MainInfluenza A virus (IAV) remains a global health threat1. Its replication relies on protein–protein interactions (PPIs) between up to 14 viral proteins2 and host factors, often confined to cellular compartments and organelles. Understanding these host–IAV PPIs in context is essential for elucidating viral strategies and therapeutic targets.Viral RNA (vRNA) synthesis is confined to the nucleus and catalysed by the vRNA-dependent RNA polymerase complex (RdRp; PA, PB1 and PB2) within viral ribonucleoproteins (vRNPs), in which RdRp and vRNA wrap around nucleoprotein (NP) filaments2. Transcription and replication also depend on recruited host factors. Nuclear vRNA processing is modulated by context-specific PPIs between NS1 and NP and host RNA-binding proteins3 that remodel nuclear speckles4. The PA-X nuclease, expressed from an alternative PA reading frame5, degrades nuclear host transcripts6 that organize subnuclear compartments in uninfected cells7. How protein and RNA interactions in infected cells influence subnuclear compartments and replication remains poorly understood. Beyond the nucleus, vRNPs are exported by viral proteins M1 (ref. 8) and NEP9, trafficked via host RAB11A protein10 at the reorganized endoplasmic reticulum (ER)11, while viral glycoproteins haemagglutinin (HA) and neuraminidase (NA) are synthesized, glycosylated and trafficked through the ER and Golgi12 before HA, NA, vRNPs, M1 and the ion channel M2 assemble into budding virions at the plasma membrane.Systematic native IAV–host PPIs maps are lacking. Existing affinity purification–mass spectrometry (AP–MS) and yeast two-hybrid studies13,14,15 rely on lysed cells, disrupt native cellular architecture, often use non-infected cells and provide no structural information on interaction sites, which remain largely uncharacterized at the atomic resolution for IAV–host complexes.Here, we applied in-cell cross-linking mass spectrometry (XL-MS)16 combined with AlphaFold-based structural modelling and functional analysis to map IAV–host PPIs in infected human lung epithelial cells, yielding residue-to-residue contact sites (within 40 Å, based on the distance constraint of the disuccinimidyl sulfoxide (DSSO) cross-linker used) across hundreds of viral–host pairs.Integrating XL-MS with structural modelling and functional screens, we uncovered previously unrecognized virus–host interactions and host factors, compartment-specific HA contacts along the ER–Golgi maturation pathway, identified LAT1 as a membrane-associated M2 interactor and revealed that IAV disrupts paraspeckles. We propose that NP and NS1 interactions, PA-X-mediated NEAT1_2 degradation and RNA polymerase II (Pol II) inhibition drive this disassembly. Together, our findings provide a spatially resolved snapshot of IAV–host contact sites and reveal mechanisms of host subversion.ResultsMapping IAV–human contact sites using in-cell cross-linkingTo map PPIs and contact sites in cells infected with the A/WSN/33 (H1N1) strain of IAV (hereafter, WSN), we applied structural host–virus interactome profiling (SHVIP)16, a recently developed approach that combines in-cell XL-MS with bioorthogonal unnatural amino acid labelling to enrich newly synthesized viral proteins and overcome the sensitivity limits of conventional in-cell XL-MS (Fig. 1a).Fig. 1: In-cell cross-linking during IAV infection.Full size imagea, Schematic of in-cell cross-linking using SHVIP. IAV-infected cells are labelled with HPG from 5 hpi to 14 hpi, followed by DSSO cross-linking. After quenching, L-HPG-incorporated proteins are enriched via click chemistry, digested with trypsin and cross-linked peptides are analysed by MS. b, Virus-centric cross-linking network at a 2% FDR threshold. Viral proteins are shown in orange, and host proteins are coloured by their functional category. Only host proteins linked to viral proteins are shown with names omitted for clarity. The width of each circle is proportional to the protein sequence length. The network including host names is shown in Extended Data Fig. 2. c, Interaction between PB2 and RAB11A/B, consistent with the established binding domain of PB2. Interlinks are shown in green, while intralinks are omitted for clarity. d, Cross-links between NP and established host factors involved in influenza virus infection. Colour shading represents distinct structural domains, with names shown only for those forming cross-links for clarity. Interlinks are shown in green, while intralinks are omitted for clarity. e, Cross-links between subunits of vRdRp. Interlinks are shown in green, while intralinks are in violet. Numbers in c–e indicate amino acid residue positions in the protein sequence. Diagrams in a created in BioRender; Kotova, I. https://biorender.com/uzohrtj (2026). C-terminal, carboxy-terminal; N-terminal, amino-terminal.SHVIP labels viral proteins during host shutoff, when viral synthesis dominates16. As IAV translation peaks early and declines later17, we tested labelling windows and cross-linking times (Extended Data Fig. 1a–e) and chose labelling from 5 hours post-infection (hpi) to 14 hpi with DSSO cross-linking at 14 hpi (Methods and Extended Data Fig. 1c). This captures late-stage vRNP export, cytoplasmic transport and virion assembly18 without affecting cell viability or infection progression (Extended Data Fig. 1f–h).To confirm SHVIP enriches the viral proteome, we compared host and viral protein abundance in enriched versus input samples. Viral proteins were tenfold more abundant across three replicates (Extended Data Fig. 1i–l). XL-MS detected protein pairs and cross-links 2–10× more frequently (Extended Data Fig. 1m–o), with 47% reproducibility across replicates, matching previous studies19 and the SHVIP HSV-1 dataset16 (Extended Data Fig. 1p–r).Extensive in-cell cross-linking networkWe analysed raw data from 3 biological SHVIP replicates using a 2% false discovery rate (FDR) threshold at cross-link spectrum match, residue-pair (Supplementary Table 1a) and PPI levels (Supplementary Table 1b). This dataset included 13,588 unique residue-pair cross-links (Supplementary Table 1c) from 2,015 proteins, including 9,613 intraprotein and 3,975 interprotein cross-links. A total of 110 viral–viral and 867 viral–host cross-links were identified, involving 9 viral proteins, 139 host proteins and 198 viral–host protein pairs (Fig. 1b and Supplementary Table 1d), plus 2,626 interprotein cross-links for 914 human protein pairs (Supplementary Table 1c).At least 50 viral–host pairs involved host proteins known to bind viral proteins (Supplementary Table 2 and Extended Data Fig. 2). Cross-links also involved host factors known to function in the IAV infection cycle, including RAB11A18 (Fig. 1c), KHSRP20, TDP-43 (TARDBP)21 (Fig. 1d), ANP32A, ANP32B22, nucleosomes23, microtubules24 and chaperonins25. Among human pairs, 771 (84%) are BioGRID physical interactions (Supplementary Table 2b), supporting dataset quality.The interactome is enriched for functional Gene Ontology terms linked to gene expression, RNA processing, splicing, transport, unfolded protein response, protein synthesis, ribosome biogenesis, chromatin organization, calcium homeostasis and phosphorylation (Fig. 1b, Extended Data Fig. 2 and Supplementary Table 3), and localizes to nuclear and cytoplasmic compartments, especially the nucleolus, cytoskeleton and vesicles (Supplementary Table 3), consistent with previous influenza virus PPI studies13,26.Compared with previous IAV PPI studies, SHVIP showed a more specific localization pattern for viral proteins (Extended Data Fig. 3a) and cross-linked host proteins (Extended Data Fig. 3b) than reported by ref. 13 (Extended Data Fig. 3c), ref. 15 (Extended Data Fig. 3d) and ref. 26 (Extended Data Fig. 3e), with host proteins localizing to compartments matching the viral proteins. The earlier PPI datasets often report incompatible localizations, including widespread mitochondrial interactors for nearly all viral proteins, despite clear evidence for PB2 only (ref. 27), and numerous nuclear interactors for ER- and plasma-membrane-associated proteins such as M2, HA and NA. Such inconsistencies may reflect infection-induced localization changes or non-physiological interactions from overexpression and biochemical purification. In contrast, our in-cell XL-MS approach directly captures infection-induced proximities.XL-MS also provides distance constraints that corroborate protein structures or guide structural modelling28. Mapping host–host cross-links onto Protein Data Bank (PDB) structures showed 93% of Cα–Cα distances between cross-linked residues fall within the DSSO constraint of 40 Å (Extended Data Fig. 4). For viral complexes such as RdRp, satisfaction rates are difficult to calculate because of conformational and oligomeric heterogeneity29. Nonetheless, the RdRp cross-linking pattern (Fig. 1e) aligns with purified RdRp29. No PDB structures captured resolved IAV–human cross-links, underscoring the difficulty of structurally capturing these interactions. Several IAV–human interactions were modelled confidently using AlphaFold2-multimer30, AlphaFold 3 (ref. 31), and the cross-link-guided protocols AF3x32 and GRASP33 (Extended Data Fig. 5 and Supplementary Table 4). Many pairs could not be modelled, probably because host–pathogen systems lack the co-evolutionary signals and same-species sequence pairing that AlphaFold relies on30.Overall, the overlap of identified cross-links with functional host protein categories, known host factors, IAV-characteristic PPIs and AlphaFold models, together with consistent subcellular localization, demonstrates that SHVIP delivers a structurally informative IAV–host interactome and captures interactions in their native cellular context.In situ validation of selected host–IAV interactions by proximity ligation assayWe validated viral–host interactions in IAV-infected cells by proximity ligation assay (PLA; Duolink) at 14 hpi on cross-linked protein pairs, selected from those for which reliable immunofluorescence-grade antibodies from different species were available. Signal intensity and puncta per cell were quantified (Fig. 2a–g and Supplementary Fig. 1).Fig. 2: In situ validation of selected viral–host proximities by PLA and functional loss-of-function screening.Full size imagea,b, PLA quantification for selected viral–host protein pairs in IAV-infected cells at 14 hpi. PLA signal is shown as total PLA intensity per cell (a) and number of PLA puncta per cell (b) for NP, NS1, HA, M1 and M2. Each dot represents one cell; cells from at least two independent experiments were pooled for visualization. The horizontal lines indicate the median. The same negative control condition was used for all PLA measurements. c–g, Representative PLA images for selected viral–host protein pairs: NP (c), NS1 (d), HA (e), M2 (f) and M1 (g). For each condition, DAPI (nuclei), PLA signal and merged images are shown. Representative images from at least two independent experiments with similar results are shown. Scale bars, 20 µm. See also Supplementary Fig. 1. h, log2-transformed relative luciferase units (log2 (RLU)) normalized to a non-targeting siRNA control (siNT). Bars are coloured by functional category of the host target as in Fig. 1b; RAB11A siRNAs (positive control) are shown in patterned grey. X-axis labels indicate the host gene targeted (format: si(gene)_(siRNA index)). Data are represented as mean ± s.d. from n = 3–5 (depending on the sample, see Source data) independent biological replicates (independent infections; unit of study). *P