Introduction

Huntington’s disease (HD) is an autosomal dominantly inherited neurodegenerative disorder caused by a pathogenic CAG repeat expansion of the Huntingtin gene.¹ HD is characterised by a progressive loss of cognitive and motor functions as well as psychiatric disturbances. The clinical onset of HD is commonly defined by the manifestation of motor symptoms, such as chorea, reduced voluntary motor control, bradykinesia, and difficulty maintaining rhythmic and paced movements.^2–4 However, changes in the brain, notably striatal atrophy in the basal ganglia^5,6 may precede the motor onset by up to 24 years^7–10 and correlate with motor and cognitive decline.^{2,5,6,11–15} While HD is rare (∼12 in 100,000), it can be seen as a model neurodegenerative disorder, due to its clear genetic cause, well-characterized disease progression, and shared features of protein abnormalities with more common disorders like Alzheimer’s and Parkinson’s disease.⁵

There is presently no approved disease-modifying therapy for HD, but numerous clinical trials are underway to test the safety and efficacy of novel therapeutics.¹⁶ The recent surge in potential disease-modifying targets has generated a demand for surrogate outcome measures that are sensitive to HD neuropathology and allow a mechanistic assessment of therapeutic effects on striatal neurodegeneration in a timely manner. Volumetric measurements from non-invasive MRI are known to be sensitive to disease progression^17–20 and have been adopted into the recently published Huntington’s Disease Integrated Staging System (HD-ISS).²¹ However, volumetric measurements do not provide information about the underlying neuropathological tissue changes that lead to striatal atrophy, such as the loss of medium spiny neurons (MSN)^22,23 and changes in glia cell density and morphology, including enlargement of reactive astrocytes and microglia.^22,24–26

Diffusion-weighted Imaging (DWI) is widely used to investigate brain tissue microstructure in vivo by exploiting apparent water displacement due to Brownian motion.^27,28 Most DWI studies in HD have used diffusion tensor imaging (DTI), which models extracellular water diffusion as a Gaussian tensor and measures diffusion properties such as mean diffusivity (MD) and the degree of diffusion anisotropy (fractional anisotropy; FA).²⁹ DTI studies in HD have consistently reported increases in MD and FA in striatal grey matter,^30,31 that likely arise from the selective neurodegeneration of medium spiny neuron connections.

Advances in multi-shell and ultra-strong gradient DWI,³² have enabled increasingly sophisticated biophysical models that require data acquisition over a range of b-values to separate extra-from intracellular diffusion signals.^33–35 Several approaches that model intraneurite space with cylinders or sticks have been put forward (e.g. Composite and Restricted Model of Diffusion, CHARMED³³; Convex optimization modelling for microstructure informed tractography; COMMIT³⁶; Neurite Orientation Dispersion and Density Index; NODDI³⁴). NODDI studies in premanifest HD have revealed reduced apparent neurite density in white mater with localised reductions in fibre orientation dispersion in the corpus callosum and basal ganglia capsules.³⁷ However, despite putamen volume loss, no NODDI-based differences in striatal grey matter were detected in gene-positive individuals long before motor onset.¹⁰

Soma And Neurite Density Imaging (SANDI)³⁸ is a novel biophysical diffusion model that extends multi-compartment approaches like NODDI to account for the more complex geometrical architecture of grey matter. SANDI requires multi-shell acquisition protocols with b-values over 3,000 s/mm² to capture restricted signal fractions from within grey matter soma, which are modelled as geometrical spheres. In this way, SANDI provides estimates of apparent soma density (f_is) and soma radius (r_s) in addition to apparent neurite density (f_in) (modelled with sticks), extracellular signal fraction (f_ec) and extracellular diffusivity (D_e). SANDI has been shown to provide highly reproducible and repeatable parameter estimation across grey matter regions in the human brain³⁹ that align closely with its known cyto- and myeloarchitecture. For instance, the gradients of apparent soma density maps were found to closely match those of Brodmann areas in human cortical regions with different soma density profiles,³⁸ and correlated in the mouse brain with cell density distributions from the Allen atlas.^38,40 These findings suggest the potential of the SANDI model for quantifying neurodegenerative processes in the grey matter of the living human brain.

Clinical applications of SANDI are currently limited to multiple sclerosis (MS), where reductions in apparent soma and neurite density and increases in extracellular signal fraction in grey and white matter align with MS-related demyelination, axonal loss, and neurodegeneration.^41,42 These microstructural abnormalities have been found to correlate with disease severity,⁴² cortical or subcortical atrophy,^41,43 and elevated levels of serum neurofilament light chain (NfL), a neuronal cytoplasmic protein marker of axonal damage,⁴³ that is also sensitive to HD progression.^44,45

The primary objective of this study was to assess whether SANDI indices were sensitive and specific to microstructural grey matter differences in the basal ganglia, compared with the thalami, in a group of individuals with premanifest and manifest HD relative to healthy controls. Secondary objectives were the exploration of the extent to which any HD-related SANDI differences accounted for basal ganglia atrophy, assessed with volumetric measurements, and performance differences in motor tasks, including speeded and paced finger tapping, known to be associated with striatal atrophy in HD. ⁴⁶ Finally, relationships between SANDI differences and disease burden were explored by employing the Huntington’s Disease Integrated Staging System (HD-ISS) ²¹ and the CAG-Age Product (CAP₁₀₀) score.⁴⁷

Materials and methods

Participants

MRI data from 56 individuals who tested gene-positive for the mutant huntingtin allele and 57 healthy age- and sex-matched individuals (healthy controls; HC) were included in the analyses. Thirty-eight of the gene-positive individuals participated in a randomised controlled feasibility trial of HD-DRUM,⁴⁸ a remote rhythmic training intervention, with ethical approval from the Wales Research Ethics Committee 2 (REC Reference: 22/WA/0147).⁵⁵ Here we report baseline MRI and behavioural data collected prior to randomisation into the trial arms. Additionally, we included MRI data from 18 gene-positive individuals and 18 age-matched HC from a previous study characterising white matter microstructure in premanifest HD (REC Reference: 18/WA/0172).⁴⁹ Further, MRI data from 25 age- and sex-matched HC from the Wales Advanced Neuroimaging Database (WAND) Study⁵⁰ (REC Reference: 18.08.14.5332RA3) and 14 from the HD-DRUM study were utilised as comparison control. All participants provided written informed consent according to the Declaration of Helsinki prior to taking part in the studies.

HD gene-positive individuals were identified and screened for eligibility in five HD clinics in the UK (Bristol, Birmingham, Cardiff, Exeter, and Liverpool). HC volunteers were recruited from online advertisements on the Cardiff University social network, Viva Engage, or in HD clinics as support partners or family members of individuals with HD. HC participants were also recruited through Healthwise Wales and by word of mouth. For the WAND study, data collection is reported elsewhere.⁵⁰

Individuals over the age of 18 years with a good command of the English language were eligible to participate. Additional inclusion criteria for individuals with HD⁴⁸ were:

• Positive for the presence of the mutant huntingtin allele (CAG length ≥ 36 repeats) and/or clinical diagnosis of HD.

• Unified Huntington’s Disease Rating Scale (UHDRS) Total Functional Capacity (TFC) score between 9 and 13.⁵¹

Exclusion criteria were:

• Contra-indication for MRI (e.g. pacemakers, stents).

• An inability to provide informed consent.

• For HD participants: A history of any other neurological condition.

• For HC: A history of any neurological or psychiatric condition, and/or alcohol or drug abuse, that have been associated with grey matter volume loss.

To characterise general cognitive functioning HD participants were assessed with the Montreal Cognitive Assessment (MOCA).⁵² Verbal intellectual ability was assessed with the Test of Premorbid Functioning (TOPF).⁵³ Disease burden was estimated by the TFC and the CAG-Age Product (CAP₁₀₀) score,⁴⁷ which was calculated using the following formula:

In addition, HD participants were stratified into one of the four HD-ISS²¹ stages using an online calculator (https://enroll-hd.org/calc/html_basic.htm):

• Stage 0: Individuals carry a mutated HD gene with CAG repeat ≥ 40.

• Stage 1: Individuals meet Stage 0 criteria and exhibit striatal atrophy.

• Stage 2: Individuals meet Stage 1 criteria and exhibit clinical signs or symptoms evidenced by changes in the scores on the United Huntington’s Disease Rating Scales (UHDRS)⁵⁴ Total Motor Score (TMS) and cognitive changes in the Symbol Digit Modalities Test (SDMT).⁵⁵

• Stage 3: Individuals meet all previous criteria and show signs of functional change in the UHDRS Total Functional Capacity (TFC) score.

Motor outcome measures

Participants of the HD-DRUM study completed a range of motor tasks from the Quantitative-Motor (Q-Motor) test-battery,^56–58 which has been shown to provide reliable assessments of speeded finger tapping performance in clinical HD trials.^59,60 Tasks included left and right 1) speeded index finger tapping using force transducers,⁴⁶ 2) speeded foot tapping, 3) paced finger tapping⁴⁶ and 4) paced foot tapping with a metronome-paced and memory-paced phase, using a fast (0.55s inter-onset interval; IOI) or slow (1.1s IOI) metronome pace as well as 5) 3D pointing to four target locations in a predefined sequence using a position-tracking stylus with the dominant hand⁶¹ and 6) 3D target pointing and speeded finger tapping dual task performed with dominant and non-dominant hand, respectively. Outcome measures for the various tasks were as follows:

• Speeded tapping: mean IOI in seconds and mean area under the curve (AUC) in Newton-seconds.

• Paced tapping: mean absolute deviation from the metronome rhythm, measured in seconds.

• Target pointing task: target frequency in Hertz (Hz).

• Target pointing-speeded tapping dual task: target frequency in Hz, mean IOI in seconds, and mean AUC in Newton-seconds.

Image acquisition

MRI data were acquired on a 3T Siemens Connectom scanner (Siemens Healthcare, Erlangen, Germany) with ultra-strong magnetic gradients (300mT/m) at Cardiff University Brain Research Imaging Centre (CUBRIC).

T1-weighted (T1w) images were acquired using a magnetisation-prepared 180-degrees radio-frequency pulses and rapid gradient-echo (MPRAGE), with the following parameters: repetition time (TR) 2,300 ms, echo time (TE) 2 ms, field of view (FOV) 256 x 256 x 192 mm, matrix size 256 x 256 x 192, resolution 1 x 1 x 1 mm³, flip angle 9°, inversion time (TI) 857 ms, in-plane acceleration (GeneRalised Autocalibrating Partial Parallel Acquisition; GRAPPA) factor 2, phase-encoding direction anterior to posterior (AP), and acquisition time of 6 minutes.

Multi-shell High Angular Resolution Diffusion Imaging (HARDI)⁶² data were obtained at b-values of 200 s/mm² (20 directions), 500 s/mm² (20 directions), 1,200 s/mm² (30 directions). 2,400 s/mm² (61 directions), 4,000 s/mm² (61 directions) and 6,000 s/mm² (61 directions) using a single-shot spin-echo, echo-planar imaging sequence with TR = 3,000 ms, TE = 59 ms, FOV 220 x 200 mm in-plane; matrix size 110 x 110 x 66; 2 mm³ resolution, gradient pulse duration - δ = 7 ms, gradient pulses separation - Δ = 24 ms in AP phase-encoding direction with an in-plane acceleration (GRAPPA) factor of 2. Fifteen non-diffusion-weighted (b-value = 0 s/mm²) images were acquired [two initial and 11 interspersed at the 33^rd volume and every 20^th volume thereafter in AP direction and 2 images in the posterior-to-anterior (PA) direction]. The HARDI acquisition time was 18 minutes.

Image processing

Diffusion-weighted image preprocessing

Multi-shell HARDI data were pre-processed and corrected for signal drift, susceptibility-induced distortions, motion and eddy current-induced distortions, gradient non-uniformity and Gibbs ringing artifacts using a custom in-house pipeline comprising tools from the FMRIB Software Library (FSL version 6.0.3),⁶³ the MRtrix software package,⁶⁴ ExploreDTI⁶⁵ (version 4.8.6) and in-house MATLAB-based scripts.³⁹

The FSL brain extraction tool⁶³ was used to mask the first non-diffusion-weighted image from each phase-encoding direction to exclude non-brain data. The diffusion-weighted MRI volumes were fitted to temporally interspersed b0 volumes to correct for within-image intensity drift by using custom code in MATLAB R2017b (MathWorks Inc., Natick, Massachusetts, USA). Slicewise outlier detection (SOLID)⁶⁶ was applied with modified Z-score thresholds of 3.5 (lower) and 10 (upper), utilising a variance-based intensity metric. FSL’s top-up tool^67,68 was used to estimate susceptibility-induced off-resonance fields from b0 images that were acquired in opposing phase-encoding directions (AP and PA) and then FSL’s eddy tool⁶⁹ was used to correct eddy current-induced distortions and subject movements. Gradient non-uniformity distortions were corrected using in-house code in MATLAB R2017b. Finally, Gibbs ringing correction was performed in MRtrix3 using the local subvoxel-shifts method.⁷⁰

For the purpose of comparing our results with the previous literature,³⁰ DTI was fitted with ExploreDTI using data with b-values of 500 s/mm² and 1,200 s/mm² to produce outcome maps for FA and MD, estimated with linearly weighted least squares regression.

SANDI analysis

Noise level estimation and denoising of diffusion MRI data were performed using a Marchenko-Pastur principal component analysis (MP-PCA)-based method^71–73 in MRtrix3 to generate noise maps for each subject’s diffusion-weighted images. These noise maps were subsequently used to fit the SANDI model³⁸ to the pre-processed multi-shell diffusion data with the SANDI MATLAB Toolbox (https://github.com/palombom/SANDI-Matlab-Toolbox-Latest-Release³⁸ using all the default settings.

The model fitting produced maps of the intra-neurite, extracellular and intra-soma signal fractions (f_in, f_ec, f_is), apparent soma size (rs; measured in μm) and intra- and extra-neurite diffusivities (D_in, D_e; measured in mm²/ms). Post-hoc sensitivity analysis of the SANDI model parameters revealed very low sensitivity to changes in D_in. Consequently, it was excluded from further analysis.

T1-weighted image preprocessing

The default FreeSurfer ⁷⁴ (v6) recon-all pipeline was utilised to segment subcortical basal ganglia ROIs of the caudate, putamen, and globus pallidum as well as of the thalamus as control ROIs. ROIs were segmented from T1w images and were identified and labelled in each hemisphere.

Extraction of microstructural metrics from regions-of-interest

Median values of each microstructural index from DTI (FA, MD) and SANDI (f_is, f_in, f_ec, r_s, D_e) models were extracted for each ROI using FSL’s fslmaths. ROI masks were aligned with the diffusion space using rigid transformation with FSL’s flirt⁷⁵ before eroding the boundaries of the subcortical masks to minimise partial volume effects and then aligning all microstructural maps with the masks.

Volumetric measures for each ROI and intra-cranial volume (ICV) were extracted from FreeSurfer v6. ROI volumes were normalised for ICV. The addition of brain volumes allowed exploration of the extent to which any HD-related SANDI differences accounted for basal ganglia atrophy.

Statistical analysis

Statistical analyses were performed in JASP (v0.18.1.0)⁷⁶, R version 4.4.1 (2024-06-14)⁷⁷ in R-studio (2024.9.0.375)⁷⁸ and SPSS (v27) (IBM Corp).⁷⁹ Data normality was assessed using the Shapiro-Wilk test, with p < 0.05 indicating non-normal distribution. Descriptive statistics for each group were reported as percentages (%), means and standard deviations (SD). Medians of each microstructural index in each ROI were compared between the groups with Mann-Whitney-U tests because of lack of normality and unequal variance between groups. Effect sizes for group comparisons were therefore reported with rank biserial correlation (r_rb). Multiple comparisons were corrected with Benjamini-Hochberg’s method to control a false discovery rate (FDR) of 0.05⁸⁰ and applied to all statistical tests that related to the same theoretical inference.⁸¹

Hierarchical linear regression analyses were conducted to test for microstructural SANDI predictors of the variance in volumetric measurements. Regression analyses were carried out for each ROI and each group separately. HD data were modelled by firstly accounting for age and TFC scores (available for all HD participants) simultaneously. This was followed by step-wise inclusion of the SANDI indices using an iterative forward selection and backward elimination method based on each variable’s F-statistic and p-value that aimed to maximise the adjusted R²-value while keeping only the most significant predictors. HC data were modelled in the same way except for the inclusion of TFC scores.

Principal Component Analysis (PCA) was carried out to reduce the dimensionality of HD participants’ Q-Motor data and hence the number of multiple correlations with microstructural SANDI indices. PCA followed established guidelines to limit the number of extracted components in relatively small sample sizes.^82,83 First the Kaiser criterion of including all components with an eigenvalue greater than 1 was applied and the Cattell scree plot was inspected to identify the minimal number of components that accounted for most variability in the data. Each extracted component was then assessed for interpretability. PCA was conducted using orthogonal Varimax rotation of the component matrix with Kaiser normalization. Loadings that were greater than 0.5 were considered to be statistically significant.

Spearman’s rho (ρ) correlations were then calculated between HD participants’ motor component scores and the CAP₁₀₀ with the SANDI indices, DTI, and volumetric measures in each ROI.

Finally, analyses to explore HD-ISS related differences in SANDI indices were conducted for SANDI indices in the basal ganglia. For these exploratory analyses a binary split combining Stage 0 and Stage 1 as "premanifest" and Stage 2 and Stage 3 as "manifest" was performed, and SANDI indices were collapsed across hemispheres by averaging. Pairwise comparisons between premanifest and HC, premanifest and manifest, and manifest and HC were conducted using Mann–Whitney U tests without additional FDR correction.

Results

Demographics

Sample characteristics, including age and sex distribution for the HD and HC groups are described in Table 1. No statistically significant differences were observed in age and sex in the HD and HC groups (p>0.05).

Table 1.

Information for HD-ISS calculation was available for 27 individuals. N = 9 were stratified into Stage 1, Ν = 5 into Stage 2, and Ν = 12 into Stage 3. One individual could not be classified due to their clinical pattern not conforming with the assumption of the HD-ISS model of motor symptoms preceding functional impairments. Twenty-nine individuals could not be classified due to missing clinical data or due to their CAG repeats falling between 36-39. Details of the demographic and clinical information per HD-ISS stage and for participants who could not be classified are summarised in Supplementary Table 1.

Table 2 displays descriptive statistics for the Q-Motor measures of the HD gene-positive participants.

Table 2.

Imaging analysis

HD-related differences in volumetric measures

Table 3 summarises the volumetric group differences across the eight ROIs. Figure 1 illustrates the effect size maps of volumetric differences between HD and HC, as well as raincloud plots for each ROI. Mann-Whitney tests identified significantly reduced volumes in all basal ganglia ROIs in HD gene-positive compared to healthy individuals. In contrast, only a marginally significant reduction was present in the right thalamus (FDR-p = .048) and a non-significant reduction in the left thalamus (FDR-p = .063).

Figure 1.

Table 3.

HD-related differences in microstructural measures

Descriptive and statistical microstructural results are shown in detail in Table 4. Figure 2 provides effect size mappings and raincloud plots of the different microstructural measures for each ROI comparing the two groups.

Figure 2.

Table 4.

The SANDI measures revealed between-group differences in the basal ganglia but not thalami for apparent soma density and size, and extracellular signal fraction (Fig. 2). HD gene-positive individuals exhibited reduced apparent soma density (Fig. 2A) across all basal ganglia ROIs, while apparent soma size was elevated bilaterally in the caudate and putamen but decreased in the pallidum (Fig. 2B). Elevated extracellular signal fraction was observed in the basal ganglia (Fig. 2C), while extracellular diffusivity was higher in the putamen and pallidum but showed no significant differences in the caudate (Fig. 2D). No differences were observed for apparent neurite density f_in in any of the ROIs (FDR-p ≥ 0.167, absolute r < 0.2).

Regarding the DTI measures, increased FA and MD (FDR-p < 0.01) were observed in the basal ganglia (Fig. 2E-F) with the exception of MD in the left pallidum (FDR-p = 0.854).

No differences were found in the thalami for any of the microstructural metrics.

Correlations between BG microstructure and motor performance in HD

Motor measures

PCA extracted one principal component that explained 64% of the Q-Motor data with high loadings (> 0.5 or <-0.5) from all variables (Table 5). Spearman’s rho correlation between Q-motor component scores and the disease burden CAP₁₀₀ score revealed a positive correlation (ρ = 0.61, p = 0.002, N = 24), i.e., higher disease burden was associated with higher scores in the Q-Motor component reflecting slower and less accurate motor performance (Figure 3).

Figure 3.

Table 5.

Spearman’s rho correlations between Q-motor component scores and SANDI microstructural measures from all ROIs and scatter plots are displayed in Figure 4. Figure 5 displays correlations and scatter plots of the Q-Motor component with DTI and volumetric indices. Overall, negative correlations were observed between principal Q-Motor component scores and apparent soma density in the basal ganglia, apparent soma size in the pallidum, and volumetric measurements in all ROIs, reflecting that lower apparent soma density and volumes were associated with impaired motor performance, reflected as higher scores in the Q-Motor component. Conversely, positive correlations were present between Q-Motor component scores and apparent soma size, extracellular signal fraction, and extracellular diffusivity in the basal ganglia, i.e. larger apparent soma size and extracellular signal were associated with impaired motor performance (Figure 4 and 5).

Figure 4.

Figure 5.

Microstructural predictors of BG volumes

Hierarchical linear regression analyses of the healthy control data (Table 6, Figure 6) revealed that 17% of volumetric variation in the left caudate was accounted for by age and apparent soma density while 11% of variation in the right caudate was explained by age and extracellular diffusivity (D_e). Age accounted for 25% of volumetric differences in the left putamen, and together with D_e for 29% of differences in right putamen. Age alone explained 17% of left and 11% of right thalamic volume variation. No significant regression models were present for bilateral globus pallidus.

Figure 6.

Table 6.

In contrast, regression analyses for the HD data (Table 7, Figures 7 & 8) showed that 60% of volume variation in the left caudate and 51% of variation in the right caudate were accounted for by age and apparent soma density and size. Similarly, 57% of variation in the right putamen volume were explained by age and apparent soma size, while 63% of volume differences in the left putamen were explained by age, apparent soma size as well as extracellular signal and diffusivity. Comparable to the healthy control results, age alone accounted for 30% of volume variation in the left and for 35% in the right thalamus while no age effects were present for bilateral globus pallidus. However, in HD patients 42% of volume variation in the left pallidum was explained by apparent soma size and extracellular signal and 27% of volume variation in the right pallidum by extracellular signal fraction only. No significant contributions of TFC were present.

Table 7.

Figure 7.

Figure 8.

Correlation of disease burden with microstructural and volumetric measures

The CAP₁₀₀ score as an index of disease progression was correlated with brain measurements to explore the relationship between disease burden and microstructural and volumetric differences (Figure 9). Figure 9A summarises correlation coefficient strengths and levels of significance. Figures 9B-J display scatter plots of significant correlations between the CAP₁₀₀ and microstructural and volumetric measures. Increased CAP₁₀₀ was negatively correlated with apparent soma density, and volume size in bilateral caudate and putamen, as well as with pallidal apparent soma size. Positive correlations were observed between CAP₁₀₀ and extracellular diffusivity and signal fraction, apparent soma size in caudate, putamen, and thalamus, apparent neurite density in putamen, FA in the basal ganglia, and MD in caudate and putamen.

Figure 9.

Exploratory pairwise comparisons between HD-ISS premanifest and manifest HD individuals

Exploratory comparisons of apparent soma density, apparent soma size, extracellular signal fraction and diffusivity averaged across left and right basal ganglia ROIs (Table 4), were carried out between individuals at premanifest HD-ISS Stage 1 (n = 9), individuals at manifest Stages 2 and 3 (n = 17), and healthy controls (Supplementary Table 1).

Compared to the premanifest group, manifest HD individuals exhibited reduced apparent soma density in the striatum, smaller apparent soma size in the pallidum, and increased extracellular diffusivity in the putamen (r_rb = 0.428–0.513, p = 0.01–0.031). Compared to controls, individuals with premanifest HD showed increased extracellular signal fraction in the basal ganglia, larger apparent soma size in the caudate and putamen, and reduced apparent soma density and extracellular diffusivity in the pallidum (rrb = 0.243–0.459, p < 0.001–0.02). The manifest group differed from controls in all measures. Descriptive statistics and statistical analysis results are in Supplementary Table 2. Bar plots displaying effect sizes and 95% confidence intervals for all pairwise comparisons are shown in Supplementary Figure 1.

Discussion

This is the first study to investigate HD-related microstructural differences in the basal ganglia with SANDI, a novel diffusion MRI technique, that was devised to probe grey matter microstructure. The objective of the study was to explore SANDI indices as potential non-invasive in vivo MRI markers of HD neuropathology that may provide more specific information about disease-related tissue abnormalities than volumetric measurements. SANDI indices of apparent soma density and size were found to be sensitive to HD pathology in the basal ganglia and explained, together with age, up to 63% of striatal atrophy in HD. Furthermore, SANDI indices correlated with motor impairments and CAP100 disease burden in HD. This pattern of results suggests the potential of SANDI indices for future imaging biomarkers of disease progression and of the neural effects of novel disease-modifying therapeutics in HD and other neurodegenerative diseases.

Microstructural and Volumetric Alterations in HD

Well-established patterns of significant volume loss accompanied by increases in FA and MD in the basal ganglia were replicated in HD compared with healthy controls. FA increases in the caudate and putamen are thought to occur due to the selective degeneration of medium spiny neuron connections.^20,84 No microstructural differences and only trends for volumetric reduction were present in the thalami. This pattern of macro- and microstructural differences in the basal ganglia is in accordance with previously reported changes in the basal ganglia associated with premanifest and early manifest HD stages.^31,85

Importantly, the application of SANDI revealed novel information about HD-related microstructural differences in the basal ganglia. In gene-positive compared with healthy individuals, apparent soma density was reduced and accompanied by increases in extracellular signal fraction and diffusivity across the caudate, putamen, and pallidum but not the thalami. In addition, apparent soma size was increased in the caudate and putamen and reduced in the globus pallidus in individuals with HD.

HD-related reductions of apparent soma density in the basal ganglia are consistent with the loss of striatal medium spiny neurons, the histopathological hallmark of HD,^22,23 and downstream degeneration of pallidal neurons likely due to the loss of striatal projections. Furthermore, changes in apparent soma size may reflect shifts in the proportion of neural and glial cell density and/or changes in cell morphology, including astrocyte and microglia swelling in response to neurodegeneration^22,24–26 and soma shrinkage prior to neuronal cell death.^86,87

Exploratory analyses using stratification into premanifest HD-ISS Stage 1 and manifest Stages 2 and 3 revealed that apparent soma density reductions in the striatum became more pronounced with disease progression. Regional variability of differences in apparent soma size were observed with early increases in the striatum and later reductions in the globus pallidus. Increases in extracellular signal fraction with disease progression were apparent across all basal ganglia ROIs while increases in diffusivity were particularly pronounced in the putamen and pallidum at manifest stages. These preliminary findings suggest that SANDI indices may be sensitive to neuropathological processes associated with different stages of HD.

Microstructural predictors of basal ganglia atrophy in HD

Regression analyses testing for microstructural predictors of HD-related atrophy in each ROI demonstrated that SANDI indices accounted for a significant proportion of atrophy in the basal ganglia but not the thalami.

Up to 63% of HD-related striatal atrophy was predicted by apparent soma density and size and age. This dropped to 27% and 42% of atrophy in right and left pallidum explained by apparent soma size and extracellular signal fraction, while age alone accounted for volume differences in the thalami. The latter finding in the thalami mirrored the pattern of results in the healthy control data where age was the most significant predictor across all ROIs (except bilateral pallidum) and the only predictor in the thalami, while diffusivity contributed to right-lateralised basal ganglia regions and apparent soma density to left caudate only.

Together these results demonstrate that SANDI indices of apparent soma density and size may capture HD-related differences in striatal grey matter microstructure in vivo. As outlined above, striatal reductions in apparent soma density and increases in apparent soma size are in accordance with the HD characteristic loss of medium spiny neurons and reactive gliosis and explained a significant proportion of HD-related atrophy in the caudate and the putamen.

Associations Between SANDI microstructural indices and Motor Function

The observed correlations between SANDI metrics and motor measures provide novel insights into the functional implications of microstructural alterations in HD. Consistent with the role of the basal ganglia in motor initiation and coordination,⁸⁸ the present study demonstrated that microstructural differences in HD, notably reduced apparent soma density in the striatum and reduced apparent soma size in the pallidum were associated with poorer Q-Motor performance, which in turn was associated with larger disease burden. This is evidenced by increased IOI and ACU in speeded tapping tasks, as well as difficulties in the paced metronome tapping task. Similarly, increased apparent soma size in the striatum, as well as elevated extracellular signal and diffusivity across all three basal ganglia regions, and striatal FA and MD, were linked to motor impairments. These findings suggest that microstructural differences due to basal ganglia neurodegeneration and associated glial reactivity are directly linked to subtle motor impairments.

Clinical Implications and Future Directions

The present study acquired multi-shell (max b-value = 6,000 s/mm²) DWI data on a non-clinical ultra-strong gradient (300mT/m) 3T MRI system. Ultra-strong gradient imaging has the advantage of improving the signal-to-noise-ratio at high b-values due to shorter echo times (TE), which in turn provides enhanced sensitivity to small water displacement,³⁹ and minimises bias due to inter-compartmental exchange.^89,90 However, it is important to note that it has been demonstrated feasible to acquire multi-shell DWI data for SANDI modelling on standard clinical 3T MRI systems,⁹¹ and that such acquisitions have been shown to be sensitive to neuropathology in MS.⁴³ Thus, it is possible apply SANDI indices clinically.

Our findings highlight the potential of SANDI-derived metrics as future markers for tracking disease progression and assessing therapeutic efficacy in HD and more common neurodegenerative diseases like Alzheimer’s and Parkinson’s disease. The sensitivity of apparent soma density and size, and extracellular water signal to microstructural changes in HD offers a complementary perspective to volumetric measures, which are currently the most widely employed imaging modality in clinical trials. Furthermore, the associations between SANDI metrics and motor measures that were correlated with disease burden scores underscore their relevance in evaluating the efficacy of emerging disease-modifying treatments.

Conclusion

In our study, we demonstrate the utility of SANDI for characterising microstructural abnormalities in HD, providing a detailed view of basal ganglia pathology and its implications for motor functions. By bridging the gap between histopathological findings and in vivo imaging, SANDI offers a promising avenue for advancing HD research and clinical care.

Data availability

This research utilised baseline data from the HD-DRUM project that has been endorsed by the Enroll-HD Scientific Oversight Committee (SOC) (14/11/2022). At the end of the HD-DRUM project, the coded study data will be shared and made accessible to the research community via the Enroll-HD specific data request process.

Supplementary figures

Supplementary figure 1.

Supplementary table 1.

Supplementary table 2.

Acknowledgements

We would like to thank Amy Dangerfield, Allison Cooper and Sonya Foley-Bozorgzad for their help with data collection as well as Derek Jones and John Evans for their advice and support with regards to the implementation of MRI data acquisition protocols. We would like to thank the following clinical and administrative staff at the participating patient identification centres for their help with identifying suitable patients for the study: Eileen Donovan, Kim Munnery, and Jane Davies from the Cardiff HD clinic; Jessica Prado Mota from the Royal Devon University Healthcare NHS Foundation Trust in Exeter; Jenni Burns from the Walton Centre NHS Foundation Trust in Liverpool; Natalie Rosewell, Anya Soonderpershad, and Dr Liz Coulthard from the Bristol Brain Centre; Claire Tilley and Dr Hugh Rickards from the Birmingham and Solihull Mental Health NHS Foundation Trust. In addition, we would like to thank all Public Involvement contributors and the members of the Enroll-HD Scientific Oversight Committee for their input into the study as well as all participants for their generous time commitment to help us conducting this research.

Additional information

Funding

This work was supported by a National Institute for Health Research (NIHR) and Health and Care Research Wales (HCRW) Advanced Fellowship to CM-B (grant number: NIHR-FS(A)-2022). The Centre for Trials Research at Cardiff University receives infrastructure funding from HCRW. MP is supported by the UKRI Future Leaders Fellowship MR/T020296/2. CC was funded by a Wellcome Trust PhD studentship (204005/Z/16/Z) and LL by a PhD studentship of the School of Psychology at Cardiff University. The WANDI project was funded by a Wellcome Trust Investigator Award (096646/Z/11/Z), a Wellcome Trust Strategic Award (104943/Z/14/Z), and a Wellcome Discovery Awards (227882/Z/23/Z and 317797/Z/24/Z).

Funding

National Institute for Health Research (NIHR-FS(A)-2022)

• Claudia Metzler-Baddeley

Health and Care Research Wales (NIHR-FS(A)-2022)

• Claudia Metzler-Baddeley

UKRI (MR/T020296/2)

• Marco Palombo

Wellcome Trust

https://doi.org/10.35802/204005

• Chiara Casella

Cardiff University

• Lucy Layland

Wellcome Trust

https://doi.org/10.35802/096646

Wellcome Trust

https://doi.org/10.35802/227882

Wellcome Trust (317797/Z/24/Z)

Wellcome Trust

https://doi.org/10.35802/104943