IFITM3 was identified in a functional genomic screen as mediating resistance to influenza A virus, dengue virus and West Nile virus infection in vitro1. However, the role of the IFITM proteins in anti-viral immunity in vivo is unknown. Therefore, we infected mice that are homozygous for a disruptive insertion in exon 1 of the Ifitm3 gene that abolishes its expression8 (Ifitm3−/− ) with a low-pathogenicity murine-adapted H3N2 influenza A virus (A/X-31). Low-pathogenicity strains of influenza do not normally cause extensive viral replication throughout the lungs, or cause the cytokine dysregulation and death typically seen after infection with highly pathogenic viral strains9, at the doses used (Fig. 1a). However, low-pathogenicity-infected Ifitm3−/− mice became moribund, losing >25% of their original body weight and showing severe signs of clinical illness (rapid breathing, piloerection) 6 days after infection. In comparison, wild-type littermates shed <20% of their original body weight, before fully recovering (Fig. 1a, b). There was little difference in virus replication in the lungs during the first 48 h of infection. However, virus persisted and was not cleared as quickly in Ifitm3−/− mice, whose lungs contained tenfold higher levels of replicating virus than the wild-type mice at 6 days post-infection (Fig. 1c). No viral RNA was detected in the heart, brain or spleen of infected wild-type or Ifitm3−/− mice over the course of infection, revealing that systemic viraemia was not occurring. Full-genome sequencing of virus removed from the lungs of wild-type and Ifitm3−/− mice showed no genetic variation. We demonstrated that IFITM3 protein expression after influenza infection was absent in Ifitm3−/− mice but increased substantially in wild-type controls (Fig. 1b and Supplementary Fig. 1). Infection of wild-type and Ifitm3−/− mice with a human isolate of pandemic influenza A H1N1 (pH1N1/09) resulted in the same severe pathogenicity phenotype in the Ifitm3−/− mice (Fig. 1a, b). Mouse embryonic fibroblast (MEF) lines generated from multiple matched littermates demonstrated that Ifitm3−/− cells are infected more readily in vitro, and lack much of the protective effects of interferon (IFN). Importantly, the stable restoration of IFITM3 conferred wild-type levels of restriction against either the X-31 strain, or the more pathogenic Puerto Rico/8/34 (PR/8) influenza strain (Fig. 1d and Supplementary Fig. 2). In addition to the role of IFITM3 in restriction of high-pathogenicity H5N1 avian influenza7, we also show that it limits infection by recent human influenza A virus isolates and influenza B virus (Supplementary Fig. 3). Therefore, enhanced pathogenesis to diverse influenza viruses is attributable to loss of Ifitm3 expression and consequential changes in immune defence of the lungs.

Figure 1: Influenza A virus replicates to higher levels in Ifitm3 −/− mice.
figure 1

a, b, Change in body mass (a) and survival (b) of wild-type (filled circles) and Ifitm3−/− (open squares) mice following intranasal inoculation with A/X-31 and pandemic H1N1/09 Eng/195 influenza (n > 5). b, Absence of Ifitm3 expression was verified in the Ifitm3−/− mice at all time points, but was seen to increase in wild-type mice. c, A/X-31 viral load in the lungs of mice (n > 4) was calculated over the course of infection by plaque assay. p.f.u., plaque-forming units. Ifitm3−/− murine embryonic fibroblasts (n = 3 per condition) stably expressing Ifitm3 (+), or the empty vector (−) were left untreated (blue), or incubated with IFN-α (red) or IFN-γ (green), then challenged with either A/X-31 or PR/8 influenza. d, Twelve hours after infection, the cells were assessed for either haemagglutinin expression (PR/8), or nucleoprotein expression (A/X-31) IFITM3 expression was determined to be present (+) or absent (−) by western blotting (Supplementary Fig. 2). Results show means ± s.d. Statistical significance was assessed by Student’s t-test (**P < 0.01; ***P < 0.001).

PowerPoint slide

Examination of lung pathology showed fulminant viral pneumonia with substantial damage and severe inflammation in the infected Ifitm3−/− mice. Lung pathology was characterized by extensive oedema and red blood cell extravasation, as well as pneumonia, haemorrhagic pleural effusion and multiple, large lesions on all lung lobes (Fig. 2a, b and Supplementary Fig. 4). We note that this pathology is similar to that produced by infection of mice and primates with 1918 H1N1 virus9,10,11. Given the higher viral load in Ifitm3−/− mice and increased replication of influenza A virus in Ifitm3-deleted cells in vitro (Fig. 1d), we examined both viral nucleic acid and protein distribution in the lung. Influenza virus infection penetrated deeper into the lung tissue in Ifitm3−/− compared to wild-type mice whose infection was primarily restricted to the bronchioles, with minimal alveolar infection. Influenza virus was detected throughout the entire lung in Ifitm3−/− sections, spreading extensively in both bronchioles and alveoli (Fig. 2c). Histopathology showed marked infiltration of cells and debris into the bronchoalveolar space of Ifitm3−/− mice (Fig. 2b and Supplementary Fig. 4b). The extent and mechanism of cell damage was investigated by TdT-mediated dUTP nick end labelling (TUNEL) assay, showing widespread cellular apoptosis occurring 6 days post-infection in Ifitm3−/− mice, whereas apoptosis in wild-type lungs was very limited (Supplementary Fig. 4c). Together, the Ifitm3−/− mouse pathology is consistent with infection by high-pathogenicity strains of influenza A virus, where widespread apoptosis occurs by day 6 post-infection, whereas lungs from low-pathogenicity infections were similar to those of wild-type mice, displaying minimal damage9,12,13.

Figure 2: Pathological examination of infected lungs.
figure 2

a, b, Wild-type mice showed few visible signs of external damage on lung lobes at day 6 post-infection, whereas Ifitm3−/− mice showed several large lesions (a, left, ventral view of intact lungs, right, all lobes displayed) resulting from severe oedema and hemorrhagic pleural effusion (b), as well as a markedly higher infiltration of cells and proteinaceous debris into the alveoli and bronchioles. c, Localization of virus within the lungs on day 6 indicated that virus penetrated deeper and more extensively into the lung tissue in the Ifitm3−/− mice, as determined by immunohistochemistry for total influenza protein and detection of virus nucleic acid (virus, red; cell nuclei, blue; A, alveolus; B, bronchiole). Original magnifications were ×5 (b) and ×20 (c).

PowerPoint slide

Analysis of cell populations resident in the lung tissue on day 6 post-infection showed that Ifitm3−/− mice had significantly reduced proportions of CD4+ (P = 0.004) and CD8+ T cells (P = 0.02) and natural killer (NK) cells (P = 0.0001), but an elevated proportion of neutrophils (P = 0.007) (Fig. 3a). Despite the extensive cellular infiltration (Supplementary Figs 4b, 5a), the absolute numbers of CD4+ T-lymphocytes in the lungs of the Ifitm3−/− mice were also lower and neutrophils increased compared to wild-type mice (Supplementary Fig. 6). The peripheral blood of infected Ifitm3−/− mice showed leukopenia (Supplementary Fig. 5c). Blood differential cell counts indicated marked depletion of lymphocytes on day 2 post-infection in the Ifitm3−/− mice (P = 0.04) (Fig. 3b), reflecting changes observed previously in high-pathogenicity (but not low-pathogenicity) influenza infections in both humans and animal models9,12,14,15. Heightened cytokine and chemokine levels are also hallmarks of severe influenza infection, having been observed in both human and animal models9,16. We observed exaggerated pro-inflammatory responses in the lungs of Ifitm3−/− mice with levels of TNF-α, IL-6, G-CSF and MCP-1 showing the most marked increase (Fig. 3c and Supplementary Fig. 7). This is indicative of the extent of viral spread within the lungs, as TNF-α and IL-6 are released from cells upon infection17. Consistent with the immunopathology data above, these changes are comparable in level to those seen with non-H5N1 high-pathogenicity influenza infections9. Neutrophil chemotaxis, together with elevated proinflammatory cytokine secretion, has previously been reported as one of the primary causes of acute lung injury18.

Figure 3: Altered leukocyte and cytokine response to influenza A infection in Ifitm3 −/− mice.
figure 3

a, Cytometric analysis of proportional resident cell populations in the lungs of mice (+/+, black; −/−, white) showed evidence of lymphopenia in Ifitm3−/− mice 6 days post-infection. b, Systemic lymphopenia was confirmed through differential analysis of peripheral blood cell counts, which showed a significant depletion of lymphocytes on day 2 post-infection of Ifitm3−/− mice (monocytes, black; lymphocytes, grey; polymorphonuclear leukocytes, white). NK, natural killer. c, Levels of pro-inflammatory cytokines were also recorded as being elevated in Ifitm3−/− lungs over the course of infection (+/+, black; −/−, white). Results show means ± s.d., n = 5. Statistical significance was assessed by Student’s t-test (*P < 0.05, **P < 0.01, ***P < 0.001).

PowerPoint slide

To investigate further the extensive damage observed with low-pathogenicity influenza A virus infection in the absence of IFITM3, we infected both wild-type and Ifitm3−/− mice with a PR/8 influenza strain deficient for the multi-functional NS1 gene (delNS1)19,20. NS1 is the primary influenza virus interferon antagonist, with multiple inhibitory effects on host immune pathways20,21. We found that delNS1 virus was attenuated in both wild-type and Ifitm3−/− mice, and whereas the isogenic PR/8 strain expressing NS1 showed typical high pathogenicity in all mice tested, lower doses of PR/8 influenza (although lethal in both genotypes of mice) caused accelerated weight loss in Ifitm3−/− compared to wild-type mice (Supplementary Fig. 8). As delNS1 influenza A virus retains its pathogenicity in IFN-deficient mice19, this suggests that Ifitm3−/− mice can mount an adequate IFN-mediated anti-viral response without extensive morbidity, and that IFITM3 blocking viral replication occurs before NS1-mediated IFN antagonism7. Therefore, unchecked lung viral replication and an enhanced inflammatory response accounts for the profoundly deleterious effects of viral infection in Ifitm3−/− mice.

The human IFITM3 gene has two exons and is predicted to encode two splice variants that differ by the presence or absence of the first amino-terminal 21 amino acids (Fig. 4a). Currently, 13 non-synonymous, 13 synonymous, one in-frame stop and one splice site acceptor-altering single nucleotide polymorphisms (SNPs) have been reported in the translated IFITM3 sequence (Supplementary Table 1). Using tests sensitive to recent positive selection, we can find evidence for positive selection on the IFITM3 locus in human populations acting over the last tens of thousands of years in Africa (Fig. 4b, c). We therefore sequenced 1.8 kilobases of the IFITM3 locus encompassing the exons, intron and untranslated regions from 53 individuals who required admission to hospital as a result of pandemic H1N1/09 or seasonal influenza virus infection in 2009–2010. Of these, 86.8% of patients carried majority alleles for all 28 SNPs in the coding sequence of the gene, but 13.2% possessed known variants. In particular, we discovered over-representation in cases of the synonymous SNP rs12252, wherein the majority T allele is substituted for a minority C allele, which alters the first splice acceptor site and may be associated with the IFITM3 splice variant (ENST00000526811), which encodes an IFITM3 protein lacking the first 21 amino acids due to the use of an alternative start codon.

Figure 4: Single nucleotide polymorphisms of the human IFITM3 gene.
figure 4

a, Multiple single-nucleotide polymorphisms have been identified within the coding region of the human IFITM3 gene. One such SNP, rs12252 (red), encodes a splice acceptor site altering T/C substitution mutation and may be associated with a truncated protein with an N-terminal 21 amino acid deletion. Therefore two transcripts are predicted to be expressed from the IFITM3 gene. b, c, Positive selection analysis using a haplotype-based test, the cross population extended haplotype homozygosity test, maximum value (|XP-EHH-max|, b), where data points above 2.7 in the YRI (Africa) (red), 3.9 in the CEU (Europe) (blue) and 5.0 in the CHB+JPT (China and Japan) (green) populations are in the top 1% of values, and using a combination of three allele frequency spectrum-based test statistics (c), namely the composite likelihood ratio (CLR)23,24,25 on 10-kb windows along chromosome 11 encompassing the IFITM3 locus. Evidence for positive selection is seen only in the YRI. d, Mutations recorded through sequencing of patients hospitalized with influenza virus during the H1N1/09 pandemic showed an overrepresentation of individuals with the C allele at SNP rs12252, relative to matched Europeans. e, A549 cells transduced to express either full-length (IFITM3) or truncated (NΔ21) IFITM3 (cell nuclei, blue; virus, green; ×4 magnification) show a reduction in viral restriction when the N-terminal 21 amino acids of IFITM3 are removed, relative to vector controls (Vector). Alignment of the N termini of full-length (IFITM3, top) and truncated IFITM3 (NΔ21, bottom). Values represent the mean of the percentage of infected cells ± s.d. (n = 3).

PowerPoint slide

The allele frequencies for SNP rs12252 vary in different human populations (Supplementary Table 2). The ancestral (C) allele, reported in chimpanzees, is rare in sub-Saharan African and European populations (derived allele frequency (DAF) 0.093 and 0.026–0.036, respectively), but more frequent in other populations (Supplementary Table 2). SNP rs12252 is notable for its high level of differentiation between Europeans and East Asians, although the fixation index (FST, a measure of population differentiation) does not reach statistical significance. The genotypes associated with rs12252 in Caucasians hospitalized following influenza infection differ significantly from ethnically matched Europeans in 1000 Genomes sequence data and from genotypes imputed against the June 2011 release of the 1000 Genomes phased haplotypes from the UK, Netherlands and Germany (Wellcome Trust Case Control Consortium 1 (WTCCC1, UK): P = 0.00006, Netherlands: P = 0.00001, Germany: P = 0.00007; Fisher’s exact test). Patients’ genotypes also depart from Hardy–Weinberg equilibrium (P = 0.003), showing an excess of C alleles in this population (Fig. 4d). Principal components analysis of over 100,000 autosomal SNPs showed no evidence of hidden population structure differences between WTCCC controls and a subset of the hospitalised individuals from this study (Supplementary Fig. 9a, b).

To test the functional significance of the IFITM3 rs12252 polymorphism in vitro, we confirmed the genotypes of HapMap lymphoblastoid cell lines (LCLs) homozygous for either the majority (TT) or minority (CC) variant IFITM3 alleles (Supplementary Fig. 9c). We next challenged the LCLs with influenza A virus and found that the minority (CC) variant was more susceptible to infection, and this vulnerability correlated with lower levels of IFITM3 protein expression compared to the majority (TT) variant cells (Supplementary Fig. 10). Although we did not detect the IFITM3 splice variant protein (ENST00000526811) in the CC LCLs, we nonetheless investigated the possible significance of its presence by stably expressing the N-terminally truncated (NΔ21) and wild-type proteins to equivalent levels in human A549 lung carcinoma cell lines before infection with influenza A virus (A/WSN/1933 (WSN/33)). We found that cells expressing the NΔ21 protein failed to restrict viral replication when compared to wild-type IFITM3 (Fig. 4e), consistent with previous data showing that the amino-terminal 21 amino acids of IFITM3 are required for attenuation of vesicular stomatitis virus replication in vitro4. Similar results were obtained using other virulent viral strains (A/California/7/2009 (pH1N1), A/Uruguay/716/2007 (H3N2) and B/Brisbane/60/2008) (Supplementary Fig. 3).

We show here that IFITM3 expression acts as an essential barrier to influenza A virus infection in vivo and in vitro. The fulminant viral pneumonia that occurs in the absence of IFITM3 arises because of uncontrolled virus replication in the lungs, resulting in profound morbidity. In effect, the host’s loss of a single immune effector, IFITM3, transforms a mild infection into one with remarkable severity. Similarly, the enrichment of the rs12252 C-allele in those hospitalized with influenza infections, together with the decreased IFITM3 levels and the increased infection of the CC-allele cells in vitro, suggests that IFITM3 also plays a pivotal role in defence against human influenza virus infections. This innate resistance factor is all the more important during encounters with a novel pandemic virus, when the host’s acquired immune defences are less effective. Indeed, IFITM3-compromised individuals, and in turn populations with a higher percentage of such individuals, may be more vulnerable to the initial establishment and spread of a virus against which they lack adaptive immunity. In light of its ability to curtail the replication of a broad range of pathogenic viruses in vitro, these in vivo results suggest that IFITM3 may also shape the clinical course of additional viral infections in favour of the host, and may have done so over human evolutionary history.

Methods Summary

Mouse infection

Wild-type and Ifitm3−/− mice8 (8–10 weeks of age) were intranasally inoculated with 104 p.f.u. of A/X-31 (H3N2) influenza, 200 p.f.u. of A/England/195/09 (pH1N1) influenza, or 50–103 p.f.u. of A/PR/8/34 (PR/8) or the PR/8 NS1 gene deletion mutant (delNS1)20 (H1N1) in 50 μl of sterile PBS. Mouse weight was recorded daily as well as monitoring for signs of illness. Mice exceeding 25% total weight loss were killed in accordance with UK Home Office guidelines. Infected lungs were collected on days 1–6 post-infection and quantified for viral load by plaque assay and RT-qPCR with primers to influenza matrix 1 protein.

Pathology of infected Ifitm3−/− mice

5-μm sections of paraffin-embedded tissue were stained with haematoxylin and eosin and microscopically examined. Apoptosis was assessed by TUNEL using the TACS XL DAB In Situ Apoptosis Detection Kit (R&D Systems). Viral RNA was visualized by QuantiGene viewRNA kit (Affymetrix), with a viewRNA probe set designed to the negative stranded vRNA encoding the NP gene of A/X-31 (Affymetrix). Lung tissue was embedded in glycol methacrylate (GMA) and viral antigens stained using M149 polyclonal antibody to influenza A, B (Takara). Single cell suspensions from the lung were characterized by flow cytometry for T-lymphocytes CD4+ or CD8+, T-lymphocytes (activated) CD4+CD69+ or CD8+CD69+, neutrophils CD11bhiCD11c-Ly6g+, dendritic cells CD11c+CD11bloLy6glo MHC class II high, macrophages CD11b+CD11c+F4/80hi, natural killer cells NKp46+CD4CD8.

Sequencing and genetics of human IFITM3

The 1.8 kb of human IFITM3 was amplified and sequenced to identify single nucleotide polymorphisms (SNPs). SNP rs12252 was identified and compared to allele and genotype frequencies from 1000 Genomes sequencing data from different populations including 1000 Genomes imputed. SNP rs12252 allele frequencies were determined in the publicly available genotype data sets of WTCCC1 (n = 2,938) and previously published data sets genotyped from the Netherlands (n = 8,892) and Germany (n = 6,253)22.

Online Methods

Mouse infection

Background-matched wild-type (>95% C57BL/6) and Ifitm3−/− mice8 8–10 weeks of age were maintained in accordance with UK Home Office regulations, UK Animals Scientific Procedures Act 1986 under the project licence PPL80/2099. This licence was reviewed by The Wellcome Trust Sanger Institute Ethical Review Committee. Groups of >5 isofluorane-anaesthetized mice of both genotype were intranasally inoculated with 104 p.f.u. of A/X-31 influenza in 50 μl of sterile PBS. In some experiments A/X-31 was substituted with 200 p.f.u. of A/England/195/09 influenza, or 50–103 p.f.u. of A/PR/8/34 (PR/8) or an otherwise isogenic virus with a deletion of the NS1 gene (delNS1)19, made as described26. Their weight was recorded daily and they were monitored for signs of illness. Mice exceeding 25% total weight loss were killed in accordance with UK Home Office guidelines. Littermate controls were used in all experiments.

Influenza virus quantification

Lungs from five mice per genotype were collected on days 1, 2, 3, 4 and 6 post-infection, weighed and homogenized in 5% weight/volume (w/v) of Leibovitz’s L-15 medium (Invitrogen) containing antibiotic-antimycotic (Invitrogen). Samples were quantified for viral load by plaque assay in tenfold serial dilutions on Madin–Darby canine kidney (MDCK) cell monolayers overlaid with 1% Avicell medium27. Lungs were subjected to two freeze-thaw cycles before titration. Virus was also quantified by quantitative PCR with reverse transcription (qRT–PCR), wherein RNA was first extracted from lung, heart, brain and spleen using the RNeasy Mini Plus Kit (Qiagen). Purified RNA was normalized by mass and quantified with SYBR Green (Qiagen) using the manufacturer’s instructions and 0.5 μM primers for influenza matrix 1 protein (M1) forward: 5′-TGAGTCTTCTAACCGAGGTC-3′, reverse: 5′-GGTCTTGTCTTTAGCCATTCC-3′ (Sigma-Aldrich) and mouse β-actin (Actb) forward: 5′-CTAAGGCCAACCGTGAAAAG-3′, reverse: 5′-ACCAGAGGCATACAGGGACA-3′. qPCR was performed on a StepOnePlus machine (Applied Biosystems) and analysed with StepOne software v2.1 (Applied Biosystems).

Western blotting

Lungs were homogenized in 5% w/v of Tissue Protein Extraction Reagent (Thermo Scientific) containing cOmplete Protease Inhibitor (Roche). Total protein was quantified by BCA assay (Thermo Scientific) and was normalized before loading into wells. Proteins were visualized with the following indicated primary antibodies: anti-mouse IFITM2 rabbit polyclonal was purchased from Santa Cruz Biotechnology (catalogue no. sc-66828); anti-Fragilis (Ifitm3) rabbit polyclonal antibody was from Abcam (catalogue no. ab15592). The IFITM3 and NΔ21 western blot using the A549 stable cell lines were probed with the anti-IFITM1 antibody from Prosci (catalogue no. 5807), which recognizes a conserved portion of the IFITM1, IFITM2 and IFITM3 proteins which is still present even in the absence of the first twenty one N-terminal amino acids. The LCL blots (including the A549 cell line lysate controls) were probed with either an antibody which is specific for the N terminus of IFITM3 (rabbit anti-IFITM3 (N-terminal amino acids 8–38) (Abgent, catalogue no. AP1153a)), or with anti-IFITM1 antibody from Prosci (catalogue no. 5807), as well as rabbit anti-MX1 (Proteintech, catalogue no. 13750-1-AP) and mouse anti-GAPDH (clone GAPDH-71.1) (Sigma, catalogue no. G8795). For the LCL immunoblots all antibodies were diluted in DPBS (Sigma) containing 0.1% Tween 20 (Sigma) and 5% non-fat dried milk (Carnation) and incubated overnight at 4 °C. All primary antibodies were consequently bound to the corresponding species-appropriate horseradish peroxidase-conjugated secondary antibodies (Dako). Actin antibody was purchased from either Abcam or Sigma, mouse monoclonal, catalogue no. A5316.

Pathological examination

5-μm sections of paraffin-embedded tissue were stained with haematoxylin and eosin (Sigma-Aldrich) and were examined and scored twice, once by a pathologist under blinded conditions. The TUNEL assay for apoptosis was conducted using the TACS XL DAB In Situ Apoptosis Detection Kit (R&D Systems).

Immunofluorescent tissue staining: protein

Lung tissue was embedded in glycol methacrylate (GMA) to visualize the spread of viral protein, as described previously28. Briefly, 2-μm sections were blocked with 0.1% sodium azide and 30% hydrogen peroxide followed by a second block of RPMI 1640 (Invitrogen) containing 10% fetal calf serum (Sigma-Aldrich) and 1% bovine serum albumin (Invitrogen). Viral antigen was stained using M149 polyclonal antibody to influenza A, B (Takara) and visualized with a secondary goat anti-rabbit antibody conjugated to alkaline phosphatase (Dako). Sections were counterstained with haematoxylin (Sigma-Aldrich). Murine IFITM1 and IFITM3 protein expression in lung sections from either uninfected mice, or those 2 days post-infection with A/X-31, were immunostained with either anti-IFITM1 antibody (Abcam, catalogue no. ab106265) or anti-fragilis (anti-Ifitm3) rabbit polyclonal antisera (Abcam, catalogue no. ab15592). Sections were also stained for DNA with Hoechst 33342 (Sigma).

Immunofluorescent staining: RNA

Viral RNA was visualized in 5-μm paraffin-embedded sections using the QuantiGene viewRNA kit (Affymetrix). Briefly, sections were rehydrated and incubated with proteinase K. They were subsequently incubated with a viewRNA probe set designed against the negative stranded viral RNA encoding the NP gene of A/X-31 (Affymetrix). The signal was amplified before incubation with labelled probes and visualized.

Flow cytometry

Single-cell suspensions were generated by passing lungs twice through a 100-μm filter before lysing red blood cells with RBC lysis buffer (eBioscience) and assessing for cell viability via Trypan blue exclusion. Cells were characterized by flow cytometry as follows: T-lymphocytes CD4+ or CD8+, T-lymphocytes (activated) CD4+CD69+ or CD8+CD69+, neutrophils CD11bhiCD11cLy6g+, dendritic cells CD11c+CD11bloLy6glo MHC class II high, macrophages CD11b+CD11c+F4/80hi, natural killer cells NKp46+CD4CD8. All antibodies (Supplementary Table 3) were from BD Bioscience, except CD69 and F4/80, which were from AbD Serotec. Samples were run on a FACSAria II (BD Bioscience) and visualized using FlowJo 7.2.4. Data were analysed statistically and graphed using Prism 5.0 (GraphPad Software).

Peripheral leukocyte analysis

Mice (n = 3 per genotype per day) were bled on days 0, 1, 2, 3, 4 and 6 by tail vein puncture. Leukocyte counts were determined by haemocytometer, whereas blood cell differential counts were calculated by counting from duplicate blood smears stained with Wright–Giemsa stain (Sigma-Aldrich). At least 100 leukocytes were counted per smear. All blood analyses were conducted in a blinded fashion. Data were analysed statistically and graphed using Prism 5.0 (GraphPad Software).

Cytokine/chemokine analysis

Lungs were collected and homogenized on days 0, 1, 2, 3, 4 and 6 post-infection from four mice of each genotype. G-CSF, GM-CSF, IFN-γ, IL-10, IL-1α, IL-1β, IL-2, IL-4, IL-5, IL-6, IL-9, IP-10, KC-like, MCP-1, MIP-1α, RANTES and TNF-α were analysed using a mouse antibody bead kit (Millipore) according to the manufacturer’s instructions on a Luminex FlexMAP3D. Results were analysed and quality control checked using Masterplex QT 2010 and Masterplex Readerfit 2010 (MiraiBio). Data were analysed statistically and graphed using Prism 5.0 (GraphPad Software).

Murine embryonic fibroblast generation, transduction and infectivity assays

Adult Ifitm3−/− mice8 were intercrossed and fibroblasts (MEFs) were derived from embryos at day 13.5 of gestation, as described previously1. MEFs were genotyped by PCR (Thermo-Start Taq DNA Polymerase, ABgene) on embryo tail genomic DNA using primers and the cycle profile described previously8 to detect the presence of the wild-type allele (450 base pairs band) and the targeted/knockout allele (650 bp band). MEFs were cultured in DMEM containing 10% FBS, 1× MEM essential amino acids, 1× 2-mercapto-ethanol (Gibco). MEFs were transduced with vesicular stomatitis virus G (VSV-G) pseudotyped retroviruses expressing either the empty vector control (pQXCIP, Clontech), or one expressing Ifitm3, as previous described1. After puromycin selection the respective cell lines were challenged with either A/X-31 virus (multiplicity of infection (m.o.i.) 0.3–0.4) or PR/8 (m.o.i. 0.4). For PR/8 infections, after 12 h the media was removed and the cells were then fixed with 4% formalin and stained with purified anti-haemagglutinin monoclonal antibody (Hybridoma HA36-4-5.2, Wistar Institute). For A/X-31 experiments, cells were processed comparably as above, but in addition were permeabilized, followed by immunostaining for NP expression (NP (clone H16-L10-4R5) mouse monoclonal (Millipore MAB8800)). Both sets of experiments were completed using an Alexa Fluor 488 goat anti-mouse secondary antibody at 1:1,000 (A11001, Invitrogen). The cells were imaged on an automated Image Express Micro microscope (Molecular Devices), and images were analysed using the MetaMorph Cell Scoring software program (Molecular Devices). Cytokines: cells were incubated with cytokines for 24 h before viral infection. Murine interferon α (PBL Interferon Source, catalogue no. 12100-1) and IFN-γ (PBL Interferon Source, catalogue no. 12500-2) were used at 500–2,500 U ml−1, and 100–300 ng ml−1, respectively.

A549 transduction and infectivity assays

A549 cells (ATCC catalogue no. CCL-185) were grown in complete media (DMEM (Invitrogen catalogue no. 11965) with 10% FBS (Invitrogen)). A549 stable cell lines were made by gamma-retroviral transduction using either the empty vector control virus (pQXCIP, Clontech), the full-length human IFITM3 complementary DNA, or a truncated human IFITM3 cDNA which is missing the first 21 amino acids (NΔ21). After puromycin selection, expression of the IFITM3 and NΔ21 proteins were confirmed by western blotting using an 18% SDS–PAGE gel and an anti-IFITM3 antibody that was raised against the conserved intracellular loop (CIL) of IFITM3 (Proteintech). A549 cell lines were challenged with one of the following strains: A/WSN/33 (a gift of P. Palese), A/California/7/2009, A/Uruguay/716/2007 and B/Brisbane/60/2008 (gift of J. Malbry) for 12 h, then fixed with 4% paraformaldehyde (PFA) and immunostained with anti-HA antibody (Wistar collection) or anti-NP antibodies (Abcam), or Millipore clone H16-L10-4R5 anti-influenza A virus antibody). Per cent infection was calculated from immunofluorescent images as described for the MEF experiments above. Alternatively, cells were transduced with lentiviral vectors to express green fluorescent protein (GFP) or IFITM3 and were stained with anti-NP antibody (Abcam) and analysed by flow cytometry following challenge with B/Bangladesh/3333/2007 virus (NIMR, England). For the immunofluorescence-based viral titring experiments, virus-containing supernatant was collected from the indicated A549 cell line cultures after 12 h of infection with WSN/33 (part one). Next this supernatant was used to infect MDCK cells (ATCC) in a well by well manner (part two). Both the A549 and MDCK cells were then processed to detect viral HA expression as described above.

LCL infectivity assays

LCL TT and LCL CC cells were grown in RPMI-1640 (Sigma-Aldrich) containing 10% FCS, 2 mM l-glutamine, 1 mM sodium pyruvate, 1× MEM non-essential amino acids solution, and 20 mM HEPES (all from Invitrogen). For infectivity assays, LCL cells were either treated with recombinant human IFN-α2 (PBL Interferon Source, catalogue no. 11100) at 100 units per ml or DPBS (Sigma-Aldrich) for 16 h. The LCL cells were then counted, resuspended at a concentration of 5 × 105 cells per ml, and plated on a 96-well round-bottom plate (200 μl cell suspension per well). The cells were then challenged with WSN/33 influenza A virus (m.o.i. 0.1). After 18 h, the cells were washed twice with 250 μl MACS buffer (DPBS containing 2% FCS and 2 mM EDTA (Sigma-Aldrich)). The cells were fixed and permeabilized using the BD Cytofix/Cytoperm Fixation/Permeabilization Kit (BD Biosciences), following the manufacturer’s instructions. Briefly, the cells were resuspended in 100 μl of Cytofix/Cytoperm Fixation and Permeabilization solution and incubated at 4 °C for 20 min. The cells were then washed twice with 250 μl 1× Perm/Wash buffer and resuspended in 50 ml 1× Perm/Wash buffer containing a 2 μg ml−1 solution of a fluorescein isothiocyanate (FITC)-conjugated mouse monoclonal antibody against influenza A virus NP (clone 431, Abcam, catalogue no. ab20921). The cells were incubated in the diluted antibody solution for 1 h at 4 °C, washed twice with 250 μl 1× Perm/Wash buffer, resuspended in 200 μl MACS buffer, and analysed by flow cytometry using a BD FACS Calibur (BD Biosciences).

Ethics and sampling

We recruited patients with confirmed seasonal influenza A or B virus or pandemic influenza A pH1N1/09 infection who required hospitalization in England and Scotland between November 2009 and February 2011. Patients with significant risk factors for severe disease and patients whose daily activity was limited by co-morbid illness were excluded. 53 patients, 29 male and 24 female, average age 37 (range 2–62) were selected. 46 (88%) had no concurrent co-morbidities. The remaining 6 had the following comorbid conditions: hypertension (3 patients), alcohol dependency and cerebrovascular disease (1 patient), bipolar disorder (1 patient) and kyphoscoliosis (1 patient). Four patients were pregnant. Where assessed, 36 patients had normal body mass (69%), one had a body mass index <18.5 and 10 had a body mass index between 25 and 39.9 and one a body mass index >40. Seasonal influenza A H3N2, influenza B and pandemic influenza A pH1N1/09 were confirmed locally by viral PCR or serological tests according to regional protocols. Consistent with the prevalent influenza viruses circulating in the UK between 2009 and 2011 (ref. 29) 44 (85%) had pH1N1/09, 2 had pH1N1/09 and influenza B co-infection, 4 had influenza B and 2 had non-subtyped influenza A virus infection. Of the adults, 24 required admission to an intensive care unit (ICU) and 1 required admission to a high dependency unit (HDU). The remainder were managed on medical wards and survived their illnesses. The GenISIS study was approved by the Scotland ‘A’ Research Ethics Committee (09/MRE00/77) and the MOSAIC study was approved by the NHS National Research Ethics Service, Outer West London REC (09/H0709/52, 09/MRE00/67).

Consent was obtained directly from competent patients, and from relatives/friends/welfare attorneys of incapacitated patients. Anonymized 9-ml EDTA blood samples were transported at ambient temperature. DNA was extracted using a Nucleon Kit (GenProbe) with the BACC3 protocol. DNA samples were re-suspended in 1 ml TE buffer pH 7.5 (10 mM Tris-Cl pH 7.5, 1 mM EDTA pH 8.0).

Sequencing and genetics

Human IFITM3 sequences were amplified from DNA obtained from peripheral blood by nested PCR (GenBank accession numbers JQ610570 to JQ610621). The first round used primers forward: 5′-TGAGGGTTATGGGAGACGGGGT-3′and reverse: 5′-TGCTCACGGCAGGAGGCC-3′, followed by an additional round using primers forward: 5′-GCTTTGGGGGAACGGTTGTG-3′and reverse: 5′- TGCTCACGGCAGGAGGCCCGA-3′. The 1.8-kb IFITM3 band was gel-extracted and purified using the QIAquick Gel Extraction Kit (Qiagen). IFITM3 was Sanger-sequenced on an Applied Biosystems 3730xl DNA Analyzer (GATC Biotech) using a combination of eight sequencing primers (Supplementary Table 4). Single-nucleotide polymorphisms were identified by assembly to the human IFITM3 encoding reference sequence (accession number NC_000011.9) using Lasergene (DNAStar). Homozygotes were called based on high, single base peaks with high Phred quality scores, whereas heterozygotes were identified based on low, overlapping peaks of two bases with lower Phred quality scores relative to surrounding base calls (Supplementary Fig. 9). We identified SNP rs12252 in our sequencing and compared the allele and genotype frequencies to allele and genotype frequencies from 1000 Genomes sequencing data from different populations (Supplementary Table 3). In addition, we used the most recent release of phased 1000 Genomes data30 to impute the region surrounding SNP rs12252 to determine allele frequencies in the publicly available genotype data set of WTCCC1 controls (n = 2,938) and four previously published data sets genotyped from the Netherlands (n = 8,892) and Germany (n = 6,253)22. In the imputation, samples genotyped with Illumina 550k, 610k and 670k platforms were imputed against the June 2011 release of 1000 Genotypes phased haplotypes using the Impute software31, version 2.1.2. Only individuals with European ethnicities (Europe (CEU), Finland (FIN), Great Britain (GBR), Spain/Iberia (IBS), Tuscany (TSI)) were included from the 1000 Genomes reference panel. Recommended settings were used for imputing the region 200 kb in either direction from the variants of interest, along with 1 Mb buffer size. The statistical significance of the allele frequencies was determined by Fisher’s exact test.

We assessed for population stratification by principal component analysis. Genotype data from the WTCCC1 1958 Birth Cohort data set were obtained from the European Genotype Archive with permission, reformatted and merged with genotype data from the GenISIS study to match 113,819 SNPs present in both cohorts. Suspected strand mismatches were removed by identifying SNPs with more than 2 genotypes and using the LD method as implemented in Plink (v1.07)32, resulting in 105,362 matched SNPs. Quality control was applied in GenABEL version 1.6-9 to genotype data for these SNPs for the GenISIS cases and 1,499 individuals from WTCCC. Thresholds for quality control (deviation from Hardy–Weinberg equilibrium (P < 0.05), minor allele frequency (MAF) < 0.0005, call rate <98% in all samples) were applied iteratively to identify all markers and subjects passing all quality control criteria, followed by principal component analysis using GenABEL. We tested for positive selection using both a haplotype-based test (|XP-EHH-max|) and allele frequency spectrum-based test statistics, namely the CLR23,24,25 on 10-kb windows across the entire genome as described previously30,33. The three statistics were combined and the combined P value was plotted corresponding to the 10-kb windows.