*2.1. Participants*

Participants included in these analyses were enrolled in the Kids-HD and Kids-JOHD studies [2,13–15]. These were longitudinal neuroimaging studies that ran in parallel to one another. The Kids-HD study recruited participants between the ages of 6–26 who were at risk for inheriting the gene that causes HD based on their family history (i.e., a parent or grandparent with confirmed HD). The Kids-HD study also recruited healthy control participants. All participants in the Kids-HD study underwent genetic testing for research purposes only. For the present analyses, the control group consisted of participants from the Kids-HD study who had molecular confirmation of a CAG repeat length < 36, ensuring that our control group did not include pre-symptomatic patients with AOHD. The Kids-JOHD study recruited participants who had been deemed to have Juvenile-Onset HD (JOHD) by their neurologist and had molecular confirmation of having a CAG repeat length of 36 or above. For these analyses, a participant was considered to have JOHD if they had a total motor score from the Unified Huntington's Disease Rating Scale (UHDRS) [16] of ≥20 prior to the age of 21. The UHDRS is sensitive to developmental motor changes such that younger children will show higher scores than older children. Therefore, even in a large cohort of children at risk, but who did not inherit the gene expansion, the UHDRS can be as high as 15 [15]. Therefore, a cuto ff of 20 on the UHDRS was used to ensure that all participants had confirmed motor-manifest JOHD.

Given the longitudinal nature of these studies, some participants had more than one neuroimaging study conducted. Specifically, there were 11 participants with JOHD that made up 13 visits. Our control group in these analyses consisted of 38 participants that had the necessary neuroimaging done, which included both anatomic (T1- and T2-weighted) and metabolic (T1ρ) images. This consisted of participants who were at risk for inheriting the gene mutation that causes HD but who were found to not carry this gene, as well as healthy control participants without a family history of HD. There were 39 neuroimaging studies amongs<sup>t</sup> this group of 38 participants.

Clinical measures were collected on all participants. As noted previously, all participants received a total motor score as measured using the UHDRS [16]. Motor symptoms were also quantified in the JOHD group using the modified Juvenile HD Motor Rating Scale (JOHDRS) [17]. This rating scale provides additional evaluation of the unique hypokinetic symptoms associated with JOHD. We also calculated a disease burden score (Age × CAG—35.5)) [18] and disease duration (age at time of assessment—age at time of JOHD clinical diagnosis) for JOHD participants.

Signed informed consent was obtained before beginning the study, per the Institutional Review Board at the University of Iowa. All experiments were performed following the guidelines outlined in the Belmont Report. Genetic testing was done for research purposes only. The results were made available to one research team member, and all other team members were blind to these genetic results. The genetic testing results were not revealed to participants or their families.

#### *2.2. Data Acquisition*

High-resolution magnetic resonance images were collected on a 3T Siemens TIM Trio scanner (Siemens Medical Solutions, Erlangen, Germany) using a 12-channel receiver head coil. Whole-brain T1- and T2-weighted anatomical acquisitions with a 1.0 mm isotropic spatial resolution were acquired first. T1-weighted images were collected using a 3D magnetization-prepared rapid gradient echo sequence with the following parameters: Coronal orientation; field-of-view = 25.6 × 25.6 × 25.6 cm; sampling matrix = 256 × 256 × 256; repetition time (TR)TR/echo time (TE)/inversion time (TI) = 2530/2.8/909 ms; flip angle = 10◦; bandwidth = 180 Hz/pixel; and Acceleration Factor (R) = 2 GeneRalized Autocalibrating Partial Parallel Acquisition (GRAPPA). T2-weighted images were collected using a 3D variable flip angle spin-echo sequence with the following parameters: TE = 430 ms; TR = 4800 ms; number of echos = 137; bandwidth = 592 Hz/pixel; matrix = 256 × 256 × 170; field-of-view (FOV) = 25.6 × 25.6 × 22 cm; and R = 2 GRAPPA. Next, quantitative parametric imaging was conducted to acquire T1ρ relaxation times. T1ρ mapping was performed using a coronal segmented three-dimensional (3D) gradient echo sequence with spin-lock pulses (TE = 2.5 ms; TR = 5.6 ms, FOV = 220 × 220 × 200 mm; sampling matrix = 128 × 128 × 40; fractional anisotropy (FA) = 10 degrees; integrated parallel acquisition techniques (IPAT) = 2; spin-lock frequency = 330 Hz; spin-lock times = 10 and 55 ms).

#### *2.3. Image Analysis*

The BRAINS AutoWorkup was used to perform the anatomical image analysis by combining information available from the T1- and T2-weighted images as described in Pierson et al. [19]. Briefly, the BRAINS AutoWorkup includes the following steps: (1) AC-PC alignment; (2) bias field correction; (3) tissue classification; and (4) anatomical labeling. The anatomical regions-of-interest used for this study include the caudate, putamen, globus pallidus, thalamus, hippocampus, and anterior cerebellum. These regions-of-interest are defined automatically using a neural network-based segmentation [20]. These regions have been shown to have a degree of reliability with a manual rater (Jaccard index ~0.80).

To estimate the T1ρ map, the individual spin-lock images were co-registered using a rigid registration with the Advanced Normalization Tools (ANTs) software [21] to account for subject motion between the two acquisitions. The resulting aligned spin-lock images were then used to calculate a T1ρ map by fitting the 10 and 55 ms spin-lock time (TSL) image signals (S0 and STSL) to the following mono-exponential decay model:

$$\mathbf{S}\_{\rm TSL} = \mathbf{S}\_0 \left( \mathbf{e}^{-\rm TSL/T\_{1\rho}} \right) \tag{1}$$

The decay model was fit using the "MR Parameter Map Suite" implemented using Insight Segmentation and Registration Toolkit (ITK) [22] and available from the InsightJournal [23]. The resulting rigid body transform was then used to resample the T1ρ map to a 1 mm isotropic resolution using linear interpolation. The T1ρ maps were then thresholded at 400 ms to remove the contribution of cerebrospinal fluid. The defined regions-of-interest generated from the BRAINS AutoWorkup were then used to estimate the mean T1ρ relaxation times for each region from non-zero voxels in the thresholded T1ρ relaxation time maps.

#### *2.4. Statistical Analysis*

The primary outcomes were mean differences in T1ρ relaxation times from the defined regions-of-interest between groups. We used linear mixed effects models to investigate estimated mean differences in T1ρ relaxation times of the six brain regions above between groups. Our models were controlled for age and sex and included a random effect per participant.

For the secondary analyses, we identified any of the regions-of-interest that demonstrated significant group differences in T1ρ relaxation times from the primary analysis. Amongst those regions-of-interest, we used linear mixed effects regression analyses to investigate the relationship between CAG repeat length and regional brain volumes and T1ρ relaxation times. These models were controlled for age and sex and included a random effect per participant. The models investigating the relationship between brain volume and T1ρ relaxation times were also controlled for intracranial volume (ICV). All of these analyses were performed amongs<sup>t</sup> the JOHD participants only. We then performed a similar analysis to assess the relationship between T1ρ relaxation times and disease burden scores while controlling for sex and a participant random effect. Age was not included as age is used to calculate the disease burden score [18]. Next, we assessed whether T1ρ relaxation times predicted the total motor score, as assessed by the UHDRS and the JOHDR [17], while controlling for age and CAG repeat length. Lastly, we investigated the relationship between the calculated disease duration and T1ρ relaxation times. Again, we performed linear mixed effects regression analyses that controlled for age and sex and a random effect per participant. RStudio (Version 3.6.2, RStudio, PBC, Boston, MA, USA) was used for all statistical analyses and a *p*-value of <0.05 was considered significant for all analyses.
