OUP user menu

Downregulation of apoptosis-inducing factor in Harlequin mice induces progressive and severe optic atrophy which is durably prevented by AAV2-AIF1 gene therapy

Aicha Bouaita , Sébastien Augustin , Christophe Lechauve , Hélène Cwerman-Thibault , Paule Bénit , Manuel Simonutti , Michel Paques , Pierre Rustin , José-Alain Sahel , Marisol Corral-Debrinski
DOI: http://dx.doi.org/10.1093/brain/awr290 35-52 First published online: 26 November 2011


The Harlequin mutant mouse, characterized by loss of function of apoptosis-inducing factor, represents a reliable genetic model that resembles pathologies caused by human mitochondrial complex I deficiency. Therefore, we extensively characterized the retinal morphology and function of Harlequin mice during the course of neuronal cell death leading to blindness, with the aim of preventing optic atrophy. Retinas and optic nerves from these mice showed an isolated respiratory chain complex I defect correlated with retinal ganglion cell loss, optic atrophy, glial and microglial cell activation. All of these changes led to irreversible vision loss. In control mice, retinas AIF1 messenger RNA was 2.3-fold more abundant than AIF2, both messenger RNAs being sorted to the mitochondrial surface. In Harlequin mouse retinas, there was a 96% decrease of both AIF1 and AIF2 messenger RNA steady-state levels. We attained substantial and long-lasting protection of retinal ganglion cell and optic nerve integrity, the preservation of complex I function in optic nerves, as well as the prevention of glial and microglial responses after intravitreal administration of an AAV2 vector containing the full-length open reading frame and the 3′ untranslated region of the AIF1 gene. Therefore, we demonstrate that gene therapy for mitochondrial diseases due to mutations in nuclear DNA can be achieved, so long as the ‘therapeutic gene’ permits the accurate cellular localization of the corresponding messenger RNA.

  • apoptosis-inducing factor
  • optic atrophy
  • AAV-mediated gene therapy
  • respiratory chain complex I
  • retinal ganglion cells
  • Harlequin mice


Mitochondrial disorders, thought initially to be rare, now appear to be relatively common (Schaefer et al., 2008). Despite spectacular progress in elucidating the molecular basis of mitochondrial respiratory chain diseases in the last decade, a substantial amount remains to be done regarding therapy (Kerr, 2010; Schon et al., 2010). Thus, innovative approaches, including gene therapy, which may offer the real prospect of developing treatments that improve the underlying condition of patients, should be explored. Various obstacles may limit success in this goal: (i) the scarcity of animal models that truly resemble human diseases; (ii) conditional tissue targeting by gene therapy so as to minimize systemic distribution and avoid harmful side-effects; (iii) difficulty of efficient mitochondrial delivery of the gene product; and (iv) as for any Mendelian disorder, the choice of vectors, delivery method and dosage.

Since the eye is particularly suitable as a target organ for gene therapy (Colella et al., 2009) and ocular involvement is a frequent feature in mitochondrial respiratory chain diseases (Carelli et al., 2009; Yu-Wai-Man et al., 2009), we decided to develop a gene therapy strategy for preventing optic atrophy in the Harlequin (Hq) mouse. Hq mice exhibit common characteristics of human neurodegenerative mitochondriopathies due to respiratory chain complex I deficiency, such as the degeneration of the cerebellum, retina, optic nerve, thalamic, striatal and cortical regions (Klein et al., 2002; Vahsen et al., 2004; Benit et al., 2008). The Hq phenotype is caused by a severe reduction of apoptosis-inducing factor (AIF) gene expression due to a retroviral insertion in the first intron of the gene (Klein et al., 2002). AIF is a flavoprotein with nicotinamide adenine dinucleotide (NADH) oxidase activity localized to the mitochondrial inner membrane and involved in respiratory chain complex I biogenesis and/or stability. Interestingly, AIF was first discovered as a caspase-independent death effector able to induce features of apoptosis in isolated nuclei (Susin et al., 1999; Ye et al., 2002). Recently, a homozygous trinucleotide deletion in exon 5 of the human AIF gene, leading to the ablation of an arginine residue at position 201 of the AIF polypeptide, has been described as responsible for a severe X-linked mitochondrial encephalopathy and oxidative phosphorylation (OXPHOS) failure in two infant patients (Ghezzi et al.). By extensively evaluating Hq mouse eyes, we determined that retinal ganglion cells reached up to 36% of loss relative to control mice, associated with the disappearance of optic nerve fibres. Furthermore, the residual fibres presented a clear respiratory chain complex I deficiency. The overall injury evidenced in the ganglion cell layer of Hq mice led to progressive vision loss; thus, 10-month-old mice were permanently blind.

At least four different isoforms of AIF are possibly generated by alternative splicing of the messenger RNA precursor; in the nervous system AIF1 and AIF2 (Hangen et al., 2010) are the most abundant. Here we show that in Hq retinas, AIF1 and AIF2 messenger RNA steady-state levels were 96% inferior to control animals; in controls, AIF1 messenger RNA is 2.3-fold more abundant than AIF2. Both AIF1 and AIF2 messenger RNAs are sorted to the mitochondrial surface, as previously reported for several nuclear transcripts encoding mitochondrial proteins in human cells (Sylvestre et al., 2003). In an attempt to prevent retinal ganglion cell loss and optic nerve atrophy in Hq mice we have constructed a recombinant adeno-associated virus type 2 (AAV2/2_AIF1) which incorporates the mouse AIF1 open reading frame with its full-length 3′-UTR (untranslated region), ensuring the sorting of the messenger RNA to the mitochondrial surface and the efficient mitochondrial translocation of the protein. This point is particularly important because of the pro-apoptotic properties of AIF if localized to the cytosol (Scovassi et al., 2009). The prevention of retinal ganglion cell degeneration and the preservation of respiratory chain complex I activity in optic nerve were durably achieved in eyes injected intravitreally with AAV2/2_AIF1. Hence, we established the proof of principle that rAAV2/2-mediated gene therapy can lead to the prevention of respiratory chain complex I defects and optic atrophy in Hq mice.

Materials and methods


Hemizygous (Hq/Y) males were obtained by mating Hq/X females with either Hq/Y or wild-type males from the colony shipped from The Jackson Laboratory at ∼2 months of age.

The Hq strain was B6CBACaAw-J/A-Pdc8Hq/J (http://jaxmice.jax.org/strain/000501.html). All mice used in this study were F1 mice bred from founders having a mixed genetic background. Only hemizygous (Hq/Y) males received evaluations and gene therapy; they were compared exclusively with littermate males from the colony. The mice were housed from one to four per cage in a temperature-controlled environment, with a 12-h light/dark cycle and free access to food and water. Studies were conducted in accordance with the statements on the care and use of animals in research of the guidelines issued by the French Ministry of Agriculture and the Veterinarian Department of Paris (Permit number DF/DF_2010_PA1000298), the French Ministry of Research (Approval number 5575) and the ethics committees of the University Paris 6 and the INSERM, Institut National de la Santé et de la Recherche Médicale (Authorization number 75-1710).

Adeno-associated viral vectors and route for ocular administration

The entire Mus musculus apoptosis-inducing factor, mitochondrion-associated 1 (Aifm1) messenger RNA sequence (http://www.ncbi.nlm.nih.gov/nuccore/NM_012019) of 1926 base pairs (bp) was synthesized by Genscript Corp, encompassing the original 87 bp of the 5′-UTR, the entire open reading frame encoding a 612 amino acid-long protein and two restriction sites at the extremities: EcoR1 at the 5′ and XhoI at the 3′ for cloning into the pAAV internal ribosome entry site-humanized recombinant green fluorescent protein (hrGFP) vector (Stratagene) in which we had earlier replaced the hGH [human growth hormone 1 (MIM 139250)] polyadenylation signal with the 176 bp full-length AIF1 3′-UTR (http://www.ncbi.nlm.nih.gov/nuccore/NM_012019) by using BglII and RsrII unique restriction sites. AIF1 transcription is under the control of the CMV promoter and the β-globin intron to ensure high levels of expression. The open reading frame is in frame with the 3× FLAG® sequence at the C-terminus. The pAAV-internal ribosome entry site-humanized recombinant GFP vector also contains a dicistronic expression cassette in which the humanized recombinant GFP is expressed as a second open reading frame translated from the encephalomyocarditis virus internal ribosome entry site. The final vector, named AAV2/2_AIF1, contains adeno-associated virus type 2 (AAV2) inverted terminal repeats, which direct viral replication and packaging. The expression cassettes, flanked by the two AAV2 inverted terminal repeats, were encapsidated into AAV2 shells. The final construct was sequenced for accuracy (Genome Express).

In a previous report, we demonstrated that intravitreal injection of vector DNA followed by electroporation is successful in transducing rat retinal ganglion cells and leads to a sustained transgene expression for up to 90 days (Ellouze et al., 2008). In vivo electroporation of the AIF1 vector DNA in control mice was efficient in transducing retinal ganglion cells, which express AIF1 for at least 2 months without any harm to retinal architecture, optic fibre density or visual function (data not shown). Nevertheless, since our aim is to prevent durably the absence of AIF in Hq mice and its deleterious effect on retinal ganglion cell integrity, we need a protocol permitting both the efficient transduction of retinal ganglion cells and the long-lasting expression of the AIF1 transgene. It has been demonstrated that, to date, AAV2/2 represents the best choice in terms of overall transduction efficiency and tropism for adult rodent retinal ganglion cells (Harvey et al., 2009; Hellstrom et al., 2009). Moreover, it is well known that AAV2 shows strong dependence for transduction on heparan sulphate proteoglycans that are present in retinal ganglion cells and in the inner limiting membrane. The inner limiting membrane is a meshwork of extracellular matrix proteoglycans located at the interface of the vitreous and the ganglion cell layer. Thus, this first physical barrier may also effectively localize the virus and prevent it from being cleared from the vitreous via the trabecular meshwork, thereby facilitating subsequent retinal ganglion cell transduction (Dalkara et al., 2009). For all these reasons, we decided to target retinal ganglion cells of Hq mice with the AAV2/2_AIF1 vector, by administration into the vitreous body of mouse eyes.

Vectors were produced in the Vector Core at the University Hospital of Nantes (http://www.vectors.nantes.inserm.fr). The AAV titres were determined by dot blot and expressed as vector genomes per millilitre. During the course of this study we used two independent productions with the following titres: 7.5 × 1010 and 2.37 × 1011 vector genomes/ml. For intravitreal injections, after dilatation of the pupil with topical 1% tropicamide (CibaVision), mice were subjected to anaesthesia with isoflurane (40 mg/kg body weight). The tip of a 15-mm 33-gauge needle, mounted on a 10 µl Hamilton syringe (Hamilton Bonaduz AG) was advanced through the sclera ∼1 mm posterior to the corneoscleral limbus in the superior region of the eye, and 2–3 µl of vector suspension was injected intravitreally, avoiding retinal structure disruption, bleeding or lens injury. All the injections were performed under visual control using an ophthalmic surgical microscope. Viral particles were mixed with 1/10 000 of sterile fluorescein to follow the homogenous dissemination of the suspension into the vitreous. Forty-three mice (4- to 8-weeks old) were subjected to AAV2/2_AIF1 administration; 33 animals received 2.25 × 108 vector genomes/eye (lowest titre batch) and 10 received 7.11 × 108 vector genomes/eye (highest titre batch). Two animals developed a cataract due to the surgery and three animals died ∼7 weeks after eye surgery from natural causes; these five animals were discarded from the study.

Tissue homogenate preparation and respiratory chain enzymatic assays

After rapid and careful dissection, optic nerves or retinas were kept frozen (−80°C). Samples were prepared at ice-melting temperature by homogenization of tissues using a 1-ml hand-driven glass–glass Potter in 200 μl of extraction buffer (0.25 mM sucrose, 40 mM KCl, 2 mM ethylene glycol tetraacetic acid (EGTA), 1 mg/ml bovine serum albumin and 20 mM Tris–HCl, pH 7.2). Large cellular debris was spun down by a low speed centrifugation (3100 rpm, 8 min) and tissue homogenates were used immediately. Respiratory chain complex I and V activities were measured using a Cary 50 spectrophotometer equipped with an 18-cell holder maintained at 37°C (Varian Australia) as previously described (Benit et al., 2008). Each assay was made in triplicate with 50 µl of the homogenates obtained. All chemicals were of the highest grade from Sigma-Aldrich. Complex I (NADH-ubiquinone oxidoreductase) activity values were either normalized by complex V or converted to specific activities after protein quantification by the Bradford method. Optic nerves were dissected from 40 mice subjected to AAV2/2_AIF1 injection; optic nerves from seven mice were subjected to histological studies. Respiratory chain assessments obtained from only 28 mice were included in Table 3, since we failed to measure complex I and complex V activities from five mouse eye samples due to technical problems encountered during the assessment.

Fundus imaging by confocal scanning laser ophthalmoscopy and optical coherence tomography

A confocal scanning laser ophthalmoscope (Heidelberg Retina Angiograph, Heidelberg Engineering) with green laser illumination was used to examine mice before treatment and each month until euthanasia. To monitor the nerve fibre layer at each session, up to four pictures, each from an average of eight images, corresponding to cardinal areas of the eye fundus were performed. The built-in software was used for post-processing the images, including alignment, adjustment of contrast, construction of a mean image and/or of a composite image (Paques et al., 2006). Spectral-domain optical coherence tomography was performed using the Spectralis® system from Heidelberg Engineering. The acquisition time per tomographic image, which includes summation of 10 raw images, was in the order of 1 s (Grieve et al., 2004). Neural retina thickness for control and Hq eyes was compared by measuring the distance from the anterior boundary of the retinal nerve fibre layer and the posterior boundary of the photoreceptors layer. Throughout the examination process, a quiet ambiance with dim illumination was maintained. Pupil dilation was performed with topical 1% tropicamide (CibaVision). Careful slit-lamp examination before the procedure ruled out the presence of any corneal or lens opacities. Mice were manually held in front of either apparatus, in an upright position. As a rule, restraint and rest periods alternated approximately every 30 s. The examination was interrupted if the animal showed any sign of tiredness.

Retinal and optic nerve histology

Animals were sacrificed under deep ketamine (50 mg/kg) xylazine (10 mg/kg) anaesthesia. After the ocular globes were removed, they were pierced in the cornea with a needle and fixed for 1 h in 4% paraformaldehyde at 4°C. After removal of the cornea and lens, eyecups were post-fixed overnight in 4% paraformaldehyde at 4°C, cryoprotected by overnight incubation in phosphate-buffered saline containing 30% sucrose (Sigma-Aldrich) at 4°C. Then, retinas were embedded in Optimal Cutting Temperature medium (Neg 50; Richard-Allan Scientific), frozen in liquid nitrogen, and stored at −20°C. Optic nerves were carefully collected and fixed overnight in 4% paraformaldehyde at 4°C. Next day, after washes in phosphate-buffered saline, tissues were cryoprotected by incubation overnight in 30% sucrose in phosphate-buffered saline. Subsequently, optic nerves were incubated in a 7.5% gelatin solution from porcine skin; Type A (Sigma-Aldrich) and 10% sucrose. Samples were frozen between −50°C and −60°C in 2-methyl-butane solution and then stored at −20°C. Sections of retinas and optic nerves (transverse and longitudinal, near the globe) with a thickness of 10 µm were cut on a cryostat (Microm Microtech) at −20°C and mounted on SuperFrost®Plus slides.

For immunochemistry, sections of retinas and optic nerves were washed with phosphate buffer pH 7.4, permeabilized with 1% Triton X-100 in phosphate buffer for 15 min at room temperature and treated with 10% normal goat serum, 3% bovine serum albumin, 0.5% Triton and 0.05% Tween-20 in phosphate buffer for 1 h. They were then incubated with primary antibody overnight at 4°C. The next day, sections were washed three times in phosphate buffer and incubated with the appropriate secondary antibodies and DAPI (4′,6-diamidino-2-phenylindole; 2 μg/ml) for 90 min at room temperature with 3% normal goat serum, 3% bovine serum albumin, 0.5% Triton and 0.05% Tween-20 in phosphate buffer. Finally, they were washed three times in phosphate buffer, rinsed with sterile water and mounted on a glass slide. Primary and secondary antibodies used are shown in Supplementary Table 1.

Microscopic observations and retinal ganglion cell counts

Image acquisition of fluorescence labelling was monitored with a fluorescence microscope (Leica DM 6000 B) using the MetaVue software for image acquisitions or a confocal laser scanning microscope Olympus FV1000. Images were captured with Olympus software (Olympus). Images were then analysed with Adobe Photoshop® CS and ImageJ (National Institutes of Health) software. Slides from retinas and optic nerve sections were also scanned with the Hamamatsu Nanozoomer Digital Pathology (NDP) 2.0 HT, its fluorescence unit option (L11600-05) and the NanoZoomer's 3-CCD TDI camera (Hamamatsu Photonics). A resolution of 0.92 µm/pixel (×40) was routinely used.

Retinal ganglion cell counts were performed from each retina on three entire consecutive cryostat sections with depth at ∼400 mm from the optic nerve, using BRN3A immunolabelling as previously described (Ellouze et al., 2008). BRN3A is a transcription factor expressed in ∼80% of all retinal ganglion cells in mouse retina (Badea et al., 2009) and >90% in rat retina (Nadal-Nicolas et al., 2009). Manual counting of immunopositive cells was performed after capturing each whole section in 13–15 images; the identity of each mouse was unknown for the counting. BRN3A-positive cell numbers were further validated, for a subset of animals in each group, by performing two additional immunostainings of different retinal sections (four sections apart from the one used for the first labelling). Retinal ganglion cell numbers was estimated as a mean for 20 control eyes, 26 eyes from Hq untreated mice and both eyes from 16 Hq mice subjected to AAV2/2_AIF1 administration in one of their eyes (12 out of 16 received the vector in their left eyes).

Optomotor tests

The head-tracking method is based on an optomotor test (Thaung et al., 2002). Vertical black-and-white lines of three varying widths, subtending 0.0625, 0.125 and 0.25 cycle/°, were presented to the animal and rotated alternatively clockwise and counterclockwise, each for 60 s with an interval of 10 s between the two drum rotations. Head movements were recorded with a video camera mounted above the apparatus. Animals were scored only when the speed of the head turn corresponded with the speed of rotation of the stripes (12°/s). Light levels were kept constant (240 lux).

RNA extractions and real-time quantitative polymerase chain reaction assay

RNA preparations from mitochondrion-bound polysomes and free-polysomes were obtained from 12 brains (cortical region) from 8-week-old C57Bl/6NCrlBR male mice with some modifications of the protocol previously published for HeLa cells (Sylvestre et al., 2003). Briefly, after euthanasia, tissues were rinsed in phosphate-buffered saline before homogenization with a glass Potter homogenizer in the freshly prepared extraction buffer [0.32 M sucrose, 30 mM Tris–HCl pH 7.5, 5 mM AcMg, 100 mM KCl, 1% (vol/vol) protease inhibitor cocktail (Sigma), 5 mM β-mercapthoethanol, 1 mM PMFS (Fluka), 0.1% (w/vol) bovine serum albumin fraction V in ultra-pure distilled water]. The homogenate was centrifuged (1000 rpm, 10 min, 4°C) to eliminate nucleus and unbroken cells. The supernatant was conserved at 4°C and the pellet was subjected to two additional homogenizations for optimal cell lysis. The three supernatants were pooled and were centrifuged at slow speed twice to remove nucleus and unbroken cells. The supernatants obtained were then centrifuged at 12 500 rpm for 30 min at 4°C; the pellet contained mitochondria and the supernatant represented the cytosolic fraction. The mitochondrial fraction was washed three times in the extraction buffer (lacking bovine serum albumin). The last mitochondrial pellet, containing polysomes bound to the outer mitochondrial membrane, was conserved at −80°C. The cytosolic fraction was gently deposed on a SW41 tube that contained two layers of 2.0 M sucrose and 0.5 M sucrose prepared in the extraction buffer. SW41 tubes were centrifuged at 4°C for 20 h at 40 000 rpm. The pellet represented the free-polysome fraction.

Total RNAs were prepared from mitochondrion-bound polysomes, free-polysome fractions as well as from wild-type or Hq mouse retinas using the RNA isolation kit (Qiagen). RNA concentration was determined with the Nanodrop ND-2000 Spectrophotometer (Thermo Fisher Scientific) and RNA integrity was verified using agarose gel electrophoresis. Real-time polymerase chain reaction (PCR) analyses using mitochondrion-bound polysome and free-polysome fractions were performed with the Superscript III one-step real-time PCR kit with Platinum Taq (Invitrogen, Life Technologies).

For real-time quantitative PCR analyses, 1 µg of total RNA from wild-type or Hq mouse retinas was reverse-transcribed with oligo-dT in a 20 µl final reaction volume using Superscript® II Reverse Transcriptase (Invitrogen, Life Technologies) following the manufacturer's instructions. The mouse AIF1, AIF2 and AIF1tr (AIF1 messenger RNA produced from the vector) primers were customized to be highly specific for each messenger RNA specie. RNAs were purified from 21 mice subjected to AAV2/2_AIF1 intravitreal injection, 13 received the lowest vector dose and eight received the highest vector dose. One of the samples from the lowest dose was discarded, since a mixture with another sample led to a confusion of its real identity. Thus, the results for 20 mice subjected to AAV2/2_AIF1 administration were presented. In treated mice, the measurement of AIF1 corresponds to the overall amount of AIF1 messenger RNA (endogenous gene and AAV2/2_AIF1 driven expression) and the measurement AIF1tr represents only the abundance of messenger RNA transcribed from the vector since one of the primers used recognizes the FLAG epitope sequence. The primers used in this study are shown in Supplementary Table 2 and were synthesized by Invitrogen, Life Technologies.

Quantitative real-time PCR reactions were performed using ABI 7500 Fast (Applied Biosystems). For microplate experiments, 1/100 and 1/500 of the reverse transcription product was used per gene as template for quantitative PCR reactions with Power Sybr® green PCR Master Mix (Applied Biosystems, Life Technologies,). Real-time PCR amplifications were carried out as follows: 10 min 95°C incubation followed by 40 cycles of 15 s at 95°C and 1 min at 60°C. Each biological sample was subjected to the assay in triplicate. Obtained Ct values for each sample allowed AIF1, AIF2, AIF1tr (AIF1 messenger RNA synthesized from the AAV2/2_AIF1 vector), BRN3A and gamma-synuclein (SNCG) messenger RNA quantifications using 7500 v.2.0.4 software (Applied Biosystems). The mitochondrial ATP6 gene was selected as the most stable, regarding messenger RNA steady-state levels, to normalize the messenger RNA species evaluated and to obtain relative messenger RNA abundance in Hq retinas. Several reports indicated that BRN3A and SNGC messenger RNA levels can be used to assess retinal ganglion cell number (Soto et al., 2008; Surgucheva et al., 2008; Torero Ibad et al., 2011). Therefore, their estimation in both Hq and control retinas aimed at completing retinal ganglion cell counts obtained by histological evaluation of retinal sections.

Statistical analyses

Values are expressed as mean ± SEM. Data were compared using Student's t-test (http://www.u707.jussieu.fr/biostatgv/index.html). P-values of <0.05 were considered as significant, and <0.01 as highly significant.


Phenotypic hallmarks of Harlequin mouse eyes

We first studied the viability of retinal ganglion cells in Hq mice since their long axons form the optic nerve. We examined eye fundus at different ages using a full-field optical coherence tomograph (Grieve et al., 2004) (Fig. 1). In control mice of any age, the nerve fibre bundles converge towards the optic nerve disc and radiating vessels were noticed in the xy images (Fig. 1A). By contrast, stacks of Hq retina xy images collected from animals older than 6 months showed a significant diminution of the number of nerve fibre bundles converging to the optic nerve disc (Fig. 1B). In reconstructed xz retinal images (Fig. 1A and B; right), the nuclear and plexiform layers were recognizable. Retinal thickness (µm ± SEM) was 234.5 ± 14.5 µm in control mice independently of their age (n = 4); while in Hq mice values were 205.75 ± 21.75 µm and 143.5 ± 9.5 µm for 2- and 13-month-old animals, respectively (n = 3 per age group). The significant reduction of retinal thickness in old Hq mice compared with age-matched controls may reflect the retinal neuron loss in all cell layers previously described in Hq animals older than 8 months (Klein et al., 2002).

Figure 1

In vivo analysis and retinal layering with spectral domain optical coherence tomography imaging of control and Hq retinas. Representative fundus photographs (left) of control (A) and Hq (B) mice aged between 2 and 14 months. The right panels of A and B illustrate cross-sectional optical coherence tomography images distal to the optic disk. Retinal thickness was measured by positioning built-in callipers between the anterior boundary of the retinal nerve fibre layer (NFL) and the posterior boundary of the photoreceptor layer excluding the retinal pigment epithelium (RPE).

To quantify retinal ganglion cell somas in Hq mice, cryostat sections of retinas were counted for BRN3A-positive cells in the ganglion cell layer, since BRN3A is a transcription factor expressed in ∼80% of all retinal ganglion cells in mouse retina (Badea et al., 2009) and >90% in rat retina (Nadal-Nicolas et al., 2009). We found that retinal ganglion cell somas were significantly reduced in Hq retinas: 149 ± 27 (n = 26) relative to 233 ± 39 in control retinas (n = 20), a diminution of >36% of the total amount of BRN3A-positive cells in the ganglion cell layer (P =2.2 × 107). Accordingly, the total number of cells in the ganglion cell layer was also significantly lower in Hq mice: 502.7 ± 50.8 relative to 608.4 ± 76 in control retinas (P = 0.00021).

Glial cell activation is an important factor contributing to retinal ganglion cell death in experimental models of glaucoma (Jiang et al., 2007; Zhou et al., 2009). In retinas from 1-month-old control and Hq mice, glial fibrillary acidic protein (GFAP), which is a sensitive marker of glial activation immunofluorescence, was confined exclusively to the ganglion cell layer, presumably corresponding to the end-feet of Müller cells and a few astrocytes (Fig. 2A). An obvious increase in GFAP immunoreactivity across the entire thickness of the retina was observed in Hq retinas from 7 - and 13-month-old mice; while 7- and 13-month-old control retinas showed GFAP immunostaining restricted to the ganglion cell layer (Fig. 2B). Interestingly, in retinal sections from 7- and 13-month-old Hq mice, the diminished number of BRN3A-positive cells in the ganglion cell layer and the thinness of both the inner nuclear layer and outer nuclear layer are noticeable, confirming the neuron cell loss noted in eye fundus images (Fig. 1) and reported earlier (Klein et al., 2002). The reconstruction of whole retinal sections confirmed the extensive glial response in 5- and 8-month-old Hq mice relative to age-matched controls as well as the retinal thickness reduction in Hq eyes (Fig. 2C). Next, optic nerve cross-sections were immunolabelled to observe axonal profiles and to search for myelin pathology in Hq mice. Figure 2D illustrates a recognizable reduction in immunopositive dots in 7-month-old Hq mice relative to age-matched controls when two antibodies against integral axonal proteins were used (β-III tubulin and NF200). Moreover, the immunostaining with the antibody against the myelin basic protein, one of the most abundant proteins of the myelin sheath, shows a very weak signal in Hq mice; whereas intact and strong labelled red rings (myelin basic protein) were observed in age-matched control (Fig. 2D). Thus, AIF depletion in the optic nerve induces severe loss of axons along with oligodendrocyte death; this latter could ultimately lead to impaired myelination. We also evaluated longitudinal sections of wild-type and Hq mice for gliosis and microglial activation. For the latter, we used IBA1 (ionized calcium binding adaptor molecule 1) since it is upregulated in activated microglia and is involved in cell migration and phagocytosis (Sasaki et al., 2001). We observed a weak immunoreactivity for GFAP in optic nerves from a 12-month-old control, while a strong signal was apparent in optic nerves from an age-matched Hq mouse. Further, it appears that IBA1-positive cells were present in higher numbers and with an increased staining intensity in optic nerves from Hq eyes compared with the few scattered IBA1-positive cells observed in optic nerves dissected from an age-matched control (Supplementary Fig. 1). Additionally, microglial cells in control optic nerves were consistently more ramified (resting state); while IBA1 immunostaining in optic nerves from Hq mice revealed both ramified cells and round somata cells with simplified processes (activated state). Hence, similar to rodent models of optic nerve injury (Bosco et al., 2008; Sivilia et al., 2009; Morrison et al., 2010) optic nerves from Hq mice hosted activated astrocytes, microglial cells, as well as structural disruption of myelination.

Figure 2

Immunofluorescence analysis of retinal and optic nerves from control and Hq mice. (A) Immunostaining for GFAP (green) and BRN3A (red) of retinas from control and Hq mice aged 1 month; GFAP expression was restricted to the ganglion cell layer (GCL). The nuclei were contrasted with DAPI (blue). (B) With ageing, Hq mice exhibited a progressive increase in GFAP expression as illustrated in Hq mice aged 7 and 13 months, respectively. By contrast, age-matched controls only showed a GFAP signal restricted to the ganglion cell layer. It also appears in this figure that with ageing the Hq mouse inner nuclear layer (INL) and outer nuclear layer (ONL) have thinned considerably. Scale bar = 50 µm. (C) Low magnification images of retinas from 5-month-old and 8-month-old Hq mice and age-matched controls, GFAP (green) and DAPI (blue) staining are shown. Scale bar = 1 mm. In merge reconstitutions, a 3-fold zoom of the same region is also illustrated. (D) Independent proximal optic nerve transversal sections from 8-month-old control and Hq mice were immunolabelled with antibodies against the heavy chain (200 kDa) subunit of neurofilaments, (NF200, green), the tubulin β-III, a microtubule protein (β-III, red) and the basic myelin protein (MBP, red). Immunoreactivity for the two axonal markers (NF200 and tubulin β-III) shows that axons are largely lost in Hq mice relative to age-matched controls. Myelin basic protein immunoreactivity was also significantly reduced in Hq mouse optic nerves, suggesting myelin alterations due to the absence of AIF1 in these mice. Scale bars = 100 and 50 µm. IPL = inner plexiform layer.

Studies of respiratory chain function

First, we assessed complex I and complex V activities and estimated the complex I/complex V ratios in retinas from eight 8-month-old Hq mice and seven age-matched controls (Table 1). A significant 60% decrease relative to control values was observed, in both complex I-specific activity and complex I/complex V ratio, while complex V activity was identical between Hq and control mice. Secondly, optic nerves were collected either from animals sacrificed before 8 months of age or older than 8 months. In control optic nerves, complex I, complex V and their ratio did not change with age, being similar to the ones measured in retinas (Table 1). Optic nerves from Hq mice aged <8 months already had an ∼50% decrease of complex I activity, a decrease that worsened with age (>70% decrease after 8 months of age). Complex V activity was not different between controls and Hq mice in either age group examined. Thus, in Hq mice not only did a significant proportion of nerve fibre bundles disappear (Fig. 1) but complex I enzymatic activity in residual nerve fibres was severely compromised.

View this table:
Table 1

Detection of respiratory chain complex I deficiency in retinas and optic nerves from Hq mice

Ocular tissuesSpecific activities (nmol/min/mg protein) ± SEMActivity ratios ± SEM
Complex IComplex VComplex I/V
Retinas from 8-month-old mice (number of eyes tested)
    Controls (14)16.44 ± 3.768.48 ± 8.240.278 ± 0.071
    Harlequin (16)6.91 ± 0.8875.73 ± 11.80.090 ± 0.014
    P-value controls/Harlequin5.41 × 10−50.139.15 × 10−6
Optic nerves (age; number of eyes tested)
    Controls (2–7 months; 32)14.12 ± 4.5372.24 ± 20.90.235 ± 0.05
    Harlequin (2–7 months; 28)7.22 ± 1.9170.64 ± 29.370.143 ± 0.025
    P-value controls/Harlequin0.00250.8987.35 × 10−9
    Controls (8–13 months; 20)13.94 ± 4.1870.93 ± 17.40.243 ± 0.055
    Harlequin (8–13 months; 28)3.78 ± 1.6472.33 ± 16.90.0706 ± 0.0176
    P-value controls/Harlequin4.67 × 10−60.8912.58 × 10−10
  • Measurements of complex I and V activities were obtained by spectrophotometry. Complex I and complex V activities were expressed as nanomoles of oxidized NADH/min/mg protein. In retinas and optic nerves, a complex I defect was shown in young Hq mice; the activity declined in Hq with age. Values represent the mean of triplicates per each eye sample evaluated.

Harlequin mouse visual function

We next studied the visual function of the Hq mice by the optokinetic drum test (Thaung et al., 2002; Davies et al., 2007). We first examined sighted littermate controls (B6CBAC) at two different ages (>3 months and between 6 and 12 months). They were capable of tracking the moving acuity square-wave gratings of the three frequencies examined, by making several head-tracking movements in the same direction and speed as the drum (Table 2). Hq mice younger than 3 months showed head-tracking of all three square-wave gratings, indicating that they were not functionally blind despite their reduced visual performance. Hq mice aged between 4 and 8 months of age spent much less time tracking across the test period at all three square-wave gratings relative to age-matched controls. Thus, by this age there was a significant effect of genotype in the three grating frequencies, for both the clockwise and counterclockwise drum rotations (Table 2). Finally, eight Hq mice older than 11 months failed to make head-tracking movements at any of the grating frequencies in either clockwise or counterclockwise drum rotation, and could be considered totally blind. The complete loss of visual function in animals older than 1 year is certainly the consequence of retinal ganglion cell degeneration, the disappearance of photoreceptors (Figs 1 and 2B) and the disruption of myelin in the optic nerves (Fig. 2D).

View this table:
Table 2

Optomotor tests of control and Harlequin mice

Animals Tracking featuresLittermate control mice (<3 months of age) n = 10Littermate control mice (6–12 months of age) n = 10Young Hq mice (<3 months of age) n = 12Hq mice (4–8 months of age) n = 10
Gratings frequency (cycles/°)0.06250.1250.250.06250.1250.250.06250.1250.250.06250.1250.25
Clockwise4.3 ± 1.95.3 ± 1.76 ± 29.8 ± 29.3 ± 2.58.9 ± 2.22.6 ± 1.64.25 ± 24.8 ± 22.1 ± 1.782.35 ± 1.73.3 ± 1.75
Counter-clockwise5.4 ± 2.16.3 ± 0.858 ± 1.410.4 ± 2.58 ± 2.39.6 ± 2.24.6 ± 2.13.75 ± 1.753.7 ± 1.71.75 ± 1.552.4 ± 1.582.7 ± 1.9
  • Four different groups of mice were evaluated for head tracking movement at an angular speed identical to that of the drum rotation.

  • The mice were presented with a 2°, 4° and 8° grating (corresponding to 0.25, 0.125 and 0.0625 cycles/°, respectively). They made several head tracking movements in the same direction and speed as the drum. Head movements shown in Table 2 are provided per minute. Data collected from Hq mice aged between 4 and 8 months at the three grating frequencies for both the clockwise and counter-clockwise drum rotations were significantly different to those of littermate control mice (6–12 months of age): Student's t-test; P = 3.8 × 10−6, 3 × 10−6, 0.0015, 0.00094, 0.0106, 0.00019. Eight Hq mice of about 1 year of age were also subjected to the test three times within a 2-week time interval: they totally failed to make head tracking movements at any of the grating frequencies for either clockwise or counter-clockwise drum rotation.

Relative abundance of AIF1 and AIF2 messenger RNAs in Hq retinas

The relative abundance of AIF1 and AIF2 transcripts (Hangen et al., 2010) in retinas from control and Hq mice was determined by quantitative real-time PCR using the comparative ΔΔCt method and the mitochondrial ATP6 as a reference gene, since its messenger RNA steady-state levels remained almost unchanged in all the samples evaluated (data not shown). RNA preparations from 24 retinas of littermate controls aged between 6 and 7 months were examined. AIF1 messenger RNA is 2.3-fold more abundant than AIF2 in retinas from control mice (P = 8.2 × 1012). The difference between AIF1 and AIF2 messenger RNA relative abundance was preserved in 31 Hq retinas, while the total amount of both messenger RNAs fell to <4% of control values (Fig. 3A). As SNCG messenger RNA is expressed at high levels in most if not all retinal ganglion cells and in no other cells within the adult mouse retina (Soto et al., 2008; Surgucheva et al., 2008), we determined the relative amount of SNCG and BRN3A messenger RNAs (Fig. 3B). The steady-state levels of these two specific markers of retinal ganglion cells were reduced to ∼50% of control values in Hq retinas (P-values of 2.3 × 108 and 1.8 × 1011 for BRN3A and SNCG, respectively), confirming our data on retinal ganglion cell loss and nerve fibre degeneration (Fig. 1).

Figure 3

AIF1, AIF2, SNCG and BRN3A messenger RNA (mRNA) steady-state levels in control and Hq retinas. Relative fold variations were calculated using the comparative Ct method and normalized to the mitochondrial ATP6 messenger RNA steady-state levels, since no variations for the latter were observed among all the RNA preparations evaluated. Data obtained (A and B) correspond to three independent quantifications (mean ± SEM) of RNA preparations from mouse retinas (control n = 24, Hq n = 31). AIF1 and AIF2 messenger RNA steady-state levels are significantly different (Student's t-test; P = 8.2 × 10−12). The steady-state levels of both BRN3A and SNCG messenger RNAs were significantly diminished in Hq retinas relative to age-matched controls (P-values of 2.3 × 10−8 and 1.8 × 10−11, respectively). PCR amplifications were performed using specific primers for each gene (Supplementary Table 2).

Design of a gene therapy strategy for preventing retinal ganglion cell and optic nerve injury

To efficiently prevent retinal ganglion cell loss and optic nerve atrophy in Hq mice, AIF protein must be translocated to the mitochondrial inner membrane. By studying RNAs isolated from mitochondrion-bound polysomes and free cytosolic polysomes isolated from mouse brains we found an enrichment of AIF1 and AIF2 messenger RNAs in mitochondrion-bound polysomes; indeed, ∼70% of the overall signal of AIF1 or AIF2 messenger RNA was found in this fraction (Fig. 4). Next, we selected AIF1 for gene therapy since its messenger RNA is 2.3-fold more abundant than AIF2 messenger RNA. We constructed a recombinant AAV vector containing both the full-length AIF1 open reading frame and the 3′-UTR of the mouse gene to ensure messenger RNA sorting to the mitochondrial surface. The recombinant vector was generated using the genome and capsid proteins from serotype 2, which has been described as able to efficiently transduce retinal ganglion cells in mice (Guy et al., 2009; Zhou et al., 2009). For preventing optic atrophy, 4- to 8-week-old animals were subjected to a single intravitreal injection since at this age the extent of retinal ganglion cell injury did not lead to noticeable nerve fibre degeneration (Fig. 1) or vision loss (Table 2).

Figure 4

Subcellular distribution of AIF1 and AIF2 messenger RNAs in mouse brains. Steady-state levels of different messenger RNAs were compared in the mitochondrion-bound polysome (M-P) and free-polysomes (F-P) fractions depending on the gene examined, as follows: mitochondrial ATP6: 50 ng of RNA for each fraction and 18 cycles; glyceraldehyde-3-phosphate dehydrogenase, GAPDH: 50 ng for each fraction and 23 cycles; adenylate kinase isoenzyme 4, AK4: 100 ng for each fraction and 30 cycles; apoptosis-inducing factor, mitochondrion-associated 1 (AIF1) encoding the isoform AIF1: 250 ng for each fraction and 30 cycles; apoptosis-inducing factor, mitochondrion-associated 2 (AIF2) encoding the isoform AIF2: 250 ng for each fraction and 30 cycles. PCR amplifications were performed using specific primers for each gene (Supplementary Table 2) and three independent RNA purifications. Each specific primer leads to the amplification of a PCR product of 150 bp. Densitometric analyses were performed with signals revealed by electrophoresis (top: 1/10 of the real-time PCR reaction for the ATP6 messenger RNA and 1/5 for the other messenger RNAs). Bar graphs (bottom) represent the amount of each messenger RNA species in the mitochondrion-bound polysome fraction relative to the total signal measured in mitochondrion-bound polysome and free-polysome fractions. The intensity of each signal was quantified using the Quantity One® 1-D Analysis Software (Bio-Rad laboratories).

Effect of rAAV2/2_AIF1 administration on retinal ganglion cell integrity

Four to five months after AAV2/2_AIF1 treatment, we evaluated the relative amount of the transduced AIF1 messenger RNA by real-time quantitative PCR in retinas from 20 Hq mice. Twelve mice received 2.25 × 108 vector genomes; the remaining eight were treated with 3.16-fold more of AAV2/2_AIF1. We detected specific messenger RNA species in retina preparations of all the injected eyes; the steady-state levels of the transduced AIF1 (AIF1tr) messenger RNA contributed to a 14% increase of the overall amount of AIF1 messenger RNA when 2.25 × 108 vector genomes were used. Remarkably, mouse retinas from eyes injected with the highest dose of the vector showed 600-fold more of the transduced AIF1 messenger RNA. Consequently, the total amount of AIF1 messenger RNA in retinas from transduced eyes reached 25.25% of the control value (Fig. 5A). We next estimated the relative amounts of BRN3A and SNCG messenger RNAs in retinas from treated and untreated eyes. Independently of the AAV2/2_AIF1 dose used, a consistent but not statistically significant increase of 13.5 and 20%, respectively in their relative abundance was measured between treated and untreated eyes, whereas rhodopsin (RHO) messenger RNA abundance remained unchanged (Fig. 5B). The high variability observed could be ascribed to the well-known interindividual phenotypic heterogeneity, unevenness due to transduction efficiency, but also to the use of whole retinas in which retinal ganglion cells represent <5% of the overall cell population. Therefore, we decided to estimate retinal ganglion cell number in retinal sections from 16 mice treated in one of their eyes with AAV2/2_AIF1 and euthanized ∼4 months after vector administration. Retinal ganglion cell number in treated eyes was 31.6% higher than in contralateral untreated eyes from the same animals (P = 1.4 × 109, Fig. 5C). AIF1 expression driven by the AAV2/2 vector in Hq mice led to an increase of ∼39% for retinal ganglion cells and 15% for ganglion cell layer cell population relative to the values obtained in the overall untreated Hq eye population (P = 1.2 × 108 and 0.00034, respectively). Thus, AAV2/2_AIF1 intravitreal administration led to a remarkable preservation of retinal ganglion cell integrity since their number reached 89% of the control value despite remaining statistically different (P = 0.042).

Figure 5

Evaluation of AIF1 transgene expression and its impact on retinal ganglion cell integrity. (A) Relative quantification of endogenous AIF1 messenger RNA and AIF1 messenger RNA transcribed from the AAV2/2_AIF1 vector in Hq retinas. For vector administration, intravitreal injections were performed in one eye with either 2.25 × 108 vector genomes (VG) or 7.11 × 108 vector genomes. (B) The impact of AAV2/2_AIF1 treatment on retinal ganglion cell integrity was estimated by the evaluation of SNCG and BRN3A messenger RNA steady-state levels in untreated or treated Hq eyes. AAV2/2_AIF1 administration did not influence the relative RHO messenger RNA amounts as shown. Data obtained correspond to three independent experiments per RNA preparation (mean ± SEM) of mouse retina [control n = 20, Hq untreated or treated eyes (2.25 × 108 vector genomes; n = 12) and Hq untreated or treated eyes (7.11 × 108 vector genomes; n = 8)]. (C) Retinal ganglion cell (RGC) number estimations were performed after immunolabelling for BRN3A antibody and DAPI staining; this latter allowed estimation of total nuclei in the ganglion cell layer (GCL). The histograms showed the means ± SEM obtained after all of the total and BRN3A cells in the ganglion cell layer were counted in three independent retinal sections per animal. Thirty-two Hq mouse eyes were evaluated: 16 were subjected to AAV2/2_AIF1 administration in one eye (12 left eyes and four right eyes); their fellows were untreated. Animals were euthanized between 16 and 20 weeks after vector administration. Values were compared with retinal ganglion cell number obtained for 20 controls and 26 Hq mice that were not subjected to gene therapy.

AIF1 expression prevents nerve fibre loss in Hq mice

To determine whether AAV2/2_AIF1 administration could lead to a better preservation of nerve fibres in Hq mice, eyes fundus images were collected using confocal scanning laser ophthalmoscopy (Paques et al., 2006) and optical coherence tomography. Experiments were performed before vector administration and monthly until euthanasia, permitting us to follow over time the disappearance of nerve fibres and look for any change related to AIF1 transgene expression in treated eyes. Striations of the nerve fibre layer radiating from the optic nerve disc were clearly visible for different areas of the eye fundus (nasal, temporal, inferior, superior). Figure 6A shows a mouse eye fundus image before (6 weeks) and at different times after vector administration. A diffuse loss of nerve fibre bundles is clearly visible in the untreated eye of the mouse at the age of 22 weeks, especially in the superior and temporal superior areas. In the bottom panel of Fig. 6A, images collected for the eye that received AAV2/2_AIF1 are illustrated, for the areas shown, tracks of axons were preserved. Thus, axon loss gives the impression of being less important than in the untreated eye, indicating that AIF1 expression protected against nerve fibre degeneration. Figure 6B illustrates fundus imaging obtained by optical coherence tomography in Hq mouse before euthanasia; it is noticeable that nerve fibre bundles are more abundant in the treated eye than in the untreated eye. Moreover, the reconstructed xz retinal images show irregularities in the retinal structure in the untreated eye, especially in the nerve fibre layer and ganglion cell layer which appear as broader regions of high scattering intensity, while the disorganization of cell architecture is less notable in the treated eye. Haematoxylin and eosin staining of AAV2/2_AIF1 treated Hq eye and contralateral untreated eye confirms the optical coherence tomography results since it appears that the nerve fibre layer is thicker in the treated eye, as if more retinal ganglion cell axons are converging to the optic nerve head. The axonal profiles, detected by immunohistochemistry for NF200 and β-III tubulin, in optic nerve cross-sections confirmed that treated Hq eyes presented an increase in neurofilament and microtubule-immunopositive signals relative to untreated eyes (Fig. 6C). This indicates that AAV2/2_AIF1 administration prevents loss of axons within the nerve and substantiates retinal ganglion cell number estimations in eyes subjected AAV2/2_AIF1 treatment (Fig. 5C). Conversely, no changes in the myelin basic protein-immunostaining were noticed in treated eyes relative to untreated ones. This result indicates that oligodendrocytes residing in the optic nerve cannot be transduced by the vector after intravitreal administration.

Figure 6

Protection against retinal ganglion cell damage in Hq eyes treated with AAV2/2_AIF1. (A) Eye fundus images obtained by scanning laser ophthalmoscopy from an Hq mouse, in which one eye has been subjected to AAV2/2_AIF1 injection when the animal was 6 weeks old. Images were collected from both eyes before vector injection and at different times until euthanasia, performed 20 weeks after vector administration. Different regions of the retina were illustrated: face up, NI (nasal inferior), T (temporal), S (superior) and TI (temporal inferior). It appears that the untreated eye may have undergone a more severe degenerative process of the nerve fibres relative to the treated eye. (B) Spectral domain optical coherence tomography and light microscopic images of treated (left) and untreated (right) eyes from one Hq mouse in which AAV2/2_AIF1 administration was performed at the age of 1 month; the animal was sacrificed 16 weeks later. En face images show more nerve fibres bundles in the treated eye than in the untreated one (left). Low-magnification images from the treated and untreated eyes of the same Hq mouse after haematoxylin and eosin staining of retinal sections are also illustrated. Note that more retinal ganglion cell axons converged to the optic nerve head in the treated eye relative to the untreated eye. (C) Immunofluorescence analysis of optic nerve cross-sections from control and Hq mice. Independent proximal optic nerve transversal sections (near the globe) from a 7-month-old control animal and an Hq mouse were immunolabelled with antibodies against the heavy chain (200 kDa) subunit of neurofilaments, (NF200, green), the tubulin β-III, a microtubule protein (β-III, red) and the basic myelin protein (MBP, red). The Hq mouse shown was subjected to a single intravitreal injection of AAV2/2_AIF1 (7.11 × 108 vector genomes) in its left eye at 8 weeks of age and euthanized 5 months later. Immunoreactivity for the two axonal markers (NF200 and tubulin β-III) shows a noticeable prevention of axon loss in the treated eye relative to the untreated one. Indeed, the density of either green (NF200) or red (β-III) dots was similar to the one observed in control mouse. Conversely, myelin basic protein immunofluorescence signal did not change by the AAV2/2_AIF1 administration. Scale bars = 100 and 50 µm.

Remarkably, optic nerve sections immunolabelled for GFAP and IBA1 showed weaker fluorescence signals in the treated eye relative to the untreated one. Immunofluorescence for both GFAP and IBA1 in the optic nerve section shown were comparable between the eye subjected to vector administration and a control sample from a mouse of approximately the same age (Supplementary Fig. 2). Accordingly, AIF1 expression driven by AAV2/2_AIF1 appears to lead to a beneficial effect on optic fibre integrity and a diminution of gliosis and microglial activation in optic nerves.

Complex I activity is maintained in optic nerves from AAV2/2_AIF1-treated animals

We next assessed complex I enzymatic activity in optic nerves from mice expressing the AIF1 transgene to determine whether the beneficial effects observed in both retinal ganglion cell overall number and nerve fibre density could result from preserved complex I activity. Twenty-eight mice were subjected to intravitreal injection of AAV2/2_AIF1 at ages between 4 and 8 weeks; eight of these animals received the highest dose of the vector. All the mice were euthanized at 16–20 weeks after vector administration. AIF1 expression driven by AAV2/2_AIF1 led to a significant protection against complex I defect, since both complex I activity and complex I/complex V ratios in treated eyes attained the values measured in age-matched controls and were statistically different to those measured in the untreated fellow eyes (Table 3). By contrast as expected, values measured in untreated eyes were very similar to the ones measured in untreated Hq mice of approximately the same age (group 2–7 months; Table 1) confirming that the observed difference is a direct consequence of the increased AIF protein level in treated eyes. Noticeably, in all the samples assessed, complex V enzymatic activities were unchanged. Thus, AAV2/2_AIF1 administration to Hq mouse eyes, before an extensive retinal ganglion cell loss, neutralized the deleterious effect of AIF depletion on complex I function; and this was independent of vector dose used, showing that a 14% increase in the overall amount of cellular AIF1 messenger RNA (Fig. 5A) is sufficient to protect against complex I deficiency.

View this table:
Table 3

Complex I and complex V activity assessment in optics nerves from Harlequin mice treated with AAV2/2_AIF1

Optic nerves (n = 28)Complex I ± SEMComplex V ± SEMComplex I/V ± SEM
Untreated eyes7.43 ± 1.9869.12 ± 14.30.128 ± 0.042
Treated eyes12.08 ± 3.0472.21 ± 11.20.20 ± 0.061
Student test: P-values
  • The results from 28 animals subjected to intravitreal administration of AAV2/2_AIF1 at an approximate age of 6 weeks are shown. The average time for euthanasia was 4–5 months after the ocular intervention; thus mice were all younger than 8 months at the time of the assessment. Complex I and complex V activities were expressed as nanomoles of oxidized NADH/min/mg protein. Data were compared using Student's t-test (http://www.u707.jussieu.fr/biostatgv/index.html). The comparison between animals subjected to vector administration and the whole control or Hq population was performed using the values shown in Table 2 for the group of animals younger than 8 months.


Because the above experiments target the multi-faceted AIF protein, a number of conclusions can be drawn from this study. They first deal with AIF function. Throughout the study, it is clear that AIF depletion in the Hq retinas and optic nerves, far from reducing cell death processes, rather generates retinal ganglion cell death; this latter process is robustly counteracted by AAV2/2_AIF1. Besides, the consequence of AIF depletion seems to be directly correlated with severely defective respiratory chain complex I activity. Thus, these results fully confirm the previous observation that the alternative function of AIF, as an assembly factor for the respiratory chain, determines the in vivo mouse phenotype associated with AIF depletion (Vahsen et al., 2004; Benit et al., 2008). A similar observation was recently reported in humans, where a deleterious mutation in AIF triggers a typical mitochondrial respiratory chain disease hallmarked by severe encephalopathy (Ghezzi et al., 2010).

It has been suggested that the AIF2 isoform mediates physiological function only in the differentiating neurons (Hangen et al., 2010). We observed that increased levels of AIF1, the major isoform we have identified in mouse retinas, are sufficient to protect retinal ganglion cell and optic nerve integrity in Hq mice. So far, in the context of eye diseases resulting from mitochondrial dysfunction, two approaches have been used to get round the absence of available experimental models. First, mitochondrial dysfunction has been triggered by blockade of complex I by rotenone. Complex I inhibition was subsequently counteracted by the allotopic expression of the yeast Ndi1 protein which successfully counteracts the effect of rotenone by providing a bypass for matrix NADH oxidation (Marella et al., 2010). Next, impairing complex I activity by the allotopic expression of a mutant ND4 gene was shown to be readily neutralized by expressing its wild-type counterpart (Ellouze et al., 2008). While these approaches demonstrate the feasibility of gene therapy in these particular models of complex I-associated eye diseases, the Hq mouse strain provides an authentic genetic model for a mitochondrial respiratory chain complex I disease affecting the eye.

Down regulation of AIF gene expression in Hq mice is responsible for a severe respiratory chain complex I defect and leads to the progressive degeneration of neurons in the cerebellum, thalamus and striatum with a massive glial reaction (El Ghouzzi et al., 2007). We have shown in Hq mice that, despite variable phenotypic progression across mice, the loss of retinal ganglion cells reached up to 36% of control values in Hq mice aged 6–9 months. The decrease of retinal ganglion cell number is associated with a noticeable diminution in the density of optic fibre bundles converging to the optic nerve disc and within the optic nerve the number of axons was greatly reduced. As in the brain (El Ghouzzi et al., 2007), Hq retinas and optic nerves exhibited a strong gliosis reaction that occurred early in the course of retinal ganglion cell degeneration. We also demonstrated a marked activation of microglia in Hq optic nerves. In this respect, Hq optic nerve pathology bears a resemblance to rodent models of glaucoma induced by acute intraocular hypertension; in these animals an early and widespread GFAP overexpression in both retina and brain occurred (Zhang et al., 2009) as well as a marked activation of microglia in the retina, optic nerve and optic tract (Ebneter et al., 2010). It can be hypothesized that impairment of mitochondrial function is a key event during the course of glaucoma as suggested recently in the DBA/2 J mouse model. Hence, their optic nerves possess a lower bioenergetic reserve that correlates with exposure to raised intraocular pressure (Baltan et al., 2010). We also demonstrated that the deleterious impact of AIF depletion in Hq optic nerves starts in young animals, as the respiratory chain complex I defect was significantly reduced in mice younger than 8 months relative to age-matched controls (∼50% of control values); the deficiency reached its greatest level in mice older than 8 months (∼30% activity relative to controls). The significance of these observations is 2-fold: (i) depletion of the AIF protein in retinal ganglion cells leads to early cell death correlated with the disappearance of optic fibres; and (ii) when animals presented up to 36% of retinal ganglion cell loss, in the residual axons the respiratory chain complex I defect was aggravated, suggesting that, eventually, the majority of retinal ganglion cells and their axons could completely disappear. Retinal ganglion cell injury, optic nerve degeneration and respiratory chain complex I defect certainly contributed to vision loss; indeed, visual capabilities of Hq mice progressively decline with age and 1-year-old mice can be considered completely and irreversibly blind. Moreover, in adult Hq mice (6–8 months of age); myelin basic protein expression is strongly diminished in optic nerves, indicating an impaired mylination process that may cause disruption of the optic nerve integrity and contribute to visual function loss.

Interestingly, patients harbouring an AIF mutation (Ghezzi et al., 2010) had abnormal visual evoked potentials: a delay in conduction and abnormal cortical waves with distortion of the recorded potentials (G. Uziel and M. Zeviani, personal communication). Since visual evoked potential measures the conduction of the visual pathway from the optic nerve to the occipital cortex, abnormal amplitudes can reflect retinal ganglion cell loss and optic atrophy, as observed in Hq mice. Our data on retinal ganglion cell physiopathology and onset were an incentive to envisage a gene therapy in Hq eyes before the disappearance of retinal ganglion cells and their axons become measurable.

A number of conclusions emerge from our successful attempt at gene therapy in Hq mouse optic atrophy. First, we strengthened our previous observation that the specific sorting of the ‘therapeutic gene’ messenger RNA to the mitochondrial surface works efficiently to ensure the delivery of the therapeutic protein inside the organelle. This is particularly obvious in the case of AIF, whose abnormal cytosolic distribution is known to readily trigger apoptosis (Scovassi et al., 2009). In keeping with this, since AIF messenger RNAs localize preferentially to the mitochondrial surface in mouse brain, we have designed an AAV2/2 vector specifying the AIF1 protein but also possessing the full-length 3′-UTR of the AIF gene, to ensure the sorting of the corresponding messenger RNA to the mitochondrial surface. Then by using serotype 2 we confirmed numerous reports demonstrating that it displays a high retinal ganglion cell transduction efficiency after intravitreal injection (Harvey et al., 2002; Leaver et al., 2006; MacLaren et al., 2006; Hellstrom et al., 2009). Thus, the present study describes the first successful example of the ocular targeting of a mitochondrial protein in the context of AAV-mediated gene delivery. AIF1 gene expression driven by the vector appears stable for up to 5 months post-injection without any negative effect on retinal architecture or function. AIF1 messenger RNA steady-state levels in transduced retinas increased depending on the vector doses used; the low dose induced an ∼14% increase in the overall messenger RNA AIF1 abundance while with the highest dose AIF1 messenger RNA reached ∼25% of the value measured in control mouse retinas. Consequently, we assume that retinal ganglion cell transduction yield was high and generated sustained AIF1 gene expression. Importantly, AIF1 transgene expression was effective in: (i) improving retinal ganglion cell survival; (ii) preserving nerve fibre integrity; (iii) reducing microglial activation and gliosis reaction; and (iv) stabilizing respiratory chain complex I function in optic nerves. Undeniably, retinal ganglion cell loss was almost completely prevented since eyes subjected to AAV2/2_AIF1 administration presented a retinal ganglion cell population reaching ∼89% of the value measured in control mice. Moreover in optic nerves from treated eyes, which axon number was clearly superior to that in untreated animals, respiratory chain complex I activity was not statistically different compared with the one measured in age-matched controls (P = 0.33; Table 3).

In conclusion, the work presented here corroborates our previous study in a rat Leber's hereditary optic neuropathy (LHON) model in which replacement gene therapy, based on messenger RNA sorting to the mitochondrial surface, was effective in preventing the deleterious effects on retinal ganglion cell and optic nerve integrity caused by a mutation in the mitochondrial ND4 gene (Ellouze et al., 2008).

With a view towards future clinical trials in patients with mitochondrial impairment, the main objective of this study was to test and validate preventive gene therapy of AIF1 driven by AAV2/2. This has proven largely successful, as the vector clearly forced expression of AIF1 in the recipient Hq mouse retinas, protected mitochondrial respiratory chain complex I function and prevented retinal ganglion cells and their axons from degeneration.

We have thus established the proof of principle that: (i) messenger RNA sorting to the mitochondrial surface represents a powerful tool for ensuring the efficient delivery of therapeutic proteins inside the organelle; and (ii) the prevention of OXPHOS failure is sufficient to protect against retinal ganglion cell and optic nerve degeneration in Hq mouse strain; which is a faithful model of human disease. Pivotally, since ocular involvement is a frequent feature in mitochondrial diseases (Fraser et al., 2010; Yu-Wai-Man et al., 2010) and since clinical trials based on AAV-mediated gene therapy for retinal dystrophies are ongoing in three laboratories (Bainbridge et al., 2008; Maguire et al., 2008; Cideciyan, 2010) our study opens the door to counteracting mitochondrial dysfunction in future clinical studies.


INSERM; the CNRS; l'Agence Française pour la Recherche (ANR); l'Association Française contre les Myopathies (AFM); l'Agence Française contre les Maladies Mitochondriales (AMMi); la Fondation Voir et Entendre.

Supplementary material

Supplementary material is available at Brain online.


The authors thank Dr Aubrey de Grey for careful reading and feedback on final version of the manuscript. We are grateful to David Godefroy and Stéphane Fouquet (Cellular Imaging Facility of the Institut de la Vision), respectively, for their help in Nanozoomer® slides scanning, confocal laser scanning microscopy and for useful discussions and comments on the histological evaluations of the study.


  • *These authors contributed equally to this work.

apoptosis inducing factor
glial fibrillary acidic protein
green fluorescent protein
polymerase chain reaction


View Abstract