Patterns of progressive atrophy vary with age in Alzheimer's disease patients

Age is not only the greatest risk factor for Alzheimer's disease (AD) but also a key modifier of disease presentation and progression. Here, we investigate how longitudinal atrophy patterns vary with age in mild cognitive impairment (MCI) and AD. Data comprised serial longitudinal 1.5-T magnetic resonance imaging scans from 153 AD, 339 MCI, and 191 control subjects. Voxel-wise maps of longitudinal volume change were obtained and aligned across subjects. Local volume change was then modeled in terms of diagnostic group and an interaction between group and age, adjusted for total intracranial volume, white-matter hyperintensity volume, and apolipoprotein E genotype. Results were significant at p < 0.05 with family-wise error correction for multiple comparisons. An age-by-group interaction revealed that younger AD patients had significantly faster atrophy rates in the bilateral precuneus, parietal, and superior temporal lobes. These results suggest younger AD patients have predominantly posterior progressive atrophy, unexplained by white-matter hyperintensity, apolipoprotein E, or total intracranial volume. Clinical trials may benefit from adapting outcome measures for patient groups with lower average ages, to capture progressive atrophy in posterior cortices.


Grey Matter Volume change without correction for APOE e4 and WMH (model 2)
GM volume change (outcome) was investigated with a main effect of group, a linear interaction term between group and baseline age, and with TIV as a covariate. See main paper for more information. GM volume change; inferences in age-diagnostic group interaction without correction for WMH or APOE genotype 2 White Matter Volume change with correction for APOE e4 and WMH (model 3) WM volume change was also investigated, with a main effect of group, a linear interaction term between group and baseline age, and with WMH, APOE e4 allele presence and TIV as a covariate. See main paper for more information.

White Matter Volume change without correction for APOE e4 and WMH (model 4)
WM volume change was also investigated, with a main effect of group, a linear interaction term between group and baseline age, and TIV as a covariate. See main paper for more information.

Regression of Baseline Brain Volume and Age
Brain volumes were estimated from the 1.5T volumetric T1-weighted images using a multi atlas template brain segmentation method (Leung et al., 2011).
To analyse the cross-sectional relationship between baseline brain volume and age at baseline a multiple linear regression was performed with baseline brain volume as the outcome, age as the predictor variable, and diagnosis and TIV as covariates. See supplementary table 2 for results.

Analyses in CSF confirmed dataset
CSF data was collected in a subset of ADNI participants (n=352), using a previously described method (Shaw et al., 2009). We classified patients by amyloid status using the abeta 1-42 cut off of 192pg/ml (Shaw et al., 2009); controls were selected for analysis if they were abeta negative (>192pg/ml) and MCI and AD patients selected if abeta positive (<192pg/ml). The final subset for analysis was 270 participants (see supplementary figure 3).
VBM steps were completed as per the pipeline described in the main paper section 2.2.1 until the DARTEL stage, in which the DARTEL registration was limited to the subset with CSF data (n= 270). The volume change maps of the subset were aligned to this CSF specific template.
The main model of GM change with age as a predictor was run in this CSF subset, WMH, TIV and APOE status (presence/absence of an e4 allele) were used as covariates, see the main paper for more details regarding the model (model 1, section 2.2.1). Models of BSI in this subset were also fitted in the CSF subset, as in main paper (section 2.2.2).

Relationships between age and cognition
We fitted multilevel linear mixed-effects regression models for repeated measures of MMSE (Frost et al., 2004). Interval in years between baseline and follow up was included as a fixed effect, in order that the resulting coefficient represented change in MMSE per year (outcome). The following covariates were included as main effects and as interaction terms with interval, in order that their inclusion could affect mean MMSE and how this changed over time: diagnostic group, an interaction between baseline age and diagnostic group, WMH, and APOE e4 carrier status (presence/absence of an e4 allele). Participant-level random effects for intercept and time since baseline MMSE measurement were included to permit between-participant heterogeneity in baseline MMSE and in rate of change in MMSE. Different random intercept and slope terms were fitted for control, MCI and AD groups, as the variability in MMSE is often higher in AD patients. In MCI and AD groups unstructured covariance of the random effects was used to allow for a correlation between baseline MMSE and rate of change in MMSE. A separate residual variance was fitted for each diagnostic group.

Analyses using age at onset
For AD patients, age at symptom onset was calculated by subtracting the year of estimated symptom onset from the date of the baseline visit, in order to give years since AD onset. Year of estimated symptom onset was estimated by the study partner, the individual accompanying the patient to visits, with 10 hours or more contact per week with the patient. Years since AD onset was then subtracted from the patient's baseline age to give age at onset.
Relationships between baseline age and atrophy rate, or age at onset and atrophy rate, were run in ADs only (as there is no equivalent for variable for age at onset for controls). Four AD participants were excluded as there was no year of AD onset information. Analyses were run to estimate change in GM, predicted by either age at baseline or age-at-onset, see main paper section 2.2.1 model 1, adjusted for APOE (presence/absence of an e4 allele), WMH and TIV and for multiple comparisons after bootstrapping, family wise error correction (p<0.05). Results using age at onset and results using baseline age were qualitatively compared.

Linearity analyses
To examine the concept of non-linearity in the age relationship a quadratic term was added to the models of GM change predicted by age for BSI and VBM (age*age*diagnostic group), (main paper model 1, section 2.2.1). Models were adjusted for WMH, age and TIV. The outcome of the non-linear term investigated whether the effect of age on atrophy rates differed for a 10 year increase in age. GM volume change; inferences in age-diagnostic group interaction without correction for WMH or APOE genotype 4

Supplementary
Total brain volume, ml 1084 (106) 1054 (97) 1096 (111) 1025 (107) 1038 (123)  GM volume change; inferences in age-diagnostic group interaction without correction for WMH or APOE genotype 5 Differences in baseline brain volume with age Supplementary Table 2: Results from the regression model assessing the relationship between cross-sectional brain volume and age. Estimates are shown for difference in brain volume (ml) with (p values) and [95% confidence intervals] for, a diagnosis of MCI a or AD b , and an increase in age of 10 years from the mean age c , conditional on intracranial volume. The effect of age c is the effect of age in controls (normal aging), the effect of age in MCI d and AD e are differences from the effect in controls.

Baseline Brain Volume
Mean Brain volume adjusted for TIV, ml GM volume change; inferences in age-diagnostic group interaction without correction for WMH or APOE genotype 6 Supplementary Figure 1: Results of the F Test to test the age-group interaction term to predict volume change (A). Clusters in the images represent voxels in which there is a significant difference in the relationship between age and atrophy rate across the three groups. Results of the T tests to directly compare the age*group interaction between controls and AD patients (B). Clusters indicate regions in which the relationships between age and atrophy are different between groups. Red clusters in the control vs AD comparison signify regions in which there is greater atrophy at younger ages in ADs, whilst for controls there is little age-atrophy relationship. Blue clusters indicate voxels which expand more at younger ages in ADs, whilst controls expand more at older ages. There were no differences between control and MCI patients. Analyses are corrected for multiple comparisons, FWE p<0.05, and TIV.  GM volume change; inferences in age-diagnostic group interaction without correction for WMH or APOE genotype 9 CSF confirmed VBM results

Supplementary Figure 4:
Results of an age-by-group interaction between controls and AD patients in the subset with confirmed abeta 1-42 status (no amyloid pathology in controls and confirmed amyloid pathology in AD) (a) and the full dataset (b).Clusters indicate regions in which the relationships between age and atrophy are different between groups, i.e. differences in age-by-group interaction. Red clusters signify regions in which there is greater atrophy at younger ages in AD patients, whilst for controls there is little age-atrophy relationship. Analyses are corrected for multiple comparisons, FWE p<0.05, and are also corrected for APOE genotype (presence/absence of an e4 allele), TIV and WMH volume.   Table 3: Results from the regression model assessing the relationship between change in brain and hippocampal volume (left and right summed) and age by each diagnostic group (estimated using an age-by-diagnostic group interaction) in the subset with abeta status confirmed by CSF. Average brain and hippocampal atrophy rates with (p value) and 95% confidence intervals [95% CI] are shown in ml/year. Age interaction estimates (a) represent an increase in atrophy rate for a ten year increase in baseline age (ml/year/decade), adjusted for total intracranial volume, APOE genotype (presence/absence of an e4 allele) and WMH volume. For MCI and AD groups, age interaction estimates are given after subtraction of the estimate effect in controls (to account for normal aging), p values for MCI and AD indicate whether the age-atrophy relationship is significantly different from controls (*).    Table 6: Results from a regression model investigating a non-linear effect of age (predictor) on atrophy rates (outcome). Estimates are shown for an increase in atrophy rate (ml/year), (p value), and [95% confidence intervals]: a 10 year increase in age (constant age effect) a , for a change in slope of the age-atrophy rate relationship for a 10 year increase in baseline age (quadratic age effect) b . Models are adjusted for age, total intracranial volume, presence/absence of an APOE e4 allele and WMH.