Anatomical variability, multi-modal coordinate systems, and precision targeting in the marmoset brain

Abstract

Localising accurate brain regions needs careful evaluation in each experimental species due to their individual variability. However, the function and connectivity of brain areas is commonly studied using a single-subject cranial landmark-based stereotactic atlas in animal neuroscience. Here, we address this issue in a small primate, the common marmoset, which is increasingly widely used in systems neuroscience. We developed a non-invasive multi-modal neuroimaging-based targeting pipeline, which accounts for intersubject anatomical variability in cranial and cortical landmarks in marmosets. This methodology allowed creation of multi-modal templates (MarmosetRIKEN20) including head CT and brain MR images, embedded in coordinate systems of anterior and posterior commissures (AC-PC) and CIFTI grayordinates. We found that the horizontal plane of the stereotactic coordinate was significantly rotated in pitch relative to the AC-PC coordinate system (10 degrees, frontal downwards), and had a significant bias and uncertainty due to positioning procedures. We also found that many common cranial and brain landmarks (e.g., bregma, intraparietal sulcus) vary in location across subjects and are substantial relative to average marmoset cortical area dimensions. Combining the neuroimaging-based targeting pipeline with robot-guided surgery enabled proof-of-concept targeting of deep brain structures with an accuracy of 0.2 mm. Altogether, our findings demonstrate substantial intersubject variability in marmoset brain and cranial landmarks, implying that subject-specific neuroimaging-based localization is needed for precision targeting in marmosets. The population-based templates and atlases in grayordinates, created for the first time in marmoset monkeys, should help bridging between macroscale and microscale analyses.

Keywords: Marmoset; brain; coordinates; cranium; neurosurgery; subject variability.

Fig. 1.. Design and construction of non-invasive head holder equipped with multi-modal markers.

(a) A multi-modal marker container (outer diameter 3.2 mm, inner diameter 2.0 mm). (b) A head holder with six markers (yellow cylinders) (left) was fixed to the marmoset’s head noninvasively (right).

Fig. 2.

Registration pipeline between coordinates of scanner, anterior-posterior commissure (AC-PC), stereotactic, and grayordinates. (a) The workflow describes registration of CT and MR images between different coordinate systems. It contains three main processes: 1) alignment between subject’s CT and MR images, 2) cortical surface reconstruction and registration to AC-PC coordinates using MRI images and the HCP-NHP pipelines, 3) alignment to stereotactic coordinates using the CT image-derived cranial landmarks. From left top: the original CT image was first registered to the original MRI image using a marker-based fiducial registration (MBFR). Then, the CT was transformed to the AC-PC native coordinates and to the AC-PC template coordinates (#1 and #2). On the top right, the original MR image was registered to the AC-PC coordinates in native and the template coordinates using the HCP-NHP pipeline, which generates three transformations: rigid-body matrix (dashed arrows #1), nonlinear warpfield (#2), and resampling to the 4k vertices of the standard-mesh marmoset cortical surface plus the subcortical parcellation (19 parcels, 0.8 mm isovoxels). (#3). The matrix #1 is a linear (rigid-body) registration between the scanner and AC-PC coordinates, the warpfield #2 is a non-linear registration between the AC-PC native and the template coordinates. Finally, the AC-PC-aligned CT in native coordinates was aligned to the image-based stereotactic coordinates using cranial landmarks including external auditory canals and infra-orbital ridge (#4). (b) A midline sagittal T1w image aligned in the AC-PC native coordinates. (c) Maximum intensity projection of the CT image aligned in stereotactic coordinates, which are orthogonal to the horizontal zero plane passing through both sides of infra-orbital ridges and the interaural line. (d) The midthickness surface vertices of 4k CIFTI ‘grayordinates’ in the AC-PC template coordinates. Abbreviations: HCP-NHP, non-human primate human connectome project; BBR, boundary-based registration.

Fig. 3.

Comparison of accuracy and precision between registration methods. (a) Comparison of the marker registration error (MRE) between normalized mutual information (NMI), boundary-based registration (BBR) marker-based fiducial registration (MBFR) and MBFR followed by BBR methods. Registration between CT and MR images of marmosets (N = 10) was performed with NMI and BBR using full and partial field-of-view (FOV), and MBFR. Note that MBFR provided significantly smaller MRE in comparison to other approaches (p-values ** ≤ 0.01, **** ≤ 0.001). The threshold of failure was defined as 1 mm (dotted line). (b) The results of MBFR between CT and T2w. Note the small but clearly visible error in registration (yellow arrows) of inner cranium boundary (red line) extends into the brain parenchyma, (c) MBFR followed by BBR tuneup shows more precise alignment of cranium inner surface to the outline of cortical surface. Note also that the outer cranium boundary was also well aligned to the signal loss boundary of the T2w (aqua arrows). Study ID: (CT: 19042302, MRI: A19042302).

Fig. 4.

Marmoset intersubject variability in AC-PC and stereotactic coordinates.

Fig. 5.

The stereotactic positioning bias and reproducibility. (a) Cranial contours of five repeated positioning in five subjects in manually positioned device-based stereotactic coordinates. Each colour indicates a subject’s cranial contour, and the crosshair is placed at the centre of the tip of the ear bar. The cranial contours demonstrated variability both in intrasubject (e.g., experimenter’s positioning reproducibility) and in intersubject (e.g., animal’s cranial shape and size). Note that the variability of the contours is particularly evident in the dorsal convexity of the cranium, which is distant from the ear canal. (b) Cranial contours of five subjects in ‘true’ image-based stereotactic coordinates. Note that the locations of the cranial contours are highly variable across subjects although the fixation points (ear canal and orbital ridge) are well colocalized across subjects. In (a) and (b), the colours indicate different subjects. Inset denotes two lines in each colour, one for the outer cranium and the other for inner cranium boundary. (c) Bias of the device-based stereotactic coordinates in rotations (left) and translations (right) relative to the ‘true’ image-based stereotactic coordinates (N = 5). (d) Intra-class correlation (ICC(1,1)) of rotations and translations of device-based stereotactic coordinates (N = 5, n = 5, total 25 experiments).The error bars indicate the 95% confidence interval. Study ID: (CT: 19051001, 19082102, 19082103, 19082104, 19082105).

Fig. 6.

Variability of cranial sutures and bregma in marmoset. Each panel shows the maximum intensity projection of CT images in the x-y plane (N = 9), demonstrating the coronal and sagittal cranial sutures and the estimated location of bregma (black arrow). The lines indicate the y-axis in the midline (red line), x-axis of the interauricular line (green) and AC origin (blue). The bregma was defined as the midpoint of the curve of best fit along the coronal suture (Paxino and Franklin 2019). Note that the coronal suture splits in some of the subjects (e.g., #3) and in other subjects turn sharply (e.g., #5, 6) just before midline, which makes determining the bregma ambiguous based on extrapolation of coronal sutures from lateral to medial. The right panel of the animal #9 shows a photograph of the dorsal view of exposed cranium and sutures. In one of the subjects (#10), cranial sutures and bregma could not be reliably identified (data not shown). Study ID: (CT: 19021301, 19022601, 19022602, 19022603, 19022701, 19022702, 19042301, 19042302, 19060401).

Fig. 7.

Landmark variability across common marmosets. (a) Cranial landmark (bregma, inion, rhinion, zygion-left, and zygion-right.) intersubject variability (N = 10) displayed on the AC-PC template pial surface. (b) The deepest points of the intraparietal sulcus (IPS) and their intersubject variability displayed on the white matter surface of the AC-PC template (white nodes, N = 20). The inset shows the macroscopic view of the ex vivo brain, demonstrating the IPS in both hemispheres. (c) Midthickness surfaces in the AC-PC native coordinates in an animal with moderate IPS (left) and with negligible IPS (middle). The right panel shows all the midthickness surfaces (N = 20) warped into the AC-PC template coordinates (right), demonstrating the cross-subject variability of midthickness surfaces around IPS even after non-linear volume registration to the AC-PC template coordinates. Study ID: (CT: 19021301, 19022601, 19022602, 19022603, 19022701, 19022702, 19042301, 19042302, 19060401, 19060402, MRI: A19021301, A19022601, A19022602, A19022603, A19022701, A19022702, A19042301, A19042302, A19060401, A19060402, A17051101, A17041201).

Fig. 8.. The MarmosetRIKEN20 multi-modal templates (version 1.0) in AC-PC and grayordinates.

(a) The multi-modal templates with T1w (greyscale, N = 20) overlaid with thresholded CT (red-yellow, N = 10). Each subject’s CT image was registered to T1w in AC-PC native coordinates using the MBFR + BBR, warped to the AC-PC template coordinates using an MRI warp field, and averaged across subjects. Note that physiological calcifications were found bilaterally in the globus pallidus (green arrow, z = 0) and dentate nucleus (cyan arrow, z = −5). (b) the subcortical grey matter atlas of MarmosetRIKEN20 including 21 subcortical grey matter regions, and anterior and posterior commissures (colour coded, outlined by black line) overlaid on the T1w template image (grey colour). Readers can find annotations of each region by accessing the data at BALSA database. (c) T1w divided by T2w myelin map overlaid on the average midthickness surfaces of MarmosetRIKEN20. (d) Surface version of marmoset cortical parcellation atlas of Paxinos, Watson, Petrides, Rosa and Tokuno (Paxinos et al., 2012) including 116 cortical areas. (e) Outlines of the cortical parcellations overlaid on the myelin map (left, lateral; right, medial view). Note that high myelin contrast is colocalized with the parcellations at MT, somatomotor sensory areas (4ab, 3a, 3b) and visual cortex (V1), and relatively high myelin with the area 8av, frontal eye field (FEF). Data at BALSA: https://balsa.wustl.edu/study/p005n

Fig. 9.

Neuronavigation strategy for cortical and subcortical targets in marmosets. For cortical surgery, the target is first identified on the vertex in cortical 2D coordinates (in either native or template midthickness surface). Next, the 3D coordinates corresponding to the vertex number of interest are read in the subject’s AC-PC native coordinates. For subcortical surgery, the target is identified in the 3D coordinate system (either in template or native coordinates). When the template coordinate system is used, the target’s 3D coordinates are warped to the subject’s AC-PC native coordinates. The robot-guided neurosurgery utilises these 3D coordinates by transforming from subject’s AC-PC native to the stereotactic coordinates. Study ID: (MRI: A17051101).

Fig. 10.

Exemplar application of neuronavigation system. (a) The targets were set in the substantia nigra (SN) and the caudate nucleus (Cau) in the AC-PC template coordinates. Template is overlaid with Paxinos atlas (Paxinos et al., 2012).The target locations were non-linearly warped to the subject’s AC-PC native coordinates (left panel to middle panel). The subject’s MR and CT images are pre-registered by the marker-based fiducial registration (MBFR) and then fine-tuned by the boundary-based registration (BBR) methods (middle and right panels). The neuronavigation robot imports the MR or CT images in subjects’ AC-PC native coordinates to navigate to the target in the surgery space. (b) The cannula-insertion positions were identified with respect to the location over the surface of the cranium. Presurgical trajectories aiming for Cau (pink) and SN (red) are shown in native MRI space. Note that the cannula-insertion trajectory to SN avoided Cau and ventricles (middle panel). (c) Postoperative MR images. Confirmation of guide cannula insertion to SN and to Cau. The tip of the cannula was planned at a position of 1.6 mm from targets, because the needle used for the drug administration extends 1.6 mm from the tip of the cannula. (d) Subcortical atlas (in color) registered to AC-PC native coordinates with the target position of the SN (cyan point, middle panel), the planned trajectory in the preoperative MR image (green line, middle panel), and the position of the cannula position (arrow head) and planned trajectory (green line) in the postoperative MRI (right panel). (e) Subcortical atlas (in colour) registered to AC-PC native coordinates with the target position of the Cau in AC-PC native coordinates (cyan point, middle panel), the planned trajectory in the preoperative MR image (green line, middle panel), and the position of the cannula position (arrowhead) and planned trajectory (green line) in the postoperative MRI (right panel). Study ID: (CT: 19060401, MRI: A19060401, A19081902).

Fig. 11.

Variability of the bregma and brain volume across species. The variability of the bregma is larger in marmosets compared to rats and mice. Thus, the bregma is not recommended as a cranial landmark in marmoset neurosurgery. The isometric scale ratio of the 3D brain volume (or variability) was calculated using the ratio between the cube root of each species’ brain volume (or its variability) relative to that in mice. The isometric scale ratio of bregma variability in the Y-direction was calculated using the ratio between bregma variability in each species relative to that in mice. All the marmoset data is from the current study, whereas the rodent bregma data from Paxinos and Watson, 2017 and Paxinos and Franklin, 2019, and brain volume data from Hasegawa et al., 2010 and Ma et al., 2008. Note that all the data is from animals housed for experimental use, but sampling is not harmonised (e.g., age and degree of in-breeding) across species.