Comparative Analyses of Transcriptional Profiles in Mouse Organs Using a Pneumonic Plague Model after Infection with Wild-Type Yersinia pestis CO92 and Its Braun Lipoprotein Mutant

We employed Murine GeneChips to delineate the global transcriptional profiles of the livers, lungs, and spleens in a mouse pneumonic plague infection model with wild-type (WT) Y. pestis CO92 and its Braun lipoprotein (Δl p p) mutant with reduced virulence. These organs showed differential transcriptional responses to infection with WT Y. pestis, but the overall host functional processes affected were similar across all three tissues. Gene expression alterations were found in inflammation, cytokine signaling, and apoptotic cell death-associated genes. Comparison of WT and Δl p p mutant-infected mice indicated significant overlap in lipopolysaccharide- (LPS-) associated gene expression, but the absence of Lpp perturbed host cell signaling at critical regulatory junctions resulting in altered immune response and possibly host cell apoptosis. We generated a putative signaling pathway including major inflammatory components that could account for the synergistic action of LPS and Lpp and provided the mechanistic basis of attenuation caused by deletion of the lpp gene from Y. pestis in a mouse model of pneumonic plague.


Introduction
The gram-negative bacterium Yersinia pestis is the etiological agent of plague. Y. pestis is transmitted to humans through the bite of an infected flea or inhalation of the organisms, resulting in bubonic, pneumonic, or septicemic forms of plague [1]. Y. pestis has attracted much interest recently because of its potential as a weapon of bioterrorism. Following entry within a host, Y. pestis evades the host immune system and replicates in the lymph nodes, ultimately leading to lymph node necrosis and death if untreated [2][3][4]. Histological evidence indicates that bacteria within neutrophils are killed, while bacteria within macrophages and dendritic cells survive and go on to express various virulence determinants, which allow bacterial growth and their eventual release from the macrophages [5][6][7]. For example, F1 (capsular) antigen [8] and type III secretion system (T3SS) effectors [9] are expressed only at 37 • C and have been shown to modulate the host response so that Y. pestis becomes resistant to subsequent phagocytosis. The use of these antiphagocytic mechanisms has led researchers to suggest that Y. pestis is predominantly an extracellular pathogen in the mammalian host [9,10]. However, a strong cell-mediated immune response to Y. pestis infection is seen in immunized mice, suggesting that immune cells are also needed to clear either intracellular bacteria or extracellular Y. pestis that have been opsonized. A T-cell component of protection against Y. pestis, in the absence of antibody, has been established [11,12]. In unvaccinated individuals, low doses of Y. pestis can be resolved following combined treatment with the T helper1-(Th1-) associated cytokines interferon (IFN)-γ and tumor necrosis factor (TNF)-α [13]. These studies suggest that cell-mediated immune responses are important for protection against Y. pestis.

Comparative and Functional Genomics
The ability of Yersinia species to infect and replicate within a host is primarily due to the bacterial expression and implementation of the T3SS [14]. T3SS is comprised of a molecular syringe-like complex that injects effector molecules into the target host cell enabling the bacteria to inhibit innate and acquired immune functions as well as to induce apoptosis. There are specific Yersinia outer membrane proteins (Yops) that have been studied extensively and characterized as inhibitors of specific biological processes that promote the survival of Yersinia species within the host. Specifically, the proteins YopE, -H, -J, -M, -O, -P, and -T disrupt cytoskeletal dynamics, inhibit innate and acquired immune functions, and promote apoptosis [15,16].
The outer membrane of gram-negative bacteria is comprised of many different proteins that help maintain the structural integrity of the bacterial cell envelope. One particularly abundant lipoprotein, designated murein (or Braun) lipoprotein (Lpp), is associated with the outer membranes of bacteria within the family Enterobacteriaceae [17]. Earlier studies indicated that Lpp (6.3 kDa) from enteropathogenic bacteria not only synergized with lipopolysaccharide (LPS) to induce septic shock but also evoked the production of TNF-α and interleukin 6 (IL-6) in both LPS-responsive and LPS-nonresponsive mice and in mouse peritoneal exudate macrophages, suggesting an alternative signaling mechanism for Lpp [18]. In fact, a subsequent study showed that Lpp signals through Toll-like receptor-2 (TLR-2) and not TLR-4, which LPS utilizes for cell signaling [19]. Our more recent data provided evidence that Δlpp mutants of Y. pseudotuberculosis and Y. pestis KIM/D27 were attenuated in mice, an effect that could be complemented [20]. In the latter strain of Y. pestis KIM/D27, a 102-kb pigmentation locus (pgm) was deleted, resulting in the attenuation of the WT bacteria [20]. Importantly, immunization of mice with this mutant provided protection to animals against pneumonic plague invoked by intranasal inoculation of Y. pestis CO92 [20].
Most bacterial virulence genes are regulated upon entering the host. Global regulators have the ability to modulate multiple operons that belong to different metabolic pathways and are important for bacteria to adapt to new conditions. Recently, several laboratories have established an intranasal mouse model of the pneumonic plague infection and have investigated the host-pathogen interaction by pathological survey and bacterial gene expression microarrays [3]. Liu et al. [21] examined the transcriptional profile of mice infected with Y. pestis strain 201, which is avirulent in humans, and reported upregulation of host cytokines that might mimic what would be observed during human infection [21]. In our study, we investigated the transcriptional profiles of mice challenged by the intranasal route with Y. pestis CO92, a clinical isolate that is virulent in both mice and humans that would presumably better model human disease. We also examined the transcriptional profile of a Y. pestis CO92 Δlpp mutant and compared the results to mice infected with WT bacteria and found that Y. pestis CO92 Δlpp mutant infection of the lung caused upregulation of many genes encoding major proteins of the host immune system. Interestingly, we found a number of unique genes which were expressed differently in all three tissues of mice infected with the Δlpp mutant but were not altered by WT Y. pestis CO92 infection. This study provided new information on the dynamic of the liver, lung, and spleen host transcriptional responses to infection with WT Y. pestis CO92 and its Δlpp mutant.

Materials and Methods
2.1. Bacterial Strains. WT Y. pestis CO92 was obtained from the Centers for Disease Control and Prevention (CDC, Atlanta, GA) and maintained in our restricted access biosafety level-(BSL-) 2 laboratory. The construction and characterization of the strain deficient in the expression of the lpp gene were previously described in detail [20]. All bacteria were grown in Brain Heart Infusion broth (BHI, Difco, Voigt Global Distribution Inc, Lawrence, KS) at 28 • C prior to infection of mice.

Animal Studies.
Swiss-Webster female mice (Charles River Laboratories, Wilmington, MA) 5-6 weeks of age were infected intranasally with 5 LD 50 of either WT or Δlpp mutant of Y. pestis CO92 [20]. Uninfected mice were used as controls. At either 12 or 48 hours post infection (p.i.), 3 mice per group were euthanized and the lungs, livers, and spleens were harvested and homogenized in 1 mL of RNALater (Ambion/Applied Biosystems, Austin, TX) using 50 mL tissue homogenizers (Kendell, Mansfield, MA). RNA was isolated from the tissue homogenates and purified using RNAqueous (Ambion). After an overnight precipitation, the RNA was resuspended in 20 μL of diethylpyrocarbonate-(DEPC-) treated water and hybridized to Affymetrix GeneChip Mouse Genome 430 2.0 arrays, performed by the Molecular Genomics Core at UTMB Galveston, Texas, per manufacture protocols. The arrays had 45,000 probe sets representing more than 39,000 transcripts derived from ∼34,000 well-substantiated mouse genes. The experiments were performed in triplicate (biological replicates), generating a total of 45 arrays.

Normalization and Initial Characterization of Arrays.
Data were Robust Multichip Average (RMA) normalized and log transformed using GeneSifter software (VizX Labs, Seattle, WA). Based on regression analysis of experimental replicates, there was an acceptable level of variation between each array (R 2 = 0.96 ± 0.03). Raw and processed data (a total of 45 arrays) were deposited in the Gene Expression Omnibus (GEO) online (http://www.ncbi.nlm.nih.gov/geo) database (Accession GSE18293).

Analysis Methods.
After normalization, data were considered separately for the three tissue types of mice. Data were further separated, based on time post infection and gene expression alterations that occurred in response to the WT Y. pestis CO92 and its Δlpp mutant. This resulted in four analyses per tissue type (uninfected versus WT-infected animals and WT-infected versus mutant-infected animals at 12 hours and 48 hours p.i.). ANOVA was performed for each comparison, and only genes with P values of ≤ .05 were considered for further analyses. Subsequent filtering was performed dependent upon group comparison types, as detailed below. Hierarchical clustering was employed on normalized and log transformed signals using GenSpring GX 10.0 (Agilent Technologies, Santa Clara, CA).

Data Analysis of Uninfected Controls versus WT-Infected
Animals. For each time point, normalized signal values were averaged and pairwise comparisons were performed using GeneSifter. Only alterations (control versus WT-infected) of at least 2.0-fold were considered for further studies. Student's t-test with Benjamini and Hoshberg correction was also performed using GeneSifter. However, only the P value without correction was used to filter data (P ≤ .05), because natural biological variation was greater for some tissues than for others. All possible individual pairwise comparisons were performed using Spotfire DecisionSite 9.0 software (Spotfire, Inc., Sommerville, MA). An alteration of at least 1.5-fold was expected for each of the 9 possible comparisons between controls versus WT-infected samples (for each time point). Any alteration observed between uninfected and WTinfected animals was expected to be at least 50% greater than the fold change calculated for each uninfected control (C1 versus C2, C2 versus C3, and C1 versus C3).

Data Analysis of WT versus Δlpp Mutant-Infected Animals.
For each time point, normalized signal values were averaged and pairwise comparisons were performed using GeneSifter. Only alterations (WT-versus Δlpp-infected) of at least 1.5-fold were considered for further analysis. Student's t test was performed using GeneSifter, with the expectation of a P ≤ .05. All possible individual pairwise comparisons were performed using Spotfire DecisionSite 9.0 software (Spotfire, Inc.). An alteration of at least 1.5-fold was expected for each of the 9 possible comparisons between WT-infected and Δlpp-infected samples for each time point. Any alterations observed between WT-infected and Δlpp mutant-challenged animals were expected to be at least 50% greater than the fold change calculated for each uninfected control (C1 versus C2, C2 versus C3, and C1 versus C3). An alteration of at least 2.0-fold (on average) was expected between either uninfected versus WT-infected or uninfected versus Δlpp mutant-infected samples. This step was intended to eliminate any presumably spurious alterations observed between WTinfected and Δlpp mutant-challenged animals that was not normally affected by infection or altered in response to the Δlpp mutant as compared to healthy animals.

General Gene Expression Changes in All Tissues from WT Y. pestis CO92-and Its Δlpp Mutant-Infected Mice.
The host transcriptional responses to infection with WT Y. pestis and its Δlpp mutant in an inhalational mouse model of pneumonic plague were studied. Mice were infected for 12 or 48 hours, and RNA was isolated from livers, lungs, and spleens for processing and hybridization to Affymetrix GeneChip Mouse Genome 430 2.0 arrays.
Uninfected animals served as controls, and experiments were performed in triplicate, which generated a total of 45 arrays (see summary in Supplementary Table 1 available online at doi:10.1155/2009/914762). A stringent data analysis method was employed, including analysis of replicate samples and subsequent elimination of naturally variable transcripts, to increase the reliability of the results and to greatly minimize false positives. Each experimental group was separately compared to appropriate uninfected control tissues, and the overall results of the analysis are shown in Table 1. In our initial GeneChip experiments, we also used times points of 24, 36, and 60 hours (data not shown); however, optimal transcriptional profiling changes were observed at 12 and 48 hours p.i.. Consequently, we focused only on these two time points.
Three general trends were apparent regarding the overall response of the host to Y. pestis infection: (1) host transcriptional responses increased dramatically between 12 hours and 48 hours p.i., (2) the liver transcriptome was more profoundly perturbed, compared to spleen or lung tissues, in WT Y. pestis-infected mice, and (3) there was a drastic difference in host transcriptional responses of mice infected with the Δlpp mutant, dependent on time course and tissue. For example, many genes were differentially expressed in Δlpp mutant-infected lungs (109 transcripts) and livers (256 transcripts) at 48 hours p.i., compared to WT Y. pestis-infected animals, whereas only modest differences were observed at the earlier time point (12 hours p.i.) in the spleens (25 transcripts) and livers (11 transcripts) of mice infected with the Δlpp mutant compared to WT bacteria (Table 1).
Hierarchical clustering of normalized and log transformed signal values for genes that were differentially expressed between the various tissue and infection types likewise indicated that the majority of gene expression differences between uninfected animals and mice infected with WT Y. pestis CO92 occurred at 48 hours post infection (Figure 2(a)). Moreover, there were some transcripts that represented a "generalized" host response at this latter time point, which is demonstrated in Figure 2(a) by the clustering of all 9 samples representing liver, lung, and spleen replicate samples from mice infected with the WT bacteria (bright red, right-hand side of Figure 2(a)). These higher expressed transcripts were further separated based on tissue type, as expected (Figure 2(a)), indicating that there was a high correlation between replicate samples for these differentially expressed transcripts. A similar concordance was obtained when the signal intensity values from Y. pestis CO92 Δlpp mutant-infected tissues were clustered. There were distinct transcriptional changes that characterized the livers ( Figure  2    A, and genes encoding guanylate binding proteins were observed to increase across the liver, lung, and spleen of WT infected mice at 12 or 48 hours p.i. (Supplementary Table 2). There were 30 genes whose expression was increased in all three tissues (liver, lung, and spleen) in response to WT Y. pestis infection, compared to uninfected control animals at 48 hours p.i. (Table 2). These genes also represented mainly immune and stress response functions. Importantly, no genes were altered across all tissue types at 12 hours p.i.
with WT Y. pestis CO92. Although there were substantial differences in individual gene expression changes observed in the three different tissue types (liver, lung, and spleen) in response to WT Y. pestis infection (Figure 2(a)), overall functional processes, as determined using Ingenuity pathway analysis software, were remarkably similar ( Figure 3).  The three replicate samples representing the three experimental conditions (uninfected animals and mice infected with WT Y. pestis CO92 or its Δlpp mutant) are labeled as control, WT, and mutant, respectively. Note that mice infected for 48 hours with WT Y. pestis CO92 exhibited a collection of altered gene expressions that were common to all three tissues examined (panel a). Most notably, the three replicates representing tissue (liver, lung, or spleen) infected with the mutant clustered together, and mutant-infected samples clustered apart from uninfected controls and mice infected with WT Y. pestis for each tissue examined (panels b-d). The vertical dendrograms indicate relative similarity between samples (columns), while the horizontal dendrograms indicate clusters for genes (rows). Bright red indicates the highest normalized intensity value, bright blue the lowest, and yellow represents median values.   Base. The red color indicates transcriptional upregulation, based on microarray results. Those signaling molecules which were not colored (e.g., NF-κB complex) were not transcriptionally altered; however, microarray data suggested they were activated non-transcriptionally.  Table 2). Many of the upregulated genes seen at 48 hours, compared to uninfected control, were due to the host INF-γ response, as evident by Ingenuity pathway analysis of upregulated genes ( Figure 3). Specifically, there were 18 transcripts (e.g., CXCL6, IL-1β, and IL1R2), in addition to IFN-γ itself, that directly participate in IFN-γ signaling and were statistically upregulated in WT Y. pestis-infected mouse spleens. The mitogen-activated protein kinases ERK1/2 and JNK, transcription factors NF-κB, CREB, and Akt, and apoptosis-associated caspases, all of which are typically regulated via nontranscriptional mechanism (e.g., phosphorylation), are integral components of IFN-γ signaling and were thus likely activated ( Figure 3).

Comparison of WT Y. pestis CO92-Infected Mice to a Previous Study Utilizing a Strain (Y. pestis Strain 201) That Is Avirulent in Humans.
Our study is the first to examine host global transcriptional responses to Y. pestis CO92 using an inhalation mouse model. However, Liu et al. [21] performed a similar study in Balb/c mice using a strain that is highly virulent in mice but not humans (Y. pestis strain 201). Because the entire gene expression data sets were not made publically available, we performed a comparison of the published results, which consisted primarily of cytokines and cytokine-related signaling molecules in order to gain insight into the potential differences and similarities in most responses to these two different strains. Despite the fact that different mice, array platforms, and analysis methods were employed, the majority of genes reported by Liu et al. as altered in response to infection with Y. pestis stain 201 were also identified as altered in our experiments (Table  3). However, there were some fundamental differences that could be contributed by the differences in virulence of the two strains. Most notably, all three tissue types (liver, lung, and spleen) responded similarly to infection with WT Y. pestis CO92 (Figure 2), particularly when major functional processes were considered (Figure 1), rather than individual gene alterations. However, some genes were altered differently depending on tissue type in mice infected with Y. pestis strain 201. Cd9, for instance, which was upregulated (2.8-fold) in our study in the livers of WT Y. pestis CO92infected mice was upregulated in the liver (2.4-fold) and downregulated (−2.6-fold) in the lung of mice infected with Y. pestis strain 201 [21]. Although Cd9 was not deemed significantly altered in our study in the lung or spleen, it was upregulated on average in both tissues (1.5-fold, and 2.1fold, resp., data not shown). Likewise, Icam2 was reported as upregulated (1.9-fold) in the liver and downregulated (−3.1fold) in the lung of mice infected with Y. pestis strain 201, whereas we found Icam2 to be upregulated (2.8-fold) in the liver (Table 3) and unaffected in the lung in response to infection with WT Y. pestis CO92 (data not shown). Most notably, IFN-γ, which we found to be upregulated in the liver, spleen, and lung of WT Y. pestis CO92-infected mice (Supplementary Tables 2 and 3) and also identified as a critical signaling pathway based on Ingenuity analysis of the entire gene expression dataset (Figure 3), was reported as downregulated in Balb/c mouse lungs in response to infection with the 201 strain [21]. While our study is the first to examine the entire host transcriptome in response to WT Y. pestis CO92, Lathem et al. previously performed a cytokine analysis of lung homogenates from C57BL/6 mice that were infected with Y. pestis CO92 via the intranasal route [3]. They found that WT Y. pestis infection induced upregulation of IL12p70, TNF, IFN-γ, MCP-1 (also called CCL2), and IL-6. As shown in Supplementary Table 2, we also detected statistically significant upregulation of IFN-γ (4.3-fold), MCP-1 (also called CCL2, 3.4-fold), and IL-6 (34.3-fold) in the lung of WT Y. pestis-infected mice. Likewise, we detected an increase in IL-12a (2.6-fold), IL12b (1.9-fold), and TNFα (1.9-fold), although these differences were not deemed statistically significant (data not shown).

Gene Expression Profiling of the Liver, Lung, and Spleen of Mice Infected for 12 Hours with the Δlpp Mutant of Y. pestis CO92. For each experimental infection with WT
Y. pestis, an experiment was also performed using the Y. pestis Δlpp mutant to determine the contribution of bacterial Lpp to host transcriptional responses. Based on a direct comparison of WT Y. pestis-infected mice and animals challenged with the Δlpp mutant, very few gene expression differences were observed at 12 hours p.i. (Table 1). In the liver of infected mice 12 hours p.i., the Δlpp mutant induced upregulation of 5 transcripts and downregulation of 6 transcripts, most of which were bacterial metabolic genes included on the array as controls (i.e., 28 probe sets representing 6 different genes, Supplementary Table 4). These alterations most likely represented differences in bacterial load at 12 hours in the livers of mice infected with the Δlpp mutant compared to WT Y. pestis, which is consistent with histopathological analysis of liver tissue [20]. On the host side, 4 genes were upregulated in response to infection with the Δlpp mutant but not WT Y. pestis, including apoptosis inhibitor 5 (Api5), which suggests that Lpp might influence the host apoptotic response to infection (Table 4) and is consistent with our recently published data [22]. No genes were detected as differentially expressed in the lungs of mice infected for 12 hours with the Δlpp mutant, compared to WT Y. pestis (Table 1). In contrast, transcriptional differences in the spleen between WT Y. pestis-infected and Δlpp-challenged mice were limited to the earlier time point (i.e., differences were observed at 12 hours only). Although these alterations were few (only 25 genes were differential between WT Y. pestis-infected and Δlpp-challenged mice), the differences were profound. For instance, 18 probe sets (Affymetrix repeated transcripts) representing 16 different genes that were upregulated in WTinfected mouse spleens were not recapitulated by infection with the Δlpp mutant (Supplementary Table 5). Most of these genes are involved in the regulation of cell growth, including stress-associated cell proliferation (e.g., cyclin D3 and ERBB receptor feedback inhibitor 1). Only one gene was uniquely altered in the absence of Lpp (transthyretin that encodes a prealbumin carrier protein associated with acute phase response [23]), which was upregulated in Δlpp mutant-infected mice, compared to control animals. Of greater interest, immune-specific transcriptional responses (e.g., antihuman CD37 antibody, immunoglobulin kappa chain variable 28, and nemo-like kinase [Nik]) were downregulated in WT Y. pestis-infected mice and upregulated in Δlpp mutant-challenged animals, compared to uninfected controls (Table 4). Conversely, apoptosis-associated transcripts (e.g., Bak1 and BCL2L1) were downregulated in WT Y. pestis-infected mice and upregulated in Δlpp mutant-challenged mice, compared to uninfected controls (Table 4).  Table 6). However, there were also more profound differences, in which genes were altered uniquely by either WT Y. pestis-or Δlpp mutant-infected mice (Figure 1).
In mouse livers at 48 hours p.i., there were 27 genes that were specifically upregulated in Δlpp mutant-infected mice but not in animals challenged with WT bacteria. Induction of these genes, which included those involved in immunespecific signaling, inflammation, and the regulation of apoptosis, is therefore presumably repressed in the presence of Lpp. There were also 41 genes that were downregulated in the livers of Δlpp mutant-infected mice but not in WT Y. pestis-infected animals, compared to control animals (Supplementary Table 6). Two of these genes (i.e., complement factor H-related 1 and V-set immunoglobulin domain containing 4) ( Table 4) are involved in the regulation of host immune responses, but the majority are associated with various metabolic processes (e.g., amino acid metabolism and gluconeogenesis).
There were 109 genes that were differentially expressed in the lung between WT Y. pestis-infected mice and animals challenged with Δlpp mutant bacteria ( Table 1). Seventy of these genes were modestly upregulated (1.5-fold to 34.3-fold, mean = 3.9 ± 4.6) in response to WT Y. pestis infection but more profoundly upregulated (3-fold to 172-fold, mean = 19.8 ± 32.4) in response to infection with the Δlpp mutant (Supplementary Table 7). In contrast to what was observed at 12 hours or at 48 hours in liver tissue, the majority of differentially expressed genes in lungs were those critical for immune and stress responses, inflammation, and apoptosis. For instance, IL-6 and CXCL2 (also called Mip-2α) were upregulated in the lungs of WT Y. pestis-infected mice 34.3fold and 18.2-fold, respectively, compared to uninfected mice. In Δlpp mutant-infected mice, on the other hand, IL-6 and CXCL2 were upregulated 172-fold and 169.1fold, respectively (i.e., 5-fold and 9.3-fold larger inductions), compared to uninfected control mice.
A total of 39 genes were upregulated exclusively in the lungs of mutant-infected mice after 48 hours of infection (i.e., not altered in the lungs of the WT Y. pestis group of infected mice) (Supplementary Table 7). Most of these genes were associated with apoptosis, inflammation, immune responses, and signaling pathways critical for immune cell activation, including apoptosis regulators Birc3 and Bcl2a1a, CD53, CXCL14 (also called Mip-2γ), coagulation factors III and X, IL-22, early growth response 1, leukemia inhibitory factor, and prostaglandin E synthase (Supplementary Table 7).
Based on the transcriptional profiles of liver, lung, and spleen of mice, the most profound differences between animals infected with WT Y. pestis versus the Δlpp mutant, literature searches, and known signaling pathways available in various online databases (e.g., NCBI, Biocarta, and The Protein Lounge), we created a putative Lpp-associated signaling pathway (Figure 4). For instance, we inferred that the most likely pathway for the production of the multiple cytokines that were identified as increased based on microarray results is phosphorylation and activation of NF-κB and JNK via TLR-2 and TLR-4 induced activation of mitogen-activated protein kinases (MAPKs) (Figure 4). IFN-γ signaling, which is peripherally associated with this same pathway, was also inferred to be activated in response to WT Y. pestis and would possibly complement LPS signaling via TLR-4 to explain the induction of cytokines, albeit somewhat blunted, in the absence of the lpp gene ( Figure 4). Reduction in some host responses, identified as expression alterations in Δlpp mutant-infected animals, could be partially explained by Lpp-mediated inhibition of leukemia inhibitory factor (Lif) and Dusp16 (Table 4), which downregulate activation of NF-κB and JNK, respectively. The most directly affected process, based on WT versus Δlpp-infected animals, was apoptosis, possibly via inhibition of prostaglandin E synthase (Ptges) ( Table 4) [24,25] and perturbation of relative ratios of mitochondrial factors (e.g., Bcl-2 family members) ( Figure 4). Overall, the three main signaling pathways induced by WT Y. pestis were TLR-4, TLR-2, and INF-γ signaling, which culminated in the production of multiple inflammatory cytokines, also detected as upregulated in infected mice in all three tissues examined.

Discussion
In the present study, a transcriptional ontological assessment of significantly modulated genes in the liver, lung, and spleen from WT Y. pestis CO92-infected mice revealed a number of up-and downregulated transcripts that were associated with immune mechanisms. For example, an increase in CD14 transcript was observed across liver, lung, and spleen of mice infected with WT Y. pestis CO92 at 48 hours p.i., but the gene encoding IL-10 was not upregulated. CD14 exists as a membrane-bound or soluble form and serves as a coreceptor with TLRs or LPS-binding protein to associate with LPS from Gram-negative bacteria [26]. During Y. enterocolitica infection, CD14 complexes with TLR-2 on macrophages and subsequently binds low calcium response antigen V (LcrV), which leads to a reduction in TNF-α and an increase in IL-10 [27]. This IL-10 induction by LcrV through binding to TLR-2/CD14 plays a key role in Y. enterocolitica immune evasion and pathogenicity [27]. However, previous studies on Y. pestis indicated that IL-10 was not produced in the lungs of mice infected intranasally, and TLR-dependent IL-10 induction by LcrV did not contribute to the virulence of Y. pestis [28]. Our results are consistent with these findings and suggest that IL-10 suppression might be an important virulence mechanism for enteropathogenic yersiniae. The majority of transcriptional alterations identified in the liver, lung, and spleen of WT Y. pestis mice were those important for host immune responses, as expected. In addition to CD14, TLR-4 and TLR-2 were upregulated p.i. (2.6-and 9.2-fold, resp., Supplementary Table 2), as were several downstream targets of these two TLR signaling pathways (Figure 4). We also identified the IFN-γ signaling pathway as a central player in the host response to WT Y. pestis infection (Figure 3). INF-γ, which was induced at 48 hours p.i. in all of the tissues, is produced by activated natural killer cells and T cells and is critical for a successful immune response to intracellular pathogens [29][30][31]. Also upregulated in WT Y. pestis-infected mice were the IFN-γ-regulated serine proteases Serpina3g and Serpina3n (Supplementary,  Table 2), which can inhibit caspase-independent death [32] and assist in the development of memory CD8 T cells [33]. Likewise, we noted WT Y. pestis-induced upregulation of suppressor of cytokine signaling 1 (socs1, 8-fold in the liver) and socs3 (4-15-fold in all three tissues), which regulate JAK-STAT signaling, and TNF-α-induced protein 3 (tnfaip3b, 2.7-15.4-fold in the liver and lung), which is essential for negative regulation of I-κB kinase/NF-κB cascade (Supplementary Table 2).
Other IFN-γ-regulated molecules that were induced in response to WT Y. pestis infection included several guanylate binding proteins (GBP2, 4, 7, and 6/10), which were upregulated in all of the tissues collected from WTinfected mice. This IFN-γ-induced family of proteins has been poorly characterized, but they have been shown to regulate endothelial cell proliferation during infection, possibly by slowing cell-to-cell spreading [34]. IGTP (IRGM3) and TGTP (IRGB6), members of the p47 GTPases family, were also increased in all of the tissues of WT bacteriainfected mice (Supplementary Table 2). These molecules are similar to the GBPs but do not require de novo synthesis of transcription factors [35]. Functionally, they have been shown to localize to infected vacuoles in a Toxoplasma gondii infection [36], which is followed by vesicle formation, disintegration of the vacuole, and the subsequent demise of the parasites [37]. Consequently, these guanylate binding proteins could perform a similar function during Y. pestis intracellular infection.
Consistent with a strong host inflammatory response to infection, multiple cytokines and chemokines were upregulated in WT Y. pestis-infected animals in all three tissues examined (Supplementary Tables 2 and 3). For instance, CXCL10 and CCL2, which were profoundly upregulated (6-30.5-fold) in response to WT Y. pestis infection, are chemottractants for monocytes, T cells, and dendritic cells. Likewise, neutrophils, important to the amelioration of early bacteremia, are attracted by CXCL6 [38], which was upregulated in the liver (23.1-fold), lung (6.4-17.5-fold), and spleen (14.6-fold) post infection (Supplementary Tables 2  and 3). Induction of some of these inflammatory chemokines (e.g., CCL3) would specifically attract monocytes, which may benefit Yersinia by providing a safe haven for replication [38,39]. The compendium of host responses identified in this study supports a strong host inflammatory response that culminates in the activation of immune effectors downstream of TLR-2 and TLR-4 and subsequent amplification of the inflammatory responses via production of IFN-γ.
We noted an upregulation of Lipocalin 2 (Lcn2) from 6.5-fold at 12 hours to 67.3-fold at 48 hours and downregulation of the HFE2 gene (8.1-fold) in the livers of WT-infected mice at 48 hours p.i. (Supplementary Table 2). Both Lcn2 and HFE2 are associated with iron regulation, and mutation in the HFE2 gene is causative for hematochormasis, which is characterized by iron overload [40]. The increase in Lcn2 by WT bacteria in the liver might check bacterial growth by binding to siderophores and could be a mechanism of mediating innate immune response. No change in its level, as observed in the Δlpp mutant, would cause normal bacterial growth in the liver. Based on our recent results (19), the Δlpp mutant grew normally in liver but not in the spleen or blood.
In conjunction with our assessment of host transcriptional responses in WT Y. pestis-infected mice, we also investigated the effects of an Δlpp mutant on gene expression. Recently, we demonstrated that in Y. pestis, deletion of the lpp gene from the pgm-locus KIM/D27 background strain further attenuated its virulence. However, minimal differences were noted in pathogenicity between the WTand the Δlpp-mutant strain of CO92 in a pneumonic plague mouse model, probably because Y. pestis CO92 strain is highly virulent and deletion of one gene causes only increases in mean time to death [20,41]. Interestingly, when groups of mice infected with either the WT CO92 or its Δlpp mutant were given a subinhibitory dose of levofloxacin, we observed a significantly higher survival rate, less severe histopathological changes, and reduced cytokine/chemokine levels in the Δlpp mutant-infected group compared to WTinfected mice [20]. These data indicated that Lpp contributed to virulence of Y. pestis CO92 and was dependent on bacterial load. We used an intranasal mouse model of infection to study host gene expression alterations in the liver, lung, and spleen at 12 hours and 48 hours p.i. that demonstrates the distinctions of virulence and pathogenic mechanism(s) between WT and Δlpp mutant strains of Y. pestis CO92 in a pneumonic plague model.
Our first observation of mice infected with the Δlppmutant strain of Y. pestis CO92, compared to WT-infected animals, was that transcriptional responses that could be due to TLR-4 activation via LPS (e.g., chemokines, JAK-STAT signaling molecules, etc.) were blunted in the absence of lpp gene expression (Supplementary Tables 4-7), which supports a synergistic role for Lpp and LPS to induce septic shock as well as the LPS-like signaling previously observed in an LPS-nonresponsive background strain of mice [18]. More interesting were transcriptional responses that were completely perturbed in the absence of lpp, such as activation in WT Y. pestis-infected animals but not in those infected with the Δlpp-mutant. These results provided much greater insight into Lpp-specific host signaling in the context of Y. pestis infection and allowed us to propose a putative signaling pathway ( Figure 4) that could explain the intertwined roles of LPS and Lpp and also how Y. pestis might survive inside host cells.
As shown in Figure 4, WT Y. pestis induces the upregulation of TLR-4, TLR-2, and CD14 independently of Lpp (i.e., these molecules were also upregulated in mice infected with the Δlpp-mutant). However, the LPS and Lpp share a common downstream signaling pathway, and even in the absence of Lpp, these intermediate inflammatory effectors (e.g., Myd88, IRAK, mitogen activated kinases, STATs, NF-κB, c-Jun and Fos, and various proinflammatory cytokines) were increased during Y. pestis infection (Supplementary Table 2 and Figure 4). Nontranscriptional events (e.g., Nik-mediated phosphorylation of IKK and subsequent degradation of IκBs and nuclear location of NF-κB) that are likely to have occurred based on the transcriptional profiles of Y. pestisinfected mice and classical signaling pathways are included for clarity. In the context of this WT model of infection, Lppspecific signaling events were also apparent. Nik, for instance, is a crucial regulatory point downstream of TLR and cytokine receptor engagement, and its upregulation in WT Y. pestis infected mice was not recapitulated when the Δlpp-mutant was used. Other mechanisms of IκB phosporylation and degradation would presumably occur in the absence of Lpp, since proinflammatory cytokines are still produced in the absence of the lpp gene.
Cell death was a major process identified as statistically overrepresented in all three tissue types, based on Ingenuity pathway analysis of altered genes (Figure 3). The balance of proapoptotic and antiapoptotic factors often determines cell fate, and apoptosis regulators can also function differently depending on cell type. Its regulatory complexity makes apoptosis-related transcriptional responses difficult to interpret. However, the absence of the lpp gene clearly perturbed the effects of the WT Y. pestis infection by subtly altering some apoptotic related transcription responses and specifically inducing or depressing others. For instance, the expressions of two genes encoding for Bcl2 family proteins (Bak1 and Bcl2l1) that function to induce apoptosis [42] were suppressed in the spleen of WT-infected mice but not in animals infected with the Δlpp-mutant (Table 4). Likewise, Hk1 was uniquely downregulated in only WTinfected mice, suggesting that its suppression requires the presence of bacterial Lpp. Whereas suppression of Bak1 and Bcl2l1 would likely be cytoprotective, cytochrome c release is inhibited by Hk1 [43], and therefore its decrease could lead to increased apoptosis [22].

Conclusions
This study provided the first comprehensive assessment of the host transcriptional profile in the lung, liver, and, spleen of mice intranasally infected with a highly virulent strain of Y. pestis CO92. We further investigated the contributions of bacterial Lpp to host transcriptional responses and presented a putative host signaling pathway that plausibly explained the synergistic actions of LPS and Lpp in the context of Y. pestis infection. Our results supported a model in which Y. pestis induced a strong inflammatory response, mediated by both LPS and Lpp, but evaded immune clearance, possibly by Lppinduced inhibition of host cell apoptosis.