Sequential disruption of SPLASH-identified vRNA–vRNA interactions challenges their role in influenza A virus genome packaging

Abstract A fundamental step in the influenza A virus (IAV) replication cycle is the coordinated packaging of eight distinct genomic RNA segments (i.e. vRNAs) into a viral particle. Although this process is thought to be controlled by specific vRNA–vRNA interactions between the genome segments, few functional interactions have been validated. Recently, a large number of potentially functional vRNA–vRNA interactions have been detected in purified virions using the RNA interactome capture method SPLASH. However, their functional significance in coordinated genome packaging remains largely unclear. Here, we show by systematic mutational analysis that mutant A/SC35M (H7N7) viruses lacking several prominent SPLASH-identified vRNA–vRNA interactions involving the HA segment package the eight genome segments as efficiently as the wild-type virus. We therefore propose that the vRNA–vRNA interactions identified by SPLASH in IAV particles are not necessarily critical for the genome packaging process, leaving the underlying molecular mechanism elusive.


INTRODUCTION
Zoonotic influenza A viruses (IAVs) from avian and mammalian species can cause se v ere disease in humans. Typically, such spillov er e v ents ar e rar e and limited to a few infected individuals due to the poor adaptation of zoonotic IAVs to humans ( 1 ). Occasionally, howe v er, zoonotic IAVs emerge that replicate efficiently in humans and cause pandemics ( 2 , 3 ). These pandemic IAVs can arise through genetic reassortment when different IAV strains exchange parts of their segmented genomes in co-infected cells ( 4 ). These novel genome constellations facilitate viral adaptation and lead to pandemics with high morbidity and mortality, as occurred in 1957, 1968 and 2009 ( 5-7 ). Genetic reassortment is thought to be controlled by a coordinated genome packaging process in which eight unique genome segments are selecti v ely incorporated into a viral particle ( 8 , 9 ). Howe v er, despite its central role in the emergence of pandemic IAVs, the molecular mechanisms underlying the genome packaging process are poorly understood.
Many studies have postulated the existence of a vRNA-vRNA interaction network that controls the assembly of a supramolecular complex consisting of eight different vRNPs ( 18 , 31 , 34 , 37-41 ). This has led to the suggestion that TPS form specific vRN A-vRN A interactions between the vRNPs, but despite some supporting evidence ( 32 , 42 ), specific base pairings of the TPS that are critical for genome packaging have not yet been revealed. In contrast, two functional interactions involving vRNA regions beyond the TPS have been identified ( 38 , 40 , 43 ). However, apart from these two validated interactions, the ar chitectur e of the genome packaging interaction network remains unresolved.
The recent de v elopment of techniques combining RNA-RNA cross-linking with next-generation sequencing has enabled high-throughput mapping of vRNA-vRNA contacts in viral particles ( 37 , 38 , 44 ). Using sequencing of psoralen cross-linked, ligated, and selected hybrids (SPLASH) ( 45 , 46 ), Dadonaite et al . demonstrated the existence of complex and redundant vRN A-vRN A interaction networks in purified virions of several IAV strains ( 38 ). Importantl y, a previousl y proposed ( 43 ) vRN A-vRN A interaction between the PB1 and NA segments of H3N2 viruses was confirmed by SPLASH, and its functional importance in IAV genome packaging was v alidated b y m utational anal ysis ( 38 ). Howe v er, it remains unclear whether SPLASH is able to detect functional vRNA-vRNA interactions on a large scale across IAV strains, as a systema tic evalua tion of the detected networks is still lacking.
Here, we systema tically evalua ted the functional significance of SPLASH-identified vRN A-vRN A interactions in the genome packaging process of an H7N7 virus by generating viral mutants lacking se v eral prominent HA segment interactions. In addition, we tested the relevance of SPLASHidentified vRN A-vRN A interactions involving TPS of the HA and NA segments.

Generation of recombinant influenza A viruses
Recombinant SC35M and chimeric PR8 viruses were generated using an eight-plasmid re v erse-genetics system ( 48 ) (Supplementary Table S1). Subconfluent HEK 293T cells in 6-w ell plates w ere transfected with 500 ng of each pHW2000 plasmid using Lipofectamine 2000 (Thermo Fisher Scientific; 11668019) according to the manufacturer's protocol. Six hours (h) after transfection, the Opti-MEM medium was replaced with DMEM containing 0.2% bovine serum albumin (BSA; AppliChem A1391) and 1% pen / strep. After 48-72 h, the viral supernatant was collected and purified by plaque assay on MDCK-II cells. Individual plaques were picked and amplified on MDCK-II cells to generate clonal SC35M virus stocks. For chimeric rPR8:Wy-PB1 / N A and rPR8:Wy-PB1 / N A Ud-sub viruses, the rescue supernatant was first amplified on MDCK-II cells in the presence of 1 l per milliliter (ml) TPCK-treated trypsin (Thermo Fisher Scientific; 20233) and then plaque-purified. Mutations in the viral genome were confirmed by Sanger sequencing. Briefly, 200 l virus supernatant was mixed with 600 l Trizol reagent (Ambion; 15596026) and purified using the Direct-zol RNA Miniprep Kit (Zymo Research; R2050). 5 l of RNA was re v erse-transcribed and amplified using segment-specific primers (Supplementary  Table S1) and the OneStep RT-PCR Kit (Qiagen; 210212). Purified DNA was sequenced by Eurofins Genomics or Genewiz.

Multicycle virus replication kinetics
Confluent MDCK-II cells in 6-well plates were washed with PBS and infected with wild-type or mutant SC35M Nucleic Acids Research, 2023, Vol. 51, No. 12 6481 virus at a multiplicity of infection (MOI) of 0.001 plaqueforming units (PFU) per cell in DMEM containing 0.2% BSA and 1% pen / strep. For chimeric PR8 viruses, the infection medium contained 1 l per ml of TPCK-treated trypsin. Supernatants were collected at 24 h post infection (hpi) and PFU titers were determined by plaque assay on MDCK-II cells.

Measurements of relative HAU-to-PFU ratios
Confluent MDCK-II cells in 6-well plates were washed with PBS and infected with wild-type or mutant SC35M virus at an MOI of 0.001 PFU per cell in DMEM containing 0.2% BSA and 1% pen / strep. Supernatants were harvested at 24 or 36 hpi. PFU titers were determined by plaque assay on MDCK-II cells. Hemagglutination titers were determined by HA assay as described previously ( 49 ). Briefly, chicken erythrocytes (Labor Merk; E-200) were diluted to 0.75% (v / v) in PBS and added to a 1:2 serial dilution of viral supernatant in a 96-round well plate. After 30 to 60 minutes (min) of incuba tion a t room tempera ture, individual w ells w er e monitor ed for hemagglutination. The HA titer (in hemagglutination units (HAU) per 50 l) was the lowest virus dilution that produced hemagglutination. The relati v e log 2 HAU to PFU ratio of the mutant (mut) compared to the wild-type (WT) virus was calculated as follows: lo g 2 HAU to PFU ratio = ( HA U mut − HA U WT ) − ( lo g 2 PF U mut − lo g 2 PF U WT ) .

Relative quantification of packaged vRNAs by RT-qPCR
Confluent MDCK-II cells in 6-well plates were washed with PBS and infected with wild-type and mutant SC35M viruses at an MOI of 0.001 PFU per cell in DMEM containing 0.2% BSA and 1% pen / strep. Supernatants were collected at 24 or 36 hpi. For RNA isolation, 200 l of virus supernatant was mixed with 600 l Trizol and purified using the Direct-zol RNA Miniprep kit. 4 l of RNA was re v erse tr anscribed using r andom hexamer primers and the Re v ertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific; K1621) according to the manufacturer's protocol. 1:100 dilutions of the cDNA products and 400 nM of segment-specific primers (Supplementary Table S1) were used for quantitati v e PCR (qPCR) using the Sensi-Fast SYBR Hi-ROX kit (Bioline; BIO-92020) according to the manufacturer's instructions. Reactions were performed in technical triplicate using the Quant Studio 5 Real-Time PCR System (Thermo Fisher Scientific; A34322). After excluding a packaging defect of the PB2 segment, vRNA levels in the mutants were normalized to those of the wild type and then to the relati v e PB2 vRNA le v els as follows:

SPLASH
SPLASH was performed as previously described ( 38 , 45 , 46 ) with minor modifications. Briefly, confluent MDCK-II cells in T75 flasks were washed with PBS and infected with wildtype or mutant SC35M virus at an MOI of 0.001 PFU per cell in DMEM containing 0.2% BSA and 1% pen / strep. For chimeric PR8 viruses, the infection medium contained 1 l per ml of TPCK-treated trypsin. Supernatants were collected at 24 to 36 hpi, cleared of debris by centrifugation (30 min, 2000 rpm, Eppendorf rotor S-4-104, 4 • C), and concentrated by ultracentrifugation (1.5 h, 25 000 rpm, Beckman Coulter rotor SW32 Ti, 4 • C) through a 30% (w / v) sucrose cushion. Purified virus was added to a 24-well plate and treated with 200 M EZ-Link Psoralen-PEG3-Biotin (Thermo Fisher Scientific; 29986) and 0.01% digitonin (Sigma-Aldrich; D141) for 5 min at 37 • C. The samples were cross-linked with UV light at 365 nm for 20 min on ice in a UVP CL-1000L cross-linker (Analytik Jena). Samples were digested with 0.25 mg / ml of proteinase K (Thermo Fisher Scientific; E00491) in proteinase K buffer (0.5% SDS, 0.1 M Tris-HCl pH 7.5, 50 mM NaCl, 10 mM EDTA) for 20 min at 37 • C with shaking. Samples, were solubilized in Trizol LS reagent (Ambion; 10296028) and chloroform. RNA was extracted from the aqueous phase using the RNA Clean & Concentrator-5 Kit (Zymo Research; R1013). Purified RNA was fragmented using the NEB Next Magnesium RNA Fragmentation Module (NEB; E6150S) at 94 • C for 4 min. RNA fragments were purified using the RNA Clean & Concentrator-5 Kit, and enriched f or biotin ylated RNA by binding to Hydrophilic Streptavidin Magnetic Beads (NEB; S1421S). Briefly, 100 l of bead suspension was washed twice with lysis buffer (50 mM Tris-HCl pH 7, 10 mM EDTA, 1% SDS). The beads wer e r esuspended in 60 l lysis buffer supplemented with 3.5 l per ml of Superase-In (Invitrogen; AM2694), and the purified RNA was added to the beads along with 600 l supplemented lysis buffer and 1200 l hybridization buffer (750 mM NaCl, 1% SDS, 50 mM Tris-HCl pH 7, 1 mM EDTA, 15% formamide). After shaking for 30 min at 37 • C, the beads were washed fiv e times with wash buffer (10% SSC (Sigma-Aldrich; S6639-1L), 0.5% SDS) by shaking for 5 min at 37 • C. To remove 3 -cyclic phosphate groups, the beads were washed twice with cold PNK-T buffer (

Bioinformatical analyses of SPLASH datasets using RNAswarm
The workflow described below is a mixture of automated steps and manual curation. The script package used is called RNAswarm and can be found in our GitHub repository ( https://github.com/gabriellovate/ RNAswarm/r eleases/tag/v0.1.0 ). We first pr e-processed the raw Illumina SPLASH data by trimming adapters and filtering for quality using fastp ( 50 ) (version 0.23.2, default parameters). The corresponding fastqc files ( 51 ) (version 0.11.8, default parameters) before and after trimming are available in our OSF repository ( https://osf.io/berj7/ ). To detect chimeric reads resulting from cross-linking of intraand intersegmental RN A-RN A interactions, we ma pped the pre-processed sequencing data to all eight segments of the corresponding viral reference genomes (Supplementary  Table S1) using segemehl ( 52 , 53 ) (version 0.3.4) with the '-S flag' to enable split-read alignment mode. The trns.txt output files were used as input to our in-house script plot heatma ps.py, w hich selects and counts the intersegmental chimeric reads and generates heatmaps of pairwise segment interactions. Multiple trns.txt files from biological r eplicates wer e used as input to gener ate 'merged' inter action heatmaps. We then extracted interaction clusters from these 'merged' heatmaps by manual inspection. 'Merged' heatmaps and extracted interaction clusters are accessible from the OSF repository. For each interaction cluster, we determined the peak position and selected interaction regions by extending the peak position by 19 nucleotides in each direction. The number of chimeric reads mapping to the resulting 39-by-39 nucleotide window was used to generate a count table of the interaction regions using the in-house script make counttable.py. We normalized these raw chimeric read counts to the total number of mapped chimeric reads using the in-house script get library size.py, r esulting in r eads per million (RPM) values. For comparati v e analysis of two different viruses, the raw chimeric read count tables were used as input to a modified DEseq2 ( 54 ) script called run DESeq2.r. Of note, this analysis r equir es at least two biological replicates and was ther efor e not perf ormed f or samples i-iii shown in Figure 6. To visualize our vRNA interactome data, we used circos ( 55 ) (v0. . RPM values are provided in Supplementary Table S2. DEseq2 data are provided in Supplementary Table S3.

Classification of high-and low-frequency vRN A-vRN A interactions
To categorize the vRN A-vRN A interactions detected in each virus (or condition) into low-and high-frequency contacts, we defined cutoff points using a previously published analytical method ( 56 ). Cutoff points are indicated for all samples in Supplementary Figure S2.

RNA structure prediction and calculation of MFE values
To pr edict RNA structur es and minimum fr ee energy (MFE) values of the 39-by-39 nucleotide interaction regions (described in the previous section), we used RNAcofold from the Vienna RNA package ( 57 ) (version 2.5.0) with '-noLP -C' options and otherwise default parameters. The structur e pr ediction script is available at https://github.com/ desiro/Jakob2023 structur es . MFE differ ences ( MFE) between wild-type and mutant interactions were calculated as follows: MFE mut − MFE WT . MFE and MFE values are provided in Supplementary Table S3.

Prediction of synonymous mutations
Synonymous mutations aimed at disrupting SPLASHidentified vRN A-vRN A interactions wer e pr edicted using Nucleic Acids Research, 2023, Vol. 51, No. 12 6483 the SilentMutations (SIM) tool ( 58 ). Parameters were varied to increase the MFE of the targeted interaction by at least 40% whene v er possib le.

Next-generation-sequencing (NGS) of virus stocks
To sequence viral stocks, RNA was extracted from cell culture supernatants using the Quick-RNA Viral Kit (Zymo Research; R1034). Ribosomal RNA was depleted with RiboMinus Eukaryote Kit v2 Kit (Thermo Fisher Scientific; A15020). P air ed-end libraries wer e then pr epar ed using the NEBNext Ultra II Directional RNA Library Prep Kit for Illumina (NEB, E7760). A total of 8 pM library was sequenced on an Illumina MiSeq instrument (MiSeq Reagent Kit v2, 300 cycles, Illumina; MS-102-2002). The de-multiplexed raw reads were subjected to a custom Galaxy pipeline. Briefly, the raw reads were preprocessed using fastp ( 50 ) (version 0.23.2) with default parameters and mapped to the wild-type SC35M r efer ence genome (Supplementary Table S1) using BWA-MEM ( 59 ) (version 0.7.17) with default parameters. Variants (i.e. SNPs and INDELs) were called using the ultrasensiti v e variant caller LoFreq ( 60 ) (version 2.1.3.1), which r equir es a minimum base quality of 30 and coverage of at least 20-fold. The called variants were then filtered based on a minimum variant frequency of 5% and the strand bias support. Finally, consensus sequences were constructed using bcftools ( 61 ) (version 1.9). Regions with low coverage ( < 20x) or variant frequencies between 30 and 70% were masked with Ns. An ov ervie w of the mutations detected is provided in Supplementary Table S1.

Statistical analyses
Statistical analyses of PFU titers, relati v e HAU-to-PFU ratios, and relati v e vRNA le v els were performed using Graph-Pad Prism 9.4.1, unless otherwise noted. An unpaired ttest was used to compare viral titers. Statistical analysis of relati v e HAU to PFU ratios and relati v e vRNA le v els was performed using a one-sample t-test. Two-tailed P values wer e corr ected by the Bonferroni-Holm method using StatistikGuru ( 62 ).Two-tailed P values < 0.05 (*), < 0.01 (**), and < 0.001 (***) were considered significant. Pearson correlations between the top 30 interactions or the 15 HA segment interactions found in cell-attached, released, and ultracentrifuged virions were performed using GraphPad Prism 9.4.1 with the assumption that the log 2 -transformed (RPM + 1) values were normally distributed.

Global detection of vRN A-vRN A inter action netw ork changes using RNAswarm
Similar to the previous identification of TPS by mutational analysis ( 20 , 24 , 26-28 , 30-34 ), we hypothesized that disruption of SPLASH-identified interactions by mutagenesis should af fect coordina ted genome packaging ( Figure  1 ). To test this idea, we needed a SPLASH workflow capable of detecting global changes in the vRN A-vRN A interaction networks of a mutant virus compared to the wild-type virus. Since no such workflow existed, we adopted the original SPLASH workflow ( 38 , 45 , 46 ) and added RNAswarm, a bioinformatics pipeline that compares the interaction frequencies between different SPLASH-identified networks. We evaluated our new workflow using two chimeric A / PR8 (H1N1) viruses, containing PB1 and NA segments of A / Wyoming (H3N2), as described by Dadonaite et al . ( 38 ). One of these viruses, rPR8:Wy-PB1 / NA, carries wildtype Wyoming PB1 and N A segments, w hile the other, rPR8:Wy-PB1 / NA Ud-sub , harbors four nucleotide substitutions in the NA segment deri v ed from A / Udorn / 1972 (H3N2). These nucleotide substitutions are critical for forming a vRN A-vRN A interaction with the Wyoming PB1 segment, which is detected by SPLASH and promotes co-packaging of both genome segments ( 38 ). By a ppl ying our SPLASH workflow to these chimeric viruses, we confirmed this interaction in rPR8:Wy-PB1 / NA Ud-sub and its absence in rPR8:Wy-PB1 / NA (Supplementary Figure  S1A and B). Importantl y, RN Aswarm detected that this interaction was highly enriched in rPR8:Wy-PB1 / NA Ud-sub compared to rPR8:Wy-PB1 / NA (Supplementary Figure  S1C), consistent with its thermod ynamic stabiliza tion due to the introduced mutations (Supplementary Figure S1D). Thus, our new workflow allowed us to (i) identify a previousl y validated vRN A-vRN A interaction that is critical for IAV genome packaging and (ii) to measure global variations in interaction frequencies between different networks.

An HA segment lacking high-frequency vRN A-vRN A interactions is properly packaged
With this SPLASH workflow in hand, we set out to map vRN A-vRN A interactions for functional evaluation. We performed se v en SPLASH replicates on the H7N7 subtype strain SC35M (rWT) and identified more than 1500 vRN A-vRN A interactions that varied in frequency over three orders of magnitude (Supplementary Figure S2 and  Supplementary Table S2). Ranking all interactions according to their detected frequencies re v ealed a biphasic distribution that allowed us to categorize them into low-and high-frequency interactions using a previously published cutoff point identification method ( 56 ). Thus, we identified 62 high-frequency vRN A-vRN A interactions (Supplementary Figure S2 and Supplementary Table S2), of which the top 30 ones formed an interconnected network with interaction loci distributed along the entire length of the vRNAs (Figure 2 A), similarly to networks previously detected for other IAV strains ( 37 , 38 ).
We hypothesized that removal of m ultiple vRN A-vRN A interactions from a gi v en genome segment should result in a packaging defect. To test this, we focused on the HA segment which plays an important role in the emergence of pandemic IAVs ( 63 ). Within the HA segment we detected ten high-frequency interactions (  Figure S3A and B). We speculated that these interactions could mediate efficient packaging of the HA vRNA and ther efor e designed synonymous mutations to disrupt four contacts to the PB2, NP, and M segments in rHA 4x . To limit the risk of adding unwanted contacts to the HA segment by mutagenesis, we introduced synonymous nucleotide exchanges at the interaction loci of the partner segments. The corresponding mutant virus, rHA 4x + 4x, lost a total of eight high-frequency HA segment interactions and retained only four high-frequency interactions of lower rank as determined by SPLASH and RNAswarm analysis ( Although vRNAs with mutated packaging signals are often inefficiently packaged per se ( 20 , 24 , 26-28 , 30-34 ), in some cases they are only inefficiently packaged when competing with their cognate wild-type segment for virion incorporation ( 38 , 40 ). We ther efor e performed a 7 + 2 rescue assay and co-transfected HEK-293T cells with two plasmids, encoding the HA wt and HA 4x segments and the se v en plasmids encoding the remaining wild-type vRNAs ( 38 , 40 , 64 ). After viral particle titration, we analyzed the HA segment of ten individual plaques by Sanger sequencing. While four plaques contained the HA wt segment, the other six plaques contained the HA 4x segment, indica ting tha t the vRN A-vRN A interaction losses do not affect HA 4x vRNA packaging in a competiti v e conte xt.
We hav e pre viousl y shown that m utation of functional packaging signals may result in 'hidden' packaging defects, in which reduced packaging of the mutant vRNA is prevented by redundant packaging signals present in other vR-NAs ( 34 ). Howe v er, a combina tion of two muta ted packaging signals can e v entually pro vok e reduced packaging of both mutant segments. We specula ted tha t the HA segments of rHA 4x and rHA 4x + 4x might hold such 'hidden' packaging defects and would only be inefficiently packaged when combined with a segment harboring a mutated 5 terminal packaging signal, such as the PB1 TPS vRNA described previously ( 34 ). To test this scenario, we generated mutant viruses carrying the PB1 TPS segment in the rHA 4x and rHA 4x + 4x backgrounds, designated rHA 4x + PB1 TPS and rHA 4x + 4x + PB1 TPS , respecti v ely. The rPB1 TPS virus carrying only the PB1 TPS vRNA showed reduced infectivity compared to rWT (Figure 3 A) and an increased production of noninfectious virions (Figure 3 B), associated with reduced packaging of the PB1 segment (Figure 3 C), as shown previously ( 34 ). Howe v er, contrary to our hypothesis, the combinatorial mutants rHA 4x + PB1 TPS and rHA 4x + 4x + PB1 TPS replica ted as ef ficiently as rPB1 TPS (Figure 3 A), did not produce more noninfectious viral particles (Figure 3 B), and did  not inefficiently package any genome segment other than the PB1 vRNA ( Figure 3 D and E). These results show that an HA vRNA lacking multiple high-and low-frequency vRN A-vRN A interactions detected by SPLASH is efficiently packaged, e v en in the presence of a vRNA that is known to re v eal 'hidden' packaging defects.
To exclude that the mutant viruses r estor ed coordinated genome packaging via re v ersions or compensatory mutations, we performed NGS on viral stocks obtained from rHA 4x , rHA 4x + 4x and rHA 4x + 4x + PB1 TPS (Supplementary Table S1). All mutant stocks harbored the introduced synonymous nucleotide substitutions. rHA 4x did not acquire any unintended mutations. In the rHA 4x + 4x + PB1 TPS virus stock, we detected a mutation (A609G) in the NA segment that caused an asparagine-to-serine exchange at position 196 (N196S) in the neuraminidase (NA) protein. rHA 4x + 4x also contained a mutation (T614C) in the NP segment that resulted in a v aline-to-alanine ex change at position 190 (V190A) in NP protein. We did not investigate these mutations further because the NA protein is not known to be involved in IAV genome packaging, and the residue 190 in NP is not exposed on the protein surface, unlike other NP protein residues known to be involved in genome packaging ( 49 , 65 ). Furthermore, none of these mutations was found in rHA 4x , indicating that they are unlikely to compensate for potential underlying packaging defects in the HA segment. Taken together, these results confirm that the disrupted vRN A-vRN A interactions of the HA segment identified by SPLASH are collecti v ely dispensab le for coordinated genome packaging.

vRN A-vRN A contacts inv olving TPS ar e dispensable f or coordinated genome packaging
Most of the packaging signals identified to date are located at the genome segment termini (18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35)(36). Howe v er, it remains to be demonstra ted tha t TPS coordina te IAV genome packaging by forming functional vRN A-vRN A interactions. To test this idea, we screened the HA segment for bona fide TPS and next performed SPLASH analysis to investigate whether the TPS would be involved in coordinating vRNA-vRNA interactions. Pre vious studies hav e discov ered functional TPS at the 5 end of the HA segment in different H1N1 viruses ( 30 ). Since the 5 HA vRNA end is not conserved among HA subtypes, we targeted the 5 end of the SC35M HA segment with three different sets of synonymous nucleotide substitutions between position 1647 and viruses, rHA TPS1 , rHA TPS2 and rHA TPS3 (Supplementary Tab le S1). Howe v er, none of these mutant viruses showed reduced viral stock titers compared to rWT (Supplementary Figure S4A), suggesting either that the targeted vRNA r egions ar e not critical for coordinated genome packaging or that their functional deficiency is compensated by redundant TPS ( 34 ). To distinguish between these possibilities, we genera ted combina torial virus mutants carrying the PB1 TPS segment in addition to the HA TPS1 , HA TPS2 or HA TPS3 segments. Notably, rHA TPS1 + PB1 TPS showed reduced viral stock titers compared to rPB1 TPS (Supplementary Figure  S4B), suggesting that the HA TPS1 region, comprising nucleotide 1647 to 1677 (3 to 5 vRNA), is critical for coordinated genome packaging. We confirmed that viral replication (Figure 4 A) and the number of non-infectious virus particles (Figure 4 B) of rHA TPS1 were comparable to those of the wild-type virus. This mutant also packaged all eight genome segments efficiently (Figure 4 C). In contrast, the combinatorial mutant rHA TPS1 + PB1 TPS was se v erely attenuated compared to rPB1 TPS (Figure 4 A) and produced more non-infectious virus particles (Figure 4 B). Importantly, rHA TPS1 + PB1 TPS showed a different packaging defect than rPB1 TPS, as it largely failed to package the PA, HA and NA segments in addition to the PB1 segment ( Figure  4 D). Thus, the HA TPS1 region in the 5 -end of the HA segment is a functional packaging signal.
SPLASH analysis re v ealed that the rHA TPS1 virus had an overall similar vRN A-vRN A interaction network compared to the wild-type virus (Figure 4 E); howe v er, RNAswarm detected the loss of two low-frequency interactions of the mutated HA TPS1 locus with the M and PB1 segments (Figure 4 F), possibly due to reduced thermodynamic stability of these interactions in the mutant (Figure 4 G). We specula ted tha t one or both of these interactions are critical for coordinated genome packaging and that their loss would cause the packaging defect observed in rHA TPS1 + PB1 TPS . To test this hypothesis, we first introduced disrupti v e mutations into the M segment interaction loci (rM mut ). We confirmed the loss of the targeted HA segment interaction in rM mut by SPLASH and RNAswarm analysis (Supplementary Figure 5A-C). Similar to rHA TPS1 , rM mut replicated to comparable PFU titers as the wild-type virus (Figure 4 H), produced similar le v els of noninfectious virions (Figure 4 I), and efficiently packaged all eight vRNAs (Figure 4 J). Howe v er, the combinatorial mutant, rM mut + PB1 TPS , replicated to onl y slightl y lower titers than rPB1 TPS (Figure 4 H), produced similar le v els of noninfectious virions (Figure 4 I) and showed a packaging defect only in the PB1 segment but no other vRNAs (Figure 4 K). We next tested whether the packaging defect of rHA TPS1 + PB1 TPS is caused by the combined loss of the vRN A-vRN A interactions of the HA 5 -TPS with the M and PB1 segments. Ther efor e, we generated rM mut + PB1 TPS+mut by introducing synonymous mutations into the PB1 interaction loci which increased the minimum free energy (MFE) of the targeted interaction to the HA 5 -TPS from −28.5 to −9.5 kcal per mol. Howe v er, rM mut + PB1 TPS+mut did not exhibit a more pronounced growth deficit than rPB1 TPS (Figure 4 H), did not show increased production of noninfectious particles (Figure 4 I), nor did it lose any segment other than the PB1 vRNA (Figure 4 L). Collecti v ely, these data suggest that the packaging defect of rHA TPS1 + PB1 TPS is not due to the loss of the vRN A-vRN A interactions of the HA TPS1 region with the M and PB1 segments.
To extend our evaluation of TPS, we searched the rWT inter actome for inter actions involving the HA segment and terminal coding regions of other vRNAs. We found a highfrequency interaction of lower rank between the HA segment and the 5 -end of the NA segment (Figures 2 B and  5 A). To test whether this interaction is critical for coordinated genome packaging, we first targeted the NA segment interaction loci at positions 1375-1405 (3 to 5 vRNA) with synonymous mutations and rescued the corresponding mutant virus, rNA TPS . SPLASH and RNAswarm analysis confirmed the loss of the targeted interaction ( Figure 5  Howe v er, SPLASH and RN Aswarm anal ysis of rHA mut also re v ealed one nov el high-frequency contact between the NA and HA segments that was not present in rNA TPS or rWT ( Figure 5 G and H and Supplementary Figure S6B). We hypothesized that this interaction may have compensated for the loss of the interaction involving the 5 -TPS of the NA segment to mediate efficient NA segment packaging in rHA mut . If so, this interaction should also restore coordinated genome packaging when grafted onto the rNA TPS virus. To test this, we generated the mutant virus rHA mut + NA TPS , carrying the potential compensatory interaction in the background of the lost interaction involving the N A-vRN A 5 -TPS. Contrary to our hypothesis, this mutant virus replicated less efficiently than the wild-type virus ( Figure 5 D), produced more noninfectious viral particles ( Figure 5 E), and still packaged the NA segment poorly (Figure 5 J), suggesting that the grafting of this vRNA-vRNA interaction into the rNA TPS background cannot correct the packaging defect caused by the mutated 5 -TPS of the N A vRN A. Taken to gether, these results demonstrate that the packaging defect in rNA TPS is unrelated to the loss of the vRN A-vRN A interaction between the 5 -TPS of the NA segment and the HA vRNA. Thus, the 5 -TPS of the HA and NA vRNAs achie v e coor dinated genome packaging by means other than the vRN A-vRN A interactions studied.

Virion release and ultracentrifugation have little effect on SPLASH-identified vRN A-vRN A inter action netw orks
In the current SPLASH workflow, RN A-RN A crosslinking is performed on digitonin-treated virions that have been released from infected cells and then purified by ultracentrifugation. Howe v er, pre vious electron microscopy and tomo gra phy studies suggest that virion release and ultra-centrifugation are delicate processes that could potentially affect vRN A-vRN A interactions. For example, it has been specula ted tha t bacilliform viral particles collapse into a spherical shape upon release, which would distort the longer vRNPs ( 66 ) and potentially alter their vRNA-vRNA interactions with other vRNPs. Similarly, ultracentrifugation induces shear stress that damages IAV particles ( 67 ) and potentially disrupts intersegmental RNA-RNA interactions. In contrast, it has been proposed that budded virions that are still attached to cells contain well-arranged genome complexes with presumably intact vRNA-vRNA contacts ( 39 , 41 ). Ther efor e, we hypothesized that the vRN A-vRN A interaction networks currently mapped by SPLASH are distorted by the processes of viral release and ultracentrifugation, and tha t cell-a ttached virions would be better suited to identify vRN A-vRN A interactions that are critical for IAV genome packaging.
To test this hypothesis, we de v eloped CAPTIVE (Catching Assembled Particles Via Timely Induced Virus Egress), an experimental approach that allows RNA-RNA crosslinking in specific preparations of either cell-attached, released or ultracentrifuged viral particles (Figure 6 A). CAP-TIVE consists of three main steps, namely an initial enrichment of ne wly assemb led virions on the surface of infected cells by vir al neur aminidase inhibition, followed by a synchronized release of the budded particles induced by an exogenous bacterial sialidase ( Supplementary Figure S7A), and a final purification of the released virions by ultracentrifugation. This modular design allowed us to fix vRN A-vRN A interactions in different virion populations, simply b y v arying the timing of the RNA-RNA cross-linking step. In addition, CAPTIVE is largely compatible with the original SPLASH workflow, requiring only a few adjustments. Specifically, RNA-RNA crosslinking is performed with a clickable psoralen-TEG-azide, which is smaller than the originally used biotin-psoralen and penetrates membranes without digitonin treatment ( 68 ). Later in the workflow, a biotin-sDIBO alkyne is added to the psoralen-TEG-azide via click chemistry, allowing enrichment of crosslinked vRNAs on streptavidin beads.
To compare vRN A-vRN A interactions between cellattached, released and ultracentrifuged virions, we performed CAPTIVE coupled to SPLASH on wild-type SC35M virus. The resulting networks were very similar with the top 30 interactions (Figure 6 B) and the top 15 HA segment interactions (Figure 6 C) being highly correlated across all three conditions (Pearson correla tion coef ficients ranging from 0.74 to 0.98) (Supplementary Figure S7B and C). Comparison of the CAPTIVE-deri v ed networ ks with those obtained with the original workflow (using digitonin treatment and biotinylated psoralen) also re v ealed high correlation among the top 15 HA segment interactions (Pearson correla tion coef ficients ranging from 0.61 to 0.87) (Supplementary Figure S7C). Notably, all HA segment interactions mutated in this study were consistently detected at high frequency (Supplementary Figure S7B and C, red dots), and only one novel high-frequency HA segment interaction with the PA segment emerged in the CAPTIVE-deri v ed networ ks (Supplementary Figure S7B and C, blue dot). Apart from this minor difference, the only major difference was that the CAPTIVE workflow efficiently detected intrasegmental interactions of the 3 and 5 vRNA termini (data not shown), suggesting that the psoralen-TEG-azide used in CAPTIVE detects vRNA promoter structures bound by the viral polymerase ( 17 ), which are largely missed by the bulkier biotinpsoralen. Taken together, these results demonstra te tha t virion release and ultracentrifugation have only little ef-fect on the vRN A-vRN A interaction network detected by SPLASH.

DISCUSSION
The prevailing mechanistic model proposes that the eight IAV genome segments are selecti v ely co-packaged into virions by a network of specific intersegmental RN A-RN A interactions. Recently, e xtensi v e and strain-specific vRNA-vRNA interaction networks have been detected in purified viral particles by SPLASH ( 38 ). Howe v er, whether these networks control genome packaging has not been conclusi v ely demonstrated. In this study, we used mutational analysis to systema tically evalua te the relevance of SPLASH-identified vRN A-vRN A interactions for coordinated genome packaging in A / SC35M (H7N7). We found that combined disruption of multiple HA segment interactions does not affect genome packaging. Similarly, we show that interactions involving the TPS of the HA or NA segments are dispensable for coordinated genome packaging. These results suggest that the studied vRN A-vRN A interactions are at most of minor importance for coordinated genome packaging compared to previously described packaging signals.
Since SPLASH was previously used to identify a high-frequency vRN A-vRN A interaction mediating copackaging of the PB1 and NA segments in H3N2 viruses, our results may seem surprising at first glance. Howe v er, it should be noted that the loci involved in this interaction wer e alr eady suggested by pr evious m utational ma ppings ( 43 ), which may have facilitated the identification of this particular interaction. In the present study, we mainly evaluated high-frequency interactions of the SC35M HA segment. Although our in-depth mutational analysis did not re v eal any nov el bona fide vRN A-vRN A interactions involved in genome packaging, it is possible that some of the remaining high-or low frequency interactions are critical for HA segment packaging. These interactions may be thermodynamically stable but poorly cross-linked by psoralen, which is known to pr efer entially r eact with interstacked pyrimidines ( 69 ). Poor cross-linking of these vRN A-vRN A interactions may result in their artificial underr epr esentation in the detected networks, making it difficult to identify them without additional information from mutational mapping.
Alternati v ely, demonstrating the actual importance of particular vRN A-vRN A interactions for coordinated genome packaging may be complicated by functionally redundant interactions in the netw ork. If tw o m utuall y exclusi v e vRN A-vRN A interactions were able to media te ef ficient HA segment packaging, it is possible that disruption of one would be compensated for by the other, resulting in no detectable packaging defect. Thus, despite tremendous interaction losses in our rHA 4x and rHA 4x + 4x viruses, the remaining vRN A-vRN A interactions might be sufficient to maintain structural integrity of the network.
Although the abundance of detected vRN A-vRN A interactions might indeed reflect to some degree a functional network redundancy, it could also point to systema tic limita tions in the SPLASH workflow that lead to falsepositi v e interactions. RN A-RN A cross-linking by psoralen is a two-step reaction with a psoralen-RNA monoadduct as an intermediate ( 70 ). Like cross-linked RNAs, these monoadducts could be enriched on streptavidin beads and hybridize with free RNAs to form artificial interactions that end up as false-positi v e chimeric reads in the r ecover ed networks. Although we did not test this possibility here, a previous SPLASH study detected only 4% false-positi v e chimeric reads ( 46 ). Alternati v ely, many of the vRNA-vRNA interactions detected by SPLASH could be byproducts of the genome packaging process. During packaging, the vRNPs are inserted into a narrow viral particle, which may induce vRNP collisions and the formation of a specific set of vRN A-vRN A interactions predetermined by the vRNP arrangement in the (7 + 1) genome complex ( 41 , 71 , 72 ). Since we detected similar interaction networks between cell-attached, released and ultracentrifuged viral particles, these budding-induced vRN A-vRN A interactions may be robust against further rearrangements. Moreover, such compaction of the viral genome during packaging may disrupt those vRN A-vRN A interactions that are actually critical for viral genome assembly, rendering them largely undetectable within virions. Thus, future studies could examine vRN A-vRN A interactions within infected cells, although this may be challenging if the critical interactions are only transient and restricted to currently poorly defined cellular compartments ( 73 , 74 ).
Finally, it is possible that intersegmental RNA-RNA interactions are not fundamental to IAV genome packaging and are therefore rare and difficult to detect. Indeed, only two functional vRNA-vRNA interactions involved in genome packaging have been identified to date, and these are strain-specific ( 38 , 40 , 43 ). This continued lack of validated vRN A-vRN A inter actions, particular ly of those involving the TPS, raises the possibility that other mechanisms control selecti v e vRNP packaging. Pre vious studies suggest that an interplay between the TPS and viral NP is central to coordinated genome packaging ( 34 ). Although speculati v e, it is possib le that the TPS form RNA structures that make specific contacts with another vRNP through intersegmental vRNA-NP interactions.
In conclusion, our study questions the utility of the SPLASH workflow alone for the de novo discovery of functional vRN A-vRN A interactions tha t coordina te IAV genome packaging. We propose that the vRN A-vRN A interaction networks identified by SPLASH are insufficient to e xplain coor dinated vRNP packaging and call for expanded strategies to re v eal the 'true' genome packaging machinery.

DA T A A V AILABILITY
The data underlying this article are available in the Sequence Read Archi v e (SRA) at https://www.ncbi.nlm.nih. gov/sra , and can be accessed with the BioProject accession number PRJNA939935.