Progressive alterations in white matter microstructure across the timecourse of Huntington's disease

Abstract Background Whole‐brain longitudinal diffusion studies are crucial to examine changes in structural connectivity in neurodegeneration. Here, we investigated the longitudinal alterations in white matter (WM) microstructure across the timecourse of Huntington's disease (HD). Methods We examined changes in WM microstructure from premanifest to early manifest disease, using data from two cohorts with different disease burden. The TrackOn‐HD study included 67 controls, 67 premanifest, and 10 early manifest HD (baseline and 24‐month data); the PADDINGTON study included 33 controls and 49 early manifest HD (baseline and 15‐month data). Longitudinal changes in fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity, and radial diffusivity from baseline to last study visit were investigated for each cohort using tract‐based spatial statistics. An optimized pipeline was employed to generate participant‐specific templates to which diffusion tensor imaging maps were registered and change maps were calculated. We examined longitudinal differences between HD expansion‐carriers and controls, and correlations with clinical scores, including the composite UHDRS (cUHDRS). Results HD expansion‐carriers from TrackOn‐HD, with lower disease burden, showed a significant longitudinal decline in FA in the left superior longitudinal fasciculus and an increase in MD across subcortical WM tracts compared to controls, while in manifest HD participants from PADDINGTON, there were significant widespread longitudinal increases in diffusivity compared to controls. Baseline scores in clinical scales including the cUHDRS predicted WM microstructural change in HD expansion‐carriers. Conclusion The present study showed significant longitudinal changes in WM microstructure across the HD timecourse. Changes were evident in larger WM areas and across more metrics as the disease advanced, suggesting a progressive alteration of WM microstructure with disease evolution.

and across more metrics as the disease advanced, suggesting a progressive alteration of WM microstructure with disease evolution.

K E Y W O R D S
diffusion tensor imaging, Huntington's disease, longitudinal, presymptomatic, symptomatic INTRODUCTION Loss of white matter (WM) organization is a key feature of Huntington's disease (HD), with myelin thinning, reduced expression of myelin-related genes, andincreased density of oligodendrocytes in the tail of the caudate nucleus occurring early on (Gómez-Tortosa et al., 2001;Teo et al., 2016;Xiang et al., 2011). Morphological magnetic resonance imaging (MRI) studies provide clear evidence of progressive striatal atrophy that rapidly extends to WM during premanifest (pre-HD) stages (Tabrizi et al., 2011(Tabrizi et al., , 2012. However, investigation at the microstructural level can help characterize the changes in WM organization and their relationship to the HD phenotype. Diffusion tensor imaging (DTI) is commonly used to infer the coherence of WM tracts and WM organization in vivo. Although biological properties of brain tissue, such as the presence of crossing fibers, limit their interpretation, DTI metrics remain the most widely used method to characterize WM microstructure (Alexander et al., 2007).
Cross-sectional studies using DTI have provided robust evidence of disorganization in deep and superficial WM (Casella et al., 2020;Filippi & Agosta, 2016;Phillips et al., 2014;Wu et al., 2017) and the corpus callosum (Di Paola et al., 2012;Rosas et al., 2010) in HD many years before symptom onset (Liu et al., 2016). Cross-sectional analyses of diffusion metrics have also shown correlations with motor, cognitive, and functional scales .
The composite UHDRS (cUHDRS) is a multidimensional measure of progression in HD encompassing functional, cognitive, and motor subscales that has been used as a primary outcome in the phase 3 clinical trial with tominersen, an antisense oligonucleotide targeting HTT mRNA. A recent study demonstrated a significant correlation between the cUHDRS and diffusivity metrics ).
An increasing number of clinical trials in HD are using diffusion imaging measures as biomarkers for drug efficacy and safety, as well as to understand the mechanisms underlying treatment response (Tae et al., 2018). Therefore, longitudinal investigation of WM microstructure is paramount. Apart from correlations with clinical function, investigating the exact point at which a certain biomarker is sensitive and determining the longitudinal sensitivity over short periods of time are crucial to defining the role of diffusion imaging metrics in clinical trials.
Unfortunately, only a limited number of studies have evaluated change in WM microstructure over time, with findings generally mixed Hobbs et al., 2015;Poudel et al., 2015;Sritharan et al., 2010;Vandenberghe et al., 2009;Weaver et al., 2009). However, evidencing significant longitudinal change over a clinical trial period is a necessary step toward showing change in the natural trajectory of a biomarker in response to a therapy.
It is likely that due to the methodological limitations analyzing multiple time points of diffusion data, the subtle changes that can occur in a slowly progressive degenerative disease over time may remain undetected. In the current study, therefore, we have employed a longitudinal pipeline that seeks to reduce misalignment between multiple time point scans by creating individual participant templates to which diffusion data from multiple visits are registered; this then ensures that the location of voxels is consistent across the data (Engvig et al., 2012).
We investigated WM microstructural change over the trajectory of HD, analyzing longitudinal diffusion data from two HD cohorts with different disease burden: TrackOn-HD, composed mainly of premanifest HD (pre-HD) and controls, with change over 24 months; and PADDINGTON, including early manifest HD and controls, with change over 15 months. For each cohort, we used the novel longitudinal pipeline in conjunction with whole-brain voxelwise tract-based spatial statistics (TBSS) (S. M. Smith et al., 2006) to compare change in diffusion metrics between HD expansion-carriers and controls. We also correlated baseline scores in clinical measures with change in diffusion metrics to test the predictive ability of these scales for microstructural degeneration. We hypothesized that HD expansioncarriers would display evidence of progressive WM disorganization from pre-HD to manifest HD, coupled with a clear correlation between baseline function and WM degeneration.

Participants
Participants with DWI data for baseline and final (third) visits were recruited from the TrackOn-HD (time interval 24 months) and PADDINGTON (time interval 15 months) cohorts (Hobbs et al., 2013;Klöppel et al., 2015).

TrackOn-HD
The TrackOn-HD cohort recruited 239 participants (106 pre-HD, 22 early HD, and 111 controls), from four study sites (London, Leiden, Paris, and Vancouver) evaluated annually over a period of 2 years. Gene expansion carriers were required to have ≥40 CAG repeats in the HTT gene and a disease burden score >250 (Penney et al., 1997). Control participants were gene-negative volunteers and family members. At  Figure S1). There were no statistically significant differences between included/excluded participants.

TA B L E 1 Baseline demographics of participants
Given the small number of early HD participants, we combined premanifest and manifest HD participants into a single group of HD gene-expansion carriers. We grouped together pre-HD and manifest HD participants as imaging and histological studies have shown that pathology in HD is progressive, rather than dichotomic (Tabriz et al., 2009;Vonsattel et al., 1985). In addition, differences in disease severity between the examined cohorts is better quantified through the disease burden score (Penney et al., 1997), showing significant differences between TrackOn-HD and PADDINGTON (Table 1).  Figure S2). There were no statistically significant differences in baseline characteristics between included/excluded participants.

Clinical scales
For our correlation analysis, we used two cognitive scales: Symbol-Digit Modalities Test (SDMT) (Parmenter et al., 2007)

and Stroop Word
Reading (SWR) (Stroop, 1935), one functional scale, the TFC, and one motor scale, the UHDRS-Total Motor Score (TMS) (Huntington Study Group, 1996). In addition, we performed correlations with the disease burden score, a function of age and CAG repeat length (disease burden score = Age × (CAG − 35.5)), which measures the amount of time that a subject has been exposed to the effects of mHTT (Penney et al., 1997) and with the cUHDRS.
The cUHDRS is a multidomain scale including the SDMT, SWR, TFC, and TMS (Equation 1), which tracks disease progression and has excellent sensitivity to disease stage (Schobel et al., 2017). Higher scores in all scales except for TMS and disease burden score indicate better performance.

MRI acquisition
Scanning protocols were standardized between sites, and inter-

Data preprocessing
DWI images were first brain extracted with FSL BET and motioncorrected using eddy in FSL (www.fmrib.ox.ac.uk/fsl) (Andersson & Sotiropoulos, 2016). FSL DTIFIT was used to apply the diffusion ten- The templates and skeletons were created and analyzed in each cohort, grouping together healthy controls and expansion carriers to investigate the differences between HD expansion carriers and controls.
F I G U R E 1 Summary of processing pipeline. Raw FA data from the two visits of all subjects in each study were registered using FSL FLIRT to create halfway images. These images were averaged to create a subject-wise FA mid-space template. TBSS automatically aligned all mid-space images to standard space. The warped FA templates were used to create the mean FA map. This map was thresholded at FA >0.2 to create the midspace FA skeleton for all subjects. All halfway image skeletons from the two visits for all DTI metrics were projected to the standardized FA skeleton. Finally, the skeleton for visit 3 was subtracted from the skeleton for visit 1 and statistical analysis was performed on the subtraction images to compare between HD gene expansion carriers and controls. The same process was repeated to obtain a template only in HD expansion carriers from each study in order to run the correlations in the subtracted images (adapted and reproduced with permission from Engvig et al., 2012). DTI, Diffusion Tensor Imaging; FA, fractional anisotropy; HD, Huntington's disease; TBSS, tract-based spatial statistics.
The same process was performed exclusively in HD gene expansion carriers for the evaluation of correlations between clinical scales and structural connectivity.

Statistical analysis
Demographic variables between groups were evaluated using t-tests for mean comparisons between the two groups. One-way analysis of variance (ANOVA) was used for mean comparisons between more than two groups. Chi-squared test was used to compare proportions where appropriate. Statistical analysis of demographic variables was performed using Stata v12.0 (StataCorp, College Station, TX, USA).
All analyses were performed using non-parametric permutation-

RESULTS
Both cohorts had similar mean age in the control groups. Age in the gene-expansion groups was significantly higher in PADDINGTON, which is consistent with manifest HD participants being on average older than pre-HD. There were no differences in male/female proportions between controls and expansion-carriers in either cohort. Mean number of CAG repeats was comparable between both cohorts, while as expected, expansion carriers from TrackOn-HD had significantly lower disease burden score and lower TMS values than expansion carriers from PADDINGTON.

Group differences
In the TrackOn-HD cohort, expansion-carriers compared to controls

Correlations with clinical scales
For the TrackOn-HD cohort, there was a significant positive correlation between MD, AD, and RD and baseline TMS scores, indicating that larger increases in diffusivity were associated with higher TMS scores (worse motor symptoms) at baseline ( Figure S3). Brain regions associated with the TMS involved the corpus callosum and corona radiata bilaterally ( Figure S3). There was no correlation between FA and TMS scores. Baseline SWR was negatively correlated with MD in the genu of the corpus callosum and left anterior corona radiata. There were no significant correlations between any DTI metric and disease burden score, SDMT, TFC, or cUHDRS in TrackOn-HD.
For the PADDINGTON cohort, there was a significant, counterintuitive positive correlation between FA in the right posterior corona radiata and baseline TMS scores, indicating that larger decreases in FA were associated with lower TMS scores (less severe motor signs) at baseline in a single cluster of 854 voxels ( Figure S4). There were also widespread correlations in the predicted positive direction between baseline TMS scores with MD ( Figure 4) and AD ( Figure S4) and a positive association between RD and baseline TMS limited to the left anterior corona radiata ( Figure S4). Baseline SDMT correlated negatively with MD and AD in the corpus callosum and corona radiata and baseline TFC with AD only in the genu of the corpus callosum, indicating that larger scores in these scales were associated with smaller decreases in structural connectivity.
Similarly, the cUHDRS also correlated negatively with MD and AD in the corpus callosum, corona radiata and subcortical WM tracts ( Figure 4). A also suggesting that more severe symptoms at baseline were associated with larger longitudinal increases in diffusivity. There were no correlations between any of the DTI metrics and disease burden score and SWR in PADDINGTON.

DISCUSSION
In the current study, we have shown robust, longitudinal changes in WM organization across the time course of HD comparing HD expansion-carriers and controls. Using diffusion MRI, we applied a whole-brain approach, optimized for longitudinal analysis, to two well- Moreover, mHTT aggregates and neuronal loss do not entirely colocalize with atrophy, suggesting the interaction of cell autonomous and non-autonomous factors alongside different protein isoforms, underlying neuronal death (Ast et al., 2018;Hackam, 1999;Raj & Powell, 2021;Ross et al., 2014). Therefore, unbiased whole-brain imaging analysis can detect alterations in WM microstructure in areas a priori not expected to change.
The main limitations of whole-brain longitudinal analyses of DWI data are registration failures between baseline and final visit scans leading to misalignment and potential spurious results. However, our use of a midspace template minimizes registration failures. Subse-quently, detailed quality control in our analyses showed excellent alignment of baseline and follow-up DTI maps with the midspace templates.
Importantly, here, we have applied the same longitudinal pipeline to two populations with different disease burden suggesting progressive decreases in structural connectivity as the disease develops. The changes we observed are largely consistent with those previously identified in WM studies for both pre-HD (Harrington et al., 2016) and manifest HD Poudel et al., 2015;Weaver et al., 2009), in addition to previous pathological findings (Vonsattel et al., 2011). Interestingly, FA decreases over time were present only in the pre-HD cohort (at our chosen statistical threshold), perhaps indicating that FA may be less sensitive than other DTI metrics to change in manifest HD Sritharan et al., 2010;Sweidan et al., 2020). (D) Scatterplot, linear trend, and 95% confidence interval depicting the association between baseline composite UHDRS and change in mean diffusivity within the largest significant cluster from (C). TBSS results are shown on the FA skeleton (green), overlaid on the MNI standard brain template. All analyses presented are adjusted by age, sex, and study site; thresholded at p < .05 (TFCE cluster-corrected). The color bar (yellow: red, higher: lower) represents p-values above the statistical threshold for significance. HD, Huntington's disease; TBSS, tract-based spatial statistics; TFCE, threshold-free cluster enhancement; UHDRS-TMS, Unified Huntington's Disease Rating Scale-total motor score 2017; Odish et al., 2015) suggesting that initially subtle changes in microstructural disorganization may be asymmetric although these tend to coalesce with disease progression. Importantly, these changes expanded toward neighboring areas, becoming bilateral when applying lower statistical thresholds.
To understand the impact of WM microstructural change over time on function and behavior, we correlated baseline clinical scores and change in diffusion measures. Increased pathological burden is a likely explanation for the progressive changes that we identified. We found that TMS was routinely associated with longitudinal changes in WM across both cohorts, indicating increased levels of WM disorganization as motor symptoms worsened. Even for the pre-HD group, where motor symptoms were less discernable, TMS was a good predictor of WM change. This is consistent with previous studies showing significant relationships between WM microstructure and TMS (Harrington et al., 2016;Hong et al., 2018) with motor scores being able to differentiate pre-HD from healthy controls (Georgiou-Karistianis et al., 2013).
There were counterintuitive correlations between the TMS and FA in the PADDINGTON cohort suggesting that in this population milder motor symptoms at baseline are associated with larger longitudinal decreases in FA. However, FA was the only DTI metric that did not show significant longitudinal change in PADDINGTON in our analysis, indicating that it may not be particularly sensitive in manifest participants. In addition, the significant area was very localized and not associated with other clinical scales suggesting that the positive correlation between FA and the TMS may be a consequence of longitudinal changes in crossing fibers caused by volumetric decreases. Crossing fibers are not modeled through DTI indices but may be present in 90% of brain voxels, being particularly problematic in areas with complex WM configurations such as the corona radiata, where counterintuitive associations were found in our study (Jeurissen et al., 2013).
Newer advanced multishell diffusion images and myelin-sensitive modalities such as quantitative magnetization transfer (qMT) imaging have shown myelin impairments from premanifest stages of the disease. Similarly, quantitative susceptibility mapping is sensitive as a myelin marker and could provide further insights into structural connectivity in these areas (Casella et al., 2020(Casella et al., , 2022Heath et al., 2018). In contrast, neurite orientation dispersion and density imaging (NODDI) analysis showed decreased neurite density and increased orientation dispersion in expansion carriers compared to healthy controls (H. Zhang et al., 2012). These techniques could overcome some of the limitations from DTI metrics in future studies investigating longitudinal change in large observational cohorts.
Interestingly, there were no correlations between disease burden score and WM microstructure, indicating that for future studies TMS is a more robust predictor of change (Gregory et al., , 2020Poudel et al., 2015).
The cUHDRS is a multidomain measure that was generated using data from 1600 early HD participants. It has good test-retest reliability and longitudinal signal-to-noise ratio (Schobel et al., 2017). In addition, it is associated with clinically meaningful change (Trundell et al., 2019), Changes in WM reflect the distribution of pathology from postmortem histological studies, with alterations starting in deep brain areas and progressing toward the cortex during pre-HD stages (Vonsattel et al., 2011). Therapies that potentially decrease the pathological load of HD, such as mHTT lowering therapies, may also slow WM disorganization, which can be tested in an exploratory fashion as part of a clinical trial. ROI-based approaches bear sufficient effect sizes to show treatment benefit over a clinical trial period but require a preselection of the structures being analyzed Rosas et al., 2006;Georgiou-Karistianis, Scahill et al., 2013).
Here, we have shown that a reproducible, unbiased whole-brain methodology could detect change over a clinical trial period without the need for an a priori selection of brain regions. In consequence, this methodology could be applied to investigate treatment effects. We considered it important to apply this longitudinal pipeline to investigate the changes in standard DTI metrics, as these have been extensively in cross-sectional studies . However, longitudinal change in newer diffusion metrics such as fixel-based analysis or NODDI metrics could be explored in future analyses.
A whole-brain approach is a significant advantage since invasive delivery of newer mHTT lowering therapies will influence the spatial distribution of neuronal changes in response to it. Avoiding preselection of brain areas for investigation could increase the likelihood of detecting significant results without introducing selection bias.
There are some limitations to our study. Although the TrackOn-HD cohort focused on (and was predominantly composed of) pre-HD, there was a small proportion of participants after clinical motor onset. However, the distinction between pre-HD and manifest HD is based on the diagnostic confidence level scale, being highly subjective (Oosterloo et al., 2021). In contrast, newer classifications of HD include objective measures in clinical scores and from imaging biomarkers and are focused on the initial stages of the disease . In addition, the progression of HD from the imaging perspective is continuous rather than dichotomous (Tabrizi, 2009) The disease burden score differed between both cohorts in our study, with higher disease severity in PADDINGTON. The disease burden score depends on age and CAG repeat length, the main determinants of disease severity and age at onset (Wexler et al., 2004) being therefore more appropriate to estimate the severity of the disease than the proportion of pre-HD and manifest HD participants. Therefore, our results indicate progressive WM disorganization across the disease time course.
In addition, longitudinal follow-up was shorter in PADDINGTON

CONCLUSION
To conclude, we have shown progressive widespread longitudinal changes in diffusivity and FA in two cohorts with different disease load.
These changes parallel the distribution of pathology in HD and are associated with clinical outcomes, including the cUHDRS in manifest participants, being currently used in clinical trials with HTT-lowering therapies. In consequence, our methodology could be applied in future clinical trials and may be relevant for the development of future therapies directly targeting disease-causing mechanisms.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study and the analysis code will be made available upon reasonable request.