IntroductionRNA viruses are constrained in genome size as a result of limitations from packaging and mutational burden1. Due to these constraints it is quite common for viral genomes to encode multifunctional proteins to maximize their use of limited genomic real estate.Influenza A virus (IAV) is a negative-sense, segmented, single-stranded RNA virus. IAV’s eighth, and smallest, genomic segment encodes two multifunctional non-structural proteins, NS1 and NS2. NS2 is also referred to as nuclear export protein (NEP) due to its role in facilitating genome egress from the nucleus to the cytoplasm2,3. This protein also aids in other functions such as viral replication and budding4,5,6,7,8,9,10.Once IAV viral ribonucleoproteins (vRNPs) are imported into the nucleus of an infected host cell, the viral polymerase at the terminus of each vRNP interacts with host RNA polymerase II to cap-snatch and cleave a nascent transcript to act as a primer for positive-sense viral mRNA transcription11,12,13,14. After viral mRNAs are translated, polymerase proteins are shuttled into the nucleus and initiate viral replication. These additional viral polymerases use the viral genomes (vRNA) as a template to synthesize positive-sense complementary RNA (cRNA), which in turn becomes a template for further production of vRNAs15,16.NS2 has been shown to interact with the heterotrimeric viral polymerase and alter the ratio of vRNA, cRNA, and mRNA in minimal replication (also called minireplicon) assays—coordinating an increase in cRNA synthesis and decrease in mRNA transcription4,5,8,17. Additionally, polymorphisms in NS2 can lead to variation in the frequency of defective viral genome accumulation, further supporting its role in controlling genome replication18,19. Lastly, NS2 influences the response of avian influenza polymerases to ANP32A, a host dependency factor and key determinant of species tropism of avian IAV20,21,22.Using a combination of length-variant libraries in HA and PB1, and sequence-variant libraries in PB1, we now establish that NS2 modulates IAV replication to a greater degree than perhaps previously appreciated. Most critically, we now know that the canonical viral promoters, considered to be both essential and sufficient for genome replication, cannot by themselves fully explain the dynamics of replication in the presence of NS223,24,25. Specifically, in our sequence-variant libraries in PB1 we identify sites that are required for efficient viral replication in the presence of NS2 which are not conserved between IAV’s genomic segments, in contrast with the canonical viral promoter. As these sites differ between segments, they may explain how IAV fine-tunes replication between its genomic segments, which exhibit different dynamics despite sharing identical core promoter sequences26,27,28,29,30.ResultsDivergent length selection in minireplicons and infectionsSince the original development of reverse genetics there have been numerous studies of selection pressures on the genome sequence of IAV. These pressures include both broad, architectural, selection on elements that are constrained by polymerase processivity and/or nucleoprotein coating, as well as site-specific constraints such as a need to retain the minimal viral promoter and vRNP bundling sequences31,32,33,34,35,36. These steps in the life cycle shape the fitness of any given viral genome and place fundamental limits on viral evolution.These selection pressures exist alongside, and potentially in conflict with, a need to encode functional protein products. This makes it difficult to study selection pressures on genomic sequence independent of protein coding—does a mutation lead to loss of fitness because it changes amino acid sequence, or because it influences genome replication or packaging? However, this difficulty only strictly applies to replication-competent viruses. Viral populations also contain defective interfering particles (DIPs), which encode a non-standard, or defective, viral genome (nsVG, DVG)37,38. For IAV these genomes generally (but not exclusively) contain large internal deletions in at least one of three genomic segments encoding the heterotrimeric polymerase, PB2, PB1, and PA39,40,41,42. For our purposes, DVGs, including deletions as observed in IAV, reduce or remove selection on protein coding (other than the encoding of detrimental products), while retaining selection during genome replication and packaging43. Thus these mutations can reveal selection pressures that are otherwise masked.In a previous publication from our group (Mendes and Russell36), we generated artificial length-variant barcoded, DVG vRNA libraries in the HA (haemagglutinin) and PB1 (polymerase basic 1) segments of the IAV strain A/WSN/193336. In that study, we transfected these libraries alongside plasmids encoding the minimal viral replication machinery—PB2, PB1, PA, and NP—and tracked how length of a vRNA influences replication. This work revealed that smaller variants were able to replicate much faster than their longer counterparts, and that this selection was balanced with selection for longer fragments during packaging (Fig. 1a, b). Therefore, at least for the lengths we tested, polymerase processivity appears to be the key rate-limiting step in influenza genome replication. In contradiction to this simple model, a concurrent publication, Alnaji et al., found that in a single-round viral infection competition assay, a 395nt variant of the PB2 segment was outcompeted by a full-length counterpart during genome replication44.Fig. 1: Prior models incompletely describe genome replication during viral infection.a Our previous work found that 400nt variants of the PB1 and HA segments outcompete their full-length counterparts during genome replication, but during packaging, full-length variants have an advantage. More comprehensive library-based analyzes of genomic length showed that smaller sizes correlate with better replication efficiency and longer sizes better packaging efficiency. Inconsistent with this model, Alnaji et al. described a phenomenon where a short variant of the PB2 segment was outcompeted during genome replication. b A schematic describing our model derived from measuring thousands of length-variants in Mendes and Russell36. c RNA genomes of barcoded 200, 400, 800, and 1600nt variants of A/WSN/1933 PB1 and HA bearing equal sequence length from 5’ and 3’ ends were generated from co-transfected plasmids using minimal replication machinery and rescued into virions by coinfection with wild-type virus. Viral supernatant was used to infect A549 cells at an MOI of 25, and qPCR was used to analyze the proportion of each variant before infection and within infected cells at 8 hours post-infection. 200, 400, and 800nt variants are assessed by their frequency relative to the 1600nt species. Asterisks indicate significantly different values, ANOVA p