Structure and function of the divalent anion/Na+ symporter from Vibrio cholerae and a humanized variant

Integral membrane proteins of the divalent anion/Na+ symporter (DASS) family translocate dicarboxylate, tricarboxylate or sulphate across cell membranes, typically by utilizing the preexisting Na+ gradient. The molecular determinants for substrate recognition by DASS remain obscure, largely owing to the absence of any substrate-bound DASS structure. Here we present 2.8-Å resolution X-ray structures of VcINDY, a DASS from Vibrio cholerae that catalyses the co-transport of Na+ and succinate. These structures portray the Na+-bound VcINDY in complexes with succinate and citrate, elucidating the binding sites for substrate and two Na+ ions. Furthermore, we report the structures of a humanized variant of VcINDY in complexes with succinate and citrate, which predict how a human citrate-transporting DASS may interact with its bound substrate. Our findings provide insights into metabolite transport by DASS, establishing a molecular basis for future studies on the regulation of this transport process.

Since citrate, succinate and other C 4 -dicarboxylates are important intermediates and/or regulators of the energy metabolism, the modulation of DASS function expectedly impacts fatty acid synthesis, energy expenditure and life span. Particularly, a handful of mutations found in a DASS-encoding gene nearly doubled the average adult life span in fruit flies, likely by promoting a metabolic state that mimics caloric and dietary restriction 14,15 . In addition, the knockdown of either one of the genes encoding NaDC2 and NaCT in worms decreased their body size and fat content, and/or increased their life span 16,17 . More recently, deletion of the gene-encoding NaCT protected mice from the adiposity and insulin resistance induced by high-fat feeding and aging 18 . Altogether, mounting evidence supported the utility of DASS proteins, particularly NaCT, as therapeutic targets for tackling obesity, type 2 diabetes and other metabolic diseases 1,2 .
Despite such clinical significance, the molecular mechanism underlying DASS-mediated co-transport remains poorly understood. The 3.2-Å resolution crystal structure of citrate-bound VcINDY, a DASS from Vibrio cholerae, elucidates the transporter architecture and implicates the amino acids in Na þ and citrate binding 19 . However, like other well-characterized bacterial DASS proteins, VcINDY is known to transport succinate and other C 4 -dicarboxylates, rather than citrate (C 6 -tricarboxylate) 13,19 .
Furthermore, although VcINDY was suggested to catalyse the co-transport of three or more Na þ ions and C 4 -dicarboxylate 13 , only one Na þ -binding site was observed crystallographically in the published work 19 . Although a second Na þ -binding site was predicted 19 , no direct structural evidence was available and the assignment of this predicted site remained speculative. Therefore, previous studies fell short of addressing a central question in DASS-mediated transport: how does the transporter recognize substrate and multiple Na þ ions?
Herein we present the structure of succinate-bound VcINDY at a resolution of 2.8 Å, which, to our knowledge, offers the first molecular view of a substrate-bound DASS. Our data enabled us to elucidate a previously undiscovered Na þ -binding site in VcINDY as well as how this transporter utilizes helix dipoles and Na þ binding to interact with trans-C 4 -dicarboxylate. Furthermore, we report the structure of a citrate-bound VcINDY alongside those of the succinate-and citrate-bound MT5, a humanized variant of VcINDY, all determined to 2.8-Å resolution. Collectively, our results bring to light how citrate inhibits VcINDY-mediated succinate transport and how a DASS distinguishes between C 4 -dicarboxylate and C 6 -tricarboxylate, thereby recasting the conceptual framework for understanding how a DASS specifically recognizes and transports its anionic substrate.

Results
Structure of a substrate-bound VcINDY. We crystallized VcINDY in the presence of succinate and Na þ , and determined the structure by combining molecular replacement and multiple isomorphous replacement and anomalous scattering (MIRAS) phasing (Table 1, Supplementary Table 1). Although not essential for solving the phase problem, the MIRAS phases allowed us to model substantially more amino acids than what had been achieved previously 19 (445 versus 398 out of 462 residues), to locate the bound succinate with confidence, and to compare different co-structures without interference from model bias. Moreover, the density-modified MIRAS maps revealed that parts of the published VcINDY structure 19 , which are directly related to the binding of Na þ and citrate, had been incorrectly determined (see below). Using the updated structures and resulting F o -F c electron density maps the other co-structures were well-determined by and also supported by the unbiased MIRAS maps. Our structure reveals a homodimeric arrangement (Fig. 1a), with each VcINDY protomer comprising eleven membranespanning helices (TM1-TM11), two re-entrant helix-turn-helix hairpins (HP in and HP out ) and two interfacial helices (H4c and H9c; Supplementary Fig. 1). Viewed in parallel to the membrane, the VcINDY dimer is shaped like an inverted bowl with its concave mouth facing the cytoplasm, allowing the aqueous solution to reach the midpoint of the membrane (Fig. 1b). In each VcINDY protomer, the inter-helical loops in HP in and HP out , and the intra-helical loops within the discontinuous TM5 and TM10 meet approximately halfway across the membrane, forming a cleft that opens toward the cytoplasm.
Na þ -binding sites in VcINDY. Within this cleft, the binding sites of two Na þ ions, designated as Na1 and Na2, were observed ( Supplementary Fig. 2). We assigned the Na þ -binding sites based on the following lines of evidence. First, the electron density ascribed to Na þ is surrounded by electronegative oxygen atoms, which is consistent with a cation coordination shell. Second, Na þ was used throughout the VcINDY purification and was the only inorganic cation included in the crystallization solution. Third, the distances between the ions and liganding atoms range from 2.3 to 2.7 Å, which are appropriate for Na þ coordination but too short for H-bonds formed between water molecules and VcINDY 20 . Indeed, we performed the valence test on the two sites, which is suggestive of the binding of Na þ ions rather than water molecules 21 . Fourth, the two putative cations are pentacoordinated, which represents one of the most common Na þ coordination arrangements 20 . CheckMyMetal, a structure validation server for metal-binding sites 22 , supported the assignment of Na1 and Na2 as the binding sites for Na þ , rather than K þ or Ca 2 þ . Fifth, a single amino-acid change in either Na1 or Na2 affected the Na þ -dependence of VcINDYmediated transport (see below).
Previously, Na þ in Na2 was not identified in VcINDY (ref. 19), likely due to errors in model building ( Supplementary  Figs 3 and 4). In each of the two Na þ -binding sites now elucidated in VcINDY, Na þ is penta-coordinated to two aminoacid side-chain and three backbone carbonyl oxygen atoms. Specifically, Na þ in Na1 is coordinated to the side-chain hydroxyl of S146 and side-chain carbonyl of N151, in addition to the main-chain carbonyls of S146, S150 and G199 (Fig. 2a); whereas Na þ in Na2 binds to the side-chain hydroxyl of T373 and side-chain carbonyl of N378, besides the backbone carbonyls of T373, A376 and A420 (Fig. 2b). Overall, the two Na þ ions are separated by 413 Å and bound to the pseudo-symmetry-related HP in and HP out , respectively. However, this symmetry is broken when detailed coordination interactions are inspected. In particular, S150 in HP in binds to Na þ , whereas its counterpart in HP out , that of S377, does not. In addition, I149 in HP in makes no contact with Na þ , but its equivalent in HP out , A376, does.
Functional importance of the Na þ -binding sites. Moreover, the mutations of S146, N151 and N378, impaired the transporter activity of VcINDY in a whole-cell based assay, even though these mutations did not substantially affect the expression level of the transporter 19 . Furthermore, we replaced S146 and T373 individually with alanine and reconstituted the detergent-purified mutants into proteoliposomes. We then compared the function of the mutants with that of the wild type transporter. Under optimal conditions, VcINDY exhibited robust succinate transport activity in a counter-flow assay 13 (Fig. 3a). Our kinetic studies on VcINDY revealed that the K M and V max for succinate transport are 0.22 mM and 254 nmol mg À 1 min À 1 , respectively. At succinate concentration below the K M , we found the VcINDYcatalyzed transport to be highly dependent on Na þ concentrations, with a Hill coefficient of 2.7 and a K 0.5-Na of 15 mM, respectively (Fig. 3b). Significantly, mutations of S146 and T373 increased the K M for succinate transport by more than 11-and 1,227-fold and elevated the K 0.5-Na to 37 and 4800 mM, respectively (Fig. 4).
Our data suggested that VcINDY binds at least two Na þ ions during transport and that Na1 and Na2 play pivotal roles in the succinate transport. Indeed, mutation S146A appeared to diminish the ability of membrane-embedded VcINDY to bind Na þ , because higher concentrations of Na þ were required by the mutant to effect maximal stimulation of succinate transport than VcINDY (Fig. 4b). In addition, mutation T373A abolished the ability of Na þ to stimulate VcINDY-mediated transport within the tested Na þ concentration range (Fig. 4b). Furthermore, mutation S146A or T373A raised the K M for succinate, albeit to a different extent (Fig. 4a), likely by indirectly weakening the binding of substrate to VcINDY within the membrane, as  Na þ coordination stabilizes the bound succinate in VcINDY (see below).
Of particular interest, T373A had more pronounced effect on the transport function than the structurally analogous S146A (Fig. 2), arguing that Na2 has a more important role than Na1 during transport and the two sites are functionally non-equivalent. In addition, mutations of human NaDC3 S143 , NaDC3 N144 and NaDC3 N484 , which are equivalent of VcINDY S150 , VcINDY N151 and VcINDY N378 , respectively, exerted deleterious effects on the transport function 23 . In particular, NaDC3 S143A exhibited impaired transport function, whereas NaDC3 N144A or NaDC3 N484A had almost no measurable transporter activity. These data further suggested that Na1 and Na2 are preserved in NaDC3 and our structure is a valid model for studying the mechanism of human DASS.
Succinate-binding site in VcINDY. Within the Na þ -binding cleft in VcINDY, the electron density for succinate was also observed ( Supplementary Fig. 5). The bound succinate makes contacts with VcINDY mainly through H-bonding interactions and is partly exposed to cytoplasm (Fig. 1b), indicating that the transporter adopts an inward-open conformation. Specifically, the side-chain amide of N151 and the side-chain hydroxyl of T152 donate H-bonds to one carboxylate in succinate, whereas the side-chain hydroxyl S377 and the side-chain amide of N378 make contacts with the other succinate carboxylate (Fig. 5a). Moreover, the side chain of T379 interacts with the aliphatic portion of the succinate through van der Waals interactions, and S200 and T421 form H-bonds with N151 and N378, respectively, which may further stabilize the interactions between succinate and VcINDY.
Significantly, the bound succinate adopts an extended conformation, structurally mimicking fumarate, a C 4 -dicarboxylate with a central trans double bond. This structural mimicry implied that fumarate interacts with VcINDY similarly to succinate, thus explaining why fumarate effectively inhibits VcINDY-mediated succinate transport 13,19 . By contrast, maleate, the cis isomer of fumarate, exerts less inhibitory effect on the succinate transport than fumarate 13 , arguing that VcINDY is specific for C 4 -dicarboxylate in a stretched conformation. Notably, most of the succinate-interacting amino acids are conserved, implying that the preference for trans-dicarboxylate is not limited to VcINDY but also shared by other DASS proteins.
Furthermore, the alanine substitution of N151 or N378 reduced the binding of succinate to VcINDY and gave rise to severely impaired transporter activity 19 . By contrast, the mutation of T379, which makes van der Waals interactions with the succinate, had only moderately deleterious impact on the transport function 19 . However, T379 was suggested to play a role in substrate recognition 19 and is often replaced by valine in The initial rates of succinate transport at 200 mM external substrate concentration (B8% of the K M for S146A) and pH 7.4 were plotted against the common logarithm of external Na þ concentrations. The data for S146A were averaged to fit to the Hill equation, yielding a Hill coefficient of 3.2 and a K 0.5-Na of 36.9 mM. T373A-mediated transport of succinate lacked Na þ -dependence within the tested range (up to 800 mM), suggesting that the K 0.5-Na for T373A is 4800 mM. Similar observations were made with 500 mM (B20% of the K M ) and 4 mM external succinate for S146A and T373A, respectively. At 500 mM succinate, the data for S146A revealed a Hill coefficient of 3.1 and a K 0.5-Na of 47.6 mM. For comparison, the data for VcINDY were plotted to highlight the functional consequences of mutations in Na1 and Na2.
Insights into the symport mechanism. Previous studies suggested that Na þ ions bind to DASS before it can bind its substrate [9][10][11][12] , implying that the coordination of Na þ promotes substrate binding. In the succinate-binding VcINDY, the Na þ ions in Na1 and Na2 coordinate several succinate-binding amino acids and thus stabilize the conformation of these amino-acid side chains. This arrangement helps to explain why the transport of succinate and Na þ is strictly coupled, as they bind to a common subset of amino acids, and the binding or unbinding of one likely affects that of the other. Moreover, the Na þ ions may attract succinate, which carries two negative charges 13 , through longrange electrostatic interactions within the low-dielectric intramembrane environment 24,25 . Furthermore, the amino ends of four short helices from HP in , HP out , TM5 and TM10, which possess localized positive dipoles, all point toward the bound succinate and stabilize its two negative charges (crystallization pHB7). Notably, the stabilization of negative charges by the opposing, positive helix dipoles within inverted structural repeats may represent a general theme for achieving anion selectivity within the lipid bilayer by membrane proteins [26][27][28][29] . The succinate-bound VcINDY structure also implies that the substrate is released from the intracellular-facing transporter prior to the dissociation of Na þ ions from Na1 and Na2, since the bound substrate is more accessible to the cytoplasm than the two Na þ ions (Fig. 1b). Furthermore, Na2 appears functionally more important than Na1, implying that the binding of Na þ to Na2 in an outward-facing VcINDY precedes as well as promotes the Na þ coordination at Na1. Previous studies also implied that VcINDY catalyses the co-transport of three or more Na þ ions and succinate 13 , although only two Na þ -binding sites are apparent in our structure. One possibility is that one or more Na þ ions may have already been released from VcINDY before it adopts the current inward-open conformation and evaded detection by our structural study.
Structure of the citrate-bound VcINDY. To uncover how VcINDY distinguishes between C 4 -dicarboxylate and C 6 -tricarboxylate, we determined the structure of a citrate-bound VcINDY (Table 1, Supplementary Table 2). Citrate inhibits the VcINDY-mediated succinate transport, likely as a competitive inhibitor 13,19 . We found the citrate-bound VcINDY structure to be similar to the succinate-bound form, with a root mean squared deviation (r.m.s.d. of o1 Å for 445 Ca positions. In the citratebound VcINDY, the side-chain amide of N151 and the side-chain hydroxyl of T152 donate H-bonds to a terminal carboxylate (pro-R) of the bound citrate 30 , whereas the side-chain hydroxyl of S377 and the side-chain amide of N378 form H-bonds with the central carboxylate of citrate (Fig. 5b). Moreover, the side chains of P201, V322 and T379 make van der Waals interactions with the bound citrate. One terminal carboxylate (pro-S) and the hydroxyl of citrate, however, make no interaction with VcINDY and project towards the solvent.
Notably, the positioning of citrate in VcINDY ( Supplementary  Fig. 6) is different from that in a published structure 19 . To examine whether this difference is real, we refined our structural model against the published X-ray data, which gave rise to better protein stereochemistry and a simultaneous drop in R free (42%). Significantly, our analysis suggested that the citrate had been incorrectly modelled previously (Supplementary Fig. 7). Despite the difference in citrate placement and incorrectly modelled amino acids ( Supplementary Fig. 4), the two citrate-bound structures could be superimposed onto each other to yield a r.m.s.d. of o1 Å for 393 common Ca positions, indicating that both structures captured VcINDY in the inward-open conformation.
More importantly, our co-structure indicated that the citrateand succinate-binding sites overlap substantially in VcINDY, and that citrate inhibits the transport of succinate by preoccupying the substrate-binding site, that is, as a competitive inhibitor. In our VcINDY structures, the two carboxylates of succinate occupy virtually the same position as the central and pro-R carboxylates of citrate. This observation indicated that HP in , HP out and the unwound region in TM10 constitute a 'trans-dicarboxylaterecognition' module in DASS. A prominent feature of this module is the absence of any protonatable or positively charged amino acids, starkly contrasting the succinate-binding water- In vivo, at least two carboxylates in citrate are deprotonated and negatively charged 13 . In the citrate-bound structure, the pro-S carboxylate, which probably carries a full negative charge (crystallization pHB7), makes no interaction with VcINDY. Rather, P201, V322 and T379 pack against the bound citrate and steer the pro-S carboxylate away from the transporter. By contrast, both carboxylates in succinate are stabilized by the H-bonding interactions made with VcINDY. Therefore, the negative charges in citrate may not be fully 'neutralized' by its interactions with VcINDY in the membrane bilayer. This finding may explain why citrate is less effective in inhibiting VcINDYmediated succinate transport than C 4 -dicarboxylates and why citrate preferably binds to the inward-facing VcINDY (refs 13,19). Assimilating the data from published studies 36,37 and our VcINDY structures, we argued that the interactions between citrate and the inward-or outward-facing VcINDY are similar but not identical: in particular, the pro-S carboxylate is less solvent-exposed in the outward-facing state and less stabilized by solvation than that in the inward-facing protein. Consequently, it would be energetically unfavourable to form the citrate-VcINDY complex in the membrane bilayer due to its negative charge surplus, especially in the outward-facing state.
Structures of a humanized variant of VcINDY. To validate this idea as well as to gain new insights into the mechanism of human DASS, we replaced eight amino acids surrounding the citratebinding cleft by their counterparts in NaCT ( Supplementary  Fig. 1), which primarily transports citrate 6 . Of note, such a 'multimutational' approach was previously employed to study the interactions between antidepressants and a bacterial homologue of biogenic amine transporters 38 . We crystallized this octuple mutant of VcINDY, denoted MT5, in complex with citrate and determined the crystal structure (Table 1, Supplementary Table 3). Although the MT5 structure remains similar to that of VcINDY, one important difference centres on the pro-S carboxylate of the bound citrate ( Supplementary Fig. 8). Specifically, the side-chain hydroxyl of S200T and the backbone amide of P201G in MT5 each donate an H-bond to the pro-S carboxylate, whereas the side chain of T379V makes van der Waals interactions with the citrate (Fig. 6a). In contrast to that in VcINDY, the pro-S carboxylate latches onto the amino ends of TM5b and the second helix in HP out in MT5, with its putative negative charge stabilized by the positive helix dipoles (crystallization pHB7). Since NaCT transports the trianionic citrate 6 , our structure may foretell the interactions between NaCT and its bound substrate.
We also determined the succinate-bound MT5 structure (Table 1, Supplementary Table 4), which revealed that MT5 binds succinate in the same way as VcINDY except that T379V in MT5 makes no contact with the bound substrate (Fig. 6b). Our transport assay further showed that MT5 retained the succinatetransporting activity, with the K M and V max values both similar to those of VcINDY (Supplementary Fig. 9). These data implied that T379 is not essential for the succinate transport and NaCT may also interact with a trans-dicarboxylate. Although VcINDY and MT5 interacted with succinate similarly, pronounced differences between them were found in the extent to which citrate inhibited the transporter-meditated uptake of succinate (Fig. 7). Specifically, 95 mM citrate reduced the succinate transport rate by o20% in VcINDY, whereas a mere 10 mM citrate decreased the transport rate by 440% in MT5 (at pH 7.4).
Since VcINDY and MT5 were inserted into the liposome membrane in presumably two orientations 13 , our data suggested that the membrane-embedded MT5 interacts with trianionic citrate more strongly than VcINDY in both the inward-and outwardfacing states. This stark difference is likely attributed to the altered pose of the pro-S carboxylate seen in MT5. In support of this notion, the temperature factors for the bound citrate in MT5 are lower than those in VcINDY ( Table 1 ), arguing that the citrate was more firmly bound to MT5 than VcINDY at pHB7. On one hand, such results should be viewed with the caveat that we were unable to elicit any appreciable citrate-transporting activity in VcINDY or MT5, which implied that further experiments would be required to convert MT5 into an efficient citrate transporter. On the other hand, given the amino-acid sequence similarity between VcINDY and NaCT as well as the homologue swap mutations carried by MT5, MT5 likely recapitulates the substrate-binding properties of NaCT considerably.
Substrate-recognition modules in DASS. Taken together, we assert that the amino ends of TM5b and the second helix in HP out form yet another substrate-recognition module in DASS for differentiating C 6 -tricarboxylate from C 4 -dicarboxylate. In a C 4 -dicarboxylate-specific VcINDY, this module includes a Pro and a Thr (Fig. 8a), which selects against citrate by pushing away its pro-S carboxylate and likely giving rise to negative charge surplus within the membrane. In a C 6 -tricarboxylate-transporting NaCT, however, the Pro and Thr are superseded by Gly and Val, respectively, which enables the recognition of pro-S carboxylate via H-bonds and the stabilization of its potential negative charge ( Fig. 8b). Thus, DASS appears equipped with two newfound substrate-recognition modules: one selective for trans-C 4 -dicarboxylate and the other for C 6 -tricarboxylate. Notably, Na þ also contributes to the binding of C 4 -dicarboxylate to DASS by stabilizing the first module. Our findings hence elucidate how a DASS recognizes its substrate and offer a new angle to understand protein-mediated anion transport in general.

Discussion
As noted previously 19 , the N and C domains of VcINDY are structurally related, with an r.m.s.d. of 3.5 Å for 101 common Ca positions, even though the two domains possess opposite membrane topology and share rather modest amino-acid sequence similarity (22% identity). Perhaps surprisingly, the transmembrane domain of VcINDY also exhibits striking structural resemblance to the dimeric AbgT transporters 39,40 despite a lack of patent amino-acid sequence homology. Specifically, the structure of VcINDY can be superimposed onto those of AbgT to yield r.m.s.d. of 3.1 and 3.5 Å for 294 and 305 Ca atoms, respectively. Moreover, VcINDY bears 18 and 13% amino-acid sequence identity to the two AbgT transporters of known structure, respectively. Interestingly, the two AbgT transporters appear to function as antibiotic efflux pumps 39,40 and are likely antiporters or exchangers, whereas most of the known DASS proteins are symporters 1,2 . The structural similarity between the AbgT and DASS protein families adds a new superfamily of secondary membrane transporters with shared dimeric organization and structural fold, notwithstanding their unrelated physiological functions and distinct transport mechanisms. Importantly, a Na þ -binding site similar to Na2 in VcINDY was also found in an AbgT transporter, and alanine substitutions of the cation-coordinating amino acids impaired the transport Positive helix dipoles are highlighted by plus signs, whereas the negatively charged carboxylates in succinate or citrate are indicated by minus signs. The charged state of succinate or citrate is deduced based on the crystallization pH (B7). Both the helix dipoles and Na þ appear to contribute to the anion binding. Furthermore, P201 and T379 may enable VcINDY to select for succinate but against citrate. In MT5, P201 and T379 are replaced by Gly and Val, respectively, which may allow the membrane-embedded transporter to bind citrate more strongly than VcINDY.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15009 ARTICLE function 40 . These data argued that at least one functionally important Na þ -binding site is conserved between the DASS and AbgT protein families. Furthermore, the substrate-binding site in VcINDY lacks any positively charged amino acid, for example, Lys or Arg. DASS has evolved such a scheme probably because positively charged amino acids would discourage the binding of Na þ in their vicinity due to electrostatic repulsion and/or cause the transporter to bind C 4 -dicarboxylate much too tightly, thereby impeding the dissociation of substrate from the inward-facing transporter. Besides VcINDY, at least five membrane transporters of known structure also select for substrates that carry net negative charges: SeCitS from the citrate-sodium symporter family 37 , Glt Ph and Glt Tk from the excitatory amino-acid transporter family 41,42 , NarK and NarU from the nitrate/nitrite porter family [43][44][45] . In contrast to VcINDY, all these transporters utilize one or two Arg residues to bind the negatively charged groups in the substrate.
This seemingly unexpected difference may be understood in light of the substrate-binding site and/or the coupling mechanism. In Glt Ph and Glt Tk , two Na þ -coupled symporters, an Asp residue is located in close proximity to the substrate-binding Arg residue. Therefore, the side-chain carboxylate of this Asp may neutralize the positive charge on the Arg side-chain and weaken the electrostatic attraction between the Arg and substrate, thereby facilitating the release of the negatively charged substrate. In NarK and NarU, both of which are exchangers, the negatively charged nitrate and nitrite can compete against each other for the two substrate-binding Arg residues, thus enabling the binding and unbinding of the substrate as well as the counter-transported ligand. Notably, these four proteins have entirely different fold from that of VcINDY, and their substrate-binding sites appear to have co-evolved with their transport mechanisms.
SeCitS, on the other hand, has a similar but not identical protein fold to that of VcINDY and catalyses the symport of Na þ and citrate. Based on the published results, the transport mechanism of SeCitS may be similar to that of VcINDY 36,37 . However, the SeCitS-mediated transport of citrate is electroneutral 37 , whereas VcINDY-catalyzed succinate transport appears electrogenic 13 . This difference implies that VcINDY interacts with three or more Na þ ions while SeCitS binds only two Na þ ions during transport. Probably, the higher number of co-transported Na þ ions enables VcINDY to recognize and transport negatively charged substrates without engaging any positively charged Arg or Lys.
The lack of charged amino acid in the substrate-and Na þbinding sites in VcINDY also begged the question of how a DASS stabilizes its unloaded or apo state. In Glt Tk and Glt Ph , a substrate-binding Arg occupies the substrate-binding site in the empty transporter 46,47 and apparently acts as a substrate surrogate to stabilize the apo state. Upon the release of substrate and Na þ ions into the cytoplasm, we suspect that water molecules may interact with the substrate-and/or Na þbinding amino acids in VcINDY to stabilize its apo state, since both the substrate-and Na þ -binding sites are solvent-accessible in the inward-facing transporter (Fig. 1). In VcINDY, water molecules may be sufficient to compensate for the local charge imbalance caused by the dissociation of substrate and Na þ ions, since no charged amino acid in the vacant substrate-and Na þbinding sites needs to be 'neutralized'. If this notion holds true, then VcINDY may also behave as a water 'carrier' in the cell 48 .
Protein crystallization and crystal derivatization. Prior to crystallization, VcINDY or MT5 was concentrated to B10 mg ml À 1 and dialyzed extensively against 100 mM NaCl, 100 mM sodium succinate or citrate (pHB7), 20% glycerol, 0.14% DM and 1 mM TCEP at 4°C. Crystallization experiments were performed using the hanging-drop vapor-diffusion method at 22°C. The protein samples were mixed with equal volume of a crystallization solution containing 100 mM NaCl, 100 mM sodium succinate or citrate (pHB7), 36-40% (wt/vol) PEG1000, 0.14% DM, 20% glycerol and 1 mM TCEP. Protein crystals usually appeared within two weeks and continued to grow to full size in a month. For heavy atom derivatization, protein crystals were incubated with 20-50 mM heavy metal compounds for 45 h at 22°C.
Structure determination and refinement. X-ray diffraction data were collected on the frozen crystals at the beam-lines 23-ID and 22-ID at Argonne National Laboratory. X-ray data were processed using the programme suite HKL2000 (ref. 49 ) and further analysed using the CCP4 package 50 unless specified otherwise. All the structures were solved by combining molecular replacement and MIRAS phasing. The protein model (PDB 4F35) was placed into the unit cell by using the programme PHASER 51 . Heavy metal-binding sites were identified by difference Fourier analysis and MIRAS phases were calculated using the programme SHARP 52 . The resulting electron density maps were improved by solvent flattening, histogram matching, non-crystallographic symmetry averaging and phase extension. Model building was carried out using the programme O (ref. 53). Structure refinement was conducted using the programme REFMAC with experimental phases as restraints 54 .
Proteoliposome reconstitution and transport assay. DDM-purified VcINDY variants were reconstituted into liposomes at 0.1 mg ml À 1 using Escherichia coli polar lipids and 1-palmitoyl-2-oleoylphosphatidylcholine at a 3:1 (w/w) ratio 13 ]-succinate. The reaction conditions were varied for S146A and T373A, which were specified in the figure captions. For all reactions, at 1-min time point, which lay well within the linear portion of the transport curve, a 60-ml sample was withdrawn and diluted in 0.5 ml of ice-cold quench buffer consisting of 100 mM Tris-HEPES, pH 7.4, 0-800 mM NaCl (same as that of the reaction buffer) and 1 mM EDTA. The quenched reaction was immediately filtrated through a nitrocellulose membrane (0.22 mm pore size, Millipore) and followed by three swift washes each with 1 ml of ice-cold quench buffer. Filters were dissolved in acetic acid and the radioactivity was measured by liquid scintillation. Transport assays were performed in duplicate and repeated more than three times.
Data availability. The coordinates of the models and structure factors have been deposited into the PDB data bank under the accession codes 5UL7, 5UL9, 5ULD and 5ULE. The PDB structure model 4F35 was used in this work. All the other data supporting the findings of this study are available within the article and its supplementary information files are available from the corresponding author upon reasonable request.