Comprehensive profiling of translation initiation in influenza virus infected cells

Translation can initiate at alternate, non-canonical start codons in response to stressful stimuli in mammalian cells. Recent studies suggest that viral infection and anti-viral responses alter sites of translation initiation, and in some cases, lead to production of novel immune epitopes. Here we systematically investigate the extent and impact of alternate translation initiation in cells infected with influenza virus. We perform evolutionary analyses that suggest selection against non-canonical initiation at CUG codons in influenza virus lineages that have adapted to mammalian hosts. We then use ribosome profiling with the initiation inhibitor lactimidomycin to experimentally delineate translation initiation sites in a human lung epithelial cell line infected with influenza virus. We identify several candidate sites of alternate initiation in influenza mRNAs, all of which occur at AUG codons that are downstream of canonical initiation codons. One of these candidate downstream start sites truncates 14 amino acids from the N-terminus of the N1 neuraminidase protein, resulting in loss of its cytoplasmic tail and a portion of the transmembrane domain. This truncated neuraminidase protein is expressed on the cell surface during influenza virus infection, is enzymatically active, and is conserved in most N1 viral lineages. We do not detect globally higher levels of alternate translation initiation on host transcripts upon influenza infection or during the anti-viral response, but the subset of host transcripts induced by the anti-viral response is enriched for alternate initiation sites. Together, our results systematically map the landscape of translation initiation during influenza virus infection, and shed light on the evolutionary forces shaping this landscape.


Author summary
When viruses such as influenza infect cells, both host and viral mRNAs are translated into proteins. Here we investigate the sites in these mRNAs that initiate protein translation during influenza infection. In particular, we explore whether some of this translation initiates at codons other than the canonical ones used to produce the primary protein product of each gene. Using computational analyses, we find that mammalian influenza viruses evolve to reduce the number of codons that can initiate such alternate translation initiation products. We next use the comprehensive experimental strategy of ribosome profiling to identify sites of translation initiation across all influenza and host mRNAs. We find a number of sites of alternate initiation on both influenza and host mRNAs. We study in detail one such alternate start site on in-frame CUG content of sequences from human H5N1 viruses (which mostly result from 96 one-off infections with avian viruses) is similar to that of the avian and 1918 sequences. 97 Notably, we did not see a comparable decrease in frequency of CUH (H = A/C/U) 98 leucine codons in any of the five lineages (Fig. 1A, lower panel), indicating that the 99 depletion of in-frame CUG codons is not attributable to a general selection against CUN 100 leucine codons. This depletion is also not due to selection against the amino acid leucine 101 or biochemically similar amino acids (Fig. S1). The depletion against CUG codons is to increase the number of such codons (high CUG NP) (Fig. S4). Our rationale for 165 recoding CUG codons in this gene is that NP is a major source of CD8 T-cell 166 epitopes [62,63], and previous work suggested that CUG codons may initiate peptides 167 that contribute to the immune epitope pool [5,6]. Using this virus mix allowed us to 168 have an internally controlled experiment: if a CUG codon in the high CUG NP variant 169 can initiate translation, we should be able to preferentially detect it against the low 170 CUG NP variant background even in the presence of end-specific sequence biases in the 171 ribosome profiling method [64].  (Fig. 2C), the negative sense reads did not show a characteristic peak around the size of 178 a ribosome footprint (Fig. S5D), and likely arise from co-purification of influenza 179 genomic RNAs that are protected from nuclease digestion due to their association with 180 the viral NP protein [65,66]. Therefore, we only considered reads that mapped to the 181 positive sense strand of influenza transcripts for the remaining analyses. Reads mapping 182 to influenza transcripts accounted for between 29-39% of the mapped reads in the 183 virus-infected samples (Fig. 2B), consistent with our use of high MOI. As expected, 184 interferon-β pretreatment reduced productive infection, and reads mapping to influenza 185 transcripts only accounted for 6-13% of aligned reads in the +ifn +vir sample (Fig. 2B). 186 We performed additional checks to ensure adequate quality of our Ribo-seq data.

187
Ribo-seq reads mapping to either human and influenza transcripts had read lengths 188 between 29-34 nucleotides (Fig. 2C, Fig. S5B). This length distribution was on average 189 a few nucleotides wider and longer than observed in other ribosome profiling single-coding region of the same segment (Fig. 2G, S5G). This dual-coding signature 204 was present but less pronounced in the known dual-coding regions of M (45 nucleotides) 205 and PB1 (261 nucleotides) segments (Fig. 2G, S5G), likely because M2 and PB1-F2 are 206 expressed at lower levels during infection [70][71][72]. 207 Finally, we used our sequencing data to evaluate the quality of the virus stocks used 208 in our experiment. Virus stocks contain virions that range in biological activity, 209 including virions defective in replication (defective particles) [73][74][75][76][77]. For influenza 210 virus, defective particles often contain large internal deletions in the polymerase 211 segments [73][74][75][76][77][78]. The presence of defective particles with internal deletions would 212 diminish our ability to detect alternate initiation sites within the deleted regions. 213 However, the burden of defective viral particles can be reduced by growing virus for a 214 short amount of time and at a low MOI, as defective viral particles increase in frequency 215 when viruses are grown at a high MOI where complementation of deleterious genotypes 216 7/53 can occur [79]. The even read coverage across the polymerase segments in our data 217 (Fig. S6B) indicates that we had a low burden of defective particles. This low burden 218 was particularly evident when we compared our data to a previous ribosome profiling 219 study of influenza virus infected cells (Fig. S6A) [27]. 220 Translation initiation sites in the viral genome 221 We next used the Ribo-seq and Ribo-seq + LTM measurements to annotate candidate 222 translation initiation sites (TIS) in the influenza genome. Our general strategy to 223 identify candidate TIS was to find peaks in Ribo-seq + LTM coverage within each 224 influenza transcript that was significantly higher than both the Ribo-seq coverage at the 225 corresponding location and the background Ribo-seq + LTM coverage distribution for 226 that transcript (Fig. 3A). Specifically, we used a zero-truncated negative binomial 227 distribution (ZTNB) to statistically model the background distribution of Ribo-seq and 228 Ribo-seq + LTM counts in each transcript [80,81]. Candidate start sites were identified 229 based on the following criteria: The ZTNB-based P-value for the Ribo-seq + LTM 230 count at that location must be <0.01 and 1000-fold higher than the P-value of the 231 Ribo-seq counts at the same location (Fig. 3A, left panel), or must have an absolute 232 value less than 10 −7 . In addition, the Ribo-seq + LTM counts must be a local 233 maximum within a 30 nucleotide window (Fig. 3A, right panel), and the Ribo-seq 234 counts must be non-zero. To account for the 1-2 nucleotide positional uncertainty of the 235 P-site density in our Ribo-seq and Ribo-seq + LTM measurements, we pooled the read 236 counts in 3 nucleotide windows before applying the above criteria. 237 We applied our candidate TIS identification strategy to each influenza transcript 238 (Fig. S7) separately for the +vir and +ifn +vir samples. We assigned the candidate TIS 239 peaks to any near-cognate AUG codon (at most one mismatch from AUG) if that codon 240 was within 1 nucleotide of either side of the TIS peak. We identified a total of 25 241 candidate TIS across both samples (Fig. S8). Fourteen of the 25 identified TIS 242 overlapped between the two samples ( Fig. 3B), and we used this overlapping subset as a 243 high-confidence set of candidate TIS for downstream analyses. We did not detect a 244 higher number of candidate TIS in our +ifn +vir sample compared to the +vir sample 245 (4 vs. 7, Fig. 3B), suggesting that anti-viral response as mediated by interferon-β 246 induction does not result in detectably higher number of alternate translation initiation 247 sites in influenza transcripts [6]. An important caveat to this observation is that the 248 Ribo-seq + LTM assay is likely to miss TIS that have a low frequency of initiation, but 249 could still be detectable by sensitive immunological assays [5,25].

250
The high-confidence set of 14 candidate TIS had multiple features consistent with 251 being bona fide TIS. First, this set included 7 of the 8 annotated TIS (aTIS) for the 8 252 segments in the influenza genome (Fig. 3E). Only the annotated TIS of the PB2 253 segment was not identified, and this is due to the dense Ribo-seq + LTM coverage and 254 lack of clear peaks in this segment (Fig. 3D, bottom right panel). Second, our set also 255 correctly identified the start codon for two previously annotated protein products 256 generated by initiation in an alternate reading frame of the PB1 and M genes. Initiation 257 in the +1 reading frame at nucleotide 118 in the PB1 gene generates PB1-F2 (Fig. 3D, 258 arrow). Initiation in the +1 reading frame at nucleotide 113 in the M segment is also a 259 known to occur [21]. Third, 13 of the 14 candidate TIS had an AUG codon within 1 260 nucleotide even though less then 3% of trinucleotides in the influenza genome are AUG 261 (Fig. 3F, top vs. bottom). Fourth, both the annotated and the novel candidate TIS in 262 our set had an over-representation of A at -3 nucleotide and G at +4 nucleotide 263 positions (Fig. 3C), which are known to be optimal contexts for translation initiation in 264 vertebrate cells [56]. Finally, the candidate TIS are enriched towards the 5 end of the 265 transcripts, with 13 of the 14 candidate TIS located in the initial third of the influenza 266 transcripts (Fig. 3G). This last observation is consistent with the initiation of candidate 267 8/53 . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under   . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The alternate dTIS in our high-confidence set are distributed across multiple 276 influenza segments: M, NA, NP, PB1, and PA (Fig. 3D, Fig S9). Ten of the 13 277 high-confidence TIS initiating at AUG are located in the canonical reading frame 0 278 (Fig. 3H). This set includes 7 annotated AUG starts, as well as three in-frame 279 downstream AUGs in NA, NP, and PA segments that would result in N-terminally 280 truncated forms of the annotated proteins. The three out-of-frame candidate dTIS are 281 the known PB1-F2 ORF [14], the known start codon at nucleotide 113 of the M 282 gene [21], and a short ORF of length 2 in NA (Fig. 3I, top panel). Excluding the 283 well-characterized PB1-F2 ORF, the two candidate out-of-frame candidate ORFs have 284 lengths that are typical of out-of-frame AUG-initiated sequences in the influenza 285 genome (Fig. 3I, lower panel).

286
Among the other alternate translation initiation sites in the influenza genome that 287 have been described previously, we did not find evidence for the initiation sites 288 previously noted in PA and PB1 (see Fig. S10) [14,15,[17][18][19][20]. This could be because of 289 low initiation frequency at these sites under our infection conditions. Re-analysis of the 290 harringtonine-treated ribosome profiling data from Ref.
[27] revealed P-site count peaks 291 that coincided with all the 7 annotated TIS as well as 3 of the 7 downstream TIS 292 identified in our experiments (Fig. S11). Among the 4 dTIS that could not be clearly 293 associated with a harringtonine peak, 3 are >200 nucleotide from the 5 end of the 294 transcripts, and could have been potentially affected by the high fraction of defective 295 viral particles in [27] (Fig. S6A). Interestingly, the dTIS in our dataset without a 296 corresponding harringtonine peak is an in-frame AUG in the NA segment that is 297 mutated to the near-cognate CUG codon in the PR8 influenza strain used in [27].

299
As mentioned previously, our infections were performed with a mix of two otherwise 300 isogenic viruses that were recoded to have different numbers of CUG codons in the NP 301 gene (Fig. S4). This experimental design allows us to sensitively look for CUG initiation 302 that cannot be detected by the start-site calling method in the previous section, but 303 might still be detectable via an internally controlled comparison of the two NP variants. 304 Specifically, the high CUG NP variant has 20 leucine codons that were 305 synonymously mutated to a non-CUG codon in the low CUG NP variant (Fig. 4A). We 306 examined if there was evidence of enhanced initiation at any of these codon sites in the 307 high CUG NP variant. Between 35% and 42% of reads mapping to NP could be 308 uniquely assigned to either the high or low CUG NP sequences based on polymorphisms 309 due to the synonymous mutations. These reads mapped to the two variants at a nearly 310 equal (1:1.3 ratio) overall proportion (Fig. 4A). The ratio of P-site density between the 311 two variants at individual NP sites obtained from the uniquely-mapping reads varied 312 over a 500-fold range (Fig. 4B). Among the sites with an excess of high CUG NP P-site 313 density, the CUG codon 322 nucleotides from the annotated TIS displayed the largest 314 such excess with 16-fold more reads for the high CUG NP variant than the low CUG 315 NP variant (Fig. 4B, red point). This excess P-site density was present in both the 316 Ribo-seq and the Ribo-seq + LTM data (Fig. 4C). The length distribution of the reads 317 generating this excess was consistent with them being derived from ribosome-protected 318

10/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018.  Coverage of Ribo-seq + LTM, Ribo-seq, and RNA-seq reads that can be uniquely aligned to either the high CUG NP variant or the low CUG NP variant, along with reads that cannot be uniquely assigned. P-site counts are shown for Ribo-seq and Ribo-seq + LTM assays. 5'-end counts are shown for RNA-Seq. Data are plotted as a stacked bar graph. Locations of the 20 CUG codons that are present in high CUG NP and synonymously mutated in low CUG NP are indicated by arrows. (B) The ratio of high CUG NP to low CUG NP coverage from panel A is plotted against their sum along the horizontal axis. (C) The green-highlighted region in panel A around the CUG322 codon is shown at greater horizontal magnification. Data shown for +vir sample. See Fig. S14 for +ifn +vir sample. fragments (Fig. S12). The excess P-site density at CUG322 did not arise from 3 ends of 319 reads mapping to the nearby recoded CUG328 (Fig. S13). The +ifn +vir sample 320 displayed a similar excess of density at site 322 for the high CUG NP variant (Fig. S14). 321

11/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018. ; https://doi.org/10.1101/326967 doi: bioRxiv preprint To examine whether CUG322 could initiate translation in the high CUG NP variant, 322 we used Western blots of heterologously expressed NP variants. However, we were 323 unable to resolve any protein fragment of appropriate size that was present only in the 324 high CUG NP variant (Fig. S15). It is possible that there is no initiation at this site, or 325 that initiation occurs but at a very low level, consistent with the low overall Ribo-seq + 326 LTM P-site density at CUG322 compared to the other locations on NP. In that case, 327 more sensitive immunological assays [5,25] might be necessary to identify initiation at 328 CUG322 and other CUG codons in the high CUG NP variant. Another important 329 caveat is that LTM treatment may not effectively arrest CUG-initiating ribosomes.

330
Indeed, evidence for CUG-based initiation being refractory to several translation 331 inhibitors has been noted in previous reporter-based studies [5].  The AUG43 in NA has a favorable translation initiation context, with a G at +4 340 nucleotides (Fig. 5B). The function of NA is to mediate viral egress by cleaving the viral 341 receptor sialic acid from the cell surface [84,85]. NA is a type II membrane 342 protein [86, 87], meaning that the cytoplasmic tail and the transmembrane domain are 343 at the N-terminus and the ectodomain is at the C-terminus (Fig. 5B). The truncated 344 NA protein resulting from translation initiation at AUG43 would lack the cytoplasmic 345 tail and the first few amino acids of transmembrane domain. Interestingly, classical 346 studies of type II membrane proteins characterized a series of artificial mutants of NA in 347 an effort to determine the motifs required for membrane insertion of the protein [88,89]. 348 One of these NA mutants was an N-terminal deletion in which the first 42 nucleotides 349 were removed, effectively creating the NA43 protein that would be generated by 350 translation initiation at AUG43. In the context of a protein expression plasmid, this NA 351 N-terminal deletion was efficiently expressed and localized to the cell surface, indicating 352 that the signal and anchor domains are reasonably intact [88,89]. On the basis of this 353 prior work, we hypothesized that any NA43 protein generated by alternate translation 354 initiation at AUG43 of the wildtype NA gene would also create a cell-surface NA 355 protein lacking the cytoplasmic tail and a portion of the transmembrane domain. 356 We analyzed the sequences of NA genes across different influenza lineages to 357 examine if the AUG43 codon is evolutionarily conserved. Most N1 avian influenza and 358 human pdmH1N1 NAs have an AUG at nucleotide 43, as do the majority of N1 swine 359 influenza NAs (Fig. 5C). Some human seasonal H1N1 NAs and N1 swine influenza NAs 360 lack an AUG at nucleotide 43, but these strains usually have another near-cognate start 361 codon at the site instead (Fig. 5C). Therefore, the candidate downstream start site at 362 NA nucleotide 43 is present in most N1 influenza strains. 363 We next sought direct experimental evidence for the NA43 protein initiated by the 364 downstream start site. Towards this, we generated a series of mutants of the NA from 365 the WSN strain of influenza. In these constructs, we mutated one or both of the translation, then these constructs should encode both, just one, or neither of the 373 full-length NA and NA43 proteins. 374 We first used these constructs to test whether NA43 is produced and localized on 375 the cell surface. We did this by transfecting 293T cells with protein expression plasmids 376 encoding the various NA mutants with a C-terminal V5 tag, which does not disrupt NA 377 folding or function and can be detected by flow cytometry [90]. NA protein was 378 detected on the surface of cells transfected with plasmids encoding mutants that lacked 379 an AUG at the canonical start codon (

393
We also quantified the levels of NA's enzymatic activity. As has been described  As an additional check for initiation at AUG43 in the presence of upstream starts, 431 we also generated constructs that introduce a frameshift just preceding position 43, such 432 that any initiation that occurs 5 to the frameshift will generate products that are not 433 detected by our V5 antibody. Only the construct that contains an AUG at position 43 434 (and not GUA at position 43) produces a NA43 band (lane 5 versus lane 6 of 435 Fig. S16B). Therefore, we conclude that the AUG at position 43 can initiate translation 436 in both the presence and absence of the canonical AUG start codon. 437 We next examined the extent to which NA43 could complement NA during viral 438 replication. To do this, we used reverse genetics [97] to attempt to generate WSN 439 influenza viruses with NAs that lacked one or both of the start sites. By far the highest 440 viral titers were obtained by using the wildtype NA (Fig. 5F). However, we also 441 obtained low but detectable viral titers when using NA that lacked the canonical start 442 codon but had the AUG at position 43, but obtained no detectable viral titers when 443 both start codons were mutated (Fig. 5F). The much lower viral titers when using the 444 mutant that just produces NA43 could be due to a combination of reduced expression 445 and the possible failure of NA protein lacking the cytoplasmic tail and a portion of the 446 transmembrane domain to effectively localize to regions of viral budding [98][99][100][101][102][103]. 447 Notably, there was no detectable change in viral titer if we mutated the AUG at 448 position 43 but maintained the canonical start codon. Therefore, although NA43 can 449 weakly complement for full-length NA, it is not important for viral growth in the 450 presence of full-length NA in cell culture.

451
The results in Fig. 5F suggest that NA43 does not substantially contribute to viral 452 growth in cell culture, but simply titering viral supernatants generated by reverse 453 genetics is a relatively insensitive way to quantify how mutations affect growth. We 454 therefore performed competition assays between viruses with wildtype NA and NA 455 lacking the start site at position 43, since such assays are a more sensitive way to detect 456 small differences in viral fitness [104]. We performed these assays by mixing virus with 457 wildtype NA with either the 1 wt 43 GUA or 1 wt 43 UUA mutant in roughly equal 458 proportion, and then allowing the viruses to compete in low-MOI infections in cell 459 culture. We then used deep sequencing to quantify the relative ratio of the wildtype and 460 mutant NAs both prior to extensive secondary replication (10 hours post-infection) or 461 after many rounds of replication (72 hours post-infection). If NA43 is important for 462 viral fitness in cell culture, then we would expect virus with the wildtype NA (which has 463 the start codon for NA43) to become enriched relative to either mutant. However, there 464 was no enrichment of virus with the wildtype NA (Fig. 5G), indicating that NA43 is not 465 detectably important for viral fitness in cell culture in the presence of full-length NA. 466 We next considered the possibility that NA43 might contribute to viral fitness only 467 in vivo. For instance, it is known that NA has additional roles that impact fitness in 468 vivo, such as helping the virus access cells in the presence of mucins [105]. We therefore 469 performed similar competition assays in mice, sequencing the viral population in the  interferon-stimulated genes are induced during the host anti-viral response [106]. We 485 sought to use our Ribo-seq and Ribo-seq + LTM data to examine translation initiation 486 on host transcripts during influenza infection and the host anti-viral response.

487
Our strategy for calling translation initiation sites (TIS) on host transcripts was 488 similar to the one we used for influenza transcripts (Fig. 3A). We used less stringent 489 criteria to account for the lower read coverage of host transcripts compared to influenza 490 transcripts (see Methods). We validated our TIS calling strategy for host transcripts 491 using data from a previous study [31]. Our calling method resulted in slightly higher 492 number of annotated TIS and lower number of downstream and upstream TIS than in 493 the TISdb database [83] created using the same data (Fig. 6A). This observation is 494 expected from our conservative strategy of using Ribo-seq data in addition to Ribo-seq 495 + LTM data to decrease false positives resulting from library preparation biases.

496
Application of our TIS calling strategy to the four samples in our study identified 497 around 2000 candidate TIS in each sample (Fig. 6B). Across all four samples, over 75% 498 of the called TIS were within 1 nucleotide of AUG codons, with CUG and GUG being 499 the next most abundant codons at the called TIS (Fig. 6B). This proportion of AUG 500 and near-cognate AUG codons was similar to those observed with data from HEK293T 501 cells (Fig S17A) [31]. 875 of the called TIS were shared between all 4 of our samples 502 (Fig. 6C), which we designate as high-confidence TIS for further analyses. Among these 503 high-confidence TIS, over 60% corresponded to annotated canonical start codons 504 (Fig. 6D), further validating our start site calling strategy. Less than half of the high 505 confidence TIS identified in our study were shared with those identified from earlier 506 work on HEK293T cells (Fig S17B). This modest overlap in called TIS likely reflects the 507 distinct gene expression landscape between the A549 cells used in our study and 508 HEK293T cells (Fig S17C). The uTIS were highly enriched for near-cognate AUG presentation of cryptic peptides [6]. We first sought to test the generality of this 514 observation at the genome-wide level using our called TIS set. Comparison of number of 515 called TIS between our four samples (Fig. 6B) did not reveal a globally higher 516 proportion of non-AUG TIS upon influenza infection, interferon-β stimulation, or under 517 both stimuli relative to the uninfected control sample (P > 0.05 for all three treatments, 518 binomial proportion test). Since non-AUG TIS tend to be enriched among uTIS and 519 dTIS in comparison to aTIS (Fig. 6E), consistently, we also did not find a higher 520 proportion of uTIS or dTIS upon any of the stimuli relative to the uninfected control 521 sample (P > 0.05 for all three treatments for both uTIS and dTIS, binomial proportion 522 test; Fig. S17E). 523 We then considered the possibility that the degree of increased non-AUG initiation 524 during influenza infection or interferon-β stimulation might be too weak to be detected 525

16/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under

17/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018. ; https://doi.org/10.1101/326967 doi: bioRxiv preprint as a globally higher number of non-AUG start codons in our TIS set called using the 526 Ribo-seq + LTM data. However, even a slightly higher degree of non-AUG initiation 527 during these inflammatory stimuli might favor evolutionary selection for a higher 528 proportion of non-AUG TIS in genes that are specifically up-regulated under these 529 conditions. To test if there is a higher proportion of non-AUG TIS among induced 530 genes, we first identified genes that were induced greater than 2-fold (average of 531 Ribo-seq and RNA-Seq counts) under each of the three treatments in our study.

532
Interferon-β treatment or interferon-β treatment followed by influenza infection resulted 533 in up-regulation of around 300 genes (>2-fold, Fig. 6F). As expected, the most highly 534 induced genes are well-characterized interferon-stimulated genes such as MxA or IFIT1. 535 The genes induced upon either interferon-β treatment or interferon-β treatment 536 followed by influenza infection show a high degree (> 90%) of overlap ( Fig S17F). By 537 contrast, influenza infection on its own resulted in up-regulation of a small set of 16 538 genes (>2-fold, Fig. 6F) including only a few interferon-stimulated genes. This 539 observation is consistent with recent work showing that activation of immune pathways 540 by influenza virus is rare at the single cell level during infections with viruses that have 541 relatively few defective particles [61]. 542 We examined the TIS codon identity and type in induced genes, and compared it to 543 a control set of all TIS that were called in any of our samples. This analysis revealed 544 that genes that were induced upon either interferon-β treatment or interferon-β 545 treatment followed by influenza infection had a higher proportion of near-cognate AUG 546 TIS relative to AUG TIS [34% (+ifn induced genes) / 33% (+ifn +vir induced genes) vs 547 26% (all genes), P = 0.02, binomial proportion test, Fig. 6G]. uTIS and dTIS were also 548 present in higher proportion than aTIS among the interferon-induced genes, as expected 549 from the over-representation of non-AUG codons among these TIS types [59% (+ifn 550 induced genes) / 58% (+ifn +vir induced genes) vs 47% (all genes), P < 0.01, binomial 551 proportion test, Fig. 6H]. We did not find the proportion of near-cognate AUG TIS or 552 uTIS and dTIS in the influenza-induced genes to be significantly different from those in 553 all genes (P > 0.05, binomial proportion test, Fig. 6G). This lack of statistical 554 significance is likely due to the meager number of called TIS (N=5) in the set of 555 influenza-induced genes. One important caveat that could bias these observations which 556 are based off our Ribo-seq + LTM data is the runoff of elongating ribosomes caused by 557 the lactidomycin treatment. The increased proportion of free ribosomal subunits under 558 these conditions could lead to artifactual detection of initiating ribosomes at start sites 559 that are not normally used in the absence of lactidomycin treatment. However, there is 560 no a priori reason why such artifactual detection will favor an increased proportion of 561 non-AUG initiation specifically on interferon-β induced transcripts over the remaining 562 transcripts. Thus our observations are consistent with the hypothesis [6] that 563 near-cognate AUG initiation at non-canonical start sites might be more common in an 564 anti-viral cellular state induced by interferon-β treatment. However, in our data this 565 increase arises from the presence of more non-canonical start sites in transcriptionally 566 induced genes, rather than increased non-canonical initiation on existing transcripts. . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018. ; https://doi.org/10.1101/326967 doi: bioRxiv preprint and NA. 576 We biochemically validated one of the new viral alternate start sites, a downstream 577 AUG at codon 43 of NA. We showed that the NA43 protein initiated by this start site is 578 produced even in the context of the canonical start site, is expressed on the cell surface, 579 and is enzymatically active. In addition, this NA43 protein is capable of supporting low 580 levels of viral growth even in the absence of full length NA. The AUG codon at 43 is 581 conserved in most N1 viral lineages. However, we were unable to find any evidence that 582 NA43 impacts viral fitness in cell culture or mice, at least at the relatively low 583 resolution with which viral fitness can be measured in laboratory settings. The 584 N-terminal cytoplasmic tail of NA helps localize the protein to areas of viral budding on 585 the cell surface [98][99][100][101][102][103]. Therefore, the N-terminally truncated NA43 could localize to 586 different regions of the cell surface than full-length NA. It is interesting to speculate 587 whether such altered localization of the NA43 protein could have some phenotypic 588 significance, such as reducing viral coinfection [107].

589
One of the motivations for our study was to search for evidence of alternate 590 translation initiation at CUG codons in influenza virus. We hypothesized that initiation 591 at such codons might lead to immune recognition of influenza virus, since CUG codons 592 can initiate immune epitopes [8][9][10][11][12][13], and initiation at CUG codons is thought to be 593 upregulated during conditions of cellular stress including viral infection [5][6][7]. We found 594 evolutionary signatures in the viral genome that were consistent with selection against 595 CUG-mediated translation initiation in lineages of influenza that have adapted to 596 mammalian hosts. However, we found minimal experiment support for CUG initiation 597 on influenza transcripts in our ribosome profiling experiments. One possibility is that 598 CUG initiation does generate evolutionarily relevant immune epitopes, but that the 599 levels of CUG initiation are too low to be detected by our experimental design. For 600 instance, CUG initiation could be partially refractory to capture by the translation 601 initiation inhibitor lactidomycin used in our experiments [5], or perhaps only occurs in 602 certain types of cells. In addition, even extremely low levels of translation initiation 603 that are difficult to detect by most experimental methods can generate peptides that 604 can still be recognized by T-cells [25]. A second possibility is that the selection against 605 CUG codons is due to some pressure unrelated to translation initiation. There are other 606 evolutionary signatures of influenza adaptation to mammalian hosts with uncertain 607 origin, such as the decrease in GC content of influenza genome during viral adaptation 608 to mammalian hosts [42]. 609 We also did not find significant shifts in global start site usage towards non-AUG 610 translation initiation during either viral infection of interferon treatment, despite the 611 fact that CUG initiation has been shown to be enhanced in these conditions using 612 biochemical and immunological assays [6]. However, the subset of transcriptionally 613 induced genes upon interferon-β treatment did have a significantly higher number of 614 non-AUG translation initiation sites than other genes. Whether these alternate start 615 sites serve any biological function will require further study. One interesting possibility 616 is that the peptides generated from these non-AUG start sites on anti-viral genes could 617 harbor T-cell epitopes that are relevant during the host immune response.

618
Our ribosome profiling experiments used the translation initiation inhibitor 619 lactidomycin to identify candidate start sites in influenza and host transcripts. One 620 potential concern is that the extended lactidomycin arrest of initiating ribosomes could 621 cause a traffic jam of scanning pre-initiation complexes [108] and lead to promiscuous 622 initiation. Indeed more stringent initiation profiling methods have been developed to 623 address this concern [81]. However, our goal was to detect start sites that might have pre-initiation complexes is not a major problem in analysis of the viral data. If 629 anything, downstream start sites might be more prone to being missed by our method 630 due to blocks caused by the initiating ribosome at the canonical start codon.

631
Overall, our work provides the first comprehensive analysis of translation initiation 632 during influenza virus infection. We identified several new alternate translation 633 initiation sites, one of which produces a functionally active viral protein that has not 634 been previously described. However, we found little evidence for large-scale initiation of 635 translation at non-canonical start codons such as CUGs on viral transcripts, and only a 636 modest increase in the proportion of non-canonical and non-AUG start codons on host 637 transcripts upregulated during the anti-viral response. The relative paucity of evidence 638 for virally induced alternate translation initiation in our ribosome profiling experiments 639 is seemingly at odds with many studies [5][6][7][8][9][10][11][12][13] highlighting its potential role in  [109]. Viruses from the 2009 swine-origin H1N1 pandemic were excluded. We 666 subsampled our sequences so that we kept at most 5 randomly chosen strains per year. 667 For human H5N1 influenza, we obtained sequences from 1997 to 2011, and did not 668 subsample our sequences due to the low number of sequences.

669
For duck influenza, we obtained sequences from 1956 to 2011. We subsampled our 670 duck influenza sequences so that we kept at most 5 randomly chosen strains per year.

671
For chicken influenza, we obtained sequences from 1934 to 2011. We subsampled our 672 chicken sequences so that we kept at most 5 randomly chosen strains per year. 673

20/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under In Fig. 1A, Fig. 1B, Fig. S1, and Fig. S2B we counted the number of the indicated motif 682 (eg. CUG codons) in the indicated reading frame for PB2, PA, NP, NS1, M1, M2, and 683 NS2 protein coding sequences for the viral strains contained in the human, duck, 684 chicken, swine, H5N1 sequence sets. 685 We also calculated the motif odds ratio (OR) as in [42]. The OR accounts for the 686 individual nucleotide content of a sequence, and therefore removes the effect of any 687 underlying changes in nucleotide content. The OR is defined as follows (shown for the 688 codon CUG): The OR is defined as the frequency of a given motif (CUG) in a sequence divided by 690 the product of the frequency of each nucleotide that comprise the motif (C, U, and G) 691 the sequence. Frequency of a motif such as CUG is defined as the number of codons in 692 a sequence that are CUG divided by the total number of codons in a sequence.

693
Frequency of a nucleotide is the number of a given nucleotide divided by the total 694 number of nucleotides in a sequence. A value greater than 1 indicates there is an excess 695 of the motif given the nucleotide content, and a value less than 1 indicates that there 696 are fewer of the motif given the nucleotide content.

697
In Fig. S2A, we calculated the odds ratio for CUG for the indicated reading frame in 698 each set of protein-coding sequences for the human, duck, chicken, swine, H5N1 699 sequence sets.

701
For the non-canonical reading frames 1 and 2 of influenza we also examined selection 702 against putative CUG initiation by examining whether there was a host-specific 703 difference in putative ORF lengths (Fig. 1C). We selected a timeframe where the human 704 and swine influenza strains should be reasonably host adapted (1970 to 2011), and took 705 sequences for the human, duck, chicken, swine, H5N1 sequence sets from that time. We 706 calculated the length of all putative ORFs initiated by CUG or CUH (where H is A, U 707 or C) in reading frames 1 and 2 for each influenza sequence set. To plot the data, we   To specifically examine putative alternate initiation at the CUG codons that are 727 selected against in reading frame 0 of influenza, we recoded the PR8 NP to contain 728 either few (low CUG PR8 NP) or many CUG codons (high CUG PR8 NP) (sequences 729 are in S4 File). We specifically chose to recode the CUG content of NP because NP 730 contains many CD8 T-cell epitopes [62,63] and CUG initiation can lead to the 731 generation of CD8 T-cell epitopes [5,6]. Furthermore, we chose PR8 NP as the CD8 732 T-cell epitopes in PR8 NP for murine models of infection are well 733 characterized [62,111,112]. To generate low CUG PR8 NP, we depleted PR8 NP of the 734 most common alternate start codons AUG, CUG, and GUG as much as possible in all 735 reading frames. We did this because we used low CUG PR8 to generate high CUG PR8, 736 and we wanted to begin with a low background of possible alternate initiation sites. We 737 generated high CUG PR8 NP by adding 20 CUG motifs that occur in reading frame 0 738 of natural influenza NP sequences to low CUG NP. The mutations we introduced to 739 generate low CUG PR8 NP and high CUG PR8 NP are made with the following 740 constraints: changes must be synonymous with regards to reading frame 0, any 741 synonymous codons that are introduced are chosen to be as frequent as possible in 742 natural existing sequences, and codons must exist in at least 100 of the sequences. The 743 sequences we used for this analysis are all full-length influenza A NP coding sequences 744 from the Influenza Virus Resource. Sequences were aligned using MUSCLE [110], and 745 alignments are included in S2 File.

746
NA43 plasmids for MUNANA and NA surface expression 747 To measure NA surface expression and NA MUNANA activity in Fig. 5D, we generated 748 constructs by placing NA into an HDM plasmid in which expression of the insert is 749 driven by the CMV promoter. Following the CMV promoter, we included the NA viral 750 5 UTR, NA coding sequence, a C-terminal V-5 epitope tag (used for surface staining), 751 and an internal ribosomal entry site (IRES) GFP (used for calculating transfection 752 efficiency). 753 We made the following set of mutagenized constructs for WSN NA:   We also made a construct that lacks the V5 tag as a negative control for background 771 V5 staining (pHDM-WSN-NA-IRES-GFP).

NA43 plasmids for Western blots 773
To examine whether there is initiation at site 43 of WSN NA (Fig. 5E), we generated 774 constructs in an HDM plasmid in which expression of the insert is driven by the CMV 775 promoter and the 5 UTR of WSN NA. The protein coding sequence consists of codons 776 1-40 of WSN NA fused to Histone-2B, followed by a C-terminal V5 epitope tag (for 777 immunoblot detection). The stop codon is followed by the remainder of NA and IRES 778 GFP. We made the following constructs:

23/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018. ; https://doi.org/10.1101/326967 doi: bioRxiv preprint construct that lacks the 5 viral UTR and begins at the AUG at coding nucleotide 322 812 which serves as a size control for NP322 (pHDM-322start-wt-PR8-FLAG). By HA staining 860 We infected A549 cells in WSN growth media for 10 hours, collected cells, and stained 861 for HA using H7-L19 antibody [115,116] at 10 ug/ml, followed by goat anti-mouse IgG 862 secondary antibody conjugated to APC, and analyzed by flow cytometry to determine 863 the fraction of cells that were HA positive. We used the Poisson equation to calculate 864 viral titer with respect to HA expressing units.

865
By TCID50 866 We titered virus by TCID50 on MDCK-SIAT1 or MDCK-SIAT1-TMPRSS2 expressing 867 cells and calculated viral titer using the Reed-Muench formula [117] implemented here: 868 https://github.com/jbloomlab/reedmuenchcalculator.  where explicitly indicated otherwise. To obtain higher resolution mapping of ribosome 932 protected fragments, we performed variable trimming based on fragment length, reads 933 that were between 25 and 32 nucleotides in length were assigned to a P-site at 14 934 nucleotides from the 5 end of the read, and reads between 33 and 39 nucleotides were 935 assigned to a P-site at 15 nucleotides from the 5 end of the read.  Proportion of Ribo-seq reads aligning to each reading frame 950 Normalized P-site density across reading frames was calculated for the coding regions of 951 the non-redundant set of transcripts included in the analysis used for read aggregation 952 plots. The single-and dual-coded regions of M, NS and PB1 were parsed from S5 File, 953 and the normalized P-site density across each of the regions was plotted.

954
Calling and analysis of candidate TIS 955 We used a zero-truncated negative binomial distribution (ZTNB) to statistically model 956 the background distribution of Ribo-seq and Ribo-seq + LTM counts in transcripts with 957 more than 50 positions with non-zero counts [80,81]. We first added the Ribo-seq and 958 the Ribo-seq + LTM read counts of the two neighboring positions to each position in 959 the genome (referred to as pooled counts below). This was done in order to account for 960 the +/-1 nucleotide uncertainty in the P-site assignment.

961
Candidate start sites were identified based on the following criteria: For influenza 962 transcripts, the ZTNB-based P-value for the Ribo-seq + LTM pooled counts at that 963 location must be <0.01 and 1000-fold higher than the P-value of the Ribo-seq pooled 964 counts at the same location (Fig. 3A, left panel), or must have an absolute P-value less 965 than 10 −7 . For host transcripts, we required the Ribo-seq + LTM P-value to be only 966 100-fold higher than the P-value of the Ribo-seq pooled counts. Additionally, for host 967 transcripts, we required that the read counts be greater than an absolute threshold 968 across all transcripts. This threshold was estimated by requiring P<0.05 in a ZTNB 969 model fit to the bottom 99% of all non-zero P-site Ribo-seq + LTM pooled counts 970 across all transcripts. Only the highest pooled counts within each 30 nucleotide window 971 was called as a candidate TIS. From the called TIS, we assigned the identity of the start 972 codon by looking at a window -1 to +1 nucleotides from the TIS peak and assigning the 973 start codon based on following hierarchy: AUG, CUG, GUG, UUG, AUA, AUC, AUU, 974 AAG, ACG, AGG, and other. If there are multiple near cognate codons in the window, 975 the codon was assigned based on the order in the above list.

976
For Fig. 3H and 3I, we consider the canonical frame defined by the aTIS as frame 0, 977 and we designate any ORF out of frame with the aTIS as an out-of-frame ORF. For 978 candidate host TIS, we designate the start codon of the canonical transcript of each gene 979 (as defined above) as the annotated TIS. uTIS, dTIS, and the frame of their ORF are 980 identified with respect to the annotated TIS. Host genes that have a median-normalized 981 fold-change in average Ribo-seq and RNA-seq counts greater than 2-fold upon +ifn, 982 +vir, or +ifn +vir treatment are considered as induced genes under that condition.

983
For analysis of data from [31], we downloaded the raw sequencing data from the 984 Sequence Read Archive, BioProject PRJNA171327, and analyzed it using the same 985 pipeline that we used for our data. The same pipeline was also used for analysis of the 986 raw sequencing data from [27] (NCBI Gene Expression Omnibus, GSE82232). The only 987 difference from the analysis of our dataset is that we identified P-site as the 13th 988 nucleotide from the 5 end of the read for both these datasets.

989
Binomial proportion test for comparing proportions of different TIS was done using 990 the R function prop.test with the alternate hypothesis set to greater.

991
Examining initiation at CUG codons in high and low CUG NP variants 992 In Fig. 4A, we considered reads that map with zero mismatches to either low or high 993 CUG NP or both NP variants. We designated reads as belonging to one of the variants 994 if it spans a SNP (recoded CUG codon), and thus could be identified uniquely. Coverage 995 of non-unique, low CUG, and high CUG variants is shown as a stacked bar blot. For 996 Fig. 4B, we plot the ratio of P-site count between the high and low CUG variants at 997 27/53 . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018. ; https://doi.org/10.1101/326967 doi: bioRxiv preprint each NP position to the mean value of the same quantity across the two variants. All 998 P-sites have a pseudocount of 1 added to both the low and high CUG read counts.

999
NA surface expression and MUNANA NA activity assay 1000 We performed NA surface expression and NA activity assays (Fig. 5D), as described  . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018. ; https://doi.org/10.1101/326967 doi: bioRxiv preprint anti-GAPDH at a 1:2500 dilution (RD systems AF5718), followed by a 1:2500 1043 Alexa-Fluor 680 donkey anti-goat secondary (Invitrogen A-21084). We used anti-H3 at 1044 a 1:10000 dilution (abcam 1791), followed by a 1:10000 Alexa-Fluor 680 goat anti-rabbit 1045 secondary (Invitrogen A-21109). Western blots were imaged using the LI-COR imaging 1046 system.

1048
To examine the conservation of the AUG at coding site 43 in Fig. 5C, we downloaded all 1049 full-length N1 NA protein-coding sequences from the Influenza Virus Resource. We used 1050 phydms [123] to construct a codon-level alignment in reference to the WSN sequence 1051 used in the ribosome profiling experiment. The alignment is subsampled such that all 1052 sequences have at least 2 amino acid differences relative to other sequences. This is to 1053 avoid having the alignment dominated by many highly similar sequences that are 1054 heavily represented in the database. We parse sequences into the following four sets: 1055 human seasonal H1N1 sequences isolated before 2009, human pandemic H1N1 sequences 1056 isolated after 2009, all avian sequences, and all swine sequences. We further subset 1057 human seasonal H1N1 sequences so that we only keep 1 sequence per year. Sequences 1058 are in S3 File. We then count the number of each codon at coding nucleotides 43-45 for 1059 our set of sequences. We consider CUG, GUG, UUG, ACG, AGG, AUC, AUU, AAG, 1060 AUA as near cognate start codons.

1061
Competition assays to examine impact of NA43 on viral fitness 1062 For the competition assays ( Fig. 5G and H) to examine if NA43 conferred a viral fitness 1063 advantage, we made 6 mastermixes of virus containing 1:1 mixes (as measured by  . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018. ; https://doi.org/10.1101/326967 doi: bioRxiv preprint were 6-8 weeks old, from Jackson labs. For infections, mice were first anesthetized with 1088 0.2 mg ketamine and 20 µg xylazine. Mice were weighed daily and lungs were harvested 1089 and flash frozen on day 4 post infection. 1090 We homogenized whole lung in 2.4 ml buffer RLT using the gentleMACS dissociator. 1091 The homogenate was clarified by centrifugation, and 700 µl of supernatant was used for 1092 RNA extraction using the Qiagen RNeasy kit, following the manufacturer's protocol.

1118
The resulting DNA libraries were sequenced on an Illumina Hi-Seq. We processed 1119 the sequencing reads to determine the frequency of the wildtype NA and mutant NA in 1120 each competition as follows. To determine the frequency of wildtype NA, we aligned our 1121 samples to the 1 wt 43 wt sequence, and counted the number of reads containing AUG at 1122 coding nucleotides 43-45. To determine the frequency of mutant NA, we aligned our wildtype to mutant at the 10 hour cell-culture timepoint. We compared the endpoint to 1134 the 10 hour timepoint instead of assuming that the initial ratio is 50:50 as the precision 1135 of TCID50 is less than that of deep sequencing, and the 10 hour timepoint allows us to 1136 30/53 . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under indicates an enrichment of wildtype over time. To assess significance, we performed two 1138 one-sided paired t-tests (mouse 1 wt 43 GUA and mouse 1 wt 43 UUA ). For the t-test, we 1139 used the log transformed ratios of wildtype to mutant at each timepoint. For the mouse 1140 assay, we first took the mean of the 3 technical replicates (using the log transformed 1141 value), and paired the average mouse value for each biological replicate to the 10 hour 1142 cell culture timepoint. The sequencing counts of codons and the resulting ratios used in 1143 analysis are included in S1 Table. 1144 8. Boon T, Van Pel A. T cell-recognized antigenic peptides derived from the cellular genome are not protein degradation products but can be generated directly by transcription and translation of short subgenic regions. A hypothesis. Immunogenetics. 1989;29(2):75-79.

31/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under   . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under   . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018. ; https://doi.org/10.1101/326967 doi: bioRxiv preprint

44/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under  Ribo-seq coverage along influenza transcripts for +ifn +vir sample. P-site counts from Ribo-seq and Ribo-seq + LTM assays are shown for all 8 influenza genome segments for our +ifn +vir sample. The counts from the two assays are shown as stacked bar graphs for ease of comparison. The candidate annotated TIS (circle) and downstream TIS (triangle) shared between the +vir and +ifn +vir samples are indicated below the coverage plots.

46/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018. ; https://doi.org/10.1101/326967 doi: bioRxiv preprint The ratio of high CUG NP to low CUG NP coverage from A is plotted against their sum along the horizontal axis. There were no RNA-seq reads with counts at 322, so this point is not highlighted. (C) The green-highlighted region in A around the CUG322 codon is shown at greater horizontal magnification. See Fig. 4 for +vir sample.

49/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under Blue arrow corresponds to full length NP and orange arrow corresponds to expected size of NP fragment due to initiation at nucleotide 322. The blot was overexposed to sensitively detect truncated peptides, leading to saturation of the the full length NP band (shown in cyan).

50/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018.

52/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018. ; https://doi.org/10.1101/326967 doi: bioRxiv preprint S1 Table. Deep sequencing from NA43 competition. Sequencing counts and ratios calculated for cell culture and mouse 1 wt 43 wt verses 1 wt 43 GUA and 1 wt 43 UUA virus competitions. S4 File. Influenza genome. This file contains the influenza genome used for our ribosome profiling analysis, including low and high CUG PR8 NP sequences. S5 File. Influenza GTF. This file contains annotations for influenza used for our ribosome profiling analysis.

53/53
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted May 21, 2018. ; https://doi.org/10.1101/326967 doi: bioRxiv preprint