Hepatic gene replacement improves energy metabolism and survival in a mouse model of neonatal mitochondrial disease GRACILE syndrome

Abstract

Preclinical gene therapy studies of mitochondrial diseases remain limited due to the typically multi-organ manifestations and the scarcity of physiologically relevant animal models. Mutations in BCS1L, a nuclear gene encoding an assembly factor for mitochondrial complex III (CIII), are the most common cause of CIII deficiency. The most severe phenotype, GRACILE syndrome, is caused by a homozygous Finnish founder mutation (c.A232G, p.S78G). The corresponding Bcs1l^p.S78G knockin mouse model recapitulates the human disease, with juvenile-onset hepatopathy, tubulopathy, growth restriction, segmental progeria, and short survival. Here, we performed liver-targeted recombinant adeno-associated virus (rAAV)-mediated gene replacement in this model. A single intraperitoneal injection of rAAVs encoding wild-type Bcs1l restored CIII assembly and activity in the liver, preventing hepatopathy. Hepatocyte-specific correction was sufficient to alleviate hypoglycemia, improve growth, normalize systemic metabolism, and extend survival by nearly 2-fold, despite persistent CIII deficiency in other tissues. Remarkably, restoring CIII activity in the liver robustly corrected the skeletal muscle transcriptomic changes, particularly those linked to altered energy substrate utilization. These results underscore the central role of the liver in systemic energy homeostasis and growth regulation in multi-organ mitochondrial diseases and demonstrate the therapeutic potential of hepatocyte-directed gene replacement in phenotypes with prominent hepatopathy.

Graphical abstract

Banerjee and colleagues performed liver-targeted recombinant adeno-associated virus (rAAV)-mediated gene replacement in a mouse model of mitochondrial complex III deficiency. rAAVs restored CIII assembly and activity in the liver, preventing hepatopathy. This was sufficient to alleviate hypoglycemia, improve growth, normalize systemic metabolism, and extend survival by nearly 2-fold.

Introduction

Mitochondrial disorders are a heterogeneous group of inherited metabolic conditions, often affecting oxidative phosphorylation.^1,2 They can arise from mutations in either mitochondrial or nuclear genes, manifest at any age, and involve almost any organ, although most are myopathies, encephalopathies, or encephalomyopathies. CIII deficiencies constitute a relatively small fraction of mitochondrial diseases.³ The most severe CIII deficiency phenotype, GRACILE syndrome (growth restriction, aminoaciduria, cholestasis, liver iron overload, lactic acidosis, and early death), is caused by a homozygous BCS1L missense mutation (c.A232G, p.S78G).^4,5 A knockin mouse model carrying the GRACILE syndrome mutation recapitulates many features of the syndrome, including hepatopathy, kidney tubulopathy, and systemic metabolic crisis.^6,7 On a mitochondrial DNA (mtDNA) background harboring a spontaneous mt-Cyb^p.D254N variant, the mutant mice succumb to metabolic crisis by 1 month of age,⁸ quite closely resembling the neonatally lethal human phenotype. In contrast, on a wild-type mtDNA background, the early disease presentation is similar but less severe, and the mice ultimately succumb to dilated cardiomyopathy by 5–6 months of age.^8,9 The liver is one of the most prominently affected organs in GRACILE syndrome patients and Bcs1l^p.S78G mice. How the systemic phenotypes, such as growth restriction and loss of white adipose tissue, relate to the liver disease remains unclear.¹⁰ In general, inter-organ communication between the affected and unaffected organs in mitochondrial diseases is poorly understood, largely due to the lack of suitable physiologically relevant models.^11,12

In gene replacement therapy, the wild-type version of a mutated gene, or other therapeutic gene, is introduced into the affected tissues using suitable delivery strategies, typically to treat genetic conditions.¹³ Among the available viral vectors for gene therapy, recombinant adeno-associated viruses (rAAVs) are widely used because of their safety and broad tissue tropism.^14,15 While efficient systemic gene delivery remains challenging, current technologies can target specific cells or tissues to achieve therapeutic effects. In clinical trials of genetic diseases, rAAVs have shown efficient gene delivery across a variety of organs and tissues, including the liver (ClinicalTrials.gov: NCT00377416, NCT02082860, and NCT02484092), eye (ClinicalTrials.gov: NCT02946879, NCT03001310, and NCT02781480), and brain (ClinicalTrials.gov: NCT00229736, NCT05603312, and NCT00195143). Organ-targeted gene therapy has shown robust effects in some preclinical models of mitochondrial disorders.^{16,17,18,19,20} Leber’s hereditary optic neuropathy is, however, currently the only primary mitochondrial disease undergoing clinical trials for gene replacement therapy.²¹ Liver-targeted gene therapy was shown to rescue lethality in a zebrafish model of Leigh syndrome (mutated LRPPRC) and in a mouse models of mitochondrial neurogastrointestinal encephalomyopathy (mutated TYMP) and hepatocerebral mtDNA depletion syndrome (DGUOK knockout).^22,23,24,25

Here, we delivered rAAVs of serotype 9 encoding wild-type mouse Bcs1l under a broadly active or hepatocyte-specific promoter²⁶ into presymptomatic Bcs1l^p.S78G mice. We used mice of the genotype Bcs1l^p.S78G;mt-Cyb^p.D254N throughout this study. As primary outcomes, we assessed growth and survival. We show that the rAAV-based gene replacement is effective and that hepatocyte-specific rescue of CIII function is sufficient to ameliorate both hepatic and systemic manifestations and double the survival of the mice.

Results

Liver-specific gene therapy extends the survival of Bcs1l^p.S78 mice

We utilized rAAVs with hepatocyte-specific ApoE enhancer and α1-antitrypsin (AAT) promoter²⁷ to drive the expression of enhanced green fluorescent protein (EGFP) or mouse BCS1L in vivo. Moving forward, we will refer to rAAVs carrying the Bcs1l or EGFP transgene as AAT-Bcs1l and AAT-EGFP, respectively. The protocol we used for the rAAV administration and assessment of disease progression is shown in Figure 1A. We assessed transduction efficiency by viral copy number qPCR and expression from the AAT promoter by analyzing EGFP expression in the liver, kidney, and skeletal muscle (quadriceps) with western blot and immunohistochemistry. Whereas the EGFP expression in the liver was abundant, kidney and skeletal muscle showed no significant EGFP protein expression (Figures S1A and S1B). Due to the lack of a good antibody against mouse BCS1L, we quantified viral and total Bcs1l mRNA to verify the Bcs1l gene replacement. AAT-Bcs1l delivery resulted in high viral copy number (5 × 10⁵ to 1 × 10⁶ viral genomes [vg]/μg genomic DNA, mean 4.4 vg/diploid genome) and viral Bcs1l mRNA expression in the liver (Figure S1C). The viral genome number (transduction) was approximately 10-fold lower in the kidney and quadriceps, with minor Bcs1l qPCR signal in the quadriceps (Figure S1C). This translated into an approximately 15-fold increase in total Bcs1l mRNA in the liver with no significant increase in kidney or quadriceps (Figure S1C). Strikingly, 1 week after the injection, the mutant livers that received AAT-Bcs1l appeared visually as healthy as the wild-type livers (Figure 1B).

Figure 1.

rAAV-based gene replacement rescues the growth and doubles the survival of CIII-deficient mice

(A) Schematic presentation of the experimental setup and the timeline of the investigations. (B) Macroscopic appearance of the livers at postnatal day 28 (P28). (C) Blood glucose values at P28 (n = 9–17/group). The dotted line indicates the critical level of glucose (<2.5 mmol/L) predicting spontaneous death. (D) Sex-segregated growth rate of the mice from P25 to P28 (n = 6–13/group). (E) Survival curves of rAAV-EGFP-injected mutant mice and those of mutant mice injected with three different Bcs1l-expressing rAAVs (n = 5–8/group). (F and G) Respiratory exchange ratio (RER) at age P26–P27. Circadian cycle of the RER and the average 12 h of daytime and nighttime measurements are shown (n = 4/group). WT, wild type. Statistics: one-way ANOVA followed by the selected pairwise comparisons with Welch’s t statistics (C, D, and F) and log-rank test (Mantel-Cox) (E). ∗p indicates comparison between survival of AAT-Bcs1l and CAG-Bcs1l group. The error bars represent standard deviation. All data points derive from independent mice.

The Bcs1l^p.S78G mice in the mt-Cyb^p.D254N mtDNA background succumb to metabolic crisis with hypoglycemia at approximately 1 month of age (postnatal day 30 [P30] to P35).^6,8,10 At P28, upon AAT-Bcs1l injection, most mutants were able to maintain normoglycemia, and, importantly, none of the treated mice showed extreme hypoglycemia (blood glucose less than <2.5 mmol/L) (Figure 1C). While the AAT-EGFP-treated mutants lost weight after P25, the mutants treated with AAT-Bcs1l gained weight (Figure 1D). The AAT-EGFP-injected mutant mice reached the criteria for euthanasia (>15% weight loss) between P28 and P30 (Figure 1E). In contrast, the mutants injected with the hepatocyte-specific rAAV-Bcs1l showed no signs of terminal deterioration or spontaneous deaths before the age of P53. Their median survival was 58 days (Figure 1E). Toward the end of the extended survival, the glucose levels became critically low again (Figure S2A).

Being an episomal vector system, expression from rAAVs can dilute or fade away over time. Compared to the livers from P28 mice (Figure S1C), Bcs1l mRNA expression indeed decreased by the end stage (Figure S2B). To assess whether vector dilution accounted for the eventual deterioration of the AAT-Bcs1l-treated mice, we co-injected a PiggyBac transposase-encoding rAAVs with the AAT-Bcs1l virus for persistent long-term expression via genomic integration.²⁷ This strategy resulted in significantly higher end-stage hepatic Bcs1l expression (Figure S2B), but the survival did not extend further (Figure 1E). This result suggests that the eventual deterioration was due to the disease progressing in the other affected organs. To interrogate the effects of extrahepatic transduction, we used a construct with the broadly active CAG promoter (CAG-Bcs1l) with an identical dose and intraperitoneal injection. The broader Bcs1l expression did increase the survival further, albeit relatively modestly (15%), to a median of 66 days (Figure 1E). This vector also led to high end-stage hepatic Bcs1l expression (Figure S2B).

Indirect calorimetry revealed a significant alteration in whole-body substrate utilization in the mutant mice (Figures 1F and 1G). Consistent with our previous findings,^8,9 the respiratory exchange ratio (RER) of <0.9 indicated a shift from glucose to fatty acid oxidation (FAO), a consequence of the hypoglycemia and glycogen depletion. AAT-Bcs1l-treated mutants had significantly higher RER values, demonstrating a systemic preservation of glucose utilization and energy metabolism (Figures 1F and 1G).

Restoration of hepatic CIII assembly and activity

As we have previously extensively characterized the respiratory complex assembly and activities in the Bcs1l^p.S78G and Bcs1l^p.S78G;mt-Cyb^p.D254N mice,^6,8,28,29 here we only assessed CIII assembly and activity in the liver. Control mutants had a decreased amount of fully assembled CIII dimer (CIII₂). Almost all the residual fully assembled CIII₂ was in the CI-CIII₂ supercomplexes, with free CIII₂ being absent. AAT-Bcs1l efficiently increased the levels of fully assembled CIII₂ (Figure 2A). In the liver mitochondria of untreated mutants, the mean CIII activity was 26% of wild type. AAT-Bcs1l increased the mean CIII activity to 74% of wild type (Figure 2B) and improved hepatic ATP levels (Figure 2C). We have previously estimated that the symptoms appear at an approximately 50% residual CIII activity in the liver.⁸ Even though the correction of the hepatic CIII assembly and activity was partial, it still clearly exceeded this ∼50% threshold required to delay disease onset.

Figure 2.

Hepatocyte-targeted gene replacement corrects hepatic CIII assembly and activity

(A) Representative blue native PAGE blot of UQCRFS1 (RISP) and UQCRC1 in free CIII₂ and supercomplexes (SCs) from isolated P28 liver mitochondria, and RISP per total UQCRC1 ratio (n = 4/group). The presence of RISP subunit marks the fully assembled CIII dimers. (B) CIII activity normalized to total protein in isolated liver mitochondria at P28 (n = 4/group). (C) Total ATP in liver at P28 (n = 6–7/group). (D) mRNA expression of Ppargc1a (PGC-1α), from P28 liver (n = 7–8/group). (E) Western blot quantification and representative blots of HSP60 and VDAC1 as markers of mitochondrial mass in P28 liver lysates (n = 6–8/group). (F) Representative images of liver sections immunostained for the mitochondrial marker HSP60 (n = 3–5/group). WT, wild type. Statistics: Mann-Whitney U test (A and C) and one-way ANOVA followed by the selected pairwise comparisons with Welch’s t statistics (B, D, and E). The error bars represent standard deviation. All data points derive from independent mice. Scale bar represents 100 μm.

Mitochondrial dysfunction can lead to increased mitochondrial mass, which we have shown in Bcs1l^p.S78G mice.^28,30 Here, the mitochondrial biogenesis-driving transcriptional regulator Ppargc1a (Pgc-1α) (Figure 2D) and the mitochondrial mass markers VDAC1 and HSP60 were increased in the untreated mutant livers, and AAT-Bcs1l treatment partially prevented these changes (Figure 2E). Immunohistochemistry showed frequent hepatocytes with irregular, abnormally strong staining for the mitochondrial marker HSP60 in the mutant livers as compared to the fine granular cytoplasmic mitochondrial staining in wild-type livers. The AAT-Bcs1l fully prevented the abnormal HSP60 staining pattern (Figure 2F).

Prevention of liver pathology and correction of energy metabolism

Histopathological analysis of the P28 mutant mice showed hepatopathy characterized by incipient expansion of portal areas, ductular reactions, and increased cell death (Figures 3A–3D). AAT-Bcs1l fully prevented these changes (Figures 3A–3D) as well as the upregulation of the mitochondrial dysfunction-associated mitokine growth differentiation factor 15 (Gdf15) (Figure 3E). A gene-expression signature derived from transcriptomics data of human liver disease and multiple mouse liver disease models³¹ was fully normalized by AAT-Bcs1l (Figure 3F). In the 2-month-old treated mutants, the hepatic ductular reactions had increased slightly (Figures S3A–S3D) and Gdf15 expression significantly (Figure S3E), suggesting that the liver disease eventually progressed despite the remaining viral Bcs1l expression.

Figure 3.

Gene therapy prevents mitochondrial hepatopathy and improves kidney cortex thickness in CIII-deficient mice

(A and B) Representative images of H&E-stained liver sections (n = 6–7/group), showing tissue morphology and expansion of portal areas (indicated by dotted lines) and liver sections immunostained for the apoptotic cell marker cleaved caspase-3 (CC3) (n = 5/group) at P28. (C and D) Quantification of hepatic ductular reactions and apoptotic cells from H&E-stained (n = 6–7/group) and CC3-immunostained sections (n = 5/group), respectively. (E) Gdf15 mRNA expression in P28 liver (n = 7–8/group). (F) Heatmap visualization of the top 20 most differentially expressed genes linked to liver disease (n = 5–6/group). (G and H) Representative images of H&E-stained kidney sections (n = 4–7/group) showing tissue morphology and kidney sections immunostained for CC3 (n = 5/group) at P28. (I and J) Quantification of kidney cortex thickness in the kidney from P28 mice (n = 4–7/group) and kidney sections immunostained for CC3 (n = 5/group) at P28. WT, wild type. Statistics: Mann-Whitney U test (C and D) and one-way ANOVA followed by the selected pairwise comparisons with Welch’s t statistics (E, I, and J). The error bars represent standard deviation. All data points derive from independent mice. Scale bar represents 100 μm.

The kidney is another major affected organ in GRACILE syndrome and in Bcs1l^p.S78G mice. Proximal tubulopathy with loss of cortex volume is the main histological manifestation (Figures 3G–3J).^4,6,9 Even though the number of apoptotic cells was not decreased by AAT-Bcs1l (Figures 3H and 3J), the kidney cortex thickness was slightly increased (Figures 3G and 3I), in line with the improved overall growth. The treated mice showed normal kidney cortex thickness even at the end stage, despite increased cell death (Figures S4A–S4F). The typical albuminuria of the mutantmice²⁹ was not affected by AAT-Bcs1l (Figure S4G).

The metabolic status of the Bcs1l^p.S78G mice resembles starvation,^4,29 characterized by hypoglycemia (Figure 1C) and depleted glycogen stores (Figures 4A and S5A). Another hallmark of the disease is the near-complete absence of white adipose tissue deposits,^4,29 reflecting reliance on lipids for fuel. Body composition analysis showed that AAT-Bcs1l did not correct fat mass at P28 (Figure S5C), possibly because of a persistent need for fuel. Interestingly, by P60 (endpoint of the survival analysis), the treated females, but not males, showed normal fat mass (Figure S5C).

Figure 4.

Liver energy metabolism reflects the restored CIII function in hepatocytes

(A) Representative images of glycogen stained with PAS on liver sections (n = 5/group) from P28 mice. (B) Western blot quantification of the phosphorylation status of AMPKα in P28 liver lysates (n = 6–8/group). (C) PCA of the liver transcriptome (n = 5/group) from P28 mice. (D and E) Heatmap visualization of the top ten most differentially expressed genes related to TCA cycle and mitochondrial fatty acid oxidation in P28 liver (n = 5–6/group). (F–I) Pdk4, Cd36, Cpt1a, and Pparα mRNA expression from P28 liver (n = 7–8/group). (J) Representative images of oil red O staining of liver cryosections (n = 4/group) showing lipid accumulation at P28. Statistics: one-way ANOVA followed by the selected pairwise comparisons with Welch’s t statistics (B and F–I). Confidence ellipses indicate 95% confidence intervals for each group (C). The error bars represent standard deviation. All data points derive from independent mice. WT, wild type. Scale bar represents 100 μm.

As an established marker for energy balance, we examined AMP-dependent protein kinase (AMPK), a central regulator of ATP and glucose availability. The level of the AMPK α subunit was decreased in the mutant liver (Figure 4B), as also previously published.^6,29 However, the activated form (Thr172-phosphorylated AMPKα [p-AMPKα]) was increased, leading to a higher p-AMPKα/AMPKα ratio, a sign of energy deficiency (Figure 4B). This was prevented by AAT-Bcs1l (Figure 4B), indicating normalized energy status in hepatocytes. Principal-component analysis (PCA) of transcriptome data showed a near-complete overlap of liver gene-expression profiles between wild-type and AAT-Bcs1l-treated mutants (Figure 4C), with only 65 differentially expressed genes between them (Figure S5D). For example, gene-expression changes related to the tricarboxylic acid (TCA) cycle and FAO were upregulated and downregulated, respectively, in the control mutant livers (Figures 4D and 4E) but not in the AAT-Bcs1l-treated livers (Figures 4D and 4E). The upregulation of Pdk4 (inhibition of the pyruvate dehydrogenase complex) and Cd36 (increased fatty acid uptake from circulation), pointing to an attempt to shift from glucose to FAO upon hypoglycemia and glycogen depletion, was also prevented (Figures 4F and 4G). Despite the Cd36 upregulation in the control mutants, the expression of the key FAO genes Cpt1a (controlling mitochondrial fatty acid import)³² and Pparα (a major FAO regulator) was markedly decreased (Figures 4H and 4I). Together with the microvesicular steatosis (Figures 4J and S5B), these findings suggest a mismatch between fatty acid uptake and FAO capacity in the mutant liver. AAT-Bcs1l preserved the hepatic glycogen stores and prevented fat accumulation, as well as basically all energy metabolism-related gene-expression changes (Figures 4A and 4F–4J). These improvements were also reflected in the restored expression of Igf1 and Ghr, consistent with the improved growth (Figures S5E, S5F, and 1D).

The Bcs1l^p.S78G mice show regenerative hepatocyte proliferation, which, in the face of their depleted nucleotide pools and other biosynthetic resources, leads to DNA damage, cell-cycle arrest, and senescence.²⁹ The restoration of CIII function in hepatocytes decreased the DNA damage/senescence marker γH2AX (Figure 5A) and abolished the cell-cycle arrest marker CDKN1A (p21) (Figure 5B). AAT-Bcs1l also fully prevented hepatic c-MYC upregulation (Figure 5C), which in both cancer and normal cells can bypass cell-cycle checkpoints^33,34 as well as the upregulation of the cell proliferation markers PCNA and cyclin A2 (Figure 5D). In line with the c-MYC-driven senescence, our previous study highlighted activation of replication stress and Fanconi anemia-related genes in the mutant liver.²⁹ AAT-Bcs1l largely prevented these transcriptional changes (Figures 5E and 5F). The upregulation of cathepsin L (Ctsl), an aging and senescence marker,³⁵ was also prevented by AAT-Bcs1l (Figure 5G). Gene-expression analyses of senescence-associated secretory phenotype (SASP) showed highly increased expression of the EGF receptor ligands amphiregulin (Areg) and betacellulin (Btc) as well as increased expression of several chemokines and cytokines associated with SASP, such as chemokine ligand 3 (Ccl3) and interleukin-1 receptor agonist (Il1rn), in the mutant liver. The upregulation of all these SASP factors was prevented by AAT-Bcs1l (Figure 5H). As part of the tissue inflammation, markers for a specific fibrogenic bone-marrow-derived population of macrophages named Fab5 (CD9⁺TREM2⁺ and expressing SPP1, GPNMB, FABP5, and CD63)³⁶ were highly increased in the mutant liver, and this was essentially fully prevented by AAT-Bcs1l (Figure 5I). These changes suggest that hepatocyte damage specifically attracts circulating inflammatory cells to the liver and that local correction of the tissue damage prevents this.

Figure 5.

Restoration of CIII function prevents the upregulation of hepatic senescence markers

(A–D) Western blot quantification and representative blots for γH2AX, CDKN1A (p21), c-MYC, PCNA, and cyclin A2 from P28 liver lysates (n = 5–8/group). (E and F) Heatmap visualization of the top ten most differentially expressed genes related to replication stress response and Fanconi anemia pathway in P28 liver (n = 5–6/group). (G) mRNA expression of Cstl from P28 liver (n = 7–8/group). (H) mRNA expression of senescence-associated secretory phenotype (SASP) genes from P28 liver (n = 7–8/group). (I) Fab5 macrophage gene expression in the liver (n = 5–6/group). (J) Hepatokine gene expression in the liver (n = 5–6/group). (K) PCA of targeted metabolomics from P28 liver (n = 6/group). WT, wild type. Statistics: one-way ANOVA followed by the selected pairwise comparisons with Welch’s t statistics (A, C, D, and G–J). Mann-Whitney U test (B) ∗∗∗∗p ≤ 0.0001, ∗∗∗p ≤ 0.001, ∗∗p ≤ 0.01, ∗p ≤ 0.05. Confidence ellipses indicate 95% confidence intervals for each group (K). The error bars represent standard deviation. All data points derive from independent mice.

The mRNA levels of several hepatokines (Ahsg, Smoc1, Fgl1, Lect2, and Sepp1), which are proteins secreted by the liver into circulation for metabolic regulatory functions in other tissues, thus potentially mediating some of the effects of the gene replacement, were normalized by AAT-Bcs1l (Figure 5J).

Finally, targeted metabolomics analysis covering mainly amino acids, glycolysis intermediates, and TCA-cycle metabolites confirmed that, in line with the transcriptome changes, gene therapy led to overall normalization of hepatic energy metabolism (Figure 5K and Table S1). For example, the levels of several amino acids, which we have previously shown to be highly increased in the juvenile mutant liver,³⁷ probably reflecting protein degradation for fuel, were normalized (Table S1). Phosphoenolpyruvate accumulation and the depletion of glycolysis (glucose-6-phosphate and fructose-6-phosphate) and TCA-cycle (malate and fumarate) intermediates (Table S1) were also normalized, indicating, together with the robust Pdk4 induction (Figure 4F), a significant remodeling of the TCA cycle.

Restoration of CIII function in the liver improves skeletal muscle metabolism

The skeletal muscle is a major energy-consuming organ, and Bcs1l^p.S78G mice show low skeletal muscle CIII activity⁸ yet have no obvious histological signs of myopathy. We quantified AMPK phosphorylation in the skeletal muscle to assess the effect of the CIII deficiency on the energy status of this tissue. The p-AMPKα/AMPKα ratio was not increased in mutants and not affected by AAT-Bcs1l (Figure S6A), likely because we sampled the muscles without exercise stress. However, in line with improved systemic energy metabolism, AAT-Bcs1l treatment strikingly shifted transcriptome-wide expression changes in the skeletal muscle toward normal (Figure 6A). Upon AAT-Bcs1l, the expression of ∼96% of the dysregulated genes in the mutant skeletal muscle were significantly shifted toward wild-type levels (Figure 6B). This effect was not due to off-target expression in skeletal muscle, as confirmed by qPCR for total Bcs1l (Figure S1C). Gene set enrichment analysis of the transcriptomics data highlighted the upregulation of energy-metabolism-related Gene Ontology biological processes, such as amino acid and glucose metabolism in the mutant skeletal muscle (Figures S6B and 6C–6E). Similarly to the liver, the mutant skeletal muscle showed upregulation of Ppargc1a (Pgc-1α), which was effectively prevented by AAT-Bcs1l (Figure 6F). In response to decreased glucose availability, skeletal muscle also compensates by upregulating FAO. Consistent with this, the expression of Pdk4 and Cd36 was markedly elevated in the mutant muscle (Figures 6G and 6H), indicating a metabolic shift from glucose to fatty acid utilization. AAT-Bcs1l significantly attenuated Pdk4 and Cd36 upregulation (Figures 6G and 6H). Importantly, although targeted metabolomics data showed that changes in skeletal muscle energy metabolites (amino acids, glycolysis intermediates, and TCA-cycle intermediates) were small, global analysis clearly separated the AAT-EGFP and AAT-Bcs1l skeletal muscle, with the latter shifted toward wild type (Figure 6I and Table S2). In summary, restoring CIII function and mitochondrial respiration in the liver led to robust correction of markers for systemic and skeletal muscle energy metabolism.

Figure 6.

Liver-targeted gene replacement improves skeletal muscle metabolism

(A) PCA of differentially expressed genes in the P28 skeletal muscle transcriptome (n = 5–6/group). (B) A volcano plot showing differentially expressed genes between mutant and wild-type (WT) livers. The genes whose expression was normalized by the gene therapy are marked with orange and light blue (n = 5–6/group). (C–E) Heatmap visualization of the top ten most differentially expressed genes related to lipid, carbohydrate, and amino acid metabolism in P28 skeletal muscle (n = 5–6/group). (F–H) Ppargc1, Pdk4, and Cd36 mRNA from P28 skeletal muscle (n = 6–8/group). (I) PCA of the P28 skeletal muscle metabolomics (n = 6/group). Statistics: confidence ellipses indicating 95% confidence intervals for each group (A and I), one-way ANOVA followed by the selected pairwise comparisons with Welch’s t statistics (F), and Mann-Whitney U test (G and H). The error bars represent standard deviation. All data points derive from independent mice.

Discussion

Despite the rapidly expanding knowledge of molecular disease mechanisms in many monogenic disorders, including mitochondrial diseases, progress in pharmacological therapies has been incremental. Therapeutic strategies aimed at correcting the underlying genetic defects, such as gene therapy, hold the greatest potential for durable disease modification and even cure. Here, we present the first preclinical gene therapy trial in a mouse model of a human CIII deficiency, using a robust model based on the GRACILE syndrome patient mutation.^4,5 We show that both broad and hepatocyte-specific rAAV-based presymptomatic gene replacement prevented the early metabolic crisis and doubled the survival of mice with otherwise very rapid deterioration by 1 month of age. AAT-Bcs1l prevented liver disease and abrogated hepatocyte senescence and hepatic inflammation. These results support the view that mitochondrial dysfunction in hepatocytes alone underlies the liver disease in GRACILE syndrome and that the liver disease plays an important role in the compromised systemic energy metabolism in this disease. For example, the growth of the AAT-Bcs1l-treated mutant mice was improved, which was likely due to their improved energy metabolism.

During fasting, but also in the chronic starvation-like metabolic state in mitochondrial diseases, fatty acids are mobilized from adipose tissue to the liver.³⁸ Despite apparently normal food intake, the Bcs1l^p.S78G mice face hypoglycemia and glycogen depletion, forcing a shift from glucose to fatty acid utilization.^6,10 To facilitate this, mutant hepatocytes upregulate fatty acid uptake. However, persistent microvesicular steatosis suggests insufficient hepatic FAO, a heavily mitochondria-dependent catabolic process. Restoring hepatic mitochondrial respiration with AAV-Bcs1l reversed the markers for the glycolysis-to-FAO metabolic switch and prevented glycogen depletion and hypoglycemia. Indirect calorimetry data also reflected prevention of the glycolysis-to-FAO metabolic switch. As an example of extrahepatic effects potentially mediated by the improved systemic energy metabolism, liver-specific gene therapy slightly improved kidney cortex mass, which is associated with proximal tubulopathy in both patients and mutant mice.^4,6,9 Because the skeletal muscle shows ∼75% loss of CIII activity in juvenile Bcs1l^p.S78G mice⁸ and is a large energy-consuming organ, we wanted to assess its potential role in the effects of gene therapy. Quite astonishingly, AAT-Bcs1l prevented transcriptome-wide alterations in the skeletal muscle and had a global effect at metabolite level, underscoring a strong metabolic and/or hormonal effect from the liver. For example, AAT-Bcs1l corrected the expression of several hepatokines with known effects on skeletal muscle in other models.³⁸ Notably, liver-specific gene therapy prevented the transcriptional glycolysis-to-FAO switch also in skeletal muscle. Together, these findings demonstrate that restoring hepatic CIII function exerts systemic benefits on peripheral tissues, improving overall energy homeostasis.

Gene therapy led to robust prevention of liver disease and metabolic alterations in Bcs1l^p.S78G mice, yet the mice still deteriorated quite sharply at ∼2 months. The genomic integration of the AAT-Bcs1l cassette did not further extend the survival, suggesting that the eventual deterioration of the mice was due to non-hepatic tissues. Intraperitoneal injection of rAAVs with broader tissue expression provided only a modest additional survival benefit. Therefore, less efficient biodistribution in harder-to-transduce tissues likely limited the therapeutic effect. Optimizing delivery strategy and/or capsid may improve the biodistribution. Loss of episomal expression over time may also be a challenge, particularly in patients with much longer life span than mice. On the other hand, recent encouraging results from CI- and CIV-related encephalopathy mouse models showed long-lasting (15 months) effects with rAAV-based gene replacement in the brain.³⁹ Repeated rAAV administration is hindered by immune responses against AAV capsids.^40,41 Therefore, alternative delivery systems such as lipid nanoparticles,⁴² engineered virus-like particles,⁴³ and extracellular vesicle-derived vectors⁴⁴ are emerging as promising tools that can bypass immunogenicity, enable repeated dosing, and potentially broaden tissue tropism. Finally, with the advent of CRISPR-based genome editing, including base editing⁴⁵ and prime editing,⁴⁶ it is becoming increasingly feasible to achieve permanent correction of disease mutations, potentially overcoming the durability limitations of conventional rAAV-based gene replacement. The GRACILE syndrome mutation is, by chance, difficult to correct using base editing (our unpublished data from patient fibroblasts), but prime editing strategies should work and can be designed for further preclinical trials in the Bcs1l^p.S78G mouse model.

In summary, our results demonstrate the first successful preclinical gene therapy trial in a mouse model of mitochondrial CIII deficiency. The most striking outcome of this study was the degree of systemic rescue achieved through hepatocyte-specific restoration of the mutated gene and CIII function. The results are potentially translatable to mitochondrial disease patients with other BCS1L mutations, most of which cause clearly milder phenotypes than the GRACILE syndrome mutation.⁴⁷ The work also sheds light on the role of the liver in systemic manifestations, such as hypoglycemia and loss of adipose tissue, in CIII deficiency. These findings suggest that liver transplantation may be a rational therapeutic option in milder BCS1L-related disorders with significant liver involvement.⁴⁸ Liver-directed gene therapy may therefore represent a promising strategy for multi-organ mitochondrial diseases with prominent hepatopathy.

Limitations of the study

While AAT-Bcs1l-mediated hepatic rescue alleviated systemic metabolic stress and improved transcriptome and metabolome signatures of skeletal muscle, we could not perform functional assays of muscle performance, as the mutants are severely exercise intolerant, and thus the physiological impact of the rescue remains unknown. As rAAV-mediated expression declines over time, this needs to be taken into account while considering translation to humans with much longer life spans. Finally, gradual failure of the kidney may explain the eventual deterioration of the treated mice. Because the kidney tubular cells are quite challenging to transduce efficiently, assessing this was beyond the scope of this study.

Materials and methods

Cloning and virus production

Mouse Bcs1l coding sequence (MmBcs1l) was PCR amplified from mouse tissue cDNA (primers EcoRI-MmBcs1l: 5′-ATGAATTCACCATGCCATTTTCAGACTTTGTTCTG-3′ and MmBcs1l-STP-HindIII: 5′-ATAAGCTTTCACCTCAGAGATTCAATGTTGT-3′). The insert was ligated into pBluescript, sequenced, and subcloned into the rAAV vector pSUB-CAG-WPRE for broad expression under the chicken β-actin promoter and the cytomegalovirus immediate-early enhancer (RRID: Addgene_119227), and into the pAAV2-LSP1-PB(TR)-EGFP vector.²⁷ The latter drives hepatocyte-specific expression under the human ApoE enhancer and α1-antitrypsin promoter (generously provided by Prof. Ian Alexander, University of Sydney). The parental vector encoding EGFP was used as a control. Along with the expression cassette, the vectors also contained flanking PiggyBac transposon recognition sequences, which allow for genomic integration upon parallel expression of the PiggyBac transposase.²⁷ Serotype 9 viral particles were produced by the AAV Gene Transfer and Cell Therapy Core Facility of the University of Helsinki.

Mouse breeding and husbandry

The animal facilities of the University of Helsinki maintained the mice on the C57BL/6JCrl background (Harlan stock 000016). Bcs1l wild-type and heterozygous animals were used as healthy controls (wild type by phenotype). Both males and females were used, and data are shown separately for them only if a significant sex difference was observed or known.

The mice were housed in individually ventilated cages with a 12-h light/12-h dark cycle at a temperature of 22°C–23°C, and they had ad libitum access to water and chow (2018 Teklad or 2916 Teklad rodent diet). Mouse health was monitored by manual behavioral scoring and weighing according to the ethical permit. Samples were collected on P28 or according to the survival of the mice. In survival analysis, the mice were euthanized when weight loss was greater than 15% of the maximum weight of the individual mouse.

The animal studies were approved by the animal ethics committee of the State Provincial Office of Southern Finland (ESAVI/16278/2020 and ESAVI/31141/2023), and were performed according to Federation of Laboratory Animal Science Associations (FELASA) guidelines. The animal work and experimental setup were designed following 3R principles.

rAAV administration

Presymptomatic (P19–P23) mice were injected intraperitoneally with 100 μL of saline containing 5 × 10¹⁰ viral particles encoding EGFP or wild-type Bcs1l. rAAVs encoding EGFP were used as a control for any vector-related effects, as is common in preclinical gene therapy trials. However, because the EGFP-expressing mutant mice were comparable to the non-injected mutant mice, we determined the dose based on the literature and test intraperitoneal injections of 1 × 10¹⁰, 5 × 10¹⁰, and 25 × 10¹⁰ viral particles expressing EGFP in wild-type mice. We selected the 5 × 10¹⁰ dose based on the visual EGFP expression in the transduced liver.

Assessment of body composition and whole-body metabolism

The echoMRI-based MiniSpec Body Composition Analyzer (Bruker, USA) was used to quantify fat mass. Respiratory exchange ratio of P27–P26 mice was determined with the Comprehensive Laboratory Animal Monitoring System (CLAMS) (Columbus Instruments).

Mouse tissue sample collection

The mice were euthanized by cervical dislocation. Before sample collection, the mice underwent only a short 2-h fasting period to avoid exacerbation of hypoglycemia in the mutant mice, and all samples were obtained at a similar time of the day, during the light period. Blood glucose was measured with a quick meter (Freestyle Lite, Abbott, USA) from the blood within the body cavity while collecting the other samples. Tissues were either placed in 10% histology-grade formalin or snap frozen in liquid nitrogen and stored at −80°C.

Viral copy number quantification, qPCR, and mRNA sequencing

Genomic DNA was isolated from the tissues using SDS-proteinase K digestion followed by isopropanol precipitation. Viral Bcs1l qPCR primers were used to quantify vector DNA copy numbers.

Total RNA was extracted from the snap-frozen tissue samples with RNAzol RT reagent (Sigma-Aldrich). qPCR was performed from the cDNA using EvaGreen- and Phire II Hot Start DNA polymerase-based detection chemistry.⁸ CFX96 thermocycler and CFX Manager software (Bio-Rad) were utilized to perform the qPCR and data analysis. LinRegPCR software⁴⁹ was used to calculate PCR efficiency. All primers used are listed in Table S3. Gak and Rab11a served as reference genes.

mRNA sequencing and its primary bioinformatics analysis were performed by Novogene (Beijing). Downstream analyses were conducted in R (v.4.4.2) using the DESeq2 package.⁵⁰ Variance stabilizing transformation-normalized read counts generated with DESeq2 were used for PCA. Differential expression analysis was performed with DESeq2, and genes were considered significantly differentially expressed at an adjusted p value of <0.05 and an absolute log₂ fold change >1. Pathway analysis was conducted using Gene Set Enrichment Analysis (v.4.3.3).⁵¹ Data visualization, including bubble plots, PCA plots, and heatmaps, was performed in RStudio. For pathway bubble plot only GOBP gene sets with an FDR value <0.15, containing the term “metabolic,” were included. In the heatmaps, top 10–20 differentially expressed genes are shown for each pathway or Gene Ontology class.

SDS-PAGE, blue native PAGE, and western blot

SDS-PAGE and immunoblotting were performed as described by Purhonen et al.²⁹

For blue native PAGE, liver mitochondria were freshly isolated using a previously described protocol.⁸ The isolated mitochondria were solubilized by adding 6 mg of digitonin per mg of protein in a cold buffer comprising 50 mM Bis-Tris-Cl⁺, 50 mM NaCl, 1.4 mg/mL digitonin, 10% glycerol, 1 mM EDTA, and protease inhibitor mix (pH 7.0). The sample lysates were clarified by centrifugation at 18,000 × g at +4°C for 6 min. Coomassie blue G-250 (1.5 mg/mL) was added to the supernatants, and 10 μg of solubilized mitochondrial protein was separated using 3%–12% NativePAGE Bis-Tris gradient gels (Invitrogen) and electrotransferred onto polyvinylidene fluoride filters as described by Torres-Torronteras et al.²²

For the analysis of albumin, urine samples were mixed with Laemmli sample buffer and boiled, and an equivalent of 1 μL of urine was loaded onto 4%–20% Tris-glycine PAGE gels. The gels were stained with Coomassie G-250, and images were captured using a flatbed scanner. All the samples were randomized before running and processing for quantification. For representative blots, individual samples were pooled from each experimental group to obtain an average signal, which represents each group.

Assessment of respiratory chain enzymatic activities and quantification of ATP

CIII activity was measured using a spectrophotometric method that involves monitoring the reduction of cytochrome c, sensitive to antimycin A and myxothiazol, with decylubiquinol serving as the electron donor.⁸ The CIII activity data were normalized to protein content.

For the enzymatic quantification of ATP from the liver, we used a published method developed in our laboratory.⁵²

Tissue histology and immunohistochemistry

Formalin-fixed paraffin-embedded tissues underwent standard procedures for general histological assessment, including hematoxylin and eosin (H&E) and periodic acid-Schiff (PAS) stainings. Additionally, frozen liver sections fixed in formalin and saturated with 30% (w/v) sucrose were subjected to standard oil red O staining to detect triglycerides. The antigen retrieval of paraffin sections for cleaved caspase-3, HSP60, and EGFP staining (see Table S4 for the antibodies) was carried out by immersing the slides in 10 mM Tris-Cl (pH 9.0) and 1 mM EDTA and boiling for 15 min in a microwave oven. After incubating the sections in the primary antibody, ImmPress peroxidase conjugate or alkaline phosphatase polymer conjugate of secondary antibodies (Vector Laboratories) were added. Nitroblue tetrazolium was used to visualize alkaline phosphatase and diaminobenzidine peroxidase activity, respectively. Nuclear fast red (Sigma-Aldrich) and hematoxylin were used as nuclear counterstains for alkaline phosphatase and peroxidase staining, respectively.

Liquid chromatography-mass spectrometry-based metabolite analyses (targeted metabolomics)

Metabolites were extracted from 5–10 mg of muscle and liver tissue by adding 400 μL of cold extraction solvent (acetonitrile/methanol/Milli-Q water 40:40:20, Thermo Fisher Scientific). Subsequently, samples were homogenized using tissue homogenizer (Bertin Technologies) for three cycles (30 s at 5,500 rpm with 60-s pauses at 4°C) followed by centrifugation at 14,000 rpm at 4°C for 5 min. Supernatants were filtered through Phree Phospholipid Removal plates (Phenomenex), transferred into polypropylene tubes, and evaporated to dryness under a stream of nitrogen at 38°C. Dried samples were resuspended in 40 μL of extraction solvent and vortexed for 2 min before being transferred into high-performance liquid chromatography glass autosampler vials. Samples were analyzed using a Thermo Vanquish UHPLC coupled to Q-Exactive Orbitrap mass spectrometer using MS1 scanning mode.⁵³ The data quality was monitored throughout the run using pooled quality control (QC) sample prepared by pooling 5 μL from each suspended sample and interspersed throughout the run as every tenth sample before analysis with MS1 full-scan and MS2 modes. The data processing for 45 metabolites was performed using TraceFinder 5.1 software (Thermo Fisher Scientific) using confirmed retention times by in-house standard library (MSMLS-1EA, Merck) and their m/z (5 ppm) for 470 metabolites. The data were quality controlled for peak quality, then prefiltered with 30% relative standard deviation cutoff of the pooled QC and noise. Data were normalized to the average sum of intensity (peak areas).

Statistics

All samples were randomized before quantification and processing. Group differences were assessed using Welch’s t test or the Mann-Whitney U test, when appropriate, for comparisons between two groups. For comparisons among multiple groups, one-way ANOVA followed by preselected pairwise comparisons using Welch’s t statistics was applied. All pairwise comparisons were conducted using two-sided tests. Survival analyses were performed using the log-rank (Mantel-Cox) test. Statistical analyses were conducted with GraphPad Prism v.10 (GraphPad Software). Unless otherwise specified, error bars in figures represent the mean ± standard deviation. A p value of <0.05 was considered statistically significant. Statistical tests for RNA-sequencing data were conducted with R Studio.

Data and code availability

Any dataset published here is available from the corresponding author upon request.

Acknowledgments

We thank Vilma Wanne, Divya Upadhyay, Katariina Kemppainen and Sonja Jansson for technical assistance, and Prof. Ian Alexander (University of Sydney) for providing the pAAV-LSP1 plasmids. We thank the core facilities of the University of Helsinki (FIMM Digital Microscopy and Molecular Pathology Unit and the Finnish Center for Laboratory Animal Pathology, Faculty of Veterinary Medicine) for processing histological samples, Biomedicum Imaging Unit for microscopy services, Biocenter Finland- and HiLIFE-supported Helsinki Metabolomics Center for the metabolomics infrastructure, and the Laboratory Animal Center of the University of Helsinki for the animal husbandry. We acknowledge funding from Samfundet Folkhälsan, Jane and Aatos Erkko Foundation, Sigrid Jusélius Foundation, University of Helsinki, The Foundation for Pediatric Research, Finska Läkaresällskapet, Medicinska Understödsföreningen Liv och Hälsa, Magnus Ehrnrooth Foundation, The Finnish Academy of Science and Letters, Finnish Doctoral Programme in Oral Sciences (FINDOS), and American and European Societies of Gene & Cell Therapy (ASGCT and ESGCT) (travel grants to R.B.). Graphical illustrations were created with BioRender (https://www.biorender.com/).

Author contributions

R.B., J.P., V.F., and J.K. designed the study; R.B. wrote the first manuscript draft and prepared the figure panels; R.B., J.P., C.K., and J.K. performed the animal experiments and sample collection; and R.B., J.P., O.R., and J.K. performed the histological analyses. The contributions to the other methods were as follows: body composition analyses (R.B. and J.P.), SDS-PAGE and western blot analyses (R.B., N.S., and J.K.), blue native PAGE (R.B. and J.P.), qPCR (R.B., N.S., and O.R.), immunohistochemistry (R.B., J.K., and O.R.), ATP measurements (R.B. and J.P.), transcriptomics (R.B. and C.K.), and metabolomics (A.I.N., J.P., and J.K.). R.B. was responsible for the statistics, and R.B., V.F., and J.K. acquired funding for the project. All authors critically read and commented on the manuscript, and R.B. and J.K. revised it accordingly.

Declaration of interests

The authors have no interests to declare.