To examine whether eIF2A provides an eIF2-independent initiation mechanism for any yeast mRNAs, we first examined bulk polysome formation in a yeast mutant lacking the gene YGR054W encoding eIF2A (denoted eIF2AΔ below) and an isogenic wild-type strain (WT), both grown in nutrient-replete medium (SC) or under conditions of isoleucine/valine starvation, imposed by the drug sulfometuron methyl (SM), to induce eIF2α phosphorylation by Gcn2 and thereby reduce TC levels. We reasoned that if eIF2A can substitute for eIF2 to maintain translation initiation of a sizeable fraction of mRNAs when eIF2 function is reduced, then we might observe a depletion of polysomes in the SM-treated eIF2AΔ mutant compared to SM-treated WT cells. As we reported previously (Gaikwad et al., 2021), SM treatment of WT evoked a small reduction in the ratio of polysomes to monosomes (P/M), indicating a diminished rate of bulk translation initiation (Figure 1A(i) and (iii)). Essentially identical P/M ratios were observed in the corresponding untreated and SM-treated eIF2AΔ mutant (Figure 1A(ii) and (iv)), suggesting that eIF2A does not function broadly to compensate for reduced eIF2 function in the yeast translatome. These findings are consistent with a previous analysis in which eIF2A was depleted by transcriptional shut-off of an eIF2A allele expressed from the GAL1 promoter (Zoll et al., 2002).

We next asked whether eIF2A can provide an eIF2-independent initiation mechanism for any particular mRNAs in yeast by conducting ribosome profiling of the same eIF2AΔ mutant and WT strains under the conditions of SM treatment employed above. Ribosome profiling entails deep-sequencing of 80S ribosome-protected mRNA fragments (RPFs, or ribosome footprints) in parallel with total RNA. The ratio of sequencing reads of RPFs summed over the CDS to the total mRNA reads for the corresponding transcript provides a measure of translational efficiency (TE) for each mRNA (Ingolia et al., 2009). Owing to normalization for total read number in each library, the RPF and mRNA reads and the calculated TEs are determined relative to the average values for each strain. The RPF and RNA read counts between biological replicates for each strain and condition were highly reproducible (Pearson’s r ≈ 0.99) (Figure 1—figure supplement 1A–H).

To establish that SM treatment induced comparable levels of eIF2α phosphorylation in both the WT and eIF2AΔ mutant, we examined the ribosome profiling data for GCN4 mRNA, whose translation is induced by phosphorylated eIF2α by the specialized ‘delayed reinitiation’ mechanism imposed by the four uORFs in its transcript. Translation of the 5′-proximal uORFs (uORF1 and uORF2) gives rise to 40S subunits that escape recycling at these uORF stop codons and resume scanning downstream. At high TC levels, they quickly rebind TC and reinitiate at uORFs 3 or 4 and are efficiently recycled from the mRNA following their translation. When TC levels are reduced by eIF2α phosphorylation, a fraction of scanning 40S subunits fails to rebind TC until after bypassing uORFs 3–4, and then rebind the TC in time to reinitiate at the GCN4 CDS, inducing GCN4 translation (Figure 1B; reviewed in Hinnebusch, 2005; Gunišová et al., 2018).

Ribosome profiling revealed the expected strong induction of GCN4 translation evoked by SM treatment of WT cells, as revealed by greatly increased RPFs in the CDS with little or no change in GCN4 mRNA reads, yielding an increase in TE (ΔTE) of ~40-fold (Figure 1C, WT+SM versus WT, cf. replicate cultures _a and _b for RPFs and RNA reads). Highly similar results were observed in the eIF2AΔ mutant in the presence and absence of SM (Figure 1C, eIF2AΔ+SM versus eIF2AΔ), indicating that eIF2α phosphorylation was induced by SM at comparable levels in both the mutant and WT strains. These data argue against the possibility that eIF2A contributes to recruitment of Met-tRNA_i as an alternative to the TC by 40S subunits scanning downstream from GCN4 uORFs 1–2. If this was the case, then the eIF2AΔ mutation would impose a further delay in rebinding Met-tRNA_i to the 40S subunits re-scanning downstream from uORFs 1–2 and thereby increase the proportion that bypass uORFs 3–4 when TCs are limiting, derepressing GCN4 translation in SM-treated eIF2AΔ versus WT cells. The highly similar induction of GCN4 translation in the two strains (Figure 1C) argues against such a role for eIF2A in recruiting Met-tRNAi to 40S subunits scanning the GCN4 mRNA leader. These findings are in agreement with measurements of GCN4-lacZ reporter expression in SM-treated eIF2AΔ versus WT cells; although the induction by SM was quite modest in that study, and an auxiliary role for eIF2A might have gone undetected (Zoll et al., 2002).

We turned next to the question of whether eliminating eIF2A alters the translation of any other yeast mRNAs, reasoning that mRNAs able to utilize eIF2A in place of TC for recruitment of Met-tRNA_i would exhibit greater TE reductions evoked by SM in eIF2AΔ mutant versus WT cells. If such mRNAs could be translated efficiently utilizing TC alone, then they would require eIF2A conditionally, that is, only when eIF2 function is reduced during starvation. If instead such mRNAs rely primarily on eIF2A for efficient initiation regardless of TC levels, they would exhibit reduced TEs in eIF2AΔ versus WT cells in the absence of SM treatment. Importantly, DESeq2 analysis of the ribosome profiling data obtained for the eIF2AΔ mutant and WT strains cultured in the absence of SM revealed no significant TE reductions in the mutant, as no mRNAs exhibited TE_eIF2AΔ/TE_WT ratios < 1 at a false discovery rate (FDR < 0.25) that is appropriate for two highly correlated biological replicates (Lamarre et al., 2018; Figure 2A). This finding suggests that few, if any, mRNAs are appreciably dependent on eIF2A for translation initiation in nutrient-replete cells where TC is abundant.

In addition to induction of GCN4 translation, SM treatment of WT cells leads to a broad reprogramming of translational efficiencies as DESeq2 analysis revealed hundreds of mRNAs exhibiting increases or decreases in relative TE even at the highly stringent FDR of < 0.01 (Figure 2B). Previously, we presented evidence that many of these TE changes conform to a pattern in which mRNAs that are efficiently translated in untreated WT cells tend to exhibit increased relative TEs, whereas poorly translated mRNAs tend to show reduced relative TEs, when phosphorylation of eIF2α is induced by SM. This same pattern was evident under two other conditions in which 43S PIC assembly is reduced, impaired recycling of 40S subunits from post-termination complexes at stop codons and depletion of an essential 40S ribosomal protein. We proposed that this stereotypical reprogramming of translation arises from increased competition among mRNAs for limiting PICs wherein strongly translated mRNAs outcompete weakly translated mRNAs to skew TE increases towards ‘strong’ mRNAs (Gaikwad et al., 2021). SM treatment of the eIF2AΔ mutant also produced a broad reprogramming of TEs involving hundreds of mRNAs translated relatively better or worse on SM treatment (Figure 2—figure supplement 1A). Comparing the TE changes conferred by SM in WT versus eIF2AΔ cells for the mRNAs showing TE changes in WT revealed that the majority of transcripts showed TE changes in the same direction on SM treatment of WT and eIF2AΔ cells (Figure 2C). Indeed, the majority of mRNAs dysregulated in the eIF2AΔ mutant also showed TE changes in the WT strain (Figure 2C, Figure 2—figure supplement 1B) and a very strong positive correlation exists, with a coefficient of 0.91, between the TE changes conferred by SM in the two strains. Thus, elimination of eIF2A did not substantially alter the global reprogramming of translation produced by phosphorylation of eIF2α.

If certain mRNAs depend more heavily on eIF2A for Met-tRNAi recruitment when eIF2α is phosphorylated, we might expect to observe additive reductions in TE when combining elimination of eIF2A by the eIF2AΔ mutation with inhibition of eIF2 by SM treatment. Interrogating the large group of 879 mRNAs that showed TE reductions on SM treatment of WT cells revealed that they show a relatively smaller, not larger, decrease in median TE on SM treatment of the eIF2AΔ mutant versus SM treatment of WT (Figure 2D, column 12 versus column 9). Consistent with this finding, most of these mRNAs exhibit a small increase in TE on comparing SM-treated eIF2AΔ to SM-treated WT cells (Figure 2D, column 11). (In these and all other box plots, when the notches of different boxes do not overlap, their median values are judged to differ significantly with a 95% confidence level. As shown in columns 1–4, the median TE change for all ~5500 expressed mRNAs detected in our profiling experiments is close to unity (log₂ = 0) for all of the comparisons examined in Figure 2D). These results suggest that the majority of mRNAs whose translation is diminished by eIF2α phosphorylation in WT cells do not utilize eIF2A for a back-up initiation mechanism that would mitigate their TE reductions when eIF2 is impaired. Our finding that most mRNAs exhibit somewhat greater TEs in eIF2AΔ versus WT cells when both are treated with SM (Figure 2D, column 11) might indicate that eIF2A generally acts to repress the translation of these mRNAs rather than augmenting eIF2 function in Met-tRNAi recruitment. This would not be the case in the absence of SM, however, as the results in column 10 of Figure 2D suggest a small positive effect of eIF2A on translation of these mRNAs in non-starved cells.

To determine whether there are any individual mRNAs that show greater TE reductions in response to SM treatment when eIF2A is absent, we conducted DESeq2 analysis of the TE changes in SM-treated eIF2AΔ versus SM-treated WT cells. A group of only 32 mRNAs showed significant TE reductions in this comparison, that is, TE_eIF2AΔ+SM/TE_WT+SM < 1, FDR < 0.25 (Figure 3A). The TE reductions for this group of transcripts (designated ΔTE_{eIF2AΔ+SM/WT+SM_down}) in comparison to all mRNAs were nearly two-fold greater in SM-treated eIF2AΔ versus SM-treated WT cells (Figure 3B, column 1), which are the results expected if they utilize eIF2A as a back-up when eIF2 is impaired by phosphorylation. They also showed TE reductions in SM-treated versus untreated eIF2AΔ cells, albeit of lesser magnitude (Figure 3B, column 2), also in the manner expected if eIF2 and eIF2A play redundant roles in their translation. Functional redundancy is further supported by the findings that both SM treatment of WT cells and elimination of eIF2A from untreated cells does not reduce their median TEs (Figure 3B, columns 3–4), as only one of the two factors is impaired or eliminated in these latter comparisons. The fact that the median TE of these mRNAs increases rather than decreases on SM treatment of WT (Figure 3B, column 3) might be explained by proposing that their ability to rely on eIF2A provides them with a competitive advantage with mRNAs that depend solely on eIF2 when the latter is impaired by phosphorylation.

To determine how many of these 32 transcripts exhibit TE reductions exclusively when both eIF2 and eIF2A are impaired/eliminated, we conducted hierarchical clustering of the TE changes in the four comparisons described above, displaying the magnitude of changes with a heat map. Only 17 of the 32 transcripts (marked with ‘#’) displayed the diagnostic pattern of an appreciable reduction in TE both on elimination of eIF2A in SM-treated cells and on SM treatment of cells lacking eIF2A (red or pink hues in columns 1–2 of Figure 3C), but either a lesser reduction, no change, or increase in TE on SM treatment of WT cells and on elimination of eIF2A from untreated cells (light pink, white or blue hues in columns 3–4 of Figure 3C).

In an effort to provide independent evidence that a subset of the transcripts analyzed in Figure 3C are dependent on eIF2A only when eIF2 function is diminished by SM, we constructed luciferase reporters for 8 of the aforementioned 17 genes (CHS5, DBF4, DNA2, HKR1, MET4, PBP4, PUS7, RAD9) and for 4 other genes that satisfied only one of the two criteria for conditional dependence on eIF2A when eIF2 is impaired by phosphorylation (SAG1, SVL3, NET1, and NSI1). The promoters and 5′ UTR sequences of the candidate genes were fused to the firefly luciferase CDS and the 3′ UTR and transcription termination sequences of the yeast RPL41A mRNA and introduced into yeast on single-copy plasmids (Sen et al., 2015). Assaying these LUC reporters in both WT and eIF2AΔ cells in the presence and absence of SM showed that only the HKR1 reporter displayed a significantly greater reduction in expression in response to SM treatment in eIF2AΔ versus WT cells (Figure 3D). Hence, except for HKR1 mRNA, the LUC reporter analysis failed to support a conditional requirement for eIF2A for the candidate genes examined.

As an independent approach, we determined the distributions of native candidate mRNAs that co-sedimented with 80S monosomes or different polysomal species in cell extracts, examining HKR1 and two other mRNAs (CHS5 and DBF4) that satisfied both of the aforementioned criteria for conditional dependence on eIF2A, three mRNAs (NET1, SAG1, SVL3) that satisfied only one of the two criteria, and ACT1 and TRM44 mRNAs examined as negative controls. The amounts of each mRNA found in monosome or polysomal fractions were multiplied by the number of ribosomes per mRNA in that fraction (one for monosomes, two for disomes, three for trisomes, etc.) to calculate the abundance of ribosomes translating the mRNA and normalized to the input level of the mRNA in the unfractionated extracts to calculate the TE of the transcript in each condition. Analyzing the results from three biological replicates for WT versus eIF2AΔ cells ± SM revealed that only three transcripts, HKR1, SAG1, and SVL3, displayed a greater TE reduction in response to SM treatment of eIF2AΔ cells compared to SM treatment of WT (Figure 3, Figure 3—figure supplement 1). Thus, only the HKR1 transcript satisfied all of the criteria for conditional eIF2A dependence in ribosome profiling data in a manner confirmed by both reporter analysis and polysome profiling. We conclude that there are very few mRNAs, possibly only one, that utilize eIF2A as a back-up for eIF2 in recruitment of tRNAi when eIF2 is impaired by phosphorylation.

We next interrogated our ribosome profiling data for the effects of deleting eIF2A on translation of URE2 mRNA, reported to contain an IRES that is inhibited by eIF2A (Komar et al., 2005). We found no significant change in TE or ribosome occupancy for URE2 mRNA on deletion of eIF2A in either non-starved or SM-starved cells (Figure 4A). It was reported that yeast mRNAs GIC1 and PAB1 also contain IRESs and are subject to translational repression by eIF2A (Reineke and Merrick, 2009); however, we again found no significant alteration in their TEs between eIF2AΔ and WT cells in the presence or absence of SM treatment (Figure 4B and C). Our results do not support a role for eIF2A in repressing IRES function in yeast cells. It should be noted, however, that eliminating eIF2A would not be expected to confer a marked increase in TE if the IRESs make only a small contribution to overall translation of these mRNAs under the conditions of our experiments.

It has been reported that mammalian eIF2A is required for the translation of certain non-AUG initiated uORFs and that translation of these uORFs upregulates translation of the downstream CDS when eIF2α is phosphorylated during ER stress (Starck et al., 2016) or during tumor initiation (Sendoel et al., 2017). Hence, we examined our profiling data more closely to determine if eIF2A might affect the translation of mRNAs harboring uORFs in yeast cells. We did not restrict our attention to stimulatory uORFs, which are rare in yeast (May et al., 2023), but included all uORFs in our analyses.

In addition to GCN4 described above, CPA1 mRNA, encoding an arginine biosynthetic enzyme, contains a single AUG-initiated inhibitory uORF that attenuates translation of the main CDS in cells replete with arginine (Werner et al., 1987; Delbecq et al., 1994). In accordance with previous findings that the inhibitory effects of single uORFs can be diminished by eIF2α phosphorylation by allowing scanning ribosomes to bypass the uORF start codon (Young and Wek, 2016; Dever et al., 2023), we observed an approximately two-fold increase in the TE of CPA1 mRNA on SM treatment of WT cells (Figure 5A). However, this TE change was not significantly altered by the eIF2AΔ mutation. Similarly, for the well-characterized YAP2/CAD1 mRNA, containing two inhibitory AUG uORFs (Vilela et al., 1998), we observed a 2.6-fold TE increase conferred by SM in WT cells that, again, was not significantly altered by deletion of eIF2A (Figure 5B). The eIF2AΔ deletion also had no effect on the TEs of CPA1 and YAP2/CAD1 in non-starved cells. Thus, although we found evidence that translational repression exerted by the uORFs in CPA1 and YAP2/CAD1 mRNAs is mitigated by eIF2α phosphorylation, it appears that eIF2A has little or no role in recognizing the AUG codons of these uORFs in the presence or absence of diminished TC levels, just as we concluded above for the GCN4 uORFs.

Looking more broadly, we interrogated a previously identified group of 1306 mRNAs containing 2720 uORFs, initiating with either AUG or one of the nine NCCs, that showed evidence of translation in multiple ribosome profiling datasets from various mutant and WT strains (Zhou et al., 2020). We also analyzed a second set of 791 mRNAs containing 982 AUG- or NCC-initiated uORFs that are both evolutionarily conserved and show evidence of translation in ribosome profiling experiments (Spealman et al., 2018). These four groups of mRNAs with translated AUG- or NCC-uORFs showed little or no TE change for their CDSs on SM treatment of WT cells (Figure 6A(i)–(ii), columns 4 and 7 versus 1), indicating that, in contrast to GCN4, CPA1, and YAP2/CAD1 mRNAs, most of them do not contain inhibitory uORFs that can be bypassed in response to eIF2α phosphorylation. Both groups of mRNAs containing either AUG- or NCC-initiated uORFs exhibit very small, albeit highly significant, reductions in CDS TEs in the eIF2AΔ mutant versus WT cells in the absence of SM (Figure 6A(i)–(ii), columns 5 and 8 versus 2); however, the same was not observed for the effects of eIF2AΔ in SM-treated cells where eIF2 function is attenuated (Figure 6(i)–(ii), columns 6 and 9 versus 3).

Evidence of translation of a uORF does not guarantee that it has an appreciable impact on the proportion of scanning ribosomes that reach the downstream CDS. Recently, groups of 557 mRNAs containing AUG-initiated uORFs and 191 mRNAs with NCC-uORFs were identified in which the uORFs were shown to influence translation of downstream CDSs by massively parallel analysis of reporters containing the native 5′ UTRs in comparison to mutant reporters lacking the uORF start codons (May et al., 2023). It was reported that among the 407 mRNAs containing a single functional AUG-uORF wherein mutating the uORF significantly altered reporter expression by >1.5-fold, all but 13 of the uORFs functioned to inhibit reporter expression. Among the 144 mRNAs containing a single functional NCC-uORF, only 13 of the uORFs influenced reporter expression either positively or negatively by >1.5-fold, reflecting the weaker effects of NCC- versus AUG-initiated uORFs on downstream translation (May et al., 2023). Our analysis of the 394 mRNAs containing a single inhibitory AUG-uORF—the only group large enough for statistical analysis of TE changes—revealed little or no change in median TE in response to SM treatment or to the eIF2AΔ mutation in the presence or absence of SM (Figure 6A(iii), columns, 4–6 versus 1–3). These findings suggest that eIF2A has a minimal role in determining whether functional inhibitory AUG-initiated uORFs are translated or bypassed by scanning ribosomes in the presence or absence of eIF2α phosphorylation. These results support our findings above that the majority of mRNAs containing translated uORFs show little response to either eIF2α phosphorylation, the elimination of eIF2A, or a combination of both perturbations.

We probed more deeply into the three sets of uORF-containing mRNAs mentioned above by examining the subsets of each group that exhibit >1.41-fold increases in TE in response to SM treatment of WT cells, making them novel candidates for mRNAs controlled by inhibitory uORFs that can be overcome by eIF2α phosphorylation. These uORF-containing mRNAs exhibit median TE increases of approximately two-fold on SM treatment of WT cells (Figure 6B(i)–(iii), maroon data), similar to the findings above for CPA1 and CAD1 mRNAs (Figure 5A and B). Unlike the larger groups of uORF-containing mRNAs from which they derive (analyzed in Figure 6A(i)–(iii)), these subsets showing appreciable TE induction by SM exhibit slightly reduced median TEs in response to the eIF2AΔ mutation in SM-treated, but not untreated cells (Figure 6B(i)–(iii), green versus orange data). One way to explain these findings would be to hypothesize that these mRNAs generally contain a positive-acting uORF that functions to overcome the translational barrier imposed by a second inhibitory uORF located further downstream in response to eIF2α phosphorylation, in the manner established for the positive-acting uORF1 in GCN4 mRNA, accounting for their increased TE conferred by SM treatment of WT cells. The TE reductions conferred by eIF2AΔ exclusively in SM-treated cells would then arise from loss of eIF2A-dependent translation of the stimulatory upstream uORFs in these transcripts required for scanning ribosomes to bypass the downstream inhibitory uORFs when eIF2α is phosphorylated.

To explore this hypothesis further, we focused on the uORF-containing mRNAs that are the most dependent on eIF2A for their translation in SM-treated cells, showing a >1.41-fold reduction in TE in SM-treated eIF2AΔ versus SM-treated WT cells. Examining all such mRNAs containing either (i) annotated or conserved AUG-uORFs (Figure 6C(i)), (ii) annotated or conserved NCC-uORFs (Figure 6(ii)), or (iii) functional inhibitory AUG-uORFs (Figure 6C(iii)) revealed that all three groups show TE reductions in response to the eIF2AΔ mutation only in the presence of SM (green versus orange data) and, except for the NCC-uORF mRNAs, also show TE increases in response to SM in WT cells (Figure 6C(i)–(iii), maroon data), generally supporting the aforementioned hypothesis. This is the same pattern we observed for all of the uORF-containing mRNAs showing appreciable TE increases on SM treatment of WT cells just described (Figure 6B(i)–(iii)). Moreover, this pattern of TE changes was observed above for the group of 32 mRNAs showing highly significant TE reductions conferred by eIF2AΔ in SM-treated cells versus SM treatment of WT (Figure 3B); although, as discussed above, only 17 of those transcripts conform to this pattern of TE changes (blue in column 4 but nearly white in column 3 of Figure 3B; dubbed ∆TE_{eIF2AΔ+SM/WT+SM_down*}). Of these 17 mRNAs, only 3 (YFL027C, YKL082C, and YLR330W/CHS5) contain a functional AUG- or NCC-initiated uORF (Figure 6D), which does not represent a statistically significant enrichment for such mRNAs (p = 0.21). Moreover, our LUC reporter and polysome profiling analyses of one of these three transcripts, YLR330W/CHS5, failed to confirm a conditional dependence for eIF2A (Figure 3D and E). The one mRNA for which conditional eIF2A dependence was confirmed, YDR420W/HKR1, does not contain a functional or conserved uORF nor even one merely showing evidence of translation in ribosome profiling experiments. Thus, we could not find compelling evidence that conditional dependence on eIF2A for efficient translation in SM-treated cells is conferred by positive-acting regulatory uORFs.

As an orthogonal approach to detecting a role for eIF2A in regulating the translation of negative-acting uORFs, we reasoned that decreased translation of inhibitory uORFs in eIF2AΔ cells would increase the translation of downstream CDSs, leading to a decrease in the ratio of RPFs in uORFs relative to RPFs in the CDSs, which we termed relative ribosome occupancy (RRO). To examine this possibility, we employed DESeq2 to identify statistically significant changes in RRO values for the same groups of mRNAs analyzed above in Figure 6A and B containing annotated or conserved uORFs with either AUG- or NCC-start sites, and which also showed evidence of translation in both WT and eIF2AΔ cells. We found that no mRNAs exhibited significant changes in RRO in the eIF2AΔ mutant versus WT in the absence or presence SM, even at a relatively non-stringent FDR of 0.5 (Figure 6—figure supplement 1A–D). Furthermore, the groups of mRNAs with annotated or conserved AUG- or NCC-initiated uORFs showed little or no difference in median RRO in the eIF2AΔ mutant versus WT cells in the absence or presence of SM (Figure 6—figure supplement 1E–H, yellow versus blue data). At least for the annotated uORFs, the median RRO values are generally higher for mRNAs containing AUG-initiated versus NCC-initiated uORFs in the presence or absence of SM (Figure 6—figure supplement 1E and G, blue data), as would be expected from more efficient initiation at AUG versus NCC start codons. Thus, our results provide no compelling evidence that eIF2A frequently enhances initiation at negative-acting uORFs initiated by either AUG or NCC start codons in either starved or non-starved cells.

Translation elongation factor eIF5A acts broadly to stimulate the rates of decoding and termination, but is particularly important for certain combinations of codons, which include stretches of proline codons and various three-codon combinations of Proline (Pro), aspartic acid (Asp), and glyine (Gly) codons (Saini et al., 2009; Gutierrez et al., 2013; Schuller et al., 2017). Slower decoding rates for these codon combinations were detected by ribosome profiling of yeast cells depleted of eIF5A by computing pause scores for all ~8000 three-amino acid motifs (Schuller et al., 2017). The pause score for each motif is calculated from its 80S occupancy relative to the surrounding stretch of CDS and averaged across the translatome. Analyzing our profiling data revealed no significant increase in pause scores for any tripeptide motif in the eIF2AΔ mutant versus WT cells (Figure 7A). As a positive control for the analysis, we examined our data on WT cells treated or untreated with SM, observing a marked increase in pause scores for all tripeptide motifs with valine (Val) in the third position, placing the corresponding Val codons in the A-site of the decoding ribosomes (Figure 7B). Increased pausing at these motifs is expected from reduced levels of charged valyl-tRNA on inhibition of valine biosynthesis by SM and attendant slower decoding of all Val codons. It is unclear why increased pausing is not found at codons for isoleucine as SM inhibits an enzyme common to the Ile and Val biosynthetic pathways (Jia et al., 2000); however, we note that flux through these pathways is differentially regulated and that the Val and leucine pathways compete for a common precursor (Jones and Fink, 1982). Taken together, our findings indicate that eIF2A plays no major role in stimulating translation elongation through particular tripeptide motifs.

As noted above, a synthetic growth phenotype was observed on combining the eIF2AΔ mutation with either a temperature-sensitive mutation affecting eIF4E (Komar et al., 2005) or deletion of the FUN12 gene encoding eIF5B (Zoll et al., 2002), suggesting functional interactions between eIF2A and these initiation factors. Other genetic interactions involving the eIF2AΔ mutation were uncovered in a global analysis of synthetic phenotypes produced by phenotyping double mutants that combine deletions or point mutations in ~90% of all yeast genes (Costanzo et al., 2016; van Leeuwen et al., 2016), compiled at https://thecellmap.org/costanzo2016/. The profile of genetic interactions observed for the eIF2AΔ mutation was found to significantly resemble that of an ade1Δ mutation, including synthetic growth defects in eIF2AΔ double mutants containing deletions of genes HPT1 and FCY2, whose products function in purine base and cytosine uptake (Fcy2) or in synthesizing purine nucleotides from purine bases via the salvage pathway (Hpt1) (Ljungdahl and Daignan-Fornier, 2012). These findings suggested that eIF2A might function as a positive regulator of de novo purine biosynthesis, such that the eIF2AΔ mutation would mimic adeΔ mutations in reducing growth in combination with the hpt1Δ and fcy2Δ mutations as a result of simultaneously impairing both the de novo and salvage pathways for adenine biosynthesis (Ljungdahl and Daignan-Fornier, 2012). To test this possibility, we generated the fcy2ΔeIF2AΔ double mutant and tested it for growth on synthetic minimal (SD) or synthetic complete (SC) medium supplemented with varying concentrations of adenine. Contrary to expectations, we failed to detect any reduction in growth in the double mutant versus either single mutant or the WT strain in the presence or absence of adenine supplements (Figure 6—figure supplement 2A and B). We also interrogated our ribosome profiling data to examine the effects of the eIF2AΔ mutation on expression of the 17 yeast genes encoding proteins involved in de novo synthesis of inosine, adenine, or guanine nucleotides (Ljungdahl and Daignan-Fornier, 2012). The eIF2AΔ mutation conferred no significant reductions in the RPFs, mRNA levels, or TEs of any of 17 genes in non-starvation conditions. Moreover, assaying lacZ reporters for two such genes, IMD2 and IMD3, in cells cultured under the conditions of our ribosome profiling experiments revealed no significant reductions in expression conferred by elimination of eIF2A (Figure 6—figure supplement 2C). Our findings do not support the possibility that eIF2A promotes expression of the biosynthetic genes for de novo purine biosynthesis.

Yeast strains used in this study are listed in Table 1. The primers used for strain construction and verification are listed in Table 3. Strain SGY3 (eIF2AΔfcy2Δ) was generated through a two-step process. First, the kanMX cassette of the eIF2AΔ deletion allele, ygr054wΔ::kanMX4, in strain F2247 was swapped with a hygromycin-resistance cassette to produce SGY1 (ygr054wΔ::hphMX4) by transforming F2247 with a DNA fragment containing the hphMX4 allele that was PCR-amplified from plasmid p4430. Transformants were selected on YPD agar plates supplemented with 300 μg/mL hygromycin B. In the second step, the FCY2 gene was deleted in SGY1 by transformation with a DNA fragment containing the fcy2Δ::kanMX4 allele PCR-amplified from the genomic DNA of strain F2379. Transformants were selected on YPD agar plates containing 200 μg/mL geneticin (G418) to produce SGY3. The ygr054wΔ::hphMX4 and fcy2Δ::kanMX4 alleles in strains SGY1 and SGY3, respectively, were verified by PCR analysis of chromosomal DNA using the appropriate primers listed in Table 3.

All plasmids employed in this study are listed in Tables 2 and 3. Plasmids p6593-p6604 were constructed by LifeSct LLC by synthesizing DNA fragments containing the promoter, 5′ UTR, and first 20 codons of the main CDS of the candidate genes and using them to replace the corresponding fragment of p6029 (Sen et al., 2015), fusing the first 20 codons of the candidate genes to the F.LUC CDS.

Ribosome profiling and RNA-seq analyses were carried out in parallel on strains BY4742 (WT) and F2247 (eIF2AΔ), with two biological replicates for each strain, as described previously (Gaikwad et al., 2021). In brief, both strains were grown at 30°C in SC until reaching log phase at A₆₀₀ = 0.5–0.6 for untreated cells, or to A₆₀₀ = 0.5–0.6 in SC lacking isoleucine and valine (SC-Ile/Val) and treated with SM at 1 µg/mL for 25 min for SM-treated cells. Cells were harvested by high-speed vacuum filtration and snap-frozen in liquid nitrogen. Cell lysis was performed using a freezer mill in the presence of lysis buffer containing 500 μg/mL of cyclohexamide, and RPFs were prepared by digesting cell lysates with RNase I. The 80S monosomes were resolved by sedimentation through a 10–50% sucrose gradient. The RPFs were purified from the monosomes using hot phenol-chloroform extraction. After size selection and dephosphorylation steps, a Universal miRNA cloning linker was attached to the 3' ends of the footprints. This was followed by reverse transcription, circular ligation, rRNA subtraction, PCR amplification of the cDNA library, and DNA sequencing using an Illumina HiSeq system at the NHLBI DNA Sequencing and Genomics Core at NIH (Bethesda, MD).

For RNA-seq library preparation, total RNA was extracted and purified from aliquots of the same snapped-frozen cells lysates described above using hot phenol-chloroform extraction. Then, 5 µg of randomly fragmented total RNA was used for library generation and sequencing, similar to the steps mentioned above except that the Ribo-Zero Gold rRNA Removal Kit (Illumina; MRZ11124C) was employed to remove rRNA after linker-ligation.

Processing and analysis of sequence libraries of RPFs or total mRNA fragments, including Wiggle track normalization to the total number of mapped reads for viewing RPF or RNA reads in the IGV browser, were conducted exactly as described previously (Gaikwad et al., 2021). In brief, sequencing reads were trimmed and noncoding RNA sequences were eliminated by aligning trimmed FASTA sequences to the Saccharomyces cerevisiae ribosomal database using Bowtie (Langmead et al., 2009). The remaining reads were mapped to the yeast genome using TopHat (Trapnell et al., 2009), and DESeq2 was used for statistical analysis of mRNA reads, RPFs, and TEs (Love et al., 2014). The R script employed for DESeq2 analysis of TE changes can be found on GitHub (https://github.com/hzhanghenry/RiboProR; Zhang, 2023). Wiggle files were visualized using IGV 2.4.14 at http://software.broadinstitute.org/software/igv/ (Robinson et al., 2011).

RROs for genes containing translated uORFs were calculated by dividing the RPF counts in the uORF by the RPF counts of the main CDS. DESeq2 was employed to conduct statistical analysis on changes in RRO values between the two biological replicates of different genotypes. uORFs with mean RPF counts below 2 or mean CDS RPF counts below 32 in the combined samples (two replicates each) were excluded from the analysis. The compilations of annotated translated uORFs (Martin-Marcos et al., 2017), evolutionarily conserved translated uORFs (Spealman et al., 2018), or functional uORFs whose elimination by start codon mutation conferred a consistent and statistically significant alteration in translation of the downstream GFP CDSs in reporter constructs (May et al., 2023) were all published previously and are provided in Figure 6—source data 5–7.

Tripeptide pause scores were computed as described previously (Meydan et al., 2023). Briefly, this involved dividing the 80S reads per million (rpm) of a three-amino acid motif by the average rpm in the surrounding region (± 50 nt around each motif). Sites smaller than the ± 50 nt window were excluded from the analysis. To generate average pause scores, we calculated the mean of individual pause scores for each tripeptide motif across the translatome. Motifs represented in the genome less than 100 times were excluded to reduce noise in the analysis.

β-Galactosidase activities were assayed in WCEs using a modified version of a protocol described previously (Moehle and Hinnebusch, 1991) in which the cleavage of ONPG was determined manually after a 30 min incubation with the WCE. Expression of luciferase was assayed in WCEs as previously described (Gaikwad et al., 2021) using the Dual-Luciferase Reporter Assay System (Promega) following the supplier’s protocol, and luciferase activities were normalized to total protein levels in the WCEs determined using the Bradford assay kit (Bio-Rad Laboratories).

To conduct bulk polysome profiling, cells were cultured as described in the figure legends, harvested by centrifugation, and WCEs were prepared by vortexing cell pellets, applying eight cycles of vortexing for 30 s followed by incubation on ice for 30 s, using two volumes of glass beads in ice-cold 1× breaking buffer (20 mM Tris–HCl (pH 7.5), 50 mM KCl, 10 mM MgCl₂, 1 mM DTT, 200 μg/mL heparin, 50 μg/mL cycloheximide, and 1 cOmplete EDTA-free Protease Inhibitor cocktail Tablet (Roche)/50 mL buffer). Fifteen A₂₆₀ units of WCEs were layered on a pre-chilled 10–50% sucrose gradient and centrifuged at 40,000 rpm for 2 hr at 4°C in a SW41Ti rotor (Beckman). Gradient fractions were continuously scanned at A₂₆₀ using the BioComp Gradient Station, and polysome to monosome (P/M) ratios were calculated using ImageJ software for at least two biological replicates.

To determine polysome association of individual mRNAs and calculate their TEs, 20 A₂₆₀ units of WCEs from each of three replicate cultures were sedimented through a 10–50% sucrose gradient by centrifugation at 35,000 rpm for 2.4 hr and the gradient fractions were scanned at A₂₆₀ nm. The fractions containing 80S monosomes or different polysomal species (2-mers, 3-mers, etc.) were collected using ‘the advanced fraction’ function of the BioComp fractionator (Figure 3—figure supplement 1). RNA was extracted from 1/5th of the total volume of input WCEs and from 300 µL of each set of pooled fractions containing 80S or different polysomal species using QIAzol lysis Reagent (QIAGEN) according to the manufacturer’s protocol. Reverse transcription (RT) was done using SuperScript III First-Strand Synthesis SuperMix (Invitrogen) with random hexamers, using 5 μg of RNA for each reverse transcription (RT) reaction. Transcript levels were quantified by qPCR using the Brilliant III Ultra-Fast SYBR Green qPCR Master Mix (Agilent Technologies) on a Roche LightCycler 96 Instrument. Primers listed in Table 3 were used to measure the individual mRNAs and 18S rRNA. The absolute mRNA/18S rRNA levels were calculated using the 2^-CT method and corrected to account for differences in the proportions of the pooled fractions from which RNA was isolated by multiplying by a factor calculated by dividing the total fraction volume (in µL) by 300 µL. They were further corrected to account for differences in the proportion of the total RNA employed for RT by multiplying by a factor calculated as 25 µL (the total volume of extracted mRNA) divided by the volume of the pooled fraction (in µL) containing 5 µg of RNA. To account for potential losses during RNA extraction, RNA recovery normalization factors were calculated for each fraction by determining the proportion of total A₂₆₀ units across all gradient fractions that is present in each set of pooled fraction (calculated from the A₂₆₀ trace obtained during gradient fractionation) and dividing the results by the proportion of total 18S rRNA across the gradient found in the corresponding pooled fraction. A second normalization factor was calculated to correct for losses in recovery of monosomes/polysomes in the gradient separations of fixed amounts of input WCEs by calculating the total A₂₆₀ units found in monosomes/polysomes for each gradient, determining the mean value for all of the gradients/samples analyzed in parallel, and dividing the mean value by the value determined for each gradient. (This correction assumes that WT and eIF2AΔ cells have the same total amounts of monosomes/polysomes per A₂₆₀ of WCE, as indicated by our repeated polysome profiling of biological replicates depicted in Figure 1A). The absolute amounts of mRNAs measured in each pooled fraction were multiplied by both normalization factors to obtain the normalized mRNA levels for each pooled fraction, which was multiplied by the number of ribosomes per mRNA in that pooled fraction (i.e., one for monosomes, two for 2-mer polysomes, etc.) and summed across the gradient fractions to calculate the total number of ribosomes translating the mRNA. To calculate TEs, this last quantity was divided by amount of input mRNA measured in the starting WCE normalized to level of ACT1 mRNA. For each of three biological replicate cultures of each strain, WT or eIF2AΔ mutant, we determined the ratio of TE in the presence versus absence of SM and calculated the mean and SEM values for ∆TEs.

Notched box plots were created using a web-based tool (http://shiny.chemgrid.org/boxplotr/). Scatterplots and volcano plots were generated using the scatterplot function in Microsoft Excel. Hierarchical cluster analysis of TE changes in mutants/conditions was performed using the R heatmap.2 function from the 'gplots' library using the default hclust hierarchical clustering algorithm. Smoothed scatterplots were computed and plotted using the ggplot2 package in R. Calculation of Spearman’s correlation coefficients and Student’s t-tests were performed using built-in features of Microsoft Excel. The Mann–Whitney U test and p-values for Pearson’s correlation were computed using the R Stats package in R.