Phylogenetic analysis using Neighbor-Joining (NJ) method was carried out on the 273-protein alignment, resulting in the tree shown in Figure 3 (see Methods). To avoid biased group distribution due to the extended N or C termini typical of some CDF transporters (e.g. ScMsc2, HsZTL1, HsZnT5), only the CDF domain was taken for the phylogenetic tree reconstruction. In particular, the start was set about 40 amino acids upstream from the first TMD, which corresponds to the first amino acid of AtMTP4, and the end was set about 40 amino acids after the Cation Efflux domain, corresponding to the last amino acid of ScZrg17. At the N terminus, with respect to the alignment, about 50% of the proteins started upstream from the first methionine of AtMTP4, with an average of 122 residues. At the C terminus, the alignment showed more conserved sequences and about 80% of proteins passed after the stop codon of Zrg17, with an average of 25 residues. Additional phylogenetic analysis using the Maximum Parsimony (MP) method (MEGA 3.1) was performed and gave congruent topologies confirming the confidence of the clustering (data not shown). Based on tree topology and confidence on nodes (bootstrap values >50), we defined 13 clusters containing characterised sequences (Figure 4), which were named by referring to one characterised member (usually the first identified). The ZnT9-like cluster contained murine and human ZnT9, previously called HUEL, that also have significant homology with a DNA-binding domain and a nuclear receptor interaction motif. ZnT9/HUEL was demonstrated to be predominantly localized to the cytoplasm and to translocate to the nucleus during the S phase of the cell cycle [34]. To avoid misleading labelling, the cluster containing the Mn²⁺ transporter ShMTP1 was named MTP8 (AtMTP8 is the orthologue of ShMTP1 in Arabidopsis thaliana), following the nomenclature proposed for Arabidopsis at the PlantsT World Wide Web site [35]. Eleven of these 13 clusters clustered into three groups (Figure 3), which are detailed below. We hypothesized that proteins belonging to the same group may share the same metal specificity. Thus, for those transporters whose substrate is not known, the position within one of these groups may suggest the metal specificity.

We further searched for residues that could represent candidate sites of functional divergence of the CDF groups. Remarkably, Mn-CDF sequences could be differentiated by the consensus sequence DxxxD (x = any amino acid) in TMD V, which appears as HxxxD in all other CDF sequences from the Zn- and Fe/Zn-CDFs, D being the highly conserved aspartate residue important for CDF function ([14,29,30] and Table 1). In Zrg17-like and in ZnT9-like clusters, sequences did not match with any of these patterns, since TxxxS/T and VxxxD patterns were found in the same topological positions, respectively (Figure 4). When the TMD V region was excluded from the alignment, the tree topology and bootstrap values did not change appreciably, suggesting that the close phylogenetic relationships were not only based on TMD V amino acid conservation. Since the group-conserved H/D residue is four residues distant the highly conserved aspartate within TMD V, the two residues will interact on the same face of the α-helix. These residues may play important but different roles in the CDF groups. In EcFieF the corresponding residue, H153, is one of the four amino acids that directly coordinates ions during binding and transport and it faces the cytosolic side of the membrane [31]. The cysteine thiolate, histidine imidazole, and aspartate/glutamate carboxylate have similar properties as ligand groups for Fe²⁺ and Zn²⁺ metals, while Mn²⁺, like other hard metals, prefers carboxylate toward N-imidazole ligand group [50]. In the Zn-CDF PtdMTP1 the corresponding histidine located on TMD V was changed to aspartate. The H260D substitution in PtdMTP1 led to the loss of any detectable function in Zn²⁺ homeostasis, without changing the protein localisation, demonstrating that these aspartate and histidine residues do not have the same role (Figure 2). However the H260D variant of PtdMTP1 was not able to complement the manganese-hypersensitive strain pmr1Δ.

In TMD II a histidine (Zn-CDF)/aspartate (Mn-CDF) residue is conserved within each specific group, although in the Fe/Zn-CDF group, both aspartate (FieF-like cluster) and histidine (MMT-like and WmFieF-like clusters) are present (Figure 4). These residues reside four amino acids upstream from the highly conserved aspartate in TMD II and in EcFieF they correspond to D45, one of the four metal coordination residues that is located close to the non cytosolic side of the membrane [31].

Another key structural feature is the presence of a histidine-rich region in most CDFs, except the Mn-CDF, the ZnT9-like, FieF-like, and WmFieF-like transporters (Figure 4). Most Zn-CDF transporters usually display histidine-rich (or histidine/serine-rich) regions at N, C termini in and/or between TMDs IV and V. The function of the histidine-rich loop was demonstrated in AtMTP1, where the mutated version lacking the histidine-rich loop was unable to bind Zn²⁺ [26], while in EcZitB neither the N-terminal nor the C-terminal regions were required for function [30]. In ZnT6-like sequences the histidine-rich loop is substituted by a serine-rich loop and a histidine/serine-rich region is also present at the C-terminus. In ZnT1-like transporters extra cysteine/serine rich regions are found in the TMDs V-VI loop and at the C terminus.

Mn-CDF members present a serine-rich region at the N terminal. Sequences belonging to Zrg17-like and MMT-like clusters also have a histidine/serine-rich loop, although located between TMDs III and IV (Figure 4).

A specific feature of the Zn-CDFs is the presence of few cysteine residues conserved within some clusters: sulfhydryl moieties are known to have a role in metal binding (mostly Zn²⁺ and Cd²⁺). In the ZnT1-like cluster five cysteine residues are conserved at the cytosolic C terminus; another cysteine residue is conserved in TMD III in both ZnT1-like and Zrc1-like clusters; the Msc2-like cluster shows a cysteine residue conserved in all sequences at TMD III (by referring to the CDF domain). Other cysteine residues are conserved in the ZnT2-like cluster: in order to test the importance of these cysteine residues for function and their conservation during evolution, all five cysteine residues (C30, C35, C64, C291, and C357) were changed to serine in the poplar PtdMTP1 member. C30, C64, and C291 were conserved within the ZnT2-like cluster, and C35 and C357 were conserved only in the plant members of this cluster. Substitution of the C35 residue abolished the function of the transporter, while the C30S and C64S substitutions only partially affected PtdMTP1 function (Figure 2). The C291S and C357S substitutions did not have any effect on Zn²⁺ tolerance: the expression of these PtdMTP1 mutated versions allowed growth of the Zn-yeast hypersensitive strain zrc1Δ up to 20 mM ZnCl₂ (Figure 2, Table 1, and data not shown). All the mutant proteins still localised on the tonoplast, as shown by C terminal chimeric fusion with GFP. To gain insights into the role of cysteine residue in the oligomerisation of PtdMTP1, Western blot analyses were performed with the mutant proteins as previously described [23]. All the mutant proteins still formed homo-oligomers (data not shown). The Fe/Zn-CDF group did not have any conserved cysteine residue, while the Mn-CDF group displayed a highly conserved cysteine at the cytosolic side of TMD IV.

In the present work, we propose a modified signature that better takes into account the newly characterised CDF transporters. Retrieved at Prosite [33] with the proposed modified CDF signature, the 930 hypothetical CDFs still mostly clustered in the three groups, thus suggesting that a good representative sample of CDF sequences was originally used for analysis. It is possible also to suggest their substrate specificity: 268 were classified in the Fe/Zn-CDF group, 561 in the Zn-CDF group, and 39 in the Mn-CDF group. Few of the sequences originally selected for the protein alignment used to construct the signature were not retrieved by the new signature, since the three databases available at Prosite did not include some data from genome sequencing projects. Some amino acid residues along the signature were specifically present only in a few clusters: this is the case of the amino acid at the position 10 (H89 in PtdMTP1, Figure 1) that is always an aspartate in the Mn-CDF group and in the FieF-like cluster, while a histidine residue is present in the same position in the Zn-CDF group and in MMT-like, WmFieF-like, and ZnT9-like clusters. At position 14 (D93 in PtdMTP1, Figure 1) aspartate is the most frequent amino acid, although some sequences belonging to the Proteobacteria phylum displayed a histidine residue in that position. These two positions, along with the fully conserved glycine were the most critical signature residues, since when changed to other amino acids many non-CDF sequences were retrieved. These amino acids are essential for function, as confirmed by mutagenesis studies (Table 1 and Figure 2), suggesting that the conservation of these amino acids has been maintained by evolutionary pressure.

CDF member distribution in the phylogenetic tree was not related to organism taxonomy, but rather to substrate specificity suggesting ancient duplication events followed by subfunctionalization (metal specialization) in a common ancestor to prokaryotes and eukaryotes. More recent duplications or gene losses could have arisen in several taxa and lineages leading to a complex distribution pattern of CDF isoforms (orthologous and paralogous copies) in living organisms. Based on amino acid sequence similarities and phylogeny of a set of 273 CDF sequences, and on biochemical features of some well-characterised CDF members, the CDF family could be divided into three major groups, one containing Zn-CDF transporters (ZnT2-like, ZnT1-like, ZnT6-like, Zrc1-like, Msc2-like, ZitB-like, and DmeF-like clusters), another enclosing Fe/Zn-CDF transporters (MMT-like, FieF-like, and WmFieF-like clusters) and a third containing only Mn-CDF transporters (MTP8-like sequences) (Figure 3 and Figure 4).

An update of CDF classification is therefore suggested in order to take into better account phylogenetic and functional features. Three groups are proposed: Zn-CDF, Fe/Zn-CDF and Mn-CDF. Ten of the 11 clusters belonging to these groups were already classified in CDF subfamilies by Nies [4] or by Gaither and Eide [3]. In the particular Zn-CDF group contained CDF groups 1 and 2 [4] and CDF subfamilies II and III [3], while Fe/Zn-CDF group contained CDF group 3 [4] and CDF subfamily I [3]. MTP8-like clusters was not present in those classifications because members belonging to this cluster had not then been characterised. The Zrg17-like cluster was very distant from the Zn-CDF group and displayed only a partial signature. It was therefore not included in the Zn-CDF group, even though it had similar biochemical characteristics. The ZnT9-like cluster (previously classified in CDF subfamily I [3]) was found to be close to the Mn-CDF group, but the lack of biochemical characterisation and a weak bootstrap value (23) prompted us to exclude this cluster from the Mn-CDF group (Figure 3 and Figure 4). Sequences belonging to the ZnT9-like cluster might have an ancient origin from a common ancestor and spread in bacteria, metazoans, algae, and plants. Its absence in fungal genomes suggests that this type of CDF has been lost in fungi. The lack of conserved residues in the active site and the potential DNA binding properties of the mammalian ZnT9 proteins [34] suggest additional or different roles of these proteins that nonetheless retained high similarity with the CDF family.

The Mn-, the Zn- and Fe/Zn-CDF groups share many conserved residues, suggesting that they derived from a common ancestor. Newly identified CDF members can be assigned to one of these groups and therefore their substrate specificity can be inferred: for example in plants, only members belonging to the Zn-CDF [13,23,26,36] or to the Mn-CDF [11,48] groups have been functionally characterised. The position of AtMTP6 and orthologous proteins in the Fe/Zn-CDF group suggests a Fe²⁺ selectivity for these CDF members.

The number of transmembrane domains predicted in each transporter is roughly conserved in each cluster, most including from 4 to 6 TMDs, depending on the method used for the prediction. Notably, the Msc2-like cluster included all members with a relatively high TMD number (12 TMDs in Msc2, HsZTL1, and AtMTP12, to 15 TMDs in HsZnT5).

Subcellular localisation of CDF transporters is not always conserved within clusters. The Fe/Zn-CDF group mainly contained sequences from prokaryotes (Archaea or Eubacteria) and fungi. The mitochondrial localisation of ScMMT1 and ScMMT2 [47] corresponds to the plasma membrane location of the prokaryotic members [7,12]. Within the ZnT2-like cluster, however, plant members localised either on the tonoplast (AtMTP1 [51], PtdMTP1 [23]) or on the plasma membrane (TgMTP1, [52]), while mammalian CDF transporters were found on vesicles [37-39]. The vacuolar membrane located transporters Zrc1 and Cot1 from S. cerevisiae clustered together with the reticulum endoplasmic localised Zhf1 from Schizosaccharomyces pombe [45].

The relatively high number of Zn-CDFs found in databases compared with the number of Mn-CDFs may reflect the evolutionary pressure to which the organisms were subjected. Indeed some Zn-CDFs display a broad specificity, which gives a potentially useful trait to the organism. The only CDF transporter known to transport Ni²⁺ (over Zn²⁺) is present in the Ni-hyperaccumulator species Thlaspi goesingense [13]. Similarly DmeF, RmCzcD, and WmFieF displayed a broad specificity [5,12], and C. metallidurans has been frequently found in sediments and soils with a high content of various heavy metals from diverse geographical locations. The need for a Mn²⁺ detoxification (or homeostasis) system driven by CDF transporters could have represented a less important pressure during evolution or could act in parallel with other transport systems (for example the Ca²⁺/Mn²⁺ P-type ATPase Pmr1 in fungi and ECA1 in plants or the cation exchanger CAX2 in plants [53]).

In the present study we identified residues that are crucial for function in the model eukaryotic Zn-CDF PtdMTP1. Amino acid substitutions of conserved aspartate and histidine residues in TMDs II and V led to the loss of almost any detectable function in complementation tests. In EcFieF four amino acids belonging to TMD II (D45 and D49) and to TMD V (H153 and D159) are the metal coordination residues that directly interact with the substrate during binding and transport [31]. The corresponding amino acids are likely to have the same role in other CDF transporters, since at least in EcZitB, RmCzcD and PtdMTP1, substitutions in these residues completely abolish the function (Table 1).

Concerning PtdMTP1, all the mutant proteins were expressed and localised using GFP to the tonoplast, as found for the wild type protein [23], with the exception of the mutant D288A/E proteins that localised on vesicle-like structures (data not shown). Thus for the substitutions D288A/E (TMD VI) we cannot conclude whether the loss of function was due to the mis-localisation or to a reduced activity of the transporter. Additionally the D181E substitution in the corresponding aspartate residue in the Zn²⁺/Co²⁺ bacterial transporter RmCzcD prevented the correct membrane localisation and induced a partial degradation of the mutant protein [29]. Therefore this aspartate residue appears to be critical for CDF protein stability and localisation in organisms from different kingdoms. Likewise in the yeast Pmr1, a Golgi-localised Ca²⁺/Mn²⁺ P-ATPase transporter, the substitutions D778E and D856E in TMD VI and TMD VIII were responsible for ER retention [54]. The G118A substitution abolished function as well. The topological position of G118 (upstream from TMD III), its unique conservation pattern among the CDF transporters and its significance in protein function (at least for PtdMTP1) suggests a role in the formation of a correct protein structure. Regarding substitutions of cysteine residues, only the C35S substitution dramatically affected protein function. The four other substitutions resulted in a reduction of PtdMTP1-mediated zinc resistance (C30S and C64S) or did not affect protein function (C291S and C375S). Based on its functional role and its conservation in a well defined phylogenetic group (plant members of the ZnT2-like cluster), we can hypothesise that the C35 residue has been positively selected throughout plant evolution and has contributed to adaptation. The possible role of these amino acids in protein oligomerisation was investigated in this work by Western blot analysis. Although previous results in the presence of DTT suggested an oligomerisation mediated by disulfide bridges [23], the variant proteins still form oligomeric structures, suggesting that the cysteine residues are dispensable in forming the quaternary structure of PtdMTP1 Zn²⁺ transporter (data not shown). The conserved cysteine residues, located in the N and C termini, are likely to be involved in substrate binding rather than substrate transport. The occurrence of more than one cysteine residue might be an indication of redundant function and the presence of double-substituted cysteine residues might be necessary to abolish protein function and/or oligomerisation.

The identification of molecular determinants that may reflect metal specificity is an interesting research field with application to bioremediation. Several features may contribute to metal specificity in CDF members, nevertheless the widely conserved aspartate residues in TMDs II and V cannot alone be responsible for metal specificity, because they are conserved in transporters with different metal specificities. The mechanism of metal selectivity might be guaranteed by the coordination chemistry in the chemical context of the immediate binding site neighbourhood. Zn²⁺ usually favours a tetrahedral coordination geometry. Mn²⁺ and Fe²⁺ are mainly found in octahedral coordination even though a tetrahedral coordination geometry of ligands is also favoured by these metals. The amino acids located in TMD V, four residues upstream from the highly conserved aspartate, seemed to be candidates of choice for determination of metal specificity of the Zn- and Fe/Zn-CDFs (histidine) or of the Mn-CDFs (aspartate). At least four features suggested that these residues are likely to be important in the determination of substrate specificity: (i) they were not interchangeable, at least in the eukaryotic Zn²⁺ transporter PtdMTP1 (Figure 2), (ii) they were specifically conserved within groups (Figure 4), (iii) they resided in close contact with the highly conserved and functionally important aspartate, on the polar side of the amphipathic helix V, (iv) they faced the cytosolic side of the helix, where they can contact the substrate. These amino acids correspond to H153 in EcFieF, which is the closest coordination residue to the cytoplasm, cis to metal uptake [31]. The H260D variant of PtdMTP1 was not able to complement the Mn-hypersensitive strain pmr1Δ. We can thus infer that this histidine residue is necessary for PdtMTP1-mediated resistance on Zn²⁺, but we cannot speculate whether the presence of aspartate, instead of histidine, is sufficient for manganese transport.

The other conserved and amphipathic helices did not contain any completely group-conserved residue that may be related to metal specificity. Nevertheless this does not exclude their involvement in conferring metal specificity. Even if we cannot conclude that the presence of this distinctive amino acid (D or H) may be sufficient to discriminate the substrate, this could be a starting point for further investigations. Other mechanisms, such as the interactions with metal chelating proteins or cofactors and the involvement of histidine rich regions (Figure 4), have to be taken into account in the first step of metal binding. Nevertheless the Fe/Zn-CDFs FieF from E. coli and C. metallidurans do not have any histidine rich-loops [7,12] (Figure 4). Furthermore, some CDF transporters have been shown to be active in reconstituted proteoliposomes [6,7,26], demonstrating that at least these transporters did not require any interaction with other proteins or cofactors to sustain metal transport.

Deduced amino acid sequences, belonging to characterised or hypothetical CDF transporters, have been retrieved from Genebank and SwissProt and from a set of annotated genomes. These include the Archaea Methanosarcina acetivorans and Methanosarcina mazei, the actinobacteria Corynebacterium glutamicum and Streptomyces coelicolor, the cyanobacterium Nostoc punctiforme, the firmicutes Bacillus cereus, Bacillus subtilis, and Clostridium tetani, the proteobacteria Agrobacterium tumefaciens, E. coli, Thiomicrospira crunogena, and C. metallidurans, the alga C. reinhardtii, the dicotyledonous plants A. thaliana and P. trichocarpa and the monocotyledonous plant O. sativa, the protist E. histolytica, the Ascomycetes Ashbya gossypii (hemiascomycete), Aspergillus fumigatus, Aspergillus nidulans, Botrytis cinerea, Candida albicans, Fusarium graminearum, M. grisea, N. crassa, S. cerevisiae, Sclerotinia sclerotiorum, and S. nodorum, the Basidiomycetes Phanerochaete chrysosporium and Ustilago maydis, the Zygomycete R. oryzae; the nematode C. elegans, the insect D. melanogaster, and the mammal Homo sapiens. Sequences homologous to Zrg17 from S. cerevisiae were retrieved by BLASTP search in the above-described genomes, since these members were not annotated. Sequences that displayed significant truncation have been discarded. CDF amino acid sequences (see accession numbers in additional file 1) were aligned using Gonnet as protein weight matrix, available in MEGA (version 3.1) [55]. Molecular evolutionary analyses were conducted using MEGA (version 3.1) [32]: for the construction of the phylogenetic tree the Maximum Parsimony (MP) or the Neighbor-Joining (NJ) method and the bootstrap (1000 replicates) phylogeny test were applied on the multiple sequence alignment. Sequence distances were calculated with the Poisson correction model with uniform rates among sites.

Logo representation of the CDF signature was constructed with the web interface program WebLogo [56,57] on the basis of the alignment of CDF members listed in additional file 1, with the exception of the Zrg17-like sequences. For TMD II and V representation, alignments corresponding to each individual cluster or group were used as queries. The signature was verified at ScanProsite [33] against all the three available databases (Swiss-Prot, TrEMBL and PDB Databases), with default settings for pattern options.

The yeast strains used for the heterologous expression of PtdMTP1 were BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0), zrc1Δ (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0 ZRC1::kanMX4), and pmr1Δ (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0 PMR1::kanMX4). The mutants used derived from the parental strain BY4741 and were all obtained from Euroscarf [58]. Growth was in YPD or in synthetic defined (SD) medium with 2% (w/v) galactose, supplemented as necessary [59]. Yeast cells were transformed as described elsewhere [60]. For metal tolerance assays, yeast was grown in liquid SD medium (with 2% [w/v] galactose) at 30°C until O.D. _{600 nm} = 1, and drop assays were performed on SD plates containing different concentrations of Zn²⁺ or Mn²⁺.

Mutant PtdMTP1 plasmid clones were generated by oligonucleotide directed PCR mutagenesis, using pYES2-PtdMTP1 as a template for the PCR (Table 2). The mutated versions of PtdMTP1 were cloned in pYES2 and expressed in the Zn-hypersensitive strain zrc1Δ. Cloning, site-directed mutagenesis, construction of the expression plasmid pYES2-GFP, and Western blotting analysis were described previously [23]. The presence of targeted mutations in all plasmid constructs was verified by DNA sequencing.