Influenza A infection is associated with pathological changes throughout the respiratory tract, however the major site of impact appears to be the respiratory epithelia. Bronchoscopy of patients with uncomplicated influenza infections reveals alterations in the ciliated epithelia of the larynx, trachea, and bronchi that includes vacuolization, loss of cilia, and desquamation of columnar epithelial cells and goblet cells down to the basal cell layer. Importantly, viral antigen is found predominantly in the epithelial cells and mononuclear cells 
. Therefore, for the studies described here, we employed telomerase-immortalized human bronchial epithelial cells (HBEC30) that retain the capacity to differentiate into a polarized ciliated epithelial sheet 
. In undifferentiated cell culture, we found that 100% of HBEC30 in culture display viral protein production after a 24-hour exposure to mouse-adapted virus at an MOI of 5 (). This leads to an approximately 50% decrease in cell viability 48-hours post infection (). Given these observations, we adopted a whole genome siRNA screening strategy that involved a 48-hour incubation post siRNA transfection, followed by a 48 hour exposure to influenza A/WSN/1933 or carrier, with cell viability as the endpoint assay. Raw viability values were converted to viability Z-scores with a metric that normalized for both position and batch effects (, see methods). The dynamic range of viability scores observed under screening conditions potentially affords the opportunity to identify both enhancers and resistors of viral pathogenicity (Table 1 in Supporting Information S1). Considering siRNA pools associated with Z-scores that were equal to or greater than 3 standard deviations above (resistors) or below (sensitizers) the mean of the population, 53 candidate resistors and 182 candidate sensitizers were identified (Table 2 in Supporting Information S1). A representative sample of targets was further tested for consequences on viral protein accumulation and viral replication. As might be expected, the majority of the siRNA pools that deflect a viral cytopathic response resulted in reduced viral protein accumulation, as detected by quantitation of viral proteins at single cell resolution, and reduced production of infectious particles (). Among these, IVNS1ABP and the splicing factor SFPQ directly interact with the viral pathogenicity factor NS1, presumably reflecting a positive role in support of viral corruption of host machinery for viral protein production 
. Of interest in this class is RRAGD, a small G-protein that supports the amino-acid responsiveness of mTOR as a component of the “ragulator” 
. Several reports have highlighted the importance of viral induction of mTOR for viral replication, but the mechanism is not fully elaborated 
. Given the participation of endosomes as a viral entry mechanism 
, it is tempting to speculate that RRAGD is a limiting host factor for viral corruption of mTOR regulation. Additional factors in this group are involved with the host defense response, p53-mediated cell death and vesicle maturation and trafficking. To test for false positives arising from off-target effects of siRNA treatment, we retested 88 siRNA pools as four individual oligos. Approximately 60% of siRNAs retested with two or more oligos reproducing the original phenotype (Figure S1).
Identification of Host Modulators of Influenza Infection.
Characterization of Viral Replication Response.
Among the most potent members of the sensitizer class were the previously described proviral host factor IFITM3 and its homolog IFITM1 (Table 7 in Supporting Information S1). IFITM3 has been reported to be required for restriction of viral infection and is thought to inhibit viral entry 
. These gene products are interferon responsive, and depletion was associated with enhanced viral pathogenicity and enhanced viral protein production at limiting MOIs (1 and 0.1) as compared to controls (). Unexpectedly, cells depleted of IFITM3 produced fewer infection competent viral particles as determined by secondary infection of MDCK cells with cell culture supernatants ( A, E). For these assays, HBEC30 cell cultures were infected with an MOI of 5 for 48 hours post transfection with siRNA pools. Supernatants were collected 24 hours post infection and used to infect confluent MDCK cell cultures. Notably, we observed enhanced frequency as well as enhanced amplitude of viral protein accumulation in IFITM3 depleted cells during primary infection. Reduced production of infectious particles, in the face of enhance viral protein production, may therefore be a consequence of either limiting host factors or disruption of viral protein/host factor stoichiometry required for assembly of viable viral particles. Of interest, the viral cytophathic effect was greatly enhanced upon IFITM3 depletion in the presence or absence of the virulence factor NS1, a viral protein known to block many of the innate immunity responses 
(). However, deletion of NS1 results in complete failure of infectious particle production even upon IFITM3 depletion (). These observations would place IFITM3 function early in the viral life cycle and independent of NS1 function, consistent with reports that indicate IFITM3's antiviral activity is at the level of viral entry 
Depletion of the cell cycle/DNA damage checkpoint proteins CDC2 and CHEK1, like IFITM3, appeared to promote viral protein production and cytopathic response, while impairing assembly of infection-competent viral particles. A global comparison of the candidate modulators of H1N1 pathogenicity identified here with two whole-genome siRNA screens for modulators of cell cycle progression revealed a significant intersection (). However, CDC2 and CHEK1 depletion show quite distinct consequences on G1 versus G2 arrest suggesting their contribution to H1N1 infection may be independent of cell cycle control. CHEK1 has not been previously isolated in viral pathogenicity or viral replication screens, including those performed with the same siRNA library employed here (, Tables 3 and 4 in Supporting Information S1). To investigate additional biological processes that may be associated with CHEK1 modulation of viral infection, we assembled a context-specific protein-protein interaction sub-network defined by the genomic Z-score distribution of the primary screen (Figure S7). This subnetwork revealed the circadian gene Timeless, recently defined as a master regulator of the host defense response 
, within the first-degree neighborhood of CHEK1 (). Given this association, we investigated the consequence of chemical inhibition of CHEK1 on H1N1 infection.
Functional Classification of Candidate Hits.
We employed SB218078, an investigational CHEK1 inhibitor similar to one currently in clinical trials as an anti-neoplastic agent, with an in vitro IC50 of 0.015 μM and a Ki,app.
of 15±4 
. Pretreatment of cultures with 1 uM or 100 nM SB218078 for 12 hours resulted in significant inhibition of viral protein accumulation together with a marked virus-specific death response by 24 hours (). While some viral infection was detected at the 100 nM dose, viral protein production was severely limited at single cell resolution (). These observations suggest that SB218078 is releasing a cell death response to viral detection that would otherwise be suppressed during the viral replication cycle. Viral-induced cell death was also observed upon siRNA-mediated CHEK1 depletion (). The seemingly contradictory increase in infection frequency upon CHEK1 depletion may therefore be an indirect consequence of infection of low density residual cell populations with hypomorphic CHEK1 activity. Remarkably, SB218078 had no consequence on H1N1 replication in A549 cells, a cancer cell line often employed to test for modulators of viral replication and host responses 
(). However, a nontransformed, telomerase-immortalized bronchial epithelial cell line, derived from a different patient, HBEC3 
, was identical to HBEC30 in its responsiveness to SB218078 (). These observations indicate intervention targets may be available in non-tumorigenic cells that are uncoupled from host regulatory networks in cancer cells, and potentially explain why CHEK1 was not identified in other efforts to date that have universally relied on cancer lines as screen hosts 
We next queried the behavior of gene depletions identified here that modulate H1N1 cytopathic effects to those in 4 whole-genome siRNA screens which measured influenza virus replication as the end-point assay 
. This allowed us to parse collective ‘hits’ into four functional classes (, Table 5 in Supporting Information S1). Class 1: genes that, when depleted, enhance bronchial epithelial cell survival upon H1N1 exposure, and are required for viral replication. This class presumably represents host factors that facilitate viral infection and/or are required to support viral replication. Class 2: genes that, when depleted, reduce bronchial epithelial cell survival upon H1N1 exposure, and are required for viral replication. This, initially unanticipated but very intriguing class, likely represents host factors that deflect cell death checkpoint responses that would otherwise engage upon detection of viral infection. Class 3: genes that, when depleted, reduce bronchial epithelial cell survival upon H1N1 exposure and enhance viral replication relative to controls. This class presumably represents antiviral restriction factors that normally oppose infection. Class 4: genes, that when depleted, enhance bronchial epithelial cell survival upon H1N1 exposure and enhance viral replication as compared to controls. Of note, Class 2, which may represent novel intervention target opportunities, includes TRRAP, a large multidomain protein of the phosphoinositide 3-kinase-related kinases (PIKK) family that is a component of many histone acetyltransferase (HAT) complexes. TRRAP was recently identified as a bona fide oncogene in melanoma through cancer genome resequencing efforts, however, its transforming mechanism is unknown 
By nature, a challenge to siRNA-screening efforts is false negatives that derive from weak phenotypes due to suboptimal depletion of what are otherwise key factors in the biological process under investigation. One opportunity to help meet this challenge is to employ coherent behavior of gene sets to identify key biological processes supporting a phenotype rather than relying solely on an arbitrary scoring threshold for each individual gene. We employed Netwalk 
here to facilitate identification of such gene sets based on overrepresentation of functionally coherent subnetworks within the graph (Figures S1, S2, S3, S4, S5, S6, S7, S8 and S9). One such subnetwork implicated prenylation of Rab-family GTPases in support of H1N1 replication (Figure S4 and ). To test this we employed 3-IPEHPC, a specific inhibitor of the type II Geranylgeranyl-transferases (IC50 of 1.27 μM and a Ki
of 0.211 μM for Rab1a modification 
). As such, 3-IPEHPC specifically inhibits modification of Rab-family proteins with a carboxy-terminal CC motif as opposed to the carboxy-terminal CAAX motif 
. HBEC-30 cells pretreated with 3-IPEHPC for 24 hours were significantly refractory to infection by H1N1 (). Inhibitory activity was observed at concentrations as low as 125 nM (). Unlike SB218078, A549 cells were also responsive to 3-IPEHPC (). While the use of a mouse-adapted virus facilitates large-scale screening and allows comparisons with other published screening efforts, the extent to which results translate to seasonal or highly pathogenic strains is not established. Importantly, 3-IPEHPC was protective against infection with the avian strain H5/N1 and the recent pandemic swine flu strain H1/N1 ().
A stark limitation of arrayed siRNA screens is the requirement for “single gene” phenotypic penetrance. This can limit sensitivity of detection of relevant molecular entities due to insufficient protein depletion and/or the presence of functionally redundant gene products. As a mechanism to potentially reveal combinatorial contributions of gene function to viral replication and cytopathic effects, we repeated the original screen using a library of 426 human microRNA mimics. These reagents have the advantage of inducing multigenic perturbations, though accurate assignment of target space is a significant challenge. This effort identified a small cohort of miRNA mimics that either enhanced or deflected H1N1-induced cell death (, Table 6 in Supporting Information S1). 11 of these were further examined for consequences on H1N1 viral protein production in HBEC30 cells, which identified both sensitizers and resistors that enhanced or repressed viral replication (). Of note, a test for “hits” that also have activity against the recent pandemic strain Cal/04/09 identified two miRNA mimics that impair Cal/04/09 protein production in A549 cells (hsa-miR-495 and hsa-miR-519a, ). To infer biological processes that may be engaged by the miRNAs that can impair H1N1 replication, we examined the intersection of predicted miRNA targets and single-gene perturbations that behaved similarly to the subject miRNA. Candidate miRNA target genes were selected based on seed sequence presence in 3′ UTRs as defined by Target Scan context scores. These predictions were intersected with siRNA data from this study and those of the 4 whole-genome siRNA screens that measured influenza virus replication 
. When considered as a heuristic, this analysis produced three subnetworks that may correspond to the miRNA mode of action, namely the glycosylphosphatidylinositol transamidase, viral and host protein ubiquitylation 
and alternative mRNA splicing ().
Here we have focused on isolation of H1N1 pathogenicity response modifiers in human bronchial airway epithelial cells (HBEC). This cell type was selected as tissue culture model that may be enriched for conservation of cell autonomous biological features representative of the viral target tissue. These cells resist plaque formation, but are highly sensitive to single cycle infection. From whole-genome siRNA and miRNA mimic screening, both candidate sensitizer and resistor response modifiers were identified. A key deliverable from this analysis was the identification of gene products that apparently serve to restrain cell death responses that would otherwise engage upon detection of viral infection. Though not required to support cell viability in the absence of viral challenge, depletion of genes in this class enhanced the death response to H1N1 infection concomitant with restraining H1N1 protein production. As such, this class may represent targets for interventions that restrain propagation of multi-cycle infection by facilitating suicide of infected cells prior to production of new infectious particles. A chemically addressable member of this class, CHEK1, showed strong activity in multiple HBEC lines but not in A549, a non-small cell lung tumor derived line commonly employed to model influenza virus infection. This suggests that intervention targets may be available in normal epithelial cells that are uncoupled from host regulatory networks in cancer cells.