Research ArticleSystems Immunology

Coordinate actions of innate immune responses oppose those of the adaptive immune system during Salmonella infection of mice

See allHide authors and affiliations

Sci. Signal.  12 Jan 2016:
Vol. 9, Issue 410, pp. ra4
DOI: 10.1126/scisignal.aaa9303

Assembling the immune system jigsaw

Analysis of individual components of the immune response to infection solves only small parts of the immune system puzzle. Hotson et al. analyzed the variable immune responses of Salmonella-infected mice over time by measuring the numbers and signaling states of multiple immune cell types and combining these data with measurements of serum cytokine concentrations, antibody responses, and bacterial burden. Mathematical analysis of this multiparametric data set revealed response elements that clustered together into patterns that showed how different components of the immune system interacted with each other. For example, neutrophils inhibited the functions of B cells during the course of infection. Thus, this systems biology approach enables the assembly of interconnected sections of the immune system puzzle.

Abstract

The immune system enacts a coordinated response when faced with complex environmental and pathogenic perturbations. We used the heterogeneous responses of mice to persistent Salmonella infection to model system-wide coordination of the immune response to bacterial burden. We hypothesized that the variability in outcomes of bacterial growth and immune response across genetically identical mice could be used to identify immune elements that serve as integrators enabling co-regulation and interconnectedness of the innate and adaptive immune systems. Correlation analysis of immune response variation to Salmonella infection linked bacterial load with at least four discrete, interacting functional immune response “cassettes.” One of these, the innate cassette, in the chronically infected mice included features of the innate immune system, systemic neutrophilia, and high serum concentrations of the proinflammatory cytokine interleukin-6. Compared with mice with a moderate bacterial load, mice with the highest bacterial burden exhibited high activity of this innate cassette, which was associated with a dampened activity of the adaptive T cell cassette—with fewer plasma cells and CD4+ T helper 1 cells and increased numbers of regulatory T cells—and with a dampened activity of the cytokine signaling cassette. System-wide manipulation of neutrophil numbers revealed that neutrophils regulated signal transducer and activator of transcription (STAT) signaling in B cells during infection. Thus, a network-level approach demonstrated unappreciated interconnections that balanced innate and adaptive immune responses during the dynamic course of disease and identified signals associated with pathogen transmission status, as well as a regulatory role for neutrophils in cytokine signaling.

INTRODUCTION

Flow cytometric analysis with antibodies specific for phosphorylated proteins (phospho-flow) enables phospho-proteomic measurements of signal transduction at the single-cell level (1, 2). By combining the detection of the activated, phosphorylated forms of signaling proteins with phenotypic cell surface markers that distinguish immune cell lineages, cell subset frequencies and signaling states are revealed. In addition, how disease perturbs the functional responsiveness of cells may be evaluated by studying immune cell signaling in response to stimulation with cytokines (3, 4). Cell type–specific signaling states can then be correlated across a population to identify signaling nodes that act in concert (5). Here, immune cell population frequencies and their signaling capacities, as well as antibody and cytokine production and bacterial loads across multiple organs in mice infected with Salmonella, were analyzed using a systems biology approach to elucidate activation of the immune network in response to chronic bacterial infection.

Infection with a given pathogen may elicit strikingly different outcomes in different individuals. For example, Salmonella Typhi causes potentially lethal typhoid fever in most infected people, but a small subset of individuals (most famously, Typhoid Mary) become asymptomatic carriers capable of bacterial transmission (6). Likewise, less than 20 colony-forming units (CFUs) of Salmonella enterica serovar Typhimurium (S. Typhimurium) cause lethal disease in C57BL/6 and BALB/c mouse strains (7), whereas Nramp1+/+ 129sv mice survive infection but are persistent carriers of the bacterium (8). Nramp1 (natural resistance–associated macrophage protein 1) is localized to the lysosomes of macrophages and dendritic cells and is necessary for induction of proinflammatory cytokine and chemokine production in S. Typhimurium–infected 129sv mice and for resistance to Salmonella-induced cell death (9). However, the genetic variation of hosts does not fully dictate disease outcome, because even within the inbred 129sv strain, the amount of bacteria shed in the feces varies by 10 orders of magnitude (10). The mice with the highest fecal loads of S. Typhimurium also exhibit gastrointestinal inflammation and are termed “super-shedders” because they can transmit infection to uninfected mice (10). The factors that dictate disparate disease progression remain unclear, although the natural microbiota certainly plays a role because pretreatment with antibiotics causes all infected mice to become super-shedders (10).

The host immune response to S. Typhimurium involves a complex interplay between innate and adaptive immunity, as well as negative feedback to temper the immune response. Persistent infection enables us to analyze the interaction between the innate and adaptive arms of the immune response because they are modulated by the pathogen over time. The innate immune response to pathogenic assault is comprised of microbial recognition and clearance as well as cytokine release, which shapes the adaptive immune response. Neutrophils are critical for containing bacteria at disease onset in susceptible mice, as exemplified by neutrophil-depleted mice having higher systemic loads of bacteria (11). In addition, S. Typhimurium induces a T helper 1 (TH1)–biased adaptive response, and neutralization of the key TH1 cytokine interferon-γ (IFN-γ) causes increased bacterial burden in multiple organs of chronically infected mice (8). At the same time, regulatory T cells (Tregs) can abrogate the proliferation of CD4+ TH1 cells (12). In addition, antibody production by B cells is required for a protective response against S. Typhimurium (13).

Immune system studies have almost exclusively focused on single-cell analyses by flow cytometry and correlating a few cell types at a time with a given mechanistic or clinical outcome. The advent of phospho-flow and other intracellular staining techniques added signaling to the repertoire of cellular events that can be measured; however, it is clear that the immune system works as a concerted network of dynamically diverse players controlled over time to lead to a given outcome. High-throughput data collection of immune system attributes, such as gene expression profiles, could potentially be used to construct correlation networks of cellular attributes to attempt to discern systems-level functionalities. In this manner, we would not look at the immune system as a set of cells with attributes, but rather as a set of cellular attributes organized into correlated functions. Here, we applied this principle to S. Typhimurium infection as a model to determine the potential of the general approach.

In a previous study, we compared the host immune response between super-shedder and non–super-shedder mice and found that super-shedders had greater neutrophilia (increased numbers of neutrophils) and a dampened TH1 response (14). Here, both the dynamic progression of the host immune response during chronic Salmonella infection and the differences in connectivity between an infected and uninfected host immune network were examined. Markers of a TH1 response, antibody response, cytokine signaling, and innate immunity were measured and correlated across mice with established infection and, separately, across uninfected mice to evaluate how infection perturbs the immune system network. Infection induced innate immunity attributes that were greatest in mice harboring high fecal bacterial loads, whereas mice with better controlled gastrointestinal infection exhibited an enhanced adaptive immune response and cytokine signaling responses. Interventions to manipulate neutrophils were used to validate an infection-specific neutrophil regulation of B cell signaling. This approach reveals new intercellular relationships as well as previously unappreciated systems-level interactions that comprise a network of functional immune response characteristics.

RESULTS

Persistent S. Typhimurium infection induces sustained activation of both innate and adaptive immune responses

Mice were infected with S. Typhimurium, and over a 28-day time course, splenocytes were isolated for staining with a panel of antibodies against phenotypic cell surface markers for analysis by traditional fluorescence-based flow cytometry. The cell populations were delineated and organized using the spanning-tree progression analysis of density-normalized events (SPADE) algorithm (15). Specific cell types were organized on the basis of the intensity of surface marker staining to identify cell lineages (Fig. 1, A and B).

Fig. 1 Rapid accumulation of innate cell types in the spleen precedes the adaptive immune response.

(A to C) Splenocytes from mice infected with 5 × 108 Salmonella bacteria were prepared for flow cytometry and analyzed by SPADE. Each bubble represents a group of cells defined by a specific lineage marker expression profile and is called a cell circle. The coloring of each cell circle represents the surface marker staining intensity (A and B) or the change in cell frequency (C) from five mice sacrificed at each time point over the 29-day time course. Each plot was generated from an equal number of cell events merged in silico from five mice. The number of cells in each cell circle is represented by the size. (A) In the SPADE plot, cell circles are colored by CD8 staining intensity, demonstrating the grouping of cell circles into cell types by lineage-specific markers (represented by a magenta box). (B) Cell circles are colored by Ly-6C staining intensity, as an example of a gradient of phenotypic marker intensity across a cell type (represented with a black arrow). (C) This figure uses color to show the change in cell frequency across the infection time course. Cell circles are colored by the change in the frequency of cells occurring in each circle at an infection time point compared to those in uninfected mice. The formula used to color cell circles was as follows: log10(percentage of cells in the cell circle during infection/percentage of cells in the cell circle in uninfected mice). An increased frequency is indicated by orange, no change by olive, and a decrease by purple. pDC, plasmacytoid dendritic cell.

SPADE enables visualization of the entire hematopoietic cell map, including cells with intermediate amounts of surface markers, which reflect the transition states of cell progression (16). To determine how cell subset frequencies were modulated by infection, we normalized the number of cells in each cell type cluster to those in the uninfected state (Fig. 1C). At 4 days postinfection (d.p.i.), there was an increase in the numbers of cells in the granulocyte and macrophage populations relative to those in uninfected mice, which became fully established by 8 d.p.i. (Fig. 1C). Activation of cells of the adaptive immune system (the adaptive compartment) also occurred during the time course, with the expansion (increase in number) of Ly-6C+ plasma cells (17, 18), which was apparent at 6 d.p.i. (Fig. 1C and fig. S1A). CD4+ T cells positive for CD44, CD49d, Ly-6C, and Ki-67 were increased in number at 8 d.p.i. (Fig. 1C). Long-term memory CD4+ T cells express Ly-6C on the cell surface (19), CD44 is a marker for effector/memory cells, CD49d is increased in abundance as activated T cells undergo cellular divisions (20, 21), and Ki-67 is expressed during the antigen-specific induction of T cell proliferation (22). In chronically infected mice, TH cells underwent continued expansion; the CD44+ Ly-6Cmid-hi population increased continuously, from 8% of the total CD4+ T cells at 8 d.p.i. to 21% at 29 d.p.i. (Fig. 1C and fig. S1B).

Infection-induced CD4+ T cells were TH1-biased as indicated by detection of the transcription factor T-bet and of the phosphorylation (and activation) of signal transducer and activator of transcription 4 (STAT4) in response to stimulation with interleukin-12 (IL-12) (fig. S1, C and D). In addition, infection induced a population of T-bet+Foxp3+CD4+ T cells that were highly responsive to IL-12. This phenotype is consistent with that of TH1-like Tregs (23). CD4+ T cells with antigen specificity toward S. Typhimurium continuously expanded through 30 days of infection (fig. S1D). To identify phenotypic markers associated with antigen-specific TH1 cells, we cultured splenocytes from infected mice with bone marrow–derived macrophages that were primed to present Salmonella antigen. Intracellular cytokine staining revealed that within the CD44+ effector/memory TH cell population, IFN-γ production increased with Ly-6C abundance (fig. S1E). Thus, Ly-6C could be used in conjunction with CD44 to define TH cell subsets enriched for the potential to secrete IFN-γ. Together, these data suggest that Salmonella infection induces a rapid, sustained innate immune response in the host. The proliferation of activated T cells was detected a week after infection, and their numbers increased through the first month. Thus, the homeostatic state during chronic Salmonella infection was characterized by increased numbers of innate myeloid cells, plasma cells, and activated TH cells. This opens two key questions: What are the relationships between the functions of these changes to cell subsets, and to what extent is there a dynamic set of relationships that precisely define such changes?

Persistent S. Typhimurium infection remodels the signaling potentials of multiple splenocyte subsets

During infection, the ability of cells to respond to cytokines may be directly modulated by the pathogen or as a secondary effect of the host immune response and cytokine production. To evaluate the cell signaling potential during chronic Salmonella infection, we exposed splenocytes isolated from mice at 30 d.p.i. to a set of cytokines that activate the maximum number of cell subsets with distinct roles in immune processes: IFN-γ (TH1-biased), IL-6 (acute-phase response), IL-10 (anti-inflammatory), and IL-21 (antibody response in B cells). The ability of cells to respond to these cytokines was determined by measurement of the relative abundances of phosphorylated STAT1 and STAT3 proteins (pSTAT1 and pSTAT3, respectively).

We observed that S. Typhimurium infection resulted in changes to cytokine responses in nearly every major cell type. For example, IL-21–induced STAT1 and STAT3 phosphorylation was greater in lymphocytes from infected mice than in lymphocytes from uninfected controls [P < 0.05 for populations of B cells, CD8+ T cells, and naïve CD4+ T cells; Kolmogorov-Smirnov (K-S) test] (Fig. 2A). Cells from infected mice also exhibited enhanced responsiveness to IL-10. IL-10 is primarily classified as an antiinflammatory cytokine, and it acts through STAT3. B cells and T cells from infected mice demonstrated enhanced IL-10–mediated STAT3 phosphorylation compared to the same populations from uninfected mice (P < 0.5, K-S test) (Fig. 2A). IL-10 also evoked stronger activation of pSTAT1 in macrophages from infected mice than in macrophages from naïve mice (P < 0.01, K-S test). As such, overall IL-10 and IL-21 responses were augmented during infection.

Fig. 2 Salmonella infection induces variability in the signaling responses of splenocytes to cytokines.

(A) Splenocytes from 12 mice infected with Salmonella for 30 days and from 4 uninfected mice were left unstimulated (unstim) or were stimulated for 15 min with IFN-γ, IL-6, IL-21 (all at 40 ng/ml), or IL-10 (80 ng/ml) before being prepared for flow cytometric analysis. The heat map shows the MFIs of pSTAT1 and pSTAT3 for the listed cell types. Mem/eff, memory/effector. (B) Correlation between the IL-6–induced phosphorylation of STAT3 in B cells and the IFN-γ–induced phosphorylation of STAT1 in effector/memory CD4+ T cells.

Note that the cells from uninfected mice responded quite uniformly; for each cell type, the resting and stimulated intensities of pSTAT1 and pSTAT3 were consistent across mice [coefficient of variation (CV) >10% for 4 of 50 conditions]. In contrast, the responses of splenocytes from infected mice exhibited marked variability (CV >10% for 49 of 50 conditions), both in their amounts of pSTAT under unstimulated conditions and in their responses to cytokines (Fig. 2A). Consistent with these data, the variance in STAT phosphorylation was statistically significantly greater across infected mice than uninfected mice for 32 of 50 conditions (Levene test). For example, the variation in the median fluorescence intensity (MFI) of pSTAT3 in IL-6–stimulated B cells was significantly (P < 0.01, Levene test) lower in uninfected mice (MFI range of 79 to 98) than in infected mice (MFI range of 71 to 168). Similarly, in IFN-γ–stimulated samples, the MFI of pSTAT1 of effector/memory CD4+ T cells ranged from 61 to 69 in uninfected mice, but expanded to between 42 and 246 in infected mice. Furthermore, infected mice were uniformly hyperresponsive either in both of these signaling nodes or in neither (Fig. 2B), demonstrating correlated responses (P < 0.01, Spearman’s correlation test) to cytokines with specialized functional outcomes: IL-6 promotes B cell growth and antibody production, whereas IFN-γ reinforces TH1 biasing in TH cells. Therefore, even at the observation level, it was apparent that signaling events associated with distinct immune consequences were correlated within infected mice. This provoked us to ask whether deeper or more extensive correlations across the immune system existed that could enable a more thorough, or systems-level, interpretation of immune actions in infected mice.

Correlation maps demonstrate extensive Salmonella-driven remodeling of host immune response

Salmonella infection, whether acute or chronic, would be expected to initiate co-regulated changes to cytokine responsiveness across various cell types. We hypothesized that the diversity of bacterial loads and infection outcomes across genetically identical mice would reveal variable immune responses, with specific immune attributes being more or less activated in more highly infected mice. To objectively illustrate coordinated immune events during persistent Salmonella infection, we determined correlation values between measurements of 125 attributes of the immune system including STAT signaling, cell frequencies across organs, markers of T cell and B cell activation, serum cytokine concentrations, and bacterial load. The intensities of each of these attributes were correlated across 20 uninfected mice and, separately, across 19 mice infected with Salmonella for 30 to 35 days. For each node-node pair (wherein a node is a cell frequency, bacterial load, serum cytokine value, or a cellular attribute such as STAT1 phosphorylation in a given cell type in response to a perturbing cytokine), it was determined whether the correlation was statistically significantly positive, significantly negative, or insignificant using data from two processing batches (see Materials and Methods). Of 7750 possible correlations, 2223 positive and 443 negative significant correlations were observed in infected mice (fig. S2A), whereas there were only 835 positive and 318 negative attribute correlations in uninfected mice (fig. S2B) as representing the “basal” state. This finding suggests that infection induces coordinated activity of the immune system above the basal state.

Many correlations between immune attributes were similar between uninfected and infected mice. Of the correlations observed in uninfected mice, 44% (512 of 1153) were maintained during infection, whereas only 5.8% (68 of 1153) of the correlations in uninfected mice were of the opposite direction in infected animals (Fig. 3A). The correlations that reversed direction during infection were associated with four nodes: splenic Treg frequency, splenic neutrophil frequency, basal STAT3 phosphorylation in B cells, and IL-21–induced STAT1 phosphorylation in granulocytes. These findings suggest that the regulatory units that control the coordinated actions of multiple immune functions are primarily conserved and amplified during infection, although a small subset of the immune response, such as Tregs, becomes differentially regulated.

Fig. 3 Transmission status and splenic neutrophil frequency share a similar correlation pattern during Salmonella infection.

(A) The map of correlations that reversed directions between uninfected and infected mice (20 uninfected mice and 19 mice infected for 30 to 35 days). Green circles are nodes with four or more reversals. Blue lines represent negative correlations during infection that were positive in uninfected mice, and red lines represent positive correlations during infection that were negative in uninfected mice. (B) The correlation map of infection shows all correlations with fecal bacterial load (feces CFU) and splenic neutrophil frequencies (Gr1) (filled magenta circles) obtained from data from 19 mice infected for 30 to 35 days. Red lines signify positive correlations, and blue lines signify negative correlations. Outlined blue or red circles share correlations with either fecal CFU or splenic neutrophil frequency, solid blue circles share negative correlations with both, and gray circles are not correlated with either. int, intermediate; hi, high; cDC, conventional dendritic cell; Macs, macrophages.

A large number of the negative correlations observed in the infected state involved either fecal bacterial load or splenic neutrophilia (Fig. 3B), demonstrating that these factors generally opposed other measured events of the immune system. For example, both fecal CFU and neutrophil frequencies were negatively correlated with splenic TH1 frequencies and with cytokine-induced STAT1 and STAT3 phosphorylation in multiple splenic cell subsets. In addition, fecal Salmonella CFU was negatively correlated with CD4 Ki-67, a measure of T cell proliferation, and Ly-6C, indicative of IFN-γ production, whereas neutrophil frequencies were associated with reduced plasma cell frequency, dampened IL-2 and IL-12 activity in CD4+ T cells from mesenteric lymph nodes (mLNs), and increased Treg numbers. Together, these data reveal that increased bacterial fecal shedding is associated with neutrophilia, reduced Janus-activated kinase (JAK)–STAT signaling, and a suppressed adaptive immune response. Thus, correlation analysis revealed that the immune network that forms during persistent Salmonella infection was balanced between an enhanced neutrophil response and an enhanced responsiveness of T cells.

Known immune interactions are identified through correlation analysis

To establish whether the correlated attributes of the host immune response to Salmonella infection were reflective of biologically meaningful relationships, we compared a subset of the correlation network to known infection-induced immune connections. In the Nramp1+/+ model of Salmonella infection, Tregs suppress the proliferation of antigen-specific T cells (12). In addition, increased fecal bacterial load induces increased numbers of neutrophils; both of these factors also cause a reduction in the numbers of TH1-biased cells and dampen STAT1 responsiveness to IL-6 in naïve CD4+ T cells, whereas neutrophils inhibit the IL-2–mediated proliferation of activated TH cells (14). These known causative relationships between attributes of the host-microbe interaction were constructed into a network for comparison with the correlation network derived herein (Fig. 4). Our correlation network identified the established positive relationship between fecal CFU and neutrophils, as well as five of six negative relationships (Fig. 4). CD44+Ly-6ChighCD4+ T cells were used as a measure of T cell proliferation because this subset contains antigenically expanded TH cells during infection (Fig. 1 and fig. S1). Additionally, our analysis identified three correlations within this subnetwork that, to our knowledge, have not been previously reported (designated as unreported correlations): a negative correlation between bacterial shedding and T cell proliferation, and positive correlations between TH1 frequency and both T cell proliferation and IL-6 responsiveness (Fig. 4). Functional intervention and measurement of the node pairs in these three correlations have not been reported, to our knowledge; thus, it is unknown if there is a direct causal relationship between these pairs, or if an alternative regulatory element coordinates their activity.

Fig. 4 Discovered connectivity shares relationships with the known regulatory network.

This subset of network relationships from the Salmonella-infected murine system illustrates the overlap between known relationships and those reported in this work. Lines with arrowheads show previously reported causal relationships as determined through experimental intervention. The solid red and blue arrows represent known positive and negative relationships, respectively, that were present in our correlation data; the dotted blue arrow is a known relationship that was not found. Thick lines without arrowheads indicate previously uncharacterized (unreported) correlations reported here.

Clustering correlations reveal functional immune cassettes

Evaluation of paired correlations is useful for determining how a given node connects within a network as a whole (Fig. 3B). However, it may be challenging to deduce the architecture of the entire network from a series of pairwise correlations. We hypothesized that master regulator functions of the immune system (cell types or cytokines, for example) would control the activity of multiple downstream functions or elements; thus, the network would be composed of modules of co-regulated elements. Furthermore, the interactions across immune system pathways would organize both master regulators and activated elements into biologically functional cassettes that could each act in coordination or opposition to each other.

To assess whether key factors of Salmonella infection such as neutrophils, TH1 response, antibody production, cytokine signaling, and fecal CFU shared common correlations, we constructed a correlation matrix. All 125 measured attributes of the immune system were organized by hierarchical clustering. Clustering revealed four groups of attributes that demonstrated similar correlation patterns; the attributes within each group shared functional commonalities that could be considered as a module of biological activity (Fig. 5A). For example, the Innate Immunity Cassette consisted of high fecal bacterial load and IL-6 serum concentrations, high frequencies of splenic antigen-presenting cells and Tregs, and reduced effector/memory CD4+ T cell representation, which are all reported characteristics of super-shedders (14). The other cassettes identified were the STAT Signaling Cassette (consisting mostly of pSTAT1 and pSTAT3 responses and certain B cell response characteristics), the Total STAT Protein Cassette, and the Adaptive Immunity Cassette (consisting of TH cell and B cell responses). There were also relationships observed between the cassettes of biological function. For example, the relationship between the STAT Signaling Cassette and the Innate Immunity Cassette was enriched for negative correlations between their individual elements (P < 0.01, t test) (Fig. 5A, gray box), indicating that the two cassettes act in opposition to each other. The Innate Immunity Cassette was also negatively correlated with the Adaptive Immunity Cassette (P < 0.01, t test) but was not significantly correlated with the Total STAT Protein Cassette (Fig. 5B). STAT signaling was positively correlated with the adaptive immune response and with the Total STAT Protein Cassette (Fig. 5, B and C). Therefore, mice harboring the greatest fecal Salmonella loads and transmission capacity demonstrated an enhanced innate immune response that actively opposed adaptive immunity and STAT signaling.

Fig. 5 Clustering of the Salmonella infection correlation matrix reveals functional cassettes comprised of attributes with shared correlations.

(A) The correlation network from 19 infected mice shown in fig. S2B is given in matrix form. The matrix was clustered and attributes that clustered together (black boxes) were identified and annotated on the basis of function. Red represents a positive correlation, white represents no correlation, and blue represents a negative correlation. The gray box highlights the intersection of two clustered cassettes. (B) The average value within a cluster is represented colorimetrically from blue (negative) to red (positive). The rectangle at the intersection of two cassettes represents the relation between the two cassettes. (C) A pictorial representation of the relationships between the cassettes. A red line represents a positive correlation, whereas a blue line represents a negative correlation.

The presence of these cassettes in uninfected immune system networks was then investigated (Fig. 6A). A χ2 test was used to determine whether cassettes defined by the analysis of data from infected mice were statistically significantly enriched for correlations in data from uninfected mice compared to the overall correlation rate. Although there were fewer correlations in data from uninfected mice, the cassettes identified by the analysis of data from infected mice were largely statistically recapitulated (albeit at a weaker level). The STAT Signaling, Total STAT Protein, and Adaptive Immunity Cassettes were significantly enriched for positive correlations, suggesting that the elements of these cassettes functioned in concert in uninfected mice. Furthermore, the intersections of these cassettes were also enriched for positive correlations. In the absence of infection, there was no significant correlation within the Innate Immunity Cassette, reflecting the absence of a regulated innate response in uninfected mice (Fig. 6A). A positive correlation between the Total STAT Protein Cassette and the Adaptive Immunity Cassette existed only in the uninfected condition (Fig. 6B).

Fig. 6 Uninfected mice exhibit co-regulation of adaptive, but not innate, immunity markers.

(A) The correlation network data obtained from 20 uninfected mice in fig. S2A is shown in matrix form. The matrix was arranged in the same order as that for infected mice in Fig. 5. n.s., not significant. (B) Representation of the relationships between cassettes.

B cells are regulated by neutrophils during infection

Of the four functional cassettes identified, only the Innate Immunity Cassette (containing neutrophil frequency) was unique to infected mice (Fig. 6A), and correlations between neutrophil frequency and other nodes were reversed during infection relative to correlations in the data from uninfected mice (Fig. 3A). Given the observations that neutrophil frequency switched to a negative correlation with the extent of phosphorylation of STAT1 in B cells during infection and that B cell basal pSTAT3 abundance also switched correlations with many nodes (Fig. 3A), we investigated B cell STAT signaling. Mice were administered granulocyte colony-stimulating factor (G-CSF) to induce granulopoiesis and sacrificed 3 days later, and then splenocytes were analyzed. In both infected and uninfected mice, G-CSF dampened baseline STAT1 and STAT3 phosphorylation, as well as inhibited the IL-6– and IL-10–dependent phosphorylation of STAT3 in splenic B cells (Fig. 7A). Furthermore, neutrophil frequencies were increased in infected mice, and granulocyte depletion by a neutralizing antibody led to augmented STAT signaling. Neutrophil depletion increased the basal phosphorylation of STAT1 and STAT3 and enhanced the IL-6–induced phosphorylation of STAT3 in splenic B cells from these mice relative to that in B cells from untreated mice (Fig. 7B). Together, these results suggest that neutrophils suppress STAT signaling in B cells and that neutrophils are abundant enough during infection to exert this effect.

Fig. 7 Neutrophils regulate STAT signaling in B cells.

(A) Uninfected and 30-day Salmonella-infected non–super-shedder mice were left untreated or were given 1 μg of G-CSF intraperitoneally daily for 3 days before being sacrificed on the fourth day. Data are means ± SD of four mice per group. The splenocytes were left untreated or were treated with IL-6 (40 ng/ml) or IL-10 (80 ng/ml) for 15 min. Fixed and permeabilized cells were stained for phenotypic markers, and the relative abundances of pSTAT1 and pSTAT3 in B cells were quantified by flow cytometric analysis. *P < 0.05 and **P < 0.001, two-sided t test. (B) Infected mice were treated with PBS as a control or were treated with 1 μg of anti–Ly-6G (clone IA8) or anti-Gr1 (clone RBC-8C5) antibody intraperitoneally daily for 3 days and then were sacrificed on the fourth day. Data are means ± SD of three mice per group. The splenocytes were left untreated or were treated with IL-6 (40 ng/ml) for 15 min. Fixed and permeabilized cells were stained for phenotypic markers, and the relative abundances of pSTAT1 and pSTAT3 in B cells were quantified by flow cytometric analysis. *P < 0.05 and **P < 0.001, two-sided t test.

DISCUSSION

Systems biology approaches reveal coordinated immune activity during infection

This work used the structured variation in a continuum of immune responses to Salmonella infection to discover organized biologically functional immune cassettes that were associated with infection severity. The values representing multiple attributes of immune system function were evaluated during chronic Salmonella infection and in uninfected mice. By clustering the correlation values of these measurements of immune activation into functional cassettes, elements of immunity that acted in concert during bacterial challenge were identified. Infection induced the amplification of certain aspects of the naïve immune network, whereas the relationships among key regulators, such as neutrophils and Tregs, and the system were altered upon infection. In addition, differences in the immune system associated with transmission status were detected because super-shedders exhibited a globally increased innate response and reduced adaptive response and STAT signaling compared to those of moderate shedders of bacterial load in feces. In addition, the identified relationships among aspects of immunity enabled us to hypothesize and show by perturbation that the onset of neutrophilia during infection in turn regulated B cell STAT signaling.

Cassette intensities predict super-shedder immune phenotype

Mice with high fecal loads of S. Typhimurium can transmit disease and are known as super-shedders. Compared to infected mice that do not shed large amounts of bacteria, super-shedders have a suppressed adaptive immune response (as measured by reduced numbers of TH1 and plasma cells and increased numbers of Tregs), a dampened STAT Signaling Cassette, and increased neutrophils, serum IL-6, and nuclear factor κB (NF-κB) pathway activation (characteristics of an activated innate immune system). Therefore, measurement of representative markers from the identified cassettes provided information on intestinal disease severity and potential for pathogen transmission. This could prove useful in cases of host-adapted pathogens, where identification of a transmitter is of key importance to halt the spread of disease, but consistent detection of the pathogen is difficult. Identification of transmitters by representative biomarkers would enable more reliable detection and faster intervention. Additionally, this systems analysis approach could likely be extended to other disease systems to guide the identification of peripheral blood markers that are predictive of severity and prognosis.

Healthy and diseased networks differ

In comparing the correlation network that formed the basis of immune homeostasis in uninfected mice to Salmonella infection, notable differences in both pairwise correlations and relationships between functional cassettes were detected. The number of node correlations increased during infection, likely because of the activity and variability of the immune response. Four of 125 nodes switched the directionality of multiple correlations after the onset of infection. For example, splenic neutrophil frequency was positively correlated with B cell STAT1 signaling and plasma cell frequencies in uninfected mice, but these factors were inversely related during infection. Granulocyte depletion experiments demonstrated that neutrophils affected these aspects of B cell biology. Furthermore, neutrophils reduced the basal pSTAT3 abundance in B cells, and this node reversed the directionality of many correlations during infection. This finding suggests that B cell pSTAT3 and other nodes are co-regulated in the uninfected state, but that during infection, the influence of neutrophilia on B cell signaling overrides alternative regulation. Indeed, the discovery by Puga et al. (24) of B helper neutrophils (NBH) demonstrates that neutrophils affect splenic B cell function. The authors implicate mucosal bacteria in driving NBH production, which in turn secrete IL-21, BAFF (B cell–activating factor), and APRIL (a proliferation-inducing ligand) to support B cell antibody production. This finding shows that neutrophils influence B cell activity and may provide a basis for the regulation of B cell signaling observed in the context of infection.

Salmonella infection also influenced the relationship between groups of immune system attributes that worked in concert. First, the correlations among the markers identified as the Innate Immunity Cassette were observed only when the immune system was challenged by infection, presumably because of the absence of a basal innate immune response in uninfected mice. Second, the Adaptive Immunity Cassette and Total STAT Protein Cassette were positively correlated with each other in unchallenged mice, but this association was lost during infection. Therefore, we infer that STAT1 and STAT3 proteins are co-regulated with adaptive immunity in the basal state, but that during infection, additional mechanisms disrupt this correlation.

Crosstalk occurs between immune system organs

During infection, some extracellular Salmonella exit the host through the intestinal tract, and a reservoir of intracellular bacteria persists in macrophages located in the spleen and mLN. Here, the bacterial load in each of these organs was associated with a separate cassette. Fecal CFU was part of the Innate Immunity Cassette, splenic CFU was incorporated within the Total STAT Protein Cassette, and CFU in the mLN was not included in a functional cassette. Therefore, the local bacterial burden in the spleen was correlated with the increased abundance of STAT1 and STAT3 proteins in splenocytes, yet the increased signaling potential of these transcription factors in splenocytes was associated with low quantities of Salmonella in the lower intestine. IL-2–induced STAT5 phosphorylation in CD4 T cells from the mLN was correlated with splenic CFU and total amounts of STAT1 and STAT3 proteins. IL-2 induces the proliferation of activated T cells, suggesting that the ability to mount an adaptive response in the mLN is associated with bacterial load in the spleen, but not in the intestine or the mLN itself. Serum cytokine concentrations were also associated with the immune status of individual organs. For example, neutrophils in the spleen were positively correlated with the acute-phase concentrations of IL-6 but were inversely correlated with the amount of the anti-inflammatory cytokine IL-10. The immune network of uninfected mice also exhibited correlations between organs. In one example of these interactions, the frequency of neutrophils in the colon shared 18 negative correlations with splenic STAT signaling. Thus, in the basal state, there was co-regulation between neutrophil recruitment to the colon and cytokine signaling potential in the spleen. Furthermore, this relationship between colonic neutrophils and splenocytes was abrogated during infection. Thus, infection disrupts the underlying systemic regulation of the immune system.

A network-level approach reveals interlocking relationships that balance innate and adaptive immunity

Salmonella causes a TH1-biasing infection, and the immune response to this organism requires both IFN-γ and T-bet (25). Salmonella-specific CD4+ T cells produce IFN-γ ex vivo (26), and we detected these cells at 8 d.p.i., and they continued to increase in number through the first month of infection in the model used here. Characterization of the CD4+ T cells that produced IFN-γ indicated that cytokine production increased with Ly-6C abundance in the effector/memory population. Ly-6C is a marker for professional memory CD4+ T cells that reside primarily in the bone marrow (19). The increase in the number of cells in this population in the spleens of infected mice relative to that in uninfected animals supports the sustained recruitment of memory CD4+ T cells to infection sites during persistent infection. Although the clonality of the CD4+ T cell populations was not investigated here, a steady increase in the number of antigen-specific, IFN-γ–producing memory CD4+ T cells was observed after infection (fig. S1D).

In summary, markers of the TH1 response, antibody response, cytokine signaling, and innate immunity were used to evaluate how infection perturbed the immune system network. Infection induced innate immune responses in infected animals, which were greatest in mice harboring high fecal bacterial loads. These mice had systemic neutrophilia and high serum concentrations of IL-6, as well as fewer plasma cells and TH1 cells and greater numbers of Treg, compared with infected mice with lower fecal bacterial loads. Splenic neutrophils were negatively correlated with B cell signaling only in infected mice, and direct interventions revealed that neutrophils regulated B cell signaling potential.

Common approaches to understanding heterogeneity in biological systems involve analyses of individual correlations to uncover testable relationships, but often fail to take full advantage of natural variation within the data. Our approach to this unmet need was to gain additional insights by clustering natural variation as correlations that revealed a series of interacting cassettes (which further served to model a system-wide view of the host immune response). When perturbations (such as infection) induced the appearance of new cassettes or new relationships between preexisting cassettes, key drivers of the immune response were identified. Additionally, elements of the host immune response that were preserved or amplified through infection could be distinguished from those that were altered or reversed because they acted in concert. We believe that such approaches will be useful in understanding and structuring so-called heterogeneity in a wide range of biological systems.

MATERIALS AND METHODS

Mice

Female 129X1/SvJ mice (The Jackson Laboratory) were 8 to 10 weeks old at the time of infection. All mice were handled in accordance with Stanford University animal care guidelines. Neutrophil intervention experiments were performed as previously described (14). For neutrophil induction experiments, mice were injected intraperitoneally with 1 μg of PEGylated G-CSF (GenScript, Z00393-50) on each of three consecutive days and were sacrificed on the fourth day. For neutrophil depletion experiments, mice were injected intraperitoneally with 1 μg of anti–Ly-6G (clone IA8, BioXcell, BE0075-1) or anti-Gr1 (clone RB6-8C5, BioXcell, BE0075-1) on three consecutive days and then were sacrificed on the fourth day.

Bacterial infection

The S. Typhimurium strain SL1344 was grown overnight in LB. Bacteria were washed and diluted. Inoculations were performed either by oral gavage of 108 CFU in 200 μl of phosphate-buffered saline (PBS) or by feeding bread containing 108 CFU.

Cytokine stimulation

mLNs and spleens were harvested and mechanically dissociated into single-cell suspensions at a concentration of 1 × 107 cells/ml in RPMI-1640 containing 10% fetal bovine serum (FBS) and penicillin, streptomycin, and glutamine (RPMI-10). Bone marrow was isolated from mouse femurs and tibias by flushing the bones with RPMI-10. Dissociated cells were allowed to recover for 15 min at 37°C. Cells were then left unstimulated or stimulated with IFN-γ, IL-2, IL-4, IL-6, IL-12 (all from BD Biosciences), or IL-21 (R&D) (all at 40 ng/ml) or with IL-10 (BD Biosciences; 80 ng/ml). Exposure to cytokines was performed at 37°C for 15 min. Cells were then fixed with 1.6% paraformaldehyde (PFA) at room temperature for 10 min, washed, resuspended in methanol at 4°C, and stored at −80°C. Blood was collected by cardiac puncture with needles preloaded with heparin (Sigma-Aldrich) and was left unstimulated or was treated with either IL-6 or IL-12 (40 ng/ml) for 15 min. Cells were then fixed with Lyse/Fix Buffer (BD Biosciences) according to the manufacturer’s guidelines. The cells were then washed and resuspended in cold methanol and stored at −80°C.

Bacterial counts

Fecal pellets were weighed and dissociated in 1 ml of sterile PBS. mLNs and spleens were weighed and dissociated in RPMI-10. Serial dilutions were plated on LB agar plates containing streptomycin. The CFU of S. Typhimurium per weight of organ or feces was then calculated.

T cell activation

Bone marrow–derived macrophages were plated at 2.5 × 105 cells in a 24-well dish (Corning) and infected overnight with S. Typhimurium at a multiplicity of infection of 5 at 37°C in RPMI-1640 with 10% FBS and gentamicin. After 18 hours, 5 × 106 splenocytes from infected mice in RPMI-10 without antibiotics were added to the culture. After 1 hour, brefeldin A (10 μg/ml) was added, and cells were cultured for an additional 3 hours. For the last 10 min of culture, Pacific Blue maleimide (0.01 ng/ml) was added to the culture to stain membrane-permeabilized dead cells. Cells were then fixed with 1.6% PFA and collected for intracellular cytokine staining.

Flow cytometry

For staining of cell surface markers and transcription factors, fixed cells were washed twice with staining medium (PBS containing 0.5% bovine serum albumin and 0.05% sodium azide). Cells were first stained with a panel of antibodies specific for cell surface markers for 30 min in staining medium. The antibodies used were Cy7-phycoerythrin (PE)–conjugated anti-CD8, Texas Red (TR)–PE–conjugated anti-B220, Cy7-allophycocyanin (APC)–conjugated anti-CD4, Pacific Orange–conjugated anti-CD44, biotinylated anti–CD62 ligand (CD62L), Cy5.5–peridinin chlorophyll protein (PerCP)–conjugated anti-CD25, fluorescein isothiocyanate (FITC)–conjugated anti–Ly-6C, and PE-conjugated CD49d (BD Biosciences). Cells were permeabilized with staining medium containing 0.3% saponin and subsequently stained with the following antibodies: Alexa fluorophore (Ax) 647–anti–T-bet, Ax700–anti-FoxP3 (both from eBioscience), Pacific Blue–conjugated anti–Ki-67 (BD Biosciences), and streptavidin quantum dot (QD) 605 (Invitrogen). A macrophage antibody cocktail consisting of TR-PE–conjugated anti-B220, Cy5.5-PerCP–conjugated anti-CD11b, APC-conjugated anti-CD11c, Cy7-APC–conjugated anti-Gr1, FITC-conjugated anti-CD86 (all from BD Biosciences), and Cy7-PE–conjugated anti-F4/80 (eBioscience) was also used in some experiments. After staining, cells were washed with staining medium containing saponin and then were analyzed on an LSR II flow cytometer equipped with 407-, 488-, and 633-nm lasers. Digital data were acquired with BD FACSDiva software, with >50,000 size-gated cells collected per sample. Data were analyzed with FlowJo software (Tree Star). For intracellular cytokine staining, cells were first stained with antibodies against surface markers: Ax700–anti-CD4, Cy7-PE–anti-CD8, FITC–anti–Ly-6C, Pacific Orange–anti-CD44, biotinylated anti-CD62L, and Cy5.5-PerCP–anti-CD11b (all from BD Biosciences). Anti-CD44 was conjugated to Pacific Orange succinimidyl ester (Invitrogen) in house. Cells were washed and permeabilized with saponin staining medium. Cells were then stained with APC–anti–IFN-γ, PE–anti–IL-2, and streptavidin QD605. For staining of phospho-proteins, cells stored in methanol were washed twice with FACS staining buffer and then stained with the following antibodies: Ax700–anti-CD4, Pacific Blue–anti-CD8, TR-PE–anti-B220, Cy5.5-PerCP–anti-CD11b, Cy7-APC–anti-Gr1, Pacific Orange–anti-CD44, Cy7-PE–anti-CD62L, and Cy7-PE–anti-F4/80. In addition, cells were stained with a pair of phospho-specific antibodies, either Ax488–anti-pSTAT1 and Ax647–anti-pSTAT3 or Ax647–anti-pSTAT4 and Ax488–anti-pSTAT5 (all from BD Biosciences), or with PE–anti–total STAT1 (BD Biosciences) or Ax647–anti–total IκBα (inhibitor of NF-κB α) (Cell Signaling Technologies). Cells were washed and then analyzed on an LSR II flow cytometer.

Serum cytokine and antibody quantification

Blood was collected by cardiac puncture and centrifuged; the serum supernatant was collected and stored at −80°C. To determine the concentrations of IL-6, IL-10, and IFN-γ, the Cytometric Bead Array Flex Set (BD Bioscience) was used according to the manufacturer’s recommendations. For Salmonella-specific antibody detection, serum was incubated with plate-bound Salmonella lysate. Plates were washed, and horseradish peroxidase–conjugated goat anti-mouse immunoglobulin G antibody was applied. Plates were analyzed with a plate reader for relative quantification of antibody.

Visualizations

SPADE clusters flow cytometry events on the basis of marker staining intensity and displays the clusters in a minimum spanning tree, as previously described (15, 27). To create a SPADE tree, each time point and an equal number of events per flow cytometry standard (FCS) file of each of five mice were concatenated with the flowCore package (www.bioconductor.org/packages/2.3/bioc/html/flowCore.html). User-defined cell subsets were identified on the basis of the surface marker intensity within a set of cell clusters. Figure 1 was generated with a beta version of SPADE, and the public version is now available at (http://cytospade.org). The flow cytometry plots in fig. S1C were generated with MATLAB (MathWorks), whereas other flow cytometry plots were generated with FlowJo (Tree Star). Figures 2B and 6 were made with Prism (GraphPad). The cell signaling heat map in Fig. 2A was generated with Cytobank (www.cytobank.org). The network representations in fig. S2 were generated with Cytoscape (www.cytoscape.org). R software (www.r-project.org) was used for clustering using Ward’s method and generating matrices. The matrices in Figs. 4 and 5 were adjusted to a red-blue color scheme in Adobe Illustrator.

Statistical analysis

The K-S test was used to test for statistically significant differences in STAT phosphorylation between uninfected and infected mice (Fig. 2A), and the Levene test was used to calculate significant differences in variance compared to the median. The χ2 test was used to test for the enrichment of correlations within cassettes in uninfected mice. Pairwise comparisons between groups were performed with a t test, with P < 0.05 considered a statistically significant difference. For analyses of correlations, Spearman’s rank correlation test was performed on batches. Data from two groups of uninfected mice (10 mice each) and two groups of infected mice (8 and 11 mice each) were processed. To establish a random distribution of correlation values for each group for each measurement, the value was assigned to a random mouse, and then the correlation values between each of the other mice were determined. This was repeated 50 times to generate a bell-shaped distribution of correlation values. Next, the probability of every correlation occurring was determined on the basis of the random distribution curve. A correlation was deemed significant if it met two criteria: (i) the product of the probabilities was ≤0.025, and (ii) the probability was in the top third in each group. This process was also repeated with two experiments of 12 and 16 infected mice each, and the correlation map was very similar to the one presented.

SUPPLEMENTARY MATERIALS

www.sciencesignaling.org/cgi/content/full/9/410/ra4/DC1

Fig. S1. The adaptive immune response to Salmonella increases through the first month of infection.

Fig. S2. The connectivity of the immune network is increased during Salmonella infection.

REFERENCES AND NOTES

Acknowledgments: We thank K. Sachs for guidance in determining correlations, R. Bruggner for aid in creating computational figures, A. Trejo and A. Jager for assisting in laboratory execution, and W. O’Gorman for critical reading of the manuscript. Funding: This work was supported by NIH grant HHSN272200700038C and National Heart, Lung, and Blood Institute grant N01-HV-00242 (to G.P.N.) and a Burroughs Wellcome Fund Investigator in the Pathogenesis of Infectious Diseases (to D.M.). A.N.H. was funded by Molecular and Cellular Immunobiology training grant 5 T32 AI07290. S.G. was funded by the Stanford Graduate Fellowship and supported by NIH grant RO1 A1095396. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Author contributions: A.N.H., S.G., G.P.N., and D.M. conceptualized the experiments and wrote the manuscript; A.N.H. and S.G. performed the experiments; M.N. and A.K. provided statistical support; and R.F. assisted in data interpretation and figure generation. Competing interests: The authors declare that they have no competing interests.
View Abstract

Navigate This Article