CoQ imbalance drives reverse electron transport to disrupt liver metabolism

Abstract

Mitochondrial reactive oxygen species (mROS) are central to physiology^1,2. Excess mROS production has been associated with several disease states^2,3; however, the precise sources, regulation and mechanism of generation in vivo remain unclear, which limits translational efforts. Here we show that in obesity, hepatic coenzyme Q (CoQ) synthesis is impaired, which increases the CoQH₂ to CoQ (CoQH₂/CoQ) ratio and drives excessive mROS production through reverse electron transport (RET) from site I_Q in complex I. Using multiple complementary genetic and pharmacological models in vivo, we demonstrate that RET is crucial for metabolic health. In patients with steatosis, the hepatic CoQ biosynthetic program is also suppressed, and the CoQH₂/CoQ ratio positively correlates with disease severity. Our data identify a highly selective mechanism for pathological mROS production in obesity, which can be targeted to protect metabolic homeostasis.

Mitochondria process nutrient-derived substrates to generate metabolites essential for cellular function. This activity inherently produces superoxide and hydrogen peroxide (H₂O₂)—collectively termed mROS—which are crucial for signalling but harmful in excess^3,4. Chronic metabolic stress from obesity⁵ increases mROS production, which contributes to insulin resistance and type 2 diabetes^6–10. Yet, over the past two decades, it has become clear that nonspecific removal of these oxidants through the systemic administration of broad and nonselective antioxidants is not effective in treating metabolic pathologies, as seen by the failure of large-scale clinical trials^11–13. Therefore, a more mechanistic approach directed to the specific source of mROS and underlying mechanisms may present a safer and more effective intervention strategy¹⁴. Identifying these molecular pathways and the role of mROS in metabolic disease in vivo remains a formidable challenge¹.

mROS production is not a single process. In intact cells and in vivo, the rate of mROS generation is the sum of the activity of at least 11 distinct sites associated with the electron transport chain (ETC) and the oxidation of matrix substrates (Fig. 1a). These 11 sites have substrate specificity and show distinct capacities to generate mROS². Notably, complex I is important not only as one of the major producers of superoxide but also because it can operate bidirectionally^15–17, which has been associated with different physiological and pathological outcomes^18–24.

Fig. 1 |. RET at complex I drives excess superoxide and H₂O₂ production in livers from obese mice.

a, Left, general overview of mROS generation. Specific sites and mechanistic detail pinpointing mitochondrial sources are not shown. Right, illustration of the 11 sites that form superoxide, which is subsequently reduced to H₂O₂ by SOD2, shown as red circles and a black star. NAD-linked sites are in the dehydrogenases of branched-chain 2-oxoacids (BCOADH, site B_F), 2-oxoadipate (OADH, site A_F), pyruvate (PDH, site P_F) and oxoglutarate (OGDH, site O_F). CoQ-linked (Q) sites are in complex III (site III_Qo), in the dehydrogenases of succinate (site II_F), glycerol phosphate (GPDH, site G_Q), dihydroorotate (DHODH, D_Q) and in the electron transport flavoprotein ubiquinone oxidoreductase (ETF:QOR, site E_F). Complex I generates superoxide during FET and RET and may have one or two sites that prematurely reduce oxygen^15,16, the flavin site I_F and the Q-binding site I_Q. b, H₂O₂ levels in vivo in mitochondria of livers from lean wild-type (WT) and leptin-deficient obese (ob/ob) mice assessed by MitoB oxidation. n = 8 mice per group (**P = 0.0094, unpaired t-test). c, Superoxide levels in primary hepatocytes from WT (n = 281 cells) and ob/ob (n = 138 cells) mice assessed by MitoSOX oxidation. Two mice per group (****P < 0.0001, unpaired t-test). a.u., arbitrary units. d, Maximum capacity of superoxide and H₂O₂ production from mitochondria isolated from livers of WT and ob/ob mice. n = 13 mitochondrial isolations from n = 13 mice per group (*P < 0.05, ****P < 0.0001, multiple paired t-test not adjusted for multiple comparisons). e, Maximum capacity of superoxide and H₂O₂ production from mitochondrial isolated from livers of WT mice fed a chow diet (CD) or a 60% HFD for 17 weeks. n = 6 mitochondrial isolations from n = 6 mice per group (*P = 0.04, multiple paired t-test not adjusted for multiple comparisons). Sites are described in a. Data are individual values and the mean ± s.e.m. All t-tests were two-tailed.

Classically, complex I oxidizes NADH to transfer the electrons to CoQ during forward electron transport (FET). However, complex I also catalyses the reverse reaction, which is known as RET. During RET, electrons from the highly reduced CoQ pool (CoQH₂) are forced back into complex I. This process is driven by a high protonmotive force (Δp) that generates significantly more superoxide than observed during FET^2,25. High levels of CoQH₂ are also influenced by substrate availability, such as succinate. Although RET was once considered an in vitro artefact because of its unusual thermodynamic requirements, growing evidence shows that RET occurs in vivo and that mROS generated during RET is physiologically relevant^{18–22,26,27}. However, how these individual sites and their activity states relate to metabolic health remains to be determined.

In this study, we systematically explore each of the 11 sites of mROS production (described in Fig. 1a) in the livers of lean and obese mice. We show that excess hepatic mROS in genetic and diet-induced obesity is site-specific and exclusively produced through RET from site I_Q in complex I. Furthermore, this unexpected biochemical alteration is due to ubiquinone imbalance. Using multiple gain-of-function and loss-of-function models in vivo, we demonstrate the importance of RET-induced site I_Q mROS for whole-body metabolism and its significance for human fatty liver diseases.

Hepatic mROS is increased in obese mice

Obesity increases hepatic lipid accumulation and inflammation^28,29 (Extended Data Fig. 1a). Evidence of elevated mROS in obesity comes mainly from ex vivo and in vitro studies^3,4. To investigate hepatic mROS production in vivo, we first used the mitochondrial-targeted ratiometric probe MitoB, which rapidly accumulates in the mitochondria in vivo and is oxidized to mitoP by H₂O₂ and peroxynitrite^30,31. Genetically obese mice showed a higher MitoP to MitoB ratio in their livers than lean control mice, which indicated that the obese mice had increased mitochondrial H₂O₂ levels in vivo (Fig. 1b). Markers of lipid and protein oxidation (4-HNE and PRDX3) were also elevated in the livers of obese mice (Extended Data Fig. 1b–f). Moreover, primary hepatocytes from obese mice showed increased MitoSOX oxidation (Fig. 1c), a result that was consistent with our previous findings³² showing cell-autonomous mROS elevation. Hence, obesity increases liver mitochondrial oxidation.

We next asked whether the obesity-driven increase in mROS stems from a general rise in the maximum capacity of all 11 sites or specific sites. First, mitochondria were isolated from livers of genetically obese and diet-induced obese mice and their lean controls. These samples were tested for their maximum capacity of superoxide and H₂O₂ production by pharmacologically isolating each of the 11 sites and providing their appropriate substrate in the presence of oligomycin (Fig. 1a). Superoxide and H₂O₂ rates were collectively measured as H₂O₂, which could be detected from all 11 sites, with sites I_Q (complex I) and II_F (complex II) showing the highest capacity in lean mice (Fig. 1d,e). Notably, only site I_Q was consistently upregulated in both obesity mouse models; therefore, we focused on this site.

Excess mROS production in obesity is site-specific

Complex I is one of the major sources of mROS and its reaction is bidirectional^15–17, which results in superoxide generation when the enzyme acts in either the FET or RET mode^33,34 (Fig. 1a and Extended Data Fig. 1g). Multiple reports have indicated that more superoxide is generated during RET than during FET^2,33–35. Here the maximum capacity of superoxide production from the flavin site of complex I (site I_F) was measured during FET. To that end, mitochondria isolated from liver were incubated with malate to generate NADH through the tricarboxylic acid cycle, and with rotenone or piericidin A to prevent electrons from moving down the ETC and therefore maintain the flavin in a highly reduced state. Aspartate and ATP were added to suppress the contribution of mROS from site O_F in oxoglutarate dehydrogenase (OGDH)³⁶ (Extended Data Fig. 1g). No difference in H₂O₂ generation from site I_F was detected in liver mitochondria between lean and obese mice (Fig. 1d,e, site I_F). To measure the maximum capacity of mROS formation from site I_Q through RET, mitochondria were incubated with succinate (to increase CoQH₂ levels and the CoQH₂/CoQ ratio) and with or without rotenone, a site-I_Q-specific inhibitor. RET is defined as the difference between these two rates (Extended Data Fig. 1g,h). Notably, superoxide generation from site I_Q through RET was differentially regulated in both genetically obese and diet-induced obese mice, with the capacity increased by >80% and 40%, respectively (Fig. 1d,e, site I_Q). We also measured succinate-driven mROS generation with S1QEL, a noncanonical site I_Q suppressor that has recently been shown to work in vivo in obese mice^37,38. Piericidin A and FCCP were also used, which inhibit RET by blocking site I_Q or collapsing the Δp, respectively (Extended Data Fig. 1g,i,j). Consistently, the rate of superoxide generation from site I_Q was higher in liver mitochondria from obese mice regardless of the compound used to define RET (Extended Data Fig. 1k–m). Notably, the increased superoxide production through RET in obese mice was not observed in mitochondria from the skeletal muscle of these mice (Extended Data Fig. 1n). Altogether, these data show that in mitochondria isolated from the livers of obese mice, excess mROS originates specifically from site I_Q in complex I through RET (Fig. 1d,e) and that dysregulation is site-specific.

CoQ imbalance drives mROS formation through RET

mROS generation from site I_Q through RET depends on the magnitude of the mitochondrial Δp and on the CoQ redox state³⁹. Therefore, we systematically explored the thermodynamic forces underlying increased mROS production from site I_Q through RET in obesity. First, we measured the mitochondrial membrane potential as a proxy for Δp with safranin O, which accumulates in energized mitochondria and results in fluorescence quenching (Extended Data Fig. 2a). Succinate led to mitochondrial polarization to the same degree in mitochondria isolated from wild-type lean and obese mice (Extended Data Fig. 2b). Increased RET in obesity was also independent of changes in the activities of complexes I, II and II/III or a blockade in the ETC (Extended Data Fig. 2c–f). We then evaluated whether obesity affects the protein levels of ETC components. We detected a small increase in some subunits that constitute complex I, in particular the ND6 subunit, which has been implicated in mROS production through RET⁴⁰ (Extended Data Fig. 2g,h). Next, we investigated whether the conditions for increased mROS formation through RET were also present in vivo. To that end, we measured CoQ abundance and the CoQH₂/CoQ ratio directly in in situ freeze-clamped liver tissue samples by liquid chromatography and tandem mass spectrometry (LC–MS/MS).

CoQ₁₀ is the predominant form in humans, whereas mice synthesize both CoQ₉ and CoQ₁₀ (ref. 41); however, their functional differences are not yet well understood^41–43. Nevertheless, they are considered bioenergetically equivalent⁴⁴ and each tissue maintains a specific CoQ₁₀/CoQ₉ ratio⁴¹. Therefore, we evaluated both CoQ₉ and CoQ₁₀ levels and their redox states. CoQ₉ levels (CoQ₉H₂ and CoQ₉) were significantly decreased in livers from obese mice (Fig. 2a). Although CoQ₁₀ levels were not significantly decreased, total CoQ content (CoQ₉ and CoQ₁₀) remained decreased in livers of obese mice (Extended Data Fig. 2i,j). Moreover, we observed significant increases in both the CoQ₉H₂/CoQ₉ and CoQ₁₀H₂/CoQ₁₀ ratios and in the fraction of reduced CoQ₉ and CoQ₁₀ (CoQH₂/total CoQ) in livers from obese mice (Fig. 2b,c and Extended Data Fig. 2k,l). This result suggested that there was a shift towards a more reduced CoQ pool state in obesity. Measurement of reduced CoQ (CoQH₂) levels in isolated mitochondria maybe unreliable; therefore, we analysed total CoQ content in the mitochondrial fraction. The total liver CoQ₁₀ remained unchanged; however, in obese mice, CoQ₁₀ levels in the mitochondrial fraction were decreased tenfold (Extended Data Fig. 2m). Conversely, mitochondrial CoQ₉ levels slightly increased 1.5-fold. These alterations led to a 94% decrease in the mitochondrial CoQ₁₀ to CoQ₉ ratio in obese mice, whereas the whole liver ratio remained unchanged (Extended Data Fig. 2m,n). This result suggests that in obesity, CoQ₉ and CoQ₁₀ trafficking may also be differentially regulated.

Fig. 2 |. The CoQH₂/CoQ ratio is increased in livers of obese mice.

a, CoQ₉ content (CoQ₉H₂ + CoQ) in the livers of WT and ob/ob mice. b,c, The CoQ₉H₂/CoQ₉ ratio (b) and per cent of reduced CoQ₉ (CoQ₉H₂/total CoQ₉) (c) in the livers of WT and ob/ob mice. n = 9 livers per group (*P = 0.016, ***P = 0.0005, ****P < 0.0001, two-way analysis of variance (ANOVA)). d, CoQ chemical structure showing the head and tail precursors. e, Relative levels of phenylalanine (Phe), tyrosine (Tyr) and 4-hydroxibenzoate (4HB) in the livers of WT and ob/ob mice. n = 9 for WT vs n = 11 for ob/ob (**P = 0.011, ***P = 0.0001, unpaired t-test). f, Metabolite levels of the mevalonate (blue) and cholesterol (red) pathways and CoQ in the livers of ob/ob mice were normalized to WT livers. n = 9 mice per group except cholesterol (n = 13), lanosterol (n = 4) and CoQ₉ (n = 18) (*P = 0.22, **P = 0.005, one sample t-test). g, Relative expression of CoQ biosynthetic genes in the livers of WT and ob/ob mice. Expression levels were normalized to WT. n = 16 mice per group (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, one sample t-test). h, Kinetics of ²H-enrichment in the CoQ₉ pool in the livers of WT (n = 14) and ob/ob mice (n = 9) (***P = 0.0005, two-way ANOVA). i, Newly synthesized hepatic CoQ₉ in WT and ob/ob mice after 24 h of ²H₂O administration in the drinking water (4% v/v). n = 6 mice per group (****P < 0.0001, unpaired t-test). j, Mass enrichment in the CoQ₉ isoprenoid tail of the different isotopomers (M1–M4) in the livers of WT and ob/ob mice after 24 h of ²H₂O administration in the drinking water (4% v/v). n = 6 mice per group (***P = 0.0003, two-way ANOVA). k, Mass enrichment in the CoQ₉ head in the livers of WT and ob/ob mice after 24 h of ²H₂O administration in the drinking water (4% v/v). n = 6 mice per group (**P < 0.003, two-way ANOVA). Data are individual values and the mean ± s.e.m. All t-tests were two-tailed.

Metabolomic analyses of in situ freeze-clamped liver tissue samples did not show accumulation of metabolites that could drive RET by feeding electrons into the CoQ pool, including glycerol phosphate, dihydroorotate, acyl-carnitines or succinate. Succinate in particular has been shown to promote RET in other contexts^18,19,24 (Extended Data Fig. 2o–s). These observations rule out substrate accumulation as the cause of elevated CoQH₂ levels and support a model whereby an elevated CoQH₂/CoQ ratio may directly be the main driver of increased RET in obese livers.

Next, we explored the mechanisms underlying the decreased hepatic CoQ levels in obesity by performing metabolomics, gene expression and metabolic flux analyses. CoQ is composed of a redox-active quinone head attached to an isoprenoid tail of various lengths^45,46 (Fig. 2d). The precursor for the head group is phenylalanine, which is converted to tyrosine and then 4-hydrobenzoate⁴⁵, whereas the precursors for the tail are derived from the same mevalonate pathway as cholesterol^45,46. All three metabolites required for biosynthesis of the quinone head were decreased in the livers of obese mice (Fig. 2e). To investigate the potential nodes of regulation, we measured the levels of metabolites in the mevalonate pathway and in downstream branches that lead to cholesterol and CoQ synthesis. Acetyl-CoA, an intermediate of fatty acid and glucose oxidation and a common precursor for de novo lipogenesis and the synthesis of sterols and non-sterols through the mevalonate pathway, was increased in the livers of obese mice (Fig. 2f). Conversely, farnesyl pyrophosphate (FPP), the last common metabolite in the synthesis of all products of the mevalonate pathway and a direct CoQ precursor⁴⁷, was decreased in the livers of obese mice (Fig. 2f, FPP P = 0.06). The gene-expression profile of the mevalonate pathway did not show consistent alterations that would explain the changes in CoQ synthesis (Extended Data Fig. 2t). By contrast, genes directly responsible for CoQ biosynthesis, including Pdss1, which heterodimerizes with Pdss2 to control the lengthening of the CoQ tail, and Coq3, Coq5, Coq8a and Coq9, which are part of the mitochondrial complex Q and are necessary for CoQ head synthesis, were significantly decreased in livers of obese mice (Fig. 2g). Proteomic analysis from mitochondria isolated from livers of wild-type lean and obese mice also revealed small decreases in COQ5 and COQ7 protein levels (Extended Data Fig. 2u–x). Together, these data suggest that in obesity, a decreased CoQ biosynthetic program contributes to decreased hepatic CoQ levels.

To assess the in vivo kinetics of CoQ synthesis and flux through the CoQ synthetic pathway in the liver, we provided lean and obese mice with ²H₂O-supplemented drinking water for up to 48 h and collected livers for analyses. We observed a significantly slower rate of ²H incorporation over time into newly synthesized CoQ₉ and CoQ₁₀ in the livers of obese mice (Fig. 2h and Extended Data Fig. 2y), which provided direct evidence for decreased CoQ synthesis in vivo. Next, we analysed flux through the CoQ synthetic pathway 24 h after ²H₂O administration. The newly synthesized fraction of CoQ was substantially decreased in the livers of obese mice, as measured by ²H enrichment of CoQ₉ and CoQ₁₀ normalized to body water 24 h after ²H₂O administration (Fig. 2i and Extended Data Fig. 2z–aa). At this time point, the incorporation of ²H into the CoQ₉ isoprenoid tail and head and the CoQ₁₀ isoprenoid tail were also significantly decreased in livers from obese mice (Fig. 2j,k and Extended Data Fig. 2ab). Moreover, flux through the cholesterol branch was not altered in livers of obese mice despite increased cholesterol accumulation. This result suggests that the regulatory point in CoQ synthesis occurs downstream of the mevalonate pathway (Extended Data Fig. 2ac–ad). Taken together, these findings indicate that CoQ biosynthesis is suppressed in livers from obese mice and is associated with an increased CoQH₂/CoQ ratio, a pivotal thermodynamic requirement to promote RET and mROS production from site I_Q.

mROS disrupts glucose homeostasis through RET

We next tested whether excess mROS produced specifically through RET contributes to obesity-driven metabolic pathologies. To that end, we first pharmacologically induced complex-I-mediated superoxide formation using mitoparaquat (MitoPQ)^31,48 and then tested whether this was sufficient to impair glucose metabolism in vivo. MitoPQ significantly enhanced H₂O₂ production through RET in a dose-dependent fashion when added to mitochondria isolated from the livers of wild-type mice (Fig. 3a). This effect was RET-specific, as MitoPQ had no effects on H₂O₂ production during FET or on the other mROS-producing sites that feed electrons into the CoQ pool (Fig. 3b and Extended Data Fig. 3a). MitoPQ-driven superoxide production was sensitive to S1QEL only during RET, a result that further supported the conclusion that this process originates from site I_Q through RET^37,38 (Extended Data Fig. 3b–d). Notably, MitoPQ at the concentrations used here did not affect cellular bioenergetics, as treatment of primary mouse hepatocytes with different concentrations of MitoPQ for 60 min did not change basal or maximum oxygen consumption rates (Extended Data Fig. 3e).

Fig. 3 |. mROS generation through RET increases hepatic glucose production and impairs glucose homeostasis.

a, Effect of MitoPQ on mROS production through RET. n = 4 mitochondrial isolations (n = 4 mice) (***P < 0.0003, ****P < 0.0001, one-way ANOVA Dunnett’s post hoc test). b, Effect of MitoPQ on mROS generation through FET and RET. n = 1 mitochondrial isolation. n = 6 (RET) or 3 (FET) replicates. Each dot is an independent measurement. (**P = 0.0015, two-way ANOVA). c, MitoPQ treatment in vivo. d, Glucose tolerance tests (GTTs; 1 g kg⁻¹) in 16-h fasted WT mice treated with 4 nmol MitoPQ or DMSO. Inset, area under the curve (AUC). n = 16 mice per group (*P = 0.045, two-way ANOVA and **P = 0.003, unpaired t-test). e,f, Immunoblot (e) and quantification (f) of proteins in total liver lysates from WT mice 1.5 h after 4 nmol MitoPQ or DMSO treatment. n = 8 mice per group (*P = 0.026, **P < 0.01, unpaired t-test). p, phosphorylated. g, Lactate–pyruvate tolerance tests (1.5 and 0.15 g kg⁻¹, respectively) in 16-h fasted WT mice treated with 4 nmol MitoPQ or DMSO. Inset, AUC. n = 22 mice per group (^##P = 0.0008, two-way ANOVA, **P = 0.006, unpaired t-test). h, Gluconeogenesis assays in primary hepatocytes from WT mice 1.5 h after MitoPQ or DMSO treatment. n = 11 hepatocyte isolations per group (*P = 0.041, paired t-test). i,j, Immunoblot (i) and quantification (j) of hepatic PDSS2 levels in Pdss2^loxP/loxP WT (n = 3), Alb/Cre,Pdss2^WT/loxP heterozygous (Het) (n = 2) and Alb/Cre,Pdss2^loxP/loxP knockout (KO) (n = 4) normalized by tubulin (****P < 0.0001, one-way ANOVA, Dunnett’s post hoc test). k, Hepatic CoQ₉ + CoQ₁₀ content in Pdss2 WT (n = 3), Het (n = 2) and KO (n = 4) mice. (****P < 0.0001, one-way ANOVA, Dunnett’s post hoc test). l, Site I_Q mROS production through RET in Pdss2 WT (n = 3) and Het (n = 2) mice. m, Lactate–pyruvate tolerance tests (1.5 and 0.15 g kg⁻¹, respectively) in 16-h fasted Pdss2 WT (n = 9) and Het (n = 2) mice. Inset, AUC (**P = 0.010, two-way ANOVA and *P = 0.019, unpaired t-test). Data are individual values and the mean ± s.e.m. All t-tests were two-tailed. NS, not significant.

To assess the in vivo relevance of excess mROS production from site I_Q through RET, we administered MitoPQ intraperitoneally into healthy lean mice (Fig. 3c), which was well tolerated up to a dose of 4 nmol. Acute administration of MitoPQ increased mROS production in vivo, as shown by the elevated levels of the oxidative stress marker PRDX3 (Extended Data Fig. 3f), and impaired glucose tolerance in a dose-dependent manner (Fig. 3d and Extended Data Fig. 3h). MitoPQ-induced mROS through RET also decreased hepatic insulin signalling in vivo, as indicated by the reduced levels of phosphorylated insulin receptor (IR), AKT and GSK3α in the liver (Fig. 3e,f). Moreover, hepatic histological sections did not show overt changes (Extended Data Fig. 3g). The in vivo effects of MitoPQ were liver-specific, as insulin signalling in muscle and white adipose tissue were unchanged, and systemic insulin tolerance remained comparable to controls (Extended Data Fig. 3j–l). Taken together, these data support a model whereby excess mROS generated by site I_Q through RET in vivo is sufficient to cause hepatic insulin resistance, thereby recapitulating the obese phenotype.

Gluconeogenesis uses a high proportion of substrates that are routed through the mitochondria, such as pyruvate, lactate and glutamine⁴⁹. Other substrates, such as glycerol, provide gluconeogenic precursors relatively proximal in the pathway to the final product of glucose and downstream of mitochondrial function. Excess mROS production through RET induced a marked increase in glucose excursion during lactate–pyruvate tolerance tests after MitoPQ administration (Fig. 3g and Extended Data Fig. 3i), which indicated a substantial effect on lactate-mediated gluconeogenesis. By contrast, glycerol-mediated gluconeogenesis was not stimulated by MitoPQ (Extended Data Fig. 3m). This result further confirmed that the effect of mROS production through RET on hepatic glucose production strictly depends on mitochondrial metabolism. Furthermore, these effects were cell-autonomous, as primary hepatocytes isolated from mice treated with MitoPQ, but not vehicle, retained the capacity to produce more glucose when given substrates routed through mitochondria, such as lactate, pyruvate and glutamine, but not glycerol (Fig. 3h and Extended Data Fig. 3n).

Having established that pharmacological induction of mROS through RET with MitoPQ impairs glucose homeostasis, we next investigated whether impaired CoQ biosynthesis per se could drive excess mROS generation through RET, which in turn stimulates hepatic glucose production. To that end, we used a mouse model with conditional Pdss2 deletion⁵⁰. PDSS2 is a prenyltransferase that heterodimerizes with PDSS1 to define the number of isoprenoid units in the CoQ tail⁴⁵. We measured hepatic PDSS2 protein levels by western blotting in three different genotypes: Pdss2^loxP/loxP (wild-type); Alb/Cre,Pdss2^WT/loxP (heterozygous); and Alb/Cre,Pdss2^loxP/loxP (knockout) (Fig. 3i,j). Total CoQ levels were 17-fold lower in the knockout mice compared with the wild-type mice (Fig. 3k, WT vs KO), far exceeding the difference observed between lean and obese mice (Fig. 2a). In heterozygous mice, in which Pdss2 is partially deleted, CoQ levels were decreased to the same range as that observed in obese mice (Fig. 3i,k, Het). Therefore, we compared the rate of hepatic mitochondrial H₂O₂ generation through RET between Pdss2 wild-type and heterozygous mice fed standard chow. Decreased CoQ synthesis in this setting significantly increased RET and hepatic glucose excursion during lactate–pyruvate tolerance tests (Fig. 3l,m). These effects were not associated with changes in body weight or liver histological phenotype (Extended Data Fig. 3o,p). Taken together, we conclude that pharmacological or genetic interventions that increase mROS through RET impair glucose homeostasis in lean mice.

Targeting RET improves metabolism in obesity

Next, we used multiple genetic loss-of-function models to test whether specifically targeting excess hepatic mROS production through RET would have a positive metabolic outcome in obesity (Fig. 4a). In our first approach, we transfected two liver cell lines, Hepa 1–6 and AML-12, with Ciona intestinalis alternative oxidase (Aox). Aox is a cyanide-insensitive oxidase that provides an alternative route to oxidize excess CoQH₂ to decrease the CoQH₂/CoQ ratio and prevent mROS production through RET^39,51,52 (Extended Data Fig. 4a). Consistently, Aox expression conferred cyanide-resistant oxygen consumption in these cells (Extended Data Fig. 4b,c). Next, we isolated primary hepatocytes from obese mice and treated them with adenoviral (Ad) particles either expressing Aox or GFP. In these experiments, total H₂O₂ release was measured in the presence or absence of 5 µM S1QEL to assess the specific contribution of site I_Q through RET without blocking oxidative phosphorylation. H₂O₂ production from Ad-GFP-expressing hepatocytes was significantly more sensitive to S1QEL than that of Ad-Aox-expressing hepatocytes (Extended Data Fig. 4d). Therefore, the S1QEL-sensitive rate, referred to as mROS through RET (Fig. 4b), was significantly reduced in Aox-expressing hepatocytes from obese mice. This result was consistent with the high CoQH₂/CoQ ratio observed in obesity that promotes excess mROS generation through RET. Furthermore, Aox expression was sufficient to improve insulin sensitivity in hepatocytes from obese mice, as demonstrated by the enhanced phosphorylation of IR, AKT and GSK3α (Extended Data Fig. 4e,f).

Fig. 4 |. Suppressing RET in vivo improves metabolism in obese mice.

a, Summary of loss-of-function obese mouse models used to suppress RET in vivo and improve glucose homeostasis. b, H₂O₂ release assay in primary hepatocytes expressing GFP or Aox from obese mice. n = 7 hepatocyte isolations per group (*P = 0.028, paired t-test). c, Immunoblot of anti-HA in tissue lysates from ob/ob mice expressing Aox or GFP (n = 2 mice per group). WAT, white adipose tissue. d,e, Immunofluorescence (d) and quantification (e) of colocalization of Aox (anti-HA) and MitoTracker Deep Red in primary hepatocytes from ob/ob mice expressing Aox or GFP. Scale bar, 50 µm. Box plots show median, interquartile range (IQR) and 1.5× the IQR. n = 6 fields per group (*P = 0.015). f, Six-hour fasting blood glucose levels in ob/ob mice expressing Aox or GFP. n = 8 mice per group (*P = 0.030). g, GTTs in ob/ob mice expressing Aox (n = 32) or GFP (n = 31). Inset, AUC (*P = 0.014, two-way ANOVA, **P = 0.0065, one-tailed unpaired t-test). h, Site I_Q mROS production through RET in isolated mitochondria from WT (n = 4) and Nd6^P25L (n = 4) mice. (*P = 0.025, paired t-test). i, Six-hour fasting blood glucose levels in WT (n = 13) and Nd6^P25L (n = 11) mice fed a HFD for 15 weeks. (*P = 0.0289). j, GTTs in WT (n = 13) and Nd6^P25L (n = 11) mice fed a HFD for 10 weeks. Inset, AUC (^#P = 0.0111, two-way ANOVA, *P = 0.0117). k, CoQ₁₀ content in ob/ob mice. n = 6 livers per group (****P < 0.0001). l, CoQ₁₀H₂/CoQ₁₀ ratio in ob/ob mice. n = 6 livers per group (**P = 0.002). m, Site I_Q mROS production through RET in isolated mitochondria from ob/ob mice treated with vehicle (n = 8) or CoQ₁₀ (n = 8). (*P = 0.011, paired t-test). n, GTTs in ob/ob mice treated with vehicle or CoQ₁₀. Inset, AUC. n = 15 mice per group (^##P = 0.0052, two-way ANOVA, **P = 0.0013). For the GTTs, 0.5 g glucose per kg was administered for all mice. Values are individual values and the mean ± s.e.m. All comparisons were unpaired two-tailed t-tests unless otherwise specified.

Next, we tested whether preventing mROS production through RET in vivo could improve metabolic outcomes in obese mice. We used adeno-associated virus (AAV) to achieve hepatocyte-specific expression of Aox (Fig. 4c and Extended Data Fig. 4g). Immunofluorescence and cellular fractionation demonstrated that Aox was specifically targeted to the mitochondria (Fig. 4d,e and Extended Data Fig. 4h). In agreement with the role of RET in modulating glucose homeostasis (Fig. 3), hepatocytes isolated from obese mice expressing Aox displayed significantly lower glucose production in response to gluconeogenic substrates compared with mice expressing GFP (Extended Data Fig. 4i). Consistent with these improvements at the cellular level, Aox-expressing mice exhibited significantly lower fasting blood glucose levels, improved intraperitoneal and oral glucose tolerance and decreased liver glycogen content compared with GFP-expressing control mice (Fig. 4f,g and Extended Data Fig. 4j–l). Plasma insulin levels and systemic insulin tolerance (Extended Data Fig. 4m,n) were not changed, which indicated that the systemic benefits of hepatic Aox expression were due to improved liver metabolism. Notably, these metabolic improvements were independent of any changes in liver lipid levels, weight gain, body composition, energy expenditure, respiratory exchange ratio (RER) or liver metabolites in Aox-expressing animals compared with GFP-expressing control mice (Extended Data Fig. 4o–u).

As an additional genetic model to test the role of RET-mediated mROS production in promoting metabolic dysfunction during obesity, we used a mouse that is incapable of generating mROS through RET^40,53. The mitochondrial DNA mutation G13997A causes a proline to leucine substitution in position 25 of the Nd6 gene (Nd6^P25L), and mice with this mutation cannot generate mROS through RET^40,53 (Fig. 4h). Notably, protein levels of the ND6 subunit were increased in the livers of obese mice (Extended Data Fig. 2h); therefore, this model gives us an additional opportunity to explore the effect of genetically preventing RET on glucose homeostasis in animals with diet-induced obesity. In lean control mice fed a chow diet, the Nd6^P25L mutation had no effect on fasting blood glucose levels, systemic glucose tolerance or body weight (Extended Data Fig. 5a–c). However, when placed on a high-fat diet (HFD) for 10–15 weeks, Nd6^P25L mice exhibited lower fasting blood glucose levels and were significantly more glucose tolerant than wild-type controls fed a HFD (Fig. 4i,j). Histological analyses also revealed a mild improvement in hepatic lipid accumulation in Nd6^P25L mice on a HFD (Extended Data Fig. 5d). These differences were independent of changes in body weight (Extended Data Fig. 5e). Taken together, our results show that mROS production through RET has a causative role in disrupting glucose homeostasis in obesity in vivo.

CoQ₁₀ is involved in steatosis and RET-produced ROS

The prospect that CoQ deficiency drives mROS through RET and contributes to the pathological consequences of obesity raises the possibility that CoQ supplementation, if properly delivered to liver mitochondria, could prevent or ameliorate some obesity-related pathologies. To investigate this question, we first treated obese mice with a CoQ₁₀ emulsion formulation. CoQ₁₀ treatment (10 mg kg⁻¹, intraperitoneal injection) every other day for 20 days significantly increased CoQ₁₀ levels in the liver, particularly in the mitochondrial fraction, of obese mice, which in turn lowered the CoQ₁₀H₂/CoQ₁₀ ratio (Fig. 4k,l and Extended Data Fig. 5f). Consequently, CoQ₁₀ treatment significantly decreased the rate of H₂O₂ generation through RET (Fig. 4m), which was sufficient to improve whole-body glucose tolerance (Fig. 4n), insulin sensitivity and hepatic lipid accumulation in obese mice (Extended Data Fig. 5g,h). Although CoQ₁₀ supplementation slightly increased mitochondrial CoQ₉ levels, it did not change whole liver CoQ₉ content or the CoQ₉H₂/CoQ₉ ratio (Extended Data Fig. 5f,i,j). These data, together with the results shown in in Extended Data Fig. 2m, suggests that CoQ₁₀ is also an important modulator of mROS production through RET in obesity. Notably, CoQ₁₀ supplementation had no effect on glucose tolerance in wild-type lean mice. CoQ₁₀ supplementation also did not alter body weight in lean or obese mice (Extended Data Fig. 5k–m).

To investigate the relevance of the CoQ synthetic pathway and the CoQH₂/CoQ ratio in humans with obesity and liver disease, we analysed samples from two cohorts of individuals with obesity and with various degrees of metabolic dysfunction-associated fatty liver disease (MAFLD, steatosis grade S0–S3) (Extended Data Tables 1 and 2 and Extended Data Fig. 6). In the first cohort, we measured the expression of eight genes responsible for CoQ synthesis, and the analysis revealed that transcripts from the CoQ synthetic pathway were significantly lower in patients with steatosis than healthy control individuals (Fig. 5a and Extended Data Table 1). Indeed, the expression levels of most of the genes analysed (COQ2, COQ3, COQ6, COQ7 and PDSS2) were consistently decreased across all steatosis stages.

Fig. 5 |. CoQ–RET axis in hepatic steatosis in humans.

a, Volcano plots showing the transcript levels of enzymes from the CoQ biosythetic pathway in liver biopsy samples from patients with different stages of steatosis (S0–S3) compared with healthy control individuals. Thresholds for the log₂[fold change] and −log₁₀[P] were set to 1.3, dotted vertical and horizontal lines, respectively. Grey quadrants, non-significant; light purple and blue, significantly lower in steatosis and significantly higher in steatosis, respectively. b, Correlation between the CoQ₁₀H₂/CoQ₁₀ ratio and the stages of steatosis in liver biopsies from patients (n = 18). c, Comparison of the CoQ₁₀H₂/CoQ₁₀ ratio between patients with low-grade and high-grade steatosis. n = 16 for low-grade and n = 3 for high-grade steatosis. **P = 0.009, unpaired two-tailed t-test. Values are individual values and the mean ± s.e.m.

Next, we measured the CoQ₁₀H₂/CoQ₁₀ ratio in hepatic tissues in a second cohort of individuals with obesity and MAFLD and different grades of steatosis (Extended Data Table 2). To reliably capture the in vivo CoQ₁₀H₂/CoQ₁₀ ratio, tissue biopsies (around 2 mg) were immediately placed in liquid nitrogen after surgery. In these samples, we found a strong positive correlation between the CoQ₁₀H₂/CoQ₁₀ ratio and the steatosis grade (Pearson r = 0.69, P = 0.0013) (Fig. 5b). Moreover, when stratified by steatosis grade, individuals with more severe steatosis (S2–S3) had a significantly higher CoQ₁₀H₂/CoQ₁₀ ratio than those with lower grades (S0–S1) (Fig. 5c). These data suggest that excess mROS production through RET can also occur in the human liver of individuals with obesity, and this may be due to lower CoQ synthesis and an increased CoQ₁₀H₂/CoQ₁₀ ratio. Finally, given that CoQ deficiency may drive metabolic disease, pharmacological interventions that disrupt CoQ₁₀ synthesis may worsen disease progression, whereas interventions that replenish CoQ₁₀ may be promising therapeutics for metabolic disease.

Discussion

Although mROS are essential for numerous cellular processes, their excessive production is widely recognized as an important factor in metabolic dysfunction and chronic disease. However, the precise sources of mROS generation and mechanisms of production are unclear, which has limited translational progress. Here we identified in obese mice a previously unknown mechanism whereby impaired hepatic CoQ biosynthesis elevates the CoQH₂/CoQ ratio to generate the thermodynamic drive for mROS production in complex I through RET. Similarly, in the livers of patients with obesity, the transcription of CoQ biosynthetic genes was decreased, and the CoQ₁₀H₂/CoQ₁₀ ratio strongly correlated with hepatic steatosis severity. We showed that excess mROS production through RET was sufficient to impair metabolism in lean mice, and preventing hepatic RET in obese mice improved glucose and lipid homeostasis (Extended Data Fig. 7). These findings may have translational relevance for several human diseases, offering potential therapeutic targets for dyslipidaemia, insulin resistance, fatty liver, cardiometabolic diseases and diabetes. Our findings also highlight the need to reimagine CoQ supplementation strategies and develop targeted formulations for effective tissue delivery. Moreover, this mechanism may help to explain the increased risk of type 2 diabetes in individuals who use statins^54,55, which is due to CoQ imbalance, and suggests that targeted hepatic replenishment could be beneficial.

The biosynthesis of CoQ remains incompletely understood^41,45,46. Although primarily synthesized in mitochondria, CoQ is distributed to other cellular membranes through trafficking mechanisms that are just beginning to be explored^56,57. Outside mitochondria, reduced CoQ serves mainly as an antioxidant, suppressing lipid peroxidation and ferroptosis through FSP1 (refs. 45,58,59). We found a divergent pattern in CoQ₉ and CoQ₁₀ levels between whole liver lysates and isolated mitochondria. Although both CoQ forms were decreased in liver lysates, only CoQ₁₀ levels were decreased in isolated mitochondria, with CoQ₉ showing an increase. These findings indicate that obesity may differentially affect the intracellular trafficking of CoQ₉ and CoQ₁₀ and reveal a previously unrecognized role for CoQ₁₀ in mROS production through RET. These results also support the notion that different CoQ forms may have distinct biological functions^42,43.

The absence of studies comparing CoQ₉ and CoQ₁₀ levels in whole liver and mitochondrial fractions in obesity has meant that the changes observed here are previously unrecognized (Supplementary Table 2). Moreover, the inability to reliably measure reduced CoQ levels in isolated mitochondria that reflect the in vivo levels, limits our ability to firmly conclude that a decrease in mitochondrial CoQ₉ levels drive an increase in the CoQ₉H₂/CoQ ratio and mROS through RET. Nevertheless, our results suggest that CoQ₁₀ has a previously unrecognized role in mROS generation through RET. Future studies should explore these possibilities.

In conclusion, our finding that excess hepatic mROS generation in obesity is site-specific may open new avenues for therapeutic strategies aimed at reducing RET, increasing CoQ levels or both.

Methods

General animal care, study design and animal models

Mouse husbandry and experiments were performed in compliance with and approved by the Harvard Medical Area Standing Committee on Animals under the protocol #IS00000800–6. Mice were bred and housed at 22 °C, 40–60% humidity for up to 20 weeks of age on a 12-h light–dark cycle, with lights on at 7:00 in the Harvard T. H. Chan School of Public Health pathogen-free barrier facility. Mice had free access to water and standard laboratory chow diet (PicoLab Mouse Diet 5058, LabDiet), unless otherwise indicated. All animals used in this study were male, and no specific power analysis was used to estimate sample sizes. Mice were randomly allocated to experimental groups. In vivo studies were not blinded, and we used sex-matched and age-matched mice for all comparisons. Experimental and control samples were processed together using the same conditions.

Animal models

Leptin-deficient B6.Cg-Lep^ob/J (ob/ob) mice on a C57BL/6J genetic background were used as the mouse models of obesity, and age-matched and sex-matched ob/+ heterozygous were used as controls (ob/ob and ob/+, Jackson Laboratories, 000632). Animals were given free access to water and food (PicoLab Mouse Diet 5058, LabDiet). For the diet-induced obesity model, C57BL/6J mice (Jackson Laboratories, 000664) were placed on a HFD (D12492i 60% kcal from fat, Research Diets) for 18 weeks starting at 6 weeks of age. Control mice were placed on a low-fat chow diet (PicoLab Mouse Diet 5053, LabDiet). The mouse strain Nd6^P25L, which carries the single point mutation in the mitochondrially encoded Nd6 gene, was a gift from D. Wallace, Children’s Hospital of Philadelphia, University of Pennsylvania. As female mice with the mutation were backcrossed over several generations onto C57BL/6J background, we used wild-type mice from the same background as controls. To generate the cohort used in this study, female mice with the Nd6 mutation and wild-type C57BL/6J females were crossed with wild-type C57BL/6J male mice to generate the cohort used in this study. Wild-type and Nd6^P25L offspring were placed on a control chow diet or a HFD for 15 weeks starting at 6–7 weeks of age.

Pdss2^loxP/loxP mice were a gift from A. Jurisicova, University of Toronto. These mice were originally generated by D. Gasser, University of Pennsylvania⁵⁰. Male Pdss2^loxP/loxP mice were crossed with albumin-Cre females (Jackson Laboratories, 003574) to delete Pdss2 from hepatocytes. Pdss2^WT/loxP Alb/cre^WT/tg females were then crossed with Pdss2^loxP/loxP males. The females and males from this generation were crossed to generate the three genotypes presented in this study: Pdss2^loxP/loxP (control), Alb/cre,Pdss2^WT/loxP (heterozygous) and Alb/cre,Pdss2^loxP/loxP (knockout).

Isolation of mitochondria

Mouse livers were minced with a razor blade in a glass Petri dish containing ice-cold STE buffer (250 mM sucrose, 5 mM Tris-HCl and 2 mM EGTA, pH 7.4 at 4 °C). Liver pieces were rinsed in ice-cold STE and homogenized in a glass homogenizer with a loose-fit glass pestle using six to nine strokes. The homogenate was filtered through a double layer of gauze, and about 15 ml STE buffer was added to the filtrate, which was then centrifuged at 1,000g for 3 min. The supernatant was centrifuged at 10,000g for 10 min to pellet the mitochondria and the pellet was washed twice in STE medium. The final crude mitochondrial pellet was resuspended in 1 ml STE buffer and underwent a further purification step by Percoll density centrifugation using a discontinuous density gradient of 2 ml 80%, 4.71 ml 52% and 4.71 ml 26% Percoll in 2 ml 4× STE buffer to a final Percoll density gradient of 1.117, 1.08 and 1.047 g ml⁻¹, respectively, at 15,500 r.p.m. for 45 min at 4 °C, using a 40Ti swing bucket in an XL-90 Beckman centrifuge. The mitochondrial fraction was collected between the 52 and 26% interfaces and washed once in STE buffer at 10,000g for 10 min to obtain purified mitochondria. Mitochondria were resuspended in 1 ml STE medium and immediately assayed.

Mitochondria from mouse skeletal muscle were isolated as previously described⁶⁰. In brief, skeletal muscle was dissected from the entire hind limbs of wild-type and obese mice and placed in 4 °C Chappell–Perry buffer (CP1, 0.1 M KCl, 50 mM Tris, 2 mM EGTA, pH 7.4) and CP2 (CP1 supplemented with 0.5 % w/v fatty-acid-free BSA (FAF-BSA), 2 mM MgCl₂, 1 mM ATP and 250 U 0.1 ml⁻¹ subtilisin protease type VIII, pH 7.4) and mitochondria were isolated as previously described⁶¹, resuspended in CP1 medium and immediately assayed. Protein concentrations were determined using the bicinchoninic acid assay.

Superoxide and H₂O₂ measurements

Rates of superoxide and H₂O₂ production were collectively measured as rates of H₂O₂ production, as endogenous and exogenous superoxide dismutase (SOD) convert two superoxide molecules to one H₂O₂ molecule. H₂O₂ generation in purified liver mitochondria (0.1 mg ml⁻¹) was measured during state 4 (oligomycin 2.5 µg ml⁻¹) in KHE assay buffer (120 mM KCl, 5 mM HEPES and 1 mM EGTA) supplemented with 2 mM MgCl₂, 10 mM KH₂PO₄ and 0.3% w/v FAF-BSA, pH 7.1 at 37 °C. The detection system consisted of 0.2 U ml⁻¹ horseradish peroxidase (HRP), 50 µM Amplex UltraRed (Molecular Probes), 100 µM PMSF and 24 U ml⁻¹ SOD. H₂O₂ generated during substrate oxidation will either react with the detection system or be neutralized by the redox-buffering networks. If redox-buffering capacity was different between mitochondria from lean and obese mice, this would result in a different maximum capacity for superoxide and H₂O₂ generation from all the sites, which was not the case. Therefore, we assumed that mitochondria from lean and obese mice have the same matrix antioxidant capacity. In mitochondria isolated from liver, the addition of PMSF was required to inhibit the non-specific conversion of Amplex UltraRed to resorufin⁶². Mitochondria isolated from skeletal muscle (0.3 mg ml⁻¹) were resuspended in KHE medium supplemented with 1 mM MgCl₂, 5 mM KH₂PO₄ and 0.3% w/v FAF-BSA, pH 7.1 at 37 °C in the presence of 2.5 µg ml⁻¹ oligomycin, but PMSF was omitted from the detection. Fluorescence was monitored using a microplate reader (SpectraMax Paradigm, Molecular Devices) at λ_excitation = 605 nm, λ_emission = 640 nm after wavelength optimization and calibrated using H₂O₂ standards after each run under the exact conditions of the assay. The substrates and inhibitors used to define each site are detailed in Supplementary Table 3.

Mitoparaquat effects on RET and FET

Liver mitochondria from wild-type mice were isolated, and the rate of H₂O₂ production through RET or FET was measured in the presence of the respective substrates (Supplementary Table 3) and different concentrations of MitoPQ in a 96-well plate. Rotenone was used to block site I_Q, which prevents MitoPQ-induced H₂O₂ production through RET. Where indicated, 10 µM S1QEL 2.2 was added together with MitoPQ to prevent MitoPQ-induced H₂O₂ production through RET. MitoPQ, at any concentration tested, was not able to increase the rate of H₂O₂ production when substrates to measure site I_F (FET) were used. The effect of MitoPQ on sites II_F, G_Q, D_Q and E_F were measured according to Supplementary Table 3 in the presence of 1 µM MitoPQ. The buffers, detection system and plate reader settings were identical as described above.

Measurement of mitochondrial membrane potential with Safranine O, and complex I and complex II activities

The mitochondrial membrane potential was estimated in purified liver mitochondria (0.1 mg ml⁻¹) in the presence of 2.5 µg ml⁻¹ oligomycin during succinate oxidation in KHE assay buffer supplemented with 2 mM MgCl₂, 10 mM KH₂PO₄ and 0.3% w/v FAF-BSA, pH 7.1 at 37 °C. The detection system consisted of 2 µM Safranine O and fluorescence was monitored at λ_excitation = 495 nm, λ_emission = 587 nm. Mitochondria membrane polarization was measured as the changes in fluorescence before and after succinate addition, when the signal stabilized between 250 and 400 min. For assessment of complex I activity, freeze–thawed mitochondria (2.5 mg ml⁻¹) were incubated in 250 mM sucrose, 0.2 mM EDTA, 50 mM Tris-HCl (pH 7.0) and 5 µM cytochrome c. NADH oxidation (150 µM) was monitored at 340 nm. Complex I activity was expressed as the rotenone-sensitive rates of NADH oxidation. Complex II activity was measured using a MitoCheck Complex II Activity Assay kit (Cayman). Complex I–III activity was measured by incubating 50 µg freeze–thawed mitochondria in 100 mM phosphate buffer (pH 7.4) in the presence of 50 µM cytochrome c and 2 µM rotenone, with or without 1 mM KCN. Succinate (5 mM) was added immediately before the measurements. Cytochrome c reduction (ε = 19 mM⁻¹ cm^{− 1}) was monitored at 550 nm, and activity was expressed as KCN-sensitive rates. All measurements were performed using a microplate reader (SpectraMax Paradigm, Molecular Devices).

In vivo H₂O₂ levels with MitoB

The levels of H₂O₂ in vivo in liver mitochondria were determined according to a previous study³⁰. Lean and obese mice (7 weeks old) were injected with 0.8 µmol kg⁻¹ MitoB (Cayman Chemical) through the tail vein. Six hours after injection, livers were collected and immediately frozen. MitoB and MitoP were extracted from 100 mg tissue, and labelled internal standards d-₁₅ MitoB and d-₁₅ MitoP were spiked into each sample³⁰. The samples were analysed using multiple reaction monitoring on an Agilent 6460 LC–MS/MS triple quad mass spectrometer. LC conditions were slightly modified from the published method we followed³⁰. Five microlitres of each sample was injected into a phenyl-hexyl column (Dikma Platisil, 5 µm, 150 × 4.6 mm). Mobile phases were A (water with 0.1% formic acid) and B (acetonitrile with 0.1% formic acid). The gradient was 5% B for 2 min, to 25% B in 1 min, to 60% B in 3.5 min and then to 100% B in 3 min. After 6.5 min at 100% B, the column was re-equilibrated at 5% B for 3 min. The flow rate was maintained at 0.3 ml min⁻¹. MitoP and MitoB concentrations were determined using calibration curves from pure standards prepared similarly to the samples, using the d-₁₅-labelled compounds as internal standards.

CoQ isolation and CoQH₂/CoQ ratio measurement

CoQ was extracted from the liver of two cohorts of wild-type and obese mice after cervical dislocation. Livers were extracted under freeze-clamp conditions using a Wollenberger clamp to ensure that the CoQH₂/CoQ ratio accurately reflected the in vivo state. In the first cohort, whole livers were kept at −80 °C for about 3 weeks before CoQ extraction. In the second cohort, livers were collected, and CoQ was extracted immediately. The results obtained from these cohorts were pooled for analysis. Once CoQ was isolated, it was maintained at −80 °C for less than 24 h before analysis. It has been demonstrated that there is no substantial CoQ oxidation during this period⁴⁴. Approximately 50 mg frozen tissue was homogenized in 250 µl ice-cold acidified methanol (0.1% HCl) and 250 µl water-washed hexanes in a bullet blender (TissueLyser II, Qiagen) with 5 mm stainless steel beads (Qiagen) at 30 Hz for 3 min at 4 °C. The samples were spiked with 200 pmol CoQ₈ as an internal standard and centrifuged at 17,000g for 5 min. The upper layer containing CoQ was dried under argon and then resuspended in methanol containing 2 mM ammonium formate and analysed within 24 h. The same protocol was used to extract CoQ from human liver biopsy samples; however, the average starting material was 2 mg. To extract CoQ from the mitochondrial fractions, 50 µg protein was incubated in 200 µl ice-cold acidified methanol and 200 µl water-washed hexanes followed by vortexing. Samples were spiked with CoQ₈ as internal standard according to the procedures described above. The reduced CoQ standards were prepared according to published methodology⁴⁴. Samples were analysed on an Agilent 6460 Triple Quad mass spectrometer coupled to an Agilent 1290 LC (Agilent Technologies). Five microlitres of sample was injected into an Eclipse Plus C18 column (Agilent, 2.1 × 50 mm). Mobile phases were A (methanol and 5 mM ammonium formate) and B (isopropanol). The gradient was as follows: 9 min at 20% B, then to 100% B in 0.1 min. After 4 min at 100% B, the column was re-equilibrated to 20% B for 3 min. The flow rate was 0.3 ml min⁻¹ for the first 9 min and then switched to 0.4 ml min⁻¹. Compounds were quantified using the signal of the high-resolution mass of the [M + NH₄]+ adducts. Quantification was done using standard curves and CoQ₈ as the internal standard. Multiple reaction monitoring was used to detect CoQs in positive-ion mode. Transitions used were as follows: CoQ₈ (IS) (744.8 > 197.1), CoQ₉ (795.6 > 197), CoQ₉H₂ (814.7 > 197), CoQ₁₀ (863.7 > 197) and CoQ₁₀H₂ (882.7 > 197.1).

²H₂O treatment

Wild-type and obese (ob/ob) mice were given with deuterated water (²H₂O) for 48 h. Initially, mice were injected with a bolus of saline solution 0.9% (w/v) prepared in ²H₂O at a dose 27 µl g^–1 body weight. Mice were then supplemented with ²H₂O (4% v/v) in their drinking water. Liver tissue samples were collected at 4, 6, 12, 24 and 48 h and prepared for CoQ₉ extraction and measurement of H²-enrichement in the CoQ tail and head.

CoQ enrichment measurement

The method for CoQ enrichment measurement by LC–MS/MS is detailed in the Supplementary Methods. Data are provided in Supplementary Tables 4 and 5.

Measurement of cholesterol synthesis by Orbitrap MS

Cholesterol synthesis was measured based on the incorporation of ²H into cholesterol (after natural abundance correction) following the administration of D₂O to mice. Approximately 20 mg liver was weighed and homogenized in 1 ml methanol and dichloromethane (DCM) (1:1, v/v) in a 2 ml pre-filled Bead Ruptor tube. The tube was washed twice with 1 ml methanol–DCM and all solutions were combined. Samples were vortexed and centrifuged for 5 min at 1,635g. A known amount of d7-cholesterol was spiked into 2 mg supernatant as an internal control, and the sample was dried under N₂. Dried extracts were saponified with 1 ml 0.5 M KOH in methanol at 80 °C for 1 h. Lipids were extracted with DCM–water and then evaporated to dryness. The dried lipid extract was then derivatized with 100 µl acetyl chloride for 1 h at 75 °C. The sample was dried under N₂ and redissolved in 100 µl iso-octane.

The ²H‐enrichment of cholesterol (m0, m1, m2 and m3 isotopologues of deuterated cholesterol) was determined using a Q Exactive GC-Orbitrap mass spectrometer (Thermo Scientific). One microlitre of sample was injected into a HP-5ms capillary column (60 m × 0.32 mm i.d., 0.25 µm film thickness) in split mode. Helium gas flow rate was initiated at 1 ml min^–1 for 14 min, followed by 0.4 ml min^–1 for 5 min and then back to 1 ml min^–1. The gas chromatography injector temperature was set at 250 °C and the transfer line was held at 290 °C. The column temperature program was 1 min at 200 °C, then increased by 20 °C min^–1 to 320 °C for 16 min. Samples were analysed at 70 eV in electron ionization mode using the targeted selected ion-monitoring method at 240,000 mass resolution (FWHM, m/z 200). Tuning and calibration of the mass spectrometer was performed using perfluorotributylamine to achieve a mass accuracy of <0.5 ppm. The quadrupole was set to pass ions between m/z 246.24 and 252.24. The Orbitrap automatic gain control target was set to 5 × 10⁴ with a maximum injection time of 54 ms. Cholesterol concentrations were calculated from the ratios of the areas of the peaks corresponding to cholesterol (m/z 247.242) and to D7-cholesterol internal standard (m/z 254.286) with full scan mass ranges of 240–260 m/z. Extraction of individual high-resolution m/z values representing each isotopomer ion was done using TraceFinder 4.1 (Thermo Scientific) with 4 ppm mass tolerance. The contribution of cholesterol synthesis was determined using the following equation, where n represents the number of exchangeable hydrogens, which was assumed to be 20: Newly made cholesterol (%) = cholesterol enrichment × (water enrichment × n)^–1 × 100.

Mitochondrial peptide labelling with Tandem Mass Tag 6-plex and proteomic analysis

Samples were prepared according to instructions of a TMT Mass Tagging kit (Thermo Scientific, 90064). Three 6-plex reactions were set up in parallel, with each containing mitochondria isolated from the livers of three lean mice and three obese mice. Nine independent isolated mitochondria from lean and nine from obese were sampled. All solutions are reported as final concentrations. In brief, 100 µg mitochondrial protein were solubilized in EasyPep lysis buffer (Thermo Scientific, A45735). Protein reduction, alkylation and precipitation reactions were performed according to the protocol of the kit. After 4 h at −20 °C to allow proteins to precipitate, acetone-precipitate pellets were resuspended in 100 µl of 50 mM triethyl ammonium bicarbonate (TEAB) with 2.5 µg trypsin, and digestion proceeded overnight at 37 °C. The next day, TMT reagents were equilibrated to room temperature and solubilized in anhydrous acetonitrile as recommended by the kit. Samples were labelled as follows: pure mitochondria from lean mice (TMT 126, 128 and 130) and pure mitochondria from obese mice (TMT 127, 129 and 131) and then combined at equal amounts. Digested and labelled peptides were cleaned-up with C18 spin tips (Thermo Scientific, 87784) before LC–MS/MS analysis.

Samples were analysed by microcapillary reversed-phase LC–MS/MS using a high-resolution Orbitrap Exploris 480 (Thermo Fisher Scientific) mass spectrometer in positive-ion DDA mode (top 12) through higher energy collisional dissociation (HCD) coupled to a Thermo EASY-nLC 1200 nano-UHPLC. A 75 µm i.d. × 10 cm microcapillary column packed 3 µm C18 beads and a 100 µm i.d. × 3 cm trapping column packed with 3 µm C18 beads (ESI Source Solutions) were used for LC–MS/MS separation. Buffer A consisted of 0.1% formic acid and buffer B consisted of 100% acetonitrile. The flow rate was held constant at 315 nl min^–1. The resolution was 60 K in MS and MS2 modes in the mass range 385 m/z to 1,500 m/z in MS, DDA-triggered MS2 mass range. The external calibration was performed with Thermo Flex Mix calibration solution. MS/MS data were searched against the UniProt Human protein database (containing 82,485 entries) using Mascot 2.7 (Matrix Science). Data were analysed using Scaffold Q+S 5.0 protein identification software (Proteome Software). Peptides and modified peptides were accepted if they passed a 1% false-discovery rate threshold.

LC/MS metabolomics

For small polar metabolite separation and data acquisition in extracted tissues, a Vanquish Horizon (Thermo Fisher Scientific) ultra-high LC system coupled to a Thermo Q Exactive HF Orbitrap mass spectrometer was used. For separation, a Waters XBridge BEH Amide (2.5 µm, 2.1 × 150 mm) column fitted with a VanGuard (2.5 µm, 2.1 × 5 mm) guard column was used. The mobile phases were as follows: phase A (95% water and 5% acetonitrile) and phase B (20% water and 80% acetonitrile) with 10 mM ammonium acetate and 10 mM ammonium hydroxide in both phases. The flow rate was held constant at 0.3 ml min^–1 and the following gradient conditions were used: 0 min, 100% B; 3 min, 100% B; 3.2 min, 90% B; 6.2 min, 90% B; 6.5 min, 80% B; 10.5 min, 80% B; 10.7 min, 70% B; 13.5 min, 70% B; 13.7 min, 45% B; 16 min, 45% B, 16.5 min, 100% B; and 22 min, 100% B. The samples were kept at 4 °C, the injection volume was 5 µl and the column was maintained at 25 °C. The separated metabolites were analysed in both positive and negative ionization modes in the same run (switching mode). The mass spectra were acquired using a resolution of 120,000 in the 70–1,000 m/z range. The ElectroSpray Ionization source parameters for both modes were as follows: capillary temperature 300 °C, spray voltage 3.5 kV, sheath gas 40 (a.u.), auxiliary gas 10 (a.u.), probe heater temperature 30 °C and S-Lens RF level 45 v. Metabolite annotation and documentation are available in Supplementary Table 6. Raw files are accessible from Metabolomics Workbench repository⁶³ with the project ID ST003846 (https://doi.org/10.21228/M8WJ9W).

Cell culture

Hepa 1–6 cells (American Type Culture Collection (ATCC), CRL-1830), a mouse liver hepatoma cell line, were cultured in DMEM with 10% CCS. AML12 cells (ATCC, CRL-2254), a mouse normal hepatocyte cell line, were cultured in DMEM–F12 medium with 10% FBS, 1× STI and 40 ng ml⁻¹ dexamethasone. All cells were cultured at 37 °C in a humidified incubator with 10% CO₂ and tested negative for mycoplasma contamination. Cells were validated based on their morphology (examined by microscopy) and their ability to respond to insulin. Cell lines tested negative for mycoplasma contamination and no commonly misidentified cell lines were used in this study.

Primary hepatocyte isolation

Mice were anaesthetized with 200 mg kg⁻¹ ketamine and 20 mg kg⁻¹ xylazine and livers were then perfused with 50 ml of buffer I (11 mM glucose, 200 µM EGTA, 1.17 mM MgSO₄ heptahydrate, 1.19 mM KH₂PO₄, 118 mM NaCl, 4.7 mM KCl and 25 mM NaHCO₃, pH 7.32) through the portal vein until the liver turned pale, using a peristaltic pump set to an infusion rate of about 4 ml min⁻¹. The rate was then gradually increased to around 7 ml min⁻¹. After about 5 min, when all of buffer I had been infused, 50 ml buffer II (11 mM glucose, 2.55 mM CaCl₂, 1.17 mM MgSO₄ heptahydrate, 1.19 mM KH₂PO₄, 118 mM NaCl, 4.7 mM KCl, 25 mM NaHCO₃, 7.2 mg ml⁻¹ FAF-BSA and 0.18 mg ml⁻¹ type IV collagenase (Worthington Biochemical, LS004188)) was infused. BSA and collagenase were added to buffer II immediately before infusion. The buffers were kept at about 37 °C during the entire perfusion process. Next, the liver was excised, and primary hepatocytes were carefully released and sedimented at 500 r.p.m. for 2 min, washed twice in DMEM with 10% CCS and suspended in Williams E medium supplemented with 5% CCS and 1 mM glutamine (Invitrogen). To separate live from dead cells, the hepatocytes were layered on a 30% Percoll gradient and centrifuged at about 1,500 r.p.m. for 15 min. The healthy cells were recovered at the bottom of the tube and plated in Williams E medium and kept overnight until experimentation.

Oxygen consumption

Hepatocyte oxygen consumption rates (OCRs) were monitored using a XF-24 and XF-96 extracellular flux analyzer (Seahorse Bioscience; Agilent). Primary hepatocytes were seeded on collagen-coated 24-well Seahorse plates at 4 × 10⁴ cells per well and kept overnight in normal growth medium. The next day, cells were rinsed twice and kept in XF Base medium (Seahorse Bioscience; Agilent) supplemented with 0.2% FAF-BSA, 10 mM glucose and 2 mM pyruvate, pH 7.4, for 1–2 h before the run. MitoPQ was added in port A at the indicated concentrations followed by 1 µM FCCP in port B and 1 µM rotenone–antimycin A in port C. Hepa 1–6 cells were seeded at 3 × 10⁴ cells per well and kept in normal growth medium until adhering. Next, DMEM without glucose (Sigma) supplemented with 20 mM galactose, 4 mM pyruvate, 2 mM glutamine and 10% CCS was added, and cells were kept overnight. The following day, cells were rinsed twice and kept in XF Base medium supplemented with 0.3% FAF-BSA with 20 mM galactose, 4 mM pyruvate, 2 mM glutamine, pH 7.4, for 1–2 h before the run. S1QEL 2.2 (ChemDiv) was added at the indicated concentrations in port A. For all experiments, basal OCR was monitored for 15 min and MitoPQ and S1QEL 2.2 effects were plotted as the per cent change of baseline in vehicle-injected wells. OCR was measured in freshly isolated liver crude mitochondria (before Percoll purification). In brief, 2.5 µg mitochondrial protein in 20 µl ice-cold KHE medium supplemented with 0.3% FAF-BSA, 1 mM MgCl₂ and KH₂PO₄ (respiration buffer) were seeded in a XF-24 plate and centrifuged at 2,000g for 15 min. Then, 480 µl respiration buffer supplemented with 2 mM ADP at 37 °C was added to each well. Baseline rates were monitored for 20 min followed by 5 mM succinate and 1 µM rotenone or 5 mM malate or 5 mM pyruvate injections from port A. The OCR during state 3 was monitored for 20 min, and 1.5 µM oligomycin was added to port B to induce mitochondrial state 4o. The OCR of AML12 and Hepa 1–6 cells expressing Aox were measured using a Clark-type electrode fitted in a water-jacketed chamber (Strathkelvin Instruments). Cells were permeabilized with 0.02% digitonin in KHE buffer supplemented with 2 mM MgCl₂, 10 mM KH₂PO₄ and 0.3 % w/v FAF-BSA, pH 7.1 at 37 °C. Recording started in the presence of 5 mM succinate, 1 µM rotenone and 1 mM ADP. At the indicated time points, potassium cyanide (KCN, 100 µM) and n-propylgallate (50 µM) were injected into the chambers to inhibit complex IV and Aox, respectively.

Overexpression of Aox

The C. intestinalis Aox (accession XP_018672179.1) gene sequence was codon-optimized for mammalian expression (IDT Codon Optimization Tool). The codon-optimized Aox with a carboxy-terminal HA tag synthesized by IDT was cloned into pLv-EF1a-IRES-puro (Addgene, 85132), pAAV.TBG.PI.eGFP.WPRE.bGH (Addgene, 105535) and pAdv-MCS-IRES-GFP (Vector Biolabs), respectively, for the production of viral particles. The AAV was produced by the Penn Vector Core (University of Pennsylvania). The adenovirus was produced by Vector Biolabs. AML12 and Hepa 1–6 cells were infected with lentiviral particles collected from the medium of HEK293T cells transiently transfected with pLv-EF1a-Aox-HA-IRES-puro, psPAX2 and pMD2.G. Cells with Aox-HA stable expression were selected by treatment with puromycin (2 µg ml^–1) for 2 weeks.

CoQ₁₀ emulsion preparation

Emulsions were prepared with CoQ₁₀ owing to reagent availability. A total of 100 mg CoQ₁₀ (oxidized form, Sigma-Aldrich) and corn oil (2:1, w/w) were dissolved in 1 ml chloroform and methanol (2:1, v/v). Next, 150 µl sodium oleate (2 mg ml⁻¹) and 500 µl phosphatidylcholine (Avanti Polar lipids) (25 mg ml⁻¹, in chloroform) were mixed together. The mixture was dried under argon for 30 min at room temperature and desiccated under vacuum overnight at 4 °C according to a published protocol⁶⁴. The next day, the viscous mixture was warmed at 60 °C for 5 min and mixed with pre-warmed 0.9 ml 2% glycerol in PBS with 0.25 mM EDTA, pH 8.4. The solution was then sonicated, 1-s pause/on, for 10 min at 50 °C. The sonicated emulsion was centrifuged for 30 s to remove impurities. The CoQ₁₀ emulsion was stored in a glass vial at room temperature. The CoQ₁₀ concentration was determined using the molar extinction coefficient ε₂₇₅ = 14.6 mM⁻¹ cm⁻¹ (CoQ₁₀).

In vivo treatments

Lean male mice aged 9–12 weeks (23–34 g) were acutely treated with MitoPQ (Cayman Biosciences) through intraperitoneal (i.p.) injection 1–2 h before experimentation. MitoPQ was titrated from 1–6 nmol per mouse, which is equivalent to 0.04–0.2 mg kg⁻¹ body weight. Experiments were performed with 0.16 mg kg⁻¹ (4 nmol per mouse) to avoid MitoPQ toxicity. For the experiments conducted with AAV-Aox or AAV-GFP, 5-week-old ob/ob mice received a dose of 1.5 × 10¹¹ viral particles by retroorbital injection while under anaesthesia (1–3% isoflurane). Glucose and insulin tolerance tests were performed at 2 and 4 weeks and 3 weeks after AAV administration, respectively. Liver tissue samples were collected 5 weeks after AAV administration. For chronic CoQ₁₀ administration, ob/ob mice received an i.p. injection of CoQ or control emulsion (10 mg kg⁻¹) every other day starting at 5 weeks of age for a total of 4 weeks. Glucose and insulin tolerance tests were performed 2 and 4 weeks, respectively, after the start of the treatments.

Glucose, lactate–pyruvate, insulin and glycerol tolerance tests

For all tolerance tests, animals were fasted for 16 h overnight, except for insulin tolerance tests, for which animals were daytime fasted for 6 h. For GTTs, lean mice acutely treated with MitoPQ or DMSO vehicle received an i.p. injection of 1 g kg⁻¹ glucose. Obese mice expressing AAV-Aox or AAV-GFP or chronically treated with CoQ₁₀ or a control emulsion, received an i.p. injection of 0.5 g kg⁻¹ glucose. For the insulin tolerance tests, lean and obese mice received an i.p. injection of 0.7 U kg⁻¹ or 1.5 U kg⁻¹ insulin, respectively (Humulin R U-100, Lilly, prepared in PBS containing 0.2% FAF-BSA). For the lactate–pyruvate tolerance tests, lean mice acutely treated with MitoPQ received an i.p. injection of 1.5 mg kg⁻¹ of a solution containing 10:1 sodium lactate to sodium pyruvate dissolved in PBS at 500 mg ml⁻¹. For glycerol tolerance tests, lean mice acutely treated with MitoPQ received an i.p. injection of 1 g kg⁻¹ glycerol. All substrates were injected in a volume of around 250 µl diluted in PBS. Blood glucose and lactate levels were monitored before MitoPQ injection (time –60 min) immediately before substrate injection (time 0 min) and for up to 2 h following injection using a glucometer (Bayer Contour Next EZ) and a lactate meter (Nova Biomedical Lactate Plus) from a superficial nick at the tip of the tail.

In vivo insulin signalling

Mice were fasted for 16 h and then received 4 nmol MitoPQ or DMSO vehicle by i.p. injection. Two hours later, mice were anaesthetized with 100 mg kg⁻¹ ketamine and 10 mg kg⁻¹ xylazine and then injected with 0.75 U kg⁻¹ insulin (HumulinR, Lilly), prepared as described above, into the portal vein. Livers were collected 3 min later and immediately frozen in liquid N₂. Epididymal adipose tissue and gastrocnemius were next collected and frozen until protein extraction. Obese mice expressing AAV-Aox or AAV-GFP received 3.5 U kg⁻¹ insulin.

Histology

Tissues were fixed in 10% zinc formalin overnight and then transferred to 70% ethanol for prolonged storage. Tissue processing, sectioning and staining with haematoxylin and eosin and periodic acid-Schiff (PAS) for glycogen detection were performed by Histowiz. Positive PAS staining intensity was identified using Halo image analysis software from Indica Labs, and areas were quantified using the algorithm (v.2.1.3). The algorithm was set-up to measure two stains by first defining the settings to identify the magenta PAS stain and the blue haematoxylin counterstain. The algorithm uses a colour deconvolution step to separate the two stain colours, and this is then followed by setting thresholds for the PAS stain to detect weak, moderate and strong (Halo threshold settings 0.3876, 0.6199, 0.9518, respectively) staining. Data were expressed as the per cent of total area positive for PAS staining.

MitoSOX oxidation

Oxidation of MitoSOX (Invitrogen) was used as indicator of superoxide levels in primary hepatocytes. MitoSOX oxidation was measured after 30 min of incubation with 1 µM MitoSOX in DMEM without phenol red and supplemented with 2 mM pyruvate and 0.2% FAF-BSA, according to the manufacturer’s specifications.

Glucose production

Primary hepatocytes were isolated and maintained in Williams E medium supplemented with 0.1 % CCS overnight. The next day, cells were washed in warmed PBS and incubated in DMEM without phenol red and glucose, with 0.2 % FAF-BSA in the presence of the gluconeogenic substrates 2 mM pyruvate, 2 mM glutamine and 20 mM lactate, or 20 mM glycerol. After 4 h, medium was collected for measurement of glucose levels, and the cells were washed and collected for protein assay. Glucose was measured using 5 U ml⁻¹ glucose oxidase, 50 µM Amplex Ultra Red and 10 U ml⁻¹ HRP after 30 min of incubation. Fluorescence was monitored using a microplate reader (SpectraMax Paradigm, Molecular Devices) at λ_excitation = 530 nm, λ_emission = 590 nm. Fluorescence was calibrated using glucose as standard.

H₂O₂ release and insulin signalling from primary hepatocytes

Primary hepatocytes from obese mice were isolated as described above. Cells were seeded (2 × 10⁴) in a 96 well-plate. At 4 h after plating, they were washed and incubated overnight in Williams E medium while exposed to adenoviral particles expressing Aox or GFP at a multiplicity of infection of 100. For the H₂O₂ production assay, cells were washed in assay buffer (120 mM NaCl, 3.5 mM KCl, 1.8 mM CaCl₂, 0.4 mM KH₂PO₄, 20 mM TES, 5 mM NaHCO₃, 1.2 mM Na₂SO₄ and 1 mM MgCl₂) supplemented with 0.1% FAF-BSA. The detection system consisted of 0.2 U ml⁻¹ HRP, 50 µM Amplex UltraRed, 100 µM PMSF and 24 U ml⁻¹ SOD, and pyruvate and glutamine were used as substrates. S1QEL 2.2 (2.5–5 µM) was used to determine the contribution of site I_Q, and H₂O₂ generated through RET was calculated by subtracting the rate of H₂O₂ generation with substrate alone from the rate in the presence of S1QEL. Fluorescence was monitored using a microplate reader (SpectraMax Paradigm, Molecular Devices) and was calibrated using H₂O₂ as standard. For insulin signalling, primary hepatocytes (the day after isolation) were incubated in serum-free Williams E medium for 4 h. Next, 3–18 nM insulin was added to the medium for 3 min, after which, cells were washed twice with 1× PBS and immediately frozen in liquid N₂.

Protein extraction and immunoblotting

Liver, epidydimal fat and skeletal muscle were homogenized with a TissueLyser II in NP-40 buffer (50 mM Tris-HCl (pH 7.4), 2 mM EGTA, 5 mM EDTA, 30 mM NaF, 10 mM Na₃VO₄, 10 mM Na₄P₂O₇, 40 mM glycerophosphate, 1% NP-40 and 1% protease inhibitor cocktail) supplemented with 10 nM okadaic acid. Homogenates were incubated for 10 min on ice and then centrifuged at 14,000g for 10 min. Supernatant was collected and protein concentrations were determined using the bicinchoninic acid method with albumin as a standard. Samples were diluted in 6× Laemmli loading buffer containing β-mercaptoethanol and heated at 95 °C for 5 min. Protein was separated by 4–12% NU-PAGE gradient gels using 1× MOPS buffer (Invitrogen) or by 4–20% Criterion TGX-stain free gels (Bio-Rad) and using 7.5% SDS–PAGE using 1× Tris/glycine–SDS buffer (Bio-Rad). Blots were incubated with primary antibody overnight, washed 3 times and then incubated with HRP-conjugated secondary antibody (Cell Signaling Technologies) for 1 h at room temperature. Membranes were visualized by chemiluminescence (Roche Diagnostics) using a Bio-Rad ChemiDoc MP imaging system. For the non-reducing gels, samples were prepared as described above with the exception that the lysis buffer was supplemented with 100 mM N-ethylmaleimide and β-mercaptoethanol was not added. Band intensities were calculated using Fiji (ImageJ), and the intensities of the bands corresponding to the proteins of interest were normalized by the intensities of a housekeeping protein used as the internal loading control, for example, VDAC or tubulin. When analysing phospho-protein, we used the non-phosphorylated protein as the normalizer (for example, phosphorylated AKT/total AKT). For the analysis of 4-HNE levels between wild-type and ob/ob mice, the entire lane intensity of each replicate was normalized to the respective lane intensity of the Ponceau-stained membrane; this approach was used to find the 4-HNE levels in wild-type and ob/ob samples. The area used to calculate the bands intensities are marked by a red dotted box in Supplementary Fig. 1.

Endogenous protein staining and confocal imaging

Primary hepatocytes were seeded on 35 mm round glass-bottom imaging dishes in Williams medium with 5% CCS and maintained overnight at 37 °C, 5% CO₂. The following morning, cells were washed and incubated in DMEM supplemented with 2 mM pyruvate and 2 mM glutamine and 50 nM MitoTracker Deep Red (Invitrogen) for 45 min and then fixed with 4% paraformaldehyde for 10 min at room temperature. The cells were then washed 3 times in PBS and permeabilized for 20 min with 0.2% Triton-X100 in PBS at room temperature. Anti-HA primary antibody (3724, Cell Signaling Technologies) diluted in 1:200 in PBS was used to visualize AAV-Aox-HA. The cells were incubated in primary antibody solution overnight at 4 °C and the next day, were washed 3 times with PBS, including one long wash for more than 10 min. Cells were then incubated with 1:1,000 diluted secondary antibody in PBS for 1 h at room temperature in the dark. The cells were washed 3 times, including one long wash. Cells were imaged with a Yokogawa CSU-W1 spinning disk confocal system, with a fixed 50 µm diameter pinhole (Andor Technology) and a Nikon Ti-E inverted microscope (Nikon Instruments). Imaging was performed with either a ×60 or ×100 Plan Apo objective lens (NA 1.4 for both), using a Zyla cMOS camera and NIS Elements software was used for acquisition parameters, shutters, filter positions and focus control. The Laser was DPSS –561 nm, 100 milliwatts, and images were acquired using a constant laser power for both channels (34% for excitation wavelength 561 nm, and 22% for excitation wavelength 640 nm). Image analysis was performed using Fiji software. At least 30 cells from each genotype were analysed.

Primary and secondary antibodies

Primary antibodies used were anti-β-tubulin (ab21058), anti-NDUFS1 (ab169540), anti-VDAC (ab14734), OXPHOS cocktail (MS604; ab110413), anti-peroxiredoxin 3 (PRDX3; ab73349) and anti-4 hydroxynonenal (4HNE; ab46545) from Abcam. Anti-ND6 (A32848) was from Boster Biological Technology. Anti-phospho-IR (pTyr972) (I1783–1VL) was from Sigma. Anti-NDUFV2 (SC-271620), anti-IRβ (SC-711), anti-NDUFA6 (SC-86755) and anti-PDSS2 (SC-515137) were from Santa Cruz Biotech. Anti-phospho-AKT (pThr308) (cs-9275), anti-phospho-AKT (pSer473) (cs-4060), anti-pan-AKT (cs-4691), anti-phospho-GSK3α/β (cs-9331), anti-HA (cs-2367) and anti-calreticulin (cs-12238) were from Cell Signaling Technologies. Anti-Aox was a gift from E. Dufour, University of Tampere, Finland. Anti-GSK3α/β (44610) was from Thermo Fisher Scientific. Secondary antibodies used were anti-rabbit IgG-HRP (7074), anti-mouse IgG-HRP (7076) and anti-rabbit IgG Alexa Fluor 488 (4412) from Cell Signaling Technologies. Uncropped raw blots are presented in Supplementary Fig. 1. Molecular weight markers were Page ruler (266616) and Magic Marker (LC5602) (Thermo Fisher Scientific).

Gene expression

RNA was isolated from liver tissues from 9–10-week-old wild-type and ob/ob mice. Tissue samples were homogenized in TRIzol (Invitrogen) using a TissueLyser II (Qiagen) at 30 Hz for 3 min at 4 °C. RNA was extracted using Nucleospin RNA kits (740955.250, Macherey-Nagel). cDNA was synthetized using an iScript RT Supermix kit (Bio-Rad). Quantitative PCR was run in triplicate on a ViiA7 RT-PCR system (Applied Biosystems) using SYBR green and custom primers or primer sets based on sequences from the Harvard Primer Bank. The cycle threshold (C_t) values of target genes were normalized to hypoxanthine phosphoribosyltransferase (HPRT1) C_t values to calculate expression levels using the 2^−ΔΔCt method. For the genes in the mevalonate and CoQ biosynthesis pathway, we used the average of HPRT C_t values between four independent experiments to normalize the values of the target genes.

Primer list

Coq2 (L-GACCCAGGTTGTTTTCCAGA; R-TGGAAGGTCGAAATG TCTCC), Coq3 (L-TCGTGGCTTCTGAAGTTGTG; R-CCCACGTAT GAGTGCCTTTT), Coq5 (L-CCACGGTCTGTGACATCAAC; R-GCCTGGT CAATGTGTGTGAC), Coq7 (L-TTTGGACCATAGCTGCATTG; R-ATGCG GTTTGCTCCATATTC), Coq8a (L-GAAGTCTGGGCTGCAGTAGG; R-GAA GCCTGCCTTTTTGTCTG), Coq9 (L-CCTGTCAAAATCCCCTGAGA; R-GCTCAGCACAACTGTCCAAA), Pdss1 (L-CCGCGACTTTCAGACTTGA; R-TTGGGGAGGCAGACATTAAA), Pdss2 (L-TGGTGCATCGTGGGATAGTA; R-TACTCCATGCACCAAGTCCA), HMGCS primer set 1 (L-CTCTGT CTATGGTTCCCTGGCT; R-TCCAATCCTCTTCCCTGCC), HMGCS primer set 2 (L-TGATCCCCTTTGGTGGCTGA; R-AGGGCAACGATTC CCACATC), HMGCR1 (L-CCGGCAACAACAAGATCTGTG; R-ATGTA CAGGATGGCGATGCA), HMGCR2 (L-ATCCTGACGATAACGCGGTG; R-AAGAGGCCAGCAATACCCAG), MVK (L-CTCTGCTTGCCTTTCTC TAC; R-TCGGGAGTGTCCTGAAATA), PMVK (L-CTGTTTAGCGG GAAGAGAAA; R-GGCAGAGCTACATCTTCATAG), MVD primer set 1 (L-AAGCAGACGGGCAGTACAGT; R-CCTGGAGGTGTCATTGAGGT), MVD primer set 2 (L-CTGCACCAGGACCAGCTAAA; R-CTGAGGCT GAGGGGTAGAGT), IPP isomerase (L-GACGTCAGGCTTGTGCTAGA; R-CTAGAACACAGAGATTCCGGCT), FPP synthase (L-TCCAGGTC CAGGACGACTAC; R-CGCCTCATACAGTGCTTTCA); GGPP synthases (L-GACAAGCTACAGATTATCATTGAAGTG; R-ATCCGGGTGATCAA GGGTTA) and HPRT1 (L-CCAGCGTCGTGATTAGCG; R-CCAGCAGGTCAG CAAAGAAC).

Body composition and the comprehensive lab animal monitoring system (CLAMS)

Body composition of 7-week-old ob/ob mice 14 days after AAV-Aox or AAV-GFP injection was measured using magnetic resonance imaging (EchoMRI). For CLAMS, 7-week-old ob/ob mice expressing AAV-Aox or AAV-GFP for 12 days were housed individually and acclimatized for 1 day. Oxygen consumption, carbon dioxide release, energy expenditure and activity were measured using a Columbus Instruments Oxymax-CLAMS system (Flow Max 210) as previously described, according to guidelines for measuring energy metabolism in mice⁶⁰.

Human participants

Patients who were overweight or diagnosed as obese and with MAFLD from the Liver Unit of Hospital Clínic in Barcelona were studied. The study was approved by the Hospital Clinic IRB (HCB/2019/0458). All patients gave written informed consent, and two specimens of liver biopsy were collected: one for histopathological diagnosis and one for microarray analysis (n = 14) or CoQ determination (n = 13). Approximately 2 mg liver biopsy samples for CoQ measurements were immediately frozen in liquid N₂ after collection. Clinical, demographic and laboratory data were collected in all participants at the time of liver biopsy, and nonalcoholic fatty liver disease was diagnosed according to clinical guidelines^65,66. All liver biopsy samples were reviewed by the pathologist of the Hospital Clinic of Barcelona, and steatosis and fibrosis grade were defined according to the NASH CRN criteria⁶⁷. Samples used for the microarray and CoQ assays were from healthy individuals who were liver donors from the Liver Unit Transplant Program in the Hospital Clinic, from which liver biopsies were performed at the time of living-donor liver transplantation. Characteristics of participants at the time of inclusion are provided in Extended Data Tables 1 and 2.

Patient sample mRNA isolation and gene-expression analysis

mRNA for the transcriptome analysis was isolated from fresh human liver tissues using TRIzol, according to the manufacturer’s instructions (Invitrogen). RNA samples were hybridized to GeneChips and Affymetrix Human Genome U219 arrays. All microarray data used in the current study have been deposited and are accessible from the public repository of NCBI Gene Expression Omnibus (GEO) (GSE139602 and GSE224645). In article, we only analysed the subset of genes involved in CoQ biosynthesis.

Statistics and reproducibility

Statistical significance was assessed using GraphPad Prism (v.9 and v.10.4.1) and Microsoft Excel (v.16.66.1). All data are individual values and mean ± s.e.m. n indicates independent values per group unless otherwise indicated in the figure. Student’s t-test was used to calculate the significance between two groups. For multiple comparison tests, one-way ANOVA or two-way ANOVA followed by Tukey’s, Sidak’s or Dunnett’s post hoc test correction was applied when appropriate. P < 0.05 was considered significant. The AUC was calculated using the trapezoid rule, as implemented in GraphPad Prism, using zero as the baseline. The exact P values and the specific test used in each panel are provided Supplementary Table 1.

Extended Data

Extended Data Fig. 1 |. ROS generation by RET from site I_Q is increased in the liver but not the skeletal muscle of obese mice.

(A) Liver sections from wildtype (wt) and ob/ob mice stained with H&E. Obese hepatocytes contain several lipid vacuoles (arrow) and small foci of inflammation (asterisk). Scale bar, 200 µm. (B) Immunoblot analysis and (C) quantification of 4-HNE as an oxidative stress marker in liver homogenates from wt and ob/ob mice. n = 6 livers per group (***p = 0.001, unpaired t-test). Area used for quantification is shown in Supplementary Fig. 1. (D) Immunoblot analysis and (E) quantification of peroxiredoxin 3 (PRDX3) in liver homogenates from wt and ob/ob mice. n = 4 per group (**p = 0.002, unpaired t-test). (F) Quantitative proteomics of PRDX3 in liver isolated mitochondria from wt and ob/ob mice. n = 9 mito isolations from n = 9 mice per group (*p = 0.011, unpaired t-test). (G) Schematic to illustrate the substrates and inhibitors used to assess the two modes of mROS generation from complex I: forward electron transport from site I_F (FET, left) and reverse electron transport (RET, right) from site I_Q. (H) Representative Amplex UltraRed traces showing that rotenone blocks mROS during succinate oxidation. The rate difference between minus and plus rotenone defines RET, which is higher in mitochondria isolated from ob/ob livers. Representative of n = 13 independent experiments. (I) Representative Amplex UltraRed traces and (J) quantification showing that 5 µM S1QEL 2.2, 2 µM rotenone, 2 µM piericidin A, and 1 µM FCCP decrease the rate of mROS production during RET induced by succinate oxidation in the presence of oligomycin in isolated mitochondria. n = 5 independent experiments, except piericidin A (n = 4) and S1QEL (n = 3) (****p < 0.0001, two-way ANOVA, Dunnett’s post hoc test) (K) Relative mROS production by RET from wt and ob/ob liver mitochondria using the compounds in (J). n = 3 mito isolations from n = 3 mice per group (*p < 0.05, multiple paired t-test not adjusted for multiple comparisons). (L) S1QEL 2.2 (0.15−10 µM) does not inhibit Hepa 1–6 oxygen consumption rates (OCR). n = 5 independent experiments, except DMSO (n = 8) and S1QEL 0.15 and 0.3 µM (n = 4). ns, p = 0.3268 two-way ANOVA, Dunnett’s post hoc test). (M) Effect of S1QEL 2.2 (0.01−10 µM) on mROS production during RET in liver-isolated mitochondria from wildtype and ob/ob mice. n = 3 mito isolations from n = 3 mice per group (*p < 0.0001, two-way ANOVA). (N) Maximum capacity of superoxide/H₂O₂ production from skeletal muscle-isolated mitochondria from lean wt (n = 12) and ob/ob mice (n = 12). Pooled from four independent experiments (*p = 0.019, multiple unpaired t-test not adjusted for multiple comparisons). Data are individual values and means ± SEM. All t-tests were two-tailed.

Extended Data Fig. 2 |. The thermodynamic forces driving mROS via RET.

(A) Representative traces and (B) quantification of mitochondrial membrane potential from lean wildtype (wt) and obese (ob/ob) livers. n = 5 mito isolations from n = 5 mice per group. AU, arbitrary units. (C-E) Complex I, II and II/III activities in wt and ob/ob liver isolated mitochondria. C, n = 3; D, n = 9; E, n = 4 independent mito isolations per group (ns=p > 0.05, unpaired t-test). (F) Oxygen consumption rate (OCR) of wt and ob/ob liver isolated mitochondria oxidizing FAD- and NAD-linked substrates under phosphorylating (state 3) and non-phosphorylating conditions (state 4). n = 4 mito isolations from n = 4 mice per group (*p = 0.01, **p = 0.007, multiple unpaired t-tests not adjusted for multiple comparisons). (G) Immunoblot (top) and quantification analysis (bottom) of complex II-V of the electron transport chain (ETC) in the liver lysates of wt and ob/ob mice normalized by VDAC, run on a separate gel (bottom of panel H). n = 3 liver lysates per group (*p = 0.029, multiple unpaired t-test not adjusted for multiple comparisons). (H) Immunoblot (left) and quantification analysis (right) of complex I subunits in the livers of wt and ob/ob mice normalized by VDAC. n = 3 liver lysates per group, except ND6 which is n = 10 per group (*p < 0.05, **p = 0.004, ****p < 0.0001, multiple unpaired t-tests not adjusted for multiple comparisons). (I) CoQ₁₀ content (CoQ₁₀H₂ + CoQ₁₀), (J) Total CoQ content (CoQ₉ + CoQ₁₀), (K) Ratio of CoQ₁₀H₂/CoQ₁₀, and (L) % of reduced CoQ₁₀ (CoQ₁₀H₂/total CoQ₁₀) in the livers of wt and ob/ob mice. n = 9 mice per group (*p < 0.05, ***p = 0.0006; ns, p > 0.05, two-way ANOVA). (M) CoQ₉ and CoQ₁₀ content in liver isolated mitochondria from wt (n = 9) and ob/ob mice (n = 10). Each mito isolation represents one mouse (**p = 0.005, ****p < 0.0001, multiple unpaired t-tests not adjusted for multiple comparisons). (N) CoQ₁₀/CoQ₉ ratio. n = 9 mice group [liver] and n = 9 for wt vs n = 10 for ob/ob [mitos] (*p < 0.0001, ns=p > 0.05, multiple unpaired t-tests not adjusted for multiple comparisons). (O) Illustration of the enzymes that can generate mROS and feed electrons into the CoQ pool. (P-S) Quantification of the levels of glycerol phosphate, dihydroorotate, acyl-carnitines and succinate in the livers of wt and ob/ob mice. n = 9 for wt vs n = 11 for ob/ob (**p = 0.0084, unpaired t-test). (T) Relative expression levels of the genes in the mevalonate pathway in the livers of ob/ob mice relative to wt. n = 16 mice (*p < 0.05, **p = 0.0085, ****p < 0.0001, one sample t-test). (U-X) Quantitative proteomics of enzymes in the CoQ synthetic pathway, COQ5, COQ7, COQ8a and COQ9. n = 9 mito isolations from n = 9 mice per group (*p = 0.044, **p = 0.003, unpaired t-test). (Y) Kinetics of ²H-enrichement in the CoQ₁₀ tail in the livers of wt (n = 16) and ob/ob (n = 11) (***p = 0.0006, Two-way ANOVA). (Z) Newly synthesized CoQ₁₀ in the livers of wt and ob/ob mice after 24 h of ²H₂O administration in the drinking water (4% v/v). n = 6 mice per group (****p < 0.0001, unpaired t-test). (AA) Total ²H-water enrichment in wt and ob/ob livers. n = 6 mice per group (****<0.0001, unpaired t-test) (AB) Mass enrichment in the CoQ₁₀ isoprenoid tail of the different isotopomers (M1-M3) in the livers of wt and ob/ob mice after 24 h of ²H₂O administration in the drinking water (4% v/v). n = 6 mice per group ***p = 0.0003, Two-way ANOVA). (AC) ²H-enrichment in cholesterol and (AD) total cholesterol in the liver of wt and ob/ob mice 24 h after ²H₂O administration in the drinking water (4% v/v). n = 6 mice (****<0.0001, unpaired t-test). Data are individual values and means ± SEM. All t-tests were two-tailed. ns, not significant.

Extended Data Fig. 3 |. Mitoparaquat promotes mROS via RET and impairs glucose homeostasis in the liver.

(A) Effect of 1 µM MitoPQ on superoxide/H₂O₂ production from the sites linked to the Q-pool (sites I_Q, II_F, G_Q, D_Q and E_F). n = 4 mito isolations from n = 4 mice (**p = 0.004, multiple paired t-tests not adjusted for multiple comparisons). (B) Effect of 10 µM S1QEL 2.2 or 2 µM rotenone on the rate of MitoPQ-induced superoxide/H₂O₂ production by RET. MitoPQ (n = 6), S1QEL (n = 3) and rotenone (n = 1) mito isolations. Each mito isolation represents one mouse (*p = 0.0125, **p = 0.009, Two-way ANOVA). (C) Effect of 10 µM S1QEL 2.2 on the rate of MitoPQ-induced superoxide/H₂O₂ production by FET. n = 3 mito isolations from n = 3 mice. (D) MitoPQ-stimulated rate of superoxide/H₂O₂ production via RET is suppressed by 2.5 µM S1QEL 2.2. n = 3 mito isolations from n = 3 mice (**p = 0.0044, One-way ANOVA, Dunnett’s post hoc test). (E) Effect of MitoPQ on the oxygen consumption rate (OCR) of wt primary hepatocytes. MitoPQ, port A; 1 µM FCCP, port B; and 2 µM rotenone/antimycin A, port C. n = 2 hepatocytes isolations from n = 2 mice (ns, Two-way ANOVA). (F) Immunoblot analysis and quantification of PRDX3 levels in liver homogenates from DMSO or MitoPQ treated mice for 1.5 h and normalized by ponceau from the same samples on a different blot. n = 9 mice per group (**p = 0.006, unpaired t-test). (G) Liver section from wt mice treated with DMSO or MitoPQ stained with H&E, bars 200 µm. (H) Blood glucose levels during i.p. glucose tolerance test (0.5 g • kg⁻¹) in wt mice treated with 2–4 nmol mitoPQ. Inset is area under the curve. n = 28 mice per group, except MitoPQ 2 nmol (n = 4) (*p = 0.0252, Two-way ANOVA. **p = 0.006, One-way ANOVA, Dunnett’s post hoc test). (I) Blood glucose levels during i.p. lactate: pyruvate tolerance test (1.5:0.15 g• kg⁻¹) in wt mice treated with 1–4 nmol mitoPQ. n = 9 mice per group, except mitoPQ 4 nmol n = 8 (*p = 0.0023, Two-way ANOVA, **p = 0.005, One-way ANOVA Dunnett’s post hoc test). (J) Immunoblot analysis and (K) quantification of in vivo insulin signaling in the gastrocnemius muscle (left) and epididymal fat (right) of wt mice 1.5 h after 4 nmol MitoPQ or DMSO treatment. n = 8 mice per group (*p = 0.047, **p = 0.007 and ns, unpaired t-test). (L) Blood glucose levels during insulin tolerance test (0.7 U insulin • kg⁻¹) in 6 h fasted mice treated with 4 nmol MitoPQ or DMSO. Inset: area under the curve. DMSO (n = 15) and mitoPQ (n = 14) mice (ns, Two-way ANOVA, inset: ns, unpaired t-test). (M) Blood glucose levels during i.p. glycerol tolerance test (1 g • kg⁻¹) in 16 h fasted mice treated with 4 nmol mitoPQ or DMSO. Inset is area under the curve. DMSO (n = 10) and MitoPQ (n = 13) mice (ns, Two-way ANOVA, inset: ns, unpaired t-test). (N) Gluconeogenesis assay in primary hepatocytes from wt mice 1.5 h after MitoPQ or DMSO treatment. 20 mM glycerol as substrate. n = 11 hepatocyte isolations from n = 11 mice per group (ns, unpaired t-test). (O) Bodyweights from pdss2 wt, het and kd mice fed regular chow. pdss2 fl/fl-wt (n = 11), het (n = 3) and KO (n = 6) mice. (P) Liver section from pdss2 wt and het mice stained with H&E, bars 200 µm. Data are individual values and means ± SEM. All t-tests were two-tailed. ns, not significant.

Extended Data Fig. 4 |. Ectopic expression of Aox in hepatocytes decreases mROS generation via RET and improves systemic glucose homeostasis.

(A) Illustration showing mROS production in ob/ob mice ± Aox expression. Aox oxidizes excess CoQH₂ and decreases mROS by RET. (B) Representative traces of cyanide-insensitive oxygen consumption in Hepa 1–6 and (C) AML12 cells expressing Aox. N-propyl gallate (n-PG) inhibits Aox activity. Top, immunoblot confirming Aox expression. Representative data from n = 3 experiments. (D) Representative traces of Amplex UltraRed oxidation to show that Aox-expressing primary hepatocytes generate less mROS by RET. Ad., adenovirus. Representative data from n = 7 experiments. (E) Immunoblot analysis and (F) quantification of 18 nM insulin action in isolated hepatocytes from 9–10-week-old obese mice incubated with ad.Aox-HA (n = 15) or ad.GFP (n = 15) for 24 h. Pooled from five independent experiments (*p = 0.0298, **p = 0.0092, ****p < 0.0001, one-tailed unpaired t-test). (G) Representative immunoblot analysis of tissue homogenates of Aox expressing mice (n = 1 mouse). (H) Immunoblot analysis of different cellular fractions from the liver of GFP or Aox mice. ndufs1 and VDAC, mitochondria; calreticulin, endoplasmic reticulum and tubulin, cytosol (n = 1 mouse per group). (I) Glucose production from primary hepatocytes isolated from obese mice expressing Aox or GFP using 20 mM lactate, 2 mM pyruvate, and 2 mM glutamine as substrates. n = 9 mice per group (*p = 0.0178, one-tailed unpaired t-test). (J) Blood glucose levels during oral glucose tolerance test (OGTT) (0.75 g • kg⁻¹) in ob/ob mice. expressing aav.GFP (n = 7) or aav.Aox (n = 8). Inset, area under the curve. (*p = 0.0496, one-tailed unpaired t-test). (K) Liver sections from ob/ob mice expressing Aox or GFP stained with PAS, bars 200 µm. (L) Quantification of liver areas positive for PAS staining of glycogen in obese mice expressing aav. Aox (n = 7) or aav.GFP (n = 5). (*p = 0.0159, unpaired t-test). (M) Plasma insulin levels during OGTT (GFP, n = 7 vs Aox, n = 8 mice). (N) Blood glucose levels during insulin tolerance test (3.5 U of insulin • kg⁻¹). Inset, area under the curve (GFP, n = 7 vs Aox, n = 8 mice). (O) Liver sections from ob/ob mice expressing Aox or GFP stained with H&E, bars 200 µm. (P-U) Metabolic profile of obese 10 days after Aox expression. (P) Bodyweights following aav.GFP or aav.Aox administration at 6 weeks of age (n = 8 per group). (Q) Body composition of 7-week-old ob/ob mice following AAV administration (n = 6 per group). (R) Energy expenditure (EE) as a function of bodyweight (Aox, n = 9 vs GFP, n = 7). (S) Respiratory exchange ratio measured during metabolic cage housing (Aox, n = 9 vs GFP, n = 7). (T) Analysis of RER during light and dark cycles (Aox, n = 9 vs GFP, n = 7). (U) Metabolite levels in the livers of ob/ob mice expressing Aox vs. GFP (n = 6 mice per group). Panels D and I were created with BioRender.com. Values are individual values and means ± SEM. ns, not significant.

Extended Data Fig. 5 |. Suppressing RET in lean mice does not change bodyweight or glucose tolerance.

(A) Six hour fasting blood glucose levels in wildtype (wt) and ND6^P25L mice fed chow diet for 15 weeks (n = 10 mice per group). (B) Blood glucose levels during glucose tolerance test (0.5 g • kg⁻¹) in ND6^P25L (n = 10) and wt (n = 11) mice on chow diet for 10 weeks. Inset, area under the curve. (C) Weight gain of wt and ND6^P25L mice over 15 weeks on chow (n = 10 mice per group). (D) Liver section from wt and ND6^P25L on HFD stained with H&E, bars 200 µm. (E) Weight gain of wt (n = 13) and ND6^P25L (n = 11) mice over 15 weeks on HFD. (F) CoQ₉ and CoQ₁₀ content in isolated mitochondria from ob/ob mice treated with vehicle or CoQ₁₀ emulsion. n = 5 mito isolations from n = 5 mice per group (****p < 0.0001, multi unpaired t-test not adjusted for multiple comparisons). (G) Blood glucose levels during insulin tolerance test (1.5 U of insulin • kg⁻¹) in 6 h fasted ob/ob mice treated with 10 mg • kg⁻¹ CoQ₁₀ (n = 8) or vehicle (n = 7) every other day for 23 days. Inset, area under the curve (****p < 0.0001Two-way ANOVA and *p = 0.019, two-tailed unpaired t-test). (H) Liver sections from ob/ob mice treated for 20 days with CoQ₁₀ or vehicle stained with H&E, bars 200 µm. (I) CoQ₉ content and (J) CoQ₉H₂/CoQ₉ ratio in the liver from leptin-deficient ob/ob mice treated with vehicle (n = 6) or CoQ10 emulsion (n = 6). (K) Blood glucose levels during i.p. glucose tolerance test (0.5 g • kg⁻¹) in lean wt mice treated with 10 mg • kg⁻¹ CoQ₁₀ or vehicle every other day for 23 days. Inset, area under the curve (n = 4 mice per group). (L) Bodyweights of lean wildtype mice treated with CoQ₁₀ (n = 3) or vehicle (n = 4). (M) Bodyweights of ob/ob mice treated with CoQ₁₀ or vehicle (n = 12 per group). Values are individual values and means ± SEM. ns, not significant.

Extended Data Fig. 6 |. Histological changes in patients with hepatic steatosis.

Liver section from patients with MAFLD and different grades of steatosis stained with H&E. S0, no steatosis grade 0 (less than 5% hepatocyte occupied by fat); S1, grade 1, mild (5–33% hepatocyte occupied by fat); S2, grade 2, moderate (34–66% hepatocyte occupied by fat); S3, grade 3, severe (above 66% hepatocyte occupied by fat). Histological images are representative of the steatosis scores observed in the livers of the 27 patients included in this study. Left panel bars, 500 µm, and right panel bars, 100 µm.

Extended Data Fig. 7 |. Model Overview of Coenzyme Q (CoQ) Synthesis Imbalance in Obese Livers.

Illustration of the model detailing the impact of CoQ synthesis deficiency on the CoQH₂/CoQ ratio, resulting in increased mitochondrial reactive oxygen species (mROS) production via reverse electron transport (RET) and subsequent impairment of glucose homeostasis (black boxes). Different genetic and pharmacological interventions were utilized to modulate specific nodes within the model, ranging from CoQ synthesis and levels to the direct generation of mROS via RET. Approaches highlighted in blue indicate interventions that improved glucose homeostasis, while those in red denote interventions that worsen it. To investigate whether obesity-driven CoQ deficiency contributed to the observed alterations in CoQH₂/CoQ ratio, obese mice were supplemented with CoQ₁₀. This treatment restored CoQ₁₀ levels, lowered CoQH₂/CoQ ratio, and mitigated mROS production via RET, thereby improving glucose homeostasis. Similarly, manipulating CoQ redox state through ectopic expression of Ciona intestinalis alternative oxidase (Aox) in the livers of obese mice resulted in decreased mROS via RET and improved glucose homeostasis. The point mutation in the ND6 gene (ND6^P25L) was found to directly hinder complex I-mediated mROS generation via RET, these mice had enhanced glucose homeostasis in high-fat diet. Conversely, impairing CoQ biosynthesis in lean mice (via pdss2 knockdown) or directly inducing mROS via RET with mitoPQ led to compromised glucose homeostasis. This figure provides a comprehensive overview of the intricate interplay between hepatic CoQ synthesis, redox state, mROS production via RET, and their collective impact on glucose homeostasis in the context of obesity. Created with BioRender.com.

Extended Data Table 1 |. Characteristics of subjects undergoing liver biopsies for gene expression analysis.

Transcriptome cohort 1 (n=14)
Age, yr	65 (58–70)
Sex, male	8 (57%)
Diabetes Mellitus	10 (71%)
Arterial Hypertension	9 (64%)
Dyslipidemia	4 (28%)
BMI (Kg/m2)	31 (28–36)
ASAT (U/L)	44 (35–72)
ALAT (U/L)	42 (21–78)
GGT (U/L)	87 (65–195)
Glucose (mg/dL)	127 (110–145)
Total cholesterol (mg/dL)	176 (149–186)
Triglyceride (mg/dL)	110 (69–140)
Serum albumin (g/L)	38 (32–42)
Platelet count (109 cells/L)	94 (82–114)
Steatosis grade (n, %)
S0	3 (21%)
S1	5 (36%)
S2	4 (29%)
S3	2 (14%)
Fibrosis grade (n, %)
F0	0
F1-F3	5 (36%)
F4	9 (64%)

Extended Data Table 2 |. Characteristics of subjects undergoing liver biopsies for CoQH₂/CoQ ratio analysis.

Cohort 2 CoQ10 redox state (n=13)
Age, yr	64 (55–65)
Sex, male	12 (92%)
Diabetes Mellitus	2 (15%)
Arterial Hypertension	4 (31%)
Dyslipidemia	5 (38%)
BMI (Kg/m2)	32 (28–36)
ASAT (U/L)	40 (22–52)
ALAT (U/L)	46 (23–65)
GGT (U/L)	44 (31–77)
Glucose (mg/dL)	104 (94–120)
Total cholesterol (mg/dL)	164 (115–198)
Triglyceride (mg/dL)	120 (89–211)
Serum albumin (g/L)	46 (41–47)
Platelet count (109 cells/L)	224 (136–239)
Steatosis grade (n, %)
S0	2 (15%)
S1	7 (53%)
S2	3 (23%)
S3	1 (7.7%)
Fibrosis grade (n, %)
F0	4 (31%)
F1-F3	5 (38%)
F4	4 (31%)

Supplementary Material

The online version contains supplementary material available at https://doi.org/10.1038/s41586-025-09072-1.

Data availability

All data necessary to evaluate the conclusions of this study are present in the article, extended data and/or the Supplementary Information. All the data and raw materials that support the findings of this study have been deposited into FigShare (https://figshare.com/s/0b856e078932282d73d4)⁶⁸. Metabolomic data are available with the study identifier ST003846 at the NIH Common Fund’s National Metabolomics Data Repository website, the Metabolomics Workbench (https://doi.org/10.21228/M8WJ9W). Source data are provided with this paper.