The participants were recruited from the Metropolit Danish Male Birth Cohort [20, 21]. The cohort consists of all males born in 1953 in the Copenhagen metropolitan area (n = 12,270), and were still alive in 1968 [20]. The participants have been followed from childhood and assessed using various questionnaires, tests and data from birth certificates, school surveys and conscription board examinations. The original aim of the cohort was to investigate the social mobility, and especially intergenerational mobility. In 2001, the cohort was revived to examine the influence of early life circumstances on health in later life [22]. In 2009, approximately 400 cohort members were selected to undergo longitudinal neuropsychological and neurophysiological examinations and neuroimaging. Participants for the current study were chosen from individuals who had previously undergone these neurophysiological examinations. A total of 107 subjects were approached for participation, with 64 ultimately choosing to take part. Individuals with severe ongoing illnesses including neurodegenerative diseases, severe psychiatric diseases, addictions, or other conditions that rendered participation impossible were excluded, otherwise, the selection was random.
The participants underwent MRI to examine cerebrovascular function and PET scanning to assess Aβ accumulation using the [11C]Pittsburgh Compound-B (PiB) radiotracer. All subjects were also PET scanned using the [18F]Fluorodeoxyglucose (FDG) radiotracer to measure the relative cerebral glucose metabolism. The scans were conducted at the Department of Clinical Physiology and Nuclear Medicine, Rigshospitalet, Denmark in the period from August 2020 to July 2022. There was on average 198 ± 110 days between the MRI scans and the FDG PET scans and 269 ± 136 days between the MRI scans and PiB scans. Other assessments were done on the day of the MRI scan. The patients underwent cognitive testing from which dementia or mild cognitive impairment (MCI) could be precluded for all participants (Intelligenz-Struktur Test (IST), Mini-Mental State Examination (MMSE) and Addenbrooke’s Cognitive Examination (ACE)).
Study on the age-related change in cognitive function and Aβ accumulation, cerebral glucose consumption or permeability of the blood-brain barrier in the same subjects has been published elsewhere [23, 24]. Studies based on other subsamples of the Metropolit 1953 Danish Male Birth Cohort have examined age-related changes in brain structures[25, 26] and electrical activity [27, 28].
MRI protocol
All MRI scans were acquired on a Philips 3 T dSTREAM Achieva MRI scanner (Philips Medical Systems, Best, The Netherlands) using a 32-channel phased array head coil.
Structural images
High resolution anatomical brain images were acquired by a 3D T1-weighted turbo field echo sequence (echo time (TE) = 5.11 ms; repetition time (TR) = 11.2 ms; flip angle = 8°, field of view (FOV) = 240 × 256 × 180 mm3; voxel size = 0.70 × 0.76 × 0.70 mm3). The brains were segmented based on the Desikan-Killiany Atlas using the FreeSurfer software package (Martinos Center for Biomedical Imaging, Massachusetts, USA) [29, 30]. The cortical thickness was found by identifying the white matter to gray matter boundary and pial surface also using the FreeSurfer package [31].
White matter hyperintensities and a Fazekas score [32] were assessed by T2 weighted brain images acquired using a fluid attenuated inversion recovery (FLAIR) (TE = 125 ms, TR = 11000 ms, flip angle = 90°, FOV = 230 × 142 × 182 mm3, voxel size = 0.60 × 0.60 mm2, 31 slices, slice thickness = 3.5 mm).
Cerebrovascular function
The change in CBF from neuroactivation (DCBFVis.Act) was measured using a 2D gradient-echo dual-echo pseudocontinuous arterial spin labelling (pCASL) sequence (16 slices, FOV = 240 × 140 × 95 mm3; acquired voxel size = 2.75 × 2.75 × 5 mm3; reconstructed voxel size = 1.875 × 1.875 × 5 mm3; TR = 4550 ms; TE1 = 13 ms; TE2 = 31.7 ms; flip angle = 90°; label distance = 90 mm; label duration = 1800 ms; post label delay = 1800 ms, 45 dynamics, total duration = 6 min 30 s; SENSE factor = 2.3). CBF weighted images were calculated by subtracting blood-labelled and non-labelled images. Blood-oxygen-level-dependent (BOLD)-weighted maps were acquired by using the nonlabelled images acquired from the second echo. Examples of BOLD-weighted and CBF images are provided in Fig. 1B + D. For neuronal activation the subjects were visually stimulated by a flickering (8Hz) black and white checkerboard shown on a screen in the MRI-scanner. A block paradigm was used with four activation blocks lasting 45 seconds each interspaced with a black screen also lasting 45 seconds in between (Fig. 1A).
The brain areas activated by the visual stimulation were found by voxel-vise analyses of the BOLD images by linear regression models using the FSL software package [33]. The regressor of the model was created as the block paradigm convoluted with the hemodynamic response function model as a gamma distribution function (mean lag = 6 s, SD = 3 s). Before entering the modelling, the BOLD images were spatially filtered using a 5 mm Gaussian filter. The significantly activated areas were found as voxels in the z-score map with values above 3.1 (Fig. 1C). Voxels in the upper 10 percentile of Z-scores in the activated region were used to generate a functional region of interest (ROI) describing the brain region with maximal activation (ROIVis.Act.).
The ROIVis.Act was hereafter applied to the CBF maps, and the median time series of the voxels in the region were calculated. The median CBF time series were then modelled by a linear regression model using the block paradigm stimulation as the regressor, similarly to the regressor used to locate the activated brain regions (Fig. 1E). The regression coefficient of the model describes the change in CBF from the visual stimulation in percentage (DCBFVis.Act). Measurements of CBF changes from visual stimulation using a similar ASL MRI technique has been validated against accepted reference standard positron emission tomography (PET) imaging using radio-labelled water as tracer [34].
Resting regional CBF maps were additionally calculated by averaging the frames from the ASL acquisition without visual stimulation in the block paradigm. The regional maps were then quantified to ml/100g/min by normalizing to the total amount of blood entering the brain through the feeding cerebral arteries (the carotids and basilar artery) measured using velocity-sensitive phase-contrast mapping (PCM) MRI technique (velocity-encoding turbo field gradient-echo sequence; 1 slice; voxel size = 0.75×0.75×8 mm3; TE = 7.33 ms; TR = 27.63 ms; flip angle = 10°; velocity encoding = 100 cm/s; without cardiac gating).[35–37] The total amount of blood going to the brain was calculated by adding the flow from each artery and dividing with the brain mass acquired from anatomical MRI and assuming a brain density of 1.05 g/ml to achieve CBF values in ml/100g/min [38]. The calculation and post-processing of CBF data have been described in details previously [39].
PET Image acquisition
The participants were scanned using a Siemens Biograph Vision 600 PET/CT scanner (Siemens Healthcare). The participants were PET scanned two times on separate days. Brain Aβ accumulation was assessed using the [11C]Pittsburgh Compound-B (PiB) radiotracer. The PiB radiotracer concentration was measured by a 20 min static acquisition initiated 40 min after injection of radiotracer. The activity of the radiotracer was approximately 250 MBq PiB. On a seperate day, the relative cerebral glucose consumption was measured using the [18F]Fluorodeoxyglucose (FDG) radiotracer according to recent guidelines.[40] The FDG radiotracer tissue concentration was measured by a static 10 minutes acquisition initiated 45 minutes after injection of FDG. The injected activity was approximately 200 MBq.
The PET data was reconstructed using a 3D Ordered Subset Expectation Maximization (OSEM) algorithm with 4 iteration, 5 subsets and 5 mm gaussian Gaussian reconstruction filters for the PiB data, and 12 iterations, 5 subsets and 3 mm gaussian reconstruction filters for the FDG data. The reconstructed images had in-plane matrix size of 440 x 440 and 159 slices resulting in voxel sizes of 0.825 x 0.825 x 1.6 mm3. The PET data were also corrected for randoms, scatter, dead time and decay. A low dose radiation CT scan (120kV, 30 mAs, 3-mm slice thickness, 512 x 512 matrix size, voxel size 0.59 x 0.59 x 3 mm3, H19s convolution kernel) was additionally acquired for attenuation correction of each PET scan.
Statistical analysis
Imaging data postprocessing was performed blinded regarding subject. The anatomical MRI images were segmented into cortical regions based on the Desikan-Killany atlas using FreeSurfer.[30] The PET images were co-registered to the anatomical MRI images also using FreeSurfer. Similarly, the BOLD-images and CBF-weighted images from the ASL analysis were co-registered to the anatomical images. The ROIVis.Act. from the fMRI modelling was transferred to the CBF images and PET images. The mean CBF and cortical thickness in ROIVis.Act. were extracted using FreeSurfer. Similarly, the mean FDG and PiB values in the cortical regions in the ROIvis.act were extracted and standardized uptake value ratio (SUVr) was calculated using the cerebellar cortex without vermis as reference region. PiB PET scans were classified by visual reading as Aβ positive or negative by an experienced specialist in nuclear medicine (IL) as previous described with global PiB SUVr values above 1.4 as a supplementary criteria [41, 42].
The correlation between DCBFVis.Act and PiB SUVr in the activated area (ROIVis.Act) was examined using a linear regression model and Pearson’s correlation coefficient (r) (Fig. 2A). The correlations between DCBFVis.Act and FDG SUVr or cortical thickness in ROIVis.Act in the activated area were similarly examined to test whether a reduced cerebrovascular response could be a result of cerebral atrophy (Fig. 2B-C). Resting CBF in ROIVis.Act was included as a covariate in the regression models to correct for its known negative correlation with BOLD and CBF response to neuroactivation.
Lastly, correlations between PiB SUVr and resting CBF in the activated area, in the whole occipital lobe and in the total cortex were assessed by linear regression models to test whether Aβ accumulation affected the resting cerebral perfusion (Fig. 3).