A Systems Analysis Identifies a Feed-forward Inflammatory Circuit Leading to Lethal Influenza Infection

Multiplex Perturbation and Modular Transcriptome Analysis Strategy to Identify Influenza-induced, Lethality-associated Biological Processes on the Organ Level

Our top-down system approach combined an extensive matrix of viral strains, infectious doses, time points, and measurements of the host response, including transcriptional studies of infected lungs, flow cytometry, automated image analysis of infected tissue, classical virologic studies, and measurements of physiologic status. We compared infection with the H1N1 virus A/Texas/36/91 (Tx91), which is non-lethal in C57Bl/6 mice at 10⁶ infectious particles and causes transient morbidity seen as minor weight loss (Figure S1A) with infections involving sublethal and lethal doses of the highly pathogenic H1N1 virus A/Puerto Rico/8/34 (PR8), which causes death in less than a week at 600 infectious particles (Figure S1B). PR8 infection in mice is a widely used model system (Allen et al., 2009; Ichinohe et al., 2009; Tate et al., 2011a; Thomas et al., 2009) that mimics pathological features seen with the 1918 virus, (Taubenberger and Morens, 2008; Watanabe and Kawaoka, 2011), including efficient alveolar spread, fatal viral pneumonia, acute pulmonary hemorrhage, and extraordinarily strong host responses.

To obtain a broad view of the biological processes associated with lethal as opposed to non-lethal influenza infection, we first generated gene expression microarrays using RNA extracted from lung samples under 19 experimental conditions. Multiple mice were either sham infected, infected with TX91, or infected with various sublethal or lethal doses of PR8. Lung tissues were collected at various time points (Figure 1A) and whole lung RNA from individual animals used for analysis. We aimed to identify the biological processes that distinguished lethal from non-lethal influenza infection. Because transcripts with a coordinate expression pattern are likely enriched for genes that act together to drive biological processes, we used a modular approach to data analysis. Dynamic genes were defined based on statistical filtering of the microarray data for features that change in at least one of all possible pair-wise comparisons across the 19 experimental conditions (Table S1, Methods). To define putative co-acting biological processes among the 8291 dynamic features identified in this manner, we clustered the features using an optimization strategy that generated 50 modules consisting of gene sets with co-ordinate patterns of regulation when considered over the entire data set. Each module was annotated based on GO-term enrichment and manually curated (Table S2, Methods). Next, we calculated the fraction of genes within each module with differential expression between infectious condition (I_i) and sham condition (S) (Figure 1B), similar to (Chaussabel et al., 2008) and displayed these data as color-coded maps.

Broadly Shared Anti-viral but Condition-specific Innate Inflammatory Signatures Characterize the Response to Influenza Infection

GO-annotated modules (Table S2) were categorized based on changes associated with the temporal phase of the host response and the pathophysiologic status of the infected animals (Figure 2A). Cross-correlation analysis of changing gene expression at early times of infection (days 1 to 3) identified two distinct patterns associated with innate immunity, each involving several modules. One pattern was associated with viral infection per se and did not distinguish influenza types or infecting doses (Figure S2A), whereas the other was closely associated with fatal disease (Figure S2B). The first signature represented gene sets involved in anti-viral responses, such as type 1 interferons and interferon γ, and genes induced downstream of these cytokines or downstream of viral pathogen-associated molecular pattern (PAMP)-driven Irf activity (Figure S2A). The earliest changes within this signature were detected in Module B-7 and reflected increases in both regulators and targets of anti-viral and interferon signaling networks (Table S2, S3).

In contrast, the gene sets uniquely associated with lethal infection (Figure S2B) were distinctly characterized by inflammatory signaling cascades. There were changes in pathways associated with TLR4 and pro-inflammatory cytokine signaling, such as NF-kB, MAPK, and JAK-STAT cascades (Figure S2B). This signature was visible well before the onset of severe morbidity or mortality in mice given lethal doses of PR8 virus (Figure S1B). The signature did not include all gene products previously reported to be connected to pathogenic influenza infection such as mRNAs for ligands binding to CCR2, CCR5, and CXCR3 (Chakrabarti et al., 2010; Chang et al., 2011; Kash et al., 2006) (Figure S2B).

Besides the increase in inflammatory and anti-viral transcripts, transcript decreases were also observed. Decreases in modules D-1 to D-10 and E-1 to E-10 were delayed as compared to the two innate modular patterns and were not detected before day 2 p.i. with lethal doses of PR8 or after day 3 p.i. with sublethal but pathogenic doses of PR8. These genes correspond to proteins identified in a recent meta-analysis of lung transcriptome data to decrease during highly pathogenic viral infections of the respiratory system, including genes with key roles in maintaining pulmonary homeostasis (Chang et al., 2011). We also observed decreased transcripts for metabolic and tissue repair genes, suggesting that the lower respiratory tract was broadly affected in its capacity to cope with noxious substances and regenerate from injury (Table S2). The delay between the appearance of the two innate signatures (A-8 and B-7) and the modular states involving down-regulated transcripts (D-1 to D-10 and E-1 to E-10) indicated that the onset of the innate response preceded and was not dependent on this loss of pulmonary homeostasis.

Deconstruction of the Organ-level Transcriptome Analysis Using mRNA from Sorted Cells

We next focused on the earliest processes that distinguished lethal from sublethal infections (Figure 2B). The first changes within the global “fatal signature” were observed in module A-8. Transcripts encoding major components of the inflammatory signaling cascade (regulators) and mRNAs of prototype inflammatory response genes (targets) populated this module (Figure S2C, Table S2, Table S3). Based on module A-8 composition, increased activation of inflammatory signaling pathways, neutrophil recruitment, myeloid cell differentiation, production of pro-inflammatory cytokines, and endothelial cell activation were highly correlated with death of the infected animals (Figure S2C). The clear segregation of inflammatory features in lethality-associated module A-8 from the anti-viral infection-associated features in module B-7 shared in all infectious conditions suggested that the two innate pathways were activated in distinct manners, with the inflammatory gene set uniquely linked to infections with lethal viral doses.

The influx of leukocytes into infected lungs complicates the interpretation of organ level changes in transcript representation, which may result from changes in cell composition, transcriptional induction within a cell population, or both. Consistent with previous studies (Lin et al., 2008; Perrone et al., 2008), flow cytometric analyses of cells from infected lungs revealed a large increase in neutrophils, inflammatory Ly6C^high monocytes, and related myeloid cells (Ly6C^high macrophages and DCs) early after infection (Figure 3A). In accord with the data associating genes in module A-8 with lethal infection, notably those encoding pro-IL-1b and neutrophil chemokines, a robust neutrophil infiltrate was detected before 2 days p.i. in animals given lethal doses of PR8 virus. This neutrophil influx preceded the onset of a moribund state and death by several days (Figure 3A, S1B). The rapid decline in pulmonary neutrophils after 5 days during PR8 infections as compared to the Ly6C^hi mononuclear myeloid infiltrate (Figure 3B) agrees with other reports (de Bruin et al., 2012; Nandi and Behar, 2011). High numbers of inflammatory monocytes and related mononuclear cells were also observed for up to 6 days in lethal conditions. In contrast, only modest increases in Ly6C^hi mononuclear myeloid cell infiltrates were seen with TX91 infection (Figure 3B). An increased signal for the constitutively-expressed neutrophil-specific chemokine receptor CXCR2 in lungs of lethally infected animals (Figure 3C) agreed with the dynamics of the neutrophil infiltrate and exemplified changes in transcript frequencies as a consequence of changes in cellular composition.

Given these dramatic cell changes, we repeated our transcriptional analysis using RNA extracted 48h post infection from 5 different populations (Figure S3A) purified by flow sorting from individual lungs (Figure S3B). One population consisted of non-hematopoietic cells (CD45neg), 45–55% of which were pulmonary epithelial cells, the primary target cell for viral replication. The others were alveolar macrophages, lymphocytes (T cells, B cells, and NK cells), neutrophils, and inflammatory monocytes together with related mononuclear myeloid cells.

Individual samples of the same cell population from different animals showed a high correlation across the transcriptome (45,281 features) but a substantial divergence in the level of expressed genes was seen among the distinct cell types (Figure S3C). Within the “SortedCell” microarray samples, 3746 dynamic features were defined based on statistical filtering of the data from the 5 sorted cell populations across 5 experimental conditions (Table S4, Figure S3D, S3E, S3F, Methods). A summary of top-ranking annotations attributed to shared or condition-specific transcript increases for CD45neg cells and for each myeloid cell type is provided in the supplemental information (Table S5).

Anti-viral and Interferon Pathways Are Robustly Activated in Multiple Cell Types

We next performed a principal component analysis (PCA) for infection-associated module B-7 (anti-viral) features derived from the “SortedCell” sample data (Figure 3D) to determine if the increase in these transcripts in the lungs resulted from preferential transcriptional induction in specific cell types or transcript import with infiltrating cells. The major distinction between samples on the x-axis accounted for 58% of the differences detected for this module and depended on whether samples were obtained from infected vs. sham infected animals, indicating that the transcriptional response to infection per se, as opposed to sample cellular composition, accounted for the principal changes. This implies that on the organ-level, the status of module B-7 largely reflected induced transcriptional changes contributed by all cell types. The finding that these major differences in transcriptional activity were not uniquely associated with cells hosting viral replication (CD45neg cells) indicated that this anti-viral response also depended on paracrine interferon activity.

We next visualized for B-7 features the normalized mean expression levels derived from “SortedCell” samples in a quantitative spider-plot (Figure 3E). Elevated average expression of B-7 transcripts was observed for all cell types from infected as opposed to sham animals, in agreement with the PCA analysis. Unsupervised hierarchical clustering of both “SortedCell” samples and module B-7 features further suggested that phagocytic cells shared a more similar anti-viral response pattern compared to CD45neg cells and lymphocytes (Figure 3F, Table S6). Overall, the analysis of these data by three different methods (Eigenvalues [PCA], average expression levels [Spider plot representation], and unsupervised hierarchical clustering [heat map visualization]) revealed that the activation of anti-viral and interferon pathways was not dominated by the cell type hosting viral replication but shared by all cell types.

Neutrophil Accumulation in Infected Lungs as a Major Driver of the Lethality-associated Gene Signature

We likewise conducted PCA on the “SortedCell” data for the features of lethality-associated module A-8. Here amplitude in X largely depended on whether samples were obtained from neutrophils or not (X: 54%; Figure 4A, S4A), revealing that on the organ level the distinctive signature monitored in the A-8 gene-set was highly sensitive to changes in neutrophil numbers infiltrating the lung. This agreed with the high mean expression level of A-8 transcripts in neutrophils (Figure 4B), whereas anti-viral module B-7 transcripts were largely independent of the cellular composition in lungs (Figure 3D, Figure S4B). Thus, the mRNAs related to inflammatory signaling components appeared to largely originate from neutrophils (Figure S2C, Figure 4C). Elevated expression of these signaling components influences signaling output. High expression of the known target genes of these pathways was observed in microarray samples and β-actin normalized qPCR samples of neutrophils (Figure 4D and not shown), indicating the relevance of the signaling component mRNA levels.

Many A-8 features were abundantly expressed in neutrophils from sham-infected animals (Figure 4E, Table S7). Therefore, both import of transcripts by neutrophils (Figure 4E subgroup 1) and additional transcriptional induction in these cells (Figure 4E subgroup 2) contributed to signals in the key lethality-associated module A-8. Importantly, transcriptional induction of gene products co-clustering in subgroup 2 was most prominent in neutrophils from lethal infections and suggested that enhanced activation of infiltrating neutrophils distinguished lethal from sublethal infections.

Constitutive pro-IL-1b expression in neutrophils was detected on both the mRNA and protein level (Figure 4F, 4G), with flow cytometry revealing high pro-IL-1b protein levels in all lung-derived neutrophils. Constitutive pro-IL-1b expression was readily detected in mature bone marrow neutrophils (Ly6G^high, c-kit⁻,CD11b^high) but absent in immature neutrophils (Ly6G^high, c-kit⁺,CD11b^med) (Figure S4C) in both germ-free and non-germ-free animals (Figure S4D). In contrast, pro-IL-1b protein was largely absent in Ly6C^high monocytes of the bone marrow. Pro-IL-1b protein was also expressed more highly in neutrophils upon in vitro stimulation with ultrapure LPS compared to inflammatory monocytes (Figure S4E), challenging the view that IL-1b activity derives predominantly from myelomonocytic cells and macrophages (Goldbach-Mansky and Kastner, 2009). The constitutive neutrophil pro-IL-1b expression we observed was unexpected, as pro-IL-1b expression is generally though to be controlled by transcriptional induction dependent on pro-inflammatory signaling.

Intriguingly, neutrophil pro-IL-1b expression was associated with expression of transcripts encoding major negative regulators of inflammatory signaling cascades (i.e. A20, NFKBIA, and ZFP36), all previously reported to act in transcriptionally-controlled, NFkB-driven negative feedback loops to control inflammatory responses in PAMP or cytokine-activated mononuclear myeloid cells (Boone et al., 2004; Hayden and Ghosh, 2008; Qiu et al., 2012; Shembade et al., 2010; Sun et al., 1993; Turer et al., 2008; Werner et al., 2008; Wertz et al., 2004). Such heightened expression of key negative regulators in neutrophils suggested that transcription-dependent inflammatory output from this cell type might require a stronger input signal as compared to mononuclear myeloid cells. In support of this hypothesis, induction of TNFa protein was readily detected in inflammatory monocytes at 10ng/ml LPS, whereas neutrophils required 100ng/ml (Figure S4F). A similar difference in the response initiation threshold between inflammatory monocytes and neutrophils was observed in germ-free or IL-1R1−/− animals (Figure S4G), implying a pre-programmed elevated response initiation threshold in neutrophils.

In addition to very high levels of pro-IL-1b mRNA (Figure 4E, among subgroup 1 features), neutrophils also expressed Tnfa and Il1a transcripts when activated (Figure 4E, among subgroup 2 features), but not high levels of Il6 transcripts (Figure 4E, among subgroup 3 features). IL-6 was instead produced by resident lung cells, both CD45neg non-hematopoietic cells and CD45pos alveolar macrophages, as well as by infiltrating inflammatory monocytes and their differentiated progeny. qPCR assays (Figure 4F) confirmed the microarray data. Thus, a full type 1 inflammatory response in the lung requires the conjoint action of diverse cells types producing different cytokines.

The constitutive expression level of the anti-viral B-7 features was substantially lower than the steady state level of inflammatory A-8 features in the whole lung samples (Figure S4H). The differential basal expression level and the accumulation of prototype TLR4-responsive immediate early response genes in A-8 agree with recent findings suggesting that inflammatory responses often reflect quantitative modulation of basal levels of gene expression rather than qualitative transitions from “off” to “on”” (Escoubet-Lozach et al., 2011). In terms of the inflammatory features in A-8, only very large increases in transcript frequency may thus be detectable because of the very high baseline expression, whereas the low basal level of anti-viral or interferon-driven transcripts in B-7 permits detection of even modest transcriptional increases.

Two Independent Chemokine Feed-Forward Circuits Regulate the Magnitude of Myeloid Cell Infiltration

The disproportionate contribution of neutrophils to the early elevated inflammatory response, and the early neutrophil infiltrate characteristic of lethal but not sublethal PR8 infections prompted us to explore how the infiltrate was controlled. Chemokines are responsible for the accumulation of high neutrophil numbers in inflamed lungs of influenza A infected mice. To identify the origin of chemokines involved in neutrophil and monocyte recruitment, we performed a hierarchical clustering on all features representing chemokines or chemokine receptors in the “SortedCell” microarray samples (Figure 5A). This revealed that most neutrophil-attractive chemokines are produced at the highest levels by neutrophils themselves (Figure 5B), whereas monocyte-attractive chemokines are predominantly produced by monocytes (Figure 5A, S5B). Neutrophils from sham-infected animals expressed high levels of transcripts for a single neutrophil chemokine CXCL2 (Figure S5A). Low amounts of mRNA for the monocyte chemokine CCL2 were found in steady-state inflammatory monocytes without detectable CCL2 protein (Figure S5B, S5C). Strong induction of additional chemokine transcripts was seen with cells obtained from infected lungs, again mainly in a cell-type reflexive manner (Figure S5A, S5B). In infected mice a high level of CCL2 mRNA expression in inflammatory monocytes was reflected in CCL2 protein readily detectable by flow cytometry in only these cells (Figure S5C).

If neutrophils and monocytes amplify their own recruitment in a feed-forward manner, one would expect this to be evidenced in situ by focal leukocyte accumulations. We therefore stained sections from lungs collected 24 or 36h p.i. for virus proteins and leukocytes (Figure 5C, 5D). Regardless of widespread infection in the lower respiratory tract, accumulations of leukocytes appeared to be tightly clustered and restricted to select localizations (Figure 5D). Lethal infections were characterized by very early dense neutrophilic foci (Figure S5D). Consistent with the predominant interstitial localization of these infiltrates, neutrophil depletion or CCR2 deficiency did not result in major changes of bronchoalveolar lavage (BAL) cytokine levels (Figure S5E; (Teijaro et al., 2011). This suggests that early activation of infiltrating myeloid cells and subsequent release of chemokines (and other inflammatory mediators) by these cells was largely confined to the pulmonary interstitial tissue. Indeed, TNFa and CCL2 were detected in association with both the interstitial, peri- and intravascular infiltrates, and with cells in the lining of the broncho-alveolar space, with the latter cells presumably producing the cytokines detected in the BAL fluid (Figure 5E, 5F, S5F).

Overall, these data provide strong evidence for the operation of a feed-forward circuit in which recruited myeloid cells, especially neutrophils but also inflammatory monocytes, are triggered in the infected site to immediately release pre-formed protein (CXCL2) in the case of neutrophils and to produce even more attractants that amplify the interstitial cellular inflammatory process.

Poorly Controlled Infectious Spread During the Innate Phase Sets the Stage for Lethality-associated Neutrophil Activation

Innate inflammatory responses are initiated by exposure to pathogen or damage associated conserved molecular cues and by pro-inflammatory cytokines. To further understand how highly pathogenic PR8 virus promotes the lethality-associated early recruitment and activation of a large neutrophil population, we investigated whether virus abundance itself or associated tissue damage correlated with neutrophil recruitment and physiological dysregulation. Virus titers from whole lung homogenates were compared to infectious spread in situ (Figure 6A), quantified using newly developed automated image analysis software (Figure 6B and Methods).

Conventional TCID50 lung virus titers did not discriminate between pathogenic (PR8) and non-pathogenic (TX91) infections (Figure 6A), in line with recent observations (Qi et al., 2011). Image analysis of infectious spread, however, very clearly distinguished between sublethal and lethal infections (Figure 6A). These imaging data were further supported by β-actin normalized qPCR assays on pulmonary CD45neg cells for viral NS-1 mRNA, which is expressed in infected cells but not in virions (Figure 6C). Viral NS-1 mRNA in CD45neg cells (Figure 6C) and the proportion of infected lung tissue (Figure 6D) both measure the infectious spread in the lung, in the contrast to measurements of infectious particles using TCID50 assays. These analyses reveal that at the same infectious particle load PR8 has spread much more through the tissue as compared with Tx91, indicating that PR8 has a higher capacity to infect host cells but a lower capacity to generate new infectious particles per infected cell.

We extended our analysis of the relationship between viral PAMPs and fatal disease course by collecting, staining, scanning, and analyzing lung sections collected from approx. 250 animals, every 6h up to 36h p.i. and at broader time intervals thereafter (Figure 6B). For infections with 10 and 100LD50PR8, the in situ dynamics showed an initial peak in the airways around 24h p.i., similar to the timing after infections with 10⁶ PFU TX91 but at higher levels (Figure 6D, 6E). The time course of airway infection ratios further indicated that infections with 10 and 100LD50 PR8 are more difficult to contain during the innate phase as compared to TX91. Similar to lethal doses, infection loads at sublethal doses of PR8 did not substantially decrease by 72h p.i. (Figure 6E) despite lack of the early initial infectious peak at 24h p.i.

These data could reflect lethal doses overwhelming innate host defense mechanisms, whereas due to the low infection load during the initial 36h p.i., sublethal doses do not produce enough immunogenic signals to show the decline typical of TX91. However, poorly contained infection in the course of co-infections with 0.6LD50PR8 and 10⁶ PFU TX91 indicate that irrespective of an early infectious peak and associated immune stimulus, infections with inocula as low as 40 PFU PR8 are difficult to contain during the innate phase (Figure S6A). The dose-independent limitation in containment of PR8 thus pointed to a strain inherent property related to either increased infectivity and/or efficient evasion of innate defense mechanisms.

Together with the preceding data on chemokine expression, these findings on viral spread indicate that continuous high infectious loads with PR8 during the innate phase are associated with a feed-forward circuit in which neutrophils are triggered in the infected site to release or/and produce more attractants that then guide more neutrophils to where previously arriving neutrophils have been effectively activated. The poorly constrained spread of the highly pathogenic PR8 virus provides a clue to how neutrophils, with a higher threshold for activation but the potential for greater toxicity if triggered, could be primarily associated with the lethal signature.

Viral NS-1 mRNA expressed in infected host cells correlated with the infectious spread in situ (Figure 6C, 6D) but not with the infectious particle load (Figure 6A). We compared NS-1 mRNA levels during PR8 versus TX91 infections (Figure S6B) in individual cell types purified from lungs of infected animals to screen for PAMP interactions. This revealed markedly greater amounts of NS-1 mRNA in infiltrating myeloid cells purified from animals infected with lethal as compared to sublethal doses of PR8. These data were consistent with the idea that other, immunogenic viral RNA species (potent PAMPs) were also available to myeloid cells and correlated with enhanced neutrophil activation and high levels of early infiltration. The discrepancy between NS-1mRNA levels measured in alveolar macrophages vs. neutrophils most likely relates to the different compartments to which the two cell types are localized and accordingly, to differences in exposure to pathogen-derived material (Figure 6C).

We then measured viral proteins by flow cytometry to obtain single cell level statistics (Figure S6C). Histograms showed a homogenous staining shift of the entire population for influenza A proteins in neutrophils and alveolar macrophages at lethal but not sublethal doses, revealing that abundant interactions between pathogen-derived material and the majority of neutrophils distinguished lethal from sublethal conditions (Figure 6C). Signals for viral proteins in immune cells were only obtained by intracellular staining and contrasted with CD45neg cells where viral proteins were readily detected by cell surface staining (Figure S6D), indicating that macrophages and neutrophils did not contribute substantially to virus production. Phagocytosis of infected dead cells and debris (Hashimoto et al., 2007) containing a wealth of viral proteins (Figure S6E) and viral RNA most likely accounts for the pathogen-derived material within neutrophils and macrophages (Figure 6C, S6B, S6C). Together with observations showing that detection of influenza virus-associated PAMPs by endosomal TLR7 triggers activation of neutrophils (Wang et al., 2008), these findings support the hypothesis that poor innate immune containment of PR8 virus contributes to infectious dose-dependent triggering of invading neutrophils through signals derived from excess phagocytosis-associated PAMP encounter and possibly also DAMPs from damaged cells.

Deleterious Neutrophil Responses Promote Fatal Outcomes at Low Lethal Doses

Lung-recruited, highly pro-inflammatory neutrophils are a major correlate of lethal infection and when activated by viral PAMPS, readily release pre-formed chemokines while making additional chemokines that bring more neutrophils into the infected tissue, a sequence of events driven most strongly by pathogenic virus PR8. The question still remained, however, whether this potent inflammatory response was a consequence or a cause of lethal disease. Previous data clearly showed that total ablation of neutrophils promoted mortality among infected animals, but the feed-forward nature of the recruitment and activation of these cells as revealed in our data suggested that attenuation of this explosive process might provide protection against lethal effects caused by excess inflammatory responses while preserving essential innate host defense activities. We tested whether this excessively amplified innate inflammatory response was causal in PR8 lethality by reducing pulmonary neutrophil numbers in animals infected with low lethal doses of PR8 (3.5LD50 PR8; Figure S7A, B) using limiting amounts of a neutrophil-depleting antibody (clone 1A8). Infected animals treated with two doses of 5ug or 50ug of 1A8 (Figure 7A, 7B, S7E), or a single dose of 50ug (Figure 7C, S7E) showed increased survival as compared to animals treated with the isotype control reagent (clone 2A3). A single injection of 5ug (Figure 7C, S7E) did not improve or prolong survival. This contrasted with two doses of 5ug, which produce a more sustained partial depletion and yielded only 2–8% neutrophils among lung leukocytes at 120h p.i. (Figure S7C). Depletion treatments with doses of 100ug or above (Figure 7A) resulted in premature death, consistent with a recent report (Tate et al., 2011b). The narrow therapeutic window of neutrophil depletion regimens raised the possibility that high dose regimens depleted other cell types with non-redundant beneficial functions and cells expressing low levels of the depletion epitope Ly6G were indeed readily detected in the CD11b^high fraction (Figure S7D).

Beneficial effects through neutrophil depletion could arise through a reduction in immunopathologic effects or from unexpectedly lower infection loads and a reduction of direct virus-related epithelial damage. We therefore compared infectious spread in animals receiving a single dose 50ug depletion regimen or isotype control treatment. No statistically significant differences were observed at 96h and 144h p.i. in terms of airway or alveolar infection ratios (Figure 7D). Thus, low dose neutrophil depletion regimens increased survival through reduction of the damaging effects of neutrophil activation and not by limiting direct viral epithelial cytopathicity.

To further test the role of myeloid cell recruitment and activation as a causative factor in fatal influenza infection, we took advantage of observations showing that myeloid cell-intrinsic Hif-1a activity is essential for early myeloid infiltrate-dependent inflammation (Cramer et al., 2003; Jacobs et al., 2010). Lys2Cre^+/wtHif-1a^fl/wt and wt animals were infected with a low lethal dose of 3.5LD50 PR8 (Figure 7E, S7E). In accord with our model and the results of the antibody depletion studies, animals with reduced levels of myeloid cell Hif-1a indeed showed prolonged survival as compared to wt animals. Surprisingly, myeloid Hif-1a reduction in Lys2Cre^+/wtHif-1a^fl/wt mice did not have major effects on the inflammatory myelomonocytic infiltrate while still resulting in a dramatic reduction of neutrophilic accumulations (Figure 7F).

Animals, Infections, and Lung Harvest

5 to 8-week-old male C57BL/6, CCR2 deficient, or Lys2Cre^+/−Hif1a^fl/wt (Jackson Laboratories) mice were intranasally infected with different doses of H1N1 influenza A viruses. Bone marrow cells of wt and IL1R1−/− (Glaccum et al., 1997) (Jackson Laboratories), or SFP and germ-free mice (kind gifts of Y. Belkaid, NIAID, NIH) were used for detection of intracellular cytokines. Further information is detailed in Supplemental Material. All animal procedures were performed under IACUC guidelines and approved NIAID, NIH Animal Care and Use Committee study protocols.

Antibodies, and Viruses

Antibodies are listed in the corresponding table in Extended Experimental Procedures. Influenza H1N1 A/PR/8/34 and A/Tx/91 seeds (a kind gift from J. R. Bennink, NIAID, NIH) were propagated in low-passage MDCK cultures.

RNA Isolation, Microarrays, qPCR

Methods for RNA preparation, amplification, hybridization and use in qPCR assays are detailed in Supplemental Material.

Bioinformatics

Detailed information on microarray analysis is provided in Extended Experimental Procedures.

Virus Detection and Quantification

Viral titers from lung homogenates, measurement of NS-1 viral RNA, staining for viral proteins, and automated image analysis are described in detail in Supplemental Material.