Generalising XTRACT tractography protocols across common macaque brain templates

Abstract

Non-human primates are extensively used in neuroscience research as models of the human brain, with the rhesus macaque being a prominent example. We have previously introduced a set of tractography protocols (XTRACT) for reconstructing 42 corresponding white matter (WM) bundles in the human and the macaque brain and have shown cross-species comparisons using such bundles as WM landmarks. Our original XTRACT protocols were developed using the F99 macaque brain template. However, additional macaque template brains are becoming increasingly common. Here, we generalise the XTRACT tractography protocol definitions across five macaque brain templates, including the F99, D99, INIA, Yerkes and NMT. We demonstrate equivalence of such protocols in two ways: (a) Firstly by comparing the bodies of the tracts derived using protocols defined across the different templates considered, (b) Secondly by comparing the projection patterns of the reconstructed tracts across the different templates in two cross-species (human-macaque) comparison tasks. The results confirm similarity of all predictions regardless of the macaque brain template used, providing direct evidence for the generalisability of these tractography protocols across the five considered templates.

Keywords: Comparative anatomy; Connectivity; Diffusion MRI; F99; INIA; NHP; NMT; Yerkes19.

Fig. 1

Axial, sagittal, and coronal views of population percentage atlases of a subset of the 42 XTRACT WM bundles for the human (adult and neonate) and the macaque brain. Reproduced from (Warrington et al. 2022). The full list of tracts is provided in Table 2

Fig. 2

Nonlinear registration of our F99 protocol ROIs to each of the other template spaces. Iterative manual modification of ROIs within each template space was used, where needed, to obtain the final protocols. These protocols were used to produce population-averaged tract atlases in each template space

Fig. 3

KL divergence in our WM-anchored common space gives the (dis-)similarity between the connectivity patterns of cortical parcels across brains and allows for identification of corresponding cortical regions between the human and the macaque. A Minimisation of KL divergence maps to find the corresponding region in the macaque brain to human cortical ROIs (sensorimotor area shown here). B Use KL divergence maps to project the human myelin map to the macaque brain

Fig. 4

Subject-wise correlation values between path distributions of tracts reconstructed using the F99-based protocols and each of the other template-based protocols. A subset of 12 (of the 42) tracts is shown comprising of the ones that required manual refinement following registration from the F99 template. For each tract and template, six samples are shown for “Original” (red asterisks, protocols are simply the F99 ones registered to new template space) and six for “Manually Adjusted” (blue circles, protocols have been refined in each template space), corresponding to the six macaque datasets we used. Adjustment was performed to improve tractography results and increase consistency across templates

Fig. 5

Examples of tracts obtained using protocols that were manually adjusted following the nonlinear registration from F99 to each other template space. A CBP and CST in NMT space, before (denoted as “Original”) and after adjustment of the registered protocol ROIs (from F99 space) (denoted as “Manually Adjusted”). B AR and CST in F99 space, before and following adjustment. Coronal sections are shown using radiological convention: R right, L left, A anterior, P posterior

Fig. 6

Group-average WM tracts from the final protocols using the five templates (F99, D99 INIA, YRK, NMT). Qualitative comparison between templates for all major tracts considered. For each tract, normalised path distributions were thresholded at 0.1% and binarised, and then averaged across the six animals. The above maps show group percentage averages of these distributions (colour codes depict from 30% to 100% of the group)

Fig. 7

Correlation (mean ± std) across six subjects for each tract between F99-derived path distributions and those derived using each other template considered. The correlation values were used for template comparison and assessment of protocol robustness and generalisability. For each tract and template space, normalised path distributions were thresholded at 0.1% and the Pearson correlation against the F99 respective path distribution was obtained within a binary mask covering the F99 thresholded path distribution

Fig. 8

Within-tract diffusion measures of microstructure — tract-wise fractional anisotropy (FA), tract-wise mean diffusivity (MD) — were used to assess protocol robustness and generalisability across templates, as more biologically relevant measures compared to correlation. A The median FA and MD values per WM tract across subjects for each template space. Values are comparable across templates. B Percent difference of FA and MD of each new template considered compared to F99. The majority of absolute differences are less than 2%

Fig. 9

Distributions of minimum KL divergence quantifying the connectivity-based (dis-)similarity between F99 and each other macaque template blueprint, as well as with the human blueprint. For each WGB location in F99 space, the divergence with the best matching vertex in other template spaces (i.e. minimum KL divergence) has been calculated. The boxplots represent the distribution of these minimum KL divergence values across the whole brain. The inset is the log₁₀ of the same KLD values. The cross-species dis-similarity (i.e. between the human and the macaque) strongly surpasses the intra-species dis-similarity (i.e. amongst macaque templates)

Fig. 10

KL divergence maps quantifying the connectivity-based (dis-)similarity between two human regions (primary sensorimotor-left and MT-right) and the macaque brain, when using tracts defined by protocols across the different macaque templates. Macaque regions with the most similar connectivity patterns to the human regions (i.e. lowest KL divergence) are shown in purple. The macaque primary sensorimotor and MT areas from the Paxinos atlas are shown for reference (bottom row). In all template spaces, the resulting patterns closely resemble the expected location of the corresponding regions in the macaque, with regions of lowest divergence overlapping with the atlas-based ROIs for M1 + S1 and MT respectively

Fig. 11

Prediction of macaque myelin map from a measured (T1w/T2w ratio) human myelin map for each template space considered (left). The KL-divergence of connectivity patterns to the reconstructed WM tracts between human and macaque is used as a transformation field. The right column shows absolute difference between each myelin map prediction and a measured (T1w/T2w ratio) macaque myelin map