Yellow Fever Virus Interactomes Reveal Common and Divergent Strategies of Replication and Evolution for Mosquito-borne Flaviviruses

Abstract

Pathogenic mosquito-borne flaviviruses infect mosquito and human hosts, relying on host protein interactions to replicate, evade immunity, and mediate pathogenesis. Prior proteomic studies mapped such interactions for some flaviviruses, but yellow fever virus (YFV)-a pathogen of resurgent concern-remains understudied. Here, we map YFV interactomes in human and mosquito cells to identify interactions common among divergent flaviviruses or unique to YFV. Functional assays reveal a previously unrecognized YFV restriction factor: RBBP6 inhibits YFV genome replication by interacting with the viral polymerase NS5. We enhance the identification of dual-host interactions using structural modeling and holistic network integration. Extending our holistic approach to other flavivirus interactomes, we distinguish conserved mechanisms of host targeting from those unique to YFV. Contrary to expectations that conserved viral proteins lead to conserved protein interactions, we find that Capsid, a divergent structural protein, shares more host interactions than NS5, a conserved enzyme. Integrating proteomics with complementary analyses defines new principles of host-targeting strategies across flavivirus and host evolution, offering a versatile resource for navigating the complex landscape of flavivirus biology.

Keywords: flaviviruses; integrative proteomics; systems biology; virus-host evolution; virus-host interaction; yellow fever virus.

Figure 1.. YFV-human proteomics identifies virus-host PPIs related to flavivirus biology.

(A) Schematic representation of experimental AP-MS pipeline and proteomic scoring. (B) Virus-host interactions were scored with both MiST and SAINT to produce the plotted number of HC-PPIs per bait. (C-D) Enrichment analysis was performed on the set of HC-PPIs for each viral bait. The heatmaps summarize the degree of enrichment (adjusted p-value < 0.05) for (C) GO: Biological Processes and Reactome pathways relevant to flavivirus replication, and (D) GO:cellular compartment. The full enrichment results are available in Table S1. Viral baits are ordered according to their position in the viral polyprotein.

Figure 2.. YFV-human PPI network.

YFV-Human HC-PPIs were identified after a tiered scoring approach using both MiST (M > 0.67 | M > 0.60 & protein complex member) and SAINT (score > 0.95 & BFDR ≤ 0.05). Viral baits (grey rectangles) are mapped to human proteins (light yellow ovals). Human proteins are annotated as host-dependency factors from a contemporary YFV CRISPR screen (gold with bolded borders). Additionally, human proteins were grouped by GO:BP (grey underlay) and CORUM protein complex (black outline). Full scoring results are available in Table S1. Ribosomal proteins (42) were not included for visual clarity.

Figure 3.. RBBP6 inhibits YFV genome replication through an RNA-dependent interaction with full-length NS5.

(A) Violin plot showing the percent knockdown efficiency of all tested gRNAs in the CRISPRi screen. gRNAs achieving achieved >50% knockdown are shown in yellow. RBBP6-targeting gRNAs are shown in bronze. (B) Robust Z-score distribution of viral titers from the CRISPRi screen for gRNAs achieving >50% knockdown (panel A, yellow), identifying RBBP6 as a top restriction factor (bronze). Full results are available in Table S2. (C) YFV-17D replication in Huh7 cells with RBBP6 knockdown. Plaque assay quantification of viral titers at 48 hours post-infection (hpi) across a range of multiplicities of infection (MOI = 0.01, 0.10, 1.00). Data represent mean ± SD of three biological replicates, with p < 0.005 (***), p < 0.01 (**), and p < 0.05 (*) by one-sided t-test. (D) Confocal immunofluorescence microscopy of RBBP6 and YFV NS5 in transfected HeLa cells. 2xStrep-tag for NS5 and HA-tag for RBBP6 (yellow). Hoechst staining for nucleus (cyan). Scale bars: 20 μm (main panels), 10 μm (zoom). (E) Western blot for co-IP of NS5 with endogenous RBBP6 or IgG in YFV-infected cells with/without RNaseA treatment. GAPDH is a loading control. (F) Schematic of RBBP6 domain architecture and truncation constructs. (G) Western blot for whole cell lysate and Strep-AP of YFV 2xStrep-tagged NS5 co-transfected with HA-tagged RBBP6 constructs. EV represents an empty vector for negative control. (H) Western blot for whole cell lysate and Strep-AP of YFV 2xStrep-tagged NS5 domain truncations. NS3 and EV are negative controls. (I) Renilla luciferase activity of YFV replicon in RBBP6 and control CRISPRi knockdown Huh7 cells. Data represent mean ± SD of three biological replicates, with p < 0.005 (***), p < 0.01 (**), and p < 0.05 (*) by one-sided t-test.

Figure 4.. Predicted structures and holistic analysis facilitate interolog identification.

(A) Number of HC-PPIs identified per viral bait from mosquito proteomics. (B) Network of structure- and evolutionary-based interologs for both YFV and DENV. The node label is the human gene symbol. The mosquito UniProt ID is listed under the node. Enriched GO:BP categories and CORUM complexes are based on human annotations. Bolded gene IDs indicate proteins found to have positive selection. (C) Hierarchical clustering of flavivirus-mosquito and flavivirus-human PPIs based on MiST scores. Interactions were only considered for viral baits consistently studied across all datasets. An interaction only included if it was a HC-PPI in at least one study and was evaluated for number of studies in which it was considered HC-PPI (cutoffs met shown in blue scale bar). K-means clustering optimized the number of clusters to select a hierarchal cutoff to define clusters (red dashed line). (D) Enrichment network for GO:BP and CORUM complexes from YFV and DENV pan-host cluster, representing interactions conserved across all interactomes analyzed. The color indicates the viral bait associated with the enrichment term and pie sector represents the fractional magnitude of the log₁₀p-value of that enrichment. A monochromatic point indicates it was only significant (p < 0.05) for a single viral bait. (E) Box and whisker plots showing the sequence identity values of human-A. aegypti homolog pairs of indicated groups of proteins. Solid horizontal lines indicate the median. Dashed horizontal lines indicate the mean. Adjusted p-values were calculated using a Wilcoxon rank-sum test with Bonferroni correction: adjusted p < 0.0001 (****), adjusted p < 0.001 (***), adjusted p < 0.01 (**), adjusted p < 0.05 (*), adjusted p > 0.05 (n.s.).

Figure 5.. Holistic integration of flavivirus-host interactomes reveals pan-flavivirus and divergent mechanisms of host targeting.

(A) Approximate evolutionary relationship between DENV, ZIKV, WNV, and YFV. TBEV was included to represent a true outgroup compared to mosquito-borne flaviviruses. (B) Hierarchical clustering of flavivirus-human PPIs based on MiST scores. Interactions were only considered for viral baits consistently studied across all viruses. An interaction was only included if it was a HC-PPI in at least one study and was evaluated for the number of studies in which it was considered HC-PPI (cutoffs met shown in yellow scale bar). K-means clustering optimized the number of clusters to select a hierarchal cutoff to define clusters (red dashed line). (C) Enrichment network for GO:BP and CORUM complexes from cluster 3, representing pan-flavivirus interactions conserved across all interactomes analyzed. The color indicates the viral bait associated with the enrichment term, and the pie sector represents the fractional magnitude of the log₁₀p-value of that enrichment. A monochromatic point indicates it was only significant (p < 0.05) for a single viral bait. (D) Schematic of differential interaction score (DIS) calculation. (E) Plot of YFV DIS distributions, separated by the viral bait. Dot colors indicate the virus used for DIS comparison. (F) Heatmap showing the degree of enrichment (log₁₀p-value, yellow scale bar) for unique terms in clusters associated with YFV specifically (cluster 1) or excluding YFV (cluster 8), grouped by viral bait. Enrichment terms were excluded if they showed significant enrichment for other clusters. Other clusters are included in the heatmap to demonstrate the enrichment is unique to clusters 1 or 8. In each tile, the number of interactions associated with that enrichment term is bolded if it was significant (p < 0.05).

Figure 6.. Integration of virus sequence and structure reveals a complex relationship with PPI conservation

(A) Schematic of PPI, sequence, and structural analysis. PPIs were analyzed according to viral baits across all four flavivirus-host PPI datasets. (B) Pearson’s correlation between MiST score profiles and viral protein sequence conservation. For each viral bait, six pairwise correlations were calculated for MiST profiles for each possible pairwise combination between four viruses. The same was done for pairwise sequence identity. Pearson’s correlations between MiST correlations and sequence identity were then calculated for each viral bait and the entire dataset. (C) Pearson’s correlation between relative HC-PPI conservation and viral protein sequence conservation. For each viral bait, mean values of relative HC-PPI conservation and pairwise sequence identity were plotted. (D-E) Visualization of YFV NS5 and Capsid protein structures with protein sequence conservation. Residues with known enzymatic activity are colored with a yellow overlay. Structures were generated using the YFV-Asibi protein sequence and AlphaFold2. (F) Violin plot of solvent accessible surface area (SASA) binned by amino acid conservation for NS5 and Capsid. Yellow dots indicate residues with known enzyme activity. Solid horizontal lines indicate the median. Dashed horizontal lines indicate the mean. Adjusted p-values were calculated using a Wilcoxon rank-sum test with Bonferroni correction: adjusted p < 0.0001 (****), adjusted p < 0.001 (***), adjusted p < 0.01 (**), adjusted p < 0.05 (*), adjusted p > 0.05 (n.s). (G) Stacked bar plot of the distribution of amino acid identity. Only NS5 amino acids with similar SASA as Capsid were considered. P-value was calculated using a Wilcoxon rank-sum (Mann-Whitney U) test: p > 0.05 (n.s).