key: cord-0276370-bobyse2w
authors: Oosterheert, Wout; Xenaki, Katerina T.; Neviani, Viviana; Pos, Wouter; Doulkeridou, Sofia; Manshande, Jip; Pearce, Nicholas M.; Kroon-Batenburg, Loes M. J.; Lutz, Martin; van Bergen en Henegouwen, Paul M. P.; Gros, Piet
title: Implications for tetraspanin-enriched microdomain assembly based on structures of CD9 with EWI-F
date: 2020-06-03
journal: bioRxiv
DOI: 10.1101/2020.06.02.130047
sha: 55e629e14908343aefb1479380a8dce6f07fbd84
doc_id: 276370
cord_uid: bobyse2w

Tetraspanins are ubiquitous eukaryotic membrane proteins that contribute to a variety of signaling processes by spatially organizing partner-receptor molecules in the plasma membrane. How tetraspanins bind and cluster partner receptors into so-called tetraspanin-enriched microdomains is unknown. Here we present crystal structures of the large extracellular loop of CD9 in complex with nanobodies 4C8 and 4E8; and, the cryo-EM structure of 4C8-bound CD9 in complex with its prototypical partner EWI-F. The CD9 - EWI-F complex displays a tetrameric arrangement with two centrally positioned EWI-F molecules, dimerized through their ectodomains, and two CD9 molecules, one bound to each EWI-F single-pass transmembrane helix through CD9-helices h3 and h4. In the crystal structures, nanobodies 4C8 and 4E8 bind CD9 at the C and D loop, in agreement with 4C8 binding at the ends of the CD9 - EWI-F cryo-EM complex. Overall, the 4C8 - CD9 - EWI-F - EWI-F - CD9 - 4C8 complexes varied from nearly two-fold symmetric (i.e. with the two CD9 - 4C8 copies in nearly anti-parallel orientation) to ca. 50° bent arrangements. Since membrane helices h1 and h2 and the EC2 D-loop have been previously identified as sites for tetraspanin homo-dimerization, the observed linear but flexible arrangement of CD9 - EWI-F with potential CD9 - CD9 homo-dimerization at either end provides a new ‘concatenation model’ for forming short linear or circular assemblies, which may explain the occurrence of tetraspanin-enriched microdomains.

The spatial organization of proteins and lipids in the plasma membrane controls cellular processes such as signaling, trafficking, cell adhesion and fusion (1) . Members of the tetraspanin superfamily of transmembrane proteins act as molecular organizers of the plasma membrane by clustering specific partner proteins in cis to arrange their assembly required for function (2) (3) (4) (5) . Tetraspanins form dynamic platforms termed tetraspanin-enriched microdomains (TEMs), also referred to as the tetraspanin web (3, 4, 6) . TEMs were formerly described as large assemblies comprising multiple tetraspanin family members and partner proteins (6) (7) (8) , but the most recent model derived from super-resolution microscopy experiments states that TEMs are nanoclusters of approximately 120 nm in diameter that contain fewer than ten copies of a single tetraspanin homolog (9) .

Among the 33 human tetraspanins, CD9 (also known as Tspan29) is an extensively studied tetraspanin that is expressed in a wide variety of tissues and is also highly abundant on the membranes of extracellular vesicles (10) . CD9 participates in processes like cell adhesion, motility and differentiation (4, 11, 12) . Additionally, it regulates cell-cell fusion events in myoblasts (13, 14) , osteoclasts (15) , macrophages (16) and sperm-egg cells (17, 18) . Besides its physiological functions, CD9 is widely associated with diseases: both the upregulation and downregulation of CD9 are linked with poor prognosis in several types of cancer, such as melanoma, leukemia and gastric, lung, breast, colon and prostate malignancies (19) . Moreover, CD9 modulates HIV-1-induced membrane fusion (20) and promotes MERS-coronavirus entry by clustering host-cell receptors and proteases (21, 22) .

At the molecular level, tetraspanins adopt a common architecture with four transmembrane helices (h1 -h4), intracellular N-and C-termini, a small extracellular loop between membrane helices h1 and h2 (EC1) and a large extracellular loop between membrane helices h3 and h4 (EC2). Initial structural studies on the soluble EC2 domain of CD81 disclosed an EC2-fold with five helical regions (termed A -E), with A, B and E forming a 'stalk' domain and C and D forming a 'head' domain (23) . Crystal structures of full-length tetraspanin CD81 (24) and, recently, CD9 with a truncated EC2 (25) , revealed an inverted-cone shape arrangement of the transmembrane domain (TMD), with the EC2 closing off the TMD-cone like a lid, although molecular dynamics simulations suggested that the EC2 lid can also adopt open conformations (24, 25) . Previous studies have established that both the EC2 and TMD regions of tetraspanins participate in homo-and hetero-oligomeric interactions, indicating that tetraspanins employ separate domain regions to bind to different partners proteins (5) . CD9 interacts with numerous single-span transmembrane proteins, including integrins (26) (27) (28) (29) , immunoglobin superfamily (IgSF) proteins (30) (31) (32) , heparin-binding EGF-like growth factor (33) and metalloprotease ADAM17 (11, 34) . IgSF protein EWI-F, also known as CD9-partner 1 or prostaglandin F2 receptor negative regulator, is a prototypical primary CD9-binding protein that also interacts with its close-homolog CD81 in stoichiometric amounts in several cell types (30, 32) . EWI-F comprises six extracellular, heavily glycosylated Ig-like C2 domains (35) , a single transmembrane (TM) helix and a small C-terminal cytoplasmic tail. Functionally, EWI-F moderates CD9 and CD81-driven cellular fusion events (14, 36) and connects TEMs to the cytoskeleton by binding to ezrin-radixin-moesin proteins (37) . Biochemical experiments have revealed that the interaction between EWI-F and CD9 or CD81 is mediated by the TMDs, with critical roles for the transmembrane helix of EWI-F and helix h4 of CD9 or CD81 (36, 38) .

Additionally, we recently found that palmitoylated membrane-proximal cysteines of CD9 are crucial for maintaining the interaction with EWI-F (39) .

Although tetraspanins have a central role in human physiology, the mechanisms through which they interact with partner proteins and the molecular principles that govern the formation and signaling of TEMs remain incompletely understood. Here, we study the structural basis for interactions in the tetraspanin web using the CD9 -EWI-F complex as a model system. We first present crystal structures of the isolated EC2 domain of CD9 in the absence and presence of anti-CD9 nanobodies 4C8 and 4E8, revealing the flexibility of the D-loop region of CD9.

We then employ nanobody 4C8 in single-particle cryo-EM studies to show that CD9 and EWI-F form a hetero-tetramer that adopts a range of conformations, with a central EWI-F dimer flanked by two CD9 molecules on each side. The structural data are in good agreement with the recently reported cryo-EM map of CD9 bound to EWI-F homolog EWI-2 (25), indicating that CD9 binds to EWI proteins through a common binding mode. Overall, the various conformations of the CD9 -EWI-F complex observed in the cryo-EM data, combined with previous biochemical interaction studies, support a 'concatenation model' for the assembly of TEMs in the plasma membrane.

Our group has previously reported the generation of several nanobodies that bind to the extracellular region of CD9 (39) . We selected nanobodies 4C8 and 4E8, which differ in aminoacid composition in complementarity-determining regions (CDRs) 2 and 3, for further characterization. 4C8 and 4E8 bind to purified, full-length CD9 with affinities of 0.9 nM and 2.0 nM, respectively, and to endogenous CD9 expressed on HeLa cells with affinities of 0.6 nM and 33 nM, respectively (Fig. S1A, B) . To determine the nanobody-binding epitopes on CD9, we solved crystal structures of the soluble EC2 of CD9 (CD9EC2) in complex with 4C8 at 2.7-Å resolution (Fig. 1A) and 4E8 at 1.4-Å resolution (Fig. 1B) , as well as the structure of 4C8 alone at 1.7-Å resolution (Fig. S2A , Table 1 ). The CD9EC2 -4C8 complex crystallized in space group I212121 with two copies in the asymmetric unit (Fig. S2B) , whereas CD9EC2 -4E8 crystallized in space group P21212 with a single copy of the complex in the asymmetric unit ( Fig. 1B) .

In both nanobody-bound structures, CD9EC2 adopts an arrangement that is globally similar to previously reported EC2-structures of CD81 (23, 24, (40) (41) (42) (43) . 4C8 and 4E8 bind nearly identical epitopes that span the C and D loops of CD9EC2 (Fig. 1A, B) . The interfaces are established through nanobody residues located in CDRs 2 and 3, whereas the CDR1 region of both nanobodies makes no contacts with CD9EC2. The surface areas buried in the CD9EC2 -4C8 and 4E8 complexes are ca. 1,280-Å 2 and 1,230-Å 2 , respectively. CD9 residues involved in the interfaces with the nanobodies are V159, E160, Q161 and S164 from the C loop; and K169, L173, T175, F176 and V178 from the D loop (Fig. 2C, D) . Both nanobodies feature an arginine residue (R101) in CDR3 that interacts with E160 through a salt bridge, and two tryptophans (W53 of CDR2 and W102 of CDR3 in both 4C8 and 4E8) that form a hydrophobic core in the interface with CD9EC2 (Fig. 1C, D) . In CD9EC2 -4C8, the D loop adopts a partially helical conformation and central residue F176 is sandwiched by 4E8 residues W59 of CDR2 and W102 and R105 of CDR3 (Fig. 1D ). In the 4C8-bound CD9EC2 structure the tip of the D loop points more outward and the Cα atom of F176 is shifted by 3 Å, such that F176 resides not at the center of the interface but at the periphery, where it forms a van der Waals interaction with 4C8-residue G105 (Fig. 1C) .

A superimposition of the nanobody-bound CD9EC2 structures onto the crystal structure of fulllength CD81 (pdb 5TCX) indicates that 4C8 and 4E8 orient away from the membrane plane in the EC2 conformation adopted by CD81 (Fig. 1E) , which is consistent with both nanobodies binding to cell-membrane embedded CD9 (Fig. S1A ).

The EC2-D loop displays a high amino-acid sequence variability among tetraspanins and has been proposed to mediate homo-and hetero-oligomeric interactions (5, (44) (45) (46) . Previously reported structures of CD81 revealed the conformational plasticity of its EC2 D-loop, with both fully helical and partially unfolded arrangements (23, 24, (40) (41) (42) (43) . Our nanobody-bound CD9EC2 structures also display a minor conformational difference in D-loop with respect to each other (Fig. 1C, D) . To further investigate the conformational changes adopted by the Dloop of CD9 upon nanobody binding, we determined the crystal structure of CD9EC2 in the absence of nanobodies at 2.0-Å resolution ( Table 1 ). The diffraction data indicated nonmerohedral twinning of the crystal with a twofold rotation around the a*+b* diagonal as the twinning operation (Fig. S3) . Moreover, we observed diffuse streaks in the a*+b* direction ( Fig. S3A ). Figure S3B shows the two twin domains in the crystal with the twinning interface in the middle. For structure refinement the data was de-twinned based on the calculated structure factors (see Experimental procedures).

CD9EC2 crystallized in space group P1 with four molecules in the asymmetric unit, arranged as a dimer of domain-swapped dimers ( Fig. 2A) . In this domain swap, the N-terminal A helix (residues 114-138) is exchanged with the A helix of its dimeric partner, resulting in an extensive interface that buries ca. 3,400-Å 2 of surface area. The two connected protein cores each strongly resemble the monomeric form of CD9EC2, which is commonly observed for structures of domain-swapped dimers (47, 48) . The domain-swapped dimers are packed in the crystal lattice through interactions between the D-loop regions, which are extended and fully unfolded ( Fig. 2A) . A comparison between all CD9EC2 molecules in the asymmetric unit indicated that the D-loop adopts two distinct conformations, both different from the nanobodybound CD9EC2 structures (Fig. 2B ). This observation suggests that the binding of nanobodies 4C8 and 4E8 to CD9EC2 changes the D-loop conformation. This may, at least in part, explain the difference in signal intensity between 4C8 and 4E8 binding to purified CD9 and to CD9 on cells (Fig. S1A, B) , assuming that the particular conformational change in the D-loop of CD9 is less efficient in the case of 4E8.

An overlay of the C-and D-loop regions of CD9EC2 structures in the absence and presence of nanobodies and full-length CD81 (pdb 5TCX) revealed a diverse set of D-loop conformations: the D-loop of CD9EC2 is extended ~11 Å compared to the compact helical arrangement observed in the structure of full-length CD81 (Fig. 2B ). In the twinned CD9EC2 structure, one of the D-loops has large B-factors (Fig. S3C) , indicating its flexibility, and it is involved in the twinning interface, apparently accommodating the deviation of true 2-fold symmetry of the two independent molecules. Additionally, the C-loops of the CD9EC2 structures adopt a more loop-like conformation compared to the helical C-loops present in CD81 structures (Fig. 2B) .

We next focused on obtaining a structural model of full-length CD9 in complex with its primary partner EWI-F. To this end, we co-expressed Strep-GFP-tagged CD9 together with EWI-F in HEK293 cells and isolated the complex in N-dodecyl-ß-D-maltoside (DDM) detergent using Strep-affinity chromatography and size-exclusion chromatography (SEC) purification steps ( Fig. 3A, B) . The complex did not dissociate during subsequent analytical fluorescence assisted SEC experiments (Fig. 3C) , indicating that the interaction between CD9 and EWI-F is preserved outside the native cell-membrane environment. We then collected a preliminary single-particle cryo-EM dataset of the purified complex. The resulting micrographs showed elongated, highly heterogeneous protein particles (Fig. 3D ). This heterogeneity is likely caused by flexibility between the six Ig-like C2 domains of EWI-F, indicating that the imaged CD9 -EWI-F complex in its full-length version is unsuitable for high-resolution structure determination. 2D-classification experiments yielded low-resolution class averages with no distinguishable micelle region present. Nevertheless, a single 2D-class average showed two rows of five domain-like densities stacked together (Fig. 3E , boxed in red), presumably corresponding to five of the six Ig-like domains of EWI-F. This suggests that EWI-F forms dimeric assemblies, which is consistent with a previous cross-linking study (38) .

Because full-length EWI-F displayed severe inter-domain flexibility, we next aimed to create a structurally more homogeneous EWI-F protein sample. For this purpose, we designed and generated N-terminally truncated EWI-F constructs with up to five Ig-like domains removed, which we termed EWI-FΔIg1, EWI-FΔIg1-2, EWI-FΔIg1-3, EWI-FΔIg1-4 and EWI-FΔIg1-5. To assess if the truncated EWI-F constructs still associated with CD9, we expressed the EWI-F variants together with Strep-GFP-tagged CD9 in small scale HEK293-cell cultures and, after Strep-affinity purification, monitored the amount of co-purified EWI-F using SDS-PAGE. The SDS-PAGE gel revealed the presence of protein bands at expected molecular weights for all EWI-F variants except for EWI-FΔIg1-4 ( Fig. 3F) , suggesting that EWI-FΔIg1-4 was possibly not expressed. Nevertheless, the co-purification of EWI-FΔIg1-5 with CD9 indicates that the first five Ig-like domains of EWI-F are not essential for maintaining the CD9 -EWI-F interaction. This is in agreement with previous studies that identified the transmembrane helix of EWI-F as the main CD9 and CD81-interacting region (36, 38) . Based on these co-purification experiments, we chose to perform our subsequent cryo-EM experiments with construct EWI-FΔIg1-5, which harbors only a single Ig-like domain (Ig6) and thus presumably has the least conformational freedom of all tested EWI-F constructs.

We next expressed Strep-tagged, full-length CD9 (without GFP) together with EWI-FΔIg1-5 and purified the complex in digitonin in sufficient quantities for cryo-EM experiments. An EWI-FΔIg1-5 dimer in complex with one or two CD9 molecules comprises <100 kDa of protein mass, making it a challenging sample for cryo-EM. We therefore investigated whether nanobody 4C8 could be employed to create a larger protein particle with more extra-membrane (or extramicelle) features. To determine if nanobody 4C8 could still bind CD9 complexed to EWI-FΔIg1-5, we incubated the complex with an excess of 4C8 and subjected it to a SEC-purification step. The resulting SEC-elution profile showed a monodisperse peak for the CD9 -EWI-FΔIg1-5 complex (Fig. 4A) ; SDS-PAGE analysis of the peak then confirmed the presence of 4C8 (Fig.   4B ), indicating that the epitope recognized by 4C8 is still accessible and that 4C8 does not disrupt the association between EWI-FΔIg1-5 and CD9. The EWI-FΔIg1-5 -CD9 -4C8 complex was monodisperse in size as assessed by Trp-fluorescence assisted SEC (Fig. 4C) , and thus suitable for analysis by single-particle cryo-EM. We collected a cryo-EM dataset and observed homogeneously sized-protein particles distributed in vitreous ice (Fig 4D) , which allowed more accurate particle picking than for the previous dataset with full length EWI-F (Fig. 3D ). 2Dclass averages of the picked particles revealed expected density features, with protein domains protruding from a disk-shaped micelle region (Fig. 4E ). Image processing in Relion yielded reconstructed density maps at a maximum global resolution of ~8.6 Å ( Fig. 5A -C, Fig. S4 , Table 2 ), although protein regions of the map exhibit a higher local resolution (Fig. S4B) .

The ~8.6-Å resolution cryo-EM map revealed a hetero-tetrameric arrangement of CD9 -EWI-FΔIg1-5, with a central EWI-FΔIg1-5 dimer flanked by 4C8-bound CD9 on each side ( Fig. 5A -C) .

Although the resolution of the reconstruction did not allow for de novo modelling, density regions corresponding to protein domains of 4C8-bound CD9 and EWI-F could be distinguished, and rigid-body fits of structures of CD9EC2 -4C8 (Fig. 1A) , the TMD of CD9 (pdb 6K4J) and homology models for the EWI-F Ig6 (chain b of pdb 1F3R) and TM-helix (pdb 5EH4) were in good agreement with the density map ( Fig. 5A -C) . Consistent with a TMD interface between CD9 and EWI-FΔIg1-5 (36, 38) , we observed that the EWI-F TM-helix resides in close proximity to CD9-helix h3 at the membrane center and to CD9-helix h4 at the cytoplasmic side of the membrane (Fig. 5D ). The two TMD regions of the complex, each comprising the four helices of CD9 and the TM helix of EWI-F, are separated and do not interact, but instead are bridged by the extracellular dimeric Ig6 domain of the two EWI-F molecules. Overall, CD9 and EWI-F share no extensive interface through their extra-membrane domains, with no major contact areas between the EC2 domains of CD9 and EWI-F-Ig6. The 4C8-bound C and D loops of the EC2 orient away from EWI-F. Compared to the structure of CD81, the EC2 is rotated upwards by ~40º, resulting in an open conformation of CD9 when in complex with EWI-F (Fig. 5E ).

Recently, a cryo-EM density map has been reported of CD9 in complex with an EWI-F homolog, EWI-2. Both the CD9 -EWI-2 map and our CD9 -EWI-F map were solved at moderate resolutions, limiting a detailed comparison of both reconstructions. However, both reconstructions reveal a similar hetero-tetrameric complex, with a central EWI-protein dimer and two CD9 molecules on each side (Fig. S5 ). Thus, CD9 interacts with its major partners EWI-F and EWI-2 in a comparable protein arrangement.

During the EM-image processing, it became apparent that the imaged EWI-FΔIg1-5 -CD9 -4C8 complex adopts a range of conformations. A typical 3D classification experiment with four major classes is shown in Fig. 6 . The 4C8 -CD9 -EWI-FΔIg1-5 -EWI-FΔIg1-5 -CD9 -4C8 composition suggests that the complex might arrange as a C2-symmetric particle. However, we observed no two-fold rotation axis in any of the four classes shown in Fig. 6 , nor in any further sub-classifications. Class #1 resembles the orientation of the map presented in Fig. 5A and shows a pseudo two-fold arrangement, with a 9º deviation in anti-parallel orientation of the two 4C8-bound CD9 copies in the complex. In contrast, classes #2 -#4 show a much larger deviation from an anti-parallel CD9 configuration, i.e. a deviation of 49º is observed in class #4. A morph of 4C8-bound CD9 models fitted in the different classes reveals the putative conformational changes of CD9 in the complex (Video S1).

Tetraspanin-mediated interactions that drive the clustering of single-pass membrane receptors into TEMs in plasma membranes have remained obscure. Structural insights into tetraspanins have so far encompassed the large extracellular EC2 loop of CD81 (23, (40) (41) (42) (43) , full-length CD81 (24) and, more recently, full-length CD9 and a low-resolution cryo-EM map of CD9 in complex with EWI-2 (25).

The three crystal structures, presented here, reveal an overall architecture of CD9EC2 similar to that of the homologous CD81EC2, except for a more extended arrangement of the C-loop and, in particular, D-loop regions. In structures of CD81, these regions adopt more helical conformations, though partially unfolded arrangements are also observed in antibody-bound CD81EC2 structures (43) . The D-loop conformations in CD9EC2 range from a partial helical arrangement in the nanobody-bound structures to an extended loop with no secondary structure elements in the domain-swapped CD9EC2 structure without nanobody (Fig. 2B) . The homodimerizing D-loops in between two domain-swapped CD9EC2 dimers ( Fig. 2A) could explain the proposed homo-dimerization interface in the homologous CD81, as mediated by D-loop residues P176 -F186 (44-46) (which correspond to P168 -T177 in CD9). Overall, the ability of the D-loop, which is sequence variable among tetraspanins, to adopt multiple conformations is preserved in the structures of CD9EC2 and CD81EC2. This observation supports the hypothesis that the EC2 head region can be tailored to facilitate interactions with specific partner proteins (49) . In the full-length crystal structure of CD81, molecules pack in an anti-parallel fashion through hydrophobic interaction of their TM regions, resulting in a non-physiological up-down arrangement (24) . Although the molecules in the crystal structure of full-length CD9 (25) align in a more physiologically relevant manner, potential homo-dimerizing contacts between EC2 fragments were abrogated due to the deletion of D-loop residues T175 -K179 (50) . When taken together, the variable D-loop conformations observed in crystal structures are consistent with the representation of the D-loop as an 'interaction hub' for homoand heterooligomerization in the head region of tetraspanins, as proposed earlier (5).

The cryo-EM map of CD9 in complex with EWI-FΔIg1-5 and nanobody 4C8 revealed a core CD9 -EWI-F -EWI-F -CD9 hetero-tetramer arranged in a linear fashion. In this complex, nanobodies 4C8 bind to the CD9 on either side without contacting the EWI-F molecules. Both CD9EC2 segments are tilted by ~40º with respect to their TMD region (Fig. 5E) , yielding an open conformation that is comparable to those predicted by MD simulation of CD81 and CD9 starting from the closed conformations observed in the crystal structures (24, 25) . CD9

predominantly contacts EWI-F through its TMD region, where CD9 helices h3 and h4 interact with the single-pass TM-helix of EWI-F (Fig. 5D) , which is in agreement with prior biochemical data that identified membrane helix h4 as a critical interface for binding EWI-F (36, 38) . The TMD regions of both CD9 -EWI-F hetero-dimers are separated by more than 10-Å distance and make no or minimal direct contacts. Instead, dimerization of the two heterodimers into a tetramer is dominated by contacts between the extracellular Ig6 domains of EWI-FΔIg1-5. Consistent with the absence of an intra-membrane interface between the two CD9 -EWI-F heterodimers and no major extra-membrane contacts between CD9 and EWI-F (Fig. 5) , the overall arrangement is highly flexible and we observed a range of linear-arranged tetrameric complexes, ranging from an extended conformation to one that is bent by ~50º (Fig. 6 , Video S1).

Most recently, Umeda et al. reported a cryo-EM map of CD9 in complex with EWI-2 (25).

Overall, the structural arrangements of CD9 -EWI-FΔIg1-5 and CD9 -EWI-2 complexes are similar (Fig. S5 ), in line with the partially overlapping cellular functions of EWI-F and EWI-2 (31, 37) . As with CD9 -EWI-F, major contacts are observed between h3 and h4 of CD9 and the single-pass TM helix of EWI-2. However, the membrane helices of EWI-2 are in closer proximity to each other than those of EWI-F, which could be correlated with the presence of two more putative palmitoylation sites in EWI-F than EWI-2. Furthermore, for CD9 -EWI-2, only one conformation was reported, whereas the distinct shape of the CD9EC2 -4C8 density enabled us to distinguish a range of straight to bent conformations for CD9 -EWI-FΔIg1-5 (Fig.   6 , Video S1). However, given the limited resolution of the cryo-EM maps of both cases, flexibility appears to be an inherent feature of the two complexes and a similar bending of CD9 -EWI-2 cannot be excluded.

CD9, CD81 and other tetraspanins interact with various single-pass TM proteins that are known to homodimerize through their extra-membrane domains. Examples of homodimerizing partners of tetraspanins (besides EWI-F and EWI-2) include: DPP4 and ADAM17 for CD9; EGFR for CD82 (51) ; and hetero-dimeric integrins for CD151 (52) (53) (54) . The observed lineararranged tetramers of CD9 -EWI-F and CD9 -EWI-2 indicate that the interactions that govern complex formation are sequentially hydrophobic-hydrophilic-hydrophobic, with tetraspanin partner-protein interactions mediated by the TMDs on either side and protein-partner homodimerization via their ectodomains in the center. Previous studies mapped a tetraspanin homodimerization site (for CD9) to membrane helices h1 and h2 (55) and, for CD81, to the variable D-loop in the EC2 head domain. Both helices h1 and h2 and the D-loops are positioned on either end of the linear-arranged tetramers and orient outwards posed for putative interactions (Fig. 5A) . Thus, the observed hetero-tetrameric arrangements (Fig. 6) suggest a simple 'concatenation model' for forming transient higher-order assemblies of end-to-end attached tetramers yielding small linear or circular structure (Fig. 7) , which may explain the occurrence of TEMs and their highly dynamic nature. This single-particle cryo-EM-derived model is in agreement with scanning-EM data on immunogold-labeled CD81 and CD9 expressed on cells, which showed both tightly packed clusters and linear tetraspanin assemblies (56, 57) . Thus, the concatenation model provides a structural basis for further studying the formation and signaling of TEMs in the plasma membrane.

All chemicals were purchased from Sigma-Aldrich unless specified otherwise.

The cDNA encoding for CD9 and EWI-F was obtained as described previously (39) . Briefly, codon-optimized cDNA for mammalian cell expression, encoding for human CD9 was purchased from Geneart. Full-length CD9 was cloned in a pUPE expression vector (U-Protein Express BV) with either a N-terminal Strep-II tag or a N-terminal Strep3-GFP tag with a TEVprotease site. The CD9-W6 construct (CD9-C9W,C78W,C79W,C87W,C218W,C219W), with all palmitoylated cysteines mutated to tryptophan, was cloned as described (39) . The construct encoding for the EC2 domain of CD9 (CD9EC2, residues K114 -N191) was generated with a PCR reaction using the Q5-PCR kit (NEB). The CD9EC2 construct was cloned in a pUPE expression vector with a N-terminal cystatin signal peptide and a C-terminal 6x-His tag. cDNA encoding for full-length human EWI-F was obtained from Source BioScience. EWI-F truncations were generated using PCR. Construct boundaries of EWI-FΔIg1 (H146 -D879), EWI-FΔIg1-2 (Q275 -D879), EWI-FΔIg1-3 (E405 -D879), EWI-FΔIg1-4 (N540 -D879) and EWI-FΔIg1-5 (T685 -D879) were based on Uniprot domain assignments (UniprotKB -Q9P2B2). All EWI-F constructs were cloned in a pUPE expression vector with a N-terminal cystatin signal peptide and a C-terminal 6x-His tag.

Phage library construction and anti-CD9 nanobody selections were carried out as described before and resulted in the selection of two nanobodies: 4C8 and 4E8 (39) . DNA sequences of clones 4C8 and 4E8 were ligated into a modified pHEN6 vector with a C-terminal thrombincleavable 6xHis tag and an N-terminal pelB leader sequence for periplasmic secretion (gift from dr. F. Opazo, University of Göttingen Medical Center, Göttingen). Large scale nanobody productions were performed overnight at 25°C using an Bioflo 115 Bioreactor (New Brunswick) as previously described (37) . Bacterial biomass was harvested by centrifugation at 5,400g and His-tagged nanobodies were purified using HisTrap™ column chromatography (GE Healthcare) followed by a buffer exchange step to PBS using a HiTrap™ desalting column (GE Healthcare). Nanobodies were stored in PBS at -20 °C until further use. Nanobody 4C8 was treated with Thrombin (0.5U/100μg; GE Healthcare) for 16 hrs at 22°C, to remove the 6xHis purification tag. 4C8 was then injected into a Superdex75 10/300 column (GE Healthcare) pre-equilibrated in buffer A (25 mM HEPES pH 7.5, 150 mM NaCl), and monodisperse fractions were collected. The tag of nanobody 4E8 was not removed prior to crystallization.

The binding affinity of 4C8 and 4E8 on CD9 was determined both on purified CD9 and on For cell binding assays, cells were fixed with 4% formaldehyde prior to the incubation with the detecting antibodies. Finally, IRDye800CW fluorescent signal at 800 nm was detected with Odyssey® Infrared Imager (LI-COR Biosciences). The measured fluorescent intensities were normalized, setting as 100% the mean of intensities detected at the higher nanobody concentrations, and plotted (mean ± SEM) over protein concentration using GraphPad Prism 7. A non-linear regression curve for one-site specific binding was fitted in order to determine apparent binding affinity (KD) of the different nanobodies.

CD9EC2 was recombinantly expressed in HEK293-EBNA cells (provided by U-Protein Express BV). Cells were grown at 37°C and harvested after six days. The cells were centrifuged at 1,000g for ten minutes, after which the medium was collected. Two different purification strategies were then employed: 1) For the protein sample used for the crystallization of CD9EC2 in the absence and presence of nanobody 4C8, the medium was loaded overnight onto a ~5 ml anti-CD9 AR40A746. 

The triclinic CD9EC2 crystal was twinned with a twofold rotation about a*+b* as the twinning operation. Consequently, two orientation matrices were used for integration with the Eval15 software (59). In the prediction of reflection profiles an isotropic mosaicity of 1.1° and a mica expansion of 0.077 along a*+b* was assumed. The resulting reflection file contained 21.5% overlapping reflections belonging to both twin domains. Initial de-twinning was performed with the TWINABS software (60) . These data were used for structure solution by molecular replacement, using PHASER (61) within the CCP4 software suite (62), using the model of CD9EC2 from the previously solved AT1412dm Fab-CD9EC2 complex (58) as model. The structure was iteratively refined using Refmac5 (63) alternated with model improvement in COOT (64) . Local non-crystallographic symmetry (NCS) restraints were maintained during refinement. The calculated structure factors from this refinement were then used for the final de-twinning of the overlapping data. The scale factor of 6.2147 * exp (-8.19544 *sin(θ/) 2 ) between the two twin domains was determined with XPREP (65), based on the nonoverlapping reflections. Interframe scaling of the de-twinned data was performed with SADABS and merging was performed with the CCP4 suite. Final refinement rounds in Refmac5 using the latest data yielded Rwork/Rfree = 23.9/27.9% and the structure was deposited in the RCSB Protein Data Bank under accession code 6RLR.

Diffraction images were processed using DIALS (66) and the integrated reflection data were truncated anisotropically using the STARANISO web server (67) . The structure was solved by molecular replacement using PHASER with the CD9EC2 structure and a nanobody homology model obtained through the SWISS-MODEL server (68) as search models. The structure was iteratively refined using Refmac5 or Phenix (69) alternated with manual model improvement in COOT. The final refinement in Phenix yielded Rwork/Rfree = 20.6/24.8% and the structure was deposited in the RCSB Protein Data Bank under accession code 6Z20.

Diffraction images were processed using Eval15 (59). The structure was solved by molecular replacement using PHASER with the nanobody in chain B of PDB 5IMK as the search model.

The nanobody residues were manually adjusted to the 4C8 amino-acid sequence in Coot. The structure was then iteratively refined using Refmac5 or Phenix, alternated with manual model improvement in COOT. The final refinement in Phenix yielded Rwork/Rfree = 20.1/23.1% and the structure was deposited in the RCSB Protein Data Bank under accession code 6Z1Z.

For the CD9EC2 -4E8 dataset, the autoprocessed and anisotropical-truncated (autoprocstaraniso) reflection data file provided by DLS was used. The structure was solved by molecular replacement using PHASER with the CD9EC2 -4C8 structure as search model. The 4C8 residues were replaced with the corresponding 4E8 residues and the CDR regions of the nanobody were manually built in Coot. The structure was then iteratively refined using 

N-Strep3-GFP-tagged CD9-W6 was co-transfected with EWI-F, EWI-FΔIg1, EWI-FΔIg1-2, EWI-FΔIg1-3, EWI-FΔIg1-4 or EWI-FΔIg1-5 in 25 ml HEK293-EBNA cell cultures (provided by U-Protein Express BV). Cells were grown at 37°C and harvested after four days. Cells were washed in PBS and then lysed in buffer containing 50 mM Tris pH 7.8, 150 mM NaCl, 1% (w/v) DDM, 0.5% (w/v) digitonin (Calbiochem) and protease inhibitor cocktail (Roche) for two hours. The supernatant was incubated with Streptactin resin (GE Healthcare) for two hours and the resin was washed with buffer E (50 mM Tris pH 7.8, 150 mM NaCl, 0.08 % (w/v) digitonin) in spin columns. Protein was eluted from the resin with buffer E supplemented with 3.5 mM desthiobiotin. Complex formation between CD9 and the co-transfected EWI-F variants was assessed by SDS PAGE.

N-StrepII-tagged, wild type CD9 and EWI-FΔIg1-5 were expressed in 3L HEK293-EBNA GNT1-cell cultures. Cells were grown at 37°C and harvested after four days. All subsequent steps were carried out at 4°C. Cells were washed in PBS and then lysed in buffer containing 50 mM Tris pH 7.8, 150 mM NaCl, 1% (w/v) digitonin 0.2% (w/v) DDM and protease inhibitor cocktail (Roche) for two hours. The lysed sample was ultracentrifuged for 45 min at 100,000 g to remove insoluble membranes and cell debris. The supernatant was incubated with Streptactin resin (GE Healthcare) for two hours and the resin was washed with 20 column volumes of buffer E (50 mM Tris pH 7.8, 150 mM NaCl, 0.08 % (w/v) digitonin). Protein was eluted from the resin with buffer E supplemented with 3.5 mM desthiobiotin. The eluted fractions were pooled, concentrated in a 100 kDa concentration device (Amicon) to ~1.5 mg/ml and incubated with EndoHF (NEB) at a volume to volume ratio of 1:20 for 2 hours to remove the N-linked glycan of EWI-FΔIg1-5. The complex was then incubated with a large excess of nanobody 4C8 (in PBS buffer with 0.08% (w/v) digitonin) and injected on a Superdex200 10/300increase column (GE healthcare) pre-equilibrated in buffer E. Fractions containing the EWI-FΔIg1-5 -CD9 -4C8 complex were pooled and concentrated to ~3 mg/ml. 

For the CD9 -EWI-FFull-Length dataset, 379 micrographs were imported in the Relion3.0beta pipeline (70) . The super-resolution mode recorded micrographs were binned 2x, gain corrected and motion corrected using MotionCor2 (71) , and GCTF (72) was used to estimate the contrast transfer function for each micrograph. 880 particles were manually picked and 2D classified.

The resulting 2D-class averages were then used as templates for automated particle picking in Relion (73) , yielding 29,216 particles, that were binned 2x upon particle extraction. These particles were subjected to numerous 2D classification runs. Further processing was not attempted due to strong heterogeneity in the particles.

For the EWI-FΔIg1-5 -CD9 -4C8 dataset, 10,724 micrographs were imported in the Relion3.1beta pipeline. The micrographs were motion-corrected and gain-corrected using MotionCor2, and GCTF was used to estimate the contrast transfer function for each micrograph. 428 micrographs were discarded based on poor CTF estimations, yielding a total of 10,296 micrographs for further processing. The motion-corrected micrographs were imported in EMAN2 (74) , and several hundred good particles and bad particles, as well as background images, were picked manually for training of the EMAN2 NeuralNet particle picker, which was subsequently used for automated particle picking. The obtained particle coordinates were imported back into the Relion pipeline and 1,148,718 particles were extracted and binned 3x (resulting pixel size 3.09 Å). The particles were 3D classified into four classes, and the 107,830 particles belonging to one obvious junk class were discarded. 1,040,888 particles were then subjected to two rounds of 2D classification into 200 classes, through which 352,979 particles were removed. The remaining 687,909 particles were 3D classified into 20 classes. The highest-populated class, comprising 354,272 particles, was 3D-auto refined using a mask and solvent flattening fourier shell correlations (FSCs), which yielded a map at a global resolution of 8.6 Å based on the gold-standard FSC = 0.143 criterion (75) . This map was sharpened with a B-factor of -1200 Å 2 and filtered based on local-resolution in Relion. Another 3D classification with the 687,909 particles (following the 2D classifications) was performed into five classes, while ignoring the CTFs until the first peak, meaning that CTF-amplitude correction was only performed from the first peak of each CTF onward. This strategy resulted in better particle separation in distinct classes. However, further sub-classifications, either with or without regions of the protein complex masked out, yielded density maps of worse quality and resolution. This suggested a continuous disorder in the protein complex rather than that few discrete conformations were adopted by EWI-FΔIg1-5 -CD9 -4C8. The particles were not un-binned as the Nyquist frequency (6.2 Å) was not reached in any refinement.

Protein models were rigid-body fitted into the sharpened 8.6-Å local-resolution filtered EWI-FΔIg1-5 -CD9 -4C8 map, as well as in non-sharpened maps obtained through 3D classifications, using the 'Fit in Map' option in UCSF Chimera (76) .

The crystal-structure figures were prepared using Pymol (Schrödinger). All figures containing cryo-EM density maps were generated using UCSF Chimera. The morph between four observed conformations of the EWI-FΔIg1-5 -CD9 -4C8 complex (Video S1) was made in UCSF Chimera and edited using Adobe Premiere Project. The cartoon models (Fig. 7) were prepared in Adobe Illustrator.

The authors declare no competing interests.

Data supporting the findings of this manuscript are available from the corresponding author Map resolution range (Å) 7.9 -17.4 Figure 6 : Flexibility of the EWI-FΔIg1-5 -CD9 -4C8 revealed by 3D classifications. Four classes obtained from a single 3D classification run are shown (see also Fig. S4 ). The density regions corresponding to the Ig6-domain of EWI-F, CD9EC2, nanobody 4C8 and the digitonin micelle are colored green, blue, orange and white, respectively. The top row depicts the four classes parallel to the membrane as a sideview. The middle row shows the classes orthogonal to the membrane from the extracellular side. The bottom row depicts the density maps in the same orientation as in the middle row, with annotated distances and angles between CD9EC2-residue K135 and 4C8-residue S121. 

Microdomains in the membrane landscape shape antigen-presenting cell function

The tetraspanin superfamiliy: molecular facilitators

Tetraspanin functions and associated microdomains

Tetraspanin-enriched microdomains: a functional unit in cell plasma membranes

Molecular interactions shaping the tetraspanin web

Lateral organization of membrane proteins: Tetraspanins spin their web

CD9, CD63, CD81, and CD82 are components of a surface tetraspan network connected to HLA-DR and VLA integrins

Characterization of novel complexes on the cell surface between integrins and proteins with 4 transmembrane domains (TM4 proteins)

The tetraspanin web revisited by super-resolution microscopy

Tetraspanins in extracellular vesicle formation and function

CD9 controls integrin α5β1-mediated cell adhesion by modulating its association with the metalloproteinase ADAM17

The tetraspanins CD9 and CD81 regulate CD9P1-induced effects on cell migration

Proteins CD9 and CD81 in Muscle Cell Fusion and Myotube Maintenance

Normal muscle regeneration requires tight control of muscle cell fusion by tetraspanins CD9 and CD81

RANKL-induced expression of tetraspanin CD9 in lipid raft membrane microdomain is essential for cell fusion during osteoclastogenesis

Tetraspanins CD9 and CD81 function to prevent the fusion of mononuclear phagocytes

Severely reduced female fertility in CD9-deficient mice. Science (80-. )

The gamete fusion process is defective in eggs of Cd9-deficient mice

CD9 tetraspanin: A new pathway for the regulation of inflammation?

Modulate HIV-1-Induced Membrane Fusion

The tetraspanin CD9 facilitates MERS-coronavirus entry by scaffolding host cell receptors and proteases

Tetraspanin assemblies in virus infection

CD81 extracellular domain 3D structure: Insight into the tetraspanin superfamily structural motifs

Structural insights into tetraspanin CD9 function

CD9 antigen is an accessory subunit of the VLA integrin complexes

Membrane-anchored heparin-binding EGF-like growth factor (HB-EGF) and diphtheria toxin receptor-associated protein (DRAP27)/CD9 form a complex with integrin α3β1 at cell-cell contact sites

Different states of integrin LFA-1 aggregation are controlled through its association with tetraspanin CD9

Tetraspanin CD9: A key regulator of cell adhesion in the immune system

FPRP, a Major, Highly Stoichiometric, Highly Specific CD81-and CD9-associated

EWI-2 Is a Major CD9 and CD81 Partner and Member of a Novel Ig Protein Subfamily

The major CD9 and CD81 molecular partner. Identification and characterization of the complexes

Heparin-binding EGF-like growth factor, which acts as the diphtheria toxin receptor, forms a complex with membrane protein DRAP27/CD9, which up-regulates functional receptors and diphtheria toxin sensitivity

The sheddase activity of ADAM17/TACE is regulated by the tetraspanin CD9

Glycosylation status of the membrane protein

The Ig domain protein CD9P-1 down-regulates CD81 ability to support Plasmodium yoelii infection

EWI-2 and EWI-F link the tetraspanin web to the actin cytoskeleton through their direct association with ezrin-radixin-moesin proteins

In situ chemical cross-linking on living cells reveals CD9P-1 cis-oligomer at cell surface

Site-specific functionality of lipidation in Tetraspanin CD9 revealed by tryptophan mimicry

Subunit association and conformational flexibility in the head subdomain of human CD81 large extracellular loop

An intramolecular bond at cluster of differentiation 81 ectodomain is important for hepatitis C virus entry

Mechanism of Structural Tuning of the Hepatitis C Virus Human Cellular Receptor CD81

Structure-Guided Combinatorial Engineering Facilitates Affinity and Specificity Optimization of Anti-CD81 Antibodies

The extracellular δ-domain is essential for the formation of CD81 tetraspanin webs

Oligomerization of the Tetraspanin CD81 via the Flexibility of Its δ-Loop

The specificity of homomeric clustering of CD81 is mediated by its δ-loop

Domain swapping: Entangling alliances between proteins

3D domain swapping: As domains continue to swap

Structure of the tetraspanin main extracellular domain: A partially conserved fold with a structurally variable domain insertion

Crystallization of the human tetraspanin protein CD9

Tetraspanin CD82 regulates compartmentalisation and ligand-induced dimerization of EGFR

Multiple levels of interactions within the tetraspanin web

Expression of the palmitoylation-deficient CD151 weakens the association of α3β1 integrin with the tetraspaninenriched microdomains and affects integrin-dependent signaling

Tetraspanins at a glance

Structural organization and interactions of transmembrane domains in tetraspanin proteins

A new panel of epitope mapped monoclonal antibodies recognising the prototypical tetraspanin CD81

In macrophages, HIV-1 assembles into an intracellular plasma membrane domain containing the tetraspanins CD81, CD9, and CD53

Unravelling the molecular mechanisms of tetraspanin CD9

EVAL15: A diffraction data integration method based on ab initio predicted profiles

Phaser crystallographic software

CCP4i2: The new graphical user interface to the CCP4 program suite

REFMAC5 for the refinement of macromolecular crystal structures

Coot: Model-building tools for molecular graphics

XPREP -Data preparation and reciprocal space exploration 66

DIALS: Implementation and evaluation of a new integration package

SWISS-MODEL: Homology modelling of protein structures and complexes

PHENIX : building new software for automated crystallographic structure determination

RELION: Implementation of a Bayesian approach to cryo-EM structure determination

MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy

Gctf: Real-time CTF determination and correction

Semi-automated selection of cryo-EM particles in RELION-1.3

EMAN2: An extensible image processing suite for electron microscopy

Optimal determination of particle orientation, absolute hand, and contrast loss in single-particle electron cryomicroscopy

Visualizing density maps with UCSF Chimera

736) Unique reflections* 22

We thank the beamline scientists of Diamond Light Source (DLS) and the European Synchrotron Radiation Facility (ESRF) for assistance during data collection; we acknowledge 

Reflections used in refinement 22