Tracts, Optic

For each subject, the dynamically acquired PET image frames were mutually coregistered to each other and to the individual's MPRAGE image using a novel method with low measured error (see Supporting Information). MPRAGE atlas transformation was computed by 12-parameter affine registration to a target image representing Talairach atlas space (Talairach and Tournoux, 1988 ) as defined by the “SN” method of Lancaster et al. (1995) . The atlas transformation for each PET frame then was computed by composition of transforms (frame→MPRAGE × MPRAGE→atlas) and the PET data were resampled to (2mm)³ atlas space (Hershey et al., 2003 (link)).
For each subject, we created an image of decay-corrected PET activity summed from 60-120 minutes after [¹⁸F]NMB injection, normalized to the mean in whole cerebellum. These images were averaged across subjects to create a composite image of averaged NMB activity during this time. Using a peak-finding algorithm, we identified regions of peak activity in the composite image mandating that local peak voxels must be separated by at least 6mm (3 voxels) and accepting only peaks that were at least 20% higher in intensity than the normalized cerebellar reference region. These parameters increased certainty that the generated list of peak regions were reasonably independent of each other. We identified anatomic labels for the peaks using the Talairach Client software (www.talairach.org) (Lancaster et al., 1997 (link), 2000 (link); Talairach and Tournoux, 1988 ). Peak regions of radioactivity were interpreted to indicate increased D2R specific binding, assuming constant non-specific uptake across the brain.
As a complementary approach, D2R binding (BP_ND) was quantified in a priori defined ROIs. The neuroimaging software Freesurfer (http://surfer.nmr.mgh.harvard.edu/) was used for segmentation of subcortical deep nuclei, frontal and temporal cortical regions, and cerebellum on individual MRs. Caudate, putamen, nucleus accumbens (NAc), thalamus, amygdala, hippocampus, various temporal and frontal cortical regions and cerebellum were identified. The cerebellum region included all gray and white matter of both hemispheres. The hypothalamus and midbrain regions were manually traced on individual MPRAGEs and added to this group of a priori ROIs. The hypothalamus was continuously traced, beginning where the optic tract merged with the optic chiasm and ending where pons was absent but mammillary bodies were present on coronal slices. The midbrain was traced on axial slices at the level of the superior colliculus, ventral to corticospinal and corticobulbar tracts and dorsally to the raphe nuclei and included substantia nigra and red nucleus regions. We eroded several structures to minimize partial volume effects on the regional PET measurements of radioactivity. For the caudate, putamen, thalamus and hippocampus regions, we combined a gaussian smoothing filter with thresholding to erode approximately one voxel from the surface of the original region. For the amygdala, we eroded one voxel from the edge on each axial slice. In this way, approximately 2 mm was removed from the surface of ROIs (Figure 1). Hypothalamus, substantia nigra, nucleus accumbens, frontal and temporal cortical ROIs were not large or thick enough to erode in this manner.
The Freesurfer- and manually defined ROIs and the cerebellar reference region were resampled in atlas space and decay-corrected tissue activity curves were extracted for each subject from the dynamic PET data. BP_ND was calculated for each ROI.

Sonications were applied transcranially under MRI guidance (see Supplementary Methods for parameters). In monkeys 1–3 (four sessions), burst sonications were delivered to individual points in the brain (35 targets overall). In the subsequent 26 sessions (monkeys 4–7), nine locations in a 3×3 grid in a single plane were targeted during each sonication (Supplementary Fig. S1B–C). During these volumetric sonications, 10 ms bursts were applied in sequence to the nine locations. The focal point was advanced to the next location every 100–400 ms, yielding an effective pulse repletion frequency at each location of 1.1–0.28 Hz. Spacing between the targets in these volumetric sonications was 2 mm, yielding a roughly cubic region of BBB disruption with dimensions of ~1 cm³.
Overall, 185 locations or volumes were sonicated in the seven monkeys. In monkeys 1–4, a range of acoustic power levels, microbubble injection/infusion parameters and brain targets were evaluated. Targets included the thalamus, putamen, cingulate cortex, visual cortex, hippocampus, and white matter structures. Sonications centered on the lateral geniculate nucleus (LGN) included the hippocampus and part of the optic tract. The third animal was tested twice over two weeks, and the fourth was tested 13 times over 26 weeks.
In the trials that targeted single locations per sonication and in 45 volumetric sonications, the microbubble USCA (Definity, Lantheus Medical Imaging) was injected as a bolus at the start of each sonication (dose: 10 µl/kg). These sonications consisted of 10 ms bursts applied at 1 Hz for 70s. Subsequent tests at 82 locations with volumetric sonication used an infusion pump (Spectra Solaris EP, Medrad) to deliver microbubbles throughout the exposures. Most (67/82) sonications with infusion used a 20 µl/kg microbubble dose and a 150 s total sonication duration; see Supplemental Methods for more details on the infusion protocol.