Multiomics Analysis of Arboviral Capsid Targets in Mosquitoes Reveals a Proviral Function of the Chromatin-Remodeling Brahma Complex

TL;DR

This study explores how arboviruses interact with mosquito cells, revealing a new role for the Brahma complex in viral replication.

Contribution

The study identifies the chromatin-remodeling Brahma complex as a proviral host factor in arbovirus replication in mosquitoes.

Findings

A multiomic analysis identified 110 novel host proteins interacting with arboviral capsids.

The Brahma complex was found to support orthoflavivirus replication in mosquito cells.

Arboviral capsids were shown to modulate chromatin accessibility in mosquitoes.

Abstract

In recent years, arboviral infections have surged dramatically because of the geographic expansion of Aedes and Culex mosquitoes, their main vector mosquitoes. Despite significant efforts to uncover arbovirus–host interactions and viral protein effector functions in mammals, systematic studies aiming to characterize virus–vector interactions in arthropods are largely missing, and the functions and cellular targets of many arboviral proteins in mosquitoes remain elusive. Here, we applied a multiomic approach to systematically evaluate the ability of arboviral capsids to interact with the Aedes aegypti proteome. This extensive multimodal atlas across 11 pathogenic arboviral species spanning three viral genera revealed shared and distinct host factor specificities, uncovering species-, genus-, and vector preference–specific patterns of host usage in mosquitoes. Functional phenotypic screening of 110 newly discovered host proteins across three prototypic arboviruses (La Crosse virus, dengue virus, and West Nile virus) identified several novel host dependency factors, including a new role for the chromatin-remodeling Brahma complex in orthoflavivirus replication. Using a combination of biochemical and sequencing approaches, we characterized the cellular determinants of these interactions and profiled their functional consequences on the chromatin landscape. Altogether, this study provides a multilayered repository to categorize and characterize arboviral capsid effector functions in invertebrates, providing important cues on novel mechanisms of transcriptional regulation via capsid-mediated modulation of chromatin accessibility in insects.

•Virus-host interaction network of 11 arboviral capsids in mosquito cells.•Identification of novel host-dependency factors for DENV, LACV, and WNV.•Novel role of the chromatin-remodeling Brahma complex in orthoflavivirus replication.•Discovery of capsid-mediated transcriptional regulation in mosquito cells.

Virus-host interaction network of 11 arboviral capsids in mosquito cells.

Identification of novel host-dependency factors for DENV, LACV, and WNV.

Novel role of the chromatin-remodeling Brahma complex in orthoflavivirus replication.

Discovery of capsid-mediated transcriptional regulation in mosquito cells.

This study by Flory et al. characterizes virus–vector interactions in Aedes mosquito cells through a systematic proteomics-based survey of capsid interactors of 11 pathogenic arboviruses. In a dsRNA-based knockdown screen, the functional relevance of 110 capsid interactors was tested for three different arboviruses revealing several new host dependency factors. This study identified and characterized the novel mosquito host dependency factor, the chromatin-remodeling Brahma complex, and showed capsid-mediated transcriptional regulation activities using a multiomics sequencing approach.

Figures

Full text

Arthropod-borne viruses (arboviruses) are a prime reason for morbidity and mortality worldwide and cause a broad spectrum of diseases ranging from hemorrhagic to neurovirulent pathological outcomes in humans and other mammals. These viruses represent a further increasing public health burden, given their continuous expansion and ability to establish new endemic areas, and the complete lack of approved antivirals for therapeutic or prophylactic use (1, 2, 3). The main factor contributing to the global spread of arboviruses is the rapid expansion of one of their main mosquito vectors. Aedes spp., in the northern hemisphere (4, 5), accompanied by a rising number of autochthonous infections by orthoflaviviruses, alphaviruses, and orthobunyaviruses in Europe and North America, and public health concerns of preparedness for potential upcoming epidemics (6). As a result of the steady increase in global temperatures and further geographic expansion of arthropod vector species, an increase in epidemics driven by arboviral infections will be seen in the next years, with two-thirds of the global population currently at risk of infections (6).

Despite remarkable differences in genome organization and replication strategies among different arboviral families, viral capsid or nucleocapsid proteins emerged as highly versatile structural proteins harboring multiple host-modulating functions beyond their role in virus entry, viral RNA uncoating, and virion morphogenesis (7). For instance, the orthoflaviviral capsid protein (C) has been reported to engage in multiple protein–protein interactions (PPIs) with host proteins in mammals, thereby modulating cellular pathways involved in lipid metabolism (8, 9), protein homeostasis (10, 11), antiviral innate immunity (12, 13, 14), nuclear transport (15, 16, 17, 18, 19, 20, 21, 22, 23), and cell death (7, 24, 25, 26, 27, 28, 29). Furthermore, previous studies described the shuttling of different orthoflaviviral capsids to the nucleus of mammalian and mosquito cells, where they modulate processes involved in ribosomal stress and apoptosis (24, 25), transcriptional regulation (24, 30), and chromatin remodeling (30, 31), also in the absence of productive viral infection. Similarly, the capsid proteins of other arboviral species, including those of the genera Alphavirus (family Togaviridae) and Orthobunyavirus (family Peribunyaviridae), have been reported to play multifunctional roles beyond their functions in virion packaging (32, 33). For instance, ectopic expression of alphaviral capsids specifically regulates host transcriptional shutoff and promotes immune evasion in a viral RNA–independent manner (33, 34, 35, 36, 37, 38, 39, 40), whereas the poorly characterized orthobunyaviral nucleocapsid has been reported to modulate cellular transcription and translation (41, 42).

In the past decade, substantial efforts have been made to uncover arbovirus-specific mechanisms of host adaptation using genomic-, transcriptomic-, and proteomic-based approaches (43, 44, 45, 46, 47). These studies predominantly focused on the mammalian host, whereas systematic studies aiming to characterize virus-induced changes and virus–host interactions in arthropods are largely missing (45, 48, 49, 50). Currently, virus–host interactions in invertebrates have been characterized only for two orthoflaviviral species (51, 52); yet, surprisingly, little is known about the determinants of host susceptibility and vector competence across pathogenic arboviruses.

Here, we employed affinity purification (AP) coupled to mass spectrometry (MS) to systematically characterize the ability of 11 prototypic arboviral capsids spanning three virus genera to interact with the Aedes aegypti proteome. This extensive PPI atlas uncovered novel mosquito host targets targeted by individual arboviral species, identifying a completely new set of PPIs exhibiting subviral and pan-viral specificities within and across arboviral genera (Orthoflavivirus, Orthobunyavirus, and Alphavirus). Leveraging a custom-made phenotypic screening platform, we further characterized 110 capsid-interacting proteins, assessing their functional relevance for the production of three human pathogenic viruses in mosquito cells (dengue virus [DENV], West Nile virus [WNV], and La Crosse virus [LACV]). This approach identified several novel host dependency factors, uncovering a new proviral role for the chromatin-remodeling Brahma complex in orthoflavivirus replication. Using a combination of biochemical assays and high-throughput sequencing technologies (assay for transposase-accessible chromatin sequencing [ATAC-Seq] and RNA-Seq), we characterized the functional consequences of these interactions on the chromatin landscape and demonstrated that orthoflavivirus capsids selectively modulate gene expression in invertebrates.

Experimental Procedures

Cells and Viruses

Ae. aegypti–derived cells Aag2, Aag2-AF5, and the cell fusing agent virus–free Aag2-C3 were cultured at 28 °C, without CO_2_, in Leibovitz L-15 medium (Gibco) supplemented with 10% fetal bovine serum (FBS) (Capricorn), 100 U/ml penicillin, 100 μg/ml streptomycin (Sigma–Aldrich), and 10% tryptose phosphate broth (Sigma–Aldrich). African green monkey Vero E6 cells (American Type Culture Collection) were grown at 37 °C with 5% CO_2_, in high-glucose Dulbecco’s modified Eagle’s medium (Sigma–Aldrich) with 10% FBS, 100 U/ml penicillin, and 100 μg/ml streptomycin. Stocks of viruses used in this study (DENV serotype 2 [DENV2] strain UVE/DENV-2/2018/RE/47099, yellow fever virus [YFV] strain Asibi, Zika virus [ZIKV] strain H/PF/2013, WNV strain UVE/WNV/UNK/CF/Ar B 3573/82, Japanese encephalitis virus [JEV] strain UVE/JEV/2009/LA/CNS769, Usutu virus [USUV] strain BNI-507/2016/Germany, LACV [PMID: 16051834], Sindbis virus [SINV] isolate BNI-10865/2016/Germany, and Chikungunya virus [CHIKV] strain UVE/CHIKV/2014/HT/Haiti_19) were produced in VeroE6 cells and stored at −70 °C. Stock titers were determined by plaque assay as described below.

Antibodies

The following antibodies were used in this study (applications and dilutions indicated in parentheses): rabbit and mouse anti-FLAG DYKDDDK tag (both Proteintech) (immunofluorescence [IF]: 1:100 dilution), horseradish peroxidase (HRP)–conjugated mouse anti-FLAG (clone M2; Sigma–Aldrich) (Western blot [WB]: 1:1000 dilution), rabbit anti-WNV capsid (ProSci) (WB: 1:1000 dilution), mouse monoclonal anti-DENV capsid (16) (WB: undiluted unpurified supernatant), HRP-conjugated goat anti-mouse and anti-rabbit (both Sigma–Aldrich; WB: 1:10,000 dilution), HRP-conjugated mouse anti–beta-actin (clone C4; Santa Cruz; WB: 1:5000 dilution), mouse antihemagglutinin (HA) (Cell Signaling) (IF: 1:200 dilution, WB: 1:1000 dilution), mouse antiflavivirus group antigen found on envelope (E) protein (clone D1-4G2-4-15; American Type Culture Collection; purified in-house; IF: 1:50 dilution), and anti-mouse and anti-rabbit antibodies conjugated with Alexa Fluor 488, 568, or 647 (all Life Technologies; IF: 1:1000 dilution).

Plasmids

Amplicons containing the HA-tagged orthoflavivirus capsid sequences were PCR amplified with Phusion polymerase (New England Biolabs) from pUC19-based vectors containing N-terminally HA-tagged capsids (synthesized by BioCat GmbH) (primers are provided in Supplemental Table S1), whereas alpha- and orthobunyaviral sequences were obtained as synthetic DNA gBlocks (Integrated DNA Technologies). Nucleotide sequences were derived from the following strains (UniProt accession numbers in parenthesis): ZIKV Asian lineage H/PF/2013 (FSS13025, A0A142I5B9), ZIKV African lineage DakAr41524 (A0A481XTV0), JEV SA-14 (P19110), Powassan virus (POWV) LB (Q04538), tick-borne encephalitis virus (TBEV) Neudoerfl (P14336), USUV Vienna 2001 (Q5WPU5), WNV NY-99 (Q9Q6P4), YFV Asibi (Q6DV88), Mayaro virus (MAYV) TRVL15537 (A0A515HFL8), SINV AR-339 (A0A7U1GI35), LACV 10846-18 (Q38PK7), as well as hepatitis C virus (HCV) JFH1 (E7ELX2). Amplicons were ligated into the pPUb-ZeoR-2A2A-V5-enhanced GFP (eGFP) vector (51, 53) via the Avrll and BsrGI (New England Biolabs) sites, thereby replacing the V5-eGFP sequences, and transformed into top 10 chemically competent Escherichia coli. To generate pPUb vectors expressing mosquito host genes, the antibiotic resistance gene ZeoR from the pPUb-ZeoR-2A2A plasmid was exchanged with the puromycin resistance gene via the NcoI and EcoRI sites, using standard molecular biology methods. Subsequently, sequences containing the N-terminally FLAG-tagged Ae. aegypti host factor Moira (VectorBase identity [ID]: AAEL022608) or eGFP were introduced into the pPUb-PuroR-2A2A vector via the Avrll and FseI sites (New England Biolabs), and the resulting recombinants were transformed into Stbl3-competent E. coli. All plasmid DNA insert sequences were verified by Sanger sequencing.

Generation of Stable Mosquito Cell Lines

To generate Aag2 cell lines stably expressing different HA-tagged arboviral capsids, 10 μg of each pPUb-ZeoR-2A2A-HA-capsid plasmid was transfected into Aag2 cells using Lipofectamine 3000 reagent (Thermo Fisher Scientific) at a 1:2:1.5 ratio of DNA:P3000:Lipofectamine 3000. Zeocin selection was initiated at 48 h post transfection (hpt) using Leibovitz L-15 medium supplemented with 10% FBS, 100 U/ml penicillin, 100 μg/ml streptomycin, 10% tryptose phosphate broth, and 100 μg/ml zeocin (Gibco). After 1.5 to 2 months of continuous passaging under antibiotic selection, stable cells were amplified for MS sample preparation. Expression and subcellular distribution of HA-tagged capsids were validated by WB and IF assays, respectively.

Experimental Design and Statistical Rationale

All proteomics experiments were performed with four biological replicates. Only proteins that were identified with at least one peptide and in at least three quantitation events in at least one condition were considered in the subsequent data analysis. Statistical analysis to identify significant high-confidence interactors of different arboviral capsids was done with a false discovery rate (FDR)–corrected Welch’s t test (S0 = 1, FDR = 0.05, n = 250 randomizations, |log2[fold change]| ≥3.5; p ≤ 0.015; control bait: HCV core), whereas differentially regulated host proteins in the effectome were called by two-sided FDR-corrected Student’s t test (S0 = 1, FDR = 0.05, n = 250 randomizations, |log2[fold change]| ≥2; p ≤ 0.05; control bait: HCV core).

All high-throughput sequencing experiments (ATAC-Seq and RNA-Seq) were done with three biological replicates. Statistical analysis to define either differentially accessible (DA) peaks or differentially expressed genes was done by using the Wald's test statistic of DESeq2. Thereby, only peaks that occur in all three replicates were included in the subsequent analysis. To define DA peaks (ATAC-Seq), a cutoff of log2(fold change) >1.5 (dsMoira compared with dsRNA targeting eGFP [dseGFP]) or log2(fold change) >0.5 (all other comparisons) was applied. Differentially expressed genes (RNA-Seq) were called by an FDR (Benjamini–Hochberg–adjusted p values) <0.05 and with no minimum log2(fold change).

Other experiments assessing viral phenotypes upon silencing of host genes were performed with at least three biological replicates in independent experiments. Statistical analysis for these infectivity assays was done in GraphPad Prism (GraphPad Software, LLC; version 10.2.2) using one-way or two-way ANOVA with Dunnett's multiple comparisons test.

Coimmunoprecipitation Assays

For AP–LC–MS/MS, stable Aag2 cells expressing individual arboviral capsids, host proteins, or control baits were grown to confluence in T175 flasks, scraped in ice-cold PBS, and stored as dry pellets at −70 °C. APs were performed as previously described (54). In brief, cell pellets were thawed on ice and lysed for 30 min in 1 ml of lysis buffer (0.5% NP-40, 150 mM NaCl, and 50 mM Tris–HCl [pH 8.0]) supplemented with 1× cOmplete Protease Inhibitor Cocktail (Roche). Cell lysates were clarified by centrifugation for 15 min at 20,000g, and protein concentrations were normalized across all samples (Pierce 660 nm Protein Assay reagent; Thermo Scientific). Ten percent of each lysate was used to determine global proteomes (effectomes) or as input, whereas the remaining were processed for immunoprecipitation using anti–HA- (Anti-HA Agarose Affinity Gel; Sigma–Aldrich) or anti–FLAG-conjugated beads (ANTI-FLAG M2-Affinity gel, Sigma–Aldrich). Beads were incubated with cell lysates for 3 to 4 h at 4 °C, and after extensive washes, proteins were eluted in SDS sample buffer (350 mM Tris–HCl [pH 6.8], 5.28% SDS, 3 mM DTT, 0.012% bromophenol blue, and 15% glycerol) for WB or 40 μl of urea/thiourea buffer (6 M urea, 2 M thiourea in 10 mM Hepes, pH 8.0) for further LC–MS/MS analysis. Samples were stored at −20 °C until further processing. For mapping of Moira–capsid interaction determinants, Aag2-C3 cells were seeded into T75 flasks (1.1 × 10^7^ cells), transfected with 15 μg of pPUb plasmid encoding FLAG-eGFP or FLAG-Moira, and at 24 hpt infected with WNV (multiplicity of infection [MOI] = 0.5) or DENV2 (MOI = 0.8). Cell pellets were collected at 48 h postinfection (hpi) (WNV) or 72 hpi (DENV2) and processed as described above using anti-FLAG beads. When indicated, nuclease treatment was carried out before immunoprecipitation by incubating cell lysates with nucleases (25 units [U] of DNase I [Thermo Scientific] and 50 μg of RNase A [Sigma]) for 20 min at 25 °C.

Capsidome Proteomics Sample Preparation

Eluates from affinity-purified complexes (interactomes) or whole-cell lysates (effectomes; 50 μg total) of Aag2 stably expressing different arboviral capsids (DENV2, YFV, ZIKV of Asian and African lineage, JEV, WNV, USUV, POWV, TBEV, MAYV, SINV, and LACV) or control baits (eGFP, HCV, and mock), each with n = 4 biological replicates per experimental group, were denatured in 40 μl U/T buffer (6 M urea/2 M thiourea in 10 mM Hepes, pH 8.0), and proteins were reduced and alkylated in 10 mM DTT and 55 mM iodoacetamide. Samples were digested with 1 μg (interactome) or 2 μg (effectome) of each LysC (FUJIFILM Wako Chemicals) and trypsin (Promega) in 40 mM ABC buffer (50 mM NH_4_HCO_3_ in water, pH 8.0) overnight at 25 °C, 800 rpm. Following digestion, peptides were purified on StageTips with three layers of C18 Empore filter disks (3M) as previously described (45, 55).

Ultra–HPLC and Trapped Ion Mobility Spectrometry Quadrupole Time-of-Flight Settings

All capsidome effectome and interactome samples were analyzed on a nanoElute (plug-in v.1.1.0.27; Bruker) coupled to a trapped ion mobility spectrometry quadrupole time of flight (timsTOF Pro; Bruker) equipped with a CaptiveSpray source as previously described (56). In brief, peptides were first injected into a Trap cartridge (5 mm × 300 μm, 5 μm C18; Thermo Fisher Scientific), followed by separation on a 25 cm × 75 μm analytical column, 1.6 μm C18 beads with a packed emitter tip (IonOpticks). The column was maintained at 50 °C using an integrated column oven (Sonation GmbH). First, the column was equilibrated using four column volumes prior to loading samples in 100% buffer A (0.1% formic acid [FA] in water). Samples were separated with a 120 min separation method and a flow rate of 400 nl/min using a linear gradient from 2% to 17% buffer B (0.1% FA in acetonitrile) over 60 min before ramping up to 25% (30 min), 37% (10 min), and 95% of buffer B (10 min) and finally sustained for 10 min. The timsTOF Pro was operated in parallel accumulation–serial fragmentation mode using Compass Hystar, version 5.0.36.0. The following settings were applied: mass range 100 to 1700 m/z, 1/K0 start 0.6 Vs/cm^2^ end 1.6 Vs/cm^2^; ramp time 110.1 ms; lock duty cycle to 100%; capillary voltage 1600 V; dry gas 3 l/min; and dry temperature 180 °C. The parallel accumulation–serial fragmentation settings were 10 tandem MS scans (total cycle time, 1.27 s); charge range 0 to 5; active exclusion for 0.4 min; scheduling target intensity 10,000; intensity threshold 2500; and a collision-induced dissociation energy 42 eV.

MS Raw Data Processing and Analysis of Capsidome

Raw MS data were processed with the MaxQuant software, v.1.6.17.0, using the built-in label-free quantification (LFQ) algorithm and Andromeda search engine (57). The search was done against the Ae. aegypti reference proteome LVP_AGWG (UniProt Proteome ID: UP000008820; 14,555 entries in total (58)) concatenated with individual arboviral capsid and eGFP sequences. In addition, the LFQ algorithm (59) and the match-between-runs option were enabled. The search was restricted to peptides with a minimum length of seven amino acids and a maximum peptide mass of 4600 Da, and we allowed a maximum of two missed cleavage sites in the digestion with Trypsin/P and LysC. In MaxQuant, carbamidomethylation was set as a fixed modification, whereas methionine oxidation and N-acetylation were set as variable modifications. All other parameters were left as default (initial search peptide tolerance = 20 ppm, main search tolerance = 10 ppm; Fourier transform MS/MS match tolerance = 20 ppm; ion trap MS/MS match tolerance = 0.5 ppm). Experiment type was set as trapped ion mobility spectrometry–data-dependent acquisition (DDA) with no modification to the default settings. Search results were filtered with an FDR of 0.01 for peptide and protein identification. The Perseus software, version 1.6.15.0 (60), was used to further process the AP and total proteome datasets. Protein tables were filtered to eliminate the identifications from the reverse database and common contaminants. In the subsequent MS data analysis, only proteins identified on the basis of at least one peptide and a minimum of three quantitation events in at least one experimental group were considered. The MaxLFQ protein intensity values were log2 transformed, and missing values were filled by imputation with random numbers drawn from a normal distribution calculated for each sample (width 0.3, down shift 1.8) (60). All samples passed the quality control assessment and were included in the downstream analysis. Significant bait-specific PPIs were determined by FDR-corrected Welch’s t test (S0 = 1, FDR = 0.05, n = 250 randomizations, |log2[fold change]| ≥3.5; p ≤ 0.015) using HCV-core samples as control. HCV core was chosen as a control bait since it has similar biochemical properties (small hydrophobic RNA–binding protein) and functions, allowing more stringent selection criteria to identify species-specific host-interacting proteins. Mock and eGFP controls were subsequently used to re-evaluate identified hits as well as to select them for downstream functional validations. Significantly regulated proteins in the effectome were identified by two-sided FDR-corrected Student’s t test (S0 = 1, FDR = 0.05, n = 250 randomizations, |log2[fold change]| ≥2; p ≤ 0.05) using HCV-core samples as control. Results were plotted as scatter plots, and heat maps were generated in Perseus (60), and PPI networks were generated in Cytoscape (61). The UpSet plot and Venn diagrams were generated in R using the packages UpSetR (62), VennDiagram (63), gplots (64), and eulerr (65) in R, version 4.4.0. The complete enrichment analysis results are available in Supplemental Table S2.

To identify patterns of PPIs shared within viral genera or underlying natural vector preferences, the following additional criteria were considered: pan-orthoflaviviral host factors (identified in at least four of eight orthoflavivirus baits), alphaviral (identified as MAYV- and SINV-specific interactors), or orthobunyaviral (identified as LACV nucleocapsid interactors). Furthermore, host factors were defined as Aedes specific if they were specifically bound by at least three of five viral species, as Ixodes specific if they were identified as both POWV and TBEV interactors and Culex specific if they were specifically bound by at least two of four Culex-borne viral capsids.

Ortholog Mapping and Gene Ontology Enrichment Analysis

Subunits of benzo[a]pyrene (BAP) and primary biological aerosol particle (PBAP) complexes were inferred from Drosophila melanogaster sequences using VectorBase BLAST as well as blastn (National Institutes of Health), whereby the following orthologs were identified based on nucleotide sequence identity: Polybromo (AAEL000181), BRM (AAEL004942), OSA (AAEL017280), SNR1 (AAEL004371), and SAYP (AAEL009381).

Global protein ortholog mapping from Ae. aegypti to Homo sapiens was done by adapting and modifying a protocol similar to Shah et al. (52). We developed a Snakemake (66) workflow for orthology inference that employs the MMseqs2 easy-rbh method (67) and the InParanoid-DIAMOND ortholog mapping algorithm (68). InParanoid-DIAMOND was run with default parameters, whereas MMseqs2 easy-rbh was run with e-value parameter set to 10, in order to increase sensitivity and capture weak but potentially biologically relevant homologies. Postprocessing scripts standardize the output of both tools by computing an orthotype label for each protein pair, classifying relationships into 1:1 (one-to-one), 1:m (one-to-many), n:1 (many-to-one), and n:m (many-to-many) orthologs. In addition, the workflow includes a filtering step that identifies orthologs unique to MMseqs2 by comparing its results with those from InParanoid-DIAMOND. This highlights candidate pairs that may represent novel or lineage-specific orthologs missed by InParanoid-DIAMOND alone. Full implementation details and source code are available at https://github.com/ntnn19/orthologue_mapping_snakemake. Ae. aegypti host factors, which were successfully mapped to human IDs, are displayed with the human gene name and the Ae. aegypti gene ID; alternatively, only the Ae. aegypti gene ID was used. A total of 82% (n = 782) of the extended list of interactors (n = 945) were successfully mapped to human IDs and used for Gene Ontology (GO) enrichment analysis in metascape.org (69) and intersections with previously published human capsid interactome datasets.

Sample Preparation of dsRNA-Knockdown Cells for Quantitative LC–MS/MS Proteomics and Ultra HPLC and MS Settings

To validate the dsRNA-mediated knockdown (KD) of selected host targets (vATP-C, ID40: BAP170, AAEL001361; ID43: NGDN, AAEL008443; and ID55: Moira, AAEL022608) by MS, Aag2-AF5 cells were seeded into 6-well plates (1.42 × 10^6^ cells/well), and gene expression was silenced by transfecting 3750 ng of dsRNA (1:5 TransIT Insect transfection reagent [Mirus]) in 250 μl of Opti-MEM (Gibco) 24 h later. At 4 to 6 hpt, the medium was changed. Cell pellets were harvested at 72 hpt. Whole cell lysate (50 μg) of Aag2-AF5 KD cells in 4% SDS (in 10 mM Tris, pH 7.5, supplemented with 1× cOmplete Protease Inhibitor Cocktail) was processed for quantitative LC–MS/MS proteomics (n = 3). Cell lysates were incubated for 10 min at 95 °C, followed by 15 cycles of sonication. Next, reduction and alkylation with 10 mM DTT and 55 mM iodoacetamide were conducted. Subsequently, cell lysates were acetone precipitated twice, followed by and denaturation in 40 μl U/T buffer (6 M urea/2 M thiourea in 10 mM Hepes, pH 8.0). Then, samples were digested with 2 μg LysC (FUJIFILM Wako Chemicals) and 2 μg trypsin (Promega) in 40 mM ABC buffer (50 mM NH_4_HCO_3_ in water, pH 8.0) overnight at 25 °C, 800 rpm. Following digestion, peptides were purified on stage tips with three layers of C18 Empore filter discs (3M) as previously described (45, 55).

Samples were analyzed on a Vanquish Neo LC system (Thermo Scientific) coupled to an Orbitrap Exploris 480 mass spectrometer (Thermo Scientific) equipped with a Nanospray Flex source (Thermo Scientific). Peptides were injected into an Acclaim PepMap 100 trap column (2 cm × 75 μm, 3 μm C18; Thermo Fisher Scientific) and next separated on a 25 cm × 75 μm column (1.7 μm C18 beads ultra HPLC column; Aurora Ultimate) with a packed emitter tip (Ion Opticks), at a constant flow rate of 300 nl × min^-1^ over 100 min linear gradient. The column temperature was maintained at 50 °C using an integrated column oven (Sonation GmbH). The column was equilibrated using three column volumes before loading samples in 100% buffer A (99.9% Milli-Q water, 0.1% FA). Samples were separated using a linear gradient from 9% to 31% buffer B (99.9% acetonitrile, 0.1% FA) over 56 min before ramping up to 43% (35 min), 100% (1 min), and sustained for 8 min. The Orbitrap Exploris 480 was operated in positive ion mode, with a positive ion voltage of 2000 V in DDA using the Thermo Xcalibur software (Thermo-Fischer Scientific; version 4.5.474.0). DDA analysis was performed with a cycle time of 1.5 s. Survey scans were acquired at 120,000 resolution, with a full scan range of 350 to 1400 m/z, an automatic gain control target of 300%, and a maximum ion injection time of 25 ms, intensity threshold of 5 × 10^3^, 2 to 6 charge state, dynamic exclusion of 90 s, and mass tolerance of 10 ppm. The selected precursor ions were isolated in a window of 1.6 m/z, fragmented by a higher-energy collisional dissociation of 30. Fragment scans were performed at 15,000 resolution, with an Xcalibur-automated maximum injection time and standard automatic gain control target. Raw MS data were processed with the MaxQuant software, version 2.2.0.0, using the built-in LFQ algorithm and Andromeda search engine (57). The search was done against the Ae. aegypti reference proteome LVP_AGWG (UniProt proteome ID: UP000008820 with 14,555 entries). In addition, the intensity-based absolute quantification algorithm and match-between-runs option were used. The search was restricted to peptides with a minimum length of seven amino acids and a maximum peptide mass of 4600 Da, and we allowed maximum two missed cleavage sites in the digestion with Trypsin/P and LysC. In MaxQuant, carbamidomethylation was set as fixed and methionine oxidation and N-acetylation as variable modifications. Initial search peptide tolerance was set at 20 ppm, and the main search was set at 4.5 ppm. Experiment type was set as DDA with no modification to the default settings. Search results were filtered with an FDR of 0.01 for peptide and protein identification.

dsRNA Synthesis

Total cellular Aag2 RNA was extracted using TRIzol (Invitrogen) as recommended by the manufacturer and reverse transcribed following the High-Capacity Complementary DNA (cDNA) Reverse Transcription Kit (Applied Biosystems) manufacturer’s protocol using oligo dT primers (sequence: 5′-TTTTTTTTTTTTTTTTTTTTVN-3′). The T7-DNA template for dsRNA synthesis was generated by PCR with Phusion polymerase (New England Biolabs) using 100 ng cDNA with target-specific forward and reverse primers containing the T7 RNA polymerase promoter sequence (GTAATACGACTCACTATAGGG) at the 5′-ends (full primer sequences are provided in Supplemental Table S1). If dsRNA synthesis primers were not retrieved from publications, sequences of gene targets were retrieved from VectorBase (70) and synthesis primers were designed to generate dsRNA sequences of 150 to 500 base pairs. For the design of the dsRNA synthesis primer pairs used in the phenotypic screen (n = 110 targets), the DKFZ E-RNAi tool was used (71). To exclude off-target effects, each of the predicted dsRNA sequences was tested by manual blasting (Nucleotide blastn, National Center for Biotechnology Information, standard settings). After PCR, the sense and antisense T7-DNA templates were gel purified from a 1% agarose gel (Macherey–Nagel Nucleospin Gel and PCR clean-up kit), and 1 μg of each was used for in vitro ssRNA synthesis (16 h, 37 °C) using the RiboMAX Large Scale RNA production kit (Promega). Next, ssRNAs were annealed by heating (70 °C for 5 min), followed by a gradual cooling to 30 °C (0.4 °C steps each minute), 15 min at 30 °C, and a final step at 25 °C (on hold). Remaining ssRNA and DNA templates were removed by incubation with 2 μl of DNase I (ThermoScientific) and 2 μl of RNase A (Sigma) in DNase I kit reaction buffer for 1 h at 37 °C. Finally, synthesized dsRNA was purified using the TRIzol RNA extraction protocol (Invitrogen), resuspended in nuclease-free water, and stored at −70 °C until further use.

Phenotypic Screen and Silencing Validation Experiments

The functional relevance of newly identified capsid-interacting host proteins on viral replication was assessed by dsRNA-mediated gene silencing. Among the significant capsid-interacting host factors, 110 were selected for a functional dsRNA-mediated KD screen with DENV, WNV, and LACV based on their virus or genus specificity or for pan-arboviral-binding patterns. This included selective species-specific interactors (DENV2, eight targets; WNV, four targets; and LACV, three targets), interactors with pan-orthoflaviviral (73 targets) and pan-alphaviral (2 targets) specificity, interactors with subviral specificities (orthoflavi- and orthobunyaviruses [13 targets], orthoflavi- and alphaviruses [three targets]); as well as four hits displaying pan-arboviral binding specificity. dseGFP was used as a negative control (51), whereas positive controls included three host dependency factors previously described (Loquacious, Loqs; (72), orthologs of the human ATP6V0C [vATP-ac39 (73, 74, 75)] and vATP-C (76), and two known host restriction factors [AGO2 (72) and Piwi4 (77)]. WNV and LACV screening was performed in Aag2-AF5 cells, whereas DENV2 screening was performed in Aag2-C3 cells, which displayed higher permissiveness to DENV2 infection.

Cells were seeded into 24-well plates (2.8 × 10^5^ cells/well), and gene expression was silenced by transfection with 750 ng of dsRNA (1:5 TransIT Insect transfection reagent) in 50 μl of Opti-MEM (Gibco) 24 h later. Medium was changed at 4 to 6 hpt. Plate layouts of dsRNA transfections were shuffled across the three biological replicates to minimize plate effects.

For dsRNA screening, cells were infected the following day with DENV2 (MOI = 0.1), WNV (MOI = 0.1), and LACV (MOI = 0.5), respectively, using 300 μl inoculum/well for 1 h at 28 °C without CO_2_. After virus absorption, inocula were removed, and cells were incubated at 28 °C without CO_2_. Virus-containing cell culture supernatants were collected at 24 hpi (LACV), 48 hpi (WNV), and 72 hpi (DENV2), and infectious titers were determined by plaque assays. Plaque detection and counting was carried out using a Nikon artificial intelligence (AI)–based method as described below. For validation experiments, Aag2-C3 cells were transfected with dsRNA against selected targets as described above and infected the following day with individual arboviruses at varying MOIs: MOI = 0.1 (DENV2, WNV, YFV, USUV, and ZIKV), MOI = 0.5 (LACV), or MOI = 0.01 (CHIKV and SINV). Cell culture supernatants were subjected to plaque assay, and cells were lysed in TRIzol reagent for quantification of intracellular viral RNA levels. KD of selected host targets was verified by RT–quantitative PCR (qPCR), and cell viability was assessed by resazurin cell viability assay at each endpoint.

Plaque Assay

Vero E6 cells were seeded into 96-well plates (2.5 × 10^4^ cells/well) or 24-well plates (2 × 10^5^ cells/well). The next day, serial dilutions of virus-containing cell culture supernatants were used to infect Vero E6 cells, overlayed with 1.5% medium viscosity carboxymethylcellulose (Sigma–Aldrich) in MEM (Gibco) containing 2% FBS, and after two (SINV), three (WNV, USUV, MAYV, and CHIKV), four (LACV), five (DENV2), or eight (YFV) days postinfection, cells were fixed and stained with crystal violet (Roth) as described before (78).

AI-Based Quantification of Plaque Assays for dsRNA Screenings

Effects of host factor silencing in the DENV2, WNV, and LACV phenotypic screens were analyzed by plaque assays in 96-well plate format. Brightfield images of each well were acquired using a fully motorized Nikon Ti and a Nikon Ti2 inverted microscope, respectively. Images were captured with a Plan Apo λ 2× objective lens using a pco.edge 4.2 camera. The JOBS module of NIS Elements (v5.42) enabled the automated image acquisition of each 96-well plate. Briefly, each 96-well assay plate was aligned at the XY stage of the microscope, and six wells (four at the edges and two in the center of the plate) were selected to define a focus surface over the entire plate using brightfield-based software autofocus. Subsequently, the Z drive of the microscope was set to follow the defined focus surface while capturing each well with a single shot, followed by an automatic shading correction. First, a representative selection of 15 to 25 training images (wells) was chosen for each dataset, and plaques were manually annotated using the binary editor functionality of NIS Elements (v5.42). These annotations were then used to train the Segment.ai and Segment Objects.ai image processing and analysis module of NIS Elements. The trained AI was iteratively optimized by applying the trained network on the original training images as well as a selection of additional training images without annotations. The annotations were corrected using preprocessing functionalities, removing tiny annotations not representing plaques, filling holes in predicted plaques, and manually correcting imprecisely predicted plaque outlines using the binary editor functionality, followed by another training procedure. Once the neuronal network was trained, the GA3 module of NIS Elements was used to batch process the entire datasets of assay plates. The calculation of viral titers according to the plaque count per well was subsequently performed using a Python (v2.9.0) script.

Cell Viability Assay

Resazurin sodium salt (Sigma–Aldrich) solution was prepared in 1× Dulbecco’s PBS (Sigma–Aldrich) (1 mg/ml) and incubated with Aag2-AF5 cells or Aag2-C3 cells in 0.5 ml/well L-15 complete medium for 1 h at 28 °C (final concentration: 0.167 mg/ml). Resorufin fluorescence emission was measured at 590 nm after excitation at 560 nm with a plate reader (TECAN Infinite M Plex). Values were normalized to dseGFP-treated control cells.

RT–qPCR Analysis

dsRNA-mediated KD of host targets or intracellular viral RNA levels were quantified by two-step RT–qPCR. First, cDNA was generated from TRIzol-extracted intracellular RNA using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems) with random primers. Then, qPCR was performed with the PowerUP SYBR Green Master Mix (Applied Biosystems) following the manufacturer's protocol (qPCR primers are listed in Supplemental Table S1). Relative gene expression data were acquired with the 7500 Fast Real-Time PCR System and 7500 Software v2.3 from Applied Biosystems, whereby expression of the gene of interest was normalized to Ae. aegypti actin or 40S ribosomal protein S7 gene expression using the 2^-(ΔΔCt)^ method (79).

Western Blotting

Aag2 cells were lysed in radioimmunoprecipitation assay buffer (50 mM Tris–HCl [pH 8.0], 1% Triton X-100, 0.25% NP-40, 0.1% SDS, and 150 mM NaCl) supplemented with 1× cOmplete Protease Inhibitor Cocktail for 20 min on ice. Normalized clarified cell lysates were boiled in reducing sample buffer for 10 min at 95 °C, and 10 μg of total protein were separated by SDS-PAGE on 15% acrylamide gels. Proteins were transferred onto Amersham Protran nitrocellulose membranes (Sigma–Aldrich) using a Bio-Rad Mini-PROTEAN Tetra Cell apparatus. Unspecific binding sites were blocked with blocking buffer (10% milk in 0.25% PBS–Tween-20), and membranes were incubated with primary and HRP-conjugated secondary antibodies. Membranes were imaged using an ECL CHEMOCAM system (INTAS) using the Western Lightning Plus-ECL substrate (Revvity).

IF Assay

Aag2 cells stably expressing HA-tagged arboviral capsids were fixed and permeabilized for 20 min at −20 °C in ice-cold 100% methanol. For all the other IF assays, cells were fixed in 4% paraformaldehyde (pH 8.6) for 1 h and permeabilized in 0.5% Triton X-100 for 10 min at room temperature. Unspecific binding sites were blocked with 5% bovine serum albumin (45 min, room temperature) and incubated with primary and secondary antibodies (1 h, room temperature) diluted in 1% bovine serum albumin. Nuclei were stained with 4′,6-diamidino-2-phenylindole (Fisher Scientific) (1:3000 dilution, 5 min, room temperature). Coverslips were mounted onto microscopic slides using Fluoromount-G (SouthernBiotech). Images were acquired employing a spinning disk microscope (Nikon Ti2–based Spinning Disk equipped with Yokogawa CSU-W1 Spinning Disk Unit with Andor iXon Ultra 888 EMCCD cameras and Omicron Laser box with 405, 488, 561, and 647 nm lasers) using the Plan Apo 100× numerical aperture 1.45 objective lens or a confocal laser scanning microscope (Nikon Ti2–based A1 R confocal laser scanning microscope with a 60× CFI Plan Apo Lambda immersion oil objective [numerical aperture = 1.42] equipped with A1-DUG-2 GaAsP Multi Detector Unit and the LU-N4/N4S 4-laser unit [405, 488, 561, and 640 nm]). Both microscopes were operated using the NIS-elements software (AR 5.42.02).

Experimental Design and Sample Preparations for ATAC-Seq and RNA-Seq

Chromatin remodeling was assessed by ATAC-Seq on mock- or virus-infected (DENV2, WNV, or YFV) Aag2-C3 cells. To assess the impact of virus infection, capsid expression, or Brahma complex regulation on the chromatin landscape, Moira KD (dsMoira) and dseGFP-treated cells were included in the analysis alongside Aag2-C3 cells overexpressing HA-eGFP, HA-YFV capsid, HA-WNV capsid, or HA-DENV2 capsid as well as YFV, WNV, DENV2, and mock-infected cells (N = 3 biological replicates). Aag2-C3 cells were seeded into 24-well plates (2.8 × 10^5^ cells/well), and the next day, transfected with 750 ng of dsRNA (dseGFP or dsMoira) or 1 μg of pPUb plasmids (HA-eGFP, HA-YFV capsid, HA-WNV capsid, and HA-DENV2 capsid) as described above. Untreated cells were infected 1 day postseeding with DENV2 (MOI = 1), WNV (MOI = 0.5), and YFV (MOI = 1.5). Seventy-two hpi or hpt, samples were processed for ATAC-Seq and RNA-Seq.

ATAC-Seq Sample Preparation, Data Acquisition, and Analysis

ATAC-Seq was performed as previously described by Grandi et al. (80), with slight modifications. In brief, Aag2-C3 cells were treated with 1000 U DNase I (Worthington) for 30 min at 37 °C, harvested by trypsinization, and resuspended in ice-cold RSB buffer (10 mM Tris–HCl, pH 7.4, 10 mM NaCl, and 3 mM MgCl_2_) for downstream ATAC-Seq processing, or stored in TRIzol reagent at −20 °C for RNA-Seq processing. For ATAC-Seq, 1 × 10^5^ cells were lysed in RSB buffer, and nuclear pellets (5 min at 500g, 4 °C) were carefully resuspended in 50 μl cold ATAC–NTD buffer (10% NP-40, 10% Tween-20, and 1% digitonin in RSB buffer) and incubated on ice for 3 min. Lysis buffer was diluted with 1 ml ice-cold ATAC-T buffer (RSB + 10% Tween-20) genomic DNA pelleted by centrifugation (10 min at 500g, 4 °C). Transposition was performed by adding 50 μl of transposition mix (2× Tagmentation DNA buffer [Illumina], 1% digitonin, 10% Tween-20, 100 nM loaded Tn5 transposase in 33% PBS), and subsequent incubation for 30 min at 37 °C in a shaker (1000 rpm). The Tn5 reaction was stopped by adding 250 μl of DNA binding buffer (NucleoSpin Gel and PCR Clean-up Kit; Macherey–Nagel). DNA was purified following the manufacturer's instructions with the following changes: 300 μl DNA wash buffer and 21 μl DNA elution buffer. Purified DNA was stored at −20 °C prior to library preparation.

Barcoding of transposed fragments was performed as described before (80) using two adapter oligonucleotides (Supplemental Table S1). The final libraries were quality controlled by accessing the fragment sizes via the TapeStation D5000 Assay (Agilent). ATAC libraries were subsequently sequenced on an Illumina NextSeq 2000 system in a paired-end mode (2 × 61 bp, NextSeq 1000/2000 P2 XLEAP-SBS Reagent Kit [100 cycles]) and demultiplexed by bcl2fastq. All samples were sequenced to 11.5 to 24.9 million read pairs, and data were quality assessed using the FastQC (81) and MultiQC (82) tools. Paired-end reads were aligned with Bowtie2 (83) using options “--very-sensitive -X 2000 --no-mixed --no-discordant,” and only high-quality reads with option “q > 30” were retained using samtools (84). Resulting BAM files were sorted with samtools, and duplicate reads were removed using Picard MarkDuplicates command (85). Read coordinates were shifted for transpose-induced insertions using deepTools (86) command alignmentSieve. Finally, MACS2 (87) was used to call peaks using the option “BAMPE --nomodel --call-summits --keep-dup all -B --SPMR -q 0.01.” For each comparison, peaks were converted to GRanges object for every replicate per condition using the R package ChIPQC (88) GetGRanges function. Consensus peaks were generated using all samples per comparison using runConsensusRegions function of the R package soGGi (89). For each comparison, peaks with a minimum occurrence of three were retained. Resulting consensus peaks from all comparisons were combined to generate the main consensus peak set, which was used for downstream differential accessibility analysis. YFV infection replicate 3 was excluded from the analysis as an outlier. Reads were counted using the main consensus peak set using R package Rsubread (90). The DESeq2 likelihood ratio test was used to identify differential accessible peaks, and null variance of Wald's test statistic output by DESeq2 was re-estimated using the R package fdrtool (91) to calculate p values (and adjusted using the Benjamini–Hochberg method) for the final list of DA peaks. For the comparison of dsMoira and dseGFP, a cutoff of log2(fold change) >1.5 was applied, whereas for all other comparisons, log2(fold change) >0.5 was used. Annotation of peaks to the nearest transcription start site (TSS) and other genomic features was performed using the HOMER annotatePeals.pl command. Peaks were assigned to the TSS of the nearest gene within a minimum distance of 50 kb. Motif enrichment for DA peaks was done using the HOMER findMotifsGenome.pl command, and to identify peaks encompassing motifs was done using findMotif.pl command (92). All DA peaks were used for motif enrichment. Quantification of genomic features and distance to TSS plots for DA peaks were generated using the R package ChIPseeker (93, 94).

RNA-Seq Data Acquisition and Analysis

Total intracellular RNA was isolated from Aag2-C3 cells using the TRIzol RNA extraction protocol described above. High RNA integrity was confirmed by applying a TapeStation RNA ScreenTape analysis (Agilent, Supplemental Table S3). Per sample, 1 μg of total RNA (quantified by Qubit RNA HS Assay; Thermo Fisher) was poly(A)-captured applying the Lexogen Poly(A) RNA Selection Kit and further processed via the RNA-Seq V2 Library Prep Kit with unique dual indexes (Lexogen) in the long insert size variant (RTL) according to the manufacturer’s instructions (applying 14 cycles of Library Amplification PCR, step 6.3, User Guide, version 171UG394V0101).

Libraries were quality controlled on a TapeStation D5000 Assay and were sequenced on an Element Biosciences AVITI instrument (2 × 150 Sequencing Kit Cloudbreak Freestyle) in a paired-end mode (2 × 155 bp). After demultiplexing via bases2fastq, 15.3 to 54.7 million read pairs were assigned to each sample. All samples passed quality control by FastQC (81) and MultiQC (82) and were subjected to downstream analysis. Paired-end reads were trimmed 12 bases and 9 bases from the start to remove the unique molecular identifier and sequencing adapter, respectively. Gene abundances were quantified with salmon (v1.10.0) (95) using Ae. aegypti Ensembl gene annotations (version 60) for AaegL5 genome assembly and imported using R package tximport (96). Count normalization and differential expression (DE) analysis were performed using the DESeq2 package (v1.44.0) (97). FDR (Benjamini–Hochberg-adjusted p values) <0.05 and no minimum log2fold change were used as criteria for the final DE gene list. GO enrichment analysis was performed with the Ensemble Metazoa BioMart database using the enricher command from the R package clusterProfiler (98). Fisher's exact test was used to test the significance of DE genes from RNA-Seq and DE genes from ATAC-Seq using R command phyper. Downstream analysis and plots were generated using ggplot2 (99) and custom R scripts using the R statistical computing language (version 4.4.1).

Results

A Pan-Arboviral Map of Capsid Protein Interactors and Targets in Ae. aegypti Cells

To map unique and conserved effector functions of arboviral capsids in invertebrates, we generated a quantitative, multilayered proteomic atlas across 11 individual viral species spanning three prototypic arboviral genera: orthoflavivirus (DENV, YFV, ZIKV, JEV, WNV, USUV, TBEV, and POWV), Alphavirus (SINV, MAYV), and Orthobunyavirus (LACV) in Ae. aegypti Aag2 cells. These 11 arboviruses encompass different primary vector specificities, including Aedes spp. mosquitoes (DENV, YFV, ZIKV, LACV, and MAYV), Culex spp. mosquitoes (WNV, JEV, USUV, and SINV), or Ixodes spp. ticks (POWV and TBEV), allowing to assess the potential differences between Aedes-specific or conserved effector functions. This approach combines systematic mapping of PPI networks (interactome) with global proteomics (effectome) to identify host proteins targeted by individual viral capsids. To this end, we generated a library of Aag2 cells constitutively expressing 12 N-terminally HA-tagged capsid proteins of individual arboviruses and three control baits (Hepacivirus HCV-core protein, eGFP, and mock-transfected cells) under the control of the polyubiquitin promoter (100, 101) (Fig. 1A). Subcellular distribution and expression of HA-tagged capsids or control baits were assessed by IF and WB analysis (Supplemental Fig. S1, A and B), confirming robust and comparable expression levels for all baits, with the exception of USUV capsid and the LACV nucleocapsid displaying relatively lower and higher expression levels, respectively. For each cell line, four replicates of affinity-purified complexes (interactomes) and whole-cell lysates (effectomes) were collected and analyzed by quantitative MS. To increase the accuracy of functional annotation, protein IDs were mapped to their human orthologs based on protein homology (Supplemental Table S2 and Experimental Procedures section). If existent, we are referring to the corresponding human gene name besides the Ae. aegypti gene ID. A total of 2171 different proteins were identified across the entire arboviral capsid interactome atlas (capsidome) (median identification rate = 1033 proteins; Fig. 1A, Supplemental Fig. S1C and Supplemental Table S2), with up to twofold differences between the lowest (WNV, USUV, and mock) and the highest number of enriched proteins across baits (HCV and POWV). Interestingly, these differences did not linearly correlate with the relative expression levels of the baits, suggesting quantitative differences in the propensity of different arboviral capsids to interact with the mosquito proteome (Fig. 1B and Supplemental Fig. S1D). Principal component analysis of the individual capsid-interaction networks confirmed high correlation within biological replicates, further highlighting subclusters within each viral family, with orthoflaviviruses associated with hemorrhage-like disease manifestations (YFV and DENV) forming a distinct cluster when compared with the neurotropic subgroup (ZIKV, JEV, TBEV, and WNV) (Supplemental Fig. S1E). Notably, the capsid of POWV, a virus naturally transmitted by ticks (102, 103), displayed the most divergence from all other species, likely reflecting the substantially higher number of interacting proteins (Fig. 1B and Supplemental Fig. S1E). Altogether, these results suggest significant intragenera and intergenera specificities in the ability of different arboviral capsids to interact with the mosquito proteome.Fig. 1The pan-arbovirus capsidome in mosquito cells.A, experimental design and workflow of the pan-arboviral capsid interaction atlas. Stable Aedes aegypti Aag2 cell lines expressing HA-tagged arboviral capsids (n = 13) were used to characterize changes in the whole cellular lysate (effectome) and to identify interacting host proteins (interactome). Total number of identified proteins and significant hits are displayed (interactome: two-sided Welch’s t test, S0 = 1, FDR = 0.05, n = 250 randomizations, with log2[fold enrichment] ≥3.5 and p ≤ 0.015; effectome: two-sided Student’s t test, S0 = 1, FDR = 0.05, number of randomizations = 250, |log2[fold change]| >2; control: HCV core). This figure was created in BioRender. Flory, C. (2026) https://BioRender.com/gzavmbs. B, total numbers of unique and shared significant interactors of each bait. Interactors were defined as shared if identified as significant interactors for at least two baits. C, combined virus–host protein–protein interaction network of arboviral capsids measured by AP–MS. Selected biological functions and processes are highlighted by colors and in squares. Only the top 50 most enriched significant interactors are shown per bait (log2[fold change] ≥3.5; p ≤ 0.015). Interactions between viral and host proteins are indicated by gray lines. Colored circles and nodes represent a manually curated selection of GO annotations. Gene names listed refer to human orthologs or *Ae. aegypti* gene IDs. AP, affinity purification; FDR, false discovery rate; GO, Gene Ontology; HA, hemagglutinin; HCV, hepatitis C virus; ID, identity; MS, mass spectrometry.

The Pan-Arbovirus Mosquito Capsidome Identifies Distinct and Conserved Networks of Host Interactions Across Species

To identify bait-specific, high-confidence PPIs, four replicates of each arboviral capsid were analyzed against the interactome of HCV core—the capsid of a human-specific hepacivirus with similar biochemical properties—using very stringent cutoff criteria (log2 [fold change] ≥3.5; two-sided Welch’s t test p value ≤0.015; S0 = 1, FDR ≤0.05, n = 250). In total, the arboviral capsidome includes 420 significant host proteins interacting with the capsids of 11 arboviral species (Fig. 1A, Supplemental Table S2), normally distributed across the full dynamic range of protein abundance in the Ae. aegypti proteome and spanning more than 13 orders of magnitude (Supplemental Fig. S1F). Among the 12 individual PPI networks, DENV and POWV displayed the highest number of significant and unique interactions, whereas the largest group of interactors was shared between YFV, POWV, and DENV (n = 57) (Figs. 1B and 2A, Supplemental Fig. S1G).Fig. 2Arboviral capsid proteins display distinct genus-, species-, and vector-level host binding specificities.A, upset plot of 420 significant capsid-interacting proteins in the capsidome (two-sided Welch’s t test, S0 = 1, FDR = 0.05, number of randomizations = 250 comparing each bait to HCV; log[fold change] ≥2^3.5^ and p ≤ 0.015). B, metascape enrichment analysis of all statistically significant host-interacting proteins from the capsidome. Human orthologs of mosquito proteins were used for the genus-based functional enrichment analysis as detailed in the Experimental Procedures section. C, identification of virus genus-specific interactors (criteria: found in at least four orthoflaviviruses, two alphaviruses, and one orthobunyavirus). D, identification of capsid interactors specific to Aedes-borne, Culex-borne, or tick-borne arboviruses (criteria: found in at least three Aedes-, two Culex-, two tick-borne viruses). E–H, identification of species-specific capsid-interacting proteins (Hawaii plot, Pearson's correlation of each bait to all other 12 baits, and three controls used in this study [referred to as “complement”], number of permutations = 100, class A: S0 = 1, FDR = 0.05, class B: S0 = 0.1, FDR = 0.05)—exemplary DENV2 (E), WNV (F), LACV (G), and MAYV (H). I, intersection of C-interacting proteins identified in this study (extended list of capsid interactors [n = 945], 82% [n = 782] mapped to human IDs) with previously reported capsid interactomes in mammalian cells (WNV–capsid interactome (47), ZIKV-capsid (45), ZIKV– and DENV–capsid interactome (52), and ZIKV-C interactome (14)). J, STRING network of Aedes aegypti capsid interactors identified in this study, previously reported as capsid-interacting proteins in human studies (found in at least three of four human data sets as orthologs of *Ae. aegypti* proteins). Factors that were found in this and all human studies are highlighted by thick black borders. Connections indicate functional and physical interactions that were experimentally determined (https://string-db.org/). DENV2, dengue virus serotype 2; FDR, false discovery rate; HCV, hepatitis C virus; LACV, La Crosse virus; MAYV, Mayaro virus; WNV, West Nile virus.

Among the strongest interactors shared across different arboviral capsids, we identified host factors associated with ribosome biogenesis and mitochondrial ribosomes, as well as chromatin remodeling, RNA processing, and nuclear transport, corroborating previous reports on DENV-C and ZIKV-C in Ae. aegypti cells (51, 52) (Fig. 1C).

To identify evolutionarily conserved patterns of host usage within each arboviral genus, the capsid interactomes of individual viral species were collapsed into the corresponding genera for functional GO enrichment analysis (Fig. 2B). Orthoflaviviral capsids displayed a significant propensity to interact with mitochondrial ribosomal proteins, as well as proteins involved in RNA processing, modification, and localization, reflecting their affinity for nucleic acids and their nucleocytoplasmic distribution (51, 52). Conversely, alphaviral capsids predominantly targeted processes involved in transcriptional regulation and protein secretion, whereas the orthobunyaviral nucleocapsid displayed selective functional enrichment in processes involved with tRNA processing, small nuclear ribonucleoprotein regulation (SMN–RNA polymerase II–RHA complex), and chromosome organization (Fig. 2B).

To further identify evolutionarily conserved binding specificities across arboviral capsids, we next stratified individual PPIs based on their conservation within viral genera (Orthoflavivirus, Orthobunyavirus, or Alphavirus specific) or natural arthropod vector species (Aedes-, Culex-, or tick borne) (Fig. 2, C and D). This analysis revealed a high divergence of host interactions among different viral genera, with only limited overlap of PPIs across viral genera or viral species transmitted by different arthropod vectors (Fig. 2C). For instance, 22 of the 25 host proteins targeted by the majority of orthoflaviviruses were specific to the genus when compared with alphaviruses and orthobunyaviruses (e.g., TAF1 [AAEL025013], NPM2 [AAEL003750], TAOK1 [AAEL000217], KIF20A [AAEL022590], MED18 [AAEL002284], METTL17 [AAEL003822], PTCD3 [AAEL010042], AURKAIP [AAEL020116], GALE [AAEL007390], and TMEM43 [AAEL001011]). Similarly, only four factors interacted with the capsids of both alphaviruses tested (MAYV and SINV), with three exhibiting Alphavirus-specific binding profiles (STXBP5 [AAEL006948], USP34 [AAEL022105], and MTCH1 [AAEL015041]).

Overlaying binding profiles on arthropod vector preferences of individual virus species (Aedes-, Culex-, or ticks specific) revealed a large cluster of interactors unique to capsids of Aedes-borne viruses (n = 32). Conversely, only four factors interacted with the capsids of both tick-borne viruses (TBEV and POWV) (EEF1B2 [AAEL000951], TAF1 [AAEL025013], MRPS6 [AAEL004034], and NPM2 [AAEL003750]) and six host proteins were selectively bound by the majority of capsids from the Culex-borne subgroup STXBP5 (AAEL006948), USP34 (AAEL022105), KIF20A (AAEL022590), NUAK1 (AAEL019473), AP2M1 (AAEL003106), and ACADM (AAEL014452), possibly reflecting the specificity for the cellular background used (Ae. aegypti derived) (Fig. 2D). Notably, only two host factors—TAF1 (AAEL025013) and EEF1B2 (AAEL000951)—were bound by the majority of Aedes-, Culex-, and tick-borne species, suggesting an evolutionarily conserved targeting of transcriptional and translational regulators by arboviral capsids.

The extended capsidome additionally allows the identification of high-confidence species-specific interactors, providing a blueprint for mechanistic and functional follow-up studies on individual arboviral species (Fig. 2, E–H). As examples of these PPIs, we identified proteins involved in lysosomal degradation of glycolipids and virus cell entry (i.e., NAGA [AAEL005460] and SDC2 [AAEL013969]; DENV), transcriptional regulation and RNA processing (i.e., ZFX [AAEL008538], GPATCH4 [AAEL011985], and THOC6 [AAEL006547]; WNV), endoplasmic reticulum (ER) protein translocation, and the RNAi pathway (i.e., SSR3 [AAEL007441], AGO2 [AAEL017251]; LACV) as well as transcriptional regulation and RNA transport (MED21 [AAEL010797], JARID2 [AAEL010971], and FMR1 [AAEL009326]; MAYV) (Fig. 2, E–H). To investigate possible differences between historical and contemporary isolates, the interactomes of two prototypic ZIKV strains, representatives of the newly emerged Asian (H/PF/2013) and the historical African lineages (DakAr41524), were also included in the analysis (Supplemental Fig. S1H). Interestingly, despite their high sequence identity (only three amino acid residues differ between the two strains), two host factors—uncharacterized protein AAEL022631 and complex III subunit 9 UQCR10 (AAEL000182)—were selectively enriched in the capsid interactome of the Asian lineage compared with the African lineage, highlighting possible mechanisms contributing to increased transmissibility or pathogenicity of newly emerged strains (Supplemental Fig. S1H; (104, 105, 106)).

To assess conservation of PPIs between human and mosquito hosts, the extended arboviral capsidome (n = 945) was mapped to the human proteome and intersected with the four human capsid interactomes of individual arbovirus species published to date (14, 45, 47, 52). This analysis revealed a surprisingly high degree of cross-species conservation, with 214 human orthologs of mosquito proteins found in at least one human interactome study (Fig. 2I), and 13 host factors consistently identified in three of four human datasets (Fig. 2J). Among these, we identified mosquito orthologs of previously reported capsid interactors in humans (52), such as NGDN (AAEL008443), GADD45GIP1 (AAEL005146), LARP7 (AAEL007853), and PPAN (AAEL009309) (Fig. 1C). Furthermore, we identified PTCD3 (AAEL010042), LARP1 (AAEL009070), KRR1 (AAEL000708), and DDX18 (AAEL005744) as highly conserved cross-species capsid interactors, indicating that post-transcriptional regulation of gene expression, translation initiation, ribosome and spliceosome assembly are essential arboviral targets in both vertebrate and invertebrate hosts (Fig. 2, I and J).

Altogether, the Ae. aegypti pan-arboviral capsidome provides an evidence-based framework to categorize species- and host-specific mechanisms of host adaptation across three human pathogenic arboviral genera and illuminates a novel set of host proteins potentially involved in arbovirus replication and transmission in mosquitoes.

Capsid Expression Selectively Modulates Cellular Proteostasis in Invertebrates

To map virus–host PPIs to effector functions of viral capsids, we additionally analyzed quantitative changes in global proteomes of Aag2 cells stably expressing individual arboviral capsids (“effectomes”) (Fig. 1A and Supplemental Table S2). Analysis of global cellular proteomes across the different cell lines identified 5453 unique proteins (Fig. 1A), with similar identification rates across all samples and replicates (Supplemental Fig. S2A).

This approach identified 124 proteins displaying altered abundance upon expression of at least one arboviral species, with USUV, JEV, ZIKV, and WNV capsid-expressing cells exhibiting the highest proportion of selectively regulated targets (Fig. 3A and Supplemental Fig. S2B). Interestingly, the majority of modulated host proteins exhibited increased abundances (Fig. 3B), including proteins involved in lipid metabolism (e.g., Orthoflavivirus-regulated AAEL006028), cellular transport (e.g., WNV-regulated transferrin [AAEL015458]), endopeptidase activity (PRSS36 [AAEL001084/AAEL001077/AAEL002254]), and apoptosis (e.g., WNV: AAEL000328 and DENV: RBBP6 [AAEL019965]) (Fig. 3, C–F). Among these, LACV nucleocapsid displayed selective increase in APTX (AAEL014945) and HNRNPDL (AAEL005049), key regulators of DNA repair and mRNA processing, respectively; and profound and specific reduction in UDP-glycosyltransferase (AAEL003091) abundance (Fig. 3I). Furthermore, TDO2 (AAEL000428) was identified as a pan-arboviral–regulated host factor (Fig. 3G), whereas the uncharacterized AAEL008502 as well as cytochrome P450 (AAEL004054)—recently involved in mosquito host responses to arboviruses (107, 108)—were specifically regulated in WNV capsid–expressing cells (Fig. 3H). Notably, overlaying the arboviral capsid effectomes and interactomes revealed limited overlap, suggesting that regulation of host protein abundance might occur through transcriptional modulation of gene expression or protein stability, independent of direct PPIs.Fig. 3**Arboviral capsids modulate abundance of selected mosquito host proteins.**A, total numbers of unique and shared host proteins displaying differential abundances upon persistent capsid protein (C) expression in Aag2 cells (two-sided Student’s t test, S0 = 1, FDR = 0.05, number of randomizations = 250, |log2(fold change)| ≥2, p ≤ 0.05; control: HCV core). Host factors were defined as shared if significantly modulated by the expression of at least two individual arboviral capsids. B, numbers of significantly upregulated or downregulated proteins in each C-expressing Aag2 cell line. C–F, identification of regulated host factors upon expression of representative arboviral capsids. Volcano plots of significantly upregulated or downregulated proteins in DENV-C (C), WNV-C (D), LACV-C (E), and MAYV-C-expressing cells (F) (two-sided Student’s t test, S0 = 1, FDR = 0.05, number of randomizations = 250, log2|[fold change]| ≥2, p ≤ 0.05). G–I, profile plots of proteins regulated by the capsids of DENV (G), WNV (H), and LACV (I). DENV, dengue virus; FDR, false discovery rate; HCV, hepatitis C virus; LACV, La Crosse virus; MAYV, Mayaro virus; WNV, West Nile virus.

Collectively, the effectome highlights the ability of arboviral capsids to remodel host protein expression, with the neurotropic clade of orthoflaviviruses (ZIKV, JEV, WNV, and USUV) exhibiting the most profound effects on the mosquito proteome.

Phenotypic Screening Identifies Novel Capsid-Interacting Host Factors Involved in DENV, WNV, and LACV Replication

To functionally characterize newly identified interactions, we selected 110 capsid-interacting proteins with binding profiles spanning diverse species, genus, or vector specificities (see the Experimental Procedures section) and assessed their relevance in viral replication via dsRNA-based phenotypic screening (Fig. 4A). To generalize findings on host factor usage across arboviral genera, Aag2-C3 cells or Aag2-AF5 cells were transfected with gene-specific dsRNAs and infected 24 hpt with a prototypic Aedes-borne orthoflavivirus (DENV), a Culex-borne orthoflavivirus (WNV), or an Aedes-borne Orthobunyavirus (LACV). Virus titers were assessed by plaque assays, using a custom-made image-based analysis pipeline leveraging AI-assisted segmentation for unbiased quantitation of plaque-forming units (Fig. 4B, Supplemental Fig. S3).Fig. 4**Functional relevance of newly identified capsid-interacting proteins across different arboviruses.**A, specificity of shortlisted capsid-interacting host proteins (n = 110 targets) across arboviral capsids selected for functional validation. B, experimental design of dsRNA-mediated knockdown screen to investigate the functional relevance of identified capsid interactors for DENV, WNV, and LACV replication in Aag2-C3 (DENV) or Aag2-AF5 (WNV and LACV) Aedes aegypti cells. Viral titers were determined by plaque assay in 96-well plates and imaged after crystal violet staining using a screening wide-field microscope (Nikon Europe BV). A proprietary convolutional neural network (Nikon segmentation artificial intelligence [AI] (154)) was trained to systematically identify and count PFUs as described in the Experimental Procedures section. This figure was created in BioRender. Flory, C. (2026) https://BioRender.com/ps1zv6b.C, the effect of dsRNA-mediated knockdown of 110 capsid interactors or controls on virus production was assessed by plaque assay. The results of each biological replicate (n = 3) of dsRNA treatment are shown as heatmaps for the three viruses, respectively (virus titers in PFU/ml relative to nontargeting dseGFP-treated control cells). Host factor binding specificities are indicated on the right (DENV, D; WNV, W; LACV, L; other orthoflaviviruses, O; alphaviruses, A). Newly identified host restriction and dependency factors for each virus species are highlighted as red and blue circles on the left, respectively (cutoff criteria: ≥50% difference in viral titers). Cell viability was determined by resazurin assay, and low cell viability (<75%) upon dsRNA treatment is indicated by a black border around the circles (complete cell viability dataset is provided in Supplementary Fig. S3). DENV, dengue virus; dseGFP, dsRNA targeting eGFP; LACV, La Crosse virus; PFU, plague-forming unit; WNV, West Nile virus.

In agreement with previous reports, silencing of known mosquito host dependency factors (vATP-c and vATP-ac39, key subunits of the endocytic machinery (73, 74, 75, 76, 109, 110, 111) and Loqs, an RNA-binding protein (72, 112)) strongly inhibited DENV and WNV replication, whereas silencing of AGO2 and Piwi4, two previously described host restriction factors (72, 77, 113), increased DENV replication by 213% and 126%, respectively (Fig. 4C). Silencing of three host targets (MADF [AAEL011301], GINS4 [AAEL024136], and MRPL52 [AAEL023209]) resulted in very low cell viability (<75%) in Aag2-C3 cells (DENV) and was therefore excluded from further analysis (Supplemental Fig. S4A). This approach identified 69 novel regulators of arboviral infections, including 68 proviral and two antiviral genes significantly modulating replication of at least one viral species (|log2[fold change]| ≥2, n = 3) (Fig. 4C and Supplemental Fig. S4, B–D). Notably, 26 of these host factors displayed a functional requirement consistent with their binding specificity (23% of DENV interactors, 56% of WNV interactors, and 20% of LACV interactors).

Among the newly identified DENV host factors, we identified two novel host restriction factors, TP63 (AAEL007595) and MED21 (AAEL010797), and a total of 26 novel proviral host factors critically required for DENV replication. Among these, silencing of BOP1 (AAEL001169), BMS1 (AAEL003555), SMARCC2 (AAEL022608), and selected mitochondrial ribosomal proteins displayed the strongest effects, reducing DENV viral replication up to 94% (Fig. 4C and Supplemental Fig. S4B). In total, 52 targets were identified as proviral factors for WNV replication, with MADF (AAEL011301), GINS4 (AAEL024136), CHCHD1 (AAEL001974), and WDR1 (AAEL014037) exhibiting the strongest effects (Fig. 4C and Supplemental Fig. S4C). Finally, 11 host targets were identified as novel host dependency factors for LACV, including WDR1 (AAEL014037), ICT1 (AAEL010339), SNRNP70 (AAEL011340), PNO1 (AAEL011832), PPAN (AAEL009309), and SNRPB (AAEL002377) (Fig. 4C and Supplemental Fig. S4D).

Amongst the newly identified host dependency and host restriction factors, 16 displayed DENV-specific effects, 33 were WNV specific, and three exhibited LACV-specific phenotypes, with a moderate degree of conservation between orthoflaviviruses (DENV and WNV: 11 shared host dependency factors). Notably, three host dependency factors (PNO1 [AAEL011832], WDR1 [AAEL014037, and SNRPB [AAEL002377]) exhibited proviral phenotypes for all three viruses, suggesting an evolutionarily conserved requirement for nuclear ribosome biogenesis, regulation of cytokinesis, and spliceosome activity in arbovirus replication in mosquitoes (Fig. 4C and Supplemental Fig. S4E).

Altogether, the combined results of the phenotypic screen with detailed information on interspecies binding specificity provide a novel set of mosquito host dependency factors for multiple arboviruses and experimental evidence for their functional role across different arboviral genera.

The Brahma Complex Exerts a Broadly Conserved Proviral Role in Orthoflavivirus and Orthobunyavirus Replication

Amongst the host proteins displaying the strongest proviral effects, we identified two components of the Brahma complex, ARID2 (AAEL001361) and SMARCC2 (AAEL022608), to be critically required for replication of both DENV and WNV, with a mean viral titer reduction of 88% upon silencing (Fig. 4C and Supplemental Fig. S4, B–D). The Brahma complexes, also known as BAP in Drosophila or SWI–SNF complex family BAF chromatin–remodeling complexes in humans, are ATP-dependent chromatin-remodeling complexes highly conserved in vertebrates and invertebrates, with individual complex members bearing different names across species (Fig. 5A). Detailed information on the composition and molecular function of Brahma-like subcomplexes in mosquitoes is currently elusive. Therefore, we used the subcomplex composition of the closest and most-studied Drosophila spp. orthologs as a blueprint for further mechanistic studies (Fig. 5, A and B). In Drosophila, two distinct subcomplexes have been characterized: the BAP and PBAP complex (Fig. 5B). Both complexes contain a BAP core consisting of a Moira dimer, an ATPase module (BRM, ACT, and BAP55), and three accessory proteins (BAP111, SNR1, and BAP60). However, the BAP complex features the additional subunit OSA (an ARID-domain–containing protein), whereas the PBAP complex is defined by the presence of BAP170, Polybromo, and SAYP. Brahma complexes play important functions in cell cycle regulation, development, and differentiation and are essential transcriptional regulators of immune and DNA damage responses (114, 115, 116, 117, 118). While both subcomplexes have partly redundant functions, including regulation of NF-κB-mediated innate immune responses, distinct functions of the individual complex subtypes have been reported, including regulations of selected interferon-responsive genes (118, 119, 120, 121, 122) and transcriptional activation through nuclear receptors (123, 124).Fig. 5**Functional relevance of Brahma complex subunits in arbovirus replication.**A, summary of Brahma complex subunit names in humans (Homo sapiens) and Drosophila melanogaster and indication of presence in BAP or PBAP subcomplexes (155, 156, 157). B, schematic representation of the Drosophila BAP and PBAP complex compositions. This illustration was created in BioRender. Flory, C. (2026) https://BioRender.com/2hph5gn. C, effect of dsRNA knockdown of Moira and BAP170 in Aag2-C3 cells on the production of different orthoflaviviruses (DENV, WNV, USUV, ZIKV, and YFV), orthobunyavirus (LACV), and alphaviruses (CHIKV, SINV) was assessed by plaque assay (mean ± SD; n = 3–4). D and E, effect of dsRNA-mediated knockdown of different Brahma complex subunits on intracellular viral RNA levels (D) and virus titers (E) of WNV, DENV, and YFV in Aag2-C3 cells. Data underlying dsvATP-C, dsBAP170, and dsMoira silencing on YFV replication by plaque assay in C were again shown here for ease of comparison. Asterisks indicate significance as assessed by two-way ANOVA with Dunnett's multiple comparisons test (p: 0.12 [not significant [ns], 0.033 [∗], 0.002 [∗∗], 0.001 [∗∗∗]; mean ± SD; n = 3–4). BAP, benzo[a]pyrene; CHIKV, Chikungunya virus; DENV, dengue virus; dseGFP, dsRNA targeting eGFP; LACV, La Crosse virus; PBAP, primary biological aerosol particle; PFU, plaque-forming unit; SINV, Sindbis virus; USUV, Usutu virus; WNV, West Nile virus; YFV, yellow fever virus; ZIKV, Zika virus.

Since we identified BAP170/ARID2 (AAEL001361) and Moira/SMARCC2 (AAEL022608) as functional capsid-interacting nodes required for productive DENV and WNV infection, we further assessed their relevance across a broader spectrum of orthoflaviviruses (DENV, WNV, USUV, ZIKV, and YFV), orthobunyavirus (LACV), and alphaviruses (CHIKV and SINV) (Fig. 5C and Supplemental Table S4). Silencing of the two Brahma complex subunits led to a significant reduction in LACV titers, as well as all orthoflaviviruses tested, with the notable exception of YFV, being unaffected or displaying a significant increase in virus replication (Fig. 5C). Conversely, alphavirus replication was largely unaffected (SINV and CHIKV) (Fig. 5C), suggesting a broadly conserved role of the Brahma complex in orthobunyavirus and orthoflavivirus replication.

To characterize which of the two putative Brahma complexes plays a role in orthoflavivirus replication, individual subunits unique to each Brahma subcomplex (OSA, Polybromo, BAP170, and SAYP) or shared between the two complexes (Moira, BRM, and SNR1) were tested for their relevance in WNV, DENV, and YFV replication (Fig. 5, D and E). Robust silencing of Moira, SAYP, SNR1, and OSA upon dsRNA-mediated KD was confirmed at mRNA and protein levels by qRT–PCR and quantitative LC–MS/MS, without overt cytotoxic effects (Supplemental Fig. S5, A–C). Interestingly, silencing of all four subunits resulted in significantly reduced intracellular viral RNA levels (Fig. 5D, Supplemental Fig. S5A) as well as infectious titers (Fig. 5E and Supplemental Table S4) of both DENV and WNV, suggesting that both subcomplexes have a functional role in viral replication. Conversely, YFV replication was largely unaffected or mildly increased across all conditions (Fig. 5, D and E), indicating that YFV might have evolved alternative, Brahma-independent mechanisms for transcriptional regulation of host genes.

Overall, these results indicate novel proviral roles of both BAP and PBAP-like complexes in the replication of specific orthoflaviviruses.

Moira Interacts With Capsid in a Nucleic Acid–Dependent Manner

To further characterize the Moira–capsid interaction in the context of productive viral infection, we engineered a full-length Moira construct carrying an N-terminal FLAG-tag (FLAG-Moira). IF analysis of transfected Aag2-C3 cells confirmed the predicted nuclear localization of Moira, which was not significantly altered in virus-infected cells (Supplemental Fig. S6A). Furthermore, we confirmed nuclear colocalization of ectopically expressed HA-tagged capsids and FLAG-tagged Moira by IF assay (Supplemental Fig. S6B). Importantly, capsids of both WNV and DENV coimmunoprecipitated with Moira in virus-infected Aag2-C3 cells, confirming a specific interaction (Fig. 6, A and B). Interestingly, this interaction was completely abrogated in the presence of nucleases, suggesting an essential role of nucleic acids in mediating the Moira–capsid interaction (Fig. 6C). Altogether, these results confirm Moira as a bona fide capsid-interacting host protein and provide biochemical evidence that the interaction requires nucleic acids.Fig. 6**Biochemical characterization of Moira–capsid interaction.**A–C, co-IP–WB (anti-FLAG) of FLAG-Moira- or FLAG-eGFP-transfected Aag2-C3 cells that were WNV-infected (48 hpi; A), DENV-infected (72 hpi; B), or WNV-infected (48 hpi) and nuclease treated upon cell lysis (C). DENV, dengue virus; eGFP, enhanced GFP; hpi, hours postinfection; IP, immunoprecipitation; WB, Western blot; WNV, West Nile virus.

Orthoflavivirus Capsids Remodel the Mosquito Host Chromatin Landscape

Brahma complexes regulate several cellular pathways and have been previously reported to influence viral replication of large DNA viruses and retroviruses, by facilitating latency and viral transcription (125, 126, 127, 128, 129, 130), as well as that of positive-sense RNA viruses such as severe acute respiratory syndrome coronavirus 2 through transcriptional regulation of viral surface receptor abundance (125, 131). To assess whether orthoflaviviral capsids modulate chromatin accessibility through interactions with the Brahma complex, we systematically analyzed changes in global accessibility of genomic DNA (ATAC-Seq) and gene expression (RNA-Seq) in virus-infected, capsid-overexpressing, and Moira KD Aag2-C3 cells. Efficient silencing of Moira (dsMoira) and comparable infection rates across DENV-, WNV-, and YFV-infected Aag2-C3 cells as well as comparable transfection rates of pPUB-HA-capsid-transfected Aag2-C3 cells were confirmed by RT–qPCR (Supplemental Fig. S7A) or IF (Supplemental Fig. S7, B–D), respectively. In addition, principal component analysis of ATAC-Seq and RNA-Seq datasets further confirmed a good level of correlation and clustering of biological replicates (Supplemental Fig. S7, E and F).

Consistent with functional inhibition of Brahma complex activity, silencing of Moira expression led to major alterations in the chromatin landscape, with 3280 DA peaks (Fig. 7A). Most of the upregulated peaks were located in promoter regions (approximately 45%) within 1 kb distance to the TSS (Supplemental Fig. S7G and 7B), whereas a smaller proportion of peaks located in distal intergenic regions (25%, Fig. 7B). Conversely, significantly less accessible peaks were mostly found in TSS-distal sites (>10 kb) (Supplemental Fig. S7G), introns, or distal intergenic regions (Fig. 7B), confirming the ability of Moira to modulate the chromatin landscape in mosquito cells.Fig. 7**Chromatin accessibility in Orthoflavivirus-infected, capsid-overexpressing, and Moira knockdown (KD) Aag2-C3 cells.**A, Violin plot of log2-transformed normalized read counts (merged three biological replicates) for all differentially and no-differentially accessible peaks per sample. dsMoira was compared with dseGFP, capsids were compared with HA-eGFP, and infection samples were compared with mock. Numbers at the bottom indicate differentially regulated ATAC-Seq peaks. B, relative distribution of upregulated or downregulated peaks across different genomic features. C, overlap between differentially expressed genes (RNA-Seq) and genes associated with differential changes in chromatin accessibility (ATAC-Seq) for each comparison. Genes were assigned to the peaks using the nearest approach. D, intersection of common genes differentially expressed in Moira KD (dsMoira), WNV capsid, and WNV-infected Aag2-C3 (RNA-Seq). E, heatmap of log2(fold change) of 66 differentially expressed genes significantly regulated in both WNV-C expressing- and WNV-infected cells and regulated in a similar manner. Fifty-four genes are upregulated and 12 downregulated. F, directional mean log2(fold change expression) of the six shared genes across RNA-Seq consistently regulated in WNV capsid and WNV infection but inversely modulated in Moira KD (dsMoira) cells. Numbers of nearby differentially accessible peaks (nearest approach distance 50 kb) are displayed on the top. G, HOMER p values of transcription factors found linked to enriched motifs in Moira KD, DENV capsid and infection, WNV capsid and infection, absent in YFV capsid and infection conditions (HOMER Fisher's exact test). H, enriched binding motif and corresponding transcription factors. Numbers of differentially expressed genes (RNA-Seq) associated with peaks displaying the enriched motifs are shown. I, heat map of log2(fold change expression) of selected genes with identified motifs that are either shared pairwise or tripletwise across Moira KD, WNV capsid, and WNV infection. ns, nonsignificant (DESeq2 Wald's test). ATAC-Seq, assay for transposase-accessible chromatin sequencing; dseGFP, dsRNA targeting eGFP; eGFP, enhanced GFP; HA, hemagglutinin; WNV, West Nile virus; YFV, yellow fever virus.

Notably, ectopic expression of either DENV, WNV, or YFV capsid alone was sufficient to trigger significant changes in chromatin accessibility, with more than 2000 DA peaks quantified in each condition (Fig. 7A). Interestingly, approximately 20% to 30% of DA peaks in both capsid-expressing or virus-infected cells (DENV, YFV, or WNV) were assigned to promoter regions, suggesting that persistent capsid expression might modulate gene expression via perturbation of the chromatin landscape (Fig. 7B).

To assess whether changes in chromatin accessibility drive changes in differential gene expression, matched samples were analyzed by RNA-Seq analysis, and results were overlayed on DA sites at the chromatin level. This approach revealed significant overlaps between differentially regulated chromatin regions and abundance in transcripts of proximal genes (Fisher’s exact test, p values: Moira KD: 1.529326e-18, WNV capsid: 0.001634847, and WNV infection: 5.691283e-37; Fig. 7C). Interestingly, WNV capsid expression and WNV infection triggered DE of a large number of transcripts (323 and 3750 genes, respectively), with 10% to 15% consistently linked to DA chromatin regions (Fig. 7C and Supplemental Fig. S8, A–C).

To identify direct links between capsid-mediated chromatin remodeling and Moira complex activity, we intersected differentially expressed genes across the three datasets and identified 997 genes consistently regulated in both WNV infection and Moira KD–regulated genes, with functions associated with transmembrane transport, endopeptidase activity, and catalytic activity (Fig. 7D and Supplemental Fig. S8D). Among these, 66 genes were regulated in both virus-infected or capsid-transfected cells, corroborating the notion that ectopic capsid expression drives differential gene expression of host targets in virus-infected cells (Fig. 7E). Among these transcripts, we identified genes associated with intracellular traffic and transport (Tango1 [AAEL022091]), lipid metabolism (ATP8B [AAEL019531]), spin ([AAEL008356]), protein translocation (SEC63 [AAEL025151]), and ER–Golgi protein transport (Fig. 7E and Supplemental Fig. S8E).

Altogether, intersecting ATAC-Seq and RNA-Seq results identify direct links between changes in chromatin accessibility (downregulated or upregulated peaks) and modulation of gene expression, providing direct hints about capsid-mediated modulation of host gene expression.

We next used a two-pronged approach to further assess causal relations between capsid-mediated changes in host transcript abundance and Brahma complex activity.

The first approach relied on the assumption that capsid expression in transfected or virus-infected cells modulates expression of selected genes via Brahma complex activity, and thus effects in Moira KD cells should be inversely mirrored in virus-infected and capsid-expressing cells. To this end, we selected genes consistently modulated in virus infection and capsid expression but displaying a reciprocal regulation in Moira KD cells (Fig. 7F). This approach identified six genes, of which four targets (protein patched [AAEL002850], zinc finger protein 704 [AAEL003014], bgm [AAEL019523], and protein FAM13A [AAEL026075]) displayed differential chromatin accessibility in nearby regions in the ATAC-Seq dataset, suggesting a potential capsid-mediated Brahma complex–dependent regulation of gene expression on selected targets (Fig. 7F and Supplemental Fig. S8F).

Since infection and capsid binding may also affect Brahma complex activity through modulation of its target gene specificities, the second approach relied on the identification of Brahma complex–associated motifs differentially enriched in virus-infected and capsid-transfected cells. Through this approach, we first identified 83 common motifs enriched across all conditions (Moira KD, WNV, and DENV infection, and capsid overexpression; Supplemental Table S5), including eight motifs exclusively enriched in Moira KD, WNV, and DENV infection, and capsid overexpression when compared with the YFV counterparts (Fig. 7G). Next, we investigated whether any of the differentially expressed genes in the RNA-Seq data contained one or more of these motifs (PHOX2A, ISRE, HAT2, ELT-3, DEAR2, and ATHB7) (Fig. 7H and Supplemental Fig. S8G). This approach identified two target genes, which were differentially expressed in both WNV infection and capsid expression: ER-protein Calumenin B scf (AAEL007725) and glycogen synthase kinase 3 beta homolog sgg (AAEL005238) (Fig. 7I). Among the other potentially relevant targets, we additionally identified a few genes, which displayed reciprocal effects in Moira KD and WNV infection. These included the major facilitator superfamily transporter 3 MFS3 (AAEL012522), serine/threonine receptor kinase tkv (AAEL012125), inositol hexakisphosphate kinase Ip6k (AAEL000275), putative C9orf172 AJM1 (AAEL018320), Arf GTPase activating protein drongo (AAEL003534), cyclin-dependent kinase Eip63E (AAEL012914), and glutamine synthetase Gs1 (AAEL001887).

Overall, these experiments demonstrate that ectopic expression of orthoflaviviral capsid selectively modulates gene expression and identify 66 Ae. aegypti transcripts that are regulated in virus-infected cells via a capsid-dependent mechanism. Importantly, using a combination of complementary sequencing approaches, we identified a subset of 15 differentially regulated genes, which could be linked to orthoflavivirus capsid expression and Brahma complex activity, suggesting a novel function of orthoflavivirus capsid in host chromatin remodeling.

Discussion

In this study, we leveraged a systematic AP–MS/MS approach to profile the interactions of 12 different arboviral capsids with the mosquito proteome. This extended atlas of capsid-interacting proteins significantly expands prior knowledge on arbovirus–vector interactions, allowing systematic mapping of unique and conserved interactions across different arboviral species. This large network revealed broadly conserved targeting of cellular pathways involved in ribosome biogenesis, RNA processing, chromatin remodeling, and nuclear transport, generalizing previous reports on individual viruses such as ZIKV (51) and DENV (52) (Supplemental Fig. S1G). In agreement with previous virus–host interaction studies in human cellular backgrounds, arboviral capsids specifically target (post-)transcriptional regulators, hijacking host processes involved in translation initiation, spliceosome assembly, and nuclear transport and suggesting a core of evolutionarily conserved proteins targeted by arboviruses in both human and vector hosts. As an example of cross-species capsid-interacting proteins, we identified here a conserved antiviral role for the TP53-protein family member (AAEL007595), a mosquito ortholog of the human p53 protein previously described to trigger apoptosis in response to DENV-C in mammals (25) (Figs. 1C and 4C). Similarly, we observed a conserved proviral role for alpha-galactosidases (NAGA [AAEL005460]) and syndecan (SDC2 [AAEL013969]) in DENV and WNV replication in Ae. aegypti cells, in agreement with previous reports describing their human counterparts as orthoflaviviral host dependency factors (52, 132, 133, 134) (Fig. 4C). We also confirmed that in mosquitoes, capsid binds to GADD45GIP1 (AAEL005146)—a highly conserved capsid interactor in mammals (14, 45, 47, 52)—further demonstrating that it functions as a host dependency factor across orthoflaviviruses. Altogether, these pan-viral host factors could represent promising targets in both therapeutic and vector control settings.

Overall, our phenotypic dsRNA-mediated screen assessing the functional relevance of 110 conserved and unique arboviral capsid-interacting proteins across three viral genera identified two DENV host restriction factors as well as 26 (DENV), 11 (LACV), and 52 (WNV) host dependency factors, uncovering a completely novel group of mosquito host factors required for productive alphavirus, orthobunyavirus, and orthoflavivirus replication (Fig. 4, Supplementary Fig. S3). In agreement with previous studies using PPI data as entry points for phenotypic screens, the capsidome achieved a higher efficiency in functional nodes identification compared with genome-wide screens, highlighting the value of PPI networks in streamlining de novo viral target discovery (45, 135, 136, 137).

Among these, we further revealed a proviral role for the Piwi-interacting RNA pathway member FKBP6 in DENV replication, as well as conserved pan-viral functions for three poorly characterized mosquito proteins (PNO1 [AAEL011832], WDR1 [AAEL014037], and SNRPB [AAEL002377]) in DENV, WNV ,and LACV replication (Fig. 4C). Moreover, these experiments provide additional cues into a number of mosquito proviral factors previously reported for individual viral species, generalizing the role of Loqs (AAEL008687), a dsRNA-binding protein reported to play essential roles in the replication of DENV, YFV, and ZIKV (72, 112), and the endocytic machinery (vATPase subunits vATP-ac39 [AAEL021342] and vATP-C [AAEL000291]) (73, 74, 76, 109, 110, 111) as pan-orthoflavivirus host dependency factors in mosquito cells (Fig. 5C). In agreement with earlier reports, AGO2 significantly restricted DENV replication (72, 113), while exhibiting no effect on WNV and LACV replication, highlighting the heterogeneity of the RNAi response across different arboviruses in Ae. aegypti cells (53). Our ongoing studies will further elucidate the relationship between each of these arboviral capsids and their targets, uncovering additional mechanisms by which each species manipulates the mosquito host cell machinery.

Among the newly identified PPIs, we focused on the interaction of orthoflaviviral capsids with two components of the Brahma complexes (also known as (P)BAF or mSWI–SNF complexes), BAP170/ARID2 (AAEL001361) and Moira/SMARCC2 (AAEL022608), and showed that the Brahma complex plays a critical role in orthoflavivirus replication in mosquito cells.

The physiological function and exact composition of the Brahma complex in mosquito cells are currently elusive. In humans, three main subtypes of BAF complexes have been identified, whereas in Drosophila, two distinct subcomplexes (BAP and PBAP) have been described to date, with shared and distinct roles in cell cycle control, specific immunoregulatory functions, and transcriptional activation through nuclear receptors being characterized in both species (114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124). In this study, through a combination of genetic, biochemical, and sequencing approaches, we provide multiple lines of evidence demonstrating a chromatin-remodeling function of the Brahma complex in mosquito cells. Notably, we show that both putative Brahma subcomplexes are critically required across arboviruses, including those not primarily transmitted by Ae. aegypti in endemic areas (WNV, USUV, and SINV), and confirm that this interaction occurs upon ectopic capsid expression as well as in virus-infected cells (Fig. 6A)

To our knowledge, this is the first study describing a functional role for the Brahma complex in arbovirus replication in invertebrates. However, specific hijacking of cellular pathways involved in chromatin remodeling has been previously described in the context of infections by positive-sense RNA viruses in mammalian cells. For instance, the human BAF45b, a subunit of the neuronal BAF complex, was identified as a pan-orthoflavivirus host dependency factor, although the underlying molecular mechanisms and viral determinants remain elusive (138). Similarly, SMARCA4 (human ortholog of BRM), ARID1A (human ortholog of OSA), SMARCE1 (human ortholog of BAP111), SMARCB1 (human ortholog of SNR1), and SMARCC1 (human ortholog of Moira) were also reported as severe acute respiratory syndrome coronavirus 2 host dependency factors in mammalian cells (131, 139). Interestingly, silencing of Brahma complex components did not affect YFV and SINV replication, suggesting that some orthoflaviviruses and alphaviruses may have evolved alternative mechanisms to regulate host gene expression, such as histone mimicry (31) or capsid-induced transcriptional shutoff (35, 40, 140, 141, 142, 143).

Biochemical characterization of the molecular determinants underlying the capsid–Moira interaction revealed a nucleic acid dependency (Fig. 6). This suggests that interaction of Moira with the orthoflaviviral capsids could be directly mediated by nucleic acids or might require an intact Brahma complex, as complex stability and structure are strongly modulated by the presence of nucleic acids (144, 145, 146).

To shed light on the functional consequences of the capsid–Moira interaction on host gene expression, we systematically profiled changes in the chromatin landscape and cellular transcriptome using a combination of high-throughput sequencing approaches. These experiments revealed profound modulation of gene expression and chromatin accessibility in both virus-infected and Moira-depleted cells (Fig. 7). Interestingly, we identified 66 genes consistently regulated at the transcriptional level in both virus infection and ectopic capsid expression, suggesting a direct role of orthoflaviviral capsid in the modulation of gene expression. These genes include the Notch-ligand neurogenic locus protein delta (AAEL025606), previously shown to be dysregulated upon ZIKV infection in human neuronal progenitor cells (147) and the Tango1 gene (AAEL022091), involved in ER–Golgi transport (148) and potentially linked to virus assembly and release. Furthermore, among the dysregulated genes, we identified Sec61 and other accessory components of the translocon machinery, previously described as a DENV host dependency factor in mammals (52). Moreover, the lipid transport factors, ATPase phospholipid transporting 8B (AAEL019531), the lysolipid transporter protein spinster (AAEL008356), as well as M13 family metallopeptidase neprilysin 3 (AAEL011369), were amongst the newly identified targets significantly upregulated by capsid expression. Notably, overall transcriptional changes observed upon capsid expression were milder than those observed in virus-infected cells. These differences might result from reduced capsid expression levels in transiently transfected cells or the presence of additional viral or host proteins differentially regulated upon productive virus infection. Collectively, the moderate impact of capsid expression on global proteostasis (Fig. 3) and the coherent regulation of cellular transcripts in virus-infected and capsid-transfected cells suggest that orthoflaviviral capsids modulate gene expression at least in part through interactions with the Brahma complex in the nucleus (Supplementary Fig. S5, Fig. 7D).

To identify causal relations between capsid-mediated changes in infection and Brahma-regulated genes, we intersected dysregulated genes at the RNA level with DA chromatin regions in Moira-silenced, virus-infected, and capsid-expressing cells. This approach led to the identification of six genes potentially regulated by the Brahma complex upon interaction with the capsid. These include upregulation of the Hedgehog signaling pathway member protein patched (AAEL002850), the transcription factor zinc finger protein 704 (AAEL003014), the acyl-CoA synthetase bubblegum family member 1 (AAEL019523), and protein FAM13A (AAEL026075). Interestingly, the Hedgehog signaling pathway has been implicated in suppression of host immune responses upon RNA and DNA virus infections (149), suggesting a possible proviral role for this gene in orthoflavivirus infections in mosquitoes.

To characterize further potential differences in Brahma complex activity upon association with capsid, we additionally performed transcription factor binding motif enrichment, focusing our analysis on genes displaying DA peaks in Moira KD, DENV, and WNV capsid expression and virus infection. As specificity control, we used YFV-infected cells, which did not exhibit any Moira-dependent phenotypes in our study (Fig. 5, D and E). This approach allowed the identification of capsid-regulated, Brahma complex–associated genes that might influence the orthoflaviviral replication cycle, including the ER-localized Calumenin B scf (AAEL007725) and glycogen synthase kinase 3 beta homolog shaggy (AAEL005238). In line with our findings, pharmacological inhibition of the human GSK-3 ortholog has previously been shown to negatively affect DENV replication and may also modulate the nuclear localization of JEV capsid (150, 151).

Altogether, the identification of host genes linked to Brahma complex activity and coherently regulated upon virus infection and capsid expression suggests that orthoflavivirus capsids might interfere with Brahma complex–mediated chromatin remodeling, modulating the expression of selected host genes with proviral functions. Further studies will be needed to explore the functional consequences of the Brahma complex for orthoflavivirus replication in mosquito cells.

Collectively, this study provides a comprehensive network to rationalize arboviral capsid effector functions in mosquito cells, shedding light on a new set of invertebrate host proteins functionally promoting replication of prototypic arboviruses in insects. Furthermore, we identified a novel and conserved role for the orthoflavivirus capsid in the modulation of gene expression and chromatin remodeling, exemplifying the versatility of viral capsids’ functions in virus replication in invertebrates.

Data Availability

The MS-based proteomics data have been deposited at the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository with the following dataset identifiers: PXD064586 (capsidome interacting data) and PXD070116 (dsRNA-KD cell dataset). The corresponding MS-Viewer (152) annotated spectra can be accessed via the following search keys and URLs: 7mjjzswdyn (URL: https://msviewer.ucsf.edu/prospector/cgi-bin/mssearch.cgi?report_title=MS-Viewer&search_key=7mjjzswdyn&search_name=msviewer) and lvxlrtcjwp (URL: https://msviewer.ucsf.edu/prospector/cgi-bin/mssearch.cgi?report_title=MS-Viewer&search_key=lvxlrtcjwp&search_name=msviewer). ATAC-Seq and RNA-Seq data have been deposited to the European Nucleotide Archive repository with the following identifier: PRJEB89854. The code used for the ortholog mapping is available on https://github.com/ntnn19/orthologue_mapping_snakemake.

Supplemental Data

This article contains supplemental data (14, 45, 47, 51, 52, 72, 76, 80, 153, 154).

Conflict of Interest

The authors declare no competing interests.