Microarray-based analyses are increasingly being used in virology and have helped to elucidate virus-host interactions for a number of viruses (reviewed in reference 30
). Here, we performed a comprehensive analysis of the responses of a human respiratory epithelial cell line to HPIV1 infection and to IFN-β treatment. Microarray analyses have previously been applied to study infection by murine PIV1 (SeV) (12
). However, these studies had methodological issues that limited the ability to interpret the results, including that (i) human cells were infected with a murine virus, (ii) nonrespiratory cells were infected with a respiratory virus, (iii) mRNA levels were determined only at a single time point, (iv) customized arrays that included fewer than 1,000 genes and that focused on the IFN and antiviral pathways were used, and (v) the response to viral infection was not compared with the response to IFN treatment (12
). In the present study, gene expression kinetics following infection with a human respiratory virus or administration of IFN-β were determined across multiple time points in human respiratory epithelial cells, the natural target of HPIV1, by use of a microarray representing the entire human genome. We chose to analyze only genes whose expression was at least fourfold up- or downregulated, thereby excluding genes that were more modestly induced or suppressed by viral infection or IFN treatment. In order to avoid a subjective or biased analysis, functional bioinformatics tools were used to group thousands of differentially expressed genes into distinct hierarchical clusters and to identify functional pathways overrepresented in each cluster. oPOSSUM single-site analysis software was used to examine upstream sequences for predicted TFBSs within each hierarchical cluster (25
), resulting in the identification of several key transcriptional regulatory pathways that are likely involved in the cellular response to HPIV1 infection.
As a first step in our analysis, we sought to identify genes that were differentially expressed as a result of wt HPIV1 infection. This set of genes was subsequently compared to the set of genes whose expression was altered by treatment with a physiological concentration of IFN-β, allowing us to segregate and analyze IFN-responsive genes separately from IFN-independent genes in the wt HPIV1 microarray data set. IFN-β treatment was chosen for comparison because A549 cells and human airway epithelial cells produce this IFN in response to HPIV1, whereas IFN-α is not secreted at all or is barely detectable. In agreement with previous reports that established the central role of the type I IFN response in innate antiviral immunity (9
), our clustering analysis indicated that approximately 60% of all genes induced late in infection with wt HPIV1 were also induced by IFN-β treatment. Specifically, 209 of the 343 genes were upregulated by both wt HPIV1 infection and IFN-β treatment (Fig. , clusters A, B, and D). These 209 genes were overrepresented in the IFN signaling, antigen presentation, and ubiquitination pathways, as well as in the Toll-like receptor and NF-κB pathways. However, wt HPIV1 upregulated an additional 129 genes that were not upregulated by IFN-β (Fig. , column I, cluster E). Many of these genes were also under the control of the family of IRF and NF-κB transcription factors and were involved in apoptosis signaling and inflammation. These data are in agreement with and expand on a previous report on the dominant role of IRF3- and NF-κB-regulated genes in the response of human embryonic kidney cells to SeV infection at 6 h (12
). It can be concluded by inspection of the kinetic data that wt HPIV1 is able to silence the innate antiviral response for at least 6 to 12 h p.i. but that IFN-dependent and IFN-independent antiviral pathways are activated at later time points as part of the innate response to infection.
Next, we sought to determine the role of the HPIV1 C proteins in suppressing the antiviral response of respiratory epithelial cells by analyzing gene expression patterns following infection of A549 cells with two C mutant viruses. Previous studies had established a role for the HPIV1 C proteins and SeV C proteins as inhibitors of IFN induction and signaling, as well as inhibitors of apoptosis (19
). Two mutant viruses, one with a single point mutation in C and one with a deletion of all four C proteins, CF170S
and P(C−), respectively, were selected for this analysis. wt HPIV1 inhibits the induction of IFN and signaling of IFN through its receptor, but CF170S
and P(C−) each fail to inhibit these two activities. In addition, P(C−) infection induces apoptosis, whereas CF170S
and wt HPIV1 infections do not, and P(C−) is much more restricted in replication in AGMs than CF170S
). Using the CF170S
and P(C−) viruses, we sought to compare the host responses following infection with these mutants to that following wt HPIV1 infection. We found that, following infection with either C mutant virus, the number of differentially expressed genes expanded dramatically, implicating a critical role for the HPIV1 C proteins in blunting the host's antiviral response. Using whole-genome microarrays, we found that with CF170S
infection, 392 and 2,322 genes were differentially regulated more than fourfold at 24 and 48 h p.i., respectively, while with wt HPIV1 infection, only 54 and 277 genes were differentially expressed at the same time points. Thus, approximately 1% of the over 41,000 unique probes analyzed in our study were altered by HPIV1-CF170S
at 24 h p.i., and only 14% of that subset of genes were altered by wt HPIV1, indicating that the wt HPIV1 C proteins inhibited modulation of the majority of antiviral genes that would otherwise be upregulated or downregulated as a result of HPIV1 infection. Importantly, this striking suppression of gene expression was associated with mutation in the C proteins but not with differences in virus replication. CF170S
and wt HPIV1 exhibited identical kinetics of replication in vitro, and although P(C−) was approximately 100-fold restricted in replication at 24 and 48 h p.i. compared to wt HPIV1 and CF170S
(see Fig. S5 in the supplemental material), P(C−) infection was capable of suppressing gene expression just as well as CF170S
infection. Strähle et al. examined gene expression in human fibrosarcoma cells 24 h p.i. in response to SeV-wt or SeV-CF170S
infection by using a customized array designed to probe approximately 150 genes with overrepresentation of the IFN pathway (62
). In their study, infection with SeV-CF170S
at an MOI of 20 induced 15 of 150 genes (10%) more than twofold at 24 h p.i., while SeV-wt (Ohita M strain) failed to induce the expression of any gene. We also found that the expression of IFN-β-inducible genes was greater following infection with CF170S
or P(C−) than following treatment with 300 pg/ml of IFN-β, the concentration of IFN induced by CF170S
infection of A549 cells (66
). The greater induction seen with C mutant infection is due in part to direct viral recognition by RIG-I-like receptors and subsequent activation of IRFs and NF-κB. Activated IRF3 not only stimulates the IFNB promoter to produce IFN-β but also can independently activate ISRE to augment the production of IFN-stimulated genes (ISGs). In contrast, the effect of exogenous IFN is limited to signaling through the IFN receptor and activation of the ISGF3 transcription factor complex to upregulate the expression of antiviral effectors and other ISGs. The wt C proteins likely block the RIG-I-like receptor and IFN signaling pathways, as evidenced by the lack of IRF3 and STAT activation and IFN production during infection with wt HPIV1, and this antagonism is ablated as a result of mutation or deletion of the HPIV1 C proteins (6
). In another similarly designed microarray study, Hartman et al. compared the host response to wt Ebola virus with the response to a highly attenuated Ebola virus containing a single amino acid change, namely, R312A, in the IRF3-inhibitory domain of VP35 (22
). Like wt HPIV1, wt Ebola virus was remarkably effective in suppressing the activation of cellular antiviral and IFN-responsive genes. The single VP35R312A
mutation in Ebola virus, however, reversed the inhibition of only 39 genes, whereas the single CF170S
mutation in HPIV1 reversed the inhibition of over 2,000 genes (Fig. ). The Ebola virus VP35 protein confers virulence during infection by inhibiting IRF3 activation and IFN-β production, whereas the VP24 protein inhibits STAT1 nuclear translocation and IFN signaling (21
). Unlike Ebola virus, which utilizes two separate viral proteins to antagonize IFN production and signaling, HPIV1 encodes the C proteins from a single gene segment that blocks both arms of the viral recognition and IFN pathways and probably suppresses the expression of many more genes than VP35 alone.
Following the comparison of levels of gene expression between wt HPIV1 and the two C mutant viruses, we next compared the cellular responses to CF170S and P(C−) infections to determine whether differences in gene expression profiles were associated with the observed differences in the mutants' apoptosis and attenuation phenotypes. Surprisingly, the cellular response to CF170S infection was comparable to the response to P(C−) infection, both qualitatively and quantitatively (Fig. ). Although the microarray-based analysis may have missed differences between CF170S and P(C−) due to a decreased sensitivity compared to that of RT-qPCR, this limitation occurred only with very highly upregulated genes, where gene induction is already qualitatively clear. For example, IFNB, which was induced 853,062-fold as measured by RT-qPCR and 628-fold by microarray analysis, showed 1,358-fold greater sensitivity by qPCR due to saturation of the IFNB probe on the microarray. However, less strongly upregulated genes, such as NFKB1, which was induced 68-fold as measured by RT-qPCR and 11-fold by microarray analysis, showed only 6-fold greater sensitivity in detecting differential expression. The difference is significantly less pronounced since genes that are upregulated less do not suffer as much from microarray probe saturation. Thus, although the microarray-based analysis may have missed small differences in gene expression between viruses, we have shown that both technologies have comparable sensitivities at lower expression levels. Based on the phenotypic differences between the two mutants, with P(C−) being the more attenuated in vivo and a strong inducer of apoptosis in vitro, we expected to find transcriptional differences in at least the apoptosis pathway. However, not a single mRNA was differentially expressed by use of our predefined criteria, i.e., a fourfold change and a P value of <0.01. To detect any subtle differences between infections with the two C mutant viruses, we tested severalfold-change cutoffs. After relaxing our change cutoff to twofold and subsequently performing a t test with multiple testing correction, we identified two genes, caspase 3 and the TRAIL receptor 2 (TNFRSF10B), that could potentially contribute to P(C−)-induced apoptosis. P(C−) infection upregulated both genes earlier, at 24 h p.i., than CF170S infection. However, in both cases, the magnitude of upregulation during CF170S infection caught up by 48 h p.i., and the TRAIL ligand (TNFSF10) itself was induced at comparable levels by both viruses at 24 h p.i.
It was surprising that the CF170S
mutant, which contains a single point mutation at amino acid 170 of the C protein, induced a transcriptional profile that was almost indistinguishable from that induced by the P(C−) mutant, a virus that does not express any of the C genes. Although P(C−) infection induces apoptosis much more rapidly and extensively than CF170S
), the expression kinetics of death ligands, death receptors, or antiapoptotic factors that could account for P(C−)-induced cell death did not differ significantly by fourfold from those of CF170S
at any time (Fig. ). Since knockout of the C gene causes cell death whereas expression of the wt C gene does not, we hypothesized that the wt HPIV1 and CF170S
C proteins prevented cell death through apoptosis inhibition, i.e., either through inhibition of proapoptotic pathways or through activation of antiapoptotic pathways. Previous studies of SeV suggested that a delicate balance between proapoptotic and antiapoptotic pathways during viral infection exists. SeV infection activates the cellular phosphatidylinositol 3-kinase pathway and, through AKT activation, prevents apoptosis (49
). Thus, the differences in the apoptosis phenotypes of the two C mutants likely lie in the interaction of one or more of the C proteins of CF170S
and wt HPIV1 with one or more constitutively expressed factors, which results in the suppression of the apoptosis response. Both wt C proteins and point-mutated C proteins could potentially activate apoptosis inhibitory proteins or antagonize proapoptotic proteins that would otherwise be triggered during HPIV1 infection. These two possibilities can be tested once the mechanism of interaction with the host apoptosis machinery has been defined. The greater level of attenuation of P(C−) than of CF170S
likely reflects a contribution of apoptosis to the decreased replication of P(C−), but the loss of a yet-undefined function of C required for replication in vivo may contribute as well.
Although wt HPIV1 suppressed the upregulation of a broad array of cellular genes that were induced by P(C−) and CF170S
infection, global shutoff of host transcription did not occur. In fact, only 234 of 2,612 differentially expressed genes were downregulated (Fig. , panel II, cluster H). Interestingly, both C mutant viruses downregulated these genes to a greater extent than wt HPIV1 (see Table S2 in the supplemental material). Thus, the wt C proteins blunted both induction and suppression of a specific set of host genes that would otherwise react to viral infection. This is in stark contrast to what has been described for vesicular stomatitis virus, which shuts down transcription globally through M protein-mediated inhibition of the three cellular RNA polymerases I, II, and III (1
), or for Rift Valley fever virus, which, through inhibition of TFIIH transcription factor complex assembly via its NS proteins, leads to the rapid destabilization of host cellular RNA synthesis (40
). However, other members of the order Mononegavirales
, such as the Marburg and Ebola filoviruses, suppress the antiviral and type I IFN responses without shutting down host cell functions globally, and the degree of antiviral suppression by these viruses seems to correlate with virulence in vivo (34
). Similarly, our data indicate that wt HPIV1 C protein function, i.e., suppression of antiviral pathways early in infection, is an important virulence factor and that the loss of this activity in both CF170S
and P(C−) leads to attenuation in vivo.
Following the global analysis of virus-modulated pathways, we attempted a more detailed analysis of the IFN and apoptotic pathways in order to understand the different in vivo and in vitro phenotypes of wt HPIV1, P(C−), and CF170S
. We found that by 48 h p.i., wt HPIV1 induced IFNB expression 180-fold as measured by microarray and 1,200-fold as measured by qPCR (Fig. ). Previously, we reported that IFNB mRNA was not detectable at any time point in response to HPIV1 infection (66
). Reexamination of that data identified a 5,600-fold induction of IFNB mRNA following CF170S
infection versus a 12-fold induction following wt HPIV1 infection at 48 h p.i., i.e., a 458-fold reduction in IFNB transcription following wt HPIV1 infection compared to that following CF170S
). In the present study, CF170S
induced 695 times more IFNB mRNA than wt HPIV1 at 48 h p.i., confirming that IFNB mRNA is strongly induced by CF170S
but not by wt HPIV1 (see Table S3 in the supplemental material). At the protein level, we previously were unable to detect IFN-β in the supernatant of wt HPIV1-infected A549 cells by IFN bioassay, not even after 96 h p.i. (66
). In contrast, Bousse et al. detected 380 IU/ml of IFN-β by enzyme-linked immunosorbent assay in MRC-5 cells 48 h p.i. with wt HPIV1 strain C-35 (GenBank accession no. M74081) (6
). This difference in IFN-β secretion could be due to the use of different cell lines (MRC-5 versus A549) and/or different virus strains. A protein sequence alignment of the C protein of the C-35 strain of HPIV1 used by Bousse et al. with our Washington 1964 wt HPIV1 strain revealed a CQ38P
substitution in the C gene that could potentially be responsible for a distinct IFN phenotype. The CF170S
virus serves as an example for the role a single amino acid substitution in the C protein can potentially have in the IFN phenotype. Importantly, the deficient IFN-β response of A549 cells to wt HPIV1 infection was also seen in primary human airway epithelial cell cultures, confirming the IFN production deficiency of respiratory epithelial cells infected with wt HPIV1 (5
One aim of this project was to identify cellular pathways that are targeted by the HPIV1 C proteins and to define how the C proteins exert such impressive transcriptional control of the host's innate immune system. We analyzed the gene expression profiles for all 14 transcription factors identified by bioinformatics analysis (see Fig. S6 in the supplemental material) and found that the expression levels of four genes, IRF1, ISGF3G, NF-κB1, and c-Rel, were significantly upregulated increasingly over time, confirming the importance of the IRF3 and NF-κB signaling pathways during parainfluenza infection and furthermore providing a feed-forward mechanism to amplify the host transcriptional response to viral infection. We expected to identify IRF and NF-κB TFBSs overrepresented in genes induced by IFN treatment and HPIV1 infection, but other unexpected TFBSs were also identified. For example, FOXD1 and FOXD3 TFBSs were identified in genes upregulated by the HPIV1 C mutant viruses. Only recently has an important role for FoxD1 in the regulation of immune activation through the NF-κB pathway been identified (42
). FoxD3 is well known for controlling cellular differentiation, but no role for FoxD3 in the regulation of the immune response or the response to viral infection has yet been defined. In future work, it will be important to define how the HPIV1 C proteins modify the activities of FOXD1, FOXD3, and other transcription factors and how C′, C, Y1, and Y2 differ in their abilities to alter transcriptional activity.
In summary, the HPIV1 C proteins exert remarkable control over the cellular transcriptional response to viral infection, indicating that the C proteins are important virulence factors for HPIV1. Mutations within the C gene permit the activation of a broad array of cellular genes involved in the type I IFN, IRF3, and NF-κB pathways that would otherwise be repressed by HPIV1 infection, and these mutations specify an attenuation phenotype in vivo. However, the lack of a clear differential regulation of antiapoptotic and proapoptotic genes in P(C−)- and CF170S-infected cells allows for two models of P(C−)-induced apoptosis that could individually, or in combination, play a key role in determining the apoptosis phenotype of HPIV1 infection: (i) earlier expression of TNFRSF10B and caspase 3 and (ii) alterations in posttranslational events involving constitutively expressed cellular proapoptotic or antiapoptotic factors.