Diverse regulation of functional dimerization of a sugar transporter by different interfacial lipids

Abstract

Endogenous lipids play essential roles in modulating membrane protein structure and function, yet the molecular mechanisms governing lipid-specific regulation remain elusive. Here, we combine solid-state NMR spectroscopy and molecular dynamics simulations to elucidate how distinct lipids regulate the structure and activity of a membrane protein in a native-like membrane environment. Using VsSemiSWEET as a model system, we determine its high-resolution structure with bound lipids, identifying three lipid types: phosphatidylethanolamine (PE), phosphatidylglycerol (PG), and cardiolipin (CDL). These lipids bind at the monomer-monomer interface, stabilizing the dimeric structure of VsSemiSWEET. Notably, PG and CDL exhibit differential binding modes, with CDL demonstrating a dual interaction mechanism involving both its headgroup and acyl chains that enhances both dimer stability and functional activity. These findings reveal how lipids with different physicochemical properties differentially control membrane protein oligomerization and function, providing a mechanistic framework for lipid-specific regulation.

Introduction

Specific-binding lipids play a pivotal role in modulating the function, structure, stability, oligomeric state, and conformational dynamics of membrane proteins^1,2,3, such as ammonia channel (AmtB)⁴, osmotic stress-regulated betaine transporter (BetP)⁵, multidrug transporter LmrP⁶, prokaryotic SthK channel⁷, and Na⁺/H⁺ antiporter NhaA⁸. The high lipid compositional diversity of cellular membranes, combined with the complexity of interactions between bound lipids and membrane proteins, allows for differential regulation of these proteins by multiple bound lipid species^2,9,10. Distinct structural and chemical properties of lipids, such as variations in acyl chain length and saturation, as well as differences in headgroup chemistry, contribute to the diversity of lipid–protein specific interactions^11,12.

Growing evidence highlights the critical role of specific bound lipid species in modulating the function and structure of membrane proteins. For instance, the KirBac1.1 channel is activated by anionic lipids such as cardiolipin (CDL) and phosphatidylglycerol (PG), but is inactivated by phosphatidylinositol 4,5-bisphosphate (PIP₂)¹³. Additionally, the anionic lipid phosphatidic acid (PA) and the zwitterionic lipid phosphatidylethanolamine (PE) have been reported to selectively bind to the two-pore domain potassium channel K2P4.1, with each lipid featuring non-overlapping interaction sites and distinct regulatory mechanisms¹⁴. The structural and functional roles of diverse lipid binding were further underscored by recent structural studies of mechanosensitive potassium leak channel TREK1¹⁴, pentameric ligand-gated ion channels (pLGICs)^15,16, and Mycobacterium tuberculosis efflux pump EfpA¹⁷. These studies reveal the binding of multiple lipid species within the structures of these membrane proteins, demonstrating that the binding of multiple lipids is not merely incidental but often integral to the structural integrity and biological activity of membrane proteins.

Although diverse lipid–protein interactions have been extensively studied across membrane proteins, the role of bound lipids in regulating protein structure and function are still widely debated. Currently, it is challenging to experimentally investigate the specific interactions between lipids and membrane proteins using conventional structural biology methods such as X-ray crystallography, solution NMR spectroscopy, and mass spectrometry. This is primarily because these approaches typically rely on detergent-containing membrane mimetics, which disrupt native lipid–protein interactions in non-native membrane environments. In X-ray crystallography, lipids frequently fail to adopt ordered arrangements due to their high flexibility, rendering them undetectable in membrane protein crystals^6,18. Similarly, in mass spectrometry, the complexity and overlap of lipid fragmentation spectra impede the clear identification of specific lipids and their interactions¹⁹. As a result, the specific bound lipid species, their binding sites on proteins, and their functional relevance are often poorly defined for many membrane proteins.

A representative example is the bacterial sugar transporter Vibrio sp. SemiSWEET (VsSemiSWEET), a prototypical member of the major facilitator superfamily (MFS)²⁰. The SWEET family of transporters is responsible for facilitating sugar uptake across cellular membranes, enabling solute permeation down a concentration gradient. These proteins play essential roles in various biological processes, including nectar secretion, pollen nutrition, and susceptibility to pathogens²¹. Their bacterial homologs, known as SemiSWEETs, are widely distributed among prokaryotes and represent one of the smallest known transporter families. To date, the structures of four distinct SemiSWEETs have been determined by X-ray crystallography in the lipid cubic phase (LCP), revealing three conformational states: outward-open, occluded, and inward-open^{22,23,24,25,26}. All four structures exhibit the same dimeric architecture, suggesting that SemiSWEETs consistently function as dimers and employ an alternating-access mechanism for sugar transport.

However, it is important to note that none of these reported crystal structures showed the presence of bound lipids. In contrast, mass spectrometry analysis of VsSemiSWEET in detergent identified tightly bound lipid species, including CDL, PG, and PE. Among these, only CDL was found to regulate VsSemiSWEET dimerization^1,27. The presence of detergent in these experimental conditions may disrupt the native structure of membrane proteins and interfere with protein–lipid interactions. Additionally, the use of different membrane-mimetic environments, such as detergent micelles versus LCP, may further contribute to the discrepancies in the observed results. Consequently, it is still unclear whether VsSemiSWEET harbors tightly bound lipids in its native membrane environment, where they bind, and what specific functional roles they play. Further experimental investigation is required to elucidate the nature of lipid binding and its functional implications in VsSemiSWEET²⁸. In this context, solid-state NMR (ssNMR) has emerged as a powerful technique for determining membrane protein structures and characterizing their lipid interactions at atomic resolution within native-like lipid bilayers^29,30,31,32.

In this work, we explore the molecular basis for lipid modulation in VsSemiSWEET through an integrated approach incorporating NMR, additional biochemical experimental techniques, and molecular dynamics (MD) simulations. We identify three co-purifying lipid types: PE, PG, and CDL, among which CDL exhibits the highest binding affinity. Through integrated ssNMR and MD analyses, we determine the structure of lipid-bound VsSemiSWEET in native-like lipid bilayers and resolve the distinct binding sites and interaction modes for each lipid. Our results reveal that PG and CDL exert differential effects on dimer stability and function of VsSemiSWEET via their specific acyl tails and headgroups. These findings provide a mechanistic understanding of how lipids with distinct physicochemical properties differentially regulate membrane protein oligomerization and activity.

Results

Identification of bound lipids co-purified with VsSemiSWEET

SsNMR spectral analysis of VsSemiSWEET liposomes confirmed the presence of bound lipids that co-purified with the protein. The purified protein was successfully reconstituted into total E. coli lipid extracts, and resonances for 85 out of 97 residues were assigned using a series of multidimensional ssNMR spectra³³. To identify these co-purified bound lipids, we performed a ¹³C-¹³C combined ${{{\rm{R}}}2}_{{{\rm{mathrmn}}}}^{{{\rm{v}}}}$-driven (CORD) experiment³⁴ on uniformly-¹³C, ¹⁵N-labeled VsSemiSWEET. Since the co-purified lipids were ¹³C-labeled, they generated ¹³C-¹³C correlation signals in the CORD spectrum. In contrast, the E. coli total lipid extracts used for reconstitution were unlabeled and did not produce such correlations in the spectrum. As shown in Fig. 1a, several cross-peaks were observed in the ¹³C chemical shift region of 60−75 ppm, corresponding to characteristic chemical shifts of lipid headgroups and glycerol backbone. To rule out the possibility that these signals originated from the protein itself, we employed an unfolding-refolding approach. During this process, ¹³C-labeled bound lipids were removed during the unfolding procedure, while unlabeled lipids were reintroduced to the protein during the refolding and reconstitution process. Following this treatment, the signals in this region were either absent or significantly weakened, confirming their origin from ¹³C-labeled bound lipids. Furthermore, Ser and Thr residues are the only amino acids exhibiting intra-residual correlations in this region. The signals of these residues have been unambiguously assigned, confirming that the identified signals arise from co-purified lipids rather than protein-derived signals.

Fig. 1: Identification of co-purified bound lipids in VsSemiSWEET.

a SsNMR 2D CORD ¹³C-¹³C spectra (mixing time of 50 ms) of VsSemiSWEET reconstituted into E. coli membrane extracts, comparing samples subjected to unfolding-refolding treatment (pink) with native preparations (black). The unfolding-refolding treatment replaces tightly bound ¹³C-labeled lipids with natural abundance lipids, resulting in the disappearance of lipid-specific signals (blue dotted box). All assigned Ser/Thr backbone resonances are labeled. b Molecular structure formulas of PE, PG, and CDL lipids. To validate the bound lipid signals, ssNMR 2D ¹³C-¹³C CORD spectrum of ¹³C-labeled VsSemiSWEET with bound lipids (black) were superimposed to that of pure E. coli lipids without the proteins (pink). Homology model illustrating VsSemiSWEET bound with PE, PG, and CDL lipids in lipid bilayers. c Solution NMR analysis of lipid binding in VsSemiSWEET. Left: Superimposed 1D ³¹P NMR spectra of lipids bound to VsSemiSWEET before and after delipidation. Detergent treatment selectively removes loosely bound lipids while preserving tightly bound species. Right: Homology models illustrating VsSemiSWEET in its native lipids-bound state (top) and after delipidation (bottom).

To identify the specific species of the bound lipids, we first measured the ¹³C chemical shifts corresponding to the headgroups and glycerol backbone of anionic lipid PG, zwitterionic lipids PE, and CDL using ¹H-¹³C HSQC (heteronuclear single quantum coherence) solution NMR spectra of pure lipid standards (Fig. S1 and Table S1)³⁵. We then acquired a 2D ¹³C-¹³C CORD spectrum of ¹³C, ¹⁵N-labeled VsSemiSWEET with a 500 ms mixing time, optimized to enhance correlation signal intensities from bound lipids (Fig. S2). Although minor chemical shift differences existed between the free and protein-bound states of the lipids, the characteristic α–β correlation patterns allowed us to assign the glycerol backbone and headgroup signals for PE, PG, and CDL^36,37,38. Specifically, well-resolved and symmetric correlations at 64.8–42.9, 73.4–69.2, and 72.0–69.0 ppm were unambiguously assigned to the headgroups of co-purified PE, PG and CDL, respectively (Fig. 1b). To further verify these assignments, we prepared a ¹³C-labeled total E. coli lipid extract sample containing no protein. This sample consisted of approximately 57% PE, 15% PG, 10% CDL, and 18% other minor species identified by mass spectroscopy³⁹. Notably, this lipid profile is analogous to the native lipid composition of Vibrio species⁴⁰. As illustrated in Fig. 1b, the ¹³C chemical shifts of PE, PG and CDL in the E. coli lipid extract align perfectly with those observed for the lipids bound to VsSemiSWEET. This exact signal match provides direct evidence that these three lipid species co-purify with the protein, demonstrating their stable and high-affinity interaction with VsSemiSWEET.

To further identify the bound lipids, we employed 1D ³¹P solution NMR, a technique that yields distinct chemical shifts for lipid phosphate groups in each lipid type^41,42,43. In our spectra, PE, CDL, and PG were identified by their characteristic shifts at −0.20 ppm, 0.22 ppm, and 0.30 ppm, respectively (Fig. S3). As shown in Fig. 1c, the ³¹P NMR spectra of VsSemiSWEET in detergent micelles verified the co‑purification of PE, PG, and CDL. Notably, after a wash with 1% dodecyl-β-D-maltoside (DDM), the ³¹P signals of PE and PG decreased by 80% approximately, while the CDL signal showed little reduction, demonstrating that CDL binds more tightly than PE or PG. We extended this analysis to SemiSWEET proteins from E. coli, L. biflexa, and Bradyrhizobium species. Although all three lipids co‑purified with these homologs, the binding strength of CDL was consistently lower than that observed for VsSemiSWEET (Fig. S4). This trend agrees with our MD simulations, which also identified CDL binding in these variants but with comparatively weaker interactions (Fig. S5).

Structural characterization of VsSemiSWEET in lipid bilayers using ssNMR

Next, we proceeded to identify the specific lipid-binding sites on VsSemiSWEET. Although the X-ray crystal structure of VsSemiSWEET is available²⁵, we determined a high‑resolution 3D structure of the protein in native‑like lipid bilayers to avoid potential artifacts introduced by the non‑native detergents used in X-ray crystallography. To obtain the necessary distance restraints for structure determination, we conducted 2D ¹³C-¹³C CORD experiments on sparsely ¹³C-labeled samples. During bacterial protein expression, [2-¹³C] or [1,3-¹³C] glycerol was used as the carbon source (denoted as 2-VsSemiSWEET and 1,3-VsSemiSWEET), resulting in significant ¹³C spin dilution⁴⁴. This dilution greatly improved spectral resolution and suppressed dipolar truncation effect, thereby providing numerous well‑resolved cross‑peaks corresponding to medium‑ and long‑range distance restraints^45,46,47. The “medium-range” and “long-range” correspond to distances between residues where i and j are residue numbers. For medium-range, the distance is defined as 2 ≤ |i–j | <5, while for long-range, it is |i–j|≥5. Both the 2-VsSemiSWEET and 1,3-VsSemiSWEET samples were prepared using an unfolding-refolding approach to replace the original ¹³C-labeled co-purified lipids with unlabeled ones, thereby eliminating potential lipid‑derived signal interference during restraint assignment. Comparison of the chemical shifts and signal intensities between the refolded and native VsSemiSWEET samples confirmed that the unfolding‑refolding process did not perturb the protein structure (Figs. S6 and S7).

As shown in Fig. 2a, b and Fig. S8, the CORD spectra with a 500 ms mixing time exhibit hundreds of well-resolved cross-peaks. However, significant spectral overlap complicates their unambiguous assignment. To address this, we employed a structural template to guide distance restraint determination. We constructed an experimental structural model using CS-Rosetta calculations, inputting chemical shift data and initial distance restraints determined solely based on residue-specific chemical shifts (Table S2). The resulting monomeric structure of VsSemiSWEET exhibits high convergence, with a root mean square deviation (RMSD) of 0.9 ± 0.2 Å for backbone heavy atoms among the ten lowest-energy structures, and is consistent with the AlphaFold3-predicted structure (Fig. S9)⁴⁸. To illustrate how this structural model can resolve assignment ambiguities, we provide an example (Fig. S10). The cross-peak at (70.0, 61.4) ppm in the CORD spectrum of 2-VsSemiSWEET has eight candidate assignments; however, seven possibilities are excluded because their corresponding ¹³C-¹³C distances exceed the upper limit (9 Å) for a 500 ms mixing time, leaving only the 4.3 Å correlation between T60 Cα and W72 Cα. Further assignment details regarding the assignment of distance restraints can be found in the Methods Section. In total, we successfully assigned 442 ¹³C-¹³C intra-monomer distance restraints, including 117 long- and medium-range restraints (Fig. 2a, b and Fig. S11). Structural calculations performed using Xplor-NIH software with these comprehensive intra-monomer restraints yielded an ensemble of ten structures (selected from 200) with a backbone RMSD of 1.5 Å across all α-helices of the protein (Fig. S10c).

Fig. 2: Three-dimensional structure of VsSemiSWEET in lipid bilayers determined by ssNMR.

a, b Selected regions of the 2D ¹³C-¹³C CORD spectrum of 2-VsSemiSWEET with a mixing time of 500 ms, illustrating the assignments of the long-range intra-monomer correlations. Cross peaks corresponding to long-range restraints (|i - j|> 5, where i and j represent correlated residues) are labeled in red. c Selected regions of the 2D ¹³C-¹³C CORD spectrum of 1,3-VsSemiSWEET and the 2D NcaCX spectrum of uniformly-¹³C, ¹⁵N labeled VsSemiSWEET, showing the assignments of the long-range inter-monomer correlations. d Superposition of 2D ¹³C-¹³C CORD spectra of 2-¹³C VsSemiSWEET and spin diluted (2-¹³C:NA = 1:2) VsSemiSWEET samples. Assigned long-range inter-monomer correlations are labeled. e Superposition of the ten dimer structures of VsSemiSWEET with the lowest energy in lipid bilayers, with RMSD of 0.9 ± 0.3 Å of the backbone atoms. The N-helix (residues 1–17) structure remains unconverged due to the lack of assigned chemical shifts in residues from 1 to 5 and 12 to 17, which are not shown.

VsSemiSWEET exists as a dimer in lipid bilayers, as demonstrated by our cross-linking experiments (Fig. S12). Assignment of inter-monomer distance restraints, which are critical for determining the dimer structure, is challenging due to potential overlap with intra‑monomeric contacts. To address this, we initially assigned inter-monomer distance restraints by excluding intra-monomer contacts based on the monomer structure. All possible intra-monomer distances exceeded the upper limit of 9.0 Å, allowing us to exclude them during distance restraint assignment (Fig. 2c and S13a, b). To further distinguish between intra- and inter-monomer contacts, we prepared a spin-diluted sample by mixing ¹³C-labeled monomers with natural abundance (NA) analogs at a 1:2 ratio using an unfolding-refolding approach (Fig. S14a). In this setup, intra-monomer signals are reduced to one-third of their original intensity in uniformly labeled samples, while inter-monomer signals were absent (Fig. S14b). By comparing the signal intensities in 2D ¹³C-¹³C spectra of 2-VsSemiSWEET and the spin-diluted [(1:2)2-¹³C/NA] VsSemiSWEET samples, we unequivocally identified cross‑peaks originating from inter‑molecular contacts (Fig. 2d).

The analysis of CORD spectra from a diluted sample revealed six ambiguous inter-monomer distance constraints. These constraints were used to construct an initial dimer structural model using the docking module of Xplor-NIH. Further details on the homodimer construction are provided in the Methods Section. Utilizing this initial model, thirteen unambiguous and eight additional ambiguous inter-monomer contacts were identified. The strategy for resolving intra- versus inter-monomer contact ambiguity is illustrated in Figs. S13c and S15, with examples of assignments for Q29CG-T48CA and K32CA-G43CA restraints. All inter-monomer long-range distance restraints are listed in Table S3. A total of 21 inter-monomer and 442 intra-monomer distance restraints, along with 78 backbone torsion angles predicted by the TALOS+ program using assigned chemical shifts (Table S4), were employed to calculate 1024 dimer structures of VsSemiSWEET using Xplor-NIH. For further details on the structure calculations, refer to the experimental section. Among the calculated structures, the ten lowest-energy structures exhibited high convergence, with a backbone RMSD of 0.9 ± 0.3 Å (Fig. 2e).

Hydrogen/deuterium (H/D) exchange experiments revealed a solvent-accessible cavity between the two monomers near the periplasmic region, supporting an outward‑open conformation (Fig. S16). The overall ssNMR structure is similar to the crystal structure (Fig. S17), with a backbone RMSD of 2.0 Å across well-ordered regions (residues 17–39, 46–67, and 70–92). Notably, a key structural feature in the ssNMR structure is the pronounced kink at Pro29 within the conserved PQ‑loop motif of TM1, which divides the helix into TM1a and TM1b segments. Importantly, while bound lipids remained physically present in the VsSemiSWEET structural determination samples, their structural constraints were removed by substituting ¹³C-labeled lipids with natural abundance lipids during the unfolding-refolding process. Consequently, the bound lipids are absent in the determined structures, leaving empty coordinates due to the lack of distance restraints.

Characterization of lipid interactions with VsSemiSWEET in native-like lipid bilayers

To investigate the interactions between VsSemiSWEET and its bound lipids, we performed 2D CORD spectra of uniformly ¹³C, ¹⁵N-labeled VsSemiSWEET using a long mixing time of 500 ms. The chemical shift at 72.8 ppm was uniquely assigned to the G2 nuclei of the central glycerol moiety in lipids. Therefore, any cross-peaks at this chemical shift in the 2D ¹³C-¹³C spectrum of lipid-bound VsSemiSWEET must represent either lipid–lipid or lipid–protein correlations. A ¹³C-¹³C spectrum of uniformly ¹³C-labeled E.coli lipid extracts, which contains no protein signals, serves as a reference to distinguish between lipid–lipid and lipid–protein interactions. As shown in Fig. 3a, additional peaks observed at 72.8 ppm in the VsSemiSWEET spectra correspond specifically to lipid–protein correlations. Based on the correlated chemical shifts of these peaks (Fig. S18), we identified residues S22, Q29, K32, H37, E39, H40, C76, and A84 as potential interaction sites for lipid G2 nuclei, as both the main-chain and side-chain ¹³C atoms of these residues exhibit correlation with the G2 nuclei of lipids. Critically, the observation of these lipid–protein cross-peaks in the dipolar-coupling-based experiments imply lipid binding times in the microsecond range or longer but exclude shorter lifetimes, which is the timescale of the relevant dipolar couplings that drive spin diffusion³⁷.

Fig. 3: Analysis of interactions between bound lipids and VsSemiSWEET.

a Potential lipid-binding sites identified by ssNMR. Overlay of 2D ¹³C-¹³C CORD spectra for VsSemiSWEET bound with lipids (black) and pure ¹³C-labeled E. coli lipid extracts (pink). The VsSemiSWEET spectrum exhibits additional peaks at 72.8 ppm (assigned to lipid G2), indicating lipid–protein interactions. Possible assignments of these additional peaks are labeled based on the residue-specific backbone and side-chain chemical shifts. b MD simulation-derived lipid-binding sites. Residence time histograms of CDL at distinct binding sites. The ssNMR structure of VsSemiSWEET in lipid bilayers is shown as a cartoon, with lipid-binding sites rendered as space-filling models. The error was calculated as the standard deviation from three replicate simulations. Source data are provided as a Source Data file. c Validation of lipid-binding sites by solution NMR. Superimposed 1D ³¹P spectra of wild-type VsSemiSWEET and mutants (K32A, H37A/H40A, and K32A/H37A/H40A). The arrow denotes reduced ³¹P signal intensity in mutants, implicating key residues in lipid binding. d Solution NMR analysis of protein–lipid interaction type. Superimposed 1D ³¹P spectra of wild-type VsSemiSWEET and K32R mutant. e Molecular docking of CDL with VsSemiSWEET, illustrating interaction interfaces.

To assign these lipid–protein correlated peaks, we carried out MD simulations to identify potential lipid-binding sites on VsSemiSWEET. We conducted 5-μs all-atom MD simulations using the ssNMR structure of VsSemiSWEET in lipid bilayers as the starting structure. The average residence time of different lipids at each residue of the SemiSWEETs was calculated to evaluate their binding ability (Fig. 3b and Fig. S19). The results indicated that CDLs exhibited stronger binding ability to VsSemiSWEET compared to PEs, which is consistent with our experimental observations (Fig. 1c). Analysis of the MD simulations highlighted that residues interacting with CDL were predominantly located at the monomer–monomer interface and near transmembrane regions. Notably, basic residues K32, H37, and H40 in the intracellular region demonstrated significantly stronger interactions with CDL compared to other residues (Fig. 3b).

To validate these findings, we mutated the three basic residues K32, H37, and H40 to basic-to-neutral substitutions (K32A, H37A/H40A, and K32A/H37A/H40A). We then assessed the effects of these mutations on lipid binding using 1D ³¹P solution NMR spectra of the mutated VsSemiSWEET solubilized in micelles. As shown in Fig. 3c, the signal intensities for PE, PG, and CDL lipids all decreased in these mutants, indicating that these lipids bind at the interface between the two monomers. However, while signals from PE and PG persisted, nearly no signals were observed for CDL in the K32A/H37A/H40A mutant. This suggests that CDL exclusively binds at the monomer–monomer interface, whereas PE and PG bind both at the interface and the vicinity of transmembrane regions. These findings align with a recent MD simulation study on VsSemiSWEET, which demonstrated that CDL molecules bind in sufficient numbers to the dimer interface on the cytoplasmic side²⁸.

To further elucidate the nature of the interaction between bound lipids and these basic residues, we constructed a K32R mutant, which involved a basic-to-basic substitution (Fig. 3d). As expected, the lipid signals observed in the 1D ³¹P spectrum of the K32R mutant were nearly identical to those of the wild-type VsSemiSWEET, indicating that this basic-to-basic mutation does not significantly disrupt the interaction between the residues and lipids. This finding supports the conclusion that their interactions are predominantly mediated by electrostatic forces. Moreover, in contrast to the dipolar-coupling-based signals from glycerol backbones and headgroups of the bound lipids, which indicate that these groups exhibit structural rigidity on a micro- to millisecond timescale, the signals observed in scalar-coupling-based ssNMR spectra suggest a high degree of flexibility on submicrosecond timescales⁴⁹. As shown in Fig. S20, the tail carbons of lipids in VsSemiSWEET were detected in a 2D ¹H-¹³C INEPT spectrum, indicating the flexibility of the lipid acyl chains. Finally, to generate a composite structural model that incorporates the missing lipid coordinates, we performed molecular docking. This placed two CDL molecules at the dimer interface (Fig. 3e). The docking model revealed that the lipid headgroups establish contacts with basic residues K32, H37, and H40, forming several hydrogen bonds. These findings underscore that lipid binding is predominantly mediated through electrostatic interactions between lipid headgroups and basic residues within the dimer interface.

Bound lipids influence the dimer stability of VsSemiSWEET in lipid bilayers

Dimerization is crucial for enabling VsSemiSWEET to efficiently transport sugars across cellular membranes²¹. To examine how lipid binding influences dimer stability, we conducted ssNMR, biochemical experiments, and MD simulations. As shown in Fig. S20, VsSemiSWEET maintains its dimeric state in the E. coli cellular membrane, in proteoliposomes, and when affinity-purified in detergent micelles (0.3% DM), as demonstrated by SDS-PAGE analysis (SDS-resistant). However, mutating the key interfacial basic residues K32, H37, and H40 to alanine (single or combined mutants) caused a predominant monomeric migration on SDS-PAGE gels (Fig. 4a), whereas the charge-preserving K32R mutant retained its dimeric form. Furthermore, size-exclusion chromatography of these mutants exhibited a monodisperse peak similar to that of wild-type VsSemiSWEET (Fig. 4b), indicating that these alanine mutants still elute as dimers in solution. The SDS-induced dissociation during SDS-PAGE analysis reflects a lipid-dependent stabilization of the dimer interface. Consistent with this finding, thermal stability assays revealed that lipid binding enhances the structural stability of VsSemiSWEET (Fig. S22). We also acquired ssNMR spectra of the H37A/H40A and K32A/H37A/H40A mutants to examine the conformations of the protein in the absence of lipid binding. As shown in Fig. 4c, d, the signals in the 2D ¹⁵N-¹³Cα NCA spectrum of these mutants exhibited broader linewidths than those of the wild-type, indicating conformational heterogeneity without lipids binding. Moreover, the disappearance of signals from the mutated sites and their neighboring residues, along with the loss of signals from many other residues, suggests increased conformational dynamics and flexibility in the absence of bound lipids. To further evaluate the structural consequences of disrupting lipid interactions, we carried out enhanced sampling simulations to investigate the stability of the H37A/H40A mutant. The resulting free energy landscape (Fig. S23) shows that the mutant dimer is substantially destabilized: it no longer adopts the lowest-energy conformation characteristic of the wild-type protein, and no stable inter-residue interactions persist at the dimer interface. Collectively, these results highlight the importance of lipid binding in sustaining the stability and structural integrity of the VsSemiSWEET dimer, likely through specific interactions at the monomer–monomer interface.

Fig. 4: Lipid binding modulates the oligomeric state and conformation of VsSemiSWEET.

a Oligomeric stability analysis: SDS-PAGE results for wild-type VsSemiSWEET, the K32R and lipid-binding-deficient mutants (K32A, H40A, and H37A/H40A) mutants, reconstituted in proteoliposomes. The gels shown are representative of three independent experiments. Source data are provided as a Source Data file. b Oligomeric state analysis: Size-exclusion chromatography profiles comparing wild-type VsSemiSWEET with its variants. Source data are provided as a Source Data file. c, d Conformational analysis: Superposition of 2D ¹⁵N-¹³Cα spectra of wild-type VsSemiSWEET and the H37A/H40A and K32A/H37A/H40A mutants. e MD simulation (lipid-free state) versus ssNMR structure: representative MD conformation of the dimeric VsSemiSWEET without bound lipids, overlaid with the ssNMR-derived structure (lipid-bound). f MD simulation (CDL-bound state) versus ssNMR structure: representative MD conformation of the dimeric VsSemiSWEET with bound CDL, overlaid with the ssNMR-derived structure (lipid-bound). g Schematic representation of the effect of lipid-binding-site mutations on the oligomeric stability and conformational state of VsSemiSWEET.

We further employed MD simulations with enhanced sampling to delineate the mechanism by which lipids modulate dimerization. We characterized the free energy landscapes of CDL-bound and lipid-free VsSemiSWEET (Fig. S24). Representative conformations of the lowest free energy state for both systems are shown in Fig. 4e, f, respectively. Compared to the ssNMR structure, the representative lowest-energy conformation of the lipid-free dimer showed fewer interfacial contacts and only rare, weak interprotomer interactions (Fig. S25). In contrast, the CDL-bound dimer adopted a conformation closely matching the ssNMR structure, with stable interactions across the binding interface (Fig. 4g). These results demonstrate that CDL binding enables the stabilization of a robust dimeric interface for VsSemiSWEET.

Differential regulations of interfacial CDL and PG on the structure and function of VsSemiSWEET

Since three distinct lipid types (CDL, PE, and PG) bind to the dimer interface of VsSemiSWEET, we used an unfolding-refolding approach to selectively deplete tightly bound CDL, allowing us to isolate the structural and functional effects of different lipid compositions. As shown in Fig. S26, fully denatured VsSemiSWEET can refold into dimers in the presence of either PE/PG or PE/PG/CDL mixtures. This observation indicates that both PE/PG and PE/PG/CDL lipid combinations support dimer formation, suggesting functional similarity between these lipid types in stabilizing the protein structure. However, MD simulations revealed fundamentally different binding modes for PG and CDL (Fig. 5a). Compared to PG, CDL forms more stable interactions with the protein, particularly through hydrogen bonding (Fig. S27). For example, the hydrogen bond occupancy between CDL and the H40 residue reaches 64%, whereas the maximum hydrogen bond occupancy between PG and H40 is only 23%. This disparity stems from CDL’s structure: its headgroup bears two negatively charged phosphate groups, creating a strong electrostatic surface that interacts tightly with adjacent basic residues (K32, H37, and H40) on the protein. This specific interaction underlies CDL’s high-affinity binding and its critical role in structural stabilization.

Fig. 5: Regulatory role of PG and CDL binding in VsSemiSWEET, revealed by MD simulations (a, b) and ssNMR experiments (d–g).

a Distinct lipid-binding modes of PG and CDL to VsSemiSWEET analyzed by MD simulations. Left: Residues interacting with lipid headgroups (PG/CDL phosphate groups: orange phosphorus, red oxygen). Right: Residues contacting lipid tails (highlighted in green). Regions with cavity-localized interactions are marked by yellow shading. b Contact probability differences of residues with lipids between PG-bound and CDL-bound VsSemiSWEET studied by MD simulations. Source data are provided as a Source Data file. c Schematic comparison of PG and CDL binding mechanisms. The acyl tails of PG penetrate the central interfacial cavity, disrupting dimer stability. In contrast, the acyl tails of CDL align along the outer dimer surface, reinforcing the dimeric interface. d SsNMR validation of lipid-binding modes. Superimposition of 2D ¹⁵N-¹³Cα and ¹³C-¹³C spectra of VsSemiSWEET in samples I (CDL-dominant modulation) and II (PG-dominant modulation). Assigned isolated resonances are labeled, and representative pronounced shifts in chemical shifts are indicated by blue arrows. e Quantitative analysis of the CSPs of residues induced by different lipid binding. Plot of the ¹⁵N and ¹³C CSPs between VsSemiSWEET in samples I and II as a function of residue number. Horizontal lines at 0.3 ppm indicate the threshold for significant CSPs. Residues exhibiting both significant CSPs and MD-predicted lipid contact probability differences (PG- versus CDL-bound states) are marked in red. f Superimposition of 2D ¹⁵N-¹³Cα and ¹³C-¹³C spectra of VsSemiSWEET in samples I and II, with peaks that show significantly reduced signal intensity marked with orange dotted boxes. Source data are provided as a Source Data file. g Quantitative analysis of changes in signal intensity of residues induced by different lipid binding. Relative intensities of signals resolved in ¹⁵N-¹³Cα and ¹³C-¹³C spectra of VsSemiSWEET in sample II (pink and blue bars), normalized by the corresponding signal intensities in sample I (gray bars). Dotted lines represent relative signal intensities of 0.6. Residues with relative intensities lower than 60% are colored blue, and signals that became unobservable in sample II are marked with blue asterisks. The secondary structures of VsSemiSWEET in lipid bilayers are illustrated at the top of the figure. Source data are provided as a Source Data file.

Importantly, MD simulations revealed that the tail chains of PG preferentially penetrate the cavity between the two monomers while their headgroups remain spatially confined at the dimer interface of VsSemiSWEET (Fig. 5a, c). By engaging specific residues within this cavity, the PG tails effectively disrupt inter-monomer packing. In contrast, CDL exhibits a distinct binding mode; its bulky hydrophobic tail cannot accommodate the cavity due to steric constraints (Fig. S28). Instead, CDL tails localize to the outward surface of VsSemiSWEET, promoting a tightly packed dimeric interface. The interaction strength between VsSemiSWEET residues and CDL or PG lipids at the interface was quantitatively evaluated based on contact probability differences. As shown in Fig. 5b, PG lipids predominantly interact with residues located on transmembrane helix 1 (TM1), the N-terminal region of TM2, and the C-terminal region of TM3. Notably, key residues (S22, T26, Q29, L33, S36, T48, Y52, and Y82) exhibited higher lipid contact probabilities in the PG-bound state. These residues are predominantly positioned along the inner faces of the transmembrane helices, further supporting the deep penetration of PG lipid tails into the VsSemiSWEET inner cavity. Additionally, we computed the in-plane density of lipid tail carbon atoms in the membrane. As illustrated in Fig. S29, lipid molecules in the PG-bound VsSemiSWEET system penetrate into the dimer interior and occlude the central pore of the protein.

To experimentally validate the distinct binding modes of PG and CDL, we prepared two ssNMR samples: sample I (with a combination of PE, PG, and CDL lipid binding) and sample II (with a combination of PE and PG lipid binding). The preparation procedures for both samples are shown in Fig. S30. Several critical aspects should be noted: First, the lipid bilayers of both samples are composed of PE and PG lipids to exclude the influence of other environmental lipids on the structure of VsSemiSWEET. Second, binding affinities of different lipids to VsSemiSWEET, calculated by MM/GBSA⁵⁰, indicated that the binding ability of CDL to VsSemiSWEET is ~215 times that of PG and about 3600 times that of PE (Fig. S31). These predicted results align with observations from the above delipidation experiments (Fig. 3e). Third, during the refolding, excess PE/PG or PE/PG/CDL was added to fully refold and reassemble the denatured VsSemiSWEET into a dimeric form, favoring binding of lipids with higher affinities. Together, our results indicate that CDL primarily modulates the structure of VsSemiSWEET in Sample I, while PG exhibits a more pronounced regulatory effect in Sample II.

To probe the distinct structural effects of CDL versus PG binding, we acquired ssNMR spectra for Sample I (PE/PG/CDL) and Sample II (PE/PG only). Chemical shift perturbations (CSPs) greater than 0.3 ppm were considered significant, given the high spectral resolution and reproducibility of the ssNMR spectra (Fig. S6). Figure S32 presents the superimposition of the 2D ¹⁵N-¹³Cα and ¹³C-¹³C spectra of VsSemiSWEET in samples I and II. While most peaks showed negligible CSPs, some peaks exhibited significant CSPs (Fig. 5d). As shown in Fig. 5e, multiple transmembrane residues of VsSemiSWEET exhibited significant CSPs in sample II compared to sample I, including S22, T26, Q29, L33, S36, T48, Y52, Y82, V86, and I89. These CSP patterns show strong correlation with our MD simulation results, where the same residues demonstrated the highest differences in lipid contact probability between the PG-bound and CDL-bound states (Fig. 5b). This consistency between experimental CSP data and MD simulation predictions suggests these residues participate in distinct lipid–protein interactions depending on the bound lipid species. The structural differences between CDL and PG likely contribute to their distinct interaction patterns. The larger size of CDL prevents its hydrophobic tails from drilling into the cavity of VsSemiSWEET. In contrast, PG has a smaller steric hindrance, allowing its tails to penetrate deeply into the protein’s cavity. This deeper penetration facilitates direct interactions between PG and inner cavity residues, resulting in significant CSPs in these transmembrane residues. Additionally, Fig. 5f and the bar diagram in Fig. 5g showed that the signal intensities of many residues decreased and, in some cases, became unobservable in the dipolar-coupling-based spectra of sample II. This reduction can be attributed to the enhanced mobility and reduced structural stability of these residues in the absence of CDL binding.

To link the influence of dimer formation of different lipid binding to the functional implications, we analyzed the detailed intermolecular interactions and free energy landscapes of VsSemiSWEET binding with various lipids (Fig. 6a). All-atom MD simulations revealed that while PG binding, unlike CDL, does not hinder VsSemiSWEET monomer dimerization but results in more disordered residue arrangements at the dimeric interface and weaker inter-monomeric interactions compared to CDL-bound VsSemiSWEET (Fig. 6b). Moreover, the RMSD distribution further confirms that the CDL-bound system stabilizes a dimer conformation closely matching the ssNMR-resolved structure of VsSemiSWEET, whereas the PG-bound dimer is more heterogeneous (Fig. S33). These findings align with ssNMR data (Fig. 5f, g), which indicate greater dynamics and heterogeneity in the PG-bound dimer. Furthermore, both MD simulation and experimental findings demonstrate that PE exerts a regulatory effect similar to PG, promoting dimerization but also inducing conformational heterogeneity and instability in the resulting dimers (Fig. S34).

Fig. 6: Functional regulation of VsSemiSWEET by different lipid binding.

a Molecular structures of the lipids POPG and CDL. b Representative conformations of PG-bound dimeric VsSemiSWEET, illustrating intermolecular residue contacts compared to CDL-bound VsSemiSWEET. c Free energy landscape (FEL) of CDL-bound VsSemiSWEET as a function of the open sizes of the outward gate (d_out) and inward gate (d_in). Here, d_out corresponds to the distances between the Cα atoms of residues Y62 on the two protomers, while d_in corresponds to the center-of-mass (COM) distances between the Cα atoms of residues L42–L44 on the two protomers. Representative structures of the states observed on the FEL are shown. d FEL of PG-bound VsSemiSWEET as a function of the opening size of outward gate and inward gate. e Representative conformations of the outward-open and inward-open states. The helices of the outward-open structure are colored dark gray (TM1), red (TM2), and dark green (TM3). The helices of the inward-open structure are colored light gray (TM1), light orange (TM2), and light green (TM3). f illustration showing that the tails of CDL primarily interact with TM2 of VsSemiSWEET. g Contact probability of bound CDL lipids with the residues of VsSemiSWEET. The black dashed box indicates the region with a high density of contact residues. Source data are provided as a Source Data file.

The regulatory effects of different types of lipids on the function of VsSemiSWEET was characterized by MD simulations. As proposed by Feng et al., sugar transport by SemiSWEETs follows a “free ride” mechanism²³. This mechanism involves spontaneous transitions between outward-open, occluded, and inward-open configurations, facilitating sugar passage from extracellular to intracellular environments. Our analysis of the free energy landscape identified key functional states in CDL-bound VsSemiSWEET: outward-open (states 1 and 2), occluded (states 3 and 4), and inward-open (state 5) structures (Fig. 6c). The outward‑open states correspond to the lowest free‑energy minima and are thus the most populated, whereas the inward‑open state lies at a significantly higher energy and is scarcely populated. This is consistent with our ssNMR findings, where inward-open structures were undetected. These results indicate that CDL binding enables VsSemiSWEET to transition between essential functional states, enhancing its activity via the “free ride” mechanism. In stark contrast, the free energy landscape of PG‑bound VsSemiSWEET shows no such conformational transitions (Fig. 6d). The lowest free energy minimum for PG-bound VsSemiSWEET corresponds to a nonfunctional state with an intramembrane cavity filled by PG tails. This suggests that PG creates steric hindrances that not only obstruct ligand passage but also impede the transition cycle between functional states.

The gating mechanism of VsSemiSWEET differs significantly from those observed in its homolog, L. biflexa SemiSWEET (LbSemiSWEET)^23,25. In LbSemiSWEET, Feng et al. proposed that TM1 undergoes kinking at the central proline residue, disrupting interactions with TM3 and facilitating conformational transitions between outward-open, occluded, and inward-open states. In VsSemiSWEET, however, TM2 serves as the primary driver of structural transitions between closed and open states at both extracellular and intracellular gates (Fig. 6e). The rocking motion of TM2 in opposite directions across the two protomers of the VsSemiSWEET dimer controls the opening of both the outward and inward gates (Fig. S35). This motion is closely linked to lipid interactions, as shown by our analysis of residue–lipid interactions in all-atom MD simulations. The results reveal extensive interactions between TM2 and CDL (Fig. 6f, g), where the hydrophobic tails of CDL interact with the outer surface of TM2, modulating helical orientational fluctuations. Experimental evidence supports this mechanism: residue L46 in the TM2 region, which exhibits the highest probability of contact with CDL, shows significant CSPs in sample II (lacking CDL) compared to sample I (Fig. 5g). Together, these findings underscore a critical and specific role for CDL in enabling the functional transition cycle of VsSemiSWEET.

Discussion

In this study, we have identified the bound lipids PE, PG, and CDL in VsSemiSWEET by characteristic ¹³C chemical shifts from the glycerol backbone and headgroup in ssNMR spectra, as well as ³¹P chemical shifts of phosphate groups in solution NMR spectra. These identifications were further validated through delipidation experiments. The binding of PE, PG and CDL to VsSemiSWEET aligns with the findings from previous studies that used native mass spectrometry^1,27. In contrast, no bound lipids were identified in the crystal structure of VsSemiSWEET²⁵, likely due to the high mobility of these lipids. Moreover, the choice of detergent used for solubilization can significantly affect lipid binding. Recent studies suggest that different detergent species can impede the cooperative binding of lipids to membrane proteins^17,51,52,53. It remains unclear whether the samples in DDM or LCP detergent for the crystal characterization of VsSemiSWEET contain co-purified lipids. While we have demonstrated the binding modes of individual lipids, such as PG and CDL, to VsSemiSWEET, the actual binding interactions are likely more complex. Based on both ssNMR and solution NMR experimental results, it is plausible that the binding involves a combination of two or more different types of lipids at the interface of the monomers. This complexity warrants further investigation to fully understand the lipid interactions with VsSemiSWEET.

Lipid–protein specific interactions play a critical role in regulating the sugar transport activity of SemiSWEETs, which function as transporters by assembling into dimers and cycling through outward-open, occluded, and inward-open conformations⁵⁴. CDL, a phospholipid characterized by its dimeric structure with four acyl chains and two phosphate groups, has been shown to play a pivotal role in modulating the oligomerization and conformational dynamics of VsSemiSWEET. The headgroups of CDL molecules bind at the monomer–monomer interface, coordinating the dimerization process of VsSemiSWEET protomers to enable the formation of dimers with a proper and stable conformation that facilitates sugar transport. Furthermore, the four acyl chains of the CDL influence its function in two distinct ways: first, their large size prevents them from entering the transmembrane cavity, thereby avoiding interference with intermolecular interactions; second, the hydrophobic nature of these chains facilitates interaction with transmembrane helix 2, enhancing the transport turnover of VsSemiSWEET. Importantly, various studies have indicated that the interaction between proteins and CDL may depend on the specific way on properties of CDL acyl chains⁵⁵. For example, α-synuclein binding to CDL is significantly influenced by the nature of CDL’s acyl side chains: complete saturation drastically reduces affinity, whereas the removal of a double bond increases it⁵⁶. These findings underscore that CDL acts as a regulatory factor in membrane proteins not only through the stabilizing interactions provided by its headgroups but also due to the structural characteristics of its four acyl chains. Collectively, this highlights the critical importance of CDL–protein interactions in modulating SemiSWEET-mediated sugar transport activity.

Dimerization is a prerequisite for VsSemiSWEET to achieve transmembrane sugar transport, as it forms the functional unit required for substrate translocation²³. The two protomers create a central pathway that enables selective sugar movement across the membrane, with the transport mechanism relying on coordinated conformational changes between monomers, generating an alternating-access mechanism. Disruption of dimerization impairs this conformational switching, leading to a loss of transport activity, underscoring the essential role of dimerization in the function of SemiSWEETs. Notably, the dimer interface is stabilized by interactions with lipids such as CDL, which is critical for maintaining the dimer conformation of VsSemiSWEET. Extensive studies have demonstrated that lipid binding plays a universal regulatory role in membrane protein oligomerization, which is crucial for structural stability and functional integrity^1,12,57. In this study, we found that, similar to CDL, PE and PG lipids promote dimer formation of VsSemiSWEET. Further insights from ssNMR and MD simulations reveal that CDL tails localize to the outer surface of VsSemiSWEET, enhancing inter-monomer interactions and stabilizing the dimeric structure while facilitating its functional transition cycle; in contrast, PG tails penetrate deep into the protein cavity, disrupting interprotomer interactions and impeding functional conformational switching. This observation aligns with previous findings on the Na⁺/H⁺ antiporter, where CDL was identified as the optimal lipid for promoting dimer assembly, stability, and functional integrity compared to other types of lipids⁵⁸.

The different binding modes observed for PG and CDL lipids can be attributed to their molecular architectures. Cellular membranes exhibit a highly diverse lipid composition, which has the potential to modulate the structure and function of membrane proteins through various lipid–protein interactions⁵⁹. It is interesting to compare the binding modes of PG and CDL to the protein, as well as their respective influence on the structure and function of the protein. One molecule of CDL is roughly equivalent to two PG molecules, but it features a connection between the glycerol backbone of the two PG headgroups. This structural characteristic of CDL impacts distinct properties compared to a combination of two separate PG molecules. For PG, bringing two headgroups into proximity within a binding site incurs both enthalpic and entropic costs. In contrast, the structure of CDL allows for a more favorable arrangement of its phosphate groups without significant energetic penalties, which enhances its interactions with adjacent basic residues. The presence of two phosphate groups in CDL leads to tighter binding with the three basic residues, K32, H37, and H40, due to stronger electrostatic interactions, compared to PG. This observation is consistent with recent MD simulations revealing that CDL molecules exhibit preferential binding to bacterial inner membrane proteins. Notably, these simulations identified over 700 independent CDL binding sites on E.coli inner membrane proteins, demonstrating that the proximity of the two phosphate groups in the CDL molecule facilitates interactions with two to three adjacent basic residues³⁸. Additionally, the four acyl chains of CDL confer larger hydrophobic volumes compared to PG. This structural feature likely restricts access to the transmembrane cavity while simultaneously stabilizing the conformational integrity of VsSemiSWEET.

The high conservation of basic residues at the binding pockets, particularly the charged residues at positions 32 and 37 on the surface of various SemiSWEET species, highlights the importance of interactions between interfacial lipids and SemiSWEET proteins (Fig. S36). This conservation is likely a result of the co-evolution of lipids and proteins, in which organisms have developed specific lipid profiles to their environmental conditions. Consequently, these lipid profiles shape the evolution of membrane proteins, facilitating diverse lipid–protein interactions that regulate their functions. Such interactions are essential for allowing membrane proteins to adapt to various environments, thereby enhancing their functionality and stability across different physiological contexts.

In summary, our study provides structural and mechanistic insights into a fundamental question of membrane protein biology: how lipid diversity regulates protein behavior through specific interactions. Through a systematic investigation of the interactions between VsSemiSWEET and its specific bound lipids (PE, PG and CDL), our findings reveal that CDL, with its dual phosphates and four acyl chains, modulates the behavior of membrane proteins through interactions involving both the headgroup and the fatty acid. We also demonstrate that the physicochemical properties of the bound lipids play a crucial role in regulating protein conformational stability and functionality. As shown in Fig. 7, the acyl tails of PG insert into the interfacial cavity of VsSemiSWEET, disrupting both the dimer’s structure and its functional transition cycle. In contrast, the larger hydrophobic tails of CDL block access to the interfacial pore, thus avoiding disruption of the dimeric architecture. Additionally, the acyl tails of CDL interact with TM2, thereby facilitating a functional conformational transition in VsSemiSWEET. These insights highlight the complexity of specific lipid–protein interactions and emphasize how variations in the acyl tails and headgroups of bound lipids can regulate membrane protein oligomerization and functionality in different ways. Future studies could investigate whether similar lipid-mediated regulatory mechanisms exist in other membrane proteins and explore their roles under physiological or pathological conditions. Such investigations would not only expand our understanding of lipid–protein interactions but also elucidate the molecular mechanisms underlying membrane protein behavior in diverse biological contexts.

Fig. 7: Distinct lipid–protein interactions and regulatory mechanisms of CDL and PG in VsSemiSWEET.

a Molecular topology comparison: PG (single phosphatidylglycerol headgroup with two acyl chains) versus CDL (dimeric phosphatidylglycerol containing four acyl chains). b Regulatory mechanisms: In the absence of lipids, VsSemiSWEET forms unstable dimers due to repulsion between interfacial basic residues. PG acyl tails insert into the interfacial cavity of VsSemiSWEET, disrupting interprotomer interactions and inhibiting the functional transition cycle. In contrast, the acyl tails of CDL interact with TM2, stabilizing the dimeric structure and facilitating functional transitions.

Methods

Sample preparation for solid-state NMR experiments

VsSemiSWEET proteoliposome sample

Expression and purification. VsSemiSWEET was expressed in E. coli BL21(DE3) cells using a pET28a-VsSemiSWEET plasmid with an N-terminal 6×His tag. Colonies of VsSemiSWEET were inoculated into 100 mL of LB media containing kanamycin (100 μg/mL) and grown overnight at 310 K. Then, 20 mL of cell suspension was diluted into 1 L of fresh LB media and allowed to grow at 310 K. When the OD₆₀₀ reached 1.0−1.1, the cells were harvested by centrifugation and resuspended in 250 mL of M9 media (4 g/L ¹³C glucose, 1 g/L ¹⁵N NH₄Cl) and incubated at 310 K for 30 min. Notably, the 2-VsSemiSWEET and 1,3-VsSemiSWEET samples were prepared by substituting glucose in the M9 medium with the corresponding 2-¹³C– and 1,3-¹³C–labeled glycerol, respectively. Next, 1 mM IPTG (isopropyl-β-D-thiogalactoside) was added to induce protein expression at 310 K. After 5 h, the cells were collected by centrifugation and resuspended in a lysis buffer (20 mM Tris, 100 mM NaCl, pH 8.0), followed by ultrasonication until the solution became clear. The lysate was then subjected to low-speed centrifugation at 4629×g for 10 min to remove unbroken cells and larger debris. Total cell membranes were collected by high-speed centrifugation at 58,545×g for 1 h. The membranes were dissolved overnight in 1% (w/v) sodium N-lauroylsarcosinate (NLS) at 277 K. VsSemiSWEET was then purified using 2 mL of Ni-IDA resin, with 10 column volumes (CVs) of wash buffer (20 mM Tris, 100 mM NaCl, 20 mM imidazole, 0.2% (w/v) DM, pH 8.0), followed by 5 CVs of elution buffer (20 mM Tris, 100 mM NaCl, 300 mM imidazole, 0.3% DM, pH 8.0). Approximately 30−35 mg of VsSemiSWEET was purified from 1 L of M9 culture.

Reconstitution. The purified VsSemiSWEET was reconstituted into E. coli liposomes through a dialysis process. A 5 mg/mL stock of E. coli lipids was prepared by hydration and ultrasonication. The lipids were dissolved in 2% (w/v) octyl glucoside (OG) and mixed with the purified VsSemiSWEET at a protein-to-lipid ratio of 1:0.6 (w/w). The mixture was incubated at 291 K for 2 h. Subsequently, the mixture underwent dialysis at 291 K for 6 days against an external solution (20 mM Tris, pH 8.0). The VsSemiSWEET protein−liposomes were collected through ultracentrifugation at 419,832×g for 3 h, followed by freeze-drying. The dried complexes were then rehydrated with 30% (w/w) water and packed into a 3.2 mm rotor (thin-wall) for solid-state NMR experiments.

VsSemiSWEET cellular membrane sample

Expression of VsSemiSWEET. The method used for VsSemiSWEET expression was similar to that described in the preparation of the VsSemiSWEET proteoliposome sample. However, the OD₆₀₀ value of the cells before switching from unlabeled to isotopically enriched media was adjusted from 1.0−1.1 to 1.3−1.4, and the incubation time prior to adding IPTG was reduced from 30 to 10 min.

Cellular membrane isolation. The cell membranes obtained from 250 mL of culture of isotope-labeled cells were resuspended in a lysis buffer (20 mM Tris, 100 mM NaCl, pH 8.0) and sonicated until the solution became clear. To separate the inner and outer membrane fractions, sucrose gradient centrifugation was employed. Specifically, 2.7 mL of 60% (w/v) sucrose, 3.5 mL of 51% (w/v) sucrose, 3 mL of 35% (w/v) sucrose, and 4 mL of the sample were sequentially added to the buffer system (20 mM Tris, 100 mM NaCl, pH 8.0). The gradients were centrifuged in an SW-40 Ti rotor (Beckman) at 169,818×g for 16 h. The inner membranes were harvested from the interface between the 35 and 51% sucrose layers using a syringe, while the outer membranes were collected from the interface between the 51 and 60% sucrose layers. For the inner membrane sample, 2 mL was diluted with water to fill an 8.9 mL Beckman centrifuge tube, which was then centrifuged at 419,832×g for 3 h to collect the inner membranes. After freeze-drying, the inner membranes were rehydrated with 30% (w/w) water and packed into a 3.2 mm thin-wall rotor for solid-state NMR experiments.

¹³C-labeled total E. coli lipid extract

To assign the chemical shifts of the E. coli lipids, we prepared a sample of ¹³C-labeled extracts from the total E. coli lipids⁶⁰. Briefly, E. coli BL21(DE3) cells were cultured in M9 medium, using ¹³C glucose as the sole carbon source. After an overnight culture, the cells were harvested by centrifugation at 6000×g for 15 min, yielding a wet mass of 15–20 g from 1 L of M9 culture. The cell pellet (10 mg) was resuspended in 20 mL of chloroform and 40 mL of methanol, followed by filtration. Subsequently, 20 mL of chloroform and 20 mL of distilled water were added. The mixture was transferred to a separator funnel and allowed to stand at 277 K for 24 h. The ¹³C-labeled E. coli total lipids were collected from the lower chloroform phase.

Unfolding and refolding approach for VsSemiSWEET

Unfolding of VsSemiSWEET. The method used for VsSemiSWEET expression was similar to that described in the preparation of the VsSemiSWEET proteoliposome samples. The obtained 250 mL of cells were collected by centrifugation and resuspended in a lysis buffer (20 mM Tris, 100 mM NaCl, pH 8.0), followed by ultrasonication until the solution became clear. The lysate was then subjected to low-speed centrifugation at 4629×g for 10 min to remove unbroken cells and larger debris. Total cell membranes were collected by high-speed centrifugation at 58,545×g for 1 h. The membranes were dissolved overnight in 1% (w/v) NLS at 277 K. VsSemiSWEET was subsequently purified using 2 mL of Ni-IDA resin, followed by sequential washing steps with 15 mL of buffer A (20 mM Tris, 100 mM NaCl, 0.2% (w/v) NLS, pH 8.0), and 10 mL of buffer B (8 M Urea, 150 mM NaCl, and lysis buffer in a 1:1 volume ratio) to initiate the unfolding of VsSemiSWEET. Further denaturation of the protein was achieved by treating it with 140 mL of Buffer C (8 M Urea and 150 mM NaCl). Finally, to obtain a fully denatured VsSemiSWEET solution, 10 mL of Buffer D (8 M Urea, 150 mM NaCl, 1% formic acid, and 0.2% SDS) was added, and the mixture was left in a 298 K incubator overnight.

Refolding of VsSemiSWEET. An appropriate volume of 2 M Tris solution (pH 8.0) was added to the denatured VsSemiSWEET solution to achieve a final Tris concentration of 50 mM, enhancing its buffering capacity. Next, NaOH solution was gradually added to raise the pH of the protein solution from 2.8 to ~8.0. The lipids were dissolved in 2% (w/v) OG and mixed with the purified VsSemiSWEET at a protein-to-lipid ratio of 1:1 (w/w). The mixture was then incubated overnight at 310 K.

Delipidation of VsSemiSWEET

Detergent washing was employed to remove lipids loosely bound to VsSemiSWEET. In the purification of VsSemiSWEET, the membranes were dissolved overnight in 1% (w/v) NLS at 277 K. VsSemiSWEET was then purified using 2 mL of Ni-IDA resin and subjected to washing with 15 mL of wash buffer (20 mM Tris, 100 mM NaCl, 20 mM imidazole, 0.2% (w/v) DM, pH 8.0). After washing with buffer containing 1% DDM, VsSemiSWEET was eluted in an elution buffer (20 mM Tris, 100 mM NaCl, 300 mM imidazole, 0.3% DM, pH 8.0), and concentrated to ~0.5 mg/mL.

H/D exchange of VsSemiSWEET proteoliposome

The liposome pellets of VsSemiSWEET were suspended in D₂O and incubated with slow shaking at room temperature for 72 h. Subsequently, the pellets were collected through ultracentrifugation at 548,352×g for 2 h. The pellets were freeze-dried to remove moisture and then rehydrated with 30% D₂O for subsequent solid-state NMR experiments.

Solid-state NMR experiments

All solid-state NMR experiments were performed at 272 K on an 800 MHz Bruker Avance spectrometer equipped with a 3.2 mm three-channel E-free T3-HXY MAS probe, at a MAS frequency of 10.5 kHz. The typical 90° pulse length for ¹H was 3.1 μs, and the 90° pulse lengths for ¹³C and ¹⁵N were 4.5 and 6.6 μs, respectively. The chemical shifts of ¹³C and ¹⁵N were indirectly calibrated using adamantane as an external referencing standard (40.48 ppm for the downfield carbon). During the acquisition and indirect chemical shift evolution process, SPINAL-64 decoupling on the ¹H channel was set to 70−75 kHz. In the 2D and 3D solid-state NMR experiments, SPECIFIC CP⁶¹ was employed in polarization transfer between ¹³C and ¹⁵N, and ¹³C-¹³C correlations were established through dipolar-assisted rotational resonance (DARR) recoupling⁶². The 2D ¹³C-¹³C CORD experiments of 2-VsSemiSWEET and 1,3-VsSemiSWEET samples were used to acquire ¹³C-¹³C distance restraints. The 2D ¹H-¹³C INEPT experiment was performed at 298 K, with experimental conditions were optimized for a ¹_JCH coupling constant of 225 Hz. All solid-state NMR experimental data were processed using NMRPipe software⁶³ and analyzed using Sparky. Detailed parameters for these ssNMR experiments are compiled in Table S5.

Solution NMR experiments

For the ³¹P NMR study, approximately 0.15 mM of VsSemiSWEET was solubilized in DM in a pH 8.0 buffer (20 mM Tris, 100 mM NaCl, 30 mM imidazole, 0.3% DM, 10% D₂O, pH 8.0). ³¹P solution-state NMR spectra were collected on a Bruker Avance III spectrometer operating at a magnetic field strength of 11.7 T, corresponding to a proton Larmor frequency of 500 MHz. The spectrometer was equipped with a 5 mm CPP BBO 500S2 BB-H&F-D-05 Z probe and analyzed by TopSpin 4.0.7. The sample temperature was maintained at 315 K.

Extraction of structural restraints

Dihedral angle restraints

Dihedral angles were predicted using TALOS+ based on the N, CO, Cα, and Cβ chemical shifts⁶⁴. The backbone chemical shifts were directly adopted from BMRB entry 51331, as the sample preparation protocol and resulting 2D ¹⁵N-¹³Cα and ¹³C-¹³C spectra were identical to those in our prior study³³. Only angles classified as “good” by TALOS+ were used as dihedral angle restraints, with the program providing error values.

Hydrogen bond restraints

Intrahelical hydrogen bond restraints were applied when both the chemical shift index⁶⁵ and TALOS+ indicated helical structures for the corresponding residues. The intrahelical hydrogen bond restraint was defined as the distance between the donor oxygen (O_i) and the acceptor nitrogen (N_i+4), and it was set within the range of 1.5 to 3.5 Å.

¹³C-¹³C distance restraints

¹³C-¹³C distance restraints were derived from 2D ¹³C-¹³C CORD spectra with a mixing time of 500 ms for 2-VsSemiSWEET and 1,3-VsSemiSWEET samples. The signals corresponding to ¹³C-¹³C distance restraints were assigned based on the following criteria: (1) The peaks that could be assigned as ¹³C-¹³C distance constraints exhibited symmetry, were well-resolved, and had a sufficient signal-to-noise ratio (SNR ≥ 6) in the 2D CORD spectra. (2) All possible assignments of the cross peaks were collected. A tolerance of approximately 0.2 ppm was set for both direct and indirect dimensional ¹³C chemical shifts, considering the linewidth of the peaks (0.2 to 0.6 ppm) in the CORD spectra. (3) Ambiguity in the assignments was reduced through the following procedures: (i) The ambiguity in the assignment of intra-monomer correlations was notably reduced by utilizing the VsSemiSWEET monomer structure predicted by CS-Rosetta based on the assigned chemical shifts. (ii) If a peak could not be assigned as an intra-monomer correlation, it was classified as a possible inter-monomer correlation. Moreover, the assignments of inter-monomer correlations were confirmed based on the corresponding signal intensities obtained in the 2D ¹³C-¹³C spectra recorded on the [(1:2)2-¹³C/NA] VsSemiSWEET sample.

VsSemiSWEET dimer structure determination in lipid bilayers

Step 1: Monomer VsSemiSWEET structure predicted by CS-Rosetta

The initial monomeric structure of VsSemiSWEET was predicted using the standard POMONA/CS-RosettaCM protocol from the CS-Rosetta package^66,67. A set of 80 backbone ¹⁵N, ¹³Cα, ¹³CO, and 72 ¹³Cβ chemical shifts were utilized as input for the fragment and template selection process. During the subsequent structure generation, only fragments and templates derived from PDB proteins with less than 30% sequence similarity to SemiSWEETs were employed. Additionally, 28 ambiguous long-range intra-monomer distance restraints, determined based on chemical shift specificity, were incorporated for structural prediction. A total of 10,000 all-atom models were generated, from which the ten lowest-energy models were selected as the initial monomeric structures of VsSemiSWEET for further analysis. The RMSD of the backbone heavy atoms for the ten lowest-energy models was 0.9 ± 0.2 Å.

Step 2: Assignments of intra-monomer distance constraints

In the 2D ¹³C-¹³C CORD spectra of 2-VsSemiSWEET and 1,3-VsSemiSWEET samples, only isolated peaks with diagonal symmetry, suitable linewidths (0.2 to 0.6 ppm), and sufficient signal-to-noise ratios (SNR ≥6) were assigned. To reduce ambiguity in the assignment of intra-monomer distance constraints, the monomer structures of VsSemiSWEET predicted by CS-Rosetta were utilized. In this step, possible assignments involving intra-monomer distances for correlated atoms exceeding an upper limit of 9.0 Å in the monomer structure were excluded.

Step 3: Monomer VsSemiSWEET structure determination by Xplor-NIH

The VsSemiSWEET monomer structures were calculated by using the Xplor-NIH program (version 2.47)⁶⁸. Standard terms for bond lengths, bond angles, and improper angles were employed to enforce correct covalent geometry. Statistical torsion angle potentials and gyration volume terms were applied, and the hydrogen bond database term HBPot was used to improve hydrogen bond geometry. Folding calculations for full-length single-stranded VsSemiSWEET were derived from first-order sequence extensions of the chain. A total of 522 structures were generated using molecular dynamics simulated annealing in torsion angle space. The calculations involved two consecutive annealing schedules followed by final gradient minimization in Cartesian space. Each VsSemiSWEET structure contained 344 non-redundant ¹³C-¹³C distance constraints, 88 hydrogen bond constraints, and 78 torsion angle constraints derived from TALOS-N. The structural calculations began with constant temperature molecular dynamics at 3500 K, running for 800 ps or 10,000 steps, with the time step adjusted to maintain constant energy within tolerance. The temperature was gradually reduced in steps of 25 K. At each temperature, the kinetics were simulated for 2 ps or 500 steps, whichever was shorter. The force constant for the distance constraint was increased from 5 to 50 kcal mol⁻¹ Å⁻². During high-temperature kinetics at 3500 K, the dihedral angle constraint was disabled, but was enabled during simulated annealing with a force constant of 200 kcal mol⁻¹ rad⁻². The slew volume force constant was geometrically extended from 0.002 to 1. After simulated annealing, the structure underwent energy minimization using Powell’s energy minimization scheme. This methodology resulted in 200 structures, and the first 10 structures with the lowest energy exhibited an RMSD of 1.5 ± 0.4 Å for backbone atoms.

Step 4: Initial dimer structure of VsSemiSWEET generated by Xplor-NIH docking

The initial dimer structure of VsSemiSWEET in lipid bilayers was calculated using the docking module in Xplor-NIH, combined with six ambiguous inter-monomer distance constraints. The assignments of these inter-monomer correlations were confirmed by analyzing the corresponding signal intensities observed in the 2D ¹³C-¹³C spectra recorded on the 2-VsSemiSWEET and [(1:2)2-¹³C/NA] VsSemiSWEET samples. Furthermore, these correlations could not be assigned as intra-monomer correlations based on the analysis of VsSemiSWEET monomer structure. The PDBEditor tool was employed to create the initial structure for dimer docking, positioning the lowest-energy VsSemiSWEET monomer structure at two fixed points approximately 100 Å apart in a straight line. The ambiguous inter-monomer distance constraints were applied to the inter-monomer interface of the VsSemiSWEET dimer, with force constants ranging from 20 to 100 kcal mol⁻¹ Å⁻². Noncrystalline restraints were used to ensure that both monomers in the calculation were identical. This potential allowed the internal coordinates of the VsSemiSWEET monomers to remain fixed during the docking process, enabling changes in the orientation of the monomers while preserving their internal structure. During the annealing process, the hydrogen bond database potential term HBDB was used. Standard bond angles and lengths were set using the terms BOND, ANGL, and IMPR. van der Waals forces were applied with force constants ranging from 0.05 to 5.0 kcal mol⁻¹ Å⁻². Torsion angle kinetics were performed for 5000 steps at 3500 K, followed by annealing to 25 K in increments of 12.5 K for 100 steps at each temperature. After annealing, final energy minimization of the torsion angles and Cartesian coordinates was conducted. The resulting structures were sorted by energy, considering all potential energies determined from the calculations. From the final set of 100 structures, the ten structures with the lowest energy were selected for subsequent analysis.

Step 5: Assignment of the inter-monomer distance constraints

The ten initial VsSemiSWEET dimer structures obtained through docking were used to identify inter-monomer distance constraints. The assignment of intermonomer correlations in the 2D CORD spectra followed the following criteria: (I) Only the peaks that could not be assigned to intra-monomer contacts and that exhibited a noticeable decrease in intensity (>1/3) or even disappearance in the spectra of the [(1:2)2-¹³C/NA] VsSemiSWEET sample, compared to the corresponding peaks in the spectra of the 2-VsSemiSWEET sample, were assigned as unambiguous intermonomer distance constraints. (II) Only peaks with diagonal symmetry, suitable linewidths (0.2 to 0.6 ppm), and sufficient SNR (≥6) in the CORD spectra were assigned. (III) If a peak satisfies both intra-monomer and inter-monomer distance constraints, and the signal intensity in the spectrum of the [(1:2)2-¹³C/NA] VsSemiSWEET sample is significantly reduced or even disappears, it was considered an ambiguous inter-monomer constraint; otherwise, it was regarded as an intra-monomer distance constraint.

Step 6: Calculation of the VsSemiSWEET dimer structure by Xplor-NIH

The folding calculations for the full-length single-stranded VsSemiSWEET were extrapolated from first-order sequence extensions of the chain. The two monomers were constrained to maintain identical structures during the annealing process through the non-crystallographic symmetry term PosDiffPot and the translational symmetry term DistSymmPot. A total of 1000 structures were de novo generated through molecular dynamics simulated annealing in torsion angle space with two consecutive annealing schedules, followed by final gradient minimization in Cartesian space. The structure calculation utilized the obtained intra-monomer and intermonomer distance restraints derived from NMR spectra, dihedral angle restraints, and intrahelical hydrogen bonds.

The structural calculations began with constant temperature molecular dynamics at 3500 K, running for 800 ps or 10,000 steps, with the time step allowed to adjust to maintain constant energy within tolerance. The temperature was gradually reduced in steps of 25 K. At each temperature, the kinetics were simulated for 2 ps or 500 steps, whichever was shorter. The force constant for the distance constraint was increased from 1 to 300 kcal mol⁻¹ Å⁻². During high-temperature kinetics at 3500 K, the dihedral angle constraint was disabled but enabled during simulated annealing with a force constant of 300 kcal mol⁻¹ rad⁻². The slew volume force constant was geometrically extended from 0.002 to 1. After simulated annealing, the structure underwent energy minimization using Powell’s energy minimization scheme. This methodology resulted in 1024 structures, and the first 10 structures with the lowest energy exhibited an RMSD of 0.9 ± 0.3 Å for backbone atoms.

Molecular dynamics simulations

The initial structure of VsSemiSWEET in MD simulations is based on the solid-state NMR structure reconstituted in E. coli liposome. The membrane bilayer consists of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE) 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol (POPG), and cardiolipin (CDL). The lipid molar ratio is POPE:POPG:CDL = 75:13:12, which mimics the lipid composition found in E. coli membranes⁶⁹. The total number of lipids in the membrane model was 300, composed of 224 POPE, 40 POPG, and 36 CDL molecules, reflecting a physiologically relevant ratio. For the conventional MD (cMD) simulations used to analyze lipid residence times around VsSemiSWEET, three independent replicas of 3 μs each were carried out per system. The reported lipid residence times are averages derived from these replica simulations. For the PG-bound system, four additional POPG lipids were docked to the binding sites of VsSemiSWEET using MOE⁷⁰. In the CDL-bound system, two CDL molecules, each containing two phosphate groups, were docked to the binding sites of VsSemiSWEET. The membrane bilayer model and protein insertion were accomplished using the CHARMM-GUI web server⁷¹. The CHARMM36 force fields were utilized for both proteins and membranes⁷², while the TIP3P model⁷³ was adopted for water molecules.

The systems underwent energy minimization for 5000 steps using the steepest descent method. During these minimization steps, proteins and lipids were restrained with force constants of 10 kJ/mol × Ų and 2.5 kJ/mol × Ų, respectively. NVT equilibration steps, constrained to the positions of protein-heavy atoms and lipids, were conducted for 500 ps. Subsequently, several 1-ns NPT simulations were performed, during which position constraints on the protein and lipid heavy atoms were gradually reduced to equilibrate the systems. The LINCS algorithm⁵⁰ was used to constrain the bond lengths involving hydrogen atoms. A semi-isotropic scheme was employed to independently couple the lateral and perpendicular pressures. Long-range electrostatics were calculated using the Particle-Mesh Ewald method⁷⁴ with a cutoff distance of 12 Å. The temperature was maintained at 300 K using the V-rescale method⁷⁵, and pressure was controlled by the Parrinello-Rahman barostat⁷⁶. Production runs were executed with a 2-fs time step using the CUDA-accelerated version of the AMBER suite package⁷⁷.

Free energy simulations were performed using well-tempered metadynamic simulations^78,79. Two collective variables (CV), d_out and d_in, were employed to characterize the opening of the extracellular and intracellular gates on VsSemiSWEET. d_out calculated the distances between Cα atom on residue Y62 in each protomer, while d_in calculated the distances between the Cα atoms of residues L42–L44 in each protomer. The bias factor γ, which controls the added bias in a history-dependent manner, was set to 32. The height of the initial bias energy was set at 1.0 kJ mol⁻¹ and the width was set at 0.5 Å. The bias was added every 2 ps. The GROMACS2022 software⁸⁰ with the plumed 2.8 plugin was utilized to perform the metadynamics simulations. The simulation details, including the number of atoms, number of water molecules, box sizes, salt concentration, and lipid compositions, are summarized in Table S6. The convergence tests of simulations are analyzed based on the free energy surface deviation along the simulation time and are given in Fig. S37.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.