Transcriptome-wide analysis of roles for tRNA modifications in translational regulation

SUMMARY

Covalent nucleotide modifications in noncoding RNAs affect a plethora of biological processes, and new functions continue to be discovered even for well-known modifying enzymes. To systematically compare the functions of a large set of ncRNA modifications in gene regulation, we carried out ribosome profiling in budding yeast to characterize 57 nonessential genes involved in tRNA modification. Deletion mutants exhibited a range of translational phenotypes, with enzymes known to modify anticodons, or non-tRNA substrates such as rRNA, exhibiting the most dramatic translational perturbations. Our data build on prior reports documenting translational upregulation of the nutrient-responsive transcription factor Gcn4 in response to numerous tRNA perturbations, and identify many additional translationally-regulated mRNAs throughout the yeast genome. Our data also uncover unexpected roles for tRNA modifying enzymes in regulation of TY retroelements, and in rRNA 2′-O-methylation. This dataset should provide a rich resource for discovery of additional links between tRNA modifications and gene regulation.

eTOC

RNA modifications affect a multitude of biological processes. Chou et al comprehensively investigate the functions of one large set of ncRNA modifications, generating ribosome occupancy maps for mutant yeast lacking tRNA-modifying enzymes. This dataset confirms prior studies, identifies new functions for many enzymes, and provides a resource for future analyses.

INTRODUCTION

In addition to the four common nucleotides present at roughly equal abundance in RNA – A, G, C, and U – it has long been known that covalently-modified nucleotides are present at lower abundance. A classic example of such a modified nucleotide is the 7-methylguanylate cap found at the 5′ end of eukaryotic mRNAs. Although decades of investigation have identified scores of modified nucleotides in a variety of coding and noncoding RNAs, new nucleotide modifications continue to be discovered. Moreover, the functions of many nucleotide modifications remain obscure – while the chemical and structural properties of specific modified nucleotides are often well-understood, the detailed functional and regulatory consequences remain unknown for many RNA modification events in vivo.

Nucleotide modifications are particularly common in transfer RNAs (tRNAs), and it is estimated that ~20% of all tRNA nucleotides are covalently modified (Czerwoniec et al., 2009; El Yacoubi et al., 2012; Phizicky and Hopper, 2015). Modified tRNA nucleotides include relatively common species such as 5-methylcytosine (m⁵C) and pseudouridine (Ψ), as well as unusual and complex nucleotides such as the wobble modification 5-methoxycarbonylmethyl-2-thiouridine (mcm⁵s²U), which is generated in a multi-step process by the Elongator complex along with a number of additional factors (Esberg et al., 2006; Huang et al., 2005; Huang et al., 2008). At the molecular level, tRNA modifications have been implicated in a wide variety of processes, including stabilization of tRNA secondary structure (Motorin and Helm, 2010), translation initiation (Liu et al., 2016), decoding (Li et al., 1997; Nedialkova and Leidel, 2015; Zinshteyn and Gilbert, 2013), reading frame maintenance (Lecointe et al., 2002), protection of tRNAs from nuclease cleavage/degradation (Alexandrov et al., 2006; Schaefer et al., 2010), and subcellular trafficking of tRNAs (Kramer and Hopper, 2013). At the organismal level, tRNA-modifying enzymes have been implicated in processes ranging from neurodevelopment to meiotic chromosome pairing to abscisic acid signaling to early development (Phizicky and Hopper, 2010). Interestingly, some tRNA modifications can be regulated in response to environmental conditions, as for example thiolation of specific tRNAs in S. cerevisiae is responsive to the levels of sulfur-containing amino acids cysteine and methionine in the growth media, with altered tRNA thiolation affecting the translation of relevant biosynthetic proteins (Laxman et al., 2013). Although these and many other examples of tRNA modification biology have been uncovered over decades of study, the functions of many tRNA modifications remain mysterious.

A handful of recent studies have carried out transcriptome-wide analysis of translation, using ribosome profiling (Ingolia et al., 2009) to illuminate the roles for specific tRNA-modifying enzymes in translation and proteostasis (Laguesse et al., 2015; Nedialkova and Leidel, 2015; Thiaville et al., 2016; Tuorto et al., 2015; Zinshteyn and Gilbert, 2013). Here, we adopt this approach to systematically study the roles for tRNA modifications in translational regulation in S. cerevisiae, using ribosome profiling to generate ribosome occupancy maps for 57 yeast deletion strains. As expected, deletion of genes encoding enzymes that modify nucleotides in the tRNA anticodon caused the most dramatic translational phenotypes, while loss of enzymes responsible for more distant tRNA modifications often resulted in few discernible translational phenotypes. Codon-level analysis in many cases recovered the expected stalling or slowing of ribosomes at codons corresponding to relevant modified anticodons, as well as identifying codon-level translational perturbations in a number of previously unstudied mutants. Scrutiny of the dramatic translational phenotypes observed in trm7Δ cells revealed a potential role for this gene in methylation of rRNA as well as tRNAs. At the level of individual transcripts, mutant effects on translation of specific genes resulted in a variety of downstream outcomes, both expected (altered transcription of amino acid metabolism genes secondary to altered translation of GCN4 mRNA) and surprising (impaired heterochromatin-mediated gene silencing). Most surprisingly, we find a role for the Elongator complex (and other factors involved in generation of mcm⁵s²U) in maintaining expression of transcripts associated with TY1 retrotransposon Long Terminal Repeats (LTRs). Overall, our data illuminate unanticipated aspects of Trm7 and Elongator function, reveal many additional examples of translational regulation by uORFs, and provide a rich source of hypotheses for future study.

RESULTS

Ribosome footprinting in mutant yeast strains

Budding yeast encode 73 genes currently annotated to play a role in tRNA modification, of which 14 are essential. Here, we set out to characterize translational phenotypes for the remaining 59 nonessential genes. Haploid deletion mutants for 57 genes (two mutants – pcc1Δ and pus6Δ – failed quality control several times) were obtained freshly by sporulation of heterozygous mutant diploids to confirm the viability of the deletion in question and to minimize the potential for suppressor mutations. Initial studies revealed aneuploidies in a subset of the mutants, the majority of which we subsequently re-derived and confirmed to be euploid. However, for two mutants – bud32Δ and gon7Δ, both members of the conserved KEOPS complex (Daugeron et al., 2011; Downey et al., 2006; Kisseleva-Romanova et al., 2006; Srinivasan et al., 2011) – we repeatedly obtained haploid strains bearing an additional copy of ChrIX, suggesting that the KEOPS complex may be essential in our strain background. We included these mutants in the final set of 57 mutants despite this aneuploidy, as many of the dramatic ribosome footprinting phenotypes observed in these two mutants were also observed to a lesser extent in the euploid cgi121Δ mutant (which exhibits partial but incomplete loss of the t⁶A modification (Thiaville et al., 2015)), suggesting that many of the observed phenotypes accurately reflect KEOPS function. Nonetheless, we urge caution in interpreting results obtained from bud32Δ and gon7Δ cells, as observed phenotypes may be secondary to second-site mutations.

Figure 1A shows these 57 genes, grouped according to their modified nucleotide product. This list includes not only the catalytic subunits of tRNA modifying enzymes, but other factors that affect a given tRNA modification, such as the phosphatase Sit4 that is required for Elongator function in vivo. In addition, it is important to note that many of the encoded proteins are also known to affect nucleotide modifications on other RNA species such as mRNAs or rRNA, or play more pleiotropic roles in cell biology – the multimethylase-activating scaffold protein Trm112, for example, is required for methylation of tRNAs, rRNA, and elongation factors (Liger et al., 2011). When appropriate, we will discuss the potential for non-tRNA targets as the relevant mechanistic basis for translational changes observed below.

Figure 1. Overview of dataset.

(A) Nonessential genes involved in tRNA modifications in budding yeast. Encoded proteins are grouped roughly according to function, as for example the Elongator complex is grouped along with other enzymes required for formation of the mcm⁵s²U wobble modification. For each group of enzymes, the known product is shown (R indicates ribose in the tRNA backbone for modified bases), along with a tRNA cartoon showing the best-characterized modification locations. For some sets of mutants, the modification shown represents only a subset of products, as for example Elongator and associated factors also catalyze the formation of mcm⁵U, ncm⁵U, and ncm⁵Um, in addition to mcm⁵s²U as shown. Throughout the manuscript, modifying enzymes are generally color-coded as indicated here, except in cases where subsets of related factors must be distinguished.

(B) Example of RNA-Seq and ribosome footprinting data for chr3:57,000–107,000, showing strong correlations between biological replicate experiments, and also, for the majority of the transcriptome, between mutant strains. See also Supplemental Figures S1 and S2, and Tables S1–S3.

For these 57 deletion strains, we assayed the consequences of loss of tRNA modification on translational control proteome-wide, using ribosome profiling (Ingolia et al., 2009) to provide codon-resolution insight into ribosome occupancy (Figure 1B). Matching mRNA abundance data was gathered for each strain as a reference for the ribosome footprinting dataset to enable calculation of translational efficiency per transcript, and to identify any mutant effects on transcription or mRNA stability. Wild-type and mutant strains were grown to mid-log phase in rich media and processed for ribosome profiling, with biological duplicates for each strain. As the use of the translation inhibitor cycloheximide can affect the distribution of ribosomes across open reading frames (Gerashchenko and Gladyshev, 2014; Hussmann et al., 2015), we also generated replicate datasets for six mutants collected without the use of cycloheximide. Tables S1–S4 provide complete datasets for mRNA, ribosome occupancy, and translational efficiency, as well as the six no cycloheximide replicates. Overall, biological replicates were well-correlated with one another (Figure 1B, Supplemental Figure S1), with pairwise correlations ranging from 0.97 to >0.99. Moreover, the six datasets generated from cultures not subject to cycloheximide treatment exhibited similar pairwise correlations between experiments run with or without cycloheximide (Supplemental Figure S2) as observed for replicate pairs run in the presence of cycloheximide. Ribosome-protected footprints (RPFs) were predominantly 28–32 nt (Table S5), as expected, and exhibited known features such as 3 nt periodicity over coding regions, and absence of ribosomes over introns. In addition, our data recapitulated prior ribosome profiling analysis of Elongator and other mutants involved in the formation of mcm⁵s²U and related modifications (Nedialkova and Leidel, 2015; Zinshteyn and Gilbert, 2013) – see below. This dataset thus provides a high-quality resource for analysis of the roles for tRNA modifying enzymes in translation.

Below, we analyze the dataset at three levels of granularity: averaged across codons, averaged over metagenes, and averaged across individual genes.

Codon-level analysis of ribosome occupancy changes

Changes in tRNA levels or modifications can affect the dwell time of ribosomes on the relevant codon occupying the A site, and this is readily observed as changes in codon-averaged ribosome occupancy. We therefore analyzed global codon occupancy for all 57 mutants, as previously described (Nedialkova and Leidel, 2015) (Figure 2A, Supplemental Figure S3A, Table S6). Our data recapitulate recent studies of mutations affecting mcm⁵s²U formation (henceforth collectively referred to as Elongator-related mutants) that documented increased A site ribosome occupancy over AAA, CAA, and GAA codons, which are decoded using mcm⁵s²U34-containing tRNAs (Nedialkova and Leidel, 2015; Zinshteyn and Gilbert, 2013), providing further confidence in our dataset (Figures 2A–B).

Figure 2. Codon-level analysis of ribosome occupancy.

(A) Effects of all 57 mutations on average ribosome occupancy at A, P, and E sites over all 61 codons (excluding stop codons). Columns depict mutations, with key mutations identified above clusters – Supplemental Figure S3A shows an expanded view with all 57 mutants annotated, and the entire dataset is available as Table S6. Heat maps show log2 fold changes relative to the wild-type average (red=increased codon occupancy; green=decreased codon occupancy). Data for A, P, and E sites are all sorted identically, based on A site dataset clustering. Note that P and E site panels are scaled to 50% of the width of the A site panel. Yellow boxes highlight examples shown in panels (B–F).

(B) A site ribosome occupancy for all 61 codons for Elongator-related mutants, relative to wild-type average. Grey diamonds show average occupancy (zero, by definition) and standard deviation for 9 replicates of BY4741, red diamonds show average and standard deviation for 30 Elongator-related datasets (2 biological replicates for 15 deletion mutants). The expected increase in ribosome occupancy is confirmed over AAA, CAA, and GAA, as indicated.

(C–E) Data are shown as in panel (B), but here data are shown separately for known target codons (left columns, indicated) with all remaining codons then sorted alphabetically. Data for (C–D) and (E) show A and P site occupancy, as indicated.

(F) Discrepant behavior between mutants affecting yW synthesis. A, P, and E site occupancy data shown for Phe codons for the four tyw mutants, as indicated – Tyw1-3 are indistinguishable by design, while the discrepant behavior of Tyw4 is visually emphasized.

For many modifiers with known tRNA substrates, we document altered ribosomal A site occupancy at the relevant codons (Figures 2C–E, Supplemental Figures S3B–C, Table S6), consistent with the tRNA modification in question affecting tRNA stability, charging, or codon recognition. For example, loss of Mod5, which generates N6-isopentenyladenosine (i⁶A) at position 37 in a number of tRNAs (Cys-GCA, Ser-NGA, Tyr-GUA) (Laten et al., 1985), results in dramatically decreased A site ribosome occupancy over all relevant codons (Figure 2C). In the case of Tan1, responsible for generation of N4-acetylcytidine (ac⁴C) at position 12 of leucine and serine tRNAs (Johansson and Bystrom, 2004), deletion mutants exhibit dramatically decreased A site ribosome occupancy, but a corresponding increase in P site occupancy over the relevant codons (Figures 2D–E). The observation suggests that this modification could potentially play some role(s) in peptidyl-tRNA positioning, enhancing translocation of codons from the P to the E site, or delaying translocation from the A site to the P site. Curiously, loss of the dehydroyuridine synthase Dus2, which is responsible for dU20 formation in the majority of tRNAs (Xing et al., 2004), caused a similar (albeit less dramatic) reduction in A site occupancy of serine and leucine codons (Figure 2A), although it did not cause the same compensatory increase in P site occupancy. In addition to relatively specific changes occurring over known target codons, we observed more widespread changes in A site occupancy in mutants lacking Trm1 or Trm112 (Figure 2A), consistent with the broad substrate range for these methylases – Trm1 generates N2,N2-dimethylguanosine (m²₂G26) in the majority of cytoplasmic tRNAs (Ellis et al., 1986), while the Trm112 methylase scaffold is required for appropriate methylation of tRNAs, rRNAs, and translation factors (Liger et al., 2011). Interestingly, in both of these mutants, A site occupancy tends to increase over codons beginning with purines, and decrease over codons beginning with pyrimidines (Figure 2A).

In addition to these and other cases (Table S6) that confirm and extend expected aspects of tRNA modification, we uncovered a number of surprising changes in ribosome occupancy that suggest additional roles for tRNA modifications in translation (Figure 2F, Supplemental Figures S3C–E). For example, in several analyses (here and below) we observed discrepant phenotypes for the four tyw mutants, despite the fact that loss of any of the four encoded enzymes completely eliminates wybutosine (yW modification at position 37 in tRNA-Phe) synthesis in vivo (Noma et al., 2006). Here, we noted that all four deletion mutants exhibited increased ribosomal P site occupancy over relevant codons, but that A site occupancy at these codons was increased only in mutants lacking Tyw1-3 (Figure 2F). As the yW precursor “yW-72” (lacking a methyl group on the α-carboxyl group and a methoxycarbonyl group on the α-amino group of yW) accumulates in target tRNAs in tyw4Δ mutants (Noma et al., 2006), we speculate that yW-72 may be sufficient for appropriate decoding of tRNA-Phe, but that the full yW modification is required for efficient peptidyl transfer and/or translocation. As another example, although we observe significantly increased P site occupancy over valine codons GUC/GUG/GUU in mutants affecting the Trm8/Trm82 heterodimer (Alexandrov et al., 2006), these mutants also exhibit unexpected decreases in P site occupancy of UGC and UGU, which are decoded by tRNA-Cys-GCA (Supplemental Figure S3D–E).

Taken together, these analyses recapitulate previously-reported translational deficits and thereby validate the quality of our dataset, as well as illuminating additional functions or targets of various tRNA modifications, providing hypotheses for mechanistic followup.

Trm7 methylates both tRNAs and rRNAs

Turning from the relatively subtle codon-level ribosome occupancy phenotypes described above to gene-level analysis of ribosome occupancy, we noted a particularly dramatic phenotype in initial surveys of the ribosome footprint landscape. Specifically, yeast lacking the tRNA methylase Trm7 – which is required for 2′-O-methylation at positions 32 and 34 of the anticodon loop of several tRNAs (Pintard et al., 2002) – exhibit dramatic, widespread changes in ribosome occupancy at a large number of 5′ UTRs and 5′ coding regions (Figure 3A). This was unique to trm7Δ among all 57 mutations in this study (Supplemental Figure S4A), and appeared to be an artifact of globally reduced translation in this mutant – the 5′ enrichment of ribosomes was only observed in libraries prepared in the presence of cycloheximide (Supplemental Figure S4B), suggesting that an excess of free ribosomal subunits are present in this mutant which can assemble onto mRNA upstream of cycloheximide-arrested ribosomes during cell lysis. Consistent with this hypothesis, we confirmed a significantly-decreased abundance of polysomes in trm7Δ mutants (Figure 3B), as previously reported (Pintard et al., 2002).

Figure 3. Dramatic effects of Trm7 on ribosome occupancy profiles.

(A) Increased ribosome occupancy at 5′ UTRs in trm7Δ mutants. RNA-Seq and RPF data for wild-type and trm7Δ mutant yeast at characteristic genomic loci. Red arrows show examples of increased ribosome occupancy in the mutant.

(B) Polysome profiles of the indicated strains reveal a global deficit in translation in trm7Δ. WT and unrelated trm4Δ are shown for comparison. See also Supplemental Figures S4A–B.

(C) RiboMeth-Seq analysis of rRNA 2′-O-methylation in wild-type yeast. Top panel shows counts (normalized to reads per million rRNA-mapping reads) of sequencing reads starting across 54 nt of 18S rRNA. The three annotated locations are dramatically under-represented, and correspond to three well-known 2′-O-methylation sites on 18S rRNA. Bottom panel shows methylation “A scores” (Birkedal et al., 2015; Marchand et al., 2016) aggregated for 8 wild-type datasets – individual replicates are nearly indistinguishable – with * indicating previously-validated methylation sites. Our dataset also recovers known methylation sites on 25S rRNA (Supplemental Figure S4C).

(D) Trm7 effects on rRNA RiboMeth-Seq. Scatterplot compares methylation A scores for WT (x axis, n=8) and trm7Δ (y axis, n=8) strains. The five significantly differentially-methylated nucleotides are indicated with large purple points, and lose methylation in trm7Δ but are unaffected in the unrelated trm3Δ mutant (Supplemental Figures S4D–E).

(E) Normalized RiboMeth-Seq 5′ end read starts (as in (C), top panel) for 14 nt surrounding 25S rRNA C663, as indicated.

(F) Comparison of mutant effects on the five candidate Trm7 target sites in 25S rRNA, shown as the change in A score for each mutant replicate relative to the average of 8 WT replicates. For the five Trm7 target nucleotides, data are shown for WT, trm7, and trm3 mutants (n=8 each), and for trm13, trm44, trm732, and rtt10 mutants (n=2 each). Note that C663 methylation is lost in mutants affecting Trm7 as well as one of its heterodimerization partners, Rtt10, while the remaining four potential Trm7 target sites are not affected by either Trm732 or Rtt10.

This dramatic translational phenotype suggested that Trm7 may have additional substrates, particularly since neither of the two known partners of Trm7 – Trm732 and Rtt10 – had similar effects on ribosome occupancy at 5′ UTRs (Supplemental Figure S4A). As the closest Trm7 homolog in bacteria is an rRNA methylase (Pintard et al., 2002), we tested the hypothesis that Trm7 might also methylate rRNA in budding yeast. We assessed 2′-O-methylation of rRNA using RiboMeth-Seq (Marchand et al., 2016), in which limited alkaline hydrolysis of RNA is used to cleave RNAs at all positions with an unmodified 2′ hydroxyl, allowing sequencing-based identification of 2′-O-methylation sites based on a reduction in sequencing reads starting immediately downstream of the methylated ribose.

Our RiboMeth-Seq data obtained from wild-type yeast recovered all known 2′-O-methylation sites in yeast 18S and 25S rRNA, with high concordance between 8 replicate libraries (Figure 3C, Supplemental Figure S4C, Table S7). Comparing rRNA methylation in WT, trm7Δ, and the unrelated trm3Δ mutant (n=8 each), we identified five significantly hypomethylated sites, all of which were specific to the trm7Δ mutant (Figure 3D–E, Supplemental Figure S4D–E). Extending this analysis to several additional mutants, including other 2′-O-methylases as well as Trm7’s known dimerization partners, revealed that four of the five candidate Trm7 target sites were affected exclusively in trm7Δ, but also that the strongest candidate – C663 – was in addition hypomethylated in mutants lacking Rtt10, one of the two known heterodimerization partners for Trm7 (Figure 3F). Given that the observed depletion of 5′ end reads starting downstream of C663 is not complete (Figure 3E), we infer that this nucleotide is likely to be partially-methylated across the population of ribosomes in the cell, perhaps suggesting a role for Trm7/Rtt10 in context-dependent methylation of specific subpopulations of ribosomes. Thus, although the basis for the global effects of Trm7 on translation remains unclear, our data do support the hypothesis that the tRNA methylase Trm7 also methylates a subset of rRNA molecules in vivo.

mRNA abundance changes report on diverse cellular functions impacted by tRNA modifications

We next sought to assess how the loss of specific tRNA modifications affects genomic output at the levels of mRNA abundance and ribosome occupancy (and, by extension, translation efficiency) of individual genes. As expected, changes in mRNA abundance were generally reflected in the ribosome occupancy dataset (Supplemental Figure S5). We focus first on mRNA levels, as changes in translational efficiency of key regulators (such as transcription factors) can cause widespread physiological and transcriptional changes that are readily appreciated by RNA-Seq, effectively amplifying the signal for biologically important translational regulation events.

Overall, we find robust mRNA abundance changes in roughly half (27/57) of the mutants analyzed, with minimal or no effects on mRNA abundance observed for the remaining 30 mutants (Figure 4A). The vast majority of mutants exhibiting substantial impacts on cell function, as measured by gene expression changes, are known to affect tRNA modifications in the anticodon itself or immediately adjacent to the anticodon, including 1) ncm⁵U, mcm⁵U and mcm⁵s²U at the wobble position (Elongator and related factors), 2) threonylcarbamoyladenosine (KEOPS), 3) wybutosine (Tyw3), 4) isopentenyladenosine (Mod5), 5) pseudouridine (Pus3/Deg1 and Pus7), and 6) ribose 2′-O-methylation (Trm7). These findings are consistent with the expectation that anticodon modifications should have greater effects on the primary tRNA function – decoding mRNAs – than modifications that occur elsewhere in the tRNA molecule. In addition to the factors involved in anticodon modification, a small number of proteins involved in tRNA modifications distant from the anticodon – such as Rit1, which is required for the 2′-O-ribosylphosphate modification at A64 of initiator tRNA that prevents initiator tRNA from participating in translational elongation (Astrom and Bystrom, 1994) – also exhibited robust gene expression changes. Interestingly, several mutants that robustly affected codon-specific translation, such as trm82Δ and tan1Δ, had only modest effects on gene expression.

Figure 4. Effects of tRNA modifying enzymes on RNA abundance.

(A) Overview of all RNA-Seq changes across the 57 mutants in this study. Data are shown for all genes changing at least 2-fold in at least 2 mutants (filtered for average mRNA abundance > 10 rpkm). Boxes show 5 relatively coherent gene expression clusters, with prominent functional annotations enriched in each geneset indicated. See also Supplemental Figure S5.

(B) Translational upregulation of GCN4 is a common occurrence in tRNA modification mutants. Top panels show RNA-Seq and RPF data for the GCN4 ORF and its 5′ UTR, which carries 4 well-studied regulatory upstream ORFs (uORFs). Zoom-ins focusing on the GCN4 coding region show RNA-Seq and RPF data for the indicated mutants.

(C) Mutant effects on GCN4 RNA and RPF levels are shown for all mutants, sorted from high to low Gcn4 translational upregulation.

(D) RNA-Seq correlates of GCN4 translational upregulation. Rows show Gcn4 targets (genes that exhibit >2-fold increase in RNA Pol2 occupancy in (Qiu et al., 2016)) for all mutants, sorted as in (C).

Consistent with prior studies on Elongator (Deng et al., 2015; Nedialkova and Leidel, 2015; Zinshteyn and Gilbert, 2013) and KEOPS (Daugeron et al., 2011), we observe upregulation of a large group of genes, primarily involved in amino acid biosynthesis and related metabolic pathways, in these mutants. This upregulation can be attributed to translational upregulation of the nutrient- and tRNA-responsive transcription factor Gcn4 (Hinnebusch, 2005) (Figures 4B–D). Our data recapitulate these prior findings, further validating the dataset. Moreover, we find that expression of GCN4 is translationally upregulated in several additional mutants, including pus3Δ, pus7Δ, rit1Δ, trm1Δ, trm7Δ, mod5Δ, and tyw3Δ. Thus, a wide variety of aberrations in tRNA function convergently result in increased synthesis of Gcn4, presumably as a consequence of impaired translation of regulatory upstream ORFs (uORFs) in the GCN4 5′ UTR. We also find that the upregulation of a proteostasis stress response previously described in Elongator mutants (Nedialkova and Leidel, 2015) is exhibited in additional mutants, with PRE3 upregulation occurring in mutants affecting Elongator, KEOPS, and Trm112, and MSN4 upregulation occurring more broadly across the set of mutants that affect the Gcn4 response (Table S1).

While the loss of multiple distinct tRNA modifying complexes induced a common transcriptional response through GCN4 upregulation, other gene expression changes were confined to a more limited set of mutants (Figure 4A) and thus were clearly not secondary to increased cellular levels of Gcn4. Among upregulated genes, a large group of genes related to mitochondrial function and carbohydrate metabolism were upregulated in KEOPS mutants and in tyw3Δ and sit4Δ mutants. Importantly, this was not a result of these strains having lost their mitochondrial DNA, as we detected abundant transcripts for mitochondrially-encoded genes in all of these strains. Other potential explanations for the physiology underlying this gene expression program include altered mitochondrial function resulting from loss of mitochondrial tRNA modifications, or altered expression of the respiration-regulating Hap4 transcription factor. However, in several mutants it is unlikely that this gene expression signature results from loss of the relevant tRNA modifications. Most notably, although the cell cycle phosphatase Sit4 is required for formation of mcm⁵s²U (Huang et al., 2008), sit4Δ is the only Elongator-related mutant exhibiting the carbohydrate/mitochondria transcriptional phenotype. Similarly, although all four Tyw proteins are required for wybutosine formation, only tyw3Δ mutants (which accumulate tRNAs modified with the “yW-86” intermediate (Noma et al., 2006)) upregulate HAP4 and related genes. On the other hand, this phenotype might potentially reflect a bona fide consequence of loss of t⁶A in KEOPS-related mutants – although the aneuploid bud32Δ and gon7Δ strains exhibit much stronger upregulation of carbohydrate metabolism genes than do cgi121Δ mutants, cgi121Δ mutants, which maintain ~80% of wild-type levels of t⁶A, do exhibit a modest effect on these genes (Figure 4A). Given the unusual pattern of mutants exhibiting upregulation of carbohydrate metabolism genes, we did not further pursue this connection, although it may prove (at least in the case of KEOPS) an interesting area for future study.

Silencing-related phenotypes in tRNA modification mutants

We next addressed silencing-related phenotypes in the dataset, as tRNA-modifying enzymes have previously been implicated in silencing of subtelomeric and mating type reporters (Chen et al., 2011; Li et al., 2009b), and in telomere capping and recombination (Downey et al., 2006; Peng et al., 2015). Defects in various aspects of heterochromatin silencing in yeast result in separable transcriptional responses, which in turn provide robust proxies for function of the relevant silencing pathways. For example, the dramatic downregulation of haploid-specific genes (such as those encoding the pheromone response pathway) observed in bud32Δ and gon7Δ mutants (Figure 4A) is typical of the “pseudo-diploid” state of budding yeast mutants that fail to repress the silent mating loci (Rusche et al., 2003).

To systematically explore mutant effects on mating locus and subtelomeric silencing (Figure 5A), we plotted expression of haploid-specific genes (HSGs) and subtelomeric genes across all mutants in this study (Figure 5B, top and middle panels). We also included PHO genes as a separate class (bottom panel) – although several PHO genes are located near chromosome ends, we noted that both telomere-proximal and -distal PHO genes were downregulated in essentially all mutants that affect GCN4 mRNA translation (Figure 4A). Mutants in Figure 5B are sorted according to their effects on HSG expression – as is clear in Figure 4A, bud32Δ and gon7Δ exhibit the most dramatic downregulation of HSGs. A small number of additional mutants – sit4Δ, kti11Δ, trm7Δ, mod5Δ, and cgi121Δ – showed moderate (~1.5-fold) downregulation of these genes, with the majority of mutants, including most of the Elongator-related mutants, exhibiting extremely subtle changes in HSG expression. Derepression of subtelomeric genes was similarly restricted primarily to KEOPS mutants. Downregulation of subtelomeric PHO genes such as PHO89 (Figure 5A–B) is shown here to emphasize the distinction between the small subset of mutants that specifically affect silencing-related phenotypes, and the larger class of mutants exhibiting relatively nonspecific phenotypes (as in, e.g., Gcn4 upregulation).

Figure 5. Analysis of silencing-related phenotypes.

(A) RNA-Seq data for a ~15 kb locus adjacent to TEL2R. Two notable phenotypes are indicated with arrows – repression of PHO genes, observed in a wide range of mutants in this study (Figure 4A), and derepression of a subset of subtelomeric genes, which is confined primarily to mutants in the KEOPS complex.

(B) Mutant effects on expression of haploid-specific genes (a robust reporter for silent mating locus derepression), subtelomeric genes, and PHO genes, as indicated. Mutants are sorted by their average effect on haploid-specific genes.

(C) Mutant effects on translational efficiency of Sir proteins.

(D) Downregulation of TY1 expression in Elongator-related mutants. Cluster shows genes from Cluster 4 (Figure 4A) – structural genes encoded by the TY1 retroelement are indicated with orange boxes.

(E) ORFs downregulated in Elongator-related mutants are associated with TY1 long terminal repeats (LTRs). Top panels show RNA-Seq data for TYE7 for WT and a representative Elongator-related mutant. Bottom panels show genomic loci associated with the ORFs shown in panel (D).

(F) Elongator effects on target genes are not mediated via changes in TYE7 expression. Q-RT-PCR for two ORFs and for a TY1 element, as well as two normalization controls (TEF1 and TDH3), were performed in one of six strain backgrounds – wild-type, uba4Δ, elp3Δ, tye7Δ, tye7Δuba4Δ, and tye7Δelp3Δ, as indicated – in four replicates. All data are normalized to the wild-type expression levels. Left panel validates our RNA-Seq observations, while right panel shows Elongator effects on these genes in the absence of Tye7. TY1 mRNA levels are decreased in tye7Δ – as expected – but, importantly, deletion of Elongator leads to a further decrease in TY1 expression. See also Supplemental Figure S6.

As the silencing phenotype previously reported for elp mutants was ascribed to defective translation of the SIR4 mRNA (Chen et al., 2011), we next examined the translational efficiency of the SIR mRNAs in our dataset (Figure 5C). Consistent with the dramatic silencing defects observed in bud32Δ and gon7Δ, we found that the translational efficiency of SIR2 mRNA was significantly decreased in these two mutants (Figure 5C). However, outside of this connection, silencing phenotypes were otherwise poorly-correlated with SIR mRNA translational efficiency as assayed by ribosome footprinting. For example, although mod5Δ mutants did exhibit modest changes in SIR2 mRNA translation accompanied by moderate downregulation of HSGs, nearly-identical changes in SIR2 translation in pus3Δ and ncs2Δ mutants did not result in appreciable mating locus derepression (Figures 5B–C).

Taken together, our data indicate that defects in silencing of endogenous loci are relatively rare in mutants that affect tRNA modifications, with the most dramatic silencing defects being confined to the unusual case of the two aneuploid KEOPS mutant strains.

Regulation of TY1 expression by Elongator

Although robust silencing defects were largely confined to bud32Δ and gon7Δ, we uncovered a surprising silencing-related phenotype in Elongator-related mutants. Specifically, we observed substantial downregulation of TY1 retrotransposon expression occurring almost exclusively in Elongator-related mutants (Figure 5D). Interestingly, a small number of endogenous protein-coding genes were also downregulated in the same subset of mutants, and inspection of these genes reveal that all such genes are located in genomic neighborhoods in proximity to intact TY elements, solo TY1 LTRs, and tRNA genes (Figure 5E). This link was of great interest to us given the central role for tRNAs in the biology of LTR retrotransposons (Marquet et al., 1995; Weiner and Maizels, 1987), and raise the question of how the Elongator complex affects TY-linked gene expression.

Although diminished TY1 expression occurs, counterintuitively, in sir mutants (Lenstra et al., 2011), this is unlikely to explain the downregulation we document for Elongator mutants – KEOPS and other mutants that affect other aspects of Sir-dependent silencing in this dataset do not cause TY1 repression, and conversely the various Elongator-related mutants exhibit subtle or no effects on other Sir-dependent phenotypes (Figures 5B, D, and Supplemental Figures S6A–B). In addition, the highly Elongator-specific effects on TY1 regulation cannot be a result of GCN4 upregulation, which occurs in a much broader group of mutants (Figure 4). We next considered the possibility that TY-adjacent genes are affected in Elongator mutants as a secondary consequence of altered levels of the target gene TYE7 (Figure 5D–E), which encodes a known transcriptional activator of TY LTRs (Lohning and Ciriacy, 1994). However, q-RT-PCR of several target genes in a tye7Δ background revealed further decreases in mRNA abundance in tye7Δelp3Δand tye7Δuba4Δ double mutants (Figure 5F), demonstrating that altered regulation of TY1 elements does not result from Elongator’s effects on the endogenous TYE7 locus. Finally, although most phenotypes of Elongator mutants are suppressed by overexpression of a subset of its target tRNAs (Chen et al., 2011; Esberg et al., 2006; Nedialkova and Leidel, 2015; Zinshteyn and Gilbert, 2013), we found that Elongator’s effect on expression of TY-adjacent genes was unaltered by overexpression of two such tRNAs (Supplemental Figure S6C). These data suggest that Elongator’s effects on TY element expression could result from the modification of other target tRNAs, or, more intriguingly, that its control of TY expression may not require wobble nucleotide modification.

Given the many links between tRNAs and LTR element replication (Marquet et al., 1995), it will be of great interest in future studies to determine how Elongator functions to support expression of genes located near TY LTRs.

Gene-specific changes in translational efficiency reveal regulatory uORFs

We finally turn to analysis of translational efficiency in our dataset. Although mutant effects on the translational control of key regulatory genes, such as GCN4, result in an amplified response at the level of mRNA abundance, altered synthesis of many proteins is not expected to cause dramatic transcriptional phenotypes and thus mutant effects on translational efficiency must be addressed directly. We note that increased ribosome occupancy on a coding region can result from increased translation, as observed for example for GCN4, or from slowed or stalled translation. However, analysis of mutants with significant effects on codon-level occupancy (Figure 2) revealed that although in some cases genes with high levels of the affected codons exhibited increased ribosome occupancy (Supplemental Figure S7) – suggesting slowed or stalled translation – this effect was quantitatively extremely modest overall, with codon frequency in open reading frames typically explaining no more than ~1–2% of the variance in translational efficiency. Thus, slowed translation contributes modestly to overall ribosome occupancy, which instead primarily reports on translational efficiency.

Figure 6A shows clustered translational efficiency for all 57 mutants, relative to wild-type. As with mutant effects on mRNA abundance, we noted that mutants that affect nucleotide modifications at or adjacent to anticodons exhibited altered translation of many more transcripts than did mutants affecting distal nucleotides. Interestingly, although we identified a handful of relatively specific translational changes in subsets of these mutants, overall we find that the majority of mutants that affect anticodon modifications tended to affect translation of a common group of mRNAs, suggesting that many of the affected genes respond to some aspect of overall translational efficiency (such as, e.g., efficiency of uORF translation), rather than to levels or functionality of individual tRNAs.

Figure 6. Effects of tRNA-modifying enzymes on translational efficiency.

(A) Overview of translational efficiency dataset. Heatmap shows log2 fold changes, relative to wild-type, of all genes with TE changes of at least 2-fold in 2 or more mutants. See also Supplemental Figure S7.

(B) Translational regulation of SER3 by uORFs. RNA and RPF (ribosome-protected footprint) data are shown for WT, pus7Δ (where SER3 is unaffected) and elp1Δ, in which SER3 translational efficiency is increased. Notable here is a peak of ribosome occupancy over the upstream regulatory transcript SRG1 which is lost (despite no change in SRG1 RNA abundance) in mutants that translationally derepress SER3.

(C–E) Examples of genes translationally repressed in various tRNA modifying enzyme mutants. Data shown as in panel (B), with green arrows highlighting diminished ribosome occupancy of ORFs, and red arrows highlighting likely regulatory uORFs. Here, known (CPA1) or putative (CMR3, YGP1) upstream regulatory ORFs are highlighted in red in the genomic annotation.

Focusing first on translationally-upregulated genes, we identified a small group of target genes that exhibited a similar mutant profile to that of GCN4. Most notably, we found that translation of SER3 mRNA was highly correlated with that of GCN4 across our dataset. Closer inspection of the SER3 locus revealed clear evidence for ribosome occupancy upstream of the SER3 start (Figure 6B), falling within a previously-described regulatory transcript known as SRG1 (Martens et al., 2004). Although SRG1 was originally described as a sense-strand cryptic transcript that is terminated near the SER3 AUG, Martens et al. also noted the presence of long readthrough SRG1 transcripts extending to the SER3 3′ end, and we find multiple sequencing reads spanning the SRG1/SER3 junction, indicating that a subset of SER3 transcripts include the SRG1 sequence as their 5′ UTR. These results are most consistent with a model in which translation of SER3 mRNA is regulated by a uORF in a manner analogous to the intensively-studied mechanism of GCN4 regulation (Hinnebusch, 2005), and imply that SRG1 plays separable roles in regulation of SER3 at both transcriptional and translational levels.

Turning next to translationally down-regulated genes, we noted that CPA1, which is known to be translationally regulated by an upstream “attenuator peptide” (Gaba et al., 2005), was downregulated in essentially the same broad set of mutants that affect GCN4 translation (Figures 6A, C). A number of other transcripts were translationally repressed in the same set of mutants that affected CPA1, and in many cases we found evidence for uORFs that likely confer translational regulation on the downstream ORFs (Figures 6A, D, E). Together, these data provide an expanded survey of presumptive regulatory uORFs, with potential implications for understanding the distinctions between uORFs with stimulatory, vs. repressive, effects on downstream ORF translation.

DISCUSSION

This dataset provides a unique resource for understanding the roles for tRNA-modifying enzymes and their various cofactors in translational regulation. A key feature of this study is the comparison of multiple disparate mutants within the same dataset, which provides a valuable opportunity to constrain hypotheses for the mechanisms underlying translational phenotypes of interest (see, for example, Figure 5). Overall, we find that those tRNA modifications that occur at or adjacent to the anticodon have the greatest effects on ribosome occupancy, as expected. That said, some modifications that are distant from the anticodon (e.g. tRNA position 12), also have strong effects on ribosome occupancy. The absence of transcriptional or ribosome occupancy phenotypes for many of the remaining mutants involved in tRNA modification could reflect a variety of factors – regulatory feedback could maintain high levels of tRNAs that are destabilized in the absence of a given modification, or certain tRNA modifications could play important roles under alternative growth conditions that stress the proteostasis machinery.

More granular analyses at varying levels of resolution from gross transcriptional phenotypes to gene-centric and codon-centric ribosome occupancy reveal both expected behaviors of various mutants as well as unanticipated observations that inform mechanistic hypotheses for future study. We highlight several striking examples of such findings, such as the distinction between the tyw mutants in A and P site occupancy at relevant codons, and many additional related examples can be found in the various Supplemental Tables.

Most surprisingly, we discover a role for the Elongator complex and other factors required for wobble U modifications (mcm⁵s²U and related) in control of expression of both intact TY1 elements as well as multiple endogenous genes associated with solo LTRs. This finding is of great interest given the ancient links between tRNAs and LTR retroelements – tRNAs or tRNA-like RNA structures almost universally serve as primers for reverse transcriptase (Marquet et al., 1995; Weiner and Maizels, 1987), and recent studies implicate cleaved tRNA fragments in control of LTR-associated genes (Martinez et al., 2017; Schorn et al., 2017; Sharma et al., 2016). We consider a number of hypotheses for the mechanistic basis for Elongator control of TY1, ruling out roles for the Sir complex, Gcn4, or Tye7 as mediators of this effect. These findings reveal a surprising connection between a tRNA-modifying complex and control of LTR elements, and mechanistic dissection of the role for Elongator in TY transcription or mRNA stability will be of great interest.

STAR METHODS

Contact for Reagent and Resource Sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Oliver Rando (oliver.rando@umassmed.edu).

Experimental Model and Subject Details

Yeast strains and culture conditions

All strains were generated in the BY4741 background (MATa his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0). Haploid deletions were generated by sporulation and tetrad dissection of heterozygous deletions in the diploid BY4743 background, which were obtained either from the Yeast Knockout Heterozygous Collection (Dharmacon), or generated de novo via replacement of genes of interest with KanMX6 in BY4743. Haploid deletions were selected on YPD+G418, Lys minus (SD +His +Leu +Met +Ura with glucose), and Met minus (SD +His +Leu +Lys +Ura with glucose) media. MATa and deletion genotypes were verified by PCR. Cells were cultured in YPD+G418 at 30°C for small amounts and were amplifi ed in YPD at 30°C for experiments. After initial analysis of RNA-Seq from all deletion mutants, we identified clear evidence of aneuploidy (consistently elevated expression across entire chromosomes) for several strains. These strains were freshly re-made, and, with the exception of gon7Δ and bud32Δ for which all isolates obtained carried an addition ChrIX copy, all remade strains were confirmed to be euploid.

Strains tye7Δ, tye7Δ elp3Δ, and tye7Δ uba4Δ were generated by replacing TYE7 with URA3 in BY4741, elp3Δ, and uba4Δ, respectively, and selected on SC-Ura media. The cells were cultured in SC-Ura media for small amounts and were amplified in YPD at 30°C for experiments. Yeast strains with tRNA overexpression plasmids were generated using lithium acetate-based transformation and selected on SC-Leu media. The plasmids were generous gifts from Sebastian Leidel and Wendy Gilbert. All yeast strains used in this study and the oligos for making the deletion strains are listed in Table S8 and Table S9, respectively.

Method Details

Ribosome profiling and RNA-seq

Ribosome profiling was carried out as described in (Heyer and Moore, 2016), with minor modifications. Yeast strains were grown overnight to mid-log phase (OD₆₀₀=0.5–0.6) in YPD at 30°C, treated with cycloheximide (CHX) in a final concentration of 100 μg/ml for 30 seconds with vigorous shaking, and harvested by centrifugation at 4°C for 2 minutes. The cell pellets were flash-frozen in liquid nitrogen immediately. Total time from adding CHX to snap-freezing was about 4 minutes. For ribosome footprinting, cells were lysed in ice-cold lysis buffer (20 mM Tris-HCl, 140 mM KCl, 1.5 mM MgCl₂, 1% Triton X-100, 0.5 mM DTT, 100 μg/ml CHX) by glass-bead beating in an ice-cold block, and the clarified ribosome extract was obtained by centrifugation for 5 minutes at 9,500 rpm, 4°C. The ribosome extract was diluted with ice-cold lysis buffer, aliquoted to 5 A₂₆₀ units per tube, and followed by incubating with 500U of RNase I at 23°C for 1 hour. Monosomes were separated on 12 ml of 10–50% sucrose gradient (20 mM Tris-HCl, 140 mM KCl, 5 mM MgCl₂, 0.5 mM DTT, 20 U/ml SUPERase-In, 100 μg/ml CHX) in Sw40 ultracentrifuge tubes by centrifugation for 160 minutes at 35,000 rpm, 4°C. Gradients were fractioned using Brandel Density Gradient Fractionation System, and 80S monosome fractions were collected and flash-frozen immediately. RNA was extracted using TRIzol and 2-propanol precipitation. 27–34 nt ribosome footprints were isolated using denaturing 15% polyacrylamide-TBE-urea gels and then purified with Zymo ZR small-RNA PAGE Recovery Kit.

For CHX-free ribosome profiling, yeast cells were harvested by rapid filtration onto 0.8-um-pore-size mixed cellulose membrane. The cells were scraped and snap-frozen in liquid nitrogen immediately. Frozen cells and frozen pellets of lysis buffer (the same as above) were ground together into fine powder using a Retsch CryoMill with sample chambers pre-chilled in liquid nitrogen. After thawing the frozen powder on ice, the ribosome extract was obtained by centrifugation for 5 minutes at 9,500 rpm, 4°C, and the same procedure was followed.

For RNA-Seq, total RNA was extracted from the same yeast culture for ribosome footprinting using TRIzol, followed by DNase digestion and purification using Zymo RNA Clean & Concentration Kit. DNase-free RNA was depleted of rRNA using Illumina Ribo-Zero Gold Kit, followed by zinc-based fragmentation for 10 min at 70°C. RNA fragments and ribosome footprints were constructed into deep-sequencing libraries as described in (Heyer et al., 2015). Briefly, RNA 3′ ends were dephosphorylated using T4 PNK, ligated with a 5′-preadenylated adaptor using T4 RNA Ligase 2 truncated K227Q, and reverse transcribed with barcode primers. cDNA was precipitated using 2-propanol and size selected using denaturing 6% polyacrylamide-TBE-urea gels. cDNA was purified using traditional “crush and soak” method, followed by circularization with epicenter CircLigase and PCR amplification prior to sequencing.

2′-O-methylation sequencing (RiboMeth-seq)

RiboMeth-seq was carried out essentially as described in (Marchand et al., 2016), except for 3′-end dephosphorylation. 500 ng of total RNA extracted for RNA-seq was subjected to fragmentation using alkaline solution (100 mM Na₂CO₃-NaHCO₃, pH 9.2) for 10 minutes at 95°C, followed by RNA purification using Zymo RNA C lean & Concentration Kit. Purified RNA fragments were 3′ dephosphorylated using various enzymes, 5′ phosphorylated using T4 PNK in the presence of ATP, and finally constructed into deep-sequencing library using NEBNext Small RNA Library Prep. For WT, trm7Δ, and trm3Δ, we initially generated 2 replicate datasets each using T4 PNK, Antarctic phosphatase, or Shrimp alkaline phosphatase for 3′-end dephosphorylation, with no significant effects of any of these variant protocols on methylation. A second round of libraries was built using T4 PNK for 3′ end dephosphorylation, with two additional replicates for WT, trm7Δ, and trm3Δ (final n=4 for T4 PNK protocol, n=8 across all 3 protocols for these three strains), as well as 2 replicates each for trm13Δ, trm44Δ, trm732Δ, and rtt10Δ.

Polysome profiling

The preparation of ribosome extract and sucrose gradient for polysome profiling was the same as mentioned above, but with incubating 5 A₂₆₀ units of ribosome extract with SUPERase-In instead of RNase I. Fractionation and detection of ribosomes were performed using the same system and its data capture software.

Quantitative RT-PCR

Total RNA was isolated using TRIzol with glass-bead beating, followed by the purification using Zymo Direct-zol RNA MiniPrep with in-column DNase I treatment. cDNA was generated using SuperScript IV Reverse Transcriptase with random hexamers for priming. Quantitative PCR was performed using KAPA SYBR Fast qPCR Master Mix 2X Universal with ~10 ng of cDNA and ROX Low in QuantStudio 3 Real-Time PCR System. Relative fold changes in mutants were calculated as 2^(−ΔΔCt), and the gene expression was normalized to the expression of TEF1 and TDH3.

Quantification and Statistical Analysis

Analysis for ribosome profiling and RNA-seq

Sequencing read mapping and gene-level analysis

Barcoded libraries were pooled and sequenced on an Illumina NextSeq500. Raw fastq reads were de-multiplexed and removed of adaptor sequence using HOMER package (Heinz et al., 2010). RPF reads were mapped to S. cerevisiae rDNA and the mapping reads were discarded. The remaining RPF reads and RNA-seq reads were mapped to sacCer3 genome using TopHat v2.0.12 (Trapnell et al., 2009) with parameters -p 4 -I 5000 --no-coverage-search. Unique mapping reads were saved using SAMtools (Li et al., 2009a) view function and parameter -q 10. Reads in length of 27–34 nt (RPF) or ≥27 nt (RNA-seq) were used for gene-level analysis, in which reads were quantified as raw counts or reads per kilobase of transcript per million mapped reads (RPKM) using HOMER analyzeRepeats.pl function with open reading frame annotations downloaded from Saccharomyces Genome Database (SGD). The read coverage in UCSC Genome Browser tracks were generated using HOMER makeUCSCfile function with parameters -fragLength 30 for RPF or -fragLength 38 for RNA. Translational efficiency (TE) was calculated as relative RPKM of ribosome footprint density divided by relative RPKM of RNA abundance, with relative RPKM calculated as RPKM in each mutant divided by the average of RPKM in WTs grown in the same batch. Hierarchical clustering was performed using standard methods. Metagene analyses in Supplemental Figures S4A–B show normalized ribosome occupancy averaged across all genes, aligned in a defined region from 100-nt upstream to 110-nt downstream of the first nucleotide of start codon. It was analyzed using a Python-based package Plastid (Dunn and Weissman, 2016), in which it quantified the counts of the nucleotide in a read corresponding to the first nucleotide of ribosomal p-site in the defined region of each gene. Within each gene, the counts of each position in the defined region were normalized by the total number of counts in a normalization region, which is 70–100 nt downstream of the first nucleotide of start codon. Finally, the mean of normalized counts in each position across all genes was calculated.

Codon occupancy analysis

Global codon occupancy analysis was calculated as described in (Nedialkova and Leidel, 2015), with minor modifications. The P-site offsets were calculated by examining the cumulative distribution of 28–31 nt reads aligning at start codons using online package Plastid. After applying the respective offset to reads of each size, only in-frame reads were used. The first 15 and the last 5 codons of each transcript were removed from the reference. The frequency of each codon in ribosomal A, P, and E sites was calculated, and divided by the average frequency of the same codon in the three downstream codons from the A site for normalization.

Analysis for RiboMeth-seq

Indexed libraries were pooled and sequenced on an Illumina NextSeq500. De-multiplexed reads were mapped to S. cerevisiae rDNA using Bowtie2 v2.1.0 (Langmead et al., 2009) with parameters -p 4 --no-unal. 5′ ends of uniquely mapping reads were quantified using bedtools (Quinlan and Hall, 2010) genomecov function with parameters -d -5. Methylation A scores were calculated as described in (Birkedal et al., 2015; Marchand et al., 2016).

$$\mathrm{A}\text{score}\text{of}\text{position}\mathrm{i}=max\{0,1-\frac{2n_i}{0.5\mid {\mu }_l-{\sigma }_l\mid +n_i+0.5\mid {\mu }_r-{\sigma }_r\mid }\}$$

where n_i is the read counts at position i, μ_l and σ_l are the mean and standard deviation of ${\mathrm{\sum }}_{j=i-6}^{i-1}{n}_j$, respectively. μ_r and σ_r are the mean and standard deviation of ${\mathrm{\sum }}_{j=i+1}^{i+6}{n}_j$, respectively.

Data and Software Availability

All fastq files of deep-sequencing data can be found in GEO with the accession number GSE100626. UCSC Genome Browser Tracks for all RPF and RNA data can be accessed at the following links:

http://genome.ucsc.edu/cgi-bin/hgTracks?hgS_doOtherUser=submit&hgS_otherUserName=chouhj&hgS_otherUserSessionName=11Del_2015batch

This session consists of tracks for ribosome profiling and RNA-seq from 11 mutants – elp1Δ, elp3Δ, kti11Δ, urm1Δ, trm4Δ, trm7Δ, trm732Δ, rtt10Δ, rit1Δ, pus1Δ, and pus7Δ.

http://genome.ucsc.edu/cgi-bin/hgTracks?hgS_doOtherUser=submit&hgS_otherUserName=chouhj&hgS_otherUserSessionName=46Del_2016batch

This session consists of tracks for ribosome profiling and RNA-seq of another 46 mutants.

http://genome.ucsc.edu/cgi-bin/hgTracks?hgS_doOtherUser=submit&hgS_otherUserName=chouhj&hgS_otherUserSessionName=noCHX_2017

This session consists of tracks for cycloheximide-free ribosome profiling of 5 mutants – elp3Δ, mod5Δ, tan1Δ, trm7Δ, and tyw4Δ.

Supplementary Material

10. Table S1. RNA-Seq dataset. Related to Figures 1, 3, 4, 5, 6.

Sequencing depth-normalized RNA-Seq data for mutant and wild-type (columns), with rows showing the 4884 genes with at least 10 or more reads in all wild-type replicate RPF datasets.

2. Table S2. Ribosome footprinting dataset. Related to Figures 1, 3, 4, 5, 6.

As in Table S1, but for ribosome footprinting data.

3. Table S3. Translational efficiency. Related to Figures 4, 5, 6.

Mutant effects on translational efficiency, relative to wild-type, expressed as log2 fold change. Mutants effects on translational efficiency are calculated as RPF rRPKM/RNA rRPKM, with rRPKM calculated by normalizing RPKM in each mutant by the average of RPKM in WTs grown in the same batch.

4. Table S4. No cycloheximide ribosome footprinting dataset. Related to Figures 1–6.

As in Table S2, for six replicate ribosome profiling experiments carried out without cycloheximide treatment.

5. Table S5. Ribosome-protected fragment lengths. Related to Figure 1.

Fraction of each fragment size in 27–34 nt for all ribosome footprinting datasets after sequence demultiplexing, 5′ CC and 3′ adaptor trimming, and rRNA removal.

6. Table S6. Codon occupancy. Related to Figure 2.

Ribosome footprints of 28–31 nt were analyzed as described in STAR Methods. Columns show individual replicates of wild type and various mutants, while rows show A-, P-, or E-site occupancy on 61 codons (no stop codons are included in the analysis since the first 15 and the last 5 codons of each transcript were removed from the reference).

7. Table S7. RiboMeth-Seq dataset. Related to Figure 3.

RiboMeth-Seq data for WT yeast and the indicated mutants. Sheets include raw counts of 5′ read starts.

8. Table S8. Strain list. Related to STAR Methods.

All yeast strains used in this study.

9. Table S9. Oligonucleotide list. Related to STAR Methods.

Oligonucleotide sequences used in this study.

Research Highlights.

1. Systematic analysis of translational functions of tRNA modifying enzymes

2. Identification of novel translationally-regulated target genes in budding yeast

3. Unexpected role for tRNA modifying enzyme in control of LTR element expression