A stress‐induced tyrosine‐tRNA depletion response mediates codon‐based translational repression and growth suppression

Abstract

Eukaryotic transfer RNAs can become selectively fragmented upon various stresses, generating tRNA‐derived small RNA fragments. Such fragmentation has been reported to impact a small fraction of the tRNA pool and thus presumed to not directly impact translation. We report that oxidative stress can rapidly generate tyrosine‐tRNA_GUA fragments in human cells—causing significant depletion of the precursor tRNA. Tyrosine‐tRNA_GUA depletion impaired translation of growth and metabolic genes enriched in cognate tyrosine codons. Depletion of tyrosine tRNA_GUA or its translationally regulated targets USP3 and SCD repressed proliferation—revealing a dedicated tRNA‐regulated growth‐suppressive pathway for oxidative stress response. Tyrosine fragments are generated in a DIS3L2 exoribonuclease‐dependent manner and inhibit hnRNPA1‐mediated transcript destabilization. Moreover, tyrosine fragmentation is conserved in C. elegans. Thus, tRNA fragmentation can coordinately generate trans‐acting small RNAs and functionally deplete a tRNA. Our findings reveal the existence of an underlying adaptive codon‐based regulatory response inherent to the genetic code.

Ribonuclease‐generated fragments of one particular tRNA control expression of growth‐promoting genes upon stress.

Introduction

Transfer RNAs (tRNAs) are universal decoders of the genetic code. By recognizing three‐nucleotide sequences (codons) in transcripts, tRNAs enable ribosomal incorporation of specific amino acids into the growing polypeptide chain. Because there exists a larger number of codons than amino acids, the code is degenerate, with multiple “synonymous” codons encoding a given amino acid. The human genome contains over 400 tRNA genomic loci, with multiple genes encoding tRNAs that contain the same anticodon, termed isodecoders (Parisien et al, 2013). The central roles of tRNAs in translation have been defined through elegant structural and biophysical studies (Nissen et al, 2000; Ogle et al, 2002). Recent studies have, however, challenged our textbook notions regarding these essential molecules and suggest that beyond their static roles as adaptors in translation, tRNAs play additional dynamic roles in gene regulation (Schimmel, 2018).

One line of support for non‐canonical roles for tRNAs was the discovery in Tetrahymena that starvation stress caused some tRNAs to undergo endonucleolytic cleavage into smaller fragments (Lee & Collins, 2005). In humans, tRNA‐derived fragments (tRFs) were originally detected in the urine of cancer patients but were of unknown function (Gehrke et al, 1979). A variety of stresses have since been observed to generate tRFs in organisms ranging from yeast to man (Thompson et al, 2008). Specific ribonucleases have also been implicated in generating tRFs in distinct species, including Rny1p in yeast and angiogenin (ANG) in human cells (Fu et al, 2009; Thompson & Parker, 2009a). Past studies reported that tRNA fragmentation did not noticeably deplete (< 1%) the precursor tRNA molecules of the specific tRFs being studied (Saikia et al, 2012). Such observations suggested that the primary consequence of tRF generation was trans action by these small RNAs rather than translational impairment owing to depletion of the precursor tRNA. Consistent with this hypothesis, endogenous tRFs have been found to interact in trans with RNA‐binding proteins and to mediate post‐transcriptional gene repression (Goodarzi et al, 2015; Kuscu et al, 2018). Attesting to their functional roles, inhibition of tRFs has been shown to impact malignant phenotypes at the cellular and organismal levels, while deletion of ANG elicited defects in hematopoiesis (Goncalves et al, 2016).

A second line of evidence for non‐canonical roles by tRNAs was the observation that expression levels of some tRNAs become modulated in the context of malignancy and cancer progression (Pavon‐Eternod et al, 2009; Goodarzi et al, 2016). Such observations challenged the common dogma that tRNAs are static components in mammalian cells. Comparison of breast cancer cells to non‐malignant breast cells revealed a number of tRNAs to be overexpressed and others repressed, perhaps owing to genomic instability and subsequent copy number alterations of tRNA genes (Pavon‐Eternod et al, 2009; Goodarzi et al, 2016). Such alterations in tRNA content have been associated with altered cellular protein expression (Chan et al, 2010; Pavon‐Eternod et al, 2013; Pershing et al, 2015) and mRNA stability (Hoekema et al, 1987; Presnyak et al, 2015; Boel et al, 2016). Other analyses revealed similar findings of large‐scale tRNA expression alterations in proliferative cancer cells (Gingold et al, 2014) or upregulation of specific tRNA isodecoders enhancing translation of specific proteins that promoted cellular invasiveness and metastatic capacity (Goodarzi et al, 2016). Mutagenesis studies of the KRAS oncogene also provided support for distinctness among “synonymous” codons; KRAS protein expression became upregulated upon mutation of a “rare” codon (decoded by a low‐abundant tRNA isoacceptor) to an “optimal” synonymous codon (decoded by an abundant tRNA isoacceptor) (Pershing et al, 2015)—consistent with tRNA availability impacting protein expression (Gustafsson et al, 2004).

The observed tRNA modulations and the ensuing codon‐dependent gene expression effects in the context of genomic instability and cancer raise the question of whether such tRNA modulations occur in normal cells and perhaps elicit translational effects in response to exogenous signals. We herein describe the existence of an endogenous gene regulatory response to oxidative stress in human cells that is mediated by the depletion of a specific tRNA. We observed that oxidative stress rapidly induced fragmentation of tyrosine tRNA_GUA—leading to tRNA^Tyr _GUA depletion and reduced expression of growth‐promoting genes enriched in tyrosine codons. The affected genes were significantly enriched in growth and metabolic pathways. Moreover, we observe that the tyrosine tRF^Tyr _GUA that is produced interacts in trans with the hnRNPA1 and SSB RNA‐binding proteins. Additionally, oxidative stress‐induced tyrosine‐tRNA_GUA fragmentation is also conserved in the nematode C. elegans. Our findings uncover a direct relationship between stress‐induced tRNA fragmentation and tRNA modulation, with functional consequences for translation.

Results

Systematic characterization of changes in tRNAs and tRFs as a response to oxidative stress

We hypothesized that stress‐induced tRNA fragmentation may deplete the pools of specific tRNAs. To search for such tRNAs, we sought to globally profile changes in tRNA and tRF levels upon exposure of cells to oxidative stress—a stress known to robustly induce tRF generation (Thompson et al, 2008; Yamasaki et al, 2009; Thompson & Parker, 2009a). Multiple independent methods have been recently developed to profile tRNA species (Cozen et al, 2015; Zheng et al, 2015; Goodarzi et al, 2016; Gogakos et al, 2017). We employed one such method—a probe‐based tRNA capture, ligation, and deep‐sequencing approach—to profile tRNA isodecoder abundances across samples (Goodarzi et al, 2016). At both 8 and 24 h after 200 µM hydrogen peroxide (H₂O₂) treatment of human mammary epithelial cells (MCF10A), tRNA levels were globally similar to tRNA levels in the non‐treated condition (Fig 1A). However, multiple specific tRNAs (tRNA^Tyr _GUA, tRNA^Ile _UAU, tRNA^Leu _UAA, tRNA^Thr _AGU, and others) became significantly decreased over time after H₂O₂ treatment. We next exposed MCF10A cells to oxidative stress and profiled small RNA (smRNA) abundances via deep‐sequencing (Fig 1B). This revealed induction of many tRFs—consistent with previous studies describing the effects of oxidative stress on fragmentation of specific tRNAs (Thompson et al, 2008; Saikia et al, 2012). The observed tRF inductions were not artifacts of cell death, as cell viability was unchanged in treated versus control samples at the concentration of H₂O₂ used (Fig EV1A). Integration of tRF sequencing and tRNA profiling analyses identified a set of candidate tRNAs that became fragmented and depleted upon oxidative stress (Figs 1B and C, and EV1B).

Figure 1. Identification of modulated mammalian tRNAs and tRFs in response to oxidative stress.

1. Heatmap of tRNA profiling of MCF10A cells at 8 and 24 h post‐exposure to oxidative stress (200 μM H₂O₂). Biological triplicate data is depicted at each time point relative to control cells.
2. MCF10A cells were exposed to oxidative stress (200 μM H₂O₂) and processed for small RNA‐sequencing. The log₂‐fold induction levels for tRFs derived from distinct tRNAs are plotted.
3. Schematic depicts the overlap of tRNAs in (A) that decreased over time with tRFs from (B) that were induced.

Figure EV1. Fragmentation of tRNA^Tyr _GUA in response to oxidative stress is not due to cell death.

1. MCF10A cell viability 1 h after H₂O₂ treatment (200 μM) was tested by a trypan blue exclusion assay (n = 6). Horizontal bars represent the mean. A two‐tailed Mann–Whitney test was used to test for statistical significance.
2. The log₂ fold induction of tRFs for all isoacceptors in the overlapping set of tRNAs found in Fig 1C.

Oxidative stress‐induced fragmentation depletes tRNA^Tyr _GUA

Northern blot analysis confirmed that oxidative stress induces generation of tRF^Tyr _GUA and tRF^Leu (multiple isodecoders). We focused our efforts on the tRF that exhibited the strongest induction by Northern blot (tRF^Tyr _GUA) and its associated tRNA (Fig EV2A). To determine if tRF^Tyr _GUA formation is conserved, we exposed C. elegans to H₂O₂. We observed similar generation of tRF^Tyr _GUA following brief exposure (15 min) of C. elegans to H₂O₂ (Fig EV2B). We next performed time course studies in human cells. TRF^Tyr _GUA became rapidly induced (within 5 min) upon cellular exposure to H₂O₂ (Fig 2A). TRF^Tyr _GUA generation was associated with a concomitant precipitous decline in pre‐tRNA^Tyr _GUA levels, which remarkably became nearly undetectable at 1‐h post‐treatment (Fig 2B). This suggests that the majority of tRF^Tyr _GUA is generated from the pre‐tRNA^Tyr _GUA rather than the mature tRNA, similar to our observations in C. elegans. As tRNAs are one of the most stable classes of RNAs, we would expect a delayed effect on mature tRNA abundances upon acute reduction of the pre‐tRNA pool. Indeed, we observed a significant delayed reduction in the tRNA^Tyr _GUA pool, observed at 24 h post‐H₂O₂ exposure (Fig 2A and B). The mature tRNA^Tyr _GUA pool diminished to roughly half the pre‐treatment levels at 24 h post‐treatment. These observations reveal that a single stress can induce generation of a tRNA fragment and significantly deplete its corresponding mature tRNA pool.

Figure EV2. Fragmentation of tRNA^Tyr _GUA occurs with other sources of oxidative stress and additional cell lines.

1. Northern blot for tRF^Tyr _GUA and tRF^Leu _HAG in MCF10A cells at 1 h post‐oxidative stress (200 μM H₂O₂) exposure.
2. Northern blot for tRF^Tyr _GUA in C. elegans cells at 15 min post‐oxidative stress (200 μM H₂O₂) exposure.
3. A Northern blot depicting a time course experiment ranging from 5 min to 24 h for HBEC30 cells in response to oxidative stress (200 μM H₂O₂). As before, a single probe complementary to pre‐tRNA^Tyr _GUA, mature tRNA^Tyr _GUA, and tRF^Tyr _GUA was ³²P‐labeled and used for detection in (C) and (D).
4. A Northern blot depicting two time points, 1 and 24 h, after exposure to oxidative stress (200 μM H₂O₂) in MCF10A cells.
5. Quantification of tRNA^Tyr _GUA, tRNA^His _GUG, and tRNA^Glu _YUC by Northern blot analysis from two independent experiments 24 h (normalized to U6 levels) after exposure to oxidative stress (200 μM H₂O₂) are shown (n = 4).
6. Northern blot after MCF10A cells were treated with a pharmacological agent used to induce oxidative stress (menadione).
7. Quantification of tRNA^Tyr _GUA by Northern blot analysis from two independent experiments 24 h (normalized to U6 levels) after exposure to menadione are shown (n = 4).

Data information: Data represent mean ± SEM. *P < 0.05. A one‐tailed Mann–Whitney test was used to test for statistical significance.

Figure 2. TRNA^Tyr _GUA abundance is reduced while the corresponding tRF is induced in response to oxidative stress.

1. A Northern blot depicting a time course experiment ranging from 5 min to 24 h of MCF10A cells in response to oxidative stress. A single probe complementary to pre‐tRNA^Tyr _GUA, mature tRNA^Tyr _GUA, and tRF^Tyr _GUA expression was ³²P‐labeled and used for detection.
2. Quantification of pre‐tRNA^Tyr _GUA (left) and mature tRNA^Tyr _GUA levels (right) by Northern blot analysis from multiple independent experiments (normalized to U6 levels) are shown (n = 6).
3. MCF10A cells were exposed to oxidative stress (200 μM H₂O₂) once daily for five continuous days.
4. Quantification of mature tRNA^Tyr _GUA bands by Northern blot after cells were treated once daily for five continuous days (normalized to U6) from multiple independent experiments (n = 12).
5. Quantification of Northern blot analysis for pre‐tRNA^Tyr _GUA (left) and tRNA^Tyr _GUA (right) after 1 and 24 h, respectively, in HBEC30 cells upon exposure to oxidative stress (200 μM H₂O₂) as in (A) (n = 6).

Data information: Data represent mean ± s.e.m. A one‐tailed Mann–Whitney test (**P < 0.01 and ***P < 0.001) was used to test for statistical significance between the treated and control cell lines for each time point.

We next determined whether oxidative stress‐induced tRNA^Tyr _GUA depletion is a transient response or whether it could persist upon continual exposure to stress. Treatment of cells with H₂O₂ once daily for five consecutive days maintained tRNA^Tyr _GUA repression (Fig 2C and D). RNA was collected on the sixth day, 24 h after the last H₂O₂ exposure, and as before, tRF^Tyr _GUA induction was maintained over this time (Fig 2C and D). Such continuous H₂O₂ treatment repressed mature tRNA^Tyr _GUA levels by roughly half. We observed similar effects in a distinct mammalian cell line, the human bronchial epithelial cell line (HBEC30) (Figs 2E and EV2C). Despite this reduction in mature tRNA^Tyr _GUA levels following H₂O₂ exposure, other tRNAs—tRNA^His _GUG and tRNA^Glu _YUC—remained unchanged relative to the control (Figs 2C and EV2D and E). Menadione, another commonly used pharmacological agent that induces oxidative stress, yielded similar results to that of H₂O₂ (Fig EV2F and G). Thus, stress‐induced tRNA depletion can be a sustained response and that it can be elicited in multiple human cell types.

Oxidative stress‐induced tRNA^Tyr _GUA depletion represses cellular growth

Our tRNA depletion observation was unlikely to be a consequence of a non‐specific cell death phenomenon as we observed depletion of specific tRNAs rather than a global effect on all tRNAs and no reduction in cell viability was seen (Fig EV1A). Consistent with such prior studies using intermediate and sublethal doses of H₂O₂ (Martindale & Holbrook, 2002), we observed impaired growth upon H₂O₂ treatment (Fig 3A). To determine the impact of reduced tRNA^Tyr _GUA activity on cell growth, we undertook two orthogonal approaches. By depleting tRNA^Tyr _GUA through short‐hairpin RNA‐induced inactivation, we generated a stable MCF10A cell line in which endogenous tRNA^Tyr _GUA was depleted to roughly the same level as that observed upon H₂O₂ treatment of cells (Fig 3B). As an independent loss‐of‐function approach, we sought to impair cellular utilization of tRNA^Tyr _GUA via RNAi‐mediated depletion of its cognate amino acid charging enzyme—the tyrosine‐tRNA synthetase (YARS) gene (Fig 3C). Due to guanine‐uracil wobble base pairing (Crick, 1966; Ladner et al, 1975; Quigley & Rich, 1976), tRNA^Tyr _GUA can recognize both codons (UAC and UAU) that code for the amino acid tyrosine. Moreover, we could not detect the other tyrosine tRNA (tRNA^Tyr _AUA) in MCF10A cells by Northern blot. This suggests that this synonymous tRNA may play a minimal, if any, role in translation in these cells and is a rare tRNA—as reported in other mammalian cells (dos Reis et al, 2004). We observed that impairment of tRNA^Tyr _GUA function by either tRNA^Tyr _GUA depletion or YARS depletion using two independent hairpins strongly impaired growth of MCF10A cells (Fig 3D). These results phenocopy the growth impairment seen with H₂O₂‐induced physiological depletion of endogenous tRNA^Tyr _GUA. In contrast, overexpression of tRNA^Tyr _GUA led to the opposite phenotype—increased cell growth (Figs 3E and EV3A and B). We propose that stress‐induced depletion of tRNA^Tyr _GUA constitutes an endogenous growth‐suppressive stress response that contributes in part to the cellular oxidative stress response.

Figure 3. Cell growth repression upon oxidative stress and tRNA^Tyr _GUA depletion.

1. Growth curves of MCF10A cells exposed to oxidative stress (200 μM H₂O₂) relative to control cells (n = 3). Two‐way ANOVA was used to test for significance.
2. Northern blot of MCF10A cells expressing a control short‐hairpin RNA or a hairpin targeting tRNA^Tyr _GUA.
3. Western blot of MCF10A expressing a control short‐hairpin RNA or a hairpin targeting the tyrosine‐tRNA synthetase, YARS (red arrow). HSC70 was used as a loading control.
4. Growth curves of MCF10A cells expressing RNAi targeting mature tRNA^Tyr _GUA or YARS relative to cells expressing a control hairpin (n = 3). Two‐way ANOVA was used to test for significance.
5. Cell growth of MCF10A cells transiently transfected with a tRNA^Tyr _GUA overexpression vector relative to an empty control vector (n = 3). A one‐tailed Mann–Whitney test was used to test for significance at day 3.

Data information: Data represent mean ± SEM. *P < 0.05 and ***P < 0.001.

Figure EV3. Validation of YARS knockdown and tRNA^Tyr _GUA overexpression.

1. Total mRNA from MCF10A cells stably expressing a short‐hairpin targeting YARS was analyzed by qRT–PCR. The two hairpins with the best knockdown were used for subsequent experiments. A one‐tailed Mann–Whitney test (*P < 0.05) was used to test for statistical significance.
2. Northern blot of MCF10A cells transiently transfected with a tRNA^Tyr _GUA overexpression vector or an empty control vector. TRNA^Tyr _GUA overexpression is shown at 24, 48, and 72 h post‐transfection.

Data information: Data represent mean ± SEM.

TRNA^Tyr _GUA depletion represses expression of a set of growth genes

We hypothesized that stress‐induced tRNA^Tyr _GUA depletion impairs growth by reducing production of proteins enriched in its corresponding tyrosine codons. Label free mass‐spectrometric proteomic profiling of cells depleted of tRNA^Tyr _GUA or of YARS revealed a highly significant correlation (R = 0.648; P < 2.2e‐16)—consistent with a common set of downstream genes being impacted by these orthogonal methods of tRNA^Tyr _GUA loss‐of‐function (Fig EV4A). Among these, we identified 109 proteins that were also enriched for Tyr codons and are therefore likely to be tRNA^Tyr _GUA‐dependent (Fig 4A). This set of tyrosine‐enriched proteins was most significantly enriched in gene ontology (GO) functional categories (Ashburner et al, 2000; The Gene Ontology C, 2017) associated with cellular growth, including regulation of ATP synthesis, G0 to G1 cell‐cycle progression, and phosphorylation (Fig 4B).

Figure EV4. Proteomic analysis identifies a tRNA^Tyr _GUA‐dependent gene‐set.

1. A plot showing the correlation between protein abundance changes with tRNA^Tyr _GUA depletion or YARS depletion (shYARS‐2) relative to control cells. A Pearson’s two‐sided test was used to assess the statistical significance of the correlation between tRNA^Tyr _GUA and YARS depletion effects.
2. Schematic showing locations of all tyrosine residues in the coding sequence of EPCAM, SCD, and USP3.
3. Levels of mRNA expression for target genes in cells depleted of either tRNA^Tyr _GUA or YARS as measured by qRT–PCR (n = 4). A one‐tailed Mann–Whitney test was used to test for statistical significance.
4. MCF10A cells transfected with two independent siRNA targeting EPCAM, SCD, or USP3 were analyzed by qRT–PCR at the end of each growth assay. A one‐tailed Mann–Whitney test was used to test for statistical significance (n = 3 except for siSCD‐2 has n = 2).

Data information: Data represent mean ± SEM. *P < 0.05.

Figure 4. A set of growth‐promoting genes are sensitive to tRNA^Tyr _GUA depletion.

1. Cells depleted of tRNA^Tyr _GUA or YARS were processed for label free quantitation by mass spectrometry to identify proteins that were reduced by a log₂‐fold change of 0.5 or more. This set was overlapped with proteins containing a higher than median abundance of Tyr codon content to identify candidate mediators of the pleiotropic effects of tRNA^Tyr _GUA depletion.
2. GO functional analysis of the 109 candidate gene‐set from (A).
3. Quantitative Western blot validation depicting abundances of protein targets (EPCAM, SCD, and USP3) identified from (A). HSC70 was used as a loading control and is not modulated upon molecular perturbation of tRNA^Tyr _GUA.
4. Quantification of Western blot analysis in (C) (n = 4). A one‐tailed Mann–Whitney test was used to test for statistical significance.
5. Growth curves for MCF10A cells were transfected with either control siRNA or two independent siRNA targeting EPCAM, SCD, or USP3. Note that the control cell growth curve is the same in all graphs and were plotted separately for clarity and does not represent independent experiments. Two‐way ANOVA was used to test for significance (n = 3).

Data information: Data represent mean ± SEM. *P < 0.05, **P < 0.01.

Source data are available online for this figure.


We selected a small set of tRNA^Tyr _GUA‐regulated genes that exhibited some of the greatest fold reductions upon tRNA^Tyr _GUA depletion for further functional studies (Fig EV4A and B). These genes were ubiquitin specific protease 3 (USP3), epithelial cell adhesion molecule (EPCAM), and stearoyl‐CoA desaturase (SCD). Quantitative Western blotting and quantitative real time PCR (qRT–PCR) revealed significant reductions in the endogenous levels of these growth‐related proteins and mRNA transcripts upon shRNA‐mediated depletion of tRNA^Tyr _GUA or YARS (Figs 4C and D, and EV4C). In contrast, the control protein HSC70 was not significantly depleted, consistent with our proteomic findings of a subset of proteins being modulated upon tRNA^Tyr _GUA depletion. Knockdown of each of these genes by two independent hairpins repressed the growth of MCF10A cells—consistent with growth‐promoting roles (Figs 4E and EV4D). Our results reveal that repressing the function of a specific tRNA by depleting it or inhibiting its aminoacylation, and thus its use in translation, represses expression of a set of tyrosine‐enriched proteins enriched in growth‐dependent processes, which functionally promote growth. Our findings suggest that this stress‐induced tRNA fragmentation and depletion constitutes an adaptive growth‐suppressive stress response.

TRNA^Tyr _GUA depletion impairs protein translation in a codon‐dependent manner

To test if these tRNA^Tyr _GUA‐regulated proteins become repressed upon oxidative stress, we performed quantitative Western blotting 24 h after H₂O₂ treatment, a time point when tRNA^Tyr _GUA is depleted. We noted significant reductions in protein levels of these tRNA^Tyr _GUA‐regulated genes (Fig 5A and B). Despite these reductions at the protein level, the transcript levels of two of these three downstream genes were not significantly altered upon H₂O₂ treatment (Fig EV5A). Observed changes in mRNA expression levels described above with tRNA or YARS depletion may be dependent on the steady‐state effects of these experiments (Fig EV4C). In comparison, the differences seen immediately after H₂O₂ stress represents a different experimental context with fewer changes expected at the mRNA level. The finding that one downstream gene (EPCAM) is altered both at the transcript and protein levels may signify lack of translational regulation for this gene or that translational repression occurs and is accompanied by mRNA decay. Our findings are consistent with translational repression of two specific growth‐regulatory genes—USP3 and SCD—upon tRNA^Tyr _GUA depletion.

Figure 5. Global ribosome occupancy analysis from tRNA^Tyr _GUA‐depleted cells reveals reduced translation efficiency for Tyr‐enriched genes.

1. Quantitative Western blot EPCAM, SCD, and USP3 in MCF10A cells 24 h after treatment with H₂O₂ (200 μM). HSC70 was used as a loading control.
2. Quantification of Western results in (A) (n = 9). A one‐tailed Mann–Whitney test was used to test for statistical significance.
3. A schematic of the codon‐based USP3 reporter. A Myc‐tagged WT or mutant reporter with 5 Tyr codons mutated to Ala codons were cloned upstream of a Luciferase used for transfection normalization.
4. Quantitative Western blot for the Myc‐tag and Luciferase (top) with normalized fluorescent intensities (bottom) are shown (n = 3).
5. Ribosome occupancy of 20‐22nt RPFs in tRNA^Tyr _GUA‐depleted cells compared to control cells reveal greater occupancy at both tyrosine codons in tRNA^Tyr _GUA‐depleted cells using a Wilcoxon test.
6. Genes were sorted based on their changes in GC‐corrected translation efficiency (TE) values, with reduced TE in tRNA^Tyr _GUA‐depleted cells shown in left and enhanced TE shown on right. The red bars over each column depict the range of values in that bin. Distribution of genes with high tyrosine codon content across these three bins were assessed using mutual information calculation and testing (see Materials and Methods for details). For visualization, we used the hypergeometric distribution to assign P‐values to the overlap between tyrosine‐rich genes and each bin. We then defined an enrichment score as –log of p‐value, if there was a significant enrichment. If the overlap is significantly fewer than expected by chance, log of p‐value is used instead (depletion). The resulting enrichment score is then shown as a heatmap with gold depicting positive enrichment.

Data information: Data represent mean ± SEM. *P < 0.05 and ***P < 0.001.

Source data are available online for this figure.


Figure EV5. Proteomic and ribosomal profiling validation of tRNA^Tyr _GUA‐depleted cells.

1. MRNA expression levels for target genes 24 h after treatment with H₂O₂ (200 μM) as measured by qRT–PCR (n = 9). A one‐tailed Mann–Whitney test was used to test for statistical significance.
2. Examples of the mapped position of the 5′‐end of reads near the start (top) or stop (bottom) codons are shown, revealing the characteristic 3‐nucleotide periodicity previously described by (Ingolia et al, 2009).
3. Histogram of the read length distribution of RPFs observed upon ribosome profiling sequence analysis.
4. Codon usage scatter plots of the ribosomal A‐site of shTyr vs shControl from replicate ribosomal profiling experiments. The five most affected codons are labeled with UAC (Tyr) being the most affected (red arrow). Gray outline illustrates the 95% confidence interval around the regression line.

Data information: Data represent mean ± SEM. ***P < 0.001.

We next employed a codon‐dependent reporter of an endogenous target of tRNA^Tyr _GUA depletion. A Myc‐tagged USP3 coding sequence was cloned upstream of Luciferase, which acted as a transfection control. The cloned USP3 sequence was either wild‐type (WT) or a mutant variant, which contained 5 tyrosine codons mutated to alanine codons (Fig 5C). These reporters were transfected into MCF10A cells and 24 h after H₂O₂ treatment, quantitative Western blotting was performed. Consistent with our model of a codon‐dependent regulation at the level of translation, we noted a significant increase in the abundance of the mutant USP3 relative to the WT version of the protein (Fig 5D).

To directly test if ribosomal engagement of tyrosine codon‐enriched transcripts is impaired upon tRNA^Tyr _GUA depletion, we performed ribosomal profiling in control and tRNA^Tyr _GUA‐depleted cells (Fig EV5B and C) (Ingolia et al, 2009). We compared the ribosome protected fragments (RPFs) detected in cells with and without tRNA^Tyr _GUA depletion, in order to examine the global translational effects due to modulating this single tRNA (Fig EV5B and C) (Ingolia et al, 2009; Lareau et al, 2014; McGlincy & Ingolia, 2017). Analysis of the 20–22 nt RPFs (Dunn & Weissman, 2016), which represent when the A‐site of the ribosome is empty (Wu et al, 2019), showed an increase of reads containing either UAC or UAU codons that encode for tyrosine in tRNA^Tyr _GUA‐depleted cells compared to control cells (Fig 5E). Consistent with previous reports for non‐optimal or rare tRNAs, we observe ribosomal slowing or stalling upon decoding of tyrosine codons at the ribosomal A‐site (Lauria et al, 2018) when the cognate tRNA becomes reduced (Fig EV5D). Furthermore, a corrected ribosome occupancy score was calculated for each gene as a metric for active translation in control and tRNA^Tyr _GUA‐depleted cells. Genes with distinct translation efficiencies, defined as the ratio between RPFs and mRNA reads, were separated into three equally populated gene sets. Genes with higher tyrosine codons were significantly enriched in the set of genes with the lowest translational efficiency (denoted by the lowest red bar) upon tRNA^Tyr _GUA depletion (Fig 5F). These findings reveal that ribosomal translation of a set of tyrosine codon‐enriched genes is sensitive to depletion of tRNA^Tyr _GUA to physiologically relevant levels.

Tyrosine tRF interacts with the RNA‐binding proteins hnRNPA1 and SSB

In addition to causing tRNA^Tyr _GUA depletion, oxidative stress also induced generation of tRF^Tyr _GUA (Figs 1B and 2A). Significant tRF^Tyr _GUA induction was observed at 5 min, remained elevated for up to 8 h, and declined to near baseline levels at 24 h post‐H₂O₂ exposure (Fig 6A and B). Previous studies have implicated tRFs in multiple biological processes including proliferation, cell invasion, translation, trans‐generational inheritance, and cancer metastasis (Goodarzi et al, 2015; Honda et al, 2015; Chen et al, 2016; Sharma et al, 2016; Keam et al, 2017; Kim et al, 2017). Sequencing of the tRF^Tyr _GUA after gel extraction identified tRFs with a 5′ leader sequence from nearly every tRNA^Tyr _GUA genomic locus (Appendix Fig S1A and B). In order to test whether pre‐tRNA splicing machinery promotes tRF^Tyr _GUA formation, we used RNAi‐mediated knockdown for TSEN2, the catalytic subunit of the tRNA splicing complex (Paushkin et al, 2004; Hopper & Nostramo, 2019), as well as for ANG, a RNase previously described to cleave certain tRNAs at the anticodon loop (Fu et al, 2009). Depletion of neither ANG nor TSEN2 impaired tRF^Tyr _GUA formation, suggesting that these ribonucleases are not mediating this oxidative stress‐induced response (Appendix Fig S1C–E). Further evidence suggesting that a defect in tRNA splicing was not the source of the tRF induction was that equal levels of tRFs were observed following RNAi‐mediated depletion of the RNA kinase CLP1 (Appendix Fig S1F and G), a component of the tRNA splicing complex (Weitzer & Martinez, 2007).

Figure 6. Identification of proteins that interact with tRF^Tyr _GUA .

• A, B
  Quantification of tRF^Tyr _GUA induction in response to oxidative stress as a function of time in MCF10A (A) (n = 4) and in HBEC30 (B) (n = 6) with mean ± SEM shown for each cell line.
• C
  Volcano plot of mass spectrometry results from a synthetic 5′‐biotinylated tRF^Tyr _GUA co‐precipitation experiment with cell lysate. Log₂ fold enrichment values of proteins identified from tRF^Tyr _GUA relative to scrambled tRF control samples.
• D
  Western blot validation of mass spectrometry results for three of the top hits in (A), showing co‐precipitation of endogenous proteins with transfected tRF^Tyr _GUA relative to a scrambled control sequence (Scr).

Source data are available online for this figure.


Using the conserved region from the different tRF^Tyr _GUA sequences (Appendix Fig S1B), we next transfected a 37nt synthetic mimetic as a means of eliciting gain‐of‐function. TRF^Tyr _GUA transfection did not significantly alter protein levels of USP3, SCD, or EPCAM and did not impact growth (Appendix Fig S1H and I), consistent with tRF^Tyr _GUA playing a regulatory role independent of the tRNA^Tyr _GUA‐mediated response. We hypothesized that tRF^Tyr _GUA may interact in trans with an RBP, as has been observed for other tRFs (Couvillion et al, 2010; Haussecker et al, 2010; Goodarzi et al, 2015). To test this, we used a synthetic 5′‐biotinylated tRF^Tyr _GUA as bait in an in vitro co‐precipitation experiment. Proteins interacting with the mimetic from cellular lysate were identified by in‐solution digestion and mass spectrometry, and compared to proteins interacting with a scrambled mimetic. Mass spectrometry identified numerous proteins that were enriched in the tRF^Tyr _GUA co‐precipitation relative to a control oligonucleotide (Fig 6C). We selected the most significantly enriched proteins—hnRNPA0, hnRNPA1, and SSB—and validated their interactions with synthetic tRF^Tyr _GUA relative to scrambled control by Western blot (Fig 6D). These results suggest that tRF^Tyr _GUA may interact with one or more RBPs.

Endogenous tRF^Tyr _GUA interacts with endogenous hnRNPA1 and SSB

UV‐crosslinking enables assessment of direct endogenous interactions between RBPs and their cellular RNA substrates (Ule et al, 2003; Mili & Steitz, 2004) and has been coupled with deep‐sequencing methods such as HITS‐CLIP or PAR‐CLIP to identify the landscape of RNAs that interact with a given RBP (Licatalosi et al, 2008; Hafner et al, 2010). Such experiments have previously been done with Argonaute‐2, which binds microRNAs (Chi et al, 2009), and YBX1, which binds tRFs (Goodarzi et al, 2015). To determine if any of the candidate RBPs identified by mass spectrometry interact with endogenous small RNA (smRNA) populations, we included an experimental condition where the HITS‐CLIP protocol was conducted in the absence of RNase digestion to ensure that any potential small RNA‐RBP bands visualized were not a consequence of, or confounded by, RNase digestion (Fig 7A). UV‐crosslinked immunoprecipitation followed by SDS–PAGE in the presence or absence of RNase digestion revealed that endogenous hnRNPA1 and SSB interacted with an endogenous small RNA population (Appendix Fig S2A and B). We did not observe this small RNA‐RBP band for hnRNPA0 (Appendix Fig S2C). This RBP may not significantly interact with a smRNA population in vivo or our method is not conducive to identifying this interaction.

Figure 7. Functional effects of interactions between hnRNPA1 and tRF^Tyr _GUA .

1. Schematic depicting the expected visualization of a crosslinked immunoprecipitation by autoradiogram. Bands corresponding to smRNA‐RBP interactions (no RNase digestion) or a smear representing mRNA‐RBP interactions (low RNase digestion) by autoradiogram were processed for HITS‐CLIP.
2. IGV plots from an hnRNPA1 HITS‐CLIP (Huelga et al, 2012) reveals interactions with tRF^Tyr _GUA.
3. IGV plots representing SSB interacting with the tRF^Tyr _GUA in samples that were treated with low levels of RNase digestion. SSB bound tRF^Tyr _GUA reads mapped to multiple loci encoding tRNA^Tyr _GUA.
4. Similar to the IGV plots shown in (C), but depicting SSB interactions with tRF^Tyr _GUA loci in samples without RNase digestion.
5. A cumulative distribution in control and hnRNPA1 depleted cells of the stability levels for mRNA transcripts with 3′ UTR hnRNPA1 CLIP binding (Huelga et al, 2012). Transfection of tRF^Tyr _GUA led to a significant right‐shift in the expression levels of 3′ UTR bound hnRNPA1 transcripts. Statistical significance was measured using the Kolmogorov–Smirnov test.
6. Cumulative distribution as in (E). Transfection of locked nucleic acid against tRF^Tyr _GUA and treatment with 200 µM H₂O₂ led to a significant left‐shift in mRNA stability of 3′UTR bound hnRNPA1 transcripts. Statistical significance was assessed using the Kolmogorov–Smirnov test.
7. Model of tRNA^Tyr _GUA‐dependent gene regulatory response to oxidative stress.

We next sought to define the identities of the smRNAs bound by hnRNPA1 and SSB. An analysis of tRFs bound in a previously published CLIP‐seq study for hnRNPA1 (Huelga et al, 2012) validated our observations by revealing a reciprocal interaction between hnRNPA1 and tRF^Tyr _GUA (Fig 7B). Studies have previously described roles for hnNRPA1 in gene expression regulation, including as a regulator of splicing and mRNA stability via binding to 3′ UTRs (Glisovic et al, 2003; Wang et al, 2016). HnRNPA1 has also been implicated in promoting growth and multiple cancer progression phenotypes (Biamonti et al, 1993; Roy et al, 2017). We next conducted HITS‐CLIP for endogenous SSB with and without RNase digestion. SSB, also known as La, is a well characterized RNA‐binding protein known to bind the nascent 3′ ends of Pol III transcripts, including those of full‐length pre‐tRNAs (Gottlieb & Steitz, 1989; Maraia et al, 1994; Yoo & Wolin, 1997). We found that consistent with its previously described canonical role in binding Pol III transcripts, alignment and analysis of sequencing reads revealed binding of SSB to the 3′‐trailers of pre‐tRNAs (Appendix Fig S2D). Importantly, in addition to this previously described binding, we observed previously unreported interactions of SSB with the 5′ half of tRNA^Tyr _GUA arising from multiple distinct loci with and without RNase treatment (Fig 7C). The 5′ leader containing reads likely represent intermediates in the pre‐tRNA^Tyr _GUA processing reaction (Fig 7C and D) (Hanada et al, 2013). The abundant number of reads mapping to the 5′ regions of tRNA^Tyr _GUA distinguish these SSB‐tRF interactions from the previously described canonical SSB interactions with 3′‐trailers of full‐length pre‐tRNAs. This notion is further supported by our observations of SSB binding to the 5′ tRF without 3′‐trailer binding even in the absence of RNase digestion.

We next investigated if the stress‐induced tRF^Tyr _GUA might regulate the activity of hnRNPA1 or SSB as a trans factor. Using the 3′ UTR targets of hnRNPA1 previously identified by CLIP‐seq (Huelga et al, 2012), we assessed whether tRF^Tyr _GUA impacts hnRNPA1‐mediated mRNA stabilization. Following transfection of the tRF^Tyr _GUA mimetic or a scrambled control, we observed impaired stabilization of hnRNPA1 targets in an RBP‐specific manner (Fig 7E). Conversely, locked nucleic acid mediated inhibition of tRF^Tyr _GUA or a control had the opposite effect in increasing stability of hnRNPA1 mRNA targets in an hnRNPA1 specific manner (Fig 7F). These results are consistent with known reports of hnRNPA1 binding to 3′ UTRs to affect mRNA stability (Glisovic et al, 2003; Wang et al, 2016). Taken together, our results are consistent with tRF^Tyr _GUA competing with hnRNPA1 target transcripts for binding to endogenous hnRNPA1. TRF^Tyr _GUA‐dependent reduction of hnRNPA1 binding to its 3′ UTR targets decreased stability of hnRNPA1 transcripts during α‐amanitin‐stability analyses. Transfection of the tRF^Tyr _GUA mimetic did not impact expression of Pol III transcriptional targets, suggesting that the canonical nuclear role for SSB was unaffected by increased levels of the tRF^Tyr _GUA mimetic in this context (Appendix Fig S2E). Our findings describe that stress‐induced fragmentation can cause a specific tRNA to become depleted, resulting in translational consequences. This process concomitantly gives rise to a tRNA fragment that can interact in trans and functionally impact the regulon of an RNA‐binding protein known to promote growth and cancer progression phenotypes (Fig 7G).

TRNA^Tyr _GUA fragment generation is DIS3L2 dependent

In an ongoing RNAi screen focused on identifying putative regulators of tRNA fragments in C. elegans, we observed that depletion of the exoribonuclease disl2—the ortholog of DIS3L2—a tumor suppressor in Wilms tumor, impaired stress‐induced tRNA fragment levels (unpublished results). Exposure of worms to H₂O₂ for a brief period (15 min), so as to avoid starvation effects, revealed that disl2‐depleted animals produced less stress‐induced tRNA^Tyr _GUA fragments relative to control animals (Appendix Fig S3A–C). A similar effect was observed in disl2 mutant (syb1033) worms upon oxidative stress exposure. Depletion or genetic deletion of DIS3L2 also impaired, but did not completely eliminate, oxidative stress‐induced tRNA^Tyr _GUA fragment levels in human cells (Appendix Fig S3D–F). DIS3L2 depletion selectively impaired tRF^Tyr _GUA formation upon oxidative stress and did not cause pre‐tRNA^Tyr _GUA accumulation, consistent with DIS3L2 mediating processing/maturation of a tRF^Tyr _GUA precursor rather than pre‐tRNA^Tyr _GUA cleavage. These findings implicate the DIS3L2 tumor suppressor in tRF^Tyr _GUA generation.

Discussion

Aberrant tRNA breakdown products were first detected nearly 40 years ago upon analyses of the urine of cancer patients (Gehrke et al, 1979). TRFs have been implicated in processes such as hematopoiesis (Guzzi et al, 2018), cancer pathogenesis (Lee et al, 2009; Goodarzi et al, 2015), and mediate effects via interactions with RNA‐binding proteins (Couvillion et al, 2010; Goodarzi et al, 2015) or messenger RNAs (Kim et al, 2017). Studies of specific tRFs in Saccharomyces cerevisiae and mammalian cells have revealed that the specific tRNA pools studied were not impacted upon tRNA fragmentation (Thompson & Parker, 2009b; Saikia et al, 2012). By comprehensively surveying global tRF generation and tRNA abundances, we have identified a specific tRNA, tRNA^Tyr _GUA, for which significant depletion and translational repression occurs during stress‐induced fragmentation.

Our findings reveal that in non‐malignant epithelial cells, there exists an oxidative stress response pathway that represses growth via selective tRNA depletion. The consequence of this response is repressed translation of a set of growth‐promoting genes. This tRNA depletion response likely cooperates with other cellular oxidative response pathways (Martindale & Holbrook, 2002). While our work has interrogated the consequences of tRNA^Tyr _GUA depletion, future work is needed to determine the impact, if any, of depletions of other tRNAs that were observed to generate fragments. Two such examples are tRNA^Leu _UAA and tRNA^Leu _CAG. Interestingly, six isoacceptor tRNAs exist for leucine, we only observed significant concomitant tRF induction and tRNA depletion for these two. Potential tRNA fragmentation selectivity among isoacceptors may exist with potential for codon‐biased translational consequences.

While oxidative stress has been observed to repress global tRNA levels in E. coli (Zhong et al, 2015) and S. cerevisiae (Torrent et al, 2018), our results reveal that in mammalian cells, there is a selective tRNA and codon‐dependent response to oxidative stress. While we do not yet know the mechanistic basis for this selectivity, we speculate that the tRNAs that become modulated upon oxidative stress may be of lower relative abundance compared to other unaffected tRNAs or that various steps in tRNA processing might be targets of regulation. Additionally, the molecular mechanisms that govern translation in response to these tRNA changes require further study. The impact of individual variables such as tRNA abundance, location of ribosomes along the mRNA, adjacent codon pairings, or undiscovered interactions on translation remains to be studied. In humans, all tRNA^Tyr _GUA isodecoder genes have introns that require splicing to generate mature tRNAs. Our observed tRNA^Tyr _GUA‐dependent response initiates at the pre‐tRNA^Tyr _GUA level, giving rise to fragments.

Though our data indicate that tRNA splicing is not the main source of tRF induction, we cannot rule out the possibility that additional currently unknown RNases might have a role in the tRNA splicing complex. If the known tRNA splicing complex is involved, conditional genetic inactivation of components of the tRNA splicing complex or TSEN2 might be necessary to completely blunt tRF induction upon oxidative stress. Aberrantly generated tRFs derived from pre‐tRNA introns have been associated with human neuronal degeneration (Karaca et al, 2014; Schaffer et al, 2014). Furthermore, oxidative stress has been implicated or associated with a myriad of human diseases including Parkinson’s, ALS, Alzheimer’s, cardiovascular disease, and cancer pathogenesis (Lin & Beal, 2006; Piskounova et al, 2015; Cervantes Gracia et al, 2017). Exposure of mouse neurons and MEFs to oxidative stress was previously shown to elicit fragmentation of tRNAs, including tyrosine tRNA (Hanada et al, 2013). Sequencing analysis revealed generation of fragments from tyrosine pre‐tRNA—similar to our observations. This past work, however, did not assess the subsequent impact of tRF formation on pre‐tRNA or tRNA abundances in the neural cells under study. It would be of interest to determine if oxidative stress‐induced fragmentation could also functionally deplete the corresponding tRNA pool in non‐mitotic neurons in a manner similar to that we observed in mitotic epithelial cells in this study.

Our findings raise a number of questions for future study. Firstly, could tRNA^Tyr _GUA fragmentation and depletion occur during normal development and physiology to mediate growth repression? Oxidative species are generated during cellular respiration and may endogenously activate this response. Moreover, our identification of the growth‐associated gene DIS3L2 as an exoribonuclease involved in tRNA^Tyr _GUA fragmentation across multiple species suggests that this tRNA depletion response may perhaps also play a role in these other contexts. Recent elegant genetic studies revealed that the previously assumed direct substrate of DIS3L2 is not let‐7, suggesting that another RNA substrate may mediate the effects of DIS3L2 on tumorigenesis and growth (Hunter et al, 2018). Future work is needed to interrogate the role of this tRNA fragmentation response in the pathogenesis of this disorder. More broadly, our studies of tRNA modulation by oxidative stress raise the possibility that additional stresses may generate similar or distinct tRNA fragmentation and depletion effects, which may mediate coherent gene regulatory responses.

Materials and Methods

Cell culture

MCF10A cells were cultured in DMEM/F12 media supplemented with 5% horse serum and final concentrations of 20 ng/ml of EGF, 0.5 mg/ml of hydrocortisone, 100 ng/ml of cholera toxin, and 10 μg/ml of insulin. HBEC30 cells were cultured in keratinocyte‐SFM media with the included supplements of BPE and EGF. All cell lines were STR tested and routinely tested for mycoplasma contamination. Oxidative stress was induced with 200 μM H₂O₂ or 50 μM of menadione (Sigma) dissolved in DMSO. RNA was then isolated by TRIzol and isopropanol precipitation as described below.

Stable cell line generation

Lentivirus was produced in 293T cells. 3 μg of each packaging vector (pRSV‐Rev, pCMV‐VSVG‐G, and pCgpV) were transfected with 9 μg of the appropriate hairpin in a pLKO‐backbone vector using 30 μl of Lipofectamine 2000 (Invitrogen). After 24 h, the media were replaced and virus‐containing supernatant was harvested 48 h after transfection. Supernatant was filtered through a 0.45 μm filter before transducing cells with 8 μg/ml of polybrene. Antibiotic selection with 1 μg/ml of puromycin was started 48 h after transduction. Targeted sequences used for stable RNAi cell lines were: TTATGCCACAGTCCTTATAAT (YARS sh2), GTTCATCAAAGGCACTGATTA (YARS sh3), and TGGTAGAGCGGAGGACTGTAGA (tRNA^Tyr _GUA). Cells were selected for 3–5 days with non‐transduced cells as a control. Knockdown of mRNA and protein were validated by qRT–PCR and Western blot, respectively, while knockdown of tRNA was validated by Northern blot.

RNA isolation and purification

RNA was extracted from cells using TRIzol and isopropanol precipitation at −20°C overnight. After centrifugation at max speed (~21,000 g) in a refrigerated tabletop centrifuge, the RNA pellet was washed twice with ice‐cold 75% EtOH before being resuspended in RNase free water or TE‐buffer.

Northern blotting

Purified RNA was run on a 10% Urea‐PAGE gel before being transferred onto a nylon membrane and UV‐crosslinked (240 mJ/cm²). The membrane was pre‐hybridized in UltraHyb‐Oligo buffer (Ambion) at 42°C. DNA oligos were radiolabeled with [γ‐³²P]ATP using T4 PNK (NEB) and further purified by G‐25/G‐50 columns before incubating with the blot overnight. After hybridization, the blot was washed twice with SSC and SDS buffers. Probes that were ³²P labeled and used for detection were: ACAGTCCTCCGCTCTACCAGCTGA (tRNA^Tyr _GUA), ACAGTCCTCCGCTCTACCAACTGA (C. elegans tRNA^Tyr _GUA), CAGCGCCTTAGACCGCTCGGCCA (tRNA^Leu _HAG), AACGCAGAGTACTAACCACTATACG (tRNA^His _GUG), GCGCCGAATCCTAACCACTAGACCA (tRNA^Glu _YUC), CACGAATTTGCGTGTCATCCTT (U6), and CAAATTATGCAGTCGAGTTTCCCACATTTG (U1). Quantification was done using FIJI (ImageJ) where the intensity of each band over background was measured and normalized to U6 levels.

Western blotting

Cells were lysed in ice‐cold RIPA buffer containing a protease inhibitor cocktail (Roche). Cellular debris was cleared at max speed in a refrigerated tabletop centrifuge. Samples were heated with LDS buffer and reducing agent before running on an SDS–PAGE gel and transferred onto a PVDF membrane (Bio‐Rad). The membranes were blocked and then probed using target‐specific antibodies. Antibodies used were YARS (Abcam, ab150429, RRID: AB_2744675, 1:1,000), EPCAM (Proteintech, 21050‐1‐AP, RRID: AB_10693684, 1:1,000), SCD (Proteintech, 23393‐1‐AP, RRID: AB_2744674, 1:1,000) USP3 (Proteintech, 12490‐1‐AP, RRID: AB_10639042, 1:1,000), SSB (MBL, RN074PW, RRID: AB_11124309, 1:1,000), hnRNPA1 (Santa Cruz, sc‐32301, RRID: AB_627729, 1:1,000), hnRNPA0 (Bethyl Laboratories, A303‐941A, RRID: AB_2620290, 1:1,000), DIS3L2 (Novus biologicals, NBP2‐38264), Myc (Cell Signaling, 2276S, RRID: AB_331783), Luciferase (Proteintech, 27986‐1‐AP, RRID:AB_2750646), and HSC70 (Santa Cruz, sc‐7298, RRID: AB_627761, 1:2,000). Chemiluminescent signal was detected by HRP‐conjugated secondary antibodies, ECL Western blotting substrate (Pierce), and the SRX‐101A (Konica Minolta) developer according to manufacturer’s instructions. Membranes were stripped (Restore Western blot stripping buffer, Pierce), blocked, re‐probed, and re‐developed if necessary.

Quantitative Western blotting

The Odyssey® quantitative Western blotting system (LI‐COR) was used to compare protein expression levels. This method is identical to Western blotting except membranes were blocked in Odyssey® blocking buffer (PBS) and the secondary antibody used was a species‐specific fluorescent IRDye®. Membranes were then imaged using the Odyssey® Sa Infrared Imaging System. Image quantification was done using Image Studio™ Lite and an unpaired t‐test was used to determine statistical significance.

Codon reporter assays

A Myc‐tagged construct was synthesized (Genewiz) and cloned into the psiCHECK2 vector (Promega), replacing the synthetic Renilla Luciferase gene. The construct contained a 6x‐glycine linker placed between the Myc‐tag and the gene. The constructs included a wild‐type USP3 (NM_006537) or a USP3 gene with 5 tyrosines mutated to alanines. Multiple mutant constructs were cloned but only those with high levels of expression were chosen for testing. MCF10A cells were seeded to be 70% and exposed to 200 μM H₂O₂ or a control for 1 h before being transfected with the reporters using Lipofectamine 3000 (Invitrogen). Cells were lysed and compared through quantitative Western blotting 24 h post‐H₂O₂ treatment.

Transient overexpression of tRNA^Tyr _GUA

Overexpression constructs for tRNA^Tyr _GUA were made as previously reported by (Mattijssen et al, 2017). Briefly, 3 copies of tRNA^Tyr _GUA, per plasmid, were synthesized (Genewiz) into the pUC57‐KAN vector. Each copy had 150nt upstream and 90nt downstream genome sequences included but did not include the intron sequence. This vector or an empty pUC57‐KAN vector were transiently transfected using Lipofectamine 3000 (Invitrogen).

Transfer RNA profiling

MCF10A cells were 50% confluent before addition of 200 μM H₂O₂ for 8 or 24 h. RNA was isolated using the mirVana miRNA isolation kit. Transfer RNA profiling was done as described by (Goodarzi et al, 2016). Briefly, biotinylated probe‐pairs against nuclear encoded tRNAs were designed to hybridize to each half of the tRNA. A nick at the anticodon end of the DNA‐RNA hybrid was filled using T4 DNA ligase and SplintR ligase (NEB). MyOne‐C1 Streptavidin Dynabeads (Invitrogen) were used to purify the DNA‐RNA hybrids and ligated probes were eluted following RNase H and RNase A incubation. Eluted probes were PCR amplified and high‐throughput sequenced.

For computational analysis, fastq files were aligned to tRNA probe sequences using bowtie2. Reads were further sorted, indexed, and counts were generated with samtools. Raw counts were imported into R V3.4.1 and normalized with EdgeR. Linear regression tests were used to assess tRNAs that were significantly depleted at 8 and 24 h conditions relative to control conditions.

Whole‐genome ribosomal occupancy profiling

This procedure was conducted as described by (McGlincy & Ingolia, 2017). Briefly, cells were washed and flash frozen with liquid N₂ before being lysed with lysis buffer containing cycloheximide (Alfa Aesar). Lysate was digested with RNase I (Lucigen) before ribosomes were isolated through ultracentrifugation through a sucrose cushion. The ribosome pellet was resuspended in a solubilization buffer (0.5% SDS, 1 mM EDTA, and TRIzol) before RNA was extracted using the Direct‐zol kit (Zymo Research). RNA was separated on a 15% TBE‐Urea gel. RNA between 17nt and 34nt were gel extracted. Barcoded pre‐adenylated linkers were ligated using T4 RNA ligase 2 truncated K227Q (NEB) and rRNA was depleted using the Ribo‐Zero gold kit (Illumina) according to the manufacturer’s protocol. RNA was converted to cDNA using SuperScript III (Life Technologies) and the RT product was circularized by CircLigase II (Epicentre). A PCR library was amplified and sequenced using Illumina Nextseq 500 at the Rockefeller University Genomics Center.

For the ribosome footprinting data, reads were first subjected to linker removal, quality trimming (cutadapt v1.17), and then distributed among the samples based on their assigned barcodes using fastx_barcode_splitter (using ‐‐eol and ‐‐mismatches 1). Reads were then collapsed and UMIs were extracted in two steps (2 at the 5′ end and 5 at the 3′ end) using UMI Tools. The reads were then aligned against a reference database of rRNAs and tRNAs as to remove contaminants (bowtie 2.3.4.1). STAR was then used to align the remaining reads to the human transcriptome (build hg38). PCR duplicates were removed using UMI Tools. Xtail (Xiao et al, 2016) was used to count RPFs, estimate translation efficiency, and perform statistical comparisons. For RNA‐seq data analysis, reads were first subjected to quality trimming and adapter removal. STAR (v2.5.2a) was used to align the reads to the human transcriptome (hg38). The number of reads mapping to each gene was counted using htseq‐count.

Using the tRF^Tyr _GUA mimetic to identify interacting proteins

A 5′ biotinylated 37‐nucleotide tRF^Tyr _GUA mimetic (CCUUCGAUAGCUCAGCTGGUAGAGCGGAGGACUGUAG) and a scrambled control oligo (GAGACCAGGGUACGCAAUCGAGUUGUUGGGCACUCUG) were synthesized (IDT). Each mimetic was incubated at 4°C with equal amounts of MCF10A cell lysate containing protease inhibitor that had been pre‐cleared for debris and by the control oligo. Proteins bound to our mimetic were co‐precipitated by MyOne‐C1 Streptavidin Dynabeads (Invitrogen) and washed twice with a low salt wash buffer (50 mM Tris–HCl pH 7.5, 150 mM NaCl) and twice with a high salt wash buffer (50 mM Tris–HCl pH 7.5, 400 mM NaCl) before being submitted to the Rockefeller University Proteomics Resource Center.

RNA stability of hnRNPA1 bound transcripts

MCF10A cells were reverse transfected with a control siRNA or siRNA targeting hnRNPA1 with either a tRF^Tyr _GUA mimetic or a scrambled mimetic using Lipofectamine RNAimax (Invitrogen). For the converse experiment, siRNA was reverse transfected as above into MCF10A cells using Lipofectamine RNAimax. A miRNA inhibitor (LNA) targeting tRF^Tyr _GUA (GAUAGCUCAGUUGGUAGAGCGGAGGA) or non‐targeting control (UCGUUAAUCGGCUAUAAUACGC) was directly transfected 24 h later. After 48 h, half of the samples were treated with α‐amanitin (10 mg/ml, Sigma) and incubated for 8 h while the other half were processed as the 0 h time point. RNA was extracted using TRIzol and RNA‐seq libraries were prepared using QuantSeq 3′ mRNA‐seq kit (Lexogen) or TruSeq RNA Library Prep (Illumina) following the manufacturers’ instructions.

SSB HITS‐CLIP with and without RNase

HITS‐CLIP for endogenous SSB was done as described by (Licatalosi et al, 2008) with the modifications previously used for YBX1 small RNA CLIP (Goodarzi et al, 2015). MDA‐MB‐231 cells were UV‐crosslinked at 400 mJ/cm² before cell lysis. Samples with and without RNase treatment were immunoprecipitated with an anti‐SSB antibody for protein‐RNA complexes. Polyphosphatase (Lucigen) was incubated with smRNA samples before ligation and PCR amplification with primers described by (Goodarzi et al, 2015). Constructed libraries were sequenced on the Illumina HiSeq2000 at the Rockefeller University Genomics Center.

Generation of CRISPRi‐DIS3L2 cell lines

A non‐targeting control sgRNA (sequence from hCRISPRi v2 (Horlbeck et al, 2016)) and 5 sgRNAs targeting DIS3L2 (generated using the ChopChop tool (Labun et al, 2019)) were cloned into lentiGuide‐Puro (Addgene #52963). The individual sgRNAs were subsequently transduced into MCF10A cells stably expressing pHR‐SFFV‐dCas9‐BFP‐KRAB (Addgene #46911), and the cells expressing the sgRNAs were enriched for using 1.2ug/ml puromycin selection. Cells with the strongest knockdown of DIS3L2 (>90%) by Western blot were subsequently used for experiments, with the following guide sequences: sgRNA‐NC1: GCGTACGACAATACGCGCGA, sgRNA‐DIS3L2‐A: CGCGGCGTTCTAGAGAGCGA.

Statistical analysis

Statistical testing for experiments comparing blot intensities used a Mann–Whitney test while cumulative distribution plots were compared using a Kolmogorov–Smirnov test. Data shown as mean ± SEM with *P < 0.05, **P < 0.01, and ***P < 0.001.

Author contributions

SFT conceived the project and supervised all research. SFT and DH wrote the manuscript. DH, MCP, JG, SD, CG, LF, AP, HM, ZR, EAM, HA, and HG designed, performed, and analyzed the experiments.

Conflict of interest

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Expanded View Figures PDF

Table EV1

Review Process File

Source Data for Figure 4

Source Data for Figure 5

Source Data for Figure 6

The EMBO Journal (2021) 40: e106696.

Data availability

The data for high‐throughput sequencing experiments are deposited at GEO under the accession number GSE120385 GSE120385.