Structural insights into the elevator-like mechanism of the sodium/citrate symporter CitS

The sodium-dependent citrate transporter of Klebsiella pneumoniae (KpCitS) belongs to the 2-hydroxycarboxylate transporter (2-HCT) family and allows the cell to use citrate as sole carbon and energy source in anaerobic conditions. Here we present crystal structures of KpCitS in citrate-bound outward-facing, citrate-bound asymmetric, and citrate-free inward-facing state. The structures reveal that the KpCitS dimerization domain remains stationary throughout the transport cycle due to a hydrogen bond network as well as extensive hydrophobic interactions. In contrast, its transport domain undergoes a ~35° rigid-body rotation and a ~17 Å translocation perpendicular to the membrane to expose the substrate-binding site alternately to either side of the membrane. Furthermore, homology models of two other 2-HCT proteins based on the KpCitS structure offer structural insights into their differences in substrate specificity at a molecular level. On the basis of our results and previous biochemical data, we propose that the activity of the 2-HCT CitS involves an elevator-like movement in which the transport domain itself traverses the lipid bilayer, carrying the substrate into the cell in a sodium-dependent manner.

forms of KpCitS were obtained in the citrate-containing droplet. One crystal form (form 1), refined at 4.0 Å resolution, contains an outward-facing KpCitS homodimer ( Supplementary Fig. 1). In contrast, the other crystal form (form 2), refined to 3.8 Å resolution, contains two pairs of KpCitS dimers in an asymmetric unit; one has an outward-facing conformation, and the other an asymmetric conformation. Citrate-free KpCitS yielded a crystal diffracting to 3.6 Å resolution (form 3) with two inward-facing KpCitS dimers in an asymmetric unit. The dimers are practically identical, with Cα rms differences below 0.3 Å, suggesting that this state is structurally stable in the crystallization conditions. We also tried solubilizing bacterial membranes with n-decyl-β-D-maltopyranoside as used in the structure determination of SeCitS 22 . Interestingly, this method only produced a crystal of an asymmetric KpCitS homodimer in the presence of citrate (form 4), and it was refined at a resolution of 3.5 Å. For the sake of the accuracy, we selected the structure of the highest resolution dataset from each of the three major conformational states for analysis; thus, the outward-facing, inward-facing, and asymmetric conformations of crystal forms 2, 3, and 4, respectively, are shown throughout this article (Supplementary Figs 2, 3 and 4).
Our crystal structures reveal that the KpCitS protomer is formed by homologous N-terminal and C-terminal halves with inverted topology, each containing six transmembrane helices (TM 2-5, 7 and TM [8][9][10][11]13) and one helical hairpin (TM 6a/b (or HP1) and TM 12a/b (or HP2)) (Fig. 1a). TM 1 with the N-terminus toward cytoplasm is not involved in either of the two halves. KpCitS protomer can be divided into two domains; a central dimerization domain (TM 1-4 and 8-10) critical for formation of the dimeric structure, and a peripheral transport domain (TM 5-7 and 11-13) containing residues needed for interaction with citrate and sodium ions. The outward-and inward-facing structures contain a pseudo two-fold symmetry at the center (Fig. 1b). Each KpCitS protomer binds one molecule of citrate in the outward-facing and asymmetric conformations, indicating that the functional unit for citrate uptake and transport is a monomer (Fig. 1b,c). As expected from the 92% sequence identity, the CitS structures obtained from Klebsiella pneumoniae and Salmonella enterica are highly homologous. The Cα rms differences of the backbone atoms are ~1.1 Å for the outward-facing, and ~0.3 Å for the inward-facing protomers, suggesting that they have very similar overall structures and that their mechanism of substrate transport is conserved ( Supplementary Fig. 5). Throughout this article, single apostrophes are used for the residues of one protomer to differentiate them from those of the other protomer.
The dimerization domains of KpCitS in the transport cycle. The crystal structures in the three functional states show that the dimerization domains remain fixed throughout the transport cycle due to extensive hydrophobic interactions ( Fig. 1 and Supplementary Fig. 6). The dimer interface has an approximate area of 2,700 Å 2 , and contains a large hydrophobic cavity with a volume of 900 Å 3 (Fig. 2a). The internal cavity is semi-circular and is completely lined with hydrophobic residues, mainly from helices TM 2, 4, 8 and 10 (Fig. 2b). One detergent molecule of n-octyl-β-D-glucopyranoside (β-OG), used as additive in protein crystallization, fills the central cavity at the level of the membrane's inner leaflet ( Supplementary Fig. 7). Consistent with the hydrophobic nature of the cavity, the hydrophilic head group of β-OG is exposed to the cytoplasm, and the non-polar carbon tail is inserted into the cavity in an extended conformation, with a structure that is nearly identical to that of the lipid chain within the dimeric interface of SeCitS 22 . Since the cavity is large enough to bind the long lipid acyl chains, KpCitS within a native membrane environment would permit the entry of a lipid molecule, which may have some effect on the stability and activity of KpCitS by virtue of hydrophobic interactions with residues of the dimer interface 23 .
The substrate binding and its transport. The transport domain of KpCitS undergoes a large conformational change associated with substrate binding ( Fig. 1 and Supplementary Fig. 6). The most obvious changes are concentrated on the two helical hairpins HP1 and HP2 arranged vertical to the membrane (Fig. 3a). In the outward-facing state, the hairpin loops are located close to the membrane/periplasm interface, which appears to favor the entry of external substrate and ions (Fig. 3a, left). This is consistent with mutational and cross-linking data showing that the hairpin loops, which contain conserved GGxG motives, are positioned close together to form the binding sites for the citrate and sodium ions 24,25 . In the inward-facing state, whether or not citrate binding, the hairpins slide downwards relative to the dimerization domain as a result of an approximately 35° rigid-body rotation and 17 Å translocation perpendicular to the membrane; this opens the substrate transport pathway to the cytoplasm as far as the inner leaflet of the membrane (Figs 1c and 3a, middle and right).
The substrate transport pathway of KpCitS is located between the transport and dimerization domains (Figs 1b,c and 3a). In the outward-facing state the citrate is firmly fixed by essential hydrogen bonds and ionic interactions as follows (Fig. 3b, left and Supplementary Fig. 9a): (i) the hydroxyl group of the 2-hydroxycarboxylate motif of citrate forms hydrogen bonds with the hydroxyl side chain and the backbone nitrogen of S405, and with the backbone oxygen of R402, (ii) the carboxylate group of the 2-hydroxycarboxylate motif forms hydrogen bonds with the backbone nitrogens of N186 and G187, (iii) the 1-carboxylate group of the citrate makes ionic and hydrogen bonds with the side chain of R428 and the backbone nitrogen of G404, and (iv) the 3-carboxylate group of citrate makes hydrogen bonds with the side chains of N186 and R402. Y348 on TM 10 also participates in the coordination of citrate via an anion-π interaction with the 3-carboxylate group [26][27][28] . The highly conserved Q424 is not directly involved in the coordination, but contributes to orienting R428 to permit it to form strong hydrogen bonds with the 1-carboxylate group of citrate. In the outward-facing state the bound citrate is partially exposed to the periplasmic space. However, it is unlikely that it can escape backwards because it is secured by strong hydrogen and electrostatic interactions, as mentioned above. This is different from the crystal structures of other secondary active transporters such as GltPh, TtNapA, SLC26_Dg, PaNhaP, MjNhaP1, BetP and UapA, in which the substrate is completely occluded from the aqueous phase [29][30][31][32][33][34][35][36][37][38] .
In the citrate-bound inward-facing state, the backbone structure of the substrate binding site hardly changes compared to that in the outward-facing state. However, the critical side chains of the residues coordinating the citrate move substantially (Fig. 3b, left and middle). The side chain of Q424 rotates by 55°, which results in a conformational change of R428, and weakens the ionic interaction with the 1-carboxylate group of citrate. Furthermore, the entire transport domain moves downwards relative to the dimerization domain, so that Y348 on TM10 completely loses the ability to coordinate the 3-carboxylate group of citrate. Consequently, the citrate is weakly attached only to the side chain and the backbone nitrogen of S405 and the backbone nitrogen of G404 (Fig. 3b, middle and Supplementary Fig. 9b). They are consistent with the KpCitS proteoliposome data showing that citrate transport occurs in an asymmetric conformation in which one binding site has a high and the other a low affinity for substrate 4 . In the citrate-free inward-facing state, we observed that one water molecule takes the place of a citrate (Fig. 3a,b, right). Since the substrate binding site is continuous with the cytoplasm, water molecules are able to move deep into the membrane inner leaflet. In addition to the changes in amino acid conformation as described above, polar water molecules also seem to facilitate release of the substrate into the cell by breaking the electrostatic interactions between substrate and transporter. This idea is supported by the fact that the bound citrate in the inward-facing protomer is only partially hydrated in the SeCitS structure 22 .
Substrate selectivity of the 2-HCT family. In order to understand the differences in substrate selectivity between KpCitS and homologous 2-HCT proteins, we generated homology models of CitP of Leuconostoc mesenteroides (LmCitP, citrate/lactate exchanger) and MleP of Lactococcus lactis (LlMleP, malate/lactate exchanger) 1, 39 based on the KpCitS structures determined in this study. To increase the accuracy of the homology modeling, we excluded the N-terminus  and the linker region between TM 7 and TM 8 (237-265) of KpCitS, in which there is low sequence conservation. The final models of LmCitP and LlMleP had similar folds to those of KpCitS, with 31% and 33% sequence identity, and 56% and 43% similarity, respectively. In the modeled structures, S405 of KpCitS is changed to threonine at positions 402 and 384 of LmCitP and LlMleP, respectively, where it mediates the invariant hydrogen bonding with the hydroxyl group of the 2-hydroxycarboxylate of citrate (Fig. 4). This is consistent with data, showing that removal of the hydroxyl group reduces the affinity for substrate of these proteins 1 . LmCitP is known to transport both malate and citrate, but LlMleP only transports malate 39 . Also, both transporters are highly stereoselective, that is, they favor substrates with (S)-enantiomeric rather than (R)-enantiomeric configurations. The modeled structures show that N186, Y348 and R402 in KpCitS are replaced by V190, L345 and M399 in LmCitP (Fig. 4, middle) and by M173, L327 and M381 in LlMleP (Fig. 4, right). These alterations may affect the apparent affinity for substrate and substrate specificity by increasing the hydrophobicity and removing the anion-π interaction near the 3-carboxylate group of citrate, resulting in a shift of substrate preference towards (S)-malate. The modeled structure is consistent with previous biochemical data showing that mutation of N186 of KpCitS to a valine severely reduces affinity for citrate with only minor changes in Vmax 40 . It is also consistent with functional data indicating that LmCitP has ~40% higher activity for malate than for citrate 1 .   In particular, M173 of LlMleP is located in a position such that it can directly collide with the 3-carboxylate group of citrate, which accounts for why the latter is a substrate of LmCitP but not of LlMleP (Fig. 4, right) 39 .

Discussion
We have studied the molecular mechanism of CitS in the 2-HCT family by determining multiple crystal structures along the transport cycle. Our results show that the KpCitS has a homodimeric structure in which its dimerization domain remains fixed during the transport cycle due to a hydrogen bond network as well as extensive hydrophobic interactions, while the substrate-loaded transport domain undergoes an elevator-type translocation (i.e. up and down) across the membrane. Our crystal structures are fully consistent with previous functional evidence regarding the interaction between substrates and 2-HCT proteins. For example, site-directed mutagenesis demonstrated that conservative mutation of R428 to a positively charged lysine in LmCitP and in CimH of Bacillus subtilis (BsCimH, citrate/malate symporter), caused a 10-fold decrease in affinity for citrate. However, their replacement with uncharged cysteines had a more serious effect, reducing the affinity for citrate more than 20-fold 15,41 . Reaction of the cysteine mutants with MTSEA (2-aminoethyl-methanethiosulfonate), which introduces a positive charge on the thiol groups of cysteine residues, resulted in 50-fold and 2-fold increases in transport activity, respectively 14,15 . Conversely, the addition of a negative charge to the cysteine mutant of BsCimH by reaction with MTSES (2-sulfonatoethyl-methanethiosulfonate) resulted in a 2-fold decrease in activity, highlighting the importance of electrostatic interactions between substrate and transporter. Our structural observations are also consistent with previous evidence that mutations of residues in HP1 and HP2 cause substantial reductions in the transport function of KpCitS (Supplementary Fig. 10) 24,25 . Locating these residues in our KpCitS structure showed that they participate directly in either sodium binding or substrate coordination. Thus, mutation of these residues seems to result in reduced affinity for ions/substrate, or impede the elevator-like movement of the transport domain.
From the KpCitS structures, it is immediately evident how the internal substrate-binding site is alternatively accessible on one side and then the other of the membrane during the transport cycle (Fig. 1c) 42 . In the Figure 5. Alternating access mechanism of the KpCitS. (a) In the outward-facing conformation, the transport pathway is blocked by I121, L408 and M417. In the inward-facing conformation, these residues move and citrate is released to the cytoplasm. The citrate is shown in orange and red. The release of citrate in the inward-facing conformation is marked by a black arrow. (b) Role of the polar residues in the alternating access mechanism. In the inward-facing conformation, D112, E195, R205 and Y348 form a salt bridge network at the membrane/ periplasm interface, by virtue of which the substrate transport pathway is open only towards the cytoplasm. The view is rotated by 30 degrees along the vertical axis from that of (a).
outward-facing conformation, L408 and I121 are in close proximity, and serve as a latch governing access of substrate to the cytoplasm (Fig. 5a). In addition, M417 seems to act as a plug preventing substrate exit. The conformational change to the inward-facing state relocates L408 and M417, opening the substrate binding site to the cytosol. Conversely, D112, E195, R205 and Y348 form a salt bridge network in the inward-facing conformation, blocking entry of substrate from the periplasm (Fig. 5b). A change of surface charge distribution also supports the alternating access mechanism of KpCitS ( Supplementary Fig. 11); the electrostatic surface potential at the periplasmic and cytoplasmic surfaces alternates between the positive and neutral charges along the transport cycle that will assist binding of the negatively charged citrate. While the SeCitS structure revealed that two sodium ions (Na1 and Na2) are coupled to the substrate binding site 22 , we observed only Na1 between the two hairpin loops of KpCitS in all three states (Fig. 6). Na1 is coordinated by the backbone oxygens of I181, G183, M399 and N401 with either an octahedral or a trigonal bipyramidal geometry with two and one ordered water molecule, respectively. As in the SeCitS structure, the proposed binding site for Na2 in KpCitS is coordinated by side chains of S427, N401 and D407 and the backbone carbonyl oxygen of C398, but no electron density for Na2 was observed in either protomer (Fig. 6). There are three possible reasons why only one sodium ion is observed in each protomer of our structures. The first is that although Na2 is bound to the transporter, the resolution of the data is not high enough to identify its position. The second possibility is that Na1 is sufficient for binding of substrate, but not for transporting it in the outward-facing protomer.
The crystal structure of KpCitS shows five and three glycine residues, respectively, in the HP1 and HP2 loops (Fig. 6e,f). This is reminiscent of HP2 in GltPh 43 . As in that case, we imagine that the cluster of glycines is likely to confer high structural flexibility on the HP loops of KpCitS, enabling its conformation to change easily to provide a perfect fit for a substrate molecule upon binding of one sodium ion. As KpCitS has been shown to transport substrate coupled with two sodium ions 16,19 , we think it is likely that the binding of Na1 precedes or is nearly simultaneous with that of substrate, and that Na2 is necessary only for substrate transport. The third possibility is that our KpCitS structures may have been captured after Na2 had already been released from the inward-facing protomer into the cell. This idea is supported by many structures of secondary active transporters as well as their MD simulations, which show that substrate release is initiated by dissociation of a sodium ion 4,16,22,31,32,[44][45][46][47][48] .
The structural data for KpCitS, combined with the previous results of diverse experimental approaches, allow us to propose a molecular mechanism for the CitS in the 2-HCT family 4,16,19,24,42 (Fig. 7). In this model, the outward-facing conformation is in the apo state with substrate and sodium ions unbound. When the protein is occupied by sodium ions, it presents a suitable binding site for substrate, and undergoes a rotational and vertical movement across the membrane. This elevator-like movement is accompanied by changes of amino acid conformation essential for substrate binding and hydration of the substrate binding site by water influx, resulting in disruption of the protein-substrate interaction, release of substrate and sodium ions, and return to the apo state. Unfortunately, current structural data do not clearly address whether this process occurs in both of the protomers at the same time, whether it occurs in only one protomer, or whether it occurs alternately in the two protomers 16 (Fig. 7).
In summary, we report the first crystal structures of KpCitS in different functional states. Our crystal structures, together with the previously reported structural data, provide evidence that the elevator-type alternating Figure 7. Proposed transport mechanism of the CitS. Dimerization and transport domains are shown in dark and pale blue, respectively. The helical hairpins, HP1 and HP2, of the transport domain are represented as cylinders and loops in purple. Citrate is shown as an orange diamond, and sodium ions as black spheres. In the apo state, the binding site of substrate and ions is open toward the external environment. Binding of sodium ions prepares the protein to interact with its substrate, which promotes the elevator-like movement of the transport domain in either one of two protomers in the dimer or both. Dissociation of citrate and ions into the cytoplasm resets the protein into the outward-facing apo state. The crystal structures of CitS that have been determined by us and others 22 are marked with black stars. access mechanism is widespread among secondary active transporters 22,34,38,[49][50][51] . In essence we propose that CitS proteins employ a sodium dependent elevator-like movement in which the dimerization domain is held fixed during the transport cycle, and the transport domain moves up and down across the membrane bilayer to translocate the substrate. Further research is required to unravel many details of this model, such as how substrate transport by 2-HCT proteins is coupled to the ion gradient and/or ion binding 52 .