Main

Membrane transport is critical for fundamental cellular processes, including cell growth, division and homeostasis. Transporters can mediate either the active or passive transport of a wide variety of substrates across cellular membranes1,2,3. To date, several transporter folds have been identified4,5,6,7,8, including members of the ATP-binding cassette (ABC) and solute carrier families, among others. However, many membrane proteins hypothesized to be transporters remain uncharacterized, even in well-studied model organisms such as E.* coli*. Some of these hypothetical transporters may be evolutionarily related to known transporter families, but have diverged beyond recognition at the sequence level. Alternatively, these unstudied protein families may represent new kinds of transporters that await experimental characterization.

The mammalian cell entry (MCE) family of proteins has been implicated in lipid transport across the cell envelope in double-membraned bacteria9,10,11,12 and between the endoplasmic reticulum and chloroplasts in plants13,14. MCE proteins have an important role in maintaining the cell envelope of Gram-negative bacteria10,15, and scavenge host lipids, such as fatty acids and cholesterol, in Mycobacteria11. The MCE domains that define this protein family hexamerize to form rings with a central pore, the basic building block for diverse higher-order architectures that form pathways for lipid transport between membranes9,16. Associated integral membrane proteins are thought to drive lipid translocation through the MCE ring. The best-characterized MCE systems interact with ABC transporters to drive substrate translocation9,10,11,13,14,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30. However, many other MCE gene clusters do not encode components of an ABC transporter, and it is unknown whether and how lipids are translocated, or how transport is energized.

Lipophilic envelope-spanning tunnel B (LetB) is a large MCE protein, long enough to span the periplasm between the E. coli inner membrane and outer membrane (Fig. 1a). The prevailing model is that LetB transports lipids between membranes through a central hydrophobic tunnel9,15,31,32,33. How lipids enter the LetB tunnel and the direction of transport remain unknown. LetB is encoded in an operon together with LetA (Fig. 1b), a multipass transmembrane (TM) protein with no detectable homology to transporter families. In Gram-negative bacteria, proteins homologous to LetA are encoded adjacent to some classes of MCE proteins, including the paraquat inducible (Pqi) system in E. coli, suggesting that these proteins may have evolved to function together. LetA is poised to facilitate substrate translocation through the LetB tunnel and has the potential to define a new class of membrane transport proteins.

Fig. 1: LetA and LetB form a complex.

a, Model of LetA and LetB in the cell envelope. A cross-section of LetB (PDB 6V0C) is oriented in the context of the inner membrane (IM) and outer membrane (OM), with phospholipids (PL) and LPS indicated. b, Schematic of the letAB operon. c, Western blot from a pull-down assay to assess the interaction between LetA and LetB. His–LetA was used as the bait, and the interaction with untagged LetB was assessed using anti-LetA (clone 72) and anti-LetB antibodies. Three independent purifications were performed starting with three different colonies, with similar results. d, 2D class averages from negative-stain electron microscopy data for full-length LetAB or the soluble periplasmic domain of LetB alone. e, Residues in LetB that were targeted for incorporation of photocrosslinking amino acid, BPA (red sphere for the inside tunnel, and blue spheres for the outside tunnel). f, SDS–PAGE analysis of purified LetAB without BPA incorporation (WT) or with BPA incorporated at positions indicated in panel e. Samples were either UV crosslinked in vivo or uncrosslinked, and the SDS–PAGE gel was stained with Coomassie (LetB) and phosphor-imaged (32P signal). Three replicates were performed starting with three different colonies, on different days, with similar results. g, SDS–PAGE analysis of purified LetB with BPA incorporated at position F468, with or without co-expression of LetA, prepared as in panel f. Three replicates were performed starting with three different colonies, on different days, with similar results. h, Surface representation of our LetAB cryo-EM structure oriented in the context of the inner membrane and outer membrane. LetB monomers are depicted in different colours. i, Views of the LetAB complex from the cytoplasm (top) and outer membrane (bottom), shown as surface representations. The LetA surface is partially transparent (blue). Gel source data for panels c,f,g are provided in Supplementary Fig. 1a,c,d, respectively.

Here we show that LetA and LetB function together, and report the structure of the LetAB complex. We found that LetA is distantly related to the eukaryotic tetraspanin superfamily of membrane proteins, which are not known to have intrinsic transporter activity, including the transmembrane AMPA receptor regulatory proteins (TARPs), claudins and vitamin K epoxide reductase (VKOR). Our structure, together with deep mutational scanning (DMS), molecular dynamics simulations, and AlphaFold predictions of alternative states coupled with experimental validation, led to a model for how LetA may drive phospholipid transport to maintain outer membrane integrity in E. coli, providing insights into a previously uncharacterized family of transporters in bacteria.

LetA and LetB form a complex

Deletion of letA and letB together (ΔletAB) in E. coli has previously been shown to cause mild sensitivity to the bile salt, cholate and the zwitterionic surfactant lauryl sulfobetaine (LSB)31,34. Both phenotypes are exacerbated when pqiAB, a second E. coli MCE system, is also deleted (ΔpqiAB ΔletAB)31,34. To assess the relative contributions of letA and letB to cholate and LSB sensitivity, we deleted letA and letB individually in a ΔpqiAB background. Strains lacking letA or letB exhibited similar growth defects to each other and to ΔletAB mutants, which could be rescued by complementation with a plasmid carrying wild-type (WT) letAB (Extended Data Fig. 1a). These results indicate that LetA and LetB function in the same pathway.

To examine whether LetA and LetB physically interact, we co-expressed both proteins in E. coli, and found that His-tagged LetA pulls down LetB (Fig. 1c), resulting in an approximately 670 kDa complex (Extended Data Fig. 1b,c). Negative-stain electron microscopy of LetAB shows particles with seven characteristic bands of density resembling LetB31 and additional globular density at one end (Fig. 1d), which we hypothesized corresponds to LetA and the TM helices of LetB surrounded by a detergent micelle. Overall, these data show that LetA and LetB form a stable complex.

LetA facilitates lipid loading into LetB

Previous studies have shown that the soluble, periplasmic domain of LetB binds phospholipids9, and crosslinking experiments in E. coli lysates suggest that the binding sites are in the LetB central tunnel31. It is unclear, however, whether phospholipids spontaneously enter the tunnel of the full-length, membrane-embedded LetAB complex in vivo. To address this question, we used an in vivo crosslinking assay. We grew E. coli in the presence of 32P orthophosphate to label phosphate-containing molecules, including phospholipids, and over-expressed LetAB with the photocrosslinking unnatural amino acid p-benzoyl-l-phenylalanine (BPA) incorporated at specific sites in LetB. We then UV irradiated live cells to allow in vivo crosslinking of molecules in proximity to the site of the BPA probe. Following purification of LetAB complexes, we analysed the crosslinking of 32P-labelled molecules to LetB by electrophoresis and phosphorimaging. BPA was positioned inside the LetB tunnel (F468, W476, Y814 or F833) or on the periplasm-facing exterior surface (K488 and E854, negative controls; Fig. 1e), locations validated in previous work31. We detect 32P incorporation into LetB at all four BPA sites inside the tunnel, with F468B showing the highest level, but we detected minimal 32P incorporation with BPA positioned outside the tunnel (Fig. 1f). To test whether lipid access to the LetB tunnel is dependent on LetA, we assessed lipid crosslinking inside the LetB tunnel with or without co-expression of LetA, using F468B as a probe. Efficient crosslinking in the LetB tunnel is dependent on the co-expression of LetA (Fig. 1g), whereas LetB membrane localization is unaffected by LetA co-expression (Extended Data Fig. 1d). Together, these results suggest that LetA is necessary for phospholipid entry into the tunnel of full-length LetB in vivo. The simplest interpretation is that LetA is an exporter that loads lipids from the inner membrane into LetB, but an alternative model is that LetA modulates import from the outer membrane to the inner membrane by allosterically regulating the loading of LetB.

Overall structure of the LetAB complex

To understand how LetA and LetB interact, we determined the structure of the LetAB complex using cryo-electron microscopy (cryo-EM). We determined two LetAB structures, in the presence or absence of a crosslinker, glutaraldehyde. Both datasets yielded maps with similar average resolutions across the LetAB complex (2.5–4.6 Å; Extended Data Tables 1 and 2 and Supplementary Figs. 2 and 3) and yielded similar final models (Extended Data Fig. 1e). As the TM region is better resolved in the presence of the crosslinker (Supplementary Figs. 2b,d and 3b,d,f), we primarily focus our discussion on the crosslinked LetAB structure, except where noted.

The LetAB complex is an elongated assembly (approximately 290 Å long and about 90 Å wide; Fig. 1h,i) consisting of six copies of LetB and one copy of LetA, where LetB accounts for approximately 225 Å of the total length (Fig. 1h,i). As expected, each LetB copy contains seven MCE domains, which associate laterally with other LetB protomers to form seven MCE rings that create a hydrophobic tunnel31,34 (Extended Data Fig. 1f,g). Six N-terminal TM helices, one from each LetB protomer, anchor the assembly in the inner membrane. A single copy of LetA interacts with MCE ring 1 and the TM helices of LetB (Extended Data Fig. 1h). An approximately 30 Å hydrophobic belt around the LetA circumference defines the region probably embedded in the inner membrane (Extended Data Fig. 1i). Of note, the density for only four of the six LetB TM helices is apparent in the electron microscopy map, and the four resolved helices interact with LetA in two nonequivalent ways. The remaining two TM helices are not resolved (Extended Data Fig. 1h) and may not stably interact with LetA, resulting in pronounced asymmetry in the TM region of the complex.

The wall of the LetB central tunnel is formed by pore-lining loops that emerge from each MCE domain31. Previous cryo-EM structures of LetB with the TM helix deleted have shown that the pore-lining loops from MCE rings 1, 5, 6 and 7 can adopt open and closed conformations, which control the diameter of the central tunnel, thereby potentially regulating the passage of substrates31,32. LetA is positioned directly underneath the pore of LetB MCE ring 1. In the absence of LetA, MCE ring 1 of LetB is predominantly in the closed state, in which the pore through the ring is not wide enough to allow passage of a phospholipid (Extended Data Fig. 1g,j). This closed conformation of MCE ring 1 is observed both in periplasmic31,32 and full-length32 structures of LetB. In our LetAB structure, MCE ring 1 of LetB adopts an open state, suggesting that binding to LetA modulates the conformation of the LetB tunnel (Extended Data Fig. 1g,j).

Overall structure of LetA

E. coli LetA is a single polypeptide that consists of two related modules, which we term ‘LetA modules’. Each LetA module consists of a cytoplasmic zinc ribbon (ZnR) domain followed by a TM domain (TMD; Fig. 2a–c). These modules are widespread across Proteobacteria15, and are found either in a single gene encoding two LetA modules, as in E. coli, or in two adjacent genes that each encode a single LetA module (for example, in Pseudomonas aeruginosa). E. coli LetA can form a functional heterodimer when the two LetA modules are artificially split into separate genes resembling the P. aeruginosa orthologue (split-LetA; Extended Data Fig. 2a,b). The TMD of each LetA module contains four TM helices, one interfacial helix at the membrane–periplasm boundary and a three-stranded β-sheet extending into the periplasm (Fig. 2a–d). The two LetA modules, which share approximately 25% sequence identity, associate in a head-to-head manner, resulting in an intramolecular dimer with twofold pseudo-symmetry (Fig. 2e). The two TMDs form an inverted V-shape, creating a large, hydrophilic cleft that faces the cytoplasm (Fig. 2f). In this cleft, we observed a 361GRWSM-Ψ-D-Ψ-F369 motif (where Ψ denotes an aliphatic amino acid: L, I, V or M) that is well conserved in the C-terminal LetA module across a diverse set of LetA-like proteins (Extended Data Fig. 2c–e). A similar motif is also present in the N-terminal LetA module, but is less conserved. In addition, LetA contains a periplasmic pocket 174 Å3 in volume (Fig. 2f), which is amphipathic and formed primarily by residues of TMDC, along with TM3 of TMDN. This periplasmic pocket sits directly below the entrance to the LetB tunnel, with the LetA periplasmic β-sheets creating a hydrophobic bridge that connects the pocket to the pore lining loops of LetB MCE ring 1 (Extended Data Fig. 3a). By contrast, an equivalent pocket is not present in TMDN. The cleft and periplasmic pocket could potentially serve as substrate-binding sites and function as part of the substrate translocation pathway.

Fig. 2: Structural overview of LetA.

a, Schematic representation of the LetA protein domain organization. b, Cartoon representation of the LetA structure, coloured as in panel a. N-terminal and C-terminal extensions were not resolved in our density and are shown as dashed lines drawn approximately to scale. Membrane boundaries are indicated by black lines. c, Topology diagram of LetA. In addition to the secondary structure, the Zn-coordinating cysteines (yellow circles) and Zn atoms (grey circles) are shown. d, Cartoon representation of the LetA TMDN. The colours and labels are the same as in panel c; periplasmic β-strands are shown in the inset. e, Superposition of TMDN and TMDC, showing structural conservation between the two domains. TMDN is rotated approximately 160°, which results in the superposition of the two domains. f, Electrostatic potential surface of LetA shown in full (left) and in cross-section (right), highlighting the periplasmic pocket and central cleft. g, Cartoon representation of the ZnR domains. Metal-coordinating cysteines are labelled and the Zn atoms are shown as blue spheres. h,i, Cartoon representations with helices shown as cylinders (h) and corresponding topology diagrams (i) of LetA TMDN, TARPγ2 (PDB 6DLZ) and claudin-4 (PDB 7KP4). The secondary structural elements of LetA TMDN conserved with structurally related proteins are coloured as in panel c.

On the cytoplasmic side, ZnRN and ZnRC interact to form a structural unit. Each ZnR domain consists of two stacked β-hairpins (Fig. 2g) with a tetracysteine motif involved in metal binding (CXXC-Xn-CXXC, where n ranges from 11 to 18; Extended Data Fig. 2c). ZnRC connects TMDN and TMDC, and interacts non-covalently with ZnRN to form a ZnR dimer, perhaps stabilizing the association between the N-terminal and C-terminal halves of LetA. Depending on their functional role, ZnR domains can bind various transition metals35, most commonly zinc or iron. To assess whether LetA is a metal-binding protein and to profile its metal-binding specificity, we performed inductively coupled plasma mass spectrometry (ICP-MS) on purified LetA protein, which shows specific enrichment of zinc atoms (Extended Data Fig. 3b). Calibration using a standard curve (Methods) suggests that LetA binds to approximately two zinc atoms per protein molecule (n = 2, range of 1.7–2), indicating that both ZnR domains preferentially coordinate zinc under our experimental conditions. As the metal remains bound throughout the purification process, and related ZnR domains bind tightly to zinc36, we infer that LetA probably also binds to zinc with high affinity.

LetA defines a new transporter family

To assess whether LetA is evolutionarily related to known transporter families, we performed a structure-based search of the Protein Data Bank (PDB) using Foldseek37. We were unable to identify structural similarity to known transporter folds, suggesting that LetA represents a new type of membrane transport protein. However, this search revealed that an individual LetA TMD is structurally related to the tetraspanin superfamily of integral membrane proteins in eukaryotes. The LetA TMD most closely resembles TARPs and claudins, which have structurally equivalent β-sheets in their extracytoplasmic regions with 3–5 β-strands, and is more distantly related to VKOR and tetraspanin itself (Fig. 2h,i and Extended Data Fig. 3c,d). All of these proteins share a common topology in the TM helices, but only LetA contains ZnR domains, and is arranged as a pseudodimer with two consecutive tetraspanin-like domains. Functionally, the eukaryotic proteins are highly divergent, and none is known to exhibit transporter activity. Tetraspanins are involved in membrane organization via the formation of microdomains that serve to recruit binding partners, often involved in signal transduction38. TARPs regulate ion channel function in neurons39, claudins function in cell–cell adhesion40 and VKOR is involved in the recycling of oxidized vitamin K1 (ref. 41). Both tetraspanin and VKOR contain lipid-binding sites for cholesterol42 and vitamin K1 (ref. 41), respectively, roughly in regions corresponding to the periplasmic pocket in LetA, which is a possible substrate-binding site (Extended Data Fig. 3e).

To explore evolutionary relationships between LetA and proteins whose structures have yet to be experimentally characterized, we carried out a Foldseek search of the AlphaFold database of predicted protein structures. In addition to bacterial LetA-like proteins, this search revealed potential uncharacterized structural homologues of full-length LetA that are present in some parasites and marine protists (Extended Data Fig. 3f). The AlphaFold predictions resemble LetA, but lack ZnR domains. As MCE proteins are generally restricted to double-membraned bacteria and photosynthetic eukaryotes, it is unclear how LetA-like proteins function in parasites and marine protists. However, LetA-like proteins identified in kinetoplastids and dinoflagellates appear to be fused to an extracytoplasmic β-jellyroll domain (Extended Data Fig. 3f), reminiscent of the bridge-like lipid transport domains of VPS13 (ref. 43), YhdP44 and the LPS exporter45. Thus, these distantly related LetA relatives may mediate the transport of lipids in some eukaryotes via bridge-like proteins instead of MCE tunnels. Together, these analyses place LetA and LetA-like proteins in the tetraspanin superfamily, which was previously thought to be a eukaryotic innovation38, but we show to be present in prokaryotes as well.

DMS of LetA

To gain unbiased insight into functionally important residues in LetA, we used DMS, in which each position in LetA was mutated to all possible amino acids (Extended Data Fig. 4a). The effect of each mutation on LetA function in cells was assessed in the presence of LSB or cholate. Heatmaps illustrating the effect of each mutation on LetA fitness show similar patterns with cholate and LSB (Extended Data Fig. 5 and Supplementary Fig. 4). As expected, mutation of the start codon or introducing a stop codon at most positions resulted in reduced fitness. For each position, we calculated a tolerance score (Methods), ranging from 0 to 1, where 0 denotes no tolerance for mutations and 1 denotes full tolerance (Fig. 3a, Extended Data Fig. 5 and Supplementary Fig. 4). Approximately 90% of residues tolerate mutations (tolerance scores ≥ 0.7; Extended Data Fig. 4d), including an approximately 25 residue cytoplasmic extension at the N terminus of LetA (Extended Data Fig. 5 and Supplementary Fig. 4). LetA constructs truncating this region are expressed and largely rescue growth of the ΔpqiAB ΔletAB strain in the presence of cholate or LSB (Extended Data Fig. 4e,f). However, a subset of positions in LetA were less tolerant of mutation (tolerance score < 0.7; Extended Data Fig. 4d), including 53 positions for cholate and 37 positions for LSB (Fig. 3a, Extended Data Fig. 5 and Supplementary Fig. 4). The majority of these functionally important residues cluster in three regions of the LetA structure (Fig. 3b): (1) the periplasmic pocket in TMDC, (2) a polar network in TMDC, and (3) the ZnR domains.

Fig. 3: Functional regions of LetA revealed by DMS and cellular assays.

a, LetA structure coloured by mutational tolerance scores. Residues most sensitive to mutation appear as deeper shades of red and a thicker backbone trace. b, Cartoon representation of LetA displaying residues identified as functionally important by DMS (spheres mark Cα of each). Three functional regions were identified: the periplasmic pocket (yellow), the polar network (cyan) and the ZnR domains (green). c, Enlargement of the LetA periplasmic pocket, highlighting the position of the periplasmic β-sheet and functionally important residues. d, DMS data corresponding to residues in the periplasmic pocket. The vertical strips for individual LetA residues from the heatmap shown in Extended Data Fig. 5 are reproduced here. Each square represents the average fitness cost of an individual mutation relative to the WT sequence (two replicates). Squares containing an ‘X’ indicate incomplete coverage. The coloured square above each strip indicates the tolerance score, calculated as described in the Methods. e, A snapshot of the LetA coordinates from equilibrium molecular dynamics simulations, highlighting the region of the LetA structure corresponding to the polar network. Residues in the polar network with low tolerance scores from DMS experiments are shown as sticks. Water molecules from molecular dynamics simulations are shown, and hydrogen bonds between the residues and water molecules are illustrated as black dotted lines. f,g, DMS data corresponding to residues in the polar network (f) and ZnRC domain (g). Individual strips are shown for each residue (f) or the whole region corresponding to ZnRC (g), with colours and annotations as in panel d.

Periplasmic pocket

The periplasmic pocket, situated right below the LetB tunnel, is well positioned to serve as a binding site for lipids moving between the inner membrane and LetB, and is analogous to the lipid-binding sites observed in tetraspanin and VKOR. From the cholate and LSB datasets combined, approximately half of all positions with low tolerance scores clustered to this region (28 residues), suggesting that the periplasmic pocket is functionally important (Fig. 3c,d). Most of these residues are hydrophobic, and are less tolerant of mutations to polar residues, suggesting that maintaining the hydrophobic character of this pocket is critical, consistent with a role in binding to lipids or other hydrophobic molecules. Most of the 28 residues are buried within the pocket or cluster to TMDC strands β1 and β3, which may allow β1 and β3 to act as a hydrophobic ‘slide’ for lipid translocation between the periplasmic pocket of LetA and the pore of MCE ring 1 (Fig. 3c and Extended Data Fig. 3a). Together, our data support an important role for the periplasmic pocket in LetA function, potentially as a substrate-binding site involved in lipid translocation between LetA and LetB.

Polar network in TMDC

A cluster of well-conserved residues with low tolerance scores forms a polar network across the membrane, from the periplasmic pocket to the cytoplasm (Fig. 3e and Extended Data Fig. 2c). These residues (K178, D181, S321, K328, S364, D367 and T402) lie in the core of TMDC and have polar or charged side chains, which is unusual in TM regions. Two of these residues (S364 and D367) belong to the 361GRWSM-Ψ-D-Ψ-F369 motif near the central cleft of LetA, and other residues in the motif are also moderately sensitive to mutation (Extended Data Fig. 5 and Supplementary Fig. 4). Mutating each of the seven polar network residues to alanine had little effect on LetA expression or LetB binding, suggesting that the polar network has a specific role in the transport mechanism, independent of folding or stability (Extended Data Fig. 6a). Of note, similar polar networks in other membrane proteins often interact with water46 and can be important for substrate interaction or energy transduction by forming proton transfer pathways1,47. To examine the solvent accessibility of residues in the polar network, we performed equilibrium molecular dynamics simulations. Waters were observed within the core of TMDC, forming a network of hydrogen-bonding interactions with the polar residues, bridging the periplasmic and cytoplasmic spaces (Fig. 3f and Extended Data Fig. 6b,c). The core of TMDN, which lacks a polar network, remains inaccessible to water (Extended Data Fig. 6b). Thus, this conserved polar network probably interacts with water, and may be important for substrate interactions or for enabling proton shuttling as an energy source.

ZnR domains

ZnR domains are typically stabilized by metal-coordinating cysteines48. As expected, mutations in the metal-coordinating cysteines of the LetA ZnRs are not tolerated (Fig. 3g, Extended Data Figs. 5 and 6d and Supplementary Fig. 4) and result in reduced levels of LetA protein (Extended Data Fig. 6e). ZnR domains sometimes mediate protein–protein interactions35,49. However, aside from the Cys residues, no other residues in the LetA ZnR domains were sensitive to mutation (tolerance score ≥ 0.7; Fig. 3a), suggesting that the LetA ZnRs are unlikely to mediate protein–protein interactions, and instead, the overall ZnR fold may contribute to the stability of LetA. To test this, we replaced ZnRN and ZnRC of LetA with those of E. coli PqiA, which share 41% and 28% sequence identity, respectively (Extended Data Fig. 6f). The ZnR-swap mutant showed cholate and LSB resistance similar to the WT (Extended Data Fig. 6g,h), suggesting that substantial sequence divergence can be tolerated outside of the Zn-coordinating cysteines.

To examine whether the ZnRs are essential for function, we tested two ZnR deletion mutants in complementation assays. LetAΔZnRN fails to rescue growth and is unable to pull down LetB (Extended Data Fig. 6i–k and Supplementary Fig. 1k,l), suggesting that this mutation interferes with folding. LetAΔZnRC partially rescues growth despite reduced expression levels compared with the WT, and binds to LetB in a pull-down assay (Extended Data Fig. 6i–k and Supplementary Fig. 1k,l). These results suggest that the LetAΔZnRC protein is folded and at least partially functional. Consistent with the idea that the ZnRC domain may not have a key role