Mitochondrial maintenance under oxidative stress depends on mitochondrially localised α-OGG1

The mammalian mitochondrial genome consists of a circular double-stranded DNA molecule that is ∼16 kb in length and accounts for 1%–2% of the total DNA in the cell. It encodes for only 13 proteins, all of which are involved in oxidative phosphorylation and ATP synthesis, as well as for two rRNAs (12S rRNA and 16S rRNA) and 22 tRNAs that are required for mitochondrial protein synthesis. However, up to 1500 proteins encoded by the nuclear genome are also found inside mitochondria, including proteins required for replication, transcription and repair of mitochondrial DNA (mtDNA). mtDNA is highly packed in DNA–protein assemblies that constitute the mitochondrial nucleoids, which are distributed throughout the mitochondrial network (Alam et al., 2003; Legros et al., 2004). Nucleoids are anchored to the inner mitochondrial membrane and are composed of one or two DNA molecules (Kukat et al., 2011) and several core proteins involved in DNA packing, transcription and signalling. In particular, mitochondrial transcription factor A (TFAM), a key activator of mitochondrial transcription and a participant in mitochondrial genome replication, is a major component of the nucleoid and is required for mtDNA packing. Besides the core proteins, many other proteins transiently associate with mitochondrial nucleoids (Gilkerson et al., 2013).

Although many proteins known for their role in nuclear DNA repair pathways, such as the base excision repair (BER), mismatch repair, and single- and double-strand break repair pathways, are imported into mitochondria (Kazak et al., 2012), which mechanisms participate in the maintenance of mtDNA is still an open question. Mitochondrial biogenesis and dynamics, including degradation of damaged mitochondria by mitophagy, also play essential roles in preserving the stability of the mitochondrial genome. The scenario becomes even more complex when we take into consideration that each cell can contain up to several thousands of mtDNA molecules. Because of cycles of fusion and fission that promote mixing and homogenisation of mitochondrial contents, cells maintain a certain level of heteroplasmy, in which mutant and wild-type mtDNA molecules coexist (Mishra and Chan, 2014). For most of the mtDNA mutations described to date, no phenotypic alteration is detectable unless mutant mtDNA molecules exceed 60% of the total mtDNA (Gilkerson et al., 2008; Schon and Gilkerson, 2010). Considering that accumulation of mutations or deletions and loss of mtDNA are implicated in human diseases, a better understanding of the cellular processes involved in the maintenance of this molecule is of major importance.

Oxidative damage of mtDNA has been reported to be more extensive and persistent than that of nuclear DNA (Yakes and Van Houten, 1997; Richter et al., 1988). It has been proposed that the higher sensitivity of mtDNA to reactive oxygen species (ROS) could be due to the absence of histones in mitochondria, which facilitates the access of ROS to mtDNA. However, this idea is not supported by the tight packing of mtDNA around TFAM in nucleoids, resulting in a highly compacted structure. Accumulation of mtDNA damage could also be due to the proximity of the mtDNA to the continuous ROS production by the mitochondrial electron transport chain. Persistence of oxidative damage in mtDNA could lead to the accumulation of deletions or mutations, and, hence, mitochondrial dysfunction (Shigenaga et al., 1994). Indeed, mtDNA damage has been linked to loss of mitochondrial membrane potential, increased ROS generation and cell death (Santos et al., 2003).

One of the most frequent DNA alterations induced by ROS is 8-oxoguanine (8-oxoG), a product of oxidation of guanine. Several lines of evidence suggest that 8-oxoG accumulation in mtDNA contributes to the mitochondrial dysfunction observed during ageing and in neurodegenerative disorders (Druzhyna et al., 2009). Three enzymes are involved in the avoidance of the accumulation of 8-oxoG and the consequent mtDNA mutations: MTH1 (also known as NUDT1) detoxifies the pool of nucleotides by hydrolysing 8-oxo-2′-deoxyguanosine triphosphate to 8-oxo-2′-deoxyguanosine monophosphate; the DNA glycosylase OGG1 excises the 8-oxoG paired with cytosine in DNA; and MUTYH (also known as MYH), another DNA glycosylase, removes the adenine paired with 8-oxoG after replication, thus giving another opportunity to OGG1 to remove the 8-oxoG (Oka et al., 2014). These three enzymes are mostly known for their role in the nucleus, but they are also found in mitochondria and their deficiency causes mtDNA loss and mitochondrial dysfunction. Indeed, Ogg1-knockout mice accumulate high levels of 8-oxoG in mitochondria (de Souza-Pinto et al., 2001). It has also been shown that MTH1 and OGG1 play essential roles in the protection of mtDNA during neurogenesis, and that Mth1/Ogg1 double-knockout mice accumulate 8-oxoG and display reduced mitochondrial membrane potential (Leon et al., 2016). Moreover, exposure of OGG1-deficient cells to oxidative stress results in mitochondrial dysfunction, with reduced mtDNA levels, decreased mitochondrial membrane potential, increased mitochondrial fragmentation and reduced levels of proteins encoded by mtDNA, possibly linked to accumulation of oxidative mtDNA damage (Wang et al., 2011).

OGG1 initiates the BER pathway by specifically recognising and excising 8-oxoG. In mitochondria, OGG1 DNA glycosylase activity is thought to be followed by that of other BER proteins: apurinic/apyrimidinic endonuclease 1 (APE1; also known as APEX1), DNA polymerase γ (Pol-γ) and DNA ligase 3 (LIG3). All these BER proteins are encoded by the nuclear genome and are imported into the mitochondria (Kazak et al., 2012). Pol-γ and LIG3 are not only involved in BER but are also essential for mtDNA replication, and defects in these proteins result in mtDNA instability leading to dramatic mitochondrial and cellular dysfunction (Simsek et al., 2011; Trifunovic et al., 2004; Zhang et al., 2011). While for many years, Pol-γ was considered to be the only polymerase present inside mitochondria, recent reports have shown the presence of DNA polymerase β (Pol-β) inside the organelle (Sykora et al., 2017). These findings indicate that, from the recognition of the lesion to the ligation step, the very same BER enzymes could be acting in both the nucleus and the mitochondria to ensure the repair of oxidative DNA damage.

According to the information present in the National Center for Biotechnology Information (NCBI), several different OGG1 mRNAs (and potentially proteins) can be generated from the OGG1 gene by alternative splicing in human cells (Takao et al., 1998; Nishioka et al., 1999; Ogawa et al., 2015; Furihata, 2015). Experimental evidence is still lacking concerning which OGG1 isoforms are indeed produced in the cell. All potential isoforms (OGG1-1a, -1b, -1c, -2a, -2b, -2c, -2d and -2e) share the same N-terminal domain containing a mitochondrial-targeting sequence (MTS). OGG1-1a (hereafter referred to as α-OGG1) is the only isoform to have also a nuclear localisation signal (NLS) and has therefore been defined as the nuclear isoform of OGG1. Which of the OGG1 isoforms is responsible for the repair of 8-oxoG in mtDNA is still unclear. Despite the very high conservation of the DNA glycosylases between mouse and human, to date, only the isoform corresponding to the human α-OGG1 has been identified in mouse. This isoform has been proposed to ensure the 8-oxoG glycosylase activity in both nucleus and mitochondria. The artificial targeting of human α-OGG1 to mitochondria, by fusing it to the MTS of MnSOD (also known as SOD2), allows the complementation of mitochondrial dysfunction observed in OGG1-deficient cells, suggesting that this isoform can also be active in mitochondria to remove of 8-oxoG efficiently (Dobson et al., 2000; Rachek et al., 2002).

In this study, by performing a combination of biochemical and imaging techniques, we show that α-OGG1 is imported into mitochondria and that both the MTS and the NLS are functional and sufficient to determine the subcellular localisation of the enzyme. In agreement with the function of α-OGG1 in DNA repair, we describe that α-OGG associates with mtDNA in mitochondrial nucleoids independently of the cell cycle phase and the mtDNA replication status. Furthermore, our results show that the presence and DNA glycosylase activity of α-OGG1 inside mitochondria are essential to preserve mitochondrial function in human cells exposed to oxidative stress.

Use of a monoclonal antibody that specifically recognises an epitope only present in the α-OGG1 isoform revealed that this isoform was present in both nuclear and mitochondrial protein fractions from U2OS cells (Fig. 1A). Likewise, α-OGG1 tagged with the FLAG epitope (α-OGG1–FLAG) was produced in the same cells and an anti-FLAG antibody was used to monitor the subcellular localisation of the fusion protein. As for the endogenous protein, α-OGG1–FLAG was present in both the nuclear and mitochondrial fractions (Fig. 1B). Cells expressing α-OGG1–FLAG were stained with MitoTracker Red prior to fixation and immunostained with an anti-FLAG antibody. α-OGG1–FLAG was present in both the nucleus and the mitochondria (Fig. 1C).

To determine the sub-mitochondrial localisation of α-OGG1, we performed protease digestion assays on isolated mitochondria. Mitochondria from U2OS cells were treated with increasing amounts of Proteinase K to degrade mitochondrial proteins facing the cytoplasm. While the cytoplasmic region of TOM22 was completely digested, the region embedded inside the outer mitochondrial membrane was protected from degradation as were proteins such as TFAM and MPPB, which are localised inside the mitochondrial matrix. α-OGG1 was protected from Proteinase K digestion, indicating that the protein is localised inside the mitochondria (Fig. 1D). To better define the localisation of α-OGG1, sucrose gradient-purified mitochondria from HEK-293T cells were treated with either hypotonic buffer or digitonin to disrupt the outer mitochondrial membrane and allow trypsin to degrade proteins from both the outer membrane and the intermembrane space. TIM23 is localised in the inner mitochondrial membrane with a domain protruding into the intermembrane space that is cleaved when the outer membrane is disrupted with hypotonic or digitonin treatments. Proteins localised in the mitochondrial matrix, such as ETFB, and the inner membrane facing the matrix, such as SDHA, are not digested by trypsin following hypotonic or digitonin treatments. α-OGG1 was protected from trypsin even after disruption of the outer membrane, indicating that the protein was either localised in the inner mitochondrial membrane or in the mitochondrial matrix (Fig. 1E). Finally, association of α-OGG1 with mitochondrial membranes was studied by treating purified mitochondria with low or high NaCl concentrations. α-OGG1 remained insoluble even at high NaCl concentrations, indicating its tight association with the insoluble membrane fraction (Fig. 1F). Altogether, these results show that α-OGG1 is located both in the nucleus and in the mitochondria, where it is tightly associated with the inner mitochondrial membrane.

Confocal microscopy showed that α-OGG1–FLAG was not homogeneously distributed inside mitochondria (Fig. 1C). While MPPB, a matrix-soluble protein, showed a homogeneous staining, discrete spots were observed for the FLAG staining, resembling mitochondrial nucleoids (Fig. S1A). To determine whether α-OGG1 was actually a nucleoid-associated protein, colocalisation of α-OGG1–FLAG with nucleoid markers was investigated. Colocalisation of α-OGG1 with TFAM and mtDNA was observed (Fig. 2A,B; Fig. S1), with Pearson correlation coefficients of 0.8 and 0.75, respectively. The spatial organisation of α-OGG1 and TFAM in mitochondrial nucleoids was further explored by using 3D structured illumination microscopy (3D-SIM). This approach revealed that α-OGG1 clearly surrounds TFAM, a marker of the nucleoid core (Fig. 2C,D; Fig. S1C, Movie 1). The association of α-OGG1 with mitochondrial nucleoids was further confirmed by fractionation in an iodixanol gradient of detergent-solubilised mitochondria from HEK-293T cells. Southern and western blot analysis of the gradient fractions revealed that a significant amount of α-OGG1 co-fractionated with mtDNA as was the case for that the mtDNA-binding protein TFAM (Fig. 2E, fractions 9–11). The soluble glycyl-tRNA synthetase (GARS), which does not bind mtDNA and is localised in the mitochondrial matrix, was excluded from those fractions. Collectively, these results indicate that α-OGG1 associates with mtDNA in the nucleoids.

In agreement with the presence of α-OGG1 in both nuclear and mitochondrial compartments, several algorithms for the prediction of targeting signals gave a very strong probability for α-OGG1 to be imported into mitochondria and identified a canonical MTS at the N-terminus of the protein. Interestingly, a second in-frame AUG codon (AUG2) is present at the N-terminal part of the OGG1 messenger and is well conserved among mammals (Fig. 3A; Fig. S2). The first AUG codon (AUG1) has a guanine in position −3 in some species but is still fairly leaky. A stronger Kozak sequence was found for AUG2, due to an adenine in −3 position and a guanine in +1 position. Intriguingly, an initiation of translation at AUG2 would result in a protein with a truncated MTS and a reduced probability of being localised inside mitochondria (Fig. 3B). Additionally, a TOM20 recognition motif (RKYF) in positions 97–100 of α-OGG1 suggests that α-OGG1 can be imported into the mitochondria by the TIM/TOM pathway (Schulz et al., 2015; Model et al., 2008).

To define the functionality of the MTS and NLS, we expressed full-length α-OGG1 (wild-type, WT) or truncated versions, lacking either the MTS or the NLS, fused to the FLAG epitope and analysed their subcellular localisation (Fig. 3C). While, as mentioned above, full-length α-OGG1 was present in both mitochondrial and nuclear compartments, disruption of the MTS (ΔMTS) due to the initiation of translation at AUG2 resulted in a protein localised exclusively in the nucleus. Interestingly, substitution of the glycine in position 12 for a glutamic acid (G12E), a mutation identified in kidney cancer cells (Audebert et al., 2002), also impaired the mitochondrial import of the protein. Deletion of the NLS (ΔNLS) resulted in a protein mostly localised in the mitochondria (Fig. 3D–F). Taken together, these results indicate that there is both a MTS and NLS in α-OGG1 and that both signals are functional and can determine the subcellular localisation of the enzyme.

As mitochondrial import has been shown to be modulated through the cell cycle (Harbauer et al., 2014), we decided to explore whether the subcellular distribution of α-OGG1 was cell-cycle dependent. To identify cells in S and G2 phases, incorporation of the thymidine analogue EdU and staining of an antibody against the kinetochore protein CENPF were used, respectively. Cells in G1 are negative for both EdU and CENPF. As shown in Fig. 4A, wild-type α-OGG1 [α-OGG1(WT)–FLAG] was detected in mitochondria independently of the cell cycle phase, and the proportion of cells with mitochondrial staining remained unchanged throughout the cell cycle. We further investigated whether the mtDNA replication status influenced the sub-organelle localisation of α-OGG1, as is the case for several proteins involved in mtDNA maintenance (Rajala et al., 2014). EdU incorporation was used to visualise mitochondrial nucleoids undergoing replication on transiently double-transfected cells expressing α-OGG1(WT)-FLAG and MTS–Turquoise2. An antibody against TFAM was used to label mitochondrial nucleoids (Fig. 4C,D). More than 500 nucleoids were analysed to determine their replication state, and the presence or absence of α-OGG1(WT)–FLAG. As can be seen in Fig. 4C, α-OGG1–FLAG was detected in both EdU-positive and -negative nucleoids, clearly indicating that association of the glycosylase with nucleoids is independent of DNA replication. Altogether, these results show that the mitochondrial import of α-OGG1 and its association with mtDNA is independent of the cell cycle phase and mtDNA replication status.

Considering that the only known activity for OGG1 is the repair of 8-oxoG, we next studied the impact of OGG1 deficiency on the response of mitochondria to menadione-induced oxidative stress, which results in the induction of 8-oxoG in mtDNA (Oka et al., 2008). Transfection of U2OS cells with OGG1-directed siRNA resulted in more than a 90% decrease in both transcript and protein levels of α-OGG1 72 h after transfection (Fig. 5A). Cells transfected with siRNA against OGG1 or control siRNA were exposed to menadione, and mitochondrial parameters were evaluated by flow cytometry, allowing the analysis of a large number of cells within a short time. The specificity of the MitoSOX probe for the evaluation of mitochondrial ROS production was optimised and validated by confocal microscopy (Fig. S3). After menadione treatment, we found a larger increase in mitochondrial superoxide production in OGG1-deficient cells than in controls (Fig. 5B). In addition, a TMRE probe for mitochondrial membrane potential revealed that menadione treatment induced a more pronounced loss of mitochondrial membrane potential in OGG1-silenced cells, while no difference was observed between the two cell populations under basal conditions (Fig. 5C). The loss of mitochondrial activity observed was not due to a loss of mitochondrial mass, as MitoTracker Green staining was not affected by the treatment (Fig. 5C, right panels).

To monitor single-cell responses over time, we performed real-time microscopy. Cells transfected with either siRNA against OGG1 or control siRNA were treated with 50 µM of menadione for 1 h, washed and imaged for 2 h at a frame rate of one image every 2 min. Most of the mitochondria from cells transfected with control siRNA showed a stable TMRE staining over time after menadione treatment, indicating the maintenance of a positive membrane potential. In contrast, ∼40% of OGG1-deficient cells showed a progressive and fast loss of membrane potential, leading to the inactivation of the entire mitochondrial network (Fig. 5D). These results revealed that human OGG1-deficient cells show mitochondrial dysfunction when exposed to oxidative stress, with a strong production of mitochondrial superoxide and loss of mitochondrial membrane potential.

To determine whether α-OGG1 expression can rescue the mitochondrial defects observed in OGG1-deficient cells, we generated U2OS cell lines stably expressing either α-OGG1(WT)–FLAG or the exclusively nuclear α-OGG1(ΔMTS)–FLAG. The deletion of 10 amino acids at the N-terminus of OGG1 did not affect the enzymatic activity of the enzyme; expression of α-OGG1(ΔMTS)–FLAG resulted in the same level of 8-oxoG DNA glycosylase activity in total cell extracts as that in extracts from cells expressing the full-length protein (Fig. 6A). In order to specifically reduce the levels of endogenous OGG1, without affecting the expression of the exogenously expressed isoforms, silent mutations were introduced in the sequences of the latter in order to make them resistant to the siRNA against OGG1. Fig. 6B shows that the levels of α-OGG1(WT)–FLAG and α-OGG1(ΔMTS)–FLAG were unaffected by transfection with siRNA against OGG1, while the endogenous OGG1 level was strongly reduced (Fig. 6B, first two lanes).

WT U2OS cells or cells stably overexpressing the different isoforms of the protein were therefore transfected with siRNA against OGG1. At 3 days after transfection, the cells were stained with TMRE or MitoSOX, treated with menadione and analysed. Oxidative stress induced a significantly higher mitochondrial ROS production along with stronger loss of mitochondrial membrane potential in OGG1-deficient cells than in cells transfected with the control siRNA. Both parameters were rescued by expression of α-OGG1(WT)–FLAG but not by expression of the mutant isoform (Fig. 6C,D), indicating that the presence of the α-OGG1 DNA glycosylase in the mitochondria is required for the recovery of the organelle from oxidative stress.

Prominent fragmentation of the mitochondrial network was observed in U2OS cells after exposure to menadione. This fragmentation was characterised by a rapid transition from a filamentous and branched network to a fragmented one, with isolated round-shaped mitochondria (Fig. 7A). To evaluate whether overexpression of α-OGG1 could suppress the mitochondrial fragmentation induced by the menadione treatment, U2OS cells were transiently transfected with constructs expressing α-OGG1(WT)–FLAG, α-OGG1(ΔMTS)–FLAG or the catalytically inactive mutant α-OGG1(K249Q)–FLAG (Fig. 7A). Mutation K249Q results in a protein with no enzymatic activity (Van der Kemp et al., 2004). This mutation does not induce a change in the structure of the protein (Bruner et al., 2000) or in its subcellular localisation (Fig. S4). The percentage of cells with a preserved mitochondrial network was quantified under basal conditions and after exposure to oxidative stress (Fig. 7B). Interestingly, overexpression of the α-OGG1(WT)–FLAG protein, but not that of the mutant forms, resulted in a normal mitochondrial network in cells exposed to menadione. Further analysis of mitochondrial morphology parameters on binarised images confirmed that, after menadione treatment, mitochondrial aspect ratio and degree of branching were comparable in non-transfected and α-OGG1(ΔMTS)–FLAG or α-OGG1(K249Q)–FLAG overexpressing cells, while in α-OGG1(WT)–FLAG overexpressing cells the network was better preserved (Fig. 7C,D).

Taken together, these results indicate that mitochondrial α-OGG1 is essential to preserve mitochondrial morphology and function after exogenous oxidative stress. In addition, because the catalytically inactive mutant α-OGG1(K249Q)–FLAG displays the same subcellular localisation as that of the α-OGG1(WT)–FLAG enzyme and colocalises with mitochondrial nucleoids, our data strongly suggest that the DNA glycosylase activity of α-OGG1, and thus the removal of 8-oxoG, is required to protect mitochondrial physiology during oxidative stress.

Despite clear evidence indicating the presence of a functional BER pathway inside human mitochondria, which isoform of the DNA glycosylase OGG1 is responsible for the recognition and excision of 8-oxoG in the mtDNA remained unclear. Although it was proposed that these activities could be performed by OGG1-1b (β-OGG1) (Oka et al., 2008), the lack of DNA glycosylase activity in this isoform renders its involvement in the repair of 8-oxoG unlikely (Hashiguchi et al., 2004). It was recently reported that in vitro purified OGG1 isoforms 1b and 1c have some enzymatic activity and could be responsible for the removal of 8-oxoG from the mitochondrial genome (Furihata, 2015). However, there is no in vivo evidence for the existence of these OGG1 variants. It is not well known which transcripts are generated in different tissues and/or under different physiological conditions and no experimental evidence determining which protein forms are produced is available. In mice, only a single transcript corresponding to the isoform α-OGG1 (OGG1-1a) has been identified and the protein has been shown to be functional in both nuclear and mitochondrial compartments (de Souza-Pinto et al., 2001). It is still unclear why so many OGG1 transcripts are found in human cells. Because several of the potential polypeptides do not show any DNA glycosylase in vitro, they could have other, as yet unidentified, roles or they might simply not be translated. Indeed, several global analyses comparing transcriptome and proteome data suggest that a large fraction of the reported mRNAs are never translated into proteins. Proteomic studies suggest that the vast majority of genes have a single dominant splice isoform that is the most highly conserved between different species (Tress et al., 2017). These considerations would be in agreement with α-OGG1, the only OGG1 isoform conserved between mouse and human, being the 8-oxoG DNA glycosylase acting both in the nucleus and in the mitochondria in both species.

Previous studies have suggested that the MTS in α-OGG1 is not sufficient to target the protein to the mitochondria because of the presence of the very strong NLS at the C-terminus of the DNA glycosylase (Nishioka et al., 1999), despite the fact that the mouse protein, with a similarity close to 94% with α-OGG1, has been found in both the nucleus and the mitochondria. Artificial targeting of human α-OGG1 to mitochondria by addition of the MnSOD MTS showed that the human protein can be active in this organelle and improve mtDNA repair and survival in cells exposed to oxidative stress (Dobson et al., 2000; Rachek et al., 2002; Druzhyna et al., 2005; Kim et al., 2014). However, these experiments did not address the question of whether the endogenous α-OGG1 protein, without the additional targeting sequence, is imported inside mitochondria. We showed here that the isoform α-OGG1 is efficiently imported into mitochondria and that the MTS at the N-terminal part of the protein is functional and sufficient to ensure the targeting of the enzyme to the organelle. Consistent with this, several algorithms predicted a canonical MTS for α-OGG1 and a mitochondrial localisation of the protein with a very high probability, close to 90% (Fig. 3). In fact, the strength of the α-OGG1 MTS is not very different from that of the MnSOD MTS used in the chimeric MTS–OGG1 construct (Rachek et al., 2002), and our study unambiguously showed that the addition of an ectopic MTS is not required to ensure the efficient import of α-OGG1 into the mitochondria. The presence of dual-targeting signals found in α-OGG1 is clearly not an exception, as many proteins with dual functional signals for nuclear and mitochondrial localisation have been identified (Yogev and Pines, 2011; Kazak et al., 2013).

Whether and how the subcellular localisation of α-OGG1 is regulated remains an open question. We did not observe any significant change in the subcellular distribution of α-OGG1 throughout the cell cycle or as a consequence of exposure to oxidative stress. While most of the BER proteins are present in both the nucleus and the mitochondria, very little is known concerning the molecular mechanisms involved in the regulation of their subcellular localisation. The subcellular distribution of the Saccharomyces cerevisiae DNA glycosylase Ntg1 is modulated by oxidative stress and involves sumoylation of the protein (Griffiths et al., 2009; Swartzlander et al., 2010). The Ntg1-sumoylated sites have been recently identified, and two of them are conserved in the human orthologue NTH1, suggesting a similar regulation in higher eukaryotes (Swartzlander et al., 2016), although this possibility has not been further explored.

It is interesting to note that two in-frame AUG codons are present in the N-terminal part of OGG1 mRNA and that they are largely conserved among mammals (Fig. 3; Fig. S2). As suggested by the prediction algorithms for mitochondrial targeting, our results clearly show that if translation is initiated at AUG2, thus truncating the MTS, the resulting protein is exclusively localised in the nucleus. Alternative start AUG codons within a single transcript largely contribute to the diversity of the proteome (Van Damme et al., 2014) and, in particular, to the mitochondrial proteome (Kazak et al., 2013). Although our fractionation experiments did not reveal a significant shift in size between the nuclear and mitochondrial OGG1, we cannot exclude that AUG2 could be used under particular conditions to favour nuclear import. OGG1 shares two characteristics with other proteins using several AUGs in frame: (1) AUG1 has a suboptimal context for translation initiation, thus facilitating ribosomal sliding (leaky mRNA scanning) that could originate different polypeptides from the same transcript; and (2) the 5′UTR is particularly long (350 nt compared to a normal length of ∼200 nt) (Bazykin and Kochetov, 2011).

Interestingly, the three proteins dedicated to counteract the impact of the incorporation of 8-oxoG in DNA – OGG1, MTH1 and MUTYH (David et al., 2007) – share some characteristics. As for OGG1, several transcripts that originated by alternative splicing have been identified for human MTH1 and MUTYH (Oda et al., 1999; Ohtsubo et al., 2000), and some of them have two or three putative initiation codons, which are functional at least in vitro. In addition, as for OGG1, initiation of translation at the first AUG in their mRNAs gives rise to a protein containing a MTS, while initiation at the second AUG results in the loss of the MTS and yields a protein localized exclusively in the nucleus. Considering that all those proteins that are essential for the protection of mtDNA against the deleterious effects of 8-oxoG have evolved from very different families of proteins and do not have any sequence homology, it is striking to observe the very similar organization of their N-terminal sequences with the presence of several AUGs in frame. Although the molecular details have not been elucidated, alternative translation initiation has already been proposed to be the mechanism of choice for the generation of nuclear and mitochondrial isoforms of LIG3 (Lakshmipathy and Campbell, 1999). It is tempting to speculate that evolution has converged to this strategy in order to modulate the subcellular distribution of these enzymes. Further studies are required to better understand how the subcellular localisation of the DNA glycosylase α-OGG1 and other BER proteins is modulated.

Our microscopy and biochemical observations show, for the first time, the association of α-OGG1 with the mtDNA in nucleoids. The use of 3D-SIM provided a detailed view of this association and demonstrated the close proximity between the DNA glycosylase and TFAM inside the mitochondrial network. While several proteins involved in DNA metabolism, such as Pol-γ, mtSSB and Twinkle, have been found to associate with mtDNA only during replication in a highly regulated way (Rajala et al., 2014; Lewis et al., 2016), our experiments showed that the association of α-OGG1 with the nucleoid is not linked to the replication state of the mtDNA but is in fact constitutive. Considering the highly compacted structure of mtDNA in nucleoids and the fact that the binding of TFAM to DNA containing 8-oxoG inhibits OGG1 activity in vitro (Canugovi et al., 2010), other proteins may be required to facilitate access of the DNA repair proteins to the mtDNA. A role for CSB in remodelling nucleoids by removing TFAM from DNA and facilitating the access of DNA repair machineries has been proposed (Berquist et al., 2012). This could explain the increase in 8-oxoG levels in mtDNA detected in Csb^−/− mice (Stevnsner et al., 2002; Osenbroch et al., 2009). α-OGG1 is strongly associated with a mitochondrial-insoluble fraction, probably the inner mitochondrial membrane, as it resists extraction with up to 500 mM of NaCl and is protected from protease digestion coupled to an osmotic shock. This is in agreement with previous studies showing the association of BER proteins and/or activities (DNA glycosylase, polymerase and ligase and, to a minor extent, the endonuclease) with an insoluble mitochondrial fraction (Stuart et al., 2005; Boesch et al., 2010). As the mitochondrial nucleoids are anchored to the inner mitochondrial membrane (Jakobs and Wurm, 2014; Gilkerson et al., 2013), this distribution of BER proteins could facilitate the efficiency of the DNA repair process.

In our study, and in agreement with previous observations, a human cell line deficient in all OGG1 isoforms displayed a larger loss of mitochondrial membrane potential and an accumulation of mitochondrial ROS, both indicators of mitochondrial dysfunction, in response to a menadione treatment. The fact that re-expression of the α-OGG1(WT) isoform in these cells fully complemented these phenotypes (Fig. 6) suggests that this isoform is perfectly functional. The failure of the α-OGG1(ΔMTS) protein, showing an exclusively nuclear localisation, to complement the mitochondrial dysfunction of OGG1-deficient cells indicates that the presence of the protein inside mitochondria is required to maintain mitochondrial function.

Under stress conditions, an increase in fission versus fusion results in fragmentation of the mitochondrial network, thus isolating the dysfunctional mitochondria that will be selectively removed by mitophagy. Furthermore, a perfect balance between fusion and fission is essential for the maintenance of mitochondrial health. A defect in either mitochondrial fusion or fission processes results in mtDNA instability (Busch et al., 2014; Garcia et al., 2017). The overexpression of α-OGG1(WT) in human cells helps to preserve the mitochondrial network under oxidative stress conditions, as has also been observed in previous studies concerning the mouse OGG1 (Torres-Gonzalez et al., 2014), and is in agreement with the general idea that DNA glycosylase activity is the rate-limiting step in the BER pathway (Hollenbach et al., 1999). The inability of the mutant protein α-OGG1(ΔMTS) to protect the mitochondrial network clearly indicates that the presence of the protein inside mitochondria is essential and it is not an indirect consequence of a better repair of nuclear DNA. Considering the important effect of the overexpression of α-OGG1(WT) in protecting mitochondrial morphology, we cannot rule out that α-OGG1 has another role, independent of DNA repair, in mitochondrial fusion or fission. Indeed, the levels of mitochondrial proteins involved in fission, DRP and FIS1, have been reported to be reduced in cells overexpressing the mouse OGG1 (Torres-Gonzalez et al., 2014), suggesting that the decrease in fission accounts for the reduction in mitochondrial fragmentation observed. However, as fragmentation occurs as a consequence of mitochondrial dysfunction and is dependent on the loss of membrane potential, this might simply reflect the better health of mitochondria in cells overexpressing OGG1. Consistent with the latter option, while overexpression of α-OGG1(WT) protected the mitochondrial network from fragmentation after exposure to menadione, the active site mutant α-OGG1(K249Q) failed to do so. This is in agreement with the results obtained with the OGG1 mutant R229Q. This mutation was initially identified in a leukaemic cell line and results in a loss of enzymatic activity due to protein destabilisation (Hyun et al., 2000; Hill and Evans, 2007). Targeting of the human mutant protein MTS-OGG1-R229Q to the mitochondria results in decreased mtDNA integrity and cellular survival after exposure to oxidative agents when compared to the wild-type MTS-OGG1 (Chatterjee et al., 2006). The fact that catalytically inactive α-OGG1 mutants could not preserve the mitochondrial morphology in cells exposed to oxidative stress makes us favour the hypothesis that the observed mitochondrial dysfunction is the consequence of accumulation of 8-oxoG in the mitochondrial genome.

The presence of 8-oxoG has been proposed to be an important cause of mtDNA mutations and deletions, which accumulate with ageing and during disease progression (Druzhyna et al., 2009). It is well accepted that the accumulation of 8-oxoG in DNA results in an increase in mutagenesis due to the misincorporation of the adenine opposite 8-oxoG during DNA replication. An increase in mtDNA point mutations has been shown in yeast strains deficient for OGG1 (Singh et al., 2001). However, Halsne et al. reported that, in OGG1^−/− mutant or in Ogg1^−/− Mutyh^−/− double-mutant mice, mtDNA mutation rates were not different from those in wild-type animals (Halsne et al., 2012). The experiments presented here showed that menadione treatment rapidly induces mitochondrial dysfunction in OGG1-deficient cells, with ∼40% of the cells showing a massive loss of membrane potential in less than 30 min. These rapid effects could hardly be explained by an increase in mutagenesis, especially taking into account the high number of copies of mtDNA and the high degree of heteroplasmy and complementation between different mtDNA molecules. So, how can the failure to remove 8-oxoG affect mitochondrial function? It has been proposed that Pol-γ is blocked or paused by 8-oxoG (Graziewicz et al., 2007; Stojkovič et al., 2016), leading to the loss of mtDNA instead of mutagenesis. Although other replisome-associated proteins, such as Twinkle or Primpol, have been suggested to help the polymerase bypass 8-oxoG (Garcia-Gomez et al., 2013; Stojkovič et al., 2016), other studies are required to confirm the role of these proteins in processing damaged mtDNA. Several reports have shown that oxidative stress induces the degradation of mtDNA and mtRNA (Crawford et al., 1997, 1998; Abramova et al., 2000; Shokolenko et al., 2009; Rothfuss et al., 2009; Furda et al., 2012). It is interesting to note that while low levels of 8-oxoG have been detected in the circular mtDNA molecule, fragmented mtDNA has a very high 8-oxo-2′-deoxyguanosine content, which is further increased after oxidative stress (Suter and Richter, 1999). Thus, whether the accumulation of 8-oxoG in mtDNA in the absence of the DNA glycosylase activity of α-OGG1 represents a direct block for transcription or replication or whether it induces degradation of DNA molecules remains to be further explored.

U2OS and HEK-293T cells (purchased from ATCC and Life Technologies, respectively) were cultured at 37°C in Dulbecco's modified Eagle's medium (DMEM, Sigma D5671) supplemented with 10% fetal bovine serum (FBS) (Sigma, F7524) and 5% penicillin-streptomycin (Gibco 15140122) in 5% CO₂.

pmTurquoise2-Mito was from Addgene (plasmid # 36208, deposited by Dorus Gadella) (Goedhart et al., 2012). α-OGG1 fusions to the FLAG epitope were generated by PCR by including the sequence coding for the FLAG in the reverse oligonucleotide, and cloned in the plasmid pCDNA3.1(−). The open reading frame of α-OGG1 was amplified by PCR to generate the fusion proteins OGG1(WT)–FLAG, OGG1(ΔMTS)–FLAG and OGG1(ΔNLS)–FLAG. Point mutations K249Q and G12E were introduced by site-directed mutagenesis from the human OGG1(WT)–FLAG construct, using the QuikChange site-directed mutagenesis kit (Stratagene). To facilitate the generation of stable cell lines, the same open reading frames were cloned in the polycistronic pIRES2-AcGFP1 plasmid (Clontech) expressing GFP in the second position of the cistron, thus allowing selection of the transfected cells expressing similar levels of GFP by cell sorting. The oligonucleotides used to generate the different constructs are listed in Table S1.

Cells were grown on coverslips or on Ibidi µ-Slide 4- or 8-well plates for microscopy experiments and on T150 flasks for biochemical analysis. Transient and stable transfections were performed 24–48 h after seeding, using Lipofectamine 2000 (Invitrogen) and following a standard protocol with a DNA:lipofectamine ratio of 1 µg:4 µl. For stable cell lines expressing OGG1(WT)–FLAG or OGG1(ΔMTS)–FLAG from pIRES plasmids, the culture medium was supplemented with 400 μg/ml of G418.

For the siRNA-mediated depletion of hOGG1, U2OS cells were transfected with siRNA against human OGG1 (5′-GGAUCAAGUAUGGACACUG-3′) using Lipofectamine RNAiMAX (Invitrogen). AllStars Negative Control siRNA (Qiagen) was used as control.

For the induction of oxidative stress, cells were exposed to 50 µM menadione for 1 h in DMEM without FBS and antibiotics. All dilutions were done in pre-warmed DMEM immediately before use.

The software programs used for the prediction of MTSs were as follows: MitoProt (http://ihg.gsf.de/ihg/mitoprot.html) (Claros and Vincens, 1996), Predotar (https://omictools.com/predotar-tool) (Small et al., 2004), PSORT II (https://psort.hgc.jp/form2.html) and TargetP (http://www.cbs.dtu.dk/services/TargetP/) (Emanuelsson et al., 2000).

For staining of the mitochondrial network, cells were rinsed twice with DMEM and incubated with 200 nM of MitoTracker Red CMXRos and/or MitoTracker Deep Red (ThermoFisher Scientific) diluted in complete DMEM for 30 min at 37°C. Where indicated, DNA was stained with PicoGreen at 1:500 (Quant-iT™ PicoGreen™ dsDNA Reagent, ThermoFisher Scientific). Cells were fixed with 2% formaldehyde for 20 min at 37°C, rinsed with PBS and permeabilised at room temperature in PBS containing 0.1% Triton X-100 for 5 min. Cells were then incubated in blocking solution (PBS containing 0.1% Triton X-100, 3% BSA and 1% normal goat serum) at 37°C for 1 h or overnight at 4°C. Cells were incubated for 1–2 h at 37°C with primary antibody diluted in blocking solution, washed three times for 5 min in PBS containing 0.1% Triton and further incubated with secondary Alexa Fluor-conjugated antibodies (Molecular Probes) diluted in blocking solution for 45 min at 37°C. Where indicated, nuclear DNA was counterstained with 1 µg/ml 4′,6′-diamidino-2-phenylindole (DAPI). Primary antibodies against the following proteins were used: FLAG (1:2000; Sigma, F3165), TFAM (1:2000; Abcam, ab155240) and CENPF (1:500; Abcam, Ab5). 5-ethynyl-2′-deoxyuridine (EdU) was added to the medium at a concentration of 1 mM to label replicating cells and stained following the manufacturer's instructions (ThermoFisher Click-iT® EdU Alexa Fluor® 647).

Image acquisition was performed using a Leica SP8 confocal microscope with a 60× oil immersion objective or using a Nikon A1 confocal microscope with a 63× oil immersion objective (1.3 NA for both). Image analysis was performed with ImageJ (Schneider et al., 2012). Pearson correlation coefficients were calculated from eight cells (around 1000 nucleoids) using the plugin JACoP (Bolte and Cordelières, 2006). Mitochondrial morphology alterations were analysed by applying the Kernel algorithm (Koopman et al., 2005) to binarised images. This approach allows measurement of morphological parameters for individual mitochondria, such as the aspect ratio, a measure of mitochondrial length, and the form factor, which is a combined measure of mitochondrial length and degree of branching.

For the quantification of the mitochondrial localisation of OGG1, 3D stacks were acquired with a Nikon A1 confocal microscope. The thresholded MitoTracker Red signal was used to generate a mask corresponding to the mitochondrial network. The perinuclear region was removed from the analysis to avoid any contamination from the nuclear signal. The mitochondrial mask was applied to the OGG1–FLAG channel and the fluorescence intensity was quantified. The mean fluorescence was calculated for each individual transfected cell and plotted. The background corresponding to the signal detected in non-transfected cells was evaluated for each image and subtracted.

3D structured illumination microscopy (3D-SIM) was performed using a rotary-stage OMX v3 system (Applied Precision, GE Healthcare), equipped with 3 Evolve EMCCD cameras (Photometrics). Signals from all channels were realigned using fluorescent beads before each session of image acquisition. All images were acquired with a PlanApo 100×/1.4 NA oil objective at 125 nm intervals along the z-axis. The resolution on 100 nm beads was found to be 125, 125 and 260 nm in the x-, y- and z-axis, respectively. Channel alignment was performed with ImageJ using the UnwarpJ plugin (Sorzano, et al., 2005).

Whole-cell extracts were obtained from 10⁶ U2OS cells. Cell pellets were resuspended in TP lysis buffer (20 mM Tris-HCl pH 7.5, 250 mM NaCl, 1 mM EDTA, protease inhibitors) and sonicated in a Bioruptor^® bath (30 s on, 30 s off pulses for 10 min at maximum intensity). After sonication, the samples were centrifuged at 20,000 g for 30 min at 4°C, the pellets were discarded and the supernatants were kept on ice.

Nuclear protein extracts were obtained from 10⁶ U2OS cells with the nuclei isolation kit Nuclei EZ Prep (Sigma), according to the manufacturer's instructions. Crude mitochondria extracts were prepared from 1.2×10⁸ U2OS cells, as described by Wieckowski et al. (2009) with a few modifications. Briefly, cells at 90–95% confluence were detached by trypsinisation and collected by centrifugation at 600 g for 7 min at 4°C. After one wash with ice-cold PBS, the cells were resuspended in ice-cold isolation buffer 1 (225 mM mannitol, 75 mM sucrose, 0.1 mM EGTA, 30 mM Tris-HCl pH 7.5) and disrupted with 15–30 strokes with a glass dounce homogeniser on ice. The integrity of the cells was checked under the microscope. The homogenate was centrifuged at 600 g for 7 min at 4°C. The supernatant was transferred into a pre-cooled ultracentrifuge vial and centrifuged at 7000 g for 12 min at 4°C. The supernatant was discarded and the pellet containing the mitochondria was gently resuspended in ice-cold isolation buffer 2 (225 mM mannitol, 75 mM sucrose, 30 mM Tris-HCl pH 7.5) and centrifuged again at 7000 g for 12 min at 4°C. The mitochondrial pellet was resuspended in isolation buffer 2 and re-pelleted at 10,000 g for 12 min at 4°C. The crude mitochondrial pellet was resuspended in 120 µl of ice-cold mitochondrial resuspension buffer (MBR; 250 mM mannitol, 0.5 mM EGTA, and 5 mM HEPES pH 7.4).

Protein contents of the different extracts were quantified by means of the Bradford assay.

Mitochondria isolated from HEK-293T cells by differential and sucrose-gradient centrifugation were treated with trypsin prior to lysis and fractionation on an iodixanol gradient, as previously reported (Reyes et al., 2011). Briefly, cells were incubated in hypotonic buffer (20 mM HEPES pH 7.8, 5 mM KCl, 1.5 mM MgCl₂) on ice for 10 min and then homogenised with a tight-fitting homogeniser. Isotonic level was restored by the addition of 2.5× MSH (210 mM mannitol, 70 mM sucrose, 20 mM HEPES pH 7.8, 2 mM EDTA pH 8.0). Nuclei and cell debris were spun down at low speed. The supernatant was subjected to high-speed centrifugation to pellet crude mitochondria. Further purification of mitochondria was performed by loading crude mitochondria onto a discontinuous 1.5–1.0 M sucrose gradient and collecting purified mitochondria from the interface. The organelles were then lysed with 0.4% DDM in 1× MSH and loaded onto a 20–45% self-forming continuous iodixanol gradient. After centrifugation at 100,000 g for 14 h, 18 fractions of 450 µl were collected, and protein and DNA were extracted from them and analysed by Southern and western blotting, respectively. An mtDNA fragment amplified by PCR with specific oligonucleotides (Table S1) was incubated with DNA-labelling beads (GE Healthcare) in the presence of 50 mCi of [α-³²P]dCTP (3000 Ci/mmol, Perkin Elmer) for 30 min. Then, the Southern blot was hybridised to this mtDNA-specific probe by overnight incubation at 65°C in 7% SDS with 0.25 M sodium phosphate buffer (pH 7.4). Post-hybridisation washes were done with 1× SSC followed by 1× SSC with 0.1% SDS twice for 30 min at 65°C. Filters were exposed to phosphor screens and scanned using a Typhoon phosphorimager (GE Healthcare). The western blots were incubated with antibodies against TFAM, the soluble glycyl-tRNA synthetase (GARS) and α-OGG1 (Table S2).

A total of 25 µg of mitochondrial extracts were incubated for 5 min at 4°C in 1 ml of MBR buffer with 1 mg/ml of BSA, with a final concentration of 5 or 25 µg/ml of Proteinase K. Where indicated, a pre-treatment with 1% Triton- X100 for 5 min at 4°C was performed before incubation with Proteinase K. Proteinase K was inactivated on ice by adding 10 mM PMSF for 5 min. The samples were then centrifuged at 20,000 g for 10 min at 4°C and the supernatant was discarded. The resulting pellets were resuspended in 50 µl of MBR. Untreated mitochondrial extracts were used as control.

Mitochondria from HEK-293T cells (2 mg/ml) purified in sucrose gradients as described above were resuspended in either isotonic (20 mM HEPES pH 7.8, 2 mM EDTA, 210 mM mannitol, 70 mM sucrose) or hypotonic (10 mM HEPES pH 7.8) buffer. Where indicated, 100 µg/ml of trypsin was added and incubation was carried out at room temperature for 30 min. Purified mitochondria were alternatively treated with 200 or 500 mg/ml digitonin in isotonic buffer for 10 min at 4°C prior to trypsin treatment, as previously reported (Reyes et al., 2011), as an alternative method of outer mitochondrial membrane disruption and digestion of proteins from the outer membrane or the intermembrane space. After washing and pelleting mitochondria three times, the organelles were lysed with 1% SDS in isotonic buffer. TOM20, TIM23, SDHA and EFTB were used as markers for the outer mitochondrial membrane, intermembrane space, inner mitochondrial membrane and mitochondrial matrix, respectively.

Sucrose gradient-purified mitochondria from HEK-293T cells (2 mg/ml) were sonicated for 1 min on ice followed by centrifugation at 10,000 g for 10 min at 4°C. The supernatant was treated with 150 or 500 mM NaCl or 2% SDS on ice for 30 min followed by centrifugation at 122,000 g for 30 min at 4°C. Supernatants containing the soluble fractions were recovered and pellets with membrane-associated proteins were suspended in equal volumes of isotonic buffer containing 0.2% SDS, as previously described (Martinez Lyons et al., 2016). TOM20 and citrate synthase were used as controls of integral membrane and soluble protein, respectively.

Aliquots from each cell extract were denatured by heating at 95°C for 5 min. The same amount of protein content was loaded, and 10–20 µg of protein extracts were separated in 10% SDS-acrylamide gel and transferred onto a nitrocellulose membrane. Membranes were stained with Ponceau Red to confirm the amounts of protein in each lane and then blocked in 8% milk in PBS with 0.1% Tween-20 (TPBS) for 30 min at room temperature or overnight at 4°C. Membranes were then incubated for 90 min at room temperature with primary antibodies in 3% milk-PBS. All the antibodies used are listed in Table S2. After three 5 min washes with PBS, western blots were incubated for 45 min at room temperature with fluorescently labelled secondary antibodies: anti-rabbit-800 (ref 05060-250), anti-mouse-800 (05061-250), anti-rabbit-700 (05054-250) and/or anti-mouse-700 (050055-250) (Diagomics) diluted to 1:10000 in 2% milk-TPBS or HRP-conjugated secondary antibodies (Promega anti-rabbit W401B, anti-mouse W402B) diluted to 1:5000 and then washed three times with TPBS for 7–10 min. The analysis of images was performed by scanning the membrane at 800 and 700 nm simultaneously with an Odyssey^® CLx imaging system or by exposure to X-ray films.

Analysis of mitochondrial mass, mitochondrial membrane potential and intra-mitochondrial ROS production were performed by flow cytometry (BD LSRII) or by using a microplate reader (CLARIOstar - BMG Labtech).

For flow cytometry analysis, 4×10⁵ U2OS were seeded in a 6-well plate and transfected with siOGG1 or with siAllStar, as described above; 72 h after transfection, the cells were treated for 1 h at 37°C with 50 µM menadione. To evaluate intra-mitochondrial ROS, U2OS cells were incubated with 5 µM of MitoSOX^® Red mitochondrial superoxide indicator (ThermoFisher, ref. M36008) in PBS-BSA 0.4% for 10 min at 37°C protected from light, harvested with trypsin and directly analysed. For mitochondrial membrane potential analysis, cells were incubated with 100 nM of MitoTracker Green FM (ThermoFisher, ref. M7514) for 30 min at 37°C, rinsed once and stained with 100 nM of TMRE (tetramethylrhodamine, ethyl ester, perchlorate; ThermoFisher, ref. T669) diluted in PBS containing 0.4% BSA for 15 min at 37°C while cells were protected from light. Cells were then harvested with trypsin and directly analysed. For each sample, at least 50,000 cells were analysed. Experiments were performed three times with similar results.

For plate reader measurements, 10,000 cells were reverse transfected with siOGG1 or siAllStars in a 96-well microplate for fluorescence assays, which was black-walled and had a clear bottom (Greiner Bio-One, ref. 655090). The cells were treated with 25 or 50 µM of menadione for 1 h at 37°C and then rinsed once with complete DMEM. For the analysis of intra-mitochondrial ROS production after the chemical treatment, the cells were stained for 10 min maximum with MitoSOX^® Red, 5 µM and Hoechst 33342 diluted to 1:1000 in pre-warmed PBS (with Ca²⁺ and Mg²⁺). PBS was then replaced with 199 µl medium before acquiring measurements. Hoechst 33342 staining was used to normalise for cell number, and results are expressed as a ratio of MitoSOX to Hoechst. To evaluate a possible loss of mitochondrial mass or mitochondrial membrane potential, cells were stained before the chemical treatment with 100 nM of MitoTracker Green and 100 nM of TMRE for 30 min at 37°C protected from light.

A single-stranded 34-mer DNA containing an 8-oxoG at position 16 was labelled at the 5′-end with Cy5 and hybridised to the complementary oligonucleotide containing a cytosine opposite the lesion, yielding the 8-oxoG:C duplexes. Cleavage reactions were carried out at 37°C in a total volume of 14 µl containing the protein extract and 150 fmol of DNA probe. The reaction buffer was 22 mM Tris-HCl pH 7.4, 110 mM NaCl, 2.5 mM EDTA, 1 mg/ml BSA and 5% glycerol. Reaction mixtures containing the 8-oxoG:C probe were incubated at 37°C for 1 h, then NaOH (0.1 N final concentration) was added and the mixture was further incubated for 15 min at 37°C. Following incubation, the reaction was stopped by adding 6 µl of formamide dye (80% formamide, 10 mM EDTA, 0.02% bromophenol blue), followed by heating for 5 min at 95°C. The products of the reactions were resolved by denaturing 7 M urea polyacrylamide gel electrophoresis on 20% gels. Gels were scanned using a Typhoon Multi-format Imager (GE Healthcare Life Sciences).

We are grateful for the scientific and technical assistance of the Microscopy, Flow cytometry and Molecular Biology facilities of the Institute of Cellular and Molecular Radiobiology. We thank Samantha Lewis (Nunnari labortaory) for her help in the optimisation of the protocol for visualisation of replicating nucleoids. We would like to thank Paul-Henri Roméo, Christophe Carles and Chantal Desmaze for critically reading the manuscript.