Parallel evolution between genomic segments of seasonal human influenza viruses reveals RNA-RNA relationships

The influenza A virus (IAV) genome consists of eight negative-sense viral RNA (vRNA) segments that are selectively assembled into progeny virus particles through RNA-RNA interactions. To explore putative intersegmental RNA-RNA relationships, we quantified similarity between phylogenetic trees comprising each vRNA segment from seasonal human IAV. Intersegmental tree similarity differed between subtype and lineage. While intersegmental relationships were largely conserved over time in H3N2 viruses, they diverged in H1N1 strains isolated before and after the 2009 pandemic. Surprisingly, intersegmental relationships were not driven solely by protein sequence, suggesting that IAV evolution could also be driven by RNA-RNA interactions. Finally, we used confocal microscopy to determine that colocalization of highly coevolved vRNA segments is enriched over other assembly intermediates at the nuclear periphery during productive viral infection. This study illustrates how putative RNA interactions underlying selective assembly of IAV can be interrogated with phylogenetics.


28
Genetic variation is ubiquitious in RNA viruses. The rapid evolution underlying this variation 29 can occur as a result of mutation, recombination, or reassortment, with major consequences for 30 human disease (Andino & Domingo, 2015). In the case of influenza virus, these consequences 31 include poor vaccine efficacy rates, immune escape, antiviral resistance, and the emergence of 32 novel strains (Lyons & Lauring, 2018). Within the past century, influenza A virus (IAV) pandemics

39
Public health measures to limit the impact of influenza virus outbreaks would benefit from the 40 ability to predict reassortment between circulating influenza viruses. Genetic mutation is driven 41 by stochastic processes and is therefore difficult to predict (Andino & Domingo, 2015). In contrast,

52
Epistasis within and between genes imposes evolutionary constraints that can be shaped by 53 a number of factors, including the function, stability, or interactions between individual RNA or 54 protein (Sardi & Gasch, 2018). Probabilistic models have revealed several destabilizing mutations 55 in the influenza virus nucleoprotein (NP) that became fixed as a result of counterbalancing 56 epistasis that improves NP protein stability (Gong, Suchard, & Bloom, 2013). These destabilizing 57 mutations occur within T cell epitopes of NP that may be important for immune escape (Gong et

72
In this study, we set out to combine phylogenetics and molecular biology to examine parallel 73 evolution across vRNA segments genome-wide in seasonal human influenza viruses and identify 74 potential epistatic relationships between vRNA segments. Unlike previous studies, our objective 75 was to identify vRNA segments that might play key roles in genomic assembly. To evaluate phylogenetic relationships among vRNA segments we relied upon the Robinson-Foulds distance 77 (d), a measure of topological distance between trees (Robinson, 1981). This method determines 78 the number of branch partitions that are not shared between two trees (Robinson, 1981) (Table 1). However, reconstructing 105 phylogenetic trees from all available sequences was disadvantageous, as a preliminary analysis 106 of three-hundred sequences suggested that a great deal of phylogenetic variation could not be 107 statistically supported by bootstrapping (branch support < 70). Instead, we used a clustering 108 approach to select representative strains that would produce more statistically robust trees. We  (Table 1), consistent with the notion that increased sequencing has led to more 114 closely related sequences in public databases.

116
The primary objective behind clustering was to reduce variation between trees that was not 117 statistically supported by bootstrapping. The cutoff for sequence identity during clustering of the 118 species tree was therefore an important consideration because it controlled how much variation 119 remained in our trees. As expected, higher cutoffs (98-99% sequence identity) yielded species 120 trees with more clusters containing fewer members while lower cutoffs (95-96% sequence 121 identity) contained increasingly fewer clusters with more members grouped in each cluster. We

225
Comparison of the dendrograms from H1N1 and H3N2 viruses revealed some expected 226 similarities as well as differences (Fig 2D, 3D, and 5A-B). For example, PB1 and NP share a 227 common evolutionary relationship across all four sets of influenza viruses examined in this study 228 (Figures 2D, 3B, and 5A-B), which may be expected based on the shared role of the encoded 229 proteins in replication (Fodor, 2013). However, in H3N2 viruses PB1 and NP are next most closely 230 related to NA and NP (Figures 2D and 3B), while in H1N1 viruses their relationship with other 231 segments varies (Figure 5A-B). We compared the overall similarity of all four dendrograms by  Figure 3A-B). Likewise, d = 10 for the post-pandemic H1N1 virus dendrogram 237 when compared to either of the H3N2 virus dendrograms. These results are in direct contrast to 238 our previously determined distance for the H3N2 virus dendrograms to one another (d = 6).

239
Overall, these data indicate that parallel evolution between vRNA segments is distinct between 240 influenza subtypes isolated from humans within similar time scales.

241
Parallel evolution in H3N2 viruses is not driven solely by protein-coding mutations.

242
As discussed previously, a phylogenetics approach such as ours would encompass parallel 243 evolution driven by either protein and RNA relationships. We have already shown that known 244 protein relationships between PB1 and PA, two members of the polymerase complex, are 245 identified by our approach (Figure 2A). However, the observation that PB2 is more coevolved 246 with NA than with either PB1 or PA ( Figure 2C) suggests that our method also reveals protein-  (Figures 2C and 6, open diamonds).

290
We quantified colocalization using our established method for examining intracellular  complete assembly complexes (Lakdawala et al., 2014). Given that newly synthesized vRNA 308 segments are exported from the nucleus as assembly intermediates comprising greater than two 309 but fewer than 8 segments, these perinuclear assembly intermediates may function as nodes for 310 further assembly (Majarian, Murphy, & Lakdawala, 2018). Therefore, we assessed the potential 311 for PB2, NA and NS to colocalize at the nuclear periphery, where assembly intermediates first 312 begin to form. We defined localization at the nuclear periphery to within three-hundred

322
In this study, we used phylogenetics and molecular biology methods to investigate genome-323 wide relationships between vRNA segments in seasonal human influenza A viruses. We found 324 that parallel evolution varies considerably between vRNA segments, with distinct relationships 325 forming in different influenza virus subtypes (H1N1 vs H3N2) and between H1N1 virus lineages 326 that arose from distinct host origins. We further demonstrate that evolutionary relatedness 327 between vRNA segments in H3N2 viruses is largely conserved over time. Importantly, our data 328 suggest that parallel evolution cannot be attributed solely to protein interactions, and we 329 successfully predicted intracellular colocalization between two coevolved vRNA segments during 330 infection with an H3N2 virus. Thus, we present a phylogenetic approach for interrogating putative 331 RNA associations that could be broadly applied toward the study of genomic assembly and 332 reassortment in segmented viruses.

333
Selective assembly of all eight genomic segments is fundamental to the production of fully

343
The approach we present here differs from other experimental approaches in that we identify

458
In each experiment, images taken of mock-infected cells were deconvolved using the same 459 parameters as those of infected cells. 3D reconstruction and colocalization analysis of the 460 resulting images were performed using Imaris software (version 8.4.2, Bitplane AG) as previously was segmented using the 'Surfaces' and 'Cell' tools in Imaris software. DAPI signal was used to 463 mask nuclear signal from the remaining channels. The 'Spots' tool was then used to populate the reconstructed cell with four different sets of Spots corresponding to foci from each of the remaining 465 channels. In each experiment, the mock infected cell was analyzed in an identical manner and 466 the fluorescence intensity for each channel of the mock-infected cell was used to establish 467 fluorescence intensity thresholds at which 97% or more of the background signal was removed 468 prior to Spot generation. A modified Matlab extension was then used to quantify spot 469 colocalization using a distance threshold of 300 nm as previously described (Nturibi et al., 2017).

470
Colocalization data was imported into the Cell and all data was exported and analyzed in R.

482
Whitney U test was also used to determine statistical significance of FISH-IF colocalization data.