Toward quantitative neuroimaging biomarkers for Friedreich's ataxia at 7 Tesla: Susceptibility mapping, diffusion imaging, R 2 and R 1 relaxometry

Abstract Friedreich's ataxia (FRDA) is a rare genetic disorder leading to degenerative processes. So far, no effective treatment has been found. Therefore, it is important to assist the development of medication with imaging biomarkers reflecting disease status and progress. Ten FRDA patients (mean age 37 ± 14 years; four female) and 10 age‐ and sex‐matched controls were included. Acquisition of magnetic resonance imaging (MRI) data for quantitative susceptibility mapping, R 1, R 2 relaxometry and diffusion imaging was performed at 7 Tesla. Results of volume of interest (VOI)‐based analyses of the quantitative data were compared with a voxel‐based morphometry (VBM) evaluation. Differences between patients and controls were assessed using the analysis of covariance (ANCOVA; p < 0.01) with age and sex as covariates, effect size of group differences, and correlations with disease characteristics with Spearman correlation coefficient. For the VBM analysis, a statistical threshold of 0.001 for uncorrected and 0.05 for corrected p‐values was used. Statistically significant differences between FRDA patients and controls were found in five out of twelve investigated structures, and statistically significant correlations with disease characteristics were revealed. Moreover, VBM revealed significant white matter atrophy within regions of the brainstem, and the cerebellum. These regions overlapped partially with brain regions for which significant differences between healthy controls and patients were found in the VOI‐based quantitative MRI evaluation. It was shown that two independent analyses provided overlapping results. Moreover, positive results on correlations with disease characteristics were found, indicating that these quantitative MRI parameters could provide more detailed information and assist the search for effective treatments.


| INTRODUC TI ON
Friedreich's ataxia (FRDA) is a rare genetic disorder affecting the FXN gene which produces the protein frataxin within the mitochondria of the cells (Koeppen, 2011). More precisely, most patients inherit trinucleotide (GAA) expansions within the first intron of the FXN gene, and those with a higher number of repeats (>500) are more likely to present at an earlier age and with different clinical symptoms (Schols et al., 1997), whereas longer GAA repeats lead to lower frataxin levels (Sacca et al., 2011). A lack of frataxin disrupts the iron metabolism (Gibson, Koonin, Musco, Pastore, & Bork, 1996;Zhang et al., 2006) leading to degenerative processes especially in tissues with a high metabolic rate such as the heart, pancreas, as well as the nervous system (Koeppen, 2013). Within the nervous system, white matter fiber tracts, spinal nerves, and the cerebellum are affected (Lamarche, Lemieux, & Lieu, 1984). FRDA is the most common inherited ataxia presenting with early onset of clinical manifestations (Koeppen, 2011). So far, no effective treatment has been found that can stop or reverse the progress of FRDA. Therefore, it is important to assist the development of efficient medication by finding reliable imaging biomarkers that reflect the disease status and progress in FRDA patients.
An extensive body of literature of cerebral abnormalities in FRDA exists (Selvadurai, Harding, Corben, & Georgiou-Karistianis, 2018) in which functional and structural cerebral and cerebellar white and gray matter changes are assessed: Volumetric, voxel-based morphometry (VBM), and tract-based spatial statistics (TBSS) magnetic resonance imaging (MRI) studies have shown a significant atrophy of the cerebral and cerebellar cortex in FRDA patients (Cocozza et al., 2020;Selvadurai et al., 2016) and the dentate nuclei (DN;Solbach et al., 2014;Stefanescu et al., 2015), a degeneration of white matter tracts (Della Nave et al., 2008;Pagani et al., 2010;Selvadurai et al., 2020), as well as a significant correlation of those findings with disease severity or certain symptoms (Akhlaghi et al., 2011) such as the impairment of the visuospatial functions (Cocozza et al., 2020).
Another study including VBM and TBSS analyses found atrophy in several parts of the brainstem, and cerebellum, but did not observe any focal signal abnormalities on proton-density and on T 2 -weighted images (Pagani et al., 2010). A recent study (Selvadurai et al., 2020), that used multiple diffusion imaging-based metrics as well as magnetization transfer imaging to assess microstructural and T 1 -weighted imaging to assess macrostructural white matter changes found extensive disruption of cerebello-cerebral and intracerebral white matter tracts. Quantitative methods such as quantitative susceptibility mapping (QSM; Ward et al., 2019)  Ultra-high field MRI has been shown to outperform conventional clinical MRI in the delineation of very small anatomical structures (Deistung, Schafer, Schweser, Biedermann, Gullmar, et al., 2013;Straub et al., 2019), and in the detection of small pathological changes such as in epilepsy (Zeineh et al., 2009) and micro-hemorrhage (Theysohn et al., 2011) due to the higher resolution as well as the higher contrast-to-noise ratios available at higher field strengths (Ladd et al., 2018). To the best of our knowledge, no study has assessed whether quantitative MRI other than diffusion tensor imaging (DTI), magnetization transfer ratio, or spectroscopy (Blair et al., 2019) could provide complementary or additional information to evaluate independent analyses provided overlapping results. Moreover, positive results on correlations with disease characteristics were found, indicating that these quantitative MRI parameters could provide more detailed information and assist the search for effective treatments.

Significance
In five out of twelve investigated structures of the brainstem and the cerebellum, statistically significant differences were found in Friedreich's ataxia patients versus controls with quantitative magnetic resonance imaging at 7 Tesla (quantitative susceptibility mapping, R 1 , R 2 relaxometry and diffusion imaging), and statistically significant correlations with disease characteristics were revealed.
The regions in which voxel-based morphometry showed significant white matter atrophy partially overlapped with the brain regions for which significant differences between controls and patients were found in the volume-of-interest-based quantitative MRI evaluation. Quantitative MRI parameters could provide more detailed information about the disease and assist the search for effective treatments. the current stage and the course of Friedrich's ataxia. Compared to 3 T MRI, it has been shown that 7 T MRI can provide contrast-tonoise ratios up to almost fourfold higher for certain structures in the brainstem . Utilizing the higher contrast-to-noise ratio available at ultra-high field, this study investigates the degeneration of fiber tracts throughout the brainstem as well as of the DN, the red nuclei (RN), and the substantia nigra (SN) in FRDA with quantitative MR parameters (susceptibility, diffusion anisotropy, and R 2 and R 1 relaxometry) by assessing a patient and a control group, and compares its results with a VBM analysis of the same cohorts.
QSM was chosen as a contrast due to its ability to quantify tissue iron content (Langkammer, Schweser, et al., 2012) and its sensitivity to myelin (Lee et al., 2012). R 1 and R 2 relaxometry was used since its known to be influenced by myelin and iron (Harkins et al., 2016;Koenig, 1991;MacKay et al., 2006;Ogg & Steen, 1998). DTI has long been utilized to assess white matter fiber integrity (Chanraud, Zahr, Sullivan, & Pfefferbaum, 2010), and was therefore also included as a quantitative MRI parameter.
Whether the used quantitative MRI methods provide sufficiently large effects to reveal differences between FRDA patients and healthy controls as well as correlations with disease characteristics at clinical field strengths are beyond the scope of this study and left to further investigations.

| Patient and control cohorts
This study was conducted in accordance with the Declaration of Helsinki. Institutional review board approval was obtained and all subjects provided written informed consent. Inclusion criteria were diagnosis of FRDA, physical and mental capability of providing informed consent and participating in the study. Exclusion criteria involved pregnancy and contraindications for 7 T MRI. As FRDA is a rare disease, the inclusion of 10 patients was the goal to achieve a sufficiently large patient cohort for the analyses. Ten genetically confirmed FRDA patients (mean age 37 ± 14 years; four female) and 10 age-and sex-matched healthy controls (mean age 37 ± 14 years; four female) were included in the study. Patients were recruited at the Department for Neurology, Medical University of Innsbruck, Austria, and MR images were generated at the DKFZ Heidelberg, Germany. All participants finished the study. Nine patients had homozygous GAA triplet expansions and one patient was heterozygous (one GAA triplet expansion and one point mutation). For eight patients, length of GAA triplet repeats was available. Table 1 summarizes the patient cohort.

| Data acquisition
Data were acquired at a whole-body 7 T system (Magnetom 7 Tesla) with an 8Tx/32Rx-channel head coil (Nova Medical Inc., Wakefield, MA, USA) by use of an in-house-constructed butler matrix. A monopolar three-dimensional multi-echo gradient echo (GRE), a twodimensional (2D) RESOLVE (readout segmentation of long variable echo-trains; Heidemann et al., 2010;Porter & Heidemann, 2009) with 11 shots in stimulated acquisition mode (STEAM; Merboldt, Hanicke, & Frahm, 1991;Tanner, 1970) with mixing time 50 ms and with b-values 50 and 800 s/mm 2 and 20 diffusion directions, an EPI (echo planar imaging) factor of 68, echo spacing of 0.32 ms, and fat saturation, and a vendor-provided MP2RAGE (magnetization prepared 2 rapid acquisition gradient echo; Marques et al., 2010) sequence were acquired using the same parameters as in a previous study . The RESOLVE sequence is equipped with navigator-based reacquisition which was used for the acquisition of the DTI data. For dictionary-based T 2 mapping, a 2D multiecho turbo spin echo (ME-TSE) with turbo factor 5 and a turbo flash sequence for B 1 mapping (Chung, Kim, Breton, & Axel, 2010) were acquired as well. All sequences were acquired in strictly transversal orientation with phase-encoding direction right-left. Sequence parameters are summarized in Table 2

| Quantitative susceptibility mapping
For susceptibility map generation, single-channel data were com-

| Relaxometry and diffusion tensor imaging
The longitudinal relaxation rate R 1 was calculated as R 1 = 1/T 1 from the vendor-provided T 1 maps that were acquired with the MP2RAGE sequence. The vendor-provided diffusion fractional anisotropy maps were used for the diffusion data evaluation. Maps of the transverse relaxation rate R 2 = 1/T 2 were calculated from ME-TSE data and B 1 maps using a dictionary-based method (Emmerich et al., 2019): A dictionary for the echo intensities was computed for the ME-TSE sequence according to the sequence parameters (Table 2) using a combination of Bloch simulations to compute the slice profiles and subsequently extended phase graph model (Weigel, 2015) simulations to compute the resulting signal intensities. The measured signal intensities and B 1 maps were then fitted to this precalculated dictionary to obtain T 2 values.

| Volumes of interest generation
Volumes of interest ( Figure 1)  , R 2 < 1 170 1 ms , and FA < 300, respectively. In the case of susceptibility maps, a CSF mask that was generated with FSL-FAST (Zhang, Brady, & Smith, 2001) was used for automatic correction. A manual VOI assessment ensued, as well as a correction in cases in which pixels were included in the VOI that clearly did not belong to the corresponding structure, i.e. pixels that were wrongly positioned in the CSF or belonged to a different anatomical structure. As the DTI data only covered an approximately four centimeter slab, not all fiber tracts were feasible for DTI analysis, the brain nuclei were not included in DTI analysis as well.

| Voxel-based morphometry
VBM (Ashburner & Friston, 2000;Mechelli, Price, Friston, & Ashburner, 2005) was performed using the MP2RAGE data and Statistical Parametric Mapping (SPM12, https://www.fil.ion.ucl. ac.uk/spm/softw are/spm12/) in Matlab. Following (Ashburner, 2010), first, grey matter, white matter, CSF, skull, and head were segmented using default parameters and generating the input data for the next step. Then, a template was generated from all controls and patients using the "Diffeomorphic Anatomical Registration Using Exponentiated Lie Algebra" (DARTEL) algorithm (Ashburner, 2007) with default parameters. Finally, DARTEL tools were used to normalize white matter images to the Montreal Neurological Institute (MNI) space which included smoothing with an 8-mm Gaussian kernel, spatially normalizing, and scaling by the Jacobian determinant.

| Statistical analysis
All statistical analyses were performed in Matlab, and with SPM12 for the general linear model analysis for VBM evaluation. Mean susceptibility values, relaxation rates, and mean diffusion fractional anisotropy (FA) were calculated. Differences between the patients and the healthy controls were assessed using the analysis of covariance (ANCOVA) with age and sex as covariates, and effect size of the group differences between healthy controls and patients was calculated using Cohen's d. Correlations with FRDA characteristics as well as with patient age (Table 1) were assessed using the Spearman correlation coefficient. p-Values of less than 0.01 were considered statistically significant. To analyze statistically significant differences in white matter volume from the VBM evaluation between the patient and the control group, a two-sided t-test was performed in SPM12 with age and sex as covariates, first uncorrected with a statistical threshold of 0.001, and secondly using the SPM12 familywise error correction to correct for multiple comparisons with a statistical threshold of 0.05. No data were removed prior to analysis.
Data processing was performed blinded to disease characteristics.

| Differences between FRDA patients and controls
In Table 3, the effect size for the differences between the patient and the control group is given by Cohen's d. Largest effect size (above 2) is observed for R 2 relaxometry of the RN and for diffusion fractional anisotropy of the SCPs. For mean susceptibility values of the DN, and RN, for mean R 1 values of the ICPs and SCPs, and RN, for mean R 2 values of the SCPs, CPs, RN, and SN, as well as for mean diffusion fractional anisotropy of the CS, the ICPs, CPs, the effect size is larger than 1. For mean R 1 values of five other fiber tracts (MCP, ML, CP, PTR, and SS), mean R 2 values of three other fiber tracts (MCP, ML, and PTR), and diffusion fractional anisotropy of three other fiber tracts (MCP, PCT, and ML), the effect size is larger than 0.5.
In five brain regions, which coincide with the regions of largest effect size (Table 3, Cohen's d 1.46 or larger), statistically significant differences (indicated by white stars) were found between healthy controls and patients. Figure  In R 1 maps, values for RN were significantly lower in patients and significantly higher in SCP. R 2 mapping showed significant differences between controls and patients for RN, and for one fiber tract (CP), where R 2 was significantly lower/higher. Diffusion fractional was significantly lower in patients in two fiber tracts (ICP and SCP). For SN, there was a significant correlation with age.
Moreover, the differences revealed by the statistical analysis are also visually recognizable. In Figure 3, representative slices of susceptibility, R 1 and R 2 maps, as well as fractional anisotropy maps from diffusion-weighted imaging are shown for brain regions for which significant differences between healthy controls and patients were found. Images from different patients (indicated by patient number P) and age-/sex-matched controls are shown. On susceptibility maps, conspicuous differences in DN (first row) and RN (second row) can be observed with a higher susceptibility on the patient images compared to healthy controls. On R 1 maps, higher R 1 relaxation rates are visible in the SCP (third row). Higher R 2 relaxation rates can be seen in the RN, and the CP (fourth row) on R 2 maps for the patient data.

TA B L E 3 Effects size (Cohen's d) of group differences
On fractional anisotropy maps, a lower fractional anisotropy can be observed for the SCP (fifth row) in the patient data.

| Voxel-based morphometry
Compared to the age-and sex-matched controls, statistically significant, white matter atrophy was found in the patient cohort in the uncorrected t-test analysis (p < 0.001). Figure 4 shows suprath- For the corrected t-test analysis, no statistically significant clusters were found.

| Correlation analysis
In Figure 5, the correlations between the different imaging contrasts and disease characteristics are visualized. Figure 5a shows the correlation coefficients and p-values for the correlations between age at disease onset, patient age, GAA1 repeats, GAA2 repeats (Table 1) and mean susceptibility,  Table S1.

| D ISCUSS I ON
In this study, statistically significant (p < 0.01, no correction for multiple comparisons) differences between a group of 10 FRDA patients and a group of age-and sex-matched healthy controls were found in five (CP, DN, ICP, SCP, and RN) out of twelve investigated structures of the brainstem, and the cerebellum using quantitative MRI (susceptibility, relaxation rates (R 2 , R 1 ), diffusion fractional anisotropy) at 7 Tesla. It was also shown that the effect size of group differences for susceptibility, R 2 , R 1 , and fractional anisotropy values between healthy controls and patients were medium to large (0.53-2.91) for several fiber tracts and brain nuclei, indicating that the used methods in the investigated structures could be useful biomarkers in larger studies. In the present study, only VOI/contrast combinations with an effect size of 1.46 (Cohen's d) or larger for group differences provided significant differences between patients and controls.
Moreover, using VBM, statistically significant white matter atrophy was found within regions of the brainstem, midbrain, and the cerebellum that partially overlapped with brain regions for which significant differences between healthy controls and patients were found in the VOI-based quantitative MRI evaluation part of this study. The largest overlap was observed for SCP for which significant differences between patients and healthy controls were found using R 1 relaxometry and diffusion-weighted imaging indicating white matter damage in patients. The atrophy of SCP shown in FRDA previously (Akhlaghi et al., 2011;Della Nave et al., 2008;Pagani et al., 2010) is in accordance with the findings of this study. Moreover, SCP atrophy as well as the atrophy of the peridentate white matter were suggested to correlate with disease symptoms (Akhlaghi et al., 2011;Pagani et al., 2010). Using VBM, atrophy in FRDA patients was also found in the 3 T study. Gradient echo techniques such as susceptibility mapping are much less affected by B 1 field inhomogeneities that occur particularly strong at ultra-high field. However, susceptibility values calculated with standard QSM methods depend on fiber orientation (Aggarwal, Kageyama, Li, & van Zijl, 2016), which might be the reason why no significant differences were found with QSM in this study.

F I G U R E 3
Representative axial slices of susceptibility, R 1 , R 2 , and fractional anisotropy maps (top to bottom) from four different patients and controls. The patient number is indicated in each slice. On susceptibility maps, arrows point to dentate nuclei (DN), and red nuclei (RN), on the R 1 map to superior cerebellar peduncles (SCP), on the R 2 map to the red nuclei (RN), and the cerebral peduncle (CP), and on the fractional anisotropy map arrows point to superior cerebellar peduncles (SCP) [Color figure can be viewed at wileyonlinelibrary.com] Moreover, this study evaluated if significant correlations of quantitative MRI parameters of the investigated VOIs and disease characteristics were present. For all MRI contrasts significant correlations with one or more disease characteristics were found. Mean R 1 values within the ML positively correlated with age at disease onset, as well as mean R 2 values within the CP which could indicate more white matter damage in the patients with the earlier disease onset agreeing with current literature (Selvadurai et al., 2020). Mean R 2 values in the RN positively correlated with GAA1 and GAA2 repeats which are in accordance with the relation of GAA repeat length and frataxin levels (Sacca et al., 2011). Fractional anisotropy in the ICP negatively correlated with GAA1 and GAA2 repeats which also agrees with findings in the literature (Selvadurai et al., 2020) that found a correlation of ICP volume and GAA1 repeats. For the significant susceptibility/VOI correlation pairs, a significant correlation with patient age was revealed as well. Magnetic susceptibility measured with QSM is known to be sensitive for age-dependent iron accumulation within the brain nuclei (Acosta-Cabronero, Betts, Cardenas-Blanco, Yang, & Nestor, 2016;Hallgren & Sourander, 1958;Zhang et al., 2018). In this study, this could mask any disease-related changes within the patient group as higher susceptibility values due to iron accumulation could be related to higher patient age or to increased disease severity (that is, longer GAA repeats/earlier age at onset) which cannot be differentiated with the studied cohort. As the control cohort was age-matched, significant differences could very well be observed and results consequently did not suffer from an age-bias.
This study has several limitations. The low number of included patients limits achievable statistical significance. Since FRDA is a rare disease, studies are often rather small (Akhlaghi et al., 2011;Pagani et al., 2010), however, there exist studies using 3 T MRI that include approximately 30 FRDA patients (Selvadurai et al., 2020). Moreover, the patient cohort was relatively inhomogeneous consisting of three young patients (between 18 and 23 years) with early onset (before the age of 25 years) of FRDA, three patients in their 30s as well as one patient aged 62 with early onset, and three patients older than 45 years with rather late onset (older than 25 years). Sex was also only considered as covariate, and sex-related differences were not assessed separately due to the rather low patient number; however, all patients and controls were age-and sex-matched. To further evaluate the value of quantitative MRI for assessing disease status as well as progression and correlation with clinical parameters, studies including larger homogeneous cohorts at high or ultra-high field might be helpful. Moreover, due to a stronger presence of field inhomogeneities at ultra-high field, DTI is even more prone to artifacts compared to clinical field strength, therefore, a readout-segmented EPI sequence with navigator-based reacquisition was chosen at the expense of longer acquisition times. To avoid prohibitively long acquisition times, only 20 diffusion directions were used as well as a lower coverage of the brain. JHU-ICBM-DTI-81 white matter labels were co-registered to the MP2RAGE, as these possessed the largest brain coverage, data of this study and then subsequently co-registered to all other contrasts which could be prone to registration

| CON CLUS ION
It was shown in this study that two independent analyses, a VOIbased evaluation of quantitative MRI data and a VBM analysis, provided overlapping results. Susceptibility mapping provided significant differences between FRDA patients and controls, as well as R 1 , R 2 relaxometry and diffusion imaging for two brain nuclei and three fiber tracts. Moreover, positive results on correlations with disease characteristics were found, indicating that these quantitative MRI parameters could provide more detailed information than volumetric assessment alone, and may assist the search for effective treatments.

F I G U R E 5
Correlations between quantitative MRI parameters and FRDA patients' disease characteristics as well as with patient age. (a) Correlation coefficients (ρ, first row) and p-values (p, second row) for correlations between mean volumes of interest (VOIs) values for the different imaging contrasts (susceptibility , relaxation rates R 1 and R 2 , diffusion fractional anisotropy FA) and age at onset (first column), patient age (second column), GAA1 repeats (third column), and GAA2 repeats (fourth column). White stars indicate significant correlations, green stars indicate that for this contrast/VOI combination, additionally significant differences between patients and healthy controls have been found (see Figure 2). Correlations for contrast/VOI combinations, for which no correlation with patient age was found, are marked with a black circle. (b) Scatterplots for the contrasts and VOIs for which significant correlations with disease characteristic, but not with patient age were found [Color figure can be viewed at wileyonlinelibrary.com] (a) (b)

DECL AR ATION OF TR ANS PAREN C Y
The authors, reviewers and editors affirm that in accordance to the policies set by the Journal of Neuroscience Research, this manuscript presents an accurate and transparent account of the study being reported and that all critical details describing the methods and results are present.

ACK N OWLED G M ENTS
The use of the ASPIRE gradient echo sequence and corresponding ICE program for coil combination of the 7 T GRE data provided by Korbinian Eckstein and Simon D. Robinson is kindly acknowledged.

CO N FLI C T O F I NTE R E S T S
The authors report no conflict of interest.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1002/jnr.24701.

DATA AVA I L A B I L I T Y S TAT E M E N T
The authors take full responsibility for the data, the analyses and interpretation, and the conduct of the research and have full access to all of the data, of which we have the right to publish any and all data in the absence of a sponsor. Anonymized data, not published in the article, will be shared on reasonable request from a qualified investigator.

R E FE R E N C E S S U PP O RTI N G I N FO R M ATI O N
Additional supporting information may be found online in the Supporting Information section.
Table S1 All Spearman correlation coefficients and p-values for