New and Revised Subcortical Protocols.

The developed subcortical tractography protocols for the macaque and human brain. Protocols for Anterior Commissure, Fornix and Uncinate Fasciculus were revised from (31).

Tractography reconstructions of subcortical bundles in the macaque and human brain using correspondingly defined protocols.

Maximum intensity projections (MIPs) in sagittal, coronal and axial views of group-averaged probabilistic path distributions, for all proposed tractography protocols in the macaque (6 animal average) and human (average of 50 subjects from the Human Connectome Project). All MIPs are within a window of 20% of the field of view centred at the displayed slices. (A) Frontal, temporal and parietal parts of the extreme capsule (πΈπ‘šπΆπ‘“, πΈπ‘šπΆπ‘‘, πΈπ‘šπΆπ‘); frontal, temporal and parietal parts of the striatal bundle (𝑆𝑑𝐡 𝑓, 𝑆𝑑𝐡𝑑, 𝑆𝑑𝐡𝑝); and the Muratoff bundle (𝑀𝐡). (B) amygdalofugal tract (𝐴𝑀𝐹); anterior commissure (𝐴𝐢); uncinate fasciculus (π‘ˆπΉ); sensorimotor part of the striatal bundle (π‘†π‘‘π΅π‘š); Muratoff bundle (𝑀𝐡). Path distributions were thresholded at 0.1% before averaging.

Tractography mirrors tracer patterns in the macaque brain, with similar patterns in the human.

The proposed protocols were first developed in the macaque guided by tracer literature, and then transferred over to the human. Relative positioning of dMRI-reconstructed tracts was subsequently explored against the ones suggested by tracers, with good agreement in both species. (A) The dorsal-medial/ventral-lateral separation between the extreme an external capsule (here the frontal parts πΈπ‘šπΆπ‘“ and 𝑆𝑑𝐡 𝑓 shown) is present in macaque tractography, as suggested in the tracer literature. The Muratoff bundle runs along the head of the caudate nucleus. These relative positions are also preserved in the human tractography results. Tracer image modified from (52) with permission. (B) Similarly for the amygdalofugal bundle (𝐴𝑀𝐹), which runs under the Anterior Commissure (𝐴𝐢) and over the Uncinate Fasciculus (π‘ˆπΉ), we see agreement with tracer studies with respect to its location in both the macaque and human tractography (23, 26, 53). Tracer image adapted from (23) with permission. In all examples group-average tractography results are shown.

Tractography-derived connectivity patterns in the putamen resemble (for both macaque and human) termination sites identified by tracers after injections at different cortical areas (frontal, sensorimotor, parietal, temporal) in the macaque.

Left: Using macaque tracer data from 78 injections in various parts of the cortex, tracer termination sites in the putamen suggested a pattern based on the distinct cortical origin of the tracer injection sites; moving from the dorsolateral to the ventromedial putamen. Right: The path distributions of the different parts of the striatal bundle (𝑆𝑑𝐡 𝑓, π‘†π‘‘π΅π‘š, 𝑆𝑑𝐡𝑝, 𝑆𝑑𝐡𝑑) within the putamen reveal a similar pattern of connectivity to different parts of the cortex, both for macaque (top) and the human (bottom). Coronal and axial views of group-average results are shown for tractography. Cortical areas (Front: frontal cortex, Par: parietal cortex, Temp: temporal cortex, SensMot: Sensorimotor cortex) were obtained from the CHARM1 parcellation (54) in the macaque brain (for both tracers and tractography) and from the Harvard parcellation in the human (55, 56, 49).

Generalisability of proposed tractography reconstructions across data quality and individuals.

Results for the new subcortical tracts (right column) are shown against reference corresponding results for the original set of XTRACT tracts (left column), which have been widely used (31). (A) Tract similarity within and between two in-vivo human cohorts, spanning a wide range of dMRI data quality (HCP: high resolution, long scan time, bespoke setup, UK Biobank (UKB): standard resolution, short scan time, clinical scanner). Violin plots of the average across tracts pairwise Pearson’s correlations, between 1,225 unique subject pairs within and across the two cohorts, are shown. Correlations are performed on normalised tract density maps with a threshold of 0.5%. Reported πœ‡ is the mean of the correlations across tracts and subject pairs and 𝜎 is the standard deviation. (B) Tract similarity in twins, non-twin siblings and unrelated subjects. Violin plots of the average across tracts pairwise Pearson’s correlations between 72 monozygotic (MZ) twin pairs, 72 dizygotic (DZ) twin pairs, 72 non-twin sibling pairs, and 72 unrelated subject pairs from the Human Connectome Project. Heritable traits are more similar in MZ twins, equally similar in DZ twins and non-twin siblings and more than in unrelated subjects. Asterisk indicates significant pairwise comparisons between groups, as indicated by the brackets.

Identifying homologous deep brain structures (subcortical nuclei and hippocampus) across species solely by connectivity pattern similarity, obtained by the new tractography reconstructions.

Using the corresponding tracts in humans and macaques, connectivity blueprints can be calculated. These are GMxTracts matrices, with each row providing the pattern of how a GM location is connected to the predefined set of Tracts (35). Left: Starting from the average connectivity blueprints of reference human ROIs (Caud: Caudate, Put: Putamen, Thal: Thalamus, Hipp: Hippocampus, Amyg: Amygdala), KullbackLeibler (KL) divergence (or inverse similarity) maps can be computed against the connectivity blueprints of deeper subcortical regions in the macaque (Middle). The highest connection pattern similarity corresponds to the homologue macaque region of the corresponding human one. Right: Boxplots of KL divergence values between reference human regions and the five macaque ones. Blue dashed line corresponds to median KL divergence values when all white matter tracts are considered (both cortico-cortical and the new subcortical ones). Red dashed line corresponds to median KL divergence when using only cortico-cortical tracts. When cortico-subcortical tracts are included vs not, there is increased specificity/contrast in the cross-species mapping of these deeper structures. The boxplot with the lowest median divergence is shown in green in each case, indicating the best matching regions in the macaque to the human reference (i.e. caudate human reference best matches macaque caudate, putamen human reference best matches macaque putamen, etc).

Identifying homologous cortical regions across species solely by connectivity pattern similarity, obtained with and without the new tractography reconstructions.

Two pairs of neighbouring frontal regions were chosen (π‘‘π‘šπ‘ƒπΉπΆ: dorsomedial prefrontal cortex & π‘£π‘šπ‘ƒπΉπΆ: ventromedial prefrontal cortex, π‘‚πΉπΆπ‘Ÿ: rostral orbitofrontal cortex & 𝐹𝑂𝑝: frontal operculum) and their mapping from human to macaque (A) and from macaque to human (B) was explored. For comparison, we overlay in cyan the corresponding homologue regions in each species, as defined in (26). (A) Kullback-Leibler (KL) divergence maps in the macaque for a given human cortical reference region, (one region per row), representing the similarity in connectivity patterns across the macaque cortex to the average pattern of the human reference region. KL divergence maps are calculated using cortico-cortical (1𝑠𝑑 column), cortico-subcortical 2𝑛𝑑 column) and all tracts (3π‘Ÿπ‘‘ column) to highlight the effect of the cortico-subcortical tractography reconstructions in the prediction. Subcortical tracts provide larger benefits for the prediction of π‘£π‘šπ‘ƒπΉπΆ and π‘‚πΉπΆπ‘Ÿ, increasing specificity with respect to the expected borders. (B) Same as in A, but using macaque regions as reference and making predictions on the human cortex. KL divergence maps in the human for a given macaque cortical region, representing the similarity of connectivity pattern across the human cortex to the average pattern of the reference macaque region. Overall, in both species, an increased similarity to the reference regions in the homologue areas and decreased similarity across the rest of the cortex is observed, when cortico-subcortical tracts are considered (2𝑛𝑑 or 3π‘Ÿπ‘‘ column).

Connectivity patterns for neighbouring frontal region pairs, showing distinct cortico-subcortical tract contributions in macaque and human.

Considered regions are the same as in Figure 6A, i.e. 𝐹𝑂𝑝: frontal operculum, π‘‚πΉπΆπ‘Ÿ: rostral orbitofrontal cortex, π‘‘π‘šπ‘ƒπΉπΆ: dorsomedial prefrontal cortex, π‘£π‘šπ‘ƒπΉπΆ: ventromedial prefrontal cortex. Reference regions were chosen in the human cortex, shown in orange, and obtained from (26). The best matching region across the whole macaque cortex was identified by the minimum KL divergence in connectivity patterns (thresholded at the 7π‘‘β„Ž percentile in each case) and is shown in blue. Average connectivity patterns for the reference and best-matching regions are depicted using the polar plots. For each region, similarities in the connectivity patterns between the macaque and human can be observed, with the new cortico-subcortical bundles contributing to these patterns. For instance, 𝐹𝑂𝑝 has a strong connection pattern involving 𝐸 𝑀𝐢𝑓 and π‘ˆπΉ and moderately 𝑆𝑑𝐡 𝑓 and 𝐴𝐹, while its neighbouring π‘‚πΉπΆπ‘Ÿhas a stronger pattern involving 𝑆𝑑𝐡 𝑓 and π‘ˆπΉ, compared to 𝐸 𝑀𝐢𝑓. These differences are preserved across both species.

Corresponding tract protocol definitions across species.

Protocol definitions for all new (and revised) tracts in the human and macaque. Protocols were first designed in the macaque brain guided by macaque tracer literature, and then transferred over to the human. Colour-coded regions depict the seed, target and stop masks. Exclusion masks are not shown for ease of visualisation.

Improved specificity in subcortical connectivity patterns when using directly the tractography path distributions.

Subcortical 𝐺𝑀𝑠𝑒𝑏 Γ— π‘‡π‘Ÿπ‘Žπ‘π‘‘π‘  blueprints were built using: i) an intermediary whole brain tractography 𝐺𝑀 Γ— π‘Š 𝑀 matrix, multiplied by π‘Š 𝑀 Γ— π‘‡π‘Ÿπ‘Žπ‘π‘‘π‘  as done in (35) for cortical regions, ii) the intersection of the path distribution of each tract with the subcortical structures of interest. The two approaches are shown on the left and right columns for each of the macaque and human examples and for representative example tracts (rows). The latter approach resulted in improved specificity in both the macaque and human, with the tract of interest connecting more focally to the relevant subcortical nucleus. For instance 𝑆𝑑𝐡 tracts end up more specifically in the putamen, 𝑀𝐡 in the caudate, 𝐴𝐢 in the amygdala and 𝐴𝑇 𝑅 in the thalamus. All examples are shown as axial views, apart from π‘†π‘‘π΅π‘š, 𝑆𝑑𝐡𝑑, 𝑀𝐡 in the macaque that are shown in coronal views.

List of all species-matched (human & macaque) tract protocols

, grouped as cortico-cortical, cortico-subcortical and cerebellar. Columns indicate whether corresponding protocols are bilateral or not and whether they are New, Revised or not changed (Original XTRACT).

Tract reconstructions using our cortico-subcortical protocols, with good agreement between the macaque and the human.

Maximum intensity projections (MIPs) of the group-averaged path distributions for all developed tractography protocols in the macaque (6 animal average) and human (50 healthy subject average from the Human Connectome Project dataset; HCP). All MIPs are across a window (20% of the field of view) centred at the displayed slices. Thresholded path distributions are displayed with a low threshold of 0.1% (for the πΈπ‘šπΆ parts the 90π‘‘β„Ž percentile was used).

Relative positions maintained for all corresponding parts of striatal and extreme capsule bundles, across species.

Maximum intensity projections (MIPs) of the group-averaged tractography results for corresponding parts of the striatal bundle (𝑆𝑑𝐡)/external capsule and the extreme capsule (πΈπ‘šπΆ) in the macaque (6 animal average) and human (50 healthy subject average from the Human Connectome Project dataset; HCP). For each part, 𝑆𝑑𝐡 is more medial and πΈπ‘šπΆ is more lateral (with respect to each other. Tracts considered: frontal, temporal and parietal parts of the anterior limb of the extreme capsule (πΈπ‘šπΆπ‘“, πΈπ‘šπΆπ‘‘, πΈπ‘šπΆπ‘); frontal, temporal and parietal parts of the striatal bundle (𝑆𝑑𝐡 𝑓, 𝑆𝑑𝐡𝑑, 𝑆𝑑𝐡𝑝) (Table 1 in main text).

Corresponding tract reconstructions using our macaque cortico-subcortical protocols, between the two macaque standard spaces, F99 and NMT.

Maximum intensity projections (MIPs) of the groupaveraged path distributions for all developed tractography results for all developed protocols in the macaque (6 animal average) using protocols in the F99 standard space and protocols in the NMT standard space. All MIPs are across a window (20% of the field of view) centred at the displayed slices. Thresholded path distributions are displayed with a low threshold of 0.1% (for the πΈπ‘šπΆ parts the 90π‘‘β„Ž percentile was used).

Corresponding tract reconstructions using our cortico-subcortical protocols, between different data resolutions (spatial and angular) and acquisition protocols.

Maximum intensity projections (MIPs) of the group-averaged path distributions for all developed tractography results for all developed protocols in 50 Human Connectome Project (HCP) and 50 UK Biobank (UKB) subjects. All MIPs are across a window (20% of the field of view) centred at the displayed slices. Thresholded path distributions are displayed with a low threshold of 0.1% (for the πΈπ‘šπΆ parts the 90π‘‘β„Ž percentile was used).

Corresponding tract reconstructions using our cortico-subcortical protocols, across different cohort sizes.

Maximum intensity projections (MIPs) of the group-averaged path distributions for all developed tractography results for all developed protocols in 10, 50 and 339 (all unrelated subjects) Human Connectome Project (HCP) subjects. All MIPs are across a window (20% of the field of view) centred at the displayed slices. Thresholded path distributions are displayed with a low threshold of 0.1% (for the πΈπ‘šπΆ parts the 90π‘‘β„Ž percentile was used).

Individual subject tract reconstructions match the group average, whilst preserving expected topology for

𝑆𝑑𝐡 and πΈπ‘šπΆ. Showing reconstruction and relative topology for frontal and parietal parts (𝑆𝑑𝐡 𝑓 vs πΈπ‘šπΆπ‘“, 𝑆𝑑𝐡𝑝 vs πΈπ‘šπΆπ‘) for individual subjects. The subjects chosen are those corresponding to the 10π‘‘β„Ž, 50π‘‘β„Ž (median) and 90π‘‘β„Ž percentile of the distribution of tract correlations against the group average for the HCP cohort (N=50). For each subject, the mean correlation to the average across all New tracts was computed and subjects were ranked based on this mean tract correlation. In all cases, tracts reconstruct similarly to the average atlas, whilst preserving the expected topology (𝑆𝑑𝐡 more medial than πΈπ‘šπΆ).

Individual subject tract reconstructions match the group average for

𝑀𝐡, 𝐴𝑀𝐹 and π‘ˆπΉ. The subjects chosen are those corresponding to the 10π‘‘β„Ž, 50π‘‘β„Ž (median) and 90π‘‘β„Ž percentile of the tract correlations against the group average for the HCP cohort (N=50). For each subject, the mean correlation to the average across all New tracts was computed and subjects were ranked based on this mean tract correlation.. In all cases tracts reconstruct similarly to the atlas.