Toward a comprehensive modification landscape of yeast mitochondrial tRNAs using Nanopore direct RNA sequencing and dihydrouridine sequencing

Abstract

Saccharomyces cerevisiae is an invaluable model in the study of mitochondrial tRNA biology. Yet the positions of modified bases in all yeast mitochondrially encoded tRNAs (mt-tRNAs) are still not fully mapped. We performed Nanopore direct RNA sequencing (DRS) on tRNAs from the crude mitochondrial fraction of yeast to map base modifications across all 24 mt-tRNA isoacceptors. Additionally, we adapted the “D-seq” method to detect dihydrouridine sites in tRNAs, where chemical reduction of dihydrouridine causes disruptions to reverse transcription. We mapped dihydrouridine, pseudouridine, and N2-dimethylguanosine sites in mt-tRNAs using DRS, tRNA-D-seq, and knockouts of five conserved tRNA-modifying enzymes. Our results establish Dus1 and Dus2 as the enzymes responsible for D₁₄, D₁₆, D₁₇, D_17a, and D₂₀ formation in S. cerevisiae mt-tRNAs. We provide evidence of interactions between Dus1, Dus2, and Trm1-catalyzed modifications, and the influence of Ψ₅₅ promoting m⁵U₅₄ in mt-tRNAs. These findings expand our understanding of mt-tRNA base modifications and their interdependence, and advance opportunities for the yeast model to investigate defects in human mt-tRNA function.

Graphical Abstract

Graphical Abstract.

Introduction

Eukaryotic organisms contain a mitochondrial genome in the form of circular dsDNA molecules found in the mitochondrial matrix. The yeast mitochondrial genome is a highly reduced derivative of its prokaryotic ancestor [1], yet it retains genes involved in oxidative phosphorylation that are essential for respiration. These genes are transcribed, then translated within mitochondria by translation machinery that includes mitochondrially encoded rRNA and tRNA. Mitochondrial tRNAs (mt-tRNAs) are chemically modified, though typically to a lesser extent than cytosolic tRNAs [2]. These chemical modifications serve important roles in tRNA structure, stability, and promotion of translation fidelity (reviewed in Phizicky and Hopper [3]).

Pseudouridine (Ψ) and dihydrouridine (D) are the two most abundant tRNA base chemical modifications in nature. Ψ is an isomer of uridine that augments its hydrogen bonding and base stacking properties [4]. Pseudouridine synthases (Pus) catalyze the formation of Ψ at positions 27, 28, 31, 32, 38, 39, 55, and 72 in mt-tRNAs [5]. D is a chemically reduced form of uridine that alters the planarity of the base, conferring flexibility to the D-loop [6] and increasing the overall stability of tRNA tertiary structure [7]. While D has been found at positions 14, 16, 17, and 20 in mt-tRNAs [8], the enzymes responsible for D formation in the mitochondria have not been previously identified with site specificity.

Disruption to the chemical modifications of mt-tRNAs can lead to defects in cell growth and are responsible for several human diseases, referred to as “modopathies” [9, 10]. Comprehensive annotation of human mt-tRNA modifications was performed through analysis of all 22 individually purified human mt-tRNAs by LC-MS/MS [11], demonstrating an important approach to mapping base modifications, albeit with a high technical benchmark. This procedure is less amenable to investigating the impact of multiple cell or enzyme perturbations on modifications simultaneously across the majority of isoacceptors, highlighting the need for complementary approaches.

The budding yeast Saccharomyces cerevisiae has been a powerful model system for investigating the enzymes that chemically modify mt-tRNA for several reasons: (i) yeast genetics make it straightforward to perturb modification enzymes; (ii) yeast can survive by fermentation, making mitochondrial gene expression non-essential under appropriate growth conditions; (iii) many of the tRNA modifications and the enzymes that catalyze them are conserved in human mitochondria; and (iv) yeast culture volumes are easily scaled in order to obtain sufficient material for analysis. Discoveries enabled by work in yeast have helped develop a mechanistic understanding of the basis of several human modopathies [12–15]. Nevertheless, characterization of yeast mt-tRNA modifications is incomplete, with the principal modification database, Modomics, providing modification profiles for only 16 of 24 yeast mt-tRNA isoacceptors [8]. A complete annotation of all mt-tRNA modifications in yeast will further our understanding of their role in mitochondrial function and disease.

Oxford Nanopore Technologies (ONT) direct RNA sequencing (DRS) of full length tRNAs is an approach to analyze tRNA chemical modifications. This measurement can provide position-level resolution at increased throughput compared to mass spectrometric analysis. As an RNA molecule transitions through a protein-based nanopore, its chemically modified bases can lead to distinct ionic current signals compared to unmodified bases. These ionic current changes can cause basecalling programs to artifactually assign a base to the modified position that does not match the “reference,” unmodified base. This outcome is referred to as a base “miscall,” and certain modifications can produce distinct miscall patterns. For example, Ψ typically causes a U-to-C miscall coincident with the modified position. Coupled with biochemical and/or genetic controls, miscalls are informative about the positions and identities of modified bases. These approaches have been used to study base modifications in conditions that affect the cellular growth state or RNA modifying enzyme activities [16–21]. Additionally, DRS can provide a systems-level understanding of modification interdependencies across dozens of tRNAs [20].

Using RNA from the crude mitochondrial fraction of budding yeast cell lysate, we analyzed yeast mt-tRNAs by DRS. We built our DRS-based sequence analysis on a revised structure-guided alignment of S. cerevisiae mt-tRNA sequences. Using the Modomics database as a reference, we found that DRS can detect nearly all known yeast mt-tRNA modifications, some supported using genetic mutants. We orthogonally validated all D sites in mt-tRNAs, using a tRNA-specific version of D-seq, tD-seq, which led to identification of new examples of modification interdependencies. For 5-methyluridine (m⁵U), a modification undetectable by either DRS or methods involving reverse transcriptase arrests, we employed LC-MS/MS quantification to measure the influence of an adjacent Ψ site. Together, this work advances the yeast model for investigating the mechanistic basis of human modopathies caused by defects in tRNA chemical modifications.

Materials and methods

Yeast growth conditions

See Supplementary Table S1 for yeast strains used in this study and their genotypes. Yeast strains were streaked out from glycerol stocks in the −80°C freezer onto Yeast extract, Peptone, and 2% Dextrose (YPD) plates and grown for 2 days at 30°C. Individual colonies were picked to inoculate 10 ml culture tubes containing 5 ml of YPD media. The liquid cultures were grown overnight on a roller drum wheel at 30°C. The next morning the cultures were diluted 1000-fold into 50 ml of YPD media in 250-ml Erlenmeyer flasks. The cultures were grown overnight on a shaking incubator at 300 rpm and 30°C. The yeast were centrifuged in 50 ml conical tubes at 3000 × g for 5 min at room temperature. The supernatant was removed, and the pellets were resuspended in 20 ml YPE (Yeast extract, Peptone, and 2% Ethanol) media and added to 980 ml YPE in 3-l Erlenmeyer flasks with a baffled base for increased aeration. For samples subjected to the crude mitochondrial isolation, cultures were grown on a shaking incubator at 300 rpm and 30°C for ∼36 h, until an OD₆₀₀ of 1.0–1.4 was reached. The yeast were centrifuged in 1000-ml Nalgene centrifuge bottles at 3000 × g for 5 min at room temperature. The supernatant was poured off, and the yeast pellets were resuspended in 250 ml of sterile Millipore water and centrifuged again at 3000 × g for 5 min at room temperature. The supernatant was removed, and the cells were resuspended in 45 ml of cold 1× PBS (phosphate buffered saline), and then transferred to 50-ml conical tubes. The resuspended cells were centrifuged at 3000 × g and 4°C for 5 min and the supernatant was decanted. The pellets/conical tubes were immediately submerged into liquid nitrogen to flash freeze, and then stored at −80°C. For samples used for the sucrose gradient purification of mitochondria, cultures were started as described above, but were grown to an OD₆₀₀ of ∼2.0, and the mitochondrial isolation was started promptly after harvesting the crude fraction, without freezing (see below).

Samples used for 2-bromoacrylamide-assisted cyclization sequencing (BACS) and primer extension assays were streaked out on YPD and grown overnight in 5 ml YPD as described above. Then, 2.5 ml of saturated culture was transferred to 15-ml conical tubes, centrifuged at 3000 × g for 5 min at room temperature, and the supernatant was decanted. Pelleted cells were resuspended in 5 ml YPE, which was then added to 45 ml YPE in 250 ml Erlenmeyer flasks that were placed on a shaker at 200 rpm, at 30°C until an OD₆₀₀ of ∼2.0 was reached. Cultures were transferred to 50-ml conical tubes and centrifuged at 3000 × g for 5 min at 4°C, and the supernatant was discarded. Cell pellets were resuspended in 10 ml cold 1× PBS, centrifuged again, and the supernatant was discarded. The cell pellets/conical tubes were immediately submerged into liquid nitrogen to flash freeze, then stored at −80°C.

For cells grown in galactose, yeast strains were streaked out from glycerol stocks in the −80°C freezer onto YPD plates and grown for 2 days at 30°C. Individual colonies were picked to inoculate 10-ml culture tubes containing 5 ml of YPD media. The liquid cultures were grown overnight on a roller drum wheel at 30°C. The next morning the cultures were diluted in 5 ml YP 2% galactose media to an OD₆₀₀ of 0.1. The cultures were grown for ∼6 h at 30°C on a roller drum wheel to an OD₆₀₀ of ∼0.8. The yeast was centrifuged in 15-ml conical tubes at 3000 × g and 4°C for 2 min. The supernatant was removed, and the yeast was resuspended in 1 ml of cold 1× PBS and transferred to microcentrifuge tubes. The tubes were then centrifuged at 9500 rpm for 1 min at 4°C. The supernatant was poured out, and the microcentrifuge tubes were immediately placed into liquid nitrogen to flash freeze yeast in log phase. Total RNA was isolated from the cells as described below, but without mitochondrial enrichment.

Crude mitochondrial fraction isolation

The crude mitochondrial fraction was isolated from yeast cells using differential centrifugation as previously described [22, 23]. The yeast pellets were thawed on ice, resuspended in 45 ml water, and centrifuged at 3000 × g at room temperature for 5 min. The supernatant was decanted, and the wet weight of each cell pellet was recorded. The cell pellets were resuspended in 2 ml of DTT buffer [100 mM Tris-H₂SO₄ (pH 9.4), 10 mM dithiothreitol] per gram (wet weight) of cells. The tubes were placed on a shaker at 70 rpm and 30°C for 20 min. The cells were centrifuged at 3000 × g, room temperature for 5 min. The cell pellets were resuspended in 7 ml Zymolyase buffer [20 mM potassium phosphate (pH 7.4), 1.2 M sorbitol)], without Zymolyase, per gram of cells. The cells were centrifuged (3000 × g, 5 min, 20°C), and the supernatant was decanted. The cells were resuspended in the same volume (7 ml/g cells) of Zymolyase buffer and transferred to a 250-ml Erlenmeyer flask. One milligram of Zymolyase 100T powder (United States Biological Corporation) per gram of cells was added to the cell suspension. The flasks were placed on a shaker at 70 rpm for 30 min at 30°C. The resulting mixture, which contained spheroplasts from cell wall digestion by Zymolyase, was transferred to 50-ml conical tubes and centrifuged (2200 × g, 8 min, 4°C). The supernatant was decanted. The pellet was carefully resuspended in 6.5 ml ice-cold homogenization buffer [10 mM Tris–HCl (pH 7.4), 0.6 M sorbitol, 1 mM ethylenediaminetetraacetic acid (EDTA), 0.2% (w/v) bovine serum albumin] per gram of cells. The spheroplasts mixture was centrifuged (2200 × g, 8 min, 4°C) and the supernatant was decanted. The same volume of homogenization buffer (6.5 ml/g of cells) was used to resuspend the spheroplasts mixture, which was then transferred to a pre-chilled, dounce glass homogenizer on ice. With a tight (B) pestle, the spheroplasts mixture was homogenized with 15 strokes. An equal volume of ice-cold homogenization buffer was added to the homogenizer, and then the homogenate was transferred to 50-ml conical tubes on ice. The homogenate was centrifuged (1500 × g, 5 min, 4°C) to pellet undigested cells, nuclei, and other large debris. The supernatant was poured into a new 50-ml conical tube and centrifuged (3000 × g, 5 min, 4°C) to further clarify the desired material. The resulting supernatant was poured into a new 50-ml conical tube and centrifuged (12 000 × g, 15 min, 4°C) to pellet the mitochondria. The supernatant was decanted, and the pellet was resuspended in 6.5 ml ice-cold homogenization buffer per gram of cells, using trimmed pipette tips to avoid breaking organelles during resuspension. The two previous centrifugation steps (3000 × g, 5 min, 4°C and 12 000 × g, 15 min, 4°C) were repeated once to remove any remaining large debris, and to re-pellet the mitochondria. The supernatant was decanted and the crude mitochondrial fraction (also containing organelles including the endoplasmic reticulum, Golgi, and vacuoles [22]) was resuspended in 3 ml ice-cold SEM buffer [10 mM MOPS–KOH (pH 7.2), 250 mM sucrose, 1 mM EDTA]. The crude mitochondrial fraction samples were stored overnight at 4°C in SEM.

Mitochondrial samples analyzed by total nucleoside LC-MS/MS were further purified immediately after obtaining the crude mitochondrial material. Sucrose gradients were poured by layering 0.7 ml of 60%, 1.7 ml of 32%, 0.7 ml of 23%, and 0.7 ml of 15% sucrose in EM buffer [10 mM MOPS/KOH (pH 7.2), 1 mM EDTA] into 13 × 51 mm Ultra Clear tubes (Beckman Coulter). Approximately 1.3 ml of crude mitochondrial material was layered on top of the gradients, which were then centrifuged for 33 min at 134 000 × g, 4°C in a Beckman SW55 Ti rotor. The mitochondrial band at the 60%/32% sucrose interface was removed using a cut P1000 pipette tip and placed in a new 13 × 51 mm tube. The tubes were filled with cold SEM buffer and then centrifuged in a SW55 Ti rotor at 100 000 × g for 30 min at 4°C. The supernatant was decanted, and the mitochondrial pellets were resuspended in 1 ml SEM buffer and flash frozen in liquid nitrogen.

tRNA isolation

The crude mitochondrial fractions were transferred to microcentrifuge tubes and centrifuged (12000 × g, 15 min, 4°C) to pellet the mitochondria. The supernatant was removed, and the mitochondria were resuspended with 400 μl TES buffer [10 mM Tris (pH 7.5), 10 mM EDTA, 0.5% SDS]. Four-hundred microliters of acidic phenol was added, and the tubes were incubated in a water bath at 65°C for 60 min, with 10 s of vortexing every 15 min to lyse the mitochondria. The lysates were placed on ice for 5 min, and then centrifuged (13 000 rpm, 10 min, 4°C). The aqueous layer (upper layer) was transferred into a new microcentrifuge tube. Four-hundred microliters of chloroform was added and the tube was vortexed for 10 s. The solution was centrifuged again (13 000 rpm, 10 min, 4°C), and the chloroform extraction was repeated. The aqueous layer was transferred into a clean microcentrifuge tube, then one-tenth volume of 3 M sodium acetate (pH 5.5) and 2.5 volumes of cold 100% EtOH were added. The tubes were mixed by inversion and placed at −80°C for ≥1 h. The tubes were centrifuged (13 000 rpm, 10 min, 4°C) and the supernatant was removed. Five-hundred microliters of cold 70% EtOH was added, and the tubes were once again centrifuged (13 000 rpm, 10 min, 4°C). The supernatant was removed, and the tubes were left open for 5 min to evaporate remaining EtOH. The pellet was resuspended in 20 μl of water and stored at −80°C.

Using the SequaGel 19:1 Denaturing Gel System (National Diagnostics), an 8% TBE–Urea gel was cast. The gel was pre-run at 45 mA for 30 min. Between 21 and 100 μg of the mitochondrially enriched RNA was combined with an equal volume of 2× RNA Loading Dye (NEB) and incubated at 70°C for 8 min to denature RNA secondary structure. The samples were loaded onto the gel, which ran at 60 mA for 1 h in 1× TBE (pH 8.3). The gel was removed from the apparatus, stained with SYBR Gold (Invitrogen) for 10 min, and imaged with an Amersham Typhoon (Cytiva). The tRNA was detected by UV-shadowing and gel segments containing RNA ∼70–100 nt in length were excised. The gel slices were placed into microcentrifuge tubes containing 450 μl of 0.3 M NaCl and incubated overnight at 4°C on a tube inverter. The next day, the liquid was transferred to a new microcentrifuge tube and 1.05 ml of 100% EtOH was added. The samples were incubated at −80°C for ≥1 h and centrifuged (13 000 rpm, 30 min, 4°C). The supernatant was removed, and the pellet was air dried for 5 min to evaporate remaining EtOH. The pellet was resuspended in 10 μl of water, the RNA concentration was measured using the Qubit RNA BR assay kit (Thermo Fisher Scientific), and the tRNA was stored at −80°C.

tRNA used in APM-PAGE experiments was isolated using Nucleobond AX-100 columns (MachereyNagel) as previously described [24].

Nanopore mt-tRNA sequencing

The RNA and DNA oligonucleotides for the splint adapters were ordered from IDT. Separate 10 μM (7.5 μM for ACCA) stock solutions of the four double-stranded splint adapters were assembled in 1× TNE (10 mM Tris, 50 mM NaCl, 1 mM EDTA) by combining equimolar concentrations of the common adapter that ligates to the 3′ end of the tRNA, and one of four adapter strands complementary to the 3′ tRNA overhang of ACCA, GCCA, UCCA, or CCA (see Supplementary Table S2 for sequences). The splints were annealed by heating at 75°C for 1 min and then cooling to room temperature by leaving the tubes on the bench for 10 min before placing them on ice.

The library preparation was performed as previously described [20], with a few modifications. Briefly, 250 ng of mitochondrial-enriched tRNA was ligated to four different double-stranded splint adapters specific to ACCA (12 pmol), GCCA (8 pmol), UCCA (8 pmol), and CCA (4 pmol) tRNA 3′ overhangs (see Supplementary Table S2 for oligonucleotide sequences). The splint adapter cognate for the CCCA 3′ overhang was excluded as no mt-tRNAs in our reference contain this overhang (Supplementary Table S5). In the first ligation, the splint adapters were ligated to tRNA by T4 RNA Ligase 2 (NEB, stock concentration 10 000 U/ml, a total of 10 units per reaction). The second ligation of the ONT RMX motor adapter to the first ligation product was performed with T4 DNA ligase (NEB, 2 000 000 units/ml). The first and second ligation reaction were each purified with magnetic beads (RNA Clean XP, Beckman Coulter), using 1.8× and 1.5× volume of beads added to the reactions, respectively. The elution of the library, flow cell priming, and loading of MinION R9.4.1 flow cells followed the SQK-RNA002 protocol. All sequencing was done on the MinION platform with live basecalling turned off to circumvent MinKnow from discarding reads approximating the length of tRNA reads.

IVT mt-tRNA construction and sequencing

The DNA oligonucleotides for the in vitro transcribed (IVT) constructs were purchased from IDT. Their sequences are shown in Supplementary Table S2. Yeast IVT mt-tRNAs were generated using the HiScribe T7 Quick High Yield RNA Synthesis Kit (NEB, E2050), as previously done for yeast cytosolic tRNAs [20]. Five-hundred nanograms of RNA from the IVT reaction was sequenced following the SQK-RNA002 protocol for mRNA sequencing.

Bioinformatic methods for DRS

Analysis of Nanopore DRS data was done essentially as described [20]. Basecalling of ionic current files was performed with Guppy v3.0.3, and the resulting FASTQ files were aligned to a reference—containing all 24 mitochondrial and 42 cytosolic tRNA isoacceptors—using BWA-MEM (parameters ‘-W 13 -k 6 -x ont2d’) [25]. Alignments were filtered using SAMtools [26] to retain those with a mapping quality score >1 (Q1). marginAlign and the subprogram marginCaller [27] were used to generate alignment and error models, and calculate posterior probabilities, respectively. The IVT alignment model, which was derived from a pooled dataset containing mitochondrial IVT tRNA reads (this study) and cytosolic IVT tRNA reads [20], was used as the “error model” in marginCaller in all experiments. Heatmaps containing the posterior probabilities were generated using matplotlib [28].

Mismatch probability [P(A′)], or the probability of any alternative nucleotide call, is complementary to the reference match probability (RMP) [P(A)]:

Random subsets of 105 Q1 aligned reads per mt-tRNA isoacceptor were ran through marginCaller and used to calculate mismatch probabilities for all experiments. Miscalls were assigned to positions with a mismatch probability of ≥0.3 for biologically derived mt-tRNA samples. To determine novel Pus2-dependent Ψ sites, positions U₂₇/U₂₈ with a mismatch probability ≥0.3 in wild type and <0.3 in pus2∆ were considered. Changes in mismatch probability [P(A′)_∆] were considered significant if the difference was larger than an empirically determined minimum difference threshold [P(A′)_{∆, min}]:

The minimum difference threshold was calculated by taking the mean P(A′)_∆ value from the seven known Ψ₂₇/Ψ₂₈ sites [8] and subtracting 1 standard deviation. The value of this threshold for Pus2 sites was 0.140:

For Dus2 sites, a three-position window centered on U₂₀ was examined, but only the position with the largest P(A′)_∆ for a given isoacceptor was used in proceeding steps. The minimum difference threshold was calculated as described above from the 16 annotated D₂₀ sites and was determined to be 0.223.

tRNA Dihydrouridine sequencing (tD-seq)

Total RNA isolated from crude mitochondrial fractions was treated with NaBH₄ as previously described [29]. For each sample, 2 μl of 100 mg/ml NaBH₄ in 10 mM KOH and 1.6 μl of 500 mM Tris–HCl, pH 7.5, was added to 1 μg of total RNA in 16.4 μl of water. No-treatment control samples were prepared similarly except for the omission of NaBH₄ in the 2 μl of 10 mM KOH. Reactions were mixed by pipetting and incubated on ice for 1 h, and then neutralized with 4 μl of 6 N acetic acid. RNA was precipitated by adding one-tenth volume of 3 M sodium acetate (pH 5.5), 2.5 volumes of cold 100% ethanol, and 2 μl RNA grade glycogen (Thermo Fisher Scientific) to the neutralized reactions, and then placed at −80°C for ≥1 h. The samples were centrifuged (13 000 rpm, 30 min, 4°C) and the supernatant was removed. The pellets were washed with cold 70% ethanol, and the tubes were once again centrifuged (13 000 rpm, 10 min, 4°C). The supernatant was removed, and the tubes were left open for 5 min to evaporate remaining ethanol. The RNA pellet was resuspended in 7.8 μl of water.

Ligation of the 3′ reverse transcription adapter was performed as described in the “Sequence-specific Direct RNA protocol” (Oxford Nanopore Technologies) to capture NCCA 3′ tRNA overhangs. Oligo A (RTA A) was ordered without modification, while custom versions of Oligo B (RTA B NCCA) that complement ACCA, UCCA, GCCA, and CCA tRNA ends were used (see Supplementary Table S2 for oligonucleotide sequences). RTA A and RTA B NCCA oligos (1:1) were separately annealed in buffer (10 mM Tris–HCl, pH 7.5, 50 mM NaCl) by heating at 95°C for 2 min in a thermocycler, then slowly cooled (0.1°C/s) to 4°C and placed on ice. The following was added to the 7.8 μl of RNA: 3 μl NEBNext Quick Ligation Reaction Buffer, 0.9 μl ACCA RTA (10 μM), 0.6 μl UCCA RTA (10 μM), 0.6 μl GCCA RTA (10 μM), 0.6 μl CCA RTA (5 μM), and 1.5 μl T4 DNA Ligase (NEB, 2M U/ml). Ligation reactions were mixed by pipetting, incubated at room temperature for 10 min, and immediately used in the following step without cleanup. Reverse transcription was started by adding 9 μl water, 2 μl of 10 mM dNTP solution, 8 μl of 5× First Strand Buffer, 4 μl of 0.1 M DTT, and 2 μl SuperScript III (Invitrogen, 18080093) to the ligation reaction. Reactions were incubated in a thermocycler at 50°C for 50 min, and then 70°C for 10 min. RNA was degraded by adding 4 μl of 1 M NaOH and boiling at 95°C for 5 min, and then 4 μl of 1 M HCl was added to neutralize the solution. The resulting cDNA was ethanol precipitated as described above and resuspended in 8 μl water.

Precipitated cDNA was mixed with an equal volume of 2× RNA loading dye, then heated at 65°C for 5 min, and loaded onto an 8% Urea–PAGE gel (8 × 8 cm), which had been pre-run for 20 min at 200 V. The gel was run at 200 V in 1× TBE for ∼45 min until the bromophenol blue dye was at the bottom of the gel, and then stained with SYBR Gold (1× final concentration in 1× TBE) for 5 min with gentle rocking. Bands corresponding to full length and truncated RT products from tRNAs (∼70–140 nt) were excised and placed in tubes containing 400 μl DNA elution buffer (300 mM NaCl, 10 mM Tris, pH 8.0), and then incubated overnight at 4°C on a tube inverter. The eluted cDNA was ethanol precipitated and resuspended in 5 μl of water. Next, 0.8 μl of 3′ adapter (80 μM) and 1 μl dimethyl sulfoxide (DMSO) was added to the cDNA, incubated at 75°C for 2 min, and placed on ice for 2 min. The ligation reaction was started by adding 2 μl RNA ligase buffer, 0.2 μl of 0.1 M ATP, 6.5 μl of 50% PEG-8000, 3.6 μl water, and 0.5 μl T4 RNA Ligase 1 (NEB, 10 000 units/ml), and then incubated overnight at 22°C. The reaction was cleaned up with 1.8× volume of magnetic beads (RNA Clean XP, Beckman Coulter) following manufacturer instructions and eluted in 10 μl water. Q5 High-Fidelity DNA Polymerase (NEB, M0491S) was used for library PCR in a final reaction volume of 50 μl, with the following thermocycler program: (i) 98°C 30 s, (ii) 98°C 10 s, (iii) 64°C 30 sec, (iv) 72°C 20 s, and (v) 72°C 2 min. Steps (ii)–(iv) were repeated for six cycles. Excess primers were removed by magnetic bead cleanup with 1.8× volume Mag-Bind TotalPure NGS beads (Omega Bio-tek) according to manufacturer instructions and eluted in 30 μl water. PCR products with unique i5 and i7 indices were pooled and sequenced on the Illumina MiSeq v3 platform using a custom index 1 primer.

Demultiplexed reads were processed as previously described [29], including trimming of adapter sequences, PCR duplicate collapsing, UMI removal, read mapping, and 5′ read end position gathering. The same tRNA reference library was used for mapping reads from both the DRS and tD-seq datasets, but for tD-seq the RNA bases of the 5′and 3′ DRS splint adapters were not included. Sequencing and alignment statistics for tD-seq Illumina sequencing data are shown in Supplementary Table S3.

Thresholds for D-site calling were empirically determined by examining the “relative stops” and misincorporations at D and non-D uridine (including other modified uridine) sites annotated in the Modomics database. For a position p, the “relative stops” was calculated from the number of mapped reads ending at p + 1 (i.e. not covering p) relative to the number of mapped reads covering p. The difference in relative stops and misincorporations was calculated between matched wild-type biological replicates that were either treated or not treated with sodium borohydride. These differences were averaged across three biological replicates to determine the “mean change in misincorporations” and “mean change in relative stops” for every uridine/modified uridine in the Modomics database. Each site was classified as a “D” or “non-D U” and then ROC curve analysis was performed to determine the optimal thresholds for calling a D-site based on the “mean change in misincorporations” or “mean change in relative stops” (Supplementary Fig. S1A). The distribution of these sites relative to the thresholds is plotted in Supplementary Fig. S1B. Next, these thresholds were used to identify D-sites in all mt-tRNAs. If a site was over at least one threshold in WT and under both thresholds in dus1∆dus2∆, it was categorized as a D. Positions 1–13 of all mt-tRNA isoacceptors were excluded due to low sequencing coverage at the 5′ end. A site was determined to be catalyzed by Dus1 or Dus2 if it was above one threshold in wild-type and below both thresholds in the single deletions (dus1∆ or dus2∆).

Nucleoside hydrolysis and quantification by LC-MS/MS

The method for nucleoside quantification was adapted from a previously described method [30]. In brief, purified tRNA (200 ng) from pus4Δ, trm2Δ, and wild-type budding yeast cells, or sucrose gradient purified yeast mitochondria, were first hydrolyzed to mononucleotides with 300 U/μg Nuclease P1 (NEB, 100 000 U/ml), 100 mM ammonium acetate, and 100 μM zinc sulfate, in a 10 μl total volume, at 37°C overnight. Samples were then dephosphorylated with 50 U/μg bacterial alkaline phosphatase (BAP, Invitrogen, 150 U/μl), 100 mM ammonium bicarbonate, and 50 mM zinc sulfate, in a 20 μl total volume, at 37°C for 5 h. Two microliters of internal standard (¹⁵N4-inosine, 1000 ng/ml) was combined with 18 μl of sample to reach a final ¹⁵N4-inosine concentration of 100 ng/ml.

Samples were analyzed using an Acquity UPLC I-Class liquid chromatography system coupled to a Waters Xevo TQ-XS triple quadrupole mass spectrometer. Samples (6°C, injection volume 1 μl) were separated on a Waters Acquity UPLC HSS T3 column (100 Å, 1.8 µm, 2.1 × 150 mm) at 40°C. Mobile phase A was LC-MS grade water with 0.1% formic acid, and mobile phase B was 60% acetonitrile with 0.1% formic acid. The flow rate was 0.2 ml/min and the LC gradient is displayed in Supplementary Table S4. Samples were run in positive mode with multiple reaction monitoring. The following MS source settings were used: desolvation gas temperature 400°C, desolvation gas flow 800 l/h, cone gas flow 150 l/h, and a capillary voltage of 2.8 kV. Calibration curves of the canonical bases and modified nucleosides contained 100 ng/ml of ¹⁵N4-inosine were used to quantify sample nucleoside concentrations. For each nucleoside, modification/main nucleoside % was calculated by taking the ratio of the concentration of that nucleoside to the sum of the concentrations of all nucleosides corresponding to the canonical base.

APM–PAGE and northern blotting

A denaturing gel containing 8% polyacrylamide, 7.5 M urea, and 0.05% [N-(acryloylamino)phenyl]mercuric chloride (APM) was cast and pre-run at 200 V for 15 min in 0.5× TBE. Column-isolated, mitochondria-enriched tRNA samples (50 ng) were mixed with an equal volume of 2× RNA Loading Dye (NEB) and heated at 80°C for 5 min. Samples were loaded onto the gel, which ran at 200 V until the lower dye front ran off. The RNA was transferred onto a neutral nylon membrane (GVS) in a Trans-Blot Turbo Transfer System (Bio-Rad) for 35 min at 400 mA with 0.5× TBE. RNA was crosslinked to the membrane at 0.12 J/cm² in a UV Stratalinker 1800 (Stratagene). Blots were pre-hybridized in 12.5 ml hybridization buffer [5× SSC, 0.02 M Na₂HPO₄ (pH 7.2), 7% SDS, 2× Denhardt] for 4 h at 50°C with 500 μg sheared salmon sperm DNA (Thermo Fisher Scientific). The hybridization buffer and sheared salmon sperm DNA was replaced, and 50 pmol of IR-800 labeled DNA probe (IDT) specific to mt-tRNA^Lys(UUU) (sequence from [12]) was added. Blots were hybridized in the dark overnight at 50°C, then washed twice (for 10 and 30 min) in wash buffer [3× SSC, 0.02 M NaH₂PO₄ (pH 7.5), 5% SDS, 10× Denhardt], and once in stringent wash buffer (1× SSC, 10% SDS) for 8 min. IR-800 signals were collected using an Amersham Typhoon (Cytiva) and densitometry was performed in Fiji-ImageJ2 (The Fiji Project).

Modified BACS for pseudouridine detection

BACS was performed as described [31], but with several modifications. A splint adapter complementing 24 nt at the 3′ end of mt-tRNA^iMet(CAU) was used to ligate the “common adapter strand” (used above for DRS) to tRNA^iMet(CAU). The two adapters (0.8 μl each of 10 µM stocks) were annealed to 100 µg of total RNA in a 15 μl solution containing 1× TNE by heating at 95°C for 2 min in a thermocycler, and then slowly cooled (0.1°C/s) to 4°C and placed on ice. Next, 4 μl of NEBNext Quick Ligation Reaction Buffer (5×) and 1 μl of T4 RNA Ligase 2 were added and the ligation was incubated at room temperature for 45 min. To separate ligated products from excess adapters, ligated tRNA^iMet(CAU) (104 nt) was separated on an 8% Urea–PAGE gel, eluted from gel slices, and ethanol precipitated as described for tRNA isolation (above), and resuspended in 12.5 μl of nuclease-free water. The treatment with 2-bromoacrylamide, column clean-up, reverse transcription (using a custom primer), and RNA degradation steps were performed as described [31]. The cDNA was cleaned up with 1.8× RNAClean XP beads following manufacturer instructions and eluted in 13 μl water. A custom adapter was ligated to the 3′ end of the cDNA as previously described [31], and the reaction was cleaned up with 1.8× RNAClean XP beads and eluted in 15 μl water. PCR was performed with Q5 High-Fidelity DNA Polymerase in a final reaction volume of 25 μl containing 5 μl of 3′ adapted cDNA, with the following thermocycler program: (i) 98°C 30 s, (ii) 98°C 10 s, (iii) 64°C 30 s, (iv) 72°C 20 s, and (v) 72°C 2 min. Steps (ii)–(iv) were repeated for 18 cycles. The reactions were then separated on a 6% TBE mini-gel, and bands corresponding to the expected product size (201 bp) were excised, eluted overnight at room temperature in DNA elution buffer, ethanol precipitated, and resuspended in 10 μl water. DNA concentrations were determined by the Qubit dsDNA HS assay (Thermo Fisher Scientific), and samples were submitted to Genewiz (Azenta) for Sanger sequencing.

Primer extension analysis of cmnm⁵U

Total RNA (20 µg in 328 μl water) was treated with NaBH₄ by adding 32 μl of 500 mM Tris–HCl (pH 7.5) and 40 μl of 100 mg/ml NaBH₄ in 10 mM KOH. The reaction was mixed and incubated on ice for 1 h, and then quenched with 80 μl of 6 N CH₃COOH. The samples were ethanol precipitated, resuspended in 6 μl water, and concentrations were determined by the Qubit RNA HS assay. Seven-hundred fifty nanograms of NaBH₄-treated RNA and 1 pmol of a IR800-labeled DNA oligo complementary to mt-tRNA^Lys(UUU) were annealed in buffer [10 mM Tris–HCl (pH 7.7) and 1 mM EDTA] by incubating at 80°C for 2 min, and then at room temperature for 2 min before being placed on ice. Reverse transcription proceeded by adding 0.75 μl water, 2 μl of 5× First Strand Buffer, 0.25 μl dNTP mix, 1.5 μl of 25 mM MgCl₂, and 0.5 μl SuperScript III. The reaction was incubated for 1 h at 55°C, then the RNA was degraded by adding 0.5 μl of 4 M NaOH and boiling at 95°C for 5 min. Samples were neutralized by adding 1 μl of 1 M HCl, 1 μl of 1 M Tris, and 12.5 μl of 2× RNA loading dye, and then denatured at 65°C for 5 min and placed on ice. The primer extension products were separated on a 12% Urea–PAGE gel, the gel was stained with 1× SYBR Gold, and IR-long and SYBR Gold scans were collected using an Amersham Typhoon.

Results

Structure-guided alignment of all 24 S. cerevisiae mitochondrial tRNAs

We first generated an updated reference list of mt-tRNA sequences and a corresponding structure-guided sequence alignment. Prior to our study, the Modomics database contained only 16 of the 24 mitochondrial-encoded tRNA sequences from S. cerevisiae [8]. The eight isoacceptors not currently in the Modomics database are mt-tRNA^Ala(UGC), mt-tRNA^Asn(GUU), mt-tRNA^Asp(GUC), mt-tRNA^Cys(GCA), mt-tRNA^Glu(UUC), mt-tRNA^Gln(UUG), mt-tRNA^Thr(UGU), and mt-tRNA^Val(UAC). To ensure the accuracy of our reference sequences, we performed pairwise comparisons of mt-tRNA sequences annotated from RNA-seq experiments [32], Modomics [8], and NCBI [33, 34]. The resulting mt-tRNA sequence reference is summarized in Supplementary Table S5. We focused on tRNA sequences encoded in the mitochondrial genome, and did not attempt to quantify the abundance of tRNAs imported into the mitochondria from the cytosol [35], since our enrichment protocol did not stringently purify mitochondrial material from the cytosol.

Next, we manually curated a structure-guided sequence alignment of all 24 mt-tRNAs based on conserved tRNA structural features including loop length, stem length, and tertiary contacts (Fig. 1A). We adopted a conventional tRNA numbering scheme [36], with accommodations for predicted single nucleotide bulges and additional nucleotides in the D-loop and anticodon-loop. Apart from these minor differences, our predicted secondary structures were similar to those in cytosolic tRNAs [20]. As seen in the cytosolic tRNAs, the serine and leucine mt-isoacceptors are both of the “type II tRNAs,” containing longer variable loops [37]. In mitochondria, the tyrosine tRNA also has a longer variable loop, unlike the cytosolic tyrosine tRNA. We used this structure-guided alignment and numbering scheme in the analysis and presentation of our sequencing data (Fig. 1A). We summarize the modifications found in yeast mt-tRNAs and their catalyzing enzymes, if known (Fig. 1B).

Figure 1.

Budding yeast mt-tRNA sequences and the ensemble of their modifications profiled in this study. (A) Structure-guided sequence alignment of 24 mt-tRNA isoacceptors from S. cerevisiae. Each base position of the alignment matches a box in the heatmaps presented in Figs 3–5. Asterisks indicate gaps. Left and right parentheses and colored regions show predicted helices/stems (labeled at top). Numbers (top line) indicate each base position in a repeating series of ∼10, or (below) every ∼10th position. The variable loop is up to 13 base positions in length, and alignment of these bases, as well as the bases at neighboring positions (44–48), in the leucine, serine and tyrosine isoacceptors, were anchored via predicted helices/stems in the loop. Histidine tRNA contains a “−1″ G added post-transcriptionally [8]. G•U wobbles are included in predicted stem regions and underlined. Conserved tertiary contacts are indicated underneath the alignment (R = purine, Y = pyrimidine). (B) Secondary structure-based illustration of a generic yeast mt-tRNA, highlighting overall structural segments, 10 types of known chemical modifications, and the names of single enzymes or heteromeric complexes annotated to catalyze them. Positions in blue are those that contain modifications verified in this study by their corresponding genetic mutants. Positions in yellow are those that also trigger base miscalls and are predicted to be chemically modified bases in at least one isoacceptor in this study. Position 54 (red) is annotated to be modified to m⁵U in many mt-tRNAs, but this modification does not yield a base miscall coincident with this position, as previously documented [20].

Direct sequencing of all 24 S. cerevisiae mitochondrial tRNAs

In our previous sequencing of total yeast tRNA isolated from wild-type cells grown in glucose and harvested during exponential phase [20], only 0.3% of all tRNA reads aligned to our mt-tRNA reference. Two factors resulted in low representation of mt-tRNA: (i) mt-tRNAs are significantly less abundant than cytosolic tRNAs, and (ii) yeast cells grown in glucose media contain fewer mitochondria than cells grown in non-fermentable carbon sources due to glucose repression [38]. To address these limitations, we grew yeast cells in large cultures (1 liter) of 2% ethanol media and isolated crude yeast mitochondria prior to RNA isolation (Fig. 2A).

Figure 2.

Overview of sample preparation and tRNA library preparation and sequencing. (A) Crude mitochondrial fraction from large yeast cultures is subjected to traditional RNA isolation, followed by gel purification of tRNA fraction. (B) tRNAs are ligated to a double-stranded splint adapter with T4 RNA Ligase 2 by the tRNA’s 3′ NCCA overhang. A second ligation is performed using T4 DNA Ligase with the tRNA and ONT sequencing adapters. The splint-adapted tRNA pool is sequenced and base called as described in the “Materials and methods” section. Base miscalls are analyzed to infer presence of chemical base modifications, using an IVT mt-tRNA library as a background model. (C) Representation of full length mt-tRNAs from wild-type cells, as measured by DRS, was enriched ∼300-fold in comparison with results obtained with total cellular tRNA isolated from cells grown on a fermentable carbon source [20].

We sequenced tRNA isolated from the crude mitochondrial fraction of yeast using a similar approach to our previous work on cytosolic tRNAs (Fig. 2B). We analyzed IVT mt-tRNA sequences that contained no modifications, which served as a reference for DRS signal obtained from otherwise identical tRNA sequences (see the “Materials and methods” section for details about IVT sample, library preparation, and sequencing). The number of total and aligned reads for each experiment are summarized in Supplementary Table S6.

The average percentage of mapped reads across experiments was ∼51%, which is consistent with prior tRNA DRS experiments using RNA002 kits [17, 20]. Across replicates, between 78% and 98% of all tRNA reads aligned to mt-tRNA reference sequences, a substantial increase in mt-tRNA representation (∼300-fold above reference [20]) compared to prior tRNA DRS studies performed without mitochondrial enrichment [20, 21] and DRS of mitochondrial RNA without tRNA-specific capture [39] (Fig. 2C).

For these analyses, we used the now-discontinued R9.4.1 flow cells and RNA002 sequencing kits because (i) data collection began prior to the release of ONT “RNA” flow cells and RNA004 kits, and (ii) to permit direct comparison to our recent DRS analysis of 42 yeast cytosolic tRNAs using the same chemistry and basecalling approach. The miscall-based modification detection method, using the RNA002 chemistry employed here, was validated by orthogonal LC-MS/MS experiments in our prior study [20]. Miscall-based DRS analysis of tRNA chemical modifications has provided valuable biological insights in numerous studies [16–20, 40] and will continue to do so until ionic current analysis of DRS data is robust and widely adopted.

Direct RNA sequencing reveals modification landscape of yeast mt-tRNAs

We generated heatmaps displaying reference match probabilities (RMPs) at each position in all 24 yeast mt-tRNAs to visualize base miscalls. The “RMP” at a given position in a tRNA represents the probability that the nucleotide assigned by the base caller matches the canonical unmodified reference nucleotide. These probabilities were calculated using an error-based model that accounts for standard DRS rates of mismatches, insertions, and deletions, as well as sequence context-based effects unique to the unmodified sequences that were derived from our IVT mt-tRNA data (see the “Materials and methods” section). The mt-tRNA IVT heatmap shows that nearly all positions have a high RMP (dark blue) as expected from unmodified RNA (Fig. 3A). The IVT sequences account for DRS error that can arise from certain sequence contexts of unmodified RNA bases. For example, while there was a miscall at position 42 in mt-tRNA^Asp(GUC), we verified that the DNA template for in vitro transcription of this tRNA matched the reference nucleotide (G), using Sanger sequencing (data not shown). Further inspection of a k-mer associated with this position, 5′-GGAGG-3′, in other IVT molecules revealed a similar G-to-A miscall [20], implicating this as a sequence context-dependent miscall.

Figure 3.

Heatmaps representing alignments of 24 S. cerevisiae mt-tRNA isoacceptors exhibit miscalls coincident with chemically modified positions. (A) Heatmap representing IVT tRNA sequences of 24 S. cerevisiae mitochondrial isoacceptors. A higher RMP (dark blue) corresponds to positions where the base-called nucleotide more frequently concurs with the reference nucleotide given the alignment, and a lower RMP (light yellow) corresponds to positions where the base-called nucleotide more frequently disagrees with the reference nucleotide. Sequences were aligned by the D-loop, anticodon loop, variable loop, and T-loop (gray boxes). Positions are numbered according to a conventional tRNA base numbering scheme, based on the sequence alignment in Fig. 1A. (B) Wild-type isoacceptors isolated from yeast cells, otherwise as described above.

The heatmap of wild-type mt-tRNAs isolated from yeast shows many positions with low RMPs (yellow-toned) even after correction from the error model, indicating possible modification sites (Fig. 3B). There are fewer base miscalls in these mt-tRNAs relative to cytosolic tRNAs analyzed with the same method [20], in agreement with the literature [3].

Within the 16 previously annotated mt-tRNAs present in the Modomics database, 10 unique chemical modifications were documented in 16 different base positions (Supplementary Table S7) [8]. We calculated the “mismatch” probability (equal to 1.0 minus the RMP value) for each position from a random subset of 105 aligned reads per mt-tRNA isoacceptor. In the program used to calculate posterior probabilities, marginCaller, the default posterior probability threshold to call an alternative nucleotide was empirically determined and set at 0.3 [27]. Importantly, mismatch probability does not necessarily reflect modification stoichiometry. Positions with a mismatch probability <.3 in IVT and ≥.3 in wild type were classified as true miscalls. Of the 16 documented modification sites, 15 had a mismatch probability of ≥.3 in at least one mt-tRNA annotated to have a modification at that position (Supplementary Table S8). The only annotated modification that was not detected on any mt-tRNA isoacceptor by DRS was m⁵U₅₄, a result consistent with our prior sequencing of cytosolic yeast tRNAs [20]. Given the correlation between mt-tRNA modifications as documented on Modomics with miscall-based signals in our DRS data, we extrapolated to the other eight mt-tRNAs whose ensemble of modifications were not previously documented. We predict these contain approximately five to eight modifications each, a density resembling that of the other 16 isoacceptors. An exception was the sparsely modified mt-tRNA^Asp(GUC) that only produced miscalls at two positions (Supplementary Table S8). Most of these eight mt-tRNAs likely also contain Trm2-catalyzed m⁵U₅₄, as 15 of 16 mt-tRNA isoacceptors listed in the Modomics database contain this modification [8].

Among known mt-tRNA modifications, Ψ was most robustly detected by DRS. Ψ can be detected as a U-to-C miscall at the modified position and does not typically impact miscalls at neighboring, unmodified positions [41]. In wild-type cells, Ψ sites at positions 27, 28, 31, and 55 were recorded as miscalls in 100% of previously annotated positions, while Ψ sites at positions 32 and 39 were recorded as miscalls in 5 of 6 isoacceptors previously annotated to contain them (Supplementary Table S7). The only previously annotated Ψ site that our DRS data did not recapitulate was position 72 of mt-tRNA^iMet(CAU) [42]. To address this discrepancy, we used a modified version of BACS [31]. The method detected Ψ₇₂ in mt-tRNA^iMet(CAU) in samples grown in the same media as the samples we had used for DRS (Supplementary Fig. S2), excluding the possibility that different growth conditions were responsible. We propose that a low stoichiometry, or the sequence context of this site, or both, may preclude detection by DRS. Finally, we predicted 21 Ψ sites in the 8 unannotated mt-tRNAs, based on comparison to Ψ positions known in the 16 annotated mt-tRNAs. We validated 11 of these 21 sites through genetic knockouts of Ψ synthases.

Pus4 pseudouridylates U₅₅ in 23 of 24 mt-tRNAs and promotes m⁵U₅₄

Pus4 catalyzes formation of U₅₅ to Ψ₅₅ in 41 of 42 nuclear-encoded tRNA isoacceptors in yeast [20]. Pus4 also modifies mt-tRNAs [43] and all 16 of the yeast mt-tRNAs with known modification profiles contain Ψ₅₅ [8]. We sequenced tRNAs from the crude mitochondrial fraction of pus4∆ yeast to verify that the low RMP at U₅₅ in wild-type mt-tRNAs (Fig. 3B) was due to Pus4-catalyzed Ψ. In mt-tRNAs isolated from pus4∆ cells, there was a high RMP at position 55 (Supplementary Fig. S3A), indicating a lack of modification. The subtractive heatmap highlights differences in RMP between mt-tRNA isolated from wild-type and pus4∆ yeast strains, where blue-colored squares represent a decrease in modification in the mutant (Fig. 4A). Out of the 24 yeast mt-tRNAs, we found that 23 were modified at position 55 by Pus4 (Supplementary Table S9). The exception was mt-tRNA^Asp(GUC), which has a guanosine at position 55 and therefore cannot be pseudouridylated.

Figure 4.

Pus4-based pseudouridylation of 23 of 24 yeast mt-tRNA isoacceptors and its impact on m⁵U levels. (A) The change in RMPs between pus4Δ and wild-type aligned isoacceptors. A positive change in RMP (blue squares) indicates a decrease in base miscalls for pus4∆, and a negative change in RMP (red squares) indicates an increase in base miscalls for pus4∆. White-toned squares indicate that RMPs do not differ between the two strains. The scale was determined by the maximum change in RMP. (B) Total nucleoside LC-MS/MS of tRNAs from wild-type and pus4∆ sucrose gradient purified mitochondria (n = 3). Modification abundances of Ψ, m⁵U, m⁵C (5-methylcytidine), and m¹A (1-methyladenosine) are shown. Data for all measured modifications are available in Supplementary Table S10. Statistical significance was determined by Šidák’s multiple comparisons. Modification/main nucleoside % was calculated as [modification/(canonical base + all modifications to corresponding canonical base) × 100]. (C) Illustration of the interaction between Ψ₅₅ and m⁵U₅₄ in the T-loop of mt-tRNAs.

Some tRNA modifications are involved in “circuits” where one modification promotes or represses the formation of other modifications [44]. Two well-studied modification circuits involve Pus4-catalyzed Ψ₅₅ promoting addition of m⁵U₅₄ and m¹A₅₈ in the T-loop of many yeast cytosolic tRNAs [20, 45, 46]. Since DRS is unable to detect m⁵U, we used total nucleoside LC-MS/MS on tRNAs isolated from sucrose gradient-purified mitochondria to measure m⁵U levels in pus4∆ yeast (Fig. 4B). The levels of m⁵C and m¹A were minimal in our pure mt-tRNA samples, compared to total cellular tRNA (Supplementary Fig. S4). These modifications are only known to be found in cytosolic tRNAs [8]; therefore, our mt-tRNA samples did not contain a significant amount of cytosolic tRNA contamination that would confound our measurement. m⁵U levels were reduced from 4.1% of all uridines in wild-type to 3.2% in pus4∆, a 22% reduction. This confirmed that Ψ₅₅ promotes the addition of m⁵U₅₄ in mt-tRNAs (Fig. 4C). Reduction of m¹A₅₈ levels in cytosolic tRNAs upon the loss of Ψ₅₅ has been observed by DRS [20]. However, m¹A₅₈ had not been detected in any mt-tRNAs previously [8], as we also observed by LC-MS/MS of our sucrose gradient-purified mitochondria, and we also did not detect miscalls at A₅₈ in the 24 mt-tRNAs isolated from a crude mitochondrial fraction.

Identification of Pus2-dependent Ψ sites in S. cerevisiae mt-tRNAs

In budding yeast, Pus2 is a mitochondrially localized enzyme that modifies U₂₇ and U₂₈ of mt-tRNAs [47]. Seven of the 16 yeast mt-tRNAs currently in the Modomics database are annotated to contain Ψ at position 27 or 28 [8], but only two sites have been determined to be Pus2-dependent [47]. We mapped Pus2-dependent modification sites on all 24 mt-tRNAs by sequencing tRNA isolated from the crude mitochondrial fraction of pus2∆ cells, (Fig. 5A and Supplementary Fig. S3B). We assigned Pus2-dependent sites by comparing the mismatch probabilities between pus2Δ and wild-type (Supplementary Table S8). We classified Ψ sites as Pus2-dependent if they met the following criteria: (i) the mismatch probability was equal to or greater than the 0.3 threshold in the wild-type data, (ii) the mismatch probability was <0.3 in the pus2∆ data, and (iii) the difference in these mismatch probabilities met or exceeded an empirically determined minimum difference threshold (see the “Materials and methods” section). This analysis corroborated all seven previously annotated Ψ₂₇/Ψ₂₈ sites across different mt-tRNAs and showed that Pus2 modifies six of these sites (Supplementary Table S9). Additionally, we inferred the presence of a Pus2-dependent Ψ₂₇ for three mt-tRNAs that have not been previously annotated: mt-tRNA^Asn(GUU), mt-tRNA^Thr(UGU), and mt-tRNA^Val(UAC) (Supplementary Table S9). For mt-tRNA^Thr(UGU), we found a Pus2-dependent miscall at U₂₈, which would make it the only mt-tRNA to have Ψ at both positions 27 and 28, a feature that exists in Pus1-catalyzed sites in cytosolic tRNAs [8, 20]. Orthogonal validation of this proposed Ψ₂₈ site in mt-tRNA tRNA^Thr(UGU) is warranted.

Figure 5.

Heatmaps reveal both positions with modifications catalyzed by Pus2 and Dus2 and other impacted sites. The heatmaps show the change in RMPs between (A) mt-tRNAs from pus2∆ and wild-type cells, and (B) mt-tRNAs from dus2∆ and wild-type cells. Otherwise as described in Fig. 4A. The scales were determined by the maximum change in RMP.

Pus2-catalyzed Ψ₂₇ and Ψ₂₈ impact miscalls at other positions

Our sequencing of mt-tRNA from pus2∆ yeast revealed that loss of Ψ₂₇/Ψ₂₈ induced changes in base miscalls at other sites. These changes suggest Pus2 is involved in a modification “circuit,” an instance in which one modification promotes or represses addition of a second, different modification [44, 48]. In some cases, loss of Ψ₂₇/Ψ₂₈ led to an associated increase in miscalls, or putative modification, at a different site. We observed this pattern in mt-tRNA^Lys(UUU), mt-tRNA^Thr(UAG), and mt-tRNA^Thr(UGU), where the RMP was decreased at position 31 or 30, when cells were missing Pus2. Position 31 is annotated as Ψ in mt-tRNA^Thr(UAG) and mt-tRNA^Lys(UUU) [8], however it is unclear whether modification levels of Ψ₃₁ are increased in pus2∆ for these isoacceptors or whether Ψ₂₇ impacts the basecalling of Ψ₃₁, as we have observed for Ψ in other contexts [49]. Further orthogonal validation is warranted to distinguish between these models. However, position 30 of mt-tRNA^Thr(UGU) is a guanosine, which is a position not known to be modified in any yeast tRNA. Therefore, the source of this signal is unknown. For mt-tRNA^Lys(UUU), the RMP at positions 34 and 35 were lower in pus2∆ compared to wild-type, suggesting that loss of Ψ₂₈ leads to an increase in a modification elsewhere. We hypothesized this was caused by an increase in levels of the annotated modification cmnm⁵s²U at position 34 of the mt-tRNA^Lys(UUU) anticodon. A heterodimeric complex composed of Mto1 and Mss1 forms cmnm⁵U, while Slm3 independently forms s²U at position 34 of mt-tRNAs [12]. We used APM–PAGE followed by northern blotting to measure 2-thiolation levels of mt-tRNA^Lys(UUU) in wild-type and pus2∆ strains (Supplementary Fig. S5A and B). The loss of Ψ₂₈ does not significantly alter 2-thiolation levels of mt-tRNA^Lys(UUU). We next examined the possibility that the increase in miscalls was due to increased levels of the non-thiolated portion of this modified uridine (cmnm⁵U), which is present in the anticodon of tRNA^Lys(UUU) at a relatively low stoichiometry in wild-type cells [12]. Coincidentally, we found that NaBH₄ treatment induces stops at U₃₄ during reverse transcription in several cmnm⁵U-containing mt-tRNAs (see below) and performed primer extension on mt-tRNA^Lys(UUU) from wild-type and pus2∆ strains (Supplementary Fig. S5C and D). The percentage of RT stops at U₃₄ were similar between wild-type and pus2∆ samples treated with NaBH₄, indicating that cmnm⁵U₃₄ levels are not altered upon the loss of Ψ₂₈. In summary, we did not find an explanation for the increase in miscalls at positions 34 and 35 of mt-tRNA^Lys(UUU) in pus2∆.

In mt-tRNA^His(GUG), there was a large increase in RMP at A₂₁ in pus2∆ relative to wild-type (Fig. 5A). One model is that addition of Ψ₂₇ by Pus2 may promote the formation of an unknown modification at or near A₂₁ in wild-type cells. But we did not uncover any further evidence of such a modification. The miscall at A₂₁ was not dependent on the presence of D₂₀ (see below) and therefore cannot be attributed to a change in D levels at this proximal position. Moreover, this site did not cause stops or misincorporations during reverse transcription in our tD-seq data from wild-type samples (see below), suggesting that it is not a modified adenosine that interferes with processivity of the SuperScript III reverse transcriptase. There are no modifications known to occur at A₂₁ in any tRNAs in any organism, and therefore further experiments are needed to identify the source of this signal.

Identification of Dus2 sites in S. cerevisiae mt-tRNAs by DRS

All 16 mt-tRNA isoacceptors currently in the Modomics database are annotated to contain D at position 20, yet the enzyme that catalyzes this modification has not been confirmed experimentally. It has been presumed that Dus2, which is responsible for dihydrouridylating position 20 in yeast cytosolic tRNAs [50], also modifies mt-tRNAs at the same position. This presumption is supported by the detection of Dus2 in the yeast mitochondrial proteome [51, 52]. Our DRS analyses of mt-tRNAs from dus2∆ yeast provide experimental evidence that Dus2 is necessary for D₂₀ formation in mt-tRNAs. In dus2Δ cells, we observed an increase in RMP at position 20 and/or neighboring positions 19 and 20a/20b/21 for many mt-tRNAs, consistent with reduced modification (Fig. 5B, blue squares; and Supplementary Fig. S3C). Therefore, a three-position window, centered on position 20, was considered for our analysis of mismatch probabilities, such that a threshold change at any one or more of these positions led us to assign a D₂₀ (see the “Materials and methods” section for details). Of the 16 mt-tRNAs in the Modomics database annotated to contain D₂₀, DRS corroborated 13 sites as Dus2 targets (Supplementary Table S11), indicating that DRS can detect D. The three mt-tRNAs annotated to contain D₂₀ sites that were not detected by DRS were mt-tRNA^Pro(UGG), mt-tRNA^His(GUG), and mt-tRNA^Tyr(GUA). Of the eight mt-tRNAs not currently in the Modomics database, DRS detected Dus2-dependent changes in six and did not detect D₂₀ in mt-tRNA^Asp(GUC) and mt-tRNA^Cys(GCA), though both contain a uridine at position 20. In these five tRNAs for which DRS did not predict changes in D₂₀ in the mutant, the modification was detectable by an orthogonal, D-specific method as Dus2-dependent (see below).

Identification of all D sites by tRNA Dihydrouridine sequencing (tD-seq)

Above, we analyzed D using DRS supported by data from a genetic knockout of a D synthase. In addition to changes that we observed coincident with the annotated modification site of Dus2 (U₂₀), we observed unexpected and significant changes at positions 15/16 in many mt-tRNAs (Fig. 5B). As D occurs at other positions in the D-loop that could be impacted by loss of Dus2, we sought to clarify our DRS-based predictions of D sites in mt-tRNAs using an orthogonal method. There are several existing sequencing-based detection methods for D including “D-seq” [29, 53], “Rho-seq” [54], “AlkAnilineSeq” [55], and the recently published “CRACI” [56]. “D-seq” and “Rho-seq” were developed to map D-sites across the nuclear-encoded transcriptome in S. cerevisiae and S. pombe, respectively. These methods use sodium borohydride (NaBH₄) to chemically reduce D into tetrahydrouridine [57], which may either cause cDNA terminations (RT stops) one nucleotide 3′ of the D site [53, 54], or base misincorporations coincident with a D site [58], during reverse transcription. While some cytosolic tRNA reads were present in the libraries of these previous studies, the presence of modifications that produce NaBH₄-independent RT-stops 3′ of D-loop sites—such as m^2,2G₂₆, m¹G₃₇, and m¹A₅₈—precluded analysis of most D sites in these tRNAs. Additionally, neither study was designed to specifically capture tRNAs nor enrich for mitochondrial RNAs.

To address this gap, we adapted the “D-seq” method to map D sites in mt-tRNAs. The chemistry of this protocol was based on the original “D-seq” protocol [29], but the library preparation method was modified to enrich tRNAs, and therefore we named it “tD-seq” (Fig. 6A). Due to the lower frequency of RT-stop causing modifications in mt-tRNAs relative to cytosolic tRNAs, this method yields sufficient coverage of the D-loop of most mt-tRNA isoacceptors without demethylation. In samples from wild-type yeast not treated with NaBH₄ (Supplementary Fig. S6A), RT stops occurred at or adjacent to established m^2,2G₂₆ and m¹G₃₇ sites, which impair cDNA synthesis [59]. Upon NaBH₄ treatment of wild-type RNA (Fig. 6B and Supplementary Fig. S6B), RT stops appeared at several positions in the D-loop of most isoacceptors, as anticipated for this treatment, and unexpectedly within the anticodon loop of several isoacceptors where D is not known to exist. In contrast, the combined knockout of DUS1 and DUS2—which encode the only D-synthases found in yeast mitochondria [51, 52]—led to the complete loss of NaBH₄-induced stops in the D-loop (Fig. 6C and Supplementary Fig. S6C). This demonstrates that tD-seq can detect D sites in tRNA.

Figure 6.

tD-seq detects D sites in mt-tRNAs. (A) Overview of tD-seq protocol, where RNA is treated with NaBH₄ that converts dihydrouridine into tetrahydrouridine. A 3′-reverse transcription (RT) primer is ligated to tRNAs with T4 DNA ligase prior to reverse transcription with SuperScript III RT. The 3′ adapter is ligated to cDNA using T4 RNA Ligase I and then PCR is performed prior to Illumina sequencing. (B) RT stops induced by NaBH₄ treatment, relative to the coverage of the position 1 nt 3′ of the specified site. Each square point represents the average relative stops for a single mt-tRNA isoacceptor at the specified site. Points from the same isoacceptor are connected by black lines. On the x-axis, NaBH₄ treatment is specified with (+) or (−) next to the genotype. The position 17 and 34 plots have fewer isoacceptors because not all mt-tRNAs have a base and/or a uridine at these positions in the structural alignment. (C) NaBH₄-induced stops in the D-loop of mt-tRNAs are absent in dus1∆dus2∆, while the stops at U₃₄ are observed independently of Dus activity. (D) Misincorporation levels compared between matched wild-type samples with and without NaBH₄ treatment. (E) Misincorporation levels compared between three biological replicates of wild-type and dus1∆dus2∆.

We next considered the unanticipated NaBH₄-induced RT stops within the anticodon loop, which coincide with known cmnm⁵U₃₄/cmnm⁵s²U₃₄ sites. Besides D, NaBH₄ chemically reduces several modified bases including ac⁴C, m⁷G, and m¹A [60, 61]. We performed primer extension on mt-tRNA^Lys(UUU) from wild-type and mto1∆ strains to assess the impact of NaBH₄ on reverse transcription stops at cmnm⁵U₃₄ (Supplementary Fig. S5C and D). Reverse transcription stops at U₃₄ were strongly reduced in the absence of Mto1 in an NaBH₄-dependent manner, indicating that NaBH₄ chemically interacts with cmnm⁵U to disrupt reverse transcription.

While tD-seq produced robust stops for some D sites (D₁₄, D₁₆, D₁₇), D₂₀ sites yielded less robust stops. Base misincorporations coincident with D sites generally followed a similar positional-pattern to RT stops, when comparing wild-type (untreated), wild-type (treated), and dus1∆dus2∆ (treated) samples (Fig. 6D and E, and Supplementary Fig. S7A–C). Therefore, we considered the change in both RT stops and misincorporations (between matched treated and untreated wild-type samples) to facilitate our tD-seq D site annotations, along with assignment of their catalyzing enzymes (Dus1 or Dus2). Cutoff thresholds were empirically determined by comparing values from “D” and “non-D Uridine” sites from annotations in the Modomics database (see the “Materials and methods” section for details on D-site calling).

We used tD-seq to map D-sites in all 24 mt-tRNAs and found that each tRNA contains between one and four dihydrouridines within their D-loops. Comparison of D-site annotations between the Modomics database, DRS, and tD-seq showed a high level of agreement (Supplementary Table S11). We examined D-sites that were detected by tD-seq, but not DRS, to determine whether there was a common feature such as sequence context or stoichiometry between these sites that could explain this discrepancy. There were modest differences in sequence context between D₁₆ sites detected and not detected by DRS (Supplementary Fig. S8). Several sites not detected by DRS produced robust RT stops or misincorporations in tD-seq, suggesting that these sites are not low stoichiometry.

Dus1 modifies cytosolic tRNAs at positions 16 and 17 [50], and is found in the yeast mitochondrial proteome [51]. tD-seq demonstrated that Dus1 catalyzes modifications at four different positions in mt-tRNAs: D₁₄, D₁₆, D₁₇, and D_17a (Fig. 7A and B, and Supplementary Fig. S9A and C, and Supplementary Fig. S10A), establishing its mitochondrial activity and expanding the list of its target sites. The remaining Dus enzymes in yeast, Dus3 and Dus4, were not detectable in the mitochondrial proteome [51, 52], and our data provided no evidence of their canonical activities at positions 47 or 20a/b, respectively, in mt-tRNAs.

Figure 7.

D sites in mt-tRNAs impact the modification levels of other nearby D sites and m^2,2G₂₆. (A, B) Relative stops and misincorporations at D-loop positions 16, 17, and 20 in wild-type and dus1∆ mt-tRNAs. (C, D) Relative stops and misincorporations at D-loop positions 16, 17, and 20 in wild-type and dus2∆ mt-tRNAs. (E) Interactions between dihydrouridines at positions 16, 17, and 20 in mt-tRNAs, based on tD-seq data herein. Arrows represent a positive interaction, while lines ending with a short perpendicular line represent a negative interaction. (F, G) Levels of m^2,2G₂₆ in mt-tRNAs upon the loss of Dus1, Dus2, and/or Trm1 activity was determined by relative stops (left) and misincorporations (right) at position G₂₆. The mean of three biological replicates is plotted, and the error bars show ± one standard deviation. All samples included in these plots were treated with NaBH₄.

DRS and tD-seq reveal interactions between Dus1, Dus2, and Trm1

The loss of Dus2-catalyzed D₂₀ was accompanied by increases or decreases in base miscalls at nearby sites for most mt-tRNAs in our DRS data. For nearly every mt-tRNA with a decrease in miscalls at position 20 in dus2∆, there was an increase (red-colored boxes) in miscall levels at positions 15/16 (Fig. 5B). Based on the DRS results, we initially hypothesized that D₂₀ represses formation of D₁₆/D₁₇. We used tD-seq to semi-quantitatively measure D levels at Dus1-catalyzed sites in dus2∆. The modification levels at D₁₆ and D₁₇ generally decreased in dus2∆ (Fig. 7C and D, and Supplementary Fig. S9B and D), indicating that D₂₀ promotes D₁₆/D₁₇ formation in most mt-tRNAs (Fig. 7E, left and middle diagrams), the opposite of our initial DRS-informed hypothesis. However, there were exceptions showing increased modification levels at: D_17a for mt-tRNA^Ala(UCG), D₁₇ for mt-tRNA^Leu(UAA) and mt-tRNA^Tyr(GUA), and D₁₄ for mt-tRNA^Ser(GCU) in dus2∆. In these instances, D₂₀  represses modification by Dus1 (Fig. 7E, left and middle diagrams), which suggests that the effect D₂₀ has on other D sites depends on the isoacceptor. Since tD-seq shows D₁₆/D₁₇ levels were not generally increased in dus2∆, the cause of the broadly increased miscalls at positions 15 and 16 in our dus2∆ DRS data remains unclear. However, we found a similar pattern in cytosolic tRNAs from the same yeast strain (data not shown). The impact of D on DRS ionic current signal, and therefore the change when it becomes absent, deserves close examination through training with synthetic and biological RNAs. Our use of tD-seq bypassed this concern, however, since it represented a direct measurement of the D modification.

In contrast to the variety of outcomes observed upon loss of the DUS2 gene, loss of DUS1 led to more consistent changes in Dus2-catalyzed modification levels. The dus1Δ mutant showed increased modification levels at D₂₀ for several mt-tRNAs in the tD-seq dataset (Fig. 7A and B, and Supplementary Fig. S9A and C). This indicates that Dus1-catalyzed D₁₄/D₁₆/D₁₇/D_17a represses D₂₀ formation in mt-tRNAs (Fig. 7E, right diagram).

Several isoacceptors in our dus2∆ DRS data had increased mismatch probabilities at position 26. We hypothesized this was caused by an increase of the annotated m^2,2G₂₆ modification, catalyzed by Trm1 [62]. We applied tD-seq to determine how the loss of DUS1 and DUS2 impacts m^2,2G₂₆ levels, as this modification disrupts canonical G-C base pairing [63, 64] and thus reverse transcription. The dus1∆ and dus2∆ strains had increased m^2,2G₂₆ levels for most isoacceptors containing the modification (Fig. 7F and G). This confirmed that the increased DRS-based miscalls at G₂₆ for several isoacceptors in dus2∆ (Fig. 5B) were due to increased m^2,2G₂₆ levels. While Dus1 does not modify mt-tRNA^Pro(UGG), this isoacceptor had elevated levels of m^2,2G₂₆ in the dus1∆ strain (Fig. 7F and G). This suggests that interactions between Dus1 and Trm1 may be independent from Dus catalytic activity but dependent on Dus binding to mt-tRNAs. In fact, recent findings show that catalytic activity of a Dus1 ortholog, DusB, is not required for its role in an oxidative stress response in Vibrio cholerae [65]. Surprisingly, the dus1∆dus2∆ mutant had m^2,2G₂₆ levels comparable to wild-type (Fig. 7F and G). Collectively, these results suggest an interplay between Dus enzymes and Trm1 that may involve non-catalytic Dus activities.

Discussion

As DRS is increasingly applied to study tRNAs from a variety of species and cell-types, it remains important to define reference sequences for each case, while allowing for the de novo discovery of new sequences that may be captured. After we updated and expanded the current list of aligned mt-tRNA sequences from S. cerevisiae, we used this reference to profile chemical modifications using DRS, identifying the positions that are modified by three enzymes: Pus4, Pus2, and Dus2. We adapted an existing method to map D sites in tRNAs, determining all modification sites of Dus1 and Dus2 in mt-tRNAs. These results, together with other observations, advances toward a near-comprehensive map of S. cerevisiae mt-tRNA modifications.

Our results revealed novel examples of modification circuits in mt-tRNAs, in which the presence of one modification either stimulates or represses modifications at other sites. In yeast cytosolic tRNAs, the disruption of a single Dus enzyme has not lead to observed changes at other D sites [50]. However, here we show that Dus1 represses D₂₀ formation and Dus2 can either promote or repress D₁₄/D₁₆/D₁₇/ D_17a formation in mt-tRNAs. While we cannot yet explain why Dus1 and Dus2 influence each other’s activities in mt-tRNAs and not in cytosolic tRNAs, there are several contextual differences. Relative to cytosolic tRNAs, mt-tRNAs have lower GC content [66] and fewer modifications, influencing the stability of their tertiary structures. This could make mt-tRNAs more sensitive to changes in individual modifications, resulting in possible compensatory pathways. Cytosolic tRNAs also contain modifications that are not present in mt-tRNAs, which may interfere with the interplay between Dus1 and Dus2 activities that we observed in mt-tRNAs. We also found that the presence of D generally repressed m^2,2G₂₆ formation by Trm1 in several mt-tRNAs. The contribution of these interactions to mt-tRNA structural stability, and mitochondrial translation, warrants further investigation.

An alternative model that we considered to explain the decrease in modification levels at Dus1 sites in dus2∆ cells is that Dus1 and Dus2 have functional redundancy, such that both enzymes modify sites at positions 14, 16, 17, and 17a, and thus loss of Dus2 could lead to decreases in D levels at these positions. Importantly, such functional redundancy was reported for DusB1 and DusB2 in Bacillus subtilis [67]. Our tD-seq data demonstrate, however, that in dus1∆ cells, there are no detectable levels of D at positions 14, 16, 17, and 17a, indicating that Dus2 does not modify these sites in the absence of Dus1, leading us to disfavor this model in S. cerevisiae mt-tRNAs. Redundancy between Dus1 and Dus2 may be present in other conditions, such as overexpression in vivo or high enzyme concentrations in vitro, which expanded the substrate specificity for B. subtilis DusB2 [67].

Recent technical developments to Nanopore DRS hold promise for improved tRNA modification detection. This includes improved sequencing accuracy with RNA004 kits and modification-aware basecallers. However, these modification-specific basecalling models are currently limited by high false positive rates [68], the inability to distinguish between chemical isomers such as m¹A and m⁶A [69], and are currently only available for four RNA modification types, as advertised. While these models continue to be developed and improved, miscall-based analysis of DRS data remains a suitable approach for detecting tRNA modifications. Our study establishes that DRS can detect D in tRNA using a genetic mutant of a Dus enzyme. This revealed that D causes a miscall-based signal “window” of three neighboring positions. The majority of annotated D sites were indeed detected by DRS, however we also found false negatives in D sites that were detected by tD-seq and reported in the Modomics database, but not detected by DRS. False positives in DRS data were also noted, such as changes in miscalls upon the loss of Dus2 that were not recapitulated by tD-seq. Therefore, as we have stressed elsewhere [20, 49], it remains important to pair DRS with orthogonal methods for the most accurate RNA modification detection.

Recent review articles have compared different techniques to measure RNA modifications, including in tRNA (see references [70–73]). Methods of detection include a variety of structural approaches, LC-MS/MS, reverse transcription-based assays, and Nanopore DRS. Use of multiple methods for detection of previously unreported sites is generally recommended, and the most optimal method of detection will depend on the type of modification and type of RNA molecule being investigated. Nanopore DRS has been particularly well-suited for Ψ in tRNA, as it tends to produce a signal coincident with the modified site and not beyond, although its presence can affect interpretation of neighboring modifications [49].

mt-tRNAs provide an ideal test case for the advancement of direct tRNA-sequencing because they contain many conserved tRNA modifications, but at a lower density compared to cytosolic tRNAs. Therefore, the miscall or ionic current signatures of chemically modified bases are more isolated from neighboring modifications and can be used to advance DRS-based identification of specific modifications. Additionally, each mt-tRNA molecule is transcribed from a single gene copy in the mitochondrial genome, thus bypassing the requirement for analysis of isodecoders that are common among nuclear-encoded tRNA genes. Our work lays a foundation for this type of systematic analysis.

In our DRS-based analysis, we identified positions where miscalls in native wild-type tRNAs did not correspond to any known modification, however only six sites met these criteria: A₂₁ in mt-tRNA^HIs(GUG), U₄₁ in mt-tRNA^Arg(UCU) and mt-tRNA^Cys(GCA), C₇₀ in mt-tRNA^His(GUG) and mt-tRNA^Lys(UUU), and U_49a in mt-tRNA^Met(CAU). Further characterization is needed to determine whether these miscalls represent novel modification sites in mt-tRNAs. This relatively short list nonetheless predicts that there are few modifications left to be discovered in the S. cerevisiae mt-tRNAs.

Like mt-tRNA, mt-rRNA is also chemically modified, yet to an extent that remains understudied. Mitochondrial rRNA contains fewer modifications than cytosolic rRNA, yet its modifications still play an important role in mitochondrial translation [74]. While Pus4 is documented to modify the yeast 15S mt-rRNA [75], in addition to mt-tRNA, we are not aware of examples catalyzed by the other enzymes addressed in our study (Pus2, Dus1, Dus2, Trm1, Trm2), in yeast, even while Trm2 orthologs have been found to modify human mt-rRNA and bacterial rRNA in vitro [76]. While we did not specifically capture mt-rRNA reads here, Nanopore DRS of yeast mt-rRNA, coupled with LC-MS/MS, could reveal the full picture of mt-rRNA modifications in a future study.

Our approach could be used to model the effects of human disease-associated mutations in tRNA modification enzymes known to impact mitochondrial function. Most tRNA enzyme functions are conserved from yeast to humans [10]. Testing human-derived mutations at orthologous sites in yeast proteins, or replacing the yeast proteins with their human counterparts, will expand our understanding of how these mutations affect mt-tRNA function and translation. Our findings also highlight how reduction or loss of a modification can lead to collateral effects on other modifications, and the ensemble of these changes must be considered in context of the resulting molecular and physiological phenotype.

Data availability

The sequencing data described in this manuscript are deposited at the European Nucleotide Archive. The study accession number is PRJEB89858.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported by the National Institutes of Health (Grants R35 GM143125 and R01 HG013876 to D.M.G., and T32 GM149387 to J.L.R.). Funding to pay the Open Access publication charges for this article was provided by the National Institutes of Health R01 HG013876.