Tissue compartmentalization enables Salmonella persistence during chemotherapy.

Antimicrobial chemotherapy can fail to eradicate the pathogen, even in the absence of antimicrobial resistance. Persisting pathogens can subsequently cause relapsing diseases. In vitro studies suggest various mechanisms of antibiotic persistence, but their in vivo relevance remains unclear because of the difficulty of studying scarce pathogen survivors in complex host tissues. Here, we localized and characterized rare surviving Salmonella in mouse spleen using high-resolution whole-organ tomography. Chemotherapy cleared >99.5% of the Salmonella but was inefficient against a small Salmonella subset in the white pulp. Previous models could not explain these findings: drug exposure was adequate, Salmonella continued to replicate, and host stresses induced only limited Salmonella drug tolerance. Instead, antimicrobial clearance required support of Salmonella-killing neutrophils and monocytes, and the density of such cells was lower in the white pulp than in other spleen compartments containing higher Salmonella loads. Neutrophil densities declined further during treatment in response to receding Salmonella loads, resulting in insufficient support for Salmonella clearance from the white pulp and eradication failure. However, adjunctive therapies sustaining inflammatory support enabled effective clearance. These results identify uneven Salmonella tissue colonization and spatiotemporal inflammation dynamics as main causes of Salmonella persistence and establish a powerful approach to investigate scarce but impactful pathogen subsets in complex host environments.

In difficult-to-treat bacterial infections, including tuberculosis, deep-seated Staphylococcus aureus infections, and invasive salmonellosis, adequate antimicrobial chemotherapy can initially clear most pathogen cells and resolve clinical symptoms. However, even extended therapy often fails to eradicate the pathogen, even in the absence of relevant antimicrobial resistance. The persistence of a small subset of pathogen cells can cause relapsing diseases and accelerates the emergence of antimicrobial resistance (1–4). Eradication of such difficult-to-treat infections represents an urgent medical need.

Development of effective treatments requires a detailed understanding of the underlying mechanisms. Various mechanisms have been proposed to explain pathogen persistence during antimicrobial chemotherapy. First, antimicrobials might not reach all bacterial cells in sufficient amounts because anatomical permeation barriers limit drug access to bacteria in certain tissue areas (5, 6). Second, bacteria might adopt physiological states in host tissues that enable them to tolerate antibiotic exposure. Such tolerant states might be triggered by stresses imposed on the pathogen by the host immune system (7–13). Moreover, limited nutrient supply and stress conditions can slow bacterial proliferation, which increases tolerance to most antibiotics (14–18), in some conditions due to low ATP levels (19). Third, pathogen heterogeneity has been suggested as a major cause of treatment failures (1, 2, 20–22). Some bacteria might stop replication either because of stochastic internal processes or in response to external triggers. Such nonreplicating “persisters” can survive exposure to otherwise lethal antibiotic concentrations. Other forms of heterogeneity might also contribute to treatment failures, including asymmetric cell division (23–25), uneven partitioning of efflux pumps among daughter cells (26), heterogeneous expression of prodrug-activating enzymes (27), transient gene amplification (28), and heterogeneous induction of specific stress responses (29). The relevance of these various mechanisms for pathogen persistence in host tissues remains unclear because supporting data have been obtained almost exclusively using in vitro models (1–3, 20–22, 30). In vivo data are critical because bacterial susceptibility depends on environmental factors, and complex and diverse microenvironments in infected tissues are difficult to mimic in vitro (31).

One suitable small-animal infection model for in vivo studies is systemic salmonellosis in mice as a model for human invasive Salmonella infections. Such infections, including typhoid and paratyphoid fever (enteric fever) as well as nontyphoidal Salmonella (NTS) bacteremia, are a major health problem worldwide (32, 33). Antimicrobial chemotherapy frequently fails to eradicate Salmonella, resulting in relapsing disease, even when the bacterial strain is susceptible to the drug (34–37). Mouse models of invasive salmonellosis recapitulate these eradication challenges (38–41). We showed previously that clinically relevant doses of fluoroquinolone antibiotics clear Salmonella from mouse tissues only slowly because Salmonella replicates slowly in vivo with generation times around 6.5 h (42). Nevertheless, clearance continues with a monophasic exponential decay of colony-forming units (CFUs) for at least 5 d. Here, we continued treatment for longer intervals and observed declining clearance rates and eradication failure at later time points. We aimed at unraveling the mechanisms underlying this antibiotic persistence of Salmonella.

In vivo studies of pathogen cells surviving chemotherapy have been thwarted by difficulties in localizing and characterizing few, sparsely distributed, micrometer-sized pathogen cells in entire centimeter-sized host organs. We addressed these limitations by adopting serial two-photon tomography (STP) (43) for detecting individual Salmonella cells across entire spleens of infected mice. We resolved interfering tissue autofluorescence, developed automated pipelines for identifying Salmonella cells in terabytes (TB) of imaging data, and validated accurate detection of single Salmonella cells at densities as low as one bacterium in 100 mm3 tissue. We used STP to localize Salmonella surviving treatment with two clinically relevant antibiotic classes. We combined STP with laser-capture microdissection, flow cytometry, Salmonella reporter constructs, and adjunctive therapies, to determine the relevance of nonreplicating Salmonella persisters, stress-triggered drug tolerance, uneven drug delivery, and tissue infiltration by neutrophils and inflammatory monocytes. We found that Salmonella colonization of spleen in untreated mice was inhomogeneous. A minor Salmonella subset colonized the white pulp (WP) and triggered only limited local infiltration of inflammatory monocytes and neutrophils. These Salmonella-killing host cells supported Salmonella clearance during chemotherapy, but their density collapsed during chemotherapy in response to receding local Salmonella loads resulting in 70-fold better Salmonella survival in the WP compared to other, initially more colonized and inflamed spleen compartments. By contrast, Salmonella dormancy, stress-induced antimicrobial tolerance, or inefficient antibiotic delivery had all minor relevance. Thus, host tissue architecture and the topology and dynamics of host–Salmonella interactions caused locally divergent antimicrobial activities that ultimately resulted in eradication failure.

Results

Chemotherapy Fails to Eradicate Invasive Salmonellosis.

We infected genetically susceptible BALB/c mice by oral administration of Salmonella enterica serovar Typhimurium. Once clinical symptoms emerged, we started chemotherapy with recommended doses of the fluoroquinolone antibiotic enrofloxacin. This treatment prevented mouse mortality, resolved clinical symptoms, and diminished Salmonella loads in the major target-organ spleen ∼200-fold with slow, monophasic exponential kinetics over 6 d (42) (Fig. 1 ). However, clearance decelerated thereafter, even though Salmonella survivor cells retained in vitro susceptibility (minimal inhibitory concentration, MIC, 0.03 to 0.06 mg per liter). After 10 d of treatment, Salmonella survived in the spleen (∼2,000 CFUs), the liver (∼2,000), the mesenteric lymph nodes (∼1,000), and the Peyer’s patches (∼200 in the last small-intestinal patch) (SI Appendix, Fig. S1). Reservoirs of surviving Salmonella in these and other organs have been previously observed after treatment of intravenously infected mice (38) and in a streptomycin-pretreated enteritis model with high initial Salmonella loads in the cecum (44).

Fig. 1.

Spleen-wide monitoring of Salmonella survival during antimicrobial chemotherapy. (A) Disease scores of orally infected mice. From day 5 to day 14, mice received daily 5 mg/kg enrofloxacin. Arithmetic means and SDs are shown for 3 to 26 mice. (B) Salmonella loads in the spleen of orally infected mice as determined by plating (triangles), flow cytometry (diamonds), or serial two-photon tomography (circles). Plating and flow cytometry results are shown as geometric means and SDs for groups of 3 to 20 mice (yellow labels indicate previously reported values) (42). Each tomography data point represents an individual mouse. The dashed line from days 2 to 5 shows an exponential fit; the line from days 5 to 15 shows an exponential decay with y offset; and the line from days 15 to 18 connects the geometric means. (C) Principle of STP tomography. A vibratome removes the surface of a tissue block. A two-photon microscope scans the blockface, before the vibratome removes the next section. (D) Side and top views of collagen fibers (trabeculae) detected by STP tomography of a 5-mm spleen slice. (E) Detection of GFP-Salmonella by STP tomography in one optical section (Left). Neutrophils were stained by in vivo injection of a phycoerythrin-labeled Ly-6G antibody. Green fluorescent particles were confirmed by staining the corresponding vibratome section with an antibody to Salmonella LPS and imaging by confocal microscopy (Right). (Scale bars, 3 μm.) (F) GFP fluorescence intensity distribution of single Salmonella cells in the spleen at day 5 after oral infection as analyzed by flow cytometry (n = 7,685) or STP tomography (n = 681). Intensities were normalized to the respective median. (G) Examples of single Salmonella cells and microcolonies in the spleen as detected by STP tomography. The numbers on the Right represent estimates for the number of Salmonella cells based on total GFP fluorescence. (Scale bar, 5 μm.) (H) Size distributions of Salmonella objects in orally infected mice before (day 5) and during antimicrobial chemotherapy. The data represent pooled data from three mice per group and 39 to 362 Salmonella objects per mouse. The Inset shows the fraction of single Salmonella among all Salmonella objects with each circle representing an individual mouse.

The Salmonella survivors caused disease relapse upon treatment discontinuation in seven out of eight tested mice (Fig. 1 and ). A second regimen of enrofloxacin again reduced clinical symptoms to baseline () indicating that Salmonella remained susceptible to treatment in vivo. At the end of the second regimen, we still found surviving Salmonella in the spleen, the liver, and the mesenteric lymph nodes but not in the last Peyer’s patch (). Administration of enrofloxacin at doses 60- to 120-fold higher than recommended (to avoid adverse effects), still requires 5 wk for Salmonella eradication (39, 45, 46).

Similarly, after 10 d of treatment with the beta-lactam ceftriaxone, 480 ± 190 Salmonella cells survived in the spleen (). Relapsing disease upon treatment discontinuation occurred in all four tested mice and could be treated with a second regimen of ceftriaxone (). Thus, two major classes of antibiotics that are widely used to treat human invasive salmonellosis, failed to eradicate Salmonella in orally infected mice.

Serial Two-Photon Tomography Detects Sparse Salmonella in the Spleen.

Our goal was to identify survival mechanisms and host microenvironments of the persisting Salmonella. Currently available methods (47–51) are impractical for characterizing scattered micrometer-sized Salmonella cells in centimeter-sized host organs. However, STP tomography provides a robust method for a similar task—the detection of thin neuronal axons with submicron resolution in entire mouse brain (43) (Fig. 1). We were thus interested to adapt STP for infectious diseases. Our initial attempts showed that standard STP protocols were hampered by substantial tissue autofluorescence in the spleen that became particularly strong during infection () and masked the fluorescence signal of GFP-expressing Salmonella. We were unable to boost GFP levels because this would diminish Salmonella in vivo fitness (52), but we discovered that storage of perfusion-fixed spleen in freeze protectant at −20 °C effectively suppressed the interfering tissue autofluorescence (). As another challenge, illumination was uneven and varied between samples (), precluding subtraction of a single reference background image as can be done for brain datasets. Applying “corrected intensity distributions using regularized energy minimization” (CIDRE) (53) resolved this issue ().

Imaging 5-mm slices of spleen at 0.435-μm horizontal and 10-μm vertical resolution (Fig. 1) revealed Salmonella-like particles with spectral characteristics of GFP (Fig. 1). A single three-dimensional (3D) image stack for a 5-mm slice of spleen contained ∼0.5 TB of data, prohibiting manual analysis. Simple threshold-based Salmonella detection was hampered by residual autofluorescence and imaging artifacts. To resolve this issue, we utilized machine learning, including support vector machines and a deep convolutional neural network that discriminated GFP-Salmonella from background with >99% sensitivity and specificity. We validated the identification of GFP-Salmonella by antibody staining of Salmonella lipopolysaccharide in spleen sections retrieved from the tomograph (Fig. 1) and detected no signals classified as GFP-Salmonella in the spleen of mice infected with yellow fluorescent protein (YFP)-Salmonella indicating specific detection. Collagen second-harmonic signals (54) and in vivo staining of host cell surface antigens with fluorescence-labeled antibodies revealed the tissue context of Salmonella cells (Fig. 1 ). We focused on infected spleen in this study but STP also identified GFP-Salmonella in liver, mesenteric lymph nodes, and small intestinal Peyer’s patches ( and Movies S1–S3).

Single Salmonella cells had a narrow GFP fluorescence-intensity distribution consistent with flow cytometry data for the same GFP-Salmonella strain in the spleen (55) (Fig. 1), indicating reliable fluorescence quantification by STP. For unresolved Salmonella microcolonies, we divided total fluorescence by the median single-cell GFP value to estimate the number of Salmonella cells per microcolony (Fig. 1). The results showed that ∼40% of Salmonella resided alone with no conspecifics in their immediate (<5 μm) neighborhood consistent with previous findings (55, 56) (Fig. 1). Based on number and size of microcolonies and the imaged tissue volume, we estimated the total number of Salmonella cells per spleen. These data were consistent with plating results and flow cytometry of independently infected mice (Fig. 1). Thus, STP tomography reliably localized and quantified individual Salmonella cells at population densities down to one cell in 100 mm3 tissue.

Comparison of Salmonella objects before and after 10 d of chemotherapy with either enrofloxacin or ceftriaxone revealed unaltered microcolony-size distributions. We did not detect a preferential survival of isolated single Salmonella cells during antimicrobial clearance of 99% of all Salmonella, which was inconsistent with the proposed exclusive survival of single nonreplicating Salmonella persister cells (57).

Antimicrobial Clearance of Salmonella Is Ineffective in the Splenic WP.

STP tomography enabled us to localize scarce surviving Salmonella during antimicrobial chemotherapy at the whole-organ level. The spleen has three major compartments with distinct physiological functions (58): the red pulp (RP) that serves as a blood reservoir and removes aged erythrocytes; the WP containing high densities of B and T lymphocytes for inducing adaptive immunity to bloodborne pathogens; and the intermittent marginal zone (MZ) that captures bloodborne particles and pathogens. We identified these three regions by in vivo staining of CD169 (also called Siglec-1 or Sialoadhesin), a marker for a macrophage subtype in the MZ (59) (Fig. 2 ). We noticed that Salmonella clearance kinetics in RP resembled those in the MZ. We, therefore, combined data for these two compartments (RP/MZ) and compared them with those for WP.

Fig. 2.

Compartmentalized Salmonella clearance during antimicrobial chemotherapy. (A) Tomogram of a 5-mm spleen slice stained in vivo with an anti-CD169 antibody. (B) Localization of Salmonella cells (cyan, arrowheads) in spleen compartments (RP; MZ, marginal zone containing CD169-positive macrophages; WP) of an orally infected mouse. (C) Salmonella loads in spleen compartments of orally infected mice during enrofloxacin chemotherapy. Each circle represents STP data of an individual mouse. The lines are fits of monoexponential decays with y offsets. (D) Ratio of Salmonella loads in WP vs. RP/MZ during chemotherapy with enrofloxacin (Enro.; P value for deviation of slope from zero after linear regression of log-transformed data, shown as dashed line). Data for 10 d of treatment with ceftriaxone (Ceftr.) are also shown. Each symbol represents STP data of an individual orally infected mouse calculated from the data shown in C. (E) Localization of GFP-expressing Salmonella in CD68-positive macrophages at the periphery of B220-positive B cell zones in the splenic WP of orally infected mice after 10 d of treatment with enrofloxacin. (Scale bars, 5 μm.)

Before treatment, ∼95% of the ∼200,000 Salmonella cells resided in RP/MZ (60, 61) (Fig. 2). Enrofloxacin cleared Salmonella from these compartments with monophasic exponential kinetics with a rate of 0.44 ± 0.23 log per day for ∼7 d, after which clearance decelerated markedly. After 10 d, ∼300 Salmonella cells or ∼0.15% of the initial load survived. By contrast, at start of therapy only ∼10,000 (∼5%) Salmonella cells resided in WP. Initial clearance of this minor Salmonella subset was 2-fold slower (0.21 ± 0.12 log per day) than in RP/MZ and decelerated markedly already after 4 d of treatment (day 9 postinfection, p.i.) with Salmonella loads declining less than 3-fold over the next 6 d. After a total of 10 d of treatment, ∼1,000 Salmonella cells or ∼10% of the initial load in the WP survived, a 70-fold larger fraction than in RP/MZ (Fig. 2 ).

Ceftriaxone, which differs from enrofloxacin by targeting peptidoglycan transpeptidases instead of DNA topoisomerases, and by lacking activity against nonreplicating bacteria (62–65), had similar inhomogeneous activity against Salmonella in the WP compared to RP/MZ (Fig. 2), indicating that the diverging Salmonella survival did not depend on a particular antimicrobial mode of action or activity against nonreplicating bacteria.

Immunohistochemistry showed that surviving Salmonella in WP resided intracellularly in CD68+ Ly-6C− F4/80− CD21/35− FDC− ER-TR7− gp38− CD3− CD19− WP macrophages (59) at the boundary between T and B cell zones (Fig. 2). We did not find Salmonella cells in B or T cells, follicular dendritic cells, dendritic cells, inflammatory monocytes, lymphatic epithelial cells, or reticular fibroblasts. Our data differ from previous studies that attempted to identify host cells of surviving Salmonella cells by analyzing tissue homogenates with flow cytometry. Homogenization disrupts most Salmonella-infected cells, even when done gently (55). Released bacteria can bind to other host cells in the tissue homogenate such as B cells (66), which do not harbor Salmonella in intact tissues (61).

Together, these data indicate that antimicrobial chemotherapy was particularly ineffective against a minor Salmonella population in WP.

Enrofloxacin Reaches the WP in Sufficient Amounts.

The poor antimicrobial clearance of Salmonella from WP might reflect inefficient drug delivery (67). Enrofloxacin effectively penetrates tissues (64); however, the splenic WP receives blood solutes only by diffusion through a conduit system (68), in contrast to the direct blood supply to the RP/MZ (69). In addition, oral Salmonella infection caused extensive thrombosis in spleen veins (70) (Fig. 3) that reduced blood circulation to and within the spleen as revealed by ultrasound imaging (Fig. 3 ).

Fig. 3.

Role of drug delivery and nonreplicating persisters. (A) Autofluorescence micrograph of a spleen section with a highly autofluorescent thrombus in a large vein. Similar images were obtained for all orally infected mice. (B) Ultrasound 3D power Doppler-mode images of blood flow in the spleen of naive and orally infected mice. (Lower) Corresponding blood flow velocities in the splenic artery as determined by pulsed-wave Doppler ultrasound. (C) Mean velocities in splenic arteries and veins of naive and orally infected mice. Each circle represents one mouse (repeated-measures two-way ANOVA comparing naive and infected mice). (D) Enrofloxacin concentration in different spleen compartments of orally infected mice. Each circle represents one mouse (repeated-measures two-way ANOVA comparing RP and WP). The MIC of enrofloxacin for the Salmonella strain as measured in Mueller–Hinton broth is shown for comparison. (E) Salmonella replication-rate distributions in different spleen compartments. The graphs represent pooled data for three orally infected mice per time point. The few nondividing Salmonella cells included enrofloxacin-killed Salmonella corpses that retain the TIMERbac protein. Previously reported in vivo data for the ssrB mutant, which shows a replication arrest in mouse spleen (42) are shown for comparison (gray). (F) Mean Salmonella division rates with each circle representing an individual orally infected mouse (P value: repeated-measures two-way ANOVA comparing red vs. white pulp).

To assess drug delivery, we isolated WP and RP/MZ using laser-capture microdissection () and determined drug concentration with an enrofloxacin-specific ELISA. Enrofloxacin levels were slightly higher in RP/MZ than in WP but were >1 mg mL−1 at day 9 p.i. in both compartments (Fig. 3), thus exceeding more than threefold the pharmacokinetic–pharmacodynamic target for effective bacterial killing (peak concentration: MIC ratio >12) (71–74). Killing assays with Salmonella cultures growing in chemostat cultures under in vivo mimicking conditions with generation times around 6 h (42) confirmed that enrofloxacin concentrations above 1 mg mL−1 caused saturated killing () (73). Fluoroquinolones accumulate in phagocytes (75), suggesting even higher enrofloxacin exposure of Salmonella residing intracellularly in macrophages (Fig. 2) compared to bulk average concentrations in the WP. These data suggested that sufficient drug was locally available for effective killing.

To track Salmonella killing by enrofloxacin in vivo, we treated mice at day 5 postinfection with enrofloxacin. Before administration and various times after, we killed mice and prepared spleen homogenates in detergent-containing buffer, which liberates >90% of all Salmonella from host cells and tissue fragments as a suspension of single bacteria (76). We sorted the Salmonella using flow cytometry and determined their survival by comparing CFUs on plates (live Salmonella) and flow cytometric counts (live and drug-killed Salmonella) (42). To ensure sufficient Salmonella numbers for rapid sorting, we used intravenously infected mice at day 5 postinfection (first enrofloxacin dose). Salmonella have indistinguishable replication rates and enrofloxacin susceptibility after oral or i.v. infection (42). The results show that enrofloxacin killed 90 to 95% of all Salmonella within the first hour (42) consistent with pharmacokinetic–pharmacodynamic target attainment (see above). However, little further killing occurred in the following 3 h (), consistent with results for chemostat cultures growing under in vivo mimicking conditions () and reported in vitro data for enrofloxacin and other fluoroquinolones (74, 77). This biphasic killing is probably a consequence of a bacterial SOS response to fluoroquinolone exposure (“drug-induced tolerance”) (77).

Overall, killing of slowly replicating Salmonella in mice and chemostats was rather limited compared to rapidly replicating bacteria in rich broth, where similar fluoroquinolone concentrations kill several logs of bacteria within 1 h of exposure (42, 74, 77). Killing activity of antibiotics against invasive Salmonella in human patients is also moderate (78, 79), and Salmonella with slightly reduced susceptibility to fluoroquinolones are associated with longer fever clearance times and increased risk of failure (37, 80), while other Enterobacterales with similarly decreased susceptibility can still be treated effectively (Clinical & Laboratory Standards Institute, CLSI, susceptibility breakpoints for ciprofloxacin: Salmonella, ≤0.06 μg/mL; other Enterobacterales, ≤0.25 μg/mL; http://em100.edaptivedocs.net/Login.aspx).

Enrofloxacin levels declined 5- to 10-fold in both RP/MZ and WP from 1 h to 4 h after drug administration (Fig. 3), consistent with a terminal elimination half-life of enrofloxacin in mice in the range of 1 h (74, 81, 82). Progressively declining enrofloxacin levels will eventually stop enrofloxacin-mediated killing and even enable some Salmonella regrowth when enrofloxacin levels drop below the MIC (see below). Sub-MIC trough levels are known to occur for recommended doses in mice (74) (and occur also in human patients) (83) but are likely less problematic for treatment of more rapidly growing pathogens that are more efficiently killed by peak enrofloxacin levels early after administration of each dose. Indeed, the Cmax/MIC ratio is a marker of effective treatment of various other pathogens with fluoroquinolones, while the fraction of time above the MIC is not (84). Higher or more frequent doses than recommended might reduce the fraction of time with sub-MIC enrofloxacin levels but would also increase adverse reactions (85).

Together, these results show pharmacokinetic–pharmocodynamic target attainment in both spleen compartments. Thus, insufficient drug delivery did not explain the sluggish Salmonella clearance from day 9 p.i. in WP. However, killing was mostly confined to the first hour after drug administration and overall rather limited because of Salmonella’s low replication rates.

Surviving Salmonella Cells Continue to Replicate.

It has been proposed that only nonreplicating Salmonella survive extensive antibiotic treatment in vivo (57). That study used continuous administration of enrofloxacin doses that were 30-fold higher than recommended for rodents. Salmonella viability after treatment was assessed indirectly based on fluorescence changes upon tissue lysis with detergent and incubation in rich medium. The observed ∼65% fluorescence decline could reflect dilution of preloaded GFP by one to two Salmonella divisions during the 12-h incubation, which was taken as evidence for viability (57), but the data do not exclude partial GFP leakage from damaged Salmonella corpses as an alternative interpretation. Two other studies did not detect any viable colony-forming Salmonella under identical experimental conditions (42, 45), suggesting that the nonreplicating Salmonella particles are largely nonviable posttreatment. Other studies suggested survival of slowly replicating Salmonella in mice treated with more clinically relevant doses of fluoroquinolones (44, 86).

To determine the replication status of surviving Salmonella in our model, we used STP tomography to quantify fluorescence of Salmonella expressing TIMERbac, a reporter protein with differential maturation kinetics of green and orange fluorophores. The green-to-orange fluorescence ratio depends on the overall protein dilution associated with bacterial growth and division and thus indicates the single-cell replication rate (42). TIMERbac data for Salmonella in infected mice are consistent with independent measurements of in vivo replication rates based on dilution of a nonreplicating plasmid. TIMERbac data are also consistent with net growth of Salmonella wild-type (WT) and attenuated mutants in mouse lines with differential resistance to Salmonella infection (42, 87). STP tomography of TIMERbac-Salmonella revealed heterogeneous replication rates (42) that were higher in RP/MZ than in WP. Median rates at 24 h after the last enrofloxacin dose declined during treatment in both compartments because of preferential killing of fast dividing subsets (42) (Fig. 3 ). However, the majority of Salmonella cells in both compartments continued to divide, consistent with sub-MIC trough levels (see above), and TIMERbac patterns remained distinct from those of a ssrB mutant, which shows replication arrest after a few initial divisions in vivo (42, 88). This was even true at day 9 p.i., when clearance essentially stopped in WP (Fig. 2). The few “nondividing” Salmonella cells included enrofloxacin-killed Salmonella corpses with lethal DNA damage but intact envelopes retaining the TIMERbac protein (42). Ongoing Salmonella replication was also supported by direct observation of dividing cells after 10 d of treatment in tissue sections (day 15 p.i.; Fig. 2 and ). Moreover, we did not detect any preferential survival of isolated single Salmonella cells during clearing of 99.5% of all bacteria (Fig. 1). Together, these data indicated that replication arrest was not a critical factor of Salmonella survival during treatment.

Host-Imposed Stresses Have Minor Impact on Salmonella Survival.

Host-imposed stresses, such as reactive oxygen species (ROS) and nitric oxide (NO), can promote bacterial tolerance to antibiotics (7–13). Salmonella are exposed to diverse stresses in infected spleen that are highly heterogeneous at the single-cell level (89). This heterogeneity provides unique opportunities to compare directly Salmonella subsets with high or low stress in the same host tissue in terms of their antimicrobial susceptibility. To test this idea, we infected mice with various Salmonella stress-reporter strains. At day 5 postinfection, we administered enrofloxacin and 1 h later, we prepared detergent-treated spleen homogenates to release >90% of all Salmonella from host cells and tissue fragments as a suspension of single bacteria (76). We sorted stressed and nonstressed Salmonella subsets using flow cytometry and determined their survival by comparing CFUs on plates (live Salmonella) and flow cytometric counts (live and drug-killed Salmonella) (42) (Fig. 4). To ensure sufficient Salmonella numbers for rapid sorting, we used intravenously infected mice at day 5 (first enrofloxacin dose). Salmonella have indistinguishable replication rates and enrofloxacin susceptibility after oral or i.v. infection (42).

Fig. 4.

Role of stress-induced tolerance. (A) Strategy for determining survival of stressed Salmonella subsets during chemotherapy of intravenously infected mice. Salmonella carrying a GFP-reporter plasmid for oxidative stress show heterogeneous fluorescence in the spleen at 1 h (Left) (61). This signal is not affected by 1 h of in vivo enrofloxacin exposure or various mutations (). GFPlow and GFPhigh subsets are sorted and survival rates are determined by comparing flow cytometry counts (live and dead Salmonella) with plate counts (live Salmonella). (B), Relative survival of Salmonella subsets for different reporter constructs and Salmonella genotypes. Each symbol represents an individual i.v. infected mouse (*P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; one-sample t test of log-transformed data with Holm–Šídák correction for multiple comparisons).

The results showed that a ∼10% of the Salmonella subpopulation with high PkatG-gfp activity indicating exposure to host-derived ROS (61), survived enrofloxacin approximately twofold better than GFPlow Salmonella with low oxidative stress (Fig. 4 and ). Salmonella required acrB, but not katG, for this protective effect, suggesting that host ROS antagonized enrofloxacin activity by stimulating AcrB-mediated drug efflux (90) and not by altering redox physiology (61, 91).

Salmonella cells with high PhmpA activity indicating exposure to host-derived NO (61), had ∼1.5-fold lower survival than nonexposed Salmonella (Fig. 4 and ). A Salmonella mutant (“Δ4,” ΔhmpA Δhcp-hcr ΔytfE) with detoxification defects that increase the NO concentration in exposed bacteria (61), had similar survival patterns. Thus, physiological NO may not increase killing itself, but rather represent a marker for microenvironments that enhance enrofloxacin activity.

Salmonella cells with high PphoP-1 (92) activities had ∼2.5-fold lower survival than cells with low PphoP-1 activity (Fig. 4 and ). PphoP-1 correlated with Salmonella replication rate (), which is associated with increased enrofloxacin susceptibility in vivo (42) (Fig. 4). However, monitoring replication rate and PphoP-1 activity together revealed an additional ∼2-fold replication-independent effect of high PphoP-1 activity (Fig. 4 and ). Thus, microenvironments that stimulate PphoP-1 enhance enrofloxacin activity through both replication-promoting and replication-independent factors. Neither low magnesium nor low pH, stimuli known to induce PphoP-1 (93), are likely to be associated with enhanced enrofloxacin activity, because the magnesium-nonresponsive Salmonella phoQ H157R mutant (94) showed similar killing patterns, and low pH diminished enrofloxacin activity (MIC values: pH 7.4, 0.03 mg per liter; pH 5.5, 0.06 mg per liter; pH 5.0, 0.24 mg per liter). Antimicrobial peptides, which also induce PphoP-1 (93), can promote antibiotic action (95) but testing their role in mice is difficult because the only described antimicrobial peptide-nonresponsive Salmonella mutant has additional signaling defects (96).

Together, these data showed complex modulation of Salmonella susceptibility by heterogeneous host-imposed stresses. However, effect sizes were moderate and some of the stresses were actually associated with enhanced antibiotic activity, arguing against host stresses as a critical factor supporting Salmonella survival during antimicrobial chemotherapy.

Declining Inflammation Diminishes Salmonella Clearance.

Salmonella infection triggers an inflammatory host response with infiltration of neutrophils and inflammatory monocytes. These cells kill Salmonella by oxidative damage (61, 97, 98) and envelope disruption (99), which partially controls disease progression (100, 101). Host inflammation supports antimicrobial clearance of bacteria (44, 66, 102–107) and may thus be a relevant factor influencing Salmonella survival.

Inflammation was not constant during treatment. Instead, various parameters all showed declining inflammation: clinical symptoms resolved during chemotherapy (Fig. 1) together with declining serum concentrations of the inflammatory cytokine IFNγ (). In the spleen, infiltration by neutrophils and inflammatory monocytes (Fig. 5 ) and the abundance of inflammation-associated transcripts Ifng, Tnfa, and Mpo all declined (Fig. 5). A similar decline in inflammation has previously been observed during high-dose ciprofloxacin treatment of a Salmonella-enteritis mouse model (44).

Fig. 5.

Inhomogeneous and declining inflammation impairs Salmonella clearance. (A) Immunohistochemistry of spleen sections of mice orally infected with WT Salmonella or mixtures of wild-type and enrofloxacin-resistant Salmonella (ER/WT). Enrofloxacin treatment commenced at day 5. Ly-6G stains neutrophils and Ly-6C stains inflammatory monocytes. Fluorescence micrographs were inverted for improved visibility. (Scale bars, 1 mm.) (B) Neutrophil density at different time points after oral infection. Each symbol represents an individual orally infected mouse. The bars represent means (**P < 0.01; two-way ANOVA for single-strain vs. mixed infections at day 9). (C) Transcript abundance in different spleen compartments (red, RP/MZ; blue, WP) of mice infected with WT Salmonella or mixtures of WT and enrofloxacin-resistant Salmonella (ER/IR). Each symbol represents an orally infected mouse (two-way ANOVA for comparison of log-transformed data for single vs. mixed infections at day 9). (D) Salmonella loads in the spleen at various time points after oral infection with mixtures of WT and ER Salmonella. The dashed red line represents the fit for mice infected only with WT Salmonella shown in Fig. 1. Enrofloxacin treatment commenced at day 5. Each symbol represents an individual mouse (circles, data from STP tomography; triangles, plating results; *P < 0.05; ****P < 0.0001; one-way ANOVA of log-transformed data with Holm–Šídák correction for multiple comparisons). (E) Clinical scores of mice infected orally with only WT Salmonella (same data as shown in Fig. 1), or mixtures of wild-type and enrofloxacin-resistant Salmonella (ER/WT), during enrofloxacin chemotherapy (mixed-effect analysis with Geisser–Greenhouse correction). (F) Serial two-photon micrograph of a spleen from a mouse at day 5 after oral infection with a mixture of wild-type (WT, GFP) and enrofloxacin-resistant (ER, YFP) Salmonella. (G) Distribution of distances to nearest neighbors of the same or another Salmonella strain at day 5 after oral infection with WT Salmonella or mixtures of WT and enrofloxacin-resistant Salmonella (ER/IR). Data were pooled form three mice per group. (H) Median nearest neighbor distances for comparisons shown in G. Each symbol represents an orally infected individual mouse (**P < 0.01 for comparison of WT-ER vs. all other groups; one-way ANOVA with Holm–Šídák correction for multiple comparisons). (I) Impact of adjunctive therapy with a neutralizing antibody to TGFβ or an isotype antibody (Ctrl.) on Salmonella tissue loads after 6 d of treatment with enrofloxacin and adjunctive anti-TGFβ therapy for the last 5 d. Each symbol represents CFU data from an orally infected mouse (t test of log-transformed data). (J) Clinical scores of mice shown in I. The magenta line indicates the anti-TGFβ treatment (enrofloxacin started at day 5).

In addition, inflammation was also inhomogeneous across spleen compartments with markedly higher neutrophil densities and transcript abundance in RP/MZ than in WP (Fig. 5 ). In particular, abundance of Mpo transcripts, a marker for neutrophils and inflammatory monocytes, dropped >1,000-fold in WP from days 5 to 9 p.i., consistent with the local depletion of these cell types as observed by immunohistochemistry. These data showed inhomogeneous and declining inflammation during antimicrobial chemotherapy, probably in response to local Salmonella loads (100, 108–110). In particular, inflammation was already weak at the start of treatment in the WP, which was scarcely colonized by Salmonella, and declined in both compartments with falling Salmonella loads.

Comparison of the spatiotemporal inflammation dynamics and Salmonella clearance indicated that at times and places with low neutrophil and monocyte densities, clearance was particularly inefficient. An important role of neutrophils and monocytes was confirmed by antibody-mediated depletion of these cells, which further impaired Salmonella clearance (66) (). These data suggested that the fading support from neutrophils and other inflammatory processes during treatment might have caused diminishing clearance rates and eradication failure.

To test this idea, we aimed at delaying the decline of inflammation. We coinfected mice orally with a GFP-Salmonella WT strain (MIC, 0.03 μg/mL) and a YFP-Salmonella enrofloxacin-resistant (ER) strain with decreased susceptibility (MIC, 0.25 μg/mL, which is still below the CLSI enrofloxacin susceptibility breakpoint of ≤0.5 μg/mL, http://clsivet.org/Login.aspx). The ER strain should be cleared more slowly during chemotherapy because of its decreased susceptibility, thus maintaining stimulation for inflammatory responses, while the WT strain would enable monitoring the clearance of susceptible Salmonella under these conditions. At the start of treatment (day 5 p.i.), mice infected with the strain mixture had similar Salmonella loads and clinical scores compared to mice infected with WT alone (Fig. 5 and E). Enrofloxacin treatment cleared YFP-Salmonella ER less efficiently than WT in single-strain infections as predicted, and this was associated with a slower decline of disease symptoms (Fig. 5 ), serum IFNγ (), neutrophil and monocyte infiltration, and inflammation-associated transcripts (Fig. 5 ).

In the presence of this sustained inflammation, antimicrobial chemotherapy successfully eradicated the coinfecting susceptible WT strain, in marked contrast to the GFP-WT single-strain infections where Salmonella eradication was impossible (Fig. 5). This supported the idea that the declining contribution of neutrophils and monocytes to Salmonella clearance caused treatment failures. Salmonella ER maintained high loads under the same conditions, suggesting that their approximately eightfold lower sensitivity to enrofloxacin resulted in poor clearance even with sustained inflammation support. Treatment failures occur also in human patients infected with S. enterica serovar Typhi strains with moderately decreased susceptibility to fluoroquinolones (37, 111).

Direct competition between ER and WT Salmonella strains did not explain the accelerated clearance of WT in mixed infections, because the two strains did not coinfect the same host cells but remained separated by several noninfected cells (Fig. 5 ) consistent with clonal spreading of infection foci without mixing (56). Limited inter-Salmonella competition in the spleen was supported by unchanged Salmonella single-cell replication rates across a hundredfold changes in Salmonella tissue loads during acute infection (42) and the comparable Salmonella loads in mixed vs. single-strain infections at day 5 (Fig. 5).

Our data are in general agreement with previous observations that stimulating inflammation can reduce antibiotic persistence of Salmonella in a mouse enteritis model (44). However, in our model injection of killed Salmonella or purified lipopolysaccharide (LPS) did not accelerate Salmonella clearance of the susceptible strain. Injected LPS is captured in the MZ (and other organs) with limited spread to splenic RP and WP (112, 113), which might limit its impact on locally persisting Salmonella.

In a complementary experiment, we determined Salmonella clearance at low Salmonella, neutrophil, and monocyte densities. For this purpose, we already started chemotherapy on day 1 p.i., when Salmonella loads were still low, neutrophils and inflammatory monocytes still scarce (), and clinical symptoms absent. We used intravenous infection for these experiments to reduce experimental variation. Early-onset enrofloxacin treatment prevented disease progression but failed entirely to reduce Salmonella loads (). Salmonella surviving 3 d of treatment had indistinguishable replication rates compared to start of treatment (), again arguing against a relevant role of nonreplicating cells.

In in vitro cell-culture infections, Salmonella survival of antimicrobial chemotherapy requires the type III secretion system 2 (T3SS-2) and the T3SS-2 effector protein SteE/SarA to induce an antiinflammatory, M2-like polarization of the infected macrophage (114). By contrast, early-onset treatment in infected mice was similarly ineffective for WT Salmonella, Salmonella ΔssrB defective for T3SS-2, and Salmonella ΔsteE (), indicating that neither T3SS-2 activity nor SteE/SarA is required for antibiotic persistence in vivo. Salmonella cell-culture infections differ fundamentally from the in vivo situation in various aspects (115), including the dominance (in vitro) vs. almost complete absence (in vivo) of nonreplicating persisters (42). The mutant phenotypes also suggest mechanistic differences between Salmonella in vivo survival during chemotherapy and Salmonella persistence in chronic infections (which requires SteE/SarA) (116), consistent with other differences in host cell types (Fig. 2) (116, 117) and Salmonella-control mechanisms (antibiotic vs. CD4 T lymphocytes and TNFα) (102, 116, 118).

Together, these data show that Salmonella are efficiently cleared with sustained contributions from neutrophils and other inflammatory mediators, but not without such contributions. Thus, the always weak inflammation in WP and the further decline of inflammation during antimicrobial chemotherapy was the main cause of diminishing clearance rates and Salmonella persistence. Uneven drug delivery, nonreplicating persisters or slowly growing Salmonella subsets, stress-induced tolerance, and immunomodulatory capabilities of Salmonella must all have limited impact, because none of these factors could prevent Salmonella clearance when sustained inflammation supported chemotherapy. This was consistent with our direct analysis of these factors (see above).

Adjunctive Immunotherapy Safely Facilitates Salmonella Clearance.

Maintaining inflammation during chemotherapy using continuous bacterial stimulation was effective for eradication of susceptible Salmonella. However, stimulation at this level was also associated with prolonged clinical symptoms, which would be unacceptable for improving patient treatment. A previous study found reduced antibiotic persistence of Salmonella in mice treated with lipopolysaccharide or CpG nucleotides (44) but these reagents have also unacceptable safety liabilities. More targeted interventions might disconnect the supportive role of some inflammatory responses from exacerbating disease symptoms. One possibility could be targeting the major cytokine TNFα, which plays a key role in Salmonella control during acute infection. However, TNFα has apparently no impact on antimicrobial clearance of Salmonella (102).

To identify more effective and safe adjunctive therapies, we considered that infiltrating neutrophils and inflammatory monocytes, which capture and destroy Salmonella, provided a major contribution to antimicrobial clearance of Salmonella (Fig. 5 ). We aimed at stimulating neutrophil functions using strategies with favorable safety profiles that are currently in clinical development as adjunctive cancer therapies. Specifically, we inhibited the suppressive effect of TGFβ on neutrophil function (119) using a neutralizing antibody. TGFβ inhibition is currently under clinical development (120). We also tested IL-15 superagonist, an IL-15 complex with improved pharmacokinetics, to stimulate the natural killer cell/neutrophil axis that leads to superior antimicrobial properties of neutrophils (121). Various formulations of IL-15 are under active clinical development (122). We observed that both anti-TGFβ and IL-15 superagonist accelerated antimicrobial clearance of Salmonella by 85 to 95% (Fig. 5 and ), without exacerbating or prolonging clinical symptoms (Fig. 5 and ). Thus, safe adjunctive immunotherapies facilitate chemotherapy of difficult-to-treat Salmonella infections.

Discussion

In this study, we developed STP tomography to localize and characterize scarce Salmonella cells in whole host organs. Using this approach, we discovered that a minor Salmonella subset in the splenic WP survives antimicrobial chemotherapy orders of magnitude better than the Salmonella bulk population. Common assumptions could not explain this observation: the antibiotic reached the WP in sufficient amounts, Salmonella continued to replicate, and host stress-induced drug tolerance of Salmonella and Salmonella’s immunomodulatory capabilities had only minor impact. These findings demonstrate that concepts derived from laboratory models of bacterial persistence may have limited relevance for in vivo conditions in complex host tissues.

Instead, antimicrobial clearance of Salmonella required support of inflammatory cells such as neutrophils and monocytes, in part because antibiotics have poor activity in vivo against slowly replicating Salmonella in mice (42). Infiltration of neutrophils and monocytes followed local Salmonella density, which was high in the RP and MZ but low in the WP (this Salmonella tropism might reflect the low abundance of target phagocytes) (61). While large numbers of neutrophils and monocytes supported continuous antimicrobial clearance from RP and MZ until late in the treatment, the low and declining number of such cells in the WP supported only slow clearance over a few days. Eventually, local Salmonella numbers went below the threshold of pathogen-associated molecular patterns to recruit inflammatory cells and, therefore, the bacteria largely evaded killing but survived and replicated slowly. The critical role of inflammatory cell infiltration was confirmed by adjunctive therapies sustaining inflammation, which enabled successful eradication. Thus, complex tissue architecture, Salmonella tropism, and associated spatiotemporal host–pathogen dynamics determined success and failure of antimicrobial chemotherapy in vivo. These mechanisms required high-resolution, large-scale tissue imaging for discovery and would be difficult to reproduce under laboratory conditions.

The crucial role of host immunity for the outcome of antimicrobial chemotherapy is also revealed by previous studies with mouse strains that are genetically resistant to Salmonella infection. Although 10 or 12 days of treatment with clinically relevant doses of ciprofloxacin fail to eradicate Salmonella from spleen and liver (as for susceptible mice), resistant mice can control the surviving Salmonella without relapsing disease (39, 123). As many as 8% of humans in endemic areas for typhoid fever carry Salmonella in liver, gall bladder, and/or gall stones without showing typhoid disease symptoms (124–126), suggesting persisting but controlled infection. It is thus possible that some human typhoid fever patients show clinical cure despite microbiological eradication failure similar to what is observed in genetically resistant mice. Eradication failures that lead to clinically detected disease relapses might happen more frequently in patients with weak immunity. This could be tested in future clinical studies.

Our results show that STP tomography is a robust method to track and characterize scarce, micrometer-sized pathogen cells in their whole-organ host context. We focused our analysis on spleen but observed surviving Salmonella survivors also in Peyer’s patches, mesenteric lymph nodes, and liver. All these reservoirs might contribute to relapses. The immediate increase of Salmonella in the spleen following treatment discontinuation in our model suggests local regrowth, but we cannot rule out migration between different sites. STP can be used to analyze all relevant organs, opening opportunities to identify and compare host microenvironments of surviving Salmonella and Salmonella fates during relapse in anatomically and functionally diverse tissues.

However, STP has also important limitations. The optical resolution of two-photon microscopy prevents detection of individual microbial cells within microcolonies. Currently available light-sheet systems still have inferior optical resolution for large tissues (127), but future improvements might ameliorate this issue. In vivo antibody staining can reveal surface-exposed epitopes in RP and MZ, but not intracellular structures or any epitopes in WP which excludes circulating antibodies in intact animals (68). Tissue-clearing methods enable antibody staining and visualization in permeabilized organs (127) and might be compatible with STP.

Our findings suggest opportunities to improve treatment of relapsing human invasive salmonellosis. In particular, adjunctive immune therapies that are already in clinical development for other medical indications, facilitated Salmonella clearance. Future studies might unravel optimal dosing of such therapies. Further insights into inflammatory mechanisms contributing to antimicrobial clearance of Salmonella could improve specific targeting with minimal adverse reactions. Our strategy and approaches might be applicable to elucidate mechanisms of bacterial persistence in other difficult-to-treat infections such as tuberculosis, which is also associated with heterogeneous bacterial colonization, slow but ongoing mycobacterial in vivo replication even after prolonged chemotherapy, poor activity of antimicrobials, and dynamically changing inflammatory responses (128).

Materials and Methods

Study Design.

In this study, we developed a method, STP tomography, to study persisting Salmonella cells in infected animals. Mice were randomized for experiments comparing treatments. Investigators were not blinded to allocation during experiments and outcome assessment. However, tomography and flow cytometry were performed automatically. We estimated sample size by a sequential statistical design. We first infected two to three mice each based on effect sizes and variation observed in our previous studies (129), and used the results to estimate group sizes for obtaining statistical significance with sufficient power. The number of replicates is indicated where applicable.

Salmonella Strains.

Salmonella strains used in this study were based on S. enterica serovar Typhimurium SL1344 hisG xyl (130, 131). GFP- or YFP-expressing strains carried the fluorescent protein–encoding genes in the sifB locus that has homogenous high activity in vivo (55). TIMERbac was expressed from the PybaJ promoter (132) on a low-copy episomal pSC101 derivative (42). Reporter strains carried pSC101 derivatives with constitutive expression of mCherry or mtagbfp2 from the PybaJ promoter and transcriptional fusions of candidate promoters to gfp-ova coding for degradable fluorescent protein variants (half-life in the range of 30 min) (133). Salmonella mutants with gene deletions were obtained by two consecutive single crossovers with positive selection for resistance to kanamycin and negative selection for levan sucrase–mediated sensitivity to sucrose (134).

To mimic in vivo conditions, we cultured Salmonella in 100 mM 2-(N-morpholino)ethanesulfonic acid (MES), 5 mM KCl, 15 mM NH4Cl, 0.5 mM K2SO4, 1 mM KH2PO4, 50 μM MgSO4, 0.02% casamino acids, 0.02% glycerol, 0.0042% N-acetyl-glucosamine, 0.003% glucose, 0.0018% glucosamine, 0.005% histidine, 25 mM NaHCO3, pH 5.5, and with a constant stream of 10% O2/5% CO2 in minichemostats (135) at a generation time of 6 h. Growth conditions were maintained for 72 h before addition of antibiotics.

Animals.

We used 10- to 16-wk-old female BALB/c mice (Charles River Laboratories). Mice were housed in groups on a 12-h light/dark cycle in individually ventilated cages. The mice had ad libitum access to food and acidified drinking water. All animal experiments were approved (license 2239, Kantonales Veterinäramt Basel) and performed according to local guidelines (Tierschutz-Verordnung, Basel) and the Swiss animal protection law (Tierschutz-Gesetz).

Age-matched, 10- to 16-wk-old female BALB/c mice (Charles River Laboratories) were deprived of food and water for 4 h and then infected orally with around 5 × 107 CFUs. At day 5 postinfection, 50 to 80% of mice showed clear symptoms of disease (score 4 or 5, see below). To minimize the impact of variable incubation times after oral infection, we analyzed only those diseased mice. For some experiments, mice were infected by tail vein injection of about 1,000 CFUs. Intravenously infected mice show less variation in Salmonella tissue loads between individual mice and thus require fewer experimental animals for detecting differences. No i.v. infected animals were excluded from analysis. Salmonella have indistinguishable replication rates, net growth, and enrofloxacin susceptibility in the spleen after oral or i.v. infection (42).The inoculum size was determined by plating for each infection.

Details of antibody injections, scoring of disease signs, and organ preparation are described in .

Flow Cytometry.

The spleen was homogenized in ice-cold phosphate-buffered saline containing 0.2% Triton X-100. All samples were kept on ice until analysis. Large host cell fragments were removed by centrifugation at 500 × g for 5 min. The supernatant contains more than 90% of the total Salmonella population as dispersed single cells (76). Relevant spectral parameters were recorded in a FACS Fortessa II equipped with 405-nm, 488-nm, and 561-nm lasers (Becton Dickinson), using thresholds on side scatter (SSC) to exclude electronic noise. Experimental details are described in .

Serial-Section Two-Photon Tomography.

We embedded the fixed tissues in 5% oxidized agarose, which was freshly prepared by incubating type I agarose (Sigma-Aldrich, A0169) in 10 mM NaIO4 (Sigma S1878) for 2 h. Tissues were covalently cross-linked to the agarose by a 2.5-h incubation in an excess of borohydride buffer (10 mM NaBH4, Sigma-Aldrich 452882, in 50 mM borax, Sigma 221732/50 mM boric acid, Sigma B6768, at pH 9.0, stirred overnight at 4 °C to dissolve). Spleens infected with TIMERbac-expressing Salmonella were embedded in nonoxidized 5% type I agarose (Sigma-Aldrich, A0169) without cross-linking to prevent alteration of TIMERbac emission spectra. We imaged embedded tissues using a two-photon whole-organ tomograph (TissueCyte 1000, TissueVision). We cut physical sections every 50 μm (spleen), 20 μm (liver), or 30 μm (Peyer’s patches and mesenteric lymph nodes) at a speed of 0.3 mm/s (0.05 mm/s for liver and noncross-linked spleen). After each sectioning, we imaged layers 20 and 30 μm (all tissues), 40 μm (all except liver), and 50 and 60 μm (spleen) below the blockface in three channels (blue channel: 450 to 500 nm; green channel: Chroma Technology ET510/20m, 500 to 520 nm for GFP and green TIMERbac emission, inbuilt 500 to 560 nm for GFP/YFP; red channel: 560 to 650 nm) using 940 nm excitation laser light (Mai Tai eHP, Spectraphysics). We imaged with a resolution of 0.435 μm in x and y by using a 20× objective with numerical aperture (NA) 1.0 (Plan Apo, Zeiss 421452-9880-000).

We used the CIDRE method for retrospective illumination correction (53) for each channel and optical plane with further correction for imaging depth. Each imaged section was formed from 800- × 800-μm “tiles” with an x/y overlap of 10%. There was no overlap in z. Illumination-corrected tiles from the entire spleen scan were stitched based upon their expected position in the x/y tile grid (no registration was conducted at tile overlap regions) using the MATLAB-based software package, StitchIt (https://github.com/SainsburyWellcomeCentre/StitchIt) (136). We segmented GFP- and YFP-expressing Salmonella cells with supervised machine learning by training the MATLAB support-vector machine classifier based on blue and green channels for GFP and all three-color channels for YFP. To remove autofluorescence among particles classified as “YFP,” we trained an eight-layer deep convolutional neural network constructed with the MATLAB layers function. All analyses were conducted using MATLAB 2017b and 2018b. The custom code is available through GitHub (https://github.com/BumannLab/Li_BumannLab_2020).

Immunohistochemistry.

Fifty-micrometer-thick spleen sections retrieved from the TissueCyte were stored at −20 °C in cryoprotectant (300 g sucrose, 10 g polyvinyl-pyrrolidone 40, 400 mL ethylene glycol, 400 mL PB buffer (10 mM NaH2PO4, 40 mM Na2HPO4), after dissolving, PB buffer was added to 1 liter) (137). Sections were washed thrice for 5 min with Tris-buffered saline pH 7.4 (TBS). For antigen retrieval, sections were incubated in 5 mL prewarmed 10 mM sodium citrate, pH 8.5, at 37 °C for 45 min. Sections were washed thrice with TBST (0.1% Triton X-100 in TBS), blocked for 1 h with 1% BSA-fraction V, 2% mouse serum in TBST, and washed again thrice with TBST. Sections were incubated in 10 µg/mL primary antibody in TBST at 4 °C overnight, washed thrice with TBST, and stained with 10 µg/mL secondary antibody for 1 h at room temperature (antibodies are listed in ). Sections were washed thrice with TBST and mounted in 70 µL mounting medium (Dako) and examined using a Leica SP8 confocal microscope.

Laser-Capture Microdissection, Enrofloxacin ELISA, and RT-PCR.

Freshly prepared unfixed spleen was wrapped with aluminum foil and put into a 50-mL tube precooled in liquid nitrogen. Frozen spleen was stored at −80 °C. Frozen 50-µm-thick sections were cut at −14 °C with a cryotome (Leica, CM1950), placed on a PET FrameSlide (Zeiss, 415190-9101-000), and dried overnight in a 50-mL tube containing 10 mL silica gel orange (Sigma, 13767). Dried sections were cut with a laser-capture microdissection microscope (Zeiss PALM-Microdissection) with a 5× objective, 100% laser power, 68% focus, and three cut cycles. The area of all the cuts was recorded and summed. The cut pieces were collected using fine forceps under a macroscope (Olympus SZ60). Enrofloxacin was quantified with the MaxSignal Enrofloxacin ELISA test kit (Bioo Scientific 1017-01) and converted into concentration values based on corresponding microdissection areas and the section thickness of 50 µm.

For RT-PCR, frozen 50-µm-thick sections were cut at −14 °C with a cryotome (Leica, CM1950), placed on PET FrameSlide (Zeiss, 415190-9101-000), and stored in a 50-mL tube with 45 mL pure ethanol (Sigma, 1117271000). Before laser dissection, slides were taken out and put into a 1-L bottle containing 500 mL silica gel orange (Sigma, 13767) for 30 min. Dried sections were cut with a laser-capture microdissection microscope (Zeiss PALM-Microdissection) with a 5× objective, 100% laser power, 68% focus, and three cut cycles. Microdissected tissue was collected in TRIzol Reagent (Fisher Scientific, 15596026). RNA was extracted with a Quick-RNA Microprep kit (Zymo, R1050). RNA was reverse transcribed with a RT-PCT kit (Promega, A6010) and random oligonucleotide hexamers. cDNA was quantified with a real-time PCR system (Applied Biosystems StepOnePlus) using glyceraldehyde 3-phosphate dehydrogenase (GAPDH) as internal reference. The following oligos were used (GAPDH, TGTGTCCGTCGTGGATCTGA, CCTGCTTCACCACCTTCTTGAT; Ifng, AACGCTACACACTGCATCTTGG, GACTTCAAAGAGTCTGAGG; Tnfa, GGATACCAACTATTGCTTCAGCTCC, AGGCTCCAAATATAGGGGCAGGGTC; and MPO, TCCCACTCAGCAAGGTCTT, TAAGAGCAGGCAAATCCAG).

Ultrasound.

Blood flow was measured in mice under isoflurane-anesthesia using Vevo3100 equipped with a MX550D transducer at 32 MHz (Visualsonics).

Quantification and Statistical Analysis.

Statistical tests were performed with GraphPad Prism 8.0.2 as indicated in the figure legends. Error bars indicate mean ± SD unless otherwise noted. Statistical tests and associated P value are reported in Results or the figure legends. All statistical tests were two sided.