Study Design
This investigator-sponsored phase I/II trial was a single center, exploratory clinical trial (NCT03119961) initiated at Hôpital de la Pitié Salpêtrière (Paris, France). The primary objective was to evaluate changes on PET imaging on β-amyloid and glucose in the region of interest (ROI) targeted by the ultrasound device.. Secondary objectives were to assess the radiographic and clinical tolerance of repeated BBB opening by ultrasound and to examine the opening of the BBB on T1-weighted (T1w) MRI, and to study the evolution of cognitive decline. The study was approved by the Paris VI Ethical Committee. Informed consent was obtained from all participants.
Patients between the ages of 50-85, with early-stage AD (mini-mental state examination [MMSE] 20-26) were eligible. Inclusion was based on cognitive assessment (24) and an MRI showing one of the three most frequent phenotypic presentations of the disease (hippocampal amnesia, logopenic aphasia, or posterior cortical atrophy syndrome). Diagnosis was confirmed by the presence of cerebrospinal fluid levels of ptau/Αβ1−42 > 0.11 (25).
No control subjects were included in the study, but 45 controls were sampled from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu) through a matching procedure taking into account age, gender, MMSE score and diagnosis (mild cognitive impairment/AD).
Ultrasound device
The SonoCloud-1 implantable ultrasound device (CarThera, Paris, France) was used for sonications (Figure 1). This investigational device was previously used in a Phase 1/2a study in patients with recurrent glioblastoma who had monthly repeated ultrasound-mediated BBB opening prior to receiving carboplatin chemotherapy (17,18). The device consisted of a 10-mm diameter, 1 MHz ultrasound implant encapsulated in a biocompatible housing. The device was placed in a 12-mm diameter burr hole in the left parietotemporal junction targeting the left supramarginal gyrus using a neuronavigation system under local anesthesia. To activate the device, it was connected using a transdermal needle to a radiofrequency generator, with the first activation occurring at least 15 days after device implantation. During sonications, a 25,000-cycle pulse was used every second for a duration of four minutes in combination with intravenous injection of SonoVue® microbubbles (0.1 mL/kg, Bracco). The device was activated every two weeks over the course of seven sessions after patient inclusion (3.5 months) The acoustic pressure, initially set at 0.9 MPa was increased after the first sonication session to 1.03 MPa. At nine months after implantation, the device was explanted.
Safety Assessments
Safety assessments included physical and neurologic examinations, and the collection of adverse-event (AE) data according to the Common Terminology Criteria for Adverse Events (CTCAE) v4.0.
Cognitive Evaluation
The neuropsychological evaluations were performed at baseline, four months and eight months. The evaluation comprised the MMSE (26), Clinical Dementia Rating Scale Sum of Boxes score (CDR-SB) (27,28), Frontal assessment battery (FAB) (29), the free and cued selective reminding test (FCSRT) (30), trail making test (TMT) (31), verbal fluencies (32), praxis (33), Rey’s figure (34), State-Trait Anxiety Inventory: STAI (Form Y) (35) and Montgomery-Åsberg Depression Rating Scale (MADRS) (36).
MRI/PET Imaging Acquisition
MRI imaging was performed following the BBB opening procedure during the first and third sessions on a 3T Prisma Fit (Siemens, Erlangen, Germany) using a 64-channel head coil for signal reception. T2-FLAIR weighted images (1 mm isovoxel) and diffusion weighted images (2 mm isovoxel) were acquired for monitoring for any potential edema induced by the BBB disruption procedure. Quantitative susceptibility mapping images (1 mm isovoxel) were obtained using multi-echo T2*-weighted images to detect any potential hemorrhages. To evaluate BBB disruption, T1 maps (1 mm isovoxel) were then obtained with the MP2RAGE sequence before and seven minutes after a bolus injection of 0.2 mL/kg gadolinium-based contrast agent (Gd-DOTA, DOTAREM, Guerbet, France). These images were planned to be performed 60 minutes after sonication.
PET imaging to examine amyloid and glucose in the brain was performed at 0, 4, and 8 months after subject inclusion. PET acquisitions were performed on the PET/MR SIGNA 3T system (GE Healthcare) after implantation of the device. The two acquisitions took place 48 hours apart. Amyloid PET imaging started 50 minutes after intravenous injection of 370 MBq of 18F-Florbetapir and FDG PET imaging started 30 minutes after intravenous injection of 2 MBq / kg of 18F-fluorodeoxyglucose (FDG). During the period of FDG tracer uptake, participants were at rest with eyes open but ears closed to minimize MRI scanner noise. The 50-minute post-injection start time for amyloid PET was used to maximize a pseudo-equilibrium state. For both radiotracers, acquisition parameters were as followed: simultaneous PET/MRI acquisition with i) 20-minute PET acquisition ii) acquisition of four T1 DIXON sequences: in-phase, opposed-phase, fat-only, and water-only and a zero echo time sequence to capture bone information (37); the five images are combined to create a µ map, used for attenuation correction of the images. iii) 3D T1w anatomical sequence.
BBB disruption efficiency
To evaluate BBB disruption efficacy, the map of Gd-DOTA concentration was calculated from the difference of the registered T1 maps, considering a T1 relaxivity of 4.5 mM−1.s−1 (38). As a metric for BBB disruption efficacy, the total quantity of Gd-DOTA in sonicated brain tissues was calculated in a 15x55mm cylindrical ROI covering the ultrasound beam generated by the implant and compared with the Gd-DOTA quantity in a symmetric contralateral control ROI (Supp Fig. 3). The volume of brain voxels with an enhanced concentration of Gd-DOTA was also calculated in the ROI, using a concentration threshold automatically adjusted such that less than 5% of the control ROI was classified as enhanced. An ultrasound-mediated BBB opening was considered successful if the quantity of Gd-DOTA in the ROI was greater than the quantity in the symmetric control ROI plus two standard deviations of all control ROIs. BBB opening was also visually assessed as in our previous study (17).
Image processing
Scans (timeframes already averaged for PET) were visually inspected for anatomical completeness, subject motion, and other artifacts, and converted to NIFTI format. MRI scans (0.488 x 0.488 x 1.2 mm) were resliced to 1x1x1 mm and processed using CorInsights MRI, which uses Freesurfer 6.0 and other algorithms for segmentation. PET scans were co-registered to their corresponding resliced volumetric MRI scans as produced by Freesurfer. PET scans obtained at four and eight months were additionally coregistered to their corresponding initial scans. Baseline MRI scans were spatially transformed to template space using SPM12, and the transformation applied to the co-registered PET scans.
Amyloid PET analysis
To confirm the presence of amyloid at baseline and to assess longitudinal changes in brain regions that are typically amyloid positive in Alzheimer’s disease, values were measured in posterior cingulate, precuneus, lateral temporal, frontal, and anterior cingulate regions. SUVRs were evaluated using whole cerebellum and eroded subcortical white matter as comparative reference regions. SUVRs in global cortex and a relatively large temporoparietal region were also measured using additional processing and reference region approaches as described under Additional Analyses below. A visual read was also performed at baseline.
To evaluate local sonication effects, a custom volume of interest was created for each participant centered at the implant location and extending inward approximately perpendicular to the skull at the position of the implant, with initial dimensions of 10x10x40 mm3. A thresholded version of each volume was created to eliminate cerebrospinal fluid (CSF) from the measured boundaries. Additional custom volumes of interest were created to measure distal tissue in the same coronal slices within the same hemisphere, as well as in the opposite hemisphere, serving as comparator ROIs (Figure 2). The ROIs local to the implant were additionally restricted to include only gray matter to assess the effects of including white matter (which provided a slightly larger ROI less vulnerable to technical or motion-related variability) upon measured values.
Standardized Uptake Value Ratios (SUVRs) were calculated as the ratio of the value in the implant ROI divided by the value in each of the comparator ROIs as the reference (same hemisphere and opposite hemisphere, separately). This approach minimized technical variability that can arise from using a reference region located in distant slices of the brain, while maximizing similarities in tissue kinetics. SUVRs were also calculated relative to the overall bilateral parietal region. This provided a comparison to similar tissue at a similar general spatial location within the brain but with larger volume to reduce technical noise. Whole cerebellum and white matter were also evaluated as reference regions but this was for information only given technical noise associated with the cerebellar reference in longitudinal measurement (39) and the better spatial and tissue type match obtained using adjacent and opposite hemisphere tissue as the reference.
The reliability of the target region amyloid measures for each subject was assessed by determining whether unacceptable embedded head motion had occurred during the scan. Motion would be indicated by spiral artifact in the MRI scan (acquired in the same session and position as the PET scan) and/or by longitudinal change in regions distant from the implant that were well beyond the range expected over 4 and 8 months physiologically based upon numerous studies (39).
FDG PET analysis
FDG PET scans were evaluated using the same standard and custom ROIs and reference region approaches that were defined for analysis of the amyloid scans. In addition, voxel-based multivariate machine learning classification software was applied to explore patterns discriminating baseline, month 4 (M4), and month 8 (M8) states. Briefly, the spatially normalized, smoothed FDG scans for the three timepoints were grouped into three (N) training classes. Using the NPAIRS software framework, principal components (PCs) were determined for the data set, after which canonical variates analysis (CVA, a form of linear discriminant analysis) was used to mathematically combine the most significant PCs into N-1 (two) patterns of hypometabolism and hypermetabolism or preservation relative to whole brain. The data set was split into halves numerous times, each time using each half to generate a model (patterns) and generating a reproducibility metric as the correlation between the patterns, and a prediction metric based on the classification of one half from the other half’s model (40). Consensus patterns were derived based upon these metrics, and the scores for the primary pattern (CV1) compared across groups and individuals.
Global Scale Analyses & Comparison With An External Control Cohort
Potential effects upon amyloid and FDG PET were additionally evaluated by comparing SUVRs for whole cortex and a temporoparietal ROI (comprising the angular, supramarginal and superior temporal gyri) in the hemisphere of the implant versus the opposite hemisphere and a slightly different set of processing steps described in supplementary materials. This provided a comparator to the standard ROI analyses that had been performed using somewhat different technical approaches and reference regions.
As no control group was recruited in the study, we selected patients from the ADNI database (41). The selection was made from ADNI subjects who had at least two sessions (on average 30 months apart) with T1w MRI, FDG and amyloid PET data, and with an MCI or AD diagnosis at the considered sessions (42,43). The images were processed using the same procedure as the one used for the study subjects but restricted to two time points. A 5:1 patient matching was performed on age, gender and MMSE (caliper =2) using the nearest-neighbor matching method without replacement.
Statistical analysis
Evolution in regional PET SUVR was tested between M0 and M8 for both the FDG and amyloid tracers using the Wilcoxon signed-rank test. Evolution in neuropsychological scores were compared between M0 and M4; M4 and M8 as well as between M0 and M8, using Wilcoxon signed-rank tests. To correct for multiple testing, we used the Benjamini-Hochberg method. We compared our population to the control ADNI population in terms of demographic characteristics such as age and MMSE and in terms of regional PET SUVR computed at baseline in both the large and small ROIs using Kruskal-Wallis H-test. We also compared the annualized percent change in cognitive and PET SUVR between these two groups. The Benjamini-Hochberg method was also used to correct for multiple testing. All patients with at least one sonication performed were analyzed for efficacy. Safety was described on all included patients.