Multimodal Analysis of the Visual Pathways in Friedreich's Ataxia Reveals Novel Biomarkers

Optic neuropathy is a near ubiquitous feature of Friedreich's ataxia (FRDA). Previous studies have examined varying aspects of the anterior and posterior visual pathways but none so far have comprehensively evaluated the heterogeneity of degeneration across different areas of the retina, changes to the macula layers and combined these with volumetric MRI studies of the visual cortex and frataxin level.

neurological function, optical coherence tomography to determine retinal nerve fibre layer thickness and macular layer volume and volumetric brain MRI. Results: We demonstrate that frataxin level correlates with peripapillary retinal nerve fibre layer thickness and that retinal sectors differ in their degree of degeneration. We also shown that retinal nerve fibre layer is thinner in FRDA patients than controls and that this thinning is influenced by the AAO and GAA1. Furthermore we show that the ganglion cell and inner plexiform layers are affected in FRDA. Our MRI data indicate that there are borderline correlations between retinal layers and areas of the cortex involved in visual processing.
Conclusion: Our study demonstrates the uneven distribution of the axonopathy in the retinal nerve fibre layer and highlight the relative sparing of the papillomacular bundle and temporal sectors. We show that thinning of the retinal nerve fibre layer is associated with frataxin levels, supporting the use the two biomarkers in future clinical trials design. ©

Introduction
Friedreich's ataxia (FRDA) is the commonest autosomal recessive ataxia affecting about 1 in 30,000 individuals in Western Europe. The causative mutation in most patients is a GAA triplet repeat expansion in intron 1 of the FXN gene. Disease onset is usually in the second decade of life and correlates with the length of the shorter allele (GAA1). Clinical features include spinocerebellar and sensory ataxia, dysarthria, hypertrophic cardiomyopathy and scoliosis. 1 FRDA progresses slowly over decades with increasing dependence on assistance with activities of daily living.
FRDA affects both the afferent and efferent visual pathways. Ocular motor abnormalities resulting from disruptions in the brainstem-cerebellar circuitry are well documented, the most frequently encountered being square wave jerks followed less frequently by horizontal nystagmus, 'broken' smooth pursuit along with hypo-and hypermetria of saccades. 2 Optic atrophy is seen in 30.4% of patients 3 though this seldom results in appreciable visual loss. The anatomical correlate of visual loss is pathology of the anterior visual pathways with thinning of the retinal nerve fibre layer, and sometimes optic tract thinning. 4 Optical coherence tomography (OCT) has been applied to many neurodegenerative conditions. 5 It is rapid, safe, reproducible and provides cross-sectional and volumetric measurements of the macula and optic nerve head with enough resolution to measure the thickness of the individual retinal layers. 6 Studies of FRDA using OCT have consistently demonstrated thinning of the peripapillary retinal nerve fibre layer (pRNFL) and that this thinning correlates with ataxia severity. 7-9 Macular volume was found to be normal in one study, 7 while two others reported it to be reduced in FRDA. 10,11 Different macular layers have been found to be affected in various neurodegenerative diseases 12 and it has been proposed that these retinal changes may be employed as monitoring biomarkers. 13 In this study we investigate the visual system from the macular layers examined by OCT to the occipital cortex assessed by volumetric MRI, combining these with visual functions tests to better define the changes seen in FRDA. We then explore the relationship of the various measures with clinical scales and determinants of disease severity, and thus suitability as biomarkers.

Subjects & Ethical Approval
Sixty-two genetically confirmed FRDA patients aged over 16 years old with no other neurological diagnoses were recruited at the Ataxia Centre of London, based at the National Hospital for Neurology and Neurosurgery (NHNN) (n = 53), and in Oxford at the John Radcliffe Hospital neurogenetics clinic (n = 9). All patients had their eyes examined by a consultant neuro-ophthalmologist and those with significant ocular pathology excluded from the study. We recruited 42 healthy control subjects, 29 of which had an OCT and 13 an MRI. Subjects were recruited between July 2016 and October 2019. Ethical

Clinical Investigations
Severity of ataxia was quantified using the Scale for Assessment and Rating of Ataxia (SARA) (0 to 40), with higher values indicating more severe ataxia. 14 Extra-cerebellar involvement was assessed using the Inventory of Non-Ataxic Signs (INAS), which provides a count of non-ataxia signs such as changes in reflexes and other motor, sensory or ophthalmic signs. 15 Functional impairment was assessed using the Activities of Daily Living (ADL) questionnaire from the modified Friedreich Ataxia Rating Scale (FARS). 16 Clinical examinations were performed by GTB at the London site and AHN at the Oxford site. All patients had their fundi examined by a consultant ophthalmologist (FB at the London site, SMD at the Oxford site).

Genetic Characterisation
DNA samples from all patients were sent to the NHNN neurogenetics laboratory. These were analysed with long polymerase chain reaction (PCR) as well as a triplet primed PCR in accordance with the European Molecular Genetics Quality Network best practice guidelines for molecular analysis of FRDA (http://www. emqn.org/emqn/digitalAssets/0/234_FRDA.pdf).

Frataxin Analysis
All patients had 12 ml of blood drawn for peripheral blood mononuclear cell (PBMC) isolation using density gradient media (Ficoll-Paque PLUS (GE Healthcare, Chicago, IL, USA)). Samples were stored at À80 C and batch processed at Imperial College London, using the Abcam Frataxin Protein Quantity Dipstick Assay Kit (Abcam, Cambridge, MA, USA) according to the manufacturers instructions. All samples were analysed in triplicate. Capture zones on developed dipsticks were quantified with a Hamamatsu immunochromato reader (MS1000 Dipstick reader).

Assessment of Visual Function
Monocular best corrected visual acuity (BCVA) was measured using the logarithm of the Minimum Angle of Resolution (logMAR) chart at four meters. Colour vision was checked with the 17 plate Ishihara test. Visual fields were assessed by Goldmann manual kinetic perimetry using the isopters I4e, III4e and V4e. Volumes of islands of vision were quantified using the method described by Christoforidis. 17

Retinal Structure
Optical coherence tomography using Fourier Transform Spectral Domain OCT with TruTrack software (Heidelberg Spectralis, Heidelberg, Germany) was carried out in both eyes of all patients. All the tests were performed by one operator (GTB) at the London/ NHNN and one operator (JB for OCT and JKJ for Goldmann) at the Oxford site.
pRNFL thickness measurements were made by conducting circle scans at a scanning angle of 12 , assuming a standard corneal curvature of 7.7 mm; thus, projecting a 3.5 mm diameter circle onto the retina. This full RNFL circle scan contained 768 A-scans along a peripapillary circle of 360 . The volumetric macular protocol of the Spectralis SD-OCT device was used to measure macular thickness and volume. This uses an internal fixation source and centres on the patient's fovea. The protocol consists of 25 vertical line scans at a resolution of 1536 (scanning angle: 20 Â 20 , density: 240 μm, 4.7scans/s, automatic real-time frames: >11). pRNFL and retinal acquisitions were obtained using TruTrack eye-tracking technology that recognises, locks onto and follows the patient's retina during scanning. We employed the OSCAR IB OCT quality control criteria 18 and OCT data were collected and reported according to APOSTEL guidelines. 19

OCT Analysis
Scans with a quality score < 15 (the minimum standard recommended by the manufacturer and OSCAR IB) were excluded. pRNFL was divided by the Spectralis software into seven regions: temporal superior, temporal, temporal inferior and nasal superior, nasal, nasal inferior, papillomacular bundle and overall pRNFL thickness was also provided (referred to hereafter as pRNFL global) (Fig. 1A). Automatic layer segmentation of the macula was applied, via Heidelberg software v.1.10.20, to compute the volume and thickness of the layers investigated (Fig. 1C). The retinal thickness map analysis protocol on the Spectralis SD-OCT is divided into inner, middle, and outer rings with diameters of 1.0 mm, 2.22 mm, and 3.45 mm centred on the fovea (Fig. 1D). Scans were reviewed to rule out gross segmentation abnormalities and the subject was rescanned if willing or manually corrected by an ophthalmologist not connected to the study if not.

MRI Acquisition and Analysis
Subjects underwent brain MRI on a 3 T Philips Ingenia CX scanner (Philips Healthcare, Best, Netherlands) equipped with a maximum gradient strength of 65 mT/m. For reception of the signal from the brain, a 32-channel receive-only RF coil was used. The brain MRI protocol consisted of the following acquisitions: (1)  (optic chiasm, cuneus, lingual gyrus, inferior occipital gyrus, middle occipital gyrus, superior occipital gyrus, occipital fusiform gyrus and occipital pole) using the GIF framework 20 and the regional volumes were adjusted for the total intracranial volume (TIV).

Statistics
Unless otherwise indicated, for pRNFL thickness the mean of both eyes was used, and data from one eye if the other's data was unavailable. Comparison of measures between patients and controls used multiple regression of the measure on a subject type indicator with age and sex covariates, and in addition, for brain MRI measures only, TIV. Associations within patients were assessed using Pearson correlation except for predictors of disease progression, for which multiple regression was used adjusting for gender and age at either baseline or onset. Diagnostic plots revealed no problematic nonnormality or heteroscedasticity. Results were reported as significant if P < 0.05. Analyses were performed in Stata 16.1 (Stata Corporation, College Station, Texas, USA).

Data Availability
All data are available on request from Prof Giunti.

Cohort Demographics
Patient and control cohort demographic details by AAO are given in Table 1. The FRDA cohort (n = 62) was 58% female with a median age of 31.9 (range 17.0, 73.4). Fifty-three patients were recruited from the London site and nine from the Oxford site. One person carried a point mutation (c.11_12dupTC, p. (Gly5fs)) and one expanded allele. As the table suggests, AAO decreased with greater GAA1 length, and this was statistically significant: in fact AAO was 34.0, 21.8, 12.5 and 11.3 years respectively in GAA1 categories ≤200, 201-500, 501-800, >800 GAA triplets, P < 0.001. Age at diagnosis and age at baseline were similarly associated with GAA1. There was no evidence of GAA1 association with ADL (r = 0.01, P = 0.947) or INAS (r = 0.11, P = 0.402), but greater GAA1 length was associated with higher SARA score (r = 0.29, P = 0.028).
Subject characteristics for those patients and controls that underwent additional MRI scanning are presented in Table S1.

Visual Data
Twenty-eight patients (44.4%) wore glasses to correct refractive error and eight (12.6%) had optic disc pallor on fundoscopy. Table 2 shows visual data by AAO. No scotomata were identified during kinetic perimetry. No patient had deficient colour vision as assessed by the 17 plate Ishihara test and none were categorised as blind or having low vision by World Health Organization definitions (respectively, VA > 1.3 logMar and VA > 0.6 to 1.3 logMAR in the better eye or in either eye) ( Table 2).

Macular data
In the macula, the volumes of the GCL and IPL were significantly lower than those of controls (Table 3). There was no difference between control and patient IRL and ORL.
The GCL, IPL and RNFL are anatomically related, and correlation coefficients between the RNFL and macula layers were both above 0.5, and between the two macula, 0.9, all three with P < 0.001.

Anatomical Measures and Predictors of Disease Progression
Younger AAO was associated with thinner pRNFL global, r = 0.48, P < 0.001, (Fig. 2). Adjusting for age at baseline and gender, longer disease duration (DD) was associated with thinner pRNFL global (DD regression coefficient À 0.59, P = 0.002; age at baseline 0.44, P < 0.001), the association with DD disappeared (À0.15, P = 0.356) when instead adjusting for AAO, suggesting that AAO rather than DD (or age at baseline) was the active predictor. Gender was not significantly associated with pRNFL global. Thinner pRNFL global was also associated with greater GAA1, r = À0.38, P = 0.004. Higher frataxin levels were associated with a higher pRNFL global: 1mAU higher Optic disc pallor on fundoscopy (n) 4 3 1 8 Other ocular pathology (n) 1 0 0 1 frataxin was associated with an increase in pRNFL global thickness of 0.02 μm, P = 0.035.
There was no evidence of association between either BCVA or Ishihara score, and either AAO, disease duration, gender or GAA1. Higher BCVA, however, was associated with both higher SARA (r = 0.31, P = 0.020) and ADL (r = 0.31, P = 0.022) scores. There was no evidence of association between GVF I4e and gender, AAO or GAA1. Higher GVF I4e was associated with shorter DD (r = À0.42, P = 0.001) and though there was a univariable association with age at baseline, this disappeared in a model with DD. Higher GVF I4e was also associated with lower SARA (r = À0.50, P < 0.001) and ADL (r = À0.48, P < 0.001) scores. The same pattern was seen with GVF III4e, though with slightly weaker associations; but there was no evidence of these associations with v4e.

Correlation between Retinal Measurements and MRI Findings
The volumes of areas of the brain involved in vision (see Methods) were compared between patients and controls, adjusting for age, sex and TIV. There was no evidence of difference except for a 10.1% higher volume of the left occipital pole in patients (P = 0.020).
To investigate whether degeneration of the anterior visual pathway was associated with cortical regions involved in visual processing, correlations between mean pRNFL global, GCL, IPL and IRL volumes and the mean volumes of the aforementioned brain regions were examined. Borderline significant positive correlations were found between the mean lingual gyrus volume and the pRNFL global (r = 0.43, P = 0.073), the superior occipital gyrus and GCL (r = 0.45, P = 0.061) and IPL (r = 0.42, P = 0.079) and the occipital pole and IRL (r = 0.42, P = 0.079).

Discussion
In this study we provide evidence that the fundamental cause of Friedreich's ataxia, and therefore key target engagement biomarker, the frataxin protein level correlates with the RNFL. This finding has obvious implications for future clinical trials, especially given the fact that there are compounds, and indeed gene therapy approaches, with the aim of augmenting frataxin levels now progressing towards phase II/III trials. These findings underline the importance of targeting frataxin levels.
The multimodal approach used in our study of the visual system in the largest FRDA cohort has allowed us to deep phenotype this relevant anatomical system. We first confirmed the association between the predictors of disease progression AAO and GAA1, and pRNFL thickness. We then confirmed relationship between pRNFL and the measures of disease severity, SARA, ADL and INAS. Given the above, the RNFL may be an informative imaging biomarker in future FRDA clinical trials. Central vision (measured by LogMAR BCVA) and colour vision in our cohort was well preserved despite the loss of cells in the GCL and their axonal projections from the macula. Conditions with a similar loss of cells from the retina such as multiple sclerosis 21 present with much more severe changes in visual acuity. The sparing of central vision may be related to the milder degeneration in the PMB than in other areas of the retina, and indeed in our study the mean standardised difference was lowest in the temporal sector followed by the PMB. Alternatively, this may represent a degree of redundancy built into the visual pathway so that central visual acuity is rather robust to neuronal loss, i.e. loss of acuity is only seen when neuronal loss dips below a threshold level. In contrast, peripheral vision was seen to be significantly affected with concentric reductions in the total volume of vision correlating with disease duration, pRNFL global, GCL and IPL. Our results corroborate the data of Fortuna et al. 22 concerning loss of peripheral vision but take their findings further by demonstrating the relationship between visual field volume, DD, SARA and ADL scores.
Though all pRNFL sectors are affected in FRDA, there is significant heterogeneity between them (Table 3). We demonstrate that it is the superior and inferior sectors that are most affected with relative sparing of the temporal sector and PMB, confirming the findings of smaller studies. 8,22,23 Our findings are similar to other neurodegenerative conditions e.g. multiple system atrophy 24 and Alzheimers disease, that also preferentially spare the temporal retina. 5 Interestingly we find that the mean of the temporal superior RNFL sector has a better correlation with SARA score than any other sector or the mean of all sectors (pRNFL global). Our data suggest that this sector could be of value as a biomarker in future longitudinal studies of RNFL thickness.
As the pattern of superior and inferior pRNFL thinning is seen in conditions that predominantly affect the magnocellular cells of the GCL [24][25][26] we propose that the loss of volume in the GCL in FRDA may be driven by magnocellular cell loss. To our knowledge only one study has looked at post mortem histopathology of the retina in FRDA; they reported ganglion cell loss but did not comment on whether there was one cell subtype preferentially affected. 27 Further histopathological studies of FRDA retinas would resolve this question.
In many neurodegenerative conditions, a reduction in volume of the GCL and IPL is observed 28 and we demonstrate that FRDA is no exception. Our study analysed the individual layers of the macula and found that the volumes of GCL and IPL were significantly decreased compared to controls (Table 3) and strongly correlated with each other and the RNFL. This is unsurprising given the anatomical links between the layers with retinal ganglion cells receiving inputs from the axons of bipolar cells and amacrine cell processes in the IPL. Interestingly, in light of the fact that GCL and IPL both correlate well with measures of disease severity, we did not find the volume of either layer to be influenced by any known predictor of disease severity.
Many conditions that cause a thinning of the retina also result in thinning of the cortex [29][30][31] and indeed this has been observed in FRDA. 32,33 Though this was not seen in our study, in some areas the relative difference between patients and controls was high, with the right cuneus and right lingual gyrus 14.9% and 10.1% lower respectively. These differences however were not significant when adjusted for age, sex and TIV (P = 0.096 and P = 0.092, respectively). Of note, the left occipital pole was significantly larger in patients (P = 0.022), possibly indicating a compensatory hypertrophy of this area in response to the loss of input from non-macula areas of the retina. Despite our small sample size, a trend towards positive correlations were found between cortical areas and OCT measures. The moderate correlations seen indicate that these results may well achieve significance were more subjects to be enrolled.
There were some limitations to our study, firstly it must be noted that some of our subjects were unable to read the chart not due to issues with visual acuity but rather due to fixation instability. We therefore may overestimate the extent to which retinal disease leads to loss of central visual acuity in some of our cohort. This fixation instability and the presence of square wave jerks in some patients also made the OCT examination technically more challenging and therefore longer then normal to perform. In total only 3 out of 65 (4.6%) patients were unable to be assessed with OCT.
The data presented in this study furthers what is currently known about the retina and its connections in FRDA. Our data highlight new biomarkers for exploration in clinical trials and reveal novel findings about the visual pathway in FRDA.

Conclusions
We demonstrate for the first time the positive association between frataxin levels, a target engagement T H E V I S U A L P A T H W A Y I N F R I E D R E I C H ' S A T A X I A biomarker, and pRNFL thickness. We then provide the most detailed description yet of the distribution of the axonopathy in the RNFL and highlight the relative sparing of the PMB and temporal sectors giving an anatomical basis for the preservation of central visual acuity seen in our cohort. Our sectoral analysis of the pRNFL demonstrates differential thinning across sectors and we demonstrate that the temporal superior sector is the most well correlated with the SARA score. These data combined with the tolerability, low cost and widespread availability of OCT mark it out as a very strong candidate for use as a surrogate endpoint in future trials. We confirm the long-suspected presence of a retinal neuronopathy by analysing the different layers found within the macula and demonstrate their relationship with clinical and visual measures. Furthermore, our MRI data demonstrate changes to the occipital lobe in patients and the possible connection between the volumes of cortical and retinal areas. These data combined with our OCT findings raise interesting questions concerning the pattern of degeneration in the visual pathway. Future studies looking at longitudinal change in these metrics and their relationship to other features of the disease will be of great importance in defining their suitability as biomarkers for future clinical trials.
T H E V I S U A L P A T H W A Y I N F R I E D R E I C H ' S A T A X I A