NDUFS4 Regulates Cristae Remodeling in Diabetic Kidney Disease

The mitochondrial electron transport chain (ETC) is a highly adaptive process to meet metabolic demands of the cell, and its dysregulation has been associated with diverse clinical pathologies. However, the role and nature of impaired ETC in kidney diseases remains poorly understood. Here, we generated diabetic mice with podocyte-specific overexpression of Ndufs4, an accessory subunit of mitochondrial complex I, as a model to investigate the role of ETC integrity in diabetic kidney disease (DKD). We find that these conditional mice exhibit significant improvements in cristae morphology, mitochondrial dynamics, and albuminuria. By coupling proximity labeling with super-resolution imaging, we also identify the role of cristae shaping proteins in linking NDUFS4 with improved cristae morphology. Taken together, we discover the central role of NDUFS4 as a powerful regulator of cristae remodeling, respiratory supercomplexes assembly, and mitochondrial ultrastructure in vitro and in vivo. We propose that targeting NDUFS4 represents a promising approach to slow the progression of DKD.

reveal that ETC integrity determines the stability of RSCs and plays a central role in cristae and 23 mitochondrial dynamics in podocytes.

ETC remodeling of podocytes in DKD.
ETC is known to exhibit significant remodeling in 2 response to the metabolic demands of the cell 16 . However, it remains unclear how ETC adapts 3 to the diabetic milieu in podocytes. To provide a deeper insight into the possible dynamics of 4 individual ETC complexes in the diabetic environment, we performed a comparative 5 mitochondrial proteome profiling focusing quantitatively on the protein abundance of ETC 6 complexes in primary podocytes isolated from diabetic C57BL/6-Ins2 Akita /J (Ins2 Akita/+ ) mice, an 7 established model of type 1 diabetes, and their nondiabetic littermates ( Fig. 1a and Extended 8 Data Fig. 1a). We identified 62 out of the 73 known subunits of mouse ETCs corresponding to a 9 recovery rate of 85% (Fig. 1b). Notably, we found that the abundance of several subunits of CI 10 was significantly reduced in the podocytes of diabetic mice (Fig. 1b,c and Extended Data Fig. 11 1b). Consistent with this finding, we also observed reduced CI enzymatic activity in enriched 12 mitochondrial samples from podocytes in both type 1 (Ins2 Akita/+ ) and type 2 (Lepr db/db ) diabetic  Fig. 1f). Using Nephroseq database, we found a 20 positive correlation between NDUFS4 mRNA in glomeruli and estimated glomerular filtration 21 rate (eGFR) in subjects with DKD (Fig. 1h). We validated these findings by 2i,j), mesangial matrix expansion (Fig. 2k), glomerular basement membrane (GBM) thickening 7 (Fig. 2l), podocyte foot process effacement and podocytes loss (Fig. 2i,m). We also crossed 1 Ndufs4 podTg mice with obese Lepr db/db mice, an established model of type 2 diabetes, and found 2 similar phenotypes with a significant improvement in albuminuria in diabetic Lepr db/db ;Ndufs4 podTg 3 independent of body weight gain and blood glucose levels (Extended Data Fig. 2g-k). Ndufs4 overexpression improves mitochondrial morphology. We reasoned that the 6 underlying molecular mechanism of Ndufs4 podTg -mediated improvement in DKD could be 7 associated with improved mitochondrial respiration in podocytes. To test this, we measured 8 mitochondrial respiration using a Seahorse Analyzer. Whereas the oxygen-consumption-rate 9 (OCR) measurements were similar between primary podocytes from WT and Ndufs4 podTg mice, 10 podocytes from diabetic Ins2 Akita/+ mice exhibited a significantly reduced basal, maximal, ATP-11 linked, and spare OCR values . In contrast, OCR values were markedly improved in 12 podocytes from diabetic Ndufs4 podTg mice ( Fig. 3a-e). We next tested the susceptibility of these 13 podocytes to rotenone, a CI-specific inhibitor. We found that although podocytes from diabetic 14 Ins2 Akita/+ mice had a significantly lower OCR suppression curve and rotenone IC50 values, 15 primary podocytes from diabetic Ndufs4 podTg mice had much higher IC50 and OCR suppression 16 curve, almost similar to those in WT mice (Fig. 3f). Consistent with these findings, we also found 17 improved CI activity in the glomeruli of the diabetic Ndufs4 podTg mice compared to that in 18 podocytes from Ins2 Akita/+ mice (Fig. 3g). 19 We have previously shown that DKD progression is associated with an excessive 20 mitochondrial fission in podocytes 28,29 . Thus, we next explored the contributions of Ndufs4 podTg 21 on mitochondrial dynamics. We confirmed that podocytes from type 1 Ins2 Akita/+ and type 2 22 Lepr db/db diabetic mice exhibited enhanced mitochondrial fission with altered cristae morphology 23 . However, overexpression of NDUFS4 restored the 24 tubular and interconnected mitochondrial morphology as shown by improved mitochondrial 25 aspect ratio, circularity, roundness, and ferret measurements in podocytes  whereas transfection of podocytes with NDUFS4 OE led to improved cristae integrity and 23 mitochondrial morphology (Fig. 4e). To correlate these results with RSCs assembly, we 24 examined the role of NDUFS4 OE on the abundance of intact RSCs in digitonin-treated 25 mitochondrial samples by blue native polyacrylamide gel electrophoresis (BN-PAGE) 32,33 . The abundance of intact RSCs on BN-PAGE gels was reduced in HG conditions (Extended Data 1 Fig. 4f,g) whereas cultured NDUFS4 OE podocytes in HG media exhibited significantly higher 2 protein abundance of RSCs (Fig. 4f). Western blot analysis of the same samples with OXPHOS 3 cocktail antibodies provided similar results (Extended Data Fig. 4f,g). Consistent with these 4 results, the CI in-gel activity as shown by reduction of nitro-blue tetrazolium in the presence of 5 NADH, was reduced in podocytes treated with HG as compared to those treated with NG 6 (Extended Data Fig. 4h), whereas NDUFS4 OE restored the CI in-gel activity even under HG 7 conditions (Fig. 4g,h). Taken together, our findings suggest that NDUFS4 OE regulates not only 8 cristae morphology but also RSCs assembly and mitochondrial dynamics in kidney podocytes. 9 10 NDUFS4 interaction with cristae regulatory proteins. The regulatory effect of NDUFS4 on 11 cristae morphology, RSCs integrity and mitochondrial dynamics raises several questions 12 regarding the underlying protective molecular mechanism of NDUFS4 OE. It is known that 13 cristae integrity is necessary for the proper spatial distribution of RSCs 34,35 . However, whether 14 changes in ETC integrity could result in cristae remodeling is not well understood. We 15 suspected that the effects of NDUFS4 on cristae remodeling could be through its interaction 16 with one or several cristae regulatory proteins. To this end, we first assessed the abundance of 17 some of the key cristae regulatory proteins by Western blots in podocytes from WT, Ndufs4 podTg , 18 diabetic Ins2 Akita/+ , and diabetic Ndufs4 podTg mice. The cristae regulatory proteins, including 19 STOML2 (Stomatin-like protein 2), IMMT/MIC60 (Inner membrane mitochondrial protein), 20 ATAD3A (ATPase family AAA domain containing 3A) and OPA1 (OPA1 mitochondrial dynamin 21 like GTPase) were all reduced in the podocytes of diabetic Ins2 Akita/+ mice (Extended Data Fig.   22 5a). Conversely, the abundance of these proteins was significantly higher in podocytes from 23 diabetic Ndufs4 podTg (Extended Data Fig. 5a). We next adopted a proximity labeling approach to 24 interogate a possible interaction of NDUFS4 with cristae regulatory proteins (Fig. 5a). To this 25 end, we engineered a podocyte cell line that stably express a DOX-inducible NDUFS4-APEX2 26 chimeric protein (Extended Data Fig. 5b). Biotin-labeled proteins in proximity to NDUFS4-1 APEX2 activated with H2O2 were isolated using streptavidin-coupled beads and identified by 2 LC-MS/MS. NDUFS4-APEX2 transfected podocytes without H2O2 activation or without DOX 3 induction were used as controls. Out of 2152 proteins identified, 357 were mitochondrial 4 proteins, and 46 of them were significantly enriched in DOX+H2O2 podocytes (Fig. 5b,c and 5 Extended Data Fig. 5c). Among them, we found several cristae regulatory proteins, including 6 STOML2, IMMT/MIC60, and ATAD3A ( Fig. 5c and Extended Data Fig. 5c). The close proximity 7 of NDUFS4 to the cristae regulatory proteins was further validated by immunoblot analysis using 8 specific primary antibodies whereby the biotinylated STOML2, ATAD3A, and IMMT/MIC60 were 9 efficiently pulled down by streptavidin beads when treated with H2O2 ( Fig. 5d). Surprisingly, 10 OPA1 was not a closely associated protein with NDUFS4 based on both the proximity labeling 11 and strepavidin pulldown assays. Taken together, these findings suggest that the cristae 12 regulatory proteins could potentially form complexes with NDUFS4. However, proximity labeling 13 assay is unable to differentiate between a direct interaction or a mere close association. 14 Additionally, it does not specify if the association of NDUFS4 with these proteins occurs in the 15 context of individual complexes or RSCs. Consequently, we performed complexsome profiling 16 on enriched mitochondrial samples from HG-treated cells and HG-treated NDUFS4 OE 17 podocytes to explore whether the cristae forming proteins interact with NDUFS4 in the context 18 of RSCs and therefore, comigrating within RSCs (Fig. 5e). We separated ETC complexes on 19 BN-PAGE, sliced five distinct bands representing distinct RSCs (labeled 1-5) after Coomassie 20 blue staining, and performed mass spectrometry (LC-MS/MS). A careful analysis of each band 21 revealed that several cristae organizing proteins, including STOML2, ATAD3A and 22 IMMT/MIC60, comigrate with RSCs, suggesting that they are in close association with RSCs 23 structures ( Fig. 5f). To further test whether STOML2, ATAD3A and IMMT/MIC60 are integrated 24 within RSCs, we performed immunoblotting with cocktail antibodies against OXPHOS or cristae 25 organizing proteins (Fig. 5g). We observed that whereas STOML2 was mainly colocalized with RSCs and its abundance was markedly increased with NDUFS4 OE, ATAD3 was not 1 significantly colocalized with RSCs, and IMMT displayed a more wide-spread distribution within 2 and outside of RSCs (Fig. 5g). Since STOML2 comigrated with RSCs in a NDUFS4 OE-3 dependent manner and considering our previous proximity labeling and pulldown assays, we 4 decided to further pursue its role as a link between NDUFS4 OE and cristae integrity.

5
To further corroborate the physical interaction of NDUFS4 with STOML2, we employed a 6 combination of biochemical and super-resolution imaging approaches. We first performed co-7 immunoprecipitation (Co-IP) experiments in HEK293T cells transiently transfected with 8 NDUFS4-FLAG construct. Among pulled down proteins with FLAG antibody, both STOML2 and 9 NDUFS4 were detected, but not with IgG antibody (Fig. 5h). We then performed a GST affinity 10 pulldown assay in vitro in which a GST-NDUFS4 fusion protein was incubated with cell lysates 11 transiently overexpressing a STOML2-HA fusion protein. Immunoblotting showed that STOML2 12 interacted with GST-NDUFS4, but not with GST control in vitro, consistent with a potential 13 interaction between NDUFS4 and STOML2 (Fig. 5i).
14 To further establish whether NDUFS4 and STOML2 are spatially in close physical 15 proximity, we employed two super-resolution imaging approaches, the stimulated emission 16 depletion (STED) and the stochastic optical reconstruction microscopy (STORM) providing 17 nanoscale spatial localization at a single-molecule resolution 36 (Fig. 5j,k). Similar results were confirmed in the analysis of colocalization based on 24 the distance (Fig. 5j,k). Additional colocalization analyses of 3D images with single molecule 25 super-resolution imaging obtained from STORM validated the spatial interaction between 26 NDUFS4 OE and STOML2 (Fig. 5m). Specifically, among molecules with the inter-molecular 1 distance <500 nm (Extended Data Fig. 5d), podocytes under the HG condition exhibited a 2 significantly longer distance between NDUFS4 and the nearest neighboring STOML2, with a 3 median nearest neighboring distance (NND) that was also significantly longer compared to 4 podocytes under the NG or HG-treated NDUFS4 OE conditions (% colocalization=38.9±1.3% 5 for HG-DOX vs. 49.8±1.8% for NG and 49.5±1.8% for HG+DOX; NND= 65.3±3.6 nm for HG-6 DOX vs. 40.3±2.9 nm for NG and 41.6±3.0 nm for HG+DOX) (Fig. 5n,o). These findings suggest 7 that within each mitochondrion, Ndufs4 and STOML2 have the highest proximity in NG or with 8 Ndufs4 overexpression in HG conditions. In contrast, they exhibit the least spatial proximity in 9 HG conditions.  (Fig. 6b). We also observed that HG reduced the mitochondrial cristae 19 density in podocytes, but NDUFS4 OE restored the cristae structure and abundance in HG 20 media and improved mitochondrial morphology ( Fig. 6c-f). In the NDUFS4 OE podocytes in 21 which the STOML2 was deleted, however, the cristae restoration was no longer observed and 22 mitochondrial morphology was distorted in HG conditions ( Fig. 6c-f). Thus, these data suggest 23 that STOML2 is an essential component of the NDUFS4 OE-mediated improved cristae 24 integrity, RSC assembly, and mitochondrial morphology.

13
The full-length mouse STOML2 protein is 353 amino acids long and contains four 1 functional domains: the N-terminal mitochondrial-targeting sequence (MTS), a hydrophobic 2 hairpin (HP) domain, a conserved stomatin (STOM) domain, and the C-terminal coiled-coil 3 domain (CTD) (Fig. 6g). To address how NDUFS4 interacts with STOML2, we created a series 4 of STOML2 deletions and observed that the C-terminal deletion mutant was co-5 immunoprecipitated with FLAG tagged-NDUFS4, but not the other mutants at the N-terminal 6 domain which included HP and STOM domains, suggesting that the N-terminal domain of 7 STOML2 is the key domain for NDUFS4 binding (Extended Data Fig. 6a,b). We next generated

21
The mitochondrial ETC harnesses the chemical energy of nutrients in the form of high-22 energy electrons to generate an electrochemical proton gradient leading to the reduction of 23 molecular oxygen to water. Beyond its role on the mitochondrial respiration, electron flow across 24 inner mitochondrial membrane is also crucial for the bioenergetics properties of mitochondria 1 through synthesis of ATP and critically involved in the biosynthetic and signaling properties of 2 mitochondria. Importantly, ETC dysfunction has been associated with several human 3 pathologies 23-26 . However, the impact of impaired ETC assembly and the link between ETC, 4 RSCs formation, and cristae integrity in the pathogenesis of DKD remained unknown.

5
The findings of this study provide a mechanistic link between ETC remodeling in 6 podocytes and the pathogenesis of DKD, a major complication of diabetes. Using an integrated 7 in vitro and in vivo experimental approach, we find a causal link between NDUFS4 deficiency,  These findings represent a major paradigm shift in the current management of DKD by 10 suggesting that targeting ETC remodeling could be a promising approach for developing 11 therapies to mitigate the progression of DKD.

12
Our findings also uncover the central role of ETC integrity as a defining feature of 13 mitochondrial dysfunction and a powerful regulator of cristae remodeling and mitochondrial 14 dynamics in the diabetic milieu. We discovered an unexpected role of Ndufs4 as a major culprit 15 in maintaining cristae morphology and mitochondrial shape beyond its previously well-16 established role in CI assembly and stabilization 38,39 . Our study also highlights the benefits of 17 rescuing Ndufs4 deficiency on mitochondrial dysfunction and DKD progression (Fig. 7).

18
While previous studies have identified the critical role of cristae shaping proteins, 19 including STOML2, in cristae remodeling and RSCs assembly 34,40,41 , the regulatory role of ETC 20 integrity on cristae structure and the interaction between NDUFS4 and cristae shaping proteins 21 remained unknown. A first hint that NDUFS4 deficiency could have a key role in defining 22 mitochondrial dysfunction in DKD came from our initial comparative proteomic profiling revealing 23 a consistent downregulation of several subunits of CI in diabetic podocytes. We reasoned that 24 one possible explanation for the CI remodeling in podocytes could be an initial adaptive 25 mechanism to the diabetic environment. However, as hyperglycemia persists, these chronic changes might have become maladaptive and pathogenic in nature resulting in biochemical and 1 structural alterations in mitochondria. To this end, we argued that forced expression of the 2 NDUFS4 subunit, consistently downregulated in the podocytes of the type 1 and type 2 diabetic 3 mice as well as in the glomeruli of the DKD patients, might overcome the mitochondrial 4 maladaptation in DKD and provide significant insights into molecular mechanisms of its 5 progression. Indeed, we found that overexpression of the NDUFS4 in podocytes improved 6 mitochondrial respiration and CI activity and prevented mitochondrial fragmentation in 7 podocytes of diabetic mice. We also linked NDUFS4 overexpression with the protection of both 8 RSC assembly and cristae morphology.

9
How does NDUFS4 overexpression restore cristae organization and mitochondrial 10 dynamics? We had previously shown that enhanced mitochondrial fission is implicated in the 11 diabetes-induced mitochondrial dysfunction in podocytes 28,29 . However, the role of ETC integrity 12 on cristae morphology and mitochondrial fission was incompletely understood. Our qualitative 13 and quantitative approaches in this study including tomography clearly suggest aberrant cristae 14 morphology and a significant loss of their native lamellar morphology. Remarkably, we find that 15 these alterations in cristae structure were significantly improved with the forced expression of 16 Ndufs4. By coupling proximity labeling, streptavidin pulldown assays, complexsome profiling, 17 and super-resolution imaging approaches, we identified a possible interaction between 18 STOML2, a 39 kDa cristae shaping protein, and NDUFS4 in the context of improved RSCs 19 assembly as the main explanation for the NDUFS4-mediated improvement in cristae 20 morphology. We further validated this interaction and found that two regions of the -pleated 21 sheet structures in STOM domain of STOML2 are crucial for its binding to NDUFS4. These properties, these cristae shaping proteins also regulate RSCs integrity and mitochondrial 25 dynamics. However, while the effect of cristae organizing proteins on cristae integrity has clearly 26 been established, our findings uncover a novel role of Ndufs4 in regulating cristae and RSC 1 integrity and ultimately mitochondrial morphology and function. We speculate that 2 overexpression of Ndufs4 subunit stabilizes and improves its interaction with STOML2 leading 3 to proper cristae formation, RSCs assembly and improved mitochondrial respiration and 4 dynamics in the diabetic milieu.

5
The NDUFS4 subunit is localized between the N-and Q-module and it is known that 6 pathogenic mutations in the nuclear DNA-encoded NDUFS4 gene, the most extensively studied 7 mutations in CI subunit, cause a severe form of Leigh-like Syndrome in pediatric populations, a 8 rare and heterogeneous disorder that affects central nervous system 26,42 . Similarly, mice lacking 9 full length NDUFS4 protein develop Leigh (like) disease with postnatal lethality at ~50 days 43 .

10
Several tissue-specific Ndufs4 knockout mouse models are also developed to further examine 11 the role of NDUFS4 as a model to study mitochondrial diseases [44][45][46][47] . Comprehensive analysis of 12 cells derived from these mice revealed that tissue-specific NDUFS4 mutations were commonly 13 associated with increased ROS, altered mitochondrial ATP homeostasis and mitochondrial 14 morphology 44,47 . However, it is important to emphasize that our data suggest that the diabetes-

22
Our study has unraveled many unexpected aspects of pathobiology of NDUFS4 in 23 podocytes and its role in the progression of DKD, however, our findings also raise several 24 important questions that remain to be fully addressed. For example, what upstream signaling 25 pathways are required to initiate the cascade of events that lead to reduced expression of NDUFS4 in podocytes in DKD? Furthermore, additional experiments are needed to determine 1 the extent of NDUFS4 deficiency in other cells and tissues. It would also be meaningful to 2 examine the role of other subunits of CI on progression of DKD. The interplay between NDUFS4 3 and STOML2 is also complex and further studies are needed to test whether STOML2 can be 4 regarded as a specific and necessary binding partner of NDUFS4 in maintaining cristae 5 structure. Indeed, to better understand the pathobiology and structural integrity of NDUFS4 in 6 the diabetic milieu, it would be important to understand the molecular relationship between 7 NDUFS4 with other cristae organizing proteins as they relate to the recruitment of the 8 supercomplexes and improved mitochondrial dynamics. Finally, one important question is 9 whether our findings are relevant to humans. Our data indicates that reduced levels of 10 glomerular NDUFS4 expression correlate with albuminuria and eGFR in subjects with DKD.

11
Importantly, we find that NDUFS4 staining in glomeruli is progressively reduced with worsening 12 of DKD histology suggesting that Ndufs4 may play an important role in progression of DKD.

13
However, mice and humans present major differences that might influence the dynamics of the 14 events described in this study and further studies are required to validate our results in 15 individuals with DKD.

16
In summary, our study suggests that reduced levels of NDUFS4 expression leads to 17 compromised CI and RSCs formation with a significant effect on bioenergetic capacity, cristae 18 integrity, and mitochondrial morphology of podocytes promoting DKD progression. We 19 discovered that forced expression of NDUFS4 in the diabetic environment, however, leads to 20 significant improvement in RSCs assembly and cristae and mitochondrial morphology mitigating 21 DKD progression. We propose that strategies aimed at improving NDUFS4 expression in DKD 22 could emerge as a paradigm shifting intervention for ameliorating progression of DKD. presented as mean ± SEM except for (c). *P< 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001 28 by Mann-Whitney test, FDR Q< 0.01 (c), unpaired two-sided t test (d,g), one-way ANOVA with 29 post-hoc Tukey's test (k,l), test for trend analysis for different classifications of DKD (l)    Median NND (n=4). Data are presented as mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001.

28
One-way ANOVA with post-hoc Tukey's test (k,l,n,o).   34 Western blot assays were performed as described previously 28 . In brief, cells or purified 35 mitochondria were resuspended in RIPA buffer (TEKnova) containing 1% protease inhibitor 36 cocktail (Sigma). Protein concentration was determined using BCA protein assay (Pierce). To  Table 3.

45
Podocyte and tubular cell isolation 46 Podocyte and tubular cell isolation from mouse kidneys were performed as previously described 47 with slight modifications 51,52 . In brief, podocytes were ex vivo selected by biotin-labeled anti-

48
Kirrel3 and podocalyxin antibodies (R&D Systems BAF4910 and R&D Systems BAF1556, 49 respectively), then isolated using magnetic, streptavidin labeled Dynabeads (Thermo Fisher). To 50 isolate tubular cells, dissected and minced kidneys were digested with collagenase type II in 51 RPMI media for 30 mins at 37 o C. Cells were sieved first through a 100 m nylon mesh, then 1 through a 40 m nylon mesh, followed by centrifugation at 500 g for 10 min. The pellet was 2 resuspended in red blood cell lysis buffer (R&D Systems) and incubated on ice for 10min. After 3 centrifuge at 500 g for 10 min, cells were resuspended in RIPA buffer (TEKnova) containing 1% 4 protease inhibitor cocktail (Sigma) and stored at -80 o C freezer for experiments.

6
Mitochondrial isolation 7 Mitochondrial isolation from tissue was performed using Percoll density gradient 8 centrifugation 53 . Mouse isolated podocytes were resuspended in mitochondrial isolation buffer 9 (MIB1: 10 mM HEPES, 250 mM Sucrose, and 1 mM EDTA, pH 7.4, at 4 o C), and homogenized 10 by a glass homogenizer, followed by centrifuging homogenate at 1,300 g at 4 o C for 3 min. After 11 two cycles of homogenization and centrifugation, the pooled supernatant was centrifuged at 12 21,000 g at 4 o C for 10 min. The resultant pellet was resuspended with 15% Percoll in MIB1  according to the manufacturer's instructions. The intensity of MitoSOX Red was analyzed by 1 flow cytometry as previously described 28 .
14 15 CI activity 16 CI activity in tissue sections was assessed by NADH diaphorase staining using kidney frozen 17 sections from mice based on the previous report 43 . NADH oxidoreductase was assayed by 18 incubating kidney sections in 50 mM Tris-HCl (pH 7.4), 0.8 mg/ml NADH (Sigma), and 1 mg/ml 19 nitro blue tetrazolium (NBT, Sigma) for 1hr at RT. After washing in distilled water three times, 20 samples were washed with 3 exchanges of the 30, 60, 90% acetone solutions in increasing then 21 decreasing concentration to remove unbound NBT. After a rinse in distilled water, slides were 22 mounted with the aqueous mounting medium. The intensity of CI activity was measured using 23 image J. To assess in-gel activity for CI, BN gels were incubated in the assay buffer consisting 24 of 5 mM Tris-HCl (pH 7.4) with 0.1 mg/ml NADH (Sigma) and 2.5 mg/ml NBT (Sigma) for 15 min 25 at RT as described previously 32 .

27
Scanning electron microscopy 28 Scanning electron microscopy was conducted as previously reported 29 . In brief, tissue samples  Mitochondrial morphology 9 Mitochondrial morphological measurement was performed as previously described 29 . Briefly, 10 mitochondrial aspect ratio was defined as the major and minor axes of the ellipse expressed as 11 a fraction. Circularity was 4π x (mitochondrial area (Am) per [perimeter (Pm)] 2 ), and roundness 12 was (4 x Am)/ (π x [major axis] 2 ). Feret was the longest distance between any two points along 13 mitochondrial perimeter.

15
Cristae morphological assessment 16 A total of 60 mitochondria in TEM micrographs were analyzed using Image J. To quantify 17 mitochondrial cristae abundance, inner/outer mitochondrial membrane perimeter ratio, total 18 cristae length per mitochondrial area, and cristae junction number per mitochondrial area were 19 measured as previously described 57,58 .

27
Coverslips were washed 3 times in TBS and mounted onto slides. Images were captured by 28 FV1200 MPE confocal microscope (Olympus). Quantification was carried out using Image J.

29
Antibodies and dyes used in this experiment are summarized in Supplementary Table 3. from WT and Ins2 Akita/+ mice were excluded from the analysis, because these low abundant 50 proteins greatly exaggerated the ratios between Ins2 Akita/+ and WT. For complexsome profiling, digitonin-solubilized mitochondria proteins were subjected to BN-GEL analysis. Five putative SC 1 bands, determined as described previously 32 and in Extended Data Fig. 4f, were excited and cut 2 into 1x1 mm pieces followed by in-gel digestion using LysC and trypsin enzymes. The peptides 3 were dried in a speed vac and dissolved in 10 µl of 5%methanol containing 0.1% FA buffer. LC-4 MS/MS analysis was conducted in the same way as described above. The peptides identified 5 from mascot result file were validated with 5% false discover rate (FDR). The gene product 6 inference and quantification were done with label-free iBAQ approach using 'gpGrouper' 7 algorithm 60 . 8 9 APEX proximity labeling 10 APEX proximity labeling was performed as previously described 61,62 . In brief, 80-90% confluent 11 cells were pretreated with 500 M biotin-tyramide for 30 min, followed by 1 mM H2O2 for 1min.

12
After quenching and washing with PBS containing 5 mM Trolox and 10mM sodium ascorbate,  26 Co-IP of NDUFS4 and STOML2 was carried out as previously described 63  GST pull-down assay 36 GST pull-down was carried out as previously described 51 . C-terminus HA-tagged STOML2 wild 37 type and deletion mutants were transiently transfected into HEK 293T cells and then lysed in 38 TNTE buffer (10 mM Tris HCl, pH 7.8, 150 mM NaCl, 1 mM EDTA, and 1.0% Nonidet P-40). and STAR-RED excitation used for STOML2 at 640 nm and the emission collected at 640 nm 1 with the same time gating. The two channels were acquired sequentially, using pixel dwell times 2 of 17 s with a 15 nm pixel size. Mander's coefficients 64 were calculated as the indexes of 3 intensity based colocalization, and distance of particle's center between NDUFS4 and STOML2 4 were obtained using JACoP plugin of ImageJ 65 . Object-based colocalization was defined as the 5 distance less than 140 nm 66 . For STORM imaging, imaging experiments were conducted with a Huygens Essential (Scientific Volume Imaging).

21
Cryo-ET 22 Cryo-grids were prepared by plunge-freezing in a liquid ethane using a Vitrobot Mark IV 23 (Thermo Fisher) that was set to 100 % humidity at 4°C. 2 µl of purified mitochondria sample was 24 applied to 200 mesh, R 2/1 Quantifoil copper grids (Quatifoil) and blotted with Whatman filter 25 paper for 4 sec. Cryo-ET data collection was performed using a Titan Krios G3 300 keV FEG 26 transmission electron cryo-microscope (Thermo Fisher). Cryo-ET images were acquired using a 27 BioQuantum energy filter (Gatan) with slit width set to 20 eV. Images were recorded on a 4k x 28 4k K2 Summit direct electron detector (Gatan) operated in counting mode at nominal 29 microscope magnifications of 26,000x, 33,000x or 19,500x corresponding to pixel sizes of 5.32 30 Å, 4.20 Å and 7.50 Å for NG, HG and HG-DOX, respectively. SerialEM software 68 was used for 31 imaging. Each tilt series was collected from -50° to +50° with increment of 2° using the low dose 32 functions for tracking and focusing. The cumulative dose of each tilt-series was 80-90 e -/Å 2 .

33
Defocus values were set between -8 m and -10 m. For tomogram reconstruction and 34 segmentation, raw movie frames of tilt-series were corrected for beam-induced motion using 35 MotionCor2 69 . The aligned micrographs were imported into EMAN2 70 and were compiled into 36 tilt-series. Automated alignment was performed and 1k x 1k 3D tomograms (bin4) were 37 generated using e2tomogram.py in EMAN2 software package 70 . The references containing 38 features of interest were manually boxed out in tile images of 64 x 64 pixels using EMAN2 39 graphical tool and then used for training Convolutional neural network (CNN). Once a CNN has 40 been trained to recognize a certain feature, it was applied to annotate the same feature in a 41 tomogram. Molecular graphics and visualization were performed with UCSF Chimera 71 .

43
Protein structure prediction and molecular docking 44 Human STOML2 structure was predicted by Contact-guided Iterative Threading ASSEmbly Group data are expressed as mean ± SEM or median ± IQR. Comparisons between two groups 2 were performed using two-sided unpaired Student's t-test for normally distributed data and two-3 sided Mann-Whitney test for non-normally distributed data. Comparisons of multiple groups 4 were performed using one way-analysis of variance (one-way ANOVA) followed by Tukey's 5 multiple comparisons test or Kruskal-Wallis followed by Dunn's multiple comparisons test based 6 on the sample distribution. Multiple Mann-Whitney test (Fig. 1c) and Student's t-test (Extended 7 Data Fig. 1c-e) were followed by two-stage linear step-up procedure with the specific FDR 8 indicated in the figure legend. All tests were performed with GraphPad version 9.3.1 (Graphpad 9 Software), and P values <0.05 were considered to be statistically significant.