Streptococcus pyogenes forms serotype and local environment-dependent inter-species protein complexes

Streptococcus pyogenes is known to cause both mucosal and systemic infections in humans. In this study, we used a combination of quantitative and structural mass spectrometry techniques to determine the composition and structure of the interaction network formed between human plasma proteins and the surface of different S. pyogenes serotypes. Quantitative network analysis revealed that S. pyogenes form serotype-specific interaction networks that are highly dependent on the domain arrangement of the surface-attached M protein. Subsequent structural mass spectrometry analysis and computational modelling on one of the M proteins, M28 revealed that the network structure changes across different host microenvironments. We report that M28 binds secretory IgA via two separate binding sites with high affinity in saliva. During vascular leakage mimicked by increasing plasma concentrations in saliva, the binding of secretory IgA was replaced by binding of monomeric IgA and C4BP. This indicates that an upsurge of C4BP in the local microenvironment due to damage of the mucosal membrane drives binding of C4BP and monomeric IgA to M28. The results suggest that S. pyogenes has evolved to form microenvironment-dependent host-pathogen protein complexes to combat the human immune surveillance during both mucosal and systemic infections.


38
Bacterial pathogens have evolved to express a multitude of virulence factors on their surface to 39 establish versatile host-pathogen protein-protein interactions (HP-PPI) 1 . These interactions 40 range from binary interactions between two proteins to the formation of multimeric interspecies 41 protein complexes that enable bacterial pathogens to hijack and re-wire molecular host systems 42 to circumvent host immune defenses. One prominent example is Streptococcus pyogenes, a isolates obtained from the blood of GAS infected patients at Lund university hospital and 166 serotyped by clinical microbiology department of the hospital. These bacteria were grown on 167 blood agar plates, and single colonies were isolated and grown in Todd-Hewitt (TH) broth 168 supplemented with 0.6 % yeast extract at 37°C, in 5% CO2 16 hours. Bacteria from the 169 overnight culture were sub-cultured in TH broth with 0.6% yeast extract at 37°C, in 5% CO2 170 till mid-logarithmic phase (OD600 nm 0.4-0.5). The cells were harvested by centrifugation at 171 3500 g for 5 min. The pellets were washed in HEPES-buffer (50mM HEPES, 150 mM NaCl, 172 pH 7.5) twice and re-centrifuged at 3500 g for 5 min. The washed cells were re-suspended in 173 HEPES-buffer to a 1% solution. These cells were further used for SA-MS experiments. 174 175 Bacterial surface adsorption of human plasma proteins 176 To capture human plasma proteins on S. pyogenes surface 400 µl of pooled normal human 177 plasma was added to 100 µl of 1% bacterial solution in six biological replicates for each strain. 178 The samples were vortexed briefly, and incubated at 37°C at 500 rpm for 30 min. Cells were harvested by centrifugation at 5000 g for 5 min, and washed three times with HEPES-buffer 180 followed by centrifugations of 5000 g for 5 min, respectively. The cells were finally re-181 suspended in 100 µl HEPES-buffer. For limited proteolysis of surface-attached bacterial and 182 human proteins, 2 µg of 0.5 µg/µl sequencing grade trypsin (Promega) was added, and the 183 digestion was allowed to proceed at 37°C, 500 rpm for 60 min. The reaction was stopped on 184 ice, and the supernatant collected by centrifugation at 1000 g for 15 min at 4°C. Any remaining 185 bacteria in the supernatants were heat-killed at 85°C at 500 rpm for 5 min, prior to sample 186 preparation for mass spectrometry. mixed environment experiments 100 µl saliva-plasma dilutions were made for 100% saliva, 1% 195 plasma (99µl saliva + 1µl plasma), 10% plasma (90µl saliva + 10µl plasma) and 100% plasma 196 and incubated with the protein-charged beads at 37 °C, 800 rpm, 1 h. The beads were washed 197 with 10 ml ice-cold 1x PBS (for plasma) and 4 ml ice-cold 1xPBS (for saliva and saliva-plasma 198 dilutions) at 4 °C, before eluting the proteins with 120 µl 5 mM biotin in 1xPBS at room 199 temperature (RT). To remove biotin from the eluted protein mixture tri-chloro acetic acid 200 (TCA) was added to a final concentration of 25% and incubated in -20C for 16hours. The 201 protein mixture was centrifuged at 13000 rpm for 30 min at 4 °C. The pellets were washed two 202 times in 500 µl and once in 200 µl ice-cold acetone by centrifuging at 13000 rpm 10 min at 4 203 °C. These pellets were then prepared for mass spectrometry.  The SA-MS and AP-MS DIA data were processed using the OpenSWATH pipeline 46 . For DIA 276 data analysis, spectral libraries from the above DDA dataset were created in openBIS 42 using 277 SpectraST (version 5.0, TPP v4.8.0 PHILAE, build 201506301157-exported (Ubuntu-x86_64)) 278 in TPP 47 . For DIA data analysis, raw data files were converted to mzXML using msconvert and analyzed using OpenSWATH (version 2.0.1 revision: c23217e). The RT extraction window 280 was ±300 s, and m/z extraction was set at 0.05 Da tolerance. RT was then calibrated using iRT 281 peptides. Peptide precursors were identified by OpenSWATH (2.0.1) and PyProphet (2.0.1) 282 was used to control the false discovery rate of 1% at peptide precursor level and at 1% at protein 283 level. Then TRIC 48 was used to align the runs in the retention time dimension and reduce the 284 identification error by decreasing the number of missing values in the quantification matrix. 285 Further missing values were re-quantified by TRIC 48 . Resulting DIA data sets were analysed 286 using Jupyter Notebooks (version 3.1.1). For the DIA data analysis proteins identified by more 287 than 3 peptides and enriched with a log2 fold enrichment of >1 (two-fold) with an adjusted P-288 value <0.05 using the Student's t-test were considered has true interactors. However, for the 289 saliva-plasma dilution DIA data TRIC was not enabled. The intensities of the proteins were 290 estimated by summing the intensities of the most intense three peptides for each protein relative 291 to the total peptide intensities (without iRT) for that protein. The CM5 chip was docked into the instrument and the chip surface was activated following 303 EDC/NHS protocol with PBS buffer as the running buffer before the immobilization procedure. 304 The ligand (M1/M28) was injected for 7 min (flow rate: 10 µL/min) at a concentration of 0.01 305 mg/mL (in 10 mM acetate buffer, pH: 5.0) followed by an injection of 1.0 M ethanolamine for 306 7 min (flow rate: 10 µL/min) in order to deactivate excess reactive groups. Once the targeted 307 immobilization level (≈ 2500 RU) was achieved no further immobilization was carried out. The 308 flow channel_2 (active channel) was used for the ligand immobilization while the flow 309 channel_1 (reference channel) was used as a reference to investigate non-specific binding. The data collected for each experiment was analysed according to 1-1 fitting model using the 325 kinetic fitting programs that yields ka, kd and KD values and also fitting the data to 326 heterogeneous binding model. Equilibrium binding analysis were performed by plotting the RU 327 values measured in the plateau versus each concentration series. 328 329 First the binding was tested for the simplest 1-1 Langmuir binding model, which follows the 330 equation: 331 where A is the analyte, B is the ligand, AB is the complex. The ka (rate of association, M -1 s -1 ) 333 is measured from the reaction in the forward direction while the kd (dissociation rate, s -1 ) is 334 measured from the reverse reaction. 335 The binding was also tested for heterogeneous ligand model where the same analyte binds 336 independently to multiple ligands or to several binding sites on the same ligand. Heterogeneous (2) 340 where A represents the analyte, B1 and B2 represent two different ligands or two different 341 binding sites on the same ligand, respectively, AB1 and AB2 represent the first and second 342 complexes formed after the binding of the analyte to the surface, ka1 and ka2 are the association 343 rates of the first and second complexes while kd1 and kd2 represent the dissociation rates.

TX-MS data analysis and computational modelling 346
The UniProt accession numbers used for the S. pyogenes M28 protein, human C4BPa, C4BPb, 347 IGHA1, and IGHA2 were W0T1Y4, P04003, P20851, P01876, and P01877, respectively. The 348 tertiary structure of the M28 protein was characterized using Rosetta comparative modeling 349 (1 mm x 15mm) and washed with 0.1% FA for 60s. Thereafter, the trap column was switched 386 in-line with a reversed-phase analytical column, Hypersil GOLD, particle size 1.9 µm, 1 x 50 387 mm, and separation was performed at 1°C using a gradient of 5-50 % B over 8 minutes and 388 then from 50 to 90% B for 5 minutes, the mobile phases were 0.1 % formic acid (A) and 95 % 389 acetonitrile with 0.1 % formic acid (B). Following the separation, the trap and column were 390 equilibrated at 5% organic content, until the next injection. The needle port and sample loop 391 were cleaned three times after each injection with mobile phase 5%MeOH and 0. Peptides identified by PEAKS with a peptide score value of log P > 25 and no modifications 410 were used to generate peptide lists containing peptide sequence, charge state and retention time version 3.01 (Sierra Analytics Inc., Modesto, US). Due to the comparative nature of the 413 measurements, the deuterium incorporation levels for the peptic peptides were derived from the 414 observed mass difference between the deuterated and non-deuterated peptides without back-415 exchange correction using a fully deuterated sample. HDX data was normalized to 100% D2O 416 content with an estimated average deuterium recovery of 75%. The peptide deuteration was 417 determined from the average of all high and medium confidence results, with the two first 418 residues of each peptide set to be unable to retain deuteration. The allowed retention time 419 window was set to ± 0.5 minutes. Heatmaps settings were uncoloured proline, heavy smoothing 420 and the difference heatmaps were drawn using the residual plot as significance criterion (±1 421 Da). The spectra for all timepoints were manually inspected; low scoring peptides, e.g. obvious 422 outliers and peptides were retention time correction could not be made consistent were 423 removed. 424

425
Human plasma protein interaction networks with S. pyogenes surface proteins 426 M proteins are long extended surface attached proteins with various combinations of A, B, C 427 and D domains (Fig. 1A) that allow the M proteins to engage in numerous protein interactions 428 simultaneously. While the protein interaction network formed around A-C patterns is relatively 429 well described 16 , less is known about the protein interaction network organized around E pattern 430 strains. Here, we combined quantitative and structural mass spectrometry techniques, to 431 determine how the different M protein domains within E and A-C patterns influence the 432 composition and structure of the human plasma-S. pyogenes interaction network (Fig. 1B).  Fig. 2A). The quantitative data matrix across each strain show that there are marked 443 differences in the HP-PPI networks formed on the streptococcal surface between the serotypes 444 producing A-C or E type M proteins ( Fig. 2A). The A-C pattern strains typically bind fibrinogen 445 and components of the complement system, whereas the E pattern typically bind with various 446 apolipoproteins, immunoglobulins and components from the complement and coagulation 447 system such as C4BP and vitamin K-dependent protein S (PROS) (Fig. 2A). The analyzed 448 strains were furthermore capable of forming distinct serotype-specific interaction networks, 449 also within their respective A-C or E patterns. To objectively determine the major components 450 of these networks, we used co-expression network analysis (Fig. 2B). This analysis revealed 451 four highly connected protein clusters (highlighted using semi-transparent circles) of strongly 452 correlating human proteins (blue lines, r 2 > 0.9) that bind to one or two of the strains. The 453 highest correlating proteins associated with M49 and M89 networks were several 454 are several proteins that are negatively correlated (r 2 < -0.6) as indicated by the red lines in 458 Figure 2B. To further visualize these binding patterns, correlation plots were plotted for 459 selected proteins pairs from each protein cluster (Fig 2C). As expected, strong correlations were 460 observed between proteins belonging to the same protein cluster in Figure 2B such as APOH-461 APOC4, FIBB-FIBA, C4BPA-C4BPB, APOB-PROS, IGHA1-IGHA2, CO3-CO4A, C4BPA-462 PROS and IGHA1-C4BPA. In contrast, other proteins appear to bind significantly more to some 463 strains such as fibrinogen and C4BP, where M1, M3 and M5 bind fibrinogen but not C4BP, 464 while M28, M49 and M89 bind C4BP but not fibrinogen (Fig. 2C). Moreover, high levels of 465 fibrinogen were related to low levels of several other proteins of the M28, M48 and M89 466 network such as IgA, PROS and components of apolipoproteins such as APOH (Fig. 2C). These 467 results demonstrate that each strain can assemble strain-specific HP-PPI network and that there 468 are substantial differences in the interaction networks between emm types. 469

Human plasma protein interaction networks with S. pyogenes -M proteins 470
To understand to what degree differences in the domain arrangement of the M proteins (Fig.  471 1A) mediates the differential binding patterns of human proteins to the S. pyogenes strains, we 472 applied protein affinity purification mass spectrometry (AP-MS) 16 as schematically shown in 473 proteins emm type E (M28, M49 and M89) (Fig. 3B) in a similar fashion as observed in the 486 SA-MS results above and as previously suggested by Sanderson et. al. 13 . To detail the properties 487 of the differential binding patterns, we constructed another correlation network plot for the 32 488 proteins across the five different M proteins (Fig. 3C). Similar to the SA-MS results, the one or two of analyzed M proteins. Several of the serotype specific proteins shown above such 491 as fibrinogen, apolipoproteins, C4BP and IgA, are strongly associated with particular M 492 proteins. In addition, we can confirm that binding of some proteins seems to result in lower 493 binding of other proteins (r 2 < -0.5) such as fibrinogen-PROS and fibrinogen-C4BPα, 494 demonstrating that the interactions captured above using SA-MS are to a large degree mediated 495 by the M proteins. To visualize the core-interaction network between the analyzed M proteins, 496 we selected the highly enriched protein interactions (log2 > 3 compared to GFP) to plot a 497 schematic interaction network graph (Fig. 3D). The network graph reveals that albumin, IgG1 498 and IgG4 are equally associated with all analyzed M proteins. Albumin is known to bind the 499 conserved C-repeats 14,20,23,24 of the M protein, thus making the association of albumin with all 500 M proteins logical. In addition, IgA2 is enriched in all M proteins although significantly more 501 enriched to M28, which is also coupled to C4BPα, IgA1, alpha-1-antitrypsin (A1AT) and to 502 lesser degree PROS. The cysteine residue on the C terminus of α chain of monomeric IgA has 503 been shown to form disulfide bonds with A1AT 52 and C4BP is known to form complex with 504 PROS 53 . We speculate that these proteins form a larger complex mediated via human-human 505 protein interactions on M28. In contrast, M1 and M3 typically bind fibrinogen and fibronectin, 506 whereas M49 binds several components of the complement system, and both M49 and M89 507 bind PROS. In conclusion, the results from the AP-MS analysis demonstrate that M proteins 508 play a major role in shaping the serotype-specific HP-PPI networks observed in SA-MS. 509 Although the E type M proteins are substantially smaller compared to A-C types, their 510 interaction networks with human plasma proteins are still surprisingly complex. As there are 511 no structural model for any E type M interspecies protein complex, we selected M28 for further 512 structural characterization with a particular focus on the binding with IgA and C4BP as outlined 513 in Figure 1B. 514

Characterization of the M28 IgA-C4BP interaction in different local microenvironments 515
As we observed that IgA was significantly enriched on M28, we measured the affinity between 516 M28 and IgA by using surface plasmon resonance (SPR). The binding of M28-IgA was 517 compared to M1-IgA binding, which according to our observation showed very low or no IgA 518 binding. We immobilized the M proteins (ligand) on the sensor chip and injected IgA (analyte) 519 over them to mimic the M proteins protruding out from the bacterial surface and the 520 immunoglobulins floating in the plasma. The kinetic analysis showed the best fit to a 521 heterogeneous ligand model compared to a 1-1 model (Fig. S3, A & B). Surface heterogeneity analyte. Thus, an explanation for the deviations from a 1-1 fitting model could be that IgA has 524 multiple binding sites on M28. Calculated affinity constants showed that IgA had a 3 log higher 525 affinity for M28 as compared to M1 (KD1≈10 -10 M and KD2≈10 -8 M for M28 and KD1≈10 -7 M 526 and KD2≈10 -7 M for M1) (Fig. 4A-I & II). The differences in KD1 and KD2 values of IgA 527 towards M28 support two different binding sites on M28 for IgA, one with high and the with 528 lower affinity. We also performed SPR analysis of the interaction with C4BP and M28 since 529 C4BP was significantly enriched on M28 in our SA-MS and AP-MS experiments above. In this 530 case, kinetic analysis showed a better fitting to a 1-1 model (Fig S3, C). Affinity constants 531 calculated from the sensorgrams resulted in a KD of 1.88 x 10 -10 , suggesting one single binding 532 site with a high affinity between M28 and C4BP (Fig. 4A-III). 533 Most IgA produced in the human body is secreted into the mucus membrane thereby acting as 534 a first line defense against infections 27 . To understand how an IgA rich microenvironment alters 535 the S. pyogenes M28 protein network, we quantified the protein interaction network of M1 and 536 M28 in pooled normal human saliva by AP-MS. These experiments showed that IgA binding 537 from saliva only occurs on M28 but not on M1 (Fig. 4B). Additionally, we observed co-538 enrichment between IgA and polymeric immunoglobulin receptor (PIGR) and IGJ (Fig. 4B). 539 PIGR is known to bind polymeric IgA and IgM at the basolateral surface of epithelial cells. or plasma mixtures, although the concentration of IgA1 is lower in plasma (Fig 4C-I). In 552 contrast, IgA2 binding to M28 predominantly occurs in saliva and decreases with the decreasing 553 & IV). These results imply that M28 can bind IgA in both sIgA and monomeric form where 557 the former is pronounced in saliva. The higher levels of AP-purified IGJ and PIGR compared 558 to the input pool in 10% plasma environment, with nearly 18 times higher plasma protein 559 concentration compared so saliva, suggest that the sIgA binds with high affinity which is in 560 contrast to previously published results 56 . The mixed saliva-plasma enrichment comparison of 561 M28 additionally revealed elevated levels of C4BP on M28 only at high plasma concentrations 562 ( Fig. 4C-V). Interestingly, although there were detectable levels of C4BPA in 1% plasma, there 563 was no strong enrichment of C4BPA to M28 at this low plasma concentration. These results 564 are surprising as the SPR analysis showed that the affinity between M28 and C4BPA was in 565 the sub nanomolar range. Possibly this could be accounted to the fact that the levels of secretory 566 IgA were still high at 1% plasma. The levels of IgG1 enriched on M28 seemed to increase with 567 increase in plasma concentration (Fig. 4C-VI). Collectively, these results show that M28 binds 568 secretory IgA in saliva and C4BP binding on M28 only becomes accentuated in the absence of 569 secretory IgA. 570

Structural determination of M28 with IgA and C4BP 571
To understand how the shorter E type M28 binds secretory IgA in saliva and monomeric IgA 572 and C4BP in plasma, we performed targeted cross-linking mass spectrometry (TX-MS 17 ) and 573 hydrogen-deuterium mass spectrometry (HDX-MS) of M28 in complex with C4BP or the Fc-574 domain of IgA. As the input for TX-MS-based structural modelling, we first generated a 575 computational model of the full-length M28, which was determined using the Rosetta 576 comparative modeling protocol 49 based on the previously reported model of the M1 protein 26 . 577 This model was further used to provide protein-protein docking decoys using structures 578 deposited in protein data bank for IgA (PDB 6LXW) and C4BP (PDB 5HYP). For TX-MS, 579 the affinity-tag of M28 was removed, and the untagged protein was cross-linked individually 580 in solution to either C4BP or the Fc-domain of IgA. The cross-linked peptides observed 581 between M28 and C4BP overlapped with the interaction interface resolved using X-ray 582 crystallography 57 (Fig. 5A-B, Fig. S4, ST2). These cross-links were observed between two 583 C4BP residues (K28 and K67 in PDB 5HYP; corresponding residues K72 and K111 in the full-584 length C4BPα chain) and K50 on our M28 construct (Fig. 5A-B, ST1). No cross-links from 585 C4BP were observed to the crystallized M28 segment (Fig. 5A-B), most likely due to the lack 586 of stereo chemical favorable lysine residues at the interaction interface. For IgA, we identified 587 two different cross-linked sites by TX-MS, the first one was supported by four inter-protein (Fig. 5A, C-I, Fig. S4 & ST2). In addition, TX-MS also identified a novel IgA-Fc interface in 590 the middle of M28 supported by eight high-confident inter-protein cross-links ( Fig. 5C-II, Fig.  591   S4, ST2). The two binding sites between IgA and M28 could result in the binding of either two 592 single IgA-Fc's ( Fig. 5CI-II) or one sIgA molecule, where a dimeric IgA is bridged by a J-593 chain and a secretory component (Fig. 5C-III). Using a recently determined structure for sIgA 58 594 as input for TX-MS, we showed that the binding between sIgA and M28 is supported by five 595 unique inter-protein cross-links (Fig. 5C-III). The binding of sIgA onto two separate and 596 possibly synergistic binding sites on M28 could explain why sIgA binding was more 597 pronounced in the AP-MS experiments compared to C4BP as shown above (Fig 4C). interacting with C4BP (Fig. 5A, D), enclosed between the C4BP-binding site in the crystallized 602 complex (PDB 5HYP), and the cross-linked site (K50) identified by TX-MS (Fig. 5A). HDX-603 MS analysis of the M28-IgA Fc-domain interaction showed strong protection to deuterium 604 uptake at two distinct sites. At an M28 to IgA ratio of 1:1 the reduction in deuterium uptake 605 was observed at the SAP-peptide and the overlapping region identified by TX-MS (Fig. 5A, C-606 I, E-I). A reduction in deuterium uptake was furthermore observed for residues 112-128 at an 607 M28 to IgA ratio of 2:1, especially at short labeling times (Fig. 5A, E-II), indicating a lower 608 affinity site as suggested by the SPR data (Fig 4B). Importantly, this latter M28 site protected 609 to deuterium uptake overlaps with the IgA-Fc interface identified by TX-MS (Fig. 5A, C-II). 610 Taken together, our AP-MS data in combination with the integrative structural mass 611 spectrometry approach, allowed us to propose two distinct models for the M28 interactions. In 612 one model two single IgA Fc-monomers and a C4BP-moculecule would simultaneously bind 613 to M28 (Fig. 6A) and in the other one the IgA-binding sites would be occupied by sIgA alone 614 (Fig. 6B)  The SAP peptide derived from M22 27 was previously shown to harbor an IgA binding site. In 648 our study, M28 is the only M protein which contains the SAP-peptide sequence (Fig. S1). IgA 649 is the most abundant immunoglobulin on the mucosal surface. As S.pyogenes are known to 650 localize in mucosal surfaces, hence strong binding of IgA to M28 could be warranted and likely mucosal surface. In fact, the M28 serotype has been reported to be one of the leading causes of 653 puerperal sepsis 59-62 . Persistent infections of the mucosal membrane by S.pyogenes can induce 654 vascular leakage thereby providing access of the bacterium to human plasma. Here we mimic 655 a localized infection condition followed by a systemic infection and we observe that under such 656 circumstances, the M28 interaction network gradually changes its composition from 657 predominant binding of secretory IgA in saliva to monomeric IgA and C4BP in plasma. This 658 change is driven by the differences in protein concentration in the host microenvironment. 659 However, even at higher plasma concentrations (10% plasma), secretory IgA is enriched to a 660 higher extent to M28 compared to the input sample, whereas C4BP is not enriched to the same 661 extent under these conditions. Typically, bacteria-host relationships are well-balanced. Sepsis 662 is a relatively rare condition compared to uncomplicated local infections, implying that the 663 evolution of bacteria-host relationships is predominately taking place in local host 664 microenvironments and not in blood as previously proposed 25 . In local microenvironments, 665 secretory IgA is the major immunoglobulin. Our results support the following three models in 666 a mucosal niche: i) one dimeric IgA occupying both the IgA binding sites on M28; ii) two 667 dimers binding separately to the two sites and lastly; iii) one dimer and one monomer could be 668 engaged on M28 (Fig. 6C) IdeS 65 to circumvent IgG effects. This change in local microenvironment may therefore drive 674 binding to C4BP along with monomeric IgA (Fig. 6C). It has been reported that binding of both 675 IgA and C4BP to a M protein is crucial in inhibiting phagocytosis 31 . C4BP is known to bind to experiments. We propose that M28 binds either secretory IgA or monomeric IgA and C4BP depending whether they cause a localized infection or a systemic infection (Fig. 6C)   using a log2 fold enrichment of > 1 with an adjusted P-value of 0.05 using the student t-test. 744 The red dots represent high-confident interactors while the grey did not pass the above filtering  close-up view of the cross-linked site identified between M28 (grey helix) and C4BP (blue). 773 The interaction interface on the crystallized M28 segment (PDB 5HYP) is shown in cyan, and 774 the SAP-peptide interacting with the IgA Fc-domain in yellow. Cross-links are observed 775 between lysine residues K72 and K111 (numbered based on the full-length C4BPα chain) and 776 K50 on our M28 construct. The cross-links are depicted as dotted lines, with the labels 777 corresponding to a given spectrum in Figure S4  with the labels corresponding to a given spectrum in Figure S4 and supplementary table 2 784 (ST2). (II) The novel interaction site between M28 (grey helix) and the IgA Fc-domain (red). 785 The cross-links are depicted as dotted lines, with the labels corresponding to a given spectrum 786 in Figure S4  The cross-links are depicted as dotted lines, with the labels corresponding to a given spectrum 789 in Figure S4 M1  M1  M1  M1  M3  M3  M3  M3  M3  M5  M5  M5  M5  M5  M5  M28  M28  M28  M3  M28  M28  M28  M89  M89  M89  M89  M89  M89  M49  M49  M49  M49  M49  M49   APOA  IGHG3  IGHG2  IGHG4  APOE  APOA4  SPTA1  SPTB1  APOA1  THRB  PLMN  C4BPA  PROS  C4BPB  IGHG1  IGHA2  IGHA1  APOB   PON1  APOA2  APOH  APOC4  APOC2  FBLN3  MYH10  C1QB  AFAM  F13A  ANK1  GELS  F13B  FBLN1   ITIH1  KNG1  AMBP  HPT  TRFE  CO8G  HBB  AACT  CERU  A1AG1  A1AG2  HEMO  A1BG  FETUA  ITIH2  HPTR  ALBU  VTDB  CFAH  FHR1  C1R  HRG  FINC  B3AT  ITIH4   CO9  CO8B  CO5  CO8A  CO7  PROP  CFAB  A2MG  PZP  CLUS  CO6  CO3  CO4A  CO4B   ACTB  ACTG  FIBB  ACTS  ACTC  ACTA  ACTH  ITB3  FIBG  VTNC  FHR5  APOL1  CD5L   MYH9  TSP1  FLNA  ITA2B  FIBA  C1S  FA5