

Effect of Sinorhizobium fredii vutD and vutE Mutations on Growth on Valine and Sensitivity to Acrylate
To confirm that VutD and VutE function in the valine oxidation pathway in S. fredii itself, we made insertional, genomic mutations in vutD (NGR_c37150) and vutE (NGR_b20860) of S. fredii, using pBIO1879, a Spc R -resistant derivative of the widely used suicide plasmid pK19mob (see Materials and Methods). One ratified mutant for each of the genes, designated strains J554 (VutD 2 ) and J555 (VutE 2 ), was compared with the wild type, as follows. Whereas wild type S. fredii NGR234 grew well on L-valine (3 mM) as sole carbon source, to a final OD 600 of ,0.3, neither the VutD 2 nor the VutE 2 mutants showed any signs of growth on this amino acid, although both grew normally on succinate, to an OD 600 of 0.91 and 0.88 respectively.
As predicted, the S. fredii VutD 2 and VutE 2 mutants were more sensitive to acrylate and to methacrylate than the wild type, which grew in the presence of either compound at concentrations of . Predicted reactions for enzymes conferring an Acr R phenotype and related fatty acid biosynthesis enzymes. Reaction A shows the conversion catalysed by AcuI [20]. Reaction B shows the conversion catalysed by FabI [40] and FabK [37]. Reaction C represents the ArkAmediated conversion of acrylyl-CoA to propionyl-CoA predicted in this study. doi:10.1371/journal.pone.0097660.g003 1 mM; in contrast, the VutD 2 and VutE 2 mutants (J554 and J555) showed no growth at 0.1 mM ( Figure 6).
Taken together, these results confirm that the VutD and VutE enzymes of S. fredii are required for valine oxidation, and that mutations in either of these genes also affect the ability of S. fredii to tolerate the toxic effects of acrylate or methacrylate.

Relationship of the VutD and VutE Enzymes to the Previously Identified AcuK Acrylyl-CoA Hydratase
VutD is in the same super-family (Pfam PF00378) of enoyl-CoA hydratase/isomerases as AcuK, an enzyme described in two bacteria that catabolise DMSP, namely Halomonas sp. HTNK1 [12] and Alcaligenes faecalis M3A [13], in both of which acuK is clustered with the ddd genes required for DMSP catabolism. The ability of these two species to grow on acrylate (via DMSP cleavage) requires AcuK, which is predicted to convert acrylyl-CoA to 3HP-CoA, the same as that described for VutD (see above). Although we noted that VutD does not lie within the subgroup of PF00378 polypeptides that includes ratified AcuK enzymes, VutD and AcuK are similar in their sequences (ca 60% identical). AcuK (in Halomonas) and VutD (in S. fredii) also show similarity to the acrylyl-CoA hydratase (SPO0147) identified in the DMSP catabolic pathway of R. pomeroyi [48], being 55% and 63% identical respectively to the sequence of this enzyme, which has been shown to convert acrylyl-CoA to 3HP-CoA. These observations prompted us to test if, like VutD, AcuK can confer resistance to acrylate, as follows. A plasmid, pBIO1729, containing acuK of Halomonas sp. HTNK1 [12] was introduced into the E. coli AcuI 2 mutant strain J522. The resultant transformant displayed a slight increase in acrylate tolerance, similar to that seen (above) with the cloned vutD. However, when the cloned acuK gene was introduced into the E. coli mutant that already harboured vutE (in plasmid pBIO2174), this fully restored acrylate resistance (data not shown). This was analogous to what was seen, above, with the synergistic effects of vutD plus vutE.
Interestingly, the ddd/acu gene clusters of both Alcaligenes faecalis M3A and Halomonas sp. HTNK1 also include an acuI gene, as well as acuK (see above), so they may have two separate protective systems against acrylate-mediated toxicity. The reason for this is unclear but one explanation could be that AcuI is a dedicated detoxification system for acrylyl-CoA, whereas AcuK is mainly used to catabolise acrylate as a source of carbon, but the way(s) in which such functional compartmentalisation may occur are not clear.

Discussion
Almost by definition, functional metagenomics, as first envisaged by Jo Handelsman, provides a powerful method to isolate ''novel'' genes that would not have been recognised on the basis of scanning the sequences of metagenomes (or genomes). This approach has been most widely aimed at isolating novel genes that confer functions of biotechnological relevance (e.g. cellulases or lipases [49,50]). But, as shown here, it can also uncover  unexpected genes that may be involved in more fundamental metabolic function(s), as well as extending the diversity of previously identified genes that confer the function under study (acuI in this case).
The most frequently encountered metagenomic genes that corrected the acrylate sensitivity of the E. coli AcuI 2 mutant were themselves related to acuI. These varied considerably in their sequences, and likely came from the genomes of a wide range of bacterial taxa. Nevertheless, all these seven AcuI polypeptides were squarely placed within the medium chain reductase/ dehydrogenase MDR012 sub-family [23], as had been all the previously ratified AcuI polypeptides that confer an Acr R phenotype [22]. It remains to be seen if all MDR012 polypeptides confer acrylate resistance, but, judging by the MDR012 sequence space that is occupied by those for which this has been confirmed (Figure 2), and the taxonomic range of the bacteria in which these occur, it would be surprising if this were not the case.
For those bacteria that catabolise DMSP, generating large amounts of acrylate and its even more toxic CoA-ligated derivative, the presence of an enzyme that converts acrylyl-CoA to a less noxious product makes good sense, especially if the corresponding gene is closely linked to the ddd genes that are responsible for making the acrylate and if the expression of that gene is markedly enhanced by the acrylate co-inducer. And this is exactly what occurs in several taxonomically diverse bacteria, in which their acuI is clustered with the ddd or dmdA genes [3,22]. But what of those many other bacteria, which on the face of it, do not encounter acrylate? As a caveat to that last query, we would say that although we do not normally associate acrylate as part of the chemical milieu of many bacteria (notably E. coli), this may not necessarily be true. Many marine animals (from krill to penguins to seals) do eat DMSP-containing vegetation either directly or indirectly, and, indeed, seaweeds contribute to the human diet in many parts of the world. Acrylate was shown to affect the gut microflora of penguins [10], and DMSP-catabolising bacteria were readily obtained from guts of herrings [51]. So, enteric bacteria, at least, might encounter acrylate, at a global level, sufficiently often to maintain a selective advantage on those cells that can detoxify it.
However, if exogenous acrylate does not provide the selection pressure to maintain acuI, two possible scenarios must be considered. The first is that acrylyl-CoA is generated endogenously, even though the presence of this molecule has only been documented in a few cases over and above those bacteria that catabolise DMSP such as those cases, described above, in Chloroflexus, Clostridium and some Archaea. It has also been predicted that acrylyl-CoA is formed from lactoyl-CoA in lactate metabolism [52][53][54], propionyl-CoA in propionate metabolism [55] and the b-oxidation pathway of glucose fermentation [56], and b-alanyl-CoA in b-alanine metabolism [57], but these suggestions have not been directly verified.
A second general explanation is that acrylyl-CoA may not be the bona fide ''natural'' substrate. Certainly, we are unaware of any report that describes acrylyl-CoA in E. coli cells, despite the very strong affinity of the AcuI (YhdH) enzyme for this molecule in this species [21]. A more detailed study of the acuI (yhdH) mutants of E. coli might shed light on the ''proper'' role of these genes. To date, their only recognised phenotype is their marked sensitivity to acrylate, at concentrations as low as 20 mM, some two orders of magnitude less than in the wild type [22]. A more wide-ranging and subtle search of any other phenotypic changes, including those that may alter the metabolome, and performed under various environmental conditions might reveal the bona fide pathway that involves AcuI in these bacteria. Of course, such an approach could also be applied to other, genetically amenable bacteria that contain AcuI-like enzymes. Given the relatively wide diversity of the amino acid sequences in these different AcuI-type enzymes, the ''correct'' substrate/pathway need not necessarily be the same in all bacteria, even though they can all recognise acrylyl-CoA. In connection with this, it is noteworthy that in three metagenomic clones, in which the cloned DNA was likely derived from genomes of different delta-, alpha-and gammaproteobacteria, their acuI gene was immediately 59 of a gene that likely encoded a CoAligase. Furthermore, we also noted close linkage of acuI-like genes and a gene that encodes a CoA-ligase in the genomes of several known bacteria (e.g. the betaproteobacteria Azoarcus sp. KH32C and Dechloromonas aromatica RCB, and the Actinomycete Frankia). Although in the same polypeptide family, these various CoAligases are not very closely related to each other (,30% identical). However, the partial amino acid sequence of the ligase from pBIO2154 showed 47% identity (over ,500 amino acids of its Nterminus) to the only known (so far) acrylyl-CoA ligase [48]. In these cases, the products of these cloned genes may ligate CoA to an acrylate-like molecule, prior to its reduction by the product of the neighbouring acuI-like gene. However, the diversity of these ligases, together with the known low specificity of enzymes of this type [58], again preclude any confident predictions on what the natural substrate might be; again, an analysis of the corresponding mutants may be needed.
As already noted, many bacterial strains, genera and even whole phyla lack an acuI-like gene and this had underpinned our use of functional metagenomics to establish if this patchy distribution might be due to the role of AcuI being undertaken by other enzymatic systems. In terms of proof of concept, this was successful. Despite its lack of sequence similarity to AcuI, the arkA gene conferred an Acr R phenotype and likely catalyses the same overall reaction, namely the reduction of acrylyl-CoA to its propionyl derivative (Figure 3), though by a different molecular mechanism, in which NAD is the predicted coenzyme, in contrast to the NADP that occurs in AcuI [59].
The same questions that were raised with regard to the ''real'' function of acuI in (for example) E. coli also apply to the strains of Bacillus that harbour arkA. We do not know if the bona fide substrate is acrylyl-CoA or the pathway in which the ArkA enzyme participates. Whatever its function, it does not seem to be involved in fatty acid synthesis, despite the significant sequence similarity of ArkA and FabK, since arkA of B. megaterium did not rescue E. coli from its sensitivity to triclosan. Also, several strains of Bacillus that contain arkA, including B. megaterium, contain fabL, which encodes the bona fide enzyme that catalyses the final step in fatty acid biosynthesis in many strains of Bacillus [60]. As discussed above for acuI, the isolation and characterisation of an ArkA 2 mutant in a genetically amenable Bacillus strain (e.g. B. subtilis) might provide useful insights, but this has not yet been done.
Close relatives of the metagenomically obtained arkA (and its neighbouring genes) are restricted to some strains of Bacillus and very closely related genera of Firmicutes. But, as with acuI, the distribution of arkA is sporadic, being absent from several Bacillus species, including the well-known B. cereus, B. anthracis and B. thuringiensis. Do these strains not need an ArkA-type enzyme activity -or can it be substituted by yet another enzyme?
With the discovery of the vutD and vutE genes of Novosphingobium and Sinorhizobium, our complementary approach, in which we mined the genomes of bacteria that lack arkA or acuI for genes that conferred an Acr R phenotype, also proved successful. In this case, though, unlike that for arkA, pre-existing knowledge of the functions of these genes provided a ready explanation for the Acr R phenotype. Thus, the conversion, in two steps of the valine oxidation pathway, of the highly toxic [46] MAC-CoA to 3hydroxyisobutyryl-CoA and then to 3-hydroxyisobutytrate would correspond to the generation of 3-hydoxypropionate if cells were supplied with exogenous acrylate ( Figure 5). Our finding that maximal protection against the effects of acrylate requires both vutD and vutE provides strong evidence that a CoA-ligated intermediate (in this case, the VutE substrate, 3-hydroxypropionyl-CoA) is significantly more cytotoxic than the unmodified 3HP itself, but that both are less harmful than the acrylyl-CoA.
The valine oxidative pathway is widespread, in many bacteria, animals and plants [46,61,62]. The repercussions of the toxic intermediate MAC-CoA may even underpin an unusual feature of this pathway ( Figure 5), in which an activated acyl group (in 3hydroxyisobutyryl-CoA) is destroyed, only for another such molecule (propionyl-CoA) to be made later as a means of rapidly removing the potentially harmful MAC-CoA [46]. Certainly, genetic defects in this pathway cause serious pathologies in humans [63] and in plants [64].
Despite the extension of the different types of enzyme that can confer an Acr R phenotype, there are still many bacteria that contain none of the polypeptides that resemble AcuI, ArkA, or VutDE described above, or the AcrABC enzyme that converts acrylyl-CoA to propionyl CoA in the Clostridia [30]. These may simply have no need for such an enzymatic activity but, if these have yet other ways of dealing with acrylate, it should be relatively straightforward to identify these by making fresh genomic libraries from such 'null' bacteria and to screen these, in turn, for any cloned, functional genes that confer an Acr R phenotype.
Similarly, more wide-ranging screenings of metagenomic libraries that sample a wider range of environments (Including marine ones) may uncover yet other types of genes that confer an Acr R phenotype. Although the 'natural' functions of any such genes may not be forthcoming -as in the case of arkA -this approach could nonetheless point to the general functions of at least a few of the massive numbers of genes of unknown function that crowd the databases of genomes and metagenomes.
To test growth of S. fredii NGR234 on different carbon sources, late log phase cells were washed, diluted 1/100 into 5 ml RM containing the appropriate carbon source, then incubated with shaking at 28uC; growth was monitored at OD 600 . Sensitivity of E. coli to acrylate was tested by placing 10 ml aliquots of various dilutions of freshly grown cells onto LB agar plates containing appropriate concentrations of acrylate or other test compounds, then incubated for 16 hours. S. fredii cultures were incubated for 48 hours in 5 ml TY, then diluted 1/100 into 5 ml RM medium containing 10 mM succinate as carbon source, plus different concentrations of the test compounds. After 24 hours shaking incubation, the OD 600 was determined.

In vivo and in vitro Genetic Manipulations
Transfer of plasmids by conjugation, using helper plasmid pRK2013 [70], and transformation of E. coli, were done as in Wexler et al. [71] and by electroporation, essentially as in Hanahan [72]. E. coli strain 803 [73] was the usual host for transformation of most plasmids; JM101 [74] was used as the host for pBluescript. PCR primers containing appropriate restriction sites are shown in Table S11.

Construction of Mutant Strains
An insertional mutation into acuI (yhdH) of wild type E. coli J521 was made using the lambda red recombinase system [75]. Primers designed to amplify the Spc R cassette from plasmid pHP45V [76] also contained 40 bp flanking regions, corresponding to sequences immediately 59 and 39 of yhdH (Table S11). The resulting PCR product was electroporated into E. coli J521 carrying a lambda red recombinase expression plasmid, then plated on LB Spc plates and incubated at 30uC. One Spc R colony, termed strain J557, was picked and the insertional mutation in yhdH was confirmed by PCR, and by the mutant's acrylate-sensitive phenotype.
Insertional mutations in the S. fredii NGR234 genome were made using the pK19mob suicide vector system [77]. Fragments internal to the S. fredii genes NGR_c37150 and NGR_b20860 were each amplified and cloned into pBIO1879 [78], a Spc R derivative of pK19mob, to form plasmids pBIO2176 and pBIO2177 respectively. These plasmids were each transferred to S. fredii NGR234 by triparental conjugations. Single crossover events, in which the incoming plasmid had inserted into the corresponding target genes, were selected on TY plates containing Str (to kill E. coli donor strains), Spc and Kan (resistances conferred by pBIO1879). Strains with verified mutations in the genes NGR_c37150 (vutD) and NGR_b20860 (vutE) were termed J554 and J555 respectively.
The pLAFR3-based cosmid pBIO2170, which contains cloned N. tardaugens ARI-1 genomic DNA, including the bau/vut gene cluster, was randomly mutated using transposon Tn5lacZ. Plasmid pBIO2170 DNA was transformed into E. coli strain A118, which has a chromosomal copy of Tn5lacZ [79]. One transformant was then used as the donor in a triparental conjugational mating with R. leguminosarum bv. viciae 3841 as the recipient, selecting for Rhizobium transconjugants that harboured pBIO2170::Tn5lacZ on Str (Rhizobium), Kan (Tn5lacZ) and Tet (pLAFR3 cosmid). Cosmid DNA was prepared from ca. 200 transconjugants and used to transform E. coli strain 803, an efficient host for transformation with large plasmid/cosmids. Approximately 100 individual E. coli transformants were individually used as donors in triparental matings to the E. coli Rif R AcuI 2 mutant strain J523. Any mutant cosmids that no longer corrected the acrylate sensitivity of the J523 mutant were studied further.

Library Construction
The vector for the genomic library of Novosphingobium tardaugens ARI-1 was the wide host-range, mobilisable, cosmid pLAFR3 [80], which can accommodate inserts of ,25 kb. These libraries were made essentially as described by Curson et al. [19].
Metagenomic libraries [34] from a wastewater treatment plant (WWTP) in Whitlingham, Norfolk, UK, also used pLAFR3 as the vector and metagenomic small-insert libraries in pCR-XL-TOPO were made as in Nacke et al. [81]. Details of the libraries and their construction are shown in Table S1.

Bioinformatics
Sequences were compared to the NCBI sequence databases using BLASTX and BLASTP [82]. Protein alignments and neighbour-joining phylogenetic trees employed MEGA v5.10 [83] with 100 bootstrap replications.