Coding sequences for NS3 and NS5 Flaviviruses full-length proteins or NS3 helicase, NS3 protease, NS5 polymerase and NS5 methyltransferase functional domains were provided in pDONR207 entry vector (Gateway, Invitrogen) by Bruno Coutard (Architecture et Fonction des Macromolécules Biologiques, UMR6098, Marseille) and referenced in ViralORFeome database 17 . The viral ORFs were isolated from the following viruses: dengue virus serotype 1 (strain D1/H/IMTSSA/98/606), Alkhurma virus (strain 1176), West Nile virus (Strain paAn001), Japanese Encephalitis virus (strain Beijing1), Kunjin virus (MRM61C) and Tick borne encephalitis virus (strain 263). Cellular ORF coding for AZI2 was purchased from Invitrogen (clone IOH41551) and coding sequences for NFKBIA, and TRAF4 were obtained from the Human ORF Collection (OHS4187, Open Biosystems). Viral and cellular coding sequences were subsequently transferred by in vitro recombination from pDONR207 into different Gateway-compatible destination vectors following manufacturer's recommendation (LR cloning reaction, Invitrogen). To perform yeast-two hybrid experiments, human prey coding sequences were recombined into pACT2 (Invitrogen) to be expressed in fusion downstream of the activation domain of Gal4 (Gal4-AD) and viral bait coding sequences into pGBKT7 to be expressed in fusion downstream of the DNA binding domain of Gal4 (Gal4-BD). In mammalian cells, GST-tag and 3xFLAG-tag fusions were achieved using pDEST27 (Invitrogen), or pCI-neo-3XFLAG (kindly provided by Y. Jacob Institut Pasteur) vectors, respectively.

Viral cDNAs cloned into bait Gal4-BD vector pGBKT7, were transformed into AH109 yeast strain (Clontech) and used to screen by mating human cDNA libraries from liver, brain, spleen and bronchial epithelia cloned in the GAL4-AD pACT2 vectors, and transformed into prey Y187 yeast strains. The mating between baits and prey yeast cells was performed on a selective medium lacking histidine and supplemented with 10 mM 3-amino-triazole (3-AT; Sigma-Aldrich). After 6 days of culture on selective medium, [His+] diploids colonies were isolated and further selected over 3 weeks by culture on selective medium to eliminate false-positives colonies. After selection, yeast colonies were treated with zymolyase in order to digest their cell walls, and AD-cDNAs were amplified by PCR using primers that hybridize within the pACT2 regions flanking cDNA inserts (Fwd: gacggaccaaactgcgtataacg, Rev: ccaaacctctggcgaagaagtcc). PCR products were sequenced (GATC Biotech) and cellular interactors were identified by BLAST analysis as previously described 18 .

Interactions retrieved from literature, describing binary interactions between cellular and flavivirus proteins, were extracted from VirHostNet knowledge base 19 after PubMed extensive curation. Briefly, VirHostNet is an up to date knowledge base for the management and the analysis of proteome-wide virus-host interaction networks available at http://pbildb1.univ-lyon1.fr/virhostnet. A total of 16 protein-protein interactions were retrieved and added to our experimental data set.

The overall statistical significance of the interconnectivity (number of protein-protein interactions) between flaviviruses interactors was assessed by a random resampling testing procedure (n = 10, 000 permutations). For each permutation, we randomly extracted as many proteins as the number of flaviviruses interactors from the human interactome, and the value of interconnectivity was assessed. The randomization procedure was weighted and corrected according to the connectivity of proteins in order to prevent inspections bias on highly studied proteins. A theoretical distribution was computed for the 10, 000 resampled values. From this distribution, an empirical p-value for the random resampling test was computed by counting the number of resampled values greater than the observed value.

Molecular functions, biological processes and cellular components from Gene Ontology (GO) database 20 were used to annotate the human proteins targeted by the flaviviruses. Briefly, for each GO term, we determine if the set of annotated proteins interacting with the flavivirus proteins is significantly enriched in comparison with the set of proteins annotated with this term within the whole proteome. For each GO term, the enrichment analysis was performed by using an exact Fisher test (p-value < 0.05) followed by the Benjamini and Yekutieli multiple test correction 21 . The analysis was conducted with the web-based software GOEAST 22

Alignments were performed with the tool « Align » from EMBOSS http://www.ebi.ac.uk/Tools/emboss/align/.

Human HEK-293 null cells were maintained in growth medium consisting of Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% heat-inactivated fetal bovine serum (FBS), 100 U/ml penicillin G, 100 μg/ml streptomycin, at 37°C under 5% CO2.

Pull downs and cell lysates (15 μg of protein extracts) were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis on 4-12% NuPAGE Bis-Tris gels with MOPS running buffer (SDS-PAGE) (Invitrogen) and transferred to nitrocellulose membrane (I-Blot, Invitrogen). 3XFlag- and GST-tagged proteins were detected with a mouse monoclonal peroxidase-conjugated anti-FLAG M2 antibody (A8592, Sigma) and a rabbit polyclonal anti-peroxidase-conjugated anti-GST antibody (A7340, Sigma) and revealed with ECL detection reagent (pico West, Amersham).

To unravel new protein-protein interactions between flavivirus and human proteins, we sub-cloned sequences encoding NS3 and NS5 flaviviruses proteins into yeast-two-hybrid (Y2H) vectors. All available viral proteins listed in additional file 1 were expressed in yeast either as full-length proteins or as functional domains, namely NS3 helicase, NS3 protease, NS5 polymerase and NS5 methyltransferase. They were then used as viral baits against human cDNA libraries. Viral ORFs coding for NS3 and NS5 proteins were isolated from distinct human pathogens belonging to major flavivirus evolutionary lineages: (i) aedes-borne pathogen: DENV; (ii) culex-borne pathogens: WNV (including the Kunjin Australian variant (KUNV)) and JEV; (iii) tick-borne pathogens: Tick-borne encephalitis (TBEV) and Alkhurma (ALKV) viruses. Protein sequence comparison study revealed that the functional enzymatic domains of NS3 are highly conserved amongst these viruses (Additional file 2).

At least three independent screenings against human cDNA libraries were performed for each viral bait. Eighty-five percent of the identified cellular targets of each bait were then tested pairwise against all the viral proteins baits including the original bait using an array-based Y2H strategy which confirmed 90% of the interactions identified in the initial screens. Furthermore, the bait panel versus selected targets strategy used in the array cross experiment enabled us to identify 69 additional, novel virus-host interactions not detected in the first screen. Repetition and confirmation of our Y2H experiment by the array strategy allowed us to be very stringent in obtaining a high quality set of 108 human proteins that interacted with one or more of the viral protein baits (Additional file 3). In one of our previously published studies using the same Y2H screening settings, the validation rate obtained by co-affinity purification reached 85% 12 . We conducted GST-pull down assays to further validate our Y2H data (Additional file 4). An extensive literature curation allowed us to finally complete our set of data by 16 previously published interactions, 15 of which not identified by our screen (Additional file 3).

Based on our high-throughput Y2H screen and literature search, we created the flavivirus NS3 and NS5 proteins interaction network composed of 186 interactions involving 120 distinct human proteins, 108 from our screen and 13 from the literature (Table 1, Figure 1, additional files 3 and 5). We emphasize that among the 186 interactions, 171 were obtained from our Y2H screen and only 16 from previously published work. Despite the conserved amino acid patterns within the different viral ORFs that we used as viral baits, only one third of the cellular targeted proteins identified in our study interacted with two or more flaviviruses (Table 2). Moreover, only five cellular proteins (CAMTA2, CEP250, SSB, ENO1, and FAM184A) were found to interact with both NS3 and NS5 proteins (Figure 1, additional file 5).

To further describe the topological properties of the flavivirus interaction network in relation to the whole human interactome, we then took advantage of the VirHostNet knowledgebase which includes an extensive assembly of human-human and viral-human interactions 19 . We thus calculated the local (degree) and global (betweenness) centrality measures of the human proteins targeted by NS3, NS5 or both flavivirus proteins integrated into the human interactome (Table 3). Briefly, the degree of a protein in a network refers to its number of direct partners and is therefore a measure of local centrality. Betweenness is a global measure of centrality, as it measures the number of shortest paths (the minimum distance between two proteins in the network) that cross a given protein. The 120 identified human proteins interacting with NS3 and NS5 were shown to have a higher average degree i.e. local connectivity (22, 93 versus 10, 43) and betweenness i.e. global centrality (4, 02.10^-4 versus 1, 30.10^-4) in comparison with the human proteins belonging to the human interactome (Table 3). In addition, the degree and the betweenness distributions of human proteins interacting with NS3 and NS5 are significantly distinct from the proteins belonging to the human interactome distributions (U-test, all p-values < 10^-12, additional file 6). This indicates that NS3 and NS5 have a strong tendency to interact with proteins that are highly connected and central within the human interactome. This latest observation is in accordance with previous virus-host interactome features 11 12 23 . Furthermore, we found that a total of 47 cellular proteins (39%) out of 120 are cellular targets for other viruses as well, including HIV, herpes, hepatitis C and papilloma viruses (Additional file 7, exact Fisher test, p-value = 1, 2.10^-12). This observation reinforces our findings since different viruses, and possibly other pathogens, are expected to interact with common cellular targets as a consequence of possible common strategies adopted by viruses for infection and replication 23 .

We then performed an enrichment analysis using Gene Ontology (GO) database on the 120 proteins targeted by the flaviviruses in order to characterize the cellular functions significantly over-represented in the pool of proteins interacting with the flavivirus NS3 and NS5 proteins. Briefly, each cellular protein identified in our analysis and listed in the GO database was ascribed with its GO features. For each annotation term, a statistical analysis evaluated a putative significant over-representation of this term in our list of proteins compared to the complete list of the human annotated proteins. The most significantly over-represented GO annotation terms are listed in Table 4. It is noteworthy that among the enriched functions identified, some are associated with already known function of NS3 and NS5 viral proteins namely RNA binding and viral reproduction (Table 4, molecular function). One may thus put forward the hypothesis that among the cellular proteins listed for these two particular processes some might be key cellular partners for the viral life cycle. We also identified structural components of the cytoskeleton as cellular partners of NS3 and NS5 and we will discuss their putative implication in the viral infectious cycle thereafter in the discussion (Table 4, cellular component). Finally, our analysis revealed that the flaviviruses interact with cellular proteins involved in the Golgi vesicle transport and in the nuclear transport, suggesting that the NS3 and NS5 proteins might be able to interfere with these two cellular functions (Table 4, biological process). In addition, it is important to underline that both regulators of type I interferon-mediated signaling pathway and of innate immune response were found to be significantly enriched in the statistical analysis as well. We will discuss the implication of the functional enrichment profile of the cellular proteins identified in our screen and how these proteins affect the virus replication and assembly.

Only 1/3 of the cellular proteins are represented in the human-human protein-protein interactome, suggesting that most of the cellular proteins are not connected 19 . We observed that the human proteins targeted by the flavivirus NS3 and NS5 were highly overrepresented in the human interactome (108/120, exact Fisher test, p-value < 2, 2.10^-16). This implies that most of the cellular proteins targeted by the flaviviruses are connected with other human proteins. An analysis of the level of interconnectivity of the 108 proteins revealed that they are indeed highly connected to each other (84 protein-protein interactions), and that this interconnectivity is highly significant compared to the theoretical interconnectivity computed from resampled networks (resampling test, n = 10, 000, p-value < 10^-4, additional file 8). All together these results, in accordance with our functional enrichment analysis, emphasized the fact that the flaviviruses are targeting closely related cellular proteins, which are likely to share common functional features.

Figure 2 represents the sub-network of all the cellular proteins connected into the human protein-protein network and targeted by the flavivirus replication complex NS3 or NS5 proteins. These interacting proteins form a relatively compact connection web with a central core of 35 proteins, the majority of which has been shown to interact with other viruses (Figure 2 and additional file 7). Interestingly, among these central proteins, several are important components of the cytoskeleton. These include in particular VIM, MYH9, ACTB, ACTG1, LMNA and GOPC (Table 2). NS3 and NS5 are interacting with two smaller functional units: one is composed by 4 proteins belonging to the interferon signalling cascade (PRMT5, TYK2, STAT2 and IFNAR2) and the second one is made up by 3 molecules involved in vesicular transport (TSG101, GGA1 and TOM1L1).