In Humans, fMRI Reveals That Striosome‐Like and Matrix‐Like Striatal Voxels Are Engaged in Different Phases of Movement

ABSTRACT

The striatum is organized into two neurochemically and anatomically distinct compartments, the striosome and matrix, that play specialized roles in motor and cognitive functions. While extensive animal research has elucidated compartment‐specific contributions to reward, learning and motor control, direct evidence for compartment specialization in humans is lacking. We defined human striatal voxels as striosome‐like or matrix‐like based on biases in structural (diffusion) connectivity. Then we investigated functional activation patterns in those compartment‐like voxels using task‐based functional MRI (tfMRI) during pre‐movement cue and five motor conditions (left/right hand, left/right foot, and tongue movements). Functional activation was strikingly segregated: striosome‐like voxels were preferentially engaged during the cue phase, while matrix‐like voxels dominated activation during motor execution, especially for tongue and foot movement. Motor tasks elicited robust bilateral striatal activation, with contralateral activation dominating during limb movements. Activation was more lateralized in matrix‐like than in striosome‐like voxels. Both striosome‐like and matrix‐like voxels exhibited strong activation at the onset of task execution (e.g., within the first few seconds post‐cue). However, activation in matrix‐like voxels declined modestly over the course of the movement phase, while striosomal activation dropped sharply at task termination, suggesting a role in behavioral transitions. These findings are consistent with the role of the striosome in anticipatory evaluation and dopaminergic modulation, and matrix specialization for executing automatized routines. This study provides the first task‐based fMRI evidence of temporally and functionally distinct striatal compartment dynamics in humans, offering novel insights into striatal microcircuitry in motivated behavior and the planning and execution of movements.

Key Points

• Striatal medium spiny neurons develop in two interdigitated tissue compartments, the striosome and matrix, that are embryologically, pharmacologically, and anatomically distinct. Inter‐compartmental differences in function have been identified in animals but never in humans.

• We found that in humans, the compartments differed in functional activation during movement tasks: during the task cue, activation was greater in striosome‐like voxels, while matrix‐like activation was greater during each of five distinct types of movement.

• Both compartments were active at the beginning of movement, but at the termination of movement, striosome‐like activation fell to below baseline, suggesting a role for the striosome in behavioral transitions.

(A) Connectivity‐based parcellation (diffusion tractography) identified voxels whose anatomic features matched those of the striatal compartments, striosome and matrix. (B) During the cue portions of motor fMRI tasks, activity was higher in striosome‐like voxels. (C) During movement, activity was higher in matrix‐like voxels, and striosome‐like voxels deactivated at movement termination.

1. Introduction

The basal ganglia are a subcortical cluster of interconnected nuclei, historically implicated in motor control, but now recognized to contribute to skill acquisition and action selection in executive, emotional, learning, and reward tasks (Bamford and Bamford 2019; Rocha et al. 2023). While dysfunction of these circuits was classically associated with movement disorders such as Parkinson disease, Huntington disease, dystonia, and dyskinesias (Wichmann and Dostrovsky 2011), striatal pathology has also been implicated in a wide range of neuropsychiatric and cognitive disorders, including obsessive‐compulsive disorder OCD; (Burguiere et al. 2015), Tourette syndrome TS; (Albin 2006), addiction, and depression (Dunlop and Nemeroff 2007; Kalivas 2008).

The striatum is the major input nucleus of the basal ganglia and receives convergent cortical, thalamic, and brainstem projections that support action selection and motor execution (Figure 1). Its principal neurons are organized into two neurochemical compartments: striosome and matrix that differ in embryologic origin, histochemical markers, and afferent and efferent connectivity (Crittenden and Graybiel 2011). Striosomal regions receive proportionally greater input from limbic cortices, whereas matrix regions are preferentially innervated by sensorimotor and associative cortices (Lévesque and Parent 2005; Watabe‐Uchida et al. 2012). Both compartments receive dopaminergic projections, but their connectivity with midbrain structures and their neuromodulatory responses differ, suggesting partially specialized functional roles.

FIGURE 1.

Striosome and matrix compartments in the human striatum. Limbic cortices preferentially innervate striosomes, whereas sensorimotor and associative cortices preferentially target the matrix. Within each compartment, D1‐ and D2‐expressing spiny projection neurons give rise to direct and indirect pathways. Striosomal projections form a distinct inhibitory pathway to the substantia nigra pars compacta (SNc) that inhibits midbrain dopaminergic (DA) neurons. Matrix neurons comprise the classical basal ganglia direct and indirect pathways via the globus pallidus externa (GPe), subthalamic nucleus (STN), and globus pallidus interna/substantia nigra pars reticulata (GPi/SNr), ultimately influencing thalamocortical output. Dopaminergic (DA) input from the SNc innervates both compartments, providing compartment‐specific neuromodulation that links limbic, cognitive, and motor control loops.

Animal studies support distinct contributions of the compartments to behavior. Striosomal circuits have been associated with processes such as affective or value‐related evaluation, whereas matrix circuits have been more strongly linked to sensorimotor integration and movement execution (Trytek et al. 1996; Reinius et al. 2015; Weglage et al. 2021). Recent work further indicates that both compartments participate in motor behaviors but do so through separable circuits (Okunomiya et al. 2025). Despite this extensive experimental literature, whether comparable compartment‐level functional distinctions are detectable in humans during active motor behavior has not been established.

Evidence from disease models also supports compartment‐specific motor functions. In the YAC128 mouse model of Huntington disease, striosomes show greater loss of volume and SPN count than the matrix, and this striosome‐selective degeneration strongly predicts motor coordination deficits (Lawhorn et al. 2008). This does not contradict the matrix's role in executing movement; rather, it suggests that striosomal modulation of dopaminergic tone and circuit balance is essential for maintaining smooth motor control. Recent work further demonstrates complementary roles in different phases of movement, with striosomal dSPNs suppressing dopamine at the end of movement epochs and matrix dSPNs sustaining activity during execution (Okunomiya et al. 2025). Beyond neurodegeneration, imbalanced activation between the compartments predicts repetitive, inflexible motor patterns following psychostimulants (Canales and Graybiel 2000). Together, these findings show that dysregulated striosome–matrix interactions, rather than dysfunction of either compartment alone, can drive motor impairments, underscoring the need to study the distinct contributions of each compartment to the control of movement.

Decades of prior animal studies suggested a functional division of labor between the compartments. The striosome is thought to regulate anticipatory and evaluative processes such as action selection, informed by reward potential or limbic state, whereas the matrix is considered critical for executing well‐learned, habitual motor actions (Graybiel 2008; Hikosaka et al. 2014; Friedman et al. 2015). While this framework was formulated based on structural connectivity patterns (Donoghue and Herkenham 1986; Flaherty and Graybiel 1993; Eblen and Graybiel 1995), subsequent functional studies provided direct behavioral evidence for striosome involvement in reward evaluation and decision‐making. For example, placement of stimulating electrodes in the striosome, but not in the matrix, led to habitual self‐stimulation (White and Hiroi 1998). Similarly, striosomal activation has been linked to pessimistic decision‐making under emotional conflict (Amemori et al. 2021) and suppression of movement under threat (Friedman et al. 2015), highlighting the compartment's role in affectively guided behavioral control.

We recently demonstrated that connectivity‐based parcellation (probabilistic diffusion tractography) can identify voxels with striosome‐like and matrix‐like properties (e.g., biases in structural connectivity, intrastriate location) in living humans (Waugh et al. 2022). This method enables, for the first time, individualized mapping of compartment‐like voxels in vivo, based on their distinct cortical and subcortical connectivity profiles. We refer to the parcellated striatal voxels as “striosome‐like” or “matrix‐like” to emphasize the inferential basis of this approach and to clarify that these voxels are not direct equivalents of the striosome and matrix compartments identified by immunohistochemical analyses. However, our MRI‐based parcellations closely mirror key anatomical features of the striosome and matrix compartments observed in tissue. First, the relative abundance of matrix‐like and striosome‐like voxels matched histological estimates, with matrix‐like voxels comprising approximately 85% of the striatum and striosome‐like voxels approximately 15%, a ratio consistent with prior human and primate tissue studies (Desban et al. 1993; Holt et al. 1997; Waugh et al. 2022; Funk et al. 2023, 2024; Sadiq et al. 2025). Second, the spatial distribution of compartment‐like voxels conformed to the distribution of the compartments in tissue, with striosome‐like voxels preferentially localized to the rostroventral striatum, and matrix‐like voxels more evenly distributed, especially in the dorsolateral and caudal striatum (Graybiel and Ragsdale Jr 1978). Third, projections to striosome‐like and matrix‐like voxels are organized somatotopically (Funk et al. 2023), just as they are in animals (Eblen and Graybiel 1995). Fourth, matrix‐like voxels are more likely to occur in large, contiguous clusters, while striosome‐like voxels are more likely to occur in isolated clusters. Finally, connectivity‐based parcellations are highly reproducible between repeated MRI scans, with a 0.14% test–retest error rate (Waugh et al. 2022). These connectivity‐based parcellations lay the foundation for investigating human striatal compartmental function in both rest and task‐based fMRI experiments (Sadiq et al. 2025).

However, the extent to which these compartment‐specific functions translate to human behaviors remains largely unknown. We previously demonstrated that striosome‐like and matrix‐like voxels participate in distinct structural networks (Funk et al. 2023, 2024) and are embedded in segregated resting‐state fMRI networks (Sadiq et al. 2025). This compartmental segregation aligns with prior hypotheses that striosomes and matrix may be differentially implicated in neuropsychiatric and movement disorders. In particular, (Crittenden and Graybiel 2011) reviewed evidence suggesting that disorders such as Huntington disease, Parkinson disease, depression, and OCD may exhibit selective vulnerability or dysregulation of the striosome system, potentially due to its unique connectivity with limbic networks and nigral dopaminergic neurons. However, direct evidence of functional dissociation during active behavior in humans has not been demonstrated previously, to the best of our knowledge. To address this gap, we leveraged our connectivity‐based parcellation method to examine functional activation in striosome‐like and matrix‐like voxels during simple motor tasks. Using these compartment‐specific masks as seeds in a task‐based fMRI design, we tested whether striosome and matrix compartments show distinguishable activation profiles during motor behaviors. We aimed to answer three critical questions: (1) Do striosome and matrix compartments exhibit functionally distinct activation patterns in the human brain during motor tasks? (2) Are these differences specific to phases of the task, for example, during cue‐related or motor execution periods? (3) Do these dynamics support the proposed roles of the striosome in anticipatory processing and of the matrix in habitual motor control?

By comparing activation in compartment‐like voxels during the cue, initiation, plateau, and termination phases of movement, we provide the first evidence that striosome and matrix differ in both the amplitude and timing of task‐evoked functional responses in humans. These findings have important implications for understanding striatal microcircuit contributions to motor planning, action execution, and the pathophysiology of conditions in which dysfunctional striatal signaling has been implicated, such as OCD, dystonia, and tic disorder.

This study provides the first direct evidence of functional dissociation between striosome‐ and matrix‐like activation in the human striatum during motor behaviors, offering a new framework for linking microcircuit‐specific dynamics to complex motor and neuropsychiatric symptoms.

2. Materials and Methods

Overview: we used a combination of structural and functional connectivity to investigate compartment‐specific activation during motor tasks in humans (Figure 2). We first utilized connectivity‐based parcellation (probabilistic diffusion tractography) to identify striatal voxels with striosome‐like or matrix‐like biases in structural connectivity. Notably, we have previously described striatal parcellation in 14 distinct human neuroimaging datasets (Waugh et al. 2022; Funk et al. 2023, 2024; Sadiq et al. 2025)—in each dataset, compartment‐like voxels followed the spatial distribution, relative abundance (striosome:matrix volume ratio), and extra‐striatal connectivity patterns of striosome and matrix tissue identified through prior animal and human histological studies. Next, we examined functional activation within these compartment‐like striatal masks, modeling each subject's BOLD response to task events using a general linear model (GLM), a standard statistical approach that estimates brain activity related to specific task phases (e.g., movement initiation or execution). For each voxel, we computed contrast parameter estimates (COPEs) that quantified the difference in activation between task and baseline conditions, yielding compartment‐specific activation metrics. This approach allowed us to directly compare the amplitude and timing of activation within striosome–matrix‐like voxels across different motor tasks.

FIGURE 2.

Overview of the experimental workflow for assessing compartment‐ and phase‐specific activation in the human striatum. We used structural connectivity‐based parcellation (probabilistic tractography based on diffusion tensor imaging [DTI]) to define striosome‐like and matrix‐like compartments for each individual. Note that the precise location of the striosome varies between individuals—a uniform region‐of‐interest approach will not accurately distinguish the striatal compartments. These compartment masks served as regions of interest for evaluating compartment‐specific activation using task‐based functional MRI (fMRI) during motor performance. We compared cue‐related and execution‐related activity across compartments to test hypothesized roles of the striosome in anticipatory processing and the matrix in motor execution. Activation values are presented in arbitrary units (AU), reflecting relative BOLD signal change.

2.1. Study Population

This was a secondary analysis of MRI data from the Human Connectome Project (HCP) S1200 release (Van Essen et al. 2013), which included comprehensive behavioral and neuroimaging datasets from a large cohort of healthy young adults. From the original sample of 1206 participants, we selected individuals who had complete diffusion MRI and task‐based (motor) fMRI datasets. We excluded participants if they had any lifetime history of illicit or addictive substance use (cocaine, hallucinogens, cannabis, nicotine, opiates, sedatives, or stimulants) based on HCP diagnostic screening. Additionally, we excluded individuals who met the DSM‐5 criteria for Alcohol Use Disorder (either Alcohol Abuse or Dependence) or who reported consuming more than four alcoholic drinks per week on average during the year prior to scanning. After these exclusions, our final study cohort comprised 701 healthy adults (mean age = 29.3 years, SD = 3.8), including 409 females and 292 males. Handedness was assessed using the Edinburgh Handedness Inventory (Oldfield 1971). Our sample comprised 640 right‐handed and 61 left‐handed participants. All participants provided written informed consent at enrollment with the HCP study (Van Essen et al. 2012). In a recent study whose experimental cohort overlapped with the cohort utilized here, we identified the resting state functional‐networks that covaried with striosome‐like versus matrix‐like voxels (Sadiq et al. 2025).

2.2. MRI Acquisition Protocols

Task‐based functional MRI (tfMRI) and diffusion tensor imaging (DTI) data were acquired as part of the Human Connectome Project (HCP) S1200 release using 3 T scanners with harmonized imaging protocols across multiple sites. The tfMRI sessions used the same echo‐planar imaging (EPI) parameters as resting‐state fMRI (rfMRI), ensuring consistency in spatial and temporal resolution. Each tfMRI scan was acquired with a multiband gradient‐echo EPI sequence with the following parameters: TR = 720 ms, TE = 33.1 ms, flip angle = 52 ^° , FOV = 208 × 180 mm, matrix = 104 × 90, with 72 axial slices and a multiband factor of 8. Echo spacing was 0.58 ms, and the bandwidth was 2290 Hz/pixel. The tfMRI resolution was 2.0 mm isotropic, which we have previously demonstrated is sufficient to resolve striosome‐like and matrix‐like structural connectivity (Waugh et al. 2022). Each motor task condition included two runs of 284 frames, each lasting approximately 3 min and 34 s. The short scan duration and fast TR enabled high temporal resolution for task‐evoked activity mapping. DTI data for S1200 subjects was acquired at 1.25 mm isotropic resolution using 200 directions (14 B0 volumes, 186 volumes at noncolinear directions) with the following parameters: repetition time = 3.23 s, echo time = 0.0892. DTI scans included both anterior–posterior and posterior–anterior acquisitions, allowing for correction of susceptibility artifacts.

2.2.1. Motor‐Task Design

The motor task paradigm was adapted from protocols developed by Buckner and colleagues (Buckner et al. 2011; Yeo et al. 2011) to robustly activate motor and somatosensory cortical and subcortical areas. Participants were visually cued to perform a series of specific movements, isolated to one body part at a time. Specifically, in separate blocks, participants tapped the left fingers, right fingers, moved the left toes, right toes, and tongue. Each block included a 3‐s cue phase to prepare for the movement task, then 12 s of performing one of 10 discrete movements. Each run included 10 movement blocks: two tongue blocks, four hand blocks (two left, two right), and four foot blocks (two left, two right). Within each run, three 15‐s fixation blocks served as baseline condition. All motor task events were modeled relative to the average of the three 15‐s fixation blocks, providing a stable baseline estimate for contrasting movement‐related activation.

2.3. fMRI Preprocessing

We used the minimally preprocessed HCP motor task‐fMRI data (Glasser et al. 2013). The HCP minimal preprocessing pipeline includes gradient–distortion correction, motion correction (volume realignment), EPI distortion correction using spin–echo field maps, and registration of the functional images to each subject's T1‐weighted anatomical scan and to MNI space via the fMRIVolume pipeline. The data also undergoes ICA‐FIX denoising, which removes structured noise components including motion‐related artifacts. No additional preprocessing (e.g., temporal filtering or spatial smoothing) was applied prior to first‐level analysis.

To assess residual motion, we computed framewise displacement (FD) (Power et al. 2012) from the HCP‐provided movement regressors for both LR and RL motor runs. We then calculated the mean FD across all subjects, which was low for the sample (group mean FD = 0.18 mm). Because the dataset exhibited low overall motion and the HCP minimal preprocessing pipeline already includes realignment and ICA‐FIX denoising, no subjects were excluded based on individual FD values.

2.4. Striatal Parcellation

We previously established a technique to identify striosome‐like and matrix‐like voxels in the human striatum based on their distinct in vivo connectivity profiles (Waugh et al. 2022). We use the terms “striosome‐like” and “matrix‐like” to remind readers that these parcellations are inferential and are not the equivalent of immunohistochemical staining, the gold standard for identifying striosome and matrix in tissue. We constructed composite target masks of regions whose striatal structural connectivity was biased toward one compartment, as demonstrated through prior studies that utilized injected tract tracers in animals, or regions whose structural connectivity biases were demonstrated in human diffusion tractography (Waugh et al. 2022). Striosome‐favoring regions included the posterior orbitofrontal cortex, anterior insula, basolateral amygdala, basal operculum, and posterior temporal fusiform cortex. Matrix‐favoring regions included the inferior frontal gyrus pars opercularis, primary motor cortex, supplementary motor area, primary somatosensory cortex, and superior parietal cortex.

We conducted striatal parcellation using the FSL tool probtrackx2, evaluating the relative connectivity of each striatal voxel to striosome‐favoring versus matrix‐favoring target masks. We performed tractography in each subject's native diffusion space and utilized standard parameters: curvature threshold = 0.2; steplength = 0.5 mm; number of steps per sample = 2000; number of samples per seed voxel = 5000; distance correction, to prevent target proximity from influencing connection strength. For each striatal voxel, we compared the number of streamlines reaching striosome‐favoring versus matrix‐favoring target regions. The resulting ratio of these seed‐to‐target streamline counts served as an index of compartmental bias. Each striatal voxel therefore had a bias probability of p = 0–1. We defined compartment‐specific connectivity bias as p > 0.55 toward striosome‐favoring or matrix‐favoring target masks. This voxelwise comparison yielded a continuous map of striosome‐like and matrix‐like connectivity, specific to each subject and hemisphere. Notably, since the striosome is uniquely located in each individual, striosome‐like and matrix‐like masks must be uniquely located for each participant; standardized striatal region‐of‐interest masks are inadequate for investigating the striatal compartments. This underscores the necessity of individualized, connectivity‐informed parcellation approaches for studies aiming to replicate or extend compartment‐specific findings in the human striatum.

Since each diffusion voxel has the potential to include both striosome and matrix, many striatal voxels have only modest compartment‐like connectivity bias. To maximize the contrast between striosome‐like and matrix‐like voxels, we excluded voxels with indeterminate or low bias by applying an iterative thresholding approach to the compartment‐like probability maps. For each subject and hemisphere, we selected voxels from the most‐biased end of the distribution, representing either striosome‐like or matrix‐like connectivity, and gradually reduced the threshold until the total mask volume matched our target volume. We set this target at 13% of the original striatal mask, corresponding to 1.5 standard deviations above the mean in a Gaussian distribution. To ensure that subsequent connectivity measures were not biased by differences in mask size, we assured that striosome‐like and matrix‐like masks for each subject and hemisphere had equal volume. We have previously shown that these high‐bias, equal‐volume masks recapitulate the known spatial distribution and connectivity of striosome and matrix compartments, as established in histological studies (Waugh et al. 2022; Funk et al. 2023, 2024; Sadiq et al. 2025).

Diffusion data were acquired at 1.25‐mm isotropic resolution (voxel volume = 1.95 mm³), while fMRI data were acquired at 2‐mm isotropic resolution (voxel volume = 8 mm³). Therefore, the same anatomical tissue volume contains fewer voxels in fMRI space than in diffusion space. Using a volume‐based conversion factor of ~4.1 (8 ÷ 1.95), 311 diffusion voxels correspond to approximately 76 voxels at fMRI resolution. This conversion provided a consistent, biologically meaningful reference for comparing compartment volumes across modalities and ensured valid interpretation when these masks were projected into fMRI space. Then we registered each high‐bias compartment‐like mask from diffusion space to the subject's structural (T1‐weighted) image using FSL's flirt, and then non‐linearly transformed into fMRI space using the precomputed functional‐to‐structural registration matrices from the HCP pipeline. All transformations were visually inspected for accuracy. The resulting striosome‐like and matrix‐like masks in fMRI space were used for subsequent functional connectivity analyses.

2.5. Validation of Striatal Parcellation

To evaluate whether our parcellated voxels reflected known histological compartmentalization patterns, we measured the Cartesian position of every voxel within the equal‐volume 1.5SD masks. For each subject and hemisphere, the Cartesian coordinates (x, y, z) of each voxel were extracted and referenced to the centroid of the corresponding nucleus (caudate or putamen). Spatial distribution was quantified by calculating within‐plane dispersion and root‐mean‐square (RMS) distance from the nucleus centroid, providing a direct voxelwise measure of how striosome‐like and matrix‐like voxels were organized within three‐dimensional striatal space. Notably, we have previously demonstrated that striosome‐like voxels were consistently enriched in rostral, medial, and ventral regions of the striatum, in line with prior histological characterizations (Graybiel and Ragsdale Jr 1978; Goldman‐Rakic 1982; Donoghue and Herkenham 1986; Ragsdale and Graybiel 1990; Desban et al. 1993; Eblen and Graybiel 1995; Waugh et al. 2022).

Striosomal branches are embedded within the surrounding matrix (Graybiel and Ragsdale Jr 1978; Holt et al. 1997). In coronal tissue sections, striosome branches appear as discrete “islands” amid a contiguous “sea” of matrix tissue. To quantify the spatial distribution of striosome‐like and matrix‐like voxels, we applied the fsl‐cluster command, using a higher bias threshold of p > 0.87 to isolate voxels with high compartment‐specific bias. We focused on the largest cluster within each compartment, which represented the dominant spatial organization of striosome‐ and matrix‐like patterns.

Our goal of matching striosome‐like and matrix‐like volume in compartment specific masks left the less‐biased portions of the matrix‐like distribution unsampled in each subject. Note that since the striosome makes up only 15% of striatal volume, our selection of the uppermost 13% of striatal voxels to represent the striosome left very few, if any, striosome‐like voxels unsampled in each subject. We undertook a secondary analysis to evaluate these unsampled portions of the matrix‐like distribution. We divided the desired mask volume by the number of z‐axis planes and selected that number of voxels in each plane. We then replicated our motor task fMRI analyses using these topographically dispersed matrix‐like voxels.

We previously found that minor shifts in voxel location were enough to disrupt compartment‐like structural connectivity patterns, indicating that such biases were dependent on precise voxel locations rather than their “neighborhood” (Funk et al. 2023). We later demonstrated that shifting compartment‐like voxels by 2–3 mm was sufficient to eliminate any compartment‐like biases in resting state functional connectivity (Sadiq et al. 2025). We hypothesized that compartment‐specific biases in task‐based functional connectivity would also be dependent on precise voxel location. To evaluate the spatial specificity of task‐evoked compartmental activation, we jittered the locations of striosome‐ and matrix‐like voxels by ±0–3 voxels in each anatomical plane, at random and independently for each voxel. For each subject, we confirmed that although individual voxels were shifted, the mean location of the jittered masks remained nearly identical to the original compartment‐specific maps. Importantly, jittered voxels did not overlap with any voxels from the original compartment‐like masks. The average root‐mean‐square shift in individual voxel position was small: 2.9 voxels for the randomized‐striosome mask and 3.1 voxels for the randomized‐matrix mask. This calculation was based on the absolute magnitude of the shift. When examining the average shift within each anatomical plane (including both positive and negative displacements), the movement was minimal: 0.37 voxels on average, with a range of 0.07–1.2 voxels across all planes. As intended, random voxel shifts produced small but noticeable displacements at the individual voxel level yet did not meaningfully alter the overall spatial location of the full striatal masks. On average, the jittered masks still occupied the same “neighborhood” as the original striosome‐like and matrix‐like masks. We then used these location‐shifted voxels as a negative control, comparing task‐evoked activation in the original striosome‐ and matrix‐like voxels (used in our main analyses) to that in the location‐shifted voxels.

2.6. Analysis of Task‐Based Functional MRI (tfMRI)

We conducted a task‐based fMRI analysis to examine condition‐specific activation patterns within high‐bias striosome‐like and matrix‐like striatal masks during motor task execution. First‐level statistical analysis was performed using a general linear model (GLM), in which each motor task condition (hand, foot, or tongue movements) was modeled as a 12‐s block, based on onset timings in the HCP‐supplied explanatory variable (EV) files. Given the availability of both left‐to‐right (LR) and right‐to‐left (RL) phase‐encoding acquisitions for each subject, we performed a second‐level analysis to average across these runs, thereby reducing potential biases due to encoding direction. We extracted the contrast of parameter estimates (COPEs) for each task condition and computed the mean activation within striosome‐like and matrix‐like masks from these COPE images. Activation values represent COPE estimates derived from subject‐level GLMs. These values are expressed in arbitrary units (AU), reflecting the relative amplitude of BOLD signal change associated with each condition.

In contrast to resting‐state functional connectivity, which assesses spontaneous whole‐brain correlation patterns, this task‐based approach targeted localized, condition‐locked activation and its temporal characteristics. This allowed us to directly assess how striosome‐ and matrix‐like voxels differentially engage across the various phases of motor behavior.

2.7. Defining Initiation, Plateau, and Termination Phases in Motor Tasks

To assess the temporal dynamics of compartment‐specific activation during movement, we segmented each motor task trial into discrete phases using the EV files provided by HCP. These files specify the onset and duration of each 12‐s movement event (e.g., hand, foot, or tongue movement) for each participant. The task includes a 3‐s cue period preceding each movement block, but we did not subdivide the cue block as its short duration would not allow for sufficient temporal resolution. Instead, we focused exclusively on the 12‐s movement blocks, dividing each into three equal‐duration, non‐overlapping, four‐second windows to capture the evolving trajectory of neural activity. Specifically, the Initiation phase (0–4 s) captures activity related to movement preparation and execution; the Plateau phase (4–8 s) reflects sustained motor activity; and the Termination phase (8–12 s) represents the final portion of active movement, during which neural responses may begin to decline or transition in anticipation of rest. Notably, participants were not given a countdown or cue to signal the upcoming end of the task and were expected to continue performing movements uniformly until the end of the 12‐s block. Thus, “Termination” refers to the neural dynamics associated with the tail end of the movement phase, rather than post‐task deactivation. This division of each block into three task epochs enabled us to examine whether striosome‐like and matrix‐like compartments exhibited distinct activation profiles across the different phases of movement execution, a key step in linking compartment‐like function to behavioral timing.

2.8. Phase‐Specific Activation Slope Analysis

To quantify how activation evolved across the motor block, we computed temporal slopes between the three task epochs (4‐s each) defined above (Initiation, Plateau, Termination). For each subject and compartment, we quantified the activation during each epoch using the COPE maps (contrast of parameter estimates) generated by the task GLM. We extracted mean COPE values from all voxels within each compartment for each epoch.

Slope is defined as:

$$\text{Slope}=\frac{∆\text{Activation}}{∆\text{Time}}$$

where ΔActivation is the difference in mean activation between epochs and ΔTime is the separation in seconds (4 s for I–P and P–T, and 8 s for I–T).

Specifically:

Initiation → Plateau

$$\mathit{\text{Slope}}=\frac{\mathbf{P}-\mathbf{I}}{\mathbf{4}\mathbf{s}}$$

Plateau → Termination

$$\mathit{\text{Slope}}=\frac{\mathbf{T}-\mathbf{P}}{\mathbf{4}\mathbf{s}}$$

Initiation → Termination

$$\mathit{\text{Slope}}=\frac{\mathbf{T}-\mathbf{I}}{\mathbf{8}\mathbf{s}}$$

2.9. Statistical Analysis

We assessed the accuracy of our striatal parcellations—the intra‐striate position, clustering, and mean bias of our compartment‐like voxels using a series of two‐tailed paired‐samples t‐tests. We compared the volume within streamline bundles seeded by compartment‐like voxels using two‐tailed paired‐samples t‐tests.

Next, we evaluated task‐induced activation levels across cue and motor conditions: cue, left foot, right foot, left hand, right hand, or tongue movement. For each task, we extracted activation values from striosome‐like and matrix‐like voxels separately in the left and right hemispheres. These were averaged across hemispheres to create a single mean activation value for group‐level comparisons, but hemisphere‐specific activation values were also retained to enable direct left–right comparisons (see Section 3.4). This resulted in two values per task per subject: one for the striosome‐like compartment and one for the matrix‐like compartment. For statistical comparisons of activation across movement types and cue‐related activity, all analyses were performed on ROI‐based mean activation values, not voxelwise statistical maps. For each condition, we extracted the mean COPE value from the equal‐volume striosome‐like and matrix‐like masks (76 voxels each), yielding one activation value per compartment per task for each participant. Because the statistical tests were performed on these extracted mean values, the false discovery rate was controlled across the set of condition‐wise comparisons rather than across voxels. Specifically, we applied the Benjamini–Hochberg false discovery rate (BH‐FDR) correction (Benjamini and Hochberg 1995) across the family of seven paired tests (cue, left foot, right foot, left hand, right hand, tongue, and the averaged motor‐activation map) using q = 0.05, resulting in a corrected significance threshold of p = 1.0 × 10⁻⁵.

For the hemispheric differences analysis, we used paired t‐tests to compare activation between the left and right hemispheres for each motor task. Limb movements (hands and feet) were labeled as ipsilateral or contralateral relative to the side of movement, while tongue movement was treated as a midline task. To account for multiple comparisons across the five motor conditions, we applied the Benjamini–Hochberg false discovery rate (BH‐FDR) correction with Q = 0.05, resulting in a corrected significance threshold of p = 0.014. To further quantify hemisphere‐specific activation differences, we computed Laterality Indices (LI) for each motor task, defined as: LI = (Contralateral—Ipsilateral)/(Contralateral + Ipsilateral).

Where “contralateral” refers to the hemisphere opposite the moving limb and “ipsilateral” refers to the same‐side hemisphere. LI values range from −1 to +1, with positive values indicating contralateral dominance, values near zero reflecting equal bilateral or non‐activation, and negative values (not observed in this study) indicating ipsilateral dominance. LI scores were computed separately for each participant and task condition, then averaged at the group level to assess task‐specific lateralization trends.

Finally, we analyzed phase‐specific modulation of striatal activity by dividing task‐related activation into three temporal epochs: initiation, plateau, and termination. Although we computed activation values for all three phases, no significant differences between striosome‐like and matrix‐like voxels were observed during the initiation phase. Therefore, our primary analyses and results focus on the plateau and termination phases, where robust compartmental effects were present. For each compartment, we computed average activation within each phase and compared striosome‐like and matrix‐like activation levels using two‐tailed paired‐samples t‐tests. To further characterize temporal dynamics within each compartment, we calculated linear activation slopes across phase transitions: plateau‐to‐termination and initiation‐to‐termination. We then compared these within‐compartment slope values between striosome‐like and matrix‐like compartments using paired‐samples t‐tests.

3. Results

3.1. Comparing MRI‐Parcellated Voxels to Striosome and Matrix in Tissue

To validate our MRI‐based parcellations, we first examined whether striosome‐like and matrix‐like voxels recapitulated their known spatial distribution within the striatum. Across diverse species from rodents to primates, the spatial distribution of striosome and matrix compartments is conserved, with striosome enriched in the rostral, ventral, and medial striatum, and matrix compartments enriched in the caudal, dorsal, and lateral striatum (Graybiel and Ragsdale Jr 1978; Goldman‐Rakic 1982; Donoghue and Herkenham 1986; Ragsdale and Graybiel 1990; Desban et al. 1993; Eblen and Graybiel 1995; Waugh et al. 2022). Our voxelwise location analysis recapitulated this compartment‐specific location bias in both hemispheres. In the caudate, we found that matrix‐like voxels were significantly more lateral (1.5 mm; p = 3.7 × 10⁻⁵¹), caudal (−8.9 mm; p < 1 × 10⁻²⁶⁰), and dorsal (8.7 mm; p < 1 × 10⁻²⁶⁰) than striosome‐like voxels. In the putamen we found that matrix‐like voxels were more caudal (−5.8 mm; p < 1 × 10⁻²⁶⁰) and dorsal (5.9 mm; p < 1 × 10⁻²⁶⁰) than striosome‐like voxels; their medial‐lateral position was not significantly different. Striosome‐like voxels in the caudate were 12.2 mm medio‐rostro‐ventral to the centroid (Cartesian distance), while matrix‐like voxels were 13.5 mm latero‐caudo‐dorsal to the centroid. Striosome‐like voxels in the putamen were 10.2 mm medio‐rostro‐ventral to the centroid, while matrix‐like voxels were 8.0 mm latero‐caudo‐dorsal to the centroid.

Histological assessments in both animal and human tissue estimated that the striosome and matrix make up approximately 15% and 85% of the striatal volume, respectively (Johnston et al. 1990; Desban et al. 1993; Holt et al. 1997). We evaluated compartment‐like volume at a range of bias thresholds to assure that our MRI‐based parcellations approximated the ratio of striosome:matrix found in tissue. Given that diffusion voxels sample the striatum in 1.25 mm cubes, independent of the underlying compartment architecture, many voxels will include both striosome and matrix tissue. Utilizing higher bias thresholds excludes more of these blended voxels and thus shifts relative volume assessments to voxels that are more striosome‐like and more matrix‐like. For each subject, we quantified the total number of voxels strongly biased (p > 0.87) toward striosome‐like or matrix‐like connectivity. Striosome‐like voxels made up 5.9% of highly biased voxels, while matrix‐like voxels made up 94.1%. This high cutoff excludes most striatal voxels, with 82% of the total striatal volume falling into the indeterminate category (p ≤ 0.87), reflecting an intermediate bias indicative of mixed striosome‐ and matrix‐like connectivity. At the lowest threshold for compartment‐like bias (p ≥ 0.55), striosome‐like voxels constituted 25.5% and matrix‐like voxels 74.5% of biased striatal volume. This lower cutoff included a greater portion of the striatum, with 45% of total striatal volume showing sufficient bias to be classified as striosome‐ or matrix‐like, and the remaining 54% falling into the indeterminate range (p ≤ 0.55). The probability threshold that most‐closely approximated the striosome‐to‐matrix ratio in tissue (15:85) was p ≥ 0.7, with 17.5% of voxels classified as striosome‐like and 82.5% as matrix‐like. This threshold included 32.2% of the total striatal volume, while the remaining 67% of voxels fell into the indeterminate category (p ≤ 0.7). These proportions varied with bias threshold but consistently reflected the marked predominance of matrix‐like connectivity across the striatum. These results agree with histological evidence from both human and animal studies, demonstrating a clear preponderance of matrix‐like voxels in the living human striatum.

Finally, we assessed whether striosome‐like and matrix‐like voxels differed in their clustering patterns. In histologic sections, each striosome branch is surrounded by contiguous matrix tissue (Graybiel and Ragsdale Jr 1978; Holt et al. 1997). In contrast, our diffusion MRI voxels sampled the striatum using a rigid, non‐adaptive grid that did not align with the individual‐specific architecture of striosome. This misalignment introduced partial volume effects, leading to spatial blurring of striosome‐like signal across adjacent voxels. Though it was not possible to resolve the fine‐grained connectivity biases of individual striosome branches at this resolution, we set out to assess whether striosome‐like and matrix‐like voxels differed in their tendency to occur as isolated voxels or to blend into clusters. Striosome‐like voxels were found in smaller, dispersed clusters, while matrix‐like voxels occupied much larger and contiguous clusters. In the right hemisphere, matrix‐like clusters were on average 4.3 times larger than striosome‐like clusters, while in the left hemisphere, the ratio was 3.4‐fold. In the right hemisphere, the mean volume of the largest striosome‐like cluster was 232 mm³ [SEM ±7.0; 95% CI (218, 245)], significantly smaller than the mean matrix‐like cluster volume of 987 mm³ [SEM ±13.2; 95% CI (962, 1013)]; p = 1.2 × 10⁻¹⁹⁹. Similarly, in the left hemisphere, striosome‐like clusters had a mean volume of 299 mm³ [SEM ±7.9; 95% CI (284, 315)], while matrix‐like clusters were significantly larger, with a mean volume of 1007 mm³ [SEM ±14.7; 95% CI (979, 1036)]; p = 5.6 × 10⁻¹⁶². These findings are consistent with histology: in tissue, the striosome is spatially dispersed and surrounded by extensive, contiguous matrix tissue.

3.2. Task‐Specific Activation Differed in Striosome‐ and Matrix‐Like Voxels

Cue‐evoked and motor‐evoked activation patterns showed a clear functional dissociation between striosome‐ and matrix‐like compartments. In the cue condition, mean activation was significantly higher in striosome‐like voxels (57.6 AU) than in matrix‐like voxels (48.4 AU; p < 1.7 × 10⁻¹¹; Figure 3, Figure S1). In contrast, motor‐related activation was consistently higher in matrix‐like voxels across all tasks. In the group‐level average of all motor tasks, activation was 2.4‐fold higher in matrix‐like than in striosome‐like voxels (matrix: 16.4, striosome: 6.9; p = 5.4 × 10⁻³²). Motor‐evoked activation (including both ipsilateral and contralateral hemispheres) was higher in matrix‐like voxels across all movement types: left foot movement (matrix = 25.4, striosome = 15.1; p = 4.6 × 10⁻¹⁹); right foot movement (matrix = 22.3, striosome = 11.9; p = 1.5 × 10⁻²⁴); left hand movement (matrix = 6.3, striosome = 2.2; p = 6.8 × 10⁻⁷); right hand movement (matrix = 11.6, striosome = 7.5; p = 1 × 10⁻⁵); tongue movement (matrix = 16.3, striosome = −2.1; p = 5.4 × 10⁻³²).

FIGURE 3.

Activation differences in striosome‐like and matrix‐like voxels across motor tasks and cue presentation. The left‐most section shows cue‐evoked mean activation levels within matrix‐like (blue) and striosome‐like (red) compartments. Both sets of compartment‐like voxels had much higher activation during cue than during motor execution, but striosome‐like voxels consistently had stronger cue‐evoked responses than matrix‐like voxels. Note the discontinuity in the y‐axis (hashed line). The right section shows motor‐evoked mean activation levels within matrix‐like and striosome‐like voxels. Matrix‐like voxels consistently had stronger motor‐evoked responses across all tasks. Notably, tongue movement was the only task that elicited below‐baseline activation in the striosome‐like compartment. Error bars represent the standard error of the mean (SEM). *, p < 10⁻⁵. **, p < 10⁻¹⁵. This data are also described in Figure S1.

Both matrix‐like and striosome‐like voxels exhibited their strongest activation during the cue period, with substantially lower activation during motor execution across all effectors (Figure 3). This pattern was highly similar between the two compartments. Within this shared response profile, however, we observed reliable compartment‐specific differences: striosome‐like voxels showed greater activation than matrix‐like voxels during cue presentation, whereas matrix‐like voxels showed greater activation than striosome‐like voxels during movement execution. Although these differences are modest in absolute magnitude compared to the cue‐versus‐movement contrast, between‐compartment differences were substantial: in the average of all motor tasks, activation in matrix‐like voxels was 2.4‐fold larger than activation in striosome‐like voxels.

3.3. Sensitivity Analysis Using Topographically Dispersed Voxels

Our primary analyses relied on voxels with the strongest connectivity biases, and we matched the volume of striosome‐like and matrix‐like masks to avoid connectivity biases induced by differences in mask size, which left roughly three quarters of the striatum unsampled in each subject. We wished to ensure that our findings were not dependent on a spatially constrained subset of the striatum, so we defined a second set of matrix‐like voxels that were distributed throughout the striatum. For each axial plane we identified the most biased voxels, dividing the mask volume equally across all z‐axis planes. Note that our original striosome‐like masks included nearly all potential striosome‐like voxels; a very small portion of the striosome‐like distribution went unsampled. Topographically selected matrix‐like voxels were still highly biased toward matrix‐favoring bait regions (mean: p = 0.95); bias was reduced by only 0.49% relative to the original matrix‐like masks, which were defined by bias throughout the striatum, independent of location. Topographic matrix‐like masks were substantially more dispersed: the volume of the largest matrix‐like cluster was reduced by 60.9% (topographic, 56.2 voxels; original, 143.6 voxels; p = 1.8 × 10⁻²⁵⁸), and by definition, these new masks were represented in every axial plane throughout the striatum. We performed a sensitivity analysis to compare BOLD activation in the topographically selected matrix‐like masks versus the original striosome‐like masks.

Topographically selected matrix‐like masks replicated the principal compartment dissociations observed in our primary analysis. Matrix‐like voxels continued to show higher movement‐related activation than striosome‐like voxels across all body parts (left foot, left hand, right foot, right hand, tongue), with highly significant differences in every condition (all p < 10⁻¹⁰). The magnitude of these matrix‐dominant effects was similar to the effects measured in our original comparison, indicating that the motor‐related contrast was spatially distributed and was not dependent on the highest‐bias striatal voxels.

Similarly, the topographically selected masks had cue activation patterns that were similar to our original analysis, where striosomal activity exceeded matrix activity during the cue period. In topographically selected matrix masks, cue activation was higher in striosome‐like voxels (striosome: 54.9; matrix: 49.9; p = 3.6 × 10⁻⁵), though the difference between compartments was attenuated (Δ = 4.97) compared with our assessment in the original compartment‐like masks. Critically, despite this shift in matrix‐like voxel location, the underlying functional dissociations between cue‐related (striosome‐activating) and movement‐related (matrix‐activation) portions of the task were preserved across voxel‐definition strategies.

3.4. Influence of Precise Voxel Location on Striatal Compartmentalization

Striosome and matrix occupy different parts of the striatum, on average, and so did our striosome‐like and matrix‐like masks. It is possible that compartment‐specific differences in functional activation simply reflected the activation profiles of the parts of the striatum where striosome‐like or matrix‐like voxels are enriched—a “neighborhood” effect rather than a compartment‐specific finding. To determine whether compartment‐specific activation was dependent on the precise location of selected voxels, we compared task‐based activation levels derived from the original, precisely selected striosome‐ and matrix‐like masks (Section 3.2) to those from location‐shifted versions of those compartment‐like masks. We jittered the position of each compartment‐like voxel individually, so the effect on the mean location of the whole mask (all voxels combined) was minimal: on average, the center of gravity of the location‐shifted mask was only 0.37 voxels from the original compartment‐like mask. In matrix‐like voxels, the average activation bias dropped from 0.95 (original) to 0.83 (shifted; paired‐samples t‐test, p < 1 × 10⁻²⁶⁰). In striosome‐like voxels, the bias dropped from 0.78 to unbiased (0.47; paired‐samples t‐test, p < 1 × 10⁻²⁶⁰). These results suggest that compartment‐specific activation is not explained by regional location or neighborhood effects but rather depends on the specific structural connectivity of the precisely selected voxels in our compartment‐like masks. Shifting the location of striosome‐like voxels reduced functional activation for cue and every motor task, significantly so in 5 of 7 task conditions (Table 1). For example, these location‐shifted negative controls disrupted the expected pattern of compartmentalization: in location‐shifted striosome‐like voxels cue‐related activation dropped by 11.8% (p = 3.4 × 10⁻³¹) and tongue‐related activation decreased by 171% (p = 4.8 × 10⁻⁷). Even though location‐shifted voxels occupied the same striatal “neighborhood”, their striosome‐like function was dependent on their precise location.

TABLE 1.

Randomly shifting the positions of striosome‐like voxels by a few voxels significantly reduced functional activation in both cue and motor tasks, while shifting matrix‐like voxels did not significantly alter functional activation.

Compartment	Task	∆ activation (shifted–original)	% change	p
Striosome	Cue	−6.8	−11.8	3.4 × 10 −31
	Left‐foot	−1.4	−9.3	1.0 × 10 −3
	Left‐hand	−0.5	−22.7	2.1 × 10−1
	Right‐foot	−2.2	−18.5	3.0 × 10 −7
	Right‐hand	−0.3	−4.0	4.1 × 10−1
	Tongue	−3.6	−171.4	4.8 × 10 −7
	Average	−1.6	−23.2	5.9 × 10 −7
Matrix	Cue	1.2	2.5	3.6 × 10−2
	Left‐foot	0.7	2.8	9.8 × 10−2
	Left‐hand	−0.5	−7.9	1.2 × 10−1
	Right‐foot	−0.7	−3.1	1.0 × 10−1
	Right‐hand	−0.2	−1.7	6.1 × 10−1
	Tongue	0.3	1.8	6.2 × 10−1
	Average	−0.1	−0.6	8.5 × 10−1

Note: Task‐based activation values were compared between the original compartment‐specific masks and location‐shifted versions, across cue and five motor conditions. Delta activation (Shifted—Original), percentage change, and p‐values reflect whether voxel displacement significantly altered mean activation within each compartment. Bolded p‐values indicate significant differences (corrected for multiple comparisons). While this analysis was performed for both striosome‐like and matrix‐like voxels, only striosome‐like voxels demonstrated significant differences in functional activation. The stability of functional activity in shifted matrix‐like voxels likely reflects the abundance of matrix in the striatum; shifting the location of a matrix‐like voxel is likely to select another, slightly less‐biased matrix‐like voxel. The pronounced changes observed in shifted striosome‐like voxels highlight that compartment‐specific effects depend critically on precise voxel selection rather than the “neighborhood” where a voxel is located.

In contrast, in the matrix‐like compartment, no task condition showed a significant effect of randomization. Across tasks, Δ activation values were small (between −1.2 and +0.7) and percentage changes were consistently under 8%. This stability reflected the abundance of matrix‐like voxels: shifting the location of a matrix‐like voxel is likely to select another, slightly less‐biased matrix‐like voxel.

Taken together, these findings provide strong evidence that striosome‐like bias was spatially specific and vulnerable to randomized location shift, while matrix‐like bias was stable and resistant to location shift. This negative control thus underscores the biological specificity of compartment‐related differences: shifting striosome‐like voxels makes them non‐striosome‐like, while shifting matrix‐like voxels makes them only slightly less matrix‐like. Compartment‐specific effects depend on the precise voxel selection and are not explained by local striatal “neighborhood” activation profiles.

3.5. Hemispheric Differences in Striosome and Matrix Activation

We next investigated activation patterns separately in the left and right hemispheres across cue and motor tasks (Figure 4, Figure S2). We first combined both sets of compartment‐like voxels (to identify general patterns of hemispheric asymmetry) and then assessed each set of compartment‐like voxels for asymmetries. During the cue period, when activation values were averaged across both striosome‐like and matrix‐like voxels, the left hemisphere showed significantly greater activation (mean = 56.7) compared to the right (mean = 49.3; p < 6.8 × 10⁻¹⁰).

FIGURE 4.

Mean activation in the left and right hemispheres during cue presentation and five motor tasks (left and right foot movement, left and right‐hand movement, and tongue movement). The left side shows cue‐evoked activation was significantly greater in the left hemisphere compared to the right. The right side shows that asymmetries in motor‐evoked activation were evident in all motor tasks and consistently favored the contralateral hemisphere. No inter‐hemispheric difference was found for tongue movements. Error bars represent the SEM. **, p < 10⁻³; *, p < 0.05. This data are also described in Figure S2.

Motor‐evoked activation exhibited clear hemispheric asymmetries, particularly during hand movements (Table 2). Left‐ and right‐hand movements preferentially activated the contralateral hemisphere. Foot movements showed a similar contralateral dominance pattern, though with slightly smaller asymmetries. In contrast, tongue movement yielded relatively symmetric activation (Table 2). In addition to contralateral dominance patterns, we observed higher overall activation magnitudes during right‐sided body movements compared to left‐sided ones for hand movements (total activation: right hand = 19.1, left hand = 8.4). However, the reverse was true for foot movements, where left foot activation exceeded right foot activation (total activation: left foot = 40.5, right foot = 34.2). These asymmetries may reflect a combination of motor dominance and task‐specific factors such as effort or pacing.

TABLE 2.

Mean activation values (including all compartment‐like voxels) are shown for the left and right hemispheres during five motor tasks: Left foot, right foot, left hand, right hand, or tongue movement.

Movement task	LH	RH	Larger activation	p
Left foot	19.0 (Ipsi)	21.5 (Contra)	Contra	0.014
Right foot	20.1 (Contra)	14.1 (Ipsi)	Contra	1.0 × 10−9
Left hand	2.6 (Ipsi)	5.8 (Contra)	Contra	1.8 × 10−5
Right hand	12.2 (Contra)	6.9 (Ipsi)	Contra	1.6 × 10−10
Tongue	8.4 (Bilateral)	5.8 (Bilateral)	None	0.11

Note: For limb movements, hemispheric values are labeled as either ipsilateral (Ipsi) or contralateral (Contra) relative to the body side of the movement. For tongue movement, both hemispheres are labeled as bilateral. p‐values report paired t‐tests comparing left versus right hemisphere activations. Hand and foot movements exhibited strong contralateral dominance, while tongue movements evoked more symmetric responses.

Abbreviations: LH: left hemisphere; RH: right hemisphere.

To quantify hemispheric asymmetries, we also computed Laterality Indices (LI) for each movement condition. Consistent with known motor system organization, hand movements showed the strongest contralateral dominance (LI = 0.38 for left hand; LI = 0.28 for right hand). Foot movements demonstrated substantially weaker lateralization overall (LI = 0.06 for left foot; LI = 0.18 for right foot), though the left–right differences were of similar numerical magnitude to those observed within hand movement trials, despite not reaching significance. Tongue movements were bilaterally represented. These patterns align with prior reports of contralateral limb representation and bilateral control of midline structures (Lotze et al. 1999; Sörös et al. 2020).

Having established overall hemispheric asymmetries, we next asked whether the degree of laterality differed between striosome‐like and matrix‐like voxels. Both compartments exhibited significant contralateral dominance, but this effect was stronger in matrix‐like voxels. When comparing contralateral versus ipsilateral activation across limb motor tasks (left/right hand and left/right foot), activation was significantly greater contralaterally for both striosome‐like (Ipsi: 7.5, Contra: 10.8; p = 4.1 × 10⁻⁸) and matrix‐like voxels (Ipsi: 13.8, Contra: 18.9; p = 8.9 × 10⁻¹⁵). Contralateral activation exceeded ipsilateral activation in both compartments (Figure 5) across all motor tasks (though not significantly larger for left foot striosome). The degree of lateralization, computed as contralateral–ipsilateral difference scores, was significantly larger for matrix‐like voxels than for striosome‐like voxels (Striosome: 3.3, Matrix: 5.2; p = 0.031), indicating a stronger contralateral bias in matrix‐like voxels.

FIGURE 5.

Contralateral activation exceeded ipsilateral activation in both compartments, with a stronger effect in matrix‐like voxels. Ipsilateral versus contralateral activation in striosome‐like and matrix‐like voxels during motor tasks. Bars represent mean activation (BOLD signal change) across all tasks for each compartment and hemisphere condition. Error bars indicate SEM. *, p < 10⁻³; **, p < 2 × 10⁻⁵.

To evaluate whether handedness contributed to the observed lateralization effects, we repeated the cue‐related, left‐hand movement, and right‐hand movement analyses in right‐handed participants only (N = 640). All the contralateral activation patterns.

Observed in the full cohort were preserved, and the direction and significance of group differences were unchanged, indicating that the laterality effects were not driven by the small subset of left‐handed individuals.

We next repeated the same analyses in left‐handed participants only (N = 61) to directly assess whether hemispheric lateralization differed in this subgroup (Table 3).

TABLE 3.

Handedness‐specific hemispheric activation in left‐handed participants.

Movement task	LH	RH	p
Cue	54.8	50.5	0.29
Left foot	14.1	26.3	0.0049
Right foot	−0.6	1.6	0.35
Left hand	16.8	14.5	0.45
Right hand	12.8	9.7	0.24
Tongue	7.2	10.3	0.64

Note: Mean cue‐ and motor‐evoked activation values are shown separately for the left hemisphere (LH) and right hemisphere (RH) in left‐handed participants (N = 61). p‐values reflect within‐subject comparisons between hemispheres for each condition. Only left‐foot movements showed a significant hemispheric difference, with greater right‐hemisphere activation consistent with contralateral motor organization (bold text). Other conditions did not reach statistical significance, likely reflecting limited statistical power in the smaller left‐handed subgroup. No reversal of laterality effects was observed.

There were no reversals of the laterality effects observed in the full cohort (Figure 4). In left‐handed participants, cue‐related activation was similar to that of right‐handed participants (left hemisphere mean = 54.8; right hemisphere mean = 50.5), though this difference did not reach statistical significance (p = 0.29). For motor‐evoked activation, contralateral patterns were qualitatively preserved across all movement conditions. Left‐foot movements had significantly greater right‐hemisphere activation (right: 26.3, left: 14.1; p = 0.0049), consistent with contralateral motor organization. In contrast, left‐hand (p = 0.35), right‐foot (p = 0.45), right‐hand (p = 0.24), and tongue movements (p = 0.64) did not show significant hemispheric differences. The lack of statistical significance for several effects may be attributable to limited statistical power in the substantially smaller left‐handed cohort, rather than to a difference in hemispheric organization in left‐handed subjects.

3.6. Laterality of Cue‐Evoked Activation

To determine whether cue‐evoked activation differed when preparing for left‐ versus right‐sided movements, we compared activation for left‐ and right‐cue conditions within each compartment. Cue‐evoked activation differed significantly between right‐ and left‐sided movement preparation in both compartments, with stronger activation for cues preceding right‐sided movement in striosome‐like voxels (striosome‐right: 53.3, striosome‐left: 44.6; p = 0.015) and matrix‐like voxels (matrix‐right: 47.4, matrix‐left: 35.3; p = 0.0012).

3.7. Phase‐Specific Modulation of Striosome‐ and Matrix‐Like Activation

We found different patterns of phase‐specific activation between striosome‐like and matrix‐like voxels, particularly during the transition from plateau to termination (Figure 6). These effects reflect dynamic compartment‐specific modulation of activity during motor execution. During the initiation phase, both compartments exhibited similarly high activation (striosome: 29.7, matrix: 30.1), which may reflect equivalent recruitment at the onset of movement or residual activation from the preceding cue block, which was substantially higher than activation in the motor tasks (Figure 3).

FIGURE 6.

Matrix‐like voxels had more sustained activation during motor execution compared to striosome‐like voxels. We divided each 12 s movement block into three phases: Initiation, plateau, and termination. The initiation phase was heavily influenced by residual activation from the high‐activation cue block (Figure 3), leading us to focus on the plateau and termination phases of movement. (A) Mean activation in the plateau and termination phases for striosome‐like and matrix‐like voxels. (B) Activation slopes between task phases (plateau to termination [P–T], and initiation to termination [I–T]) illustrating dynamic changes in compartment‐like activity over time. Error bars represent SEM. **, p = 1.3 × 10⁻¹⁰; *, p < 7 × 10⁻⁴.

In the plateau phase, activation declined moderately in both compartments (striosome‐like: 14.1, matrix‐like: 15.4), reflecting waning functional activation. However, during the termination phase, activation trajectories diverged: activation in striosome‐like voxels dropped below baseline (−1.7), while activation in matrix‐like voxels remained positive (9.3; p < 1.3 × 10⁻¹⁰), indicating more sustained involvement of matrix‐like voxels at task offset.

To quantify temporal differences in activation patterns, we computed phase‐specific slopes for each compartment‐like mask, representing the rate of activation change across task phases (Figure 6). From initiation to plateau, striosome‐like and matrix‐like voxels exhibited similar declines in activation (striosome slope: −3.9; matrix slope: −3.7), and the difference between them was not significant (p = 0.7). However, from plateau to termination, the slopes diverged sharply: activation in striosome‐like voxels showed a steep deactivation (−4.0), while in matrix‐like voxels, activation declined more gradually (−1.5; p = 1.3 × 10⁻⁴). This difference in the slope of deactivation was also evident when comparing the first to the last phase (initiation to termination): striosome‐like voxels again exhibited a steeper overall decline in activation (−3.9) than matrix‐like voxels (−2.6; p = 6.5 × 10⁻⁴). These results highlight a biphasic divergence in temporal dynamics: matrix activation decreased gradually across phases, whereas striosomal activation dropped more abruptly at task termination.

4. Discussion

The fundamental differences in the embryology, pharmacology, and connectivity of the striosome and matrix suggest that their functions in behavior may also be divergent (Graybiel and Ragsdale Jr 1978; Brimblecombe and Cragg 2017; Prager et al. 2020). Indeed, distinct, compartment‐specific functional roles have been demonstrated in an array of animal behaviors: reward (White and Hiroi 1998), decision making under threat (Friedman et al. 2015; Amemori et al. 2021), habit formation (Nadel et al. 2021), learning (Jenrette et al. 2019; Xiao et al. 2020), and motor control (Weglage et al. 2021; Okunomiya et al. 2025). However, to the best of our knowledge, distinct functional roles for striosome and matrix have never been identified in living humans.

Here, using connectivity‐based parcellation combined with a well‐characterized motor task, we found that striosome‐like and matrix‐like voxels exhibited different temporal activation profiles. Striosome‐like voxels showed relatively greater activation during cue/anticipation periods, whereas matrix‐like voxels showed greater activation during motor execution. This pattern is consistent with compartmental dissociations observed in prior animal studies but does not establish a direct mechanistic mapping between species. Rather, our findings provide noninvasive, task‐evoked evidence that compartment‐specific differences in functional activation at rest (Sadiq et al. 2025) are also reflected in behaviorally driven activation.

It is important to acknowledge several limitations of the study. First, our approach identified striosome‐ and matrix‐like voxels indirectly, using biases in structural connectivity identified through probabilistic tractography. The accuracy of our functional assessments depends on the anatomical accuracy of our striatal parcellations. We previously demonstrated that striatal parcellation is highly reliable test–retest error rate of 0.14% (Waugh et al. 2022) and that these biased patterns of connectivity depend on highly precise selection of striatal voxels—shifting voxel position by just a few millimeters negates all compartment‐like bias (Funk et al. 2024; Sadiq et al. 2025). Moreover, striatal parcellation recapitulates all of the anatomical features of striosome and matrix demonstrated in human and animal tissue: their relative abundance, spatial distribution, contiguity, and extra‐striate connectivity (Waugh et al. 2022; Funk et al. 2023, 2024). However, this technique is not the equivalent of direct identification of the compartments through histological staining; compartment‐like voxels share many features with the striosome and matrix, but the degree to which connectivity‐based parcellation approximates the compartments is unknown.

A further limitation is the spatial resolution of diffusion MRI. Even though the voxel resolution used in this study (1.25 mm isotropic) matches the diameter of the largest human striosome branches (Graybiel and Ragsdale Jr 1978; Holt et al. 1997), every striosome‐like voxel will include some fraction of matrix tissue. However, it is notable that our previously‐described validation experiments—in which compartment‐like voxels matched the anatomic features of striosome and matrix identified through immunohistochemistry—were accurate in diffusion datasets with lower resolution than we utilized here (Waugh et al. 2022, Funk et al. 2023, 2024). The resolution of our fMRI dataset matches that of our prior diffusion datasets (2 mm isotropic), and we demonstrated that this resolution is sufficient to identify widespread and robust compartment‐specific patterns of functional connectivity (Sadiq et al. 2025). However, we urge readers to recall that these inferential methods cannot match the fine‐grained histological detail available in post‐mortem tissue. This limitation is inherent to all current non‐invasive anatomical mapping techniques.

The unique distribution of the striosome within each person's striatum precludes the use of region‐of‐interest based assessments. Therefore, we analyzed mean activation within the striosome‐like and matrix‐like masks for each subject, rather than conducting voxel‐wise activation analyses. While averaging within each compartment enabled meaningful between‐subjects comparisons, this approach may have obscured more spatially localized activation patterns within each compartment. Since corticostriate projections are organized somatotopically (Flaherty and Graybiel 1993; Waugh et al. 2022; Sadiq et al. 2025), the functional specializations we identified may be localized to particular parts of each striatal compartment. The somatotopic organization of the striatal compartments underscores the requirement for individualized striosome‐like and matrix‐like masks, particularly in future studies that validate or build on these findings.

It is also important to consider ways in which our experimental approaches may have distorted functional connectivity. Histological studies consistently report that matrix occupies approximately 6‐fold larger volume than striosome (Desban et al. 1993; Holt et al. 1997). However, our goal of generating equal‐volume masks, which was necessary to avoid size‐based biases in probabilistic connectivity, also led us to assess only a fraction of all matrix‐like voxels. It is possible that other matrix‐like voxels, those not among the most‐biased set, could have different functional activation patterns than the ones we identified. However, when we utilized matrix‐like voxels that were distributed throughout the striatum (Results 3.3), these differences in task‐related functional activation did not differ from our primary analyses, in which voxels were selected solely for their matrix‐like bias. The natural volume asymmetry between striosome and matrix remains relevant for understanding the anatomical context of our findings but does not diminish the functional activation patterns reported here.

Finally, several limitations arise from the design of the movement task itself.

The division of the 12‐s movement block into three equal 4‐s windows (initiation, plateau, and termination) was our operational choice rather than a physiologically defined boundary. Although this segmentation allowed us to examine temporal slope differences across movement phases, the precise onset and offset of neural processes underlying initiation and termination likely varied across individuals and may not align perfectly with these fixed windows. Because of this, our time windows should be viewed as rough approximations of different parts of the movement rather than exact neural states. Even so, the main compartment differences we observed, especially the separation between striosome‐like and matrix‐like activity toward the end of movement, were robust and consistent across participants. Future studies with higher temporal resolution or more precise timing markers could better define these movement phases.

The BOLD signal reflects a delayed hemodynamic response rather than instantaneous neuronal activity. Accordingly, the timing of striosome‐ and matrix‐related effects should be interpreted in relation to the modeled hemodynamic response, not as a reflection of precise neural events. While the block‐based design and cue period reduce overlap between preparatory and movement‐related activity, some temporal blurring is unavoidable. Importantly, all conditions were modeled using the same hemodynamic response function, allowing valid relative comparisons between compartments despite this limitation.

Our task‐based fMRI results revealed a clear difference in how striosome‐like and matrix‐like voxels respond during motor behaviors. Striosome‐like voxels showed higher activation during the cue period, whereas matrix‐like voxels were more active during movement execution. This was true in voxels that were at the upper‐most end of the bias distribution, and in voxels distributed throughout the striatum. This pattern aligns with the expected engagement of preparatory versus execution‐related processes during motor tasks and is broadly consistent with compartmental distinctions described in prior animal studies (Graybiel 2008; Friedman et al. 2015), without implying a direct mechanistic equivalence. Matrix‐like activation also varied across effectors, with stronger responses during foot and tongue movements than during hand movements. Striosome‐like voxels showed reduced, and in some cases negative, activation during movement, most notably for tongue responses. However, because the task was not designed to isolate the factors driving these differences, these effects should be interpreted cautiously. Overall, our results show that striosome‐like and matrix‐like voxels differ in their task‐evoked dynamics, with striosome‐like voxels more responsive during preparation and matrix‐like voxels more responsive during execution.

In addition to overall hemispheric asymmetries (Figure 4), we also examined whether these effects differed between striosome‐ and matrix‐like voxels (Figure 5). When averaged across compartments, cue‐related activation was stronger in the left hemisphere, potentially a correlate of prior evidence linking the left striatum to internal processing and self‐control (Zhang et al. 2017). In contrast, during motor execution, matrix‐like voxels showed task‐specific contralateral dominance, with greater right than left hemisphere activation.

To further quantify compartment‐related differences, we computed differential activation indices across effectors. Striosome‐like voxels were more active during the cue phase, whereas matrix‐like voxels showed consistently stronger activation during motor execution, with the largest execution‐related differences observed for tongue movements. These effects reflect a general division between preparatory and execution‐related phases of the task. Although prior animal studies have reported early‐trial increases in striosomal activity during tasks involving predictive or reward‐related cues (Bloem et al. 2017; Yoshizawa et al. 2018), the present motor task does not isolate such processes, and our findings should not be interpreted as evidence for these functions in humans. Instead, our results demonstrate that compartment‐like parcellations derived from structural connectivity exhibit distinct, temporally patterned activation profiles during simple movements.

Our findings revealed distinct phase‐specific activation profiles between striosome‐ and matrix‐like voxels during motor task execution. Matrix‐like voxels showed robust activation during the initiation phase, which declined only modestly through the plateau and termination phases. These findings align with prior work suggesting the dorsolateral striatum (DLS) plays a key role in the execution and initiation of habitual sequences (Jin and Costa 2010). Although Jin and Costa did not examine striosome and matrix compartments separately, the DLS is known to be matrix‐dominant (Graybiel and Ragsdale Jr 1978; Goldman‐Rakic 1982; Donoghue and Herkenham 1986; Gimenez‐Amaya and Graybiel 1990; Desban et al. 1993; Eblen and Graybiel 1995), suggesting that the observed role of the DLS in habitual behavior may be mediated largely by matrix circuits. As skills are learned and become automatic, striatal involvement shifts from associative (dorsomedial striatum) to sensorimotor (DLS) circuits, a transition supported by region‐specific synaptic plasticity, particularly in the indirect pathway of the DLS. In contrast, striosome‐like voxels showed comparable activation to matrix‐like voxels at the plateau but exhibited a pronounced drop during the termination phase. This pattern suggests that striosome engagement is maintained through initiation and plateau, with a selective reduction as the movement concludes. This temporal divergence was further quantified in slope analyses: striosome activation remained stable between initiation and plateau but dropped significantly from plateau to termination while matrix activation declined steadily across all phases (Figure 6).

Another possibility is that the pronounced drop in striosomal activation during task termination may reflect a functional disengagement once evaluative or predictive demands subside, potentially signaling behavioral transitions or the resolution of action plans. This interpretation aligns with anatomical findings that striosomal neurons receive input from limbic regions involved in internal state monitoring and project to midbrain dopaminergic areas implicated in behavioral regulation and state transitions (Fuccillo 2016; Brimblecombe and Cragg 2017). Supporting a role in action modulation, Okunomiya et al. (2025) showed that prolonged chemogenetic activation of striosomal neurons via DREADDs led to sustained motor slowing in mice, suggesting that elevated striosome activity may inhibit movement execution (Okunomiya et al. 2025). Notably, their manipulation spanned minutes, far exceeding the transient 4‐s termination window in our task. Our findings suggest a temporally confined role for striosomal disengagement during transitions rather than generalized inhibition. This divergence in slopes supports a compartmental division of labor: matrix circuits contribute to sustained execution, while striosomal circuits regulate transitions, boundary marking, and internal evaluation, possibly through their direct inhibition of nigral dopaminergic neurons (Crittenden et al. 2016).

Although our results highlight clear functional distinctions between striosome‐like and matrix‐like voxels, our findings also indicate that the two compartments are not entirely segregated in their functional roles. For example, both compartments showed increased activation at the onset of movement, consistent with shared participation in the initiation of motor actions. The divergence emerged only during movement termination, when striosome‐like activation dropped below baseline while matrix‐like activation remained elevated. This pattern suggests that while the compartments can carry complementary or opposing signals, they also cooperate during phases of behavior that rely on integrated striatal output. Such overlap aligns with anatomical evidence showing that striosome and matrix are interdigitated systems that exchange information through interneurons and share upstream and downstream cortical–basal ganglia loops, even though their major long‐range projections diverge—striosomes primarily targeting the substantia nigra pars compacta (SNc) and matrix projections terminating in the substantia nigra pars reticulata (SNr). Thus, the presence of both shared and compartment‐specific dynamics underscores a model in which striosome and matrix contribute jointly to behavioral control, with distinct biases that become most apparent during state transitions.

Importantly, we also observed hemispheric asymmetries in both compartments. Striosome activation during cue phases was stronger in the left hemisphere, consistent with the left striatum's proposed role in internal monitoring and self‐regulation (Zhang et al. 2017). The stronger left‐hemisphere activation during the cue phase may partly reflect language‐related processing, such as reading the cue, given the typical left‐hemisphere dominance for language (Tzourio‐Mazoyer et al. 2004; Price 2012). In contrast, matrix activation during motor execution exhibited task‐specific contralateral dominance, particularly for foot and hand movements, mirroring classical patterns of lateralized motor control (Draganski et al. 2008). Together, these findings reinforce the idea that the striosome and matrix compartments are not only functionally segregated in time but also organized across hemispheres to support preparatory versus execution roles in motor behavior.

Handedness can influence hemispheric motor organization, with right‐handed individuals typically showing stronger left‐hemispheric dominance. Although our sample included a small proportion of left‐handed participants, repeating the laterality analyses in a cohort restricted to right‐handed individuals yielded identical contralateral activation patterns. These findings suggest that the observed lateralization effects reflect intrinsic motor system organization rather than variability introduced by handedness.

In conclusion, this study provides the first task‐based fMRI evidence that compartment‐like voxels in the human striatum exhibit distinct, temporally patterned activation profiles during motor behaviors. While our prior resting‐state fMRI work demonstrated that these compartment‐like voxels participate in segregated intrinsic functional networks (Sadiq et al. 2025), the current findings suggest that such compartmental organization also influences activation dynamics during task performance. These results support the emerging view that striatal microarchitecture may contribute to phase‐specific functional specialization, with striosome‐ and matrix‐like voxels differentially engaged across cue processing and motor execution. Further studies will be needed to directly map these functional signatures to underlying cellular and circuit‐level mechanisms, and to explore their relevance across cognitive domains and clinical populations.

Prior work has demonstrated that the balance between striosome‐ and matrix‐like compartments is altered in several human neuropsychiatric and neurological conditions. Notably, each of these disease‐specific structural abnormalities scaled with symptom severity. Using the same connectivity‐based parcellation approach applied here, we previously reported a selective expansion of matrix‐like volume in autism spectrum disorder, resulting in a substantially increased matrix–striosome volume difference relative to typically developing controls (Waugh et al. 2025). Similarly, we identified compartment‐specific shifts in structural connectivity in major depressive disorder, with increased striosome‐like bias in the putamen and a compensatory matrix‐like shift in the caudate (Waugh and Tieu 2025). Independent work has also extended these methods to early Parkinson's disease, revealing increased matrix‐like volume and altered matrix‐favoring connectivity in de novo, treatment‐naïve patients (Marecek et al. 2024). Together, these converging structural findings across diverse neuropsychiatric disorders underscore that compartment organization is clinically relevant and susceptible to disease‐related alteration. Although our current motor task did not interrogate clinical populations, the compartment‐specific cue‐ and movement‐related responses identified here provide a mechanistic framework for generating testable predictions about how disease‐related alterations in compartment structure may influence striatal function in human diseases.

By integrating structural connectivity‐based parcellation with phase‐specific activation profiling of the motor task, our findings extend the functional characterization of the striatal compartments beyond resting‐state connectivity to real‐time functional activation during motor behaviors. This framework enhances our understanding of how striosome‐ and matrix‐specific networks support different facets of human behavior. These findings also suggest that compartment‐specific maldevelopment or injury may explain distinct motor features of neuropsychiatric disorders that involve the striatum or striatal networks.

Author Contributions

A.S. data acquisition and analysis, initial manuscript drafting, and critical revision of the manuscript. J.L.W. data acquisition, analysis, and interpretation, initial manuscript drafting, and critical manuscript revision. All authors contributed to the article and approved the final version for submission.

Funding

Dr. Waugh was supported by: the CTSA Pilot Award; the Elterman Family Foundation; NINDS grant 1K23NS124978‐01A; the Brain and Behavior Research Foundation Young Investigator Award; and the Children's Health CCRAC Early Career Award. The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of these funding agencies.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Full distribution of cue‐evoked and motor‐evoked activation in compartment‐like voxels. Violin plots illustrate the individual‐subject distributions underlying the mean activation differences shown in Figure 3. For each task condition (Cue; Left Foot; Left Hand; Right Foot; Right Hand; Tongue; and the average of all motor tasks), matrix‐like voxels (light blue) and striosome‐like voxels (pink) are shown with their respective distribution shapes, medians (horizontal bars), and interquartile ranges (blue or pink vertical rectangles). Consistent with the summary statistics shown in Figure 3, striosome‐like voxels show stronger activation during cue presentation, whereas matrix‐like voxels show stronger activation during all motor execution conditions. Asterisks denote significance thresholds identical to those shown in Figure 3 (*, p < 10⁻⁵; **, p < 10⁻¹⁵). This supplementary figure provides the complete activation distributions across all subjects and complements the compartment‐averaged results presented in the main text.

Figure S2: Full distribution of cue‐evoked and motor‐evoked activation in compartment‐like voxels, separated by hemisphere. Violin plots illustrate the individual‐subject distributions underlying the mean activation differences shown in Figure 4. For each task condition (Cue; Left Foot; Left Hand; Right Foot; Right Hand; Tongue), left hemisphere voxels (gray) and right hemisphere voxels (orange) are shown with their respective distribution shapes, medians (horizontal bars), and interquartile ranges (black vertical rectangles). Asterisks denote significance thresholds (*, p < 0.05; **, p < 10⁻³; NS, non‐significant). This supplementary figure provides the complete activation distributions across all subjects and complements the results presented in the main text.

Sadiq, A. , and Waugh J. L.. 2026. “In Humans, fMRI Reveals That Striosome‐Like and Matrix‐Like Striatal Voxels Are Engaged in Different Phases of Movement.” Human Brain Mapping 47, no. 3: e70472. 10.1002/hbm.70472.

Data Availability Statement

The data that support the findings of this study are available in Human Connectome Project at https://db.humanconnectome.org. These data were derived from the following resources available in the public domain: ‐ S1200, https://db.humanconnectome.org/data/projects/HCP_1200.