Cortico-incertal connectivity analysis workflow.

(a) The zona incerta (ZI) seed region used for connectivity analyses. Diffusion MRI data were processed to reconstruct (b) streamlines via diffusion tractography and (c) connectivity matrices quantifying the number of streamlines between each ZI voxel and cortical region as defined by the HCP-MMP1.0 atlas. (d-f) ZI gradients and clusters were computed to illustrate the principal organizations of connectivity variability among ZI voxels. (d) Significant patterns were highlighted based on inter-voxel similarity using the normalized cosine angle. (e) Diffusion map embedding and spectral clustering were used to construct the gradients and clusters, respectively. (f) Gradient-weighted cortical maps were created by multiplying each row of the initial connectivity matrices with the corresponding principal gradient value, then averaging these rows to produce a single cortical representation of each gradient. A winner-takes-all approach was used to create a cortical map with areas color-coded according to their connectivity with the ZI clusters.

Cortico-incertal structural connectivity patterns.

(a) The first two gradients of the zona incerta (ZI) based on structural connectivity shown using axial and 3D radiological views both revealed a rostral-caudal axis. (b) Similarly, spectral clustering shows a topographic organization of discrete clusters along a rostral-caudal axis, with cluster 1 positioned most rostrally. (c) Gradient-weighted cortical maps corresponding to gradients 1 and 2 and the spectral clustering results (left to right).

Cluster-wise tractography.

(a) Example cluster-wise tractograms for the left hemisphere of a single subject. (b) Group-level cluster-wise cortical connectomes, where parcels are color-coded and have varying opacity levels (linearly scaled), reflecting the number of streamlines connecting each parcel with the clusters (row-wise). (c) Boxplot shows the total number of streamlines connecting the cortex with each cluster.

Replicability and reliability of zona incerta (ZI) connectivity patterns.

(a) Axial cross-sections of gradient 1 and 2 per MRI dataset. (b) Similar to a but with ZI voxels displayed in the respective 2D gradient coordinates space. (c) Procrustes disparity scores for each comparison of 2D gradient coordinates among datasets.

Replicability and reliability of the spectral clustering results.

(a) Axial cross-sections of spectral clustering results (k=6) for each MRI dataset (left to right). (b) Comparison of cluster centroids between datasets, with results matched column-wise. Bright labels correspond to the 3T and 3T retest datasets. (c) Dice overlap scores and centroid distances for each comparison of spectral clustering results among datasets for the left and right hemispheres.

Individual subject replicability.

(a) Spatial correlation between individual and group-level structural connectivity for each of the two retained gradients. (b) Centroid distances and (c) Dice scores for the alignment between individual and group-level clusters. Results are shown for both left and right hemispheres.

Correlation analysis between cortico-incertal structural connectivity patterns and cortical properties.

The gradient-weighted cortical maps (Figure 2c) and cluster-wise connectomes (Figure 3) exhibit significant correlations with various (a) cognitive terms (e.g., movement) and (b) cortical hierarchies (e.g., sensorimotor-association axis) based on Spearman’s correlation coefficient (colored markers). Marker colors represent P-values, corrected for spatial autocorrelation using N=10k spin tests. Semi-transparent black bars show permuted values. The source file containing all Spearman’s correlation coefficients is available in the code repository referenced in the manuscript.

Mapping a deep brain stimulation (DBS) case electrode stimulation volume to ZI tractography-based gradients and clusters.

(a) DBS electrode reconstruction was performed using the Lead-DBS (v3.0.0) software and visualized in a common space overlayed with the ZI clusters derived in this work. (b) Propagated coordinates of the left and right stimulation volumes (black markers) into gradient space. Stimulation volume centroids are shown with red outlines. (c) Distribution of gradient scores within stimulation volumes and percent (%) of the DBS stimulation volume overlapping with ZI clusters.

Spectral clustering tractograms and streamline count.

(a) Tractogram based on the 7T MRI dataset shown from left vs. right views. (b) Scatter plots showing the streamline count per cortical parcel, color-coded according to the HCP-MMP1.0 atlas as well as a cortical surface representation, color-coded according to the number of streamlines. The source file containing streamline counts is available in the code repository referenced in the manuscript.

Overview of successive gradients and alternative cluster solutions.

(a) Gradients 3-10 (G3-10) shown using 3D radiological views. The scree plot shows the variance explained by each gradient as indicated by their lambda value (left y-axis). The right y-axis displays each gradient’s respective contribution (%) to the total explained variance by all gradients. (b) Other spectral clustering solutions (k=2-8).

Composite cluster-wise tractogram.

(a) Composite tractogram colored by spectral clustering labels, based on the 7T MRI dataset and for k=6 clusters solution, shown fom left vs. right, (b) superior vs. inferior and (c) anterior vs. posterior views.

Zona incerta (ZI) 2D gradient coordinate space.

(a) Gradient 1 (x-axis) and 2 (y-axis) values were used to position each ZI voxel in the corresponding 2D gradients coordinates space, color-coded using a 2D colormap. (b) Similarly as a, but ZI voxels are color-coded for the k=6 spectral clustering solution. (c-d) Projection of the 2D gradient coordinate space onto the ZI volume and cortical surface spaces, respectively.

Zona incerta (ZI) longitudinal relaxation times (T1).

(a) 3D volumetric, radiological display of ZI T1 values (msec). (b) ZI voxels in the 2D gradient coordinate space color-coded for T1 value. Voxels with high gradient 2 values are characterized by longer T1 values. (c) Comparison of the rostral ZI T1, Schaltenbrand atlas, and gradient and spectral clustering results (left to right). Rostral ZI T1 differences align with changes in gradient 2 values.

Coronal cross-sections of zona incerta gradient and spectral clustering results.

Visual display of the comparison between MRI datasets in the 2D gradient coordinate space.

(a) Each arrow in the quiver plots illustrate the shift in gradient 1 (G1) and gradient 2 (G2) values between datasets for a single voxel, color-coded for the direction of change. (b) Projection of the quiver plots into the MRI volume space for localization purposes.

Replicability and reliability of the spectral clustering results.

(a) Centroid distance (mm) and (b) Dice similarity scores, split per hemisphere and cluster.

Comparison of spectral clustering and connectivity gradients across diffusion MRI acquisition protocols.

(a) Spectral clustering results and (b) connectivity gradients obtained from the high-resolution 7T HCP dataset and an independently acquired clinical-quality diffusion MRI dataset (2 mm isotropic resolution with a reduced number of diffusion-encoding directions).

Example cortical (a) NeuroSynth (N=124) and (b) neuromaps (N=73) maps used for contextual analysis.

Each cortical map is scaled according to their minimum and maximum value.

Top 5% of connections between each zona incerta (ZI) cluster (k=6) and cortical regions.

Cortical regions from the left and right hemispheres (HCP-MMP1.0) with streamline counts exceeding the 95th percentile for each cluster are ranked in descending order. The source files used for generating these rankings are available in the code repository referenced in the manuscript.