Temperature-dependent regulation of upstream open reading frame translation in S. cerevisiae

Background Translation of an mRNA in eukaryotes starts at an AUG codon in most cases, but near-cognate codons (NCCs) such as UUG, ACG, and AUU can also be used as start sites at low levels in Saccharomyces cerevisiae. Initiation from NCCs or AUGs in the 5′-untranslated regions (UTRs) of mRNAs can lead to translation of upstream open reading frames (uORFs) that might regulate expression of the main ORF (mORF). Although there is some circumstantial evidence that the translation of uORFs can be affected by environmental conditions, little is known about how it is affected by changes in growth temperature. Results Using reporter assays, we found that changes in growth temperature can affect translation from NCC start sites in yeast cells, suggesting the possibility that gene expression could be regulated by temperature by altering use of different uORF start codons. Using ribosome profiling, we provide evidence that growth temperature regulates the efficiency of translation of nearly 200 uORFs in S. cerevisiae. Of these uORFs, most that start with an AUG codon have increased translational efficiency at 37 °C relative to 30 °C and decreased efficiency at 20 °C. For translationally regulated uORFs starting with NCCs, we did not observe a general trend for the direction of regulation as a function of temperature, suggesting mRNA-specific features can determine the mode of temperature-dependent regulation. Consistent with this conclusion, the position of the uORFs in the 5′-leader relative to the 5′-cap and the start codon of the main ORF correlates with the direction of temperature-dependent regulation of uORF translation. We have identified several novel cases in which changes in uORF translation are inversely correlated with changes in the translational efficiency of the downstream main ORF. Our data suggest that translation of these mRNAs is subject to temperature-dependent, uORF-mediated regulation. Conclusions Our data suggest that alterations in the translation of specific uORFs by temperature can regulate gene expression in S. cerevisiae.

aligned to the S. cerevisiae rRNA database using Bowtie [33]. The reads that did not align with 224 the reference database (non-rRNA reads) were aligned to the S. cerevisiae genome or to the 225 FASTA file made from the reporter sequence (pR AUG F UUG ) to generate alignments to the F-Luc 226 reporter mRNA using TopHat [34]. for the mORF were employed rather than the mRNA read counts in the uORF alone. Spearman's 249 11 correlation was calculated using an online tool at https://www.wessa.net/rwasp_spearman.wasp.

255
Wiggle tracks: A combined alignment file (Bam file) was generated using two alignment files 256 (one for each of the two biological replicates). The combined files were generated for both RPF 257 samples and total RNA samples (for 20 °C, 30 °C and 37 °C ). Wiggle files were generated from 258 this combined alignment file. Files were generated for each gene on the Watson or Crick strand.

259
The tracks were visualized using Integrative Genomics Viewer (IGV 2.4.14). The tracks were 260 normalized according to total number of mapped reads in the combined file. To normalize the 261 effects of changes in mRNA levels, the total read-normalized peaks were scaled with respect to 262 the changes in mRNA levels to reflect the changes in translation efficiency (see Results section).

263
Wiggle tracks, both with and without the scaling by changes in mRNA levels, are shown.

265
Finding upstream ORFs: We took a similar approach to identify possible uORFs and confirm 266 their translation in our experiments as that described previously [19]. First, putative translated 267 uORFs were identified essentially as described [38]. Briefly, for all open reading frames in 268 annotated 5'-UTRs that initiate from either AUGs or near cognate codons, the ratio of RPF counts 269 at the +1 position (start codon of uORF) to -1 position (upstream of the start codon) was 270 calculated. Those uORFs with ratios > 4, with >14 RPF counts at the +1 and −1 positions 271 combined, and with at least 50% of the counts reads in the 0-frame with respect to the start site 272 (i.e. the relevant line of code is -c15-r4-z0.5), were selected for further analysis. The multiple 273 ribosome profiling datasets we used to identify potential uORFs were described previously [19] 274 12 and have been submitted to the NCBI Gene Expression Omnibus and the accession numbers are 275 listed https://elifesciences.org/articles/31250/figures#supp1 in the additional file table S2.

276
To determine which of these putative uORFs were translated in our experiments, we 277 employed an ORF identification tool (RibORF) described previously [39], which uses 3-nucleotide 278 periodicity and a uniform distribution of RPFs counts across the uORF as scoring criteria. We 279 applied a moderately stringent cutoff of the probability of prediction of 0.5 and used a combined 280 alignment file generated from footprint libraries of all 6 biological replicates grown at three different 281 temperatures to identify uORFs with evidence of translation at one or more growth temperatures.

282
After excluding uORFs shorter than three codons, we identified 1367 uORFs starting with AUG 283 (N = 142) or a NCC (N = 1225). Quantifying the total mRNA and RPF counts in the 5'-UTR, uORF 284 or mORF was done as previously described [19].

285
To identify potential uORFs in all yeast mRNA 5'-UTR transcriptome, 5'-UTR sequences 286 for all mRNAs were extracted, and each AUG and NCC nucleotide triplets were searched 287 throughout the 5'-UTR. We identified the first in frame stop codon for each start codon as the end

295
We previously developed and validated a dual luciferase assay to calculate the efficiency 296 of utilization of near-cognate codons (NCCs) as translational start sites in yeast [20]. In this assay,

297
Renilla luciferase (R-Luc) and Firefly luciferase (F-Luc) are expressed using separate promoters 298 and transcription terminators from a single low-copy plasmid ( Figure 1A). R-Luc mRNA has an 299 13 AUG as the start site and acts as an internal control for cell growth, lysis efficiency and pipetting 300 inconsistency. The start codon of the F-Luc reporter is varied and can be AUG or any NCC (e.g.

301
UUG, ACG, etc.). F-Luc expression is normalized to that of R-Luc to control for any global 302 changes in gene expression or cell growth. F-Luc AUG /R-Luc AUG represents the normalized 303 expression value for a F-Luc reporter starting with AUG. The relative expression from UUG (or 304 any other near-cognate) is calculated by normalizing with respect to the normalized F-Luc AUG 305 expression (F-Luc UUG /R-Luc AUG) )/(F-Luc AUG /R-Luc AUG ).

306
We used this assay to investigate the effects of changes in growth temperature on the use 307 of NCCs as translational start sites. Yeast cells (BY4741) were transformed with the dual 308 luciferase reporter plasmid with either an F-Luc AUG or F-Luc UUG gene, cultured at various 309 temperatures, and the luciferase activity of the F-Luc UUG reporter relative to the F-Luc AUG reporter 310 was measured. (F-Luc and R-Luc assays in cell extracts were performed at 24 ˚C regardless of 311 the temperature at which the yeast cells were cultured.) Elevating the growth temperature from 312 30 °C to 37 °C led to ~1.5-fold increase in the normalized expression of F-Luc UUG , while lowering 313 the growth temperature from 30 °C to 25 °C or 20 °C led to ~1.6-fold and ~2.5-fold reduction, 314 respectively, in the normalized F-Luc UUG expression ( Figure 1B). These findings suggest that at 315 lower growth temperatures the efficiency of use of the F-Luc UUG start codon is decreased, and 316 that at higher temperatures it is increased.

317
To try to confirm these findings and test the generalizability of the observed effects of 318 temperature on start codon usage, we performed a similar experiment with two otherwise identical 319 HIS4-lacZ fusion reporters with an AUG or UUG start codon [40] (Additional file 1: Figure S1A).

320
The normalized expression of a HIS4-lacZ reporter with UUG as a translational start site was 321 reduced ~5-fold at 20 °C with respect to 30 °C and was elevated ~2.8-fold at 37 °C (Additional 322 file 1: Figure S1B). These results are consistent with the findings from the dual luciferase reporter 323 that the use of a UUG relative to an AUG start codon is reduced at 20 ˚C and increased at 37 ˚C.

324
Frequently, the finding that two orthogonal reporter assays give similar results might be taken to 325 14 indicate that the observed effect is generalizable to most mRNAs. However, as described below, 326 this turns out not to be the case in this system.

327
We next tested the effects of changes in growth temperature on the normalized expression 328 from F-Luc reporter mRNAs starting with other NCCs (GUG, CUG, ACG, AUA, AUC and AUU), 329 all of which have been shown to be utilized as start sites in yeast cells to varying extents [4,20].

330
Like UUG, the normalized expression from all NCCs was significantly lowered (~2 fold) at 20 °C 331 indicating that the effects of lowering the growth temperature were not specific to the UUG start 332 site ( Figure 1C). On the other hand, elevation in growth temperature resulted in differential 333 changes in normalized expression, ranging from no change for ACG to ~30% increase (for AUC 334 and GUG) as compared to their expression at 30 °C. This suggested that the efficiency of use of 335 NCCs might be differentially affected at some temperatures.

336
The changes in expression of the F-Luc reporters starting with different initiation codons 337 could be due to changes in mRNA stability induced by the nonsense-mediated decay (NMD) 338 pathway triggered by altered translation of an upstream open reading frame (uORF) starting from 339 an NCC. In the absence of an AUG start codon, it is possible that an upstream or out-of-frame 340 NCC codon is used for initiation leading to premature termination, and potentially NMD [41].

341
Alternatively, it is also possible that the scanning 43S pre-initiation complexes bypass the NCC 342 start site (leaky scanning), initiate at an out-of-frame AUG or NCC in the mORF, terminate in the 343 mORF, and thereby trigger NMD [42]. To test these possibilities, we performed the luciferase 344 assay in a strain in which the UPF1 gene, which encodes a protein essential for NMD, had been 345 deleted (upf1Δ) [43]. Deletion of UPF1 did not affect the temperature-dependent changes in 346 expression observed in WT cells for the majority of the start codons tested at both 20 °C and 37 347 °C ( Figure 1D). This result suggests that the NMD pathway does not play a role in the observed 348 changes in expression from these reporters at different temperatures.

349
The alterations in the use of UUG as a start site could be attributed to changes in the 350 levels of eIF1, which has been shown to be a 'gatekeeper' in the start codon recognition process 351 15 [44], helping to restrict start codon selection to AUGs and block the selection of NCCs. To test 352 the possibility that temperature-dependent changes in eIF1 expression might be responsible for 353 the observed effects of temperature on start codon usage, we assessed the levels of the factor 354 using western blot analysis of whole cell lysates from cells cultured at different temperatures.

355
Levels of eIF1 protein were not significantly altered at 20 or 37 ˚C relative to 30 ˚C (Additional file 356 1: Figure S2A), suggesting changes in start codon use are not due to changes in eIF1 357 concentration.

358
We also tested the effect of over-expression of eIF1 on the observed temperature-359 dependence of F-Luc start codon utilization. Over-expression of eIF1 from a high-copy (hc) 360 plasmid has been shown to suppress the reduced stringency of start codon recognition (Sui -) 361 phenotype caused by mutations in several initiation factors [45][46][47][48]. Consistent with its role as a 362 central gatekeeper of start codon recognition, over-expression of eIF1 (hc-SUI1) suppressed the 363 use of UUG as a start site at all three temperatures (Additional file 1: Figure S2B). The decrease 364 in F-Luc UUG expression at 20 ˚C and the increase at 37 ˚C were still observed in the hc-SUI1 365 strain, although the magnitude of the increase at 37 ˚C was reduced relative to WT cells in this 366 experiment. Consistent with these results, reducing the concentration of eIF1 relative to WT cells 367 by using a SUI1/sui1D heterozygous diploid strain, resulted in increased expression of F-Luc UUG 368 relative to F-Luc AUG at all three temperatures relative to expression in a WT diploid containing two 369 wild-type chromosomal alleles of SUI1 (Additional file 1: Figure S2C). No decrease in the 370 magnitude of the temperature-dependence of normalized F-Luc UUG expression was observed in 371 the SUI1/sui1D haploinsufficient diploid. In addition, haplo-insufficiency of eIF1A (+/tif11D) or eIF5 372 (+/tif5D), also factors involved in start codon recognition, did not significantly alter the effect of 373 temperature on expression of F-Luc UUG relative to F-Luc AUG (Additional file 1: Figure S2C). It is 374 noteworthy that, for reasons unknown, the increase at 37 °C is dampened in the WT SUI1/SUI1 375 diploid versus the SUI1 haploid strain analyzed in Figure 1A, and that the larger differences 376 resurface in the SUI1/sui1D heterozygote. Although altering the dosage of the SUI1 gene appears 377 16 to modulate somewhat the effects of 37 °C on UUG initiation, overall it appears that the effects of 378 temperature on NCC utilization are not dictated by altered cellular levels of eIF1.

379
Taken together, these results suggest that the effect of temperature on F-Luc UUG 380 expression is not due to changes in the concentrations of eIF1, eIF1A or eIF5. Additionally, when 381 we performed ribosome profiling in cells grown at different temperatures (see below) we did not 382 observe any obvious changes in the ribosome protected fragments (RPF) counts and translational 383 efficiencies for the mORF of eIF1, further confirming that the level of eIF1 does not change as a 384 function of growth temperature in WT haploid cells (Additional file 1: Figure S2D).

388
To investigate the effects of changes in growth temperature on the relative use of different codons 389 as translational start sites throughout the transcriptome, we performed ribosome profiling in yeast 390 cells cultured at multiple temperatures. WT yeast cells (BY4741) transformed with the F-Luc UUG 391 reporter plasmid were cultured in SC-Ura for 16 hours at 20, 30 and 37 °C and ribosome profiling 392 was performed ( Figure 2A) as previously described [8,27], with some notable changes (see 393 Methods section). In particular, we did not add cycloheximide to the intact cells to stop translation 394 because of the known artifacts it creates [29][30][31][32], but instead flash froze the cells in liquid nitrogen 395 and added cycloheximide to the cell lysis buffer only.

396
We calculated ribosomal-read density as the number of 80S ribosomal footprint reads 397 mapped to an mRNA sequence relative to the total number of reads in the footprint library (ribo-398 seq), and we calculated the mRNA read density by normalizing RNA-seq reads mapped to an 399 mRNA sequence relative to the total number of reads in the RNA-seq library. The translation 400 efficiency (TE) for each mRNA is calculated as ribosomal read density normalized to mRNA read

404
We mapped the ribo-seq (ribosome protected fragments, RPFs ) and RNA-seq reads to 405 the F-Luc UUG mRNA reporter to confirm our findings with the reporter assay. The expression of F-406 Luc UUG mRNA reporter was significantly altered at multiple temperatures ( Figure 2B and 407 Additional file 1: Figure S3B). The RPF read count for the F-Luc mRNA was decreased at 20 °C 408 by ~25% and increased at 37 °C by ~46% as compared to 30 °C (Additional file 1: Figure S3B).

409
In contrast, the mRNA-read count for the F-Luc mRNA was increased at 20 °C by ~300%, and 410 decreased at 37 °C by ~20%, as compared to 30 °C (Additional file 1: Figure S3B). To control for 411 the changes in mRNA abundance, we calculated the TE of the F-Luc reporter ( Figure 2C). TE for 412 F-Luc UUG mRNA was reduced at 20 °C by ~77% and elevated at 37 °C by ~80%, as compared to 413 30 °C, indicating that the translation of F-Luc mRNA is significantly altered at 20 °C and 37 °C, 414 consistent with the findings from the luciferase assay where the normalized expression of this 415 reporter (with respect to the F-Luc AUG control reporter, Figure 1A

419
To better understand the effects of changes in temperature on start site selection, we identified a 420 set of 1367 uORFs that show evidence of translation in our yeast strain at one or more 421 temperatures. To this end, we employed a two-step strategy for translated uORF discovery with an AUG or NCC on the basis of a strong peak of ribosome density at the start codon and the 425 occurrence of >50% of downstream read counts in the zero frame of the start codon. After 426 excluding uORFs shorter than three codons, we identified 6061 potential uORFs by applying this 427 algorithm to several previously reported ribosome profiling datasets (see Methods). In the second 428 18 step, we examined which of these 6061 putative uORFs show evidence of translation in our 429 ribosome profiling data using a different identification tool, RibORF [39], which is based on the 430 criteria of 3-nt periodicity (a hallmark of mRNA fragments protected by actively translating 431 ribosomes) and a uniform distribution of reads across uORF codons. This tool generates a 432 predicted translating probability ranging from 0 to 1. Lower probability values indicate skewed 433 distributions of reads and equally distributed fractions of reads at the zero, 1 st and 2 nd reading 434 frames. Higher values indicate a uniform distribution of reads and a majority of reads aligned in 435 the zero reading frame. Applying a moderately stringent probability of prediction of >0.5, we found 436 evidence for translation in our datasets for 1367 uORFs among the 6061 potential uORFs 437 detected in the first step, located on 755 different mRNAs. These uORFs were further analyzed 438 for changes in their expression at different temperatures using DESeq2 (see methods).

439
These 1367 translated uORFs start with either an AUG (~10%) or a NCC (~90%). We 440 observed a range of NCCs as start sites ( Figure 3B), with UUG the most common (~30% of all 441 uORFs) and AGG the least (~1% of all uORFs). The uORFs with near-cognate start codons with 442 2 nd base changes from AUG (AAG, ACG, AGG) contributed only 8% of all the uORF start codons, 443 with ACG at 6% and both AAG and AGG at 1%, , which is consistent with previous findings 444 indicating that AAG and AGG are the least efficiently used near-cognate start codons in yeast 445 cells [4]. Near-cognate codons with 1 st base changes (UUG, GUG, CUG) comprised ~50% of the 446 total uORF start codons indicating that they are the most efficient near-cognate start sites, also 447 consistent with previous studies [4] and our luciferase reporter analyses ( Figure 1C). Among the 448 ~10% of all uORFs with an AUG start codon, ~33% have a preferred Kozak context at the -3

452
We also looked at the overall abundance of AUGs and NCCs in the 5'-UTR transcriptome 453 after removing any that initiate predicted uORFs less than three codons in length (in order to 454 match the conditions used for identification of translated uORFs in our datasets, which also 455 eliminated potential uORFs less than three codons long; Figure 3C). Comparison of the start 456 codon distribution of the set of translated uORFs we identified ( Figure 3B) with the 5'-UTR 457 transcriptome abundance of potential uAUG and NCC start codons ( Figure 3C

465
To investigate the translatability of the uORFs we calculated translational efficiency (TE) 466 at 30 °C for all the uORFs starting with AUGs or NCCs. TE, as described above, is the ratio of 467 ribosomal footprint read density to mRNA read density. As shown in Figure 3D, uORFs starting 468 with different initiation codons were translated with differing median efficiencies at 30 °C. As might 469 be expected, the median TE for uORFs starting with AUG codons (AUG uORFs) was significantly 470 higher than the medians for uORFs starting with any NCC (NCC uORFs). In this and all other box 471 and whisker plots below, lack of overlap in the notches of two adjacent plots indicates that their 472 medians differ with >95% confidence (Chambers et al., 1983 Graphical Methods for Data 473 Analysis. Wadsworth, Bellmont). The median TE for AUG uORFs in good context was comparable 474 to that for AUG uORFs in poor context ( Figure 3D), suggesting that other features of these mRNAs 475 might modulate the effect of sequence context for this subset of uORFs.

476
The median TEs for NCC uORFs varied by a factor of about 2 to 3-fold, depending on 477 which base varied from AUG. Consistent with our analysis of the start codon distribution of 478 translated uORFs ( Figure 3B), uORFs starting with NCCs with 2 nd base changes were the least 479 efficiently translated, whereas uORFs starting with NCCs with 1 st base changes were translated 480 20 the most efficiently ( Figure 3D). These data, together with the results described in Figure 3B, 481 suggest differential recognition and utilization of NCCs as start sites for uORFs in yeast in a 482 manner consistent with previous analyses of the efficiency of different start codons for main ORF 483 translation [5,6,20].

484
We next examined the relationship between the translation efficiency of the uORFs and 485 that of the downstream main ORF (mORF). We calculated the TEs for the mORFs downstream 486 of each subset of uORFs grouped according to the uORF start codon. We observed that the 487 median TEs for the mORFs downstream of AUG uORFs were significantly lower than any other 488 group ( Figure 3E), suggesting that uORFs with AUG start sites were typically inhibitory of

517
To identify uORFs showing changes in TE that appear to be activated or repressed by a 518 change in growth temperature, we applied two criteria. First, we considered only those uORFs

519
showing an increase or decrease in TEuORF of ³ 2-fold at a given temperature with respect to 30 520 °C using a false discovery rate (FDR) of £ 0.1. 39 uORFs showed significant TE changes at 20 521 °C versus 30 °C; whereas 84 uORFs displayed such TE changes at 37 °C versus 30 °C 522 (Additional file 1: Figure S4A, S4B). We reasoned that changes in TEuORF could result because of 523 multiple mechanisms. For example, translation initiation on the mRNA as a whole could increase 524 or decrease because of changes in the efficiency of PIC attachment or scanning processivity, 525 leading to corresponding increases or decreases in the TEs of both the uORF(s) and mORF. To 526 exclude such changes in TEuORF occurring concurrently with similar changes in TEmORF, we 527 devised a term called 'relative-TEuORF' which is TEuORF/TEmORF at any given temperature.

528
Calculating changes in Relative-TEuORF (DRelative-TEuORF) helped to identify changes in uORF 529 translation not occurring simultaneously with similar changes in the TE of the mORF. Thus, 530 22 according to our second criterion, translation of a uORF was called 'regulated' if there was ³2-531 fold change (increase or decrease) in relative-TEuORF at a given temperature with respect to 30 532 °C; that is, TEuORF changed ≥2-fold more than TEmORF, or their changes were in opposite 533 directions. Applying these criteria, we identified uORFs whose translation is specifically regulated 534 by changes in growth temperature. We classified uORF translation as activated if both DTEuORF 535 and DRelative-TEuORF are ³ 2 and as repressed if both DTEuORF and DRelative-TEuORF are £ 0.5.

536
After applying these criteria, we found 36 uORFs showing temperature dependent 537 translational regulation at 20 °C. There were 16 uORFs whose translation was significantly 538 activated ( Figure 4A, red circles in panel (iii)) of which 2 were AUG uORFs (open circles) and 539 14 were NCC uORFs (solid circles). There were 20 uORFs whose translation was significantly 540 repressed at 20 °C ( Figure 4A, blue circles in panel (iii)) as compared to 30 °C, of which 5 were

542
showing temperature dependent translational regulation at 37 °C. There were 24 uORFs whose

547
were driven by changes in ribosome density (DRPF-density) and not by changes in mRNA levels 548 (DmRNA-density) for the uORFs regulated at either 20 °C or 37 °C (Additional file 1: Figure S4C), 549 as well as for all 1359 translated uORFs identified in this study (Additional file 1: Figure S4D

552
It has been reported that alternative transcription start sites can produce mRNA isoforms 553 in yeast with different translational efficiencies [51]. Thus, we also calculated Spearman 554 23 correlation coefficients between DTEuORF and the changes in reads of just the 5'-UTRs of these 555 sets of mRNAs (DmRNA5'UTR-density) and found that they are much smaller than those for DRPF-556 density and are not statistically significant (Additional file 1: Figure S4C and S4D, orange bars).

557
This result suggests that the changes in TEuORF observed are not due to temperature-dependent 558 alterations in transcriptional start sites that produce different levels of mRNA isoforms including 559 or excluding the uORFs in question. Furthermore, high Spearman correlation coefficient values 560 between DTEuORF and DRPF-density, and low coefficient values between DTEuORF and both 561 DmRNA-density and mRNA5'UTR-density were observed for all translated AUG and NCC uORFs 562 (Additional file 1: Figure S4E and S4F), indicating that for both these sets of uORFs the changes 563 in translational efficiency were driven by changes in RPF density and not in mRNA or 5'-UTR 564 density.

565
The F-Luc UUG and HIS4 UUG -LacZ reporters showed decreased translation at 20 °C and 566 increased translation at 37 °C ( Figures 1B, 2B, Additional file 1: Figure S1B) suggestive of altered 567 efficiency of use of the NCCs. In contrast to these reporters, we found changes in growth 568 temperature lead to a more diverse transcriptome-wide response of translation of uORFs starting 569 with not only NCCs but also AUGs. As described above, we identified 112 uORFs (14 AUG 570 uORFs and 98 NCC uORFs) (Additional file 2: Supplementary Table 3) on 84 different mRNAs 571 whose translation was activated or repressed in response to changes in growth temperature. It is 572 noteworthy that less than 10% of the set of 1359 translated uORFs have significantly altered 573 translation relative to changes in mORF translation at reduced or elevated growth temperatures 574 (20 or 37 °C), indicating that temperature changes in this range do not produce global effects on 575 the initiation efficiency of uORFs, but rather have specific, mRNA-dependent effects.

24
The distribution of start codons of the 112 regulated uORFs described above is shown in 579 Figure 5A and B. We separated the uORFs based on their start sites into four classes: uAUGs;

580
NCCs with 1 st base changes with respect to AUG (UUG, CUG, GUG), 2 nd base changes (AAG, 581 ACG, AGG) and 3 rd base changes (AUC, AUA, AUU). The number of uORFs in each bin is too 582 small to make inferences about the statistical significance of the differences. Thus we next 583 analyzed the changes in TE (DTEuORF) of all 1359 translated uORFs by binning them into groups 584 based on their start codon triplets without applying the criteria used to identify significant changes 585 ( Figure 5C, D). The black horizontal dotted line indicates the median DTEuORF for all uORFs 586 analyzed in this study at each temperature, which is close to unity. The NCC uORFs did not show 587 a significant difference in median DTEuORF at 20 °C versus 30 °C when compared to all uORFs 588 ( Figure 5C). In contrast, the AUG uORFs showed a significantly lower median DTEuORF at 20 °C 589 when compared to all uORFs, suggesting that use of uAUGs tends to be decreased at 20 °C.

590
We performed a similar analysis with uORF translation at 37 °C ( Figure 5D). The TE of 591 uORFs starting with NCCs with 3 rd base changes had a significant tendency to be downregulated

607
Similarly, the majority (9/14) of the AUG uORFs ( Figure 5E, green circles) are positioned in 608 Quadrant 1, indicating that translation of regulated uORFs starting with AUGs tends to be 609 decreased at 20 °C (log2TEuORF < 0) and increased at 37 °C (log2TEuORF > 0). All of the AUG 610 uORFs that met our criterion for regulated changes in translation are in Quadrants 1 or 2, 611 indicating that translation of AUG uORFs is generally increased at 37 °C relative to 30 °C. The

615
We performed a similar analysis with all translated uORFs (N = 1359) ( Figure 5F). The 616 overall distribution on this scatter plot is similar to that for the regulated uORFs in Figure 5E, 617 except that many uORFs whose translation is unaffected by temperature are present near the 618 middle of the plot. The significant numbers of points representing NCC uORFs along the vertical 619 axis between Quadrants 3 and 4 in both plots ( Figure 5E and 5F, pink circles) indicates that many 620 of these uORFs are repressed at 37 °C relative to 30 °C, but are relatively unaffected at 20 °C,

26
Recently, 982 uORFs were identified from S. cerevisiae in 791 mRNAs using a 629 comparative genomics approach to identify translated uORFs that are conserved in length or 630 sequence among yeast species [10]. Approximately 44% of these are AUG uORFs and ~31% are 631 UUG uORFs. When we interrogated this conserved uORF set, we found that, similar to our 632 observations with the translated uORFs described above, translation of the conserved AUG 633 uORFs is significantly repressed at 20 °C and activated at 37 °C (Additional file 1: Figure S5A-C, 634 green boxplots and circles). Intriguingly, the TE of the conserved NCC uORFs is on average 635 slightly elevated at 20 °C and slightly repressed at 37 °C (Additional file 1: Figure S5A-C, pink 636 boxplots and circles). Thus, a set of conserved AUG uORFs identified in a different manner than 637 was our set of translated AUG uORFs displays a similar overall response to temperature.

641
To look for possible mechanisms influencing translation of NCC and AUG uORFs, we investigated 642 whether intrinsic properties of these uORFs or their mRNAs display any significant correlations 643 with uORF translational efficiencies. We first used a dataset of ~2700 yeast mRNA 5'-UTR lengths 644 and propensities of forming secondary structures [52] to look for trends in the translated uORFs.

649
Translated AUG uORFs tend to be significantly closer to the 5'-cap (Additional file 1: Figure S6D) 650 and shorter (Additional file 1: Figure S6F) than are all translated uORFs or NCC uORFs. AUG 651 uORFs also tend to be on shorter 5'-UTRs (Additional file 1: Figure S6G uORFs (green versus gray boxplots). No statistical difference is seen between AUG and NCC 655 uORFs or all uORFs when distance from the mAUG is compared (Additional file 1: Figure S6E).

656
Next, we calculated the 'context adaptation scores' for the uORFs, as described previously 657 [19,53], quantifying the similarity between the start codon context of each uORF to that of the 658 mORF AUGs of the 2% of yeast mRNAs with the highest ribosomal loads [54]. The start codons 659 of the AUG uORFs have a significantly lower context score when compared to NCC uORFs 660 (Additional file 1: Figure S6N

674
UTR and DTEuORF at 20 °C for all NCC uORFs and all uORFs. Similar correlation was observed 675 for AUG uORFs, but they did not meet statistical significance because of the smaller number of 676 these uORFs. At 37 °C, the distance between the uORF start codon and the mAUG, the length 677 of the uORF, and the length of the 5'-UTR all had a significant positive correlation with DTEuORF 678 28 for all uORFs and NCC uORFs ( Figure 6, bottom panel). Again, similar correlations were observed 679 for these same parameters for AUG uORFs, but they did not meet statistical significance because 680 of the smaller number of these uORFs. Taken together, these data suggest that the position of a 681 uORF in the 5'-UTR and the length of the 5'UTR can influence how translation of the uORF 682 responds to changes in growth temperature.

683
In an effort to confirm these relationships between TE changes and uORF position relative

692
As shown in Figure

712
We did not observe any significant influences of start codon context, the uORF length, or

733
The simplest explanation for this behavior is that overall initiation on these mRNAs (e.g., PIC 734 loading onto the 5'-UTR) increases or decreases, leading by mass action to increases or 735 decreases in translation of both the uORF and the mORF. Alternatively, it is possible that some 736 cases in these two quadrants represent mRNAs on which re-initiation after translation of the uORF 737 is very efficient and thus an increase or decrease in uORF translation has a corresponding effect 738 on mORF translation.

739
We performed a similar analysis using the TE_up and TE_down NCC uORF sets 740 described above and in Figure 7A-C ( Figure 8C, D). As with the set of regulated NCC uORFs, 741 this set was also distributed into all four quadrants, with a preponderance in Quadrants 1 and 2.

742
We also performed this analysis for all translated AUG uORFs ( Figure 8E, F). Again, the 743 plot shows a distribution of mRNAs into all four quadrants. However, consistent with the behavior 744 of AUG uORFs described above, a majority (66%) are in Quadrants 1 and 4 (i.e., negative 745 DTEuORF) at 20 °C, whereas at 37 °C a majority (76%) are in Quadrants 2 and 3 (i.e., positive 746 DTEuORF).

747
Why TEmORF increases for the mRNAs described in Figure 8A-F more often than it 748 decreases regardless of the direction of DTEuORF is unclear, although it is possible that mRNAs 749 whose main ORF translation decreases (negative DTEmORF) are more likely to be degraded due 750 to the coupling between active translation and mRNA stability [55,56] and thus less likely to 751 appear in the ribo-or RNA-seq data.

752
In order to assess possible trends in the relative translation of the uORFs and mORFs, we 753 colored the circles corresponding to each uORF in the plots in Figure Figure S10A). At 20 °C the 781 TEs of these NCC uORFs were increased on average 1.6-fold relative to 30 °C, which had no 782 detectable effect on the TE of the mORF ( Figure 9A, compare black and blue traces, effective 783 increase in RROuORF ~2-fold). In contrast, at 37 °C the TEs of the NCC uORFs decreased ~3-fold, 784 which was accompanied by an ~2-fold increase in mORF translation (effective decrease in 785 RROuORF ~6-fold), suggesting that the NCC uORFs may be involved in regulating translation of 786 the GCN4 mORF at elevated temperatures. Further, we calculated the TEs of the AUG uORFs 787 manually as our pipeline does not allow the validation of uORFs shorter than three codons. We 788 found that these uORFs also show temperature dependent changes in TEs. At 20 °C, the TE of 789 uORF1 is decreased ~1.7-fold relative to 30 °C, while the TE of uORF3 is increased by ~1.5-fold.

790
At 37 °C, the TE of uORF2 is decreased ~3-fold relative to 30 °C, while the TEs of uORFs 3 and 791 4 are increased ~1.7-and 2.7-fold, respectively. The differing behavior of each GCN4 uORF with 792 respect to temperature seems to underscore the conclusion that no single variable is solely 793 responsible for the observed effects of growth temperature on uORF translation.

803
These data suggest that CPA1 translation may be regulated by temperature-dependent changes 804 in uORF translation, either through a direct effect of temperature on uORF TE or indirectly, for 805 example due to temperature-dependent changes in arginine levels in the cell.

33
A number of other mRNAs that were not previously known to be subject to uORF-mediated 807 translational control appear in our data as having reciprocal, temperature-dependent changes in  Figure 9D and Additional file 1: Figure S10D, compare black to blue).

816
The AGA1 and AGA2 mRNAs, which encode the subunits of the a-agglutinin receptor, are

863
Of the 1367 translated uORFs in our data set, we found ~10% exhibited changes 864 (activation or repression) in their translation at 20 and/or 37 °C relative to 30 °C that met our dual 865 criteria for temperature-regulated uORF translation ( Figure 4A Figure 3D) and reside on shorter, less structured 5'-UTRs (Additional file 1: Figure S6), has a 873 temperature dependence such that its rate increases with growth temperature. NCC uORFs, 874 which tend to be on longer, more structured 5'-UTRs (Additional file 1: Figure S6

36
The position of the uORF in the 5'-UTR exerts a significant influence on the direction and 883 magnitude of the temperature dependence of translation of uORFs. Translation of uORFs that 884 are closer to the 5'-cap than the average distance for all uORFs tends to be inhibited at 20 ˚C 885 relative to 30 ˚C, whereas translation of uORFs that are farther from the cap than the average 886 tends to be activated at 20 ˚C (Figure 6 and 7D). In addition, for NCC uORFs, the distance from 887 the mORF AUG codon also correlates with the temperature dependence of translation such that 888 those farther from the mORF AUG are more likely to be activated at 37 ˚C (Figure 6 and 7E).

889
One simple explanation for some of the observed effects could be that low temperature 890 stabilizes structure in 5'-UTRs, which is generally inhibitory towards uORF initiation, whereas 891 higher temperature tends to destabilize overall 5'-UTR structure. Such an effect could explain 892 uORFs whose translation decreases at 20 ˚C and increases at 37 ˚C, including the majority of

913
Although our data indicate that translation of a significant number of uORFs is regulated 914 by growth temperature and suggest some cases in which these effects influence expression of 915 the main ORF in the mRNA, it is also striking that this is not a general effect and that translation 916 of most uORFs is relatively insensitive to changes in temperature, at least between 20 and 37 ˚C.

930
Those uORFs for which we observe significant changes in TE are ones in which translation is 931 increasing or decreasing more than the average change in the experiment. Nonetheless, our data 932 indicate that translation of most uORFs behaves the same with respect to temperature, which 933 implies a general mechanism to prevent relative translation rates of different ORFs from diverging 934 38 when the growth temperature shifts and thereby changing global proteomic ratios in suboptimal 935 ways.

936
More studies will be required to understand the mechanistic basis of the temperature-         Gerashchenko MV, Gladyshev VN:

NS−uORF−non−NTE and TE_mORF at 30C
TE_mORF at 30C (log2) A l l X 1 s t . b a s e . c h a n g e X 2 n d . b a s e . c h a n g e X 3 r d . b a s e . c h a n g e

NS−uORF−non−NTE and TE_uORF at 30C
TE_uORF at 30C (log2) A l l X 1 s t . b a s e . c h a n g e X 2 n d . b a s e . c h a n g e X 3 r d . b a s e . c h a n g e  A ll X 1 s t _ b a s e _ c h a n g e X 2 n d _ b a s e _ c h a n g e X 3 r d _ b a s e _ c h a n g e u A U G

Supplementary Methods
The 5'-UTRs of mRNAs with uORFs tend to be longer and more structured than the genomic average.
To determine whether the general characteristics of the 5'-UTRs of mRNAs with uORFs are different than those of mRNAs that don't have evidence of translated uORFs, we examined their lengths and propensities to form secondary structures. We interrogated the dataset of ~2700 yeast mRNA 5'-UTR lengths and propensities of forming secondary structures [1]. As described previously, we identified 1367 uORFs (NCC and AUG, Figure 3A ) which were present on mRNAs (uORF mRNAs) with an average 5'-UTR length of 195 nt, which was significantly greater than the average 5'-UTR length of ~79 nt calculated for all mRNAs (All mRNAs) in the PARS dataset (Additional file 1: Figure S6A). We also examined the mRNAs that do not show evidence of an actively translated uORF (Non-uORF mRNAs) as identified by our pipeline discussed in Figure   3A. These 2157 non-uORF mRNAs had an average 5'-UTR length of 66 nt, significantly shorter than the average for All mRNAs. Figure S6A (Additional file 1) shows the cumulative fraction distribution of these three sets of mRNAs.
To examine 5'-UTR secondary structure, we analyzed the PARS (Parallel Analysis of RNA Structure) data available from the same study [1]. In this study, each nucleotide in ~ 3000 mRNAs was assigned a PARS score, which is a measure of its propensity to be in the double stranded conformation. The PARS score is based on the susceptibility of each nucleotide in the transcribed mRNAs to in vitro digestion with RNase V1 (specific to double-stranded mRNA) and RNase S1 (specific to single-stranded mRNA). Higher PARS score indicates a greater tendency of a nucleotide to be in a double-stranded conformation. We analyzed the PARS scores for each of the sets of mRNAs: All mRNAs, uORF mRNAs and non-uORF mRNAs. We considered several PARS features, each of which is an indication of an extent to which a region of the 5'-UTR or coding region near the 5'-UTR can form a secondary structure (Additional file 1: Figure S6B): sum of PARS scores of all the nucleotides (nt) present in the 5'-UTR (Total); average PARS score per nt (Average); sum of PARS scores for the first 30 nts (First30); sum of PARS scores for 30 nts surrounding the start codon (for mRNAs with a 5'-UTR of ≥15 nt; Start30); and highest total PARS score measured for a 30-nt region anywhere across the 5'-UTR (Max30). We also analyzed PARS scores downstream from the mAUG (start codon of the main ORF), with an interval of +1 to +30 (Plus15).
Interestingly, the uORF mRNAs were shown to have higher PARS scores than All mRNAs for most of the PARS features considered (Additional file 1: Figure S6C, compare red versus purple columns). Notably, the Non-uORF mRNAs showed significantly lower PARS scores for some of the PARS features compared to All mRNAs (Additional file 1: Figure S6C, compare red versus blue columns), suggesting that these non-uORF mRNAs are less structured than the genomic average. Together, these data indicate that mRNAs containing actively-translated uORFs typically have longer and more structured 5'-UTRs than those without uORFs. (A) BY4741 cells harboring F-Luc reporter (UUG) were cultured at either 20 °C, 30 °C or 37 °C to an A600 of 0.6 to 0.8 and whole cell extracts were subjected to Western blot analysis using antibodies against eIF1 and Ded1 (loading control). Extracts were loaded in two amounts differing by a factor of two. eIF1 western blot signal was normalized to that of Ded1, to calculate relative SUI1 levels, which was then set as 1 for 30 °C and the relative SUI1 expression at 20 °C or 37      A) Translated uORFs shown to be conserved previously [2] were used here for scatterplot analysis between changes in TEs at 20 °C (∆TEConserved-uORF 20 °C) and 37 °C (∆TEConserved-uORF 37 °C). The plot is divided into four quadrants (Q1 to Q4) based on the directionality of TE changes.

NS−nonNTE & Context score
Context score (nt, log2) binned with respect to their start sites. All represents all the translated uORFs identified in this study as described in Figure  A) Boxplot analysis of uORF start codon context scores for the sets of regulated uORFs described in Figure 7A. All NCCs represent all translated NCC uORFs whose context scores were available (N = 1206). The dotted horizontal line represents the median context score for All NCCs. For TE_down or TE_up (20 °C or 37 °C) sets, at least 98/100 uORFs' context scores were available.