Diffusion MRI data, sulcal anatomy, and tractography for eight species from the Primate Brain Bank

We acquired eight brains from the Primate Brain Bank (PBB)—one prosimian primate (bushbaby), four New World monkeys (night monkey, woolly monkey, white-faced saki, and tufted capuchin), two Old World monkeys (black-and-white colobus and grey-cheeked mangabey), and one great ape (chimpanzee). Demographics are listed in Table 1. All samples were returned to the PBB after scanning and remain available for future research. In addition, we used data from one rhesus macaque (Macaca mulatta, male, 4.03 years of age at death) from a previous study (Folloni et al. 2019) for comparison with this most commonly studied primate. Phylogenetic relationships between all species in this study and the human are displayed in Fig. 1.

Prosimian and monkey data were acquired locally on a 7 T magnet with an Agilent DirectDrive console (Agilent Technologies, Santa Clara, CA, USA) using a 2D diffusion-weighted spin-echo protocol with single line readout (DW-SEMS, TE/TR: 25 ms/10 s; Matrix = 128 × 128; number of slices: 128; Resolution: 0.3 × 0.3 × 0.3 mm for the galago, 0.4 × 0.4 × 0.4 mm for the night monkey, 0.5 × 0.5 × 0.5 mm for the saki monkey, 0.6 × 0.6 × 0.6 mm for the woolly, capuchin, colobus, and mangabey; diffusion data were acquired over the course of ~ 52.5 h). 16 non-diffusion-weighted (b = 0 s/mm²) and 128 diffusion-weighted (b = 4000 s/mm²) volumes were acquired with diffusion directions distributed over the whole sphere. The brains were soaked in PBS before scanning and placed in fluorinert during the scan.

Great ape data were acquired on a 7 T whole-body scanner with an 28 channel knee coil (QED) using a 3D diffusion-weighted steady-state free precession (McNab et al. 2009) protocol (DW-SSFP, TE/TR: 18 ms/22 ms; flip angle: 50°; resolution: 0.6 × 0.6 mm²; number of slices: 176; slice thickness: 0.6 mm; EPI factor: 1; diffusion data were acquired over the course of ~ 50 h). 12 non-diffusion-weighted (q-value = 20 cm⁻¹) and 240 diffusion-weighted (q-value = 300 cm⁻¹) volumes were acquired with diffusion directions distributed over the whole sphere. Accurate modelling of DW-SSFP data requires the estimation of DW-SSFP dependencies (T1, T2 and B1) (Buxton 1993). To account for this, T₁-weighted data were obtained using a turbo inversion recovery (TIR, TIs: 30, 60, 120, 240, 480 and 940 ms; TE/TR: 17 ms/1020 ms; resolution: 0.8 × 0.8 mm²; number of slices: 112; slice thickness: 0.8 mm; TIR data acquired over the course of ~ 2 h), T₂-weighted data were obtained using a turbo spin-echo (TSE, TEs: 8.7, 17, 26, 35, 52 and 70 ms; TR: 1000 ms; resolution: 0.8 × 0.8 mm²; number of slices: 112; slice thickness: 0.8 mm; TSE data acquired over the course of ~ 2 h) and B₁ maps were estimated using an actual flip angle imaging (Yarnykh 2007) sequence (AFI; TE: 2.6 ms; TRs: 7.7 and 21 ms; resolution: 3 × 3 mm²; number of slices: 32; slice thickness: 3 mm; AFI data acquired in ~ 43 s).

MGE data for surface reconstruction for the prosimian and monkey samples from the PBB were also acquired on a 7 T magnet with an Agilent DirectDrive console (Agilent Technologies, Santa Clara, CA, USA) using a 3D multi-gradient-echo sequence (MGE3D, TE1/δTE/TR = 7.5 ms/11 ms/98 ms; resolution = 0.15 × 0.15 × 0.15 mm (galago), 0.2 × 0.2 × 0.2 mm (night monkey), 0.25 × 0.25 × 0.25 mm (saki monkey), 0.3 × 0.3 × 0.3 mm (other samples); data were acquired over 1 h).

Structural scans for the great ape were acquired on a 7 T whole-body scanner with a 28 channel knee coil (QED) using a 3D true fast imaging with steady state free precession protocol (TRUFI, TE/TR: 5.06 ms/10.12 ms; flip angle: 39°; resolution = 0.4 × 0.4 mm²; number of slices: 256; slice thickness: 0.4 mm; TRUFI data acquired over the course of ~ 22 min). 4 TRUFI datasets were acquired (phase increment: 0°, 90°, 180°, and 270°) and averaged via root sum of squares.

All prosimian and monkey datasets were converted to NIFTI format, reoriented to approximate AC/PC orientation with the origin set at the middle of the anterior commissure, and preprocessed using tools from FSL (www.​fmrib.​ox.​ac.​uk/​fsl). These steps are implemented in the ‘phoenix’ module of the MR Comparative Anatomy Toolbox (Mr Cat; www.​neuroecologylab.​org). We used BedpostX to fit a crossing fiber model to each dataset (Behrens et al. 2007), allowing up to three fiber orientations per voxel. This tool produces voxel-wise posterior distributions of fiber orientations that are subsequently used for probabilistic tractography.

For the great ape DW-SSFP data, prior to analysis a Gibbs-ringing correction (Kellner et al. 2016) was performed on both the diffusion-weighted and non-diffusion-weighted datasets. To account for the DW-SSFP dependencies (T₁, T₂ and B₁), T₁ and T₂ maps were estimated from the TIR and TSE datasets assuming mono-exponential signal evolution (the Gibbs correction was also performed on the TIR and TSE data). A B₁ map was generated from the AFI data following the methodology described in Yarnykh (2007). All co-registrations between and within modalities were performed using FSL FLIRT using a 6 degrees-of-freedom (rotation and translation) transformation. These datasets were subsequently fit using custom diffusion tensor and ball and two stick models which incorporates the full Buxton DW-SSFP signal equations (Buxton 1993) using cuDIMOT (Hernandez-Fernandez et al. 2019).

Cortical surfaces were reconstructed from the structural scans using FreeSurfer v6.0 (Fischl 2012). Processing steps included tissue segmentation of grey and white matter, identification of subcortical structures, and surface reconstruction. To obtain successful reconstructions of the non-human primate brains, the FreeSurfer pipeline was complemented with tools from FSL v6.0.1 (Jenkinson et al. 2012), ANTs (Avants et al. 2011), and MATLAB (2017a, The Mathworks, Inc., Natick, Massachussets, United States).

A brain mask was used to separate brain tissue from any other structures present in the scans such as fiducial markers. Bias field correction was then applied to the images with ANTs. Voxel intensities from the original MGE and TRUFI images were inverted to obtain a T1-like contrast with low voxel intensities in gray matter voxels and high intensities in white matter voxels. For the mangabey, an average b0 volume was used as the original image instead. Next, tissue segmentation was performed using FreeSurfer. Manual corrections were made where necessary to correct segmentations of white matter, cerebellum, and subcortical structures. FreeSurfer’s default Talairach registration step was complemented by a step-by-step registration process, starting with a registration from the subject at hand to a macaque template brain (Seidlitz et al. 2018), followed by a registration from the macaque template to a chimpanzee template brain (based on data from chimpanzeebrain.org), and from the chimpanzee template to human Talairach space. The chimpanzee brain was first registered to the chimpanzee template brain and then to human Talairach space, skipping the macaque template step. This step-by-step registration process ensured that major brain structures such as the cerebellum and thalamus were correctly segmented in the smaller brains. Next, surface reconstruction was started on the high-resolution volumes. The pial surface was refined by alignment to a downsampled version of the original volumes using the high-resolution scripts from the HCP FreeSurfer pipeline (https://​github.​com/​Washington-University/​HCPpipelines/​tree/​master/​FreeSurfer/​custom). Tissue segmentations and surface reconstructions were visually checked for accuracy and consistency across datasets. It should be noted that the reliance on manual interference does produce some artifacts, especially in the territory underneath the corpus callosum. This subcortical area, however, is usually masked out when working with cortical surface data.

Surfaces were converted to GIFTI format and a midthickness surface was created by averaging the pial and white matter border surfaces. For display purposes, the midthickness surface was smoothed (strength 0.5, 15 iterations) using the Workbench Command tools of Connectome Workbench (Marcus et al. 2011).

Tractography recipes were defined using the conventions of XTRACT, a recently released tool to standardize tractography across species (Warrington et al. 2020). Each tract was defined through a seed and a target or waypoint mask, a set of exclusion masks, and potentially a stop mask. All recipes were drawn with respect to homologous sulci and gyri or as similar a location as possible. All recipes are available online (see "Data availability" statement). These tracts give a good balance between association and non-association fibers and fibers crossing through areas with and without crossing fibers.

Figure 2 shows a lateral view of a midthickness surface for the scanned prosimian and monkey samples and the macaque F99 surface (Van Essen et al. 2012) for comparison. We did not label the chimpanzee, as this great ape brain’s morphology is very different from the (pro)simian samples and a comprehensive atlas was recently published (Falk et al. 2018). Sulci were labelled predominantly using terms from the rhesus macaque or, if those were not present, the description of Connolly (1950) as a guide. As described before, the galago cortex shows very little gyrification, with the lateral and intra-parietal sulci as the only prominent sulci on the lateral surface and only dimples present in other locations. The night monkey shows only moderately more gyrification, with a quite extended lateral sulcus and a temporal longitudinal sulcus.

In the frontal cortex, most monkeys show a clear principal sulcus [previously termed the sulcus rectus, Connolly (1950)] and, except for the saki and night monkeys, a clear arcuate sulcus. The relative location of the arcuate sulcus and the presence of a spur at the back of it are variable, with some of the larger brains showing a clear posterior superior frontal dimple. On the medial wall, a cingulate sulcus running above the corpus callosum is evident in all species. However, the marginal sulcus is not present in the galago and only to a very limited extent in the night monkey.

The complexity of the occipital–temporal area also differs substantially across species, ranging from no gyrification in the galago, little in the saki monkey, to clear lunate and inferior occipital sulci in the capuchin, mangabey, and macaque. The intra-parietal sulcus is generally present, but its relationship with the posterior ends of the lateral and superior temporal sulci is variable. The most complex posterior gyrification can be seen in the woolly monkey, with distinct segments where the intraparietal sulcus is in Old World monkeys.

We were able to obtain high-quality diffusion MRI data in all species. The pattern of orientation of the principal diffusion direction allows us to identify some of the prominent landmarks of the primate white matter (Fig. 3). Between hemispheres, the corpus callosum is the most prominent bundle, but the anterior commissure was also very present and could be used as the origin of the image. The posterior commissure was sometimes more difficult to identify, but can be used to orient the images to approximate AC/PC orientation. Prominent projections fibers, especially the upcoming fibers of the corticospinal tract and internal capsule were clearly present. In addition, anterior–posterior longitudinal fibers were clearly visible, even in areas of crossing fibers where they were not evident in the primary diffusion direction.

As a first reconstructed tract, we identified the optic radiation, a projection fiber tract carrying fibers from the LGN toward the primary visual cortex. This tract was identified in all species (Fig. 4). The same was true for the dorsal part of the cingulum, a longitudinal fiber bundle running just above and along the length of the corpus callosum. However, while this fiber tract is generally a separated bundle running through the cingulate gyrus, in the smaller brains, including the galago and night monkey, it consisted of a white matter sheet that was continuous with parts of the superior longitudinal fibers (Fig. 3).

The superior longitudinal fascicles (SLF) run dorsally between the parietal and frontal cortex. These fibers run through territory with multiple crossing fibers and as a result can be difficult to reconstruct (Behrens et al. 2007; Mars et al. 2019). Indeed, parts of the SLF can be seen in the secondary fiber directions of all species, including the galago. We here sought to establish whether we could reconstruct longitudinal tracts between parietal and frontal cortex in all species. To increase the sensitivity, we aimed to reconstruct the entirety of the SLFs as a single ‘complex’, rather than distinguishing the individual subbranches as is often done in more gyrified species, including the human and macaque (Schmahmann and Pandya 2006; Thiebaut de Schotten et al. 2011). Figure 4 shows that we could identify such tracts in all species.

In the human brain, a prominent fiber bundle termed the inferior fronto-occipital fascicle (IFOF) travels through the extreme capsule, connecting parts of occipital and temporal cortex with the frontal cortex. In more gyrified species, the IFOF can be distinguished from the middle and inferior longitudinal fascicles (MdLF, ILF) running in the gyri of the temporal lobe. However, in the prosimian galago the temporal white matter consists primarily of a single sheet making it difficult to distinguish sub-tracts. We therefore reconstructed what we termed a longitudinal frontotemporal tract (LFT) that should pick up the IFOF, but might include sections of MdLF, ILF, and uncinate fascicle, by selecting the entire coronal section of the temporal white matter in the target mask. We did keep the constraint of fibers having to pass through the extreme capsule to be included in the tractogram; this was done to make it more likely we isolated fibers belonging to the IFOF. We observed such a fiber running between frontal and occipitotemporal cortex in all species studied here, including the prosimian galago. In the galago, the bulk of the tractogram runs in the middle part of the temporal white matter sheet, which is consistent with the location of the IFOF in other species.

The longitudinal frontotemporal tract and the SLF complex are association fibers, connecting parts of association cortex that have preferentially expanded in some parts of the primate lineage (Chaplin et al. 2013). As such, one might expect these fibers to occupy more of the cortical white matter in anthropoid primates than in prosimians. To account for the size of the brain itself, we compared the volume of the association tracts with those of the non-association tracts (Fig. 5). This showed, albeit qualitatively, that the SLF complex of parietal–frontal fibers was least elaborate in the galago compared to all other primates. The same relationship was not observed for the fronto-temporal fibers.