To optimize the mitoribosome profiling method, we tested two different RNases typically used for ribosome footprinting, RNase I and MNase, followed by the isolation of the 55S mitoribosome by sucrose gradient fractionation. We mapped the sequence reads to the mitochondrial genome and its thirteen open reading frames (ORFs). For each mapped sequence read, we inferred the position of the ribosomal P-site using a 13-nucleotide (nt) offset from the 5’ end, determined by the analysis of footprint distribution surrounding the initiation sites of MT-ND4, MT-ND6, and the newly reported MT-ND5-dORF. We observed coverage across mitochondrial ORFs, suggesting we captured translating mitoribosomes (Figure 1B). Furthermore, analysis of MRPF lengths revealed a distinct difference between the RNase I and MNase-digested footprint. While both libraries are enriched for reads ∼30 nt in length, MNase footprints showed a bimodal distribution, with a clear secondary peak at ∼22 nts (Figure 1C), which may indicate that MNase-treatment may have captured two distinct conformations of the ribosome. To determine the quality of the ribosome profiling data, we calculated read phasing, which occurs when a nuclease creates uniform MRPFs and the resulting ends of the MRPFs reflect the 3-nt periodicity of elongating ribosomes. While both datasets do show clear phasing, the read phasing is more biased toward position 1 of the codon in the RNase I-treated sample, indicating slightly more consistent digestion (Figure 1D, Table S1). Nevertheless, we observed similar levels of footprint abundance per gene, with the only notable exceptions being that RNase I-digested footprints were enriched on MT-ND3 and depleted on MT-ND4L and MT-ND6 (Figure 1E). While the cause of the difference is unclear, we did not expect it to interfere when comparing an antibiotic-treated sample to a control. Read coverage also appeared similar across the two methods, with 85%-87% codons having between 2 and 50 reads mapping to them (Figure 1F). Overall, these results indicated that both MNase and RNase I digestion capture mitoribosome footprints with only minor differences. Considering the improved read phasing, we selected RNase I for our subsequent mitoribosome profiling experiments.

First, we examined the impact of three antibiotics – retapamulin, josamycin, and tiamulin – on translation initiation or mitochondrial translation. Retapamulin and tiamulin are members of the pleuromutilin class of antibiotics. They work by binding to the peptidyl transferase center (PTC) of the bacterial ribosome and disrupting initiator tRNA and A site tRNA positioning (Figure 2A, left)^{22,23,31–33}. Josamycin is a macrolide antibiotic that binds in the nascent peptide exit tunnel near the PTC and arrests bacterial ribosomes at initiation by interfering with A-site tRNA positioning (Figure 2A, right)³⁴. We sought to understand whether these antibiotics also target translation initiation in mitochondria. To this end, we treated HEK293 cells with high concentrations of these antibiotics (100 µg/ml of tiamulin and josamycin, 10 µg/mL of retapamulin) for 30 minutes, performed mitoribosome profiling, and compared the treated experiments to a DMSO-treated control.

We mapped MRPFs to the 13 mitochondrial genes. On MT-CYB and all other genes, we observed that after treatment with antibiotics, reads accumulated at the 5’ end of the transcript and MRPFs across the rest of the coding region were lost or decreased (Figures 2B, S1). We quantified the percentage of MRPFs mapping at the 5’ end relative to the total ORF, excluding polycistronic mRNAs. Treatment with translation initiation inhibitors increased the percentage of those reads from 2.4 to 5.2-fold relative to the DMSO-treated cells (Figure 2C). Additionally, we quantified the cumulative footprint abundance moving from the start to the stop codon (Figure 2D). While ribosome footprints are generally evenly spread across the open reading frame in the DMSO control, MRPFs from cells treated with the initiation inhibitors showed a sharp increase in the percentage of reads found at the 5’ end. Together, these results indicate that retapamulin, tiamulin, and josamycin inhibit mitochondrial translation initiation, likely by the same mechanism as in prokaryotes.

Chloramphenicol, linezolid, and telithromycin selectively inhibit bacterial translation when ribosomes are translating specific sequences, in a process called context-dependent translation arrest^25,30. Linezolid, like chloramphenicol, induces context-dependent arrest when alanine, serine, or threonine is at the penultimate position of the nascent peptide (Figure 3A, left). Unlike chloramphenicol and linezolid, telithromycin binds in the nascent peptide exit tunnel near the PTC and adjacent to the nascent peptide³⁵. During elongation, telithromycin causes ribosomes to arrest translation at an R/K-X-R/K motif, when the identity of both the penultimate position of the nascent peptide and the incoming aminoacyl-moiety is either arginine or lysine (Figure 3A, right). We carried out mitoribosome profiling on HEK293 cells treated with either antibiotic to understand whether these context-dependent translation inhibitors arrest mitoribosomes in the same manner as they arrest bacterial ribosomes. Unlike the cells treated with initiation inhibitors, cells treated with chloramphenicol or linezolid carried ribosome density throughout the entirety of the ORF, suggesting that, as expected, they were not acting as inhibitors of initiation (Figures 3B, S1). Cells treated with telithromycin did show a slight increase in MRPFs abundance at the 5’ end of MT-CYB. However, across all genes, the levels of MRPFs at the 5’ end were not near the levels seen when using the initiation inhibitor retapamulin and were more similar to chloramphenicol, linezolid, and the DMSO control (Figure 3C). The cumulative distribution of MRPFs suggested that ribosome density was biased towards the 5’ end of the gene for chloramphenicol and telithromycin, but not significantly for linezolid (Figure 3D). These results suggest that these three antibiotics are not acting as initiation inhibitors and that linezolid and telithromycin redistribute MRPFs to the 5’ end of the ORF.

Next, we determined whether this observed redistribution of elongating ribosomes was dependent on the context of the sequence being translated. We calculated the average change of MRPFs depending on the identity of the amino acid at either the −1, P-site, or A-site positions. For both chloramphenicol and linezolid, we observed a specific increase in ribosome footprints when mitoribosomes encountered codons for alanine, serine, or threonine in the −1 position, with a 2-, 2.2-, and 1.3-fold change for chloramphenicol and a 2.3-, 1.4-, and 1.6-fold change for linezolid, respectively (Figure 3E). Additionally, we also observed context-dependent translation arrest induced by telithromycin. Unlike the R/K-X-R/K motif observed in bacteria, we found that telithromycin caused mitoribosomes to arrest at sites with a lysine positioned in the A-site, with an average 2.5-fold increase in MRPFs around those sites, and that neither residue was enriched in the −1 position. Due to the limited size of the mitochondrial genome, there are only nine instances of the R/K-X-R/K motif; nevertheless, when we observed the positions with the greatest fold change, 3 of the top 4 contained the motif with lysine in the A-site (Figure 3F), suggesting that the motif was at least partially recapitulated. To specifically look for the motif, we calculated the average fold change when we fixed arginine or lysine in the A-site and varied the identity of the −1 amino acid (Figure 3G). We found that when lysine was in the A-site while either arginine, lysine, or alanine was in the −1 position, there was strong arrest (> 5-fold change). However, when arginine was in the A-site, we did not observe a strong stalling event with any −1 position residue. Therefore, in mitochondria, telithromycin induces stalling at R/K/A-X-K motifs. This motif appears in only 6 of the 13 mitochondrially encoded genes, indicating that mitochondrial protein synthesis is unequally inhibited (Figure S2A, B). We concluded that the mechanism underlying context-dependent translation arrest for telithromycin is different in mitochondria compared to bacteria.

During the process of assigning the P-site positions of MRPFs, we observed that no ribosomes were assigned in the first 13 nucleotides of the 5’ end of the transcript. This is not surprising, considering that mitochondrial transcripts often have no or very short 5’ untranslated regions - typically less than 4 nts long - and we applied a 13-nt offset from the 5’ end of MRPFs. It reasons that as mitoribosomes begin translation on a leaderless mRNA, the 5’ end of the resulting MRPF would have a full-length region 3’ of the P-site and no sequence 5’ of the P-site. As a ribosome moves down the transcript, the 5’ end of the MRPF would increase until it is accessible for digestion, at which point it would generate 31 nt-long MRPFs (Figure 4A). As accurate mapping of mitoribosomes at initiation could be a useful tool to understand mitochondrial translation, we attempted to map the location of initiating mitoribosomes at the 5’ end using footprint length, rather than the 5’ offset. Because mitoribosome profiling with MNase generated a population of ribosomes with short MRPFs, we hypothesized that these footprints could be indicative of capturing initiating ribosomes (Figure 1C). We plotted the distribution of MRPFs from 20 to 34 nts in length that mapped to the 5’ end of the 11 leaderless or short leader ORFs and compared them to reads mapping to the whole ORF. We observed that MRPFs 22-23 nts in length were 6-fold more abundant and MRPFs >28 nts in length were depleted compared to the whole ORF (Figure 4B). As mitochondrial ORFs vary in the length of their 5’ UTR and we expect most ribosomes to be positioned at the start codon, we tested whether leaderless ORFs were enriched in shorter reads compared to ORFs preceded by a short leader. Leaderless ORFs, such as MT-CO2, MT-CO3, and MT-ND4L, contained MRPFs mainly 20-23 nucleotides in length at the 5’ end (Figure 4C, top). MT-ATP8, which contains a start codon beginning at the third nt, contained footprints that were 22-24 nt long, and MT-CO1, with a start codon at the fourth nt, contained MRPFs 23-24 nt long (Figure 4C, middle). Not only did we observe longer MRPFs in short leader ORFs, but the size of the MRPFs also appeared proportional to the location of the start codon. Together, these results indicate that the size of the footprint is directly related to the start codon position at the 5’ end of mitochondrial mRNAs.

Two genes, MT-ND1 and MT-ND5, deviated from this trend. The annotated start site for MT-ND5 is a leaderless AUA codon (a start codon in mitochondria), but we observed few 20-22 nt MRPFs (Figure 4C, bottom). Similarly, the annotated start site for MT-ND1 is also an AUA codon that begins at the third nt, and it is depleted in reads for MRPFs that are smaller than 25 nts. Instead, MRPFs derived from the 5’ end of both genes are enriched in longer reads: 27 nts-long for MT-ND5 and 28-29 nts-long for MT-ND1. Curiously, the third codon for both genes is an in-frame AUG, and the read lengths proportionally correspond to mitoribosomes positioned at these codons, suggesting mitoribosomes initiate translation at these sites. MRPFs generated from cells treated with initiation inhibitors and footprinted by MNase yield similarly sized reads at the 5’ end of MT-ND1 and MT-ND5 (Figure S3). Using protein mass spectrometry, we tested whether peptides matching the N-terminus of these proteins would correspond to initiation at the canonical start codons or alternative start codons. We identified a peptide that mapped to the N-terminal region of MT-ND1 but lacked the first two amino acids, accordant with translation initiating from the alternative start codon (Figure 4D). While we did not identify N-terminal peptides mapping to MT-ND5 from our mass spectrometry data, the protein mass spectrometry database MassIVE³⁴ covered the N-terminal region of MT-ND5 (Figure 4E). Consistent with our mitoribosome profiling data, the majority of N-terminal fragments did not include the amino acids from the annotated start or second codon but began with methionine at codon 3. We also observed that MRPFs on MT-ND6, which initiates at the first of two AUG codons immediately at the 5’ end, correlated to ribosomes positioned at both AUG codons, indicating that this gene may initiate translation from the second AUG codon as well. However, we did not observe any peptides mapping to the N-terminus of MT-ND6. Together, these data show that MT-ND1 and MT-ND5 may not initiate from their canonical AUA codons but may use alternative in-frame AUG codons. Additionally, our findings indicate that the mapping of mitochondrial translation initiation events by size, rather than 5’ offset, is an accurate strategy when used in conjunction with MNase footprinting.

During our analysis of 5’ end mapping, we did not remove reads mapping to non-coding RNAs (ncRNAs) before mapping to the mitochondrial genome and performing P-site assignment. Consequently, we observed that some tRNAs and rRNAs had an abundance of reads mapping to 13 nts away from their 5’ ends, similarly to what we observe on mRNAs for translation initiation. In addition, some of these ncRNAs contained a potential start codon near the 5’ end. These reads could either be products of random digestion of the abundant background of ncRNAs or be genuine MRPFs. We hypothesized that if these reads were the result of mitoribosomes initiating or attempting to initiate translation at the 5’ end of the ncRNAs, then we could use the same analysis of read length distribution at 5’ ends of mRNAs we performed above and determine of the sizes correlated to an AUG, AUU, GUG, or UUG as potential start codons. While GUG and UUG are not used in canonical mitochondrial initiation, they were included in the search because noncanonical initiation has been observed at GUG³⁶, and both are frequently utilized in bacterial initiation. We identified sites on four tRNAs that contained MRPFs with a size that correlated to either AUG or UUG codons (Figure 5A). The predicted ORFs and thus resulting potential translation events varied widely. On MT-TF, the start codon is followed by an immediate stop codon. The ORF on MT-TI is three codons long and terminates with the non-canonical mitochondrial termination codon AGA. On MT-TY, a 7-codon ORF is terminated by a UAA codon. MT-TS1 contains the largest open reading frame identified on tRNAs, at 20 codons, and it terminates with the U AGA sequence, as used in MT-CO1³⁷. Additionally, we identified a putative start site at the 5’ end of MT-RNR1 (Figure 5A), and mitoribosome profiling of cells treated with initiation inhibitors and digested with MNase shows similar footprint sizes (Figure S3). However, this site did not occur at a canonical AUG or AUU codon, but instead at a UUG codon. This putative start codon is followed by a 58-codon ORF and terminates at a UAG stop codon.

To determine whether translation elongation occurs at these putative ORFs, we determined the read phasing characteristic for MRPFs derived from translating ribosomes at these ORFs. However, the sites on MT-TF, MT-TI, and MT-TY are short and close to the 5’ end, hindering phasing analysis and thus do not allow us to conclude whether our sequence reads are indicative of translation. While the ORF on MT-TS1 is longer, MRPF density was low, and we did not observe read phasing and thus it is likely not translated (not shown). The reads mapping to the ORF on MT-RNR1 had a phasing profile similar to those mapping to the 13 canonical mitochondrial genes (Figure 5B). In contrast, reads mapping to the remainder of MT-RNR1 or mapping to the region of MT-RNR2 that codes for the putative mitochondrial micropeptide Humanin show no or conflicting read phasing. When compared to the amino acid composition of the mitochondrially-encoded proteome, the amino acid composition of the resulting peptide shows a stronger positive correlation (R²=0.534) (Figure 5C) to the remaining non-coding portion of MT-RNR1 (R²=0.241) (Figure 5D). Although mitoribosome profiling suggests that this putative ORF is translated, we did not detect any corresponding peptides in the protein mass spectrometry of isolated mitochondria. Nevertheless, our mass spectrometry dataset, or the MASSIVE database, does not cover many regions of the mitochondrial proteome, and thus, the absence of a spectrum is not necessarily evidence that the peptide is not translated. The read phasing analysis suggests this region of MT-RNR1 is translated and could encode a 15^th mitochondrial protein. Overall, this data shows that our mitoribosome profiling can be used as a tool to study non-canonical translation events resulting from ncRNAs.

Mitoribosome profiling was carried out in HEK293 Flp-In T-REx cells. 15 cm plates were treated with either DMSO as a control or 100 µg/mL of protein synthesis inhibitor for 1 hour, washed with 4°C PBS, and flash frozen by floating the plate on liquid nitrogen. Cells were collected by scraping on ice with 600 µL of lysis buffer (1.5x Lysis buffer (30 mM Tris–HCl pH 7.8, 150 mM KCl, 15 mM MgCl₂, 1.5 mM DTT, 1.5% Triton X-100, 0.15% NP40, 1 × complete phosphatase and protease inhibitors)) and lysed by passing through a 27-G needle 10 times. Lysate was centrifuged for 10 minutes at 21,000 g at 4°C, and the supernatant was collected. 300 µL of lysate was footprinted with either digested with 1,5000 U of MNase, 10 µL of SUPERase-In, and 5 mM of CaCl₂ at 22°C for 1 hour and stopped by adding EGTA to a concentration of 6 mM or with 8 U/µL of RNase I at 22°C for 1 hour and stopped by adding 10 µL of SUPERase-In. Titration of the digestion enzymes was not performed and is based on previous published work^29,50. The digestion was clarified by centrifugation at 21,000 g for 5 minutes. Lysates were loaded onto a 5%-30% sucrose gradient (20 mM Tris-HCl, pH 7.8, 100 mM KCl, 10mM MgCl₂, 1 mM DTT) and spun for 2.5 hours at 40,000 rpm in a SW40 rotor. 55S mitoribosome fractions were collected using a BioComp Fractionator and MRPFs were isolated by SDS/hot phenol/chloroform extraction.

Samples were run on a 15% TBE-Urea polyacrylamide gel and footprint sizes from 15-34 nts were excised and footprints were eluted in 0.3 M NaOAc, 1 mM EDTA, and 0.25% SDS followed by purification by Oligo Clean & Concentrator. Samples were treated with Quick CIP. 3’ adapters (5’-rAppNN-6 nt UMI-TGGAATTCTCGGGTGCCAAGG-L) were ligated using Rnl2(1-249)K227Q ligase for 2 hours at 16°C and pooled. Pooled samples were treated with T4 PNK for 30 minutes at 37°C. A 5’ chimeric DNA-RNA adapter (GTTCAGAGTTCTACAGTCCGACGATCrNrNrNrN) was ligated to the samples with T4 RNL1. Reverse transcription was carried out using SuperScript IV followed by PCR amplification by Taq DNA polymerase. Libraries were size selected using a 3% agarose Pippin prep and final libraries were PCR amplified. Samples were sequenced as either single- or paired-end 50 nt reads on either a NextSeq550 or NovaSeq.

Adapter sequences were removed with cutadapt and aligned with bowtie2 to a fasta containing rRNA and tRNA sequences. To identify stalling sites, reads were mapped to a custom mitochondrial transcriptome containing all 5’ UTRs, coding sequences, and 3’ UTRs for all processed mRNAs. Reads were aligned with bowtie2. A 13-nt P-site offset was assigned using Plastid and mapping reads relative to the start codons of ND4, ATP6, MT-ND5-dORF. We determined the relative ribosome density by normalizing the number of reads at each codon by the total number of reads mapping to its respective gene for each sample. Fold change was then calculated by dividing the relative ribosome density in a treated sample by that in an untreated sample, excluding sites that carried less than 0.1% of the total ribosome density of each gene. To identify putative novel translation initiation sites, we mapped our footprints to the entire mitochondrial genome and visually inspected the 3’ ends of mitochondrial tRNAs and rRNAs.

Mitochondria were isolated from 2 × 10⁷ HEK293 cells using the Mitochondria Isolation Kit for Cultured Cells (Thermo Scientific). Mitochondrial proteins were processed for trypsin/LysC digestion, and the digested peptide mixture was then concentrated and desalted using C18 column. Reconstituted desalted peptides in 25 μl of 0.1% formic acid. 12 μl of peptides from each sample was analyzed by a 110 min LC/MS/MS run. The LC/MS/MS analysis of tryptic peptides for each sample was performed sequentially with a blank run between each two sample runs using a Thermo Scientific Orbitrap Exploris 240 Mass Spectrometer and a Thermo Dionex UltiMate 3000 RSLCnano System. Peptides from trypsin digestion were loaded onto a peptide trap cartridge at a flow rate of 5 μL/min. The trapped peptides were eluted onto a reversed-phase Easy-Spray Column PepMap RSLC, C18, 2 μM, 100A, 75 μm × 250 mm (Thermo Scientific) using a linear gradient of acetonitrile (3-36%) in 0.1% formic acid. The elution duration was 60 min at a flow rate of 0.3 μl/min. Eluted peptides from the Easy-Spray column were ionized and sprayed into the mass spectrometer, using a Nano Easy-Spray Ion Source (Thermo) under the following settings: spray voltage, 1.6 kV, Capillary temperature, 275°C. Other settings were empirically determined. Raw data files were searched against the human protein sequences database and the client target protein sequences database using the Proteome Discoverer 2.5 software (Thermo, San Jose, CA) based on the SEQUEST algorithm. Carbamidomethylation (+57.021 Da) of cysteines was set as a fixed modification, and Oxidation / +15.995 Da (M), Phospho / +79.966 Da (S, T, Y), and Deamidated / +0.984 Da (N, Q) were set as dynamic modifications. The minimum peptide length was specified to be five amino acids. The precursor mass tolerance was set to 15 ppm, whereas the fragment mass tolerance was set to 0.05 Da. The maximum false peptide discovery rate was specified as 0.01. The resulting Proteome Discoverer Report contains all assembled proteins with peptides sequences and peptide spectrum match counts (PSM#) and MS1 peak area relative abundance.