Sample population
The study used data from the Australian Imaging, Biomarkers and Lifestyle (AIBL) cohort, of which the design, enrolment process, neuropsychological assessments and diagnostic criteria have been previously described [10]. Of the 1572 participants enrolled in the AIBL study, we restricted the analysis to individuals having both genotype and imaging data available (N = 780). Longitudinal data was collected every 18 months over multiple years (mean = 4.8 years, SD = 2.1). Participants were classified as those with Mild Cognitive Impairment (MCI) [11] or AD [12] when the clinical criteria for diagnosis were met. In the absence of these features a classification of Cognitively Normal (CN) was given by a clinical review panel, blinded to Aβ-PET status (see below). Ethics approval for the AIBL study and all experimental protocols was provided by Austin Health, St Vincent’s Health, Hollywood Private Hospital and Edith Cowan University. All experiments and methods were carried out in accordance with approved guidelines and regulations and all volunteers gave written informed consent before participating in the study.
MRI and PET imaging
All subjects underwent a 3 T MRI and Aβ-PET imaging. T1 MPRAGE MRI was obtained at 3 T using the Alzheimer’s Disease Neuroimaging Initiative (ADNI) magnetization-prepared rapid gradient echo (MPRAGE) protocol, with in-plan resolution of 1 × 1 mm and 1.2 mm slice thickness. Freesurfer was used to estimate all cortical volumes from the T1 [13]. All volumes were corrected for age and ICV using a regression approach and reference population composed of healthy subjects (CN, Aβ negative, MMSE> 28, CDR = 0, APOE non-ε4). Left and right volumes were averaged.
Aβ-PET imaging was performed with one of five radiotracers: [11C]-PiB, [18F]-flutemetamol (FLUTE), [18F]-florbetapir (FBP), [18F]-florbetaben (FBB), and [18F]-NAV4694 (NAV). A 20- minute acquisition was performed 50 minutes post-injection of PiB, NAV and FBP, and 90 minutes post-injection of FLUTE and FBB. Due to the difference in SUVR dynamic ranges of each Aβ tracer, the Centiloid (CL) scale was used to provide a standard scale for Aβ-PΕΤ quantification [14], with 0 representing the typical Aβ-PΕΤ in young controls, and 100 the typical Aβ-PET in mild AD patients. Values equal to or above 20 CL were considered representative of abnormal levels of Aβ deposition [15]. CL were generated using CapAIBL software [14] and estimated in three regions of interest (whole neocortex, frontal and posterior cingulate). CL values were also projected onto the individual cortical surface and then transferred to a cortical atlas where statistical analyses were performed [16].
Cognitive scores
All AIBL participants complete a battery of neuropsychological tests as previously described [10]. The resulting data were used to calculate cognitive composite scores to assess recognition memory, executive function and episodic recall memory. Briefly, the composites were computed by standardising the outcome measure for each neuropsychological test to be included, using the baseline mean and standard deviation for the cognitively normal sample, then averaging those standardised scores. Each composite consists of the following tests; recognition memory (California Verbal Learning Test Second Edition and Rey Complex Figure Test), executive function (Controlled Oral Word Association Test and Category Switching), and episodic recall memory (California Verbal Learning Test Second Edition, Logical Memory II, and Rey Complex Figure Test) (Harrington reference). Clinical Dementia Rating (CDR) sum of boxes (CDRSB) score was also used to assess clinical progression.
Age of onset definition
The age of onset of abnormal levels of amyloid deposition was determined as follows: First a progression curve giving for Aβ-amyloid deposition as a function of disease progression time was constructed as described in [17, 18]. Then, the participants’ age of onset was estimated by using the progression curve to calculate the elapsed time between a participant passing the CL threshold and reaching their mean longitudinal Aβ-amyloid levels. Finally, this value was subtracted from their mean longitudinal age to get the age at onset.
Genetic data and Polygenic hazard score
Genome wide genetic data was ascertained from the OmniExpressHumanExome+ BeadChip (Illumina, USA) as previously described [10]. A polygenic hazard score was then derived from this genetic data for each individual following the methodology described previously [9]. Briefly, the approach consists in three steps. First, a list of 1854 SNPs implicated with AD (p-values < 10− 5) was extracted from the published summary statistics (p-values and odds ratios) generated by the IGAP consortium [4]. Second, a forward stepwise regression to identify a subset of 31 SNPs, in addition to the two APOE variants, that were associated with AD age of onset. Finally, for each patient, a polygenic hazard score predicting the individual’s risk of developing AD, given their polygenic profile and age was derived. In the analyses we assessed the PHS as a continuous measure, quantifying individual risk for AD, and as a dichotomous variable (high and low). To directly replicate the findings from Tan et al. [19], we used the same grouping method, defining high PHS by 1 standard deviation (SD) above the mean and low PHS by 1 SD below the mean (Supplementary Fig. 1).
Statistical analyses
Association with cross-sectional Aβ deposition
We used linear regression to investigate the relationship between PHS and regional brain Aβ deposition at baseline. Three main areas were investigated: neocortical, frontal cortex and posterior cingulate. In this cross-sectional analysis, we controlled for age, gender, level of education in years and APOE ε4 status (0 = no ε4 allele, 1 = 1 or 2 ε4 alleles). To evaluate the contribution of PHS and APOE ε4 status in the linear regression, we used likelihood ratio tests to compare models with and without these terms.
To further assess the association between PHS and Aβ deposition, beyond the role APOE, we performed the same analysis in two sub-cohorts containing exclusively APOE ε4 carriers (N = 278; CN = 161, MCI = 58, AD = 59) and non-carriers (N = 502; HC = 412, MCI = 66, AD = 24). In these subsequent analyses, we used the same linear regression, however, we did not control for APOE ε4 status due to the lack of variation in these sub-populations. All the results were adjusted for multiple comparisons using false discovery rate (FDR). The same analysis was performed at a vertex level on a template cortical surface.
Association with regional brain atrophy
We used linear mixed-effects models to evaluate the relationship of PHS with longitudinal volume change in 33 regions of interest from the Desikan-Killiany atlas in Freesurfer [20]. As volumes were previously controlled for intra-cranial volume (ICV) and age using a healthy sub-population we only controlled for gender, level of education in years and APOE ε4 status (0 = no ε4 allele, 1 = 1 or 2 ε4 alleles). In these analyses, we also controlled for sex, education, and APOE status. We then examined the simple effects by comparing slopes of volume loss over time for individuals at high (+ 1 SD) and low (− 1 SD) levels of PHS [19, 21].
Association with longitudinal cognitive decline
For comparative purposes, we used the same linear mixed effects models as described in the original Desikan paper [19]. The only deviation made from this model was the use of the CL value from the frontal cortex instead of the standard uptake volume ratio (SUVR). Therefore, the final linear mixed effect model was defined as follows:
$${\Delta}_c={\beta}_0+{\beta}_1\ PHS\ast Time+{\beta}_2\ Centiloid(fc)\ast Time+{\beta}_3\ enthorinal\ cortex\ volume\ast Time+{\beta}_4\ Baseline\ Age\ast Time+{\beta}_5\ Sex\ast Time+{\beta}_6\ Education\ast Time+{\beta}_7\ APOE\ast Time+\left(1| Patient\right)$$
This model was used to investigate the association between PHS and cognitive decline and clinical progression rate (represented as Δc) across four measures: recognition, executive function, episodic recall and CDR-SB. In this model, Time represented the number of years since the baseline visit. The APOE term indicated the presence/absence of APOE ε4 allele, encoded as a binary variable (0 = no ε4 allele, 1 = 1 or 2 ε4 alleles) and the term (1| Patient) corresponded to the random intercept. Continuous variables were centred and scaled in all the analyses. Further, to assess the original model and limit potential over-specification, we used a stepwise variable selection approach (backward selection) and identified a reduced model based on superior model fit (Akaike information criterion).
Association with age of onset of abnormal levels of Aβ deposition
Cox proportional hazards models of survival were performed to compare the time taken to reach abnormal levels of neocortical Aβ between participants with low versus high PHS scores (threshold at 1.04), adjusted baseline age, gender and years of education. The definition for survival time was the number of years between birth and a) having a PET scan indicating abnormal levels of Aβ (classed here an event, age of onset), b) withdrawing from the study (censored), or c) the last completed follow-up examination without an event (censored). For some individuals it was necessary to impute the date at which their Aβ levels became abnormal as previously published [22]. The age at which 50% of the cohort (median age) reached abnormal levels of Aβ, was reported.