The Journal of Immunology

Role of Double-Stranded RNA Pattern Recognition Receptors in Rhinovirus-Induced Airway Epithelial Cell Responses1

Qiong Wang,* Deepti R. Nagarkar,* Emily R. Bowman,† Dina Schneider,† Babina Gosangi,† Jing Lei,† Ying Zhao,† Christina L. McHenry,† Richai V. Burgens,† David J. Miller,‡ Umadevi Sajjan,† and Marc B. Hershenson2*†

Rhinovirus (RV), a ssRNA of the family, is a major cause of the as well as and chronic obstructive pulmonary disease exacerbations. Viral dsRNA produced during replication may be recognized by the host pattern recognition receptors TLR-3, retinoic acid-inducible gene (RIG)-I, and melanoma differentiation-associated gene (MDA)-5. No study has yet identified the receptor required for sensing RV dsRNA. To examine this, BEAS-2B human bronchial epithelial cells were infected with intact RV-1B or replication-deficient UV-irradiated virus, and IFN and IFN-stimulated gene expression was determined by quantitative PCR. The separate requirements of RIG-I, MDA5, and IFN response factor (IRF)-3 were determined using their respective small interfering (siRNA). The requirement of TLR3 was determined using siRNA against the TLR3 adaptor molecule Toll/IL-1R homologous region-domain-containing adapter-inducing IFN-␤ (TRIF). Intact RV-1B, but not UV-irradiated RV, induced IRF3 phosphorylation and dimerization, as well as mRNA expression of IFN-␤, IFN-␭1, IFN-␭2/3, IRF7, RIG-I, MDA5, 10-kDa IFN-␥-inducible /CXCL10, IL-8/CXCL8, and GM-CSF. siRNA against IRF3, MDA5, and TRIF, but not RIG-I, decreased RV-1B-induced expression of IFN-␤, IFN-␭1, IFN-␭2/3, IRF7, RIG-I, MDA5, and inflammatory protein-10/CXCL10 but had no effect on IL-8/CXCL8 and GM-CSF. siRNAs against MDA5 and TRIF also reduced IRF3 dimerization. Finally, in primary cells, transfection with MDA5 siRNA significantly reduced IFN expression, as it did in BEAS-2B cells. These results suggest that TLR3 and MDA5, but not RIG-I, are required for maximal sensing of RV dsRNA and that TLR3 and MDA5 signal through a common downstream signaling intermediate, IRF3. The Journal of Immunology, 2009, 183: 6989–6997.

iral , most commonly caused by rhinovirus localized to the endosomal and plasma membranes. TLR3 senses (RV),3 are a frequent cause of asthma and chronic ob- dsRNA released from dying cells and signals through its unique V structive pulmonary disease exacerbations (1). RV is a adaptor protein-Toll/IL-1R homologous region (TIR)-domain-con- nonenveloped, positive, ssRNA virus from the Picornaviridae taining adapter-inducing IFN-␤ (TRIF; Ref. 2). The cytoplasmic family. RV is internalized by receptor-mediated endocytosis and retinoic acid-inducible gene (RIG)-I and melanoma dif- undergoes a conformational change at endosome low pH, leading ferentiation-associated gene (MDA)-5 have recently been identi- to insertion of viral RNA into the cytosol. After entry, replication fied as intracellular receptors for viral dsRNA (3, 4). RIG-I and occurs entirely in the cytoplasm, where ssRNA forms a dsRNA MDA5 are homologous cytoplasmic helicases containing two N- intermediate, the main form of viral RNA inside the cell. terminal caspase activation and recruitment domains and a C-ter- dsRNA produced during viral represents an important minal DExD/H-Box RNA helicase domain. They bind to dsRNA stimulus of the host innate immune response. It is recognized and through the helicase domain and signal through caspase activation engaged by three pattern recognition receptors (PRR). TLR3 is and recruitment domains to a common adaptor molecule, IFN-␤ promoter stimulator-1 (also called VISA; Refs. 5 and 6). Engage- ment of TLR3, RIG-I, or MDA5 initiates signaling through two † *Department of Molecular and Integrative Physiology, Department of Pediatrics and protein kinase complexes, TANK-binding kinase/I␬B kinase Communicable Diseases, and ‡Department of Internal Medicine, University of Mich- igan, Ann Arbor, MI 48109 (IKK)-␧ and IKK␣/IKK␤, leading to activation of IFN-regulated ␬ Received for publication May 4, 2009. Accepted for publication September 29, 2009. factor (IRF)-3 and NF- B, respectively (7). Transcription factor The costs of publication of this article were defrayed in part by the payment of page activation, in turn, induces expression of IFNs and proinflamma- charges. This article must therefore be hereby marked advertisement in accordance tory . with 18 U.S.C. Section 1734 solely to indicate this fact. Although all three receptors can recognize viral dsRNA, they 1 This work was supported by National Institutes of Health Grants HL81420 and appear to be specialized in their recognition of particular . 82550 (to M.B.H.). RIG-I and TLR3 are required for respiratory syncytial virus-in- 2 Address correspondence and reprint requests to Dr. Marc B. Hershenson, University duced expression of IFN-␤, 10-kDa IFN-␥-inducible protein (IP- of Michigan, 1150 West Medical Center Drive, Room 3570, MSRBII, Box 5688, Ann Arbor, MI 48109-0688. E-mail address: [email protected] 10) in airway epithelial cells (8). RIG-I-deficient mice fail to pro- 3 Abbreviations used in this paper: RV, rhinovirus; TRIF, TIR-domain-containing duce type I IFNs in response to the negative-sense ssRNA viruses adapter-inducing IFN-␤; RIG, retinoic acid-inducible gene; MDA, melanoma differ- Newcastle disease virus, Sendai virus, vesicular stomatitis virus, entiation-associated gene; IKK, I␬B kinase; IRF, IFN response factor; IP-10, 10-kDa IFN-␥-inducible protein; EMCV, encephalomyocarditis; DC, dendritic cell; BEGM, and influenza virus, and to the positive-sense ssRNA Japanese en- bronchial epithelial growth medium; ISG, IFN-stimulated gene; TIR, Toll/IL-1R ho- cephalitis virus, whereas MDA5-deficient mice fail to detect en- mologous region; PRR, pattern recognition receptor; poly(IC), polyinosinic-polyade- cephalomyocarditis (EMCV), a positive-sense ssRNA picornavi- nylic acid. rus (9). The engagement of PRRs is also cell-type specific: for Copyright © 2009 by The American Association of Immunologists, Inc. 0022-1767/09/$2.00 example, whereas MDA5 is essential for induction of type I IFNs

www.jimmunol.org/cgi/doi/10.4049/jimmunol.0901386 6990 PRRs AND RV INFECTION

after infection with EMCV in fibroblasts and conventional mM glycine, pH 8.4, with and without 1% deoxycholate in the cathode dendritic cells (DC), plasmacytoid DCs use the TLR system for and anode chambers, respectively, for 30 min at 40 mA. Samples in the ␮ viral detection (9). native sample buffer (10 g of protein, 62.5 mM Tris-Cl (pH 6.8), 15% glycerol) were applied to the gel and electrophoresed for 60 min at 25 Little is known about the contributions of the various pattern mA. Immunoblotting of IRF3 was performed as described above in recognition receptors to RV-induced responses in bronchial epi- “Immunoblotting.” thelial cells. Primary human bronchial epithelial cells express Small interfering RNA (siRNA) knockdown of RIG-I, MDA5, TLR3, and the TLR3 ligand poplyinosinic-polycytidylic acid IRF3, and TRIF (poly(IC)) elicits a strong proinflammatory response in these cells (10, 11). In 16HBE14oϪ human bronchial epithelial cells, TLR3 The 19-bp duplex of targeting siRNA or nontargeting siRNA (Dharmacon) is primarily localized in the endosomes but not cell surface (12). was transfected into subconfluent BEAS-2B cells while cell seeding, using RNAiMAX in OptiMEM (Invitrogen). A pool of double-stranded siRNAs TLR3 is partially required for RV39-induced IL-8 expression in containing equal parts of the following antisense sequences was used to 16HBE14oϪ cells (12) and RV1A-induced MUC5AC expression knock down RIG-I: 1, 5Ј-GCACAGAAGUGUAUAUUGG-3Ј;2,5Ј-CC in NCI-H292 mucoepidermoid carcinoma cells (13). However, the ACAACACUAGUAAACAA-3Ј;3,5Ј-CGGAUUAGCGACAAAUUUA- requirement of either RIG-I or MDA5 for RV-induced responses 3Ј;4,5Ј-UCGAUGAGAUUGAGCAAGA-3Ј. The nontargeting siRNA se- quence was 5Ј-CGAACUCACUGGUCUGACCdtdt-3Ј (sense), 5Ј-GGUC has not yet been tested. In this study, we found that MDA5 and AGACCAGUGAGUUCG-dtdt-3Ј (antisense). For knockdown of MDA5, TLR3, but not RIG-I, are required for RV-induced IFN responses a pool of the following sequences was used: 1, 5Ј-GAAUAACCCAUCAC in human airway epithelial cells. UAAUA-3Ј;2,5Ј-CACGAGGAAUAAUCUUUA-3Ј;3,5Ј-UGACACAA UUCGAAUGAUA-3Ј;4,5Ј-CAAUGAGGCCCUACAAAUU-3Ј. For Materials and Methods knockdown of IRF3, a pool of the following sequences was used: 1, 5Ј- CGAGGCCACUGGUGCAUAU-3Ј;2,5Ј-CCAGACACCUCUCCGGA Cell culture CA-3Ј;3,5Ј-GGAGUGAUGAGCUACGUGA-3Ј,4,5Ј-AGACAUUCUG Ј BEAS-2B human bronchial epithelial cells, a SV40-transformed airway GAUGAGUUA-3 . For knockdown of TRIF, a pool of the following Ј Ј Ј bronchial epithelial cell line, were purchased from American Type Culture sequences was used: 1, 5 -GGAGCCACAUGUCAUUUGG-3 ;2,5- Ј Ј Collection (ATCC). Cells were grown on collagen-coated (5 ␮g/cm2) CCAUAGACCACUCAGCUUU-3 ;3,5-GGACGAACACUCCCAGA Ј Ј Ј plates in bronchial epithelial growth medium (BEGM; Lonza) containing UC-3 ;4,5-CCACUGGCCUCCCUGAUAC-3 . The next morning, cells epidermal growth factor (25 ng/ml), bovine pituitary extract (65 ng/ml), were incubated in fresh BEGM for 24 h. Finally, cells were treated with the all-trans-retinoic acid (5 ϫ 10Ϫ8 M), hydrocortisone (0.5 ␮g/ml), insulin (5 relevant stimulus in BEGM medium for 1 day before harvest. ␮ ␮ ␮ g/ml), transferrin (10 g/ml), epinephrine (0.5 g/ml), triiodothyronine Data analysis (6.5 ng/ml), gentamicin (50 ␮g/ml), and amphotericin (50 ␮g/ml). Primary tracheal epithelial cells were all isolated from the tracheas of SigmaStat computing software (SPSS) was used for data analysis. Data are transplant donors, as described (14, 15). Submerged cells were grown represented as means Ϯ SEM. Normality was tested using the Kolmog- as monolayers to 80–100% confluence in BEGM containing epidermal orov-Smirnov test. Statistical significance was assessed by either one-way growth factor (25 ng/ml), bovine pituitary extract (130 ng/ml), all-trans- ANOVA or the Kruskal-Wallis one-way ANOVA based on ranks, when- Ϫ retinoic acid (5 ϫ 10 8 M), and BSA (1.5 ␮g/ml). ever appropriate. Differences identified by ANOVA were pinpointed by the Student-Newman-Keuls multiple range test. RV infection RV1B and RV39 were obtained from ATTC. Viral stocks were generated Results by incubating HeLa cells with RV in serum-free medium until 80% of the RV1B-induced IFN expression in BEAS-2B human bronchial cells were cytopathic. Viral stocks were concentrated, partially purified, epithelial cells and titered as previously described (14, 15). RV1B was irradiated with UV light at 100 ␮J/cm2 for 10 min on ice, using a CL-1000 cross-linker (UVP). To test whether RV induces an IFN response in human bronchial epithelial cells, BEAS-2B cells were infected with intact RV1B or Quantitative real-time PCR (qPCR) UV-irradiated RV-1B for1hat33°C. Cellular total RNA was Total RNA was extracted using the RNeasy kit (Qiagen) and then transci- extracted from lysates to measure the gene expression at 1, 8, 18, bed to first-strand cDNA using Taqman Reverse Transcription Reagents 24, 48, and 72 h postinfection by quantitative PCR. Compared with (Applied Biosystems). First-strand cDNA is then used to quantify the ex- ␣ ␤ ␭ ␭ replication-deficient UV-irradiated virus, intact RV1B increased pression of IFN- , IFN- , IFN- 1, IFN- 2/3, IRF-7, IP-10/CXCL-10, IL- ␤ ␭ ␭ 8/CXCL8, and GM-CSF mRNA levels by qPCR using specific primers and the mRNA expression of IFN- , IFN- 1, and IFN- 2/3, as well as probes. the expression of several IFN-stimulated genes (ISG) including IRF7, RIG-I, MDA5, and TLR3 (Fig. 1). The peak level of mRNA Measurement of protein levels expression was 24 h postinfection, except in the case of IFN-␤, BEAS-2B cells were grown to 80% confluence and infected with RV1B or which was 48 h postinfection The fold-induction varied widely, medium alone for 1 h. Inoculum was then replaced with BEGM. After 24 h, from 4-fold (TLR3) to ϳ27,000-fold (IFN-␭2/3). Fold increases in supernatant was collected for the measurement of IFN-␭1, IP-10/CXCL- 10, and IL-8/CXCL-8 protein by ELISA (R&D Systems). IFN and ISG mRNA expression tended to be higher for genes expressed at lower levels as baseline, as measured by cycle number Immunoblotting (Table I). RV1B infection also increased the protein expression of Cells were lysed with radioimmunoprecipitation buffer (50 mM Tris (pH IFN-␭1, IP-10/CXCL-10 and IL-8/CXCL-8 (Fig. 2, A–C). How- 7.4), 150 mM NaCl, 5 mM EDTA, 50 mM NaF, 1% Nonidet P-40, 10% ever, there was no induction of IRF7 protein expression (Fig. 2D). glycerol). Into each well were loaded 30 ␮g of protein lysate. Lysates were Nevertheless, these data, combined with increases in RIG-I and resolved by SDS-PAGE. Proteins were transferred to nitrocellulose or MDA5 protein abundance (see below), demonstrate that RV in- polyvinylidene difluoride membrane. Membranes were blocked in 5% milk for1hatroom temperature and probed with mouse anti-RIG-I (Alexis duces a robust, replication-dependent innate immune response at Biochemicals), goat anti-MDA5 (Santa Cruz Biotechnology), rabbit anti- both the mRNA and protein levels. RV1B infection also increased IRF3 (IBL America), rabbit anti-TRIF (Cell Signaling), or rabbit anti-IRF7 IFN and ISG mRNA expression in primary tracheal epithelial cells (Abcam). Ab binding was detected with a peroxidase-conjugated anti- (Fig. 3 and Table II). mouse, anti-rabbit, or anti-goat IgG and chemiluminescence. Native PAGE to determine IRF3 dimerization MDA5 and TLR3, but not RIG-I, are required for RV-induced innate immune responses BEAS-2B cells were lysed with native PAGE sample preparation kit (Invitrogen). Native PAGE was performed using 10% Ready Gel (Bio- Viral dsRNA generated during replication may be detected by the Rad), as described (16). The gel was prerun with 25 mM Tris and 192 PRRs RIG-I, MDA5, and/or TLR3. To determine which PRR is The Journal of Immunology 6991

FIGURE 2. RV1B-induced protein expression of IFNs and ISGs. BEAS-2B cells were grown to near confluence and infected with RV1B, UV-irradiated RV1B, or sham-infected. A–C, Medium supernatants were extracted 24 h after infection for ELISA to determine the expression of IFN-␭1, IP-10, and IL-8. D, Protein lysates were used to determine IRF7 expression by immunoblotting.

gesting that expression of RIG-I is inducible. RIG-I siRNA had a slight inhibitory effect on the expression of its homolog protein, MDA5 (Fig. 4B). However, RIG-I siRNA failed to decrease RV- FIGURE 1. Fold increase of RV1B-induced IFN and ISG responses in induced expression of IFNs or ISGs compared with nontargeting cultured airway epithelial cells. BEAS-2B cells were infected with RV1B siRNA, suggesting that RIG-I is not required for sensing RV or sham-infected (1 h at 33°C). Total RNA was extracted at 1, 8, 16, 24, dsRNA (Fig. 5). 48, and 72 h after infection. A–G, The expression of IFN-␤, IFN-␭1, IFN- Next, we knocked down MDA5 expression using MDA5 siRNA ␭2/3, IRF7, RIG-I, MDA5, and TLR3 at each time point was determined (Fig. 4C). Again, an 80–90% knockdown efficiency was achieved. by qPCR. The expression of each target gene was normalized to GAPDH. MDA5 protein expression was also induced by RV1B infection. Expression levels are represented as the ratio of the response to intact RV MDA5 siRNA had no effect on RIG-I protein expression (Fig. 4D). vs the response to sham. Data represent means Ϯ SEM for three After transfection with MDA5 siRNA, there were significant de- experiments. creases in RV1B-induced IFN-␤, IFN-␭1, and IFN-␭2/3 mRNA expression and IFN-␭1 protein expression compared with cells responsible for sensing RV dsRNA and inducing the innate im- transfected with nontargeting siRNA (Fig. 6). MDA5 siRNA also mune response, we used siRNA against RIG-I, MDA5, and the decreased RV1B-induced mRNA expression of the ISGs IRF7 and TLR3 adaptor protein TRIF/TIR domain-containing adapter mol- IP-10/CXCL10. Unlike IFNs, MDA5 siRNA treatment was asso- ecule. Forty-eight h later, cells were infected with RV1B, UV- ciated with a paradoxic increase in RV1B-induced expression of irradiated RV-1B, or sham HeLa cell lysate, and the expression of IL-8 and GM-CSF. Together, these data suggest that MDA5, but IFNs and ISGs was measured by qPCR 24 h after infection. RIG-I not RIG-I, is required for maximal RV-induced IFN responses, but expression was knocked down by 80–90% after treatment with not for RV-induced cytokine responses. RIG-I siRNA (Fig. 4A). Immunoblots also showed a significant To examine whether MDA5 is also required for major group increase in RIG-I protein expression with RV1B treatment, sug- RV-induced IFN responses, we repeated our experiment using RV39. BEAS-2B cells were transfected with either MDA5 or non- targeting siRNA and then infected with RV39, UV-irradiated

Table I. Effect of RV1B infection on mean cycle threshold (Ct) values RV39 or sham HeLa cell lysate. Twenty-four hours after infection, of IFN and ISG mRNA expression in BEAS-2B cells cellular RNA was extracted to measure the expression of IFNs and ISGs by quantitative PCR (Fig. 7). There was a significant de- Sham RV1B crease in mRNA expression of IFN-␤, IFN-␭1, and IFN-␭2/3 in cells transfected with MDA5 siRNA compared with cells trans- Ct average SE Ct average SE fected with nontargeting siRNA. MDA5 siRNA also decreased IFN-␤ 39.47 0.53 30.91 0.296 RV39-induced mRNA expression of the ISGs IRF7 and ␭ IFN- 1 34.80 0.16 24.23 0.20 IP-10/CXCL10. IFN-␭2/3 39.67 0.33 25.30 0.21 IRF7 24.54 0.20 21.38 0.15 We then sought to determine whether MDA-5 is required for the RIG-I 26.97 0.29 21.19 0.19 recognition of viral dsRNA in primary tracheal epithelial cells as MDA5 26.50 0.20 21.64 0.11 in BEAS-2B cells. Primary cells were cultured until 70% confluent TLR3 25.47 0.16 22.80 0.04 and then transfected with siRNA against MDA5 or nontargeting 6992 PRRs AND RV INFECTION

Table II. Effect of RV1B infection on mean Ct values of IFN and ISG mRNA expression in primary cells

Sham RV1B

Ct average SE Ct average SE IFN-␤ 35.97 2.14 32.75 2.72 IFN-␭1 33.68 1.69 29.06 2.02 IFN-␭2/3 35.51 1.64 28.47 2.18 IRF7 29.94 1.26 29.87 0.16 RIG-I 28.43 0.31 28.22 0.59 IP-10 30.78 1.00 28.04 0.02

and IRF3 dimerization determined by native PAGE (Fig. 10C). Poly(IC), a synthetic dsRNA that induces both IRF3 phosphory- lation and dimerization, served as a positive control. We found that RV1B, but not UV-irradiated RV1B, induced IRF3 phosphoryla- tion and dimerization. Similar results were observed in primary tracheal epithelial cells (Fig. 10D).

FIGURE 3. RV1B-induced expression of IFNs, ISGs, and in primary tracheobronchial epithelial cells. Primary tracheobron- chial epithelial cells were grown to near confluence and infected with RV1B or sham. A, Total RNA was extracted 24 h after infection, and the expression of IFN-␤, IFN-␭1, IFN-␭2/3, IRF7, IP-10, IL-8, and GM-CSF was determined by qPCR. Expression levels are represented as the ratio of the response to intact RV vs the response to sham. The y-axis is in log scale. Symbols represent the results of three different experiments. B–G, Time course of RV1B-induced responses in primary cells. siRNA. After 48 h, cells were infected with RV1B or UV-irradi- ated RV1B. Cellular mRNA was extracted 24 h after infection to determine mRNA expression by qPCR. Compared with nontarget- ing siRNA, MDA5 siRNA decreased RV1B-induced expression of ␤ ␭ ␭ IFN- , IFN- 1, IFN- 2/3, IRF7, and IP-10/CXCL10 (Fig. 8). These data confirm that MDA5 is required for sensing RV dsRNA and induction of the subsequent IFN response in primary cells. Next. we blocked TLR3 signaling using siRNA against TRIF/ TIR domain-containing adapter molecule, the TLR3 adaptor mol- ecule. Again, a high knockdown efficiency was verified by immu- noblotting (Fig. 9). Like MDA5 siRNA, TRIF siRNA abolished ␤ ␭ ␭ RV1B-induced expression of IFN- , IFN- 1, IFN- 2/3, IRF7, and IP-10/CXCL10, suggesting that TLR3 signaling is also required FIGURE 4. RIG-I and MDA5 siRNA knockdown efficiencies. A and B, for maximal RV1B-induced IFN responses. RIG-I or nontargeting siRNA was transfected into BEAS-2B cells. After transfection, cells were infected with RV1B, UV-irradiated RV1B (UV- IRF3 is required for RV1B-induced IFN responses RV1B) or sham. After infection, cell lysates were probed with Abs against RIG-I (A) or MDA5 (B). Note the inductions in RIG-I and MDA5 expres- IRF3 is a ubiquitously expressed transcription factor that regulates sion with intact RV, as well as the apparent degradation of MDA-5 fol- type I IFN production. To test whether RV1B induces IRF3 acti- lowing viral infection. C and D, MDA5 or nontargeting siRNA was trans- vation, BEAS-2B cells were infected with RV1B or UV-irradiated fected into BEAS-2B cells. After transfection, cells were infected with RV1B. Cell protein lysates were collected 12 h after infection. RV1B, UV-irradiated RV1B (UV-RV1B), or sham infected. After infec- IRF3 phosphorylation shift was determined by SDS-PAGE fol- tion, cell lysates were probed with Abs against either MDA5 (C) or RIG-I lowed by immunoblotting using anti-IRF3 Ab (Fig. 10, A and B), (D). Blots are representative of three separate experiments. The Journal of Immunology 6993

FIGURE 5. siRNA against RIG-I does not block RV1B-induced IFN FIGURE 6. siRNA against MDA5 blocks RV1B-induced IFN and ISG and ISG expression. RIG-I-specific or nontargeting siRNA was transfected expression. MDA5-specific or nontargeting siRNA was transfected into into BEAS-2B cells. After transfection, cells were infected with RV1B, BEAS-2B cells. After transfection, cells were infected with RV1B, UV- UV-irradiated RV1B (UV-RV1B), or sham infected. A–G, Total RNA was irradiated RV1B (UV-RV1B), or sham infected. A–G, Total RNA was extracted, and the expression of IFN-␤, IFN-␭1, IFN-␭2/3, IRF7, IP-10, extracted and the expression of IFN-␤, IFN-␭1, IFN-␭2/3, IRF7, IP-10, IL-8, and GM-CSF was determined by qPCR. The expression of each IL-8 and GM-CSF was determined by qPCR. The expression of each target target gene was normalized to GAPDH. Expression levels are represented gene was normalized to GAPDH. Expression levels are represented as the as the fold increase vs sham-infected, nontargeting siRNA-transfected fold increase vs sham-infected, nontargeting siRNA-transfected cells. The cells. The y-axis has been broken to show the effects of siRNA on both y-axis has been broken to show the effects of siRNA on both basal and basal and maximal gene expression. Bars, means Ϯ SEM for four exper- maximal gene expression. H, To determine the protein expression of IFN- iments; numbers over bars indicate the fold increase (ϫ) compared with ␭1, cell supernatants were collected 24 h after infection for ELISA. Bars p Ͻ 0.05 vs RIG-I ,ء .sham-infected sample within its own siRNA group represent mean Ϯ SEM for three to five experiments; numbers over bars siRNA-transfected RV1B-infected sample, ANOVA. siNT, Non-targeting indicate the fold increase compared with sham-infected sample within its siRNA; siRIG-I, siRNA against RIG-I. p Ͻ 0.05 vs MDA5 siRNA-transfected RV-infected ,ء .own siRNA group sample, ANOVA. siNT, Non-targeting siRNA; siMDA-5, siRNA against MDA-5. We then examined the requirement of IRF3 in RV-induced IFN responses using siRNA against IRF3. IRF3 protein abundance was substantially knocked down by IRF3 siRNA (Fig. 11). IRF3 RV1B, UV-irradiated RV1B, or sham HeLa cell lysate. IRF3 ␤ ␭ siRNA nearly abolished RV1B-induced expression of IFNs- , 1, dimerization was resolved by native PAGE (Fig. 12). We found and 2, 3; IRF7; and IP-10/CXCL10. However, there was no effect that MDA5 and TRIF siRNA each caused a partial reduction in on IL-8 or GM-CSF expression. Taken together, these data suggest RV-induced IRF3 dimerization compared with nontargeting that IRF3 is activated by RV1B and that IRF3 is required for siRNA-transfected cells, confirming in airway epithelial cells the RV-induced IFN responses, as well as the expression of ISGs such general notion that IRF3 functions downstream of MDA5 as IRF7 and IP-10. and TRIF.

IRF3 functions downstream of MDA5 signaling Discussion RIG-I/MDA5 and TLR3 signaling pathways converge on a com- Host pathogen recognition, as reflected by the induction of type I mon TNF receptor-associated factor 3 adapter complex, which IFNs, is mediated by activation of PRRs. The membrane dsRNA then activates two IRF3 kinases, TANK-binding kinase and IKK-␧ receptor, TLR3, and the recently identified cytoplasmic dsRNA (7). To examine whether IRF3 functions downstream of MDA5 in receptors, RIG-I and MDA5, are responsible for sensing viral airway epithelial cells, BEAS-2B cells were transfected with either dsRNA (2–4). Although all three receptors can recognize viral MDA5 siRNA or nontargeting siRNA and then infected with dsRNA, the engagement of receptor and viral dsRNA seem to be 6994 PRRs AND RV INFECTION

FIGURE 8. siRNA against MDA5 blocks RV1B-induced IFN re- sponses in primary tracheal epithelial cells. A, MDA5 siRNA or nontar- geting siRNA was transfected into primary tracheobronchial epithelial FIGURE 7. siRNA against MDA5 blocks RV39-induced IFN and ISG cells. After transfection, cells were infected with RV1B or UV-irradiated expression. A–G, Total RNA was extracted, and the expression of IFN-␤, RV1B. After infection, cell lysates were probed with anti-MDA5 Ab. Blot IFN-␭1, IFN-␭2/3, IRF7, IP-10, IL-8, and GM-CSF was determined by is typical for three experiments. B–H, Total RNA was extracted and the qPCR. The expression of each target gene was normalized to GAPDH. expression of IFN-␤, IFN-␭1, IFN-␭2/3, IRF7, IP-10, IL-8, and GM-CSF Expression levels are represented as the fold increase vs sham-infected, was determined by qPCR. The expression of each target gene was nor- nontargeting siRNA-transfected cells. The y-axis has been broken to show malized to GAPDH. Expression levels are represented as the fold increase the effects of siRNA on both basal and maximal gene expression. Bars, vs cells treated with UV-irradiated virus and nontargeting siRNA. The means Ϯ SEM for three experiments; numbers over bars indicate the fold y-axis has been broken to show the effects of siRNA on both basal and increase (ϫ) compared with sham-infected sample within its own siRNA maximal gene expression. Bars, means Ϯ SEM for three experiments; -p Ͻ 0.05 vs MDA5 siRNA-transfected RV-infected sample, numbers over bars indicate the fold increase (ϫ) compared with sham ,ء .group p Ͻ 0.05 vs MDA5 ,ء .ANOVA. siNT, Non-targeting siRNA; siMDA-5, siRNA against MDA-5. infected sample within its own siRNA group siRNA-transfected RV-infected sample, ANOVA. siNT, Non-targeting siRNA; siMDA-5, siRNA against MDA-5. cell type and virus specific. RIG-I-deficient mice fail to produce type I IFNs in response to the negative-sense ssRNA viruses New- castle disease virus, Sendai virus, vesicular stomatitis virus, and RIG-I, and MDA5 were verified by ELISA and immunoblotting, influenza virus and to the positive-sense ssRNA Japanese enceph- implying that RV significantly increases IFN responses in airway alitis virus, whereas MDA5-deficient mice fail to detect EMCV, a epithelial cells. Further, we demonstrate for the first time that positive-sense ssRNA picornavirus (9). Whereas MDA5 is essen- MDA5 and TRIF, but not RIG-I, are required for maximal sensing tial for induction of type I IFNs after infection with EMCV in RV dsRNA in human airway epithelial cells. Transfection of both fibroblasts and conventional DCs, plasmacytoid DCs use the TLR a human bronchial epithelial cell line (BEAS-2B cells) and pri- system for viral detection (9). Although these results are compel- mary tracheobronchial epithelial cells with siRNA against MDA5, ling, it seems premature to conclude that all are but not nontargeting siRNA, significantly inhibited RV1B-induced sensed by MDA5 in all cell types, because only cardioviruses have expression of a type I IFN, IFN-␤, as well as a number of ISGs. been studied. Knockdown of MDA5 also attenuated expression of the type III In this article, we found that infection of BEAS-2B and primary IFNs IFN-␭1 and IFN-␭2/3, which functionally resemble type I tracheal epithelial cells with RV induced substantial increases in IFNs (17) and are also induced by RV infection (18). These data IFN and ISG mRNA expression. In limited studies, we also found are in agreement with previous data suggesting that MDA5 is re- similar changes in gene expression after infection with major and quired for sensing picornavirus dsRNA (9). Alternatively, our data minor group virus. Due to the low level of baseline IFN expres- contrast with previous data from A549 alveolar type II epithelial sion, the fold increases in IFN expression were quite high, perhaps cells showing that RIG-I and TLR3, but not MDA5, are required artificially so. However, increases in IFN-␭1, IP-10/CXCL-10, for sensing human respiratory syncytial virus, a paramyxovirus The Journal of Immunology 6995

FIGURE 10. RV-induced IRF3 activation in cultured BEAS-2B and primary airway epithelial cells. A, BEAS-2B cells were infected with RV1B or UV-irradiated RV1B (1 h at 33°C) at a multiplicity of infection of 10. Cell lysates were collected and probed with anti-IRF3 Ab. Poly(IC) served as a positive control. IRF3 protein is visualized as two bands, an upper phosphorylated form and a lower unphosphorylated form. B, Den- sitometry of IRF3 phosphorylation is provided. Bars, means Ϯ SEM for three experiments. C, Cellular proteins were also subjected to native-PAGE to resolve the dimerization of IRF3. Poly(IC) served as a positive control. IRF protein is visualized as two bands, an upper dimer and a lower mono- mer. Blot is representative of five individual experiments. D, Primary tra- FIGURE 9. siRNA against TRIF blocks RV-induced IFN and ISG ex- cheobronchial epithelial cells were infected with RV1B or UV-irradiated pression. A, TRIF siRNA or nontargeting siRNA was transfected into RV1B. Top, Cell lysates were collected 12 h after infection and probed BEAS-2B cells. After transfection, cells were infected with RV1B, UV- with anti-IRF3 Ab. Poly(IC) served as a positive control. Bottom, Cellular irradiated RV1B, or sham infected. After infection, cell lysates were proteins were subjected to native-PAGE to resolve the dimerization of probed with anti-TRIF Ab. B–H, Total RNA was extracted and the expres- IRF3. Poly(IC) served as a positive control. Blots are representative of sion of IFN-␤, IFN-␭1, IFN-␭2/3, IRF7, IP-10, IL-8, and GM-CSF was three experiments. determined by qPCR. The expression of each target gene was normalized to GAPDH. Expression levels are represented as the fold increase vs sham- infected, nontargeting siRNA-transfected cells. The y-axis has been broken to show the effects of siRNA on both basal and maximal gene expression. pression. Indeed, expression of IL-8 and GM-CSF were often par- The blot shown is a typical representative of 3 separate experiments. Bars, adoxically increased. Previous studies have shown that RV- means Ϯ SEM for four experiments; numbers over bars indicate the fold induced IL-8 expression is strictly regulated by the transcription increase (ϫ) compared with sham-infected sample within its own siRNA factor NF-␬B (14, 19). The initial phase of IL-8 expression is also p Ͻ 0.05 vs TRIF siRNA-transfected RV-infected sample, replication independent (14, 20–22). Inhibition of dsRNA sensing ,ء .group ANOVA. siNT, Non-targeting siRNA; siTRIF, siRNA against TRIF. would therefore not be expected to reduce IL-8 expression. Fur- ther, the up-regulation of proinflammatory cytokine expression fol- lowing MDA5 and IRF3 knockdown may represent a compensa- (8). Thus, in the airway epithelium, recognition of viral dsRNA is tory mechanism by which the airway epithelial cells increase indeed virus type specific. immunosurveillance when the IFN response is suppressed. We also examined the contribution of IRF3 to RV-induced re- We previously showed that TLR3 was partially required for sponses in airway epithelial cells. IRF3 siRNA nearly abolished RV39-induced IL-8 expression in 16HBE14oϪ cells (12). In this IFN and ISG expression. MDA5 and TRIF knockdown also de- study, we could not verify TLR3 knockdown by immunoblotting creased IRF3 dimerization. These data are consistent with the idea or flow cytometry in BEAS-2B cells (not shown), leading us to use that TLR3 and MDA5 regulate IFN expression via a common siRNA against TRIF. In contrast to MDA5 knockdown, inhibition downstream intermediate, IRF3. of TLR3 signaling using TRIF siRNA inhibited both IFN and IL-8 In contrast to siRNA against MDA5, siRNA against RIG-I had expression. Based on our previous result, the complete effect of no effect on RV-induced IRF3 dimerization or IFN expression. TRIF siRNA on IFN signaling was unexpected. It is conceivable The divergent roles of RIG-I and MDA5 in the context of RV that TRIF is coupled to other, as yet unknown pattern recognition infection suggest that the two homologous helicases function dis- receptors and that the reduction in IFN expression induced by tinctly from each other. Further, although RIG-I expression was TRIF knockdown is not solely due to TLR3-linked signaling. In induced after RV infection, RIG-I apparently cannot compensate any event, the differential effects of MDA5 and TRIF siRNA on for reduced expression and/or function of MDA5. IL-8 expression suggest that, although NF-␬B and IRF3 are both Using siRNA against MDA5 and IRF3, we found that MDA5/ components of the same transcriptional enhanceosome in the reg- IRF3 signaling is required for RV-induced IFN, but not IL-8 ex- ulation of IFNs, ISGs, and inflammatory cytokine IL-8 (23, 24), 6996 PRRs AND RV INFECTION

FIGURE 12. siRNA against TRIF and MDA5 but not RIG-I reduced RV1B-induced IRF3 dimerization. A, RIG-I, MDA5, TRIF siRNA or non- targeting siRNA was transfected into BEAS-2B. After transfection, cells were infected with RV1B, UV-irradiated RV1B, or sham infected. After infection, cell lysates were probed with anti-IRF3 Ab. Poly(IC) served as a positive control. B, IRF3 dimer to monomer ratio was quantified by densitometry. Blot is representative of two experiments. siNT, Non-target- ing siRNA; siTRIF, siRNA against TRIF; siMDA-5, siRNA against MDA-5; siRIG-I, siRNA against RIG-I.

We confined most of our studies of RV-induced IFN responses to the minor group serotype RV1B. However, IFN expression by RV39, a major group virus, was also blocked by MDA5 but not RIG-I siRNA. This was not unexpected, given that RV1B and RV16, another major group serotype, have been shown to induce nearly identical patterns of gene expression in primary cultured airway epithelial cells (28). We have found that infection with RV1B and RV39 induce similar levels of Akt phosphorylation and IL-8 expression in cultured 16HBE14oϪ human bronchial epithe- lial cells, and that inhibition of PI 3-kinase blocks both RV1B- and RV39-induced IL-8 expression induced by either virus (29). Thus, FIGURE 11. siRNA against IRF3 blocks RV-induced IFN and ISG ex- there are ample data suggesting that major and minor subgroup pression. A, IRF3 siRNA or nontargeting siRNA was transfected into RVs stimulate similar airway epithelial cell signaling pathways BEAS-2B while seeding. After transfection, cells were infected with and elicit similar immune responses. RV1B, UV-irradiated RV1B (UV-RV1B) or sham. A, After infection, cell In conclusion, we have shown for the first time that MDA5, lysates were probed with anti-IRF3 Ab. Blot is representative of three TLR3, and IRF3 are each required for maximal RV-induced IFN separate experiments. B–H, Total RNA was extracted and the expression of responses. Viral infections, most commonly caused by RV, are the ␤ ␭ ␭ IFN- , IFN- 1, IFN- 2/3, IRF7, IP-10, IL-8, and GM-CSF was deter- most frequent cause of asthma exacerbations and account for a mined by qPCR. The expression of each target gene was normalized to substantial percentage of chronic obstructive pulmonary disease GAPDH. Expression levels are represented as the fold increase vs sham- exacerbations (1). Bronchial epithelial cells isolated from patients infected, nontargeting siRNA-transfected cells. The y-axis has been broken to show the effects of siRNA on both basal and maximal gene expression. with asthma have been demonstrated to have an incomplete innate Bars, means Ϯ SEM for four experiments; numbers over bars indicate the immune response to rhinovirus infection, with deficient type I fold increase (ϫ) compared with the sham-infected sample within its own IFN-␤ and type III IFN-␭ production (18, 30). Further understand- p Ͻ 0.05 vs nontargeting siRNA-transfected RV1B-in- ing of the biochemical signaling pathways regulating RV-induced ,ء .siRNA group fected cells, one-way ANOVA. siNT, Non-targeting siRNA; siRF3, siRNA IFN expression may therefore provide insight into the pathogen- against IRF-3. esis of human airway diseases and new therapeutic targets for treatment. their requirement for gene expression may vary for different target genes, perhaps depending on the organization of IFN-stimulated Acknowledgments and NF-␬B response elements in the promoter. We thank Drs. Nathalie Grandvaux and John Hiscott (McGill University, It has recently been reported that picornaviruses may develop Montreal Canada) for their advice in the measurement of IRF3 dimerization. strategies to escape host immune surveillance. Hepatitis A virus has been shown to suppress RIG-I-mediated signaling in fetal rhe- Disclosures sus monkey kidney (FRhK-4) cells (25). infection in- The authors have no financial conflict of interest. duces cleavage of MDA5 in HeLa cells (26). In A549 human al- veolar type II alveolar epithelial cells, RV14 infection fails to References induce high levels of IFN, apparently by interfering with IRF3 1. Hershenson, M. B., and S. L. Johnston. 2006. Rhinovirus infections: More than dimerization (27). In this study, RV1B induced robust IRF3 dimer- a common cold. Am. J. Respir. Crit. Care Med. 174: 1284–1285. ization and IFN responses in both BEAS-2B and primary bronchial 2. Yamamoto, M., S. Sato, H. Hemmi, K. Hoshino, T. Kaisho, H. Sanjo, epithelial cells. Also, although we did indeed observe apparent O. Takeuchi, M. Sugiyama, M. Okabe, K. Takeda, and S. Akira. 2003. Role of adaptor TRIF in the MyD88-independent Toll-like receptor signaling pathway. degradation of MDA5 following viral infection (Fig. 5), the total Science 301: 640–643. protein abundance of MDA5 was still substantially increased 24 h 3. Yoneyama, M., M. Kikuchi, T. Natsukawa, N. Shinobu, T. Imaizumi, M. Miyagishi, K. Taira, S. Akira, and T. Fujita. 2004. The RNA helicase RIG-I after RV inoculation. Taken together, these data suggest an intact has an essential function in double-stranded RNA-induced innate antiviral re- host innate immune defense against RV1B dsRNA. sponses. Nat. Immunol. 5: 730–737. The Journal of Immunology 6997

4. Andrejeva, J., K. S. Childs, D. F. Young, T. S. Carlos, N. Stock, S. Goodbourn, IFN-␭s mediate antiviral protection through a distinct class II cytokine receptor and R. E. Randall. 2004. The V proteins of paramyxoviruses bind the IFN-in- complex. Nat. Immunol. 4: 69–77. ducible RNA helicase, MDA-5, and inhibit its activation of the IFN-␤ promoter. 18. Contoli, M., S. D. Message, V. Laza-Stanca, M. R. Edwards, P. A. Wark, Proc. Natl. Acad. Sci. USA 101: 17264–17269. N. W. Bartlett, T. Kebadze, P. Mallia, L. A. Stanciu, H. L. Parker, et al. 2006. 5. Kawai, T., K. Takahashi, S. Sato, C. Coban, H. Kumar, H. Kato, K. J. Ishii, Role of deficient type III -␭ production in asthma exacerbations. Nat. O. Takeuchi, and S. Akira. 2005. IPS-1, an adaptor triggering RIG-I- and MDA5- Med. 12: 1023–1026. mediated type I interferon induction. Nat. Immunol. 6: 981–988. 19. Zhu, Z., W. Tang, J. M. Gwaltney, Y. Wu, and J. A. Elias. Rhinovirus stimulation 6. Xu, L.-G., Y.-Y. Wang, K.-J. Han, L.-Y. Li, Z. Zhai, and H.-B. Shu. 2005. VISA of interleukin-8 in vivo and in vitro: role of NF-kappaB. Am. J. Physiol. Lung is an adapter protein required for virus-triggered IFN-␤ signaling. Mol. Cell 19: Cell Mol. Physiol. 273: L814–L824, 1997. 727–740. 20. Griego, S. D., C. B. Weston, J. L. Adams, R. Tal-Singer, and S. B. Dillon. 2000. 7. Fitzgerald, K. A., S. M. McWhirter, K. L. Faia, D. C. Rowe, E. Latz, Role of p38 mitogen-activated protein kinase in rhinovirus-induced cytokine pro- D. T. Golenbock, A. J. Coyle, S. M. Liao, and T. Maniatis. 2003. IKK⑀ and TBK1 duction by bronchial epithelial cells. J. Immunol. 165: 5211–5220. are essential components of the IRF3 signaling pathway. Nat. Immunol. 4: 21. Suzuki, T., M. Yamaya, K. Sekizawa, M. Hosoda, N. Yamada, S. Ishizuka, 491–496. K. Nakayama, M. Yanai, Y. Numazaki, and H. Sasaki. Bafilomycin A(1) inhibits 8. Liu, P., M. Jamaluddin, K. Li, R. P. Garofalo, A. Casola, and A. R. Brasier. 2007. rhinovirus infection in human airway epithelium: effects on endosome and Retinoic acid inducible gene-I mediates early anti-viral response and Toll like ICAM-1. Am. J. Physiol. Lung Cell Mol. Physiol. 280: L1115–L1127, 2001. 22. Kaul, P., M. C. Biagioli, I. Singh, and R. B. Turner. 2000. Rhinovirus-induced receptor 3 expression in respiratory syncytial virus-infected airway epithelial phox cells. J. Virol. 81: 1401–1411. oxidative stress and interleukin-8 elaboration involves p47 but is independent 9. Kato, H., O. Takeuchi, S. Sato, M. Yoneyama, M. Yamamoto, K. Matsui, of attachment to intercellular adhesion molecule-1 and viral replication. J. Infect. S. Uematsu, A. Jung, T. Kawai, K. J. Ishii, et al. 2006. Differential roles of MDA5 Dis. 181: 1885–1890. and RIG-I helicases in the recognition of RNA viruses. Nature 441: 101–105. 23. Ohmori, Y., and T. A. Hamilton. 1993. Cooperative interaction between inter- feron (IFN) stimulus response element and ␬B sequence motifs controls IFN␥- 10. Muir, A., G. Soong, S. Sokol, B. Reddy, M. I. Gomez, A. van Heeckeren, and and lipopolysaccharide-stimulated transcription from the murine IP-10 promoter. A. Prince. 2004. Toll-like receptors in normal and cystic fibrosis airway epithelial J. Biol. Chem. 268: 6677–6688. cells. Am. J. Respir. Cell Mol. Biol. 30: 777–783. 24. Casola, A., R. P. Garofalo, M. Jamaluddin, S. Vlahopoulos, and A. R. Brasier. 11. Sha, Q., A. Q. Truong-Tran, J. R. Plitt, L. A. Beck, and R. P. Schleimer. 2004. 2000. Requirement of a novel upstream response element in respiratory syncytial Activation of airway epithelial cells by Toll-like receptor agonists. Am. J. Respir. virus-induced IL-8 gene expression. J. Immunol. 164: 5944–5951. Cell Mol. Biol. 31: 358–364. 25. Fensterl, V., D. Grotheer, I. Berk, S. Schlemminger, A. Vallbracht, and 12. Sajjan, U. S., Y. Jia, D. C. Newcomb, J. K. Bentley, N. W. Lukacs, J. J. LiPuma, A. Dotzauer. 2005. Hepatitis A virus suppresses RIG-I-mediated IRF-3 activation and M. B. Hershenson. 2006. H. influenzae potentiates airway epithelial cell to block induction of ␤ interferon. J. Virol. 79: 10968–10977. responses to rhinovirus by increasing ICAM-1 and TLR3 expression. FASEB J. 26. Barral, P. M., J. M. Morrison, J. Drahos, P. Gupta, D. Sarkar, P. B. Fisher, and 20: 2121–2123. V. R. Racaniello. 2007. MDA-5 is cleaved in poliovirus-infected cells. J. Virol. 13. Zhu, L., P.-k. Lee, W.-M. Lee, Y. Zhao, D. Yu, and Y. Chen. 2009. Rhinovirus 81: 3677–3684. induced major airway mucin production involves a novel TLR3-EGFR dependent 27. Kotla, S., T. Peng, R. E. Bumgarner, and K. E. Gustin. 2008. Attenuation of the pathway. Am. J. Respir. Cell Mol. Biol. 40: 610–619. type I interferon response in cells infected with human rhinovirus. Virology 374: 14. Newcomb, D. C., U. Sajjan, S. Nanua, Y. Jia, A. M. Goldsmith, J. K. Bentley, and 399–410. M. B. Hershenson. 2005. Phosphatidylinositol 3-kinase is required for rhinovirus- 28. Chen, Y., E. Hamati, P.-K. Lee, W.-M. Lee, S. Wachi, D. Schnurr, S. Yagi, induced airway epithelial cell interleukin-8 expression. J. Biol. Chem. 280: G. Dolganov, H. Boushey, P. Avila, and R. Wu. 2006. Rhinovirus induces airway 36952–36961. epithelial gene expression through double-stranded RNA and IFN-dependent 15. Papi, A., and S. L. Johnston. 1999. Rhinovirus infection induces expression of its pathways. Am. J. Respir. Cell Mol. Biol. 34: 192–203. own receptor intercellular adhesion molecule 1 via increased NF-␬B-mediated 29. Newcomb, D. C., U. S. Sajjan, D. R. Nagarkar, Q. Wang, S. Nanua, Y. Zhou, transcription. J. Biol. Chem. 274: 9707–9720. C. L. McHenry, K. T. Hennrick, W. C. Tsai, J. K. Bentley, et al. 2008. Human 16. Iwamura, T., M. Yoneyama, K. Yamaguchi, W. Suhara, W. Mori, K. Shiota, rhinovirus 1B exposure induces phosphatidylinositol 3-kinase-dependent airway Y. Okabe, H. Namiki, and T. Fujita. 2001. Induction of IRF-3/-7 kinase and inflammation in mice. Am. J. Respir. Crit. Care Med. 177: 1111–1121. NF-␬B in response to double-stranded RNA and virus infection: common and 30. Wark, P. A. B., S. L. Johnston, F. Bucchieri, R. Powell, S. Puddicombe, unique pathways. Genes Cells 6: 375–388. V. Laza-Stanca, S. T. Holgate, and D. E. Davies. 2005. Asthmatic bronchial 17. Kotenko, S. V., G. Gallagher, V. V. Baurin, A. Lewis-Antes, M. Shen, epithelial cells have a deficient innate immune response to infection with rhino- N. K. Shah, J. A. Langer, F. Sheikh, H. Dickensheets, and R. P. Donnelly. 2003. virus. J. Exp. Med. 201: 937–947.