Bacterial outer membrane proteins (OMPs) are critical players in host–pathogen interactions and environmental adaptation. Here we describe the newly developed “Gradient Enrichment of Native Targets from Lipid Environments” (GENTLE) methodology and use this approach to elucidate the structures of Campylobacter jejuni OMPs directly from native, detergent-solubilized crude membranes. We identify and solve high-resolution cryo-EM structures of PorA, OMP50, and Cj0034c from C. jejuni membranes, all of which are required for Campylobacter invasion, adhesion, and initiation of host infection. Notably, our results provide the first structural information of OMP50, revealing a two-domain architecture constructed with an all β-stranded transmembrane domain and an all α-helical periplasmic domain. This structure depicts that all tyrosine residues, many of which are expected to be critical for phosphorylation and host–pathogen interaction, are localized to the outer membrane of C. jejuni. Our studies also led to the first structure of the full-length Cj0034c protein, which assembles as a nonamer with each protomer containing a single-spanning transmembrane helix and a large periplasmic domain. The nine protomers stack side-by-side to form a channel that spans the entire lipid bilayer. However, whether Cj0034c spans the outer membrane (OM) or inner membrane (IM) of C. jejuni must await further experimental studies. In addition, we observed that the surface-exposed extracellular loop L4 of PorA is very flexible, which may be critical for the virulence of this porin. Collectively, this work provides novel structural information for functionally important OMPs and sheds light on how they assemble in native bacterial membranes. These findings further demonstrate that it is possible to obtain high-resolution structural information for targeted membrane proteins from crude native membranes without their overexpression and purification.

“Gradient Enrichment of Native Targets from Lipid Environments” (GENTLE) methodology

Application of the GENTLE methodology allowed us to identify and solve cryo-EM structures of OMPs from native C. jejuni membranes without overexpression and purification of our protein targets. We grew C. jejuni NCTC 11,168 cells, harvested their membranes, solubilized them in detergent micelles, and enriched the membrane protein components using sucrose cushion centrifugation. Mass spectrometry confirmed the presence of OMPs in this C. jejuni membrane sample (S1 Table). This enriched sample was then applied to a holey carbon grid and single-particle cryo-EM images were collected. The ensuing cryo-EM data were processed using the BaR methodology [33], where in silico purification and sorting of images were able to isolate and separate the images into different classes. 2D classifications of the images indicated that there are at least two distinct classes of images resembling the architecture of OMPs. In addition, a third class of images depicting a membrane protein that likely possess a large periplasmic domain was detected. Several iterative rounds of 2D/3D classifications allowed us to rigorously sort these images (S1 Fig). Based upon these classes of single particles, we were able to identify and solve high-resolution cryo-EM structures of the PorA/MOMP and OMP50/CjTK OM porins. Additionally, we obtained a high-resolution cryo-EM map of the protein images subsequently identified as the virulence factor Cj0034c/CjSLP (S1 Fig and S2 Table). The Step-by-step GENTLE protocol is listed as follows:

Step 1: Sample preparation. The preparation procedures entail cell growth, cell lysis, membrane isolation, solubilization of membranes in detergent (such as lauryl maltose neopentyl glycol (LMNG) or glycol-diosgenin (GDN)), and the enrichment of protein components of solubilized membranes using sucrose cushion ultracentrifugation.

Step 2: Target validation. The presence of target proteins is confirmed using mass spectrometry and proteomic analysis.

Step 3: Single-particle cryo-EM. Membrane sample is loaded onto cryo-EM grids followed by cryo-EM data collection, initial image processing, and preliminary 2D and 3D ab initio classifications.

Step 4: Build and Retrieve. Different subsets of 3D classes are vigorously cleaned and re-sorted via several rounds of 2D classifications and ab initio 3D reconstructions to build initial maps. The built maps are then used as templates for 3D heterogeneous classifications of the 2D cleaned particles, where each subset is treated separately to retrieve additional particles for higher resolution reconstruction.

Step 5: Map and model refinement. Retrieved subsets are cleaned via multiple rounds of 2D and 3D classifications, and non-uniform refinements performed to build the final maps. Symmetry can be applied to improve the resolutions of the maps. As soon as the resolutions of the cryo-EM maps reach ~3.5 Å or better, these maps are used to determine their corresponding identities using the program DeepTracer [35]. Models are then constructed using Coot [36] and refined using PHENIX [37].

Detailed procedures, including sample preparation, proteomic analysis, cryo-EM data collection and processing, building maps and retrieving single particles, and map and model refinements are described in the Methods section.

PorA/MOMP

The most abundant OM porin that we identified is the PorA membrane protein (S1 Fig). PorA was named MOMP because it is highly abundant at the OM of C. jejuni. It creates a relatively large pore on the cell surface to allow for the exchange of solutes and solvents in and out of the C. jejuni cell [26,38]. PorA appears to be a moonlighting protein that performs multiple functions. It has been reported that PorA functions not only as a porin, but also as an adhesin to anchor the bacterium at the host’s intestinal epithelium [25,38,39]. The myriad of genetic polymorphisms of the porA gene (cj1259), particularly in surface-exposed regions of expressed PorA variants, often allow the bacterium to bypass the host’s immune response [40]. The associated virulence of this porin has been highlighted by the discovery of the antibiotic-resistant and hypervirulent strain, named clone “sheep abortion” (SA) [41,42], where mutations within one of the cell surface-exposed loops of PorA are responsible for the hypervirulence [27].

Structure of PorA-I. We collected a total of 136,178 single-particle projections for this class of images. Rigorous 2D/3D classifications allowed us to sort these projections into two subclasses with similar structural features. The first subclass contains 85,621 single particles with the generated cryo-EM map at 2.47 Å resolution. The program DeepTracer [35] led us to confirm the identity of this membrane protein as full-length PorA. We then solved its cryo-EM structure, designated as PorA-I, at this resolution (S2 Table). The structure of PorA-I resolved from crude C. jejuni membranes depicts that this porin assembles as a trimer. Each protomer forms an 18-stranded β-barrel (β1-β18) protein, where this barrel creates a pore to allow for substrate transport (Figs 2A–2C and S2). Therefore, PorA possesses three pores within the trimer. This structural assembly is in good agreement with the X-ray structure this protein [26]. Superimposition of our cryo-EM structure to the X-ray structure of PorA (PDB ID: 5LDT) [26] gives rise to a root-mean-square-deviation (r.m.s.d.) of 0.6 Å (S3 Fig). Each PorA protomer also contains nine extracellular loops (L1–L9) and eight periplasmic loops (T1–T8), connecting the 18 β-strands which form the core of the β-barrel (Figs 2C and S2). Interestingly, L3, L4 and L7 fold into the middle of the pore formed by the 18-stranded β-barrel (Fig 2C). These loops also create the narrowest region of each pore (Fig 2D). Residues 171–177 of L4 create an α-helix (α2) facing toward the outer leaflet of the OM. An extra spherical-shaped cryo-EM density surrounded by L3, L4 and L7 was found in the middle of each PorA protomer. The location of this extra density overlaps with a Ca²⁺ ion found in the X-ray structure of PorA [26]. We therefore assigned this extra density as a Ca²⁺ ion, which forms bonds with D142 of L3, Q174 and D177 of L4, and E307 of L7 (Fig 2C). These negatively charged and polar residues along with helix α2 appear to be critical for ion binding. Electrostatic surface potential calculations indicate that the extracellular surface of the PorA β-barrel is predominantly neutral, but residues at locations corresponding to the outer leaflet of the OM make up positively and negatively charged patches (Fig 2E). Interestingly, the pore formed by each PorA protomer is negatively charged in nature (Fig 2E), suggesting that PorA may favor the shuttling of positively charged substrates. This result is indeed in good agreement with the observation that PorA is capable of transporting small positively charged poly-arginine peptides across the membrane [43].

C. jejuni typically colonizes within the gut of the host, but a recently emerged hypervirulent clone has shown to be capable of moving across the intestinal track, leading to bacteremia and abortion in pregnant animals [27]. Through the methodology of “directed genome evolution” [27], followed by selection in an animal disease model and whole-genome sequence analysis, it was found that mutations of the PorA membrane protein are the main cause of these abortions [27]. These results also highlight that PorA is a key virulence factor and an attractive candidate for the development of therapeutic strategies to combat Campylobacter infections. Site-specific mutagenesis indicates that amino acid substitutions on L4 of PorA are required to establish the abortion phenotype of C. jejuni [27]. Our cryo-EM structure of PorA-I depicts that L4 is a surface-exposed loop likely responsible for direct interaction with the host. In fact, of the 18 mutations observed in PorA of the hypervirulent SA strain, nine of these substitutions were found within the flexible L4. Subsequent experiments have also verified that only mutations in L4 were needed to cause abortion in guinea pig models [27].

Structure of PorA-II. The second subclass contains 50,557 single particles, which allowed us to solve the structure of this subclass, designated as PorA-II, to a resolution of 2.74 Å (Figs 2F and S1; S2 Table). The overall structure of PorA-II is very similar to that of PorA-I with an r.m.s.d. of 0.4 Å when these two structures are superimposed (S4 Fig). The structure of PorA-II indicates that this protein also exists as a homotrimer, where each PorA protomer presents as an 18-stranded β-barrel protein. When the structures of PorA-I and PorA-II are compared, the major difference arises from the flexible surface-exposed loop L4. In PorA-I, residues 171–177 of L4 constitute a short α2 helix, which appears to be important for Ca²⁺ recognition with Q174 and D177 participating in ion binding. In PorA-II, the cryo-EM densities originating from residues 171–191, which comprise the majority of L4, are missing (Figs 2F and S4). In addition, no bound Ca²⁺ ion is found in this cryo-EM structure. It appears that the secondary structural elements of L4 may not be stable and are likely unstructured in the absence of bound Ca²⁺. This observation indeed highlights the intimate coordination between PorA and Ca²⁺. It is known that host Ca²⁺ ions can trigger bacterial virulence, adhesion, host cellular damage, and host-defense resistance [44]. In addition, it has been demonstrated that the host intracellular Ca²⁺ concentration is a key factor to regulate Campylobacter virulence, where C. jejuni is capable of mobilizing the host’s intracellular Ca²⁺ to initiate an invasion into host intestinal cells [45]. To that end, our structures suggest that PorA may be able to participate in Ca²⁺ signaling to manipulate host cells and facilitate C. jejuni invasion.

OMP50/CjTK

C. jejuni OMP50 is an OM porin with a size of approximately 50 kDa [29]. Interestingly, this membrane protein is a species-specific porin expressed in C. jejuni and C. lari [32,46]. Its gene cj1170c is not ubiquitous in Campylobacter spp. OMP50 is homologically very distinct from other gram-negative OMPs of similar sizes. It caught the attention of the scientific community when it was revealed that this membrane protein is the only C. jejuni tyrosine kinase (CjTK) and also a major virulence factor of the bacterium [30]. Thus, it appears that OMP50 is also a moonlighting protein, serving as both an OM porin and a tyrosine kinase, where phosphotyrosine signaling modifies several outer membrane and periplasmic proteins required for bacterial capsule and polysaccharide synthesis [30]. Therefore, disruption of OMP50 signaling may be a promising direction to pursue in order to deliver a novel antimicrobial therapy to treat this infection.

Structure of OMP50-I. Our GENTLE approach allowed us to obtain the first structure of the OMP50 membrane protein. We collected a total of 30,483 single particles for this class of images. Like PorA, this class of particles can be divided into two subsets with similar structural features, representing two distinct conformations of this OMP. We accumulated 21,652 single particles of the first subset of images, and then constructed the cryo-EM map to a resolution of 2.82 Å (S1 Fig and S2 Table). The program DeepTracer [35] led us to identify this protein as full-length OMP50. We then solved its structure, designated as OMP50-I, to this resolution (Fig 3A–3C). The structure reveals that the architecture of OMP50-I is very distinct from that of PorA-I. Our data shows that OMP50-I is trimeric in form. Each OMP50-I protomer assembles in such a way that it creates a pore. Therefore, the trimer contains three separate pores. Each OMP50-I protomer contains two α-helices and 18 β-strands that are folded into a 16-stranded β-barrel (β1–β9 and β12–β18) with a central pore (Figs 3C and S5). Some of these β-strands, such as β4-β6, are quite short in length and can only span part of the OM. Each OMP50 protomer also constitutes eight extracellular loops (L1–L8) and seven periplasmic loops (T1–T7) that link the 16 β-strands to create the core of the β-barrel. Six of these flexible L loops, including L2–L7, are rather long, containing more than 20 amino acids. Surprisingly, our cryo-EM structure reveals that the N-terminal end of each OMP50-I protomer forms an elongated α-helix (α1) of 44 residues. Within the trimer, the three α1 helices intimately interact with each other through coiled-coil interactions to assemble an antenna-like three-helix bundle (Fig 3A–3E). The vertical dimension of this helix bundle is approximately 60 Å long, protruding into the periplasm from the periplasmic surface of the OM. Based on the structural information, this elongated periplasmic bundle may be critical for OMP50-I signaling from the outside to the inside of the bacterium. The trimeric structure of OMP50-I suggests that the linker region (residues 77–89) between α1 and β1, the transmembrane β-strands, β2-β4, and the flexible L loops, L2 and L4, create the trimeric interface.

As a tyrosine kinase, each protomer of OMP50-I possesses 27 tyrosine residues. In C. jejuni, it was detected that tyrosine phosphorylation only occurs at the OM [30]. Our structure indeed indicates that these tyrosines are located at the β barrels of the OM region of OMP50-I (S6 Fig). Most of these tyrosines are found within the β-strands, residing by the wall of the β barrel that forms a pore in the middle of each OMP50-I protomer. Therefore, many of them may be responsible for anchoring this kinase to the OM. However, some of these tyrosines should be critical in creating a network to facilitate phosphorylation signaling at the OM while the periplasmic α-helical bundle of the OMP50-I trimer may serve as a receiver to promote the transmission of signals to regulate the bacterium. Notably, residues Y147 of L2, Y204 of L3, Y360 of L6, and Y399 of L7 point toward the middle of the pore, where L3, L6 and L7 seemingly stretch into this pore to facilitate this structural arrangement (Fig 3C). Residues 349–358 of L6 creates a surface-exposed α-helix (α2) which may be important for host–pathogen interactions. In addition, L4′ (the ′ indicates structural elements belonging to the next protomer) is also observed to stretch into the center of the pore of this protomer, contributing to the formation of this pore (Fig 3B). Interestingly, the internal wall of the pore, formed by each OMP50-I protomer, is lined by Y147, Y204, Y360 and Y399, creating a path for substrate transport (Fig 3D). The entire path of the pore is very narrow with an average radius of <3 Å, where Y204 forms a constriction site and completely closes this pore. Therefore, these four tyrosines may be critically important for tyrosine phosphorylation as well as controlling the pore opening and closing for substrate transport. Indeed, it has been reported that a mutation of Y147 to a phenylalanine abolished this kinase’s function [30]. Like PorA, electrostatic surface potential calculations depict that the extracellular surface of the OMP50-I β-barrel is predominantly neutral, but residues at locations corresponding to both surfaces of the OM create positively and negatively charged patches to facilitate the anchoring of this membrane protein at the OM (Fig 3E). In addition, the internal pore formed by each β barrel of the OMP50-I protomer is mostly neutral in charge, quite distinct from the pore formed by the PorA protomer. Due to the neutrality of the inner pore combined with the arrangement of the four tyrosines (Y147, Y204, Y360 and Y399) lining the pore, one would expect OMP50 to be more favorable in shuttling neutral and/or positively charged ligands across the OM. Indeed, it has been documented that OMP50 is capable of forming a cation-selective pore with a major conductance of 50–60 ps [29].

As mentioned, the transmembrane region of trimeric OMP50-I forms three identical pores, similar to those found in PorA and other trimeric porins such as OmpF and OmpC [19]. However, a detailed inspection of the architecture of the three OMP50-I pores indicates that the trimeric organization of OMP50-I is very distinct from other porins. Residues 240–260 of each OMP50-I protomer form an elongated loop L4 (Fig 3B, top view). This random loop protrudes into the center of the β barrel of the next OMP50-I protomer, constituting part of the central pore of the next protomer and likely also participating in the opening and closing of this pore. Based on the structural information, the three pores of the OMP50-I trimer are presumed to function in a dependent manner, where they coordinate with each other to control substrate transport across the OM.

Structure of OMP50-II. We next investigated the second subset consisting of 8,831 single particles of this OMP50 class of images (S1 Fig). The constructed cryo-EM map allowed us to obtain a 2.81-Å resolution cryo-EM map (Figs 3F and S1). The map clearly indicated that these particles represent images of the OMP50 protein. However, the cryo-EM density corresponding to the N-terminal end of the periplasmic α1-helical bundle of the structure cannot be clearly seen. However, densities corresponding to these helices can be observed when lowering the contour level of the cryo-EM map (S7 Fig), indicating that the conformation of this helical bundle extending into the periplasmic space may be quite flexible in nature. We then solved its cryo-EM structure, designated as OMP50-II, to this resolution to obtain its detailed structural information.

Similar to the structure of full-length OMP50 (OMP50-I) indicated above, OMP50-II exists as a homotrimer, where each OMP50-II protomer within the trimer is identical in conformation. In comparison with OMP50-I and OMP50-II, the conformations of the OM domains of these two structures are nearly identical, where superimposition of the OM domains of these two OMP50 trimers results in an r.m.s.d. of 0.3 Å (S8A Fig). As indicated, it appears that the major difference between these two structures arises from the N-terminal α-helical coiled-coil domain coordinated by helix α1 in the periplasm (S8A Fig). In OMP50-I, a full-length trimeric antenna-like three-helix bundle made up of helix α1 (residues 36–79) from each OMP50-I protomer is established in the middle of the trimer. In OMP50-II, only the C-terminal portion of α1 (residues 63–79) forming 1/3 of the length of this trimeric antenna can be easily seen (Fig 3F), as the cryo-EM densities originating from residues 36–62 comprising 2/3 of the length of this antenna are difficult to trace from the cryo-EM map. It is possible that these residues are very flexible and likely unstructured. The flexibility of these structural elements at the N-terminal end of the antenna-like feature may be important for facilitating phosphotyrosine signaling. It should be noted that our structure of OMP50-II resembles the published structure of the Veillonella parvula OmpM trimer [47], where only 16 residues from each protomer of the periplasmic α-helical domain can be seen in the cryo-EM structures. Superimposition of the structures of trimeric C. jejuni OMP50-II and V. parvula OmpM (PDB ID: 8YBS) [47] gives rise to an r.m.s.d. of 7.7 Å (S8B Fig), indicating that the structures of these two proteins are very different from each other.

Cj0034c/CjSLP

In addition to PorA and OMP50, we identified a third class of images. We collected a total of 9,242 single particles and resolved a cryo-EM map of this class of images to a resolution of 3.13 Å (Figs 4A–4C and S1; S2 Table). The program DeepTracer [35] allowed us to reveal the identity of this membrane protein as Cj0034c, also called CjSLP, a known adhesion factor [48]. One of the key steps required to establish a successful C. jejuni infection is adhesion to the host cell. As Cj0034c appears to be a critical virulence factor, it has been identified as an attractive vaccine candidate [49]. Cj0034c has also been found to be immunogenic, and an antigen of Cj0034c has been observed to provide a significant protection against invasion in the spleen and liver [49].

Our cryo-EM structure allowed us to reveal for the first time the full-length architecture of Cj0034c and its assembly in native C. jejuni membranes. Nine protomers of Cj0034c stack side-by-side to form a doorknob-like apparatus (Fig 4A and 4B). Each protomer of Cj0034c consists of a transmembrane domain and a periplasmic domain (Figs 4C and S9). The N-terminal residues 6–34 form a single α-helical transmembrane helix (TM) that spans the entire thickness of the membrane. The majority of the amino acids, 36–230, are folded into a large periplasmic domain which can be further divided into two subdomains (D1 and D2) (Figs 4C and S9). Subdomain D1 is located right next to the TM and is composed of four β-sheets and one α-helix, whereas subdomain D2 constitutes five β-sheets and two α-helices. Different domains and subdomains are connected by flexible linkers, which also form hinges between them. Our cryo-EM structure of the periplasmic domain of each Cj0034c protomer is in good agreement with the X-ray structure of the monomeric soluble Cj0034c protein containing only D1 and D2 [48]. Superimposition of the periplasmic domain of our cryo-EM structure to the X-ray structure of Cj0034c (PDB ID: 7C51) [48] gives rise to an r.m.s.d. of 1.0 Å (S10 Fig).

It should be noted that Cj0034c was initially reported to be an OMP, but the N-terminal signaling peptide and outer membrane-spanning sequence were subsequently cleaved before protein maturation [21,50]. Therefore, the mature protein was presumed to become a soluble periplasmic protein, similar to the Brucella BP26/OMP28 homologous protein [48,49]. Surprisingly, our cryo-EM structure indicates that Cj0034c is an integral membrane protein possessing a single α-helical TM of approximately 50 Å long that spans the entire membrane thickness. The periplasmic domain of the nonameric protein creates a cage-like feature with dimensions of 75 Å long and 95 Å wide, making the entire oligomeric protein resemble a doorknob. The central core of the Cj0034c nonamer creates an open channel at both ends that directly connects both sides of the membrane of the bacterium (Fig 4D). The internal diameters of this channel at the two membrane surfaces are 20 Å and 25 Å, respectively (Fig 4B). Electrostatic surface potential calculations show that the internal wall of the nonameric Cj0034c channel is positively charged in nature (Fig 4E), suggesting that Cj0034c may be more favorable to bind negatively charged or neutral substrates.

In gram-negative bacteria, based upon all current reported structures, the IMP and OMP classes are distinct from each other. IMPs use α-helical structural features and OMPs utilize β-stranded secondary elements to constitute their TM domains. According to this structural information, the full-length Cj0034c membrane protein likely forms a typical IMP with a single-α-helical TM from each protomer to span the IM. However, recently the crystal structure of an atypical mycobacterial OMP MctB (Rv1698) has been reported [51]. The structure of this mycobacterial membrane protein shows an α-helical membrane-spanning region across the outer mycomembrane, where the N-terminal signaling peptide has dual functions. It guides the expressed protein to the outer mycomembrane as well as creates an α-helical TM domain [51]. Based on the structural information of MctB, it appears that Cj0034c could represent an unusual G. negative OMP possessing all α-helical secondary structural elements in the TM domain. If this is the case, then the N-terminal sequence of Cj0034c may also hold a dual purpose. It could lead the protein to target the OM as well as participate in bridging the OM with an all α-helical TM domain, similar to that found in MctB. Therefore, the structure of the Cj0034c nonamer may exemplify the first structure of a gram-negative OMP featuring with an α-helical TM domain. Further studies are necessary to determine the spatiotemporal nature of this membrane protein as well as its complete physiological functions.

Cell growth, membrane preparation, and cryo-EM sample preparation

C. jejuni NCTC 11,168 were grown in 3 l of Mueller-Hinton (MH) broth at 42 °C under microaerobic conditions (85% N₂, 10% CO₂, and 5% O₂) for 16 h with shaking at 120 rpm. Cells were harvested by centrifugation at 12,000g for 10 min. The collected bacteria were resuspended in buffer containing 20 mM Tris-HCl (pH 7.5), 50 mM magnesium acetate, 1 mM dithiothreitol (DTT) and 0.5 mM ethylenediaminetetraacetic acid (EDTA) then lysed using a French pressure cell. The membrane fraction was collected and washed two times with buffer containing 100 mM sodium phosphate (pH 7.2) and 10% glycerol, and then two times with buffer containing 20 mM HEPES-NaOH (pH 7.5) and 20 mM NaCl. The membrane was then solubilized in 1% (w/v) lauryl maltose neopentyl glycol (LMNG) overnight at 4 °C. Insoluble material was removed by ultracentrifugation at 186,000g for 40 min.

The membrane protein component of the solubilized membrane was enriched using sucrose cushion ultracentrifugation. Briefly, 1 ml of sucrose buffer A (60% sucrose, 20 mM Na-HEPES (pH 7.5), 20 mM NaCl and 0.001% LMNG) was added to the bottom of the centrifugation tube, and then 1 ml of sucrose buffer B (15% sucrose, 20 mM Na-HEPES (pH 7.5), 20 mM NaCl and 0.001% LMNG) was carefully added on top of the first layer. These sucrose concentrations were empirically optimized: the 60% cushion forms a dense barrier that retains membrane protein complexes above the tube bottom, while the 15% layer allows membrane proteins to migrate through but restricts entry of excess free detergent micelles. Next, a 2 ml detergent-solubilized sample was slowly added to the top of the sucrose layers and centrifuged at 100,000 g for 4 h at 4 °C. After centrifugation, the lower half of the 15% sucrose fraction, which is enriched in membrane protein complexes while depleted of free detergent, was carefully collected. It was then diluted 8-fold with buffer containing 20 mM HEPES-NaOH (pH 7.5), 20 mM NaCl, and 0.001% LMNG to reduce the sucrose concentration and facilitate subsequent pelleting of the enriched membrane proteins. The diluted fraction was then centrifuged at 100,000g for 16 h at 4°C. Subsequently, the protein pellet was dissolved with 10 µl of a buffer containing 20 mM HEPES-NaOH (pH 7.5), 20 mM NaCl and 0.001% LMNG. This protein solution was then centrifuged at 20,000g for 2 min to remove insoluble particles. The same buffer was used to adjust the protein concentration of the enriched sample to 10 mg/ml (estimated using a NanoDrop OneC instrument, Thermo Fisher Scientific).

For cryo-EM sample preparation, 2.5 μl of enriched membrane sample was directly applied to glow-discharged holey carbon grids (Quantifoil Cu R1.2/1.3, 300 mesh). The optimal blotting time was determined to be 12 s. After blotting, the sample was plunge-frozen in liquid ethane using a Vitrobot (Thermo Fisher Scientific). The resulting grids were then transferred into cartridges prior to data collection.

Data collection and data processing

Samples were collected at the Cleveland Center for Membrane & Structural Biology (CCMSB) using a Titan Krios cryo-electron transmission microscope (Thermo Fisher Scientific) operating at 300 kV, equipped with a BioQuantum energy filter (Gatan) with a slit width of 20 eV. A total of 19,448 movies were collected using SerialEM for automated data acquisition. Images were recorded at 81,000× magnification using a K3 direct electron detector (Gatan) in super-resolution mode, with a defocus range of 0.8–1.5 μm. Each movie consisted of 40 frames with a total electron dose ranging from 41 to 50 e⁻/Ų, corresponding to a super-resolution pixel size of 0.535 Å/pixel and a physical pixel size of 1.07 Å/pixel (S2 Table). Initial image stacks were binned by a factor of 2 and motion corrected using cryoSPARC v3 [60]. The patchCTF function was used to estimate the contrast transfer function [60]. The Topaz tool [61], default ResNet16 (64 U) pre-trained model, was used to pick initial particle sets from 19,448 movies, yielding 29,498,465 particles. The BaR protocol [33] was used to separate initial structure classes. During the “build” phase, the particles were sorted via multiple rounds of 2D classification to identify particles with distinct structural features. Particle stacks consisting of distinct 2D representations were each subjected to independent ab initio reconstruction, and 3 starting target models with structurally viable global features and 3 random unfeatured decoy models were then used as filters in a series of heterogeneous refinements using the original particle stack. At the “retrieve” phase, each target model retrieved additional particles relative to the initial stack, including a more diverse set of orientations. These refined particle stacks were then used to generate refined 3D models. These models were used as updated filters to retrieve additional particles belonging to each distinct class. This BaR cycle was repeated iteratively until no additional particles were incorporated and no further improvement in map quality was observed, at which point three distinct particle populations were identified: 309,511 particles, 35,032 particles and 26,588 particles. Subsequently, 3D classification was performed on each population independently to sort different conformations, non-uniform refinement and local refinement was used on the final 3D models to generate high-resolution density maps. Protein identification of the PorA, OMP50, and Cj0034c membrane proteins were done by a combination of manual building and utilizing the program DeepTracer [35]. The presence of these three membrane proteins in our sample was also confirmed by proteomics (S1 Table).

Model building and refinement

Based on their respective cryo-EM maps, model building of all proteins were performed using Coot [36]. Structural refinements were accomplished using the phenix.real_space_refine program [62] implemented in the PHENIX suite [37]. The final atomic models of these membrane proteins were evaluated using MolProbity [63]. The statistics associated with data collection, 3D reconstruction and model refinement are included in S2 Table.

Proteomic analysis

A 1× lysis buffer (5% SDS in 50 mM tiethylammonium bicarbonate (TEAB) buffer was added into 10 μg of the membrane sample to a final volume of 46 μl. The sample was digested with a trypsin solution (0.1 μg/μl in 50 mM TEAB) and transferred onto the S-trap spin column. A Bruker TimsTof Pro2 Q-Tof LC-MS system operating in positive ion mode coupled with a CaptiveSpray ion source (both from Bruker Daltonik GmbH, Bremen) and a Bruker 15 cm x 75 μm id C18 ReproSil AQ, 1.9 μm, 120 Å reversed-phase capillary chromatography column was used for peptide analysis. Five μl extract was injected, and the peptides were eluted from the column by an acetonitrile/0.1% formic acid gradient at a flow rate of 0.3 μl/min. The digests were analyzed using a Parallel Accumulation–Serial Fragmentation (PASEF) DDA method [64] to select precursor ions for fragmentation with a TIMS-MS scan followed by 10 PASEF MS/MS scans. The TIMS-MS survey scan was acquired between 0.60 and 1.6 Vs/cm² and 100–1,700 m/z with a ramp time of 166 ms. The total cycle time for the PASEF scans was 1.2 seconds and the MS/MS experiments were performed with collision energies between 20 eV (0.6 Vs.cm²) to 59 eV (1.6 Vs/cm²). Precursors with 2–5 charges were selected with the target value set to 20,000 a.u. and the intensity threshold set to 2,500 a.u. Precursors were dynamically excluded for 0.4 s.

The DDA data acquired by TimsTofs (Bruker) were converted into.mgf files using AlphaTims [65], an open-source package developed by Mann lab, before analyzing using the program Proteome Discoverer 2.4 (Thermo Fisher Scientific).

S1 Fig. Native C. jejuni membrane sample cryo-EM data processing workflow.

A total of 19,448 micrographs were collected and processed through motion correction to extract 29,498,465 particles. Following the build and retrieve (BaR) computational pipeline, including 2D classification, multi-class ab initio reconstruction, and heterogeneous refinement, particles were sorted into three major populations. Representative 2D class averages are shown for each population. Subsequent 3D classification and non-uniform refinement with local refinement yielded structures of multiple membrane proteins. Final reconstructions include: PorA-I at 2.47 Å resolution, PorA-II at 2.74 Å resolution, OMP50-I at 2.82 Å resolution, OMP50-II at 2.81 Å resolution, and Cj0034c at 3.13 Å resolution.

https://doi.org/10.1371/journal.pbio.3003961.s001

(TIF)

S2 Fig. Secondary structural topology of PorA.

The topology diagram was generated based on the cryo-EM structure of PorA-I. The outer membrane (OM) is colored yellow. The transmembrane β-strands are shown as green arrows. The α-helices located outside the membrane are colored magenta.

https://doi.org/10.1371/journal.pbio.3003961.s002

(TIF)

S3 Fig. Structural comparison of C. jejuni PorA-I with the published X-ray structure.

Superimposition of the PorA-I structure from this study (green) with the previously reported structure (PDB: 5LDT, purple). The inset shows a detailed view of the calcium binding site. Key residues involved in Ca²⁺ coordination are shown as sticks with residue numbers corresponding to this study. The Ca²⁺ ion is shown as a green sphere.

https://doi.org/10.1371/journal.pbio.3003961.s003

(TIF)

S4 Fig. Structural comparison of C. jejuni PorA-I and PorA-II.

Side and top views showing superimposition of the PorA-I (yellow) with PorA-II (magenta) structures. Loop 2 (L2) and loop 4 (L4) are indicated with black arrows. The Ca²⁺ion in the PorA-I structure is shown as a green sphere. L4 of PorA-II is missing in the structure (black dotted curve).

https://doi.org/10.1371/journal.pbio.3003961.s004

(TIF)

S5 Fig. Secondary structural topology of OMP50.

The topology diagram was generated based on the cryo-EM structure of OMP50. The outer membrane (OM) is colored yellow. The transmembrane β-strands are shown as green arrows. The α-helices and β-strands located outside the membrane are colored magenta and cyan, respectively.

https://doi.org/10.1371/journal.pbio.3003961.s005

(TIF)

S7 Fig. Cryo-EM maps of OMP50-II at different contour levels.

This figure includes the side view of the final structure of the OMP50-II trimer, and the side view of cryo-EM maps of OMP50-II contoured at 4.6 σ, 3.3 σ and 2.0 σ levels. These maps indicate that the N-terminal residues 21–62 are likely to be flexible.

https://doi.org/10.1371/journal.pbio.3003961.s007

(TIF)

S8 Fig. Structural comparison of C. jejuni PorA-I, C. jejuni PorA-II and V. parvula OmpM.

(A) Side view of the superimposition of the structures of OMP50-I (cyan) and OMP50-II (yellow). This superimposition gives rise to an r.m.s.d. of 0.3 Å. (B) Side view of the superimposition of the structures of OMP50-II (green) and V. parvula OmpM (red). This superimposition gives rise to an r.m.s.d. of 7.7 Å, suggesting that the structures of C. jejuni OMP50-II and V. parvula OmpM are very different from each other.

https://doi.org/10.1371/journal.pbio.3003961.s008

(TIF)

S9 Fig. Secondary structural topology of Cj0034c.

The topology diagram was generated based on the cryo-EM structure of Cj0034c. The outer membrane (OM) is colored yellow. The transmembrane α-helix (TM) is colored blue. The secondary structural elements of the periplasmic subdomains D1 and D2 are colored red and purple.

https://doi.org/10.1371/journal.pbio.3003961.s009

(TIF)

S10 Fig. Structural comparison of the periplasmic domain of C. jejuni Cj0034c with the published X-ray structure containing only the periplasmic domain.

Side view showing superimposition of the Cj0034c structure (blue) with a published C. jejuni Simpl-like protein structure (PDB ID: 7C51) (white).

https://doi.org/10.1371/journal.pbio.3003961.s010

(TIF)