Hypoxia determines survival outcomes of bacterial infection through HIF-1α–dependent reprogramming of leukocyte metabolism

See allHide authors and affiliations

Science Immunology  10 Feb 2017:
Vol. 2, Issue 8, eaal2861
DOI: 10.1126/sciimmunol.aal2861

Hypoxic immune cell conditioning

Oxygen deficiency, or hypoxia, has been shown to alter immune cell function. However, how these hypoxia-induced immune cell changes affect the host response to bacterial infection has remained unclear. Now, Thompson et al. report that although acute hypoxia accentuated morbidity and mortality as a result of bacterial infection in mice, chronic hypoxia before infection could actually prevent these pathological responses. This hypoxic preconditioning reduced neutrophil glucose utilization, decreasing the related pathology. If these findings hold true in humans, they suggest that immune targeting could aid patients with systemic hypoxia and chronic infections such as adult respiratory distress syndrome or chronic obstructive pulmonary disease.


Hypoxia and bacterial infection frequently coexist, in both acute and chronic clinical settings, and typically result in adverse clinical outcomes. To ameliorate this morbidity, we investigated the interaction between hypoxia and the host response. In the context of acute hypoxia, both Staphylococcus aureus and Streptococcus pneumoniae infections rapidly induced progressive neutrophil-mediated morbidity and mortality, with associated hypothermia and cardiovascular compromise. Preconditioning animals through longer exposures to hypoxia, before infection, prevented these pathophysiological responses and profoundly dampened the transcriptome of circulating leukocytes. Specifically, perturbation of hypoxia-inducible factor (HIF) pathway and glycolysis genes by hypoxic preconditioning was associated with reduced leukocyte glucose utilization, resulting in systemic rescue from a global negative energy state and myocardial protection. Thus, we demonstrate that hypoxia preconditions the innate immune response and determines survival outcomes after bacterial infection through suppression of HIF-1α and neutrophil metabolism. In the context of systemic or tissue hypoxia, therapies that target the host response could improve infection-associated morbidity and mortality.


A close and bidirectional relationship between hypoxia and inflammation is well recognized. Hypoxia can induce inflammation (for example, in acute mountain sickness), and inflamed tissues are typically hypoxic, in part, due to oxygen consumption by inflammatory cells (1). Localized hypoxia is also a feature of tissues infected with a range of different pathogens (2, 3). Systemic hypoxia (hypoxemia) is a clinical characteristic of acute conditions, such as adult respiratory distress syndrome (ARDS) (4), and of chronic diseases, such as chronic obstructive pulmonary disease (COPD) (5), but how hypoxemia modifies host responses to infection is largely unknown.


Acute hypoxia induces rapidly progressive morbidity after local infection with Staphylococcus aureus

Staphylococcus aureus, an important human pathogen, is commonly found in critical care settings (6) where patients can be profoundly hypoxemic. Host responses to S. aureus infection depend critically on the presence and functionality of myeloid cells in both animals (7, 8) and humans (9). We used a model of subcutaneous S. aureus infection to investigate the effects of systemic hypoxia on outcomes of infection. Subcutaneous S. aureus (SH1000 strain) produced a macroscopic skin lesion in normoxic mice, which developed over 7 days and was well tolerated (fig. S1, A to C). However, the same bacterial challenge in the setting of acute hypoxia (10% FiO2) caused substantial and progressive sickness behavior and hypothermia (Fig. 1, A and B, and fig. S1, D and E). These responses were accompanied by bradycardia (Fig. 1C), hypotension (Fig. 1D), and impaired cardiac function, with reduced cardiac ejection fraction (Fig. 1E) and cardiac index (Fig. 1F and fig. S1F), ultimately leading to death. Hypoxia-induced hypothermic responses to subcutaneous S. aureus were also observed with heat-killed SH1000 (fig. S2A) and replicated by intraperitoneal administration of bacterial lipopolysaccharide (LPS) (fig. S2B). Additional validation of the sickness phenotype was provided by radiotelemetry (fig. S2, C to F). The phenotype was confirmed in an outbred mouse strain (fig. S2G) and independent of whether infection was during daytime or nighttime (fig. S2, H and I). A graded phenotypic response was observed over a range of oxygen tensions (21 to 10% O2) (Fig. 1, G and H). This pathophysiological response occurred in the context of equivalent myeloid cell recruitment (fig. S3, A to C) and bacterial burden (fig. S3D), with no evidence of S. aureus extravasation into the blood or seeding to kidneys, liver, or spleen.

Fig. 1 Hypoxia induces hypothermia, sickness behavior, cardiac dysfunction, and mortality in mice exposed to regional or systemic bacterial infection.

(A to H) Subcutaneous S. aureus infection. Sickness scores (A), temperatures (B), heart rate (C), and systolic blood pressures (D) of mice were recorded 12 hours after injection of S. aureus or PBS vehicle in normoxia (N; 21% O2) or hypoxia (H; 10% O2) and after 12 hours in specified oxygen tensions (G and H). Echocardiographic measurements of ejection fraction (E) and cardiac index (F) 12 hours after injection. (A) ***P = 0.000044, Normoxia versus Hypoxia (n = 10); ***P = 0.00005, Hypoxia versus Vehicle (n = 10 Hypoxia and 5 Vehicle); (B) ***P = 0.000004, Normoxia versus Hypoxia (n = 13); ***P = 0.00001, Hypoxia versus Vehicle (n = 13 Hypoxia and 5 Vehicle), one-way ANOVA with Tukey’s posttests. (C) *P = 0.0186, N versus H (n = 5); *P = 0.0151, H versus Vehicle H (n = 5 H and 3 Vehicle H), unpaired t tests. (E) ***P = 0.000753, N versus H (n = 4); (F) *P = 0.0298, N versus H (n = 4), one-way ANOVA with Tukey’s posttests. (G) ***P = 0.000003, 21% versus 10%; ***P = 0.000036, 15% versus 10%; **P = 0.00217, 12% versus 10%; (H) ***P = 0.000005, 21% versus 10%; ***P = 0.000004, 15% versus 10%; **P = 0.00109, 12% versus 10% (21% n = 9, 15% n = 5, 12% n = 5; 10% n = 9), one-way ANOVA with Tukey’s posttests. (I to Q) Intratracheal S. pneumoniae. Sickness scores (I) and temperatures (J) of mice 14 and 24 hours after intratracheal instillation of 107 CFU of S. pneumoniae D39 or vehicle (V) and housed in normoxia (N; 21% O2) or hypoxia (H; 10% O2). (I) *P = 0.0282, 14 h N versus H (n = 21 N and 15 H); (J) *P = 0.0459, 14 h N versus H (n = 21 N and 15 H); ***P = 0.000036, 24 h N versus H (n = 8), one-way ANOVA with Tukey’s posttests. (K and L) Viable bacterial counts recovered from homogenized lung (K) or whole blood (L) at 14 or 24 hours after instillation of S. pneumoniae in specified oxygen tension. (L) *P = 0.0239, 14 h N versus H (n = 21 N and 15 H), Kruskal-Wallis with Dunn’s posttest comparisons. Total BAL cell counts (M) and neutrophil (N) and macrophage (O) counts at 14 or 24 hours after instillation of S. pneumoniae or vehicle in specified oxygen tension. (M) **P = 0.0034, 14 h N versus H (n = 21 N and 15 H), unpaired t test. (N) **P = 0.01, 14 h N versus H (n = 21 N and 15 H), unpaired t test. (O) *P = 0.0172, 14 h N versus H (n = 21 N and 15 H), unpaired t test. (P) BAL supernatant IgM concentration in normoxia (N) and hypoxia (H) 14 and 24 hours after instillation. (Q) Kaplan-Meier survival curves of mice instilled with S. pneumoniae and housed in normoxia or hypoxia. P = 0.0002 N versus H (n = 10), log-rank test. Horizontal lines (A, B, and G to O) or bars (C to F and P) are means ± SEM.

In keeping with the observed morbidity, higher serum corticosterone and creatinine (fig. S4, A and B) levels were detected in hypoxia-exposed mice. However, a full systemic inflammatory response syndrome (SIRS) response was not observed, with plasma cytokines either below the limits of detection [interleukin-1β (IL-1β), tumor necrosis factor–α (TNF-α), interferon-γ (IFN-γ), IL-10, IL-4, IL-13, and IL-22] or unchanged between hypoxic and normoxic mice (fig. S4, C to E), as were levels of the matrix metalloproteinase MMP9 (fig. S4F), plasma nitrates [NO(X)] (fig. S4G), and malondialdehyde, a marker of oxidative stress (fig. S4H). The presence of nitrosylated tyrosine oxidative products in the skin was also unaffected (fig. S4, I and J). Gross organ function was preserved under hypoxia with equivalent peripheral blood leukocyte profiles (marrow) (fig. S5A) and normal serum aspartate transaminase (liver) and lipase (pancreas); circulating metabolic and stress responses [lactate, adenosine 5′-triphosphate (ATP), prostaglandin E2, and catecholamine levels] were also equivalent (fig. S5, B to H). There was no evidence of cerebrovascular leak, with undetectable levels of Evans blue dye in the brain tissue, and no evidence of lung injury, with equivalent wet-to-dry lung weight ratios, alveolar cell counts, and lung architecture (fig. S6, A to D).

Increased mortality is observed after systemic infection with Streptococcus pneumoniae in the setting of acute hypoxia

To define whether the morbidity observed with S. aureus in hypoxia was restricted to that organism or specific to the skin, where low oxygen tensions are reported in health (10, 11), we extended our experiments to a model of bacteremic pneumonia (12). In keeping with the S. aureus phenotype, challenge with intratracheal high-dose serotype 2 Streptococcus pneumoniae in the context of acute hypoxia resulted in increased sickness responses and associated hypothermia (Fig. 1, I and J). This was again independent of bacterial burden, with equivalent lung colony-forming unit (CFU) counts at 14 and 24 hours (Fig. 1K) and blood CFU counts lower at 14 hours and equivalent at 24 hours (Fig. 1L) when comparing hypoxic and normoxic animals. Reduced numbers of neutrophils and macrophages were recruited to the airways at 14 hours in hypoxia-exposed mice, but cell counts were equivalent to normoxic mice by 24 hours (Fig. 1, M to O), and no differences in immunoglobulin M (IgM) levels, a marker of lung injury, were observed (Fig. 1P). When infection was allowed to progress beyond 24 hours, approximately half of normoxic mice cleared the infection, whereas all hypoxia-exposed mice died (Fig. 1Q). Thus, at two different sites (skin and lung) and with two different pathogens (S. aureus and S. pneumoniae), combining infection with acute systemic hypoxia resulted in severe morbidity and later fatality, which occurred despite equivalent control of bacterial infection and the absence of a typical SIRS response, oxidative damage, or multiorgan failure.

Hypoxic preconditioning before infection protects against the increase in morbidity and mortality observed with acute hypoxia

These profound acute physiological consequences of combined hypoxia and infection parallel human observations in the critical care setting, where increased mortality is described in hypoxemic patients who present with or develop bacterial infections (6, 13, 14). The clinical situation for chronic hypoxia is more complex. In COPD, for example, infective exacerbations and hypoxemia are each independently associated with disease progression (15, 16), yet bacteria are frequently present in the airways of clinically stable hypoxemic patients without severe systemic compromise (17). We therefore questioned whether more prolonged hypoxia could modify the systemic response to infection. Mice preconditioned for 7 days at 10% O2 before S. aureus infection showed marked protection from the acute hypoxia-associated systemic phenotype (Fig. 2, A and B). Hypoxic preconditioning also reversed sickness and hypothermic responses and mortality observed with intratracheal administration of S. pneumoniae (Fig. 2, C to E). Hence, preconditioning confers this protection after challenge by different bacterial pathogens, with different sites of infection, and in both localized and disseminated bacterial infections.

Fig. 2 Hypoxic preconditioning confers protection from hypothermia and sickness behavior and reverses the negative energy balance observed with infection in the setting of hypoxia.

(A and B) Subcutaneous S. aureus. Mice were preconditioned in hypoxia for 7 days and then challenged with subcutaneous SH1000, and sickness scores (A) and temperatures (B) were recorded after 12 hours. (A) *P = 0.0365, Hypoxia versus Preconditioned (n = 7), unpaired t tests. (B) **P = 0.00855, Hypoxia versus Preconditioned (n = 7), unpaired t tests. (C to P) Intratracheal S. pneumoniae. Mice were instilled with PBS control (vehicle) or S. pneumoniae (Spn) and housed in normoxia (N), hypoxia (H), or hypoxia after preconditioning (Pre H). Sickness scores (C) and temperatures (D) were recorded at 14 hours, and Kaplan-Meier survival curves were performed over a 168-hour period (E). (C) **P = 0.00114, Spn N versus Spn H; **P = 0.00114, Spn H versus Spn Pre H; (D) *P = 0.00779, Spn N versus Spn H; ***P = 0.0006, Spn H versus Spn Pre H (n = 6 Spn N, 5 Spn H, and 6 Spn Pre H), one-way ANOVA with Tukey’s posttests. (E) P = 0.0005, Normoxia versus Hypoxia; P = 0.000012, Hypoxia versus Pre H (n = 10), log-rank test. (F) Respiratory exchange ratios were calculated from indirect calorimetric analysis undertaken over a 1-hour period after a 20-hour exposure to S. pneumoniae. *P = 0.0230, N versus H; *P = 0.0181, H versus Pre H (n = 4), one-way ANOVA with Tukey’s posttest. Separately, 24 hours after infection, livers were harvested and PAS-stained for glycogen (G) and inguinal (H) and intrascapular (I) fat reserves measured by weight, serum β-hydroxybutyrate levels were quantified (J), and blood glucose levels were determined relative to survival outcome (K). For accurate glucose monitoring, a subgroup of mice was fasted for the last 6 hours of experimental procedure, and paired glucose (L) and insulin (M) levels were recorded. (H) *P = 0.0429, Spn N versus Spn H; *P = 0.0467, Spn H versus Veh H; *P = 0.0208, Spn Pre H versus vehicle Pre H (n = 4); (I) *P = 0.0041, Spn N versus Spn H; *P = 0.0078, Spn H versus Spn Pre H; *P = 0.0219, vehicle N versus vehicle H (n = 4); (J) *P = 0.0165, Spn N versus Spn H; ***P = 0.000083, Spn H versus Spn Pre H (n = 4), one-way ANOVA with Tukey’s posttests. (K) ****P < 0.000001, Survived (n = 25) versus Culled (n = 24), unpaired t test. (L) *P = 0.0101, N versus H; *P = 0.0298, H versus Pre H (n = 12 N, 9 H, and 9 Pre H), unpaired t tests. (M) *P = 0.0106, N versus H (n = 12 N and 9 H), unpaired t test. In a separate group of animals, 18F-FDG was administered 5 hours (N) and 23 hours after instillation of S. pneumoniae (Spn) (O and P). Radioactivity levels were assessed with a gamma counter on brown fat (N) and hearts (P), and in vivo standardized glucose uptake values were determined by PET (O). (N) *P = 0.0177, Spn N versus Spn H; *P = 0.0185, Spn H versus Spn Pre H (n = 4 Spn N, 4 Spn H, and 3 Spn Pre H), one-way ANOVA with Tukey’s posttest. (P) *P = 0.0307, Spn N versus Spn Pre H; *P = 0.0237, Spn H versus Spn Pre H (n = 3 Spn N, 4 Spn H, and 4 Spn Pre H), one-way ANOVA with Tukey’s posttest. (Q and R) Subcutaneous S. aureus. Echocardiographic measurements of ejection fraction (Q) and cardiac index (R) were undertaken 12 hours after injection of S. aureus in normoxia (N), hypoxia (H), or hypoxia after preconditioning (Pre H). (Q) ***P = 0.000485, N versus H (n = 4 N and 6 H); *P = 0.0132, H versus Pre H (n = 6); (R) **P = 0.00133, N versus H (n = 4 N and 6 H); ***P = 0.00041, H versus Pre H (n = 6), one-way ANOVA with Tukey’s posttests. Horizontal lines (A to D, N, and P) or bars (F, H to M, Q, and R) are means ± SEM.

Hypoxic preconditioning rescues the host from a global negative energy state

At a cellular level, adaptation to hypoxia critically depends on coordinated metabolic responses that both maintain ATP production and modify energy requirements. The consequences of hypoxia and hypoxic preconditioning on metabolism were therefore explored. Indirect calorimetry revealed that animals infected in the setting of acute hypoxia preferentially switched toward carbohydrate utilization in contrast to normoxia, where fat and carbohydrates were proportionately consumed (Fig. 2F). In keeping with the suggestion of an increased reliance on glycolysis, hypoxic mice displayed a substantial loss of liver glycogen (Fig. 2G). Infection combined with acute hypoxia resulted in a negative energy state with loss of inguinal (white) and intrascapular (brown) adipose tissue (Fig. 2, H and I), increased serum ketone production (Fig. 2J), and lower circulating glucose levels in animals that succumbed to infection (Fig. 2K). Hypoxic preconditioning rescued the animals from this negative energy state, with restoration of proportionate fat and carbohydrate consumption, liver glycogen reserves and fat mass, reduction in circulating ketones, and restoration of circulating fasting glucose and insulin levels (Fig. 2, F to M). Furthermore, although brown fat glucose uptake was suppressed (Fig. 2N), in vivo 18F-fluoro-2-deoxy-d-glucose (FDG) positron emission tomography (PET) studies identified enhanced glucose uptake by the myocardium after preconditioning (Fig. 2, O and P, and fig. S7A), with a parallel increase in cardiac function observed by measures of cardiac index and ejection fraction (Fig. 2, Q and R).

Hypoxic preconditioning responses are retained in the circulating leukocyte population, with morbidity a neutrophil-mediated response

Mice experienced 1 week of hypoxia before they were maintained in normoxia for up to 28 days prior to bacterial challenge in hypoxia to determine the longevity of the preconditioning response. Protection was maintained even after 7 or 28 days of normoxic reacclimatization (Fig. 3, A to D). Thus, the marked protection observed with hypoxic preconditioning extends substantially beyond the initial period of exposure.

Fig. 3 Hypoxic preconditioning confers long-term protection from adverse responses to infection in hypoxia and changes the transcriptome of circulating leukocytes.

(A to D) Duration of preconditioning effect. Mice were preconditioned in hypoxia for 7 days, SH1000 was injected subcutaneously after either 7 days (A and B) or 28 days (C and D) back in normoxia, and sickness scores (A and C) and temperatures (B and D) were recorded after a further 12 hours. (B) *P = 0.0203 (n = 5); (C) *P = 0.0142 (n = 6 H and 7 Pre H); (D) *P = 0.0238 (n = 6 H and 7 Pre H), unpaired t tests. (E to G) Bone marrow transfer. Mice were preconditioned in hypoxia for 7 days, and bone marrow was harvested and injected into WT C57BL/6 mice preirradiated with 12 fractions of 1 Gy (Pre H to WT), with nonpreconditioned mice used as marrow donor controls (Un to WT). After 3 weeks, reconstituted mice were challenged with either subcutaneous SH1000 in hypoxia (E and F) or intratracheal S. pneumoniae (G). Sickness scores (E) and temperature (F) were recorded after 12 hours. (E) *P = 0.0444, Pre H to WT versus Un to WT (n = 6 Pre H to WT and 5 Un to WT), unpaired t test. (G) Kaplan-Meier survival curves were undertaken in mice instilled with S. pneumoniae and housed in normoxia or hypoxia. P = 0.0007, Normoxia versus Hypoxia; P = 0.0305, Naïve BM hypoxia versus Pre BM hypoxia (n = 8), log-rank test. (H) Myeloid cell depletion. Twelve hours after injection with vehicle control (PBS), anti-Ly6G antibody (Ly6G), clodronate (Clod), or control liposomes containing PBS (Lipo), animals were challenged with subcutaneous SH1000 and rectal temperatures were measured after 12 hours. ****P = 0.00000002, Normoxia PBS versus Hypoxia PBS; *P = 0.0235, Hypoxia PBS versus Hypoxia Ly6G; *P = 0.0146, Hypoxia Ly6G versus Hypoxia Clod (n = 8 N PBS, 7 H PBS, 7 N Ly6G, 7 H Ly6G, 8 N Clod, 8 H Clod, 3 N Lipo, and 3 H Lipo), two-way ANOVA with Tukey’s posttest. Horizontal lines (A to F and H) are means ± SEM. (I to K) RNA-seq. Blood leukocytes were harvested from hypoxia-naïve mice instilled with S. pneumoniae in 10% O2 (NI) or vehicle control (NC) or after hypoxic preconditioning (PI and PC). (I) Pearson correlation heat map analysis with hierarchical clustering on total RNA-seq data sets. Pearson correlation scores are plotted from 0 (white) to 1 (dark blue) in steps of 0.1. (J) Heat map of column-normalized z scores for each gene identified as changing significantly between the NI and PI samples. Blue, z score of −2; yellow, z score of +2. (K) Signatures of differential gene expression across the data sets. Genes were selected on the basis of fold change values greater than twofold induced or repressed relative to naïve control expression status (=1). Numbers of genes per group are shown in square brackets above the plots.

The importance of leukocyte populations in regulating outcomes after hypoxic preconditioning was explored in a series of bone marrow transfer experiments. Reconstitution of hypoxia-naïve mice with hypoxia-preconditioned bone marrow, after fractional radiation (18, 19) to ensure selective targeting of the marrow population, conveyed protection from the morbidity (Fig. 3, E and F, and fig. S7, B to D) and mortality observed in both infection model systems (Fig. 3G). These results implicate circulating leukocyte populations both in driving the systemic phenotypes observed in acute hypoxia and in rescuing this response after hypoxic preconditioning. Anti-Ly6G (neutrophil) and clodronate (mononuclear cells) depletion (fig. S7E) subsequently identified the hypoxia-induced hypothermic responses to be neutrophil-dependent (Fig. 3H). To address the mechanism by which preconditioning reprograms neutrophil glucose utilization and protects against cardiac decompensation, we undertook RNA sequencing (RNA-seq) of circulating leukocytes after acute hypoxia in the presence or absence of infection with S. pneumoniae and of hypoxic preconditioning. Although global transcriptional states were largely unaltered between the four conditions (mean Pearson’s correlation score, 0.88), a number of genes displayed strong expression changes, with most of the changes a hallmark of preconditioning alone (Fig. 3, I to K).

Preconditioning represses leukocyte hypoxia-inducible factor pathway and glycolysis genes, resulting in suppression of glucose utilization and phenotypic rescue

Pathway analysis revealed unique signatures between the treatment groups (fig. S8), with suppression of hypoxia-inducible factor–1α (HIF-1α) pathway genes by preconditioning (fig. S9, A to C) further validated at both the RNA (Fig. 4, A and B) and protein level (Fig. 4C). HIF-1α is widely recognized to play a central role in coordinating cellular-adaptive responses to hypoxia (20), including energy metabolism, and in regulating myeloid cell phenotypes (19, 2124). Correlative heat map analysis and hierarchical clustering of the subset of 1274 detected metabolic genes revealed distinct metabolic gene expression signatures between preconditioned and naïve states (Fig. 4D). Metabolic pathway analysis subsequently detailed the relative suppression of glucose transporters and key unidirectional glycolytic enzymes, validated by quantitative polymerase chain reaction (qPCR) (Fig. 4E and fig. S9, D to F) and Western blot (Fig. 4C). A reduction in glycolysis after preconditioning was subsequently demonstrated by suppression of extracellular acidification rates (ECARs) (Fig. 4F), with diminished glucose uptake by circulating and recruited leukocytes confirmed in vivo by 18F-FDG PET studies (Fig. 4, G and H). Last, to validate leukocyte HIF-1α expression as the critical mediator of systemic morbidity, HIF-1α (Hif1aflox/flox;LysMcre+/−)–deficient mice were challenged with S. aureus. Myeloid-specific loss of HIF-1α (Hif1aflox/flox;LysMcre+/−), independent of effects on either local bacterial counts or systemic cytokine release (fig. S10, A to D), conveyed near-complete protection from the heightened sickness responses and hypothermia in hypoxia-exposed mice (Fig. 4, I and J). This protection was equivalent to the effect of preconditioning alone (fig. S10E). Administration of a nonselective pan hydroxylase inhibitor (dimethyloxalylglycine) was unable to suppress leukocyte HIF expression and therefore did not confer protection in this model system (fig. S10, F to H). In keeping with metabolic protection, mice with myeloid-specific targeted deletions of HIF-1α (Hif1aflox/flox;LysMcre+/−) displayed higher circulating glucose levels after infection even in the nonfasting state (Fig. 4K).

Fig. 4 Repression of leukocyte HIF-1α suppresses glucose utilization, resulting in phenotypic rescue.

(A to C) HIF-1α expression. Leukocytes were harvested from mice injected with SH1000 (A) or S. pneumoniae (Spn) (B and C) in normoxia (N; 21% O2), hypoxia (H; 10% O2), or hypoxia after hypoxic preconditioning (Pre H). RNA and protein were extracted, and relative expression of Hif1a and Hif1b RNA was normalized to Actb (A and B) with protein to P38 (C). (B) Hif1a: *P = 0.0447, H versus Pre H (n = 6 N, 7 H, and 6 Pre H), unpaired t test; Hif1b: *P = 0.0272, H versus Pre H (n = 6 N, 7 H, and 6 Pre H), unpaired t test. (D and E) Metabolic gene expression analysis. Correlation heat map and hierarchical clustering analysis was undertaken in metabolic transcripts identified within the RNA-seq data set of blood leukocytes harvested 14 hours after intratracheal installation of S. pneumoniae (Spn) in normoxia (N), hypoxia (H), or hypoxia after preconditioning (Pre H). Color scale: red, higher correlation; blue, lower correlation, with significance determined by multiscale bootstrap analysis (P < 0.01) and represented by changes in color within the dendrogram (D). (E) The log fold change of individual glycolytic enzyme transcripts was determined for preconditioned leukocytes relative to naïve controls (green, up-regulated by at least 60%; orange, down-regulated by 60%). Changes in transcript abundance were further validated in the SH1000 model by real-time PCR. Glut1: *P = 0.0168, N versus H; **P = 0.0082, H versus Pre H; Tpi1: **P = 0.0074, N versus H; Gapdh: **P = 0.0015, N versus H; *P = 0.0345, H versus Pre H; Pkm: *P = 0.0460, H versus Pre H (n = 6 N, 6 H, and 5 Pre H), one-way ANOVA with Tukey’s posttests. (F to H) Functional changes in metabolism. (F) Leukocytes were harvested from mice challenged with subcutaneous PBS or SH1000 and housed in normoxia (N), hypoxia (H), or hypoxia after preconditioning (Pre H). Glycolytic rates after infection were indirectly measured by ECARs relative to vehicle controls. ***P = 0.0007, N versus H; *P = 0.0466, N versus Pre H; *P = 0.0114, H versus Pre H (n = 3), one-way ANOVA with Tukey’s posttest. (G and H) 18F-FDG was administered 5 hours after instillation of S. pneumoniae (Spn), and radioactivity levels in harvested blood leukocytes (G) and BAL cells (H) were assessed after 1 hour by gamma counts. (G) *P = 0.0269, H versus Pre H (n = 4 N, 3 H, and 4 Pre H); (H) *P = 0.0124, H versus Pre H (n = 3 N, 4 H, and 4 Pre H), one-way ANOVA with Tukey’s posttests. (I to K) Myeloid loss of HIF-1α. Sickness scores (I) and rectal temperatures (J) were determined in WT C57BL/6 and Hif1aflox/flox;LysMcre+/− (Hif1a−/−) mice 12 hours after subcutaneous injection of SH1000. (I) **P = 0.00144, WT H versus Hif1a−/− H (n = 8 WT H and 10 Hif1a−/− H); (J) ***P = 0.000545, WT H versus Hif1a−/− H (n = 7 WT H and 10 Hif1a−/− H), two-way ANOVA with Tukey’s posttests. (K) In a separate group of animals, random glucose levels were measured in WT and Hif1aflox/flox;LysMcre+/− (Hif1a−/−) mice 20 hours after infection with S. pneumoniae in 10% O2. *P = 0.0158, WT versus Hif1a−/− (n = 21 WT, 23 Hif1a−/−), unpaired t test.


A substantial body of work using conditional knockout mice has formed the basis of our understanding of how the HIF hydroxylase pathway modifies myeloid cell function and survival. These studies have delineated context-specific roles for individual HIF family members in the regulation of macrophage invasion and motility (21), macrophage bacterial killing (21, 22), neutrophil survival (19, 23, 25), and phagocytosis (26) both in the setting of regional hypoxia and after bacterial infection. The consequences of systemic hypoxia and bacterial infection acting in concert have not previously been examined in detail but are highly relevant to clinical situations such as ARDS and COPD.

Our studies in murine models of both localized and systemic acute infection in the setting of acute hypoxemia demonstrate the catastrophic in vivo consequences of exaggerated leukocyte HIF-1α activation. Profound pathophysiological responses occur as a consequence of imbalanced glucose availability and utilization, with animals infected in the setting of acute hypoxia skewing toward carbohydrate utilization, loss of liver glycogen stores, consumption of white and brown adipose tissue, elevated ketone production, and the eventual development of a circulating hypoglycemia. Strikingly, these animals show preserved bacterial killing capacity despite metabolic compromise. This divergence between bacterial clearance and overall morbidity and mortality thus indicate the importance of targeting the host response, in combination with an antimicrobial strategy, to improve outcomes where hypoxia and infection coexist.

Hypoxia and infection can coexist chronically, such as in COPD. In an experimental model of a more chronic state of hypoxia and bacterial infection, we could ameliorate the observed increase in morbidity and mortality by previous exposure of animals to hypoxia. This protection extended beyond the initial hypoxic exposure period, was sustained over time, and was both dependent on and retained by the bone marrow compartment. These data suggest that hypoxic preconditioning changes bone marrow leukocyte populations, which, when released into the circulation after bacterial challenge, demonstrate altered behaviors that, in turn, determine the host outcome. Subsequent phenotypic rescue with anti-Ly6G depletion demonstrated that these hypothermic and sickness responses were neutrophil-mediated. Thus, short-lived circulating cells can have functional memory of previous hypoxic challenges, resulting in modified innate immune responses. Although HIF-1α–dependent training of responses to bacterial sepsis and a fungal cell wall component has been observed (27), our work provides evidence that oxygen availability is a critical determinant of morbidity and mortality outcomes after bacterial infection.

Immune cell metabolism can profoundly influence key inflammatory responses. Macrophage polarization states are, in part, driven by metabolic processes. Both antimicrobial M1 macrophages and neutrophils depend on glycolysis, yet tissue repair M2 macrophages require fatty acid oxidation and oxidative phosphorylation for ATP production, with the tricarboxylic acid (TCA) cycle dominating over glycolysis (2831). In contrast, neutrophils demonstrate a disordered TCA cycle even in the resting state (32). Consequences of metabolism for immune responses are therefore cell type–specific and vary in physiological and pathophysiological disease states. In the models studied, we saw no effect of clodronate-mediated mononuclear cell depletion on either hypoxia-induced hypothermia or sickness, whereas neutrophil depletion protected from the sickness response. We therefore propose that neutrophils are dependent on glucose and that what would be a beneficial adaptive response under normal systemic oxygenation drives a fatal response in the context of systemic hypoxia where global glucose requirements are increased, resulting in the development of hypoglycemia and cardiac failure. There are precedents that cell type–specific metabolic responses can influence the metabolism of other organs with consequences for the host. For example, skeletal muscle deletion of Phd2 protects against myocardial ischemia reperfusion injury through the regulation of hepatic kynurenic acid production (33), whereas M2 macrophages can coordinate browning of subcutaneous white adipose tissue (34).

We observed HIF-1α to be the critical mediator of the heightened sickness responses, with myeloid-specific loss of HIF-1α conveying protection from hypothermia, sickness, and hypoglycemia after challenge with S. aureus. This is of interest, given that HIF-1α deficiency has been previously linked to impaired bacterial killing (21) and negative outcomes in more fulminant bacterial models (22). Thus, the innate immune response must be exquisitely sensitive to HIF-1α levels of activity, with both insufficient (fulminant infection models) and excessive (acute infection and hypoxia) HIF-1α activation associated with poor outcomes, and hypoxic preconditioning enabling the restoration of this finely tuned response. Understanding the mechanisms that regulate HIF-1α expression, stability, and activity will therefore be critical in the development of strategies that target the innate immune response. Although stabilization of HIF-1α protein in innate immune cells is well characterized in the setting of hypoxia, regulation of HIF activity has also been described after pathogen challenge (22), exposure to bacterial LPS cell membrane fraction (35), and iron chelation (36). More recently, small metabolite regulation of HIF-1α stability in both near-haploid cells (37) and CD8+ T lymphocytes (38) has been described, with HIF-1–dependent accumulation of S-2-hydroxyglutarate (S-2HG) after T cell receptor triggering, further augmenting HIF signaling in both normoxic and hypoxic culture conditions. A further level of complexity is added with the observation that S-2HG can also inhibit the 2-oxoglutarate–dependent epigenetic modifiers that demethylate histones (Jumonji C–containing proteins) or oxidize 5-methylcytosine in DNA [ten-eleven translocation (Tet) proteins] (3941). Thus, in a model of T cell receptor triggering, changes in small metabolite abundance can both regulate HIF-1α expression and lead to alterations in epigenetic marks (38). These data, combined with the evidence of innate immune training after repeated infectious challenge, suggest that further exploration of the links between individual metabolites, epigenetic changes, hypoxia, and regulation of HIF pathway responses will be important in delineating the mechanisms by which hypoxia can reprogram the neutrophil transcriptome, a limitation of our current study and an area of ongoing research activity.

In summary, we show that outcomes of infection are profoundly regulated by neutrophil responses to oxygen and nutrient availability. We identify a mechanism by which hypoxic preconditioning induces sustained changes in leukocyte glucose requirements and utilization as a consequence of transcriptional suppression of leukocyte HIF-1α responses. This, in turn, defines survival outcomes after local and systemic bacterial challenge through a restoration of balance between glucose availability and tissue need and independent of recognized antimicrobial function. Our work highlights the potential therapeutic importance of targeting the host response in combination with antimicrobial strategies when treating bacterial infection in the setting of hypoxia.


Study design

The goal of this study was to investigate the effects of ambient hypoxia on host responses to S. aureus and S. pneumoniae infection. We also determined the mechanisms by which hypoxic preconditioning prevented adverse outcomes in these infection models. Pilot experiments were performed to define the number of animals required to detect significant differences in body temperature and sickness score between infected animals exposed to hypoxia and those maintained in normoxia. Vehicle-treated animals were also exposed to hypoxia and assessed alongside infected animals. Other end points included measurements of bacterial burden and serological and physiological parameters as stated under the subheadings below. The number of biological replicates for each experimental group is indicated in the figure legends. In vivo experiments were performed independently at least twice, unless otherwise specified. Clinical assessment of mouse sickness behavior was made by two independent observers blinded to which oxygen tension the mice had been exposed. Study design was approved with authorization of our project license in accordance with the Home Office Animals (Scientific Procedures) Act of 1986.

Murine colonies

Lysozyme M–driven cre recombinase (LysMcre) was used to target Hif1a (Hif1aflox/flox;LysMcre+/−) deletions in myeloid lineage cells. Animals were backcrossed to a C57BL/6 background (21, 42). C57BL/6 mice or littermate LysMcre−/− floxed mice were used as controls. All animal experiments were conducted in accordance with the Home Office Animals (Scientific Procedures) Act of 1986 with local ethics approval.

Subcutaneous skin infection model

SH1000, a strain of S. aureus derived from the clinical isolate NCTC 8325, was used throughout these experiments (43, 44). Mice were injected subcutaneously with live stationary phase bacteria (1 × 107 CFU) before they were exposed to hypoxia (10% O2) or maintained in room air. At indicated time points (6 and 12 hours), mice were assessed alongside normoxic controls. Clinical assessments of mouse sickness behavior were made, and rectal temperature was recorded. Mice were anesthetized and exsanguinated, and tissues were processed as described below or in the Supplementary Materials.

Blood pressure and heart rate measurements in awake mice

To obtain readings in awake mice, we trained animals for 7 days to undergo noninvasive blood pressure measurement using the BP-2000 Blood Pressure Analysis System (Visitech Systems Inc.). After training, these animals were then injected with either SH1000 bacteria or phosphate-buffered saline (PBS) and exposed to hypoxia or normoxia for 12 hours, after which time blood pressure and heart rate measurements were recorded. A minimum of 10 readings were attempted on each animal.


Twelve hours after subcutaneous injection of bacteria, hypoxic or normoxic mice were anesthetized with 5% isoflurane supplied in oxygen at 2 liters/min and placed on a warming pad. Mice were secured on the pad lying flat and supine. The fur on the chest was clipped, and hair removal cream was used to ensure good penetration of ultrasound waves. Transthoracic echocardiography was performed by an experienced operator using a VisualSonics Vevo 770 Imaging system and RMV707B scanhead (VisualSonics).

Bone marrow transplantation

C57BL/6 recipient mice were irradiated with three fractions of 1 Gy each day for 4 days before injection with 1.5 × 106 bone marrow cells from C57BL/6 mice exposed to hypoxia (10% O2) for 7 days or control C57BL/6 mice. S. aureus subcutaneous injection experiments were performed 4 weeks after injection of donor marrow, at a time point when tissue macrophages retain a native phenotype.

RNA isolation and relative quantification

Murine peripheral blood leukocytes (1 × 106 per condition) were lysed, and RNA was extracted using the mirVana total RNA isolation protocols (Ambion). For RNA quantification, samples were treated with deoxyribonuclease (DNase) (Ambion), and random hexamer complementary DNA (cDNA) was synthesized by reverse transcription. Assays-on-Demand gene expression TaqMan MGB 6FAM dye-labeled products (Applied Biosystems) were used for relative quantification of cDNA.

Intratracheal pneumonia model

Wild-type (WT) C57BL/6 mice were anaesthetized with ketamine (100 mg/kg intraperitoneally; Vetalar V, Pfizer) and acepromazine (5 mg/kg intraperitoneally; Calmivet solution injectable, Vetoquinol). The fur was shaved from the neck, and a small incision was made. The trachea was then exposed by blunt dissection and cannulated with a 24-gauge cannula (Jelco radiopaque cannula, Smiths Medical International Ltd.). Each mouse had 1 × 107 CFU of D39 type 2 S. pneumoniae instilled via the trachea. Control animals were instilled by the same method with PBS. Mice were recovered for 6 hours and then exposed to hypoxia (10% O2) or maintained in room air. At indicated time points (14 and 24 hours), mice were assessed alongside normoxic controls. Clinical assessment of mouse sickness behavior was made by two independent observers blinded as to which oxygen tension the mice had been exposed to, and rectal temperature was recorded. Mice were then culled, and tissues were harvested. For preacclimatization experiments, mice were housed in 10% oxygen for 7 days before bacterial challenge as described above. For the Kaplan-Meier plots, mice were culled once the threshold of sickness was reached.

Assessment of lung injury

Bronchoalveolar lavage (BAL) was performed via cannulation of the trachea. Total cell counts were calculated using hemocytometer counts, and differential counts were assessed on cytocentrifugation slides. Levels of IgM and elastase were analyzed using commercially available kits (Mouse IgM ELISA quantitation set, Bethyl Laboratories Inc.; EnzChek Elastase Assay Kit, Molecular Probes Europe BV).

Quantification of viable bacterial counts

Tenfold serial dilutions were performed on whole blood aliquots. Three 10-μl drops from each of six dilutions were then plated onto blood agar plates and cultured overnight in 37°C to calculate viable bacterial counts. After collection of the BAL fluid, the lungs were carefully dissected and stored in sterile tubes. The lungs were homogenized, and 10-fold serial dilutions were performed on each sample to calculate viable bacterial counts, which were normalized to count per pair of lungs.

Respiratory exchange ratios

Mice were placed individually into a precalibrated home-cage indirect calorimetry PhenoMaster system (TSE Systems) for 1 hour, and oxygen consumption, carbon dioxide production, and xyz activity by infrared beam breaking were assessed. Flow rate was set to 0.3 liter/min. Mice cages were sampled for 3 min and compared with an empty cage air reference every cycle in 15-min blocks. Respiratory exchange ratios were calculated from O2 and CO2 measurements, and the data are presented as the average of cycles 2 to 4 of the 1-hour period.

Liver histology

Livers were placed in 10% buffered formalin before processing and staining with hematoxylin and eosin or periodic acid–Schiff (PAS) reagents.

Paired plasma biochemical profiling

Animals were fasted for 6 hours before sacrifice, and blood was collected at end point for glucose, insulin, and γ-hydroxybutyrate quantification by enzyme-linked immunosorbent assay [glucose (HK), Sigma; insulin, Crystal Chem; hydroxybutyrate, Sigma].

In vivo 18F-FDG PET

Circulating glucose levels were measured in animals at the time of radiotracer injection. Animals were injected with intraperitoneal 18F-FDG (5.1 ± 2.2 MBq, 0.2 ± 0.05 ml) 5 and 23 hours after the installation of intratracheal S. pneumonia. After injection, mice were returned for 1 hour to their original ambient oxygen tensions before tissue harvesting for gamma counting or imaging in vivo under isoflurane anesthesia (1.5% oxygen, 0.5 liter/min; nitrous oxide, 0.5 liter/min) using a NanoPET/CT scanner (Mediso Medical Imaging Systems). A 30-min whole-body emission scan was obtained using a 1:5 coincidence list mode. At the end of the emission scan, a computed tomography scan was acquired (semicircular full trajectory, maximum field of view, 480 projections, 55 kilovolt peak, 300 ms, and 1:4 binning) for attenuation correction and co-registration with PET data. Three-dimensional PET data were reconstructed into 3 × 10–min frames using the Mediso iterative Tera-Tomo 3D reconstruction algorithm and the following settings: four iterations and six subsets, full detector model, normal regularization, spike filter on, voxel size of 0.6 mm, and energy window of 400 to 600 keV. PET data were corrected for randoms, scatter, and attenuation. Reconstructed whole-body PET scans were imported into PMOD 3.4 software (PMOD Technologies), and volumes of interest were drawn around organs and tissues of interest. The measured activity of the target organs and tissues was expressed as the standard uptake values corrected for circulating glucose levels. Radioactivity levels of collected tissue samples were assessed using an automatic gamma counter (Wizard 1470 Gamma Counter, PerkinElmer). Measured disintegrations per minute were converted to becquerel, expressed in percentage injected dose per gram of tissue (%ID/g), and normalized to circulating glucose levels.

Depletion of myeloid cell subsets

WT C57BL/6 mice were injected intraperitoneally with vehicle control, anti-Ly6G antibody (500 mg in 500 μl per mouse; eBioscience), and clodronate- or PBS-containing liposomes (500 μl per mouse; Twelve hours later, the mice were challenged with subcutaneous SH1000 and housed in either normoxia (21% O2) or hypoxia (10% O2). At 12 hours after the challenge, clinical assessment of sickness was performed by two separate observers and rectal temperature was recorded. Mice were then anesthetized, and 200 μl of whole blood cell was collected into vials containing 2 mM EDTA. Red blood cells were lysed twice in red cell lysis buffer (BioLegend), and cells were washed in fluorescence-activated cell sorting buffer (PBS plus 2 mM EDTA plus 2% bovine serum albumin). Cells were stained with Ly6C fluorescein isothiocyanate (BioLegend), Lineage (CD3 and CD19, both BioLegend), Siglec-F phycoerythrin (BD Biosciences), CD115 allophycocyanin (BioLegend), 7-amino-actinomycin D (BioLegend), CD45 AF700 (BioLegend), and Ly6G Pacific Blue (BioLegend). Cells were acquired using a BD Fortessa 6 laser flow cytometer.

RNA-seq and analysis

Murine peripheral blood leukocytes (1 × 106 per condition) were lysed, and RNA was extracted using Mini total RNA isolation protocols (Qiagen). Samples were treated with DNase (Ambion), and sample integrity was verified on the Agilent Bioanalyzer with the RNA Nano chip. Illumina TruSeq paired-end strand-specific sequencing was carried out on a NextSeq 550 sequencer (Edinburgh Clinical Research Facility, Western General Hospital, Edinburgh, Scotland, U.K.). Total RNA (500 ng) underwent ribosomal RNA depletion before purification, fragmentation, random hexamer cDNA generation, and purification with AMPure XP beads (Beckman Coulter). Multiple indexing adapters were ligated to double-stranded cDNA with subsequent hybridization onto flow cells and DNA fragment enrichment by 15 cycles of PCR for sequencing. Completed libraries were quantified by qPCR using the KAPA Library Quantification Kit (Illumina) before multiplexing in two equimolar pools and running on two flow cells on the Illumina NextSeq 550. The resulting FastQ files were mapped to the reference genome (mm9) using the TopHat alignment tool (V1) on the Illumina BaseSpace software, and reads per kilobase per million (RPKM) scores were calculated. Differentially expressed genes were identified using Cufflinks and the Differential expression tool on the Illumina BaseSpace software. Genes showing greater than twofold change with DESeq-generated P values of <0.05 were termed significant changers. Global analysis of total RPKM data sets to determine overall trends on a gene-to-gene basis was carried out using R values and distances calculated by Euclidean and Ward methods, with the resulting Pearson’s correlation scores plotted on a heat map. RPKM values for genes identified as being significantly different between the naïve infected (NI) and preconditioned infected (PI) data sets by Cufflinks were subsequently expressed as a z score heat map with row-normalized values. Signatures of expression change were calculated by stratifying the total data set into groups displaying significant changes (P < 0.05, twofold change) in RPKM score relative to the naïve control (NC) set as follows: up/down infection (altered expression in both naïve and preconditioned infection sets), up/down NI (altered expression only in the NI set), up/down PI (altered expression only in the PI set), and up/down preconditioning (altered expression in both the preconditioned naïve and infected states). Gene Ontology term analysis was carried out on these gene lists using the David functional annotation tool ( Analysis of the HIF-1α pathway was carried out using KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway annotations found at the KEGG pathway database ( Metabolic gene expression analysis was performed in R on the subset of metabolic genes, defined as the union of genes in KEGG metabolic gene sets and genes of the iSS1393 mouse metabolic reconstruction (PMID 23022739). Genes expressed at a level of at least 1 count per million reads in at least three of the samples were filtered out with the EdgeR package (PMID 19910308), and differentially expressed genes and their false discovery rate–corrected P values were identified by the limma package (PMID 25605792). Heat map analysis was performed using the d3heatmap package, and significant clusters were calculated via multiscale bootstrap analysis with the pvclust package (PMID 16595560). For pathway analysis, gene expression data were mapped on KEGG metabolic pathways using the Pathview package (PMID 23740750) and color-coded according to the directionality of deregulation.

Immunoblot detection of murine leukocyte protein

Murine lysates were prepared by lysis in SDS. Immunoblotting was performed with polyclonal anti-mouse HIF-1α (Cell Signaling Technology) primary antibody. Sample loading was confirmed by p38 mitogen-activated protein kinase expression (Cell Signaling Technology). All bands shown were at the predicted molecular weight for the protein of interest.

Seahorse analysis of extracellular acidification

Leukocytes were resuspended in extracellular flux assay medium at a concentration of 3 × 106/ml. Three million cells per condition were plated onto a XF24 cell plate precoated with Cell-Tak (Corning). ECAR was measured at intervals of 7 min over a 90-min cycle using a Seahorse XF24 (Seahorse Bioscience).

Statistical analysis

Data were analyzed using Prism 7.0 software (GraphPad Software Inc.). For comparison of two-sample means when cells from the same individual were used, paired t tests were performed. Unpaired t tests were used for comparisons between infected normoxic and hypoxic sample means, with one-way analysis of variance (ANOVA) and Tukey’s posttests used if the comparison also included a vehicle control group. If multiple time points or concentrations were used, repeated-measures ANOVA with Dunnett’s posttests were performed, and if comparisons between normoxia and hypoxia or WT and transgenic mice were required in these experiments, two-way ANOVA with Bonferroni posttests were performed. For bacterial counts, Kruskal-Wallis test with Dunn’s multiple comparisons test was used. Survival was analyzed using a log-rank test. Statistical significance was accepted when P < 0.05.


Materials and Methods

Fig. S1. Hypoxia induces hypothermia, sickness behavior, and cardiac dysfunction in mice infected subcutaneously with S. aureus.

Fig. S2. The phenotype observed in hypoxic infected C57BL/6 mice is reproduced by heat-killed SH1000 and by intraperitoneal LPS in an outbred mouse strain and after dawn or dusk infection.

Fig. S3. Local immune responses are not impaired by hypoxia.

Fig. S4. Mice display stress responses to hypoxia but no demonstrable systemic inflammatory or oxidative response.

Fig. S5. Mice display preserved gross organ function in hypoxia.

Fig. S6. No evidence of increased pulmonary edema or pulmonary immune infiltration in hypoxic animals infected with S. aureus.

Fig. S7. Hypoxic preconditioning alters glucose uptake into tissues, and protection is conveyed via the bone marrow.

Fig. S8. Hypoxic preconditioning profoundly changes the global transcriptome of circulating leukocytes.

Fig. S9. Hypoxic preconditioning alters the HIF-1α pathway and downstream targets in circulating leukocytes.

Fig. S10. Preserved bacterial load, cytokine responses, and preconditioning responses with myeloid-specific suppression of HIF-1α.

Western blots (pdf)

Source data (excel file)


Acknowledgments: We thank L. Murphy for help with the Illumina RNA-seq. Funding: This work was supported by the Medical Research Council (MRC) Clinical Training Fellowship (awards G0802255 and MR/K023845/1 to A.A.R.T. and R.S.D., respectively), a National Institute for Health Research (NIHR) Clinical Lectureship and an Academy of Medical Sciences starter grant (to A.A.R.T.), a Wellcome Trust postdoctoral clinical fellowship (110086 to A.M.), a Wellcome Trust Senior Clinical Fellowship award (098516 to S.R.W.), a Wellcome Trust Senior Clinical Fellowship award (076945 to D.H.D.), a British Lung Foundation Fellowship (F05/7 to H.M.M.), a Wellcome Trust New Investigator Award (WT100981MA to N.M.M.), and a British Heart Foundation Senior Basic Science Research Fellowship (FS/13/48/30453 to A.L.). E.R.C. and A.S.C. are supported by the NIHR Cambridge Biomedical Research Centre. R.H.S. is supported by the MRC. R.R.M. is supported by MRC (MC_PC_U127574433), Biotechnology and Biological Sciences Research Council, and European Chemical Industry Council grants. M.M. is supported by the European Research Council (OxyMO). The MRC/University of Edinburgh Centre for Inflammation Research is supported by an MRC Centre Grant. Author contributions: A.A.R.T., R.S.D., J.P.T., H.M.M., N.M.M., S.J.F., D.H.D., R.S.J., R.R.M., M.K.B.W., and S.R.W. designed the experiments. A.A.R.T., R.S.D., J.P.T., H.M.M., A.T., J.W., L.W., F.M., N.M.M., A. Lewis, N.A., A.M., P.D.S.C., C.D., E.R., E.W., A.G.H., J.A.P., V.F., and A.S.C. performed the experiments. A.A.R.T., R.S.D., J.P.T., H.M.M., S.F., R.H.S., A. Lawrie, M.M., P.S., J.G., F.T., P.C., A.S.C., E.R.C., R.R.M., D.H.D., R.S.J., M.K.B.W., and S.R.W. provided technical expertise and performed data analysis. All authors contributed to writing the manuscript. Competing interests: The authors declare that they have no competing interests.
View Abstract

Stay Connected to Science Immunology

Navigate This Article