Resource

Lipidomic Profiling of Influenza InfectionIdentifies Mediators that Induceand Resolve InflammationVincent C. Tam,1 Oswald Quehenberger,2,4 Christine M. Oshansky,5 Rosa Suen,1 Aaron M. Armando,3,4

Piper M. Treuting,6 Paul G. Thomas,5 Edward A. Dennis,3,4 and Alan Aderem1,*1Seattle Biomedical Research Institute, Seattle, WA 98109, USA2Department of Medicine3Department of Chemistry and Biochemistry4Department of Pharmacology

University of California, San Diego, La Jolla, CA 92093, USA5Department of Immunology, St. Jude Children’s Research Hospital, Memphis, TN 38105, USA6Department of Comparative Medicine, University of Washington, Seattle, WA 98195, USA

*Correspondence: alan.aderem@seattlebiomed.org

http://dx.doi.org/10.1016/j.cell.2013.05.052

SUMMARY

Bioactive lipid mediators play a crucial role in theinduction and resolution of inflammation. To eluci-date their involvement during influenza infection,liquid chromatography/mass spectrometry lipidomicprofiling of 141 lipid species was performed on amouse influenza model using two viruses of sig-nificantly different pathogenicity. Infection by thelow-pathogenicity strain X31/H3N2 induced a proin-flammatory response followed by a distinct anti-inflammatory response; infection by the high-patho-genicity strain PR8/H1N1 resulted in overlappingpro- and anti-inflammatory states. Integration of thelarge-scale lipid measurements with targeted geneexpression data demonstrated that 5-lipoxygenasemetabolites correlated with the pathogenic phaseof the infection, whereas 12/15-lipoxygenase metab-olites were associated with the resolution phase.Hydroxylated linoleic acid, specifically the ratio of13- to 9-hydroxyoctadecadienoic acid, was identi-fied as a potential biomarker for immune statusduring an active infection. Importantly, some of thefindings from the animal model were recapitulatedin studies of human nasopharyngeal lavages ob-tained during the 2009–2011 influenza seasons.

INTRODUCTION

Influenza virus, whether seasonal or pandemic, can cause

severe health conditions, often leading to pneumonia. While

most infections in humans are self-limiting, highly virulent strains

can cause an overexuberant inflammatory response that is detri-

mental to the host (Kash et al., 2006). Influenza virus is an envel-

oped, negative-sense, single-stranded RNA virus that contains

eight genomic segments that encode up to 13 proteins (Jagger

et al., 2012). One of the hallmarks of influenza virus is its high pro-

pensity to acquire mutations, which promotes rapid evolution to

increase fitness and evade pre-existing immunity in a host pop-

ulation. Antigenic drift, or mutations mainly of the segments en-

coding the surface proteins, fosters evasion of the host immune

system and allows a slightly varied virus to re-infect the popula-

tion. Antigenic shift, or recombination and exchange of genomic

segments between multiple viruses infecting the same cells,

allows for the generation of emerging viruses with drastic

changes in pathogenicity or host specificity. If a novel virus

with increased fitness and pathogenicity emerges, the herd

immunity of the human population may be insufficient to pre-

vent sustained transmission, creating devastating epidemic or

pandemic events. The most notorious pandemic, caused by

the 1918 H1N1 strain, cost the lives of 40–100 million people

worldwide (Taubenberger and Kash, 2010). The latest influenza

pandemic was caused by a swine-origin H1N1 virus, which is a

triple reassortment of human, avian, and swine strains (Neumann

et al., 2009). Although it is well known that the pathogenicity of

influenza strains can vary widely, the precise factors and mech-

anisms that contribute to the clinical outcome of infection have

not been defined.

Eicosanoids refer to a family of bioactive lipid mediators that

regulate a wide variety of physiological as well as pathophysio-

logical responses and often exhibit potent inflammatory proper-

ties (Quehenberger and Dennis, 2011). These mediators are

generated from arachidonic acid and related polyunsaturated

fatty acids after their enzymatic release from membrane phos-

pholipids via complex metabolic mechanisms involving over 50

unique enzymes (Buczynski et al., 2009). The three major meta-

bolic pathways for enzymatic eicosanoid biogenesis are the

cyclooxygenase pathway (COX-1 and COX-2), producing the

prostaglandins and thromboxanes; the lipoxygenase pathway

(5-LOX, 12-LOX, and 15-LOX), producing the leukotrienes as

well as numerous hydroperoxy and hydroxy fatty acids; and

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. 213

(legend on next page)

214 Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc.

the cytochrome P450 pathway, producing epoxide and corre-

sponding dihydroxy metabolites of arachidonic acids. Polyun-

saturated fatty acids are susceptible to lipid peroxidation,

especially under condition of oxidative stress during inflamma-

tion producing isoprostanes as well as other nonenzymatic

oxidation products, including hydroxy fatty acids. In addition to

arachidonic acids, the same enzymes can effectively metabolize

other polyunsaturated fatty acids, such as linoleic and linolenic

acids. Several bioactive lipid mediators have also been appreci-

ated for their anti-inflammatory activity and their participation

in the resolution phase of inflammation (Serhan et al., 2008).

Arachidonic acid-derived lipoxins via the lipoxygenase pathway

and docosahexaenoic acid (DHA)- and eicosapentaenoic acid

(EPA)-derived resolvins, protectins, and maresins have anti-

inflammatory and proresolution activities (Serhan et al., 2000).

These lipid mediators can prevent further infiltration of immune

cells to the site of infection as well as signaling the nonphlogistic

phagocytosis of apoptotic immune and epithelial cells, therefore

allowing the system to return to homeostasis after microbial

infection. While many of these lipid mediators have been studied

individually in various model systems, only a few studies have

conducted comprehensive quantification of the metabolites in

a natural microbial infection (Chiang et al., 2012; Morita et al.,

2013; Tobin et al., 2012).

We utilized two influenza viruses, PR8/H1N1 a highly patho-

genic strain in mice, and X31/H3N2, a low-pathogenicity strain,

and profiled 141 lipid metabolites and bioactive mediators dur-

ing the course of a mouse influenza infection in vivo. The induc-

tion kinetics of the protein levels of cytokines/chemokines in the

bronchoalveolar lavage and the transcript levels of immune

response genes in the host are similar between infections with

the two viruses. In contrast, the lipidomic profile provided a

more dynamic and comprehensive description of the processes

involved in the induction and resolution of inflammation. Metab-

olites derived from the lipoxygenase and cytochrome P450 path-

ways and those derived from linoleic acid, DHA, and EPA had

distinct profiles in the two viral infections during the resolution

phase of inflammation. Surprisingly, metabolites of the cycloox-

ygenase pathway showed similar levels and kinetics across

different conditions. We have further profiled the lipidome of

nasopharyngeal lavages from human influenza clinical samples

obtained during the 2009–2011 influenza seasons and validated

some of the observations from the animal model. Many of the

Figure 1. Mouse Influenza Infection Model

(A) Weight loss curves of mice infected with PR8sublethal 200 pfu (blue), X31sublemean ± SEM.

(B) Viral loads in BAL (left), CD45+ immune cells (middle), or alveolar epithelium

Mann-Whitney tests were performed to determine statistical significance (*0.01 <

between PR8lethal and both X31sublethal and PR8sublethal infections; blue and red

infections, respectively; black asterisks denote significance between PR8subletha(C) Representative H&E and anti-influenza antibody-stained sections from PR8 a

bronchioles (TB), arterioles (a), and alveoli (A) are as indicated. Boxed regions on

and IHC panels, except for PR8 at d9, where the sections are stepped; note that th

in lower right X31 IHC panel is an example of the control immunoglobulin G-stai

(D) Representative IHC from PR8 D6- and X31 D5-infected lung sections. Brown

(E) Antiviral response represented as heatmap of RT-PCR data detected in BAL

elongation factor 1 alpha (EF1a).

See also Figure S1.

metabolites and bioactive mediators induced in the high-

pathogenicity PR8 mouse influenza infection model were also

produced at significantly higher levels in individuals with

increased clinical symptoms and immune responses.

RESULTS

Mouse Influenza InfectionWe used X31 and PR8 to determine how the varying pathoge-

nicity of influenza viral strains affects the transcriptional, proteo-

mic, and lipidomic profiles of the host immune response. These

two strains differ in their hemagglutinin and neuraminidase seg-

ments and contain the identical ‘‘internal’’ genes (PB1, PB2, PA,

NP, NS, and M). PR8 is a high-pathogenicity, mouse-adapted

virus that is lethal even at a low multiplicity of infection (Francis

andDe Torregrosa, 1945). In contrast, X31 is a low-pathogenicity

virus that causes a self-limiting infection, even at high doses

(Allan et al., 1990). Three infection conditions were used to deter-

mine the lipidomic profiles of the host during influenza infections:

PR8 at a lethal dose (2 3 105 plaque-forming units [pfu];

PR8lethal), PR8 at a sublethal dose (200 pfu; PR8sublethal), and

X31 at a sublethal dose (23 105 pfu; X31sublethal). We monitored

the progress of influenza infection by measuring the percentage

of weight loss (Figure 1A). Animals infected with PR8sublethal had

a peak weight loss on day 9 postinfection of 22% and recovered

to their original weights by day 13 postinfection. Animals infected

with X31sublethal differed slightly from animals infected with

PR8sublethal; they had a peak weight loss on day 7 postinfection

of 16% and recovered to their original weights by day 10 postin-

fection. Finally, animals infected with PR8lethal experienced rapid

weight loss and succumbed 6 days postinfection.

We next determined the viral load in the bronchoalveolar

lavage (BAL) and in separated cell populations by measuring

the levels of the influenzaM gene. Cell populations included cells

isolated from the BAL, as well as CD45+ immune cells and

CD45� epithelial cells isolated from dispase-digested lungs

(Figure 1B). Viral loads were measured in cells isolated from sub-

lethally infected animals on days 3, 5, 7, 9, 11, 13, and 19 post-

infection, whereas viral loads in lethally infected animals were

measured on days 3 and 5 postinfection (Figure 1B). The viral

load peaked on day 3 postinfection for all three conditions. Not

surprisingly, animals infected with PR8lethal had significantly

higher viral loads in the alveolar epithelium and BAL on day 3

thal 2 3 105 pfu (red), or PR8lethal 2 3 105 pfu (black) on day 0. Graphs depict

(right) as measured by RT-PCR using Fluidigm. Graphs depict mean ± SEM.

p < 0.05; **0.001 < p < 0.01; ***p < 0.001; magenta asterisks denote significance

asterisks denote significance between PR8lethal and PR8sublethal or X31sublethal

l and X31sublethal infections).

nd X31 sublethally infected lungs. Days postinfection, magnifications, terminal

subgross images (1.25X) correspond to the regions shown at 20X for both H&E

e same terminal bronchiole and arteriole are present in both panels. Boxed inset

ned tissue. Black bar represents 6 mm.

staining indicates presence of antigen. Arterioles are indicated (a).

, CD45+, or alveolar epithelial cells. Scale bar represents normalized value to

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. 215

than animals infected with PR8sublethal or X31subtlethal. Animals in-

fected with X31sublethal had similar viral loads on day 3 and day 5

as animals infected with PR8sublethal but subsequently cleared

the virus after day 7 postinfection. Thereafter, the viral load in

animals infected with PR8sublethal was generally higher than

that found within animals infected with X31sublethal.

In separate experiments, we performed hematoxylin and eosin

(H&E) staining and anti-influenza immunohistochemistry (IHC) on

lung sections to evaluate the histopathologic characteristics,

distribution of changes, and localization of influenza antigens

(Figures 1C and 1D and Figure S1 available online). In order to

compensate for the slightly shifted kinetics in weight loss, we

adjusted the collection times for the X31 and PR8 infections.

Lung tissues were collected during the induction of inflammation

(day 5 for X31; day 6 for PR8), during peak weight loss (day 7 for

X31; day 9 for PR8), and during resolution of inflammation (day

13). The IHC staining indicated that, while PR8 is capable of

spreading into the deeper interstitial lung tissue, X31 is primarily

restricted to the major bronchial region with limited positive

staining in the lung (Figure 1D). H&E staining of PR8-infected

lung tissue indicated polymorphonuclear and mononuclear

leukocyte infiltration in and around major bronchial and lung

interstitial space on day 6 postinfection (Figure 1C). The infiltra-

tion spread more extensively in the lung by day 9. On day 13

postinfection, demarcated lymphocytic aggregates were

observed in both infections. For X31 infection, the infiltrating

cells were generally restricted to the major bronchial areas with

lower levels of lymphocyte infiltration at later time points during

the resolution phase than for PR8 infection.

We compared the transcript levels of several interferon-

induced genes (Mx1, Rsad2, and Ifit1) and the transcription fac-

tor Irf7, all signatures of influenza infection (Sato et al., 2000), to

assess the antiviral response (using Taqman assays listed in

Table S1). The strongest responses were detected in infiltrating

cells isolated from the BAL as well as immune cells from the

lung (CD45+) (Figure 1E). Despite the different kinetics of weight

loss in animals infected with PR8sublethal and X31sublethal, the

abundance of Mx1, Ifit1, Rsad2, and Irf7 transcripts were com-

parable and reflected the level of viral infection: PR8sublethalinduced a slightly stronger and more sustained response than

X31sublethal, and PR8lethal induced an even more exuberant

response than the sublethal infections.

Cytokine and Chemokine ResponseIn response to infectionof the lung, aconcertedsignalingcascade

is activated to induce secretion of cytokines and chemokines (as

well as lipid mediators) that promote vasodilation and a massive

infiltration of immune cells to affected tissues. We measured the

cytokine and chemokine levels in the BAL by multiplexed immu-

noassay (Luminex) to establish the levels of protein mediators

inducedduring thecourse of infection (Figures 2 andS2). The shift

in the kinetics of weight loss between PR8sublethal and X31sublethalwas mirrored by the production of interleukin (IL)-5, monocyte

chemoattractant protein (MCP)-1, and tumor necrosis factor

alpha (TNF-a). By contrast, similar levels of IL-6, IL-1b, and

macrophage inflammatory protein-1a (MIP1a) were produced

during infection with PR8sublethal and X31sublethal. Finally, infection

with PR8sublethal was accompanied by higher levels and pro-

216 Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc.

longedproduction of keratinocyte-derived chemokine (KC), inter-

feron-inducible protein-10 (IP-10), granulocyte colony-stimu-

lating factor (G-CSF), regulated on activation, normal T cell

expressed and secreted (RANTES), IL-1a, interferon g, and IL-

10compared to infectionwithX31sublethal (Figures2andS2). Inter-

estingly, despite their difference in pathogenicity, PR8lethal and

PR8sublethal produced similar levels of the cytokines and chemo-

kines described above, with the exception of MCP-1, RANTES,

and G-CSF, which were higher during infection with PR8lethal.

Lipidomic and Transcriptomic ProfilingA variety of bioactive lipid mediators play a critical role in the

regulation of inflammation, and we sought to examine their role

in the pathogenesis of influenza. Given the complexity of the

pathways and multiple routes for degradation and other modifi-

cations of these mediators, we examined the dynamic lipidome

during influenza infection using liquid chromatography (LC)/

mass spectrometry (MS). Over the course of infection, 141 lipid

species (Extended Experimental Procedures) were profiled and

52 lipid metabolites were detected and quantified. The relative

percentages of metabolites generated through the COX, LOX,

and CYP450 pathways derived from arachidonic acid, as well

as metabolites derived from linoleic acid, linolenic acid, DHA,

and EPA, were determined to assess the composition of bioac-

tive lipids in the BAL (Figures 3A and S3A). For mediators in

the COX pathway, levels of prostaglandins increased relative

to other metabolites during the early inflammatory response

(day 3) but decreased thereafter until returning to the basal level,

regardless of whether the mice were infected with PR8sublethal or

X31sublethal. In contrast, the percentage of prostaglandins relative

to other metabolites was significantly lower during PR8lethalinfection. LOX-derived metabolites also decreased during the

early phase of PR8sublethal and X31sublethal infection, while ani-

mals infected with X31sublethal had higher levels of LOX metabo-

lites throughout the resolution phase of inflammation. The

percentage of metabolites of the cytochrome P450 pathway

increased during the induction of inflammation, and the kinetics

were slightly shifted between PR8 and X31, mirroring the kinetics

of weight loss (Figures 1A and S3). In contrast to the COX and

CYP450 pathways, percentages of metabolites derived from

linoleic acid, linolenic acid, DHA, and EPA increased during the

later phase of infection. The concentrations of linoleic acid and

EPA initially decreased during the early phase of inflammation

and subsequently increased throughout the resolution phase.

Moreover, the percentage of linoleic acid metabolites was signif-

icantly higher in PR8sublethal than X31sublethal. In contrast, the low-

pathogenicity virus, X31sublethal, induced a higher percentage of

both DHA- and EPA-derived metabolites than the high-pathoge-

nicity virus, PR8sublethal (Figures 3 and S3A).

While the lipid class composition during infection offered a

view of the activation kinetics of specific lipid metabolic path-

ways, it did not shed light on individual mediators, which have

diverse physiological consequences for induction and resolution

of inflammation. The COX pathway generates prostaglandins,

and their levels were surprisingly similar under all three condi-

tions of infection (X31sublethal, PR8sublethal, and PR8lethal) (Figures

3B and 3C). Most of the prostaglandins peaked early and

decreased thereafter (Figure 3B). In general, the messenger

Figure 2. Cytokine and Chemokine Responses in Mouse Influenza Infections

(A) Cytokine and chemokine levels detected in BAL by multiplex Luminex represented as a heatmap. Scale represents the amount of protein in pg/ml.

(B) Line graphs of selected representative cytokines/chemokines depicting the mean ± SEM.

See also Figure S2.

RNA transcripts encoding the enzymes within the cyclooxyge-

nase pathway displayed similar expression kinetics to the spe-

cific metabolites that they generate (Figures 3B and S3B). One

striking difference between the PR8sublethal and X31sublethal infec-

tions was the significantly higher levels of cytochrome P450 and

linoleic acid-derived metabolites in animals infected with the

high-pathogenicity virus (Figure 3B).

DHA and EPA metabolites have recently been shown to pro-

mote the resolution of inflammation (Ariel and Serhan, 2007;

Bannenberg et al., 2005). While 17-HDoHE, a precursor of resol-

vins and protectins (D series), was found at similar levels in

PR8sublethal and X31sublethal (Figure 3B), two distinct groups of

hydroxylated DHA emerged: 4-, 10-, 13-, and 20-HDoHE were

detected at higher levels in PR8sublethal-infected animals,

while 8-, 14-, and 16-HDoHE were found at higher levels in

X31sublethal-infected animals. Moreover, for lethally infected ani-

mals (PR8lethal), 4-, 10-, 13-, and 20-HDoHE had increasing

levels from day 3 to day 5 postinfection, reaching a level similar

to the maximum level found in PR8sublethal-infected mice (Fig-

ure 3B). In contrast, 8-, 14-, and 16-HDoHE peaked at day 3

and remained at levels significantly greater than in sublethal in-

fections until day 5 (Figure 3B). Of note, several of these metab-

olites can be formed either enzymatically or through autoxidation

processes. For example, 17-HDoHE is a putative product of the

mammalian 12/15 LOX activity, which can be further metabo-

lized to protectin D1, a potential suppressor of viral replication

(Morita et al., 2013; O’Flaherty et al., 2012). Similarly, other posi-

tional isoforms of hydroxylated DHA can be formed via 12/15

LOX (Morgan et al., 2010). Even though the source of these

hydroxylated DHA metabolites are not entirely established, the

profile between PR8 and X31 infection conditions were strikingly

distinct. In general, metabolites of EPA, 12-hydroxyeicosapen-

taenoic (HEPE), and 15-HEPE were detected at a higher level

in X31sublethal-infected animals than in PR8sublethal-infected ani-

mals. The relationship between the level of a metabolite and

that of the transcript of the corresponding enzyme that generates

it can be discerned by comparing Figures 3B and S3B.

Lipidomic Profiling in Human Nasal WashesThe lipidomic profiling in the mouse influenza infection model

allowed us to obtain a comprehensive kinetic view of the lipi-

dome across a well-defined time course. Understanding the

relevance of these results to human infection is crucial. We con-

ducted lipidomic profiling of human nasopharyngeal lavages

obtained from a surveillance study at the St. Jude Children’s

Research Hospital during the 2009–2011 influenza seasons.

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. 217

Figure 3. Lipidomic Profiling of Mouse Influenza Infection

(A) Stacked bar graph representing the percentages of arachidonic acid-derived cyclooxygenase (COX), lipoxygenase (LOX), and cytochrome P450

(CYP450)metabolites, as well as linoleic acid-, linolenic acid-, DHA-, and EPA-derivedmetabolites. Each vertical line represents data from a single animal/sample

(n = 8–10).

(B) Quantified lipid mediators represented as a heatmap. Scale bar represents levels measured in pmol/ml.

(C) Line graphs of selected representative lipid mediators depicting the mean ± SEM.

See also Figure S3.

218 Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc.

Study participants were enrolled based on detection of the influ-

enza matrix gene (influenza A) or nonstructural NS1 gene (influ-

enza B) by quantitative real-time PCR. Index cases and their

asymptomatic household contacts were followed for approxi-

mately 1 month following enrollment, with contacts sampled as

cases following a positive detection of influenza by quantitative

RT-PCR (qRT-PCR). Based on hierarchical clustering using clin-

ical scores and multiplexed immunoassay of growth factors,

cytokines, and chemokines, we identified three groups within

qRT-PCR-confirmed influenza-infected subjects: (1) patients

with low clinical scores and minimal amounts of cytokines and

chemokines, (2) patients with medium clinical scores and posi-

tive for a subset of growth factors and cytokines (epidermal

growth factor, G-CSF, growth-related oncogene, IL1Ra2, IL8,

and IP-10), and (3) patients with high clinical scores and high

levels of a wide panel of cytokines (Figure S4A). A subset of

samples was selected from each group for lipidomic profiling

(n = 9–11/group, total = 30). While the human nasal cavity is likely

to be a very different physiological environment than the mouse

lung, we detected induced levels of many lipid mediators that

were identified in the mouse influenza infection model. We found

one dramatic difference in the lipid profile between the human

andmouse controls; linoleic acid comprises 68%of the lipidome

in the nasopharyngeal lavage of human low responders (asymp-

tomatic contacts of the index cases), whereas it makes up

33.7% of the lipidome in the BAL of naive mice (Figures 3A

and 4A). High levels of linoleic acid have been detected in

humans in the skin, mucosal surfaces, and nasal fluid and

contribute to antimicrobial activity at these sites (Drake et al.,

2008). The samples from patients with high clinical score and

increased levels of cytokines and chemokines also contained

significantly higher percentages of lipidmediators from the lipox-

ygenase, CYP450, DHA, and EPA pathways (Figures 4A and

S4B). Amounts of individual lipid mediators (including PGE2,

LTE4, and 17 HDoHE) were also significantly increased within

the high-response group compared to the low- and medium-

response groups (Figures 4B and 4C).

Proinflammatory and Anti-Inflammatory States duringInfluenza InfectionIn order to look at global changes in the mouse and human data,

we grouped the bioactive lipids based on their proinflammatory

or anti-inflammatory/proresolution properties and created quan-

titative indices to characterize the proinflammatory and anti-

inflammatory contributions of the host response at each stage

of the infection (see Extended Experimental Procedures). These

indices were calculated for each infected mouse or patient at

every time point/condition. To create the index for the mouse

infection model, we first calculated the fold change of individual

lipid mediators in infected animals over naive animals and only

included values greater than or equal to two. The fold changes

for each mediator were then normalized to the maximum value

across all samples, setting the maximum value to one. We sepa-

rated lipid mediators into groups with pro- or anti-inflammatory

activity and then summed the normalized values of each medi-

ator within the pro- or anti-inflammatory group to define the total

activity. The maximum activity in each group is then equal to the

total number of mediators in the group. To generate the index for

each time point, the percentage of the pro- or anti-inflammatory

mediator activity was calculated by dividing total activity by the

maximum activity in the pro- or anti-inflammatory groups.

As discussed above, the early proinflammatory response was

similar between PR8sublethal and X31sublethal infections, as indi-

cated by levels of mediators such as prostaglandins on day 3

postinfection (Figure 5A). This was reflected in a comparable

increase of proinflammatory indices on day 3 for both sublethal

infections (Figures 5B and S5A). For PR8sublethal infection, a

secondary wave of proinflammatory mediators occurred on day

7 postinfection, including linoleic acid-derived epoxide and dihy-

droxy linoleic acid. This contributed to a significant increase in the

proinflammatory indices for PR8sublethal, which began to subside

on day 9 postinfection (Figures 5B and S5A). The mediators 9,

10 EpOME; 12, 13 EpOME; 9, 10 diHOME; and 12, 13 diHOME,

which are potent chemoattractants for neutrophils (Totani et al.,

2000), were present at much lower levels in X31sublehtal-infected

animals on day 7 postinfection. For the anti-inflammatory index,

the peak of the response began on day 7 postinfection and was

not significantly different between PR8sublethal and X31sublethal.

The anti-inflammatory indices decreased slightly on day 9 postin-

fection and sustained these levels until day 13 postinfection. In

contrast to the sublethal infections, PR8lethal induced significant

increases of both pro- and anti-inflammatory indices consecu-

tively on day 3 and day 5 postinfection (Figures 5B and S5A).

Using the same approach to determine the proinflammatory

and anti-inflammatory indices of patient samples (using low clin-

ical symptoms and immune response as the control), the high

response group had significant increases in the pro- and anti-

inflammatory indices compared to the medium response group

(Figures 5C, S5B, and S5C).

Lipoxygenase PathwayWe next dissected the lipoxygenase pathway in more detail,

since it contains well-defined pro- and anti-inflammatory media-

tors. Leukotrienes synthesized by 5-lipoxygenase have long

been known as proinflammatory mediators: leukotriene B4

(LTB4) is a potent chemotactic factor for neutrophil recruitment,

and LTC4 and LTE4 are potent mediators for immediate hyper-

sensitivity, bronchoconstriction, smooth muscle contraction,

and increased vascular permeability. Recently, 12-lipoxyge-

nase- and 15-lipoxygenase-derived metabolites, such as 12-

HETE, hepoxilin, 15-HETE, and lipoxins have been increasingly

appreciated as anti-inflammatory mediators (Kuhn and O’Don-

nell, 2006). Specifically, 12-HETE and 15-HETE are potent acti-

vators of peroxisome proliferator-activated receptor gamma

(PPARg) (Kronke et al., 2009), which can dampen inflammation

via inhibition of nuclear factor-kB and modulate macrophage

function (Ricote et al., 1998). Lipoxins can block further recruit-

ment of neutrophils (Serhan et al., 1995) and stimulate nonphlo-

gistic clearance of apoptotic cells (Godson et al., 2000). While

many of these molecules have been studied individually in

various settings, a comprehensive but focused analysis of the

metabolites and mediators within this pathway would yield a

better understanding of the local inflammatory milieu during an

active microbial infection.

The levels of 5-lipoxygenasemediators were similar in the BAL

of sublethally infected mice, with the exception of LTE4, in which

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. 219

Figure 4. Lipidomic Profiling of Human Nasal Washes

(A) Stacked bar graph represents the percentages of arachidonic acid-derived cyclooxygenase (COX), lipoxygenase (LOX), and cytochrome P450 (CYP450)

metabolites, as well as linoleic acid-, linolenic acid-, DHA-, and EPA-derivedmetabolites. Each vertical line represents data from a single patient/sample, which is

categorized into either the low, medium, or high clinical score/immune response group (n = 9–11).

(B) Quantified lipid mediators represented as a heatmap. Scaled bar represents levels measured in pmol/ml.

(C) Column vertical scatter plot of selected representative lipid mediators depicting the mean ± SEM.

See also Figure S4.

220 Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc.

Figure 5. Pro- and Anti-Inflammatory Indices of Mouse and Human Lipidomic Profiles

(A) Heatmap represents the mean fold change in each lipid mediator (separated by having either pro- or anti-inflammatory activity).

(B) Line graphs of pro- and anti-inflammatory indices depicting the mean ± SEM.

(C) Column vertical scatter plot of pro- and anti-inflammatory indices for individual patient samples. Horizontal lines indicate the mean ± SEM.

See also Figure S5.

they were more abundant in PR8sublethal-infected animals on day

7 and day 9 postinfection (Figure 3C). This difference resulted in

higher percentages of 5-lipoxygenase-derived mediators in

PR8sublethal infection (Figures 6A and S6A). In concordance

with its pathogenicity, PR8lethal infection elicited much greater

amounts of LTE4, which contributed to a significant increase in

the percentages of 5-lipoxygenase-derived metabolites com-

pared to the sublethal infections (Figures 3C, 6A, and S6A). Tran-

scription of 5-lipoxygenase (alox5) was downregulated at the

onset of infection (Figure 6B), similar to the regulation observed

in macrophages stimulated in vitro with lipopolysaccharide

Kdo2-lipid A (Dennis et al., 2010). 5-lipoxygenase activating pro-

tein, alox5ap, on the other hand, was significantly increased in

the lethal infection, which correlated with the level of the lipid

LTE4 (Figures 3C and 6B).

Mediators of the 12/15-lipoxygenase pathway, 12-HETE, and

hepoxilin B3 (HXB3) had increased levels in the late phase of

X31sublethal infection when compared to PR8sublethal (Figure S6A).

These differences contributed to significantly higher percent-

ages of 12-lipoxygenase-derived mediators for X31sublethalwhen compared to PR8sublethal fromday 7 to day 13 postinfection

(Figures 6A and S6A). The level of 15-HETE was similar between

the sublethal infections. The transcriptional level of 15-lipoxyge-

nase (alox15) is significantly increased in X31sublethal infection at 9

and 11 days postinfection in comparison to PR8sublethal (Figures

6B and S6B). The discrepancy suggests that the increased

quantity of Alox15 protein may be acting on other substrates,

such as linoleic acid or linolenic acid. 8-lipoxygenase (alox8) ex-

hibited a greater level of expression in PR8sublethal-infected ani-

mals than in X31sublethal-infected animals at the late phase of

infection. This transcriptional induction in the epithelium corre-

latedwith increased detection of 8-HETE in the BAL fluid (Figures

6B and S6A). In general, a high percentage of proinflammatory

5-lipoxygenase mediators were tightly correlated with the highly

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. 221

Figure 6. Lipoxygenase Enzymes and Derived Metabolites in Mouse and Human Influenza Infection

Stacked bar graph represents the percentages of lipoxygenase 5-, 15-, 12-, or 8-derived metabolites of all lipoxygenase-derived metabolites in mouse (A)

or human (C) samples. Each vertical line represents data from a single sample (n = 8–11 per time point). (B) Line graphs of 5-lipoxygenase (alox5),

5-lipoxygenase-activating protein (alox5ap) in CD45+ cells, 15-lipoxygenase (alox15) in BAL, and 8-lipoxygenase (alox8) in alveolar epithelium depicting the

mean ± SEM.

See also Figure S6.

pathogenic phase of the PR8 virus, whereas an elevated per-

centage of anti-inflammatory 12/15-lipoxygenase mediators

were associated with the resolution phase of the X31 infection

(Figures 6A and S6A).

222 Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc.

This trend was recapitulated in studies of human nasopharyn-

geal lavage fluid obtained during the 2009–2011 influenza

seasons. Here, the percentages of 5-lipoxygenase-derived

mediators were strongly correlated with disease symptoms

Figure 7. Hydroxylated Linoleic Acid Metabolites as a Biomarker for Immune Status during Influenza Infection(A) Pathway for linoleic acid to generate 9-hydroxylated or 13-hydroxylated linoleic acid.

(B) Line graphs of 9-HODE, 13-HODE, and 13-:9-HODE in the mouse lung during influenza infection, depicting mean ± SEM (red: X31sublethal; blue: PR8sublethal;

black: PR8lethal).

(C) Column vertical scatter plot of 9-HODE, 13-HODE, and 13-:9-HODE in human nasal wash samples. Horizontal lines indicate mean ± SEM.

See also Figure S7.

from low to medium to high responders (Figures 6C and S6B).

The high response group exhibited significantly elevated levels

of 5-HETE, LTB4, and LTE4 compared to the low andmedium re-

sponders. In contrast, percentages of 12-lipoxygenase-derived

metabolites were anticorrelated with the level of disease

response (Figures 6C and S6B).

Hydroxylated Linoleic Acid Metabolites as Biomarkersfor the Status of Influenza InfectionWhile the comprehensive study of bioactive lipid profiles offered

a complete picture of the host response to influenza, the meth-

odology also allowed for identification of useful biomarkers

that indicate the immunological status of an active infection.

Linoleic acid is further metabolized into 9- or 13-hydroxylated

linoleic acid (9-hydroxyoctadecadienoic acid [HODE] and 13-

HODE). While 13-HODE is generated by 15-lipoxygenase,

9-HODE production can be catalyzed by multiple enzymes and

nonenzymatic reactions (Obinata and Izumi, 2009; Figure 7A).

9-HODE and 13-HODE have previously been proposed as

markers for lipid peroxidation in various chronic diseases

(Spiteller and Spiteller, 1997). More recently, 9-HODE has been

shown to have a proinflammatory role (Hattori et al., 2008;

Obinata and Izumi, 2009), while 13-HODE is anti-inflammatory

(Altmann et al., 2007; Belvisi and Mitchell, 2009). We calculated

the ratio of 13- to 9-HODE to determine whether it accurately de-

scribes the pro- versus anti-inflammatory state of an infection.

An advantage of using ratios ofmediators is that, while the values

of individual lipid mediators may vary greatly between different

animals and different patients, the ratio between two mediators

is normalized, as they arise from a common precursor. Another

reason for choosing 9- and 13-HODE is that both mediators

were detected robustly across all samples (29/30 clinical sam-

ples for 9-HODE and 30/30 clinical samples for 13-HODE). In

contrast, many of the other mediators were undetectable in

either the low or medium response groups, making them less

suitable as biomarkers.

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. 223

When we compared the levels of 9-HODE, 13-HODE, and the

ratio of 13-HODE/9-HODE during influenza infection in mice,

PR8sublethal infection induced a significantly increased level of

9-HODE on day 7 postinfection compared to X31sublethal infec-

tion. In contrast, both PR8sublethal and X31sublethal induced similar

production of 13-HODE. Interestingly, the ratios of 13-:9-HODE

are increased for both PR8sublethal and X31sublethal during the

resolution phase of the infection, indicating an increasingly

anti-inflammatory state. Moreover, the ratio is significantly

higher in X31sublethal, even at early time points (day 7 and 9 post-

infection), which supports the notion that the low-pathogenicity

virus induced a response skewed toward an anti-inflammatory

profile (Figure 7B).

We tested the robustness of these parameters by combining

three additional experiments performed on separate days and

animals from different vendors and harvested at different time

points. While the raw values of 9- and 13-HODE varied between

animals and experiments, the ratio of 13-:9-HODE remained

consistent (Figures S7A and S7B).

In the human nasopharyngeal lavage, levels of 9-HODE were

slightly lower in the high response group, although the difference

was not significant (Figure 7C), while levels of the anti-inflamma-

tory mediator, 13-HODE, were elevated with moderate signifi-

cance (p < 0.05). When we compared the 13-:9-HODE ratios

between the groups, however, the difference became more dra-

matic and highly significant (p < 0.001) (Figure 7C). Further mea-

surements on additional samples and clinical data will be

required to determine whether the 13-:9-HODE ratio can be

used to further stratify the severity and/or course of infection.

DISCUSSION

Lipidomic profiling of influenza infection provided a comprehen-

sive map of eicosanoids and other related bioactive lipids during

an active infection of the lung. We detected and quantified

metabolites andmediators within the cyclooxygenase, lipoxyge-

nase, and cytochrome P450 pathways derived from arachidonic

acid, as well as metabolites and mediators derived from linoleic

acid, linolenic acid, DHA, and EPA. The protein levels of cyto-

kines and chemokines and the transcript levels of numerous anti-

viral genes suggested a straightforward, positive correlation

between the level of immune response and the pathogenicity

of the infection, with the response increasing from infection

with X31, the low-pathogenicity virus, to PR8, the high-pathoge-

nicity virus (PR8sublethal to PR8lethal). In contrast, the lipidomic

profiling uncovered a more complex response, with the levels

of mediators involved in both pro- and anti-inflammatory pro-

cesses exhibiting more complex dynamics. Surprisingly, the

cyclooxygenase pathway mediators, consisting of the classical

signaling prostaglandins, were produced with similar kinetics

during infections by the different influenza strains. Concordantly,

enzymes within the COX pathway were activated to similar levels

transcriptionally. In contrast to the COX-derived lipid mediators,

metabolites derived from other pathways (LOX and CYP450) and

precursors (linoleic acid, DHA, and EPA) diverged strongly

between the different infections.

Hepoxilin and 12-HETE, mediators derived from the 12/15

lipoxygenase pathway, were detected at a higher level in

224 Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc.

X31sublethal than PR8sublethal-infected animals. Hepoxilin has

been suggested to be involved in anti-inflammatory mecha-

nisms, and stable hepoxilin analogs inhibit macrophage influx

as well as fibrosis in the lung (Jankov et al., 2002). 12-HETE

also has anti-inflammatory activities: for example, it is capable

of blocking TNFa-induced IL-6 secretion from macrophages

(Kronke et al., 2009). In contrast, 8-HETE, which is derived

from 8-lipoxygenase, was detected at a higher level in

PR8sublethal- than X31sublethal-infected animals. While the biolog-

ical activity of 8-HETE within the respiratory system is poorly

characterized, the metabolite has been detected in human lung

vasculature (Kiss et al., 2000). The increased production of

8-HETE late in infection suggests that it plays a role in epithelial

repair after influenza infection. It is important to note that mice do

not have separate 12- and 15-lipoxygenases but rather a com-

bined mouse 12/15-lipoxygenase, which is related in sequence

to the human 12-lipoxygenase as well as human 15-lipoxyge-

nase type I (reticulocyte-type) and type II (epidermis-type) en-

zymes (Kuhn et al., 2002).

Overall, while metabolites of the LOX pathways displayed a

heterogeneous profile in the mouse infection model, a clear dif-

ference between PR8 and X31 infection was found primarily in

metabolites contributing to anti-inflammatory and proresolution

activities. In human samples, however, a general increase was

observed in both the proinflammatory (5-lipoxygenase-derived)

and anti-inflammatory (12-, 15-, and 8-lipoxygenase-derived)

metabolites of the LOX pathway in high responders. Interest-

ingly, when comparing the percentages of mediators derived

from different lipoxygenase enzymes, a consistent pattern

emerged in the mouse and human data. In the mouse model,

the percentage of 5-lipoxygenase metabolites was elevated dur-

ing the pathogenic phase of the PR8 infection, whereas the

percentage of 12/15-lipoxygenase metabolites was elevated

during the resolution phase of the X31 infection. In the human

samples, increasing percentages of 5-lipoxygenase-derived

metabolites and decreasing percentages of 12-lipoxygenase-

derived metabolites correlated with increasing clinical symp-

toms and immune response. However, a direct comparison

between the mouse and human data is difficult, due to the lack

of information on the time course and pathogenicity of the viral

strains in the human infection samples as well as the unknown

contribution of host genetic diversity to susceptibility to viral

infection. When infected with influenza, 12/15-lipoxygenase

gene-deficient mice recovered more slowly than wild-type

mice, similar to the observations of Morita et al. (2013; data not

shown). However, 12/15 gene-deficient mice are of limited utility

in modeling human infections, as the function of this enzyme is

performed in humans by at least four different enzymes that

are expressed in different cell types. Detailed pharmacological

studies using specific inhibitors would be required to determine

the effect of these enzymes at different phases of infection.

Cytochrome P450 enzymes can metabolize arachidonic acid

to produce 5,6- or 14,15-epoxyeicosatrienoic acid (EET) with

subsequent metabolic transformation to diHETrE by the soluble

epoxide hydrolase (sEH). EETs have potent vasodilator activ-

ities, and they have been determined to have anti-inflammatory

activities by activating the PPARa pathway (Thomson et al.,

2012; Wray et al., 2009). CypP450 metabolites were detected

at significantly higher levels in PR8sublethal infections than in

X31sublethal-infected animals. Similar to arachidonic acid-derived

EET and diHETrE, cytochrome P450 enzymes can metabolize

linoleic acid to produce 9,10 or 12,13 EpOME, and sEH further

converts them to 9,10- or 12, 13-diHOME. The epoxides of

linoleic acid are leukotoxins produced by activated neutrophils

and macrophages. Elevated levels are associated with acute

respiratory distress syndrome (ARDS) and cause pulmonary

edema, vasodilation, and cardiac failure in animal models (Ishi-

zaki et al., 1995, 1999). These EpOME mediators and their

sEH-derivedmediators, diHOME, are chemotactic to neutrophils

and exert their toxicity by disrupting mitochondrial functions

(Sisemore et al., 2001; Totani et al., 2000). The elevated levels

of thesemediators in animals infected with the highly pathogenic

influenza strain may explain the severe damage occurring in the

lung during infection.

Recently, a new class of endogenous lipid mediators (derived

from DHA and EPA) that promote resolution of inflammation pro-

vides great promise to treat inflammatory diseases (Ji et al.,

2011). Although assays for resolvins and protectins were

included in the panel, these anti-inflammatory/ proresolution

molecules were not detected in the animal model or human clin-

ical samples. This is interesting because the production of PD1 is

suppressed in lung tissues of influenza-infected mice at early

time points (Morita et al., 2013). Additionally, PD1 and 17-HDoHE

were detected at higher levels in exhaled breath condensate

from healthy individuals compared to asthma patients (Levy

et al., 2007). Further studies will determine whether these differ-

ences arise from the different sampling methods employed (BAL

versus lung tissue; nasopharyngeal lavage versus exhaled

breath condensate). Aside from different tissue, the infection

model and the time frame of sampling is quite different between

our study and previous reports. Even though our profiling

approach was comprehensive, the biological activities of several

of the metabolites detected in our screens are unknown, and

further systematic studies will be necessary to determine their

involvement in the inflammatory process and return to homeo-

stasis within the respiratory system.

By creating quantitative pro- and anti-inflammatory indices,

we defined lipid profiles that readily distinguished infections by

the low-pathogenicity strain, X31, and the highly pathogenic

strain, PR8. X31sublethal induced a normal sequence of events,

with induction of a proinflammatory state on day 3 after infection,

followed by induction of an anti-inflammatory state on day 7,

which restored homeostasis to the system. Interestingly, while

PR8sublethal also induced a proinflammatory state on day 3, there

was a second, more robust induction of proinflammatory medi-

ators on day 7, occurring simultaneously with the induction of

anti-inflammatory mediators. This overlap of pro- and anti-

inflammatory states was also quite apparent in the PR8lethalinfection, in which both indices were significantly higher than

the sublethal infections on day 3 and 5 postinfection. The clinical

data were similar to the mouse data, suggesting that pro- and

anti-inflammatory states overlap within infected humans as

well. Further studies are warranted to determine whether this

simultaneous pro- and anti-inflammatory lipid response is bene-

ficial or detrimental to the host during an infection with a highly

pathogenic influenza strain.

From the mouse lipidome, we have identified 9-HODE, 13-

HODE, and their ratio as potential biomarkers for immune status

during an infection. While there are numerous parameters to

assess the proinflammatory status of an infection, parameters

characterizing the resolution of inflammation are lacking.

Because these metabolites can be robustly detected by ELISA

and are derived from a common precursor, making their abun-

dance ratio self-normalizing, they offer great potential as a

biomarker of immune status during influenza infection. A larger

cohort of patients and samples is required to confirm the corre-

lation between the ratio of 13-:9-HODE and the status of disease

and recovery.

Lipidomic profiling provides an additional parameter for un-

derstanding the host response during a microbial infection.

Although transcriptomic analysis of the response to infection

has been conducted routinely, our data suggest that, while

some transcript levels of the enzymes correlate with the lipid

levels, the majority of the changes seen in the lipid response

were regulated posttranscriptionally. It is likely that the response

to infection is accelerated by the uncoupling of metabolite pro-

duction from the transcriptional regulation of the specific up-

stream enzymes.

EXPERIMENTAL PROCEDURES

Mouse Influenza Infection

Female C57Bl/6 animals (8–12 weeks) (Jackson Laboratory and Charles River)

were kept in specific pathogen-free conditions. Experiments conducted were

approved by the Institute for Systems Biology Institutional Animal Care and

Use Committee. Animals were anesthetized with a ketamine/xylazine mixture

and infected intranasally with 200 or 23 105 pfu of influenza virus strain PR8 or

23 105 pfu of X31 in 30 ml sterile PBS. Mock-infected animals were inoculated

with 30 ml sterile PBS. Mice were monitored and weighed daily.

Human Clinical Samples

This study was carried out in accordance with Good Clinical Practice (GCP) as

required by the U.S. Code of Federal Regulations applicable to clinical studies

(45 CFR 46), the International Commission on Harmonization guidelines for

GCP E6, and the National Institutes of Health (NIH) Clinical Terms of Award.

Children and adults with influenza-like illness receiving medical counsel

were approached by study personnel to participate. Study participants were

enrolled based on detection of the influenza matrix gene (influenza A) or

nonstructural NS1 gene (influenza B) by quantitative real-time PCR. Individuals

with acute respiratory illness meeting the case definition for influenza (fever or

feverishness accompanied by cough or sore throat) whowere symptomatic for

96 hr or less and their household or family contacts (symptomatic or asymp-

tomatic) were eligible to participate in the study and followed for approximately

1 month postenrollment. Contacts that became influenza-positive were

sampled cases after conversion. Nasopharyngeal swabs and lavages were

collected upon enrollment and at each subsequent study visit.

Lipidomic Profiling by LC Mass Spectrometry

Lipid mediators were analyzed by LC and MS essentially as described previ-

ously (Dumlao et al., 2011; Quehenberger et al., 2010). Briefly, 0.9 ml of BAL

or nasal washes were supplemented with a mix consisting of 25 deuterated in-

ternal standards (Cayman Chemical) and lipid metabolites isolated by solid

phase extraction on a Strata X column (Phenomenex). The extracted samples

were evaporated, reconstituted in a small volume, and the eicosanoids were

separated by reverse phase LC using a Synergy C18 column (Phenomenex).

The eicosanoids were analyzed by tandem quadrupole MS (MDS SCIEX

4000 QTRAP, Applied Biosystems) operated in the negative-ionization mode

via multiple-reaction monitoring (MRM) using transitions that were optimized

for selectivity and sensitivity. Authentic standards were analyzed under

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. 225

identical conditions, and eicosanoid quantitation was achieved by the stable

isotope dilution method and comparison with 141 quantitation standards.

Data analysis was performed using the MultiQuant 2.1 software (Applied Bio-

systems). A complete list of all metabolites included in theMRMprofile is given

in Table S2.

Statistical Analysis

The two-tailed Mann-Whitney test was used to determine statistical signifi-

cance between population means (for mouse infections, red: X31sublethal;

blue: PR8sublethal; black: PR8lethal. Magenta asterisks denote significance

between PR8lethal and both X31sublethal and PR8sublethal infections; blue and

red asterisks denote significance between PR8lethal and PR8sublethal or

X31sublethal infections, respectively; and black asterisks denote significance

between PR8sublethal and X31sublethal infections). Statistical significance:

*0.01 < p < 0.05; **0.001 < p < 0.01; ***p < 0.001. For correlation studies, the

two-tailed Spearman test was used to calculate correlation significance.

Heatmaps and clustering analysis were generated using Genedata and Java

TreeView.

Data Dissemination

Raw files and processed data for lipidomics profilings are available at the

Systems Influenza Website (http://www.systemsinfluenza.org) and at Metab-

olights (http://www.ebi.ac.uk/metabolights/; MTBLS31).

ACCESSION NUMBERS

The Metabolights accession number for the lipidomics profilings reported in

this paper is MTBLS31.

SUPPLEMENTAL INFORMATION

Supplemental Information includes Extended Experimental Procedures, seven

figures, and two tables and can be foundwith this article online at http://dx.doi.

org/10.1016/j.cell.2013.05.052.

ACKNOWLEDGMENTS

We thank Alan Diercks and Peter Askovich for helpful advice and critical

reading of the manuscript. This work was supported by National Institute of

Allergy and Infectious Diseases Contract no. HHSN272200800058C, ‘‘A Sys-

tems Biology Approach to Infectious Disease Research’’, the American Leba-

nese Syrian Associated Charities (ALSAC), the NIH/NIAIDCenter of Excellence

for Influenza Research and Surveillance (HHSN266200700005C), and NIGMS

U54 GM069338 (to E.A.D.).

Received: March 7, 2013

Revised: April 8, 2013

Accepted: May 20, 2013

Published: July 3, 2013

REFERENCES

Allan, W., Tabi, Z., Cleary, A., and Doherty, P.C. (1990). Cellular events in the

lymph node and lung of mice with influenza. Consequences of depleting

CD4+ T cells. J. Immunol. 144, 3980–3986.

Altmann, R., Hausmann,M., Spottl, T., Gruber, M., Bull, A.W.,Menzel, K., Vogl,

D., Herfarth, H., Scholmerich, J., Falk, W., and Rogler, G. (2007). 13-Oxo-ODE

is an endogenous ligand for PPARgamma in human colonic epithelial cells.

Biochem. Pharmacol. 74, 612–622.

Ariel, A., and Serhan, C.N. (2007). Resolvins and protectins in the termination

program of acute inflammation. Trends Immunol. 28, 176–183.

Bannenberg, G.L., Chiang, N., Ariel, A., Arita, M., Tjonahen, E., Gotlinger,

K.H., Hong, S., and Serhan, C.N. (2005). Molecular circuits of resolution:

formation and actions of resolvins and protectins. J. Immunol. 174, 4345–

4355.

226 Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc.

Belvisi, M.G., andMitchell, J.A. (2009). Targeting PPAR receptors in the airway

for the treatment of inflammatory lung disease. Br. J. Pharmacol. 158, 994–

1003.

Buczynski, M.W., Dumlao, D.S., and Dennis, E.A. (2009). Thematic Review

Series: Proteomics. An integrated omics analysis of eicosanoid biology.

J. Lipid Res. 50, 1015–1038.

Chiang, N., Fredman, G., Backhed, F., Oh, S.F., Vickery, T., Schmidt, B.A., and

Serhan, C.N. (2012). Infection regulates pro-resolving mediators that lower

antibiotic requirements. Nature 484, 524–528.

Dennis, E.A., Deems, R.A., Harkewicz, R., Quehenberger, O., Brown, H.A.,

Milne, S.B., Myers, D.S., Glass, C.K., Hardiman, G., Reichart, D., et al.

(2010). A mouse macrophage lipidome. J. Biol. Chem. 285, 39976–39985.

Drake, D.R., Brogden, K.A., Dawson, D.V., and Wertz, P.W. (2008). Thematic

review series: skin lipids. Antimicrobial lipids at the skin surface. J. Lipid

Res. 49, 4–11.

Dumlao, D.S., Buczynski, M.W., Norris, P.C., Harkewicz, R., and Dennis, E.A.

(2011). High-throughput lipidomic analysis of fatty acid derived eicosanoids

and N-acylethanolamines. Biochim. Biophys. Acta 1811, 724–736.

Francis, T., and De Torregrosa, M.V. (1945). Combined infection of Mice with

H. influenzae and influenza virus by the intranasal route. J. Infect. Dis. 76,

70–77.

Godson, C., Mitchell, S., Harvey, K., Petasis, N.A., Hogg, N., and Brady, H.R.

(2000). Cutting edge: lipoxins rapidly stimulate nonphlogistic phagocytosis of

apoptotic neutrophils by monocyte-derived macrophages. J. Immunol. 164,

1663–1667.

Hattori, T., Obinata, H., Ogawa, A., Kishi, M., Tatei, K., Ishikawa, O., and Izumi,

T. (2008). G2A plays proinflammatory roles in human keratinocytes under

oxidative stress as a receptor for 9-hydroxyoctadecadienoic acid. J. Invest.

Dermatol. 128, 1123–1133.

Ishizaki, T., Shigemori, K., Nakai, T., Miyabo, S., Ozawa, T., Chang, S.W., and

Voelkel, N.F. (1995). Leukotoxin, 9,10-epoxy-12-octadecenoate causes

edematous lung injury via activation of vascular nitric oxide synthase. Am. J.

Physiol. 269, L65–L70.

Ishizaki, T., Ozawa, T., and Voelkel, N.F. (1999). Leukotoxins and the lung.

Pulm. Pharmacol. Ther. 12, 145–155.

Jagger, B.W., Wise, H.M., Kash, J.C., Walters, K.A., Wills, N.M., Xiao, Y.L.,

Dunfee, R.L., Schwartzman, L.M., Ozinsky, A., Bell, G.L., et al. (2012). An over-

lapping protein-coding region in influenza A virus segment 3 modulates the

host response. Science 337, 199–204.

Jankov, R.P., Luo, X., Demin, P., Aslam, R., Hannam, V., Tanswell, A.K., and

Pace-Asciak, C.R. (2002). Hepoxilin analogs inhibit bleomycin-induced pulmo-

nary fibrosis in the mouse. J. Pharmacol. Exp. Ther. 301, 435–440.

Ji, R.-R., Xu, Z.-Z., Strichartz, G., and Serhan, C.N. (2011). Emerging roles of

resolvins in the resolution of inflammation and pain. Trends Neurosci. 34,

599–609.

Kash, J.C., Tumpey, T.M., Proll, S.C., Carter, V., Perwitasari, O., Thomas,M.J.,

Basler, C.F., Palese, P., Taubenberger, J.K., Garcıa-Sastre, A., et al. (2006).

Genomic analysis of increased host immune and cell death responses induced

by 1918 influenza virus. Nature 443, 578–581.

Kiss, L., Schutte, H., Mayer, K., Grimm, H., Padberg, W., Seeger, W., and

Grimminger, F. (2000). Synthesis of arachidonic acid-derived lipoxygenase

and cytochrome P450 products in the intact human lung vasculature. Am. J.

Respir. Crit. Care Med. 161, 1917–1923.

Kronke, G., Katzenbeisser, J., Uderhardt, S., Zaiss, M.M., Scholtysek, C.,

Schabbauer, G., Zarbock, A., Koenders, M.I., Axmann, R., Zwerina, J., et al.

(2009). 12/15-lipoxygenase counteracts inflammation and tissue damage in

arthritis. J. Immunol. 183, 3383–3389.

Kuhn, H., and O’Donnell, V.B. (2006). Inflammation and immune regulation by

12/15-lipoxygenases. Prog. Lipid Res. 45, 334–356.

Kuhn, H., Walther, M., and Kuban, R.J. (2002). Mammalian arachidonate 15-

lipoxygenases structure, function, and biological implications. Prostaglandins

Other Lipid Mediat. 68-69, 263–290.

Levy, B.D., Kohli, P., Gotlinger, K., Haworth, O., Hong, S., Kazani, S., Israel, E.,

Haley, K.J., and Serhan, C.N. (2007). Protectin D1 is generated in asthma and

dampens airway inflammation and hyperresponsiveness. J. Immunol. 178,

496–502.

Morgan, L.T., Thomas, C.P., Kuhn, H., and O’Donnell, V.B. (2010). Thrombin-

activated human platelets acutely generate oxidized docosahexaenoic-acid-

containing phospholipids via 12-lipoxygenase. Biochem. J. 431, 141–148.

Morita, M., Kuba, K., Ichikawa, A., Nakayama, M., Katahira, J., Iwamoto, R.,

Watanebe, T., Sakabe, S., Daidoji, T., Nakamura, S., et al. (2013). The lipid

mediator protectin D1 inhibits influenza virus replication and improves severe

influenza. Cell 153, 112–125.

Neumann, G., Noda, T., and Kawaoka, Y. (2009). Emergence and pandemic

potential of swine-origin H1N1 influenza virus. Nature 459, 931–939.

Obinata, H., and Izumi, T. (2009). G2A as a receptor for oxidized free fatty

acids. Prostaglandins Other Lipid Mediat. 89, 66–72.

O’Flaherty, J.T., Hu, Y., Wooten, R.E., Horita, D.A., Samuel, M.P., Thomas,

M.J., Sun, H., and Edwards, I.J. (2012). 15-lipoxygenase metabolites of doco-

sahexaenoic acid inhibit prostate cancer cell proliferation and survival. PLoS

ONE 7, e45480.

Quehenberger, O., and Dennis, E.A. (2011). The human plasma lipidome.

N. Engl. J. Med. 365, 1812–1823.

Quehenberger, O., Armando, A.M., Brown, A.H., Milne, S.B., Myers, D.S.,

Merrill, A.H., Bandyopadhyay, S., Jones, K.N., Kelly, S., Shaner, R.L., et al.

(2010). Lipidomics reveals a remarkable diversity of lipids in human plasma.

J. Lipid Res. 51, 3299–3305.

Ricote, M., Li, A.C., Willson, T.M., Kelly, C.J., and Glass, C.K. (1998). The

peroxisome proliferator-activated receptor-gamma is a negative regulator of

macrophage activation. Nature 391, 79–82.

Sato, M., Suemori, H., Hata, N., Asagiri, M., Ogasawara, K., Nakao, K.,

Nakaya, T., Katsuki, M., Noguchi, S., Tanaka, N., and Taniguchi, T. (2000).

Distinct and essential roles of transcription factors IRF-3 and IRF-7 in response

to viruses for IFN-alpha/beta gene induction. Immunity 13, 539–548.

Serhan, C.N., Maddox, J.F., Petasis, N.A., Akritopoulou-Zanze, I., Papayianni,

A., Brady, H.R., Colgan, S.P., andMadara, J.L. (1995). Design of lipoxin A4 sta-

ble analogs that block transmigration and adhesion of human neutrophils.

Biochemistry 34, 14609–14615.

Serhan, C.N., Clish, C.B., Brannon, J., Colgan, S.P., Chiang, N., and Gronert,

K. (2000). Novel functional sets of lipid-derived mediators with antiinflamma-

tory actions generated from omega-3 fatty acids via cyclooxygenase

2-nonsteroidal antiinflammatory drugs and transcellular processing. J. Exp.

Med. 192, 1197–1204.

Serhan, C.N., Chiang, N., and Van Dyke, T.E. (2008). Resolving inflammation:

dual anti-inflammatory and pro-resolution lipid mediators. Nat. Rev. Immunol.

8, 349–361.

Sisemore, M.F., Zheng, J., Yang, J.C., Thompson, D.A., Plopper, C.G., Corto-

passi, G.A., and Hammock, B.D. (2001). Cellular characterization of leukotoxin

diol-induced mitochondrial dysfunction. Arch. Biochem. Biophys. 392, 32–37.

Spiteller, P., and Spiteller, G. (1997). 9-Hydroxy-10,12-octadecadienoic acid

(9-HODE) and 13-hydroxy-9,11-octadecadienoic acid (13-HODE): excellent

markers for lipid peroxidation. Chem. Phys. Lipids 89, 131–139.

Taubenberger, J.K., and Kash, J.C. (2010). Influenza virus evolution, host

adaptation, and pandemic formation. Cell Host Microbe 7, 440–451.

Thomson, S.J., Askari, A., and Bishop-Bailey, D. (2012). Anti-inflammatory

effects of epoxyeicosatrienoic acids. Int. J. Vasc. Med. 2012, 605101.

Tobin, D.M., Roca, F.J., Oh, S.F., McFarland, R., Vickery, T.W., Ray, J.P., Ko,

D.C., Zou, Y., Bang, N.D., Chau, T.T.H., et al. (2012). Host genotype-specific

therapies can optimize the inflammatory response tomycobacterial infections.

Cell 148, 434–446.

Totani, Y., Saito, Y., Ishizaki, T., Sasaki, F., Ameshima, S., and Miyamori, I.

(2000). Leukotoxin and its diol induce neutrophil chemotaxis through signal

transduction different from that of fMLP. Eur. Respir. J. 15, 75–79.

Wray, J.A., Sugden, M.C., Zeldin, D.C., Greenwood, G.K., Samsuddin, S.,

Miller-Degraff, L., Bradbury, J.A., Holness, M.J., Warner, T.D., and Bishop-

Bailey, D. (2009). The epoxygenases CYP2J2 activates the nuclear receptor

PPARalpha in vitro and in vivo. PLoS ONE 4, e7421.

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. 227

Supplemental Information

EXTENDED EXPERIMENTAL PROCEDURES

Cells and Bronchoalveolar Lavage IsolationMice were euthanized by CO2 inhalation. Broncho-alveolar lavage (BAL) was collected in 2 3 1 ml washes with HBSS buffer. Cells

from the BAL were collected by light centrifugation. Distinct cell types from the dissected lungs were isolated as described (Gereke

et al., 2007; Marsh et al., 2009; Snelgrove et al., 2008). Dispase digested lung tissue was disrupted into single-cell suspensions by

passage through 100-mm, 70-mm, and 40-mm filters. Cells were incubated with anti-mouse CD16/32 (UCSF Monoclonal Antibody

Core) and anti-mouse CD45 microbeads (Milteny) and separated by magnetic separation. RNA was extracted from distinct cell pop-

ulations using TRIzol reagent (Invitrogen).

Proteomic Quantitation by LuminexMultiplex quantitation of chemokines and cytokines was conducted using the Luminex xMAP system with a MILLIPLEXMAP human

cytokine/chemokine immunoassay panel or Mouse Cytokine/Chemokine panel as described by manufacturer (Millipore).

Histological and Immunohistochemical AnalysisDissected mouse lungs (left lobes) were fixed in 10% neutral-buffered formalin, processed routinely into paraffin and stained with

hematoxylin and eosin (H&E) or anti-influenza antibody (Meridian Life Science) followed by a rabbit anti Goat IgG secondary antibody

(Jackson ImmunoResearch). The isotype control antibody was purified Goat IgG (Invitrogen). Staining was performed using the Leica

Bond Automated Immunostainer and signal detected with 3,3-Diaminobenzidine (DAB, brown) and counterstainedwith Harris hema-

toxylin. H&E and IHC-stained slides were digitized with the Olympus Nanozoomer and images captured with Nikon Digital Pathology

viewing software.

Transcriptional Profiling Using Fluidigm RT-PCRRNA sample was treated with TURBO DNase (Ambion) before reverse transcription with Superscript II Reverse Transcriptase

(Invitrogen). For pre-amplification, 20x TaqMan Gene Expression Assays of selected genes (Table S1) were mixed together to a final

of concentration of 0.2x for each gene. This mix was used to set up amplification reactions of the reverse-transcribed samples. Each

Fluidigm 48.48 Dynamic array was first primed with Fluidigm’s IFC Controller MX, loaded with 48 pre-amplified samples, and the

primer/probe pairs for the selected genes described above and in the Extended Experimental Procedures. The array was run in

the BioMark HD System and data were collected using the BioMark Data Collection Software. Subsequent analysis was performed

with the Fluidigm Real-Time PCR Analysis software.

Calculation of Pro- and Anti-Inflammatory IndicesFold change was calculated using the median of mock-infected animals (in themousemodel) or of the low response group (in human

samples). Only values that were greater than or equal to two were included. The fold changes were then normalized to the maximum

value across all samples, setting the maximum value to one. Mediators were divided into two groups based on having either pro- or

anti-inflammatory activity. Pro-inflammatory mediators include 6k PGF1a, PGF2a, PGE2, PGD2, PGF1a, 6, 15 dk dh PGF1a, 15k

PGF1a, 15k PGF2a, 15k PGE2, dhk, PGE2, dhk PGD2, LTB4, LTE4, 5 HETE, 9 HODE, 9 10 EpOME, 12 13 EpOME, 9 10 diHOME,

12 13 diHOME. Anti-inflammatory mediators include PGA2, PGB2, tetranor 12 R HETE, PGE1, 15 HETE, 8 HETE, 12 HETE, HXB3, 16

HETE, 5 6 EET, 14 15 EET, 11 12 diHETrE, 14 15 diHETrE, 13 HODE, 13 HOTrE, 15 HETrE, 13 HDoHE, 8 HDoHE, 16 HDoHE, 20

HDoHE, 7 HDoHE, 4 HDoHE, 17 HDoHE, 10 HDoHE, 14 HDoHE, 19 20 EpDPE, 19 20 DiHDPA, 15 HEPE, 12 HEPE. We then added

together the normalized value of each sample within the pro- and anti-inflammatory groups to define the total activity. The maximum

activity in each group is equal to the total number ofmediators in that group. To generate the index for each time point, the percentage

of pro- and anti-inflammatory mediator activity was calculated by dividing the measured activity by maximum possible activity in the

each group.

We included most of the prostaglandins in having pro-inflammatory activities because of their vasomodulatory activities (Holtz-

man, 1992; Sheibanie et al., 2007). Leukotrienes have chemoattractant activities and are potent mediators for immediate hypersen-

sitivity, bronchoconstriction, smooth muscle contraction, and increased vascular permeability (Peters-Golden et al., 2006). 5 HETE

has been shown to induce airway contraction and potentiate neutrophil transcellular migration (Bittleman and Casale, 1995). 9 HODE

has been attributed to have a proinflammatory role, by activating G2A, a G protein-coupled receptor, which mediates intracellular

calcium mobilization and JNK activation (Hattori et al., 2008; Obinata and Izumi, 2009). The epoxides of linoleic acid (9,10 EpOME

and 12,13 EpOME) are leukotoxins produced by activated neutrophils and macrophages. Elevated levels are associated with

ARDS (acute respiratory distress syndrome) and caused pulmonary edema, vasodilation, and cardiac failure in animal models

(Ishizaki et al., 1999, 1995). These metabolites and their soluble epoxide hydrolase metabolites (DiHOME) are chemotactic to neu-

trophils and exert their toxicity by disrupting mitochondrial functions (Sisemore et al., 2001; Totani et al., 2000).

PGA2 and PGB2, non-enzymatic derivatives of PGE2, were included to having anti-inflammatory activities. PGA2 is known to sup-

press activation of microglia and astrocytes in the experimental autoimmune encephalomyelitis (EAE) model and activates the

intrinsic apoptotic pathway (Lee et al., 2010; Storer et al., 2005). PGB2 has been shown to induce pulmonary hypertension, an oppo-

site effect of PGE2 (Liu et al., 1994). PGE1 has been shown to inhibit the acute inflammatory edema response by activating the EP3

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. S1

receptor (Ahluwalia and Perretti, 1994). 12- and 15-HETE also have anti-inflammatory activities by blocking TNFa-induced IL-6

secretion from macrophages (Kronke et al., 2009). 16-HETE has been shown to inhibit human polymorphonuclear cells adhesion

and aggregation, as well as decreasing LTB4 synthesis (Bednar et al., 2000). HXB3 and stable hepoxilin analogs inhibit macrophage

influx as well as fibrosis in the lung (Jankov et al., 2002). EET and diHETrE have been determined to have anti-inflammatory activities

by activating the peroxisome proliferator-activated receptor alpha (PPARa) pathway (Thomson et al., 2012; Wray et al., 2009); (Deng

et al., 2010). Specifically, 5,6 EET and 11,12 diHETrE have been shown to prevent leukocyte adhesion to the vascular wall

{{Node:1999wp}}. 14,15 EET has an anti-inflammatory role by preventing TNFa-induced IkBa degradation in human airway smooth

muscle cells (Morin et al., 2008). 14,15 diHETrE is a potent activator of PPARa that contribute to its anti-inflammatory activities (Fang

et al., 2006). 13 HODE is an agonist for PPARg and having an anti-inflammatory role (Altmann et al., 2007; Belvisi and Mitchell, 2009;

Emerson and LeVine, 2004; Fritsche, 2008; Stoll et al., 1994). Lastly, we included metabolites derived from DHA and EPA into having

anti-inflammatory activities. Some of these characterizedmediators have direct and potent effects on the anti-inflammatory and pro-

resolution processes (Groeger et al., 2010; Lima-Garcia et al., 2011). While other metabolites are generated by auto-oxidation or

other non-enzymatic routes, they can also be enzymatically metabolized into generating potent mediators, such as resolvins, pro-

tectins, and maresins.

SUPPLEMENTAL REFERENCES

Ahluwalia, A., and Perretti, M. (1994). Anti-inflammatory effect of prostanoids inmouse and rat skin: evidence for a role of EP3-receptors. J. Pharmacol. Exp. Ther.

268, 1526–1531.

Bednar, M.M., Gross, C.E., Russell, S.R., Fuller, S.P., Ahern, T.P., Howard, D.B., Falck, J.R., Reddy, K.M., and Balazy, M. (2000). 16(R)-hydroxyeicosatetraenoic

acid, a novel cytochrome P450 product of arachidonic acid, suppresses activation of human polymorphonuclear leukocyte and reduces intracranial pressure in a

rabbit model of thromboembolic stroke. Neurosurgery 47, 1410–1418.

Bittleman, D.B., and Casale, T.B. (1995). 5-Hydroxyeicosatetraenoic acid (HETE)-induced neutrophil transcellular migration is dependent upon enantiomeric

structure. Am. J. Respir. Cell Mol. Biol. 12, 260–267.

Deng, Y., Theken, K.N., and Lee, C.R. (2010). Cytochrome P450 epoxygenases, soluble epoxide hydrolase, and the regulation of cardiovascular inflammation.

J. Mol. Cell. Cardiol. 48, 331–341.

Emerson, M.R., and LeVine, S.M. (2004). Experimental allergic encephalomyelitis is exacerbated in mice deficient for 12/15-lipoxygenase or 5-lipoxygenase.

Brain Res. 1021, 140–145.

Fang, X., Hu, S., Xu, B., Snyder, G.D., Harmon, S., Yao, J., Liu, Y., Sangras, B., Falck, J.R., Weintraub, N.L., and Spector, A.A. (2006). 14,15-Dihydroxyeicosa-

trienoic acid activates peroxisome proliferator-activated receptor-alpha. Am. J. Physiol. Heart Circ. Physiol. 290, H55–H63.

Fritsche, K.L. (2008). Too much linoleic acid promotes inflammation-doesn’t it? Prostaglandins Leukot. Essent. Fatty Acids 79, 173–175.

Gereke, M., Grobe, L., Prettin, S., Kasper, M., Deppenmeier, S., Gruber, A.D., Enelow, R.I., Buer, J., and Bruder, D. (2007). Phenotypic alterations in type II alve-

olar epithelial cells in CD4+ T cell mediated lung inflammation. Respir. Res. 8, 47.

Groeger, A.L., Cipollina, C., Cole, M.P., Woodcock, S.R., Bonacci, G., Rudolph, T.K., Rudolph, V., Freeman, B.A., and Schopfer, F.J. (2010). Cyclooxygenase-2

generates anti-inflammatory mediators from omega-3 fatty acids. Nat. Chem. Biol. 6, 433–441.

Holtzman, M.J. (1992). Arachidonic acid metabolism in airway epithelial cells. Annu. Rev. Physiol. 54, 303–329.

Lee, S.-Y., Ahn, J.-H., Ko, K.W., Kim, J., Jeong, S.W., Kim, I.-K., Kim, J., and Kim, H.-S. (2010). Prostaglandin A2 activates intrinsic apoptotic pathway by direct

interaction with mitochondria in HL-60 cells. Prostaglandins Other Lipid Mediat. 91, 30–37.

Lima-Garcia, J.F., Dutra, R.C., da Silva, K., Motta, E.M., Campos, M.M., and Calixto, J.B. (2011). The precursor of resolvin D series and aspirin-triggered resolvin

D1 display anti-hyperalgesic properties in adjuvant-induced arthritis in rats. Br. J. Pharmacol. 164, 278–293.

Liu, F., Orr, J.A., and Wu, J.Y. (1994). Prostaglandin B2-induced pulmonary hypertension is mediated by TxA2/PGH2 receptor stimulation. Am. J. Physiol. 267,

L602–L608.

Marsh, L.M., Cakarova, L., Kwapiszewska, G., von Wulffen, W., Herold, S., Seeger, W., and Lohmeyer, J. (2009). Surface expression of CD74 by type II alveolar

epithelial cells: a potential mechanism for macrophage migration inhibitory factor-induced epithelial repair. Am. J. Physiol. Lung Cell. Mol. Physiol. 296, L442–

L452.

Morin, C., Sirois, M., Echave, V., Gomes,M.M., and Rousseau, E. (2008). EET displays anti-inflammatory effects in TNF-alpha stimulated human bronchi: putative

role of CPI-17. Am. J. Respir. Cell Mol. Biol. 38, 192–201.

Peters-Golden, M., Gleason, M.M., and Togias, A. (2006). Cysteinyl leukotrienes: multi-functional mediators in allergic rhinitis. Clin. Exp. Allergy 36, 689–703.

Sheibanie, A.F., Yen, J.-H., Khayrullina, T., Emig, F., Zhang, M., Tuma, R., and Ganea, D. (2007). The proinflammatory effect of prostaglandin E2 in experimental

inflammatory bowel disease is mediated through the IL-23—>IL-17 axis. J. Immunol. 178, 8138–8147.

Snelgrove, R.J., Goulding, J., Didierlaurent, A.M., Lyonga, D., Vekaria, S., Edwards, L., Gwyer, E., Sedgwick, J.D., Barclay, A.N., and Hussell, T. (2008). A critical

function for CD200 in lung immune homeostasis and the severity of influenza infection. Nat. Immunol. 9, 1074–1083.

Stoll, L.L., Morland, M.R., and Spector, A.A. (1994). 13-HODE increases intracellular calcium in vascular smooth muscle cells. Am. J. Physiol. 266, C990–C996.

Storer, P.D., Xu, J., Chavis, J.A., and Drew, P.D. (2005). Cyclopentenone prostaglandins PGA2 and 15-deoxy-delta12,14 PGJ2 suppress activation of murine

microglia and astrocytes: implications for multiple sclerosis. J. Neurosci. Res. 80, 66–74.

S2 Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc.

Figure S1. Histological Analysis Score, Related to Figure 1

Lung sections stained with hematoxylin and eosin (H&E) were scored for the levels of interstitial pneumonia/alveolitis (IP), bronchial epithelium necrosis (BN) and

hyperplasia (BH), alveolar hyperplasia(AH), lymphoid aggregates (LA), perivascular cuffing by mononuclear cells (PC-M) and neutrophils (PC-N), perivascular or

alveolar edema(PE, E), hemorrhage (H), vasculitis (V) and percent of section affected in any manner (E1) or in most severe manner (E2). Lesions were scored 0 =

normal or no change; 1 = minimal; 2 = mild; 3 = moderate; 4 = severe. Extent scores were applied as: 0 = no lesions; 1 = < 5%; 2 = 6%–30%; 3 = 31%–60%; 4 >

61%. Significant histological disease is reflected by summed score > 15.

(A) Total histological scores plotted as scattered plot of uninfected control (PBS, green circle), PR8 (blue square), or X31 (red triangle) infected tissues. Plot

represents mean +/� SEM. Mann-Whitney tests were performed to determine statistical significance between PR8sublethal D6 p.i. versus X31sublethal D5 p.i.,

PR8sublethal D9 p.i. versus X31sublethal D7 p.i., and PR8sublethal D13 p.i. versus X31sublethal D13 p.i. (N.S.; not statistically significant).

(B) Mean and standard deviation of individual scores (n = 4).

(C) Representative examples of histological lesions in influenza infected lungs. A. Normal lung. Terminal bronchiole (TB), pulmonary arteriole (a) and alveolus (A)

are indicated. B. Early lesions are principally focused on the pulmonary arteriole and terminal bronchiole anatomical unit. The arteriole wall is thickened and there

are abundant intravascular marginating leukocytes and at higher magnification, reactive endothelial cells are discernable (scored collectively as vascular lesions;

V). The perivascular space (PV) is moderately edematous (scored as perivascular edema; PE) with inflammatory cells extending mildly into the adjacent inter-

stitium (scored as interstitial pneumonia; IP); D6 p.i. PR8. C. Necrosis of the bronchiolar epithelium (scored a bronchiolar necrosis; BN) lining the terminal

bronchioles (TB) is also noted during early infection. The epithelium of the terminal bronchiole (TB) is multifocally denuded and necrotic cells are sloughed into the

airway. There are moderate perivascular inflammatory cells that at higher power can be discerned into either mononuclear (lymphocytes and macrophages) or

neutrophilic (scored as PV-M and PV-N respectively) There are moderate numbers of peribronchiolar inflammatory cells and increased interstitial pneumonia and

alveolar inflammatory cells and fibrin; D6 p.i. PR8.

(D) At D9 (nadir weight loss), the lung is consolidated with severe interstitial pneumonia and the terminal bronchiole is multifocally reepithelized with deeply

basophilic and plump regenerative epithelium (scored as bronchiolar hyperplasia; BH); D9 p.i. PR8.

(E) Severe interstitial pneumonia with dense mononuclear perivascular inflammation, sloughed cells and debris within the TB and there is moderate alveolar

hyperplasia (white A) characterized by alveoli lined with cuboidal pneumocytes (scored as alveolar hyperplasia; AH); D13 p.i. PR8.

(F) Moderate peribronchiolar lymphoid aggregate (scored as lymphoid aggregate; LA) with mild perivascular edema and inflammation and relatively normal

terminal bronchioles; D13 X31. Histological lesions scored but not shown in C include alveolar edema (E) characterized by flooding of alveolar spaces with

proteinaceous fluid and hemorrhage (H) characterized by extravascular erythrocytes within airways. All panels original magnification 20 3, bar in A equals

300 mm.

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. S3

Figure S2. Cytokine and Chemokine Levels Detected in BAL by Multiplex Immunoassay Represented as Line Graphs, Related to Figure 2

y axis represents the amount of protein in pg/ml. Line graphs of selected representative cytokines/chemokines depicting the mean +/� SEM (red: X31sublethal;

blue: PR8sublethal; black: PR8lethal). Statistical significance: * 0.01 < p < 0.05; ** 0.001 < p < 0.01; ***p < 0.001; magenta asterisks denote significance between

PR8lethal and both X31sublethal and PR8sublethal infections; blue and red asterisks denote significance between PR8lethal and PR8sublethal or X31sublethal infections,

respectively; black asterisks denote significance between PR8sublethal and X31sublethal infections).

S4 Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc.

Figure S3. Percentages of Lipid Mediator Classes and Transcriptomic Analysis in Mouse Influenza Infection, Related to Figure 3

(A) Line graphs of percentages of cyclooxygenase (COX), lipoxygenase (LOX), cytochrome P450 (CYP450), linoleic acid, linolenic acid, DHA, and EPA derived

metabolites. Also depicted is the line graph of total lipid mediators (without arachidonic acid) as measured in BAL (pmol/ml). Line graphs represent the

mean +/�SEM (red: X31sublethal; blue: PR8sublethal; black: PR8lethal). Mann-Whitney tests were performed to determine statistical significance (* 0.01 < p < 0.05;

** 0.001 < p < 0.01; ***p < 0.001; magenta asterisks denote significance between PR8lethal and both X31sublethal and PR8sublethal infections; blue and red asterisks

denote significance between PR8lethal and PR8sublethal or X31sublethal infections, respectively; black asterisks denote significance between PR8sublethal and

X31sublethal infections).

(B) Heat map represents transcript levels of gene involved in lipid metabolism as measured using Fluidigm RT-PCR. Scale bar represents values normalized

to EF1a.

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. S5

Figure S4. Clustering Analysis and Percent of Lipid Mediator Classes in Human Clinical Nasal Washes Samples, Related to Figure 4

(A) Hierarchical clustering or clinical symptoms and cytokine/chemokine levels measured in human nasal washes. Three separate groups were identified: low,

medium, and high degree of symptoms and immune responses.

(B) Column vertical scatter plots of the percentages of cyclooxygenase (COX), lipoxygenase (LOX), cytochrome P450 (CYP450), linoleic acid, linolenic acid, DHA,

and EPA derived metabolites in nasal wash samples, mean +/� SEM. Mann-Whitney tests were performed to determine statistical significance (* 0.01 < p < 0.05;

** 0.001 < p < 0.01; ***p < 0.001).

S6 Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc.

Figure S5. Pro- and Anti-Inflammatory Indices of Mouse and Human Lipidomic Profiles, Related to Figure 5

(A) XY scatter plots represent the pro- versus anti-inflammatory indices of individual mice infected with PR8sublethal (blue), X31sublethal (red), or PR8lethal (black).

(B) Heatmap represents the fold change of medium or high responders over low responders for each lipid mediator (separated by pro- or anti-inflammatory

activity).

(C) XY scatter plot representing the pro- versus anti-inflammatory indices of individual clinical sample (green: medium responders; red: high responders).

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. S7

Figure S6. Lipid Mediator Levels within the Lipoxygenase Pathway, Related to Figure 6

Line graphs and column vertical scatter plots of percentages ormeasured levels of metabolites within the 5-, 15-, 12-, and 8-lipoxygenase subpathways inmouse

(A) and human (B) samples, respectively. Graphs represent the mean with SEM as error bars (in A, red: X31sublethal; blue: PR8sublethal; black: PR8lethal). Mann-

Whitney tests were performed to determine statistical significance (* 0.01 < p < 0.05; ** 0.001 < p < 0.01; ***p < 0.001; magenta asterisks denote significance

between PR8lethal and both X31sublethal and PR8sublethal infections; blue and red asterisks denote significance between PR8lethal and PR8sublethal or X31sublethalinfections, respectively; black asterisks denote significance between PR8sublethal and X31sublethal infections). Graphs represent mean +/� SEM.

S8 Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc.

Figure S7. Linoleic Acid Metabolites in a Compilation of Three Additional Experiments, Related to Figure 7

(A) Heatmap represents the level of linoleic acid derived metabolites measured in BAL of animals infected with PR8sublethal or X31sublethal. Data represents the

compilation of three additional experiments performed on separate days and on animals obtained from a separate vendor (Jackson Laboratory). Scale bar

represents levels measured in pmol/ml.

(B) Line graphs of 13:9 HODE ratios from these data, mean +/� SEM.

Cell 154, 213–227, July 3, 2013 ª2013 Elsevier Inc. S9