Intestinal epithelial NAIP/NLRC4 restricts systemic dissemination of the adapted pathogen Salmonella Typhimurium due to site-specific bacterial PAMP expression.

Inflammasomes can prevent systemic dissemination of enteropathogenic bacteria. As adapted pathogens including Salmonella Typhimurium (S. Tm) have evolved evasion strategies, it has remained unclear when and where inflammasomes restrict their dissemination. Bacterial population dynamics establish that the NAIP/NLRC4 inflammasome specifically restricts S. Tm migration from the gut to draining lymph nodes. This is solely attributable to NAIP/NLRC4 within intestinal epithelial cells (IECs), while S. Tm evades restriction by phagocyte NAIP/NLRC4. NLRP3 and Caspase-11 also fail to restrict S. Tm mucosa traversal, migration to lymph nodes, and systemic pathogen growth. The ability of IECs (not phagocytes) to mount a NAIP/NLRC4 defense in vivo is explained by particularly high NAIP/NLRC4 expression in IECs and the necessity for epithelium-invading S. Tm to express the NAIP1-6 ligands-flagella and type-III-secretion-system-1. Imaging reveals both ligands to be promptly downregulated following IEC-traversal. These results highlight the importance of intestinal epithelial NAIP/NLRC4 in blocking bacterial dissemination in vivo, and explain why this constitutes a uniquely evasion-proof defense against the adapted enteropathogen S. Tm.


INTRODUCTION
Oral bacterial infection causes localized gastrointestinal disease, but pathogen dissemination (from here on termed migration) to systemic sites can lead to life-threatening complications. Multiple host defenses therefore cooperate to limit mucosal infection and pathogen spread. 1 The intestinal mucus lining and antimicrobial peptide secretion restrict mucosal invasion. [2][3][4][5][6][7] Intestinal epithelial cells (IECs) and lamina propria phagocytes mount cell-intrinsic programs and release pro-inflammatory signals to counter pathogens that breach this first barrier. 2,[8][9][10][11] Diverse immune cell types also patrol systemic organs and prevent excessive pathogen replication. [12][13][14][15] This multilayered host defense is triggered by pattern recognition receptors (PRRs) that detect pathogenassociated molecular patterns (PAMPs), exposed by invading microbes. However, it is not fully understood which innate defense mechanisms act at which stage of the infection in vivo and what quantitative impact can be assigned to each layer of the defense.
The prototypic enteropathogen Salmonella enterica Typhimurium (S. Tm) colonizes the gut lumen, invades the mucosa, and can migrate to mesenteric lymph nodes (mLN), spleen, and liver, causing life-threatening infection, e.g. in immunocompromised hosts. 1 Luminal S. Tm expresses flagella to target gaps in the mucus layer, 4,16 a type-III-secretion system (TTSS-1) to actively invade IECs, 17 and, upon cell invasion, a second TTSS (TTSS-2) to promote IEC traversal. 18 Mononuclear phagocytes (e.g., dendritic cells (DCs), macrophages) are involved in multiple steps of the S. Tm infection cycle. They facilitate S. Tm uptake across the epithelial barrier, [19][20][21] lodge S. Tm within the lamina propria and at systemic sites, 18,22 and act as vessels for S. Tm migration between organs. 23 Consequently, restriction of systemic S. Tm infection may depend on the capacity of both IECs and phagocytes to recognize the pathogen through PRR(s) and mount appropriate counter measures. For an adapted pathogen like S. Tm, this task is complicated by the bacterium's evolved ability to evade PRR recognition, through e.g., context-dependent regulation of its gene expression or inhibition of host-cell signaling. 1,24,25 Inflammasomes are multimeric signaling complexes that assemble in the host cell cytosol upon sensing of PAMPs or cellular damage by PRRs of the Nod-like receptor family (NLRs; NLRP1, NLRP3, NLRC4), Aim2, or Pyrin. 26 Inflammasome assembly causes cleavage of pro-inflammatory Caspase-1, 27 secretion of RESULTS NAIP/NLRC4 potently restricts S. Tm migration from the gut lumen to systemic sites Our in vivo analysis of the NAIP/NLRC4-mediated defense focused on the S. Tm infection dynamics during the first 24h after orogastric inoculation in the Streptomycin pretreated mouse model. 56 Due to its highly reproducible, fast and robust kinetics, this infection setup is ideally suited to study innate immune restriction of enteropathogen dissemination to systemic sites. Within 24 h, the pathogen colonizes the gut lumen of Streptomycin pretreated mice, invades the gut mucosa, and disseminates systemically via gut-draining mLN. 57 The size of a pathogen population inside the mLN is the product of several parameters, i.e., bacterial immigration to this site, replication on the way to and within the organ, emigration to other sites and elimination of the pathogen by the host. In contrast to classical selective plating of infected organs, which merely provides a snapshot of the bacterial population size, infections with mixtures of wild-type isogenic tagged strains (WITS), combined with mathematical modeling can reveal the dynamic parameters and thereby provide essential information on pathogen restrictive mechanisms (Fig S1). [58][59][60][61][62][63] To establish the quantitative infection parameters, we infected Streptomycin pretreated mice with a mixed inoculum (5 × 10 7 total CFU per gavage), comprised of non-tagged S. Tm and seven WITS each spiked in at a 1:140 dilution (i.e., the WITS strains together made up 5% of the inoculum). By conventional plating, we detected~10-fold increased total S. Tm loads in the mLN of Nlrc4 −/−64 mice compared to their heterozygous littermate controls at 24 h post-infection (hpi) (Fig. 1a, "all S. Tm"). This phenotype was confirmed by selective plating of the WITS population (Fig. 1a, "WITS"). In line with earlier work establishing that gut luminal S. Tm loads are independent of gut inflammation during the first 24h, 65 we did not detect any changes in luminal colonization ( Fig S2A). To determine the cause of the elevated mLN pathogen loads, we analyzed the number of distinct WITS recovered from this organ. In the mLN of infected Nlrc4 −/− mice all, or close to all, of the seven WITS were recovered at 24 hpi. By striking contrast, Nlrc4 +/− littermate control mLN harbored on average only two WITS (Fig. 1b).
The higher diversity of WITS in the mLN of Nlrc4 −/− mice suggested that more bacteria seed the organ to give rise to the mLN-lodged pathogen population. To test this hypothesis, we applied a mathematical model for analysis of population dynamics, considering pathogen migration (µ) and net growth (r 59 ; see materials and methods for details). This established that the S. Tm migration rate (µ) from the cecal lumen to the mLN was elevated by 8.7-fold in Nlrc4 −/− mice compared to controls (Fig. 1c). Notably however, the presence or absence of NLRC4 did not affect the pathogen's net replication rate (r) within the mLN (Fig. 1d). We repeated these experiments in Naip1-6 Δ/Δ66 mice, which lack the receptors activating the NLRC4 inflammasome. 66 These mice phenocopied Nlrc4 −/− animals, showing a 6.7-fold increase in S. Tm migration to the mLN, but a similar within-mLN pathogen replication rate as control animals ( Fig. 1e-h, S2B).
Taken together, these findings show that the NAIP/NLRC4 inflammasome is of key importance for restricting S. Tm migration from the gut lumen to the mLN, but does not affect the pathogen's replication within this target organ. By inference, this indicates that phagocyte NAIP/NLRC4 is dispensable for controlling pathogen growth in the mLN. It remained to be established if IEC or phagocyte NAIP/NLRC4 could explain restriction of pathogen migration from the gut, and if additional inflammasomes also impact this process. NLRP3 and Caspase-11 are dispensable for control of S. Tm dissemination from the gut, even in the absence of NLRC4 The significance of the NLRP3 inflammasome and the noncanonical Caspase-11 inflammasome during oral S. Tm infection and their role in limiting pathogen levels at systemic sites are ambiguous. 11,37,39,43,49,50 To quantitatively assess the involvement of these inflammasomes, we infected Nlrp3 −/−67 and Casp11 −/−41 mice using the experimental setup described in Fig. 1. We did not observe any difference between the total bacterial loads in the cecal lumen and mLN, the WITS loads, or the numbers of WITS recovered from the mLN of either knockout line, when compared to their respective littermate controls (Fig S2C-D, S3A-B). Moreover, mathematical inference confirmed equivalent values for migration (µ) and net replication (r) in knockouts and controls ( Fig. 2a, b; values for Nlrc4 −/− mice replotted from Fig. 1c, d for comparison). Hence, neither NLRP3, nor Caspase-11, impact the dynamic parameters of oral S. Tm dissemination. Casp1/11 −/−68 mice featured an intermediate phenotype with regards to total mLN pathogen loads, which were increased~3-fold (Fig S3C), while luminal colonization (Fig S2E), migration and replication of the bacterium were not altered (Fig. 2a, b). These data agree with a previously reported partial, but non-absolute, dependence of NAIP/NLRC4 on Caspase-1. 10,11,69,70 As NAIP/NLRC4 potently limits pathogen migration (Fig. 1), we reasoned that the lack of involvement of NLRP3 and Caspase-11 might be attributable to redundancies with this inflammasome. We thus crossed Nlrc4 −/− mice with Nlrp3-or Casp11-knockout animals to obtain Nlrc4 −/− Nlrp3 +/− and Nlrc4 −/− Nlrp3 −/− littermates, as well as Nlrc4 −/− Casp11 +/− and Nlrc4 −/− Casp11 −/− littermates, for infections. However, also in an Nlrc4 −/− background, neither Nlrp3-ablation nor Casp11-ablation affected pathogen migration (µ) to the mLN, pathogen replication (r), total S. Tm loads in this organ by 24 hpi (Fig. 2c, d, Fig S3D-E), or cecal luminal growth (Fig S2F-G). In line with these data, we did not detect significant differences in mucosal inflammation between the mice (Fig S4). Thus, we conclude that in our oral infection model, NLRP3 and Caspase-11 are not involved in the control of S. Tm dissemination from the gut lumen during the first day of infection. As our approach monitors all infection steps from the gut lumen to the mLN, these data should exclude an impact of IEC and phagocyte NLRP3 and Caspase-11, alike. NAIP/NLRC4 in IECs, not in phagocytes, restricts systemic spread of S. Tm from the gut lumen The population dynamics analysis revealed that NAIP/NLRC4 restricts migration (µ) of S. Tm from the intestinal lumen to the mLN, but appears dispensable for containment of pathogen replication (r) within the mLN. The parameter µ in the model summarizes two main steps of the infection process: (i) the invasion/translocation of S. Tm across the cecal mucosa and (ii) the mice (open circles) display higher variety in WITS than those isolated from Nlrc4 +/− (circles) littermates. Analysis of mice depicted in a. Only mice with detectable WITS (plating) in the mLN were included in the analysis, remaining samples were set to 0. c The migration rate µ from the cecal lumen (of mice shown in b with ≥1 WITS-CFU per mLN, translocation events/day) to the mLN of Nlrc4 −/− mice (open circles) is 8.7-fold increased compared to Nlrc4 +/− (circles) littermates. d S. Tm growth rate r within the mLN (of mice shown in b with ≥1 WITS-CFU per mLN, depicted as growth rate per day) is independent of NLRC4. e S. Tm counts in the mLN of Naip1-6 Δ/Δ mice (open circles) are~10-fold increased at 24 hpi compared to Naip1-6 fl/Δ (circles) littermates. f The populations recovered from the mLN of Naip1-6 Δ/Δ mice (open circles) display higher variety in WITS than that of Naip1-6 fl/Δ (circles) littermates. Analysis of mice depicted in e. Only mice with detectable WITS (plating) in the mLN were included in the analysis, remaining samples were set to 0. g The migration rate µ from the cecal lumen (of mice shown in f with ≥1 WITS-CFU per mLN, translocation events/day) to the mLN of Naip1-6 Δ/Δ mice (open circles) is 6.7-fold increased compared to Naip1-6 fl/Δ (circles) littermates. h S. Tm growth rate r within the mLN (of mice shown in f with ≥1 WITS-CFU per mLN, depicted as growth rate per day) is independent of NAIP1-6. Depicted are counts of all S. Tm (dark blue, selected for with Streptomycin) and specifically of the WITS (light blue, selected for with Kanamycin, 5% of inoculum). Each circle represents one mouse. Combined data of three (e-h) or four (a-d) independent experiments. Dotted line: detection limit. Gray line: Median, in c, d, g, and h 95%-Confidence Intervals are indicated. Statistical analysis: Mann-Whitney-U Test, p-values indicated, ns: p ≥ 0.05. subsequent transport of the pathogen from the mucosa into the mLN. To specify which of the two components is impacted by NAIP/NLRC4, and in which cell type the restriction takes place, we infected mice lacking the NAIP receptors specifically in IECs (Naip1-6 Δ/ΔIEC ). Surprisingly, we found that IEC-specific ablation of NAIPs was sufficient to reproduce the pathogen migration phenotype observed in full body Naip1-6 knockouts (Fig. 3a, compare with Fig. 1), while luminal colonization was unaffected ( Fig S2H). The replication parameter (r) remained unaffected in Naip1-6 Δ/ΔIEC animals ( Fig. 3a), further supporting a role for IEC NAIP/NLRC4 specifically in preventing pathogen migration from the gut lumen.
Notably, the restriction of systemic bacterial loads by epithelial NAIP/NLRC4 was equally relevant in an infection with a different S. Tm strain (S. Tm 14028 ; Fig S5A, S2L). This suggests that the epithelial inflammasome is of general relevance for protection against S. Tm strains.
Active epithelial invasion, most prominently in the cecum, 4 is the main pathway by which S. Tm traverses the intestinal epithelium to reach the mLN in the Streptomycin mouse model. By contrast, passive pathogen transport via lymphoid follicles and/ or gut lumen-sampling DCs accounts for only~10% of the total transport. 19,57 Due to the pronounced effect of the IEC NAIP/ NLRC4 inflammasome, it remained unclear if this "alternative" sampling route for traversal might also be restricted by the NAIP/ NLRC4 inflammasome. The S. Tm mutant S. Tm Δ4 (SL1344 sipAsopBsopEsopE2) lacks the TTSS-1-delivered effector proteins necessary for active invasion into IECs, but retains the TTSS-1 structural components sensed by NAIP/NLRC4. 71 While being severely attenuated for IEC invasion, non-invasive mutants like S. Tm Δ4 can still traverse the epithelial barrier by the passive sampling route. 19 Hence, this strain only rarely passes through IECs on the way from the gut lumen to the mLN, but instead promptly enters the lamina propria phagocyte compartment. This feature allowed us to specifically analyze the impact of phagocyte NAIP/NLRC4 on the pathogen migration rate. Towards this aim, we applied the same procedure as described in Fig. 1. We infected Nlrc4 −/− mice for 24h with a mixture of non-tagged S. Tm Δ4 and spiked in the seven barcoded WITS Δ4 strains at a 1:21 dilution (i.e., the WITS Δ4 strains together made up 33.3% of the inoculum). In line with this strain only migrating through the passive sampling pathway, the absolute mLN loads of S. Tm Δ4 were~10-fold lower than in an infection with wildtype S. Tm (compare Fig. 3b with Fig. 1a) and no detectable mucosal pathology was induced at 24 hpi ( Fig S4C). This is in line with earlier work indicating that sipAsopBsopEsopE2-mediated epithelium invasion is the key trigger of acute mucosal inflammation. 72,73 Importantly, when infecting with S. Tm Δ4, , we did not observe any effect of the ablation of NLRC4 on the mLN infection dynamics and cecal lumen colonization (Fig. 3b, S2I). This was also the case for Naip1-6 Δ/ΔIEC mice (Fig. 3c, S2K). It is interesting to note that we observed a slight, but non-significant trend towards higher mLN counts in Naip1-6 Δ/ΔIEC mice. We suspect that this is attributable to some residual epithelial invasion capacity of S. Tm Δ4 19,74 and/or uptake by M-cells.
Finally, to formally exclude the involvement of DC NAIP/NLRC4 in S. Tm restriction, we infected mice lacking the NAIP receptors specifically in CD11c + cells (Naip1-6 Δ/ΔCD11c ) with wild type S. Tm as described above. In line with our previous observations, the ablation of Naip1-6 in CD11c + cells did not affect luminal colonization (Fig S2M), mLN pathogen loads, migration of S. Tm to the mLN and replication of the bacterium at this site ( Fig S5B). Altogether, these data demonstrate that i) NAIP/NLRC4 specifically in IECs acts as a firewall against pathogen dissemination from the gut lumen, whereas ii) this inflammasome has minimal impact in phagocyte populations that take up S. Tm in the mucosa, transport  Fig. 1e-h). Growth rate r within the mLN was independent of NAIP1-6 within IECs. b Streptomycin pretreated mice were orally infected with 5 × 10 7 CFU S. Tm Δ4 . S. Tm Δ4 , which bypasses IECs, is unaffected by NLRC4-mediated restriction during migration to and colonization of the mLN. This extends to epithelial NAIP1-6 (c). Depicted are counts of all S. Tm (dark blue, selected for with Streptomycin) and specifically of the WITS (light blue, selected for with Kanamycin, 5% of the inoculum) or S. Tm Δ4 (dark green, selected for with Streptomycin) and specifically of the WITS Δ4 (light green, selected for with Kanamycin, 33.3% of the inoculum). Each circle represents one mouse. Combined data of two (b, c) or three (a) independent experiments. was spiked with the seven barcoded WITS in a dilution of 1:700 (i.e., the WITS strains together made up 1% of the inoculum). Strikingly, when analyzing spleen S. Tm loads in Nlrc4 −/− mice and littermate controls, we found NLRC4 to be completely dispensable for pathogen control. In line with this, we did not detect any differences in the number of WITS recovered from the spleen of these mice. Furthermore, the migration rate µ and the replication rate r within the spleen were not altered upon ablation of NLRC4 (Fig. 4a). We also observed no effect of NLRP3, or Caspase-11, on the containment of systemic S. Tm infection, even when analyzed in an Nlrc4 −/− background (Fig. 4b, c). Casp1/11 −/− mice again featured an intermediate phenotype, similar to our observations in the oral infection model, i.e.,~3-fold increased bacterial loads in the spleen by 24 hpi (Fig. 4d, compare to Fig S3C). Mathematical modeling suggests that this effect is due to a Caspase-1 mediated restriction of initial S. Tm migration to the spleen, rather than involvement in suppression of pathogen replication within this organ (Fig. 4d, population structure, µ and   37,39,43,47,49,50,76 It should be noted that differences in the studied infection time points or distinct virulence factor expression patterns of the employed S. Tm strains may also account for disparate phenotypes. Nevertheless, in our iv infections, Caspase-1 mediated defense appears independent of NLRC4, NLRP3, or Caspase-11. It remains to be established which other activation pathway might be involved. IECs express more Naip and Nlrc4 transcripts than the remaining mucosal tissue cell types To assess the NAIP/NLRC4 sensing potential of the specific cell types that interact with S. Tm during oral infection in mucosal and systemic tissues, we analyzed Naip/Nlrc4 expression by quantitative PCR (qPCR). For assessment of Naip transcript levels in IECs vs. other cells of the mucosa (including phagocytes), and to set the baseline of the assay, we initially compared tissues from uninfected wildtype mice (Naip1-6 fl/fl ), to those of Naip1-6 Δ/ΔIEC and Naip1-6 Δ/Δ animals. Mice carrying just one intact allele of the Naip1-6 locus (Naip1-6 fl/Δ ) served as an additional control for the sensitivity of the assay. Nlrc4 and Naip1, 2, 5, and 6 expression levels were markedly higher in the cecum tissue, as compared to both mLN and spleen ( Fig. 5a-e, data from Naip1-6 fl/fl mice). Nlrc4 was expressed at~10fold higher levels in the cecal mucosa than in the mLN and around 100-fold higher than in the spleen (Fig. 5a). Similar differences could be observed for the Naip transcripts. Especially Naip1 was strongly expressed in the cecum, but~100-fold reduced in the spleen (Fig. 5b). The bulk of the cecal Naip1, 2, 5, and 6 expression was attributable to IECs, in agreement with earlier work by us and others 11,77 (Fig. 5b-e, compare Naip1-6 fl/fl with Naip1-6 Δ/ΔIEC ). Interestingly, when we analyzed the liver as a non-barrier, nonlymphoid organ, we found that the expression levels of Nlrc4 and Naip1-6 were generally~100-fold lower than in the cecal mucosa ( Fig S6A-B, compare to Fig. 5a-e). Hence, the cecal epithelium appears to be particularly loaded with NAIP/NLRC4 for detection of invading pathogens. This charging of non-immune cells with inflammasome receptors might be particularly pronounced in barrier tissues that are in close contact with microbes and engage actively in defense. 10,11,43,55,78 In addition to the major epithelial Naip transcript pool, we noted a significant contribution from other cell types to the expression of the Naip genes in the cecal mucosa ( Fig. 5b-f, compare Naip1-6 Δ/ΔIEC and Naip1-6 Δ/Δ animals). From the available dataset we could not infer to what percentage various immune cells, fibroblasts, endothelium and/or other cell types contribute to this non-IEC expression. To specifically analyze the expression of Naip/Nlrc4 in intestinal DCs-the cell type that has been reported to take up S. Tm in the cecal mucosa and transport it to the mLN 19,23 -we sorted intestinal DC subsets from wildtype LPSinjected mice and PBS vehicle controls. Whereas Nlrc4 was expressed equally in all intestinal DC subsets, CD103 + DCs expressed Naip1 and Naip2 at increased levels compared to other subsets (Fig S6C). The expression of these receptors was not boosted by exposure to the pro-inflammatory stimulus LPS. In fact, LPS-priming rather led to a downregulation of expression in some of the DC subsets, which might be explained by the immunotolerant phenotype of mucosal myeloid cells. 79 In addition, this downregulation might represent a protective mechanism by which self-destruction of infected DCs is prevented to ensure antigen presentation. Taken together, our data highlight that IECs express particularly high levels of Naip and Nlrc4 transcripts, whereas DCs encountered by S. Tm subsequent to epithelial traversal and/or upon passive sampling appear to express more modest levels. Importantly though, the failure of DCs to restrict S. Tm dissemination through NAIP/NLRC4 in vivo cannot be explained by a complete lack of this inflammasome. However, the reduced expression levels (compared to IECs) might explain why S. Tm Δ4 dissemination is not efficiently restricted by NAIP/NLRC4 (Fig. 3).
S. Tm expresses the NAIP ligand structures TTSS-1 and flagella during IEC invasion, but promptly downregulates them upon transit to the lamina propria and systemic tissues S. Tm migration to systemic sites is potently restricted by IEC NAIP/ NLRC4, but apparently not by phagocyte NAIP/NLRC4, even though both cell types express detectable levels of the required receptors (Figs. 3-5; Fig. S6C). One reason for this discrepancy might reside in the pathogen's gene expression program, which may prevent PAMP expression at certain sites. Such compartmentspecific down regulation has been observed previously in various infection models. 25,[80][81][82] S. Tm requires TTSS-1 and flagella (composed of flagellin subunits) for the initial colonization and establishment of gut infection. 16,72 Especially for invasion of IECs, which strongly express Naip1, 2, 5, 6, and Nlrc4 (Fig. 5a-e), the NAIP ligand structures TTSS-1 (for the delivery of the effector proteins SipA, SopB, SopE, and SopE2) 72,73 and flagella (to subvert gaps in the mucus layer and reach the epithelium by directed motility) 4,16 are both crucial. This could explain why infection events are efficiently sensed and restricted specifically by IEC NAIP/NLRC4. However, it remained unclear to which extent S. Tm regulates TTSS-1 and flagella expression during and after epithelium traversal.
To clarify this aspect, we assessed the expression of NAIPactivating S. Tm PAMPs during the infection process. Initially, we optimized staining procedures in S. Tm infections of cultured HeLa epithelial cells, a process that relies on TTSS-1 (TTSS-1-deficient mutants~500-1000-fold attenuated) 74 and flagella-driven motility. 16 While HeLa cells are clearly a simplistic test system riddled by genetic drift 83 they nevertheless provide an efficient test system for establishing microscopy analysis pipelines. Notably, when infected HeLa cells were immunostained for the flagella subunit FliC, we could detect a significant fraction of FliC + S. Tm also within the epithelial cells (Fig S7A). Notably,~30% of the bacteria still stained positive for FliC at 7 hpi (Fig S7B). This is in line with previous work. 84 These findings were supported by parallel analyses of infected HeLa cells using cryo-focused ion beam (cryo-FIB) milling and cryo-electron tomography (cryoET). Cryotomograms showed fully assembled flagella located between the bacterial surface and the membrane of the Salmonella containing vacuole (SCV) (Fig. 6a).
Previous work has shown that a subfraction of S. Tm may escape the SCV over time and become cytosolic. 81,85 When we analyzed the cytosolic subpopulation with the help of S. Tm localizer , carrying a reporter for cytosolic escape (pCK100, glucose-6-phosphate (Glc6P)-driven expression of mCherry), [86][87][88][89] we found that at 7 hpi, around one quarter of all the cytosolic bacteria was FliC + (Fig S7B). Strikingly, we frequently detected bacteria with protruding flagella (Fig S7A, S7C). Cryotomograms, however, indicated a tight enclosure of both bacterium and flagellum within the SCV (Fig. 6a). Indeed, cytosolic localization correlated with a protruding flagellum conformation (Fig S7C).
These data suggest that in addition to the mandatory expression of TTSS-1, a significant fraction of S. Tm carry flagella within cultured epithelial cells, and maintain or re-express these during egress into the cytosol, which is in line with observations by Knodler et al.. 84 These cytosolic S. Tm carrying the flagellum might be a potent trigger of epithelial inflammasome activation.
Results from immortalized cell lines exhibit limited reproducibility to in vivo settings due to altered cell physiology and lack of tissue environment. 83,90,91 We therefore assessed if our observations also apply in vivo during IEC infection in the mouse model. We infected mice orally with S. Tm carrying a transcriptional reporter (prgH-GFP) for TTSS-1 expression (S. Tm SPI-1-GFP ) and analyzed the pathogen populations in the cecum tissue (12 hpi) and at systemic sites (spleen, 3 dpi to obtain sufficient bacterial loads). Fluorescence microscopy was used to assess TTSS-1 expression (i.e., prgH-GFP) and flagella expression (staining with anti-FliC antibodies). Around 80% of the S. Tm lodged in IECs were SPI-1 + (Fig. 6b-d). This sharply contrasted to SPI-1 expression in the lamina propria (~20% SPI-1 + S. Tm) and the spleen (no detectable SPI-1 expression) (Fig. 6b-d). In line with the tissue culture data above,~50% of the S. Tm lodged in the cecal epithelium also stained positive for FliC ( Fig. 6e-g). This population was increased two-fold in the absence of epithelial NAIP receptors (Fig S8A), which correlated with a significantly increased fraction of cytosolic bacteria, especially with regard to microcolonies consisting ≥4 bacterial cells, in Nlrc4 −/− mice compared to Nlrc4 +/− or wildtype mice (Fig S8B). These observations are consistent with NLRC4/NAIP-dependent expulsion of IECs containing flagellated S. Tm. By contrast, we detected significantly less FliC + S. Tm in the lamina propria (Fig. 6f, g) and no FliC + bacteria in the spleen (Fig. 6e, g). It is currently not entirely clear if fully assembled flagella structures can activate the NAIP/NLRC4 inflammasome. Nevertheless, one can take the flagella as a proxy for the presence of flagellar subunits. Assuming that 90% of the lamina propria-lodged pathogens have reached this site by traversing the IECs (which necessitates the expression of PAMPs recognized by NAIP/NLRC4), our data indicate that the pathogen promptly down regulates expression of both structures during transit, thereby evading recognition by phagocyte NAIP/ NLRC4.
To formally test if PAMP down regulation explains S. Tm evasion of phagocyte NAIP/NLRC4-restriction, we analyzed the effect of forced PAMP expression. This strategy has been used in the past to assess the PAMP-specificity of individual inflammasomes. 92 To this end, we infected Nlrc4 +/− and Nlrc4 −/− littermate mice iv with a 1:1 mix of S. Tm ECV (S. Tm ΔflgB TAG13 (pZ2500); empty control vector) and S. Tm fliCind (S. Tm ΔflgB TAG1 (pEM087); expression of FliC under a Doxycycline-inducible promoter 14 ). We induced FliC expression by iv Doxycycline administration at 17 hpi, a time point when most of the bacteria should be lodged within splenic phagocytes. In NLRC4 proficient hosts, splenic S. Tm fliCind loads were~10-fold lower than those of S. Tm ECV by 24 hpi (Fig S9). Notably, this phenotype was not observed in Nlrc4 −/− mice ( Fig S9). Thus, splenic phagocytes are able to sense S. Tm via the NAIP/NLRC4 inflammasome, but S. Tm as a rule avoids recognition. This further supports that downregulation of PAMP expression upon traversal of the gut epithelium contributes to S. Tm evasion.

DISCUSSION
Earlier work has implicated NAIP/NLRC4 in host defense against S. Tm and related enteropathogens. 8,10,11,30,44,47,55 Mice globally lacking key components of this inflammasome suffer from exacerbated systemic infection and higher pathogen loads in lymph nodes, spleen and liver upon oral inoculation. 8,10,11,37,44,47 As mechanistic work on NAIP/NLRC4 signaling has focused on phagocytic cell types, 26 and as S. Tm frequently lodge within such cells at extraintestinal locations, 58 it has been assumed that     ; Fig. 3b, c). Instead, our data show that inflammasome defense against dissemination of different S. Tm strains from the gut relies specifically on intestinal epithelial NAIP/ NLRC4 (Fig. 3, S5), previously shown to drive the expulsion of infected IECs and restrict mucosal pathogen loads by >>50-fold during oral S. Tm or Citrobacter rodentium infection. 11,55 These results establish IEC NAIP/NLRC4 as the warden of not only the gut mucosa, but also of the systemic compartment, upon infection with a host-adapted pathogen. The expression levels of Naip and Nlrc4 transcripts are notably high in IECs, but also clearly detectable across systemic organs and mucosal tissue DCs (Fig. 5, Fig. S6). On top of this, the downregulation of NAIP-activating PAMPs within host tissues provides an explanation to the failure of phagocytes, but ability of IECs, to utilize NAIP/NLRC4 for combatting S. Tm in vivo (Fig. 6,  Fig S7-S8).
Following flagellaand TTSS-1-driven invasion of IECs, which by necessity exposes these PAMPs to epithelial NAIP receptors, S. Tm promptly downregulates expression of both structures, thereby preventing NAIP/NLRC4 activation beyond the epithelial barrier (Fig. 6, Fig S8-S9). 93 Other evasion mechanisms, e.g., SCV membrane shielding of PAMPs from the cytosolic NAIPs, and active interference by secreted virulence factors, might further dampen residual inflammasome activation in a cell-type specific manner. 1,24 Also, it remains to be fully shown how polymerization into the flagellum affects the NAIP/NLRC4 triggering intensity of flagellin. Anyhow, the evasion of recognition is possibly already employed during IEC-invasion, but may be hampered by high bacterial invasion rates and relatively long lifetimes of the bacterial proteins. Nevertheless, even when administered systemically, S. Tm is not recognized by phagocyte NAIP/NLRC4, except if expression of the flagella component FliC, a NAIP5-6 ligand, is artificially forced (Fig. 4, Fig S9). 92 Therefore, while phagocytes at systemic sites would indeed be capable of pathogen restriction via NAIP/NLRC4 (Fig. 5, Fig S9), our population dynamics data conclusively show that different S. Tm strains evade this defense. This is in stark contrast to less-well adapted bacteria, like Chromobacterium violaceum, which are vigorously restricted (>>1.000-fold in liver) by the NAIP/NLRC4 inflammasome in systemic organs. 94 Based on these observations, it is tempting to speculate that mechanisms for avoiding the NAIP/NLRC4 defense have contributed to the evolution of gene-regulatory circuits in host-adapted bacterial pathogens.
Other inflammasomes have previously been deemed protective during S. Tm infection in mice. 37,39,43 However, under stringent littermate-controlled conditions, we here find that neither NLRP3, nor Caspase-11, impact S. Tm transmucosal migration or early systemic replication in vivo, at least in the Streptomycin pretreated mouse model (Fig. 2). Caspase-11 can protect against systemic infection by bacteria colonizing the host cell's cytosol, e.g., Burkholderia spp. 40 Other intracellular pathogens, including S. Tm and Legionella pneumophila (L. pn), express virulence factors that stabilize the intracellular vacuole. 88,95 Mutant S. Tm and L. pn strains lacking those factors breach the vacuolar membrane with higher frequency and are efficiently detected by Caspase-11. 40,42 Our finding that Caspase-11 is dispensable for restriction of S. Tm dissemination/replication (Fig. 2, Fig S3) agrees with these reports, supporting that wildtype S. Tm evades Caspase-11 recognition after oral infection in vivo. Importantly, deletion of Casp11 in Nlrc4 −/− mice does not increase pathogen migration or systemic colonization (Figs. 2, S3, 4). This refutes the possibility that a potent epithelial NAIP/NLRC4 response masks any effect of Caspase-11. Furthermore, Casp1/11 −/− animals only partially recapitulate the elevated S. Tm dissemination of NAIP1-6 or NLRC4-deficient animals (Figs. 2, S3, 4). This points to the potential involvement of also other Caspases, e.g., Caspase-8, 10 in the execution of the IEC NAIP/NLRC4 response. The molecular wiring of this defense system will be an important topic for future work.
During acute infection, NLRP3 does not contribute to restriction of S. Tm infection neither in the gut mucosa (Fig. 2, Fig S3), nor at systemic sites (Fig. 4), not even in the absence of NLRC4 (Figs. 2,  S3, 4). This contrasts to previous reports in the literature. 37,39 We can only speculate about the origins of this discrepancy. Our results apply to different, widely used S. Tm strains (SL1344 and 14028), rendering the use of different strains an unlikely cause. It might be attributable to the huge impact of the intestinal microbiota on non-typhoidal Salmonella infection models, as well as to small differences in genetic backgrounds that have recently been revealed. [51][52][53] These could have been confounding in experiments using separately bred wildtype mice as controls rather than littermates. Again, the lack of an NLRP3-linked response may also be explained by efficient S. Tm evasion of this inflammasome. 96 Furthermore, NLRP3 and/or Caspase-11 may serve some more specific function(s) during late stages of a persistent infection or in hosts other than mice.
Taken together, our data extend previous work, 10,11,37,39,42,44,45,47,49,50,76 highlight that in vivo studies of host-pathogen interactions are highly sensitive to the experimental conditions, and may be subject to confounding effects that limit reproducibility. The general approach presented here-a combination of littermate controlled infections, host cell typespecific gene knockouts, genetically tagged bacterial consortia, mathematical modeling and high-resolution imaging of the infected cells/organs-allowed us to decipher the impact of specific host responses in vivo. Here, this led us to refute a protective function for phagocyte inflammasome defense and to uncover intestinal epithelial NAIP/NLRC4 as a firewall preventing systemic dissemination of the orally acquired adapted pathogen S. Tm. We expect the same approach to be powerful for (re-) assessing also the impact of other barriers during the step-wise progression of infectious disease, with extension to other pathogens.

METHODS
Salmonella strains and growth conditions Strains used in this study are S. Tm SL1344 (SB300, Streptomycin resistant), or derivatives thereof, except for S. Tm 14028 97 and WITS 14028 . S. Tm Δ4 lacks four TTSS-1 effector proteins (sipA, sopB, sopE, sopE2), but still expresses the TTSS-1 secretion apparatus, and was described previously. 71 The tags for the WITS strains as described in ref., 58 were introduced by P22 phage transduction. We used S. Tm carrying pM975 (S. Tm SPI-2-GFP ) 72 as a reporter for SPI-2 expression (S. Tm, SPI-2 within the SCV), S. Tm expressing GFP under a prgH-promotor (JH3010, S. Tm SPI-1-GFP ) to mark SPI-1 expressing bacteria (S. Tm SPI-1 ), S. Tm carrying pCK100 (Glc6Pdriven mCherry expression, S. Tm Glc6P-mCherry ) as a reporter for cytosolic bacteria (S. Tm Glc6P ) and S. Tm localizer expressing GFP under the ssaG promotor (JH3009) and carrying pCK100 to track SCV/cytosolic localization of S. Tm. pCK100 was constructed as follows: A 137 bp genomic fragment immediately upstream of the uhpT gene in Shigella flexneri 2a 2457T (GenBank: AE014073.1, 3869168...3869032) was cloned upstream of the ribosomal binding site of phage T7 gene 10 and mCherry on a pSC101 backbone. S. Tm carrying this plasmid showed induced red fluorescence when cultured in presence of Glc6P but not in presence of glucose. As Glc6P is specifically present in the host cell cytosol, this is a reporter plasmid for cytosolic localization, as described previously for Shigella flexneri. 98     was generated by electroporation of the plasmid pEM087 14 into S. Tm. ΔflgB S. Tm ΔflgB carrying an empty control plasmid was used as "wildtype" control in the respective experiment (S. Tm, ECV Fig S7). Genetic barcodes were introduced by P22 phage transduction into S. Tm fliCind and S. Tm ECV for relative quantification of the abundance of the two strains during competitive infection. See Table 1 58 Total abundance of the WITS strains was assessed by integrating their relative distribution analyzed by qPCR with total bacterial CFU obtained through plating.
Population dynamics analysis Population dynamics were analyzed with a previously published model, 59 based on total WITS counts in the organ of interest of each mouse. This method was developed for the estimation of the migration rate µ from the gut lumen to the mLN and the replication rate r within the mLN during oral S. Tm infection, and was directly used for the estimates of the oral infection experiments (Figs. 1-3, S5). We applied the same method to estimate the migration rate µ from the blood to and the replication rate r within the spleen for the iv infection experiments (Fig. 4). We note that the interpretation of the migration rate estimate for this application of the model differs from the original application. The migration is assumed to be constant over time in the model, however, previous research has shown that the number of bacteria in the blood exponentially decreases for iv inoculation. 99 Therefore, the migration we estimate to occur with a constant rate over one day, might have migrated only within the first few hours after inoculation. However, because we are not primarily interested in the migration rates themselves but rather the differences in migration rates between wildtype and knockout mice, this interpretational subtlety does not confound our conclusions. Calculations were performed with R Studio, R version 3.6.0.

Gene expression analysis
For gene expression analysis, tissue was snap frozen in RNAlater (Sigma-Aldrich). RNA isolation was performed with the Qiagen  ). mRNA levels were normalized to Actb and calculated with the 2 −ΔCT -method. Primers for the indicated genes were purchased from Qiagen (RT2 qPCR Primer Assay).

Histopathology
For the assessment of histopathology, cecal tissue was snapfrozen in OCT (Tissue-Tek) and stored at −80°C. Five micrometer sections were cut from the tissue and stained with hematoxylin and eosin as described before. 56 Tissue pathology was scored according to the extent of submucosal edema, epithelial integrity, goblet cell loss and infiltration of polymorphonuclear neutrophils (PMNs) into the lamina propria.
For detection of flagellated intracellular S. Tm, 20 µm sections of cecal mucosa and spleen were prepared. After rehydration (PBS, 1 min), sections were permeabilized with 0.5% Triton X-100 for 5 min (room temperature). Next, samples were incubated in blocking buffer (10% normal goat serum, Reactolab) for 30 min. Sections from JH −/− mice were stained for 40 min (room temperature) using a mouse-αFliC antibody (Abcam, 1:300). Sections from Naip1-6 Δ/ΔIEC knockout mice were incubated with a FliC-Cy3-conjugated antibody (1:300; Cy3 ® Fast Conjugation Kit Abcam; ab188287). After three washes with PBS, samples were stained for 40 min (αmouse IgG-Cy3, 1:200; Phalloidin-A647, 1:200; DAPI, 1:1000; Thermo Fischer). Finally, samples were washed three times with PBS and a coverslip was mounted on the microscope slide with 15 μL Mowiol. Image acquisition was performed with a Nikon Eclipse T1 (inverse) microscope equipped with a Yokogawa CSU-W1-T2 spinning-disk confocal unit (Visitron), a sCMOS camera (Orca Flash 4.0 V2) and a ×100 oil objective (PLAN Apochromat, NA 1.49). All data were analyzed in Fiji. 100 Preparation of frozen-hydrated specimens Plunge freezing was performed as previously described. 101 Briefly, grids containing infected HeLa cells, were removed from the wells using tweezers. The forceps were then mounted in the Vitrobot chamber and the grid was blotted from the backside by installing a Teflon sheet on one of the blotting pads. Grids were plungefrozen in liquid ethane-propane (37%/63%) using a Vitrobot (Thermo Fisher) and stored in liquid nitrogen. 102 Cryo-focused ion beam milling Cryo-focused ion beam (cryoFIB) milling was used to prepare samples of plunge-frozen infected HeLa cells that could then be imaged by cryo-electron tomography. 103 Frozen grids with infected HeLa cells were prepared and processed as previously described. 102 Briefly, lamellae were milled in several steps in a Helios NanoLab600i dual beam FIB/SEM instrument (Thermo Fisher). In a first step, two rectangular regions were used to generate a lamella with~2 µm thickness with the ion beam set to 30 kV and~400 pA. The current of the ion beam was then gradually reduced until the lamella reached a nominal thickness of 250 nm (ion beam set to~25 pA). The stage temperature was maintained below −154°C during loading, milling and unloading procedures. CryoFIB-processed grids were unloaded and stored in liquid nitrogen until further use.
Cryo-electron microscopy and cryo-electron tomography CryoFIB processed infected HeLa cells were examined by cryoelectron microscopy (cryoEM) and cryo-electron tomography (cryoET). 101,102 Images were recorded on a Titan Krios TEM (Thermo Fisher) equipped with a Quantum LS imaging filter and K2 Summit (Gatan). The microscope was operated at 300kV and the imaging filter was set to a 20 eV slit width. The pixel size at the specimen level was 5.42 Å. Tilt series covered an angular range from −60°to +60°with 2°increments and −8 µm defocus. The total dose of a tilt series was 120 e − /Å. 2 Tilt series and 2D projection images were acquired automatically using SerialEM. 104 Three-dimensional reconstructions and segmentations were generated using the IMOD program suite. 105