Author Correction: Uncoupling of invasive bacterial mucosal immunogenicity from pathogenicity

There is the notion that infection with a virulent intestinal pathogen induces generally stronger mucosal adaptive immunity than the exposure to an avirulent strain. Whether the associated mucosal inflammation is important or redundant for effective induction of immunity is, however, still unclear. Here we use a model of auxotrophic Salmonella infection in germ-free mice to show that live bacterial virulence factor-driven immunogenicity can be uncoupled from inflammatory pathogenicity. Although live auxotrophic Salmonella no longer causes inflammation, its mucosal virulence factors remain the main drivers of protective mucosal immunity; virulence factor-deficient, like killed, bacteria show reduced efficacy. Assessing the involvement of innate pathogen sensing mechanisms, we show MYD88/TRIF, Caspase-1/Caspase-11 inflammasome, and NOD1/NOD2 nodosome signaling to be individually redundant. In colonized animals we show that microbiota metabolite cross-feeding may recover intestinal luminal colonization but not pathogenicity. Consequent immunoglobulin A immunity and microbial niche competition synergistically protect against Salmonella wild-type infection. Virulent pathogens generally induce a stronger mucosal immunity than avirulent strains, but whether the associated inflammation is necessary for this is unclear. Here, using auxotrophic Salmonella enterica, the authors show that virulence factor function determines induction of protective IgA.

M ounting a functional anti-microbial adaptive immune response depends on concomitant induction of an innate immunogenic response through pattern recognition receptor (PRR) activation 1 . PRRs sense conserved microbial molecular structures, such as bacterial lipopolysaccharide (LPS), peptidoglycan, and flagellin, that are conserved across pathogenic and non-pathogenic microorganisms 2 . Pathogenspecific virulence factors such as type 3 secretion system (T3SS) components 3 and intracellular toxin action have also been shown to be specifically sensed by PRRs 4,5 . The integration of diverse PRR signals is believed to regulate immune responses according to the nature of the microbial threat 6,7 . Natural and artificial PRR signaling agonists are consequently exploited pharmaceutically as pro-immunogenic additives or adjuvant components of vaccines 8 . Besides the immunogenic response, PRR activation by pathogens may also drive inflammation and innate anti-microbial defense. This arm of the innate immune system is important for the control of primary pathogen infection, but is also responsible for the adverse effects of inflammation and defense that damage host tissue and symbiotic microbiota, which may be exploited by some mucosal pathogens 9 .
The intestinal mucosal membranes are colonized continuously with a diverse symbiotic microbiota and are guarded by a complex mucosal immune system. The mucosa is well adapted to stable symbiosis with non-pathogenic microbes. Multiple physical and chemical barriers as well as active immune tolerance avoid the unnecessary activation of immune defense mechanisms by harmless symbiotic microbes or food antigens 10 . Only virulent mucosal pathogens normally induce inflammatory responses. Avirulent, fully attenuated pathogens are inefficient at driving inflammation, but also tend to induce less effective adaptive immunity than virulent pathogens 11,12 . It is consequently difficult to induce protective mucosal immunity safely with adequately attenuated live vaccines-this compromises vaccination efforts in developing countries for which safe, effective, and easy-toadminister oral vaccines are urgently needed 13,14 .
While there is a clear difference between the immune responses induced by virulent and non-virulent variants of a pathogenic bacterium, it is unclear which aspects of bacterial virulence may be differentially sensed by the immune system to induce efficacious adaptive immunity. Virulence factors enable pathogens to colonize privileged body sites, overgrow host defenses, and consequently damage host tissue architecture and function. Inactivated (killed) pathogenic microbes are avirulent, because they are sterile and most virulence mechanisms (apart from, for example, stable exotoxins) are dependent on bacterial viability. Our question was whether a pathogen that combines sterility and viability, which expresses molecularly functional virulence factors in vivo but is unable to replicate, still retains its mucosal immunogenicity.
To address this question, we apply a quantitative Salmonella enterica serovar Typhimurium (STm) infection model in germfree mice in which live bacterial replication in vivo is blocked. We use auxotrophic mutants of STm (STm Aux ) that are genetically engineered to be fully replication incompetent in germ-free animals and host tissues. As we previously established in nonpathogenic enterobacteria 15,16 , STm Aux colonization in germ-free mice is limited by the quantity of the bacterial inoculum and fully transient, allowing the germ-free host to return to germ-free status.
Using germ-free mice, this experimental approach allows us to rigorously test the following issues. First, whether the mucosal immunogenic response can distinguish between virulence factor proficient and deficient intestinal bacteria also in the absence of an acute inflammatory response and pathology. Secondly, whether the remaining immunogenic response would depend on similar PRR signaling pathways as the innate immune defense. These fundamental studies are carried out in a germ-free setting, to avoid the possible confounding effect of auxotrophic metabolite crossfeeding by bacteria of the gut microbiota in vivo. Extending our results into colonized mice, we move on to show that indeed crossfeeding by the microbiota can recover efficient intestinal colonization, but not pathogenicity of STm Aux . Strictly confined by the mucosal barrier it then combines virulence factor-dependent immunogenicity and avirulence with the added benefit of pathogen niche competition.

Results
Proliferation-incompetent STm Aux induces functional immunity. Mucosal tissue invasion and virulence of STm are mediated by two type 3 secretions systems (T3SS) encoded on Salmonella pathogenicity islands, SPI1 and SPI2 (refs. 3,[17][18][19]. Activity of the SPI1-encoded T3SS induces early mucosal inflammation 20,21 . As the invading and tissue-overgrowing virulent bacteria responsible are subject to pronounced population bottle necks 3,22 , we hypothesized that a strain of live STm encoding functional virulence factors would retain its invasiveness with associated adaptive immunogenicity, despite being unable to replicate and overall avirulent. To test this hypothesis we generated an auxotrophic STm strain (STm Aux ) that strictly requires supplementation with the essential peptidoglycan constituents D-alanine (D-Ala) and mesodiaminopimelic acid (m-Dap) to grow and survive cell division ( Supplementary Fig. 1A, B). Like the homologous model in commensal Escherichia coli developed previously 15,16 , STm Aux colonized the gastrointestinal tract of germ-free mice only transiently, allowing rapid and full recovery to germ-free status, as neither host metabolism nor diet could substitute the auxotrophic requirement for these metabolites (Fig. 1a, b). Salmonella T3SS competence or deficiency had no effect on STm Aux colonization kinetics. Bacterial quantitation in small intestinal (Supplementary Fig. 2A) and cecal ( Supplementary  Fig. 2B) contents at early time points revealed small intestinal transit of STm Aux in quantities similar to wild-type STm until 2.5 h following inoculation. At 4.5 h, STm Aux had transited from the small intestine into cecum without evidence for replication (Supplementary Fig. 2A, B; compare STm Aux numbers between small intestine at 2.5 h and cecum at 4.5 h), whereas wild-type STm populations had begun to expand in the cecum. By 34 h after inoculation wild-type STm stably colonized all intestinal segments, whereas STm Aux densities had sharply declined. No spontaneous D-Ala/m-Dap-independent revertants have been isolated ex vivo during these experiments.
D-Ala/m-Dap auxotrophic bacteria depleted of D-Ala or m-Dap, analogous to wild-type bacteria exposed to beta-lactam antibiotics, remain active until self-destruction by programmed autolytic cell death occurring at the onset of cell division 16,23 . Accordingly, D-Ala-and m-Dap-depleted STm Aux displayed normal cell invasiveness, as demonstrated by immunofluorescence microscopy and gentamicin protection assay (Fig. 1g, h). In germ-free mice, following enteral administration of 10 10 colonyforming units (CFU) of STm Aux by gavage, the invasive auxotroph was found to be completely avirulent. In contrast to wild-type STm, STm Aux was rarely recoverable from mesenteric lymph nodes (mLN), liver, or spleen (Fig. 1f). It no longer induced detectable levels of typhocolitis (inflammation of the cecum, the main enteric histopathology in the non-typhoidal invasive salmonellosis mouse model 24 ) as determined either by quantification of cecal luminal inflammation marker lipocalin-2 ( Fig. 1c) or by histopathologic scoring (Fig. 1d, e). Quantification of early mRNA markers of chemokine and other innate activation signals in total cecum tissue supported the conclusion that STm Aux is avirulent ( Supplementary Fig. 2C).
Transitory intestinal mucosal conditioning by live STm Aux bacteria (Fig. 2a) induced an adaptive immune response highly protective against the re-challenge of the germ-free animals with non-auxotrophic wild-type STm. While immunity had no effect on the large intestinal luminal load of the challenge strain (Fig. 2b), it protected against its intestinal pathogenesis (Fig. 2c, d) and limited penetration to the mLNs, liver, and spleen (Fig. 2e). Protective immunity was associated with high STm-specific titers of intestinal secretory IgA measured by live bacterial flow cytometry 25 (Fig. 2f, g), and was abolished in B cell-and antibody-deficient J H −/− mice ( Fig. 2h-j). B and T cell-deficient RAG-deficient mice phenocopied J H -deficient mice (Supplementary Fig. 3A-D). Hence, B cell immunity is functionally required for STm Aux -induced intestinal protective immunity. The live STm Aux dose-response relationship was examined by comparing the mucosal conditioning with doses of 10 10 , 10 8 , and 10 6 live STm Aux , which revealed that induction of functional immunity required doses greater than 10 8 live STm Aux (see extended dataset a b  in Supplementary Fig. 4A-G). Thus, STm Aux allowed us to probe mucosal immunity in a strictly dose-dependent manner. This data showed the threshold effects of STm Aux conditioning in germ-free mice, which would not be achievable with conventional non-auxotrophic bacteria that would exponentially expand rapidly to reach high intestinal densities independently of inoculum size.
Optimal protective efficacy of STm Aux is viability dependent. We next addressed how relevant STm Aux viability is for the induction of functional intestinal immunity. The replication incompetency of live STm Aux in germ-free mice allowed us to quantitatively compare the functional effects of mucosal exposure to live versus killed STm: both live and killed STm Aux cells are sterile entities in germ-free mice. Parallel groups of germ-free mice were intestinally conditioned by gavage with STm Aux inocula administered either live or following inactivation by peracetic acid (PAA) treatment. PAA killing is highly effective and has been shown to preserve mucosally protective STm surface B cell epitopes 26,27 . Naïve germ-free animals served as negative controls. Four weeks after the first treatment, the germfree animals of all three groups were challenged orally with virulent wild-type STm and studied at days 1 and 4 after the challenge, respectively (Fig. 3a). Compared to live STm Aux -conditioned mice, PAA-killed STm Aux -induced STm-specific IgA titers were reduced at day 1 of challenge ( Fig. 3b, Supplementary  Fig. 5). Yet, by day 4 this difference was no longer apparent. However, while pretreatment with either PAA killed or live STm Aux were similarly protective against early wild-type STminduced mucosal inflammation at day 1 after challenge, only live STm Aux preconditioning provided effective protection from intestinal pathology and organ infection until day 4 ( Fig. 3c-g).
Notably, live STm Aux -induced immunity not merely delayed the onset of disease, but protected the germ-free mice from lethal STm infection. Live STm Aux -conditioned germ-free mice that were followed up for 3 weeks following challenge remained free of macroscopic evidence of severe infection and were recovering at the time of sacrifice ( Supplementary Fig. 6). None of these effects were explained by differences in fecal or cecal luminal colonization levels of the challenge strain, which were similar across all experimental groups (Fig. 3h, i, Supplementary Fig. 6B).
Salmonella type 3 secretion signifies robust immunogenicity. We next asked whether or not the viability-dependency of functional mucosal immunogenicity of STm Aux is virulence factor related. We hypothesized that host interaction through Salmonella T3SSs (whose function is energy and viability dependent) signifies the functional immunogenicity of live STm Aux . If this was true, T3SS deficiency would diminish the mucosal efficacy of live STm Aux . We tested this hypothesis by comparing the protective effect of the enteral conditioning of germ-free mice with matching doses of live T3SS-competent and isogenic T3SS-deficient mutant strains of STm Aux (STm Aux T3SS− ). Two different isogenic STm Aux T3SS− mutants were tested: a complete SPI1 and SPI2 genomic island deletion mutant (ΔSPI1 ΔSPI1) devoid of T3SS genes entirely 28 , and a ΔinvC ΔssaV mutant expressing defective T3SSs 29,30 . Mice treated with equivalent doses of PAA-killed STm Aux or naïve mice served as controls. Four weeks after the first treatment, the germ-free animals were enterally challenged with wild-type STm, and studied at days 1 and 3 after challenge (Fig. 4a). Analysis of the severity of challenge infection and mucosal pathology at day 3 revealed that live, T3SS-deficient STm Aux strains induced less robust functional protective immunity than T3SS-competent STm Aux , and their efficacy against intestinal mucosal pathology (Fig. 4b, c) and bacterial penetration to mLN (Fig. 4d) was no longer significantly better than that of PAA-killed STm Aux . Genetic deletion of the three most important SPI1 effector protein genes (sopE, sopE2, and sipA) required for early SPI1 T3SS-mediated intestinal STm pathogenesis 31 also resulted in reduced efficacy ( Supplementary  Fig. 7). This suggests that not merely immune recognition of a functional T3SS apparatus but rather mucosal pathogenesisrelated type 3 effector protein functions are driving the superior immunogenicity of the T3SS-competent STm Aux strain.
As in the previous experiment ( Fig. 3b) killed STm Aux -as well as STm Aux T3SS− -preconditioned mice displayed reduced STmspecific IgA titers at day 1 of challenge ( Fig. 4e, Supplementary  Fig. 8). Yet, by day 3 mice of all three treatment groups had similar intestinal IgA titers. Immunoglobulin repertoire sequencing analysis of small intestine and mLN revealed overlapping IgA repertoires following mucosal conditioning with live T3SScompetent STm Aux -versus T3SS-incompetent STm Aux that clustered separately from those of naïve germ-free control mice Fig. 1 Transient colonization of GF mice with STm Aux . a Mice were inoculated at day 0 with 10 10 CFU of either auxotrophic ( Aux ; red symbols) or nonauxotrophic control (black symbols) STm strains that were either type 3 secretion competent (STm/STm Aux , filled symbols) or isogenic type 3 secretiondeficient mutants (STm T3SS− /STm Aux T3SS− , open symbols). b Time course of viable bacteria of each strain recoverable from feces (STm Aux n = 32, STm Aux T3SS− n = 15, STm T3SS− n = 12, STm n = 11, animals examined over nine independent experiments). c Lipocalin-2 concentration in cecal contents at 9 h after inoculation (STm Aux n = 5, STm T3SS− n = 5, STm n = 5 animals). d Cecal histopathology score at 9 h after inoculation. Each symbol represents one individual (STm Aux n = 5, STm T3SS− n = 5, STm n = 5 animals). e Cecal histology at 9 h after inoculation with indicated STm strains. H&E staining. Scale bar: 100 μm (STm Aux n = 5, STm T3SS− n = 5, STm n = 5 animals). f Organ loads of T3SS-proficient STm Aux and STm in mLN, liver, and spleen on day 1 (mLN: STm Aux n = 18, STm n = 3; liver: STm Aux n = 6, STm n = 2; spleen: STm Aux n = 9, STm n = 3 animals examined over four independent experiments) and 3 (mLN: STm Aux n = 2, STm n = 3; liver: STm Aux n = 2, STm n = 3; spleen: STm Aux n = 2, STm n = 3 animals examined over two independent experiments) post inoculation. g Immunofluorescence of HeLa cells infected for 2 h with wild type (STm), SPI1 T3SS-deficient (STm ΔinvC ), auxotrophic SPI1 T3SS-proficient (STm Aux ), and auxotrophic SPI1 T3SS-deficient (STm Aux ΔinvC ) STm. Cells were stained with DAPI (DNA/nuclei, blue), and with anti-STm group B antiserum and labeled secondary antibodies consecutively before, and after membrane permeabilization to differentiate extracellular (red + green) and intracellular (green only) STm. Scale bar: 10 µm (six samples were examined over two independent experiments for each condition). h Quantification of gentamicin-protected intracellular STm Aux ΔinvC (black open circles, n = 6 wells examined over two independent experiments), STm ΔinvC (black filled circles, n = 6 wells examined over two independent experiments), STm Aux (red open triangles, n = 6 wells examined over two independent experiments), and STm ΔinvC (red filled triangles, n = 6 wells examined over two independent experiments) in HeLa cells 2 h after infection. Statistics: bars indicate mean (c, f, h) or median (d) values. Horizontal dotted lines indicate the lower limit of detection (b, c, f, h). Panel c was analyzed with ordinary one-way ANOVA and Tukey's test for multiple comparison. Panel d was analyzed with a two-sided multicomparison Kruskal-Wallis test and Dunn's post hoc test. Panel f was analyzed with unpaired two-tailed t-test for each day. Panel h was analyzed with two-way ANOVA (virulence and auxotrophy as the two factors) and Sidak multiple comparison correction. Source data are provided as a Source Data file. Detailed statistical metrics are available in the Supplementary Statistical Analysis file.
( Supplementary Fig. 9). Repertoire overlap was measured by calculating the geometric mean of relative overlap frequencies between CDR3 amino acid sequence usage (see Methods section). Preprocessed clonotype amino acid sequences and metadata description are available as supplementary data files (Supplementary . O-serotype specific IgA has been shown previously to be a necessary component of any intestinal immune protection induced by killed or live STm 12,26,27 . O-antigen is a dominant polysaccharide antigen and in binding assays tends to mask other surface epitopes from antibody recognition, which is the basis of O-serotyping. To specifically study O-serotype-independent Salmonella surface binding IgA, germ-free mice were preconditioned with STm Aux but challenged with the different Salmonella serotype Enteritidis (SEn) (Supplementary Fig. 10A). The resulting intestinal IgA had reduced surface reactivity towards O-antigendeficient (rough) STm compared to wild-type (smooth) STm, as expected ( Supplementary Fig. 10B, C). However, the non-Oantigen-specific IgA cross-reacted between rough STm and rough SEn ( Supplementary Fig. 10B, C). It also cross-reacted with smooth wild-type SEn, suggesting that it contributes to serotype-independent Salmonella surface reactivity (Supplementary Fig. 10B, C). Although the O-serotype-independent IgA component alone is insufficient 12 Fig. 2 Intestinal conditioning of GF mice with STm Aux induces B cell dependent functional intestinal immunity. a Germ-free mice were enterally conditioned with six doses of 10 10 CFU of STm Aux (red triangles, n = 5 animals) or were left untreated (gray-filled circles, n = 5 animals). Four weeks after the first treatment (day 0) mice were challenged with wild-type STm (10 5 CFU) and analyzed 2 days later. Each symbol represents one individual. b Shedding of wild-type STm in feces 1 day after challenge. c Cecal histopathology score at day 2 after challenge. Each symbol represents one individual. O-antigen-specific IgA in protective mucosal immunity. Supporting this idea, we found that, although both killed and T3SSdeficient STm Aux preconditioning at day 3 of challenge resulted in robust IgA titers towards smooth STm (Fig. 4e, panel Day 3), IgA binding to rough STm was significantly reduced (Fig. 4f).
These data show that Salmonella T3SS-dependent virulence functions signify the mucosal immunogenic efficacy of life STm Aux in absence of inflammation. The underlying T3SSdependent IgA B cell response is characterized by a less Oantigen-restricted bacterial surface reactivity.  PRR signaling redundancy in induction of immunity. Innate pathogen recognition through PRRs is critical in the defense against primary STm infection. MYD88 knockout mice lacking TLR and IL1R family downstream signaling are consequently severely impaired in innate immunity against mucosal STm infection 17,19 . Canonical Caspase-1-dependent and noncanonical Caspase-11 (Caspase-4 in humans)-dependent inflammasome activation have also been implicated in innate immune control of STm infection 32 . The NLRC4 inflammasome is activated by the SPI1 T3SS needle complex proteins and therefore may mediate innate recognition of T3SS-competent intestinal STm specifically 3 . Moreover, the NOD1/NOD2 nodosome has been reported to respond to bacterial pathogenicity by sensing the cytoplasmic activities of Salmonella SPI1 T3SS-1 effector proteins 4 . Are these also factors individually important for the induction of functional adaptive immunity in the absence of an inflammatory response? To address this we tested the hypothesis that deficiencies for innate recognition pathways critical in innate immune defense also affect induction of functional adaptive immunity by live STm Aux . Mice deficient in (i) TLR/IL1R family adaptor proteins MYD88 and TRIF, (ii) Caspase-1 and Caspase-11, (iii) NLRC4, and (iv) NOD1 and NOD2 were derived germ free and compared with innate immunocompetent control mice for their adaptive immune responses towards live STm Aux .
Using STm Aux avoids bacterial overgrowth of severely innate immunodeficient hosts that lack control and containment of intestinal microbes, leading to increased mucosal penetration also of attenuated, avirulent, and commensal bacteria. In MYD88/ TRIF double-deficient mice this has been shown to result in abnormally high systemic exposure to gut commensals and consequent compensatory B cell immunity. Involvement of redundant innate signaling pathways triggered by massively increased microbial loads has been postulated to be responsible but has not been characterized further 33 . The STm Aux model in germ-free mice, however, uniquely fixes the bacterial load per animal and consequently avoids bacterial overgrowth to skew immune activation.
First, germ-free MYD88/TRIF double-deficient and wild-type control mice were enterally conditioned with live STm Aux . Four weeks after the first treatment all mice were challenged orally with wild-type STm harboring an intracellularly inducible GFP reporter plasmid (pM973) 17 and studied at day 3 after the challenge (Fig. 5a). Control groups of both genotypes were challenged but without STm Aux preconditioning. Quantification of cecal mucosal (Fig. 5b, Supplementary Fig. 11A), mLN, liver, and spleen (Fig. 5c) burdens of STm[pM973] revealed that the induction of functional immunity by live STm Aux was robust even in the highly susceptible MYD88/TRIF double-deficient mice. At day 3 post challenge, bacterial loads in mLN, livers, and spleens of STm Aux -treated MYD88/TRIF-deficient mice were similar to, and in cecal mucosal tissues even lower than, those in the STm Aux -treated wild-type animals. In accordance with this relatively greater effect of STm Aux treatment in mutant than wildtype mice, a two-way ANOVA revealed a significant interaction between the effects of genotype and STm Aux treatment (p < 0.0001 for all panels; for detailed statistical metrics see Statistical Analysis file available as Supplementary Information). Quantitation of cecal luminal lipocalin-2 and histopathology scores both confirmed the protective effects of STm Aux treatment within each mouse genotype ( Supplementary Fig. 11B, C), although these two readouts themselves are MYD88/TRIF-dependent 17,34 and should therefore be compared between both mouse genotypes with caution.
We next tested the ability of live STm Aux to induce adaptive immunity in NOD1/NOD2-double-knockout mice (Fig. 6a-c), NLRC4-deficient mice, and Caspase-1/11 double-deficient mice ( Fig. 6d-f), all of which at day 3 of challenge were found to have no deficiency in mounting functional mucosal immunity towards live STm Aux conditioning (two-way ANOVA, for detailed statistical metrics see Statistical Analysis file available in the Supplementary Information).
These results show that MYD88/TRIF, Caspase-1/Caspase-11 inflammasome, and NOD1/NOD2 nodosome signaling were individually redundant for the induction of adaptive immunity by live STm Aux in the absence of inflammation. Their role in complementing adaptive immunity in pathogen clearance at later stages of secondary infection is likely functionally important, although not apparent at day 3 of challenge.
Microbiota-dependent colonization and niche competition. So far, the fully reversible germ-free mouse model uniquely had allowed the quantitative study of the immunogenicity of different phenotypes of STm Aux in a very clean system. However, in real-life situations STm Aux would interact also with the indigenous gut microbiota, which we hypothesized to provide crossfeeding of the required cell wall metabolites in vivo. This may delay STm Aux intestinal luminal clearance in colonized mice. We tested this hypothesis using a well-established gnotobiotic mouse model that is stably colonized with 12 representative murine intestinal taxa [stable defined moderately diverse mouse microbiota (sDMDMm) 35 ] all of which are fully sequenced and openly available as pure cultures from the "Deutsche Sammlung von Mikroorganismen und Zellkulturen" (DSMZ) 36,37 . The sDMDMm model has proven merit for the study of intestinal STm infection and its interaction with the commensal microbiota without the need for harsh antibiotic treatments, and shows relevant phenotypic effects such as limiting the colonization of STm 38 . Fig. 3 Optimal mucosal efficacy of STm Aux is viability dependent. a Germ-free mice were enterally conditioned with four doses of 10 10 CFU STm Aux (filled red triangles, n = 10 animals), PAA-killed STm Aux (filled blue diamonds, n = 11 animals), or were left untreated (gray-filled circles, n = 10 animals). Four weeks after the first treatment (day 0) mice were challenged with wild-type STm (10 4 CFU). Mice were sacrificed at day 1 and day 4 after challenge with wild-type STm, respectively. Each symbol represents one individual. b STm-specific titer (logEC 50 ) of intestinal secretory IgA determined by live bacterial flow cytometry. c, d Lipocalin-2 concentration in feces (c) and cecal content (d) at day 0-4 after challenge. (e) Cecal histopathology score at days 1 and 4 after challenge, respectively. Each symbol represents one individual. Data points depicted with an arrow are shown in panel f. f Cecal histology at day 1 and day 4 after challenge, respectively. H&E staining of cryosections. Scale bar = 100 μm. g Bacterial burden of wild-type STm in mLNs, spleens, and livers at days 1 and 4 after challenge, respectively. h Fecal colonization of wild-type STm at days 0-4 after challenge (inset graph: quantification of STm Aux in feces at day 0 confirming germ-free status at day 0). i Cecal luminal colonization of wild-type STm at day 4 after challenge. Statistics: bars indicate mean (b-d, g-i) or median values (e). Horizontal, dotted lines indicate the detection limit (b-d, g-i). Panel b was analyzed with an unpaired two-tailed t-test (control group excluded from test). Panels c, d, and g were analyzed with ordinary one-way ANOVA and Dunnett's post hoc test (comparison to control group). Panel e was analyzed with a two-sided multicomparisons Kruskal-Wallis test and Dunn's post hoc test. Source data are provided as a Source Data file. Detailed statistical metrics are available in the Supplementary Statistical Analysis file.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-15891-9 ARTICLE NATURE COMMUNICATIONS | (2020) 11:1978 | https://doi.org/10.1038/s41467-020-15891-9 | www.nature.com/naturecommunications Following a single inoculation with 10 7 STm Aux by gavage, sDMDMm mice showed efficient and stable colonization of STm Aux , reaching luminal densities similar to those of isogenic non-auxotrophic strains, including partly attenuated SPI2 TTSS-deficient (ΔssaV) and avirulent SPI1/SPI2 doubledeficient (ΔinvC ΔssaV) STm (Fig. 7a). STm Aux did not revert to lose its auxotrophic phenotype during these experiments (no recovery of STm growth from ex vivo intestinal samples in non-supplemented control medium). Even STm Aux re-isolated from an sDMDMm mouse after 8 months colonized germ-free mice fully reversibly. Following gavage of 10 10     per group) had recovered to germ-free status at day 2 post inoculation. Despite efficient luminal colonization and evidence for epithelial invasion of STm Aux (Fig. 7e), neither deep mucosal penetration to mLN and systemic organs nor mucosal pathology were evident in either wild type (Fig. 7b-d) or MYD88/TRIFdeficient (Fig. 7f) sDMDMm mice. Thus, while crossfeeding by sDMDMm organisms can rescue gut luminal colonization, it was insufficient to recover pathogenicity of STm Aux , which is consistent with the local intestinal luminal confinement of the crossfeeding microbiota and the activity of D-amino acid degrading enzymes in host tissues and intestinal mucus 39 .
Nevertheless, induction of STm-specific IgA was seen after 4 weeks of colonization, which, in contrast to colonization efficiency, was dependent on SPI1 T3SS competence (Fig. 7g,  Supplementary Fig. 12). As an added host benefit, stably colonizing STm Aux further provided robust niche competition to a subsequent oral challenge by wild-type STm (Fig. 7h blue symbols, and Supplementary Fig. 13B). Notably, pre-colonization with SPI1 T3SS-incompetent STm Aux ΔinvC provided only partial niche competition (Fig. 7h, black symbols, Supplementary  Fig. 13B). SPI1/SPI2 double-deficient STm Aux T3SS− showed the exact same phenotype (Fig. 7h, green symbols, Supplementary  Fig. 13B), supporting the conclusion that SPI1 T3SS-dependent virulence factors are mainly responsible. In RAG knockout mice also T3SS-competent STm Aux showed inefficient intestinal niche competition ( Supplementary Fig. 14A-D). These findings are consistent with the interpretation that STm Aux -induced host NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-15891-9 ARTICLE immunity synergizes with niche competition by STm Aux in protection against wild-type STm challenge. Measurements of cecal luminal lipocalin-2 and challenge bacterial burden in mLN, liver, and spleen at day 4 of wild-type STm challenge support this conclusion (Fig. 7i, j, Supplementary Fig. 14D). In a second context, streptomycin pre-treated conventional mice, a widely used mouse model for nontyphoid invasive salmonellosis 9,24,40 , were also permissive for extended gut luminal colonization of STm Aux (Supplementary Fig. 13C). These data show that in the colonized mouse model, microbiotasyntrophic STm Aux more closely mimics the natural pathogen in terms of intestinal luminal colonization and virulence factor-driven   induction of IgA immunity 12 . Thus, independently of germ-free conditions, also stable intestinal STm Aux colonization allows uncoupling of intestinal immunogenicity from pathogenicity, with the added benefit of luminal pathogen niche competition.

Discussion
Fully attenuated or inactivated pathogens have long been noted to be poorly protective mucosal immunogens compared to more virulent strains [11][12][13]41 . This has been attributed mainly to the capacity of virulent pathogens to induce more vigorous innate immune responses 6 and to penetrate into and overgrow inductive sites of the mucosal immune system 11 . Here we show specifically that invasive Salmonella cells expressing live type 3 secretion systems are recognized by the immunogenic response, independently of their propensity to deeply penetrate and replicate as live organisms inside host tissues or the induction of a marked mucosal inflammatory response. These data suggest that the mucosal immune system reacts not only to a damaging infection but can also recognize stereotypic activities of pathogens more directly, and thus potentially more sensitively and rapidly. Consequently, a small number of highly transient mucosal exposures with virulence factor-competent STm Aux robustly induce highly effective immunity in germ-free mice, in the absence of an inflammatory response. The underlying B cell response induced by live, virulence factor-competent STm Aux is characterized by the production of intestinal IgA with increased O-serotypeindependent Salmonella surface reactivity. Additional future work will be required to address which effector T cell activities may additionally contribute to STm Aux -induced immunity 42 . It has been described previously that bacterial viability itself is an important determinant of bacterial immunogenicity, independent of pathogenicity and replication competence 43 . Live apathogenic bacteria are more immunogenic live than killed when administered parenterally 43 . This difference was revealed to be mediated by innate immune recognition of bacterial messenger RNA (mRNA), highly unstable, hence normally viability-associated, molecules. The underlying sensing pathway for bacterial mRNA was shown to be dependent on TLR8 and TRIF in humans, and on TRIF, Caspase-1 and Caspase-11 in mice [43][44][45] .
Here we observed only a minor difference in intestinal mucosal immunogenicity between avirulent live and killed STm Aux (see Fig. 4), which may be mediated by the same mechanism. Virulence factor-competent invasive STm Aux , however, was much more efficacious. Its epithelial invasiveness may increase subepithelial live antigen delivery and consequently prime immunity more efficiently by delivering live bacteria into the tissues. However, its immunogenicity was robust even in MYD88/TRIF and Caspase-1/-11 deficient mice, and thus may not be fully explained by the same live bacterial sensing pathway. Our data confirm and extend previous findings of functional redundancy between innate and adaptive immune responses in the control of intestinal commensal bacteria 33 and the efficacy of established model vaccines with adjuvant 46 in MYD88/TRIF double-deficient mice. Here we show that this extends also to intestinal pathogenic bacteria. The remarkable robustness of this system may represent an evolutionary adaptation to pathogens that evade or alter the innate immune defense.
Long-established live STm vaccine strains like SPI2 T3SSdeficient 47 and aromatic amino acid auxotrophic aro mutants 48 of STm also are effective mucosal immunogens, but are not fully growth deficient in host tissues and consequently considered dangerous for HIV positive and other immunocompromised individuals (reviewed in ref. 14 ). This has so far ruled out approval for human application. On the other hand, peptidoglycan metabolite auxotrophic STm strains similar to the one we presented in this paper have been developed previously 49 but in this form have been considered insufficiently immunogenic because of their poor mucosal penetration. This conclusion is however predicated on the preclinical study mainly in conventional rodent models that are (like humans) intestinally colonization resistant against Salmonella 9 . In this context, when STm Aux proliferation in the intestinal lumen is inhibited by the competing microbiota, its colonization dynamics would be expected to be more similar to the germ-free mouse model, and it may consequently require very high oral doses (as we saw in germ-free mice) to be efficacious. Instead, the field has moved into the direction of developing more sophisticated strains that display regulated delayed in vivo attenuation/lethality phenotypes, allowing for transient survival, replication, and tissue invasion in vivo 11 . These highly innovative approaches are inherently more difficult to combine with safety parameters matching those of the constitutively D-Ala/Dap auxotrophic strain. The presented experiments in non-colonizationresistant mouse models highlight yet another possible strategy. The remarkably efficient gut luminal microbiota-syntrophy permitted extended mucosal stimulation with live virulence factorcompetent STm Aux , without compromising the strain's deficiency in causing pathology and systemic infection. This phenotype could potentially be exploited further by metabolic engineering of STm Aux strains to gain intestinal colonization efficiency, or by temporal reduction of colonization resistance in the host at the time of treatment (preferably other than by antibiotic treatment). However, given that our conclusions so far are based on mouse models that have laboratory levels of microbiota complexity, additional work in more relevant preclinical models will be necessary to assess potential translatability of these findings for veterinary or human medical applications. Fig. 6 NOD1/2, NLRC4 and Caspase-1/11 are individually redundant for mucosal induction of adaptive immunity by live STm Aux . a Germ-free NOD1/2double-deficient mice (open symbols) and wild-type control mice (filled symbols) were either enterally conditioned with three doses of 10 10 CFU of STm Aux (red triangles, n = 11 NOD1/2 KO animals and n = 5 wild-type animals examined over two independent experiments) or left untreated (gray circles, n = 11 NOD1/2 KO animals and n = 6 wild-type animals examined over two independent experiments). Twenty-seven days after the first treatment (day 0) all mice were challenged with wild-type STm (10 3 CFU) and sacrificed at day 3 after challenge. b Bacterial burden of wild-type STm recoverable from mLN, spleen, and liver at day 3 after challenge. c Lipocalin-2 concentration in cecal contents at day 3 after challenge. d Germ-free NLRC4 −/− mice (open diamonds), Caspase-1/11 −/− mice (CASP1/11 −/− , open triangles), and control mice (CASP1/11 +/− NLRC4 +/+ littermate control mice; filled triangles) were either enterally conditioned with three doses of 10 10 CFU of STm Aux (red symbols, n = 7 NLRC4 −/− animals, n = 7 Caspase-1/11 −/− , and n = 7 Caspase-1/11 +/− animals examined over two independent experiments) or left untreated (gray symbols, n = 5 NLRC4 −/− animals, n = 8 Caspase-1/11 −/− , and n = 7 Caspase-1/11 +/− animals examined over two independent experiments). Twenty-seven days after the first treatment (day 0) mice were challenged with wild-type STm (10 3 CFU) and sacrificed at day 3 after challenge. e Bacterial burden of wild-type STm recoverable from mLN, spleen, and liver at day 3 after challenge. f Lipocalin-2 concentration in cecal contents at day 3 after challenge. Each symbol represents one individual. Statistics: bars indicate mean (b, c, e, f). Horizontal, dotted lines indicate the lower detection limit. Panels b, c, e, f were analyzed with a two-way ANOVA (host genotype and treatment as factors) and Sidak multiple comparison correction. The data were pooled from two independent experiments. Source data are provided as a Source Data file.  Table 1. Auxotrophic strain HA135 (STm Aux , UK-1 background) [ΔmetC::TetRA Δalr ΔdadX Δasd] was generated from strain χ9052 [Δalr3 ΔdadB4 ΔasdA33] by replacing the coding region of metC with a TetRA resistance cassette by Lambda Red recombineering using recombineering plasmid pSIM5 (ref. 50 ) as described in ref. 51 . Isogenic mutant alleles ΔinvC::aphT, ΔinvC::aphT ΔssaV::cat, and Δ(avrA-invH::cat) Δ(ssaG-ssaU::aphT) were transferred into the STm and STm Aux backgrounds by phage P22-mediated transduction using the donor strains M736, M73831 and χ9650 (ref. 28 ), respectively, as described 52 . Auxotrophic strain HA623 (SL1344 background) [ΔmetC ΔalrN ΔalrP Δasd] was generated from strain SL1344 (SB300) by in-frame deletion of each gene. This was achieved be generation of four single deletion mutants in SL1344 using the plasmid pSIM6 encoded Lambda Red recombinase system 50 for allelic exchange of the coding sequence (leaving the stop codon) with a Tet selectable tetA-sacB cassette, followed by four sequential rounds of P22 transduction followed by Lambda Red recombineering mediated removal of the tetA-sacB cassette by counterselection as described 53 , leading to quadruple deletion mutant HA623. HA630 was generated by Lambda Red recombination of a tetRA resistance cassette into the asd deletion site of HA623. The mutagenesis primers used are listed in Table 2. Auxotrophic SPI1 effector gene sopE sopE2 sipA triple mutant H727 was constructed in parent strain SL1344 as described previously 31 .
Cellular invasion assays. HeLa (Kyoto) cells were seeded into 24-well dishes and were grown for 1 day until 80% confluence was obtained. HeLa cells were cultured Fig. 7 Efficient colonization and immune induction by auxotrophic STm in microbiota-associated mice. a sDMDMm mice were gavaged with a single dose of 10 7 CFU of either STm Aux (blue open circles, n = 21), STm T3SS− (green squares, n = 9), STm ΔssaV (black triangles, n = 5), or wild-type STm (red triangles, n = 6). Time course of viable bacteria of each strain recoverable from feces. b Lipocalin-2 concentration in cecal contents at day 2, 4, and 28 after inoculation with the indicated STm strains. Pictures show representative H&E stainings of ceca at day 4 post inoculation with either STm WT or STm Aux . c Representative cecal histology on day 4 after colonization with STm Aux , H&E staining, scale bar 100 µm. d Bacterial burden of indicated STm strains recoverable from mLN, spleen, and liver at day 2, 4, and 28 after initial colonization. e Confocal immunofluorescence microscopy of cecum tissue showing an epithelial cell invaded by STm Aux . Green, STm Aux harboring ssag::eGFP reporter plasmid pM973; blue, DNA (DAPI); gray, F-actin (phalloidin). Dotted yellow lines outline the border of the epithelium facing the intestinal lumen (Lu) and lamina propria (Lp), respectively. Scale bar = 10 µm (n = 6 animals). f Bacterial burden of indicated STm strains recoverable from mLN, liver, and spleen at day 3 post inoculation in MYD88/TRIF double KO mice (n = 3 animals per treatment group).     The Jackson Laboratory in the form of cryopreserved embryos and transferred into germ-free recipients in the Clean Mouse Facility, University of Bern. Germ-free and sDMDMm wild-type C57BL/6 animals, germ-free JH −/− mice 59 and RAG −/− mice 60 were maintained at the Clean Mouse Facility, University of Bern. Gnotobiotic sDMDMm mice have been generated at the Clean Mouse Facility of the University of Bern by inoculation of germ-free C54BL/6 mice with purified culture of the murine intestinal bacterial consortium Oligo-MM12 (ref. 38 ) and stably maintained in flexible film isolators under strictly axenic conditions. sDMDMm RAG −/− mice and MYD88 −/− TRIF lps/lps mice were generated by cohousing of the genetically modified germ-free mice with gnotobiotic wild-type sDMDMm mice for 4 weeks.
SPF C57BL/6 mice were purchased from Charles River (France) and maintained at the Central Animal Facility of the Department of Biomedical Research, University of Bern. For infections in the streptomycin pretreatment model 24 SPF mice were pre-treated with 20 mg of streptomycin dissolved in sterile water prior to infection with STm by gavage.
Infection and colonization experiments were performed under strict aseptic conditions. Mice were derived and maintained germ free in flexible film isolators 61 (including the duration of transient auxotrophic bacterial conditioning) or autoclaved Sealsafe-plus IVC cages (Tecniplast, Italy; during STm challenge and short-term infections) at the Clean Mouse Facility (CMF) of the University of Bern. Animals were provided with autoclaved mouse chow (Kliba 3307) and water ad libitum. Germ-free status of all animals was routinely monitored using culturebased and culture-independent methods established by the Clean Mouse Facility, DKF University of Bern. Mice were infected with 200 μL STm suspension.
Bacteria for enteral inoculation were grown under SPI1-inducing conditions. Auxotrophic STm were inoculated into 10 mL D-Ala-(200 μg/mL) and m-Dap-(50 μg/mL) supplemented LB containing 0.3 M NaCl and incubated shaking at 150 r.p.m., at 37°C for 16 h. The resulting bacterial cultures were diluted 10 8 -fold in 500 mL fresh medium and incubated under the same conditions for 15 more hours. STm were harvested by centrifugation (15 min, 4816 × g, 4°C), washed twice with cold PBS, and resuspended to the appropriate densities. For peracetic acid (PAA) inactivation, an aliquot of auxotrophic STm was resuspended in 10 mL 1% peracetic acid for 1 h at room temperature. The inactivated STm suspension was washed with PBS and resuspended to the the appropriate density. Sterility of PAAkilled inocula was confirmed by standard culture methods. Wild-type STm cultures were inoculated from a single colony in 10 mL 0.3 M sodium-chloride/LB and incubated at 150 r.p.m., at 37°C for 16 h. Wild-type STm cultures were diluted 1:20 into 40 mL fresh medium and incubated at the same conditions for 5 h.
Isolation of intestinal secretory IgA. Intestinal IgA lavages were collected by rinsing the small intestine with 5 mL of 1% soybean-trypsin inhibitor/0.05 M EDTA/PBS. The intestinal lavages were spun at 4816 × g, >20 min, 4°C. The supernatant was sterile-filtered (0.22 µm cut-off size) to remove bacteria-sized particles and stored long-term in aliquots frozen at −20°C.
Immunoglobulin repertoire sequencing. Germ-free C57BL/6 mice were orally administrated with 3 × 10 10 STm Aux or STm Aux ΔinvC ΔssaV three times at 7-day intervals. Naïve germ-free mice served as naïve controls. Twenty-eight days post last administration, ileum and MLN were dissected and snap-frozen in Trizol reagent (Life Technologies) using liquid nitrogen. Thawed tissues were homogenized (Retsch bead-beater) in 1 mL of Trizol reagent. Two hundred microliters chloroform was added to samples and centrifugation (12,000 × g, 15 min, 4°C) was performed. The upper phase containing RNA was collected, and RNA was precipitated with ice-cold isopropanol. After washing once with 75% (v/v) ethanol, RNA pellet was dried and resuspended in RNase-free water. Nanodrop2000 (Thermo Scientific) was used to quantify RNA concentrations and purity.
To prepare IgA amplicons, cDNA was synthesized by mixing 700 ng of RNA, 1 μL of 2 μM gene specific primer mix (as previously described 62 , 1 μL of 10 mM dNTP (containing dATP, dCTP, dGTP, and dTTP at a final concentration of 10 mM, Invitrogen) and topped with dH 2 O to 13 μL. Samples were heated to 65°C for 5 min, and then cooled for 1 min on ice. Four microliters 5X first strand buffer (Invitrogen), 1 μL 782 0.1 M DTT (Invitrogen), 1 μL RNaseOUT (Invitrogen), and 1 μL Superscript III RT enzyme (Invitrogen) were added to each reaction, mixed, and incubated at 55°C for 50 min. A heat inactivation at 70°C for 15 min was done to stop the reaction. Five microliters of synthesized cDNA library was used as a template DNA for amplicon PCR PlatinumTaq PCR buffer (Qiagen) following the manufacturer's instruction. Primers used in the PCR reaction listed below were described previously 63 . PCR products were electrophorized on 1.5% agarose gel and purified with the QIAquick Gel Extraction kit (Qiagen). The purified DNA was quantified using Qbit (Thermofisher). Sequencing adapters (Nextera® XT Index Kit, Illumina) were linked to each amplicon by doing a second PCR. After testing with a Fragment Analyzer™ (Advanced Analytical), amplicons with sequencing adaptors were pooled for sequencing on the MiSeq Illumina sequencer using paired 250 bp mode.
For primer sequences used view Table 3.
Antibody repertoire sequencing analysis. B cell IgA receptor heavy chain libraries were prepared as previously described 64 and sequenced on the Illumina MiSeq platform (2 × 250 cycles, paired-end). Output files were preprocessed (VDJ alignment, clonotyping) using MiXCR (v3.0.12). Clonotypes were defined by 100% amino acid sequence identity of CDR3 regions. Annotation of the different segments was defined by MiXCR according to the nomenclature of the Immunogenetics database (IMGT) 65 . MIXCR output files were further processed in the postprocessing tool-suite: VDJtools 66 . Further filtering was applied in order to keep only productive sequences if: (i) they were composed of at least four amino acids and (ii) had a minimal read count of 2 (ref. 63 ) and were in-frame. Repertoire overlap was measured by calculating the geometric mean of relative overlap frequencies between CDR3 amino acid sequence usage. The relative overlap similarity was represented on a multi-dimensional scaling (MDS) plot.
Histology and pathological evaluation of cecum tissue. For each individual, proximal and distal cecum tissue was embedded in OCT compound (Tissue Tek, DC6994583) and frozen in liquid nitrogen. Three consecutive 6 μm cryosections of each tissue were mounted on glass slides and stained with hematoxylin and eosin following standard protocols. Histopathology was scored in a blinded manner according to the severity of submucosal edema (0-3 score points), the number of polymorphonuclear granulocytes per high-power field in the lamina propria (0-4 score points), reduced numbers of goblet cells (0-3 score points), and epithelial damage (0-3 score points) resulting in a total score of 0-13 points 24 . The mean combined pathological score of proximal and distal cecum is reported.
Fluorescence microscopy of tissue invaded STm. Tissue invaded intracellular STm harboring ssaG::eGFP reporter plasmid pM973 were visualized and quantified in cecum cryosections prepared from paraformaldehyde-fixed and cryo-embedded cecum tissue as described previously 17 . Sections were stained with DAPI (Sigma, diluted 1:2000) and Phalloidin ATTO 647 (Sigma, diluted 1:500). Up to 12 nonconsecutive sections per animal were quantified visually using a Zeiss AXIO Imager.M1 microscope equipped with an EC-Plan-NEOFLUAR 40C/1.3 Oil objective and ×10/23 oculars. One high-power field measures approx. 40,000 µm 2 . Quantitation was carried out in a blinded manner. Images were recorded on a Zeiss LSM710 laser scanning confocal microscope using the Zeiss ZEN 3.1 software. Images were analyzed with the Image J Fiji package.
mRNA quantification in cecal tissue by qPCR. Cecum tissue was collected 6 h after infection. Immediately, the tissue was washed in PBS and preserved in RNAlater (Qiagen). The total RNA was extracted from approximately 15 mg tissue, using the RNeasy mini kit (Qiagen). The extraction quality was assessed with Agilent RNA 600 Nano Kit (Qiagen) and reached minimally RIN 9. In total, 5 μg mRNA samples were reversed with RT 2 easy first strand kit (Qiagen). cDNA libraries were analyzed in a Viia7 Real-Time PCR System and the Viia7 Real-Time PCR System acquisition software (Thermo Scientific) using a the RT 2 profiler PCR array quantifying murine Crohn's disease-related markers (PAMM-169Z, Qiagen) and SYBR green reagents (Qiagen). Five housekeeping genes (Actb, B2m, Gapdh, Gusb, and Hsp90ab1) were averaged and used for calculating ΔCT (=CT sample −CT houskeeping ). The upper CT limit was fixed to 35 cycles.
Live bacterial flow cytometry. Live bacterial flow cytometry quantification of bacterial-specific intestinal IgA titers (expressed as LogEC 50 values) were determined as previously described in ref. 25 . Briefly, STm were cultured under SPI1inducing conditions 67 as described in the Cellular Invasion Assays section. Subsequently, 1 mL of the culture was pelleted at 4816 × g in a Heraeus Fresco 21 centrifuge. The pellet was washed and resuspended to a density of 10 7 CFU/mL in sterile-filtered 2% BSA/0.005% NaN 3 /PBS. Intestinal IgA lavages were collected as described above. Intestinal lavages were serially diluted in sterile-filtered 2% BSA/ 0.005% NaN 3 /PBS. Serially diluted Ig-solutions and bacterial suspension were mixed 1:1 and incubated at 4°C for 1 h. Bacteria were washed twice in sterilefiltered 2% BSA/0.005% NaN3/PBS before re-suspension in monoclonal FITC-antimouse IgA (clone 10.3; Becton Dickinson) or PE-anti-mouse IgG1 (clone A85-1; Becton Dickinson) and FITC-anti-mouse IgG2b (clone R12-3; Becton Dickinson). After a further hour of incubation, the bacteria were washed once with PBS/2% BSA/0.005% NaN3/PBS and then resuspended in 2% paraformaldehyde (PFA)/PBS for acquisition on a Becton Dickinson FACSArray SORP or Beckman Coulter Cytoflex S using FSc (forward scatter) and SSc (side scatter) parameters in logarithmic mode. Flow cytometric gating strategy is shown in Fig. S15. Data were analyzed using FlowJo software (Tree Star), and titers were calculated by fitting four-parameter logistic curves 25 .
Statistics. Statistics were analyzed using Prism 8 for Windows (GraphPad software Inc.). The specific statistical tests used are indicated in the figure legends. Detailed statistical information is provided in the statistical data analysis file available online in the supplementary material.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The dataset supporting the conclusions of this article is available as a Source Data file.