Dual roles for the ER membrane protein complex in flavivirus infection: viral entry and protein biogenesis

Hundreds of cellular host factors are required to support dengue virus infection, but their identity and roles are incompletely characterized. Here, we identify human host dependency factors required for efficient dengue virus-2 (DENV2) infection of human cells. We focused on two, TTC35 and TMEM111, which we previously demonstrated to be required for yellow fever virus (YFV) infection and others subsequently showed were also required by other flaviviruses. These proteins are components of the human endoplasmic reticulum membrane protein complex (EMC), which has roles in ER-associated protein biogenesis and lipid metabolism. We report that DENV, YFV and Zika virus (ZIKV) infections were strikingly inhibited, while West Nile virus infection was unchanged, in cells that lack EMC subunit 4. Furthermore, targeted depletion of EMC subunits in live mosquitoes significantly reduced DENV2 propagation in vivo. Using a novel uncoating assay, which measures interactions between host RNA-binding proteins and incoming viral RNA, we show that EMC is required at or prior to virus uncoating. Importantly, we uncovered a second and important role for the EMC. The complex is required for viral protein accumulation in a cell line harboring a ZIKV replicon, indicating that EMC participates in the complex process of viral protein biogenesis.


Results
A genome-scale RNAi screen identifies DENV host factors. We conducted a genome-scale RNAi screen to identify host proteins that are necessary for robust DENV2 (New Guinea C strain) infection using a siRNA library targeting 22,909 mRNAs in human HuH-7 hepatoma cells (Fig. 1A). HuH-7 cells were reverse-transfected with siRNAs, incubated for 52 hours and inoculated with DENV2 at low multiplicity of infection (MOI). At 42 hours post-infection, cells were fixed and stained using an anti-flavivirus envelope protein antibody and Hoechst stain 30 . The fraction of cells infected in each well was ascertained by high-content imaging and analysis (Fig. 1A). Each plate had three independent negative control siRNAs (AllStars, Nonsilencing and GFP) and a positive control siRNA targeting a subunit of the vacuolar ATPase (ATP6V0C), which is required for endosomal acidification and efficient virus infection 15,16,30,31 . Figure 1B shows individual fields and percent infection for the average from the population of the negative (AllStars = 86.22% infected, Nonsilencing = 46.93% infected, and GFP = 74.96% infected) and positive control (ATP6V0C = 6.95% infected) siRNAs. The distribution of the four control siRNAs throughout the screen is shown in Fig. 1C and in Table S1. The significant differences between the three negative control siRNAs likely reflect different off-target effects for each of the siRNAs. Importantly, however, all three negative controls were significantly different and easily distinguished from the ATP6V0C positive control (One-way ANOVA, P < 0.0001, Tukey's post-test, P < 0.01).
In the screen, each gene product was targeted by at least four unique siRNAs grouped into two distinct pools (Set AB or Set CD), generating two separate measurements of infection rate for each gene (Fig. 1D) 15,16,30 . The complete experimental data for the screen are presented in Table S1. The distribution of infection rates spanned the range of the assay (0 to 100%) with the majority of data points skewed towards higher rates of infection (Fig. 1E), suggesting that our screen could effectively identify hits as strong as the positive control. We identified 455 hits that had a Z-score less than or equal to 3 relative to the ATP6V0C siRNA positive controls (see "hits" tab in Table S1). Similar to a reported RNAi screen for YFV17D host factors 16 , the most potent candidate host factors identified in this screen included ribosomal proteins (predominantly of the 60S subunit), components of the vacuolar proton pump (e.g., ATP6V0D), proteins required for protein translocation across the ER (e.g., SEC. 61A1), and proteins known to participate in biogenesis of flavivirus proteins (e.g., SPCS2). Among very interesting novel hits is the poorly annotated RNA-binding domain containing protein C1ORF144, which associates with SRP9 and SRP14 32 , two components of the signal recognition complex. Importantly, EMC3, EMC2 and EMC5 were identified as potential DENV2 proviral host factors (Table S1). Identification of these known and suspected flavivirus host factors suggested that the DENV RNAi screen was robust.
Meta-analysis of DENV and YF17D RNAi screen identifies novel candidate Flavivirus proviral factors. The screen for DENV proviral factors reported herein was performed with the same siRNA library, cell line and primary antibody as a previously published screen for YFV17D proviral factors 16,33 , permitting a robust meta-analysis of the two screens. The YFV and DENV screen datasets were aligned, and after 2,034 genes were removed due to low cell number in any of the interrogated wells, we analyzed the data for 20,875 genes. The percent of cells infected by either YFV17D or DENV2 was ranked, and a nonparametric summation of ranks was adopted that permitted comparison between the screens. A permutation analysis of the ranks was performed and identified a p ≤ 0.00137 to give a false discovery rate of less than 30 genes. Using this criterion, we identified 274 common proviral factors (Table S2), 49 of which had not been identified by either screen alone (Fig. 1F).
The meta-analysis reinforces some concepts that have emerged from previous screens. First, there is a dramatic overrepresentation of proteins of the large ribosomal subunit: 21 canonical large subunit proteins vs 6 proteins of the small ribosomal subunit (see large ribosomal subunit proteins in red font and small subunit proteins in purple font in Table S2). Additionally, we found several subunits of the vATPase and factors involved in translation of ER resident proteins (SEC61, SRP54).
Among the 274 common hits were TTC35, TMEM111 and TMEM32 (MMGT1/EMC5) which are EMC2, EMC3 and EMC5, respectively 23 . No other EMC subunit was identified by our siRNA screens. Interestingly, the YFV17D proviral factor hit list identified EMC2 and EMC3 but did not pick up EMC5, however, the meta-analysis, combining the power of testing related viruses using a common screening paradigm, suggested that EMC5 is also a proviral factor for YFV17D. The meta-analysis also predicted that C1orf9, which is the human homologue to yeast Slp1p, a protein that was suggested to have a role in the yeast EMC pathway 24 is a host factor for DENV2 and YFV. Consistent with our YFV screen 16 and more recent screens of others [18][19][20] , our data identified a subset of the EMC subunits and a protein associated with the EMC pathway as YFV and DENV2 candidate host factors. Given that the EMC was observed to strongly impact several flaviviruses we decided to study this complex further.
Validation of EMC as a proviral factor. We initially validated the YFV screen results using gene editing technology 34 by transfecting pools of HuH-7 cells with plasmids expressing Cas9 and sgRNAs targeting various EMC subunit genes and infecting with YFV17D at a MOI 0.1. Approximately 42 hours post infection (pi), a time point that allows for multiple viral lifecycles, the infection was terminated and cells were labeled by immunofluorescence targeting the viral E protein. The fraction of cells positive for E protein (% infected) was quantified by automated imaging. Relative to the two GFP controls, one sgRNA targeting EMC1, and two independent sgRNAs each targeting EMC3, EMC4, or EMC5 reduced infectivity 5-to 20-fold (Fig. 2). While we do not know why www.nature.com/scientificreports www.nature.com/scientificreports/ EMC3 sgRNA#3 failed to inhibit, we conjecture that this sgRNA failed to edit the EMC3 locus. The combined results from the RNAi-based screens and gene editing validation experiments established EMC1-5 as YFV proviral host factors. The requirement for many subunits of the EMC for efficient infection strongly suggested that the entire human EMC is a flavivirus proviral factor, and is consistent with the model, first proposed by Schuldiner and colleagues, that the EMC is a multi-protein complex and loss of an individual subunit disables the entire complex 24 .
We used gene editing to establish HuH-7 lines where EMC4 was tagged with an N-terminal HA epitope (Figs 2B and S1 for uncropped gel). Two independent EMC4 null cell lines were then generated from one of the HuH-7(HA-EMC4) clones and selected for further analysis. Endogenous EMC2 was expressed at equivalent levels in all cell lines while EMC4 was modified with the HA tag or knocked out. Importantly, the parental HuH-7(HA-EMC4) and daughter HuH-7(HA-EMC4 KO) cell lines were propagated similarly over many passages and were morphologically indistinguishable, demonstrating that the EMC is not essential for HuH-7 cell viability and proliferation.
We next quantified production of extracellular virus in HuH-7(WT), HuH-7(HA-EMC4) and HuH-7(HA-EMC4 KO) cell lines infected with multiple flaviviruses. We analyzed the kinetics of infectious virus production for YFV17D and ZIKV (P6-740) viruses. For YFV, the initial viral burst occurred by 20 hr post-infection (pi) in the HuH-7(WT) and HuH-7(HA-EMC4) cell lines and virus production continued to increase up to 33.5 hr pi (Fig. 3A). In the EMC4 KO cell lines virus production was significantly delayed with no increase at 20 hr pi and up to 3 log 10 reduction in virus titer relative to the parental HuH-7(WT) and HuH-7(HA-EMC4) cells lines at 33.5 hr pi (Fig. 3A). For the Asian lineage ZIKV (P6-740), we detected 5.4 log 10 FFU/mL or 4.8 log 10 FFU/mL in HuH-7(WT) and HuH-7(HA-EMC4) cell lines, respectively; however, no increase in infectious ZIKV was observed over the time course in EMC4 KO cells (Fig. 3B), indicating that EMC4 was absolutely required for productive ZIKV infection.
Additionally, we screened a panel of flavivirus strains at single time points after infection. We asked if EMC was required for efficient replication of the pathogenic YFV Asibi strain, which was used to derive the 17D strain 35 , and observed that HuH-7 and HA-EMC4 cells produced over 3 log 10 FFU/mL more YFV Asibi than EMC4 KO cell lines (Fig. 3C). These data established the EMC as a critically important proviral factor supporting YFV replication in human cells. We analyzed infection of DENV strains from two different serotypes in EMC4 KO cell lines. DENV2-NGC and DENV4-TVP360 replicated to approximately 4 or 5 log 10 FFU/mL, respectively, in HuH-7(WT) and HA-EMC4 cell lines, but virus production was reduced to below detectable levels in the two EMC4 KO cells lines (Fig. 3D,E), validating EMC as proviral factor for DENV. Finally, we tested the distantly related virus WNV, which, unlike DENV, YFV and ZIKV, is transmitted by Culex rather than Aedes mosquitoes. www.nature.com/scientificreports www.nature.com/scientificreports/ virus produced by genetically modified cell lines infected at an MOI 1 is shown. Virus-containing media was harvested at 0, 9.8, 20 and 33.5 hours pi. Data shown are representative of 3 independent assays. The limit of detection is 1.5 log 10 (FFU/mL). (B) ZIKV virus produced by genetically modified cell lines infected at an MOI 10 is shown. Virus-containing media was harvested at 3, 10.5, 19 and 27.3 hours pi. This assay was performed one time. The limit of detection for this assay is 1.5 log 10 (FFU/mL). (C) YFV Asibi virus produced by genetically modified cell lines infected at an MOI 1 is shown. Virus-containing media was harvested 42 hours pi. Data are representative of 2 independent assays. The limit of detection for this assay is 1.7 log 10 (FFU/mL). (D) DENV2 virus produced by genetically modified cell lines infected at an MOI 5 is shown. Virus-containing media was harvested at 42 hours pi. Data shown are representative of 3 independent assays. The limit of detection for this assay is 1.5 log 10 (FFU/mL). (E) DENV4 virus produced by genetically modified cell lines infected at an MOI 2 is shown. Virus-containing media was harvested at 42 hours pi. Data shown are representative of 2 independent assays. The limit of detection for this assay is 1.5 log 10 (FFU/mL). (F) WNV virus produced by genetically modified cell lines infected at an MOI 0.5 is shown. Virus-containing media was harvested 48 hours pi. This assay was performed one time. The limit of detection for this assay is 1.7 log 10 (FFU/mL). Each bar or time point represents the mean and standard deviation for 3 replicate wells. Abbreviations: WT = HuH-7(WT), HA-EMC4 = HuH-7(HA-EMC4) cell line, KO #1 and KO #2 = HuH-7(HA-EMC4 KO) cell lines clones 1 and 2.
www.nature.com/scientificreports www.nature.com/scientificreports/ Work of Ma and colleagues 17 suggested that WNV did not required EMC for viral replication. Indeed, pathogenic WNV-NY99 virus production by EMC4 KO cells lines was similar to that for the control parental cell lines (Fig. 3F). Together, these results indicated that the EMC is required by pathogenic flaviviruses transmitted by Aedes mosquito species. The magnitude of the effects observed for DENV2, DENV4 and ZIKV was impressive and indicated an essential role for the EMC in the replication of these viruses.

The EMC is required for efficient infection of Aedes aegypti mosquitoes.
Previously, we reported a large overlap between dipteran and human proviral factors 15 and we wondered whether the EMC is a proviral factor in mosquitoes. The EMC is highly conserved between human and Aedes aegypti, which is the principal insect vector for urban cycles of YFV, ZIKV and DENV 4,36 . Aedes aegypti EMC subunits EMC2, EMC3 and EMC4 were readily identified. EMC2 (aaEMC2; XP_001661937), EMC3 (aaEMC3; XP_001652133), and EMC4 (aaEMC4; XP_001657467) are 44%, 65% and 51% identical based on protein sequence comparison to the respective human EMC subunits EMC2 (NP_055488), EMC3 (NP_060917), and EMC4 (NP_057538). These three EMC subunits were selected for RNAi mediated depletion using an established model of knockdown in mosquitoes 37 .
Aedes aegypti mosquitoes were injected with dsRNAs targeting GFP as a negative control or EMC subunits and infected with DENV2-NGC by ingestion of a virus-containing blood meal (Fig. S2). Fifty-seven (of fifty-nine) mosquitoes injected with dsRNA targeting GFP were productively infected with DENV, with a median of 4.31 log 10 PFU/midgut (Fig. 4). In the experimental groups, 61 (of 61) mosquitoes treated with dsRNA targeting aaEMC2 and aaEMC3 subunits established productive infections and 57 (of 60) mosquitoes in aaEMC4 subunit. Silencing efficiency was determined at the day of DENV infection. EMC gene expression ranged from 49%, 73%, and 68% for EMC2, EMC3, and EMC4, respectively, compared to GFP-injected controls. The median virus production for mosquitoes treated with dsRNA targeting aaEMC2, aaEMC3 or aaEMC4, was 3.98 log10 PFU/midgut, 3.93 log10 PFU/midgut, and 3.80 log10 PFU/midgut, respectively (Fig. 4). Results show that dsRNA targeting aaEMC3 (P = 0.0257) and aaEMC4 (P = 0.0010) significantly reduced infection, while dsRNA targeting aaEMC2 (P = 0.1115) slightly reduced infection, but was not statistically significant. We also observed high mortality in EMC-silenced group vs. GFP controls suggesting a fitness cost of EMC silencing, reducing the number of mosquitoes with high levels of silencing (data not shown). These results provide the first evidence that the EMC is an evolutionarily conserved DENV proviral host factor for both humans and mosquitoes.
The EMC is required for efficient virus entry. We set out to investigate the mechanism(s) by which the EMC promotes virus infection. For these experiments we chose to focus on DENV2 and ZIKV as these viruses were both profoundly affected by EMC4 KO (Fig. 3). Brass et al. reported that DENV and ZIKV attachment were unaffected by EMC knockdown 19 . In agreement with this, we failed to observe any difference in YFV17D attachment to EMC4 knockout and parental cell lines (Fig. S1). We next evaluated DENV2 NS3 accumulation at early time points after infection. We infected HuH-7(HA-EMC4) and HuH-7(HA-EMC KO clone #2) cell lines with DENV2-NGC (MOI = 10). Cells were pretreated with NITD008, an RNA dependent RNA polymerase (NS5) inhibitor 38 , to inhibit RNA synthesis. To establish the background signal in these experiments we pretreated cells with cycloheximide (CHX) to block synthesis of viral proteins. The NS3/β-actin protein ratio in HA-EMC4 cells increased over the course of 4 hr pi. In contrast, normalized NS3 levels in infected EMC4 KO cells were significantly (p </=0.05) reduced at all time points tested relative to the parental cell line (Fig. 5A). We further analyzed infection of these cells with an infectious DENV2 encoding Renilla luciferase (RLuc) 39 . Luciferase levels were monitored at 1, 1.5 and 2 hr pi in the presence or absence of CHX to control for contaminating RLuc present in virus stocks. This analysis revealed significant differences in RLuc signals between the infected HA-EMC4 and www.nature.com/scientificreports www.nature.com/scientificreports/ EMC4 KO cells (Fig. 5B) without detectable differences in viral RNA levels (Fig. 5C). With CHX, the levels of input RLuc decayed over the time course. Interestingly, in EMC4 KO cells the RLuc signal produced by infection was above the CHX control at 1.5 and 2 hr pi, suggesting that the block imposed by lack of EMC4, although www.nature.com/scientificreports www.nature.com/scientificreports/ profound, is not absolute. Together, these data indicated that EMC promoted an early stage of DENV2 infection, at or before the step of viral protein biogenesis, as suggested by Brass and colleagues 19 .
We developed a new method to ascertain effects of EMC4 on early steps in the viral lifecycle. This method, termed VIR-CLASP, depends on uncoating of viral genomes and crosslinking of exposed viral genomes to host RNA-binding proteins (RBPs) in the cytoplasm. For this protocol, stocks of ZIKV were prepared in cells labeled with 4-thiouridine (4SU), allowing viral genomes to incorporate 4SU. The thiol group allows for efficient RNA-protein crosslinking upon exposure to UV light at 365 nm. We infected HA-EMC4 and EMC4 KO cells with labeled ZIKV (PRVABC59) at a high MOI (500) for one hour at 4 °C, the cells were washed with cold PBS to remove free virus, further incubated for 30 min at 37 °C, and then irradiated cells with UV 365 to induce RNA-protein crosslinking. Cells were then lysed and viral ribonucleoprotein complexes stringently isolated (see Materials and Methods) for analysis using solid-phase purification of these complexes under denaturing condition.
Three important controls were incorporated into these experiments: (i) infection with unlabeled ZIKV, (ii) pre-treatment of cells with bafilomycin which prevents endosome acidification 40 and fusion of viral envelope, and (iii) UV irradiation of virus prior to infection. Silver stain analysis revealed multiple proteins crosslinked to ZIKV RNA only in HA-EMC4 cells infected with 4SU-labeled virus, whereas bafilomycin pre-treatment or irradiation of virus prior to infection reduced protein crosslinking as expected (Fig. 6). Importantly, the level of proteins recovered by VIR-CLASP was dramatically reduced in EMC4 KO cells (Fig. 6). We discovered, by Western blot analysis, that the RBPs FMRP, YTHDF1 and to a lesser extent ELAVL1, efficiently came down with ZIKV RNA in HA-EMC4 cells infected labeled virus but not in EMC4 KO cells (Fig. 6). The non-RBP, TUBA4A, was not present in any of the VIR-CLASP samples. These results show that EMC4 is required for ZIKV RNA to access the host cytoplasm and implicate the EMC as a host complex required for efficient virus entry or uncoating.
After viral entry the EMC is required for efficient ZIKV protein accumulation. The data presented above, in agreement with work by Brass and colleagues, indicates a likely indirect role for the EMC requirement in a very early viral step, however, given that the EMC has been implicated in protein biogenesis 24,25,27,29 we wondered whether this complex could also be required for the efficient expression of viral proteins. To test this, we used a HuH-7 cell line harboring an autonomously replicating, subgenomic ZIKV replicon (ZIKV RepNeo) 41 . The replicon includes the ZIKV (strain FSS13025) 5′ UTR, nonstructural proteins and 3′ UTR, which is interrupted www.nature.com/scientificreports www.nature.com/scientificreports/ by an IRES driven neomycin resistance cassette (Fig. 7A). The structural proteins have been replaced with RLuc, permitting interrogation of viral lifecycle steps related to viral protein biogenesis and RNA synthesis, in the absence of viral entry, assembly and exit.
The ZIKV RepNeo cell line was reverse transfected with either of two independent negative control siRNAs or anti-EMC4 siRNAs. Four days after transfection, the total cell-associated RNA from the RNAi-treated cells was collected and RT-qPCR was used to quantify endogenous GAPDH mRNA and viral replicon RNA. Viral replicon RNA levels for negative control siRNAs A and B were not statistically different, while anti-EMC4 siRNA transfection reduced replicon RNA by 55% and 54% for siRNAs A and B, respectively (Fig. 7B), consistent with an important post-entry role during ZIKV replication.
We also analyzed the expression of several replicon proteins after EMC4 knockdown. The ZIKV RepNeo cell line was reverse transfected siRNAs as above and protein samples were collected four days later for analysis of www.nature.com/scientificreports www.nature.com/scientificreports/ EMC4, RLuc, NS2B, NS3, NS4B and β-Actin levels. EMC4 protein expression for the negative control siRNAs was similar, while EMC4 protein expression was reduced by 73% or 94% by EMC4 siRNAs A and B, respectively (Fig. 7C,D). Relative to the negative control siRNAs, expression of RLuc and all ZIKV proteins assayed was reduced by EMC4 knockdown, consistent with an important post-entry role for the EMC in ZIKV replication (Fig. 7C,D).
We noted interesting differences in the effects of EMC4 depletion on levels of the analyzed ZIKV proteins and RLuc. Specifically, NS3 levels were reduced more significantly than RLuc or NS4B, while the effect of EMC4 knockdown on NS2B was intermediate (Fig. 7C,D). We reasoned that these differential effects may be related to inherently different rates of turnover for these proteins, which would implicate a role for EMC in the biogenesis of viral proteins. To address this, we analyzed the levels of each protein in CHX-treated cells over the course 32.5 hours (Fig. 7E,F). The ZIKV proteins exhibited different rates of degradation and this was particularly evident at 21 hours post-CHX treatment when 61% of NS4B, 43% of NS2B and 27% of NS3 remained intact (Fig. 7F). The relative stabilities of these proteins correlated with the effects of EMC4 knockdown which had the largest effect on NS3 and smallest effect on NS4B (Fig. 7D). These observations strongly suggest that the EMC promotes the biogenesis of ZIKV proteins. Interestingly, RLuc was the most unstable protein examined as only 13% remained at 6 hours post-CHX treatment (Fig. 7F), but was relatively unaffected by EMC4 depletion (Fig. 7D). The dichotomy between levels of RLuc and levels of the viral proteins after EMC depletion suggests that this complex impacts the biogenesis of viral proteins at a stage subsequent to initiation of translation. Importantly, our data clearly indicate an EMC requirement for a post-entry effect that is fully consistent with the model proposed by Hedge and colleagues 25,26,29 wherein the EMC coordinates the insertion of transmembrane domains and can protect polytopic proteins from premature degradation.
All of the data presented above leads to the conclusion that EMC is required twice for infection of Aedes transmitted flaviviruses, first, at a step subsequent from attachment but prior to viral uncoating, and second, at a step subsequent to uncoating, and very likely a step after translation initiation, but required for the biogenesis of viral proteins.

Discussion
We screened a library of siRNA pools for the ability to reduce DENV2 infection of a human cell line at a time-point that permitted at least one complete viral lifecycle. We identified DENV proviral factors that were required to a degree similar to the positive control, a subunit of the vATPase. Consistent with the biology of DENV infection 9 , our high confidence hit list included three subunits of the vATPase (ATP6V0C, ATP6V0D1, and ATP6V1F), many subunits of the ribosome and translational apparatus (n = 31), and factors necessary for translocation and processing of the viral polyprotein (SRP54, SEC61A1, SPCS2). Among the high confidence hits there are many that are novel, and mechanistic insights are still not available into how the majority proviral factors, novel and previously known, impact flaviviruses. Nonetheless, this screen, previous ones from our laboratory 15,16 and those from other groups [17][18][19][20]42 have opened important lines of investigation for these viruses.
A common theme from all the aforementioned screens is a reliance on host factors that reside in or impact the ER. Previously, our lab identified that drosophila fly gene CG33129 and its human homologue TMEM214 as DENV proviral factors 15 , although the mechanistic details were not further pursued. The current screen concurred since TMEM214/FLJ20254 was once again identified as a DENV proviral factor. TMEM214/ FLJ20254, although largely unexplored, was reported to localize to the ER and may function in the ER-stress response 43 . The current screen also supports the hypothesis that the TRAPP complex may have a conserved pro-flaviviral role since TRAPPC1 was identified here and in our prior published screen for YFV proviral factors along with TRAPPC11/FLJ12716 16 . TRAPPC4 was also identified in a recent screen for flavivirus proviral factors 19 . Depletion or mutation of TRAPPC11 (and other subunits) disrupted ERGIC and Golgi morphology, prevented export 44 and N-linked glycosylation of a model protein 45 suggesting that the mammalian TRAPP complex is involved in ER-to-golgi trafficking and post-translational glycosylation. Further investigation of the DENV-TRAPP complex relationship may inform on the virus-host interaction, the biological role(s) of the TRAPP complex and may broaden our understanding of rare human genetic disorders. ERI3, a recently identified DENV-3′UTR interacting protein, was shown to support viral replication 46 , and appears herein as a DENV proviral factor, demonstrating how orthogonal screening technologies converged on a common DENV proviral factors ER resident signalases mediate specific steps in polyprotein processing 47 , and recent screens identified multiple subunits of the signalase as flavivirus proviral factors [18][19][20] . The Diamond lab created renewed interest here by suggesting that the composition and activity of the signalase(s) involved in viral polyprotein processing may vary 20 . Our screen identified signalase subunit SPCS2 as a DENV proviral factor, and future investigations may evaluate the SPCS2-dependent signalase activity in the context of viral polyprotein processing. The OST complex is a recently identified pan-flavivius proviral factor 18-20 that supported viral replication through an enzyme-independent mechanism that remains to be clearly deciphered 18 , and our screen also identified that the OST complex subunits DAD1 and RPN2 are a DENV proviral factor.
Among the most critically required proviral factors we have investigated is the EMC. This ER associated 10 subunit complex is exquisitely required for the replication of DENV2, DENV4, YFV, ZIKV (this work and [16][17][18][19][20]. Like many other proviral factors 15 we show that EMC is required for viral replication in both human cells and Aedes aegypti mosquitoes. The EMC requirement is not shared by WNV. Wu and colleagues reported that WNV infection was only slightly delayed in an EMC2 knockout cell line, relative to the parental cell line 17 . Diamond and colleagues reported that knockout of EMC4 reduced WNV infection 12 hr pi approximately 75% by one sgRNA, but reported a second sgRNA resulted in a statistically insignificant WNV reduction. In addition, WNV was reduced at least two fold for two different sgRNAs targeting EMC6 20 . Although, WNV was not affected by EMC4 knockout in our experiments, it could be that a delay would have been missed in our experiments. Nonetheless, www.nature.com/scientificreports www.nature.com/scientificreports/ it is clear that there is a marked difference between the EMC requirement for DENV2, DENV4, YFV, and ZIKV versus that for WNV and this difference deserves further investigation.
What are the roles of the EMC on the flaviviral lifecycle? Brass and colleagues reported that DENV and ZIKV attachment were not blocked by EMC knockdown 19 . Nevertheless, the pattern of internalized DENV E protein in DENV infected cells was altered shortly after infection of EMC knockdown cells 19 which suggested a defect up to and including viral RNA synthesis. For ZIKV, attachment was not altered between control and EMC knockdown cells, however, there was a suggestion that entry was blocked leaving virus on the surface of the cell. Our data using VIR-CLASP, which can directly detect interactions between incoming viral genomes and cellular proteins and thus uncoating, supports a role for the EMC in ZIKV entry. A possible explanation for the effect on entry is that EMC facilitates the biogenesis and/or assembly of required receptors in the plasma membrane or endosome resulting in an indirect effect on viral infection 27,28,[48][49][50] .
An important finding of this study is that the EMC promotes a post-entry phase of the ZIKV lifecycle. Depletion of EMC subunits reduced ZIKV protein biogenesis in a replicon system where virus entry is not required. Our data also suggest that the EMC is acting at a step after translation initiation of the RLuc ORF and self-cleavage of the luciferase from the ZIKV polyprotein. We conclude that the EMC plays a role in the biogenesis of the flaviviral proteins, which requires the synthesis, membrane insertion, modification, and maturation of the polytopic viral polyprotein. This role is consistent with the reported roles of the EMC on cellular proteins 24,25,[27][28][29] . EMC is important for the biosynthesis of multi-pass membrane proteins 26,28 , a function likely mediated by the complex's transmembrane domain insertase activity 25 . This EMC function appears to be essential to stabilize client polytopic proteins from degradation 29 . Given its structural and topological complexity the flaviviral polyprotein would be expected to directly require EMC. Perhaps the surprise is that WNV manages to circumvent this requirement. Given the difficulty of creating chimeric flavivirus ORFs it will be challenging, but interesting, to use this differential requirement to further delineate the properties that make a nascent protein sensitive to EMC function. The fact that the EMC is required for more than one step in the viral lifecycle can explain the profound inhibition we observe for many flaviviruses.

Methods
Genome-scale RNAi screen. The siRNA screen was conducted at the Duke RNAi Screening Facility using the Qiagen Human Genome siRNA Library v1.0. For a detailed description of the methods and materials, see reference 30 . For each host gene, four independent targeting siRNAs were combined into two separate pools named Set AB and Set CD 33  Data analysis for genome wide RNAi screen. We selected an analysis strategy in which high-confidence DENV host factors were identified when the biological duplicate siRNA pools each reduced DENV at least as much as a control siRNA that targets a subunit of the vATPase. In order to identify hits, the mean and standard deviation of the % infection for all ATP6V0C siRNAs (Qiagen # SI00307384) from each respective set of the plates (Set AB compared to Set CD) was calculated. The Z-score for each well in Set AB was calculated using the mean and standard deviation from the ATP6V0C control siRNAs for all plates. These Z-scores are presented in Supplemental Table 1. The 455 high confidence hits have Z-scores equal to or less than 3.00.

Meta-analysis of the DENV and YFV RNAi screen data sets.
The genome-wide data sets for the YFV17D 16 and DENV2 siRNA (this manuscript) screens were collated. Any gene target for which one or more wells from any contributing dataset had low numbers of cells/well (400 and 245 or less over two fields for the YFV and DENV screens, respectively) was filtered from further analysis resulting in removal of 2,034 genes. 20,875 gene targets were analyzed in this meta-analysis. The percent of cells infected by either YFV17D or DENV2 was ranked for each gene. A nonparametric summation of ranks was adopted in order to combine disparate screen datasets. In total, four YFV genome-wide siRNA screen datasets were combined with two DENV datasets. In order to compensate for the four YFV datasets, relative to the two DENV datasets, all rank values for the YFV data sets were divided by 2. Finally, the total screen ranks were summed for each of the 20,875 genes with a minimum possible sum rank of 4 and maximum of 83,500. A mock permutation of the ranks followed by summation of the permuted ranks (n > 1 × 10 7 permutations and summations) was performed and identified that a p ≤ 0.00137 was equal to a summation of 13,100. The calculated false discovery rate for this p-value is < 30 genes. The sgRNAs targeting EMC1, 4 and 5 were designed using sequences predicted and published in reference 34 . EMC1 was targeted by sequence human_exon_crispr_v1_003453. EMC4 was targeted by sequences human_exon_crispr_ v1_120084 and human_exon_crispr_v1_120085. EMC5 was targeted by sequences human_exon_crispr_v1_188532 and human_exon_crispr_v1_188533. The sgRNAs targeting EMC3 target the sequences 5′-CCGCCACTAC GTGTCCATCCTGC-3′, 5′-CCACTACGTGTCCATCCTGCTGC-3′, and 5′-CCTGCTGCAGAGCGACAAGAA GC-3′. The plasmids expressing sgRNAs that target EMC1, 3, 4, and 5 were constructed following the procedure outlined as "Option B" in the "gRNA Synthesis Protocol" in 34 .
Gene editing. Briefly, HuH-7 cells grown on 12-well plates were transfected with 0.5 μg of pcDNA3.3-TOPO CAS9 and 0.5 μg of the sgRNA cloning vector using Lipofectamine 2000 (LF2K) (Invitrogen) following manufacturer's recommendations. The following day, transfected cells were passaged into 10 cm dishes containing growth media plus 1 mg/mL G418. Cells were carried in G418-containing media for 3-5 days after which cells were cultured in media without G418 for an additional 7 days. The pools of edited cells were seeded onto collagen-coated 96-well plates (Corning COSTAR) at a density of 2 × 10 4 cells/well prior to infection.
After transfection cells were cultured as described above and then passaged onto collagen coated 15 cm plates at sparse density. After 12 to 14 days colonies were selected by eye and transferred to individual wells in 96 well format. HA-tagged colonies were initially screened by replica plating the 96 well plates and staining one plate using primary anti-HA antibody (Cell Signaling). Subsequent SURVEYOR (Transgenomic) assays and amplicon sequencing confirmed knockin of the HA-tag was inserted into EMC4 loci.
HuH-7(HA-EMC4 KO) cell lines were derived from a common parental HuH-7(HA-EMC4) cell line by co-transfection with pcDNA3.3-TOPO CAS9 and sgRNA # human_exon_crispr_v1_120085. Candidate knockout colonies were identified using an analogous procedure as described above with the exception that HA-negative colonies were selected for expansion. Subsequent amplicon sequencing and enzymatic SURVEYOR assays confirmed that the HA-tag contained frame shift mutations. Virus production assays. Cell lines were seeded onto 48 well assay plates in growth media. The following day, cells were inoculated at the indicated MOI at 37 °C for approximately 1 hour. After virus absorption, cells were rinsed 1x with PBS plus MgCa, 0.25 ml DMEM (5% FBS, antibiotic, 0.01 M HEPES) was added. At the indicated time post absorption, virus containing media was collected and stored at −80 °C. Virus production was quantified by focus formation assay as previously described 15 .
RNAi-mediated EMC subunits silencing in Aedes aegypti mosquitoes. RNAi-mediated gene silencing, mosquito infection and plaque assays were carried out as previously described 37,51,52 . Briefly, Aedes aegypti mosquitoes Rockefeller strain (Johns Hopkins University) were maintained on 10% sucrose solution at 27 °C and 80% relative humidity with a 14:10 h light:dark cycle. Three-to four-day old female mosquitoes were cold-anesthetized and injected with 200 ng of dsRNA per mosquito targeting EMC2, EMC3, or EMC4 and mosquitoes injected with GFP dsRNA were used as controls. The dsRNA was synthesized using the HiScribe ™ T7 In Vitro Transcription Kit (New England Biolabs). The primer sequences (5′-3′) used for dsRNA synthesis are the following (lowercase letters correspond www.nature.com/scientificreports www.nature.com/scientificreports/ of clarified supernatants through a 20% sucrose cushion in TNE buffer (50 mM Tris-HCl [pH 7.2], 0.1 M NaCl, and 1 mM EDTA) at 125,000 × g for 4 hr in a Beckman SW32Ti rotor. To remove remaining free 4SU, virus pellets were washed three times with TNE buffer, and then resuspended in virus dilution buffer (DEME medium containing 10 mM HEPES [Gibco] supplemented to contain 1% FBS), aliquoted, and stored at −70 °C. Virus titers were determined by plaque assay using Vero cells.
For VIR-CLASP, cells were infected with 4SU labeled virus for 1 hr at 4 °C and uninfected virus was washed away with cold PBS. The infected cells were incubated for 30 min at 37 °C, prior to 365 nm ultraviolet irradiation. Irradiation using UV 365nm allows for covalent crosslinking of interacting host proteins to the incoming RNA genome but not to cellular RNAs not labeled with 4SU. To irradiate with UV 365nm , the growth medium was removed and washed with PBS. Cells were irradiated on ice with 365 nm UV light (0.6 J/cm2 × 2 times) in a Stratalinker 2400 (Stratagene). Cells were scraped off in 2.5 ml PBS per plate.
Cells were lysed in denaturation buffer (50 mM Tris-HCl, pH 6.8, 10% glycerol, 2.5% SDS, 0.66% NP-40), incubated for 10 min at 95 °C and subsequently slowly cooling them to 25 °C. Crosslinked RNA-protein complexes were purified by Solid-Phase Reversible Immobilization (SPRI) 54 beads (GE Healthcare, cat# 65152105050250) under denaturing SPRI buffer (50 mM Tris-HCl, pH 6.8, 10% glycerol, 2.5% SDS, 0.66% NP-40, 1 M NaCl, 8% PEG-8000). This allows for quantitative recovery of RNA and enrichment of crosslinked proteins since only those proteins can be pulled down under high SDS conditions. To each sample, 0.66× (e.g. 660 μl of beads for 1 ml of sample) of SPRI beads were added. The SPRI beads and complexes were washed 5 times with denaturing SPRI buffer. The crosslinked RNA-protein complexes were eluted in denaturation buffer. To reduce non-specific binding on the beads, SPRI purification was repeated. An equal volume of 4x Benzonase buffer (80 mM Tris-HCl, pH 7.5, 600 mM NaCl, 20 mM MgCl 2 , 4 mM DTT, 40% Glycerol) and 2x volume of water were added to eluted samples, followed by the addition of Benzonase (EMD Millipore, cat# 70746-4) to a final concentration of 50 U/ml, and incubation for 2 hr at 37 °C. Proteins were precipitated by methanol and chloroform and then re-suspended in 2x NuPAGE LDS Sample Buffer (Thermo-Fisher, cat# NP0007) with 50 mM DTT.
Availability of materials. Materials, data and associated protocols described in this manuscript will be promptly available to readers.

Data Availability
All data generated or analyzed during this study are included in this published article (and its Supplementary Information files).