To visualize how the metal substrate is coordinated in Nramps, we determined crystal structures of DraNramp using lipid-mesophase based techniques (Supplementary file 1a). We obtained a structure of wildtype (WT) DraNramp in an occluded state bound to Mn²⁺ at 2.38 Å by soaking crystals with Mn²⁺ (WT•Mn²⁺; Table 1, Figure 1A–B). We resolved a comparable structure using co-crystallization with Mn²⁺ and the inward-locking mutation A47W (A47W•Mn²⁺; Supplementary file 1b; Cα RMSD=0.47 Å; all pairwise RMSD values listed in Supplementary file 1c; Bozzi et al., 2016b). The similarity of both structures, including a nearly identical Mn²⁺-coordination sphere (Figure 1B, Figure 1—figure supplement 1A), suggests that the observations we make based on these two structures are robust. Both structures superimpose best with the published occluded metal-free G45R structure (Bozzi et al., 2019b). Although the inner vestibule is occluded in both structures, WT•Mn²⁺ and A47W•Mn²⁺ differ in their TM1a position, with WT•Mn²⁺ nearly identical to G45R whereas the A47W•Mn²⁺ TM1a is displaced within the inner vestibule, likely to accommodate the bulky tryptophan sidechain. Therefore, we generally used the WT•Mn²⁺ structure for analysis of the occluded state. As in the metal-free G45R, the Mn²⁺-bound occluded structures have a completely sealed outer vestibule and a partially closed inner vestibule, with the Mn²⁺ occluded from bulk solvent (Figure 1A).

Anomalous difference Fourier maps confirmed presence of Mn²⁺ at the canonical, orthosteric Nramp metal-binding site between the unwound regions of TM1 and TM6 (Supplementary file 1d, Figure 1—figure supplement 1A). In WT•Mn²⁺, the Mn²⁺ is coordinated by conserved residues D56, N59, and M230, and backbone carbonyls of A53 and A227 (Figure 1A, Supplementary file 1e). A53 and A227 are pseudosymmetrically related in the inverted repeats of the LeuT fold of DraNramp (Figure 1—figure supplement 1B). The coordination sphere is completed by a water bridging Mn²⁺ with Q378, a residue previously proposed to directly coordinate Mn²⁺ in the occluded state (Bozzi et al., 2019b). This yields a coordination number of 6, typical for Mn²⁺, and a largely dehydrated metal-binding site with a distorted octahedral Mn²⁺-coordination geometry (Supplementary file 1f), as often observed in other Mn²⁺-protein complexes (Barber-Zucker et al., 2017; Couñago et al., 2014; Dudev and Lim, 2014).

At the mouth of the outer vestibule, an additional Mn²⁺ bridges D296 and D369 at the N termini of extracellular helix 2 (EH2) and TM10, respectively (Figure 1—figure supplement 1A). We denote this site as the ‘external site’ and the canonical substrate-binding site as the ‘orthosteric site’. Corroborating the structures, isothermal titration calorimetry (ITC) measurements reveal an endothermic mode of binding and are best fitted with a two-site model for WT (K_d1=190±30 µM, K_d2=1970±520 µM; Figure 1C) and A47W (K_d1=125±5 µM, K_d2=2450±650 µM; Figure 1—figure supplement 2A; all K_d values are in Supplementary file 1g; see Appendix 1 for a description of our ITC data analyses). To determine the affinity of the orthosteric site, we mutated the external-site aspartates. The D296A and D369A substitutions have little impact on Mn²⁺ transport (Figure 1—figure supplement 2B). The D296A and D369A variants each bind one Mn²⁺ with K_d=370±30 µM and 420±30 µM respectively (Figure 1—figure supplement 2A), which is closest to K_d1 of WT. Hence, the affinity for Mn²⁺ at the orthosteric site is higher than at the external site. A metal ion is present at the external site in all inward-open and occluded metal-bound DraNramp structures, but not outward-open structures, as opening the outer vestibule separates D296 and D369 and disrupts the site (Figure 1—figure supplement 3A). D296 and D369 are not conserved across Nramps, but they are more conserved within bacterial clade A, and there is a general abundance of acidic residues in the corresponding loop regions across all clades (Figure 1—figure supplement 3B–C). At present, our results provide little evidence of a biological role for this previously unidentified external site; perhaps the concentration of acidic residues at the mouth of the outer vestibule (Figure 1—figure supplement 3D) provides electrostatic attraction for metal cations.

We also determined high-resolution DraNramp structures in metal-free occluded (WT) and Mn²⁺-bound inward-open (M230A•Mn²⁺) states and re-refined a Mn²⁺-bound outward-open conformation (G223W•Mn²⁺; Table 1, Supplementary file 1a and b). Along with the published structures of outward-open metal-free G223W and inward-open metal-free ‘Patch’ (which has a patch of mutations in intracellular loops) (Bozzi et al., 2016b; Bozzi et al., 2019b), these new structures allow us to map the entire Mn²⁺ transport cycle to three major conformations, each in Mn²⁺-bound and metal-free states (Figure 2A–B). By ordering and comparing these six structures, we outline a molecular mechanism by which metal substrate binds, induces conformational change, and is released.

TMs 1, 5, 6, and 10 move the most as the Mn²⁺-binding site accessibility switches from outward to inward across the conformations (Figure 2B–C, Figure 2—figure supplement 1), as also highlighted in previous studies (Bozzi and Gaudet, 2021; Bozzi et al., 2019b; Ehrnstorfer et al., 2017). As Mn²⁺ binds to the outward-open state, TM10 tilts toward TM1b, the upper half of TM5 toward TM7, and TM6a toward TM11 to seal the outer vestibule and yield an occluded state. The lower half of TM5 also moves away from TM1a, allowing it to swing upward to open the inner vestibule in the following transition. This swing of TM1a is the only noteworthy difference between the Mn²⁺-bound occluded and inward-open states, allowing release of the Mn²⁺ into the inner vestibule. TM1a swings to a similar angle in the new inward-open M230A•Mn²⁺ as in the previous low-resolution inward-open metal-free structure (Bozzi et al., 2016b), and structures of the homologous Eggerthella lenta Nramp-related magnesium transporter (EleNRMT), LeuT, and serotonin transporter (Coleman et al., 2019; Krishnamurthy and Gouaux, 2012; Ramanadane et al., 2022). Thus, most of the structural reorganization in Nramps occurs in the shift from outward open to occluded. The three metal-free DraNramp conformations are similar to their corresponding Mn²⁺-bound structures, suggesting that once Mn²⁺ is released, the conformational transitions are reversed, including passing through an occluded metal-free intermediate, to reach the outward-open metal-free conformation ready to accept Mn²⁺ (Figure 2A–B, Figure 2—figure supplements 1 and 2A).

The substrate-binding TM1 and TM6 are well-resolved in our structures (Figure 2—figure supplement 2B) and pairwise superpositions reveal how their motions contribute to the conformational changes across the transport cycle (Figure 2C). TM6a tilts 22° and the unwound region of TM6 becomes more helical as it moves toward the Mn²⁺ to close the outer gate. The central unwound regions of TM1 and TM6 are closest in the occluded state, resulting in an almost dehydrated Mn²⁺-coordination sphere. Finally, the inner vestibule opens when TM1a tilts upward by 32°, increasing the distance between TM1a and TM6b by ~8 Å (Figure 2C). TM6b is largely static relative to the protein core, although in the inward-open structure it has high B-factors (Figure 2D), indicating that the interaction with TM1a stabilizes TM6b to close the inner vestibule.

Our Nramp structures provide snapshots of the complete Mn²⁺-coordination sphere geometries in each conformation (Figure 3). Two new structures of DraNramp point-mutants reveal the coordination of Mn²⁺ in the inward-open state: M230A•Mn²⁺ and D296A•Mn²⁺ (Cα RMSD of 0.42 Å). The Mn²⁺ is in the same location of the orthosteric site as in the occluded state, as confirmed by anomalous diffraction for D296A•Mn²⁺ (Supplementary file 1d, Figure 3—figure supplement 1). As in the occluded state, the Mn²⁺ binds D56, N59, and the A227 carbonyl, with a water replacing M230 in M230A•Mn²⁺ (Figure 3A, Supplementary file 1e). However, with TM1a displaced, the A53 carbonyl no longer coordinates Mn²⁺; instead, the Y54 carbonyl approaches Mn²⁺ at a longer distance of 3.1 Å. Two more waters, one bound to Q378 and another from the inner vestibule, complete a seven-coordination sphere resembling a pentagonal bipyramidal geometry with substantial distortion (Supplementary file 1f). Seven coordination is infrequent but found in Mn²⁺-coordinating proteins like MntR (Chen and He, 2008; Glasfeld et al., 2003). Our inward-open structures provide the first evidence that Y54 participates in the Mn²⁺ transport cycle.

The new occluded and inward-open Mn²⁺-bound structures have a monodentate coordination of D56 with Mn²⁺. For consistency, we reinterpreted the outward-open G223W•Mn²⁺ map (PDB ID: 6BU5) with a monodentate coordination of D56 with Mn²⁺ instead of previously modeled bidentate interaction (Bozzi et al., 2019b); the local and global model statistics are very similar to the original structure (Supplementary file 1e). Re-refined G223W•Mn²⁺ has six Mn²⁺-coordinating ligands: D56, N59, M230, carbonyl of A53 and two waters (Figure 3A, Supplementary file 1e); the overall geometry resembles a distorted octahedron as in the occluded structure (Supplementary file 1f). The coordination spheres in all conformations of the transport cycle are well defined as confirmed by the 2F_o-F_c maps of the closeup snapshots of their Mn²⁺-bound orthosteric site (Figure 3—figure supplement 1C, Figure 5—figure supplement 2).

Comparing the three Mn²⁺-bound conformations (Figure 3A), the carbonyls of A53 in TM1a and A227 in TM6b alternately coordinate Mn²⁺ in the outward- and inward-open structures respectively, and both residues interact with Mn²⁺ in the occluded structure. The pseudosymmetrically related A53 and A227 may thus act as hinges altering the Mn²⁺-coordination sphere as TM1 and TM6 move in turn to open the gates during Mn²⁺ transport. Furthermore, as DraNramp switches from outward- to inward-open, Y54 progressively swings downward, acting as a gate in concert with TM1a’s upward swing to open the inner vestibule and allow metal release (Figure 3A–B). Our inward-open structures also suggest that the Y54 carbonyl may participate in Mn²⁺ release through interaction with the metal ion (Figure 3A). All 3796 Nramp sequences in our alignment have either a tyrosine or a phenylalanine at this position and Y54 is completely conserved among bacterial clades A (including DraNramp) and C, while the position is 40% and 100% phenylalanine among eukaryotes and bacterial clade B, respectively (Figure 3—figure supplement 3). To evaluate the significance of Y54 in Mn²⁺ transport, we purified and reconstituted into proteoliposomes the Y54A and Y54F variants. While Y54F has near-wildtype Mn²⁺-transport activity, Y54A nearly eliminates Mn²⁺ transport (Figure 3C), indicating that an aromatic ring is essential for the gating motion required for transport.

We also measured the Mn²⁺-binding affinity of the constructs that yielded inward- or outward- open structures, M230A and G223W, respectively. Mn²⁺ is present at the external site in the inward-open M230A•Mn²⁺ (Figure 1—figure supplement 3A), and consistently, the ITC data fit a two-site model (K_d1=215±65 µM assigned to the orthosteric site, K_d2=4600±1300 µM for the external site; Figure 3D and Appendix 1). The ITC data for G223W with Mn²⁺ fits only in a one-site model (K_d=440±15 µM; Figure 3D), which we assign to the orthosteric site because the opening of the outer vestibule displaces TM10, disrupting the external site (Figure 1—figure supplement 3A).

To compare the metal-bound states to analogous metal-free states of the transport cycle, we determined two metal-free occluded structures of wildtype DraNramp at a higher resolution than the previously reported G45R structure (Bozzi et al., 2019b), which we refer to as WT (2.38 Å; Supplementary file 1b) and WT_soak (2.36 Å; Table 1). The crystal used for WT_soak was mock-soaked (with no metal in the soaking solution). WT and WT_soak are nearly identical, confirming that the soaking process does not influence the conformational state. We analyzed WT_soak, unless otherwise noted. In WT_soak, we observed density but no anomalous signal at the orthosteric site and modeled a water molecule at the position where Mn²⁺ sits in the occluded state (Figure 3A). WT_soak is nearly identical to WT•Mn²⁺, indicating that the occluded conformation is unchanged by the presence of metal ion substrate, and the metal-binding site is instead filled by ordered water molecules.

We used previously reported inward-open ‘Patch’ and outward-open G223W metal-free structures for analysis of the Mn²⁺ transport cycle (Figures 2B and 3A; Bozzi et al., 2016b; Bozzi et al., 2019b). These structures, resolved at lower resolution than the ones described here, have no density at the orthosteric site. This is consistent with a more flexible organization of a metal-binding site open to bulk aqueous solvent. In contrast, metal-free WT has a more ordered orthosteric site, suggesting a stable occluded intermediate in the switch from inward- to outward-open.

Vestibules providing access to the orthosteric site from the extracellular or intracellular side alternately open from the motions of DraNramp’s TMs 1, 5, 6, and 10, which form the outer and inner gates during Mn²⁺ transport (Bozzi et al., 2020; Bozzi et al., 2019b). To pinpoint protein features that enable these motions, we used our Mn²⁺-bound structures to identify interaction networks with the following attributes: (i) they contain conserved polar residues from at least one of the four mobile helices; (ii) they line the gates; and (iii) they rearrange between the three resolved protein conformations (Figure 4A).

Two networks seal the outer gate. In the occluded and inward-open conformations, Q378 interacts with two waters, one coordinating the orthosteric Mn²⁺ and the other interacting with D56 and the carbonyl of T130. This network is disrupted in the outward-open conformation as Q378 and the rest of TM10 swing outward to open the outer vestibule (Figure 4B). In the second network, T228, N275, and N82 interact via a water in the occluded and inward-open states, but not in the outward-open state, where the extended unwound region of TM6a positions T228 farther from the orthosteric site and N275 and N82 (Figure 4C). This T228 network helps rearrange TM6a, closing the outer vestibule in the occluded state and generating a nearly dehydrated Mn²⁺-coordination sphere (Figure 1B). As the inner gate opens to release Mn²⁺, both networks persist, ensuring that the outer gate remains closed in the inward-open conformation.

All six residues in the Q378 and T228 networks are completely conserved across bacterial clade A and highly conserved across all Nramps (Figure 4—figure supplement 1). To investigate the robustness of these networks, we performed duplicates of molecular dynamics (MD) simulations starting in each of the three conformations and confirmed that within the first 250 ns of these simulations, the T228 and Q378 networks persist in simulations of the occluded and inward-open states and remain broken in simulations starting in the outward-open state (Figure 4—figure supplement 2A–C). This includes the coordinated waters, for example the water at the center of the Q378 network is present in more than 50% of the frames in occluded-state simulations (Figure 4—figure supplement 2E). In line with the X-ray snapshots and simulation outcomes and highlighting their key function in the conformational cycle of Nramps, mutations of any of the six residues across these networks reduces Mn²⁺ transport by DraNramp (Bozzi et al., 2016a; Bozzi et al., 2019a; Bozzi et al., 2020; Bozzi et al., 2019b).

The inner vestibule is gated by rearrangements of residues in two other polar networks, namely those of R244 and Q89 (Figure 4D–E). In the outward-open state, R244 forms an ion pair with E176 and interacts with the TM1a backbone, as does D263, keeping TM1a and TM5 close and the inner gate closed (Figure 4D; Bozzi et al., 2020). In the occluded state, the E176-R244 interaction breaks and TM5 moves away from TM6b, creating space for TM1a to swing up and open the inner vestibule in the inward-open state. Accordingly, the E176–R244 ion pair is stable in MD simulations of the outward-open state, while these residues are >10 Å apart in simulations of the occluded and inward-open states (Figure 4—figure supplement 2D). Supporting the importance of the R244 network, E176 is 100% and R244 is 85% conserved across all Nramps (Figure 4—figure supplement 1) and mutation of either residue reduces Mn²⁺ transport by DraNramp (Bozzi et al., 2020).

In the Q89 network, Q89 hydrogen-bonds with Y54 and H237 to seal the inner gate in the outward-open (Bozzi et al., 2020) and occluded states (Figure 4E). In the inward-open structure, the Q89–H237 hydrogen bond is broken and a rearranged Y54–Q89 hydrogen bond buttresses the opening of the inner vestibule (Figure 4E). As discussed above, Y54 is conserved and important for Mn²⁺ transport by DraNramp (Figure 3C and Figure 3—figure supplement 1C). Similarly, Q89 and H237 are conserved (Figure 4—figure supplement 1), and mutations to either residue impair Mn²⁺ transport in a cell-based assay (Bozzi et al., 2020). Cysteine accessibility measurements showed that mutations to Q89 or H237 render the outer vestibule solvent-inaccessible (Bozzi et al., 2020), indicating that disrupting the Q89 network likely prevents closing of the inner gate.

H232 (TM6b) is conserved across all Nramps (Figure 4—figure supplement 1), highlighting its importance. H232 sits below the orthosteric site and forms a network conserved across all conformations, with water-mediated hydrogen bonds to E134 (involved in proton transfer to the salt-bridge residues in TMs 3 and 9) (Bozzi et al., 2019a; Bozzi et al., 2019b; Ehrnstorfer et al., 2017) and T130 in TM3 (Figure 4F). Waters occupy these two sites in MD simulations in all states, especially the water coordinated between H232 and T130 (Figure 4—figure supplement 2F–G). However, in M230A•Mn²⁺ H232 flips to interact with the G330 carbonyl (TM8; Figure 4F), suggesting that it may transiently move during the conformational cycle. Indeed, while the H232 sidechain rotamer is stable in MD simulations of the outward-open state, it explores other rotamers in simulations of the occluded and inward-open states (Figure 4—figure supplement 3).

Nramps transport divalent transition metals quite promiscuously, including both physiological (Fe²⁺ and Mn²⁺) and non-physiological substrates (Cd²⁺, Zn²⁺, Co²⁺, Ni²⁺, Pb²⁺), but select against alkaline earth metals (Mg²⁺, Ca²⁺) (I Bannon et al., 2002; Bozzi et al., 2016a; Bozzi and Gaudet, 2021; Bozzi et al., 2019b; Ehrnstorfer et al., 2017; Illing et al., 2012; Sacher et al., 2001). DraNramp transports Cd²⁺ well, but without concomitant proton flux and with weaker voltage dependence (Figure 5—figure supplement 1A; Bozzi et al., 2019a; Bozzi and Gaudet, 2021; Bozzi et al., 2019b). To better understand the underlying mechanistic differences, we compared the binding, transport, and structures of DraNramp with Mn²⁺ and Cd²⁺.

In contrast to endothermic binding of Mn²⁺, ITC measurements show exothermic binding of Cd²⁺ to WT DraNramp (Figure 5—figure supplement 1B; see Appendix 1 for details of the ITC analyses). Like for Mn²⁺, the Cd²⁺ isotherm fits best in a two-site model (K_d1=55±15 µM, K_d2=220±20 µM). Both D296A and D369A—containing mutations at the external metal-binding site—showed exothermic binding but fit best in a one-site model (Figure 5A). Based on their affinity (K_d=120±1 µM for D296A and K_d=70±10 µM for D369A), we conclude that the orthosteric site has higher affinity toward Cd²⁺ than the external site (Figure 5A–B). Both the orthosteric and external site shows higher affinity toward Cd²⁺ than Mn²⁺. The affinities are comparable at the orthosteric site, in low micromolar range for both metals (K_d of 190±30 µM for Mn²⁺ vs. 55±15 µM for Cd²⁺), whereas the external site has a much higher affinity toward Cd²⁺ (K_d of 1970±520 µM for Mn²⁺ vs. 220±20 µM for Cd²⁺). Consistent with its inability to transport Mg²⁺, a representative alkaline earth metal (Bozzi and Gaudet, 2021; Bozzi et al., 2019b), DraNramp does not bind Mg²⁺ (Figure 5A). These are the first ITC measurements comparing the binding of different metals to an Nramp transporter and they show clear differences in the binding mode and affinity of different substrates toward DraNramp (Figure 5A, Figure 5—figure supplements 2 and 3). In contrast to DraNramp, previous ITC studies showed endothermic binding of Cd²⁺ to the Staphylococcus capitis Nramp homolog (ScaDMT) with 29 µM affinity (Ehrnstorfer et al., 2014). However, in the absence of ITC data with other metals, it is not known whether ScaDMT also shows differences in the mode and affinity of binding to different metals like DraNramp.

To understand the differences in binding and transport of Mn²⁺ and Cd²⁺, we determined a 2.5 Å Cd²⁺-bound structure by soaking WT DraNramp crystals with 2 mM Cd²⁺ (Table 1). In agreement with the ITC data, WT•Cd²⁺ shows Cd²⁺ ions at both the external and orthosteric sites as confirmed by anomalous signal (Figure 5—figure supplement 1C–D, Supplementary file 1d). The higher affinity for Cd²⁺ than Mn²⁺ at the external site suggests that local geometry favors non-physiological Cd²⁺ over the physiological substrate Mn²⁺ (Figure 1—figure supplement 3A). The low affinity of this site for Mn²⁺, compared to Cd²⁺, is consistent with the idea that this external site plays a role in general electrostatic attraction of substrate to the orthosteric site, rather than as a finely tuned binding site specific for Mn²⁺.

Interestingly, WT•Cd²⁺ is inward open (Figure 5—figure supplement 1C), although both mocked-soaked and Mn²⁺-soaked crystals under otherwise equivalent conditions yielded occluded structures (WT_soak and WT•Mn²⁺, respectively; Figures 1 and 3). The larger ionic radius of Cd²⁺ (~0.95 Å) compared to Mn²⁺ (~0.82 Å) (Kumarevel et al., 2005; Vashishtha et al., 2016) and different preferred coordination geometry may increase the stability of the inward-open state with Cd²⁺ bound. Compared with the inward-open M230A•Mn²⁺, D56 adopts a different rotamer in WT•Cd²⁺ and does not coordinate the Cd²⁺ bound at the orthosteric site (Figure 5C). The rest of the coordination sphere is similar and includes N59, M230, the A227 carbonyl, the Y54 carbonyl, a water that coordinates Q378 and another water from inner vestibule. Cd²⁺ has six coordinating ligands, and the distortion from ideal octahedral geometry is more pronounced compared to Mn²⁺ (Supplementary file 1f). The coordination distances are larger for Cd²⁺ than for Mn²⁺ (Supplementary file 1e), as seen in other proteins (Begg et al., 2015; Yokoyama et al., 2012) and consistent with its larger ionic radius and distinct charge distribution. Cd²⁺ is a soft metal, likely explaining why it retains coordination by the softer sulfur ligand of M230 (Cammack and Hughes, 2008) but not the hard oxygen of D56, although D56 can still provide favorable electrostatics.

ITC data analysis of different DraNramp variants highlights additional differences between Mn²⁺ and Cd²⁺ binding and how their binding affects the conformational preferences of DraNramp (Figure 5B, Figure 5—figure supplements 2–3). A general observation is that, at the orthosteric site, Mn²⁺ and Cd²⁺ binding are entropy- and enthalpy-driven, respectively, and entropy contributions (-TΔS) are smaller for Cd²⁺ than Mn²⁺ (Appendix 1), suggesting that Cd²⁺ binding conformationally constrains the protein more, leads to less solvent release, or both (Ferrante and Gorski, 2012; Olsson et al., 2008).

ITC with the orthosteric-site mutants, M230A and D56A, shows metal-specific behavior. For M230A, Mn²⁺ binding fits with a two-site model and Cd²⁺ fits only with a one-site model (Figure 5B and Figure 5—figure supplement 2, Appendix 1). M230 is thus a crucial ligand for Cd²⁺ but not Mn²⁺, which is further reflected in the transport behavior where M230A affects Cd²⁺ transport drastically but has negligible effect on Mn²⁺ (Bozzi et al., 2016a; Bozzi et al., 2019a). Data for D56A fit better with a two-site model with both metals, although D56A does not transport either metal (Bozzi et al., 2019b). This suggests that D56 is more important for catalyzing transport than for substrate binding. Data for variants with an additional mutation at the external site (D56A-D296A, D56A-D369A, M230A-D296A, and M230A-D369A) fit only with a one-site model for Mn²⁺-binding isotherms, which we assigned as binding to the orthosteric site. However, all four double mutants showed complete loss of Cd²⁺ binding, which is expected for ones with M230A—which eliminates Cd²⁺ binding at the orthosteric site on its own, as we have also shown previously (Bozzi et al., 2016a)—but more surprising for double mutants with D56A. Thus, binding of the preferred physiological Mn²⁺ substrate at the orthosteric site is more robust to perturbations than binding of Cd²⁺, a toxic metal.

The conformation-locking mutants—A47W, which is outward-closed (Bozzi et al., 2016b) and crystallized in an occluded state, and outward-open G223W (Bozzi et al., 2019b)—also show different behavior with Mn²⁺ and Cd²⁺ (Figure 5—figure supplement 3). ITC with Mn²⁺ agrees with the structural data, with two-site fit for A47W (Figure 1—figure supplement 2A), and one-site fit for A47W-D296A, A47W-D369A, and G223W (assigned to the orthosteric site). However, with Cd²⁺, the ITC data for A47W fit best with a one-site model and we observed no binding with A47W-D296A, A47W-D369A, and G223W, indicating that there is no significant affinity for Cd²⁺ at the orthosteric site of A47W and G223W. This is consistent with our inability to obtain Cd²⁺-bound structures in conformations other than inward-open, despite trying both co-crystallization and soaking with A47W and G223W. These data suggest that Cd²⁺ has highest affinity for the inward-open state and low affinity for other states, in agreement with the fact that soaking of WT crystals (which yielded the metal-free occluded WT_soak structure) produced an inward-open Cd²⁺-bound structure.

To better understand the Cd²⁺ transport mechanism given that its binding at the orthosteric site seems less optimal and robust than Mn²⁺, we compared the gating networks described above for the inward-open Mn²⁺- and Cd²⁺-bound structures. The Q89, T228, and R244 networks are essentially identical in the inward-open state, regardless of whether Mn²⁺ or Cd²⁺ is bound (Figure 5—figure supplement 4A–C).

The other two networks, both of which involve D56, have some differences. In the Q378 network gating the outer vestibule, D56 does not coordinate Cd²⁺ but does interact with the conserved water that connects the metal ion with Q378 in all outward-closed structures (Figure 5—figure supplement 4D). This preserves a connection to both Cd²⁺ and Q378 to close the outer gate. Most of the H232 network is similar in the Mn²⁺- and Cd²⁺-bound structures, except for the orientation of D56 relative to E134 (Figure 5—figure supplement 4E).

Mutations of several residues in these polar networks reduced transport in proteoliposome-based assays (Figure 5—figure supplement 4F), corroborating results from cell-based assays (Bozzi et al., 2020). Y54A, H237A, and H232A mutations cause the largest decreases, Q89A causes a moderate decrease, and Y54F is similar to WT. Mn²⁺ and Cd²⁺ follow similar trends although with less voltage dependence for Cd²⁺ than Mn²⁺. These results support the idea that although Mn²⁺ and Cd²⁺ bind differently at the orthosteric site, the flexibility of the D56 sidechain enables the networks of polar residues that gate the outer and inner vestibules to engage and enable transport of both metals at similar rates.

The DraNramp WT and mutant constructs were cloned into pET21a-N8H (Bozzi et al., 2016a). Primer sequences for the mutations are listed in Supplementary file 1h. All constructs used for crystallization had a truncation of 31 residues at the N-terminus (ΔN31, which does not impair metal transport), except for A47W which was full-length and transport deficient (Bozzi et al., 2016b). For proteoliposome-based transport assays, the full-length versions of each construct were used. For ITC, full-length DraNramp constructs were cloned into pET21-NStrep (Bozzi et al., 2016a) to avoid background signal from metals (Mn²⁺, Cd²⁺) binding to the His-tag. All mutations were introduced by site-directed mutagenesis using the Quikchange mutagenesis protocol (Stratagene) and confirmed by Sanger DNA sequencing.

Protein expression was performed as previously described (Bozzi et al., 2016a). Briefly, transformed Escherichia coli C41(DE3) (Lucigen) were induced with 0.1 mM isopropyl-β-D-thiogalactopyranoside and cultured at 18 °C for 16 hr. Cell pellets from 10 L of culture were harvested and flash frozen in liquid nitrogen.

Cells were thawed and resuspended in 50 mL load buffer (20 mM sodium phosphate, pH 7.5, 55 mM imidazole pH 7.5, 500 mM NaCl, 10% (v/v) glycerol) supplemented with 1 mM PMSF, 1 mM benzamidine, 0.3 mg/mL DNAse I and 0.3 mg/mL lysozyme and lysed by sonication on ice (six cycles of 45 s with a Branson Sonifier 450 under duty cycle of 65% and output 10). Lysates were cleared by centrifuging for 20 min at 20,000 rpm (Beckman JA-20) and membranes pelleted from the supernatant by ultracentrifugation at 45,000 rpm (Beckman type 45Ti) for 70 min. Membranes were homogenized in 70 mL load buffer using a glass Potter-Elvehjem grinder, solubilized for 1 hr in 1% (w/v) n-dodecyl-β-D-maltopyranoside (DDM), then ultracentrifuged at 35,000 (Beckman type 45Ti) for 35 min to remove insoluble debris. Pre-equilibrated Ni-Sepharose beads (2 mL; GE Healthcare) were incubated with the supernatant for 90 min at 4 °C, then washed with 20 column volumes (CV) of each of the following buffers sequentially (i) load buffer containing 0.03% DDM, (ii) load buffer containing 0.5% lauryl maltose neopentyl glycol (LMNG), and (iii) load buffer containing 0.1% LMNG. Protein was eluted in 20 mM sodium phosphate, pH 7.5, 450 mM imidazole pH 7.5, 500 mM NaCl, 10% (v/v) glycerol, 0.01% LMNG, concentrated to <0.5 mL in a 50 kDa molecular weight cutoff (MWCO) centrifugal concentrator (EMD Millipore), and purified by size exclusion chromatography (SEC) using a Superdex S200 10/300 (GE Healthcare) pre-equilibrated with SEC buffer (10 mM HEPES pH 7.5, 150 mM NaCl, 0.003% LMNG). Peak protein fractions enriched in DraNramp were combined, concentrated to ~25–40 mg/mL using a 50 kDa MWCO centrifugal concentrator, aliquoted and flash frozen in liquid nitrogen and stored at –80 °C. Purifications of each protein construct were performed at least twice and resulted in similar data.

To purify protein for ITC, harvested cells expressing strep-tagged DraNramp from 10 L of culture were resuspended in 50 mL of buffer W (100 mM Tris, pH 8.0, 150 mM NaCl), and membranes were isolated, homogenized and solubilized in 1% DDM as above. The supernatant was incubated with Strep-Tactin Superflow resin (3 mL; IBA) pre-equilibrated with buffer W+0.03% DDM and washed with the following buffers sequentially: (i) 1 CV buffer W+0.03% DDM, (ii) 2 CV buffer W+0.5% LMNG and (iii) 2 CV buffer W+0.1% LMNG. Protein was eluted with 3 CV buffer W+0.01% LMNG+2.5 mM desthiobiotin. The eluted protein was concentrated up to 2.5 mL using a 50 kDa MWCO centrifugal concentrator and buffer-exchanged into 150 mM NaCl, 10 mM HEPES, pH 7.5, and 0.003% LMNG using disposable PD-10 desalting columns (GE healthcare). Protein was concentrated to ∼2.5 mg/mL a 50 kDa MWCO centrifugal concentrator and flash frozen in liquid nitrogen and stored at –80 °C. Purifications of each protein construct were performed at least twice and resulted in similar data.

Crystallization of all constructs was performed using lipidic cubic phase (LCP). Protein was mixed with monoolein in 1:1.5 volume ratio using the syringe reconstitution method (Bozzi et al., 2019b). The protein bolus (60 nL) and 720 nL precipitant were dispensed onto custom-made 96 well glass sandwich plates using an NT8 drop-setting robot (Formulatrix). Metal-free crystals (WT) and metal supplemented (5 mM MnCl₂) co-crystals (A47W•Mn²⁺, M230A•Mn²⁺, D296A•Mn²⁺) were grown in different precipitant conditions (Supplementary file 1a), harvested within 7–10 days (after reaching their optimal size of 30–40 μm rods) using mesh loops (MiTeGen) and flash-frozen in liquid nitrogen prior to data collection. Some structures (WT•Cd²⁺, WT•Mn²⁺, WT_soak) were obtained by soaking WT DraNramp crystals grown in metal-free precipitant (Supplementary file 1a) for 7–10 days: The glass covering the wells was broken without disturbing the bolus and 2 μl soak solution (Supplementary file 1a) were added before resealing with a fresh siliconized glass coverslip, incubating overnight (16–18 hr), then harvesting and flash freezing for data collection.

Diffraction data for structure determination and refinement were collected at beamlines 24-ID-C or 23-ID-B of the Advanced Photon Source at wavelengths of 0.984 Å or 1.033 Å, respectively. We used anomalous signals to confirm the presence of metals (Mn²⁺ or Cd²⁺) in the binding site. For D296A•Mn²⁺ and A47W•Mn²⁺, the same data we used for structure refinement and collected at 1.033 Å, 0.984 Å, respectively, provided strong anomalous signal in the metal-binding sites. For WT•Cd²⁺, we were able to collect data at 1.904 Å (near the low-energy boundary for the beamline), to maximize the anomalous signal. Locations of the crystals in the mesh loops were identified by grid scanning with a 20 μm beam at 10% transmission followed by data collection with a 10 μm beam at 15% transmission. Data were indexed in XDS (Kabsch, 2010) and scaled in CCP4 AIMLESS (Version 7.0) (Evans and Murshudov, 2013; Winn et al., 2011). For datasets collected from several crystals, data from each crystal were independently indexed and integrated, then combined during scaling using CCP4 AIMLESS (Evans and Murshudov, 2013; Winn et al., 2011) to obtain complete datasets. The resolution cut-off of each structure was defined based on CC1/2 values of 0.3 and above (Karplus and Diederichs, 2012; Karplus and Diederichs, 2015). Initial phases for all structures were determined by molecular replacement in PHENIX (Version 1.17.1–3660) (Liebschner et al., 2019) using an occluded structure of DraNramp (PDB ID 6C3I chain A) (Bozzi et al., 2019b) as search model. Data statistics are listed in Table 1 and Supplementary file 1b and d.

Models were built in COOT (Version 7.0) (Emsley et al., 2010) and refined in PHENIX (Liebschner et al., 2019), with macrocycles including reciprocal space, TLS groups, and individual B-factor refinement, and optimization of the X-ray/stereochemistry and X-ray/ADP weights. For WT•Cd²⁺, ‘anomalous group refinement’ was used to improve the fit to density of the Cd²⁺ ions, with Cd²⁺ as an ‘anomalous group’ with the reference f’ and f” values suggested by phenix.form.factor (–0.462 and 2.132, respectively). Ligand restraints for monoolein and spermidine were generated in Phenix.elbow with automatic geometry optimization. All structures contain one protein molecule in the asymmetric unit. The final structures span from residues 45–48 to residues 433–436, except that residues 240–249 and 240–247 were not modeled in D296•Mn²⁺ and WT•Cd²⁺, respectively, because of lack of interpretable electron density map. Model refinement statistics are listed in Table 1 and Supplementary file 1b and d. Pairwise RMSD for all structures are listed in Supplementary file 1c. Anomalous difference Fourier maps for Mn²⁺ (D296A•Mn²⁺ and A47W•Mn²⁺) and Cd²⁺ (WT•Cd²⁺) were generated in phenix.maps using a high-resolution cutoff of 3.5–4.5 Å. Polder maps omitting the metal ions were generated in phenix.polder to appropriately define the coordination sphere for Mn²⁺ and Cd²⁺ in all structures. All software were provided by SBGrid (Morin et al., 2013).

ITC experiments were performed using MicroCal iTC200 (GE Healthcare) to determine the affinity and thermodynamic parameters of binding of divalent metals (Mn²⁺, Cd²⁺ and Mg²⁺) to DraNramp (WT and its mutants) (Leavitt and Freire, 2001; Wiseman et al., 1989). All protein and metal solutions were prepared in ITC buffer (150 mM NaCl, 10 mM HEPES, pH 7.5, and 0.003% LMNG). The sample cell containing 25 μM protein was titrated with 20 2 μL injections of 6 mM metal, with an interval of 120 s between each successive 5 s injection, with a 750 rpm stirring rate at 25 °C. To nullify the heat of dilution, the data from titration of a metal solution into ITC buffer (‘buffer blank’ runs) were subtracted from the metal-protein titration curves prior to model fitting. Data were fitted and analyzed as detailed in Appendix 1. Briefly, as per best practice when c values (association constant × molar protein concentration) are below 1 (Picollo et al., 2009; Tellinghuisen, 2008; Turnbull and Daranas, 2003), all the data reported for each construct are fitted fixing the number of sites (one-site binding model with fixed n=1 or sequential binding model with fixed n=2). The binding stoichiometry was selected based the model fits and knowledge from the crystal structures and mutational analysis. Data were fitted with Origin 7 software. The mean K_d values ± SEM from 2 to 3 repeats (from independent protein purifications) for each sample are reported in Supplementary file 1g.

Protein purification, liposome preparation, and metal transport assays were performed as described (Bozzi et al., 2016a; Bozzi et al., 2019a). Purifications of each protein construct were performed at least twice and resulted in similar data. For each construct, the presented data originates from two to three batches of reconstituted liposomes, each from an independent protein purification, represented as scatter plots of each data point and the corresponding mean.

We used 92 Nramp sequences from Pfam (El-Gebali et al., 2019) to build a seed alignment using MUSCLE (Edgar, 2004). We collected 15,451 sequences from Uniprot (Bateman et al., 2021) using HMMER (Potter et al., 2018). We used HMMER’s hmmalign, with a hidden Markov model profile from the seed alignment as an input, to align all 15,451 sequences. We applied filters to retain sequences 400–600 residues in length and sequences with under 90% pairwise sequence identity, respectively. The final alignment contains 6712 sequences and is well aligned at biologically relevant residues (Figure 1—source data 1). A maximum-likelihood phylogenetic tree was generated via RAxML-NG Kozlov et al., 2019 using the LG substitution model (Le and Gascuel, 2008), with the likeliest final tree selected from 10 parallel optimization trials (Figure 1—source data 2). The canonical Nramp clade in this tree was identified based on conservation of the ‘DPGN’ and ‘MPH’ motifs in transmembrane helices 1 and 6, respectively, and contained 3796 sequences. The Nramp-related magnesium transporters were used to root the canonical Nramp phylogeny. Sequence analysis was done with Biopython (Cock et al., 2009) and sequence logos were generated with logomaker (Tareen and Kinney, 2020) using a ‘chemistry’ color scheme inspired by WebLogo (Crooks et al., 2004).

Molecular dynamics (MD) simulations were initialized from three high-resolution structures of DraNramp: the outward-open G223W•Mn²⁺ structure (6BU5) with Mn²⁺ removed and W223 mutated back to the native glycine residue in silico, the inward-open WT•Cd²⁺ structure with Cd²⁺ removed, and the inward-occluded WT structure. Crystallographic waters were retained, and protonation states of key titratable residues were selected with PROPKA (Olsson et al., 2011; Søndergaard et al., 2011) assuming a pH of 5.0 for residues exposed to external solvent and a pH of 7.0 for residues exposed to cytosol, a condition under which DraNramp exhibits high activity. All structures were oriented in the membrane with the PPM web server and membrane systems were prepared with CHARMM-GUI (Jo et al., 2008; Lee et al., 2016). A POPC membrane of surface area 99×99 Å was constructed in the XY plane around the protein (Wu et al., 2014), the system was solvated in a 100×100×100 ų rectangular box using TIP3 waters and electronically neutralized using potassium and chlorine ions at an overall concentration of 150 mM. The overall system size was approximately 103,000 atoms.

All-atom simulations were run using GPU-accelerated NAMD (Phillips et al., 2008) and the CHARMM36m forcefield (Huang et al., 2017). Prior to simulation, the energy of each system was minimized for 10,000 steps using a conjugate gradient and line search algorithm native to NAMD. To improve simulation stability, the system was initially equilibrated using an NVT-ensemble with harmonic restraints placed on protein and lipid heavy atoms. The harmonic restraints were then incrementally relaxed over a period of 675 ps according to established CHARMM-GUI protocols (Lee et al., 2016). The system was then simulated at a constant pressure, utilizing the Langevin piston method to maintain 1 atm at 303.15 K, from anywhere between 617 and 1176 ns depending on the starting conformation. Simulations were performed using periodic boundary conditions and a time step of 3.0 fs with all bonds to hydrogens being constrained. Large integration timesteps were enabled by employing hydrogen mass repartitioning (Hopkins et al., 2015). Long-range electrostatic interactions were calculated using the particle mesh Ewald (PME) method with nonbonded interactions being cut off at 12 Å. Each simulation was performed in duplicate resulting in approximately 2 µs of total sampling for each system. Simulations are summarized in Supplementary file 1i.

RMSD, residue distance, and dihedral analyses were performed using mdtraj version 1.9.8 (McGibbon et al., 2015). For distance analysis, the minimum interatomic distances were identified between the specified residues across each frame. For water analysis, simulations were centered and wrapped in VMD and a Tcl script was used to produce water density maps. Contour maps of these densities were then visualized in PyMOL. Water occupancies of sites coordinated by specific residues were also calculated in an alignment-agnostic manner by determining for each frame in each simulation whether a water was present within 2.5 Å of both specified residues. Distinct rotamers were identified from dihedrals using spectral clustering as implemented in scikit-learn version 1.0.

Atomic coordinates and structure factors for the crystal structures reported in this work have been deposited to the Protein Data Bank under accession numbers 8E5S (WT), 8E5V (WT_soak), 8E60 (WT•Mn²⁺), 8E6H (A47W•Mn²⁺), 8E6I (M230A•Mn²⁺), 8E6L (D296A•Mn²⁺), 8E6M (WT•Cd²⁺), and 8E6N (re-refined G223W•Mn²⁺). Corresponding X-ray diffraction images have been deposited to the SBGrid Data Bank under the respective accession numbers 962 (doi:10.15785/SBGRID/962), 963 (doi:10.15785/SBGRID/963), 964 (doi:10.15785/SBGRID/ 964), 966 (doi:10.15785/SBGRID/966), 967 (doi:10.15785/SBGRID/967), 968 (doi:10.15785/SBGRID/968), 969 (doi:10.15785/SBGRID/969), and previously deposited 564 (doi:10.15785/SBGRID/564). The multiple sequence alignment and phylogenetic tree have been provided as Figure 1—source data 1 and Figure 1—source data 2, respectively. All liposome-based transport data are provided in Figure 1—source data 3. Code for analysis of molecular dynamics data, as well as the raw data plotted in Figure 4—figure supplement 2 and Figure 4—figure supplement 3, can be found at https://github.com/samberry19/nramp-md (MIT license). Raw molecular dynamics trajectory files are available on Dryad (https://doi.org/10.5061/dryad.tx95x6b2b). Source files (origin files) of all ITC experiments are provided in Appendix 1—table 1—source data 1 (Mn²⁺ isotherms), Appendix 1—table 2—source data 1 (Cd²⁺ isotherms) and Figure 5—source data 1(Mg²⁺ isotherms).