Introduction

The mucous membrane covers ~400 m2 surface of internal organs in the human body. Immunoglobulin A (IgA) is the most predominant antibody in mucosal secretions.1 In contrast to IgA in serum that is mostly monomeric, mucosal IgA is mainly present as dimers (dIgA), in which two IgA molecules are linked together by another protein designated the joining chain (J-chain).2 The J-chain is also an integral component of the IgM pentamer (pIgM).3 The heavy chains of IgA and IgM contain unique C-terminal extensions known as the tailpieces, which are essential for their oligomerization and covalent linkage to the J-chain.4,5,6 Furthermore, an additional polypeptide called the secretory component (SC) is present in mucosal IgA and IgM, and such IgA and IgM complexes are often referred to as secretory IgA and IgM (SIgA and SIgM). SC is the ectodomain of the polymeric immunoglobulin receptor (pIgR), which functions to transport dIgA and pIgM through the mucosal epithelial cells.7,8,9,10 SIgA forms a critical first line of defense against pathogens at the mucosal surface, and also likely plays an important role in regulating the homeostasis of microbiota.11,12 SIgA in breast milk is important for protecting the newborn babies until their own immune systems have developed. Despite the fact that the composition of SIgA and certain details of its assembly process have long been established, the three-dimensional structure of SIgA has remained insufficiently characterized.

Due to the critical function of SIgA in immune defenses, various pathogens have developed strategies to disrupt its function. Streptococcus pneumoniae, also known as pneumococcus, is a Gram-positive bacterium that causes millions of deaths worldwide.13,14 It is an opportunistic pathogen residing in the upper respiratory tract of many people especially young children. In individuals with a weak immune system, the bacterium can invade a wide range of organs including the brain, causing severe diseases such as pneumonia, sepsis, and meningitis. S. pneumoniae SIgA-binding protein (SpsA; also known as CbpA, PspC) is a pneumococcal adhesin that binds to SIgA.15 The binding is mediated by the SC subunit in SIgA, and likely impairs the bacterial clearance function of SIgA. Furthermore, SpsA also interacts with unliganded SC and pIgR, and the interaction with pIgR may enhance bacterial adherence and facilitate its cellular invasion.16 How SpsA selectively recognizes human pIgR/SC is unclear.

Here we report the cryo-electron microscopy (cryo-EM) structure of a complex containing the Fc region of human IgA1 (Fcα), J-chain, and SC at 3.2 Å resolution. Comparison of this structure with that of Fcμ-J-SC17 reveals a more complete J-chain and distinctive features for the interactions between Fcα, J-chain, and SC. We also investigated the interaction between SIgA and S. pneumoniae adhesin SpsA, and determined a cryo-EM structure of human Fcα-J-SC in complex with the N-terminal domain (NTD) of SpsA, which shows how human pIgR/SC is specifically exploited by SpsA to promote S. pneumoniae pathogenesis.

Results

Structure determination of the Fcα-J-SC complex

We co-expressed human Fcα with J-chain in HEK293F cells and isolated the Fcα-J complex containing the dimeric Fcα. SC was individually expressed and purified, and then incubated with the Fcα-J sample to form the Fcα-J-SC tripartite complex. We then determined its structure at 3.2 Å resolution (as judged by the FSC 0.143 criterion) by the single particle cryo-EM method (Fig. 1; Supplementary information, Fig. S1). Most regions of the EM map exhibited high enough resolution for unambiguous structural assignment and analyses (Supplementary information, Figs. S1e and S2). The statistics for cryo-EM data collection and processing, as well as structural refinement and validation, are summarized in Supplementary information, Table S1.

Fig. 1: Cryo-EM structure of the human Fcα-J-SC complex.
figure 1

The cryo-EM structure of the human Fcα-J-SC complex is shown in two views. The regions corresponding to Fcα1, Fcα2, J-chain, and SC are shown in teal, light cyan, salmon, and gold, respectively. Similar color schemes are used in all figures unless otherwise indicated. The five immunoglobulin-like domains in SC are indicated as D1–D5.

Structure of the IgA dimer and its interaction with the J-chain

The two IgA molecules in SIgA are linked via Cys471-mediated disulfide bonds to the J-chain. Earlier EM analyses show that dIgA displays a double-Y-like shape.18 Solution scattering studies suggest that the two Fcα do not dock to each other in a straight manner in the IgA dimer, but adopt a slightly bent end-to-end arrangement.19 Consistent with these results, our structure shows that the Fcα dimer has a boomerang-like shape (Fig. 2a) and resembles a portion of the pIgM structure we determined recently17 (Fig. 2b). Each tailpiece of Fcα contains a β-strand, like that of Fcμ in IgM, and four such tailpiece strands bundle together to mediate the interactions between the two Fcα molecules. The EM densities are discontinuous between the tailpieces and the rest of heavy chains in Fcα1 (Fig. 2a), yet the strands in the tailpieces can still be fully resolved, suggesting that these strands are structurally stable and serve to fasten the IgA dimer.

Fig. 2: The core structures of dIgA and pIgM.
figure 2

a Overall structure of the dimeric Fcα in complex with the J-chain. The linkers between the tailpieces and the rest of Fcα1 heavy chains are indicated with dashed lines, as the EM densities are discontinuous at these regions. The β5–β6 hairpin of the J-chain (disordered in the Fcμ-J structure) is highlighted in magenta. b Structure of the pentameric Fcμ in complex with the J-chain (PDB ID: 6KXS). Fcμ1 and Fcμ5 are shown in two shades of blue, whereas Fcμ2-4 are shown in white. The β5–β6 region of the J-chain is disordered in this structure and indicated with a dashed line.

The Fcα dimer is further stabilized by the J-chain. Compared with the Fcμ-J complex, a more complete structure of the J-chain is present in Fcα-J, due to more extensive interactions between the Fcα dimer and the J-chain. Three intrachain disulfide bridges are seen in the J-chain (Fig. 3a): Cys12J–Cys100J (superscript J indicates J-chain residues), Cys71J–Cys91J, and Cys108J–Cys133J, consistent with previous analyses.20,21 Cys14J and Cys68J form a disulfide bond with Cys471Fcα2B and Cys471Fcα1A, respectively. The central region of the J-chain contains four β-strands (β1–β4) that interact with the Fcα tailpieces (Fig. 3a). Strands β1–β3 pack onto the two tailpiece strands of Fcα2 to assemble into a β-sheet, with main chain hydrogen bonds formed between adjacent strands; whereas β4 packs onto the tailpieces of Fcα1. Robust hydrophobic interactions are present between the two β-sheets, mediated by Fcα residues Val460, Val462, and Met464, and J-chain residues Ile37J, Ile39J, Val41J, Phe60J, and Tyr62J (Fig. 3b). The β2–β3 loop, β3–β4 loop, β5–β6 hairpin, and the long C-terminal hairpin of the J-chain function as four lassos to further interact with the Fcα dimer. The β2–β3 and β3–β4 loops contact the Cα3-tailpiece junctions of Fcα2B and Fcα1A, respectively (Supplementary information, Fig. S3); whereas the β5–β6 and C-terminal hairpins form extensive interactions with the Cα2–Cα3 junctions. The β5–β6 hairpin, disordered in the Fcμ-J complex (Fig. 2b), interacts with the Cα2–Cα3 junction of Fcα2B via two hydrophobic centers (Fig. 3c): the first involves J-chain residues Val76J, Leu78J, and Val83J, and Fcα2B residues Leu258Fcα2B, Arg382Fcα2B, Leu384Fcα2B, Met433Fcα2B, and Phe443Fcα2B; and the second involves Thr86J, Leu439Fcα2B, Pro440Fcα2B, and Leu441Fcα2B. The C-terminal hairpin interacts with almost the same group of residues in Fcα1A (Fig. 3d). The way that the C-terminal hairpin interacts with Fcα1 highly resembles how it binds to Fcμ1 in pIgM (Fig. 2).

Fig. 3: The interactions between the J-chain and Fcα.
figure 3

a Overall structure of the J-chain. The sulfur atoms in Cys are depicted as orange spheres. The β-strands in the J-chain and Fcα tailpieces are indicated. b Interactions between two J-chain strands and the Fcα tailpieces. c Interactions between the β5–β6 hairpin of the J-chain and the Cα2–Cα3 junction of Fcα2B. d Interactions between the C-terminal hairpin of the J-chain and the Cα2–Cα3 junction of Fcα1A.

Interaction between dIgA and SC

Extensive biochemical and biophysical studies have established that pIgR/SC forms bidentate interactions with dIgA, with both its D1 and D5 domains involved.22,23,24,25,26,27,28 This is borne out by the cryo-EM structure (Fig. 4a). The D1 domain of pIgR/SC binds to Fcα-J using its three complementarity-determining regions (CDR)-like loops (Fig. 4b), and the molecular interactions are in many ways similar to the interactions seen in the Fcμ-J-SC complex.17 CDR1 mainly contacts the J-chain. Val29SC (superscript SC indicates pIgR/SC residues) is positioned in a pocket formed by J-chain residues Arg105J, Asn106J, and Ala132J (Fig. 4c). Asn30SC coordinates Arg105J. Arg31SC interacts with Asp136J. His32SC packs against Tyr134J. Besides interacting with the J-chain, pIgR/SC also directly contacts the Fcα molecules at several places. For example, Arg34SC in CDR1 forms a salt bridge with Glu363Fcα1B, which also packs on Tyr55SC in CDR2 (Fig. 4d). Glu53SC in CDR2 engages Arg346Fcα1A. Arg99SC and Leu101SC in CDR3 enclose Tyr472Fcα2B, the terminal residue of Fcα2B, together with Arg105J (Fig. 4c). To verify the functional relevance of these molecular interactions, we tested the interaction between Fcα-J and two SC mutants, V29N/R31S and R99N/L101T. These mutants were designed to bring bulky N-linked glycans into amino acid positions 29 and 99, respectively, which would hinder molecular interactions at CDR1 and CDR3. We previously showed that these mutants displayed greatly reduced interactions with the Fcμ-J complex.17 Likewise, they failed to bind Fcα-J (Fig. 4e).

Fig. 4: Interaction between dIgA and SC.
figure 4

a Both the D1 and D5 domains of SC are involved in binding to dIgA. Fcα and the J-chain are shown as ribbon diagrams, whereas SC is shown as a surface representation. The D1 and D5 domains of SC are highlighted in gold. The rest of SC is in white. b The three CDR-like loops in the D1 domain of SC, highlighted in thicker ribbons, engage Fcα1 and the J-chain. The surfaces of Fcα1 and the J-chain are shown. c Interactions between SC and Fcα-J at the CDR1 and CDR3 regions. Polar interactions are indicated by dashed lines. d Direct interactions between SC and Fcα1. e SC mutants display reduced interactions with Fcα-J. f The structure of the Fcα monomer (PDB ID: 2QEJ), shown in blue, is overlaid onto Fcα2B in the Fcα-J-SC structure. Compared with Cys311 in the Fcα monomer, Cys311Fcα2B flips out and can readily form a disulfide bond with Cys468SC.

The interaction between dIgA and pIgR/SC also uniquely involves the D5 domain of pIgR/SC, and a disulfide bond is formed between Cys468SC and Cys311 in the Cα2 domain of Fcα.23 In the two previously determined structures of Fcα,29,30 Cys311 is present in a hydrophobic pocket and not exposed (Fig. 4f). By contrast, Cys311Fcα2B flips out in the Fcα-J-SC structure and is located in close proximity to Cys468SC. The EM density did not allow us to model a disulfide bond between them definitively (Supplementary information, Fig. S4a). A non-reducing gel suggests that our Fcα-J-SC sample is likely heterogeneous, with disulfide bonds formed between some Fcα and SC molecules (Supplementary information, Fig. S4b). Therefore, the EM density at this region is likely derived from the average of disulfide-linked and non-disulfide-linked Fcα-SC. For this study, we produced Fcα-J and SC separately, incubated them on ice for 1 h, and then isolated the tripartite complex using size exclusion chromatography (see Materials and methods). It is likely that our experimental procedure did not allow sufficient time for the oxidation reaction to fully occur. It is also likely that the formation of this disulfide bond might be facilitated by a protein disulfide isomerase in vivo. In any event, mutation of C468SC only slightly decreased the binding between Fcα-J and SC in solution (Fig. 4e). This is consistent with previous analyses showing that the initial and primary association of dIgA with pIgR/SC is mediated by interactions at the D1 domain. Disulfide formation between dIgA and Cys468SC in D5 takes place at a later stage of transcytosis, and the mucosal transport of dIgA can proceed in its absence.31 The main function of this disulfide bond is presumably to increase the stability of SIgA at the mucosal surface and in external fluids.

Interaction between SC and S. pneumoniae SpsA

SpsA comprises a C-terminal phosphorylcholine-binding domain that interacts with pneumococcal cell wall to facilitate bacterial colonization, and an NTD that recruits host proteins including pIgR/SC and SIgA.15,32 SpsANTD contains repeats of the leucine zipper motifs termed R1 and R2, each adopting a three-helix bundle structure.33 The YRNYPT hexapeptide involved in binding to pIgR/SC is located in the loop between helices α1 and α2 in the R1 motif of S. pneumoniae strain ATCC33400 SpsA, where it was first identified.15 In many other S. pneumoniae strains, a similar hexapeptide is also present in R2 (Supplementary information, Fig. S5a). To reveal the molecular mechanism underlying the specific recognition of SIgA by SpsA, we produced S. pneumoniae strain ATCC33400 SpsANTD, reconstituted a Fcα-J-SC-SpsANTD quadruple complex, and then determined the cryo-EM structure at an overall resolution of 3.3 Å (Fig. 5a, b; Supplementary information, Figs. S1 and S6 and Table S1). The α1–α2 loop of SpsANTD, especially the YRNYPT hexapeptide, displays high-quality densities and can be clearly resolved (Supplementary information, Fig. S6b).

Fig. 5: Interaction between SC and SpsA.
figure 5

a Overall structure of the Fcα-J-SC-SpsANTD quadruple complex. b Another view of the Fcα-J-SC-SpsANTD structure. c Structure of the D3–D4 domains of SC in complex with SpsANTD. The D3–D4 domains of SC are shown in a surface representation, with D3 and D4 in gold and purple, respectively. SpsANTD is shown in lemon, and the side chains of the YRNYPT hexapeptide are depicted. d Detailed interactions between SC and SpsA.

SpsANTD specifically interacts with the D3–D4 domains of human pIgR/SC.34,35 In the cryo-EM structure, the α1–α2 loop of SpsANTD docks into a pocket at the D3–D4 junction (Fig. 5b, c), formed by the DE loop of D3 and the C–C′ strands of D4. Notably, this pocket is only present in the ligand-bound conformation of SC.17 No significant conformational changes were observed in SC upon SpsANTD binding, except for the C–C′ loop of D4, which tilted slightly to engage the YRNYPT hexapeptide (Supplementary information, Fig. S7a). Tyr198, the first residue in the hexapeptide, forms a hydrogen bond with Tyr365SC (Fig. 5d). Arg199 packs on Trp386SC, and forms a salt bridge with Asp382SC. Asn200 forms a hydrogen bond with Arg376SC. Tyr201 packs on Pro283SC. Pro202 is surrounded by hydrophobic residues including Tyr365SC, Cys367SC, Cys377SC, Leu379SC, and Leu424SC. Substitution of Tyr201 with an Asp or Pro202 with a Glu abolished the binding of SpsA to SIgA.36 Thr203 interacts with Asn282SC. Notably, most of the SC residues described here are specifically present in human (Supplementary information, Fig. S5b), explaining the fact that SpsA only binds to human SIgA.36 Other SpsA residues are also involved in interacting with SC in addition to the YRNYPT hexapeptide, including Tyr206 that forms a hydrogen bond with the main chain carbonyl group of Pro283SC, and Arg265 in helix α3 that likely bonds with Asp285SC (Fig. 5d). Both of these residues are highly conserved in SpsA (Supplementary information, Fig. S5a).

Discussion

SIgA is of paramount importance to mucosal immunity. In adults, the daily synthesis of IgA is greater than all other types of antibody combined, and most of these IgA molecules are present in mucosal secretions in the form of dimeric SIgA. Despite the long history of SIgA research, its structure has remained elusive until only recently. During the preparation of this manuscript, the cryo-EM structures of SIgA have been published by Genentech.37 Our Fcα-J-SC structure is very similar to the structure reported in this study (Supplementary information, Fig. S7b), since comparison of these two structures reveals a global root mean square deviation of only 1.6 Å. The similarity of these independently determined structures demonstrates the sturdiness of the dimeric SIgA core assembly.

IgA can induce immune signaling by binding to the IgA-specific receptor FcαRI/CD89.38,39 However, it is not entirely clear whether monomeric IgA and SIgA can elicit similar immune responses. Crystal structure study reveals a 2:1 FcαRI:Fcα complex29 (Fig. 6a). In SIgA, only one side of the FcαRI-binding site would be exposed in each Fcα (Fig. 6b). The other side is occupied by the J-chain and not available for binding. From a structural point of view, there is no apparent reason to think that SIgA would not bind to FcαRI; nevertheless, it would have to cluster FcαRI molecules in a different arrangement. Whether this altered mode of binding may account for the different immune responses elicited by monomeric IgA and SIgA remains to be investigated. On the other hand, the J-chain and SC may also contribute to the signaling function of SIgA. For example, it has been shown that the Mac-1 integrin is required as a co-receptor for SIgA to activate immune responses via FcαRI, and the interaction with Mac-1 is mediated by SC.40,41 Recently, human Fc receptor-like 3 (FCRL3) has been identified as a SIgA-specific receptor.42 It is likely that the J-chain and/or SC are also involved in the interaction between SIgA and FCRL3.

Fig. 6: A hypothetical model of the dIgA-FcαRI complex.
figure 6

a Crystal structure of the 2:1 FcαRI:Fcα complex (PDB ID: 1OW0). The two FcαRI molecules are shown in orange and yellow. The distance between the C-terminal ends of the two FcαRI molecules is indicated. b Two copies of the FcαRI:Fcα structures are overlaid onto the two Fcα molecules in the Fcα-J structure to illustrate the potential binding mode between FcαRI and dIgA (only the FcαRI molecules in FcαRI:Fcα are shown for clarity). One side of dIgA is occupied by the J-chain and not exposed for interacting with the FcαRI molecules that are shown in yellow.

S. pneumoniae is an important human pathogen. SpsA is a major adhesin of S. pneumoniae and plays a role during its infection. SpsA is a modular protein and can bind to both host proteins and the pneumococcal cell wall. Despite the fact that the DNA region encoding SpsA is highly polymorphic, the Y/R-R-N-Y-P-T motif involved in binding to pIgR/SC and SIgA is highly conserved, and is present in one or two copies in more than 70% strains of S. pneumoniae.15,43 The unique ability of SpsA to bind pIgR/SC and SIgA may contribute to the pathogenesis S. pneumoniae in several ways. First, binding to pIgR on the nasopharyngeal and lung epithelia as well as brain endothelial cells may facilitate the colonization and internalization of S. pneumoniae. Indeed, SpsA-deficient S. pneumoniae showed greatly reduced ability to adhere to, and abolished activity to invade human cells in tissue culture models.16,32 Binding to SIgA may also assist in bacterial invasion, since SIgA in complex with its antigens can be internalized by certain cells via reverse transcytosis.44,45 On the other hand, the interaction between SpsA and SIgA may impede the defense function of SIgA. By binding to SIgA using its NTD and the bacterial cell wall via its C-terminal domain, SpsA may recruit SIgA to bacterial surface and promote its degradation by S. pneumoniae IgA proteases.46 Binding of SpsA to SIgA may also sterically hinder SIgA from agglutinating and clearing the bacteria, or block the interaction between SIgA and binding partners to initiate host immune responses.

It is worth noting that SpsA evolves to bind to human pIgR/SC specifically, since it does not interact with SIgA and SC from common laboratory animals including mouse, rat, rabbit, and guinea pig.36 Indeed, residues in pIgR/SC that participate in the interaction with SpsA are not conserved in these animals (Supplementary information, Fig. S5b). These differences underscore the fact that S. pneumoniae is a human-specific pathogen and should be taken into consideration for its study. On the other hand, the unusual capability of SpsA to bind human SC with high selectivity and affinity may allow the development of recombinant SpsA protein as a tool for efficient isolation and purification of human SC, SIgA, and SIgM.

Materials and methods

Protein expression and purification

The DNA fragment encoding IgA1-Fc (residues 241–472) was cloned into a modified pcDNA vector with an N-terminal IL-2 signal peptide followed by a twin-strep tag. The DNA fragments encoding the full-length J-chain and SC were cloned into a modified pcDNA vector with a C-terminal 8× His tag as previously described.17 HEK293F cells were cultured in SMM 293T-I medium (Sino Biological Inc.) at 37 °C, with 5% CO2 and 55% humidity. The two plasmids expressing IgA1-Fc and J-chain were co-transfected into the cells using polyethylenimine (Polysciences). Four days after transfection, the conditioned media were collected by centrifugation, concentrated using a Hydrosart Ultrafilter (Sartorius), and exchanged into the binding buffer (25 mM Tris-HCl, pH 7.4, 150 mM NaCl). The recombinant proteins were isolated using Ni-NTA affinity purification and eluted with the binding buffer supplemented with 500 mM imidazole. The Fcα-J complex was further purified using a Superdex 200 increase column (GE Healthcare) and eluted using the binding buffer. SC was expressed and purified as previously described.17 To obtain the Fcα-J-SC tripartite complex, purified Fcα-J and SC were mixed in an 1:2 molar ratio and incubated on ice for 1 h. The complex was then further purified on a Superdex 200 increase column and eluted using the binding buffer.

The DNA fragment encoding S. pneumoniae (strain ATCC33400) SpsANTD (residues 38–324) was synthesized by Synbio Technologies and cloned into a modified pQlink vector with an N-terminal 8× His tag. SpsANTD was expressed in BL21(DE3)pLysS E. coli. The E. coli culture was grown in the Luria–Bertani medium at 37 °C to an OD600 of 0.8, and then induced with 0.5 mM isopropyl-β-D1-thiogalactopyranoside at 18 °C overnight for protein expression. The cells were collected by centrifugation, resuspended in the lysis buffer (50 mM Tris-HCl, pH 8.0, 300 mM NaCl, 1 mM phenylmethylsulfonyl fluoride), and then disrupted by sonication. The insoluble debris was removed by centrifugation. The recombinant protein was isolated using Ni-NTA affinity purification following standard procedure and eluted with 50 mM Tris-HCl, pH 8.0, 300 mM NaCl, and 500 mM imidazole. SpsANTD was then further purified by gel filtration chromatography using a Superdex 200 increase column and eluted using the binding buffer. To obtain the Fcα-J-SC-SpsANTD quadruple complex, purified Fcα-J-SC and SpsANTD were mixed in a 1:2 molar ratio and incubated on ice for 1 h. The complex was then purified again on a Superdex 200 increase column and eluted using the binding buffer.

Negative staining and cryo-EM

The samples for EM study were prepared essentially as previously described.17 All EM grids were evacuated for 2 min and glow-discharged for 30 s using a plasma cleaner (Harrick PDC-32G-2). For the negative-staining study, 4-μL aliquots of the Fcα-J-SC complex at 0.03 mg/mL were applied to glow-discharged carbon-coated copper grids (Zhong Jing Ke Yi, Beijing). After ~40 s, excessive liquid was removed using a filter paper (Whatman No. 1). The grid was then immediately stained using 2% uranyl acetate for 10 s and air dried. The grids were examined on a Tecnai G2 20 Twin electron microscope (FEI) operated at 120 kV. Images were recorded using a 4k × 4k CCD camera (Eagle, FEI). The Fcα-J-SC-SpsANTD sample was stained and examined similarly. To prepare the sample for cryo-EM analyses, 4-μL aliquots of Fcα-J-SC (0.3 mg/mL) or Fcα-J-SC-SpsANTD (0.2 mg/mL) were applied to glow-discharged holy-carbon gold grids (Quantifoil, R1.2/1.3), blotted with filter paper at 4 °C and 100% humidity, and plunged into the liquid ethane using a Vitrobot Mark IV (FEI). Grids screening was performed using a Talos Arctica microscope equipped with Ceta camera (FEI). Data collection was carried out using a Titan Krios electron microscope (FEI) operated at 300 kV. Movies were recorded on a K2 Summit direct electron detector (Gatan) in a super resolution mode using the SerialEM software.47 A nominal magnification of 165,000× was used, and the exposure rate was 11.668 electrons per Å2 per second. The slit width of the energy filter was set to 20 eV. The defocus range was set from –0.8 to –1.6 μm. The micrographs were dose-fractioned into 32 frames with a total exposure time of 5.12 s and a total electron exposure of 60 electrons per Å2. Statistics for data collection are summarized in Supplementary information, Table S1.

Imaging processing

For 3D reconstruction of the Fcα-J-SC complex, a total of 16,264 movie stacks were recorded. Raw movie frames were aligned and averaged into motion-corrected summed images with a pixel size of 0.828 Å by MotionCor2.48 The contrast transfer function (CTF) parameters of each motion-corrected image were estimated by the Gctf program (v1.06).49 Relion (v3.07) was used for all the following data processing.50 Manual screening was performed to remove low-quality images. A set of 475 particles was manually picked and subjected to 2D classification to generate templates for automatic particle picking. A total of 6,875,153 particles were then auto-picked, which were subjected to another round of 2D classification, resulting in 5,051,275 particles that were kept for the subsequent 3D classifications. Initial model was generated using Relion and used as a reference for 3D classification. Three of the six classes (665,589 particles) from the final round of 3D classification were selected and combined for refinement, resulting in a map with a 3.23 Å overall resolution after mask-based post-processing. Finally, Bayesian polishing and CTF refinement were applied, which yielded a density map at a resolution of 3.15 Å, based on the gold-standard FSC 0.143 criteria. The local resolution map was analyzed using ResMap51 and displayed using UCSF Chimera.52 Similar data processing strategies were used for the Fcα-J-SC-SpsA complex. The workflows of data processing are illustrated in Supplementary information, Fig. S1.

Model building and structure refinement

The structure of Fcα (PDB ID: 1OW0), as well as the structures of the J-chain and SC from the Fcμ-J-SC complex (PDB ID: 6KXS), was docked into the EM map using Phenix53 and then manually adjusted using Coot.54 The β5–β6 hairpin of the J-chain, which is disordered in the Fcμ-J-SC structure, was built de novo. The SpsANTD structure was also built de novo, using the previously determined solution structure of the R2 domain (PDB ID: 1W9R) as a guide. Residues in helices α1–α2 of SpsANTD can be unambiguously assigned. The amino acid registrations in helix α3 are not entirely reliable, since this helix is only loosely attached to SC and displays poor densities due to structural flexibility. Refinement was performed using the real-space refinement in Phenix. Figures were prepared with Pymol (Schrödinger) and UCSF Chimera.

StrepTactin pull-down assay

WT and mutant SC proteins were purified using the Ni-NTA affinity method as previously described.17 For the pull-down experiments, they were first incubated with purified Fcα-J complex on ice for 1 h. The mixture was then incubated with the StrepTactin beads (Smart Lifesciences) in the binding buffer at 4 °C for another hour. A twin-strep tag is present on Fcα. The beads were spun down and then washed three times with the binding buffer. The bound proteins were eluted off the beads using the binding buffer supplemented with 10 mM desthiobiotin. The results were analyzed by SDS-PAGE and visualized by Coomassie staining.