The identification of an integral membrane, cytochrome c urate oxidase completes the catalytic repertoire of a therapeutic enzyme

In living organisms, the conversion of urate into allantoin requires three consecutive enzymes. The pathway was lost in hominid, predisposing humans to hyperuricemia and gout. Among other species, the genomic distribution of the two last enzymes of the pathway is wider than that of urate oxidase (Uox), suggesting the presence of unknown genes encoding Uox. Here we combine gene network analysis with association rule learning to identify the missing urate oxidase. In contrast with the known soluble Uox, the identified gene (puuD) encodes a membrane protein with a C-terminal cytochrome c. The 8-helix transmembrane domain corresponds to DUF989, a family without similarity to known proteins. Gene deletion in a PuuD-encoding organism (Agrobacterium fabrum) abolished urate degradation capacity; the phenotype was fully restored by complementation with a cytosolic Uox from zebrafish. Consistent with H2O2 production by zfUox, urate oxidation in the complemented strain caused a four-fold increase of catalase. No increase was observed in the wild-type, suggesting that urate oxidation by PuuD proceeds through cytochrome c-mediated electron transfer. These findings identify a missing link in purine catabolism, assign a biochemical activity to a domain of unknown function (DUF989), and complete the catalytic repertoire of an enzyme useful for human therapy.

stereochemistry 7 , the presence of these enzymes appears to be important for the rapid elimination of the metastable intermediates of urate oxidation 8 .
The functional coupling between Uox, Urah, and Urad is reflected by the evolutionary link of the corresponding genes, which are usually present or absent together in a given genome. This link has been key to the identification of urah and urad as genes associated with uox 5 ; in prokaryotes, however, cases were also noticed in which both urah and urad were present in the absence of uox genes. This observation was followed in recent years by the identification of alternative urate oxidases-non homologous enzymes involved in the same reaction step. The uox gene encodes the first identified urate oxidase 9 , a The tree represents 431 distinct species possessing either the uox, hpxO, hpyO, or COG3748 genes or both the urah and urad genes. The presence (red) or the absence (blue) of genes in complete genomes is shown alongside the organism tree. Main taxonomic groups and organisms discussed in the text are indicated.
cofactor-less enzyme with a T-fold domain 10 that is found in all tree domains of life. The existence of an alternative urate oxidase gene, named hpxO (hypoxanthine-xanthine utilization O), has been demonstrated in Klebsiella spp. possessing urah and urad but not uox 11,12 . The HpxO enzyme has been biochemically and structurally characterized as a flavoprotein belonging to a large family of hydroxylases 13,14 . An isofunctional non homologous urate oxidase has been identified in Xanthomonas spp. and named HpyO 15 ; although HpyO belongs to a distinct family of hydroxylases, similarly to HpxO, it uses NAD(P)H as a cosubstrate and FAD as a cofactor 15 .
From a mechanistic standpoint, Uox and HpxO/HpyO represent distinct solutions to the same biochemical problem. Each of these solutions has advantages and disadvantages. Uox does not depend on cofactors for activity, but its two-electron reduction of dioxygen generates H 2 O 2 . This potent oxidant is readily detoxified by peroxisomes, explaining the organellar localization of the pathway in eukaryotes [16][17][18] . However, the formation of hydrogen peroxide by the Uox reaction can pose problems with organisms lacking peroxisomes, and is a matter of concern for the therapeutic administration of the enzyme 19 . On the other hand, the four-electron dioxygen reduction catalysed by HpxO/HpyO generates H 2 O, but requires a labile cofactor and a cosubstrate that is consumed in the reaction (Fig. 1a).
Surprisingly, the enzymatic variety above described does not fully account for urate oxidation in living organisms. It has been pointed out that homologs of uox, hpxO, or hpyO genes are not identified in completely sequenced organisms known to perform urate oxidation 15 , and our preliminary search identified numerous species (mostly proteobacteria) possessing urah and urad but not recognizable uricase genes. As Urah and Urad act on metastable products of urate oxidation, the presence of the corresponding genes could be justified only by the presence of a gene capable of urate oxidation. However, no candidate oxidoreductases can be identified by homology among the genes associated with purine catabolism. To identify this missing gene and possibly a different urate oxidation mechanism, we used here gene network analysis and a data mining technique able to discover association rules between items of an item set 20,21 . The application of this technique identified with high confidence genes belonging to COG3748 as the missing urate oxidase. In striking contrast with the known genes involved in urate oxidation, the candidate gene (puuD) was found to encode an integral membrane domain (DUF989) fused with a cytochrome c. Experiments with deleted and complemented bacterial strains provided evidence for the functional assignment and revealed a novel mechanistic solution for urate oxidation.

Results
Identification of COG3748 as urate oxidase. To find candidate genes encoding enzymes with urate oxidase activity we interrogated a database of predicted gene/protein associations (String) 22 using as input Cluster of Orthologous Groups (COGs) 23 corresponding to the Uox, Urah, and Urad proteins involved in the enzymatic pathway (Fig. 1a,b). Among the numerous associated COGs reported, the COG3748 'predicted membrane protein' ranked in fifth position in the String list ( Supplementary Fig. S1) was considered a possible candidate as a domain with unknown function exhibiting the expected relation for a protein with urate oxidase activity, that is a connection with Urah and Urad and not with Uox in the association network (Fig. 1b). Indeed, genes involved in consecutive reactions of a pathway are expected to be correlated (e.g. they are found in the same genomes or in the same operons), while no correlation or anti-correlation is expected for different genes involved in the same reaction. Visual inspection of various operons containing COG3748 genes confirmed the presence of other genes involved in purine degradation and the absence of known genes encoding urate oxidase ( Supplementary Fig. S2).
To evaluate in detail the association between COG3748 and the other genes of urate oxidation we examined gene distributions in 1689 complete genomes from different species. Using a semi-automated method for gene classification we found 431 species possessing both the urah and urad genes or either COG3748 or a known urate oxidase gene (uox, hpxO, hpyO). This distribution clearly shows that the different genes involved in the first step of urate oxidation occur alternatively in different species (Fig. 1c). The cofactor-independent urate oxidase (Uox) occurs in eukaryotes, in some bacteria (particularly Actinomycetes), and in some archaea. Alternative oxidases (HpxO and HpyO) occur in a limited number of bacteria, while COG3748 occurs in several bacterial species (mostly proteobacteria) possessing both urah and urad but not known urate oxidase genes. Among these species are organisms (e.g. Paracoccus denitrificans, Pseudomonas aeruginosa) that were previously reported to be capable of oxidative urate degradation 24,25 .
Bioinformatics methods make it possible to quantitatively measure correlated 26 or anti-correlated 27 gene distributions. Such methods were devised for pairwise associations. The generalization introduced with the analysis of logic implications 28 , however, allows one to evaluate comparisons involving any number of gene families. The gene distribution of Fig. 1c shows that the relations among urate oxidation genes are not well described by pairwise comparisons. For instance, COG3748 is positively correlated with the presence of Urad (Urad implies COG3748 ) and negatively correlated with Uox (not Uox implies COG3748). Although significant, these "association rules" 20,21 have low confidence (Table 1), reflecting the fact that in many cases urad is present in the absence of COG3748, and the absence of uox does not imply the presence of COG3748 (see Fig. 1c and Supplementary Fig. S3). When more genes were included in the comparisons, association rules with increasing confidence and significance were obtained (Table 1). Interestingly, these associations are consistent with the supposed role of COG3748, and the best association rule obtained (93% confidence; P = 1 × 10 −196 ) was exactly that expected a priori by biochemical reasoning: that is, the presence of both Urah and Urad in the absence of known urate oxidases implies the presence of COG3748.
Based on the above analysis, we predicted a urate oxidase activity for COG3748. The presence of an integral membrane domain in these proteins (confirmed by sequence analysis, see below) was initially surprising, as the known urate oxidases are soluble proteins. However, early experiments reported that the urate oxidase activity of some organisms, including P. denitrificans and P. aeruginosa, was lost in the soluble fraction after sonication 24,25 , consistent with a membrane-bound enzyme.
Urate degradation without soluble uricase activity in organisms encoding COG3748 genes. For the anticipated difficulties in the recombinant expression and purification of the integral membrane COG3748 proteins, we sought to validate the bioinformatics prediction using a genetic approach. Agrobacterium fabrum C58 (previously known as A. tumefaciens C58) was selected as completely sequenced organism 29 for which well-developed genetic tools are available 30,31 . There is no information about urate utilization by A. fabrum, although the presence of a COG3748 gene (Atu2314, verified by sequencing of our laboratory strain), urah and urad genes (see Fig. 1c), as well as other downstream genes of the catabolic pathway 32 , suggest that A. fabrum is able to use urate as a nitrogen source. While in animals purine degradation is used to eliminate excess nitrogen, in plant, fungi, and prokaryotes this pathway is important for the recovery of mineral nitrogen from endogenous or imported purines 33,34 . Accordingly, we observed growth of A. fabrum in a minimal medium supplemented with urate as the nitrogen source (Fig. 2a). With respect to the growth observed in ammonia (Fig. 2b), a longer lag was observed with urate, suggesting that in this bacterium the urate utilization system is inducible. When urate consumption was measured in highly-concentrated cell cultures, the rapid disappearance of the urate peak at 293 nm was observed after a lag of about 120 minutes (Fig. 2c). This lag was not observed with pre-induced cultures ( Supplementary Fig. S4). However, no soluble urate oxidase activity could be measured using protein extracts of urate-induced cells (Fig. 2d).

Expression of a soluble urate oxidase (zfUox) complements the phenotype of Atu2314
(COG3748) deletion mutant. Using a gene replacement vector 30 , we generated a deletion mutant of the Atu2314 gene in A. fabrum C58. No differences in the growth on ammonia were observed in the deleted strain (Fig. 2b). By contrast, growth on urate was completely abolished (Fig. 2a), and no urate consumption was observed in the deleted strain (Fig. 2c).
The uox coding sequence from the vertebrate Danio rerio (zebrafish) was cloned in a IPTG-inducible expression vector 31 and used to transform the recipient Δ Atu2314 strain. No differences in the growth on ammonia were observed in the deleted and complemented strains in the presence or in the absence of IPTG (Fig. 2a). However, growth on urate was fully restored in the complemented strain in the presence of the IPTG inducer (Fig. 2b). Similarly, urate consumption was observed in Δ Atu2314/zfUox only in the presence of the inducer (Fig. 2c). Interestingly, the growth and urate consumption curves in the wild-type and complemented strains are superimposable, suggesting that in both strains the limiting step is urate uptake of from the medium (Fig. 2a,c). As expected, a soluble urate oxidase activity was measured in protein extracts of the complemented strain grown in the presence of the IPTG inducer (Fig. 2d). By comparison with the activity of the purified zfUox protein (Fig. S5), we estimate the expression of ~1 μ g zfUox per 10 6 cells in the complemented strain, corresponding to an activity of ~3 nmol urate min −1 . This intracellular activity is ten-fold higher than the activity measured in whole cells, consistent with the notion that urate import in the cytosol is a rate-limiting step in our system.
Based on bioinformatics and biochemical evidence, we conclude that Atu2314 and homologous genes belonging to COG3748 are responsible for urate oxidation in organisms lacking known urate oxidase. As a gene responsible for urate oxidation has been previously approximately mapped on the chromosome of one of such organisms (P. aeruginosa) and named puuD (mnemonic for purine utilization D) 35 , we adopted this name for the urate oxidase identified here at the molecular level.   Fig. S8a,b) and different prediction methods ( Supplementary Fig. S8c). Similarly, consistent predictions were obtained for a N out -C out membrane topology. With this organization (Fig. 3b), the longer extra-membrane loops (specifically the loops connecting helix I to helix II and helix V to VI) are located in the cytoplasm, whereas the Cyt_c domain is located outside the plasma membrane. Cleavable signal peptides are not identified in PuuD sequences, so in these proteins helix I is assumed to function as a signal anchor for the targeting to the plasma membrane. Within the DUF989 domain, the stronger conservation is observed in helices I, II, V, VI and in the cytosolic loops I-II and V-VI. Within the Cyt_c domain, conservation pertains especially to the residues that interact with the heme cofactor. Invariant residues in all PuuD sequences are two cysteines of the CXXCH motif (Cys340 and Cys343) that covalently bind the vinyl side-chains of heme through thioether bonds plus the histidine (His344) that provides the fifth heme iron ligand. Also, invariant is a methionine residue (Met385) located about 40 residues further on towards the C-terminus, which in c-type cytochromes provides the sixth iron ligand. A search with the PuuD C-terminal domain in the Swissmodel server identified more than 50 distinct cytochrome proteins suitable as templates for 3D homology model reconstruction. The best model was obtained using as template the crystal structure of cytochrome c L from Hyphomicrobium denitrificans (PDB 2D0W), a soluble cytochrome of the periplasmic space acting as the physiological electron acceptor of the methanol dehydrogenase quinoprotein through transient protein-protein interaction 38 . Only some parts of the structure can be modelled with high confidence, reflecting the low local similarity (22%) between the template and target sequences (Supplementary Fig. S9a). The positions of residues Cys340, Cys343, His344, and Met385 in the structural model is that expected for heme coordination ( Supplementary Fig. S9b), while another invariant residue in the PuuD alignment, Arg398, is directed Role of cytochrome c in PuuD urate oxidation. As a cofactor-free oxidase, the zfUox protein used in our experiments to complement the activity of PuuD deletion mutant transfers two electrons to oxygen and generates hydrogen peroxide as a by-product of the reaction. On the other hand, the structure of the PuuD protein and the presence of an electron transfer cytochrome c domain suggest a completely different mechanism of urate oxidation. When the catalase activity was measured in protein extracts of Δ puuD/zfUox cells, a four-fold increase of the catalase activity (p < 0.001) was observed for cells grown in urate with respect to cells grown in ammonia (Fig. 4a). Conversely, no differences in the catalase activity were observed in the parental strain in the two growth conditions. Also, no differences were observed when the parental strain was transformed with the empty vector (Fig. 4a). These results can be explained with production of H 2 O 2 in the urate oxidation reaction catalysed by zfUox but not in the reaction catalysed by PuuD. The slight increase in the catalase activity observed in the complemented strain grown in ammonia with respect to the other strains (p < 0.1) could be explained by the degradation of endogenously produced urate. Consistent with induction of catalase by the Uox activity, we found that urate-grown Δ puuD/zfOux strain was more resistant than the wild-type to oxidative stress, as observed by plating cells on agar medium supplemented with increasing concentrations of H 2 O 2 (Fig. 4b).
Finally, given the presence of a cytochrome c domain in the PuuD protein we observed the effect of an inhibitor of cellular respiration (azide) on urate degradation. Azide inhibited urate utilization by A. fabrum C58 in a dose-dependent manner, with a 50% inhibition obtained with about 0.2 mM azide. However, a similar effect was observed in the Δ puuD/zfOux mutant (Supplementary Fig. S10). As the purified zfOux protein was inhibited by azide at much higher concentrations in vitro (IC 50 ~20 mM), we concluded that the inhibition observed in vivo was due to a general effect on cellular metabolism rather than a specific effect on the enzymes of urate oxidation.

Discussion
We have described the identification of a protein family of unknown function (COG3478) as an integral membrane, cytochrome c urate oxidase (PuuD). This identification was initially suggested by the analysis of the gene association network (Fig. 1b) and then confirmed by the analysis of gene distribution (Fig. 1c), as quantitatively evaluated through the use of association rules (Table 1). This method, inspired by the logic extension 28 of the widely used correlated and anti-correlated phylogenetic profiles 26,27 , can be useful in the presence of functional associations not adequately described by pairwise relations, as exemplified here. The same analysis that enabled the functional assignment of COG3748, also provides evidence that the identification of PuuD completes the genetic repertoire of enzymes involved in urate oxidation. There is a very small number of cases in which any of the four urate oxidation genes is found in the absence of urah and urad, suggesting that there are no alternative genes encoding these enzymatic activities; on the other hand, among the 1689 different species considered there are only twelve cases in which the presence of urah and urad is not explained by the presence of uox, hpxO, hpyO, or puuD genes ( Table 1). Most of these exceptions can be explained by errors of the gene identification procedure (Supplementary Table S1). Genes encoding PuuD urate oxidase have a peculiar organism distribution, being found only in aerobic bacteria with two cell membranes -diderms, approximately corresponding to gram-negative in the traditional classification 40 . In these organisms, PuuD is the prevalent form of urate oxidase, while Uox is prevalent in monoderm (~gram-positive) prokaryotes (Fig. 5).
A genetic approach was selected for the experimental validation of the activity of PuuD proteins predicted by bioinformatics. The expression and purification of integral membrane proteins is notoriously challenging, and we did not observe protein overexpression when the gene was cloned in a heterologous (E. coli) expression system (Supplementary Fig. S11). Furthermore, the presence of a cytochrome c domain suggested that the enzymatic activity of this protein could depend on other components of an electron transfer chain. Ultimately, the results obtained with deleted and complemented strains along with bioinformatics provide conclusive evidence for the assignment of a urate oxidase activity to COG3478.
PuuD proteins are characterized by the presence of DUF989, an integral membrane domain with 8 TM helices. Although this organization can be reminiscent of transporter or channel proteins, the evidence presented here indicates that the domain is involved in urate oxidation and not in urate transport. Elements of evidence are the frequent genomic association of PuuD protein with known urate transporters (see e.g. Supplementary Fig. S1) and the complete rescue of the phenotype obtained through complementation of the PuuD deletion mutant with a cytosolic Uox. The analysis of gene context and gene distribution (see Fig. 1c) indicates that all full-length DUF989 proteins have a role in purine degradation; these proteins additionally contain a cytochrome c domain. Included in the list of DUF989 members in Pfam is a group of shorter proteins characterized by a partial match to the domain HMM and the absence of the C-terminal cytochrome. These shorter DUF989 proteins contain 4 TM helices corresponding to helices I, II, V, and VI of PuuD (Supplementary Fig. S12). This suggests that helices I, II, V, and VI -the most conserved elements in the PuuD alignment (see Fig. 3) -constitute the core of the DUF989 fold. Genes encoding the shorter DUF989 variant have a genetic context completely different from that of puuD genes (Supplementary Fig. S13). Searches based on HMM-HMM comparisons do not reveal any similarity of DUF989 with known protein domains. A distant similarity is found (E-vaule = 0.15; P-value = 10 −6 ) with DUF2269, an integral membrane domain of unknown function.
PuuD proteins are also characterized by the presence of the heme-binding cytochrome c domain. Cytochrome c is an extremely widespread electron transfer domain that mostly occurs in single-domain proteins. However, as in the case of PuuD, it is also often found fused with other protein domains. The association with DUF989 is the third most common multi-domain architecture involving cytochrome c and another Pfam domain. The most common association is with the cytochrome CBB3 domain as is found in ubiquinol-cytochrome c reductases 41 . Other common associations are with the heme binding CCP_MauG domain as is found in di-heme cytochrome c peroxidases 42 , and with the COX2 domain as is found in cytochrome c oxidases 43 . All the cited examples are integral membrane or periplasmic oxidoreductases in which the electron transfer activity of cytochrome c is an essential component of the enzyme catalysis. The presence of a cytochrome c domain in PuuD urate oxidases explains the otherwise puzzling observation that in Sinorhizobium meliloti-a PuuD-encoding organism (see Fig. 1c)-mutations of the ccmC gene coding for an integral membrane heme exporter 44 impair the ability to use purines as nitrogen source 45 . As a protein with a signal anchor, PuuD will be directed to the membrane in an unfolded state, implying that the heme must be assembled with the apo-Cyt_c domain in the extra-cytoplasmic space. Together with the notion that cytochrome c proteins typically function outside the plasma membrane in prokaryotes, the purine degradation deficient phenotype of ccmC mutants provides additional support for the N out -C out topology of PuuD proteins (see Fig. 3b). According to this topology, conserved extra-membrane regions are exposed towards the cytosol, suggesting that urate oxidation by PuuD takes place on the cytosolic side of the membrane. The electrons removed from the urate molecule could be transferred to the extra-cellular (periplasmic) Cyt_c domain through physical interaction with the transmembrane DUF989 domain. The hypothesis of an oxidation reaction catalysed by DUF908 on the cytosolic side is also consistent with the cytosolic localization of the enzymes (Urah and Urad) acting on the metastable products of urate oxidation.
The identification of an integral membrane urate oxidase employing a c-type heme suggests an unsuspected link between purine catabolism and bacterial respiration. However, we did not observe specific inhibition of the PuuD activity in the presence of a classical inhibitor of the electron transport chain. Azide was not expected to inhibit urate utilization through direct binding to the PuuD protein, but through binding to the cytochrome c oxidase, the terminal acceptor of the respiratory electron chain. The interpretation of the results obtained with the azide inhibitor is complicated by the fact that bacteria have branched respiratory pathway with multiple terminal oxidases exhibiting different sensitivity to the typical respiratory inhibitors 46,47 . Another complication arises from the known ability of azide to bind the heme cofactor of catalase 48 , which is expected also to affect the activity of the H 2 O 2 -producing zfUox. An alternative hypothesis to the electron transfer can be put forward by considering the known ability of cytochrome c to act as a scavenger of the superoxide ion (O 2 − ) 49 . The oxidation of the oxopurines hypoxanthine and xanthine, by xanthine oxidase (XO), can produce superoxide according to the reaction 50 : RH + H 2 O + 2O 2  ROH + 2O 2 − + 2H + . If the PuuD reaction mechanism imitates that of the preceding enzyme in the pathway, the presence of the cytochrome c domain would be justified by the scavenging of a reactive oxygen species produce by urate oxidation. In this hypothesis, however, the cytochrome c domain would have an accessory function, while the evidence of the invariant presence of this domain in PuuD proteins and the phenotype of the ccmC mutant rather suggest that the cytochrome is an essential component of the enzyme.
With four protein families and three different reaction mechanisms, urate oxidase provides an illustrative example of the variety of solutions that can be found in nature for the same biochemical problem 51 . Among these independent inventions, the cofactor-less Uox, which is found in all three domains of life (see Fig. 1c), is probably the most ancient solution, but not necessarily the best one. It appears an ideal solution for an enzyme localized in a single-membrane compartment specialized in the detoxification of reactive oxygen species. PuuD, with its more restricted organism distribution, was apparently invented later in a diderm ancestor (possibly a proteobacteria) through a gene fusion of a preexisting integral membrane domain with a cytochrome c (see Fig. 3 and Supplementary Fig. S9). This invention likely provided a selective advantage to diderm bacteria, as suggested by the observation that in these organisms PuuD has almost completely replaced Uox as urate oxidase.

Methods
Analysis of gene associations. The predicted association network for genes involved in urate oxidation was obtained by interrogating the String database (http://string-db.org) with a list of Cluster of Orthologous Groups (COG3648, COG2351, and COG3195) corresponding, respectively, to the uox, urah, and urad gene families. The "gene neighborhood", "gene fusion" and "gene co-occurrence" evidence were used to predict functional associations. To analyse the distribution of genes involved in urate oxidation in complete proteomes, homology searches where performed using Hidden Markov Models of the corresponding protein families in Pfam (see Supplementary Table S1) and the Hmmsearch program 52 . The results for each family were clustered using Blast best reciprocal hits (BRH) 23 and the MCL program 53 with an inflation index of 3. Genes clusters comprising reference genes involved in purine degradation (see Supplementary Table S1) were considered. The entire procedure was automated using Perl scripts. In the case of hpxO and hpyO, genes belonging to large families containing many paralogs, Scientific RepoRts | 5:13798 | DOi: 10.1038/srep13798 the clusters were further refined through phylogenetic analysis (Supplementary Fig. S14). The representation of gene distributions along the species phylogeny (see Fig. 1c) was obtained with the R software using the Ape library 54 and the heatmap.phylo function (Johan Renaudie; http://stackoverflow.com/ questions/15153202).
Association rules 20 based on gene distributions were determined using the Apriori program 21 using as input vectors of a gene distribution and its logical negation. Rules implicating COG3748 as consequent (see Table 1) were ranked according to their confidence 20 , defined as the number of cases in which the rule is correct relative to the number of cases in which it is applicable. The significance of the dependency of the antecedent and consequent of a rule was computed through the Fisher's Exact Test. Note that significance is not an indication of the correctness of the logic implication, as the opposite of a rule will receive the same p-value.
Sequence and structure analysis. The ClustalW program was used for multiple sequence alignments and tree reconstruction (based on neighbor-joining clustering of Kimura-corrected distances). Domain analysis was conducted using Pfam 36 and HHpred 37 . Figures of multiple alignments decorated with structural elements were obtained with the Espript web server 55 . Transmembrane topology was determined with Phobius 56 and TopCons 57 . The representation of membrane topology of PuuD proteins were obtained with the Protter web server 58 . Homology modelling was performed using Swiss Model (http://swissmodel.expasy.org/) and the Pymol program was used to analyse structural models.
Deleted and complemented strains. The deletion of Atu2314 gene of A. fabrum C58 was obtained with a described procedure 30 . Briefly, the upstream (1800 bp.) and the downstream (1748 bp.) sequences of Atu2314 gene were cloned into pJQ200sk plasmid (ATCC: 77483) previously digested with SacI and BamHI restriction enzymes. The pJQΔ Uox2 plasmid was confirmed by sequencing and used to transform A. fabrum C58 cells by electroporation. The recombinant cells were selected by growth on gentamicin-supplemented LB agar plate and then on sucrose-supplemented LB agar plate. Deletion of Atu2314 was confirmed by sequencing. The zfUox complemented strain was obtained by transforming Δ Atu2314 with a pSRKGm plasmid 31 modified by the insertion of the coding sequence of the uox gene from zebrafish (ZDB-GENE-030826-24) within NdeI and BamHI restriction sites. The zfUox CDS was subcloned from a previously generated pET11b expression vector (Ramazzina, I., unpublished). The presence of the zfUox CDS into the pSRKGm plasmid was confirmed by sequencing.
Bacterial growth. A. fabrum C58 strains were first growth on LB medium at 28 °C. After 20 h, bacteria were spun down, washed with M9 minimal medium without any nitrogen and carbon sources and diluted to an optical density at 600 nm (OD 600 ) of 0.05 in M9 minimal medium with NH 4 Cl (0.02%) and arabinose (0.2%) as sole nitrogen and carbon sources. The cultures were grown for another 20 h, spun down, washed twice with M9 minimal medium without any nitrogen or carbon sources and then diluted to an OD 600 of about 0.1 in M9 minimal medium containing arabinose (0.2%) plus NH 4 Cl (0.02%) or uric acid (0.02%), with the addition of IPTG 1 mM where no differently specified. Cultures were incubated with shaking at 28 °C. At intervals, OD 600 measurements were taken (Ultraspec 2000, Pharmacia Biotech).
Urate utilization assays. Bacteria inoculated at OD 600 0.05 were grown in M9 medium supplemented with arabinose and NH 4 Cl as described. After 20 h, cells were spun down, washed twice with M9 medium without nitrogen or carbon sources and resuspended to an OD 600 ~1 in M9 medium containing arabinose and without any nitrogen sources. After 90 minutes, urate (0.02%) was added to the cultures, together with IPTG 1 mM where no differently specified. At intervals, 1 ml of culture was spun down and 0.1 ml of supernatant was diluted with 0.9 ml H 2 O to measure the absorbance at 293 nm (Cary 50, Varian). Urate utilization was measured in the absence and in the presence of different concentrations of sodium azide.
To measure soluble urate oxidase activity, an aliquot of each culture was taken after 5 hours after the addition of urate and IPTG, and the cells were resuspended in 50 mM NaP pH 7.0 containing 1 mM PMSF and lysed by sonication. After centrifugation (7500 rpm for 15 min), the protein concentration of the soluble fraction was determined using the Bradford colorimetric assay, with bovine serum albumin as standard. Cell-free extracts (250 μ g) were added to 0.1 ml solutions containing 0.11 mM urate and the reaction was monitored by following the decrease in absorbance at 293 nm.
Catalase activity and oxidative stress assay. Cells were grown starting from OD 600 ~0.1 in M9 medium containing arabinose (0.02%) and NH 4 Cl (0.02%) or urate (0.02%) with the addition of IPTG 1 mM where no differently specified. After 12 hours, cells were spun down and the cell-free extracts were obtained as described above. The catalase activity of each sample was measured spectrophotometrically by following the decrease of absorbance at 240 nm 59 . One unit of enzyme was defined as the amount of enzyme catalysing the turnover of 1 μ mol of substrate per minute under the assay conditions. To measure the resistance of different strains to H 2 O 2 , cells grown in urate for 12 h were spun down, washed once with 50 mM sodium phosphate buffer pH 7.0 and diluted in the same buffer to an OD 600 of 0. PuuD cloning and expression. A PuuD expression vector was constructed by cloning the Atu2314 (PuuD) coding sequence into the pET11b vector. The Atu2314 coding sequence was amplified from A. fabrum strain C58 genomic DNA using Taq DNA polymerase with primer forward 5′ -CATATGTACGAATACGCCATTGC and primer reverse 5′ -GGATCCCACGCAGCAGAACC in standard reaction conditions. The PCR product was cloned in pGEM-T easy vector (Promega), which was then digested with NdeI and BamHI (Thermo Scientific) and ligated to a linearized pET11b vector DNA with compatible ends. The correct insertion of the PuuD coding sequence was verified by sequencing ( Supplementary Fig. S11). Expression attempts were conducted in transformed BL21 cells (Novagene). Different induction conditions with IPTG 1 mM were tested, from 4 to 16 hours and from 20 °C to 37 °C.