Mitochondria are highly dynamic organelles found in most eukaryotic cells, with a fundamental role in the generation of cellular energy through oxidative phosphorylation (OXPHOS). Critical for their function, mitochondria have retained their own genome the mitochondrial DNA, mtDNA. In mammals, replication of mtDNA is ensured by the DNA polymerase POLγ, which is composed by one catalytic subunit POLγA and two accessory subunits POLγB. Mutations in the nuclear-encoded POLG gene, coding for POLγA, are a common cause of human disease leading to a spectrum of disorders characterised by mtDNA instability, thus compromising mitochondrial function. Despite being relatively frequent, the molecular pathogenesis of POLG-related diseases is poorly understood and efficient treatments are missing, partly due to the lack of relevant in vivo models. Here, I describe the generation of two mouse models: 1) the PolgA449T/A449T mouse, which reproduces the A467T change, the most common human recessive mutation of POLG and 2) the PolgWT/Y933C mouse, which reproduces the Y955C change, the most common human dominant mutation of POLG. I focused on the use of the PolgA449T/A449T mouse and complementary in vitro techniques to provide insights into the molecular pathogenic mechanism of this POLG mutation. I describe the data showing that the mouse A449T mutation impairs DNA binding and mtDNA synthesis activities of POLγ, leading to a stalling phenotype. Most importantly, the A449T mutation also strongly impairs interaction with POLγB, the accessory subunit of the POLγ holoenzyme. This allows the free POLγA to become a substrate for LONP1 protease degradation, leading to dramatically reduced levels of POLγA in A449T mouse tissues, with consequences for the pathogenesis of the disease. In the second part of the dissertation, I explore a gene therapy approach for mitochondrial diseases associated with mutations in nuclear-encoded genes. In particular, I test the use of a novel adeno-associated virus (AAV) capsid (PHP.B) as a gene therapy platform to ameliorate the neurological symptoms of a pre-clinical mouse model of mitochondrial disease, the Ndufs4 knockout (Ndufs4-/-) mouse. A single injection with AAV-PHP.B to express the human NDUFS4 in Ndufs4-/- mice, improved lifespan, body weight gain, motor coordination and several molecular and histological features of the brain. These data provide promising proof-of-concept for the use of AAV-mediated gene therapy as a therapeutic option for the number of patients with, currently incurable, mitochondrial disease.