|Home | About | Journals | Submit | Contact Us | Français|
Mycobacterium tuberculosis (MTB) expresses a set of genes known as the dormancy regulon in vivo. These genes are expressed in vitro in response to nitric oxide (NO) or hypoxia, conditions used to model MTB persistance in latent infection. Although NO, a macrophage product that inhibits respiration, and hypoxia are likely triggers in vivo, additional cues could activate the dormancy regulon during infection. Here, we show that MTB infection stimulates expression of heme oxygenase (HO-1) by macrophages and that the gaseous product of this enzyme, carbon monoxide (CO), activates expression of the dormancy regulon. Deletion of macrophage HO-1 reduced expression of the dormancy regulon. Furthermore, we show that the MTB DosS/DosT/DosR two-component sensory relay system is required for the response to CO. Together, these findings demonstrate that MTB senses CO during macrophage infection. CO may represent a general cue used by pathogens to sense and adapt to the host environment.
Within pulmonary macrophages and human granulomas, Mycobacterium tuberculosis (MTB) is exposed to an environment that is thought to include both diminished oxygen (O2) and increased nitric oxide (NO) concentrations (Gomez and McKinney, 2004). In vitro, each of these stimuli initiates a metabolic shift into a state that mimics the dormant phase of mycobacterial infection (Wayne and Sohaskey, 2001). However, granuloma from chronically infected mice appear aerobic (Tsai et al., 2006), and whether the oxygen concentration is significantly diminished within human granuloma is unknown. Likewise, while human tuberculous lesions express inducible nitric oxide synthase (NOS2) and stain for nitrotyrosine (an end-product of NOS2 activity) (Nathan, 2006), the concentration of NO within human granuloma during latency remains unclear.
Despite the uncertainty about the precise stimuli that stimulate MTB to enter a dormant state during human infection, recent work has shown convincingly that in vitro both hypoxia (Desjardin et al., 2001; Florczyk et al., 2001; Kendall et al., 2004a; Roberts et al., 2004a; Voskuil et al., 2004) and NO (Ohno et al., 2003; Purkayastha et al., 2002; Voskuil et al., 2003) induce ~50 genes known as the dormancy regulon (Voskuil et al., 2003). Many of the genes of the dormancy regulon have unknown functions, but some genes have been characterized including an electron transporter, ferredoxin (fdxA) (Ricagno et al., 2007), a heat shock protein, α-crystallin (hspX or acr) (Hu et al., 2006) and an operon involved in nitrite and nitrate reduction, narK2X (Sohaskey and Wayne, 2003).
In interferon-γ activated murine macrophages, NO is critical for inducing the MTB dormancy regulon since induction of the regulon is greatly reduced within NOS2 deficient macrophages (Schnappinger et al., 2003). In contrast, in human monocyte-derived macrophages and dendritic cells, the MTB dormancy regulon is induced during infection despite the absence of NOS2 induction (Tailleux et al., 2008), implying that an alternative signal exists within infected human phagocytes. Finally, expression of several of the dormancy regulon genes has been detected during mouse (Talaat et al., 2004; Timm et al., 2003), guinea pig (Sharma et al., 2006) and human infections (Fenhalls et al., 2002a; Fenhalls et al., 2002b; Timm et al., 2003), suggesting that the dormancy regulon is truly induced in vivo.
The dormancy regulon is activated by the DosS-DosR (also known as DevS-DevR) two-component system (Roberts et al., 2004a; Saini et al., 2004) where dosS (Rv3032c) encodes a sensor kinase and is adjacent to the gene encoding its cognate response regulator dosR (Rv3033c). Another distantly located gene called dosT (Rv2027) encodes a sensor kinase homologous to dosS that also is capable of activating the dormancy regulon (Roberts et al., 2004a). However, whether DosS/T/R-mediated induction of the dormancy regulon is vital for survival of MTB in vivo is unclear. In studies using Mycobacterium bovis BCG, a dosR mutant died after oxygen starvation-induced termination of growth in vitro (Boon and Dick, 2002). Using MTB, three studies from different labs have yielded conflicting results. In a SCID mouse model of MTB infection, a dosR mutant was hypervirulent (Parish et al., 2003). In contrast, a dosR mutant was attenuated in a guinea pig model of MTB infection (Malhotra et al., 2004). Finally, a recent study found that a dosR mutant had no virulence defect during wild type mouse infection (Rustad et al., 2008). Why these experiments differ so greatly in their results is unknown, but the different animal models used likely is one explanation.
Recent in vitro work has revealed the biochemical mechanism of NO and O2 sensing by recombinant DosS and DosT. Both DosS and DosT bind heme via their N-terminal GAF domain (Ioanoviciu et al., 2007; Kumar et al., 2007; Sardiwal et al., 2005; Sousa et al., 2007; Yukl et al., 2007) and bind NO, O2 and carbon monoxide (CO) (Ioanoviciu et al., 2007; Sousa et al., 2007). DosS and DosT are proposed to function primarily as O2 or redox sensors (Kumar et al., 2007; Sousa et al., 2007), yet both NO and CO also stimulate sensor kinase activity and have dissociation constants that are 100-fold lower than O2 (Sousa et al., 2007).
Humans and mice produce CO via the enzyme heme oxygenase (HO) (Sjostrand, 1951), which catalyzes the degradation of heme into biliverdin, iron and carbon monoxide in a reaction requiring O2 and NADPH (Maines, 2004). Of the two HO genes, HO-1 is primarily expressed within alveolar, liver and spleen macrophages, and is induced by inflammatory mediators such as lipopolysaccharide, tumor necrosis factor (TNF), interleukin-1, and oxidative stress (Slebos et al., 2003). Humans typically exhale 1-2 parts per million (ppm) of CO while individuals with asthma, bronchiectasis, cystic fibrosis, and respiratory tract infection exhale significantly more CO (up to 20 ppm) due to increased HO-1 expression (Slebos et al., 2003). Moreover, CO levels in long-term smokers are as high as 60 ppm (Wald et al., 1981). By comparison, humans exhale ~6 parts per billion (ppb) of NO, which increases to ~16 ppb in individuals with active tuberculosis (Wang et al., 1998).
To date, no studies have been performed on the CO concentration in exhaled air from individuals with tuberculosis. However, since (i) a variety of TNF-dependent inflammatory conditions result in increased HO-1 expression and CO production, (ii) MTB evokes a strong, TNF-mediated inflammatory response from macrophages and (iii) CO-dependent transcriptional adaptation has been observed in both eukaryotes and prokaryotes (Gilles-Gonzalez and Gonzalez, 2005; Roberts et al., 2004b), we hypothesized that MTB infection would induce host HO-1 and that in response to HO-1 derived CO, MTB would alter its transcriptome.
MTB infection of macrophages induces expression of a large number of host genes (Kendall et al., 2004b; Schnappinger et al., 2003; Tailleux et al., 2008; Talaat et al., 2004). In preliminary experiments, we monitored the transcriptional response of murine bone marrow derived macrophages to MTB infection by microarray. In these infections, HO-1 was induced 3.5-fold in naïve macrophages by 8 hours after infection, dropping to 2-fold by 24 hours (Fig. 1A). Western blotting with antibodies specific for HO-1 demonstrated that MTB infection induced HO-1 in naïve macrophages 24 hours after infection (Fig. 1B). To determine if HO-1 is expressed in vivo, we infected mice and assayed for HO-1 expression by immunohistochemistry (Fig. 1C). HO-1 accumulated robustly in both liver and lung granuloma, and also in resident Kupffer and alveolar macrophages respectively. Immunohistochemistry with secondary antibody alone did not demonstrate HO-1 staining (Fig. S1), and staining of uninfected lung showed minimal HO-1 expression within pulmonary epithelium (Fig. S1). To confirm the success of the infection, we stained liver sections for acid-fast bacilli (AFB) and found that infected livers displayed numerous AFB (Fig. S1). Thus, HO-1 is expressed within macrophages during an MTB infection both ex vivo and in vivo.
Given the marked induction of HO-1 in infected tissue, we reasoned that MTB might sense CO or one of the other products of the enzyme. To test this, we used microarrays to probe the transcriptional response of MTB to each of the products of HO-1 individually. In preliminary experiments with high concentrations, microarray profiling demonstrated that exposure of MTB to CO, but not to biliverdin or iron, led to transcriptional induction of the MTB dormancy regulon in a manner similar to nitric oxide (Fig. S2, S3). Biliverdin induced a very small number of genes of unknown function, while exposure to iron resulted in the regulation of ~40 genes, none of which overlapped with the CO response (data not shown).
Further microarray profiling with lower CO concentrations demonstrated that exposure of MTB to CO led to a dose dependent induction of the dormancy regulon (Fig. 2A, B). When we analyzed two canonical dormancy response genes fdxA and hspX by qPCR, we found that at high concentrations (i.e. ≥2000 ppm, dissolved CO ≥ 1.6 μM), both genes were significantly induced as determined by both microarray (Fig. 2A) and qPCR (Fig. 2B). We also found that these genes were induced at lower, physiologic concentrations of CO (i.e. 20 ppm, dissolved CO 16 nM), with a mean induction for fdxA of 2.6 (range 1.5 to 5.3, SD 1.8, n=7). As additional confirmation of the ability of MTB to respond to CO, MTB bearing a reporter construct containing the hspX promoter upstream of GFP (Purkayastha et al., 2002) was treated with CO (Fig. S4). As we observed by qPCR, 20 ppm (16 nM) CO resulted in activation of the hspX promoter and subsequent GFP production. Thus, in vitro, CO appears to specifically induce the dormancy regulon at physiologic concentrations.
Since CO can be toxic to bacteria (Lighthart, 1973; Nobre et al., 2007; Weigel and Englund, 1975), we grew MTB under aerobic conditions but in the presence of increasing concentrations of CO and analyzed growth by absorbance (Fig. S5). Both moderate concentrations (i.e. 1000 ppm, 800 nM) and high concentrations (i.e. 100,000 ppm, 80 μM) were well tolerated by MTB, although higher concentrations of CO resulted in a slightly diminished bacterial growth rate (Fig. S5). These results suggest that unlike high NO concentrations (Voskuil et al., 2003), CO is not directly toxic to MTB.
To test the hypothesis that the dosS/dosT/dosR system mediates the CO response, we tested if a strain missing dosR could respond to CO. In the ΔdosR mutant, neither NO nor CO could induce the dormancy regulon, and this effect was rescued by expression of dosR in single copy (Fig. 3A). We then tested the response of mutants lacking dosS, dosT or both to high concentrations of either CO (20,000 ppm; 16 μM) or NO (100 μM). At these concentrations, ΔdosS or ΔdosT mutants were still able to induce the dormancy regulon, but the ΔdosS ΔdosT double mutant failed to respond to CO or NO (Fig. 3B). To further characterize the response of the single ΔdosS and ΔdosT mutants, we treated individual mutants with lower concentrations of either CO or NO and found that both dosS and dosT were required for the complete induction by NO of a representative dormancy gene, fdxA (Fig. 3C). In contrast, absence of dosT markedly enhanced the induction by CO while absence of dosS diminished the induction of fdxA by CO, even at the highest CO concentration (Fig. 3D). This result demonstrates that DosS and DosT have distinct functions at physiologically relevant CO concentrations.
Having shown that MTB activates the dormancy regulon in response to CO levels comparable to those found in vivo, we next tested directly if HO-1 derived CO stimulated the pathway during an actual infection.
Because mouse macrophages activated by interferon-γ are robust producers of NO via NOS2 (Schnappinger et al., 2003), and large amounts of NO are known to trigger the dormancy regulon (Schnappinger et al., 2003), we used naïve bone marrow macrophages in these experiments. We compared mycobacterial gene expression in vitro to MTB gene expression within wild type macrophages and within macrophages deficient in HO-1 (Yet et al., 1999) or NOS2 (Laubach et al., 1995). When compared to wild type macrophages, induction of the MTB dormancy regulon was modestly reduced within HO-1-/- macrophages and severely reduced in NOS2-/- macrophages by microarray profiling (Fig. 4A). Similar results were observed when HO-1 was inhibited chemically by tin protoporphyrin (SnPP) in RAW 264.7 mouse macrophages (Fig. S6). As expected, both wild type and NOS2-/- macrophages accumulated HO-1 as determined by western blotting while no HO-1 accumulation was detected in HO-1-/- macrophages (Fig. 4C). Nitrite, an end product of NOS2 activity detectable in conditioned medium, was produced equivalently by wild type and HO-1-/- macrophages (at ~1 μM) and was undetectable in NOS2-/- macrophages (data not shown). To confirm the microarray results, we analyzed the expression of three dormancy genes within distinct operons by qPCR (hspX, fdxA and Rv2630) and found that for all three genes, HO-1 deficiency significantly reduced gene expression within macrophages relative to that found in MTB growing in vitro (Fig. 4B). This occurred despite a low level of NO production in both wild type and HO-1-/- macrophages. Conversely, there was no induction of the dormancy regulon in NOS2-/- macrophages (Fig. 4A, 4B) despite the presence of HO-1 (Fig. 4C), suggesting that low level NO production is required for induction of the dormancy regulon that can be modulated by CO. Taken together, this implies that both HO-1 derived CO and NOS2 derived NO combine in naïve MTB-infected macrophages to stimulate induction of the dormancy regulon.
Our data suggests that MTB senses the host inflammatory status at least in part by measuring and responding to the levels of CO in the macrophage. We propose that as MTB establishes its latent infection, it initiates a host immune response that includes both HO-1 and NOS2 production as well as granuloma formation, thus resulting in exposure of MTB to CO, NO and diminished O2. These three gases may then act sequentially, additively or synergistically to activate the dormancy regulon via the sensors DosS and DosT (Fig. S7). In this study, we show that mouse macrophages robustly produce HO-1 during MTB infection both ex vivo and in vivo. Whether the stimulus for inducing HO-1 by MTB is directly due to a mycobacterial product, or secondarily due to production of proinflammatory cytokines is unknown. However, since macrophages produce TNF within hours of MTB infection, it seems likely that HO-1 is induced by TNF. Some studies have shown that NOS2 activity is required for HO-1 induction (Alcaraz et al., 2001; Vicente et al., 2001), but these studies were done using cell lines rather than primary cells. We confirmed that in RAW cells (a mouse macrophage cell line), NO production is necessary for HO-1 induction by LPS (data not shown). In contrast, we found that in primary bone marrow derived macrophages, MTB infection induced HO-1 independent of NOS2 activity (Fig. 4C).
Although macrophages are likely the major source of CO production during infection, HO-1 induction is not limited to this cell type. Other cells that are associated with MTB infection that have been shown in other systems to produce HO-1 include pulmonary epithelial cells (Zhou et al., 2004), dendritic cells (Chauveau et al., 2005) and T-cells (Pae et al., 2003). In fact, in the infected mouse lung HO-1 staining was also visualized in pulmonary epithelial cells (Fig. 1C). Thus, CO could be produced during infection by infected macrophages and dendritic cells as well as bystander cells within the granuloma (such as T-cells), thus resulting in higher local concentrations. Likewise, while HO is thought to be the major source of CO in vivo (Wu and Wang, 2005), a recent study demonstrated that lipid peroxidation induces a cytochrome P450 dependent activity that produces CO (Archakov et al., 2002). Unfortunately, no studies have been performed on the CO concentration in exhaled air from individuals with tuberculosis, and the actual concentration of carbon monoxide within macrophages and granuloma is unknown. However, in olfactory neurons, the effective concentration within the cell has been determined to be 10-30 μM (Ingi et al., 1996), which would correspond to a headspace CO concentration of ~20,000 ppm (16 μM). CO concentrations of 10-30 μM may overestimate the true in vivo concentration, but even if the concentration is ten times lower inside macrophages or granuloma (i.e. 1-3 μM), that would still be in the range of ~2000 ppm (1.6 μM).
Systems for sensing carbon monoxide have been described in both prokaryotes and eukaryotes (Ascenzi et al., 2004; Roberts et al., 2004b; Ryter et al., 2004). For example, in eukaryotes the transcription factor NPAS2, implicated in regulating circadian rhythm, was shown to bind CO resulting in decreased DNA binding activity (Dioum et al., 2002). Likewise, the purple photosynthetic bacterium Rhodospirillum rubrum expresses a CO-binding transcription factor, CooA, that stimulates production of a CO oxidation system (Aono et al., 2000; Roberts et al., 2001; Youn et al., 2004). How is CO sensed? A common feature shared by CO sensors is the presence of a protein-associated heme moiety, which is not surprising given the propensity of CO to bind heme (Roberts et al., 2004b). However, there is great diversity in how the heme is protein-bound and under what conditions CO can bind heme.
Rather than use a single CO-binding transcription factor, our data shows that MTB sense CO via a two component system comprised of two sensors, DosS and DosT, and a single response regulator, DosR. In this study, we demonstrate genetically that dosR is required for CO signaling, and that either dosS or dosT can transmit the dormancy signal mediated by CO or NO at high concentrations. Kumar et al. also reported that MTB can induce hspX and fdxA by CO, but their single in vitro experiment was done using a high concentration of a chemical CO donor, CORM-3. Further, the precise amount of CO delivered by the CO donor was unclear, in contrast to the use of pure CO gas in our study. Moreover, they did not assess whether the CO sensing was dependent on dosS or dosT (Kumar et al., 2007). In contrast, we show that CO sensing occurs in vitro at physiologically relevant CO concentrations (as low as 20 ppm/16 nM of CO) and that both DosS and DosT can transmit the CO signal at high concentrations of CO. By carefully assessing the CO response of individual mutants in dosS and dosT, we also show that the sensors can differentially regulate the dormancy regulon at physiologically relevant concentrations. DosS has a 25 fold lower Kd for CO than DosT (0.036 μM versus 0.9 μM) while its Kd for NO is 4 fold higher (0.020 μM versus 0.005 μM) (Sousa et al., 2007). Interestingly, despite its higher Kd for CO, DosT undergoes more rapid autophosphorylation in the presence of CO than DosS (Sousa et al., 2007). However, neither the affinities of DosS or DosT for DosR nor the kinetics of phosphorylation and activation of DosR by active DosS or DosT are known. Thus, in wild type cells where both DosS and DosT are present, we propose that at low CO concentrations, DosT could bind DosR and maintain it in an inactive state. This initial interaction of phosphorylated DosT with DosR might attenuate the dormancy response by preventing DosS from binding to and activating DosR. Thus DosT could behave as a “low affinity” receptor for CO and an inhibitor of the CO-induced response at physiologic CO concentrations. What happens in the absence of DosT? Since the activity of DosS is maximal with CO as its ligand (Sousa et al., 2007), and since DosS is autoinduced as part of the dormancy regulon, a feedback loop could be rapidly established where unopposed binding of CO to DosS results in increased expression of both DosS and DosR.
We also observed that the response to CO is both dose and time dependent. Interestingly, the CO response differs from the response to NO in that the effect of CO is long-lasting (i.e. persists for >24h), while the NO effect is short-lived (Voskuil et al., 2003). Despite the fact that NO binding to the sensors DosS and DosT is more avid than CO binding (Sousa et al., 2007), the activation of the Dos regulon by NO is more transient, likely owing to its greater chemical reactivity (i.e. NO that dissociates from the heme of the Dos sensor can react with other molecules) or its chemical conversion to nitrate over time. Conversely, the persistent effect of CO suggests that MTB lacks the ability to eradicate or remove CO once it is bound to the heme group.
While our in vitro experiments demonstrate the feasibility of CO sensing by MTB, our data from macrophage infections demonstrates that CO sensing actually occurs inside macrophages. Using naïve mouse macrophages or a mouse macrophage cell line, we show that genetic ablation or chemical inhibition of macrophage heme oxygenase reduces CO-dependent gene induction during infection. Interestingly, despite the presence of low amounts of NO, DosS and DosT signaling was lost. Conversely, in NOS2-deficient macrophages, heme oxygenase activity alone was insufficient to trigger the dormancy response. It is unclear why this would occur, but one possibility is that lack of NOS2 activity reduces HO-1 activity, either through a decrease in available heme or by increasing the association of HO-1 with its negative regulator, caveolin-1 (Kim et al., 2004). An alternative possibility is that in vivo, the short-acting gas NO “triggers” the activation of the dosS/T/R system while the longer-acting gas CO “maintains” the activation state. In sum, it appears that both NO and CO have redundant roles in stimulating the MTB dormancy response during infection of naïve macrophages.
Thus, a plausible biochemical and physiologic mechanism is that during infection, exposure of DosS and/or DosT to CO, NO and hypoxia either sequentially or in combination stimulates histidine kinase activity and activation of DosR (Figure S7). Active DosR then converts the signal(s) into a transcriptional response. Notably, other pulmonary pathogens including Pseudomonas aeruginosa, Bacillus anthracis, Bordetella spp. and Nocardia spp. have histidine kinases with GAF domains. This suggests that the ability to integrate several gaseous signals, as illustrated by the combined activities of DosS and DosT, may be a prevalent mechanism by which pathogens adapt and respond to changes in the host immune status.
A detailed description of all procedures and protocols is available as Supplemental Data.
The wild type strain of MTB used in these studies was the Erdman strain except in the experiments defining the role of DosS, DosT and DosR. MTB H37Rv served as the wild type strain for the dos mutants described previously (Roberts et al., 2004a; Sherman et al., 2001) that were kind gifts of David Sherman (University of Washington, Seattle, WA).
Mouse bone marrow derived macrophages were infected at a multiplicity of infection (MOI) of 10, RNA harvested at various time points, amplified and hybridized to Mouse Exonic Evidence-Based Oligonucleotide (MEEBO) arrays as previously described (Stanley et al., 2007).
Macrophages were infected with MTB at an MOI =10. 24 hours after infection, macrophages were lysed in RIPA buffer and the samples boiled for 20 minutes. After transfer to nitrocellulose, HO-1 was detected with polyclonal anti-HO-1 antibodies (Assay Designs-Stressgen) at 1:5000. Loading controls were either anti-glucose-3-phosphate dehydrogenase (Chemicon International) at 1:1000 or anti-β-tubulin (Cell Signaling Technology) at 1:1000.
BALB/c mice were infected through the lateral tail vein with 106 CFU MTB in 0.1 ml PBS. Ten days after infection lung and liver were harvested, immersion fixed in formalin overnight, embedded in paraffin and sectioned at 5 μm. Sections were stained with polyclonal rabbit anti-HO antibody (Assay Designs-Stressgen) at 1:130 at 4°C. A peroxidase-based Vectastain ABC kit (Vector laboratories) was used for detection.
MTB grown in roller bottles was exposed to CO gas (Airgas) in the headspace or DETA-NO (Cayman Chemicals) in the growth media. Total RNA from mycobacterial cells was prepared as previously described (Voskuil et al., 2003). Arrays were scanned using a GenePix 4000B scanner and GenePix PRO version 6.0, and data analyzed using Acuity 4.0 (Molecular Devices). Statistically significant differences in gene expression were determined using the Statistical Analysis of Microarrays software tool (SAM) version 2.23A with a false discovery rate of 1%.
RAW 264.7 macrophages were from ATCC (Mnassas, VA). BALB/c, C57BL/6, NOS2-/- (Laubach et al., 1995) mice were from Jackson Labs. HO-1-/- (Yet et al., 1999) mice were bred from heterozygote founders kindly provided by Shaw-Fang Yet (Harvard Medical School, Boston, MA). Anupam Agarwal (Univ. of Alabama-Birmingham, Birmingham, AL) generously provided HO-1-/- femurs and tibias for preliminary macrophage experiments. Mice were housed under specific pathogen-free conditions, and mouse experiments were conducted using a University of California, San Francisco, Institutional Animal Care and Use Committee-approved protocol. Bone marrow-derived macrophages were isolated by culturing in medium containing 30% L cell supernatant for 6 days in the absence of antibiotics.
MTB RNA from inside macrophages was isolated and amplified according to a method described by Rohde et al. (Rohde et al., 2007). Tin protoporphyrin (Frontier Scientific, Logan, UT) was dissolved in 0.1 M NaOH, titrated to pH 7.4 and then added to macrophage cultures at a final concentration of 50 μM, a concentration known to inhibit HO-1 (Kampfer et al., 2001; Oh et al., 2006).
For quantitative PCR (qPCR) of MTB genes growing in vitro, RNA was prepared as described above and reverse transcribed into cDNA. qPCR used the primers shown in supplementary table 1 and were normalized to 16S RNA values. For qPCR of MTB genes from infected macrophages, aRNA was prepared as above, but dUTP was exchanged for amino-allyl dUTP. aRNA was then used as template for cDNA synthesis.
For quantitative PCR experiments, one-way analysis of variance was performed to determine overall significance, followed by the Bonferroni multiple comparison test to compare individual means using GraphPad InStat Software (Graphpad Software, Inc., San Diego, CA). For microarrays, statistically significant differences in gene expression were determined using the Statistical Analysis of Microarrays software tool (SAM) version 2.23A with a false discovery rate of 1%.
We thank D. Marciano for immunohistochemistry and for critical review of the manuscript, J. Allen and S. Batra for microarray processing, S. Stanley for mouse infection, and S. Raghavan for mouse microarrays. We thank D. Sherman, S.-F. Yet, A. Agarwal and K. McDonough for reagents. We thank S. Johnson for critical reading of this manuscript, and G. King, members of the Cox laboratory and Anita Sil laboratory for helpful discussion. This investigation was supported by an A.P. Giannini Family Foundation Fellowship to M.U.S., Sandler Program in Basic Sciences Opportunity Award to J.S.C. and by NIH grants AI63302 and AI51667. J.S.C gratefully acknowledges the support of the W. M. Keck Foundation.
Accession Numbers: All transcriptional profiles have been submitted to the GEO database at NCBI (GSE10897).