Lipidomics identifies a requirement for peroxisomal function during influenza virus replication.

Influenza virus acquires a host-derived lipid envelope during budding, yet a convergent view on the role of host lipid metabolism during infection is lacking. Using a mass spectrometry-based lipidomics approach, we provide a systems-scale perspective on membrane lipid dynamics of infected human lung epithelial cells and purified influenza virions. We reveal enrichment of the minor peroxisome-derived ether-linked phosphatidylcholines relative to bulk ester-linked phosphatidylcholines in virions as a unique pathogenicity-dependent signature for influenza not found in other enveloped viruses. Strikingly, pharmacological and genetic interference with peroxisomal and ether lipid metabolism impaired influenza virus production. Further integration of our lipidomics results with published genomics and proteomics data corroborated altered peroxisomal lipid metabolism as a hallmark of influenza virus infection in vitro and in vivo. Influenza virus may therefore tailor peroxisomal and particularly ether lipid metabolism for efficient replication.

for virus production. Host cell lipid metabolism and plasma membrane microdomains are implicated in the biogenesis of virus envelopes. Several studies have dissected the lipid inventory of purifi ed infl uenza virions ( 9,10 ), whereas others have demonstrated the requirements for de novo fatty acid and sphingolipid biosynthesis and unique cholesterol compositions for virus production at budding sites (11)(12)(13)(14).
In addition to the importance of host cell lipid metabolism for the biogenesis of infl uenza virus envelopes, recent fi ndings suggest a major role for soluble lipid mediators in antiviral responses against infl uenza virus infection in vivo ( 15,16 ). These soluble lipid mediators originate from membrane glycerophospholipids (GPLs) via phospholipase activity, and (to some extent), are metabolized in peroxisomes. For example, ␤ -oxidation in the peroxisome is crucial for the retroconversion of DHA, the precursor of the lipid mediator protectin D1, which prevents nuclear export of infl uenza virus RNAs; protectin D1 production is directly inhibited by infl uenza virus ( 15 ). The role of peroxisomes during infl uenza virus replication is further evident by interaction between infl uenza virus nonstructural protein 1 (NS1) and multifunctional protein 2 (MFP2/HSD17B4), an antiviral protein essential for peroxisomal ␤ -oxidation ( 17 ). Therefore, the collective literature indicates an apparent role for peroxisomes as the initial sites of antiviral signaling ( 18 ). Supplementary key words ether lipids • sphingolipids • glycerophospholipids • lipid metabolism • biochemistry • systems biology Infl uenza viruses hijack host cell machineries for efficient replication and acquire a host-derived lipid envelope during budding. Recent systems-scale studies have primarily addressed the individual roles of genes ( 1-5 ) and proteins (6)(7)(8) in this process, yet have failed to illustrate how they function together to generate macromolecular precursors Supernatants and cell lysates were subjected to plaque assay and Western blot at 18 hpi. Cell viability after drug treatment was assessed using Vybrant MTT cell proliferation assay kit (V13154) (Invitrogen®, Life Technologies Co., San Diego, CA) according to the manufacturer's protocol. For lipid analysis by MS, A549 cells were treated with 10 µM D609 in serum-and antibiotics-free medium for 18 hpi. Signifi cant differences were calculated by Student's t -test ( P < 0.05; two-tailed).

Infl uenza virus infection of DHAPAT defi cient CHO-K1 cells
CHO-K1 and NRel-4 cells were infected with infl uenza virus A/PR/8/34 H1N1 (MOI <1) and virus replication was assessed by Western blot and plaque assay after 18 h of infection. Antiinfl uenza virus matrix protein 2 (M2 ) and anti-GAPDH antibodies (Santa Cruz, CA) were used to determine virus protein expression. Signifi cant differences were calculated by an unpaired Student's t -test ( P < 0.05; two-tailed).

Knockdown of AGPS and Rab11a by siRNA
Two siRNA constructs targeting alkylglycerone phosphate synthase (AGPS ) (Ambion, s16248 and s16249) and Rab11a (Ambion, s16703 and s16704) were used. Full details of the reverse transfection and validation of siRNA constructs is described in the supplementary data.

Catalase assay
A549 cells were infected with infl uenza virus A/PR/8/34 H1N1 and catalase activity was measured at 18 hpi using the Catalase Assay Kit (Sigma-Aldrich, St. Louis, MO). Signifi cant differences were calculated by an unpaired Student's t -test ( P < 0.05; two-tailed).

Literature mining
To derive a potential model of lipid metabolism in infl uenza virus-infected cells, we manually incorporated our lipidomics and cell biological data with existing genomics, proteomics, and metabolomics data. Procedures and additional references can be found in the supplementary data.

RESULTS AND DISCUSSION
To systematically characterize the temporal changes of host cell membrane lipid composition during infl uenza virus infection, human lung epithelial (A549) cells were infected with purifi ed infl uenza virus A/PR/8/34 H1N1 and total cellular lipids were extracted 12, 18, and 24 hpi. A high multiplicity of infection (MOI5) was used to ensure a synchronous, one-round of infection ( 19 ). A total of 175 lipid species, representing GPL and sphingolipids (SPLs), two major membrane lipid classes analyzed in this study, were measured using established methodology based on HPLC and ESI-MS and operated in multiple reaction monitoring mode ( Fig. 1A , supplementary Table  II) ( 20,21 ). The levels of 90 lipid species (i.e., ‫ف‬ 52% of all measured lipids) were signifi cantly altered between H1N1infected and mock-infected cells (q < 0.006; Fig. 1B, C ) at either 18 hpi, 24 hpi, or both. Of these, 35 ( Fig. 1B , red pie chart) and 15 ( Fig. 1B , blue pie chart) lipid species had correlation coeffi cients of >0.9 or < Ϫ 0.9, respectively, with virus titer ( Fig. 1B , supplementary Tables I and II). Virus However, a convergent view of the role of lipid metabolic pathways during infl uenza virus replication, particularly those pathways required in the generation of membranes/ envelopes, and their contribution to virus pathogenicity, are lacking. We therefore sought to examine changes in host membrane lipid composition during different stages of infection with infl uenza virus using analytical biochemistry to explore the requirement of precursors for lipid mediators as well as other peroxisome-derived lipids. While the experimental approach is tried and tested, thus not novel by itself, our detailed and integrative analysis resulted in a comprehensive systems-scale resource of how infl uenza virus impacts lipid metabolism. Using siRNA knockdown and pharmacological inhibitors, we directly provide a "proof-ofconcept" that our resource can be useful to derive novel hypotheses in the emerging fi eld of lipid involvement during virus infections. Using this approach and in combination with meta-data for further support, we reveal novel insights into the role of a class of host membrane components with poorly defi ned functions, peroxisome-derived ether lipids, during infl uenza virus replication. Our fi ndings put the peroxisome and its lipid metabolism at the forefront of infl uenza virus infections in vitro and in vivo.

Virus strains, virus production, and purifi cation
Virus stocks were prepared by passaging egg-grown virus strains once in MDCK cells. Virus strains were purifi ed from A549 and MDCK cells as described in the supplementary data.

Lipid extraction of infected cells and purifi ed viruses
A549, Chinese hamster ovary (CHO)-K1, and NRel-4 cells were seeded into 10 cm cell culture dishes 24 h prior to infection . Cells at 80-100% confl uence were infected with a 5 ml inoculum of purifi ed infl uenza virus A/PR/8/34 H1N1 at MOI 5. Virusinfected cells and mock-infected cells were collected at 12, 18, and 24 hours postinfection (hpi) (for A549 cells) or only at 18 hpi (CHO-K1 and NRel-4 cells). Lipid extraction was conducted according to a modifi ed Bligh and Dyer protocol described in the supplementary data.

Quantitative analysis of lipids by HPLC MS/MS
Samples with spiked internal standards were analyzed by ESI-MS. Signal intensities for each lipid species were extracted according to their retention time, normalized to the representative spiked internal standards, and represented as a molar fraction of the total amount of measured lipids. Statistical signifi cance was calculated using an unpaired Student's t -test ( P < 0.05; twotailed) or a block-design three-way ANOVA corrected by a false discovery rate procedure (for more details, see the supplementary data). D609 and GW7647 were purchased from Tocris Bioscience (Bristol, UK). A549 cells were infected with infl uenza virus A/ PR/8/34 H1N1 (MOI <1), and serum-and antibiotics-free medium supplemented with D609 (10 µM and 100 µM) or GW7647 (1 µM, 2 µM, and 5 µM) were added at 12 hpi and 1 hpi, respectively.  Tables I, II). Black, red (>0.9) and blue (> Ϫ 0.9) indicate 90 lipid species altered in infl uenza virus-infected cells (q < 0.006; supplementary Table II). C: Heatplot showing fold ratios (infected/mock) of 90 lipid species with altered levels upon infection (q < 0.006). Yellow and blue indicate elevated and decreased concentrations, respectively. Lipid species, which correlate with virus titer (B), are indicated by red (>0.9) and blue (< Ϫ 0.9) fonts. Representative structures to illustrate the differences between ester-linked (D), odd chain (E), and ether-linked (F) PC lipids. Please note that we were able to determine the total carbon fatty acyl composition but did not dissect the exact carbon composition of the two fatty acyl constituents in the measured GPL species. Hence, the two fatty acyl constituents shown in the structures can vary as long as they add up to the respective total carbon fatty acyl compositions. G: Fold ratios of changes in fatty acid chain length composition of Cer, HexCer, and SM lipid species. Results in panels A, B, C, and G are from three independent experiments with three replicates each (n = 9 for each condition). Error bars in G represent ± SDs. infection lowered the levels of ester-linked GPL, mainly phosphatidylethanolamine and phosphatidylcholine (aPC) species, but increased SPL such as many SM and hexosylceramide (HexCer) species ( Fig. 1B, C ). These observed changes in lipid species were also refl ected by the increase in total amounts of ether-linked PC (ePC ), odd chain aPC and SM lipid classes and by the decrease in the total amount of Ganglioside GM3 lipid class across the three independent experiments (supplementary Figs. I, II ).

D609 and GW7647 treatment of infl uenza virus-infected cells
The decrease in the proportion of ganglioside GM3 species likely refl ected infl uenza virus neuraminidase activity ( 9 ). The reduced levels of aPC species in infl uenza virusinfected cells have been previously proposed to be related to impaired aPC biosynthesis as measured by metabolite rates of phospholipid precursors and by global gene and protein expression experiments ( 8,22,23 ). SREBP1, a major regulator of the one-carbon cycle producing the methyl donor S-adenosylmethionine required for the de novo methylation pathway for aPC biosynthesis, was signifi cantly downregulated in infl uenza virus-infected cells ( 22,24 ). These changes coincided with the increasing levels of another choline containing lipid, SM, suggesting an important correlation of infl uenza virus replication with choline lipid metabolism ( Fig. 1B,C ). Increase in SM and decrease in aPC species could possibly be explained by the activities of inter-related enzyme systems including sphingomyelin synthases (SMS1 and SMS2, which transfer the choline headgroup of PC onto a ceramide backbone to produce SM) and ethanolamine kinase 1, which is downregulated in infl uenza virus-infected cells ( 22 ). However, additional experiments such as quantitative proteomics and enzymatic assays would be required to draw such conclusions. The results presented here provide a good starting point for generation of hypotheses and such future investigations. The upsurge in the proportion of long chain (>38 fatty acyl carbons, >C38) aPC species with polyunsaturated fatty acyls ( Fig. 1D ), odd chain aPC ( Fig. 1E ), and ePC ( Fig. 1F ) suggested altered peroxisomal lipid metabolism in infected cells. Consistent with impaired peroxisomal ␤ -oxidation, we observed an enrichment in C26:0 but a decrease in C24:1 fatty acids in SPL species ( Fig. 1G ) (25)(26)(27); this provided further support for an important role of fatty acyl metabolism during infl uenza infection ( 15,16 ).
We next tested whether alterations in host membrane lipid levels are detectable also in envelopes harvested from purifi ed virus particles ( Fig. 2A ). Indeed, aPC, phosphatidylethanolamine, phosphatidylinositol, and GM3 levels were decreased in viruses whereas SM was increased when compared with levels in the producer cell ( Fig. 2B , supplementary Table III) ( 9,10 ). To characterize the enrichment of peroxisome-derived ether-linked lipids in infl uenza virus envelopes and to establish a quantitative correlate between host and virus membrane compositions, we determined the ratios of ePC/aPC in both membrane extracts. To account for the confounding effect of variations in peroxisomal activity among different host cell lines as well as differences in experimental approaches in the published literature, we decided to use the difference between the ePC/aPC ratios of virus particles and uninfected producer cells ( ⌬ Virus-Cell = ePC/aPC Virus Ϫ ePC/aPC Cell ) as a molecular proxy of PC lipid class remodelling to compare a wide variety of different studies ( Fig. 2C-F ). Envelopes of infl uenza viruses (circles in Fig. 2C, F ) had a signifi cantly higher ePC/aPC ratio than their uninfected producer cells (bars in Fig. 2C, F ), whereas other enveloped viruses (including human immunodefi ciency virus, murine leukemia virus, vesicular stomatitis virus, dengue virus, and hepatitis C virus) showed equal or lower ePC/aPC ratios (diamonds in Fig. 2D, F , supplementary Table IV). Therefore, peroxisomedependent remodelling of lipids within the abundant PC class, rather than overall changes to total PC concentration, is specifi c to infl uenza virus.
We next determined the lipid composition of two closely related H3N2 infl uenza virus strains differing in pathogenicity. The parent infl uenza A strain A/Aichi/2/68 H3N2 (P0) was adapted by ten passages in mice (P10), which ultimately showed higher virulence with enhanced replication fi tness because of nonconservative point mutations in hemagglutinin (HA) (Gly218Glu ) and NS1 (Asp125Gly) ( 28 ). NS1 regulates lipid metabolic genes in a severitydependent manner ( 22 ) and interacts with MFP2/HSD17B4 (see above) ( 17 ). The latter mutation has been shown to produce high virus titres with enhanced interferon-␤ antagonism and to differentially regulate host gene expression ( 29,30 ). We assumed a negligible effect of the mutation in HA (Gly218Glu) on host lipid metabolism because the mutation lies in a region involved in sialic acid linkage recognition, important for entry rather than infl uenza virus replication within the host cell ( 28 ). We showed that the more pathogenic H3N2 strain (P10) exhibited a ‫ف‬ 25% higher ePC/aPC ratio in its envelope than the less pathogenic P0 strain ( Fig. 2E , supplementary Fig. III, supplementaryTables IV, V), which was comparable with the variation in ePC/aPC ratios between different infl uenza viruses ( Fig. 2C ). Collectively, these results suggest strain-dependent differences with regard to PC class remodelling which might refl ect infl uenza virus pathogenicity and underscore the conserved role of peroxisomes in infl uenza infection ( Fig. 2F ).
We next sought a systems-scale perspective on the broader impact of infl uenza virus-induced perturbations of lipid metabolism ( Fig. 3 ). To do so, we used the aforementioned data from i ) A549 cell infection (host), ii ) purifi ed H1N1 virus (virus), and iii ) H3N2 P10 virus (pathogenicity) for unsupervised cluster analysis ( Fig. 3A , supplementary Table  VI). We assigned 13 clusters (AU p -value >0.9, Fig. 3B ) with unique patterns of lipid regulation ( Fig. 3C ). For instance, lipids with increased concentrations in virus-infected cells but decreased levels in virions are suggestive of intracellular requirements for virus replication as revealed for long chain fatty acid-containing ePC and two aPC species with long but saturated fatty acyls (cluster 6 in Fig. 3C, D ). Unsaturated ester-linked PE and PC species were reduced both in virions as well as in infected cells ( Fig. 3E ), consistent with the downregulation of genes implicated in ester-linked GPL metabolism ( 22 ). On the contrary, lipids enriched in both infected cells and virions are indicative of a possible function in virus morphogenesis, as seen for saturated short-chain fatty acid-containing ePC species (cluster 10 in and cell viability ( Fig. 4A ). This result strengthens the previously identifi ed requirement of intact SM biosynthesis for infl uenza virus production ( Fig. 1 ) ( 14 ), but more specifi cally, implies importance of the salvage pathway.
To further scrutinise the necessity of ether lipid metabolism during infl uenza replication, we infected wild-type CHO-K1 and ether lipid-defi cient CHO cells (NRel-4) ( 32 ) with infl uenza virus H1N1 A/PR/8/34. NRel-4 cells exhibited lower expression levels of infl uenza virus proteins NS1 and M2, and a four-to fi ve-fold decrease in virus production when compared with CHO-K1 cells ( Fig. 4B ). The lipid alterations induced by infl uenza virus in CHO cells were consistent with the changes in A549 cells (supplementary Fig. V) with the obvious exception of ePC, which cannot be generated in NRel-4 cells because of the reported impairment in their peroxisomal dihydroxyacetone phosphate acyltransferase (DHAPAT) activity. DHAPAT catalyses the fi rst committed step in ether lipid biosynthesis, attaching a fatty acid to dihydroxyacetone phosphate (DHAP) to produce acyl-DHAP.
As acyl-DHAP can also be redirected into TAG biosynthesis ( 33 ), and because the two CHO cell variants were not isogenic, we therefore decided to further investigate  Table VI). Two single clusters highlighted the importance of SPL species for virion formation. While saturated fatty acyls containing SM species were enriched in virus-infected cells as well as virions ( Fig. 3F ), saturated fatty acyls containing HexCer species were additionally enriched in the more pathogenic virus strain P10 ( Fig. 3G ). These fi ndings emphasized a general requirement of SM metabolism for virion morphogenesis, which was also refl ected by high correlation with virus titer ( Figs. 1, 3H ) ( 9,11,14 ). This comparative analysis of lipid levels in purifi ed viruses and infected host cells therefore postulates new hypotheses for the roles of lipids with similar chemistries during different stages of virus replication.
We next evaluated the contribution of ether lipid and SPL metabolism during the late stages of infl uenza virus replication using a combination of pharmacological inhibitors and genetic silencing. We fi rst treated A549 cells with the SMS inhibitor D609, which inhibits an SM biosynthesis salvage pathway rather than de novo biosynthesis ( 31 ) ( Fig. 4A , supplementary Fig. IV). D609 was added at 12 hpi and virus titres were reduced in a dose-dependent manner at 18 hpi without affecting virus protein expression  Table III). Error bars represent ± SDs calculated from three independent experiments (n = 9 for mock-infected A549 cells at 12 hpi) and two independent experiments (n = 6 for purifi ed infl uenza virions); * P < 0.05, ** P < 0.005, and *** P < 0.0005 (unpaired Student's t -test; two-tailed). C: Ether PC to ester PC (ePC/aPC) ratios of infl uenza virions (circles) compared with (D) ePC/aPC ratios of other enveloped viruses (diamonds) in relation to uninfected producer cells (bars) (supplementary Table IV); ** P < 0.002 (unpaired Student's t -test; two-tailed). E: Differences in ePC/aPC ratios between P0 and P10 H3N2 strains (supplemetary Fig. III, supplementary Table V). Results are from three independent experiments with two replicates (n = 6); * P < 0.02 (paired Student's t -test; two-tailed). F: Differences in ePC/aPC ratios in virus versus host membranes ( ⌬ Virus-Cell ) of infl uenza (circles) and other viruses (diamonds) as described for D and E. changes the fatty acid of acyl-DHAP with a fatty alcohol. Both probes led to the substantial reduction in enzyme levels ( ‫ف‬ 70%), as judged by Western blotting ( Fig. 4C , ether lipid involvement using two siRNA constructs against peroxisomal AGPS. AGPS is the immediate downstream enzyme of DHAPAT in ether lipid biosynthesis, which ex-  . For A-E, virus titer was determined by plaque assay and represented as % of control ± SEMs; * P < 0.005 and ** P < 0.0005 (unpaired Student's t -test; two-tailed). Expression of proteins was determined by Western blot (one representative blot is shown), using antibodies against AGPS, Rab11a, GAPDH, and ␣ -tubulin (positive controls), and infl uenza virus NS1 and M2. F: Distribution of host (mouse) susceptibility factors to infl uenza virus infection ( 39 ): lipid-associated genes (black circles), independently identifi ed genes ( 40 ) (fi lled black circles), such as PLA2G7, the most signifi cantly associated factor; remaining genes (open gray circles). Lipid-associated genes were manually annotated to Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (gray squares) or by Database for Annotation, Visualization and Integrated Discovery (DAVID) (black squares; * P < 0.05 (EASE score, a modifi ed Fisher Exact P -value) and ** P < 0.05 [EASE score; Benjamini corrected )]. G: Proposed lipid metabolism in infl uenza virus-infected cells based on combined lipidomics data (this study) with genomics and proteomics results (supplementary Table VII). Genes, proteins, and metabolites are depicted in bold red (proviral or increased), bold blue (antiviral or decreased) and bold black (pro-or antiviral activity not determined) fonts. Fluxes proposed to be increased (bold arrows) or decreased (dashed arrows), inhibitory interactions (round ended bold lines), chemical inhibitors (underlined bold black) and the membrane scaffold of Serinc5 (dashed red box) are also shown. Gray boxes indicate sphingolipid (1), ether lipid (2) and peroxisomal ␤ -oxidation (3) metabolic pathways; *identifi ed in this study.
supplementary Fig. VI A), to a moderate (<1.5-fold) but signifi cant decrease in ether lipid levels (supplementary Fig. VI B), and, importantly, to a 60% reduction in infectious virus production ( Fig. 4C ). These results were comparable to siRNA constructs targeting Rab11a, which is required for assembly and budding ( 34 ), and demonstrate the functional importance of ether lipid biosynthesis in infl uenza virus production.
Finally, we explored the potential in vivo relevance of peroxisomal and SPL metabolism for infl uenza virus infection through an examination of the reported susceptibility factors in mice ( 39 ). We found 23 genes associated with lipid metabolism [5% of total genes measured ( 39 )] that were enriched for peroxisomal and SPL metabolism ( Fig. 4F ). The platelet-activating factor (PAF) acetylhydrolase, PLA2G7 , which exhibits the strongest association, has been independently identifi ed as a host susceptibility factor for infl uenza infection ( 40 ). PLA2G7 hydrolyses PAF, an ePC and activator of platelets and infl ammation, producing lyso-PAF, which can be converted to ePC by lyso-PC acyltransferase 2 ( Fig. 4G ). Consistent with elevated ePC levels in infected cells, infl uenza virus infection correlated with higher expression levels of PLA2G7 and induced lyso-PC acyltransferase 2 activity in mice ( 39,41 ). These fi ndings provide a link between the metabolic pathways of peroxisomes and lipid meditators involved in infl ammation in vivo.
In summary, we present a detailed account of the temporal changes in host cell membrane lipids during infl uenza virus replication in relation to the composition of virus envelopes and virus pathogenicity. While our study does not directly expose the mechanistic actions of identifi ed lipids in the infl uenza virus life cycle in detail, the comprehensive systems-scale catalog of lipids reported here is the fi rst of its kind ( Fig. 3 ). The combination and hierarchical clustering of the different datasets provides a powerful framework to derive novel hypotheses in the emerging fi eld of lipid involvement during virus infections.
As a result, we present clear evidence that metabolism of ether lipids is functionally important for the production of infectious virions. Further integration of our fi ndings with other published genomics and proteomics data ( 1, 3-8, 15, 22 ) led to a systems-scale model of host cell lipid metabolism during infl uenza virus infection, which will serve as a reference basis for future investigations ( Fig. 4G , supplementary Table VII). Based on this analysis, we propose three major lipid metabolic pathways implicated in infl uenza virus replication: 1 ) elevated ether lipid and 2 ) elevated SPL biosynthesis required for infl uenza virus morphogenesis and 3 ) decreased peroxisomal ␤ -oxidation associated with intracellular life cycle stages ( Fig. 4G ). Peroxisomal function is a common metabolic denominator and may therefore represent a key determinant for infl uenza virus replication. Our detailed analysis represents a major step forward in uncovering that peroxisomes and especially their lipid metabolism are exploited by infl uenza viruses. Our fi ndings may open entirely new avenues with immediate exploitability for therapeutic interventions against infl uenza virus infection via peroxisome function.