Structural basis for late maturation steps of the human mitoribosomal large subunit

Mitochondrial ribosomes (mitoribosomes) synthesize a critical set of proteins essential for oxidative phosphorylation. Therefore, mitoribosomal function is vital to the cellular energy supply. Mitoribosome biogenesis follows distinct molecular pathways that remain poorly understood. Here, we determine the cryo-EM structures of mitoribosomes isolated from human cell lines with either depleted or overexpressed mitoribosome assembly factor GTPBP5, allowing us to capture consecutive steps during mitoribosomal large subunit (mt-LSU) biogenesis. Our structures provide essential insights into the last steps of 16S rRNA folding, methylation and peptidyl transferase centre (PTC) completion, which require the coordinated action of nine assembly factors. We show that mammalian-specific MTERF4 contributes to the folding of 16S rRNA, allowing 16 S rRNA methylation by MRM2, while GTPBP5 and NSUN4 promote fine-tuning rRNA rearrangements leading to PTC formation. Moreover, our data reveal an unexpected involvement of the elongation factor mtEF-Tu in mt-LSU assembly, where mtEF-Tu interacts with GTPBP5, similar to its interaction with tRNA during translational elongation.

M itoribosomal function is essential for cellular energy supply and mitoribosomal defects are coupled to a large and diverse set of human diseases 1 . Mammalian mitoribosomes assemble in a multi-step process that includes the maturation of two ribosomal RNAs (rRNAs; 12S and 16S), a structural tRNA, and incorporation of 82 mitoribosomal proteins (MRPs) 2 . Several RNA-binding proteins, including helicases, GTP-binding proteins (GTPBPs), as well as RNA modifying enzymes have been shown to be essential for correct mitoribosome biogenesis 3 . Some of these factors are conserved in ribosome biosynthesis in all forms of life, while others are specific to mammalian mitochondria.
The mitochondrial transcription termination factor 4 (MTERF4) and the rRNA methyltransferase NSUN4 were among the first characterized mitoribosome assembly factors 4 . NSUN4 methylates C1488 in the 12S mt-rRNA of the small mitoribosomal subunit (mt-SSU) 5 . However, MTERF4 and NSUN4 also form a stoichiometric complex on the large mitoribosomal subunit (mt-LSU) 4,6,7 , which is essential to facilitate monosome assembly 5 .
During mt-LSU maturation, 16S mt-rRNA undergoes three methylations in the catalytically critical peptidyl transferase center (PTC). Human methyltransferase MRM1 is responsible for 2′-O-ribose methylation at G2815 of the P-loop 8,9 , while U3039 and G3040, located in the A-loop of 16S mt-rRNA, are methylated by MRM2 and MRM3, respectively [8][9][10] . The importance of methylation for mitoribosomal function has recently been substantiated by the finding that mutations in MRM2 lead to a cellular respiratory deficiency and a clinically severe multisystemic disorder 11 .
Although assembly factors are central to mammalian mitoribosome biogenesis, their molecular mechanisms of action remain largely enigmatic. Indeed, while the assembly pathways of the mt-SSU and mt-LSU in trypanosomes have recently been visualized by cryo-EM [19][20][21] , only one structural study of native human mt-LSU intermediates is available 22 . In these assembly intermediates, over a fifth of the total 16S mt-rRNA (which in mature mitoribosome comprises the PTC and intersubunit bridges) is not yet folded, while the MRP bL36m protein is missing 22 . How and when the PTC is formed and the role of the assembly factors in this process remain to be elucidated.
In this study, to provide further insights into the process of the mt-LSU assembly in human mitochondria, we determine the cryo-EM structures of mt-LSU intermediates isolated from GTPBP5-deficient cells and complement it with the analysis of a GTPBP5-bound mt-LSU complex immunoprecipitated from cells overexpressing a FLAG-tagged version of GTPBP5. The obtained structures uncover mechanisms of action for several late-stage mt-LSU assembly factors unique to mammalian mitochondria and reveal an unexpected role for mtEF-Tu in the process. These data represent visualizations of the conformational changes essential for the final stages of mt-rRNA folding and reveal the molecular basis for a critical sequence of events leading to mature mt-LSU formation in humans.

Results
Composition of the GTPBP5 KO and GTPBP5 IP mt-LSU assembly intermediates. We purified mitoribosomes from a genetically engineered human cell line that lacks GTPBP5 (GTPBP5 KO ) and from a cell line expressing 3xFLAG-tagged GTPBP5 (GTPBP5 IP ) ( Supplementary Fig. 1a, b), and used single-particle cryo-EM to determine their structures. Both the GTPBP5 KO and GTPBP5 IP mt-LSU assembly intermediates reveal several trapped assembly factors: the MTERF4-NSUN4 complex, MRM2, MTG1, and the MALSU1:L0R8F8:mt-ACP module (Fig. 1a, b, Supplementary Figs. 2-4). Furthermore, the GTPBP5 IP mt-LSU structure features GTPBP5 and the mitochondrial elongation factor mtEF-Tu (Fig. 1b, Supplementary  Fig. 3). Comparing the GTPBP5 KO and the GTPBP5 IP mt-LSU intermediates with the mature mt-LSU 23 reveals two crucial differences in the 16S rRNA conformation (Fig. 1c, d). First, in both GTPBP5 KO and GTPBP5 IP intermediates, MTERF4 in the MTERF4-NSUN4 complex, binds an immaturely folded region of the 16S domain IV. This region (C2548-G2631)-corresponding to helices H68, H69, and H71 of the mature mt-LSU-is folded into an intermediate rRNA helical structure (pre-H68-71) (Fig. 1). The pre-H68-71 occupies a different position on the mt-LSU than H68-71 in the mature mt-LSU, where helices H68-71 and H89-90 jointly form the PTC (Fig. 1d). MTERF4 binding of pre-H68-71 partly orders the disordered rRNA in the mt-LSU assemblies' subunit interface side and thereby enables MRM2 to bind (Fig. 1d). Second, in the GTPBP5 KO -but not in the GTPBP5 IP -the junction between H89 and H90 of domain V is significantly different compared to the mature mitoribosome ( Fig. 1c, d). Specifically, at the base of H89, one helical turn remains unfolded and instead forms a loop (Fig. 1d).
MTERF4-NSUN4 complex steers the final steps of 16S mt-rRNA folding and allows for MRM2 binding. The MTERF4-NSUN4 complex, previously shown to be essential for monosome assembly 4,5 , binds at the intersubunit interface in the GTPBP5 KO and GTPBP5 IP structures (Fig. 1a, b). The C-terminal part of MTERF4 binds to NSUN4 close to the NSUN4 Nterminus in a mixed hydrophobic-polar binding interface, similar to earlier crystal structures of the isolated complex 6,7 (Fig. 2a). NSUN4 was previously shown to m 5 C-methylate the C1488 in 12S mt-rRNA 5 . In our structures, the active site of NSUN4 is turned toward the mt-LSU core ( Supplementary Fig. 5a), impeding methylation of the 12S mt-rRNA. Although the methyldonor S-adenosyl-methionine (SAM) is observed in the NSUN4 active site, no RNA substrate is present. Furthermore, in the GTPBP5 KO and GTPBP5 IP structures, the MTERF4-NSUN4 complex is bound and bent from two sides by uL2m. Specifically, the uL2m C-terminus penetrates in between NSUN4 and MTERF4 to further stabilize the MTERF4-NSUN4 binding interface and decrease the curvature of the MTERF4 solenoid relative to the crystal structures (Fig. 2a). This reforming of the MTERF4 solenoid is necessary to bind the pre-H68-71 rRNA region in the strongly positively charged concave side of MTERF4 ( Supplementary Fig. 5b). Here, MTERF4 forms an extensive network of contacts with the pre-H68-71 that stabilizes the association and promotes formation of the H68-71 premature fold ( Fig. 2b; Supplementary Fig. 5c). The mature H71 basepairing is already formed within the pre-H68-71. Thus, MTERF4 initiates the folding of this 16S mt-rRNA region. Furthermore, it also exposes the A-loop, which is obstructed by H68, H69, and H71 in the mature mt-LSU (Fig. 1d), thereby allowing MRM2 binding.
Similarly to a previously determined mt-LSU assembly intermediate 22 , there is a MALSU1-module positioned adjacent to uL14m in both the GTPBP5 KO and GTPBP5 IP (Fig. 1a, b). Furthermore, MTG1 (GTPBP7), which assists in late-stage mt-LSU maturation 17 , is bound in the vicinity of the pre-H68-71 ( Fig. 1a, b). MTG1 contacts the C-terminus of MALSU1 ( Supplementary Fig. 6a) and the region encompassing A2554-U2602 of the pre-H68-71. This region could not be modeled due to the lower local resolution, but the contact is visible in the map ( Supplementary Fig. 6b). Interestingly, the position of human MTG1 in our structures differs significantly from its bacterial and trypanosomal counterparts 19,21,24 (Supplementary Fig. 6c). Specifically, while in other systems MTG1 homologs contact the rRNA, reaching out toward the PTC ( Supplementary Fig. 6c), in the trapped intermediates described here MTG1 is unlikely to induce pronounced conformational changes of the PTC.
MRM2 2′-O-methylates U3039 in the 16S A-loop during mt-LSU assembly 9,10 and in our GTPBP5 KO and GTPBP5 IP structures, MRM2 binds in the mt-LSU intersubunit interface ( Fig. 1a, b). It features two N-terminal α-helices followed by a canonical S-adenosyl-L-methionine-dependent methyltransferase domain (SAM MTase) (Fig. 2c). In GTPBP5 KO , but not in GTPBP5 IP , the two N-terminal α-helices extend from MRM2 and insert into the rRNA core to thereby displace and retrieve the Aloop (16S mt-rRNA domain V) through a complex interaction network (Fig. 2d, e; Supplementary Fig. 7a). This places the 2′hydroxyl of U3039 close to the ideal methyl-acceptor position in the MRM2 active site (Fig. 2d). However, there is no density for either SAM or S-adenosyl homocysteine (SAH) in the MRM2 active site and there is no apparent density for a 2′-O-methyl on U3039 ( Supplementary Fig. 7b). Interestingly, G3040 that is 2′-Omethylated by MRM3 9,10 , is methylated in our structures ( Supplementary Fig. 7b). Hence, 2′-O-methylation by MRM3 takes place prior to MRM2 methylation in human mitoribosome biogenesis.
GTPBP5 promotes remodeling of the PTC. GTPBP5 consists of a glycine-rich N-terminal domain (Obg-domain) and a Cterminal GTPase domain (G-domain) (Fig. 3a). In our GTPBP5 IP structure, the G-domain has GTP in its active site and is wedged between the L7/L12 stalk and the MALSU1 module ( Fig. 1b and Supplementary Fig. 8a). The Obg-domain protrudes into the PTC (Fig. 3b), thereby displacing the A-loop from MRM2 active site and expelling the MRM2 N-terminal α-helices from the rRNA core (Fig. 3c), while the A-loop folds into the fully mature position (Fig. 3b3).
The protruding Obg-domain is positioned between H89 and H93 and occupies the space that accommodates the acceptor arm of the A-site tRNA during translation ( Supplementary Fig. 8b). Hereby, GTPBP5 binds to the mt-LSU in a manner similar to the A-site tRNA, as it has been observed for ObgE in Escherichia coli 25 . The Obg-domain contains six glycine-rich sequence motifs that form antiparallel polyproline-II helices (helices a-f) (Fig. 3b). Helices c and d bind the A-loop, while the loop between helices e and f inserts into the major groove of H93. The loop between a and b inserts at the triple-junction formed between H89-H90-H93 (Fig. 3b).
Comparison of the GTPBP5 KO and the GTPBP5 IP structures with the mature mt-LSU reveals extensive maturation of the PTC upon GTPBP5 binding. The partly unfolded H89 in the GTPBP5 KO is folded in the GTPBP5 IP (Fig. 3b1). This folding is coordinated by the joint action of GTPBP5 and NSUN4: in the presence of GTPBP5, the extended N-terminal region of NSUN4 inserts into the rRNA core and temporarily displaces the P-loop (Fig. 3b2), thereby breaking the P-loop interaction with H89 ( Fig. 3d, Supplementary Fig. 8c). As a consequence, H89 is given the space necessary to fold into a structure similar to its mature form (Fig. 3b1, lower).
The GTPBP5 IP structure shows a rotation of the L7/L12 stalk in comparison to the GTPBP5 KO structure ( Supplementary  Fig. 8d). Here, A2178 in H43 and U2205 in H44 of the 16S mt-rRNA in the L54/L11 region of the stalk form π-stacking interactions with two residues of the GTPase switch I element of GTPBP5 ( Supplementary Fig. 8a, d). Thereby, the L7/L12 stalk stabilizes the "state 2" conformation of switch I. In this way, the rotated L7/L12 stalk stabilizes the GTP-state of GTPBP5 26 and consequently, a GTP is bound in our structure ( Supplementary  Fig. 8a, d). The requirement for GTPBP5 to be in a GTP-bound state is supported by the inability of a GTPBP5 Walker A mutant (GTPBP5-S238A) to bind mt-LSU intermediates 16 . A backrotation of the L7/12 stalk, presumably by binding of another maturation factor to the mt-LSU assembly intermediate, would lead to a release of the switch I and the activation of GTP hydrolysis, followed by the release of GTPBP5 from the mt-LSU assembly intermediate. Taken together, GTPBP5 plays a direct and active role in rRNA remodeling and, together with the NSUN4 N-terminus, orchestrates the maturation of mitoribosomal PTC.
Translation elongation factor mtEF-Tu is involved in mitoribosome assembly. In the GTPBP5 IP dataset, mtEF-Tu is present in approximately 75% of the GTPBP5-containing pre-ribosomal particles. This constitutes the structural evidence for binding of mtEF-Tu to mitoribosome assembly intermediates during biogenesis and confirms earlier mass-spectrometry data on isolated GTPBP5 IP assembly intermediates as well as protein-proximity interactome analysis 16,27 .
mtEF-Tu consists of a GTPase domain (G-domain; domain I) and two structurally similar β-stranded domains (domains II and III) (Fig. 4a). It was recently shown that during translation, mtEF-Tu·GTP delivers aminoacylated-tRNA to the mitoribosome in a manner similar to its bacterial EF-Tu counterparts (Supplementary Fig. 9a 28 ). In contrast, the binding of an EF-Tu·GTP·aa-tRNA complex is sterically hindered by the MALSU1 module bound in our mt-LSU intermediates ( Supplementary Fig. 9a). Unexpectedly, mtEF-Tu binds to the mitoribosome in a unique manner in the GTPBP5 IP structure (Fig. 4b). Here, domains II and III establish extensive interactions with GTPBP5, the sarcinricin loop (SRL), and the MALSU1 stalk (Fig. 4b, c; Supplementary Fig. 9b). In addition, the G-domain switch I element, in its "state 1"/GDP conformation ( Supplementary Fig. 9c), extends and binds MALSU1 (Fig. 4c).
The G-domain of mtEF-Tu does not contact the SRL as in mtEF-Tu's canonical role in translation but instead binds to the C-terminal region of one copy of bL12m 29,30 that also contacts uL10m at the stalk base (Fig. 4b). In bacteria, homologs to bL12m and uL10m recruit and activate translational GTPases such as EF-Tu via the bL12m C-terminal domain 31,32 and stimulate GTP hydrolysis 33 . Taken together, this suggests that mtEF-Tu hydrolysis-stimulated by bL12m and uL10m-is used to accommodate GTPBP5 on the maturating mt-LSU in analogy to the canonical EF-Tu role in translation, in which aminoacylated-tRNA is accommodated on the translating ribosome (Supplementary Fig. 9a).

Discussion
We isolated mitoribosomes from a cell line with knocked-out GTPBP5 and determined the single-particle cryo-EM structure of the purified mt-LSU assembly intermediates ( Supplementary  Fig. 1a). This approach allows the capture and molecular structure determination of normally short-lived assembly intermediates that do not accumulate in large numbers under physiological conditions. In order to trap sequential steps of mammalian mitoribosome biogenesis, we also determined the molecular structure of a GTPBP5 IP complex immunoprecipitated from cells expressing a FLAG-tagged version of GTPBP5 (Supplementary Fig. 1b). The advantage of the immunoprecipitation strategy over in vitro addition of the factor to purified complexes relies on the fact that the natively expressed protein is able to bind  to its endogenous interactors in native states. Indeed, thanks to this method, we were able to detect the interaction of GTPBP5 and mtEF-Tu. Importantly, the striking similarities between the GTPBP5 KO and GTPBP5 IP structures suggest that GTPBP5 KO likely represents a true assembly stage, not an "off-pathway" intermediate accumulated due to the absence of GTPBP5.
Thanks to our approaches, we were able to uncover features unique to mammalian mitochondria and explain the essential roles of several assembly factors that together promote fine RNA rearrangements and lead to the mt-LSU completion. Based on our structural data, on the different subpopulations obtained by our analysis and on previous biochemical studies, we propose a model of the late-stage mt-LSU assembly that requires the interplay of nine auxiliary factors (Fig. 5). The first intermediate of assembly proposed in our model corresponds to an mt-LSU intermediate population obtained from the GTPBP5 KO dataset that displays a weak density at the intersubunit interface, corresponding to unstructured 16S mt-rRNA, and the associated MALSU1:L0R8F8: mt-ACP module, resembling the mt-LSU intermediate previously identified by Brown et al. 22 (Fig. 5; Supplementary Fig. 2a). This stage of assembly is followed by recruitment of the MTERF4-NSUN4 complex to the mt-LSU interface that leads to the folding of the pre-H68-71 helical structure by MTERF4, at the same time exposing the A-loop for MRM2 binding.
As a consequence of MTERF4-NSUN4 recruitment, MRM2 and MTG1 are engaged at the mt-LSU interface (Fig. 5), where MRM2 positions at the A-loop (Fig. 2d) and MTG1 contacts the tip of the pre-H68-71 ( Supplementary Fig. 6b), which is in a strikingly divergent position compared to MTG1 homologs ( Supplementary Fig. 6c). In bacteria, the MTG1 homolog RbgA binds the LSU in a similar location as MRM2 in our structures, contacting the 23S rRNA at the P-and A-site, and contributing to their maturation 24 . Likewise, Mtg1 in Trypanosoma brucei is bound in complex with the other two GTPases to 12S rRNA in a position similar to the one observed for RbgA 19,21 . Based on our structural data, it appears that human MTG1 does not induce pronounced conformational changes of the PTC or participate in the recruitment/dissociation of assembly factors, at the assembly stages captured in this study. MTG1 may instead function as a quality-control checkpoint by monitoring the folding status of H68-71 and preventing early subunit joining. Noteworthy, recent work by Chandrasekaran et al. 34 shows the structure of an intermediate of mt-LSU assembly with MTG1 bound in the same location as MRM2 in our structure, suggesting that MTG1 changes its location on the mt-LSU during the assembly process.
The next step in our model consists in GTPBP5 arrival to the mt-LSU interface (Fig. 5). In particular, we show how GTPBP5 binding pushes the A-loop away from the MRM2 active site and at the same time how it facilitates the maturation steps of the PTC region (Fig. 3b, c). In addition, NSUN4 plays a direct part in PTC formation via its N-terminal extension and in coordination with GTPBP5 (Fig. 3b). Since the NSUN4 active site faces the mt-LSU core ( Supplementary Fig. 5a), our structures strongly suggest two independent functions for NSUN4 in mitoribosome biogenesis: maturation of the PTC in mt-LSU assembly as shown in this work and methylation of C1488 in mt-SSU 5 . Whether 12S methylation activity occurs earlier or later in the assembly pathway or a separate pool of NSUN4 is responsible for this function, remains to be determined.
The arrival of GTPBP5 to the mt-LSU interface coincides with mtEF-Tu binding. mtEF-Tu is a highly conserved translational factor that uses its GTP-hydrolysis to accommodate aa-tRNAs to the A-site of the ribosome. In our structure, mtEF-Tu employs domain III to bind GTPBP5, which in turn protrudes into the PTC region, analogously to a tRNA during translation ( Fig. 4b and Supplementary Fig. 8b). It is thus tempting to speculate that mtEF-Tu brings GTPBP5 to the mitoribosome in analogy to how it brings aa-tRNA during translation (Supplementary Fig. 9a). A recent analysis of the protein proximity network in mitochondria also confirmed a strong interaction of GTPBP5 and mtEF-Tu 27 , however, further functional studies of this interaction are required to assess its physiological relevance.
Interestingly, we were able to visualize a direct interaction of mtEF-Tu with the C-terminal part of ribosomal protein bL12m (Fig. 4b). In bacteria, the L7/L12 stalk is necessary for optimal ribosomal translation 35 and a highly conserved C-terminal domain of L12 is responsible for an about 1000-fold stimulation of GTP hydrolysis by EF-Tu and EF-G 33 . Noteworthy, as previously observed in bacterial system 33 , human bL12m also contacts uL10m at the stalk base. We speculate that mtEF-Tu GTPase activity is triggered by the interaction with the bL12m Cterminal domains, which is further promoted by their binding to uL10m.
Based on our data and on the subpopulations identified in our datasets, we hypothesize that the last steps of this model include the release of MTG1, followed by the release of the GDP-bound mtEF-Tu, as well as MRM2. Later, GTPBP5, the MALSU1 module and MTERF4-NSUN4 are released in an unknown order and in a yet-to-be-defined modality. Importantly, while our manuscript was in revision, other studies 34,[36][37][38] have shown different late steps of mt-LSU assembly involving the MTERF4-NSUN4 complex and other factors. In particular, in Hillen et al. 36 , GTPBP6 is proposed to act at a later stage of assembly where MTERF4-NSUN4 and GTPBP5 are released, suggesting that GTPBP6 might be important for the release of these factors. Nevertheless, further studies are needed to completely elucidate these final steps of mt-LSU assembly.
In conclusion, our approaches provide a detailed view of the final stages of the mt-LSU biogenesis pathway. Moreover, as defects in mitoribosome biogenesis-resulting from, for example, mutations in MRM2 and GTPBP5-are increasingly implicated in mitochondrial disease 11,39 , the current work does not only describe fundamental cellular processes but may also further diagnostic and therapeutic approaches to mitochondrial diseases.

Methods
Generation of GTPBP5 knock-out cell line. The knock-out cell line (GTPBP5 KO ) was generated in the Flp-In T-Rex human embryonic kidney 293 (HEK293T) cell line (Invitrogen) using CRISPR/Cas9 technology targeted on exon 1 of GTPBP5 gene, which encodes for GTPBP5, as described in Cipullo et al. 16 . In short, two pairs of gRNAs were designed and cloned into the pSpCas9(BB)-2A-Puro (pX459) V2.0 vector to generate out-of-frame deletions. Transfection of HEK293T cell line with the pX459 variants was performed using Lipofectamine 3000 following manufacturer's instructions. The selection of transfected cells was done using puromycin treatment at a final concentration of 1.5 mg/ml for 48 h. Subsequently, cells were single-cell diluted and transferred into a 96-well plate. Selected clones were screened via Sanger sequencing and Western blotting.
Purification of the mt-LSU from GTPBP5 KO cell line via sucrose gradient centrifugation. Isolation of mitochondria was performed from eighty 150 mm dishes of GTPBP5 KO cell line as described in Rorbach et al. 10 , with some modifications. Crude mitochondria were further purified via differential centrifugation by being loaded onto a sucrose gradient (1 and 1.5 M sucrose, 20 mM Tris-HCl pH 7.5, 1 mM EDTA) and centrifuged at 107,170g for 1 h at 4°C (Beckman Coulter SW41-Ti rotor). Mitochondria forming a band at the interphase between the 1 and 1.5 M sucrose were collected and resuspended in 10 mM Tris-HCl pH = 7.5 in 1:1 ratio. After centrifugation, the final mitochondrial pellet was resuspended in mitochondrial freezing buffer (200 mM trehalose, 10 mM Tris-HCl pH 7.5, 10 mM KCl, 0.1% bovine serum albumin, 1 mM EDTA), snap frozen in liquid nitrogen and stored at −80°C.
Cryo-EM data acquisition and image processing. Prior to cryo-EM grid preparation, grids were glow-discharged with 20 mA for 30 s using a PELCO easiGlow glow-discharge unit. Quantifoil Cu 300 mesh (R 2/2 geometry; Quantifoil Micro Tools GMBH) covered with a thin layer of 3 nm carbon were used for the GTPBP5 KO sample. Carbon lacey films (400 mesh Cu grid; Agar Scientific) mounted with ultrathin carbon support film were used for the GTPBP5 IP sample. Three μl aliquots of sample were applied to the grids, which were then vitrified in a Vitrobot Mk IV (Thermo Fisher Scientific) at 4°C and 100% humidity (blot 3 s, blot force 3, 595 filter paper (Ted Pella Inc.)). Cryo-EM data collection (Supplementary Table 1) was performed with EPU (Thermo Fisher Scientific) using a Krios G3i transmission-electron microscope (Thermo Fisher Scientific) operated at 300 kV in the Karolinska Institutet's 3D-EM facility. Images were acquired in nanoprobe 165kX EFTEM SA mode with a slit width of 10 eV using a K3 Bioquantum during 1 s during which 60 movie frames were collected with a flux of 0.82 e − /Å 2 per frame. Motion correction, CTF-estimation, Fourier cropping (to 1.02 Å/px), picking and extraction in 600 pixel boxes (size threshold 300 Å, distance threshold 20 Å, using the pre-trained BoxNet2Mask_20180918 model) were performed on the fly using Warp 40 . Only particles from micrographs with an estimated resolution of 3.6 Å and under focus between 0.2 and 3 μm were retained for further processing.
For the GTPBP5 KO dataset, 704,720 particles were picked from 37307 micrographs ( Supplementary Fig. 2). The particles were imported into CryoSPARC 2.15 41 for further processing. After 2D classification, 130,289 particles were selected for an ab initio reconstruction. This reconstruction, in addition to two "bad" reconstructions created from bad 2D class averages, was used for heterogeneous refinement of the complete particle set resulting in one of the three classes yielding a large-subunit reconstruction with high resolution features (196,318 particles). After homogeneous refinement of these particles, the PDB model of a mitochondrial LSU assembly intermediate (PDB: 5OOL 22 ) was fitted in the density. The reconstruction contained the MALSU1 module and also featured weak unexplained densities for several additional components in the intersubunit interface. A 3D variability analysis was performed with a mask on the intersubunit interface and a low pass resolution of 10 Å, and subsequently used for clustering into six particle classes representing different assembly intermediates. Two of the classes (43,057 and 41,619 particles) lacked the density for the A-and P-loops, H89, helices 68-71 and the L7/12 stalk. The A-and P-loops become visible in the third class (28,001 particles). The fourth class revealed a number of biogenesis factors: MRM2, MTERF4-NSUN4, MTG1, and the structured H67-H71 rRNA region (pre-H68-71) (48,646 particles) as well as H89. All the biogenesis factors are absent in the fifth class, in which helices 68 and 71 move to the mature position (26,678 particles). H69 is nevertheless not visible. The last class contains the assembled mitoribosome (8,317 particles). Non-uniform refinement of the fourth particle set yielded a reconstruction at 2.6 Å, which was used for model building and refinement. As the density for MTG1 was weaker than for the other factors, 3D variability analysis 42 was performed with a mask on the MTG1 region and with a 10 Å low-pass filter to select particles containing MTG1 (19,254 particles, which was subsequently subjected to homogeneous refinement yielding a reconstruction at 2.9 Å). The particles not containing MTG1 were refined to 2.8 Å.
For the GTPBP5 IP dataset, 283,598 particles were picked from 112,076 micrographs using Warp and imported into CryoSPARC 2.15 ( Supplementary  Fig. 3) 41 . The complete particle set was used in heterogeneous refinement against the same three references derived from the GTPBP5 KO dataset. One of the classes (78,306 particles) yielded a high-resolution reconstruction of the mt-LSU assembly intermediate. After homogeneous refinement, additional density for GTPBP5 was visible in the intersubunit interface. 3D variability analysis was performed with a mask on the GTPBP5 region and a low pass resolution of 10 Å. Subsequent clustering into two particle clusters revealed a particle subset containing GTPBP5 NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23617-8 ARTICLE (71,834 particles), which was used for model building and refinement. This reconstruction also features densities for MRM2 and MTERF4-NSUN4. In addition, a weak density was present for mtEF-TU, the bL12m C-terminal domain, and MTG1. The refined particles were subjected to 2D classification and the bad classes were removed. The remaining particles were polished in RELION 3.1 43 and re-imported into CryoSPARC for further processing. Non-uniform refinement yielded a reconstruction at 3.1 Å, which was used for model building. 3D variability analysis was performed on these particles with a mask covering mtEF-TU, bL12m, and MTG1 and a 10 Å low pass filter. Subsequent clustering (4 clusters) revealed 2 clusters containing mtEF-TU/bL12m (17,886 particles in total), one cluster containing MTG1 (8,233 particles), and one cluster containing all three proteins (13,376 particles). The reconstructions derived from the MTG1-and the mtEF-Tucontaining particles reached a resolution of 3.2 Å.
Model building and refinement. Model building of the GTPBP5 KO and GTPBP5 IP mt-LSU assembly intermediate structures was performed using Coot 44 . The structure of a previous mt-LSU assembly intermediate (PDB 5OOL 22 ) was used as a starting model. MRM2 and MTERF4-NSUN4 were identified by modeling secondary structure elements in Coot, and using the initial models for a structural search using the DALI server 45 . MTG1, as well as mtEF-Tu and the bL12m C-terminal domain in the GTPBP5-bound mt-LSU dataset, were identified using a density-based fold-recognition pipeline 22 . Using SWISS-MODEL 46 , we generated homology models for the human GTPBP5 (template: PDB 4CSU chain 9 25 ), MTG1 (template: PDB 3CNL chain A 47 ), bL12m (template: PDB 1DD3 chain A 48 ), and mtEF-TU (template: PDB 1D2E chain A 49 ). All the models, as well as the crystal structure of the human MTERF4-NSUN4 (template: PDB 4FP9 chains A and B 6 ), were fitted into the density map using Coot JiggleFit. The MTG1 GTPase domain and the L17/12 stalk were excluded from atomic refinement and were only subjected to rigid body refinement. Metal ions and modifications were placed based on map densities. Stereochemical refinement was performed using Refmac5 50  Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.