Participants
The study included participants from the ALFA study (Alzheimer and Families) at the Barcelonaβeta Brain Research Center [14], which aims at studying the preclinical stage of AD. The ALFA study (Clinicaltrials.gov Identifier: NCT01835717) includes 2743 cognitively unimpaired participants, including a high proportion of AD patients’ offspring, aged between 45 and 75 years. In this study, a subset of 322 participants from a nested study (ALFA + ; NCT02485730) was included. ALFA + individuals were invited based on their specific AD risk profile. This AD profile was determined by an algorithm in which participants’ AD parental history, age, number of APOE-ε4, alleles, verbal episodic memory score, and CAIDE (Cardiovascular Risk Factors, Aging, and Incidence of Dementia) score were taken into consideration. These individuals are cognitively unimpaired and, therefore, are in the preclinical stage of the Alzheimer’s continuum. In terms of main demographic characteristics, the percentages do not differ from the ALFA parent cohort. In addition, all individuals included in this study have available information on APOLIPOPROTEIN E (APOE) genotype, MRI examination, CSF biomarker levels, as well as cardiovascular risk factors (Fig. 1). Notice that the average time range between MRI acquisition and CSF sampling was 44 days (± 57 days). MRI acquisition and cognitive assessment were performed at the same visit.
Image acquisition
Scans were obtained with a 3 T scanner (Ingenia CX, Philips, Amsterdam, Netherlands). The MRI protocol was identical for all participants and included high-resolution 3D T2-weighted image: Turbo Spin Echo (TSE), voxel size 1 × 1x1 mm3, repetition time/echo time (TR/TE): 2500/264 ms, flip angle = 90°. In addition, a 3D T1-weighted Turbo Field Echo (TFE) sequence was acquired (voxel size 0.75 × 0.75 × 0.75 mm, TR/TE: 9.90/4.6 ms, flip angle = 8°) as well as a 3D T2-FLAIR sequence (TSE, voxel size 1 × 1x1 mm, TR/TE/TI: 5000/312/1700 ms). Scans were visually assessed for quality and incidental findings by a trained neuroradiologist. T1w images were segmented to compute total gray matter volume and total intracranial volume (TIV) using Freesurfer 6.0 (https://surfer.nmr.mgh.harvard.edu/). In addition, white matter hyperintensities (WMH) were segmented from FLAIR images using the Lesion Segmentation Toolbox (LST; https://www.applied-statistics.de/lst.html) for SPM12 [15].
Rating of perivascular spaces
PVS were evaluated by a radiologist using the visual rating scale developed by [16] based on T2-weighted images. The radiologist was blind to clinical assessment and quantification of variables of interest used in the study. Briefly, PVSs were quantified independently in two brain regions, including BG, and centrum semiovale (CS). PVS in BG and CS were assessed in the slice and hemisphere with the highest number and rated as 0 (no PVS; degree 0), 1 (mild; 1–10 PVS; degree 1), 2 (moderate; 11–20 PVS; degree 2), 3 (frequent; 21–40 PVS; degree 3), or 4 (severe; > 40 PVS; degree 4). Examples of the PVS rating are given in Fig. 2. Participants were dichotomized according to the severity of the ePVS rating of the BG and CS (degrees 0–2 were categorized as non-severe or 0, degrees 3–4 were categorized as severe or 1).
The intra-rater agreement rate (Cohen’s Kappa) of PVS rating was evaluated by estimating the intraclass correlation of two independent ratings from the same radiologist on a random sample of 20% of the subjects in the dataset. The intra-rater agreement analysis showed substantial reliability (κ = 0.77, p = 6.02e − 08 for BG subscale; and κ = 0.76, p = 8.2e − 10 for CS subscale).
CSF collection and measurement
The collection of CSF and measurement of biomarkers in ALFA + was previously described comprehensively [12]. In brief, CSF t-tau and p-tau were measured using the electrochemiluminescence immunoassays Elecsys® Total-tau CSF and phosphor-tau(181P) CSF on a fully automated cobas e601 instrument (Roche Diagnostics International Ltd.). The rest of the CSF biomarkers were measured with robust prototype assay as part of the NeuroToolKit (Roche Diagnostics International Ltd, Rotkreuz, Switzerland) on both cobas e 601 and e 411 instruments. All measurements were performed at the Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden.
Aβ pathology positivity (Aβ +) was defined by CSF Aβ42/40 ratio. We derived the cutoffs for each of these biomarkers using a two-Gaussian mixture modeling. The cut-off was defined as the mean plus 2 standard deviations (SD) of the non-pathologic Gaussian distribution (i.e., the Gaussian with the higher mean value for Aβ42/40 ratio and the resulting cutoff was 0.071. This approach for the definition of Aβ + has been shown to be optimal for the detection of pathophysiological changes in early stages of the Alzheimer’s continuum [17]. A total of 122 individuals in the study were categorized as Aβ + .
Risk factors assessment
Sociodemographic and clinical data were collected during face-to-face interviews by trained neuropsychologists, study nurses, and clinical neurologists. Participants' systolic and diastolic blood pressure were measured twice and the second measure was used. Total cholesterol level was obtained from a blood test (lipoprotein panel). Body mass index (BMI) was derived from the height and weight measured at the time of the interview. Physical activity was measured using the Spanish short version of the Minnesota Leisure Time Physical Activity Questionnaire [18] and participants were split into two categories: “active” (more than 150 min of moderate exercise or 75 min of vigorous exercise per week as recommended by the American Heart Association) or ‘inactive’. Participants with systolic blood pressure levels above 140 mmHg, self-reported hypertension diagnosis, or current use of anti-hypertensive medication were considered hypertensive. Diabetes and dyslipidemia status were defined from participant self-reported diagnosis. Moreover, based on these cardiovascular factors, we calculated a dementia risk score CAIDE (Cardiovascular Risk Factors, Aging, and Incidence of Dementia) to be included in the analysis. Further details of the calculation of CAIDE can be found in [19]. The APOE allelic variants ε2, ε3, and ε4 were determined from allelic combinations of the rs429358 and rs7412 polymorphisms, where the ε4 allele is the combination of the C allele at both sites [20]. Individuals were classified according to the number of ε4 alleles (non-carriers, heterozygotes, homozygotes). Allele frequencies and departures from Hardy–Weinberg equilibrium were inspected.
Statistical analysis
Differences between the degree of PVS in demographic and cardiovascular variables were assessed using χ2 tests (categorical variables), one-way ANOVA (normal continuous variables), and/or non-parametric Kruskal–Wallis tests (non-normal continuous variables).
There were differences in demographics and cardiovascular factors between the degrees of PVS. Therefore, odd ratios were calculated for associations of potential risk factors with ePVS. Associations between CSF biomarkers and ePVS (higher degree of PVS) were examined in logistic regressions (BG-ePVS) and multinomial regressions (CS-ePVS) adjusting for the previous potential confounders selected through stepwise regression (backward method) to generate minimally adjusted models. All regression models were adjusted by age, sex, APOE-ɛ4 status, physical inactivity, total GM volume, and TIV. BG-ePVS models were additionally adjusted by systolic blood pressure, and CS-ePVS models were additionally adjusted by diastolic blood pressure.
Analyses with CSF biomarkers were stratified by Aβ status, as defined by CSF Aβ42/40 ratio, to assess whether the relationship between ePVS and CSF biomarker levels differed between individuals with normal (negative) and pathologic (positive) Aβ levels. We first examined associations between ePVS and a non-pathological biomarker (Aβ40) to assess whether ePVS might be associated with overall protein clearance by CSF. We then sought associations between ePVS and the rest of CSF biomarkers with and without correction for Aβ40. This has been the final model selected in the study. All these models were corrected by relevant demographic and cardiovascular factors previously identified.
Notice that, categories of ePVS with less than 20 observations were included in the former category [individuals with degree 3 in BG-ePVS (N = 19), and degree 4 in CS-ePVS (N = 10). As a post hoc analysis, we excluded these categories, and we reproduced regression models and compared them with previous analyses to evaluate the risk of overfitting (these models produced similar results and were not presented).
Statistical significance was set at P ≤ 0.05 and corrected using pair-wise correction (group comparisons) and false discovery rate (FDR) (association models). All statistical analyses and data visualizations were carried out using R version 3.6.1.