Cu Transport by the Extended Family of CcoA-like Transporters (CalT) in Proteobacteria

Comparative genomic studies of the bacterial MFS-type copper importer CcoA, required for cbb3-type cytochrome c oxidase (cbb3-Cox) biogenesis, revealed a widespread CcoA-like transporters (CalT) family, containing the conserved CcoA Cu-binding MxxxM and HxxxM motifs. Surprisingly, this family also included the RfnT-like proteins, earlier suggested to transport riboflavin. However, presence of the Cu-binding motifs in these proteins raised the possibility that they might be Cu transporters. To test this hypothesis, the genomic context of the corresponding genes was examined, and three of such genes from Ochrobactrum anthropi, Rhodopseudomonas palustris and Agrobacterium tumefaciens were expressed in Escherichia coli (ΔribB) and Rhodobacter capsulatus (ΔccoA) mutants. Copper and riboflavin uptake abilities of these strains were compared with those expressing R. capsulatus CcoA and Rhizobium leguminosarum RibN as bona fide copper and riboflavin importers, respectively. Overall data demonstrated that the “RfnT-like” CalT proteins are unable to efficiently transport riboflavin, but they import copper like CcoA. Nevertheless, even though expressed and membrane-localized in a R. capsulatus mutant lacking CcoA, these transporters were unable to accumulate Cu or complement for cbb3-Cox defect. This lack of functional exchangeability between the different subfamilies of CalT homologs suggests that MFS-type bacterial copper importers might be species-specific.


Results
Amino acid sequence similarity analyses of CcoA-like transporters (CalT). CcoA homologs, referred here as CalT, are found throughout the bacterial kingdom and also encoded in the genomes of some microbial eukaryotes 17 . The vast majority of proteins from each taxonomically distinct subfamily of CalT contain the motifs MxxxM in TM7 and HxxxM in TM8, which are required for Cu uptake and cbb 3 -Cox biogenesis in R. capsulatus 16 . These findings suggest that Cu import might be a ubiquitous function for this family of MFS transporters. As a first step in addressing this hypothesis, we performed a phylogenetic and genomic context analysis on the CalT subfamily members that are mainly from other Proteobacteria and exhibit highest similarity to CcoA from R. capsulatus and CalT-O (formerly RfnT) from O. anthropi, Based on the protein similarity network (Fig. 1) and the phylogenetic tree ( Fig. 2A), 11 distinct clusters (numbered 1 to 11, Figs 1B and 2A) were identified, and the amino acid contexts of their conserved MxxxM and HxxxM motifs are shown in Fig. 3. The three largest subunits of cbb 3 -Cox (CcoN, CcoO and CcoP) were found encoded in most, but not all, of these proteobacterial genomes (SI Fig. 1), suggesting that not all CalT are involved in supplying Cu for cbb 3 -Cox biogenesis. The CcoA from R. capsulatus 13 and R. sphaeroides 17 were found in cluster 1, which is shared with orthologous proteins from the Rhodobacteraceae family, whereas CalT-O was found in cluster 4 ( Fig. 2A). Due to sequence divergence, the Rhizobiales CalT proteins which are truncated at the C-terminus and whose corresponding genes are located next to the cbb 3 -Cox biogenesis (ccoNOQP-ccoGHIS) cluster 17 , were not connected to the network. However, when these sequences were included in the phylogenetic analyses, they were found most closely related to proteins within cluster 11 ( Fig. 2A, cluster 11B), instead of cluster 1, which contains members experimentally shown to be required for cbb 3

-Cox biogenesis.
Genomic context of CalT. Next, a neighborhood analysis was performed to identify proteins other than cbb 3 -Cox that might be functionally linked to CalT. Functionally coupled genes tend to cluster physically in bacterial genomes, and both frequency and conservation of gene clustering across evolutionarily distant genomes can be used to detect functional coupling 22 . We used a window of three genes upstream and downstream of each calT gene encoding a CalT protein from the similarity network to analyze the extent to which the neighboring genes are conserved at the genus, family, and order levels of taxonomy. At the genus level, we identified 605 protein family (Pfam) domains or domain fusions (referred to as neighbors) that were seen in at least two different genera. We ranked these domains by number of genera, excluded putative transcription factors and transporters, and further analyzed the top 17 neighbors (each found in 30 or more genera) (Methods). These neighbors could be arranged into three main neighborhoods (SI Table S1). The first neighborhood N1 (yellow squares in Fig. 2A), which contains CcoA from Rhodobacter species, was composed of one or more of nine genes including the putative DNA repair (alkA, PF00730) and esterase (ypfH, PF02230) proteins (Fig. 2B). The second neighborhood N2 (red circles in Fig. 2A) contained the genes encoding a FabG-like reductase (PF13561) and/or a putative Zn-dependent Scientific RepoRts | (2019) 9:1208 | https://doi.org/10.1038/s41598-018-37988-4 dehydrogenase (PF00107-PF08240). The third neighborhood N3 (green triangles in Fig. 2A) contained genes encoding a BamE-like outer membrane protein assembly factor (PF04355), a putative ubiquinol-cytochrome c oxidoreductase (cytochrome bc 1 complex) chaperone (PF03981), a putative thiamine-monophosphate kinase (PF00586-PF02769) and/or a putative 6,7-dimethyl-8-ribityllumazine synthase (ribH involved in riboflavin biosynthesis, PF00885). These main neighborhoods (yellow squares, red circles and green triangles) are indicated in Fig. 2A, and all neighboring genes are listed in SI Table S1 (Tree and Neighborhood sheets). Of the neighborhoods, only N2 and a putative methyl transferase from N1 were enriched at the family and order levels. The RBP protein RibH, BamE-like outer membrane protein assembly factor, and ubiquinol-cytochrome c oxidoreductase chaperone from N3 were enriched at the family, but not at the order level (SI Table S1).
Positional clustering of RfnT-like CalT proteins with RBP genes. The neighborhood N3 captured the positional clustering that originally led to the identification of RfnT in M. loti, S. meliloti and A. tumefaciens, and prediction that these might be riboflavin transporters 18 . Many more bacterial genomes have been sequenced since that original analysis, and our data show that positional clustering between RBP genes and those encoding RfnT-like CalT proteins is conserved only in Rhizobiales, in a small subset of Rhodobacterales, and in Rhodospirillales (SI Figs S1A and S2A). Current data indicate that the proximity of these rfnT-like calT to ribH is mainly observed in clusters 5 and 6, and near the base of the clusters 3, 4, 10 and 11. The core unit, seen in cluster 5, is composed of the RBP gene ribH, followed by nusB encoding a subunit of the global transcriptional antitermination complex, and finally by calT. In addition, the thiL gene encoding thiamine-monophosphate kinase (vitamin B1 biosynthesis) and some other presumably functionally unrelated genes separating rfnT from ribH-nusB (clusters 3, 4, 10, and 11) were frequently seen (SI Fig. S2B). In most cases, RBP genes other than ribH are also conserved upstream of the core ribH-nusB-rfnT unit (SI Fig. S2B). Thus, the genomic proximity of the genes encoding the RfnT-like CalT to RibH-related proteins is not general, but is only seen in a subset of the clusters and in relatively closely related bacteria. Similarly, the previously identified group of Rhizobiales CalT is the only example where calT was located next to the cbb 3 -Cox biogenesis genes (ccoNOQP-ccoGHIS) 17 .
Cu-related proteins found in neighborhoods containing CalT. Given the experimentally defined role of CcoA as being a Cu transporter 13,14 , we searched within genomic neighborhoods for genes encoding either cuproproteins or other proteins involved in Cu homeostasis. Noticeably, the previously identified group of Rhizobiales CalT (cluster 11B) was an example where calT could be found located next to the cbb 3 -Cox biogenesis genes (ccoNOQP-ccoGHIS) 17 . In addition, calT homologs were observed next to a gene containing a cytochrome_CBB3 domain (PF13442), similar to subunit III of cbb 3 -Cox, in two unclassified Pelagibacteraceae bacteria and Pelagibacter sp. HIMB1321 (cluster 7), and Bradyrhizobium sp. LMTR 3, Bradyrhizobium icense and Bradyrhizobium erythrophlei (cluster 3 in Fig. 4). In the case of Bradyrhizobium spp., this gene putatively encodes SoxX and is found in a putative sulfur-oxidizing gene cluster. This finding is significant as the SoxAX from Starkeya novella was shown to contain a mononuclear Cu 2+ center 23    ). Interestingly, the O. anthropi genome contains two calT genes in cluster 4; one corresponds to the earlier described rfnT 19 , and a paralog is located near a putative Cu chaperone gene. In addition, several clusters contained copA, csoR, cueR or mco genes that encode proteins involved in Cu-detoxification (Fig. 4). Out of the 1635 calT genes analyzed, only a few were observed proximal to additional genes also encoding Cu-responsive or Cu-homeostasis related proteins, such as CusF (Cephaloticoccus primus and Cephaloticoccus capnophilus), PCuAC (Ventosimonas gracilis), SCO1 (Pseudomonas tolaasii and Pseudomonas fluorescens), CutA1 (divalent ion tolerance protein, PFAM3091) (Pseudooceanicola marinus and Pseudooceanicola antarcticus) and Cu/Zn superoxide dismutase (Epibacterium ulvae).
Heterologous expression of Rfnt-like Calt proteins in E. coli. Although in our analysis 190 (out of 1635) CalT members from Proteobacteria are found near ribH, the phylogenomic analysis alone could not definitively distinguish putative Cu transporters from putative riboflavin transporters. Indeed, phylogenetic clusters that contained calT genes clustering with the RBP genes also contained homologs nearby the Cu homeostasis genes (cluster 3, 4 and 11). Thus, to further define the functions of CalT family members, we tested experimentally the ability of CcoA-like and RfnT-like members to transport riboflavin and Cu, respectively. We chose three Rhizobiales sequences that are encoded next to the cbb 3 -Cox biogenesis gene cluster, is shown. Background shading corresponds to separate clusters (1 to 11), and leaves corresponding to the proteins experimentally examined in this study are indicated by a red arrow. Whether a CalT is encoded by a gene found in one of the three main genomic neighborhoods is indicated with either a yellow square (N1), a red circle (N2) or a green triangle (N3) according to the legend. A star (Heavy Metal Associated, HMA) indicates that the corresponding gene is found near a putative Cu homeostasis gene. A full list of all proteins analyzed and the related information can be found in SI    Table S2) expressed Myc-tagged CalT-O, CalT-R and CalT-A (M r ranging from 37 to 40 kDa) proteins, as detected by immunoblot analysis of whole cell extracts using anti-Myc antibodies (Fig. 5A). The same E. coli strain producing R. capsulatus CcoA (running as ~37 kDa) 16 was used as a control. Similarly, whole cells and chromatophore membranes of a R. capsulatus strain lacking CcoA (ΔccoA) harboring appropriate plasmids with ccoA, calT-O, calT-R and calT-A (SI Table S2) also contained comparable amounts of the respective proteins ( Fig. 5B), indicating that they were expressed and inserted in the cytoplasmic membrane in these species.

CalT-O, CalT-R and CalT-A do not complement E. coli ΔribB mutant for riboflavin auxotrophy.
E. coli has no known riboflavin transporter, but produces riboflavin via its endogenous RBP, which includes the ribB gene encoding the 3,4-dihydroxy-2-butanone-4-phosphate synthase 20,24 . Thus, an E. coli ΔribB mutant cannot grow on LB medium unless supplemented with a large amount (500 µM) of riboflavin, which is thought to diffuse passively across the membrane 20 . In contrast, heterologous expression of an efficient riboflavin uptake transporter, such  Table S2) expressing either CcoA or RibN or the CalT-O, -R or -A. pBAD corresponds to the same cells carrying an empty expression vector (SI Table S2). In each case, the uptake assays were repeated at least three times using at least two independently grown cultures, and statistical analysis was performed using the Student's t test, with p < 0.01 as the level of significance between RibN (*) and the other strains. (E) 67 Cu uptake kinetics were performed using E. coli strain LMG194 expressing the RfnT-like CalT proteins from R. capsulatus CcoA, O. anthropi (CalT-O, formerly called RfnT), R. palustris (CalT-R) and A. tumefaciens (CalT-A), or the riboflavin transporter RibN. All uptake assays were performed at 37 °C and on ice as described in Methods, and in each case the activities detected with cells kept on ice were subtracted from those incubated at 37 °C. Of these corrected values the background activity measured with the E. coli strain carrying pBAD/Myc-His (pBAD) were subtracted and plotted in function of time. Each assay was repeated at least three times using multiple independently grown cultures, and statistical analysis was performed using the Student's t test, with p < 0.01 as the level of significance between RibN and the other strains (*). as the Rhizobium leguminosarum RibN, enables growth of an E. coli ΔribB mutant on LB medium containing low amounts (2.5 µM) of riboflavin 25 .
In order to assess whether heterologous expression of CalT-R, -O and -A could confer riboflavin uptake activity in E. coli, plasmids encoding these orthologs were transformed into the E. coli ΔribB mutant (BW25141::ΔribB, SI Table S2) using LB plates containing 500 µM riboflavin. These transformants were then tested for growth on LB plates with low concentration of riboflavin (2.5 µM), in the absence and presence (0 to 2%) of L-Ara. A plasmid expressing the R. leguminosarum RibN bona fide riboflavin importer (SI Table S2) was used as a positive control 25 . Neither the E. coli ΔribB mutant, nor its derivatives carrying the calT-O, -R and -A genes were able to grow on 2.5 to 10 µM of riboflavin containing plates, irrespective of the presence of L-Ara, unlike those carrying ribN (Fig. 5C). As these CalT proteins were expressed in E. coli (Fig. 5A), their inability to rescue the growth on low concentration of riboflavin suggested that they could not confer efficient riboflavin uptake to sustain growth of E. coli, unlike RibN. Similar results were also obtained with a plasmid (pBK68) carrying R. capsulatus ccoA, indicating that CcoA also did not have such uptake activity (Fig. 5C).
During these experiments we observed that the E. coli ΔribB mutant (SI Table S2), and its derivatives expressing various CalT yielded spontaneous revertants that regained riboflavin-independent growth ability on LB medium in the absence, or presence of 2.5 µM of riboflavin (Fig. 5C, e.g., ΔribB expressing CcoA or CalT-O). These observations suggested that similar events might have occurred during the earlier work with O. arthropi gene 19 .

Neither CcoA nor RfnT-like CalT exhibit riboflavin uptake activity in E. coli. E. coli cells producing
CcoA or RfnT-like CalT were tested for their ability to take up radioactive 3 H-riboflavin. The data showed that 3 H-riboflavin was taken up readily by the E. coli cells expressing RibN, but not by those expressing the three CalT homologs or CcoA (Fig. 5D). Moreover, in the case of CcoA, which is known to transport Cu, addition of Cu (100 μM) did not affect its inability to take up 3 H-riboflavin. We concluded that neither CcoA nor the RfnT-like CalT members exhibited any efficient riboflavin uptake activity in E. coli, in agreement with their lack of complementation of the E. coli ΔribB strain for auxotrophy at low riboflavin amounts, suggesting that the earlier observed growth with O. anthropi rfnT 19 (i.e., calT-O) might have been due to spontaneous reversion.
the Rfnt-like Calt proteins mediate 67 Cu uptake activity in E. coli cells. The conservation of the CcoA Cu-binding motifs (MxxxM and HxxxM) in the RfnT-like CalT proteins led us to investigate whether they could import Cu into E. coli cells, like R. capsulatus CcoA 15,16 . Time dependent 67 Cu uptake activities of appropriate strains were measured using whole cells grown in the presence of 0.5% L-Ara. As a control, E. coli cells expressing wild-type R. capsulatus CcoA exhibited significantly higher amounts of 67 Cu uptake than the same cells lacking CcoA (i.e., CcoA-independent 67 Cu uptake background, Methods), as reported earlier 16 . Remarkably, E. coli cells expressing CalT-O or -R or -A also showed robust 67 Cu uptake activities, whereas the same E. coli (or a ΔribB derivative) cells expressing RibN had no detectable 67 Cu uptake activity (Fig. 5E). Therefore, we concluded that RfnT-like CalT proteins have Cu, but not riboflavin, uptake activity when expressed in E. coli, similar to CcoA. We note that the amounts of 67 Cu accumulated in E. coli cells expressing different CalT proteins were slightly different. This point being out of the scope of this work, the amounts and affinities for Cu of those transporters were not studied further.

Rfnt-like Calt proteins do not complement the R. capsulatus
ΔccoA mutant for cbb 3 -Cox defect. Considering that CcoA is a Cu importer required for cbb 3 -Cox biogenesis in R. capsulatus 14,15 and R. sphaeroides 17 , and that the RfnT-like CalT proteins can also import Cu, their ability to complement the R. capsulatus ΔccoA mutant for its cbb 3 -Cox biogenesis defect was tested. Appropriate plasmids expressing CalT-O or -R or -A were conjugated into a R. capsulatus strain lacking CcoA. In parallel, a similar plasmid (pBK69) expressing wild-type R. capsulatus CcoA was used as a control (SI Table S2). The trans-conjugants were first tested for the presence of cbb 3 -Cox activity using the Nadi staining procedure (Cox activity dependent conversion of α-naphthol to indigo blue 26 ). Colonies containing CcoA turned blue (i.e., Nadi + phenotype) immediately (<30 sec), while those with the RfnT-like CalT proteins remained unstained upon longer (>10 min) exposure times (Fig. 6A). In addition, supplementation of the growth medium with 1 to 500 nM Cu, to increase Cu availability (in case of the lower uptake activities, or Cu affinities of CalT's tested) was not efficient. Unfortunately, use of higher amounts of Cu supplementation was not informative because of the phenotypic suppression of a ΔccoA mutant for cbb 3 -Cox activity caused by μM amounts of external Cu 13,15 . However, immunoblot analyses of membrane preparations from the trans-conjugants using anti-myc antibodies showed that they all contained membrane-bound RfnT-like CalT proteins at levels comparable to those of CcoA (Fig. 5B). These findings suggested that, although produced and inserted into the membrane, the RfnT-like CalT proteins were unable to yield any active cbb 3 -Cox. Indeed, the trans-conjugants expressing CalT-O or -R or -A had very low levels of cbb 3 -Cox activity (~3-5%) compared with the R. capsulatus ΔccoA complemented with CcoA (100%) (Fig. 6B). Moreover, determination of the total cellular amounts of Cu associated with cells expressing CalT-A showed no accumulation of cellular Cu, unlike those containing CcoA (Fig. 6C) (see also SI Fig. S3 for the metal contents of these cells), suggesting that it was inactive in R. capsulatus membranes. Overall data showed that although the RfnT-like CalT proteins exhibited Cu uptake activity in E. coli cells, they were unable to complement a R. capsulatus strain lacking CcoA for cbb 3 -Cox biogenesis.

Discussion
During our previous comparative genomic study of CcoA required for cbb 3 -Cox biogenesis [13][14][15][16] in R. capsulatus and R. sphaeroides 17 we noticed that a subgroup of the CcoA homologs (CcoA-like transporters or CalT) included the RfnT proteins previously predicted to transport riboflavin [18][19][20] . Moreover, the conserved (MxxxM and HxxxM) motifs of CcoA, which are associated with Cu import and cbb 3 -Cox biogenesis 16  in this subgroup. This similarity led us to further investigate this subfamily in order to probe whether the different members of the CalT family could transport different substrates such as Cu or riboflavin. We first divided the CalT family into 11 clusters based on the phylogenomic and genomic context analyses. While CcoA from R. capsulatus and R. sphaeroides belongs to a distinct cluster of proteins (cluster 1) shared with orthologs from other Rhodobacteraceae, we were unable to make a clear phylogenetic distinction between putative Cu transporters and putative riboflavin transporters. In the same protein cluster (e.g., clusters 3 and 4) we found calT genes that were located proximal to HMA-domain containing Cu chaperones involved in Cu response or detoxification, in support of a Cu-related function, whereas in other genomes their orthologs were next to RBP gene clusters. Thus, to establish the substrate specificity of different CalT subfamilies with respect to Cu and riboflavin we used an empirical approach. Three RfnT-like CalT proteins from three cbb 3 -Cox encoding proteobacterial species were introduced into appropriate E. coli and R. capsulatus mutants. Protein expression, phenotypic complementation and radiolabeled Cu and riboflavin uptake kinetics data showed that CalT-O, -R and -A from O. anthropi (cluster 4), R. palustris (cluster 3) and A. tumefaciens (cluster 5), respectively, were MFS-type Cu transporters just like R. capsulatus CcoA, and not efficient riboflavin transporters. Conceivably, currently unknown link(s) between Cu and riboflavin might exist, and these proteins may transport Cu and/or riboflavin at much higher concentrations or under specific conditions that are different than those used here. In any event, our findings validated the conservation of the MxxxM and HxxxM motifs in these CalT subfamily members, and suggested that this motif may be a good predictor of Cu importers among the MFS transporters. Currently, this point is further pursued using appropriate strains and species. Most bacteria have an active RBP and are able to synthesize riboflavin de novo 20,24 , yet some species can also take up riboflavin from their environment via specific riboflavin uptake transporters 20 . Several such transporters have been described, and among them the energy coupling factor (ECF)-type RibU 28,29 , PnuX/RibM [30][31][32] , and RibN 19,25 have been shown to transport riboflavin or its derivatives, whereas some others (e.g., ImpX and RibXY) are less studied. With the exception of the well characterized ECF-type RibU 29 , very little is known about the structural properties of bacterial riboflavin transporters and the specific motifs involved in substrate binding. Initially, rfnT was proposed to encode another riboflavin transporter based on its physical proximity to the RBP genes in Rhizobiales genomes 18,19 . However, neither the expression of CalT-O in the ΔribB mutant, nor an ability to take up riboflavin was examined 19 . During our analyses, we found that the E. coli ΔribB mutant (BW25141::ΔribB) used in previous studies reverted spontaneously to riboflavin protrophy. Similarly, the ΔribB derivatives expressing CalT-O, -R and -A yielded riboflavin prototrophic revertants, raising the issue of whether the RfnT-like CalT proteins were efficient riboflavin transporters. Indeed, 3 H-riboflavin uptake experiments showed that cells harboring these proteins (and even CcoA) were unable to take up riboflavin, unlike a bone fide riboflavin transporter (e.g., R. leguminosarum RibN 25 ). Instead, these transporters also exhibited Cu-transport activity in E. coli like R. capsulatus CcoA.
The unusual association of some calT subfamilies with RBP genes might suggest a possible, but currently unknown role for riboflavin in Cu homeostasis or Cu in riboflavin biosynthesis, or even cytochrome biogenesis in bacteria. Notably, some calT genes located in neighborhood N3 were found to be associated with a gene encoding a putative chaperone of ubiquinol-cytochrome c oxidoreductase. Moreover, a recent work using transcriptomics suggested that RibN-imported riboflavin might be involved in c-type cytochrome biogenesis in Vibrio cholerae 21 .
An unexpected finding was the inability of the RfnT-like CalT proteins to complement the cbb 3 -Cox defect of a R. capsulatus mutant lacking CcoA. Considering the successful heterologous production and membrane localization of CalT-O, -R and -A in R. capsulatus, and their Cu uptake activities seen in E. coli, the basis of this observation remains unclear. A possibility is that the RfnT-like CalT subfamily members might be inactive for unknown reason(s) for Cu uptake in R. capsulatus despite their competence in E. coli. The ICP-MS data suggested that R. capsulatus cells producing CalT-A do not accumulate Cu unlike those containing CcoA. A different possibility is that the Cu uptake and delivery pathways during cbb 3 -Cox biogenesis via the CalT family members might be species-specific. If so, these proteins (or the chemical nature of Cu cargo) might be incompatible to interact with their heterologous partner(s) to convey Cu to its ultimate destination, rendering them non-interchangeable for cbb 3 -Cox biogenesis. Similar diversity occurs with cytoplasmic Cu chaperones in lower eukaryotes 33 . Ongoing work aiming at inactivating a RfnT-like CalT member (i.e., CalT-A) in a genetically tractable species like A. tumefaciens, and defining its effect(s) on cbb 3 -Cox biogenesis and Cu transport might shed further light to some of these issues. Moreover, the role of CalT in the provision of Cu to other cuproproteins also remains a possibility as not all CalT-encoding genomes encode a cbb 3 -Cox.
Finally, the biogenesis of cbb 3 -Cox is a complex process that is not yet fully understood 12,27 . It involves an increasing number of Cu chaperones and transporters, including SenC (PrrC/Sco homolog) and PccA (PCuAC homolog) 34-36 that work collaboratively 37 with the dedicated P 1B -type transporter CcoI (also known as CtpA/ CopA2) [38][39][40] . The spatial and temporal order(s) with which these Cu chaperones handle Cu, and interact with each other, is only now emerging 27 . In the absence of a three-dimensional structure for a CalT member, it is difficult to speculate about the amino acid residues that might be responsible for the observed differences. Nonetheless, sequence alignments show salient differences located around the cytoplasmic and periplasmic loops between the TM6 -TM7 and TM11 -TM12 of CalT members, respectively (SI Fig. S4). The occurrence of amino acid residues that are conserved in the CcoA and not in the RfnT-like CalT subfamilies, and vice versa, might be important in defining their specificity.
In summary, this study further defined the extended family of CalT in Proteobacteria and demonstrated that the RfnT-like CalT subfamily members are not riboflavin transporters, but they are rather bona fide Cu importer members of the Cu Uptake Porter family of TCDB 1 . Moreover, the occurrence of the conserved MxxxM and HxxxM motifs among this family appears to be a reliable predictor of Cu import activity. Whether all members of the CalT family exclusively provide Cu to the Cu B center of cbb 3 -Cox, or to other cuproproteins as well remains to be seen.

Methods
Bacterial strains and growth conditions. The bacterial strains and plasmids used in this study are listed in SI Table S2. Standard molecular biology techniques were used 41 . E. coli strains were grown in LB medium at 37 °C supplemented as needed with ampicillin (Amp), chloramphenicol (Cm), tetracycline (Tet) and kanamycin (Km) at final concentrations of 100, 30, 12.5 and 50 μg/mL, respectively 42,43 . E. coli ΔribB strains were grown in the presence of 500 μM riboflavin, because they are unable to grow at lower concentrations (e.g., 2.5 μM) unless they express a functional heterologous riboflavin transporter (e.g., RibN) 25 . Complementation of this auxotrophic growth phenotype of the ΔribB strain was used to assess the ability of a gene product to transport riboflavin upon heterologous expression. E. coli strains containing pBAD/Myc-His A plasmid derivatives were grown overnight in LB medium with 0.5% L-arabinose (L-Ara) to express L-Ara-inducible genes. The R. capsulatus SE8 (ΔccoA) strain derivatives were grown in enriched medium (MPYE) at 35 °C supplemented with 2.5 μg/mL Tet. The L-Ara-inducible pBAD-pRK415 plasmid derivatives were conjugated into R. capsulatus by tri-parental mating using the helper plasmid pRK2013 42,44 and cells were grown overnight in the presence of L-Ara (0.5% to 2% as needed) 16 .
Construction of the expression plasmids. R. palustris calT gene (calT-R) was amplified using primers RPA-F and RPA-R (SI Table S3), and the resulting 1248 bp PCR fragment was digested with HindIII and KpnI and cloned into pBAD/Myc-His A vector, yielding plasmid pYZ02 encoding a C-terminally Myc-His-fused CalT-R (SI Table S2). Similarly, a 1224 bp PCR fragment containing the calT gene from A. tumefaciens (calT-A) was amplified using primers Atu-F and Atu-R, and cloned into pBAD/Myc-His A as above, yielding plasmid pYZ03 with a Myc-His tagged CalT-A. The calT gene (previously called rfnT 18 ) from O. anthropi (calT-O) was amplified using primers OanT-F and OanT-R (SI Table S3) resulting in a 1197 bp PCR fragment that was cloned into pBAD/Myc-His A digested with EcoRI and KpnI, yielding the plasmid pYZ09 with a Myc-His tagged CalT-O (SI Table S2). Plasmids pYZ02 and pYZ09 were digested with NsiI, and ligated into the broad-host-range vector pRK415 digested with PstI (compatible cohesive ends with NsiI), yielding pYZ07 and pYZ11, respectively (SI Table S2). As the wild-type calT-A contains an internal NsiI (ATGCAT) site, the adenine of the NsiI site was replaced by a cytosine (ATGCCT) using the Q5 Site-Directed Mutagenesis Kit (NEB, Beverly, MA). Plasmid pYZ03 and primers AtuN-F and AtuN-R were used, yielding plasmid pYZ08 that was digested with NsiI and ligated into pRK415, digested with PstI, yielding plasmid pYZ13 (SI Table S2).  45 . Protein concentrations were determined using the bicinchoninic acid assay (Sigma Inc.; procedure TPRO-562). Immunoblot analysis to detect the presence of the c-Myc epitope using either E. coli or R. capsulatus cell extracts or chromatophore membrane proteins (R. capsulatus) was done as in 16 . The presence of the CcoA or CalT in cell lysates of E. coli, and in the membrane fraction of R. capsulatus was confirmed by immuno-detection using anti-Myc monoclonal antibody and horseradish peroxidase conjugated anti-mouse IgG. Signal was detected using the Supersignal West Pico chemiluminescence substrate.

Whole cell lysates and chromatophore membrane preparation, SDS-PAGE and immunoblots.
In vivo and in vitro cbb 3 -Cox activity. Tet R derivatives of R. capsulatus SE8 (ΔccoA) containing plasmids pYZ07, pYZ11, pYZ13 and pBK69 (SI Table S2) were purified on appropriate MPYE plates under respiratory growth conditions, and their cbb 3 -Cox activities visualized qualitatively with the Nadi staining procedure 26 . Staining of the colonies was done as previously described 17 . The cbb 3 -Cox activities were measured by monitoring oxidation of reduced horse heart cyt c (Sigma Inc.) using chromatophore membranes according to 17,46 . Whole cells 67 Cu and 3 H-riboflavin uptake assays. The Cu uptake assays were performed according to 15 . Radioactive 67 Cu (half-life of ~62 hours) was obtained from the DOE-Brookhaven National Laboratory (NY). E. coli strain LMG194 containing the pBAD/Myc-His derivatives encoding R. capsulatus ccoA (pBK68) or various calT (pYZ02, pYZ03 and pYZ09), or R. leguminasorum riboflavin transporter RibN (pGRibN) 25 (SI Table S2) were grown in 10 mL of LB supplemented with 0.5% L-Ara until an OD 600 of 0.5. Similarly, the E. coli strains BW25141::ΔribB (ΔribB derivative of BW25141) 19 and BW25141::ΔribB/pGRibN were grown in the presence of appropriate amounts of riboflavin as control strains. Cells were collected, washed with 50 mM sodium citrate, pH 6.5, 5% glucose buffer (uptake assay buffer) and re-suspended in 1 mL of the uptake assay buffer. Optical density at 600 nm was determined. For each assay, a total of 7.5 × 10 8 cells per 500 µL of total assay mixture (1.0 A 600 = 5 × 10 8 cells/mL) were used. Cells were incubated for 10 min either at 35 °C or on ice, before each assay. Cu uptake was initiated by addition of 10 6 cpm of 67 Cu (determined immediately before use) to the cell suspension. At each time point (0, 1, 2, 5, and 10 min), aliquots of 50 µL of assay mixture were collected and combined with 50 µl of CuCl 2 (1 mM) and 50 µL of EDTA (50 mM, pH 6.5) to stop the uptake activity, and stored on ice. The aliquots were then centrifuged, and cells washed twice with 100 µL of ice-cold EDTA ( cell mixtures kept on ice during the assays were subtracted from those obtained at 35 °C, and plotted in function of time. For 3 H-riboflavin (Moravek Inc., Brea, CA) uptake assays, E. coli strains were grown to an OD 600 of 0.4-0.6, washed with LB medium and re-suspended in LB medium to a final OD 600 of 12. A total of 7.5 × 10 8 cells were diluted with uptake assay buffer to a final volume of 500 µL. Assay mixtures were pre-incubated either at 37 °C or kept on ice for 10 min before initiating the assay by addition of 2.5 μM riboflavin containing 2 μCi of 3 H-riboflavin. At each time point (0, 2, 5, 10, and 20 min), an aliquot of 50 µL was taken and mixed with 50 µL of ice cold stopping solution (100 μM non-radioactive riboflavin in LB medium) and stored on ice. Cells were then pelleted, washed with 500 µL of stopping solution and re-suspended in 1 mL of scintillation liquid, and counted using a scintillation counter (Tri-Carb 2900 TR, Perkin Elmer).
Determination of total cellular Cu contents using ICP-MS. Samples for determination of total cellular Cu contents were prepared as described earlier 15 . Briefly, R. capsulatus strains were grown by respiration in 1 L of enriched MPYE medium prepared with metal-free water (stirred at room temperature with Chelex100 at a concentration of 5 g/L for 1 hour) to an OD 630 of 0.8-0.9. Cells were harvested by centrifugation and washed three times with metal-free 20 mM Tris-HCl pH 8.0 and once with ice cold metal free water. Cell pellets were lyophilized to complete dryness. A total of 50 mg of dry cell powder per sample was digested in 1 ml trace-metal grade nitric acid (Sigma) at 65 °C. To obtain a corresponding blank, the volume of the cell powder was replaced by milli-Q grade water (ultrapure) and treated the same as the samples. The digested samples were then diluted with milli-Q grade water to a final concentration of 1 mg/ml cell powder. Total metal content was measured by ICP-MS (Nexion 350D, Perkin Elmer equipped with an Element Scientific prepFAST M5 autosampler) using quadruplicate digested samples for each strain.
Comparative genomic and phylogenetic analyses. The protein similarity network was constructed using the EFI-EST tool (http://efi.igb.illinois.edu/efi-est/) 47 with an alignment score of 75. CalT proteins that were not connected to the main network hub were deleted and not included in further analyses. A full list of identified CalT members was published previously 17 , and the sequences used in this study are available in SI  Table S1. The network was visualized with the yFiles organic layout provided with the Cytoscape software (http:// www.cytoscape.org) 48 . The nodes in the network were colored either by taxonomy as provided by the UniProt database 49 , by cluster as determined by the phylogenetic analysis, or by the presence of proteins containing CcoN (IPR004677), CcoO (IPR003468) and CcoP (IPR004678 or IPR032858) as determined with the Interpro database 50 . The phylogenetic analysis was performed using NCBI's COBALT 51 for sequence alignment and IQ Tree 52 as implemented on the CIPRES web portal 53 with 1000 bootstrap replicates 54 . In addition to the sequences found in the network, 12 Rhizobiales CalT sequences, which are encoded by genes found near the cbb 3 -Cox biogenesis cluster were added to the phylogenetic analysis. Before tree building, the multiple-sequence alignment was edited to remove positions with a quality score less than 826 55 and those sequences that did not contain the MxxxM and HxxxM motifs. Sequence logos were built with Skylign 56 using the same multiple-sequence alignment used for the phylogenetic analysis.
Gene neighborhoods (a window of three genes upstream and downstream of each gene encoding a CalT protein from the similarity network) were retrieved using the EFI-GNT tool (https://efi.igb.illinois.edu/efi-gnt/). At the genus level, we identified 605 protein family (Pfam) domains or domain fusions (referred to as neighbors) that were seen in at least two different genera. We ranked these domains by number of genera and set a threshold at 30 individual genera, which resulted in 19 neighboring PFam domains. Of these, transcription factors, PF07690 (MFS_1) and PF00005 (ABC_tran) were excluded from further analysis because they are particularly large multi-functional families. The remaining 17 neighboring PFam domains could be collapsed into three main neighborhoods (SI Table S1). Statistics analysis. The data are presented as means ± S.D, and statistical analysis was performed using the Student's t test, with p < 0.01 as the level of significance and indicated in the figure legends.