Mitochondria are indispensable organelles as they are responsible for the production of the majority of ATP in the cell. Most cellular ATP is generated by oxidative phosphorylation (OxPhos), a process through which electrons are extracted from reducing equivalents and transferred through four different respiratory complexes (RCs) present in the mitochondria inner membrane (CI–CIV). Electron transfer is coupled with the generation of a proton gradient through the mitochondrial membrane that drives the phosphorylation of ADP to ATP by the ATP-synthase complex (also known as complex V).

Although mitochondrial RCs (MRCs) are often depicted as isolated entities, it is now widely accepted that isolated complexes coexist with superassembled structures called supercomplexes (SCs)1,2,3 in the mitochondrial membrane. Within SCs, different MRCs physically interact with each other. There are different subtypes of SCs and their assembly may vary according to the organism, the metabolic state of the cell and the tissue type4,5; however, the main SCs in mammalian mitochondria are composed of complex I, a complex III dimer and one or more copies of complex IV2, whereas complex II is mostly found isolated. These SCs are also called ‘respirasomes’ because they can ‘respire’ by themselves (transfer electrons from NADH to molecular oxygen)6. In the current, most accepted model (the ‘plasticity model’) SCs assemble and disassemble dynamically to accommodate the cellular metabolic needs4,5; however, how the assembly and disassembly of SCs are regulated is not yet fully understood and their functional relevance is still debated7. Evidence suggests that SCs are important for organizing flux of electrons and bolster substrate channeling8,9,10, to limit the leakage of electrons and reduce reactive oxygen species (ROS) production4,11 and to facilitate the assembly and enhance the stability of the single complexes12,13,14,15. The assembly of SCs is compromised in various pathological conditions including cardiovascular diseases16,17, Barth syndrome18,19, neurological diseases20,21, diabetes22 and aging23. We have previously shown that SCs are induced by exercise and involved in exercise adaptation in human24, suggesting a possible role in muscle physiology.

COX7A2L (also known as SCAF1) is the only known SC-specific assembly factor in mammals10. In particular COX7A2L is required for the stable interaction of complex III with complex IV5, thus it plays an important role in the stabilization of the respirasome in most tissues. Previous studies, including our own, have shown that Cox7a2l promotes SC formation as well as metabolic efficiency in zebrafish and mice25,26. COX7A2L expression is induced in certain stress conditions such as hypoxia, endoplasmic reticulum (ER) and nutrient stress and its induction promotes SC formation in mammalian cells27,28,29; however, the function of COX7A2L in supercomplex formation and cellular metabolism is still controversial30,31,32.

In the present study we investigated COX7A2L function in human. We found that genetic variation significantly affects COX7A2L expression in several tissues with skeletal muscle showing the strongest signal. The most significant genetic variant, which is associated with increased COX7A2L expression specifically in muscle tissues, is a 3′ untranslated region (UTR) insertion that creates a short-repeated sequence and enhances COX7A2L expression by increasing messenger RNA (mRNA) stability. Human myotubes bearing this variant have increased amount of SCs and increased respiration. Notably, this highly frequent 3′ UTR variant is associated with lower body fat and improved cardiorespiratory fitness in humans. We validate these findings in C57BL/6J mice, which do not express Cox7a2l protein. Specific reconstitution of Cox7a2l expression in these mice leads to higher maximal oxygen consumption (VO2max), increased muscle mass and increased energy expenditure during the active phase. We further show that Cox7a2l and Cox7a2l-containing SCs are induced upon exercise specifically in the skeletal muscle. To summarize, we found that increased expression of the SC assembly factor, COX7A2L, enhances metabolic and cardiorespiratory fitness in mice and human.


Genetic variation affects COX7A2L expression in human muscle

To gain insights into COX7A2L function in humans we used the GeneBridge analysis platform ( GeneBridge integrates 1,337 datasets with over 265,000 human samples from different tissues and allows to interrogate a gene’s function through its correlation or anticorrelation with other genes for which the Gene Ontology (GO) terms and biological pathway annotations are already known. We found that COX7A2L expression correlates positively with genes belonging to many mitochondria-related ontology terms and pathways (Fig. 1a), including several that were previously unknown, such as complex I biogenesis, cristae formation, mitochondrial translation and respiratory chain complex IV (Fig. 1a and Supplementary Table 1). These associations were conserved across different human tissues (Extended Data Fig. 1a and Supplementary Table 1).

Fig. 1: Genetic variation affects COX7A2L expression in human muscle.
figure 1

a, GO terms and pathways associated with COX7A2L expression in humans. The threshold of significant gene-pathway association is indicated by the red dashed line (Gene-Module Association Score (GMAS) >= 0.268, <=-0.268). GO terms and pathways are organized by similarity. Terms below the significance threshold are shown as gray dots. Known terms assigned to COX7A2L from existing annotations are shown as black dots and new terms above the threshold are shown in red. Dot size represents the GMAS (data from https://systems-genetics.org33). b, Number of significant COX7A2L cis-eQTLs per tissue (top) and number of samples per tissue (bottom) in the GTEx database. Tissue abbreviations are listed in Supplementary Table 3. c, COX7A2L cis-eQTLs by tissue plotted according to their distances from the COX7A2L TSS (x axis). Each dot represents one cis-eQTL and dot colors represent the effects of the alternative alleles on COX7A2L expression (NES, normalized effect size; blue, negative; red, positive). Only the 20 tissues with the highest number of COX7A2L cis-eQTLs are shown. d, Pearson pairwise correlations between tissues based on COX7A2L expression. Non-significant correlations (P > 0.05) are shown as blank squares. Only the 20 tissues with the highest number of COX7A2L cis-eQTLs are shown. e, Correlation of COX7A2L normalized expression in the muscle with expression in the liver (left) or heart (right), two-sided Pearson r and P value are shown above the graphs. f, COX7A2L cis-eQTLs overlap among the five tissues with the highest number of cis-eQTLs; the total numbers of cis-eQTLs per tissue are shown in parenthesis, percentages indicate the proportion of tissue-specific cis-eQTLs. Data source, GTEx database ( v.8 release (b-f).

Given this strong association with mitochondrial function, we asked whether genetic variation affects COX7A2L expression and thus SC formation, in humans. We used the publicly available Genotype-Tissue Expression (GTEx) database v.8 release, which provides whole-genome sequencing and RNA expression data of 54 tissues across approximately 1,000 individuals34 ( We looked at expression quantitative trait loci (eQTLs) located in proximity (±1 Mb) of the COX7A2L transcription start site (TSS) (cis-eQTLs; Extended Data Fig. 1b). Considering an estimated average number of cis-eQTLs per gene of ~180 in GTEx35, we found a high number (1,427) of significant cis-eQTLs for COX7A2L (false discovery rate (FDR) < 0.05, P < 2.2 × 10-4) in 32 different tissues (Fig. 1b and Supplementary Table 2; tissue full names and abbreviations used thereafter are described in Supplementary Table 3). We observed a large variation in the number of COX7A2L cis-eQTLs per tissue (range 1–1,100; Fig. 1b). As sample size significantly affects cis-eQTL mapping35,36, we tested whether the number of COX7A2L cis-eQTLs per tissue was dependent on the sample size. No significant correlation between the number of COX7A2L cis-eQTLs per tissue and the number of samples per tissue was observed (Extended Data Fig. 1c), suggesting tissue-specific regulation of COX7A2L expression. Notably, skeletal muscle harbors the highest number (1,100) and the most significant cis-eQTLs (Fig. 1b,c). Furthermore, muscle tissues (skeletal muscle and heart) are distinct from most other tissues in the effect of the cis-eQTLs alternative allele on COX7A2L expression; the majority of the cis-eQTLs downstream of the TSS are associated with opposite changes in COX7A2L expression in muscle versus most other tissues (Fig. 1c). Indeed, when looking at how the 20 tissues with the highest number of COX7A2L cis-eQTLs correlate on the basis of COX7A2L expression levels, the muscle correlates positively with the heart and negatively with most other tissues such as thyroid, testis, spleen, brain, skin, the intestinal tract (esophagus, ileum and colon) and liver (Fig. 1d,e). To assess tissue-specific regulation of COX7A2L expression, we looked at the proportion of cis-eQTLs that overlap in the top five tissues (muscle, thyroid, testis, esophagus mucosa and skin sun exposed). Skeletal muscle has double the amount of specific cis-eQTLs (18%) compared to the thyroid (9%) and testis (8%) and almost ten times more compared to the skin (2%) (Fig. 1f). A similarly higher proportion of tissue-specific cis-eQTLs was observed in the heart (Extended Data Fig. 1d), suggesting that regulation of COX7A2L expression is specific for muscle tissues. Altogether, our analysis of COX7A2L cis-eQTLs in the GTEx database revealed that several gene variants in human affect COX7A2L expression in different tissues; the skeletal muscle has a unique cis-eQTL profile and the regulation of COX7A2L expression is muscle tissue-specific.

COX7A2L lead cis-eQTL affects mRNA stability

To investigate the mechanism of genetic variation in COX7A2L expression in humans, we focused on the most significant cis-eQTL in the GTEx database (rs4181). Throughout this manuscript, we use the term ‘lead’ cis-eQTL to indicate the cis-eQTL that has the smallest association P value with COX7A2L expression or variants in linkage disequilibrium (LD) with the lead cis-eQTL. Notably, the lead COX7A2L cis-eQTL is located within the gene. Specifically, this variant corresponds to an insertion of 10 bp in the 3′ UTR of COX7A2L (Fig. 2a). The possible genotypes are −/− (reference allele in homozygosis; hereafter REF), −/AAATACACAC (heterozygous; hereafter HET) and AAATACACAC/AAATACACAC (alternative allele in homozygosis; hereafter ALT) (Fig. 2a). This variant is very frequent in the population with a minor allele frequency (MAF) of 0.383 in the 1000 Genomes Project database37 (Supplementary Table 4). Similar MAF is observed in the gnomAD database38 (MAF = 0.393; Supplementary Table 4). The REF allele represents the minor allele in the general population with most individuals being either heterozygous or homozygous for the ALT allele (83%; Fig. 2b). To note, the frequency of this allele is not equal across populations: 80% of the individuals are homozygous for the ALT allele in the Mende population in Sierra Leone, whereas only 3% of the population has this genotype among the Peruvian in Lima37 (Extended Data Fig. 2a and Supplementary Table 4), which suggests that one of the two alleles may be favored in certain environmental conditions. We found that rs4181 is a significant cis-eQTL in 19 different tissues in the GTEx database and has the most robust effect on expression in the muscle and thyroid (P= 1.1 × 10−68 and 1.2 × 10−50, respectively) (Fig. 2c,d). Consistent with our previous observation, the effect of this variant on COX7A2L expression is different according to the tissue: the association of the 10-bp insertion with COX7A2L expression is positive in the muscle, heart and cultured fibroblasts but negative in all other tissues (Fig. 2c,d). This finding again highlights the tissue-specific regulation of COX7A2L expression in humans. To test whether the rs4181 variant is sufficient to modulate expression, we cloned a portion of COX7A2L 3′ UTR with or without the 10-bp insertion in a luciferase reporter vector (Fig. 2e). An increase in luminescence was observed in HEK-293T cells transfected with the luciferase reporter harboring the 3′ UTR with the 10-bp insertion as in carriers of the ALT allele (Fig. 2f).

Fig. 2: COX7A2L lead cis-eQTL affects mRNA stability.
figure 2

a, Schematic representation of rs4181 location and genotypes. b, rs4181 genotype frequency in all populations combined (data from 1000 Genomes Project phase 3)37. c, Normalized expression of COX7A2L in four representative tissues. Data are from the GTEx database, samples are grouped by genotype. The number of samples per tissue per genotype and the cis-eQTL P values are indicated. The statistical analysis used to determine the P value is described in d, NES of rs4181 ALT allele on COX7A2L expression in different tissues in the GTEx database. Dot color represents the directionality of the effect (blue, negative; red, positive) and dot size is proportional to the P value. The statistical analysis used to determine the P value is described in All tissues for which rs4181 is a significant cis-eQTL are shown. Tissues are ordered from bottom to top by P value. e, Schematic representation of the pGL3-promoter luciferase vectors used in f. f, Relative luminescence of HEK-293T cells transfected with pGL3-promoter-empty, pGL3-promoter-3′ UTR-REF or pGL3-promoter-3′ UTR-ALT. Cells were co-transfected with pRL-CMV vector and firefly luminescence was normalized to Renilla luminescence to correct for transfection efficiency (empty n = 8, 3′ UTR-REF n = 8, 3′ UTR-ALT n = 8). g, COX7A2L expression in seven differentiated myotubes lines (REF, n = 3; ALT, n = 4) before and 1 h after inhibition of transcription with actinomycin D (10 μg ml−1). h, Schematic representation of COX7A2L mRNA REF and ALT alleles. Exons (E1–E3) and 5′ UTR and 3′ UTR are depicted with the repeat present only in the ALT allele shown in red. Data are represented as violin plots and Tukey’s box and whiskers plots (superimposed) (c). Box plots center represents the median, lower and upper hinges represent the first and third quartile (25th and 75th percentiles). The whiskers extend to the largest and smallest value up to 1.5 × IQR from the hinge. IQR, interquartile range. Data points beyond the whiskers are plotted individually. Data are represented as mean ± s.e.m. (f,g). Statistical analysis was conducted by one-way analysis of variance (ANOVA) (f) and two-way ANOVA and Sidak’s multiple comparisons test (g).

Source data

To dissect the molecular mechanism by which the 3′ UTR insertion increases expression in cultured cells, we first explored whether the 3′ UTR insertion is a transcriptional enhancer (whether this region increases expression by looping to the COX7A2L promoter). First, we genotyped HEK-293T cells and confirmed that this line is heterozygous for the rs4181 variant (has both REF and ALT alleles; Methods). We then took advantage of the CRISPR-Cas9 technology and co-transfected HEK-293T cells with a guide RNA targeting the COX7A2L promoter and a hemagglutinin (HA)-tagged dCas9 as described39 (Extended Data Fig. 2b). We then performed chromatin immunoprecipitation using an HA-specific antibody and measured by qPCR the enrichment of COX7A2L promoter and the 3′ UTR region containing the rs4181 variant. We also tested another genomic region containing the cis-eQTL rs7572231 located ~20 kb upstream of rs4181. While we were able to immunoprecipitate and enrich for the COX7A2L promoter, no co-enrichment for the two alleles of rs4181 (REF and ALT) or rs7572231 genomic regions was observed (Extended Data Fig. 2c). We thus concluded that the 3′ UTR region of the COX7A2L gene does not loop to the promoter and may not act by increasing transcription itself.

The 3′ UTR regions harbor sequences that can be bound by RNA-binding proteins and modulate RNA levels by affecting transcript stability. To test this hypothesis, we obtained primary myoblasts from seven homozygous individuals (REF, n = 3; ALT, n = 4). These human myoblasts were cultured for 10 d in differentiation medium to induce myotubes differentiation, after which transcription was blocked with actinomycin D and RNA collected after 0 and 1 h. Myotubes homozygous for the REF allele had lower COX7A2L mRNA levels after actinomycin D treatment, suggesting increased mRNA degradation (Fig. 2g). To note, the 10-bp insertion creates a short-repeated sequence in the RNA 3′ UTR (Fig. 2h). Short-repeated sequences separated by 1–10 nucleotides are preferred by several RNA-binding proteins (RBPs) and might favor binding of RBPs harboring multiple RNA recognition motifs or cooperative binding of RBP dimers40.

Taken together, these results demonstrated that COX7A2L lead cis-eQTL is a 3′ UTR insertion that is sufficient to increase gene expression in vitro by increasing mRNA stability.

COX7A2L lead cis-eQTL affects cellular respiration

Next, we assessed the metabolic phenotype of human myoblasts harboring the REF and ALT allele in homozygosis. We first confirmed that differentiated myotubes homozygous for the ALT allele have higher expression of COX7A2L (Fig. 3a). To note, the difference in COX7A2L expression between genotypes becomes only apparent in myotubes, suggesting that it is specific for differentiated muscle cells (Fig. 3a). We then asked whether differences in COX7A2L expression in human myotubes translate in differences in SC formation. To address this, we extracted mitochondria from differentiated human myotubes cultured in glucose-free medium (10 mM galactose) for 48 h and performed blue native (BN)–PAGE. As expected, human myotubes homozygous for the ALT allele had increased SCs (Fig. 3b). We then tested whether increased COX7A2L expression and increased supercomplex formation affects cellular respiration. For this, we measured respiration by Seahorse flux analyzer (Agilent) in differentiated myotubes cultured in glucose-free medium for 48 h. Myotubes homozygous for the ALT allele had higher basal, ATP-linked and maximal respiration (Fig. 3c,d). No significant difference in cellular respiration was observed when cells were cultured in high-glucose medium (25 mM glucose) (Extended Data Fig. 3a,b). Consistent with these results, myotubes homozygous for the ALT allele had higher COX7A2L protein expression, compared to myotubes homozygous for the REF allele, when cultured in glucose-free medium (Fig. 3f), whereas no differences were observed when cells were cultured in high glucose (Extended Data Fig. 3c). Cells in glucose-free medium are forced to rely on mitochondrial respiration to produce ATP and this has been linked to increased COX7A2L expression and increased MRC rearrangement into SCs27. To rule out increased mitochondrial biogenesis driving the observed differences in the ALT allele myotubes, we measured other MRC subunits as well as mitochondrial biogenesis markers both at the protein and mRNA levels in the two genotypes (Fig. 3e,f and Extended Data Fig. 3d) and found no significant differences. Additionally, we measured mitochondrial DNA copy number as a proxy of mitochondrial content in the REF and ALT allele myotubes and found no significant differences (Extended Data Fig. 3e).

Fig. 3: COX7A2L lead cis-eQTL affects cellular respiration.
figure 3

a, COX7A2L mRNA expression in human myoblasts after 0, 1, 4, 7 or 10 d of differentiation (REF, n = 3; ALT, n = 4). b, BN–PAGE immunoblot with UQCRC2 immunostaining (left panels) in isolated mitochondria from differentiated human myotubes cultured in glucose-free medium for 48 h. Coomassie stained gel and complex II (SDHA) immunostaining are shown on the right as loading controls. Each sample is an independent biological replicate. The experiment was repeated twice with similar results. c, Oxygen consumption rate in differentiated human myotubes measured during a mitochondrial stress test using a Seahorse flux analyzer. Cells were differentiated and then grown in glucose-free medium for 48 h. O, oligomycin; F, FCCP (carbonyl cyanide-p-trifluoromethoxyphenylhydrazone); R/A, rotenone/antimycin (REF, n = 3; ALT n = 4). d, Average basal, ATP-linked and maximal respiration calculated from c. OCR, oxygen consumption rate. e, Western blot analysis of differentiated human myotube lines cultured for 48 h in glucose-free medium. f, Western blot quantification of e (REF, n = 3 independent biological replicates; ALT, n = 4 independent biological replicates). Data are represented as mean ± s.e.m. Each dot and each sample used for BN–PAGE and western blot represent a different myoblast line. Statistical analysis conducted was two-way ANOVA (a) and two-sided Student’s t-test (d,f).

Source data

These data show that differentiated myotubes with higher COX7A2L expression have higher SC formation and are able to adapt better to nutrient deprivation by increasing mitochondrial respiration.

COX7A2L lead cis-eQTL affects cardiorespiratory fitness

We next tested whether COX7A2L 3′ UTR insertion is associated with metabolic and cardiorespiratory fitness-related phenotypes in human cohorts. We first looked into the IEU GWAS database (, which contains summary statistics from, among others, the UK Biobank (UKBB), which is a biobank that contains genotype data, as well as a wide variety of phenotypic data for about 500,000 individuals42. We performed two-sample Mendelian randomization, which is a method that allows testing for causal effects of gene expression on phenotypes through a genetic variant43,44. The lead cis-eQTL for COX7A2L in GTEx, rs4181 (previously annotated as rs56873751), is a small insertion that is not present in most GWAS studies. For this reason, we investigated the second-most significant cis-eQTL, rs10183278, a single-nucleotide polymorphism (SNP) that is in high LD with rs4181 (r2 = 0.952; Fig. 4a). rs10183278 was significantly associated with several metabolic phenotypes (FDR < 0.05, P < 0.02; Supplementary Table 5 and Fig. 4b). Notably, an increase in COX7A2L expression was associated with reduced body weight, reduced body fat and body mass index (BMI) and increased usual walking pace (Fig. 4b). The UKBB has only a limited number of phenotypes that allow assessing cardiorespiratory fitness and muscle function. Examples include walking pace45, an indicator of overall physical fitness and grip strength, which is more specific for upper body strength. To validate the role of COX7A2L in overall cardiorespiratory fitness, we hence analyzed participants from the Trøndelag Health (HUNT) study. The HUNT study is a large population-based cohort study that includes questionnaire data, clinical measurements and biological samples from more than 120,000 individuals46. In a sub-study of HUNT (HUNT3 fitness study47,48), maximal oxygen uptake (VO2max) was measured in 4,631 individuals. VO2max measurement is the gold standard for assessing cardiorespiratory fitness49. Again, as the most significant skeletal muscle cis-eQTL for COX7A2L in GTEx, rs4181, is not present in the HUNT cohort, we investigated the second-most significant cis-eQTL, rs10183278 and showed it to be significantly associated with VO2max (Fig. 4c). To further validate these findings, we also analyzed participants from an additional independent cohort, the Helsinki Birth Cohort Study (HBCS). In this cohort several tests were performed to assess physical fitness in 1,078 individuals aged 67–77 years50, of which 606 took part in the Urho Kaleva Kekkonen (UKK) 2-km walk test. This test evaluates cardiorespiratory fitness without the requirements for maximal physical efforts51,52 and is a valid substitute for maximal physical efforts tests (VO2max) in elderly individuals and patients with obesity53,54. Mendelian randomization analysis of rs10183278 indicated that increased COX7A2L expression in the muscle was associated with several HBCS phenotypes (FDR < 0.2; Supplementary Table 5) with UKK showing the biggest effect (effect size of 7.27 test score units COX7A2L normalized effect size, P = 0.02; Fig. 4d and Supplementary Table 5). Specifically, individuals with higher COX7A2L expression in the muscle (alternative allele) have 11% higher fitness index (average fitness index 78.7 ± 28 (REF), 87.4 ± 23 (HET) and 87.4 ± 20 (ALT); Fig. 4e). To summarize, using three independent human cohorts, we showed that COX7A2L lead cis-eQTL is associated with metabolic and cardiorespiratory fitness-related phenotypes.

Fig. 4: COX7A2L lead cis-eQTL affects cardiorespiratory fitness.
figure 4

a, Regional plot of the COX7A2L genomic region. COX7A2L muscle cis-eQTLs within 1 Mb of the TSS are plotted according to their position on Chr2 (x axis) and their P value (y axis). Variants are color-coded according to their LD (r2) with the lead COX7A2L cis-eQTL variant rs4181. The gray vertical shade indicates the location of COX7A2L gene. b, Mendelian randomization analysis on the second-most significant COX7A2L muscle cis-eQTL (rs10183278) and metabolic and fitness-related phenotypes in the UKBB. Representative phenotypes are shown (Supplementary Table 5). c, Association between the second-most significant COX7A2L cis-eQTL (rs10183278) with VO2max measured in the HUNT3 study. d, Mendelian randomization analysis between the second-most significant COX7A2L cis-eQTL (rs10183278) and the 2-km walk test (UKK) in the HBCS (Supplementary Table 5). e, Fitness index of the HBCS participants. Individuals are grouped by genotype. The number of participants per genotype is indicated. Data are represented as mean ± s.e. (bd) and violin plots and Tukey’s box and whiskers plots (superimposed) (e). Box plots center represents the median, lower and upper hinges represent the first and third quartile (25th and 75th percentiles). The whiskers extend to the largest and smallest value up to 1.5 × IQR from the hinge. Data points beyond the whiskers are plotted individually. Statistical analysis used to determine the P value is described in Methods (bd).

A Cox7a2l mutation affects cardiorespiratory fitness in mice

Human genetic analyses have the limitation that several confounding factors, including environmental factors or other genetic variants co-inherited with the variant of interest, may affect the observed association of an SNP with an outcome. Environmental factors are difficult to control and correct for in human but can be tightly controlled in mice. To determine whether specific changes in Cox7a2l expression are associated with metabolic and cardiorespiratory fitness-related phenotypes, we took advantage of the presence of a known Cox7a2l genetic polymorphism in common mouse inbred strains. C57BL/6J mice have a six-nucleotide deletion in the Cox7a2l gene that truncates the protein by two amino acids and renders it unstable10 and incapable to interact with complex IV5. DBA/2J mice, as well as most other mouse strains and humans have the long, stable Cox7a2l form (Extended Data Fig. 4a). We thus looked into the BXD mouse genetic reference population; a large panel of well-characterized recombinant inbred strains derived from crosses between C57BL/6J and DBA/2J mice26,55,56,57 ( This population segregates for the Cox7a2l polymorphism. Specifically, the BXD strains bearing the C57BL/6J allele for Cox7a2l (BB allele) have higher basal levels of Cox7a2l mRNA, compared to the strains bearing the DBA/2J allele (DD allele), likely due to a compensatory mechanism (Extended Data Fig. 4b, left), but as expected have lower levels of Cox7a2l protein5 (Extended Data Fig. 4b, middle). In agreement with the original observation in C57BL/6J mice10, we previously showed that BXD strains carrying the Cox7a2l BB allele lack some of the SCs and that this trait maps to a QTL containing the Cox7a2l gene26. Here we observed that, on a high-fat diet, strains with the DD allele (long Cox7a2l form) have higher VO2max after training (Extended Data Fig. 4b, right). No difference in VO2max between mice carrying the BB and DD allele was observed in the BXD population at baseline or on chow diet.

C57BL/6J mice are also homozygous for a loss-of-function mutation in the nicotinamide nucleotide transhydrogenase (Nnt) gene58 that renders them more susceptible to metabolic dysfunctions59. To rule out an effect of the Nnt mutation on the observed differences in VO2max between the BXD strains with the long and short Cox7a2l allele, we checked how the Nnt allele distributes between the two groups. We found an about equal distribution of DD (wt) and BB (null) Nnt allele among the two Cox7a2l genotypes (Extended Data Fig. 4c). Specifically, among the strains with the Cox7a2l DD allele, 45% have the Nnt BB allele and 50% have the Nnt DD allele. While among the strains with the Cox7a2l BB allele, 36% have the Nnt BB allele and 57% have the Nnt DD allele. Thus, the observed difference in VO2max is not due to differences in the Nnt gene.

To address the specific contribution of Cox7a2l to phenotypic differences, we used congenic C57BL/6J mice expressing the long (functional) form of Cox7a2l (thereafter referred to as C57BL/6JCox7a2l)5 (Fig. 5a). Consistent with our previous findings in the BXD strains26 (Extended Data Fig. 4b), as well as with the phenotypic associations found with COX7A2L cis-eQTLs in the human genetic analysis of the UKBB, HBCS and HUNT datasets (Fig. 4b–e), C57BL/6JCox7a2l mice have higher VO2max (with a tendency for an increased run distance), lower body weight, more muscle mass and less fat deposits compared to C57BL/6J mice (Fig. 5b–e). To address Cox7a2l-driven metabolic differences, we housed C57BL/6J and C57BL/6JCox7a2l mice in Promethion metabolic cages with or without access to a running wheel (Methods). We did not observe changes in total activity and wheel activity between the two genotypes (Extended Data Fig. 5a,b). Notably, C57BL/6JCox7a2l mice had increased energy expenditure specifically during the active dark phase and this difference was more pronounced when mice had access to a running wheel (Fig. 5f and Extended Data Fig. 5c,d). While the body temperature of C57BL/6JCox7a2l was lower than C57BL/6J at rest, after exercise C57BL/6JCox7a2l had a higher body temperature increase (+2.6 °C versus +1 °C, respectively; Fig. 5g). This temperature shift was significantly higher also when expressing it as percentage of the initial body temperature (+7.1% versus +2.7%; Fig. 5h). These results suggest that mice with the long Cox7a2l allele can increase energy expenditure more when energy demands rise, compared to mice with the short Cox7a2l allele. Consistent with the observed increase in energy expenditure during the active phase, C57BL/6JCox7a2l mice had higher total and relative food intake (Fig. 5i), despite having lower body weight (Fig. 5c). Therefore, all together, these results confirm a specific role of COX7A2L/Cox7a2l in cardiorespiratory fitness in both human and mice.

Fig. 5: A Cox7a2l mutation affects cardiorespiratory fitness in mice.
figure 5

a, To dissect the specific role of Cox7a2l, we compared C57BL/6J mice expressing the long functional allele of Cox7a2l (red) with wild-type C57BL/6J mice (blue) expressing the non-functional short allele of Cox7a2l (ref. 5). b, VO2max and run distance (C57BL/6J n = 19, C57BL/6JCox7a2l n = 20). c, Body weight (C57BL/6J n = 17, C57BL/6JCox7a2l n = 16). d, Body composition measured by EchoMRI (C57BL/6J n = 17, C57BL/6JCox7a2l n = 16). e, Tissue weights normalized to total body weight (C57BL/6J n = 15, C57BL/6JCox7a2l n = 12). f, Energy expenditure measured using the Promethion metabolic cages in standard housing condition or with free access to a running wheel (yellow bar). Hour averages (top) and day and night averages (bottom) (C57BL/6J n = 15, C57BL/6JCox7a2l n = 15). g, Body temperature measured before and after 30 min of treadmill run (C57BL/6J n = 5, C57BL/6JCox7a2l n = 5). h, Body temperature shift after 30 min of treadmill run expressed as percentage of the initial body temperature (C57BL/6J n = 5, C57BL/6JCox7a2l n = 5). i, Food intake measured using Promethion metabolic cages. Total food intake (left) and food intake normalized to body weight (right) (C57BL/6J n = 15, C57BL/6JCox7a2l n = 15). Data are represented as mean ± s.e.m. (f, top) and box and whiskers (bi). Box plot centers represent the median and lower and upper hinges represent the first and third quartiles (25th and 75th percentiles). The whiskers extend to the largest and smallest value. Statistical analysis was conducted by two-sided Student’s t-test (be,h,i), three-way ANOVA and Sidak’s multiple comparisons test (f) and two-way ANOVA and Sidak’s multiple comparisons test (g).

Source data

Cox7a2l-containing SCs are induced upon exercise

Given the specific role of COX7A2L in cardiorespiratory fitness and energy expenditure during physical activity and as SCs are induced by exercise in humans24, we addressed the role of Cox7a2l in exercise. C57BL/6J and C57BL/6JCox7a2l mice were given free access to a running wheel for 5 weeks and then killed. Both Cox7a2l total protein and Cox7a2l-containing SCs were increased in the skeletal muscle of C57BL/6JCox7a2l mice after 5 weeks of training, but were absent in C57BL/6J mice (Fig. 6a–f). Notably, only C57BL/6JCox7a2l mice showed an induction of the higher respirasome bands and III2 + IV SC upon exercise (Fig. 6c–e). This increase was likely muscle tissue-specific as no changes were observed in Cox7a2l protein expression and the levels of Cox7a2l-containing SCs upon exercise in the liver (Extended Data Fig. 6a,b). The increase in Cox7a2l expression upon exercise was not dependent on the genetic background, as it was also observed in DBA/2J mice and was proportional to the duration of the training (Extended Data Fig. 6c). Given the observed association of Cox7a2l with other mitochondrial ontologies (Fig. 1a) we assessed possible differences in mitochondrial gene expression and content in the two genotypes. We measured levels of individual mitochondrial complexes by western blot and measured PGC1a and TFAM as markers of mitochondrial biogenesis both at mRNA and protein levels (Fig. 6a,b and Extended Data Fig. 6e). We found an overall increase in the amount of mitochondrial complexes as well as PGC1a and TFAM protein levels upon endurance exercise, this was expected as endurance exercise is known to increase mitochondrial biogenesis (Fig. 6a,b). The increase in the mitochondrial complexes, PGC1a and TFAM was equal in both genotypes. Thus, upon exercise, the observed increase in the Cox7a2l-containing respirasome bands in the mice expressing the long isoform of Cox7a2l is not simply due to an overall increase in the single complexes but depends on Cox7a2l. Furthermore, C57BL/6JCox7a2l mice, but not C57BL/6J mice, had decreased mitochondrial DNA (mtDNA)/nuclear DNA (nDNA) ratio in the muscle (but not the liver) after exercise (Fig. 6g and Extended Data Fig. 6d). This phenomenon was also observed in humans60,61 and might suggest increased mitochondrial turnover and remodeling with long-term training (Discussion). Altogether, these data suggest a Cox7a2l-dependent increase in respirasome formation upon exercise.

Fig. 6: Cox7a2l-containing SCs are induced upon exercise.
figure 6

a, Western blot analysis of the indicated proteins at rest or after 5 weeks of wheel running exercise in gastrocnemius muscle. b, Western blot quantification of the results shown in a. cf, BN–PAGE immunoblot with total OxPhos immunostaining (c), Ndufs3 immunostaining (d), Uqcrc2 immunostaining (e) and Cox7a2l immunostaining (f) in isolated mitochondria from gastrocnemius muscle at rest or after 5 weeks of wheel running exercise. Arrows represent the SCs recovered with the functional long allele of Cox7a2l. g, Mitochondrial-to nuclear DNA ratio measured in the gastrocnemius muscle at rest or after 5 weeks of wheel running exercise. Data are represented as mean ± s.e.m. (b) and box and whiskers (g). Box plot centers represent the median, lower and upper hinges represent the first and third quartiles (25th and 75th percentiles). The whiskers extend to the largest and smallest value. Statistical analysis was conducted by two-way ANOVA and Sidak’s multiple comparison. C57BL/6J sedentary n = 5, C57BL/6J exercise n = 6, C57BL/6JCox7a2l sedentary n = 5, C57BL/6JCox7a2l exercise n = 7.

Source data


The assembly of MRCs in SCs is widely accepted; however, their functional importance has so far remained elusive. While several studies have found that SC assembly may be beneficial to enhance electron transport efficiency, channel substrates, reduce ROS production and stabilize single MRCs4,8,9,11,12,13,15,62, other studies have challenged these observations63,64. COX7A2L is required for the interaction of complex III with complex IV and is important for the formation of the respirasome5,26; however, these findings have also been contested31,32,65,66,67,68. Most structural models of the mammalian respirasome come from mitochondria isolated from heart at rest; however, it is now clear that SC assembly and the role of assembly factors may be dependent on the tissue type and metabolic status of the cell. Although only a few studies have investigated the physiological function of COX7A2L, a recent study suggested that it may promote metabolic efficiency in zebrafish and mice25. Nonetheless, the possible function COX7A2L in cellular bioenergetics is still controversial30 and its role in human is unknown.

To shed light on this conundrum, we sought to determine the function of COX7A2L in human. We found that genetic variation strongly affects COX7A2L expression. We identified the highest number and most significant cis-eQTLs in the skeletal muscle. Although several other tissues also had a high number of significant cis-eQTLs with high overlap between tissues, the effect of the alternative allele was opposite in muscle tissues compared to most other tissues. This suggests a muscle-specific mechanism mediating the effect of the genetic variant on gene expression. COX7A2L lead cis-eQTL is a 3′ UTR insertion that is sufficient by itself to increase expression in cultured cells. We identify a possible mechanism by which this 3′ UTR variant affects expression. The insertion creates a short-repeated sequence that increases mRNA stability, possibly through increased binding of an mRNA stabilizing factor or through disruption of a microRNA binding site. Further investigations are needed to identify the detailed molecular mechanism and to determine why this variant might have a negative effect on gene expression in most other tissues such as the liver and thyroid. In our analysis we found other independent variants that affect COX7A2L expression, thus we cannot exclude that other independent causal variants exist and act through different mechanisms.

Increased COX7A2L expression in human myoblasts leads to increased amounts of SCs and increased cellular respiration when cells are cultured in galactose medium; this is in line with previous reports showing that SCs may play a role in certain metabolic stress conditions and that COX7A2L may be important when cells are forced to use oxidative phosphorylation to produce ATP27. Physical exercise causes metabolic stress in muscle tissues. The increased energy demand from muscle contraction leads to a surge in cellular AMP levels and to a decline in oxygen levels that triggers mitochondrial adaptations to decreased energy levels and hypoxia. We show that Cox7a2l is induced in skeletal muscle upon exercise in mice and that humans with higher expression of COX7A2L have improved metabolic and cardiorespiratory phenotypes. While some studies have demonstrated a rearrangement of mitochondrial RCs into supercomplexes upon exercise training24,69, one recent study reported that increased mitochondrial biogenesis, rather than rearrangement of mitochondrial complexes into supercomplexes is responsible for improved muscle bioenergetics upon exercise70. Here, while we observe an increase in mitochondrial biogenesis and in the amount of OxPhos complexes upon exercise in mice, we demonstrate that the absence of Cox7a2l protein and the lack of respirasome formation affect exercise performance and energy expenditure. Of note, mice expressing Cox7a2l have a decreased mtDNA/nDNA ratio after exercise. mtDNA synthesis is an energy-consuming process that may be less favored in situations of reduced cellular ATP, such as during or after physical effort. Furthermore, exercise leads to increased ROS production and oxidative stress in the muscle that may damage mtDNA71,72. Thus, it is plausible that improved mitochondrial efficiency, associated with mitochondrial turnover and remodeling, rather than overall increase in mtDNA synthesis, may be favored in certain conditions of metabolic stress. Such a decrease in mtDNA/nDNA ratio in skeletal muscle upon exercise was also observed in humans60,61, suggesting a conserved adaptive response. Our findings suggest that individuals having genetic variants associated with increased COX7A2L expression in the muscle may be able to better adapt to situations of increased metabolic stress and physical effort demand, such as during physical activity. Notably, we found that the frequency of COX7A2L lead eQTL is remarkably different in certain populations (Extended Data Fig. 2a), suggesting a possible adaptive role of SCs in different environmental conditions; however, further research is needed to support this hypothesis. Previous studies have identified several human genetic polymorphisms in nuclear and mtDNA affecting mitochondrial gene expression73, metabolism74 and exercise performance75,76 as well as exercise-induced changes in mitochondrial gene expression77. Co-inheritance of these genetic polymorphisms with the COX7A2L variants may be used in the future for metabolic risk prediction.

The findings in our study not only shed light on the genetic underpinnings of how variants in COX7A2L affect mitochondrial supercomplex formation in human muscle, but also demonstrate the importance of COX7A2L and SCs for cardiometabolic fitness and human health. Cardiorespiratory fitness is an important predictor of cardiovascular and metabolic health78,79. Improving cardiorespiratory fitness and exercise capacity is a highly promising approach to reduce all-cause and cardiovascular mortality79.


Helsinki Birth Cohort

The HBCS includes 13,345 individuals born in Helsinki between 1934 and 1944. The clinical study includes 2003 participants and the protocol was approved by the Ethics Committee of Epidemiology and Public Health of the Hospital District of Helsinki and Uusimaa. Written informed consent was obtained from each participant before any study procedure was initiated. The UKK 2-km walk test has been validated against maximal effort tests by treadmill or bicycle ergometry in multiple populations, including individuals with obesity53 and elderly individuals54. During the test, participants are instructed to walk a 2-km course on flat ground as fast as possible. The test results are expressed as a fitness index, which takes into account the individual’s age, sex, BMI, time spent in walking and heart rate. The senior fitness test, which describes the physical fitness of the study participants was performed in 695 individuals and was described previously80. Isometric grip strength of the dominating hand was tested by a Newtest Grip Force dynamometer (Newtest Oy). The maximum value of three squeezes was used in analyses of linear regression. Weight and height were measured at the time of the grip strength measurements and BMI was calculated as weight (kg) divided by height (m) squared.

DNA was isolated from blood and genotyping was carried out with the modified Illumina 610k chip by the Wellcome Trust Sanger Institute. Genotype imputation was performed using the 1000 Genomes Project Phase I integrated variant set (v.3/April 2012; NCBI build 37/hg19) and IMPUTE2 software. Before genotype imputation some quality control filters were applied: SNP clustering probability for each genotype >95%; call rate >95% individuals and markers (99% for markers with MAF < 5%); MAF > 1%; and Hardy–Weinberg equilibrium P > 1 × 10−6. Moreover, heterozygosity, sex checks and relatedness checks were performed and any discrepancies were removed. Phenotype association analyses in HBCS were performed in individuals who had available genotype and senior fitness test data, approximately 400 individuals for the analysis. The scores of the fitness tests were classified based on fifth percentile range, with a score of 1 being the worst performance (score below fifth percentile); 2 the score from the fifth to ninth percentile; and 20 the best performance (in or above the 95th percentile) as described50. We performed linear regressions with SNPtest81 assuming an additive genetic model. We adjusted all models for age, sex, highest education achieved (basic or less/upper secondary/lower tertiary/upper tertiary) and smoking (yes/no).

The Trøndelag Health study

The HUNT study and HUNT3 sub-study (HUNT3 fitness study47,48) were described previously48. The HUNT3 fitness study was approved by the Regional Committee for Medical Research Ethics (2012/1672/REK nord), the Norwegian Data Inspectorate and the National Directorate of Health and is in compliance with the Helsinki Declaration. Written informed consent was obtained from all participants. The 4,631 healthy adult participants performed a measure of VO2max. Exclusion criteria for the HUNT3 fitness study were described previously48. VO2max measurement for this study was described previously48. The 4,463 participants who reached VO2max were included in the analyses.

DNA was extracted from blood samples, as described previously82. Genotyping of the HUNT study was described previously83. In brief, genotyping was performed with one of three different Illumina HumanCoreExome arrays (HumanCoreExome12 v.1.0, HumanCoreExome12 v.1.1 and UM HUNT Biobank v.1.0) according to standard protocols. The genotypes from different arrays had quality control performed separately and were reduced to a common set of variants across all arrays. Samples that failed to reach a 99% call rate, had contamination >2.5% as estimated by BAF Regress, large chromosomal copy number variants, lower call rate of a technical duplicate pair and twins, gonosomal constellations other than XX and XY or whose inferred sex contradicted the reported sex were excluded. The genomic position, strand orientation and reference allele of genotyped variants were established by aligning the probe sequences against the human genome using the BLAT tool84. With PLINK v.1.90 (ref. 85), variants were excluded if probe sequences could not be perfectly mapped and cluster separation was <0.3, Gentrain score <0.15, showed divergence from Hardy–Weinberg equilibrium in unrelated samples of European ancestry (P < 0.0001), had call rates <99% or another array with higher call rate genotyped the same variant. Ancestry of samples was derived from projecting genotyped samples into the principal components of the Human Genome Diversity Project reference panel. Different arrays were matched by narrowing to a set of overlapping variants and eliminating variants with frequency differences >15% between datasets, or that were monomorphic in one and had MAF > 1% in another dataset. These genotype data were phased using Eagle2 (v.2.3.8)86. Imputation was performed on samples of recent European ancestry using Minimac3 (v.2.0.1; and a merged reference panel built on the Haplotype Reference Consortium panel88 (release v.1.1) together with a local reference panel, which included 2,202 whole-genome sequenced HUNT study participants. Relatively unrelated individuals (kinship coefficient <0.0884) were chosen using KING89 and FastIndep90 tools. BOLT-LMM algorithm91 was employed to execute statistical analyses while variants with MAF < 0.001192 were excluded from the analysis. This algorithm computes statistics for testing association between phenotype and genotypes using a linear mixed model. The analysis was adjusted for sex and age. Genetic principal components were included as covariates in the analysis to control for residual population structure. The study was in conformity with Norwegian laws and the Helsinki Declaration and signed informed consent was obtained from all participants.

Mouse lines

C57BL/6J mice expressing the long functional Cox7a2l allele were generated as described5. The C57BL/6J OlaHsd used to generate the C57BL/6JCox7a2l mice carried the wild-type allele of the Nnt gene. For genotyping, DNA was extracted from tails and PCR amplified using the KAPA2G mix (KK5103, Kapa Biosystems) following manufacturer’s instructions (Supplementary Table 6 lists the primer sequences). PCR products were run on a QIAxcel instrument to identify different molecular weight bands indicating presence or absence of the six-nucleotide deletion (50 or 56 bp). For all experiments, littermates of the same sex were randomly assigned to experimental groups. All experiments were performed on male mice at 10–14 weeks of age. All mice were fed a chow diet (SAFE 150) and were kept in standard housing conditions (12-h light–12-h dark cycle and temperature of 22 °C). All animals had ad libitum access to food and water. All animal experiments were performed according to Swiss ethical guidelines and were approved by the Service de la Consommation et des Affaires Vétérinaires of the Canton de Vaud (license VD3419)

Cell lines

All cell lines were cultured in a humidified incubator at 37 °C and 5% CO2. HEK-293T (CRL-3216, ATCC) cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM; Gibco, 41966029) supplemented with 10% fetal bovine serum (Gibco, 10270106) and 1% penicillin/streptomycin (Gibco, 15140122). Human myoblast lines from non-diseased individuals were obtained from the CBC BioTec Centre de Biologie et Pathologie Est, Groupement Hospitalier Est and were maintained at a maximum confluency of 70% in myoblast culture medium (Ham’s F-10 Nutrient Mix (Gibco, 31550031), 12% fetal bovine serum and 1% penicillin/streptomycin). For differentiation, myoblasts were seeded in culture medium and after 3 d, when they reached ~80–90% confluence, the culture medium was replaced with differentiation medium (DMEM/F12 (Gibco, 31330038), 2% horse serum (Gibco 16050122) and 1% penicillin/streptomycin (Gibco, 15140122)). Trypsin-EDTA 0.05% (Gibco, 25300062) was used for detachment.

DNA extraction and genotyping of cell lines

DNA extraction was performed using NucleoSpin Tissue mini kit (Macherey-Nagel, 740952). For genotyping, primers flanking the rs4181 10-bp insertion site were used for PCR amplification (primer sequences are listed in Supplementary Table 6). PCR was performed using the KAPA2G mix (KK5103, Kapa Biosystems) following manufacturer’s instructions. PCR products were run on a QIAxcel instrument to identify different molecular weight bands indicating presence or absence of the insertion (425 or 435 bp). PCR products were further purified using the PCR clean-up gel extraction kit (Macherey-Nagel, 740609) and sequenced by Sanger sequencing (Microsynth) to confirm electrophoresis results. Only homozygous myoblast lines were used for experiments.

GeneBridge analysis

Ontology terms and pathways correlated to COX7A2L in humans were identified using the publicly available GeneBridge tool ( First, an analysis was performed with all available tissues in the database. GO terms and pathways with GMAS ± 0.268 were considered significantly associated with COX7A2L; the threshold of significance was determined as previously described33 (Fig. 1a). Then, 13 main tissues were selected and the same analysis was repeated for each tissue separately. GO terms and pathways with score ± 0.268 in at least one of the 13 tissues are represented in Extended Data Fig. 1a.

Genotype-Tissue Expression analyses

For the cis-eQTL analysis we used publicly available data from the GTEx Consortium34 ( v.8 release. Only cis-eQTLs (eQTLs within 1 Mb from the COX7A2L gene) are included in the analysis. For the correlation analyses in GTEx, we obtained the publicly available gene transcripts per million and the covariates used in the cis-eQTL analysis (dbGAP study accession no. phs000424.v8). To avoid correlations due to known or unknown factors and batch effects, we used the removeBatchEffect effect function from the limma R package v.3.38.3 (ref. 93) along with covariates as calculated and released by the GTEx consortium.

Mendelian randomization

For the Mendelian randomization analysis, we made use of the cis-eQTL data from GTEx34, which are freely available from rs4181 was not genotyped in the UKBB and the HBCS, therefore, we selected the second-most-significant muscle cis-eQTL, rs10183278, as the instrumental variable for our analysis. This cis-eQTL is in high LD with rs4181 (r2 = 0.952). As with rs4181, rs10183278 is also significantly linked with EML4 expression, a gene whose 3′ terminus lies only 997 bp upstream of COX7A2L, but this association is much weaker; the effect size on COX7A2L expression in muscle is 0.44 (P = 3.7 × 10−68) versus an effect size of −0.12 for EML4 (P = 2.6 × 10−5). We also verified rs10183278 in PhenoScanner v.2 ( and found no obvious confounders. The results of this analysis are in Supplementary Table 7. Muscle gene expression of COX7A2L was used as the exposure variable. We used the available_outcomes function from the TwoSampleMR v.0.5.3 R package to obtain 188 body weight- and health-related outcome summary statistics from the IEU GWAS database ( (Supplementary Table 5). For the HBCS, we used the summary statistics of the additive genetic model (see section on Helsinki Birth Cohort). Statistical inference was performed through Wald ratio tests as implemented in the harmonise_data function of the TwoSampleMR package and P values were converted to q values with the Benjamini–Hochberg FDR procedure.

Body composition

Whole-body composition was measured by NMR using a Minispec instrument (Bruker).

Metabolic cages

Energy expenditure, VO2, VCO2, food intake, cage and wheel activity were measured using the Promethion system (Sable Systems International). Mice were housed in metabolic cages for 4 d and had free access to a running wheel. To measure energy expenditure and cage activity in standard housing conditions, the wheel was kept blocked for the first 2 d of recording.

VO2max test in mice

VO2max in mice was measured with a calorimetric treadmill (Columbus Instruments) with an incremental speed protocol. To avoid anxiety created by finding themselves in a new environment, mice were habituated to the device for 10 min before starting the test. The run distance, VO2 and VCO2 were measured. The experiment was stopped when mice were exhausted or VO2max was reached (when VO2 levels failed to increase despite increasing running velocity or when respiratory exchange ratio was =1.0).

Long-term training

For long-term training experiments mice were housed singularly and had access to a running wheel for 5 weeks. After a first estimation of the average distance run per day by the first mouse cohort, the maximum allowed distance was set on this average (1,800 m per day) after which the wheel was blocked for all mice to avoid large variability between mice and between experiments. The meters run on the wheel was monitored for each mouse and for all the duration of the experiment. There were no significant differences in the total distance run between the two genotypes. Mice in control groups were also housed singularly and had access to a wheel; however, the wheel was blocked for the entire duration of the experiment.

Treadmill exercise and body temperature measurement

Body temperature was measured using a rectal probe before and after 30 min of treadmill run at incremental speed (10 min at 20 cm s−1, 10 min at 23 cm s−1 and 10 min at 26 cm s−1). The mice were acclimatized to the treadmill for 3 d before the experiment. All animals satisfactorily reached the end of the test.

BN–PAGE immunostaining

BN–PAGE was performed as described95,96 with some modifications. Briefly, ~15–30 mg of frozen tissue or ~20 × 106 cells were homogenized to isolate mitochondria and mitochondria protein content was quantified as described96. For BN–PAGE immunoblotting, 50 μg (liver and muscle) or 10 μg (myotubes) of mitochondria extract was loaded in the gel. Electrophoresis of solubilized mitochondrial proteins was performed as described96. For immunoblotting, samples were run at 150 V for 30 min and at 250 V for additional 90 min. After electrophoresis the gel was soaked for 30 min in Tris/glycine transfer buffer (25 mM Tris, 192 mM glycine, 10% methanol and 0.1% SDS). Proteins were transferred in Tris/glycine transfer buffer on a PVDF membrane using a wet transfer system (Bio-Rad). Blocking and antibody incubations were performed in 3% BSA diluted in TBS-T. Membranes were incubated with primary antibodies: total OXPHOS Rodent WB Antibody Cocktail (Abcam, ab110413, 1:1,000 dilution), anti-NDUFS3 (Abcam, ab14711, 1:2,000 dilution), anti-UQCRC2 (Abcam, ab14745, 1:10,000 dilution), anti-MTCO1 antibody (Abcam, ab14705, 1:1,000 dilution) and anti-COX7A2L antibody (St John’s Laboratory, STJ110597, 1:1,000 dilution) to detect Cox7a2l-containing SC bands. Anti-SDHA antibody (Abcam, ab14715, 1:1,000 dilution) was used as loading control. Protein signal was detected using chemiluminescence (WesternBright ECL, Advansta) and imaged using the Fusion FX6 imaging system (Vilber).

RNA extraction and qRT–PCR

RNA was extracted using TriPure Isolation Reagent (Roche, 11667165001). For the RNA extraction from myoblasts and differentiated myotubes, two wells of a six-well plate were pooled for each sample and 1 μl of glycogen (R0551, Thermo Fisher) was added during the extraction for maximum recovery of the RNA and visualization of the pellet. For reverse transcription, PrimeScript RT Reagent kit with gDNA Eraser (Takara, RR047B) was used following manufacturer’s instructions. For qPCR TB Green Premix Ex Taq (Takara, RR420W) was used and plates were run on a LightCycler 480 instrument (Roche). Primer sequences are listed in Supplementary Table 6. All qRT–PCR fold changes were calculated using Gapdh/GAPDH as housekeeping gene. The average of three technical replicates was used for each data point.

MtDNA/nucDNA ratio

MtDNA abundance was measured with the ΔΔCt method as previously described97. Briefly, ~15 mg of frozen liver or muscle tissues were used to extract total cellular DNA using NucleoSpin Tissue kit (Macherey-Nagel, 740952). For qPCR, 20 ng of DNA was used. qPCR was carried out as described above. For mtDNA amplification, primers against the 16S rRNA were used (Supplementary Table 6). For nuclear DNA amplification primers against the β2M gene were used (Supplementary Table 6). For mtDNA copy number estimation in human myoblasts the Human Mitochondrial DNA Monitoring Primer Set (Takara, 7246) was used.

Western blot

For tissue samples, ~10 mg of frozen tissue were homogenized in RIPA buffer with protease inhibitors (Thermo Fisher, 78429) using a small pestle, mixed for 20 min at 4 °C and then spun for 20 min at 11,481 r.p.m. at 4 °C to eliminate cell debris. For myotube samples, three wells of a six-well plate were pooled together for one replicate and resuspended in RIPA buffer with protease inhibitors. Samples were incubated on ice, sonicated briefly to ensure complete cell lysis and spin for 20 min at 11,481 r.p.m. at 4 °C to eliminate cell debris. Protein amount was quantified using the DC protein assay (Bio-Rad). Proteins were separated by SDS-polyacrylamide gel electrophoresis and transferred onto PVDF membranes using a wet transfer system (Bio-Rad). Blocking and antibody incubations were performed in 5% BSA diluted in TBS-T. The following primary antibodies were used: anti-Cox7a2l (St John’s Laboratory, STJ110597 and Proteintech, 11416-1-AP, 1:1,000 dilution), total OXPHOS Rodent WB Antibody Cocktail (Abcam, ab110413, 1:1,000 dilution), anti-Vinculin (Abcam, ab129002, 1:1,000 dilution), anti-TFAM (Abcam, ab131607, 1:1,000 dilution), anti-PGC1a (Calbiochem, ST1202, 1:1,000 dilution) and anti-VDAC1 (Abcam, ab14734, 1:1,000 dilution). Protein signal was detected using chemiluminescence (WesternBright ECL, Advansta) and imaged using the Fusion FX6 imaging system (Vilber).

Luciferase reporter assay

A 100-bp fragment of the human COX7A2L 3′ UTR surrounding the rs4181 insertion variant with (3′ UTR-ALT) or without (3′ UTR-REF) the 10-bp insertion was cloned in a pUC57 vector by Genscript (for the insert sequence see Supplementary Table 6) flanked by XbaI and FseI restriction sites at the 5′ and 3′ respectively. The 100-bp fragment was excised and cloned in the pGL3-promoter vector (Promega) downstream of the luciferase gene using XbaI and FseI restriction enzymes (NEB). The obtained plasmids were sequenced to check the correct directionality of the inserted sequence (Microsynth). For the luciferase assay, HEK-293T cells were transfected in 96-well plates with pGL3-promoter-3′ UTR-REF, pGL3-promoter-3′ UTR-ALT or pGL3-promoter-empty vectors (190 ng per well) together with a pRL-CMV vector (10 ng per well) to control for transfection efficiency. Transfection was performed using JetPEI transfection reagent (Polyplus) with the reverse transfection method following manufacturer’s instructions. Briefly, cells were trypsinized, the JetPEI-DNA mix was added to the cell suspension and ~30,000 cells were then plated per well. A JetPEI-DNA ratio of 2:1 was used. At 24 h after transfection both luciferase and Renilla luminescence was measured using the Dual-Glo Luciferase Assay System (Promega, E2920). A minimum of eight wells per conditions were measured.

Seahorse cellular respiration analysis

For cellular respiration analysis myoblasts lines were seeded at 80–90% confluence in growth medium and the following day they were switched to differentiation medium (see above). Myoblasts were differentiated for at least 7 d. At 48 h before the experiment the medium was switched to glucose-free medium (DMEM, no glucose (Gibco, 11966025) supplemented with 10 mM galactose (Sigma, G0750), 1 mM sodium pyruvate (Gibco, 11360070), 2% horse serum (Gibco 16050122) and 1% penicillin/streptomycin (Gibco, 15140122)) or high-glucose medium (DMEM, high-glucose (Gibco, 11965092) supplemented with 1 mM sodium pyruvate (Gibco, 11360070), 2% horse serum (Gibco 16050122) and 1% penicillin/streptomycin (Gibco, 15140122)). OCR was measured using a Seahorse analyzer (XF96, Agilent Technologies), following the manufacturer’s instructions. For the glucose-free condition, XF base medium (102353-100, Agilent Technologies) was supplemented with 10 mM galactose (Sigma, G0750), 1 mM sodium pyruvate (Gibco, 11360070) and 4 mM GlutaMAX (Gibco, 35050061). For the high-glucose condition, XF base medium (102353-100, Agilent Technologies) was supplemented with 25 mM glucose (Sigma, G8270), 1 mM sodium pyruvate (Gibco, 11360070) and 4 mM GlutaMAX (Gibco, 35050061). After measuring basal respiration, 1 μM oligomycin was added to inhibit complex V (ATP-synthase) and measure ATP-linked respiration. Maximal respiration was measured by adding 2 μM FCCP, a strong uncoupling agent. Finally, 1 μM rotenone/antimycin A mix was added to inhibit mitochondrial respiration.

COX7A2L promoter tagging with dCAS9-HA

COX7A2L promoter tagging was performed as described earlier98 with some modifications. Guide RNAs for COX7A2L promoter were designed using the online GPP web portal tool ( using Streptococcuspyogenes PAM sequence (NGG). Two gRNAs having the best predicted on- and off-target scores were selected (gRNA1and gRNA2; see also Supplementary Table 6). Both gRNAs were cloned into phU6 plasmids using BbsI restriction enzyme (Genewiz). Insertion of the gRNA was verified by sequencing (Microsynth). HEK-293T cells cultured in 15-cm dishes were co-transfected with 75 μg of dCas9-HA (Addgene, 61355) and 27 μg of the empty phU6 plasmid (EV) or its subclone expressing the gRNA sequence. Transfection was performed using Lipofectamine 3000 (Invitrogen, L3000001). At 24 h after transfection, Opti-MEM medium (Gibco, 31985070) was replaced with normal culture medium and 24 h later cells were collected for chromatin immunoprecipitation. gRNA1 gave the highest enrichment for COX7A2L promoter and was therefore selected for the experiment in Extended Data Fig. 2c.

Chromatin immunoprecipitation followed by qPCR

Chromatin immunoprecipitation was performed as described39 with some modifications. Cells expressing dCas9-HA and gRNA or EV were washed once with 1× PBS and crosslinked with 1% PFA for 10 min at room temperature. Crosslinking was quenched by adding glycine at a final concentration of 0.125 M. Cells were rinsed twice with 1× PBS and collected. After centrifugation at 1,200 r.p.m. for 6 min at 4 °C, the cell pellet was resuspended in lysis buffer 1 (50 mM HEPES-KOH, pH 7.5, 140 mM NaCl, 1 mM EDTA, pH 8.0, 10% glycerol, 0.5% NP-40, 0.25% Triton X-100 and protease inhibitors (Thermo Fisher Scientific, 78430)) and rotated for 10 min at 4 °C. After centrifugation at 1,200 r.p.m. for 6 min at 4 °C, the cell pellet was resuspended in lysis buffer 2 (10 mM Tris-HCl, pH 8.0, 200 mM NaCl, 1 mM EDTA, pH 8.0, 0.5 mM EGTA, pH 8.0 and protease inhibitors). Following centrifugation at 1,200 r.p.m. for 6 min at 4 °C, the cell pellet was resuspended in lysis buffer 3 (10 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1 mM EDTA, pH 8.0, 0.5 mM EGTA, pH 8.0, 0.1% Na-deoxycholate, 0.5% N-lauroylsarcosine and protease inhibitors). Chromatin was sheared by sonication with diagenode bioruptor UCD-200 (three 5-min cycles of 30 s ON + 30 s OFF on middle strength). Chromatin was centrifuged at 20,000g for 30 min at 4 °C to remove cell debris. Two percent of the supernatant was stored at −80 °C to be used as input. Each sample was divided in two and incubated with 5 μg of either anti-hemagglutinin (ab9110) or IgG control antibody (sc2025). Samples were rotated at 4 °C overnight. For immunoprecipitation, Dynabeads Protein A Immunoprecipitation kit was used (Invitrogen, 10006D). Then, 50 μl of magnetic beads per sample were washed twice with 200 μl of antibody binding and washing buffer and twice with 200 μl of lysis buffer 3. Samples were added to the beads and rotated at 4 °C for 4 h. Beads were then washed three times with 150 mM wash buffer (1% Triton X-100, 0.1% SDS, 150 mM NaCl, 2 mM EDTA, pH 8.0 and 20 mM Tris-HCl, pH 8.0) and twice with 500 mM wash buffer (1% Triton X-100, 0.1% SDS, 500 mM NaCl, 2 mM EDTA, pH 8.0 and 20 mM Tris-HCl, pH 8.0). After the washes, the beads were resuspended in 120 μl of elution buffer (1% SDS and 0.1 M NaHCO3) and incubated for 1 h at 65 °C with shaking. The supernatant was collected in new tubes and both supernatant and input were incubated overnight at 65 °C to reverse crosslinking. Immunoprecipitated DNA was purified using MinElute PCR Purification kit (QIAGEN, 28004) and assayed by qPCR. For qPCR, LightCycler 480 SYBR Green I Master (Roche, 04887352001) and primers were added to 3 μl of the purified DNA. Primer sequences are listed in Supplementary Table 6. Fold enrichment over IgG was calculated based on the percent input recovered in anti-hemagglutinin and IgG samples.

RNA stability assay

The RNA stability assay in differentiated myotubes was performed as described previously99. Briefly, myoblasts were grown in differentiation medium for 10 d, at day 10 of differentiation, actinomycin D (Sigma, A1410) was added to the medium at a concentration of 10 μg ml−1 and cells were collected in TriPure Isolation Reagent (Roche, 11667165001) after 0 and 1 h of treatment. RNA extraction and qRT–PCR were performed as described above. RNA abundance at each time point was calculated relatively to time 0.

Quantification and statistical analysis

No statistical methods were used to predetermine sample size. The exact value of n, the statistical methods used to determine significance and error bars are described in the figure legends. All replicates represent biological replicates. A two-tailed Student’s t-test was used to calculate statistical differences between the means of two groups. One-way or two-way ANOVA tests were used to determine statistical differences between multiple groups. Statistical tests were performed using GraphPad Prism 7 or R. In the Promethion experiment, mice that did not spontaneously run on the running wheel were excluded from the analysis (n = 1–2 per group).

Materials availability

Plasmids generated in this study will be made available upon request to the corresponding author.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.