For in vitro translation initiation experiments, suspension-adapted and tetracycline-resistant human embryonic kidney cells lacking N-acetyl-glucosaminyltransferase I (HEK293S TetR GnTI−) were used. First, cells were grown adherently in T175 plates in DMEM + 10% FBS and 1× penicillin/streptomycin. Afterwards, 9 confluent plates were transferred to 500 ml FreeStyle 293 Expression Medium (Gibco, Life Technologies), supplemented with 5% fetal bovine serum (Gibco, Life Technologies) and 1× penicillin/streptomycin, and further cultured in 3 l Corning disposable spinner flask. The cells were passaged at a density of 3 × 10⁶ cells/ml to a density of 1.5 × 10⁶ cells/ml, by centrifugation at 1000 g for 10 min and complete exchange of the culture media. For mitoribosome preparation, a 6 l cell culture with a density of 6 × 10⁶ cells/ml was used.

To generate cell lines stably expressing the FLAG-tagged version of mtIF3 and mtIF2, an Flp-In T-Rex human embryonic kidney 293T (HEK293T, Invitrogen) cell line that permits doxycycline-inducible expression of the gene of interest in a dose-dependent manner was used. HEK293T cells were cultured at 37°C under 5% CO₂ in DMEM (Dulbecco's modified Eagle's medium) supplemented with 10% (v/v) tetracycline-free fetal bovine serum (FBS), 1× penicillin/streptomycin (Gibco), 50 μg/ml uridine, 100 μg/ml zeocin (Invitrogen), and 10 μg/ml blasticidin S (Gibco). Parental HEK293T Flp-In T-Rex cells were seeded in a 10 cm dish, one day prior to transfection, to achieve 70–90% confluency. Co-transfection of the expression plasmid pcDNA5/FRT/TO-mtIF3-FLAG or mtIF2-FLAG and the Flp-recombinase plasmid, pOG44, was carried out using Lipofectamine 3000 (Invitrogen), according to the manufacturer's instructions. Selective antibiotics, 100 μg/ml of hygromycin (Invitrogen) and 10 μg/ml blasticidin S, were added 48 h post-transfection and culture media was replaced every 2–3 days.

The stable knock-out (KO) lines in HEK293T cells were established using the CRISPR/Cas9 system (23). Small guide RNA (sgRNA) pairs (Supplementary Table S1) were designed to generate an out-of-frame deletion and cloned into pSpCas9(BB)-2A-Puro (PX459) V2.0 vector. HEK293T cells grown in a six-well plate were transfected at 80% confluency using Lipofectamine 3000, as per the manufacturer's instructions. After 48 h of puromycin selection (1.5 μg/ml), cells were collected and serially-diluted single cells were plated onto multiple 96-well plates. The obtained clones were further screened by PCR and immunoblotting for verification of the targeted deletion and absence of the targeted protein, respectively.

To test viability of the WT, mtIF2 and mtIF3 knock-out cell lines in galactose, cells were seeded at a density of 50 000 cells/well of a six-well plate. The galactose media constituted DMEM without glucose (Invitrogen) supplemented with 0.9 g/l galactose, 2 mM GlutaMAX (Gibco), 1× sodium pyruvate, 10% (v/v) FBS and 1× penicillin/streptomycin. The cells were stained with trypan blue and counted at 48 h intervals up to 6 days with EVE™ Automated Cell Counter (NanoEnTek), according to the manufacturer's instructions.

Human mitoribosomes were purified according to previously described protocols (24) with modifications. Briefly, HEK293 cells were pelleted by centrifugation at 1000 g for 10 min and washed with cold PBS. The pellet was then dissolved in Swelling Buffer (25 mM HEPES/KOH pH 7.5, 100 mM KCl, 20 mM Mg(OAc)₂, 2 mM DTT) for 20 min at 4°C. Afterwards, the buffer was supplemented with sucrose and mannitol to a final concentration of 70 mM sucrose and 210 mM mannitol, and the cells were disrupted with 10 strokes in a 100 ml glass homogenizer. Nuclei and cell debris were pelleted by centrifugation at 1000 g for 10 min, and crude mitochondria were subsequently pelleted by centrifugation at 10 000 g for 10 min. Mitochondria were dissolved in Mitochondria Isolation Buffer (25 mM HEPES/KOH pH 7.5, 100 mM KCl, 20 mM MgOAc, 70 mM sucrose, 210 mM mannitol, 2 mM DTT, supplemented with cOmplete protease inhibitors) and treated with 10 U/ml Dnase I for 20 min on ice. After washing with Mitochondrial Isolation Buffer, 2 volumes of Lysis Buffer (25 mM HEPES–KOH pH 7.45, 100 mM KCl, 20 mM MgOAc, 2% Triton X-100, 2 mM DTT supplemented with cOmplete protease inhibitors and RNase inhibitors) were added and incubated on ice for 15 min at 4°C. The membranes were centrifuged at 30 000 g for 20 min, and the mitochondrial lysate was afterwards added on top of a 10–30% sucrose gradient in Ribosome Isolation Buffer (25 mM HEPES/KOH pH 7.5, 100 mM KCl, 20 mM MgOAc, 2 mM DTT) and centrifuged for 21 h at 21 000 rpm in an SW21 rotor. The gradients were then fractionated with a Biocomp Fractionator (Supplementary Figure S1D, I), 55S monosomes were pelleted and dissolved in Dissociation Buffer (20 mM HEPES/KOH pH 7.6, 300 mM KCl, 5 mM MgCl₂, 1 mM DTT) for 5 h at 4°C, overlaid on top of a 10–30% sucrose gradient prepared in Dissociation Buffer, and centrifuged in an SW41 rotor for 21 h at 21 000 rpm. The gradient was subsequently fractionated using a Biocomp Piston Fractionator (Supplementary Figure S1D, II), the samples corresponding to the small and large ribosomal subunit were pooled individually and centrifuged in an SW60 rotor for 16 h at 55 000 rpm. To prepare re-associated ribosomes, the pellets corresponding to mitoribosomal subunits were dissolved in Re-association Buffer (20 mM HEPES–KOH pH 7.6, 30 mM KCl, 30 mM MgCl₂, 1 mM DTT). The small and large ribosomal subunits were then mixed in a 1/1 ratio and incubated for 1 h at 37°C and 15 min on ice. The re-association reaction was overlaid on top of a 10–30% sucrose gradient prepared in the Re-association Buffer and centrifuged in an SW41-Ti rotor for 21 h at 21 000 rpm. The gradient was fractionated using a Biocomp Piston Fractionator (Supplementary Figure S1D, III), and the samples corresponding to the monosome were pooled and centrifuged in a TLA100.4 rotor for 16 h at 55 000 rpm.

Codon-optimized (Genscript) DNA constructs corresponding to the mature form of human mtIF2 (amino acids 38–727) and human mtIF3 (amino acids 32–278) were cloned into a pET-24b vector (Novagen). Both constructs were expressed in Rosetta 2 cells (EMD chemicals) at 25°C for 16 h in Magic Media (Thermo Fisher Scientific). After lysis, the proteins were purified over a His-Select Ni²⁺ resin (Sigma-Aldrich) and dialyzed against H-0.2 (25 mM Tris–HCl pH 7.4, 0.5 mM EDTA, 10% glycerol, 1 mM dithiothreitol, 200 mM NaCl). Further purification was conducted over a HiLoad 16/60 Superdex 200 pg gel filtration column (GE Healthcare) in buffer H-0.2 lacking glycerol.

C-terminal His-tagged E. coli EF-Tu was cloned into a pET15b plasmid and overexpressed in BL21(DE3) cells for 3 h at 37°C. The cells were resuspended in EF-Tu purification buffer A (50 mM Tris–HCl pH 7.6, 60 mM NH4 Cl, 7 mM MgCl₂, 7 mM β-mercaptoethanol, 15 μM GDP, 15% glycerol) and lysed by sonication. After centrifugation at 20 000 rpm in an SS34 rotor, the supernatant was loaded onto a Co(II)-Sepharose resin (bed volume 3 ml), pre-equilibrated with EF-Tu purification buffer A, and incubated for 1 h at 4°C. The column was washed twice with 30 ml of EF-Tu purification buffer A,10 ml of EF-Tu purification buffer B (50 mM Tris–HCl pH 7.6, 60 mM NH4Cl, 7 mM MgCl₂, 300 mM KCl, 7 mM β-mercaptoethanol, 15 μM GDP, 15% glycerol) and 10 ml of EF-Tu purification buffer C (50 mM Tris–HCl pH 7.6, 60 mM NH4Cl, 7 mM MgCl₂, 300 mM KCl, 20 mM imidazole, 7 mM β-mercaptoethanol, 15 μM GDP, 15% glycerol). Subsequently, EF-Tu was eluted using 10 ml EF-Tu elution buffer (50 mM Tris–HCl pH 7.6, 60 mM NH4Cl, 7 mM MgCl₂, 300 mM KCl, 300 mM imidazole, 7 mM β-mercaptoethanol, 15 μM GDP, 15% glycerol). The sample was dialyzed overnight against EF-Tu storage buffer (50 mM Tris–HCl pH 7.6, 100 mM NaCl, 10 mM MgCl₂, 15μM GDP) and concentrated using Amicon Ultra ultrafiltration units (MWCO: 10,000).

The first 30 nucleotides of mitochondrially encoded cytochrome C oxidase 2 (MTCO2) mRNA with the sequence: 5′-AUGGCACAUGCAGCGCAAGUAGGUCUACAA -3′, labeled with Cy3 at the 3’ end, was purchased from Sigma-Aldrich.

The leaderless mRNA with the sequence 5’ AUG UUC AGA CAA GCG CAA GUA GGU CUA CAA 3’ used for dipeptide synthesis was ordered from Integrated DNA Technologies.

The unlabeled Escherichia coli, fMet-tRNA^fMet was generously provided by Prof. Marina Rodnina. E. coli (³⁵S) fMet-tRNA^fMet was isolated from bulk E. coli MRE600 by hybridization to immobilised complementary oligoDNA. Yeast tRNA^Phe was purchased from Sigma. Charging with the cognate fMet and Phe amino acids, respectively, was performed as previously described (25).

All in vitro translation assays were carried out in assay buffer containing 50 mM Tris–HCl, pH 7.6, 30 mM KCl, 10 mM MgCl₂, 5 mM GTP, 1 mM dithiothreitol DTT, 1mM spermidine, 0.1 mM spermine.

Reactions containing 200 nM 55S mitoribosomes or 28S subunits were supplemented with 240 nM mtIF3, 240 nM mtIF2, 240 nM E. coli fMet-tRNA^fMet and 400 nM MTCO2 mRNA labeled with Cy3 at the 3’ end, incubated at 37°C for 30 min and on ice for 15 min. To purify the initiation complex, the reaction mixture was overlaid on top of a 10–30% sucrose gradient and centrifuged for 21 h at 21 000 rpm in an SW41 rotor. The optical density of the gradients was measured at 260 nm, and the presence of the Cy3-labeled mRNA was monitored by fluorescence spectroscopy using the same settings as described below (Fluorescence measurements and analysis). To qualitatively analyze the binding of the mRNA to ribosomal complexes resolved in the sucrose gradient, the absorbance at 260 nm and maximum fluorescence emission values of each fraction were normalized to the values detected for the first collected fraction. The fractions corresponding to the 55S ribosomes or 28S subunits peaks, identified by absorbance at 260 nm, were pooled and concentrated to 50 μl using a Vivaspin 500 centrifugal concentrator. Initiation complex formation was calculated based on molar concentrations of mitoribosomes (determined from absorption at 260 nm) and mRNA (as described in ‘Fluorescence measurements and analysis’).

For initiation reaction with crosslinked mitoribosomes, no mtIF3 was added as mtIF3 has been shown to possess partial dissociation activity (26).

Ribosome complexes (50 μl) containing 200 nM 55S ribosomes or 28S subunits were assembled with (³H) E. coli fMet-tRNA^fMet in the same conditions as described above unless otherwise described. After incubation for 30 min at 37°C, the reactions were directly applied to nitrocellulose filters, which were then washed with 3 ml ice-cold assay buffer containing 20 mM Mg²⁺ to wash unbound (³H) fMet-tRNA^fMet tRNAs. The filters were dried for 15 min at 50°C and the radioactivity was measured in a liquid scintillation counter. A background corresponding to reactions in absence of mRNA was subtracted from each measured value.

To prepare dipeptide fMet-Phe, elongation reactions contained 0.2 μM 55S ribosomes, 1μM (³⁵S) E. coli fMet-tRNA^Met and (³H) Saccharomyces cerevisiae Phe-tRNA^Phe, 1 μM EF-Tu. After 1 h incubation at 37°C, the reaction was overlayed on top of a 1.1 M sucrose in assay buffer and centrifuged at 110 000 rpm for 2 h in an S140-AT rotor. The resulting pellet was washed with 1 ml cold assay buffer and dissolved in 20 μl assay buffer. The concentration of ribosomes was determined by measuring the absorption at 260 nm, and the ³⁵S and ³H signal were measured in a liquid scintillator counter.

The in vitro transcribed MTCO2 mRNA (prepared as in (15)) was biotinylated by enzymatically attaching biotinylated pCp at the 3′end using the T4 RNA ligase. To this end, 100 ng MTCO2 mRNA was incubated with a 10 times excess of bio-pCp and 1 U/μl T4 RNA Ligase, in 1× Ligase Buffer, for 16 h at 16°C. Next, the bioMTCO2 mRNA was cleaned with an RNeasy Mini Kit, and the concentration of the purified construct was measured spectroscopically with a Nanodrop.

Samples containing 50 nM Cy5 55S monosomes or 50 nM Cy5 28S subunits were incubated with 75 nM bioMTCO2 mRNA, in the same conditions as the in vitro initiation assays, with the exception that the incubation time was increased to 2 h. The initiation reactions were then incubated with 20 μl streptavidin-coated magnetic beads for 16 h at 18°C to allow binding of the biotinylated initiation complex to the beads. Next, the beads were washed three times with Initiation Buffer, and finally treated with 10 μg/ml RNase A for 1 h at 37°C to release the RNA bound to the magnetic beads. Cy5 fluorescence was afterwards measured in the Flow Through, Wash, and Eluate.

Fluorescence spectra of the Cy3 fluorophore were measured with an FL-7000 fluorometer, at 25°C, in a 50 μl cuvette using the settings: excitation wavelength 620 nm, emission wavelength 625–800 nm, excitation and emission slits: 2.5 nm. The recorded spectra were corrected by subtracting the spectra of a reaction prepared in the same conditions, using unlabeled components. In order to convert the fluorescence intensity into concentration, a calibration curve was made using solutions of Cy3-labeled MTCO2 mRNA of known concentrations, measured using the same settings (Supplementary Figure S1E).

Confocal fluorescence detection was performed on a MicroTime 200 microscope (PicoQuant). Cy3 and Cy5 fluorophores were excited with two pulsed diode lasers of 510 nm (LDH-D-C-510) and 640 nm (LDH-D-C-640) with an average power of 20 μW and an UPLSAPO 60x/1.2NA objective (Olympus, Shinjuku, Japan). Fluorescence photons passed through a dual-band dichroic mirror (ZT514/640RPC, Chroma Technology, USA), 75 μm pinhole and were split into two detection pathways by a dichroic mirror (T620LPXR, Croma Technology). The red fluorescence emission passed through a bandpass emission filter (685/80, Chroma Technology) and was collected by a single-photon avalanche photodiode (τ-SPAD, PicoQuant). The green fluorescence emission passed through a bandpass filter (585/65, Chroma Technology) and was collected by a single-photon avalanche photodiode (COUNT T, Laser Components, USA).

First, we confirmed that the Cy3-MTCO2 mRNA in the initiation reaction was bound to the mitoribosome using Fluorescence Correlation Spectroscopy (FCS) and compared the theoretical diffusion coefficients of the free Cy3, Cy3-MTCO2 mRNA and Cy3-mitoribosome (Supplementary Figure S2C) with the measured diffusion coefficients (Supplementary Figure S2D).

Single-molecule brightness-gated-two-color coincidence detection (BTCCD) was measured under picomolar concentration, the detection probability of a single molecule was N ≤ 0.01 to avoid chance coincidence detection. The concentration prior to each dilution was checked with FCS. Sample aliquots were changed every 40 min to minimize dissociation contribution. Total acquisition time for the single-molecule data set was between 2 and 4 h. All measurements were made using PEGylated glass slides to rule out unspecific binding of molecules and maintain constant concentration during BTCCD measurements. More details on BTCCD method is provided in the Supplementary Note.

Cell lysates were solubilized in lysis buffer (50 mM Tris–HCl, pH 7.4 with 150 mM NaCl, 1 mM EDTA, 1% Triton X-100) supplemented with 1× protease inhibitor cocktail (Roche) and 5 mM MgCl₂. Protein concentrations were measured using Pierce™ BCA Protein Assay Kit (Thermo Fisher Scientific). Equal amounts of proteins from the total cell lysates and protein fractions were separated on a 4–12% SDS-PAGE and transferred to the PVDF membranes by wet transfer. The membranes were blocked for 1 h using 5% non-fat Milk (Semper) with PBS and incubated overnight at 4°C with a specific primary antibody, before detection the following day with secondary HRP-conjugates using ECL (BioRad) and films or digitally using ChemiDoc Imaging System (BioRad). The list of primary and secondary antibodies is summarized in Supplementary Table S2.

In vivo labeling of the mitochondrial translation products was performed by incubating the cells in growth media lacking methionine and cysteine supplemented with 44 μCi/ml of [³⁵S]-methionine-cysteine EasyTag™ EXPRESS35S Protein Labeling Mix (Elmer-Perkin), 10% dialyzed FBS, 100× pyruvate (Gibco), 100× glutamine (Gibco) along with 100 μg/ml emetine dihydrochloride (Sigma Aldrich) to inhibit the cytosolic protein synthesis. Briefly, following 30 min incubation of cells with [³⁵S]-methionine at 37°C, labeled cells were pelleted, washed with PBS and lysed. Equal amounts of proteins (40 μg) were separated on 12% SDS-PAGE, which were then dried (2 h at 65°C) for exposure to a Phosphor screen before scanning with a Typhoon FLA 7000 (GE Healthcare) for signal quantification.

Mitochondria from the WT HEK293T cells, mtIF2 and mtIF3 KO cell lines, and cells stably expressing mtIF3.FLAG were isolated as previously described (27). Briefly, cells were harvested by centrifugation at 400 g for 3 min and swollen in an ice-cold hypotonic buffer (0.6 M mannitol, 100 mM Tris–HCl pH 7.5, 10 mM EDTA, 0.05% BSA). Additional disruption of the cells was performed mechanically with glass homogenizer. The solution was centrifuged at 400 g for 10 min to remove cell debris and nuclei from the mitochondria. Supernatant was collected and centrifuged at 11 000 g for 10 min. At this point, crude mitochondrial pellets were frozen at –80°C or used directly for the analysis.

The sedimentation of mitoribosomes was analyzed using sucrose density gradients. The isolated mitochondria were solubilized with the lysis buffer (10 mM Tris–HCl, pH 7.4, 50 mM KCl, 20 mM MgCl₂, 1% Triton X-100, 1× protease inhibitor cocktail (Roche), 40 U/μl RNase inhibitor (Agilent). 700 μg of protein was loaded onto a linear 10–30% gradient containing 10 mM Tris–HCl, 100 mM KCl and 20 mM MgCl₂ supplemented with 1x protease inhibitor cocktail (Roche) and centrifuged at 79 000 g for 15 h at 4°C (Beckman Coulter; SW41 rotor). The gradients were collected from the top into 20 fractions of 450 μl/fraction, and fractions 1–18 were analyzed by immunoblotting, with fractions 1 and 2 pooled together.

Total RNA was purified from the whole cells using TRIzol reagent as per manufacturer's instructions and quantified via NanoDrop ND-1000 UV-Vis Spectrophotometer (Thermo Fisher Scientific). Reverse transcription to initiate cDNA synthesis for qRT-PCR analysis was carried out using a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). cDNA (5–10 ng) was mixed with 1× TaqMan Universal PCR Master Mix (Applied Biosystems) and universal TaqMan-based probes (Applied Biosystems, Supplementary Table S3). The qRT-PCR was performed on a QuantStudio 6 Flex Real-Time PCR System (Applied Biosystems). The recommendations of the manufacturer were followed when performing the reverse transcription and real time-PCR.

Mitochondrial pellets were resuspended in aminocaproic buffer (1.5 M aminocaproic acid, 50 mM Bis–Tris, pH 7.0), quantified using Qubit (Invitrogen) and treated with 4% (w/v) digitonin. The lysate was incubated for 20 min on ice and centrifuged to remove insolubilized material. Equal amounts of protein (50 μg) were separated on pre-cast 3–12% Bis–Tris blue native gels (Invitrogen). In-gel catalytic activity assays were performed for mitochondrial respiratory complex I (2 mM Tris–HCl pH 7.4, 0.1 mg/ml NADH and 2.5 mg/ml iodonitrotetrazolium chloride), complex II (4.5 mM EDTA, 0.2 mM phenazine methosulfate, 84 mM succinic acid, 2.5 mg/ml iodonitrotetrazolium chloride), complex IV (50 mM phosphate buffer pH 7.4, 0.5 mg/ml 3,3’-diamidobenzidine tetrahydrochloride (DAB), 1 mg/ml cytochrome c, 0.2 M sucrose and 20 μg/ml catalase) and complex V (3.76 mg/ml glycine, 5 mM MgCl₂, 1% Triton X-100, 0.5 mg/ml lead nitrate, 2 mM ATP, pH 8.4). Staining of the gels was carried out at room temperature.

The Seahorse XFe96 Analyzer (Seahorse Bioscience) was used to measure the oxygen consumption rate. Cells (3 × 10⁵ cells per well) were plated in the XF 96-well cell culture microplate coated with poly-d-lysine (Sigma) and grown for 24 h in a standard medium. One hour before the assay, the growth medium was replaced with 40 μl of Seahorse XF DMEM medium and the plate was left to stabilize in a 37°C non-CO₂ incubator. The oxygen consumption rate (OCR) was measured for each well three times, every 3 min. OCR was normalized to protein concentration following the BCA assay conducted according to the manufacturer's instructions.

The ribosome profiling experiment was performed as described previously (28), with minor modifications. During the separation of ribosome-protected fragments (RPFs) on 15% denaturing polyacrylamide gel, RNA species migrating between 10 and 50 nt were excised from the gel.

Illumina adaptor sequences were trimmed from reads using the FASTX-Toolkit (http://hannonlab.cshl.edu/fastx_toolkit/). Trimmed reads shorter than 30 nt, or longer than 35 nt in length, were discarded. The remaining reads were mapped sequentially to nuclear-encoded ribosomal RNA (rRNA); mitochondrial rRNA (mt-rRNA); mitochondrial tRNA (mt-tRNA); mitochondrial mRNA (mt-mRNA); other nuclear non-coding RNA (ncRNA); and the nuclear genome. Read mapping was performed using bowtie version 1 (29), with parameters -v 2 –best (i.e. maximum two mismatches, report best match).

To facilitate the direct comparison of mitochondrial gene expression in WT and mtIF3 KO¹ libraries, counts of mitochondrial ribosome-protected fragments (mt-RPFs) per gene in a given library were expressed as a percentage of total mtRPFs in that library. Only mtRPFs with 5′ ends mapping in the positive-sense orientation between the first nucleotide of the start codon and 30 nt 5′ of the stop codon were counted, and regions where ORFs overlap in the bicistronic ATP8/ATP6 and ND4L/ND4 transcripts were excluded from the expression calculations due to ambiguous assignment of RPFs to ORFs in these regions.

Identification and quantification of co-immunoprecipitated proteins was carried out as described previously (27). Peptides were separated on a 25 cm, 75 μm internal diameter PicoFrit analytical column (New Objective) packed with 1.9 μm ReproSil-Pur 120 C18-AQ media (Dr. Maisch,) using an EASY-nLC 1200 (Thermo Fisher Scientific). The column was maintained at 50°C. Buffer A and B were 0.1% formic acid in water and 0.1% formic acid in 80% acetonitrile. Peptides were separated on a segmented gradient from 6% to 31% buffer B for 45 min and from 31% to 50% buffer B for 5 min at 200 nl/min. Eluting peptides were analyzed on QExactive HF mass spectrometer (Thermo Fisher Scientific). Peptide precursor m/z measurements were carried out at 60 000 resolution in the 300–1800 m/z range. The 10 most intense precursors with charge state from 2 to 7 only were selected for HCD fragmentation using 25% normalized collision energy. The m/z values of the peptide fragments were measured at a resolution of 30 000 using a minimum AGC target of 2e5 and 80 ms maximum injection time. Upon fragmentation, precursors were put on a dynamic exclusion list for 45 s. The raw data were analyzed with MaxQuant version 1.6.1.0 (30) using the integrated Andromeda search engine (31). Peptide fragmentation spectra were searched against the canonical sequences of the human reference proteome (proteome ID UP000005640, downloaded September 2018 from UniProt). Methionine oxidation and protein N-terminal acetylation were set as variable modifications; cysteine carbamidomethylation was set as fixed modification. The digestion parameters were set to ‘specific’ and ‘Trypsin/P,’ The minimum number of peptides and razor peptides for protein identification was 1; the minimum number of unique peptides was 0. Protein identification was performed at a peptide spectrum matches and protein false discovery rate of 0.01. The ‘second peptide’ option was on. Successful identifications were transferred between the different raw files using the ‘Match between runs’ option. Label-free quantification (LFQ) (32) was performed using an LFQ minimum ratio count of two. LFQ intensities were filtered for at least two valid values in at least one group and imputed from a normal distribution with a width of 0.3 and down shift of 1.8. Protein quantification was performed using limma (33). Mitocarta (34) annotations were added using the primary gene name and the first of the gene name synonyms of the oldest Uniprot ID with the highest number of peptides.

Statistical significance was analyzed using Microsoft Excel, and a two-tailed unpaired Student's t-test was performed to identify statistically significant differences between the groups. The significance threshold was set at P < 0.05; indicated as * for P < 0.05, ** for P < 0.01 and *** for P < 0.001. The experimental results are presented as mean ± standard deviation (SD).

We developed an in vitro reconstituted mitochondrial translation initiation system, using mammalian mitochondrial ribosomes, mitochondrial translation initiation factors mtIF2 and mtIF3, E. coli fMet-tRNA^fMet and a leaderless transcript (MTCO2 encoding the first 30 ORF nucleotides), which contains the start codon AUG at the 5′ end. As a source of mammalian mitochondrial ribosomes, we used HEK293 cells and bovine liver tissue. We first analyzed the translation initiation efficiency of mammalian mitoribosomes by quantifying the binding of fluorescently labeled mRNA to human 55S mitoribosomes and 28S subunits. In addition, we quantified the initiation complex formation by measuring the binding of [³H] labeled initiator fMet-tRNA^fMet to bovine liver 55S mitoribosomes and 28S subunits. We observed similar translation initiation activity for both sources of mitoribosomes (Figure 1B and C).

We used two different procedures for the purification of 55S monosomes: crude ribosomes were prepared from mammalian mitochondria using a sucrose gradient (Supplementary Figure S1A), and re-associated ribosomes were isolated by dissociation of crude ribosomes and re-association of the purified small and large subunits (Supplementary Figure S1A, D). Crude ribosomes contain native co-purified mitochondrial tRNAs and mRNAs tRNAs, while re-association affords purification of empty mitoribosomes.

To observe binding of the mRNA to human mitoribosomes, we used an mRNA labeled with Cy3 at the 3′ end. After purification of the initiation complex through a sucrose gradient, the presence of ribosomes and mRNA was detected and the fraction of ribosomes carrying a Cy3-labeled mRNA was determined (Figure 1A). We observed that 40% of human crude 55S ribosomes carry a labeled mRNA (Figure 1B). In contrast, using in vitro re-associated 55S monosomes, led to a higher percentage of active mitoribosomes (∼50%) (Figure 1B), suggesting that natively-purified mitoribosomes may carry endogenous mRNA, tRNAs, and/or translational factors.

The presence of fMet-tRNA^fMet was essential for mRNA binding to 55S (Figure 1B). Moreover, we did not observe binding of an mRNA lacking the start codon to 55S ribosomes, proving that the AUG start codon at the 5′ is the unique binding site (Figure 1B). Next, we investigated whether fluorescent labeling of mitoribosomes had any impact on their binding affinity to MTCO2 mRNA (labeled ribosomes were needed for the following experiments, as described below). After randomly labeling accessible lysines with NHS-functionalized Cy5 (Supplementary Figure S1B), mitoribosomes retained initiation activity (Figure 1B), indicating that the labeling conditions and presence of the Cy5 fluorophore did not inhibit recruitment of the mRNA. Importantly, we did not detect leaderless mRNA bound to the 28S subunit under the same conditions (Figure 1B).

To further investigate if leaderless mRNA can bind to 55S ribosomes, without their dissociation, we tested initiation activity of human mitoribosomes with covalently crosslinked subunits. Following crosslinking of accessible lysine residues (as described in (18)), most mitoribosomes remained as intact monosomes in dissociation buffer containing low Mg²⁺ and high K⁺ concentrations, while non-modified 55S mitoribosomes efficiently dissociated in these conditions (Supplementary Figure S1C). Following crosslinking, the sample containing 55S monosomes which did not dissociate in these conditions was pelleted and the MTCO2 mRNA binding efficiency was determined, as described above. The crosslinked mitoribosomes retained initiation activity, although they were only half as active as re-associated mitoribosomes (Figure 1B, 25% versus 50%). This loss in activity is not unexpected, since the random crosslinker is likely to bind the ribosomal subunits in more than one location, with some of the introduced modifications potentially targeting regions essential for translation initiation.

We next performed additional filter-binding assays measuring mRNA-dependent [³H]-fMet-tRNA^fMet binding to mitoribosomes to form an initiation complex formed with 55S ribosomes and co-sedimentation experiments to determine whether the initiation complex is elongation-competent. In these experiments we used bovine liver as a source of mitochondria, which allows convenient purification of sufficient amounts of mitoribosomes. In addition, since it is currently challenging to purify and aminoacylate mammalian mt-tRNAs in the amounts required for in vitro studies, we used E. coli [³⁵S]-fMet-tRNA^fMet and S. cerevisiae [³H]-Phe-tRNA^Phe. Here it is important to note that although mt-Met-tRNA^Met and mt-Phe-tRNA^Phe do contain mt-specific modifications, they each maintain a classical secondary cloverleaf structure, including conserved tertiary interactions between the D-stem and variable loop (35).

We found a much higher binding efficiency of [³H]-fMet-tRNA^fMet to crude 55S ribosomes than to 28S subunits (Figure 1C), paralleling the mRNA binding results and in accord with the results obtained with the cross-linked 55S ribosomes. We also used the filter-binding assay to determine the in vitro role of mtIF2 and mtIF3 in selection of the [³H]-fMet-tRNA^fMet by both crude and re-associated 55S ribosomes. Omission of mtIF2 severely inhibited [³H]-fMet-tRNA^fMet binding to monosomes, showing that mtIF2 is crucial for the formation of the mitochondrial initiation complex. Omission of mtIF3 did not decrease initiation complex formation on reassociated ribosomes (Figure 1C). However, crude monosomes, which contain native mitochondrial tRNAs and mRNAs, showed a decrease in initiation activity in absence of mtIF3 (Figure 1C). This suggests that mtIF3 might facilitate the recycling of 55S crude ribosomal complexes by removing the native bound tRNAs and mRNAs before a subsequent initiation event.

Next, we determined whether the initiation complex formed with 55S ribosomes programmed with a 30-nucleotide leaderless mRNA (AUG UUC AGA CAA GCG CAA GUA GGU CUA CAA) and [³⁵S]-fMet-tRNA^fMet is elongation-competent, by testing its ability to form ribosome-bound fMet-Phe-tRNA^Phe on mixing with [³H]-Phe-tRNA^Phe and recombinant E. coli elongation factor Tu (EF-Tu). Quantification of the [³⁵S]-fMet and [³H]-Phe co-sedimenting with 55S ribosomes, the latter corrected by subtraction of a small background value determined in the absence of [³⁵S]-fMet-tRNA^fMet, showed that that 27% of 55S ribosomes contained [³⁵S]-fMet-tRNA^fMet and 22% contained [³H]- Phe-tRNA^Phe, indicating that ∼80% of the 55S initiation complex is ribosomes was active in EF-Tu-dependent dipeptide bond formation (Figure 1D).

To exclude the possibility that the sucrose gradient centrifugation or the limited size (30 nucleotides) of the mRNA accounted for our failure to observe formation of the initiation complex on the 28S subunits, we performed initiation reactions with Cy5-labeled human 55S monosomes or Cy5-labeled human 28S subunits with an in vitro transcribed full-length MTCO2 mRNA. These experiments employed MTCO2 mRNA biotinylated at the 3′ end (bioMTCO2), allowing initiation complexes to be purified using streptavidin-coated magnetic beads. A negative control with a bioMTCO2 mRNA lacking the start codon was used to determine the non-specific binding of the transcript to ribosomes. The fluorescence intensity measured in the flow-through, wash and eluate, normalized to the total fluorescence intensity of each sample, revealed that Cy5-55S monosomes were recruited on the bioMTCO2 mRNA, while the negative control showed no binding of Cy5-55S monosomes (Supplementary Figure S1F). In contrast, the Cy5-labeled 28S did not bind to any of the bioMTCO2 mRNA constructs, indicating that, in our assay conditions, the mtSSU has a much lower affinity for the full MTCO2 mRNA than 55S monosomes. Together, our in vitro experiments demonstrate that mitochondrial leaderless mRNAs bind 55S monosomes more efficiently than the 28S subunits. Moreover, initiator fMet-tRNA^fMet is essential for loading mRNA onto the full monosome to form an elongation-competent initiation complex.

Recently, a Brightness-Gated Two Color Coincidence Detection (BTCCD) method was developed to detect and quantify the fraction of associated biomolecules in solution, using standard confocal detection of fluorophore-labeled ribosomal subunits and mRNAs (Figure 2A, Supplementary Note 1) (36). This method was applied to quantify the fraction of bacterial ribosomes that dissociate during translation initiation in vitro (36), showing that most of the initiation events in bacteria occurred via the canonical dissociative mechanism (30S binding initiation). Here, we used a similar approach to investigate the frequency of exchange of mitoribosomal subunits during translation initiation in the mitochondrial in vitro system. For this purpose, we used BTCCD to determine the fraction of unlabeled human re-associated mitoribosomes, which exchanged the unlabeled large subunit during translation initiation of the Cy3-labeled MTCO2 mRNA in presence of a 5-fold excess of Cy5-labeled mtLSU, relative to the 55S monosomes.

To estimate the maximum detected Cy3 and Cy5 coincidence in our system, we used a positive control consisting of Cy5-labeled 55S ribosomes that initiated Cy3-labeled MTCO2 mRNA (Figure 2B). The 55S initiation complex was purified through a sucrose gradient to remove the excess of unbound Cy3-MTCO2 mRNA (Supplementary Figure S2 A), and the fraction of coincident Cy5 and Cy3 bursts was determined to be 43 ± 4% (Figure 2E, Supplementary Figure S2E). This positive control accounted for incomplete labeling of the MTCO2 mRNA, bleaching, and reversible dissociation of the mRNA from the mitoribosome due to dilution of the sample to picomolar concentration, possibly below K_d during single-molecule measurements.

A negative control in which Cy5-labeled 55S mitoribosomes were incubated with Cy3-labeled MTCO2 mRNA in absence of fMet-tRNA^fMet (Figure 2C) showed a very much reduced fraction (2 ± 3%) of coincident bursts (Figure 2E, Supplementary Figure S2E), confirming that the initiator fMet-tRNA^fMet is essential for binding of the leaderless mRNAs to mitoribosomes.

Next, we determined the fraction of mitoribosomes which exchanged the large subunit during translation initiation. To be able to accurately quantify this fraction, we used a sample containing a 5-fold excess of Cy5-labeled mtLSU over unlabeled 55S ribosomes and Cy3-labeled MTCO2 mRNA (Figure 2D). Excess Cy5-labeled mtLSU was added immediately before starting the reaction by incubation at 37°C in the same conditions as described above to allow translation initiation. After purification of the initiation complex through a sucrose gradient (Supplementary Figure S2B), we used BTCCD to determine the fraction of Cy3 and Cy5 coincident bursts, which consisted of initiation complexes containing a Cy5 labeled mtLSU and Cy3 labeled MTCO2 mRNA. A complete rapid exchange of the unlabeled mtLSU with a labeled mtLSU would be predicted to give a coincident burst of (43 ± 4) × (5/6), or 36 ± 4%, but the observed value was 27 ± 4% (Figure 2E, Supplementary Figure S2E). Thus, under our conditions, rapid exchange of mtLSUs occurs prior to loading of Cy3-labeled MTCO2 mRNA on the full mitoribosome for the majority (∼75%) of translation initiation events. However, 25% of events occurred on 55S monosomes without exchange of the mtLSU, further corroborating that translation initiation can occur on the monosome.

IF2 is a key player in the canonical mechanism of prokaryotic translation initiation, contributing to the fidelity of fMet-tRNA^fMet selection and catalyzing the joining of the 50S subunit to the 30S pre-initiation complex (PIC). The role of mammalian mitochondrial IF2 has only been investigated in vitro or in heterologous bacterial systems (14,37), and the consequences of its depletion on mitochondrial translation are unknown. To investigate the function of mtIF2, we constructed mtIF2 knockout HEK293T cell lines (KO) using CRISPR/Cas9-mediated genome editing (Supplementary Figure S3A). Deletion of mtIF2 was confirmed in two clonal lines, KO¹ and KO², via western blotting and PCR (Supplementary Figure S3B). As expected, KO cells were unable to proliferate in media with galactose as the sole carbon source, which forces cells to rely on OXPHOS for ATP production, indicating a mitochondrial dysfunction (Supplementary Figure S3C).

To determine if the growth phenotype of KO cells was directly attributable to a defect in mitochondrial translation, we performed metabolic labeling of mitochondrial translation products upon inhibition of cytosolic translation with emetine. The results of the [³⁵S]-methionine/cysteine metabolic labeling indicated a near complete loss of mitochondrial translation in KO clones (Figure 3A and B), which was restored upon expression of mtIF2.FLAG cDNA (Figure 3A). The observed defect in mitochondrial translation was consistent with the steady-state levels of mitochondrially-synthesized OXPHOS components being greatly lowered in mtIF2 KO cell extracts, as analyzed by immunoblotting (Figure 3C). Additionally, an in-gel catalytic assay showed a striking deficiency in the enzymatic activity of the OXPHOS complexes I, IV and V. On the contrary, complex II which is comprised solely of nuclear-encoded subunits, displayed preserved activity consistent with wildtype levels (Figure 3D). The observed OXPHOS deficiency was also in agreement with markedly decreased oxygen consumption rates (Supplementary Figure S3D). Additionally, qPCR of mtIF2 KO cell lines revealed no significant changes to mt-mRNA or mt-rRNA steady-state levels (Figure 3E), implying that mtIF2 is exclusively involved in the regulation of translation, without affecting upstream events, including transcription.

A recent study suggested a role for mtIF2 in mtSSU assembly in Trypanosoma (37). To assess the importance of mammalian mtIF2 activity for human mitoribosome formation, we performed sucrose density gradients from mtIF2 KO cells. Interestingly, the sedimentation pattern of the mitoribosome in mtIF2-depleted cells revealed no global defect in the assembly of either individual subunits or of the full monosome (Figure 3F). Gradient centrifugations were performed either in standard conditions (20mM Mg(OAc)₂) or with lower Mg(OAc)₂ concentration (5mM, Supplementary Figure S3E) to prevent spontaneous subunit joining; however, we observed normally assembled monosomes in both conditions. Bacterial homologues of mtIF2 have long, N-terminal domains which have been shown to promote 70S initiation complex formation by stabilizing the sampling of the 50S subunit during its binding to the pre-initiation complex (13). In contrast, mammalian mtIF2 entirely lacks the domains corresponding to N terminal domains I and II of the E. coli IF2 (Figure 3G) (38). The lack of these domains may explain why mtIF2 appears not to be essential for the subunit joining.

Together, these data illustrate the necessity of mtIF2 for mitochondrial translation, most likely due to its involvement in the recognition of the initiator tRNA and stabilization of mRNA in the complete initiation complex, as previously suggested (15) (Figure 3H). In bacteria, efficient initiation with leaderless mRNAs can occur even in the absence of initiation factors (8). Our results demonstrate that this is not the case for mitochondrial translation.

A number of functions have been assigned to bacterial IF3, including prevention of premature subunit association, and discrimination between the initiator and elongator tRNAs (39). It has also been suggested that IF3 is involved in the 70S scanning mechanism, by which the second cistron of bicistronic mRNAs is initiated (18). In mammalian mitochondria, mechanistic insight into the contribution of mtIF3 to initiation of translation is limited.

In a similar manner as described for mtIF2, we generated several mtIF3 KO cell lines (Supplementary Figure S4A and B) and two clones (KO¹ and KO²) were selected for further analysis. Both KO lines showed a growth defect in galactose media and markedly reduced oxygen consumption (Supplementary Figure S4C and D, respectively), supporting that this factor is also essential for mitochondrial function, as previously shown (40). Analysis of mitochondrial translation using a [³⁵S]-methionine/cysteine incorporation assay showed a dramatic decrease in the production of ATP6, which constitutes the second ORF of the bicistronic ATP8/ATP6 mRNA (Figure 4A and B). Interestingly, synthesis of other mitochondria-encoded proteins was largely unaffected or slightly increased. This phenotype was reversed by the expression of full-length mtIF3.FLAG in KO¹ cells. Western blotting further confirmed the loss of ATP6 steady-state levels in mtIF3 KO lines, while the levels of other OXPHOS components were not changed (Figure 4C). Analysis of in-gel OXPHOS activity confirmed that mtIF3 KO cells had dramatically reduced levels of complex V (F₁F₀‐ATPase) activity, whereas complexes I, II and IV were unchanged or only slightly decreased (Figure 4D). No significant differences in the steady-state levels of mt-mRNAs in KO cells, compared to controls, were detected (Figure 4E).

Together, these results demonstrate that mtIF3 is essential for translation of ATP6, while other mitochondrial proteins can be translated in the absence of mtIF3. Thus, the mitochondrial defect observed in mtIF3 KO cells is likely due to the loss of ATP6 expression, resulting in impaired complex V generation.

ATP6 is translated from the second ORF of the bicistronic ATP8/ATP6 mRNA, with two ORFs overlapping by 46 nucleotides (Figure 4G). The region encompassing the start codon of ATP6 and stop codon of ATP8 is highly unstructured, making it accessible to mitoribosomes (24), but the mechanism of the internal start codon recognition remains unexplored. Another mitochondrial bicistronic transcript, ND4L/ND4, has a much shorter overlapping region of only 7 nucleotides (Figure 4G), suggesting that the mechanism of the second ORF recognition may be different from ATP8/6. In support of this suggestion, we did not detect major changes in KO cells either in ND4 levels using a [³⁵S]-Met/Cys translation assay (Figure 4A), or in complex I activity (Figure 4D), both results indicating that mtIF3 may not be essential for ND4 translation.

To gain more insight into the expression of second ORFs in mitochondrial bicistronic transcripts, we employed mitoribosome profiling (MitoRibo-Seq) (28). Since the steady-state levels of mitoribosomal proteins and monosome formation were not affected by the absence of mtIF3 (Figure 4F), we were able to obtain footprints of active 55S mitoribosomes for profiling analysis. In mtIF3 KO cells, we detected a pronounced reduction in the abundance of mitoribosome-protected fragments (mt-RPF) on the ATP6 ORF relative to WT, whilst the abundance of mitoribosomes was increased on the ATP8 ORF, suggesting that mtIF3 deletion can severely impact the formation of initiation complexes on the downstream ATP6 mRNA (Figure 4H). The occupancy of mitoribosomes on the other mitochondrial transcripts was variable upon mtIF3 knockout (COX2, COX3 and ND3 decreased, ND4, ND5 and ND6 increased, Supplementary Figure S4E), suggesting that mtIF3 may be involved, to some extent, in the binding of mitoribosomes to leaderless mRNAs, though without having a major effect on their translation (Figure 4A and B). However, no other transcript was affected as dramatically as ATP6 (Supplementary Figure S4E), confirming that ATP8/ATP6 bicistronic mRNA requires mtIF3 for translation of the second ORF. The second most affected transcript in this respect was ND3, suggesting that the expression of ND3 might depend on the action of mtIF3. This can also explain a slight decrease in the measured complex I activity in absence of mIF3 (Figure 4D).