Human CEACAM1 is targeted by a Streptococcus pyogenes adhesin implicated in puerperal sepsis pathogenesis

Life-threatening bacterial infections in women after childbirth, known as puerperal sepsis, resulted in classical epidemics and remain a global health problem. While outbreaks of puerperal sepsis have been ascribed to Streptococcus pyogenes, little is known about disease mechanisms. Here, we show that the bacterial R28 protein, which is epidemiologically associated with outbreaks of puerperal sepsis, specifically targets the human receptor CEACAM1. This interaction triggers events that would favor the development of puerperal sepsis, including adhesion to cervical cells, suppression of epithelial wound repair and subversion of innate immune responses. High-resolution structural analysis showed that an R28 domain with IgI3-like fold binds to the N-terminal domain of CEACAM1. Together, these findings demonstrate that a single adhesin-receptor interaction can drive the pathogenesis of bacterial sepsis and provide molecular insights into the pathogenesis of one of the most important infectious diseases in medical history.

Streptococcus (GAS) 5,6 , but the molecular basis of this disease has remained unknown. Epidemiological studies have repeatedly indicated that an antigen designated R28 7,8 , which has properties as an adhesin 9 and typically is found in emm28 strains, is associated with S. pyogenes strains causing puerperal sepsis outbreaks [10][11][12][13][14][15] . A key question therefore arises: what is the role of R28 and does it allow S. pyogenes to overcome innate defense mechanisms to cause puerperal sepsis? This problem is of interest given the global increase in puerperal sepsis cases caused by S. pyogenes 2,16 , and the lack of vaccines against this pathogen. Moreover, insights about puerperal sepsis caused by S. pyogenes may provide information relevant to the pathogenesis of puerperal sepsis caused by other bacterial pathogens.
The human immunomodulatory receptor CEACAM1 (carcinoembryonic antigen-related cell adhesion molecule 1) binds to some bacterial proteins that contain an Ig-like fold designated IgI3 17 . The CEACAM1 receptor, which belongs to the CEACAM family of proteins, is widely expressed on human epithelial, endothelial, and immune cells 18 and modulates cellular activities 19,20 (Fig. 1a). Here, we show that R28, which possesses an IgI3 domain, specifically targets CEACAM1 of human origin to promote events that would collectively favor the development of puerperal sepsis. These events include enhancing bacterial adhesion to epithelial cell barriers, interfering with epithelial wound healing, subverting the innate immune response, and promoting bacterial replication in blood. While it is classically assumed that a pathogen employs distinct molecular interactions to overcome each defense mechanism of the host, our data demonstrate that a single interaction can allow a pathogen to overcome multiple host defense barriers on the pathway to invasive disease. Fig. 1 | The R28 adhesin of puerperal sepsis S. pyogenes isolates interacts with CEACAM1. a Schematic of CEACAM1 (CC1), containing the extracellular region (composed of four Ig-like domains termed N, A1, B1, and A2) and cytoplasmic tail containing two immunoreceptor tyrosine-based inhibitory motifs (ITIM) for signaling. b Schematic of the S. pyogenes R28 and S. agalactiae Rib proteins. Each protein has a signal peptide (S) and cell wall anchor (#) domains. R28 contains the N2 domain that can be divided into N2 A and N2 B (renamed IgI3) subdomains. The percentage residue identity between domains is indicated. c Binding of recombinant (r)CC1-Fc to immobilized R28 and Rib, quantified by ELISA (mean ± s.d. of n = 3 independent experiments). Statistical significance tested by two-tailed paired Student's t test (**p = 0.0079). d Binding of rCC1-HIS variants to R28-IgI3, quantified by flow cytometry (mean ± s.d. of n = 6 independent experiments) e Core genome phylogeny of 287 representative S. pyogenes genomes. spr28 positivity (inner circle) and emm-type (outer circle) given. Scale: substitutions per site. f Expression of surface-localized R28 by S. pyogenes strains, quantified by flow cytometry (Data are represented as mean ± s.d.). g Binding of rCC1-HIS to S. pyogenes strains as in f, quantified using flow cytometry (Data are represented as mean ± s.d.). In f and g, each data point represents the mean from n = 2 independent experiments per strain (n = 30 emm28 strains, n = 11 emm2 strains, n = 10 emm77 strains). Closed circles show spr28 + strains and open circles show isogenic Δspr28 strains. Open squares show spr28strains, typed by PCR. Dashed lines indicate signal from an R28negative strain (S. pyogenes M1 5448) or a CC1-binding strain (S. agalactiae A909). h Adherence of isogenic S. pyogenes AL368 strains to CHO cell lines (mean ± s.d. from n = 8 independent experiments). Statistical significance tested by one-way ANOVA with Tukey's post-hoc test (+R28 control vs. +R28 CC1 ***p = 0.0006, +R28 CC1 vs. -R28 CC1 ****p = 0.0001).

S. pyogenes R28 adhesin binds CEACAM1
The R28 protein includes a long series of identical repeats and a unique N-terminal region 9 , in which the IgI3 fold is contained in the N2 domain (Fig. 1b). We performed biochemical analysis of the interaction between intact R28 and CEACAM1 proteins, and found that recombinant (r)CEACAM1 bound to immobilized R28 isolated from the S. pyogenes cell wall, and also to β protein of S. agalactiae as previously described 17 . In contrast, rCEACAM1 did not bind to other purified streptococcal surface proteins tested including Rib of S. agalactiae, which is closely related to R28 but lacks the N2 part and the IgI3 region (Fig. 1b, c and Supplementary Fig. 1a) 9,21 . In a complementary test, purified R28 interacted directly with immobilized rCEACAM1 in a concentration-dependent manner ( Supplementary Fig. 1b). The IgI3 domain of R28 was sufficient for the interaction ( Supplementary  Fig. 1c), and the N-terminal domain of CEACAM1 was necessary and sufficient for the interaction (Fig. 1d), as demonstrated in direct binding assays. In agreement with these findings, the binding of R28-IgI3 was specifically blocked by an antibody to CEACAM1-N (Supplementary Fig. 1c) and by the Helicobacter pylori protein HopQ, which binds to CEACAM1-N ( Supplementary Fig. 1d) 22 . These data indicate that R28 and CEACAM1 interact through their IgI3 and N domains, respectively.
As R28 is strongly over-represented among S. pyogenes strains causing puerperal sepsis 8,[10][11][12]14,15 , it was of interest to analyze whether clinical S. pyogenes isolates interact with CEACAM1. We first analyzed the distribution of the gene (spr28) encoding R28 in whole genome sequences of 3886 S. pyogenes isolates representing 77 emm-types, and found that isolates belonging to emm2, emm28, and emm48 lineages, and one of two emm77 lineages, frequently carry the spr28 gene (Fig. 1e). Among these lineages, isolates belonging to emm28 (serotype M28) are most common among strains causing outbreaks of puerperal sepsis 11 . Our analysis supports previous observations 13,23 , and agrees with the finding that the spr28 gene is located on a mobile genetic element 24 . Carriage of spr28 gene, surface expression of the R28 protein and ability to bind CEACAM1 was evaluated for a large collection of strains belonging to emm28, emm2, or emm77 lineages, which included puerperal sepsis isolates. Of these, 28/28 emm28, 9/11 emm2, and 8/10 emm77 isolates carried spr28. Expression analysis ( Supplementary  Fig. 1e) showed that all strains carrying the spr28 gene expressed R28, with highest expression in the emm28 strains (Fig. 1f), which included a puerperal sepsis isolate from the 1930s. Binding of CEACAM1 was observed for most of the S. pyogenes isolates (Fig. 1g, Supplementary  Fig. 1f), and was concentration-dependent ( Supplementary Fig. 1g). As expected, two spr28-negative S. pyogenes mutants lacked surface R28 expression and CEACAM1 binding (Fig. 1f, g and Supplementary Fig. 1e, f, h) 9 . The variation in CEACAM1 binding is likely due to differences in R28 expression, as supported by a positive correlation ( Supplementary  Fig. 1h), due to mutations in an intergenic region that impacts spr28 transcription levels 25 . To analyze binding under more physiological conditions, we studied the ability of R28-expressing S. pyogenes to adhere to the surface of Chinese Hamster Ovary (CHO) cells and found that expression of human CEACAM1 enhanced the interaction through an R28-dependent mechanism ( Fig. 1h and Supplementary Fig. 1i). These data demonstrate that R28-expressing S. pyogenes strains have an enhanced ability to adhere to epithelial cells through interaction with human CEACAM1.

R28 binds CEACAM1 with high specificity
To investigate the specificity of CEACAM1 as a receptor for the R28 adhesin, we assessed whether R28 interacts with other human surface-expressed membrane proteins. High-throughput screening of the binding of R28 and the closely related Rib protein to 3359 cell line clones (derived from 2625 unique genes; Supplementary Data 1) was detected with rabbit anti-Rib serum (Fig. 2a), which detects both R28 and Rib 9 . The screen revealed enhanced fluorescent signals for 45 cell lines compared to the control cell line (Fig. 2b). Further characterization of these hits revealed that the R28 protein, but not Rib protein, bound specifically to a CEACAM1-expressing cell line only (Fig. 2b).
Twelve CEACAM1 splice variants with different features can be expressed in human cells, including those that are membrane anchored or those that are soluble. They belong to a sub-family of CEACAM receptors, which share high sequence (~90% identity) and structural homology. However, CEACAM1-4L is the main expressed isoform and the only member containing the ITIM domains and subsequent immunomodulatory functions 18 . Because R28 did not bind to HEK293 cells expressing the CEACAM1 related receptors (CEACAM3, CEACAM5, CEACAM6, and CEACAM8; Supplementary Data 1), we tentatively concluded that R28 binds with high specificity to CEACAM1 on target cells. Biochemical and cellular assays were performed with purified preparations of R28 and the other CEACAM receptor family members. No interaction was detected between R28 and rCEACAM3, rCEACAM5, rCEACAM6 or rCEACAM8 in an ELISA (Fig. 2c) or in isothermal calorimetry assays ( Supplementary Fig. 2). Additionally, R28expressing S. pyogenes bacteria did not bind to rCEACAM3, rCEACAM5, rCEACAM6, or rCEACAM8 in flow cytometry assays (Fig. 2d). Finally, a significantly higher percentage of an R28-expressing S. pyogenes inoculum adhered to CEACAM1-expressing HeLa cells, as compared to HeLa cell lines expressing other CEACAMs (Fig. 2e). Taken together, these data reveal that CEACAM1 on epithelial cells is a highly specific receptor for the R28 adhesin of S. pyogenes.

Structure of the R28-CEACAM1 complex
Structural information is essential, not only to understand how R28 targets its cognate receptor CEACAM1, but also to understand how R28 might be targeted by anti-virulence strategies. We solved the structure of R28-IgI3 bound to CEACAM1-N at a resolution of 3.05 Å ( Fig. 3a; Supplementary Fig. 3a, b; and Supplementary Table 1). The asymmetric unit contains four copies of the (R28-IgI3)-(CEACAM1-N) complex ( Supplementary Fig. 3a, the four paired molecules are A-B, C-D, E-F and G-H). Superposition of the complexes revealed little difference between them (r.m.s.d. 0.188-0.406 Å; Supplementary Table 2), and all analyses were performed using chains E and F ( Supplementary Fig. 3b). The R28 domain has an Ig-like fold belonging to the IgI3 subset, which we recently identified as a domain in the β protein from S. agalactiae that binds CEACAM1 and CEACAM5 17 . Notably, R28-IgI3 has the following properties of the IgI3 fold: A' and A" strands, lack of C′ and C′′ strands, and a truncated C strand ( Supplementary Fig. 3c). A notable difference between the structure of the R28-IgI3 domain and the β-IgI3 domain is the presence in R28-IgI3 of a helix-turn-helix (HTH) motif between the C and D strands ( Supplementary Fig. 3b-d).
We next assessed the mechanism of interaction between R28-IgI3 and CEACAM1-N. In the complex, R28-IgI3 targets the dimerization interface of CEACAM1-N (the A′GFCC′C′′ face) using a pocket of residues (K45, Y61, K64) that have high electrostatic potential ( Fig. 3a; Supplementary Fig. 3e; and Supplementary Data 2). We further identified residues A55, E49, I52, I53, and I59 in R28-IgI3 as potential contact sites of CEACAM1-N ( Fig. 3a; Supplementary Fig. 3f; and Supplementary Data 2) and tested alanine substitutions in these and additional R28-IgI3 residues to evaluate their importance in CEACAM1-N binding using biochemical assays. Significant reductions in binding of CEACAM1-N were observed for Dynabeads coated with R28-IgI3 mutated at K45, I52, I53, and Y61 ( Fig. 3b and Supplementary  Fig. 3g). These results were confirmed by ITC analysis (Supplementary Fig. 4a and Supplementary Table 3). Next, we tested whether alanine substitutions in CEACAM1-N residues altered R28-IgI3 binding (Fig. 3a, c and Supplementary Fig. 3f). Mutation of CEACAM1-N at residues F29, I91 or L95 abolished the interaction with R28-IgI3 (Fig. 3c), while mutation of CEACAM1-N at Q89 reduced binding to R28-IgI3. Additionally, mutation of residues V96 or N97 significantly increased binding of CEACAM1-N to R28-IgI3. ITC analysis confirmed these findings (Supplementary Fig. 4b and Supplementary Data 3). Taken together, our structural and biochemical analysis indicates that R28-IgI3 docks over the CEACAM1 cis/trans dimerization site of the A′GFCC′C′′ face (residues F29, Q89, I91 and L95), and that the unique structural features of R28-IgI3 provide a site with high electrostatic charge (K45 and Y61) and high stability (I52 and I53) for CEACAM1 binding ( Fig. 3d; Supplementary Fig. 3h). Finally, we investigated the presence of sequence variation in R28-IgI3 that could influence the capacity to interact with CEACAM1. The R28-IgI3 region was conserved in the genomes of those we tested for spr28 positivity (Fig. 1e) and BLAST analysis against the entire NCBI S. pyogenes protein database confirmed a high level of conservation with just a single variant of valine to alanine at amino acid residue 26. This residue is not located in or near the CEACAM1 binding interface suggestive that all R28 proteins interact with CEACAM1.
As R28-IgI3 binds to the trans-dimerization interface (the A′GFCC′ C′′ face) of CEACAM-N, we questioned whether the binding of R28-IgI3 to CEACAM1 could prevent CEACAM1 trans-dimerization interactions. Superimposition of the crystallographic CEACAM1 trans dimer (PDB: 4WHD) with the R28-IgI3 complex shows that R28-IgI3 binding is incompatible with CEACAM1 trans-dimerization (Supplementary Fig. 5a, b). Similarly, the binding of other bacterial ligands is incompatible with CEACAM1 trans-dimerization ( Supplementary  Fig. 5c, d, e) 26 . These data suggest that infection with R28 + S. pyogenes, but not R28-S. pyogenes, would reduce cross-linking efficiency of CEACAM1.
Human and macaque CEACAM1 only differed at one of the four residues shown to be critical for the R28-IgI3 interaction (residue F29 in human CEACAM1). Simulated docking of R28-IgI3 onto wildtype and mutated CEACAM1-N domain structures ( Supplementary Fig. 6c) correctly predicted that R28-IgI3 binds to human, but not macaque CEA-CAM1 (Fig. 3g). Moreover, in silico analysis also revealed that an F29I mutation in human CEACAM1 significantly reduced the R28-IgI3 binding energy, while an I29F mutation in macaque CEACAM1 significantly increases the R28-IgI3 binding energy (Fig. 3g). Thus, the analysis indicated that the F29 residue in human CEACAM1 is a critical determinant of human-specificity, likely through interaction with residues K45 and I52 on R28-IgI3 (Fig. 3h). These results show that R28 specifically binds to human CEACAM1 and indicate that results from experimental animal models likely underestimate the role of R28 in S. pyogenes virulence.

R28 impairs wound repair
As CEACAM1 is expressed on female reproductive tract epithelia (Supplementary Fig. 7a) 18,28 , we tested the hypothesis that R28 enhances the adhesion of S. pyogenes to these epithelial surfaces through engagement of CEACAM1. Adhesion of our reference S. pyogenes strain AL368 to monolayers of the ME-180 cell line, a human cell line of cervical origin that expresses CEACAM1 (Supplementary Fig. 7b, c) 29 , was dependent on R28 9 and was specifically reduced by an antibody that recognizes the N-domain of CEACAM1 (Fig. 4a). Similar observations were made for an alternative isogenic pair of S. pyogenes strains ( Supplementary Fig. 7d). Thus, R28 promotes binding of S. pyogenes to human cervical cells by binding CEA-CAM1. Similar results were obtained in assays measuring the binding of fluorescently labeled S. pyogenes to detached ME-180 cells (Supplementary Fig. 7e, f, g).
The results obtained with the ME-180 cell line demonstrated that R28 interacts with the CEACAM1 receptor present on the surface of cervical epithelial cells. However, to cause puerperal sepsis, S. pyogenes must enter female reproductive tract tissue through the wounds generated during childbirth, subvert the innate immune response and resist killing by innate immune cells upon entry into the blood. We speculated that R28 was involved in some of these processes. To analyze the contribution of the R28-CEACAM1 interaction to the entry of S. pyogenes into tissues through wounds, we employed an in vitro scratch assay that is routinely used to investigate wound healing 30 . Notably, previous results had demonstrated that CEACAM1 regulates cell migration and wound repair [31][32][33] , although it has never been proposed that S. pyogenes (or R28) could modulate this process. Scratches inflicted to ME-180 cell monolayers closed by 40.1% (±11.7) after 24 h of culture and by 66.8% (±14.0) after 48 h of culture (Fig. 4b, c), but repair of the wound was strongly delayed after infection with R28-expressing S. pyogenes, while the R28-negative mutant had little or no effect. Furthermore, blocking access to the N-domain of CEACAM1 on the ME-180 cells restored the repair of the wound upon S. pyogenes infection (Supplementary Fig. 8a, b). Wound repair was also delayed in an R28dependent manner upon infection with an alternative isogenic pair S. pyogenes ( Supplementary Fig. 8c). To understand whether the R28-CEACAM1 interaction was sufficient to induce the delayed wound repair phenotype in the absence of other bacterial components, we investigated the capacity of purified R28 to delay wound repair. Repair of the wounds in ME-180 cell layers was delayed by incubation with purified R28, but not the closely related Rib protein of S. agalactiae (Fig. 4d). The phenotype was also elicited by rR28-IgI3 (Fig. 4e), indicating that the IgI3 domain interaction with ME-180 cells was sufficient to delay wound repair. To investigate whether the rR28-IgI3 domain acted through a CEACAM1-dependent mechanism, we compared the functional properties of rR28-IgI3 variants that bind or do not bind CEACAM1 (as revealed in Fig. 3b). Importantly, CEACAM1 binding rR28-IgI3 proteins (rR28-IgI3 and rR28-IgI3 E49A ) delayed wound repair, whilst the CEACAM1 non-binding rR28-IgI3 Y61A did not delay wound repair (Fig. 4e). Though there was not a significant difference in wound repair between rR28-IgI3 and rR28-IgI3 I53A , there was a trend. Collectively, these data demonstrate that interaction between S. pyogenes and human CEACAM1 through R28 promotes adhesion to epithelial cells and suppresses wound repair.

R28 suppresses innate immune responses
The next defense mechanisms that S. pyogenes must pass is the innate immune response of a variety of immune cells that reside within the epithelium and lamina propria. Since CEACAM1 is an immunomodulatory receptor expressed on immune cells that regulates anti-bacterial responses 19,20 , we hypothesized that expression of R28 by S. pyogenes would impair the anti-streptococcal immune response of tissues. To test this, we assessed the cytokine and chemokine profiles of wounded ME-180 cell monolayers and underlying human peripheral blood mononuclear cells (PBMCs) using transwell culture chambers. Nondividing S. pyogenes strains were added to the apical chamber and an interaction was allowed to progress for 72 h prior to analysis of cytokine and chemokine production ( Supplementary Fig. 9a). The immune response against wildtype S. pyogenes was impaired in comparison to that raised against the isogenic R28-negative strain ( Supplementary  Fig. 9b). Notably, there was a significant reduction, or trend for reduction, of the production of proinflammatory cytokines and chemokines including IL-1β, TNF-α, GM-CSF, IFN-γ, RANTES, IL-6, and MCP-2. To further investigate the immunomodulatory effect of R28 in Fig. 3 | Structural basis and human specificity of R28-IgI3 interaction with CEACAM1. a Crystal structure of N-terminal domain of CEACAM1 (CC1) domain in complex with R28-IgI3, shown as an electron density map. b Mutation of residues K45, I52, I53 and Y61 in R28-IgI3 abolishes binding to CC1-N (Mean ± s.d. of n = 6 independent experiments). Statistical significance tested between R28-IgI3 wildtype and variants by one-way ANOVA with Dunnetts's post-hoc test (WT vs. K45A *p = 0.0219, WT vs. I52A *p = 0.0210, WT vs. I53A *p = 0.0214, WT vs. Y61A *p = 0.0257). c Mutation of residues F29, I91 and L95 in CC1-N abolishes binding to R28-IgI3 (Mean ± s.d. of n = 6 independent experiments). Statistical significance tested between CC1-N wildtype and variants by one-way ANOVA with Dunnetts's post-hoc test (WT vs. F29A *p = 0.0134, WT vs. I91A *p = 0.0122, WT vs. L95A *p = 0.0152, WT vs. V96A *p = 0.0471, WT vs. N97A *p = 0.0428). d A close-up view of the R28-IgI3 binding interface of CC1-N. e Alignment of CC1-N sequences. The critical residues required for binding of R28-IgI3 to human CC1 are highlighted by red boxes. f Binding of human, macaque, mouse or rat rCC1-HIS or rCC1-Fc proteins to R28-IgI3 (Mean ± s.d. of n = 4 independent experiments). Statistical significance tested by one-way ANOVA with Tukey's post-hoc test (Fc-tag proteins, human CC1 vs. human CC8 *p = 0.0478, human CC1 vs. macaque CC1 *p = 0.0354, human CC1 vs. rat CC1 *p = 0.0353; HIS-tag proteins, human CC1 vs. human CC8 *p = 0.0489, human C1 vs. mouse CC1 *p = 0.0487). g Docking of R28-IgI3 onto human and macaque CC1-N variants, quantified as the binding-free energy (kcal/mol) (Mean ± s.d. of n = 10 in silico simulations). Binding-free energy for non-interacting proteins (R28-IgI3 and human CC1-N F29A ) is shown by a dashed line. Statistical significance tested by one-way ANOVA with Tukey's post-hoc test (human CC1 vs. human CC1 F29A ****p < 0.0001, human CC1 vs. human CC1 F29I ***p < 0.0003, human CC1 vs. macaque CC1 ****p < 0.0001, macaque CC1 vs. macaque CC1I29F ***p = 0.0005). h Close-up of the R28-IgI3 binding interface of human and macaque CC1-N. R28-IgI3 residues required are shown in gray. F29 in human CC1-N and I29 in macaque CC1-N are shown in blue and yellow, respectively. a more physiologically relevant environment, we utilized human ectocervical tissue explants 34 , which represent a valuable model to study immune responses of cervix-resident immune cells to microbes. The human ecto-cervical tissue explants were infected with S. pyogenes wildtype or isogenic R28-negative strains for 24 h and then cultured in the presence of antibiotics to kill extracellular bacteria for 24 h (Fig. 5a). The ex vivo immune response to S. pyogenes in this system was donor specific ( Supplementary Fig. 10a), but expression of R28 by S. pyogenes caused significant reductions in the production of a subset of cytokines and chemokines including IL-1β, MIP-1β, GM-CSF, SDF-1β, and IL-12 (Fig. 5b). Of note, no toxicity was observed in S. pyogenes infected tissues (Supplementary Fig. 10b). Taken together, these experiments indicate that R28-expression by S. pyogenes is associated with an impaired innate immune response.
Neutrophils are innate immune cells that provide the first line of defense against invading bacteria in the blood, by phagocytosis and/or release of antimicrobial factors. However, S. pyogenes is equipped with several surface-localized virulence factors, such as M proteins 35,36 , that suppress neutrophil-mediated phagocytosis and killing. Because CEA-CAM1 is expressed by human neutrophils upon activation 18 , we investigated whether R28 could provide yet another mechanism for S. pyogenes to overcome human defenses by enhancing resistance to neutrophil killing. Purified R28 bound to the surface of activated human neutrophils, but not to resting neutrophils (Fig. 5c). Of note, R28 does not interact with the other CEACAM receptors expressed by neutrophils (Fig. 2c, d, e), including the phagocytic CEACAM3 receptor 37,38 . This suggests that R28 expressed on S. pyogenes will bind to activated neutrophils exclusively via CEACAM1. Next, we assessed whether R28-CEACAM1 interactions promoted S. pyogenes resistance to neutrophil killing by quantifying S. pyogenes survival after 1 and 2 h of incubation with human neutrophils. Indeed, R28-expression enhanced the resistance of S. pyogenes to neutrophil killing (Fig. 5d)  consistent with previous findings 25 . These observations were echoed in whole blood killing experiments ( Fig. 5e and Supplementary Fig. 10c), where R28-expression enhanced the resistance of S. pyogenes to killing. Moreover, the increased survival of R28-expressing S. pyogenes in whole blood was dependent on ability to interact with CEACAM1 ( Fig. 5f and Supplementary Fig. 10d), as there was reduced survival when the bacteria were pre-incubated with recombinant CEACAM1. Collectively, these data indicate that the resistance of S. pyogenes to innate immune cell killing is enhanced through binding of R28 to CEACAM1.

Discussion
Puerperal sepsis is a disease with important historical context, as a driver in the development of germ theory and modern infection control 4,5 . We have now clarified the role of the R28 adhesin in the pathogenesis of S. pyogenes puerperal sepsis at the molecular and functional level. Our results demonstrate that human CEACAM1 is a highly specific receptor for R28. This single host-pathogen interaction promotes the adhesion of bacteria to epithelial layers, delays epithelial wound healing, subverts the innate immune responses, and enhances c d S. pyogenes resistance to immune-mediated killing. Thus, our studies demonstrate that a single bacterial factor can target a single host factor to assist a bacterial pathogen in overcoming multiple host barriers on the pathway to causing an invasive disease. This contrasts the established view that individual molecular host-pathogen interactions mainly drive a bacterial pathogen to overcome a single host defense mechanism. This conclusion emphasizes the importance of considering that individual molecular host-pathogen interactions can have multifaceted functions in bacterial pathogenicity. The targeting of S. pyogenes to CEACAM1 on both epithelial cells and immune cells likely favors bacterial tissue invasion, survival in blood and development of systemic infection, as summarized in Fig. 6. This model is in good agreement with data obtained for other bacterial pathogens that bind CEACAM1 and exploit its role as an immunomodulatory receptor 19,22,39 , including Neisseria gonorrhoeae, Neisseria meningitidis, Helicobacter pylori, Moraxella catarrhalis, Fusobacterium nucleatum, and Haemophilus influenzae. While our data indicate that R28 plays a key role in the pathogenesis of puerperal sepsis, additional S. pyogenes virulence factors such as M proteins, exotoxins and proteases are probably secondary determinants required to cause this invasive disease 40,41 .
The R28 adhesin exclusively binds to CEACAM1. In contrast, other bacterial adhesins bind to a combination of CEACAM1, CEACAM3, CEACAM5 and CEACAM6 17,42-46 . Our finding therefore makes S. pyogenes the first pathogen for which CEACAM binding is restricted to CEACAM1. The basis of this selectivity requires further investigation and cannot be explained by targeting of R28 to different CEACAM faces, as R28 and all bacterial ligands characterized to date bind the A′ GFCC′C′′ face 17,42-45 . Moreover, the selectivity cannot be entirely explained by differential targeting of residues in the A′GFCC′C′′ face of CEACAM1, as residues F29, Q89, I91 and L95 on CEACAM1 are targeted by R28 and other bacterial adhesins including β protein of S. agalactiae, HopQ of H. pylori and Opa of Neisseria. spp 17,42,44,45 . The unique binding mode of R28 that is highly selective for CEACAM1 requires investigation using formal biophysical assessments.
Bacterial pathogens utilize CEACAM1 to dock onto epithelial cells 47 . For S. pyogenes, the binding of R28 to CEACAM1 also provides a mechanism to delay repair of wounds in epithelial layers. These properties of R28 should be of importance in the wounded tissues present in women after delivery, as a delay in wound repair would prolong the time for the pathogen to invade tissues and to cause puerperal sepsis. The properties of R28 also focus interest on the poorly understood mechanisms by which CEACAM1 is known to regulate cell migration and wound repair [31][32][33] . Of note, the interaction between R28 and CEACAM1 is different from that identified in other CEACAM1-binding bacterial pathogens, which can use CEACAM1 to  trigger endocytosis of bacteria into the epithelial cell where they can replicate and/or transcytoses the epithelial layer [47][48][49] .
S. pyogenes has evolved sophisticated mechanisms to evade neutrophil phagocytosis and killing 50 . Activated neutrophils express not only CEACAM1, but also CEACAM3, a specialized receptor employed for phagocytosis and elimination of CEACAM3-binding bacteria 37,38 . Since R28 does not bind CEACAM3, our data imply that CEACAM3 does not affect the role of R28 in the pathogenesis of S. pyogenes infections and puerperal sepsis. This lack of interaction with CEACAM3 contrasts with most of the other bacterial ligands of CEA-CAMs, including Opa of Neisseria spp. 44,45 , HopQ of H. pylori 42 , and UspA1 of M. catarrhalis 43,51 . The high-specificity for CEACAM1, suggests that R28 has evolved to escape CEACAM3-mediated detection by neutrophils, whilst maintaining the capacity to interact with CEACAM1. This could allow S. pyogenes to subvert the anti-bacterial responses of neutrophils. A better understanding of the intracellular signaling events affected by R28 engagement of CEACAM1 on neutrophils could help to understand the progression of invasive infections and puerperal sepsis.
Since a recent study reported that human integrins may act as receptors for both the N1 and N2 domains of R28 21 , it is possible that integrins serve as co-receptors for R28. However, our in vitro and ex vivo experiments demonstrate that interaction with CEACAM1 is required and sufficient for R28 to suppress wound closure, the innate immune response and neutrophil-mediated killing. We therefore postulate that R28-integrin interactions are of low affinity and dispensable for the functional properties of R28. Similarly, integrin interactions are dispensable for the virulence of H. pylori, which targets CEACAM1 through expression of the HopQ adhesin to promote virulence 22,52 . Of note, the identification of human CEACAM1 as the cognate and highly specific receptor for R28 provides critical information that humanized animal models are required to fully elucidate the role of R28 in vivo 53 , and that experimental infections in wildtype mice or macaques with R28 strains must be evaluated with caution 9,25,27 .
While R28 is found in S. pyogenes strains of several emm types, viz. emm28, emm2, emm48, and emm77, only emm28 strains are strongly associated with outbreaks of puerperal sepsis [10][11][12] . However, emm77 strains have been reported to cause at least one outbreak 54 . There are several potential explanations for the association of emm28 strains with outbreaks of puerperal sepsis. First, this could be because emm28 strains are more common than emm2, emm77, and emm48 strains, as suggested by recent data on emm type distributions 13,55 . Second, this could be because the gene for R28, spr28, is present in all emm28 strains but not all emm2, emm77, and emm48 strains 13,23 , a fact that might be attributed to relatively recent acquisition of the mobile genetic element carrying spr28 into these lineages 24 . Thirdly, our data indicate that R28 expression is higher in emm28 strains compared to emm2 and emm77 strains. Finally, it could be that emm28 strains have enhanced transmissibility or virulence properties. Some or all these factors may contribute to the preponderance of emm28 among strains causing outbreaks of puerperal sepsis. It is also important to note that sporadic cases of puerperal sepsis are commonly caused by strains of S. pyogenes that do not express R28 11 . This may occur if such strains bypass the initial barriers to infection, although they lack R28, in which case the disease may be like that caused by R28-positive S. pyogenes strains. However, it seems reasonable to assume that a smaller infection dose is required for R28-positive strains, a factor that may favor the occurrence of outbreaks.
The most appropriate experiment to test the capacity of the R28-CEACAM1 interaction to promote development of puerperal sepsis would be to study disease progress in female mice, including transgenic human CEACAM1 +/-or +/+ , after intravaginal inoculation with S. pyogenes shortly after the birth of pups. However, such a model that follows the natural route of puerperal sepsis does not exist. Given the large sacrifice of pups and the use of females who just had pups, it is highly unlikely that permission to perform such experiments would be provided by an ethical committee. Furthermore, colonization studies in transgenic mice, in the absence of birth-induced damage, are poorly justified as S. pyogenes rarely colonizes the reproductive tract of women, even in late pregnancy 56 . These arguments highlight the need to develop new models, that must be humanized given the humanspecificity of the R28-CEACAM1 interaction, to investigate puerperal sepsis.
In summary, our results provide strong evidence that the R28-CEACAM1 interaction plays a major role in outbreaks of puerperal sepsis caused by S. pyogenes. Thus, our data expand the understanding of one of the most important nosocomial infections in human history. It is noteworthy that life-threatening invasive S. pyogenes infections other than postpartum infections have increased in incidence during recent decades and often are caused by R28-expressing strains 13 . Because R28 elicits protective immunity 9 , it may therefore become of interest to include it in a future S. pyogenes vaccine 57 . This aspect is particularly relevant because R28 is encoded on a mobile genetic element 23,24 , raising concerns about dissemination of the gene to new S. pyogenes lineages or other bacterial species. Finally, our work highlights that the combination of epidemiology with functional analysis of disease-associated virulence factors can be critical to unraveling the molecular basis of major bacterial infections.

Ethics
Human blood was obtained from healthy donors, approved by the Regional Ethics Committee and Imperial College Healthcare NHS Trust Tissue Bank (Regional Ethics Committee approval no. 17/WA/0161, Imperial College Healthcare Tissue Bank Human Tissue Authority license no. 12275, and Imperial College Research Ethics Committee no. 19IC5166). All samples were collected after receiving signed informed consent from all participants. Human female reproductive tract organ specimens (tissues of vagina, ectocervix, endocervix, uterus, Fallopian tube, and ovary) for immunohistochemistry were obtained from patients treated at the Department of Gynecology of Sumy Regional Oncology Center, Sumy, Ukraine. All tissues were collected after receiving signed informed consent from all patients. The Bioethics Commission of the Medical Institute of Sumy State University approved the experimental protocol (no. 36 from 14.05.2018). Surgically resected specimens of human ecto-cervical tissue were obtained from the York Tissue Bank, University of York, UK approved by the Yorkshire & The Humber-Leeds East Research Ethics Committee (NHS REC 20/YH/0126) or collected at St. Mary's Hospital, Imperial College Healthcare NHS Trust, London, UK, after receiving signed informed consent from all patients through the Imperial College Healthcare Tissue Bank approved by Research Ethics Committee (IRAS 17/WA/ 0161). All tissues were collected after receiving signed informed consent from all patients.

Human blood
Neutrophils and peripheral blood mononuclear cells (PBMCs) were isolated from human blood samples by Ficoll/Histopaque centrifugation as and resuspended in RPMI 1640 supplemented with 2% FCS, as previously described 58 .

Human vaginal tissue explants
Resected ecto-cervical tissue was cut into 2-3 mm 3 explants comprising both epithelial and muscularis mucosae as described previously 59 . Nonpolarized explants were maintained with complete DMEM containing 10% fetal calf serum, 2 mm L-glutamine in the presence or absence of antibiotics (100 U of penicillin/ml, 100 µg of streptomycin/ml, and 80 µg of gentamicin/ml) at 37°C in an atmosphere containing 5% CO 2 .

Bacterial strains and culture
Microorganisms used in this study are listed in Supplementary Data 3. Bacteria were cultured in Todd-Hewitt Broth (TH) for S. agalactiae, TH Broth + 1% yeast (TH-Y) for S. pyogenes, or Lysogeny Broth (LB) for E. coli. The medium was supplemented with antibiotics when required.
PCR analysis: To test S. pyogenes strains for carriage of spr28, a PCR analysis was performed using primers that target the IgI3 domain F: 5′-ACAGCTCCAACATTAACCGTC-3′ and R: 5′-TTTTGGTTCGTTGCTA TCCTT-3′, with an expected amplicon of 336 bp. PCR was performed with Phusion Plus DNA polymerase following standard reaction conditions. Thermocycling was performed as 1 cycle of 98°C for 2 min, 35 cycles of 98°C for 15 s, 60°C for 20 s, and 72°C for 20 s, and 1 cycle of 72°C for 7 min. Amplification of a product was analyzed using a 1.25% agarose gel.

Expression and purification of bacterial proteins
The R28 (Spy1336), β, α, and Rib proteins were purified from S. pyogenes and S. agalactiae bacteria as previously described 9,62,63 . Recombinant M28 was purified essentially as described for other M proteins 64 . Vectors encoding mutant forms of R28-IgI3 were generated by site-directed mutagenesis of our previously constructed pRSET-C-R28-IgI3 vector (Supplementary Data 4) 17 . Wildtype and mutant forms of the IgI3 domain of R28 (R28-IgI3) were expressed in E. coli Rosetta Gami for 4 h at 37°C following the addition of 1 mM IPTG, and purified using a Nickel column (GE Healthcare Life Sciences) and affinity chromatography (ÄKTA Pure, GE Healthcare Life Sciences), as previously described 17 .

Expression and purification of rCEACAMs
Expression vectors used in this study are listed in Supplementary Data 4. Vectors encoding mutant forms of CEACAM1-N were generated by sitedirected mutagenesis of our previously constructed pRSET-C-CEACAM1-N vector, as previously described 17 . Wildtype and mutated forms of rCEACAM1-N were expressed in E. coli Rosetta Gami for 4 h at 37°C following the addition of 1 mM IPTG, and purified using a Nickel column (GE Healthcare Life Sciences) and affinity chromatography (ÄKTA Pure, GE Healthcare Life Sciences), as previously described 17 . Tag-less rCEACAM1-N for use in isothermal titration calorimetry (ITC) and crystallization were expressed using pET21d vectors in E. coli and purified, as previously described 17 . Human rCEACAMs with His-tag were expressed in Expi293F cells (Life Technologies) cultured in Expi293 Expression Medium (Life Technologies) and purified by affinity chromatography (ÄKTA Pure, GE Healthcare Life Sciences) using a Nickel column (GE Healthcare Life Sciences), as previously described 17 . Mouse rCEACAM1-HIS was purchased from R&D systems. Human, macaque and rat CEACAM1-Fc were expressed as previously described 22 .

Western blot analysis of R28 and CEACAM interactions
Nitrocellulose membranes were blocked with 5% milk solution overnight at 4°C and probed with 10 μg/ml rCEACAM1-His. Binding of rCEACAM1-HIS to the nitrocellulose membrane was detected using horse radish peroxidase (HRP)-conjugated mouse anti-HIS-IgG (HIS.H8; Invitrogen) and developed using Pierce ECL Western Blot Substrate Reagent (Thermo Fisher Scientific).

Isothermal titration calorimetry analysis
ITC measurements were performed on an iTC200 instrument (GE Healthcare) in duplicate. 500 µm of rCEACAM-N domains were loaded into the syringe of the calorimeter and 50 µm rR28-IgI3 was loaded into the syringe. All measurements were performed at a stirring speed of 750 rpm, in 30 mm Tris-HCl, 150 mm NaCl, pH 7.5 at 25°C. Origin 7.0 software was used to analyze all data.

Measurement of R28 expression by S. pyogenes
Mouse antiserum raised against purified R28 protein was previously described 9 . In all, 6 × 10 6 of mid-logarithmic phase bacteria were incubated for 1 h at 4°C with heat-inactivated 0.1% mouse anti-R28 serum or normal mouse serum. After washing, IgG deposition was measured using PE-conjugated goat anti-rabbit-IgG (12-4010-87; Life Technologies) at 4°C for 1 h and flow cytometric analysis. Gating strategy is shown in Supplementary Fig. 11b. R28 expression by S. pyogenes was quantified as IgG deposition after incubation in mouse anti-R28 serum relative to normal mouse serum. The S. pyogenes M1 5448 strain that does not express R28 was included as a negative control.
Analysis of rCEACAM binding to S. pyogenes 6×10 6 of mid-logarithmic phase bacteria were incubated with rCEA-CAM at 4°C for 1 h and washed in PBS + 0.5% bovine serum albumin (BSA). Binding of rCEACAM on the bacterial surface was detected by incubation with FITC-conjugated anti-HIS mAb (MA181891; Thermo-Fisher) at 4°C for 1 h and flow cytometric analysis of fixed bacteria (1% PFA). CC1 binding was quantified relative to an unstained control. The S. pyogenes M1 5448 strain that does not express R28 was included as a negative control. The S. agalactiae strain expressing the CEACAM1binding β protein was included as a positive control.

Screening of receptors for R28
Receptors for R28 were identified using cell microarray technology at Retrogenix, UK. For primary screening, 3559 expression vectors covering more than 2625 different human plasma membrane proteins (Supplementary Table 1) were arrayed in duplicate across 10 microarray slides. Subsequently, human HEK293 cells were grown on top of the cDNA clones and were reverse transfected. An expression vector (pIRES-hEGFR-IRES-ZsGreen1) was spotted in quadruplicate on every slide, and was used to ensure that a minimal threshold of transfection efficiency had been achieved or exceeded on every slide. The cell microarrays were fixed onto the slides and tested for binding of 10 μg/ml purified R28 and Rib protein. After washing of the slides, the streptococcal proteins were detected with rabbit antisera (mixture of two rabbit antisera both diluted 1:1600, including rabbit anti-Rib sera) and AlexaFluor647(AF647)-conjugated anti-rabbit IgG (Invitrogen). Two replicates were performed in the primary screen. Fluorescent images of the slides were analyzed and quantitated (for transfection efficiency) using ImageQuant software (GE). A positive protein 'hit' on was defined as a duplicate spot showing a raised signal compared to background levels, which was assessed by visual inspection using the images gridded on the ImageQuant software. To confirm the hits, all vectors encoding the hits identified in the primary screen were arrayed and over-expressed onto new slides. Slides were screened with 10 μg/ml R28 or buffer followed by rabbit anti-Rib sera as described above.

Immunohistochemistry of female reproductive tract
Serial sections of 4 µm were prepared from paraffin-embedded tissue previously fixed in neutrally buffered formalin and mounted on 3aminopropyltriethoxysilane-coated slides. After overnight exposition at 50°C, tissue samples were de-paraffinized and rehydrated in graded alcohol and xylol. Heat-mediated antigen retrieval was performed utilizing a water bath and 0.01 M sodium citrate for 30 min at 97°С. Endogenous peroxidase activity was blocked by treatment with 3% H 2 O 2 for 5 min and subsequent washing with PBS. Sections were blocked with 1% BSA/PBS and incubated overnight at 4°C with 0.1 µg/ml of monospecific anti-CEACAM1 (clone C5-1X/8; LeukoCom), anti-CEACAM5 (clone 3E10-3; LeukoCom) and anti-CEACAM6 (clone 1H7-4B; LeukoCom), or an isotype control. After washing, sections were probed with biotinylated secondary rabbit anti-mouse antibody (Dako) for 1 h at room temperature. Sections were washed and incubated with VECTASTAIN ABC reagent (Vector Laboratories) for 30 min according to the Manufacturers´protocol. Staining was visualized by diaminobenzidine (DAB) substrate controlling the developing color intensity via light microscopy. DAB-negative structures were identified by additional counterstaining with hematoxylin. Stained sections were mounted and documented by microscopy.

Adhesion of S. pyogenes to cell lines
Mid-logarithmic phase S. pyogenes cultured in TH-Y were washed and resuspended (CHO and HeLa, in DMEM containing 10% FCS; ME-180 McCoy's 5A containing 10% FCS). Confluent monolayers were washed and infected with the prepared bacteria at a multiplicity of infection (MOI) of 10 or 30. Assays were commenced by centrifugation for 2 min at 100 × g and incubated at 37°C with 5% CO 2 for 30 min (CHO or HeLa cells) or 120 min (ME-180 cells). Monolayers were washed five times with PBS, detached with 0.25% trypsin and lysed with PBS containing 0.025% Triton-X. Serial dilutions of the cell lysates were plated onto TH-Y agar plates. After incubation of TH-Y plates overnight at 37°C, the number of adherent bacteria were enumerated. CEACAM-specific interactions were measured in some assays by pre-incubating cells with 5 μg/ml CC1/3/5-Sab (detects N-terminal domain of human CC1, CC3, and CC5) or isotype control mAb for 30 min.
Binding of FITC-labeled S. pyogenes to cell lines Confluent CHO or ME-180 cells were detached, washed, resuspended to 4×10 6 cells/ml and incubated with FITC-labeled S. pyogenes strains at an MOI of 30 for 30 min at 37°C. After washing and fixation in 1% PFA, the fluorescence of cells was measured by flow cytometry. Gating strategy is shown in Supplementary Fig. 11a.
Scratch assay of ME-180 wound closure In all, 2.5 × 10 5 ME-180 cells were cultured on 24-well tissue culture plates until 85% confluency. In each well, a scratch was made with the tip of a yellow pipet tip and monolayers were washed three times to remove detached cells. The cells were incubated at 37°C with 5% CO 2 for 1 h after addition of 125 μl of McCoy's 5A + 10% FCS to each well. Four photographs of each scratch, representing 0-h timepoint (t0), were collected using a Nikon D5100 Digital Camera. For infection experiments, tissue culture wells were infected with 1 × 10 8 midlogarithmic phase S. pyogenes cultured in TH-Y and subsequently resuspended in McCoy's 5A containing 10% FCS. After 24 h incubation at 37°C with 5% CO 2 , monolayers were washed three times to remove bacteria and supplemented with McCoy's 5A media containing 10% FCS containing penicillin/streptomycin. Four photographs of each scratch, representing the 24-h timepoint (t1), were collected. The cell monolayers were cultured for a further 24 h at 37°C with 5% CO 2 , prior to collection of four photographs of each scratch, representing the 48h timepoint (t2). The mean distance of each scratch across the four photographs was measured using ImageJ. CEACAM-specific interactions were measured in some assays by incubating cells with 20 μg/ml CC1/3/5-Sab (detects N-terminal domain of human CC1, CC3, and CC5) or isotype control mAb throughout the assay.
For assays employing purified R28 or Rib or rR28-IgI3 variants, wounded ME-180 cells were supplemented with McCoy's 5A media containing 10% FCS containing penicillin/streptomycin and 10 mg/ml of the specific protein. Four photographs of each scratch, representing the 0-h timepoint (t1), were collected at 10x magnification. The wounded ME-180 cells were cultured for 24 h at 37°C with 5% CO 2 , prior to collection of four photographs of each scratch, representing the 48-h timepoint (t2).
In both experimental types, the percentage of wound closure was calculated as (scratch distance at t1 or t2 ÷ scratch distance at t0) × 100, where t0 equals 0 h and t1 and t2 equal a 24 or 48 h timepoint.

Transwell assays
In all, 2 × 10 4 cells/well were cultured on Transwell (Greiner Bio-One ThinCert, pore diameter 3 μm) inserts in the apical chamber of a 24well plate until confluency. PBMCs were seeded at a density of 1.25 × 10 6 cells/well in the basal chamber of a 24-well plates. The apical chamber was challenged with γ-irradiated S. pyogenes strains at an MOI of 10 and incubated for 72 h at 37°C with 5% CO 2 . Supernatants were harvested from both chambers and centrifuged at 8000 × g to remove excess bacteria before storage at -80°C for Luminex analysis.

Human ecto-cervical explant experiments
The ecto-cervical explants were incubated for 3 h in complete DMEM with antibiotics. After four washes in PBS, the explants were transferred to fresh plates and challenged with S. pyogenes in complete DMEM with no antibiotics for 24 h. Explants were then transferred to new plates and cultured for 5 days in complete DMEM with antibiotics. Culture supernatants were harvested at days 2 and 5 post-challenge for proteomic evaluation. Viability of tissue explants following exposure to S. pyogenes was determined by measuring tetrazolium salt [3-(4,5-dimethyl-2-thiazolyl)-2,5-diphenyl-2H-tetrazolium bromide (MTT)] cleavage into a blue product (formazan) by viable cells 65 , as described previously 66 . Optical density values obtained with a Synergy-HT (BioTek, Winooski, VT) plate reader were corrected for explant dry weight. Untreated tissue was considered as positive control of viability.

Measurement of cytokine and chemokine profiles
A total of eighteen soluble immune proteins were quantified in three panels by in house multiplex bead immunoassay using a Luminex 200 System (Bio-Rad, Hercules, CA) as described previously 67 . Cytokine/ chemokine levels were normalized against total protein content as measured by a BCA protein assay (Bio-Rad, Hercules, CA). Cytokine concentrations were calculated from sigmoid curve-fits (Prism, GraphPad). All data presented fulfil the criterion of R2 > 0.7. For transwell experiments employing ME-180 cells and primary human PBMCs, ratios between cytokine/chemokine concentrations in explant cultures dosed with S. pyogenes vs. untreated tissues were established and statistically compared using unpaired t test and considered significant at p < 0.05. For human ecto-cervical explant experiments, ratios between cytokine/chemokine concentrations in explant cultures dosed with S. pyogenes vs. untreated tissues were established and statistically compared using unpaired t test and considered significant at p < 0.05.

Neutrophil and whole blood assays
Binding of R28 to neutrophils was assessed by flow cytometric analysis. 45 μl of 3 × 10 7 cells/ml were incubated with 5 μl of 10 μg/ml R28 or buffer on ice for 45 min. After washing, neutrophils were incubated with 25 μl of 0.1% mouse anti-R28 sera or normal mouse sera for 45 min on ice. After washing, neutrophils were incubated with 25 μl of PEconjugated goat anti-mouse-IgG for 45 min on ice. Fluorescence of neutrophils was measured by flow cytometry. Gating strategy is shown in Supplementary Fig. 11c. S. pyogenes strains were sub-cultured to early-log phase and resuspended in PBS + 0.5% BSA. Purified neutrophils were resuspended to 1 × 10 7 cells/ml and 180 μl was inoculated with 20 μl S. pyogenes at an MOI of 10, and incubated at 37°C with gentle rotation. After 2 h, 50 μl of sample was lysed with 0.3% saponin on ice. Lysed samples were serially diluted in PBS and plated onto TH-Y agar for CFU enumeration. For whole blood killing assays, 270 μl of human blood, freshly drawn into tubes with heparin, was inoculated with 15 μl of S. pyogenes strains (CFU = 3000) and incubated at 37°C for 3 h. Inoculated blood was serially diluted in PBS and plated onto TH-Y agar plates for CFU enumeration.
Crystallization of R28-IgI3 and CEACAM1-N complex R28-IgI3 was treated with carboxypeptidase A (1:100 w/w) for 24 h at room temperature to remove the C-terminal His-tag and part of the C-terminal region of R28-Ig3. Reaction was quenched with 5 mm EDTA before repurification on a Superdex 200 column (GE Healthcare) equilibrated with 20 mm Tris, 150 mm NaCl, pH 7.5. Purified CEACAM1-N was added in a slight excess (1:1.1) with R28-Ig3, incubated on ice for 30 min before purification on a Superdex 200 column with the same buffer described above and concentrating to 5 mg/ml by centricon. The complex was screened against several different commercial crystallization screens. Crystals were identified in the Wizard Screen (1.8 m Ammonium Sulfate, 0.1 m Sodium Citrate pH 5.5). Crystals were optimized with the best crystals produced in 1.45 M Ammonium Sulfate, 0.1 M Na/K Phosphate pH 5.5. Crystals were cryoprotected in 30% v/v glycerol, and a dataset was collected beamline 9-2 at the Stanford Synchrotron Radiation Lightsource. Data was processed and indexed with XDS before scaling and conversion to structural factors using Aimless. MOLREP was used to perform molecular replacement using CEACAM1-N (PDB: 2GK2) and β-IgI3 (PDB: 6V3P) as search models. Refinement was carried out by REFMAC5 and Phenix Refine. Rebuilding was carried out in Coot. Refinement statistics are shown in Supplementary Table 3.

Bioinformatics analysis
The presence of spr28 was confirmed in 3886 previously assembled and curated genomes of UK S. pyogenes isolates using Seemann T, Abricate (https://github.com/tseemann/abricate) to detect six regions of the spr28 gene (90-336 bp in length) 68 . A subset of genomes to represent ten isolates per emm-type and excluding any with fewer than ten (except emm48 which was represented by a single spr28 positive isolate) but including all 26 emm77, were subjected to pangenome analysis and core genome alignment using Panaroo and a phylogenetic tree constructed with RAxML and annotated in iTOL [69][70][71] .
The structure of human CEACAM1-N F29A , human CEACAM1-N F29I , macaque CEACAM1-N and macaque CEACAM1-N I29F were predicted using the human (PDB: 4QXW) CEACAM1 structures, respectively, as input in SWISS-MODEL 72 . All models passed a QMEAN score of <−4.00. The docking of the R28-IgI3 structure to CEACAM1-N structure or predicted structures was simulated using HawkDock server 73 . The binding energy of simulated complexes (kcal/mol) was compared by One-way ANOVA with Tukey's multiple comparison test.

Quantification and statistical analysis
All quantitative data were analyzed and graphed using GraphPad Prism 8.4.3 software. All data are represented as mean ± s.d. calculated using the GraphPad Prism 8.4.3 software, unless indicated otherwise. Statistical details of the experiments are provided in the respective figure legends and in each methods section pertaining to the specific technique applied.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
The co-crystal (R28-IgI3 in complex with CEACAM1-N) structural data generated in this study have been deposited in the Protein Data Bank (PDB) database under accession code 8CXJ. Source data are provided with this manuscript. Source data are provided with this paper.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/.