A protein-proximity screen reveals Ebola virus co-opts the mRNA decapping complex through the scaffold protein EDC4

Abstract The interaction of host and Ebola virus (EBOV) proteins is required for establishing infection of the cell. To identify protein binding partners, a proximity-dependent protein interaction screen was performed for six EBOV proteins. Hits were computationally mapped onto a human protein-protein interactome and then annotated with viral proteins to reveal known and previously undescribed EBOV-host protein interactions and processes. Importantly, this approach efficiently arranged proteins into functional complexes associated with single viral proteins. Focused characterization of interactions between EBOV VP35 and the mRNA decapping complex demonstrated that VP35 binds the scaffold protein EDC4 through the C-terminal subdomain, with each protein found associated in EBOV-infected cells. Mechanistically, depletion of three components of the complex each similarly inhibited viral replication by reducing early viral RNA synthesis. Overall, we demonstrate successful identification of EBOV protein interaction with entire cellular machines, providing a deeper understanding of replication mechanism for therapeutic intervention.

To beeer iden4fy host protein complexes interac4ng with virus proteins, we tagged each EBOV structural protein (except the transmembrane glycoprotein) with a promiscuous bio4n ligase (BioID2) to label host interactors via proximity-dependent bio4nyla4on 18 .Here, we show that computa4onally inter-rela4ng the host-virus hits to known human protein-protein interac4on (PPI) networks allows for the rapid iden4fica4on of host protein complexes that interact with each virus protein, such as the mRNA decapping complex, which we characterize as an interac4on partner of VP35.Our data demonstrates the power of this approach to iden4fy cellular protein complexes necessary for produc4ve infec4on.
As a control to measure nonspecific bio4nyla4on we included GFP fused to BioID2 as well as parental cells not expressing BioID2.Constructs were expressed in tetracycline inducible Flp-In T-Rex 293 cells to provide comparable levels of expression.Nonetheless, expression varied (for example between VP35 N and C terminal fusions), likely reflec4ng stability of the fusion protein (Figure 1b).Consistent with previous work 5,10 , the RNAdependent RNA polymerase, L, was expressed at the lowest levels, with the N-terminal BioID2 tagged version present at levels just above the limit of detec4on.The C-terminal fusion to L was not detected and was therefore excluded from further analysis.
Bio4nyla4on of cellular proteins in close proximity to tagged viral proteins was ac4vated by incuba4ng cells with excess bio4n for 24 hours.Cells were then lysed and bio4nylated proteins purified on streptavidin beads.Five percent of the purified protein was subjected to western blojng to confirm successful bio4nyla4on and protein capture (Figure 1c).The paeern and level of bio4nyla4on was consistent between replicates but varied between constructs.While faintly expressed constructs yielded the lowest levels of bio4nylated host proteins, others such as the N-terminal fusions to NP gave much higher levels compared to their C-terminal constructs (Figure 1b-c, Extended data figure 1).This may reflect mechanis4c restric4ons on the protein.For example, C-terminal fusions to VP35 can disrupt folding due to the posi4oning of the C-terminal end in the hydrophobic pocket of the IID region of the protein 8 , which may result in protein instability.
The remainder of the samples were subjected to mass spectrometry to iden4fy bio4nylated proteins.
Spectral counts for each protein were filtered by SAINTExpress against the CRAPome database to iden4fy nonspecifically labeled proteins and develop a set of high confidence interactors 19 .Using GFP-BioID2 and unmodified parental Flp-In T-Rex 293 cells as background and a Bayesian false discovery rate of 0.01 as the cutoff, 330 proteins were iden4fied (Extended data table 1).Of these, fiky-two host proteins were iden4fied by both N-and C-terminal BioID2 constructs of the same viral protein.Seventy-two cellular proteins interacted with mul4ple viral proteins (Extended data table 1), mainly due to overlap with NP-NT, which demonstrated the highest bio4nyla4on ac4vity of all constructs.Therefore, to iden4fy cellular proteins that were specifically labelled by individual viral proteins, we repeated the SAINTExpress analysis for each viral protein using the mass spectrometry data from all other viral proteins plus GFP and parental cells as background.The resul4ng data set was significantly reduced in size and eliminated all but one shared interac4on (Extended data table 1).

Network analysis and viral protein mapping reveals potenBal interacBons of virus proteins with host protein complexes
Combined hit lists for the BioID SAINTExpress analyses yielded 447 proteins of interest (Extended data table 2).
As has been performed in other PPI screens, a hub and spoke network model was made showing the pairwise rela4onships of iden4fied virus-host protein interac4ons (Figure 1d).However, many cellular proteins execute their func4ons as part of larger protein complexes and so we sought approaches to relate hits according to known host protein-protein interac4ons (PPI).Computa4onally inter-rela4ng human PPI with the BioID hits using tradi4onal methodologies such as STRING 20 yielded a highly interconnected mass of interac4ons involving 337 hits but lacked priori4za4on by hit score or complex likelihood (Supplemental Figure 1).As an alterna4ve we implemented the Prize-Collec4ng Steiner Forest (PCSF) algorithm 21 , a graphing methodology that limits protein connec4on paths using cost-based evalua4on of connec4ons to priori4ze higher-confidence (low cost) interac4ons from the underlying human PPI and nega4vely weigh low-confidence (high cost) interac4ons (Figure 2a).As needed, PCSF also incorporates nodes from the underlying network, known as Steiner nodes, to form connec4ons between hits to develop an op4mal network (Figure 2a) 21 .
Supplemental Figure 1: STRING network analysis of BioID hits.STRING v12.0 was used to map BioID hits to the STRING database.The thickness of the black edge lines correlate to edge confidence scoring (0.4-1.0) while grey circles represent protein hits from the screen.Grey circles on the right side and bo<om of the figure not connected by dark grey lines show hits within the STRING database that could not be connected to the final network.
SAINT scores from the viral BioID2 hits were used as node prize values (Extended data table 2) and hits were mapped onto the Human Integrated Protein-Protein Interac4on rEference v2.3 (HIPPIE), an interactome assembled and scored from experimentally validated PPIs, in which interac4ons are given a confidence metric (edge weight) based on the number of publica4ons and types of experimental approaches suppor4ng the interac4on 22 .313 hits were mapped through PCSF.The virus proteins interac4ng with each host protein were then annotated onto the network (Figure 2b leR, Figure 2c).While most interac4ons were between only one virus and host protein, 16% of hits were found associated with mul4ple viral proteins and are indicated via segmented circles with propor4ons defined by SAINT scores (Figure 2b leR, 2c, Extended data table 3).From this analysis, groups of host proteins related by a common virus protein(s) were apparent.To determine if these groups had func4onal relevance, the network was clustered based on network topology and gene list enrichment was performed on each cluster using EnrichR 23 (Figure 2b right).Interes4ngly, clusters related by common virus proteins (Figure 2c) showed strong enrichment (adjusted p-values 0.014 to 8.36 * 10 !"# ) for func4onal pathways and were annotated as an addi4onal layer onto the network (Figure 2d, Extended data table 4).
The PCSF subnetwork clusters included known EBOV-host protein interac4ons but revealed new associa4ons with func4onally related host proteins.For example, VP24 was associated with mul4ple karyopherin (KPNA) proteins 24 (Figure 2c) but also with related components of the nuclear pore complex not previously described for relevance in EBOV infec4on, such as CSE1L, which promotes the nuclear export of karyopherins 25 (Figure 2d, cluster 16, Extended data table 4).Addi4onally, markers of cell-signaling processes known to be relevant in EBOV infec4on were seen but with novel EBOV interactors.Rho-GTPase signaling is necessary for both EBOV entry and assembly 26 , but was found represented by a cluster of proteins associated with NP and VP24, which extends previous findings that ac4n polymeriza4on is necessary to drive transport of mature viral nucleocapsids to sites of viral budding 27 (Figure 2d, cluster 3, Extended data table 4).Also, while the viral glycoprotein was previously found to induce NFκB signaling 28 , our analysis suggests NFκB signaling may also correlate to a cluster of NP and VP35 associated host proteins, sugges4ng addi4onal viral mediators to this host signaling pathway (Figure 2d, cluster 15, Extended data table 4).
We focused on a cluster of five proteins, DCP1A, DCP1B, DCP2, EDC3, and EDC4, labeled by VP35 (Figure 2c inset).This cluster was enriched for mRNA decay by 5' to 3' exoribonuclease as well as processing-body (Pbody) forma4on (Figure 2d, cluster 10, Extended data table 4).Consistent with the func4onal enrichment, all five proteins are components of the host mRNA decapping complex 29 .Since bio4nyla4on can involve direct as well as indirect interac4ons within 10 nm of the BioID2 enzyme 18 we predicted this cluster of proteins represented VP35 interac4on with an intact host protein complex.

EBOV VP35 interacts with the host mRNA decapping complex scaffold protein EDC4
To iden4fy interac4ons between VP35 and each member of the decapping complex, co-immunoprecipita4ons of exogenously expressed proteins were performed from lysates of HEK293T cells transfected with plasmids encoding FLAG-tagged VP35 together with HA-tagged DCP1A, DCP1B, DCP2, EDC3 or EDC4.Under the condi4ons tested, VP35 only coprecipitated with EDC4 (Figure 3a, last lane).This outcome was consistent with spectral labeling counts from the BioID screen, showing EDC4 was the most abundantly labeled decapping complex component in the VP35 BioID (Extended figure 2).To confirm the EDC4-VP35 interac4on, tags were reversed with plasmids encoding FLAG-tagged EDC4 and HA-tagged VP35 and pulldowns were performed with both an4-FLAG and HA an4bodies (Figure 3b, top and boWom panels respecBvely).Our results suggest that VP35 interacts with the decapping complex by binding EDC4.
To confirm interac4on of EBOV VP35 and EDC4 in infected cells, each protein was stained with specific an4bodies at 10 and 20 hours post infec4on (hpi) with infec4ous EBOV.For EDC4, staining was present in dis4nct puncta in the cytoplasm.Interes4ngly, for infected cells, 20-40% of the puncta were found localized with inclusion bodies iden4fied by VP35 staining (Figure 3c-d).To gain further insight into this associa4on, z-stack images were collected from infected cells stained for VP35 and EDC4.Images were deconvolved and modeled using Imaris imaging sokware.EDC4 was confirmed to be associated with VP35 both within and proximal to VP35 inclusions (Figure 3d, Supplemental movie 1).Also, a low level of staining for EDC4 and VP35 in small puncta dis4nct from viral inclusion bodies was evident.To determine if this popula4on of proteins was interac4ng, proximity-liga4on assays were performed to detect proteins within 40 nm of each other.Despite some background signal, infected cells showed both puncta and diffuse patches of signal that was dis4nct from that seen in uninfected cells (Figure 3e).Taken together these findings indicate that during EBOV infec4on VP35 and EDC4 interact with one another.showed that EDC4 was the most abundantly labeled protein of all decapping complex proteins iden,fied.

EDC4 regulates EBOV replicaBon at an early step in viral infecBon
Since EDC4 regulates mRNA stability through the decapping complex, we hypothesized that EDC4 deple4on would effect virus RNA (vRNA) replica4on.Therefore, to understand the importance of the VP35-EDC4 interac4on to the viral infec4on cycle, we depleted EDC4 protein with mul4ple siRNAs targe4ng different regions of the EDC4 gene in HeLa cells and challenged treated cells with EBOV.The amounts and subcellular localiza4on of vRNA was then visualized by RNAFISH.Deple4on of EDC4 (>80% for each siRNA, Figure 4a) resulted in 56% and 84% loss of cells with vRNA signal over mul4ple experiments (p-value < 0.01) compared to non-targe4ng (Allstars) siRNA controls, confirming that EDC4 plays a significant role in viral replica4on (Figure 4b-c).To determine which steps in viral replica4on were affected we measured both EBOV mRNA and genomic RNA levels in cells treated with siRNA.Oligo-dT reverse transcrip4on (RT) combined with qPCR was used to specifically amplify poly-adenylated RNAs and measure NP and GP transcripts.In cells depleted of EDC4 by each siRNA, NP transcript levels were seven and fourteen-fold reduced compared to non-targe4ng siRNA controls (Figure 4d, first panel) while GP transcript levels were 3 and 6-fold reduced compared to non-targe4ng controls (Figure 4d, second panel).The reduced mRNA levels appeared to result in a propor4onal 2 to 4-fold loss of viral protein levels, demonstrated by measuring VP35 protein levels (Supplemental figure 2).Next, levels of viral genomic and an4genomic RNA were measured in cells depleted for EDC4 by a previously described qPCR assay using primers specific for the virus trailer region 30 .Genomic RNA produc4on was decreased 3 to 5-fold in EDC4 knockdown samples when compared to controls (Figure 4d, third panel), while an4genomic RNA produc4on was decreased four to six-fold in EDC4 knockdown samples (Figure 4d, last panel), indica4ng that EDC4 is required for efficient viral RNA synthesis.

The EDC4 C-terminal domain is responsible for recruitment of VP35 and regulates viral replicaBon.
EDC4 forms a scaffold on which proteins DCP1A and DCP2, the minimal proteins required for mRNA decapping 31 , assemble.EDC4 contains an N-terminal WD40 domain and C-terminal proximal-distal domain, separated by a flexible linker (Figure 5a, top).DCP1A specifically binds EDC4 at the N-terminal WD40 domain while DCP2 binds the proximal domain at the C-terminal end 31 .The distal domain, which is highly conserved, does not bind members of the decapping complex but adopts a compact seven helix-turn-helix bundle structure and plays a role in EDC4 localiza4on to cytoplasmic puncta associated with P-bodies in Drosophila 32 .To understand which part of EDC4 interacts with VP35, HA-tagged EDC4 constructs encompassing the WD40 domain plus flexible linker or flexible linker domain plus C-terminal proximal-distal region of EDC4 were transfected into HEK293T cells with FLAG-tagged VP35 (Figure 5a, lower panel above doWed line).VP35 strongly coimmunoprecipitated with full-length EDC4 or with the C-terminal domain but was only weakly detected with the N-terminal WD40-containing domain (Figure 5b).We further tested whether overexpression of C-terminal subdomains affected viral replica4on in a dominant nega4ve-like manner.HA-tagged full-length EDC4 and doWed line) were transfected into HeLa cells, which were then challenged with EBOV.Cells were fixed at 20 hpi and viral replica4on was quan4fied via vRNAFISH (Figure 5c).Although expression of the proximal-distal domain demonstrated moderate effects on infec4on when compared to full-length EDC4 transfected samples, cells transfected with the distal domain alone showed a 6-fold reduc4on in infec4on compared to full-length EDC4 (Figure 5c).For each construct the frequency of transfected cells was similar though the proximal-distal domain was more highly expressed as seen by immunofluorescence (Figure 5d).Given that the distal domain of EDC4 plays an important role for recruitment to P-bodies, we examined how infec4on altered the distribu4on of EDC4 and DDX6, which are commonly present in P-bodies.HeLa cells were challenged with EBOV and fixed 48 hpi to allow for mul4ple rounds of infec4on.In cells with smaller VP35 inclusion body size, which represent cells in earlier stages of the EBOV infec4on cycle, DDX6 and EDC4 puncta were located proximally to VP35 inclusion bodies, as seen in Figure 3c and d.However, in cells with extensive VP35 staining, which indicates late-stage infec4on 33 , large puncta of EDC4 and DDX6 could no longer be detected (Figure 5e).Overall, the data shows that VP35 interacts with the C-terminal half of EDC4, and overexpression of the distal domain disrupts infec4on.High level expression of VP35 seen in advanced stages of infec4on also caused loss of EDC4 puncta associated with DDX6, a marker of p-bodies.Further work will elucidate the importance of the EDC4 distal domain in viral infec4on.To evaluate whether EBOV replica4on depends on decapping components other than EDC4, we studied localiza4on of DCP1A and DCP2 during infec4on and tested the impact of knockdowns of DCP2 and accessory decapping protein EDC3 on infec4on efficiency.Like EDC4, a por4on of DCP1A and DCP2 was found in brightly stained puncta associated with VP35 stained inclusion bodies (Figure 6a-d, supplemental movies 2-3).Since DCP1A and DCP2 do not appear to directly bind VP35 (Figure 3a), this suggests that these components are being recruited through EDC4.Next, DCP2 and EDC3 were depleted with siRNA (Figure 6e) and challenged with virus.DCP2 deple4on resulted in a fourteen-fold reduc4on of EBOV replica4on while EDC3 deple4on resulted in a 2fold reduc4on of EBOV replica4on (Figure 6f-g).Overall, our data suggests a model in which EBOV co-opts the intact decapping complex through VP35 binding to the decapping scaffold protein EDC4.Though we observed associa4on of DCP1A and VP35, we were unable to reduce expression of DCP1A or its isoform DCP1B and excluded these proteins from further analysis.

Discussion:
Here we present a comprehensive protein proximity screen using all EBOV proteins except GP and define a way to inter-relate the virus-host interactome in the context of known host-host protein interac4ons to reveal host protein complexes.Annota4ng virus protein interactors onto the host protein-mapped hits, we iden4fied members of poten4al host protein complexes associated with specific cellular func4ons, interac4ng with common viral proteins.Out of this analysis, we iden4fied proteins that make up the mRNA decapping complex as important factors in EBOV replica4on and demonstrated that VP35 likely recruits this complex by binding to one of its members, the scaffold protein EDC4.We demonstrated that VP35 interacts with EDC4 in the same region as the enzyme DCP2, and showed that deple4on of EDC4, DCP2, or EDC3 inhibited viral replica4on in a similar manner.In support of a direct role in viral replica4on, EDC4, DCP1A, and DCP2 associated with VP35-containing inclusion bodies in infected cells.
Other groups have reported interac4on of isolated members of the decapping complex with filovirus proteins (Supplemental figure 3) but its significance was not further evaluated.Using a split-TurboID system in which an ac4ve bio4n ligase was formed upon interac4on of L and VP35, which together make a func4onal polymerase complex, Fang et al. found EDC3 in the top hits with or without VP30 while EDC4 and DCP1A were seen as secondary hits 16 .These hits were not func4onally connected and other members of the complex were not iden4fied.This may reflect the deeper coverage seen in our screen.In two independent immunoprecipita4on-based screens using single virus proteins as bait, VP35 pulled down EDC4 10,34 (Supplemental figure 3).Again, other members of the decapping complex were not seen as partners with VP35 and the role of the interac4ons with EDC4 were not revealed.In our work, the role of the complex in Ebola virus replica4on was immediately suggested by a connected set of func4onally related proteins associated by a common virus protein, VP35.This was supported by experimental verifica4on that loss of each protein in cells gave a similar impact on viral replica4on.Furthermore, the proteins are seen with similar staining distribu4ons in cells, being in puncta close to inclusion bodies containing VP35.Taken together, these data support a direct role for the en4re decapping complex, not just EDC4, in regula4on of viral infec4on.al 37 .b BioID hits were compared to gene,c deple,on screens performed for EBOV (Flint et.al 38 , Mar,n et.al 39 ).The EBOV entry screen performed by Care<e et.al 40 was not included as it iden,fied hits related to EBOV GP, which was not included in our BioID analysis.
Although the mRNA decapping complex has been implicated in the replica4on of other viruses 41,42 , our work is the first to implicate the complex in filovirus replica4on and one of the few examples linking the complex to non-segmented nega4ve sense viruses.Here we show that for EBOV, the decapping complex is required for efficient virus replica4on.For other viruses, components of the decapping complex exert an4viral effects.Rik Valley Fever Virus competes with host mRNA decapping enzyme DCP2 for the cellular pool of 5' methylated caps as part of its cap-snatching func4on 43 .XRN1, which binds EDC4 within the same region as DCP2 and degrades decapped mRNAs 31 , has been implicated as both a proviral and an4viral factor for DNA and posi4ve sense RNA viruses respec4vely 44,45 .In measles virus, a nega4ve-sense virus, XRN1 promotes infec4on and in its absence, viral dsRNA accumulates in virus inclusion bodies 46 .Although XRN1 was labeled by VP35 in our BioID screen, our network modeling mapped it to a cluster of proteins containing UPF1 and other nonsense-mediated decay factors and so was not evaluated.The posi4oning of XRN1 in the network was likely due to the stringent parameters adopted for the network analysis, discussed below, and does not mean that it is absent from the decapping complex during infec4on.Further work is needed to determine whether XRN1 promotes EBOV infec4on through the decapping complex-VP35 interac4on or through another RNA processing pathway.
Members of the decapping complex have been associated with processing-bodies (P-bodies).These dynamic cytoplasmic structures are composed of protein and RNA and may be involved in RNA processing and/or mRNA sequestra4on.P-bodies have been previously reported to fluctuate during replica4on of other viruses, such as SARS-CoV-2 47 .In HeLa cells, markers of P-bodies such as DDX6 and EDC4 normally formed up to six large puncta in the cell cytoplasm.However, later in infec4on, when VP35 demonstrated more diffuse staining in the cytoplasm, these puncta disappeared and were replaced with smaller, more numerous puncta.Though the role of P-bodies in mRNA decapping is debated 48 , the C-terminal region of EDC4 that interacts with VP35 is also responsible for targe4ng EDC4 to P-bodies and P-body assembly 32 .Several other major components of P-bodies associated with decapping complex members were iden4fied in our VP35 BioID screen in the network model, including DDX6, LSM14A, and LSM14B 48 .EDC3 interacts with the decapping complex by binding DCP1A and DCP1B in P-bodies 29,48 .While its role as an enhancer of mRNA decapping in yeast has been well characterized 48 , its role in human mRNA decapping is understudied.Whether its labeling in our screen is due to complex interac4ons between DCP1A and EDC4 or is an independent interac4on of VP35 requires further study.
The approach of combining proximity-dependent tagging with PPI-driven network modeling of the virushost interac4ons should be readily applied to other viruses and cell types.Proximity tagging through bio4nyla4on has advantages over pull down mass-spectrometry approaches as it can capture labile and transient complexes.However, once proteins are eluted for analysis, informa4on on protein partners forming complexes is generally lost.Addi4onally, the extended set of interac4ons requires further down selec4on before star4ng any labor-intensive mechanis4c evalua4on.We ra4onalized that incorpora4on of known human protein-protein interac4ons into this virus-host interactome might allow recovery of protein groupings.When network tools such as STRING 20 were used to connect hits an extremely dense and complex network was produced that provided no strong guidance for follow up (Supplemental figure 1).Edge-density clustering methods such as MCODE 49 , which predict complexes according to connec4on density, was applied to the STRING network and iden4fied some puta4ve protein complexes, but failed to incorporate many proteins that are members of well-known complexes present in screen hits (data not shown).Instead, the Prize Collec4ng Steiner Forest network algorithm, which seeks direct paths between hit nodes and takes into account confidence of associa4ons from the underlying network, gave a more limited network without significant loss of mapped protein hits.We also chose to use the Human Integrated Protein-Protein Interac4on rEference (HIPPIE), a well curated PPI database based on experimental evidence present in the literature 22 .The combina4on of both provided a greatly simplified, high stringency network that maintained cellular pathway flows.The extent of the produced network can be adjusted by tuning the stringency parameters.A balance that yielded a network of 313 members was achieved by maintaining strict edge-confidence and degree parameters in conjunc4on with relaxed hit incorpora4on.The addi4on of an annotated informa4on layer of virus protein interac4ons and gene-set enrichment for clusters provided further insight into impact on cell pathway func4on and directed us to evaluate the decapping complex and its interac4on with VP35.In addi4on, other mul4-protein complexes were seen, such as those involved in nuclear-cytoplasmic transport with VP24 or microtubule polymeriza4on with NP, that will be studied in future work.
Ul4mately, our approach of screening using BioID to label neighboring proteins followed by confidence mapped network analysis helps to model the consequences of iden4fied virus-host interac4ons in infec4on and enables priori4za4on of complexes for func4onal studies.Our work emphasizes that while most host-interac4on studies readily iden4fy 1:1 interac4ons it is important and helpful to iden4fy interac4ons with mul4ple host proteins to provide mechanis4c insight into how the virus is manipula4ng the host for its replica4on.This work will further help to iden4fy major points of cellular control of EBOV infec4on, and beeer inform therapeu4c development for this devasta4ng virus and other virus types.

Figure 1 :
Figure 1: A protein proximity screen for Ebola virus proteins iden;fies an expanded viral interactome.a Top schema,c shows arrangement of EBOV structural proteins in virion and encoding genes in the virus genome.Bo<om shows constructs used in screening with placement of the BioID2 ligase indicated as green blocks.b Expression of viral protein-BioID2 fusion constructs and GFP-BioID2 non-specific control from stable cell line lysates.Blots were probed with bio,n ligase specific an,body and loading levels controlled by detec,on of β-ac,n.c Blot of bio,nyla,on levels, measured in cell lysates aKer inducing expression of each indicated protein by tetracycline and then adding bio,n to cell medium to measure ligase ac,vity.GFP-BioID2 was used as non-specific bio,nyla,on control (second to last lane) as well as endogenous bio,nyla,on in Flp-In TREx 293 cells not expressing any BioID2 construct (last lane).One representa,ve experimental replicate of each construct is shown.d Hub and spoke network model showing iden,fied interac,ons from screening.447 host-viral protein interac,ons were iden,fied between both N-and C-terminal constructs.

Figure 2 :
Figure 2: Network analysis reveals virus interac;ons with puta;ve host protein complexes.PCSF was used to inter-relateBioID hits with the HIPPIE human protein-protein interac,on (PPI) network.a The algorithm seeks to connect hits via the shortest pathways defined by the PPI network and considers confidence of underlying interac,ons and the presence of other hits along the mapped path.Hits are represented by nodes (green circles), where node size corresponds to the BioID hit score.At leK, two paths are iden,fied to connect strongly scored hits 1 and 4. The red path is the shortest to make the connec,on but requires inclusion of two host proteins that were not hits (Steiner nodes, triangles).The blue route is longer but has higher confidence, being be<er supported by literature cita,ons, incorporates other hits from the screen and only one Steiner node.The algorithm will have preference for incorpora,ng the blue route into the final network solu,on at right.b Overview of complete network mapping and gene set enrichment process.(LeK) indicates how virus protein interac,ons from BioID screening were rescored as a por,on of total SAINT scores for an associated host protein interac,on.If a host protein was associated with mul,ple virus proteins it was labeled by a segmented circle using colors indicated in c (Right) to perform func,onal enrichment, subnetwork clusters were segregated by edge-betweenness measurements (red and purple dashed circles indicate clusters).Gene set enrichment was then performed for each cluster (red and purple circles).c Full network generated by PCSF for iden,fied BioID2 screening hits related by the HIPPIE PPI.Five proteins of the mRNA decapping complex, each bound by VP35, are indicated in the inset.d Gene ontology annotated network obtained by

Figure 3 :
Figure 3: VP35 interacts with the decapping complex through EDC4. a Pulldowns of HA-tagged mRNA decapping proteins using an,-HA beads were performed in the presence of Flag-tagged VP35.Under the condi,ons, VP35 only coprecipitated with EDC4 (lane 5).b Confirma,on of the VP35-EDC4 interac,on.Pulldowns were performed with Flag-tagged EDC4 using an,-Flag beads (top panel) or HA-tagged VP35 using an,-HA beads (bo<om panel).c-d EDC4 proximally localizes to VP35 inclusions during infec,on.Cells were challenged with EBOV at an MOI of 1 and fixed c 10 hpi and d 20 hpi.EDC4 puncta were iden,fied colocalizing with VP35 at the edges of viral inclusion bodies (triangles).Scale bar = 25 μM.e Popula,ons of EDC4 and VP35 colocalize during infec,on.Cells were challenged with EBOV and fixed 20hpi.Proximity liga,on analysis was performed with an,bodies directed towards EDC4 and VP35.Signal was observed in infected cells (leK) over uninfected cells (right).Scale bar = 25 uM.One representa,ve experimental replicate is shown.

Figure 4 :
Figure 4: EDC4 levels impact viral RNA produc;on.a siRNA deple,on of EDC4 protein levels measured by immunoblot probed with specific an,bodies.Lower panel shows quan,ta,on of a representa,ve replicate of three independent experiments measured by area under the curve and normalized to b-ac,n expression levels.b Viral replica,on was measured by RNAFISH staining using mul,ple oligonucleo,des complementary to EBOV NP and VP35 mRNA.Scale bar = 100 μM.c Quan,fica,on of 10 images for which panel b is representa,ve.P-values ** <0.01; *** <0.005.d EDC4 deple,on results in reduc,on of viral mRNA and genomic RNA synthesis.Each indicated set of RNA transcripts were quan,fied via qPCR.An,genomic RNA indicates posi,ve sense RNA and genomic RNA is nega,ve sense RNA.Where applicable, sta,s,cal significance was calculated by ANOVA.P-values were * <0.05; ** <0.01; **** < 0.0001.One representa,ve experimental replicate is shown per qPCR.

Figure 5 :
Figure 5: EBOV VP35 interacts with the C-terminal domain of EDC4. a Map of EDC4, showing binding sites for DPC1A and DCP2.Lines below the map show constructs used in co-IPs (above dashed line) and overexpression (below dashed line) experiments.b HA-tagged EDC4 constructs were co-expressed with FLAG-tagged VP35.VP35 did not co-precipitate with EDC4 when the C-terminal domain comprising amino acids 974-1401 was absent (lane 3).c EBOV infec,on was measured aKer overexpression of the indicated domain of EDC4 as determined by RNAFISH of viral mRNA (magenta) and cell nuclei stained with Hoechst 33342 (blue).*P-value >0.05; ****P-value> 0.001.d Transfec,on efficiency of EDC4 constructs as determined by HA staining (green).Scale bars = 100μM.e EDC4 puncta associated with DDX6.Cells were challenged with EBOV and fixed 48 hpi, then stained with an,bodies against VP35 (grey) to label infected cells, EDC4 (red), and DDX6 (green).Both EDC4 and DDX6 are markers of p-bodies.Cell boundaries are indicated by dashed line.Scale bar = 25μM.

Figure 6 :
Figure 6: EBOV replica;on depends on mul;ple components of the decapping complex.a-b DCP1A and DCP2 colocalize with VP35 during infec,on.Cells were challenged with EBOV at an MOI of 1 and fixed at a 10 hpi and b 20 hpi.Proximal localiza,on of DCP1A to VP35 inclusion bodies was observed (triangles).Scale bar = 25μM c-d DCP2 colocalizes with VP35 during infec,on.Cells were challenged with EBOV and fixed at c 10 hpi and d 20 hpi.Colocaliza,on of DCP2 to VP35 inclusion bodies was observed (triangles).Scale bar = 25μM e Deple,on of DCP2 and EDC3 with siRNA.f Impact of deple,on of DCP2 or EDC3 on EBOV infec,on as quan,tated through RNAFISH of viral mRNA.Samples were infected at an MOI of 0.8.Scale bar = 250 μM.g Quan,fica,on of f *P-value > 0.05; **P-value > 0.01; ****P-value > 0.0001.

Supplemental Figure 3 :
Comparison of BioID2 screen with other EBOV PPI and gene,c screens.a BioID hits for indicated virus proteins in this study (colored circles) were compared to other published EBOV PPI screens for those indicated proteins, including mass spectrometry of purified virus par,cles (Spurgers et.al 35 ), affinity-pulldown mass spectrometry using NP, VP35, VP24, VP40 and VP30 as bait proteins to iden,fy interac,ons (Garcia-Dorival et.al 13,14 , Chen et.al 36 , Pichlmair et.al 34 , Batra et al 10 ), and other BioID screens performed using VP40, VP35, and L to iden,fy proximal interactors (Fan et.al 15 , Fang et.al 16 ).No significant overlap in hits was iden,fied with the NP BioID screen performed by Morwitzer et.