Evolution of Neurodegeneration Imaging Biomarkers from Clinically Normal to Dementia in the Alzheimer Disease Spectrum

Abstract

The availability of antemortem biomarkers for Alzheimer’s Disease (AD) enables monitoring the evolution of neurodegenerative processes in real time. Pittsburgh compound B (PIB) positron emission tomography (PET) was used to select participants in the Mayo Clinic Study of Aging and the Mayo Alzheimer’s Disease Research Center with elevated β-amyloid, designated as “A+,” and hippocampal volume and ¹⁸fluorodeoxyglucose (FDG) positron emission tomography were used to characterize participants as having evidence of neurodegeneration (“N+”) at the baseline evaluation. There were 145 clinically normal (CN) A+ individuals, 62 persons with mild cognitive impairment (MCI) who were A+ and 20 with A+ AD dementia. Over a period of 1–6 years, MCI A+N+ individuals showed declines in medial temporal, lateral temporal, lateral parietal, and to a lesser extent, medial parietal regions for both FDG standardized uptake value ratio (SUVR) and grey matter (GM) volume that exceeded declines seen in the CN A+N+ group. The AD dementia group showed declines in the same regions on FDG SUVR and GM volume with rates that exceeded that in MCI A+N+. Expansion of regional involvement and faster rate of neurodegeneration characterizes progression in the AD pathway.

1. Introduction

As recently as a decade ago, disease-course models of the pathophysiological processes that comprised Alzheimer’s disease (AD) were constructed by cobbling together antemortem clinical findings and neuropathological observations. While neuropathologically-derived models (Braak and Braak, 1991; Delacourte et al., 1999; Thal et al., 2000; Braak et al., 2006) have been hugely successful in conceptualizing mechanisms in AD, the mapping from neuropathology back to live patients left many aspects of the sequence of events speculative. The recent ability to characterize individuals according to both their degree of β-amyloid burden (Klunk et al., 2004) and degree of neurodegenerative changes (as indexed by either imaging or cerebrospinal fluid biomarkers) has made it possible to establish a sequence of events in the progression of AD pathophysiological processes in living patients in real time. While the availability of Tau-positron emission tomography (PET) imaging (Johnson et al., 2015; Brier et al., 2016; Scholl et al.; Schwarz et al., 2016) offers a different route to understanding progression of disease in AD, structural MR and FDG PET might be causally more aligned with cognition than neurofibrillary tangle burden (Gomez-Isla et al., 1997; Savva et al., 2009).

The temporal order of antemortem biomarker changes provides clues about underlying neurobiological mechanisms. Elevated brain β-amyloid has been associated cross-sectionally and longitudinally with changes in neurodegenerative imaging biomarkers in non-demented individuals (Sabuncu et al., 2011; Ewers et al., 2012; Dore et al., 2013; Villemagne et al., 2013; Oh et al., 2014; Araque Caballero et al., 2015; Nosheny et al., 2015). The same has also been reported in persons with AD dementia (Sabuncu et al., 2011). However, while the prior studies (Sabuncu et al., 2011; Ewers et al., 2012; Dore et al., 2013; Araque Caballero et al., 2015; Nosheny et al., 2015) used amyloid biomarkers longitudinally to examine the consequences of elevated brain β-amyloidosis on neurodegenerative biomarkers, none conditioned the examination on the baseline state of neurodegeneration. Among clinically normal (CN) persons (Knopman et al., 2013) and in those with mild cognitive impairment (MCI) (Knopman et al., 2015), we have reported that those with elevated brain amyloid and abnormal levels of neurodegenerative biomarkers at baseline showed greater progression of atrophy and hypometabolism compared to their cognitively comparable peers without both elevated amyloidosis and neurodegenerative changes. In our prior reports, we were not able to compare across CN and MCI, nor were we able to include an AD dementia (AD DEM) comparison group. Here, we focus on the differences in neurodegeneration imaging biomarkers in persons with elevated brain amyloid across the cognitive spectrum from CN to MCI to AD DEM, to explore the consequences of the intersection of existing neurodegenerative changes and elevated β-amyloid on regional changes in cortical volume and metabolism.

2. Methods

2.1. Participants

We included participants who had at least two PET and MRI scans and a diagnosis of CN, MCI, or AD DEM at baseline. AD DEM and MCI participants were drawn from the Mayo Alzheimer’s Disease Research Center (ADRC) and the Mayo Clinic Study of Aging (MCSA), while CN participants were drawn from the latter study only. To be eligible, AD-DEM participants had to have elevated amyloid by PIB PET at baseline. The CN participants reported here included those previously reported (Knopman et al., 2013) plus a much larger group not previously reported; MCI participants were the same ones who were reported previously (Knopman et al., 2015).

The methods used for participant recruitment and diagnosis have been previously detailed for the MCSA (Petersen, 2004; Roberts et al., 2008; Petersen et al., 2010; Roberts et al., 2012; Roberts et al., 2014). A consensus diagnostic process was used to assign clinical labels of CN, MCI and dementia according to criteria established by the National Institutes on Aging-Alzheimer Association workgroups (Albert et al., 2011; McKhann et al., 2011). We allowed any neuropsychologically-defined MCI subtype. Blood was collected on all participants and APOE genotype determined.

These studies were approved by the Mayo Clinic and Olmsted Medical Center institutional review boards. Written informed consent was obtained from all participants. In the case of persons with dementia, a close family member also provided written consent.

2.2 Imaging Methods

All subjects underwent MR imaging, ¹⁸fluorodeoxyglucose (FDG) positron emission tomography (PET) and ¹¹C-Pittsburgh compound B (PIB) PET with our previously described methodology (Knopman et al., 2013; Jack et al., 2014; Jack et al., 2015; Knopman et al., 2015). Mayo ADRC and MCSA participants who agree to imaging undergo a brain MR on one day and the FDG and PIB PET scans on the following day. A CT scan was performed at the time of PET imaging for attenuation correction. The scans were done within 6 months of a participant’s clinical visit.

2.2.1. MR imaging

MR scanning was performed at 3 Tesla and the 3-dimensional magnetization-prepared radiofrequency pulses and rapid gradient echo (MPRAGE) sequences were used for quantitative analyses. The imaging parameters were: TR/TE/T1, 2300/3/900 ms; flip angle 8#x000B0;, 26-cm field of view (FOV); 256×256 in-plane matrix with a phase FOV of 0.94, voxel dimensions of 1.0156×1.0156×1.2 mm, and number of locations/slab: 170. (Jack et al., 2008).

Hippocampal volume was one of the imaging features used to classify participants at baseline as having abnormal neurodegeneration (“N+”). Hippocampal volume was measured with FreeSurfer (v5.3; https://surfer.nmr.mgh.harvard.edu/). Each subject’s raw hippocampal volume was adjusted for TIV to create a total intracranial volume (TIV)-adjusted hippocampal volume (HVa). TIV was measured using the MPRAGE MR scan in which all intracranial contents were identified. HVa was derived by calculating the residual from a linear regression of hippocampal volume versus TIV among a sample of 133 clinically normal participants aged 30 to 59 (Jack et al., 2014)).

Regional grey matter (GM) volume, both at baseline and over time, was one of the outcome measures. The TBM-SyN algorithm was used to derive GM volume estimates (Jack et al., 2014; Vemuri et al., 2015). TBM-SyN computes a symmetric deformation between each pair of scans in a subject’s time series of structural T1 images. For a given serial image pair, the late image is warped to the space of the early image, and a soft mean of the warped late image and the early image is computed, to produce a synthetic early image. In addition, the Jacobian image from the deformation is log-transformed, then annualized to account for the inter-scan interval. The synthetic early image is then parcellated into cortical ROIs, and segmented into GM using SPM unified segmentation (Ashburner, 2009) and a set of custom template, priors and anatomical ROI labels. The ROI masks are then used to quantify the amount of GM in each ROI for the synthetic early image, as well as the mean of the annualized log Jacobian in each ROI. This entire process is repeated in the opposite direction, warping the early image to the late image, to get GM estimates and rates of change in the late image space as well. This pair-wise computation is repeated on all consecutive image pairs for each subject to give estimates of GM at each time point. GM volume estimates from TBM-SyN were averaged at each time point. The values were computed for 14 cortical regions defined by the AAL atlas (Tzourio-Mazoyer et al., 2002); right and left hemisphere values were summed.

2.2.2. PIB PET imaging

PIB PET imaging was used to characterize participants at baseline as having elevated amyloid (“A+”) or not elevated (“A−“). A PIB PET scan consisted of four 5-minute dynamic frames acquired from 40–60 minutes after injection of ¹¹C-PIB. PIB values were grey matter and white matter sharpened. Partial volume correction was performed to account for voxel CSF content. The PIB PET ROIs were derived from an MR template generated from each participant. Median voxel values were extracted from each labeled cortical ROI in the AAL atlas (Tzourio-Mazoyer et al., 2002) modified in-house. A global PIB standardized uptake value ratio (SUVR) was calculated from a group of regions in both hemispheres including parietal, cingulate precuneus, prefrontal, orbito-frontal, temporal, anterior cingulate and posterior cingulate regions, with cerebellar grey matter as the reference region.

2.2.3. FDG PET Imaging

FDG SUVR was used both for baseline classification of neurodegeneration, and also as an outcome measure. FDG PET was obtained on the same day as the PIB PET scan and consisted of four 2-minute dynamic frames acquired from 30–38 minutes after injection of ¹⁸FDG. FDG values were non-sharpened and were not partial-volume corrected. FDG PET regional SUVR values were derived in a similar manner to the PIB PET scans.

Low FDG SUVR in a group of voxels typically impaired in AD was the second imaging feature used to define neurodegeneration (“N+”) at baseline (Jack et al., 2012; Knopman et al., 2013; Jack et al., 2014; Knopman et al., 2015). The voxels were located bilaterally in angular gyrus, posterior cingulate gyrus, and middle/inferior temporal gyrus. For this meta-ROI, values were normalized to the pons and vermis (Landau et al., 2011).

Change in regional FDG SUVR values were used as one of the primary outcome measures. Image values were extracted from a template created from the patient’s own MR using the AAL atlas (Tzourio-Mazoyer et al., 2002) that was modified by us. Regional values were calculated for 14 regions and averaged across hemispheres and weighted to the ROI size. For the outcome measure, FDG SUVR used pons for normalization.

2.3. Baseline Characterization of Participants

Participants were characterized as having elevated brain amyloid (A+) based on a cutpoint of 1.4 for PIB PET SUVR. This value represented the 10^th percentile of the distribution of values from a calibration sample of 75 patients with AD dementia who were age 65 or older with MRI and PET data from the Mayo ADRC or MCSA (Jack et al., 2014). A value of PIB SUVR of 1.4 was found to correspond to neuropathological Thal amyloid stage of 2 (Murray et al., 2015). Within the study cohort, 144 of 327 CN participants were A+, 62 of 96 MCI participants were A+, and all AD DEM patients were A+ by design. We also present data from the CN A− (n=183) and MCI A− (n=34) groups for completeness, but the A− groups were not the focus of this report.

The participants were further characterized as having abnormal neurodegenerative changes (N+) based on either HVa <−2.40 cm³ or FDG SUVR in the meta-ROI affected in AD <1.32. The cut-points were derived from the same AD dementia patients described in previous paragraph, and represented the 90^th percentile in AD dementia.

2.4. Outcome measures

Baseline and annualized change were examined in GM volume and FDG SUVR in 4 ROIs that are typically abnormal in AD (Whitwell et al., 2008; Dickerson et al., 2011; La Joie et al., 2012): medial temporal (MT), lateral temporal (LT), medial parietal (MP) and lateral parietal (LP). See Supplemental Figure 1 for their locations. The other 10 cortical regions were designated as non-principal ROIs.

2.5. Analyses

Linear mixed-effects regression models were fit in each ROI to assess group differences in baseline FDG SUVR and GM volumes and to assess group differences in changes over time in FDG SUVR and GM volume. In a secondary analysis, a linear mixed-effects regression model was fit to assess group differences in global PIB SUVR. Each mixed effects model included main effects of time, age at baseline, sex, group and interactions for age at baseline by time and group by time. To account for repeated measures and patient heterogeneity we included random subject-specific intercepts and slopes. In these models group differences at baseline and rates of change were adjusted for age-related baseline differences.

All outcome measures were log-transformed to reduce skewness and to allow for interpretation of slope estimates as approximate annual percent change. We report baseline and change estimates with the usual 95% confidence intervals (Cis), and also used 84% CIs because at that level, non-overlapping intervals correspond to a biomarker group difference that is significant at the P < 0.05 level (Knol et al., 2011). We did not correct for multiple comparisons. We used a parametric bootstrap approach with 5000 replicates to simulate baseline and slope coefficients to assess how much support the data provided for an ordering of average values such that AD DEM < MCI A+N+ < CN A+N+ (Gelman and Hill, 2007). The percentage of replicates whereby this ordering held can be interpreted as a posterior probability. For example, if the ordering held in 90% of the replicates, the data suggests a high likelihood that on average this ordering is true. We also assessed the correlation between FDG and GM volume at baseline and in terms of rate of change for each of the four principal regions. Valid volumetric estimates could not be obtained for five subjects; they were excluded from all GM volume analyses. All analyses were performed using R Statistical Software (version 3.1.1; R Foundation for Statistical Computing).

3. Results

The 5 A+ groups are described in Table 1. The Supplemental Table 1 includes the A− groups as well. As the CN and MCI groups were not matched on any demographic or clinical features, groups varied to a greater or lesser degree in age, education, sex as well as level of cognitive function and range of imaging abnormalities. Carriage of the APOE e4 allele was highest in AD DEM and MCI A+. The MCI A+ group was overwhelming of the amnestic type (57/62, 92%). Forty-seven percent of MCI but only 20% of CN participants were A+N+ (Supplemental Table 2). As expected the AD DEM group had the lowest cognitive test scores. All 20 of the AD DEM participants were A+ by definition and all but one was N+.

Table 1. Participant characteristics.

Statistics shown are of the form mean (SD) unless otherwise specified.

	CN A+/N− (n = 79)	MCI A+/N− (n = 17)	CN A+/N+ (n = 65)	MCI A+/N+ (n = 45)	AD DEM A+ (n = 20)
Male, no. (%)a	39 (49%)	11 (65%)	51 (78%)	29 (64%)	14 (70%)
Age, yrsa	80 (5)	79 (5)	80 (5)	81 (5)	78 (6)
Age, Min, Max	72 to 94	71 to 90	70 to 93	72 to 90	71 to 92
Education, yrsb	14 (3)	14 (4)	15 (3)	14 (3)	16 (2)
APOE e4 genotype N, (%)c	30 (38%)	10 (59%)	27 (42%)	23 (51%)	14 (70%)
PiB SUVr	1.74 (0.33)	1.95 (0.33)	1.82 (0.37)	2.11 (0.48)	2.44 (0.42)
FDG SUVR AD sign regions	1.46 (0.10)	1.44 (0.10)	1.27 (0.10)	1.25 (0.14)	1.12 (0.23)
Adj. Hippocampal Volume (HVa)	−1.13 (0.54)	−1.75 (0.55)	−2.12 (0.84)	−2.81 (0.82)	−3.14 (0.90)
Total Intracranial Volume, Liters	1.5 (0.14)	1.5 (0.14)	1.6 (0.14)	1.5 (0.13)	1.5 (0.12)
MMSE†	28 (1.4)	26 (2.0)	28 (1.2)	25 (2.7)	21 (5.2)
Global Cognitive Z-score*	0.75 (0.81)	−0.29 (0.42)	0.35 (0.76)	−0.74 (0.84)	−1.20 (0.82)
No. of visits /imaging	2 (0.5)	2 (0.5)	2 (0.5)	2 (0.7)	3 (0.9)
Range	2 to 4	2 to 3	2 to 4	2 to 6	2 to 5
No. (%) having
2 imaging visits	62 (78%)	10 (59%)	52 (80%)	33 (73%)	13 (65%)
3+ imaging visits	17 (22%)	7 (41%)	13 (20%)	12 (27%)	7 (35%)
Time in study†, yrs	2.6 (1.19)	2.2 (0.99)	2.6 (1.19)	2.1 (1.10)	1.9 (1.11)
Range	1 to 5	1 to 4	1 to 6	1 to 5	1 to 4

^*Z-scores 70+ CN & MCI, MCSA

^†Converted from STMS scores

^aNo differences in sex or age across CN (A+N− and A+N+ combined), MCI (A+N− and A+N+ combined) and AD DEM groups.

^bCompared to AD DEM, CN and MCI groups had lower education (p=0.02), but there were no differences between CN and MCI groups.

^cAD DEM group had higher proportion of APOE e4 carriage than CN group (p=0.01) but not MCI group (p=0.07). There were no differences in APOE e4 carriage between CN and MCI.

^†Time in study refers to the period between the first and last imaging studies.

Across all A+ participants, baseline values of FDG SUVR and GM volume in the 4 principal ROIs were moderately correlated (rho’s 0.32, 0.22, 0.23 and 0.31; p<0.001), and rates of change were similarly or slightly less correlated (rho’s 0.32, 0.21, 0.15 and 0.18; p≤0.002 for MT, LT, MP and LP, respectively).

Figures 1 to 3 and Tables 2 and 3 illustrate the principal findings. With a much larger sample than in our previous publication on CN, we replicated and extended our prior observations (Knopman et al., 2013) on the extent to which the CN A+N+ group’s rate of change in both FDG SUVR and GM volume in the medial temporal ROI exceeded the other CN groups. Our recent report on observations within an MCI group (Knopman et al., 2015) are also reflected in the current analyses. However, the current report focuses on between-clinical group differences.

Figure 1. Baseline FDG SUVR and GM Volume.

Group mean (SE) estimates at baseline for FDG SUVR (A) and GM volume (B) estimates with 95% confidence interval (CI) estimates (thin gray lines) and 83.5% CI estimates (thicker black lines) by biomarker group for primary ROIs: medial temporal, medial parietal, lateral temporal, and lateral parietal. The 83.5% CI allows for visual comparisons between groups where any amount of overlap indicates a lack of significance at the 0.05 level. Estimates are from a linear mixed model for a male participant age 80 years.

Figure 3. Trajectory over time, by group.

Regression plots for the 4 primary ROIs showing trajectories of FDG SUVR (top panels) and grey matter (GM) (bottom panels) over time for biomarker defined groups for a hypothetical male 80 years of age. The x-axis is time from baseline. The time scale was limited to 2 years for illustrative purposes. See Figures 1 and 2 for confidence intervals of baseline values and slopes. Color coding: CN (orange), MCI (blue), AD Dementia (green); A+N+ (solid lines); A+N− (dashed lines).

Table 2. Effect size and 95% confidence intervals for paired group differences in annual rate of change^*.

	FDG SUVR MCI A+N+ minus CN A+N+	GM Volume MCI A+N+ minus CN A+N+	FDG SUVR AD DEM minus MCI A+N+	GM Volume AD DEM minus MCI A+N+
Medial Temporal	−1.0 (−1.5, −0.5)	−0.8 (−1.1, −0.4)	−0.8 (−1.5, −0.01)	−0.6 (−1.2, −0.1)
Lateral Temporal	−1.3 (−2.0, −0.6)	−0.8 (−1.2, −0.4)	−1.3 (−2.3, −0.3)	−0.7 (−1.3, −0.03)
Medial Parietal	−1.6 (−2.5, −0.8)	−0.4 (−0.9, 0.1)	−0.8 (−2.1, 0.5)	−0.4 (−1.2, 0.5)
Lateral Parietal	−1.6 (−2.5, −0.7)	−1.0 (−1.7, −0.2)	−1.4 (−2.7, −0.1)	−0.6 (−1.8, 0.5)

^*All bolded contrasts were significant at an uncorrected p<0.05. Contrasts in yellow boxes were significant at 0.001< p≤ 0.01, and contrasts in orange boxes were significant at p<0.001.

Table 3. Percent of 5000 bootstrap replicates with the ordering AD DEM < MCI A+N+ < CN A+N+ at baseline and AD DEM > MCI A+N+ > CN A+N+ declines over time.

	FDG SUVR	GM volume
Baseline
Medial Temporal	98%	99%
Lateral Temporal	26%	96%
Medial Parietal	29%	70%
Lateral Parietal	20%	83%
Rate of change
Medial Temporal	97%	99%
Lateral Temporal	99%	98%
Medial Parietal	89%	74%
Lateral Parietal	99%	85%

AD DEM versus MCI A+N+

The AD DEM group had significantly lower baseline FDG SUVR in all 4 principal ROIs (MT, p<0.001; LT, p<0.001; MP, p=0.003; LP, p<0.001 LP) and lower GM volume in MT and LT ROIs (both p=0.01) compared to the MCI A+N+ group (Figure 1). The AD DEM group showed greater metabolic annual percent declines compared to the MCI A+N+ group in 3 of 4 principal ROIs (MT, p=0.048; LT, p=0.01; MP, p=0.22, LP, p=0.03), and greater volumetric losses in MT (p=0.04), and LT ROIs (p=0.02) (Table 2 and Figure 2). The relationship between baseline values and rate of change are depicted in Figure 3.

Figure 2. Annual Percent change in FDG SUVR and GM Volume.

Group mean (SE) estimated annual percentage change for FDG SUVR (A) and GM volume (B) with 95% confidence interval (CI) (thin gray lines) and 83.5% CI (thicker black lines) by biomarker group for primary ROIs: medial temporal, medial parietal, lateral temporal and lateral parietal. The 83.5% CI allows for visual comparisons between groups where any amount of overlap indicates a lack of significance at the 0.05 level. Estimates are from a linear mixed model for a male participant age 80 years.

The AD DEM group’s FDG SUVR declines in non-principal ROIs did not differ from the MCI A+N+ group. There were also no differences in GM volumes annual percent declines in non-principal ROIs in the AD DEM group compared to the MCI A+N+ group except in insula (p=0.02). The values for non-principal ROIs are given in the Supplemental Table 2.

MCI A+N+ versus CN A+N+

At baseline, the MCI A+N+ group had lower MT FDG SUVR (p=0.045) and lower MT GM volume (p=0.006) compared to the CN A+N+ group but no differences in other principal ROIs (Figure 1). On the other hand, the annual percent declines in FDG SUVR in all 4 principal ROIs (p<0.001) and GM volume in 3 of the principal ROIs (MT, p<0.001; LT, p< 0.001; and LP, p= 0.01) were greater among the MCI A+N+ group than in the CN A+N+ group (Table 2 and Figure 2). The relationship between baseline values and rate of change are depicted in Figure 3.

The MCI A+N+ group showed significantly greater FDG SUVR declines in almost all of the non-principal ROIs compared to the CN A+N+ group. The MCI A+N+ group showed significantly greater GM volumetric declines compared to CN A+N+ in insula (p=0.02), and precentral gyrus (p=0.03). The values for non-principal ROIs are given in the Supplemental Table 2.

Given the differences between AD DEM and MCI A+N+, and between MCI A+N+ and CN A+N+, differences between AD DEM and CN A+N+ were large. The values are given in the Supplemental Table 2.

Summary of CN and MCI A+N+ group and AD DEM differences

The findings are summarized in Table 3. For the 4 principal ROIs we used a bootstrap method with 5000 replicates to evaluate the likelihood that on average AD DEM < MCI A+N+ < CN A+N+ for baseline values of FDG SUVR and GM volume, and that on average AD DEM > MCI A+N+ > CN A+N+ for annual rate of worsening in FDG SUVR and GM volume. Figure 3 and Table 2 summarize the group differences. We found strong evidence for ordering by cognitive severity for baseline FDG SUVR in the MT ROI (98%) and for baseline GM volume in the MT (99%) and LT ROI (96%). The data also indicated moderate support for this ordering for GM volume in the MP ROI (70%) and LP ROI (83%). On rate of decline, there was strong evidence for ordering by cognitive severity across all regions for FDG SUVR (ranging from 89% to 99%) and in the temporal regions for GM volume (99% for MT and 98% for LT). The data also suggest a relatively high likelihood of ordering for GM volumes in the parietal regions (74% for MP and 85% for LP).

MCI A+N− versus CN A+N−

The CN and MCI A+N− groups also differed. They had similar mean ages but differences in distributions of APOE e4 genotype and sex. The MCI group had more elevated PIB SUVR and lower HVa. Whereas there were no baseline differences in FDG SUVR and GM volumes between CN and MCI A+N− groups (Figure 1), the MCI A+N− showed larger declines over follow-up than did the CN A+N− for both FDG SUVR (MT, p= 0.004; LT, p=0.008; LP, p=0.01) and GM volume (MT, p=0.001; LT, p<0.001; LP, p=0.04). (Figures 2 and 3).

In the non-principal ROIs, the MCI A+N− did not show greater metabolic declines compared to CN A+N−. Volumetrically, the MCI A+N− group showed greater GM volume loss compared to CN A+N− in prefrontal (p=0.02), orbitofrontal (p=0.02), and insula (p<0.001). The values for non-principal ROIs are given in the Supplemental Table 2.

Groups without elevated PIB SUVR

Neither the MCI A−N− nor MCI A−N+ groups showed greater declines in FDG SUVR or GM volume compared to their CN counterparts (see Figure 2).

Differences in PIB SUVR across groups

We compared baseline global PIB SUVR across the A+ groups (Figure 4). As expected, at baseline the AD DEM group had the highest PIB SUVR, the MCI A+ groups the next highest and the CN A+ groups the lowest. However, the rate of increase in PIB SUVR was similar among the A+ CN, MCI and AD DEM groups.

Figure 4. Global PIB SUVR at Baseline and Annual Percent Change.

Group mean (SE) estimates for global PIB PET estimates at baseline (left panel) and estimated annual percentage change (right panel) with 95% confidence interval (CI) estimates (thin gray lines) and 83.5% CI estimates (thicker black lines) by biomarker and clinical diagnostic group. The 83.5% CI allows for visual comparisons between groups where any amount of overlap indicates a lack of significance at the 0.05 level. Estimates are from a linear mixed model for a male participant age 80 years.

The role of APOE e4 genotype

We explored the relationship between carriage of the e4 allele of APOE on the progression of structural and metabolic changes across the cognitive spectrum. Because very few APOE e4 carriers were found among those who did not have elevated amyloid (Supplemental Table 3), we restricted our analyses to the groups with elevated amyloid. We found that carriers of the APOE e4 allele were generally younger, yet had higher PIB SUVR and lower HVa at baseline (Supplemental Table 4 and Supplemental Figure 4). In the regression analyses, there were no baseline group differences based on APOE e4 genotype in any ROI. However, in longitudinal analyses carriage of APOE e4 was associated with more degeneration in the MT ROI in MCI A+N+ (for FDG SUVR (p<0.001) and GM volume (p=0.01)) and in AD DEM (for FDG SUVR (p=0.003) only). In longitudinal analyses, there were no differences in comparing APOE e4 carriers to non-carriers in any other ROI in those two subgroups, and none of the other clinical subgroups showed any APOE e4+ effects in any ROIs.

4. Discussion

We found an internally consistent regionally-specific pattern of rate of decline of FDG SUVR and GM volume in A+N+ groups across the spectrum from CN to MCI to dementia. The AD-DEM group, which was A+ by design and definition, showed both lower baseline values and higher rates of neurodegenerative processes in the 4 principal ROIs plus insula compared to MCI A+N+ and to CN A+N+ (as well as to all other MCI and CN subgroups). The MCI A+N+ group also exhibited a neurodegenerative process that spread into lateral temporal and both parietal ROIs (but mainly lateral parietal). The MCI A+N+ group had less structural or metabolic impairment at baseline and marginally less worsening over time compared to the AD DEM group, but more metabolic and structural deterioration than CN A+N+. Among persons with elevated brain β-amyloid, the transition from clinically normal status to MCI to dementia mapped logically onto the progression of neurodegenerative imaging biomarkers. While clinical – neuropathological studies (Braak and Braak, 1991; Duyckaerts et al., 1997; Delacourte et al., 1999; Braak et al., 2006) have told us that disease expands regionally in temporal isocortex, parietal isocortex and insula as cognition declines, a unique contribution of antemortem serial imaging was the demonstration of an increasing rate of neurodegenerative change as cognitive decline worsened from CN to MCI to AD DEM. As we have previously reported in both CN (Knopman et al., 2013) and MCI (Knopman et al., 2015), CN A+N+ or MCI A+N+ groups showed greater declines compared to their other biomarker-defined peers. Our findings go beyond prior work (Sabuncu et al., 2011; Ewers et al., 2012; Dore et al., 2013; Araque Caballero et al., 2015; Nosheny et al., 2015), in showing that the combination of elevated β-amyloidosis and substantial neurodegeneration were jointly required for progression to occur across the AD spectrum.

While global brain β-amyloid levels rose across the spectrum from CN to MCI to AD DEM, there were no group differences in rate of change of β-amyloid accumulation. In contrast to the regional expansion of neurodegeneration from CN to MCI to dementia, the lack of difference in rate of change in brain amyloid levels once again (Jack et al., 2009) makes the point that β-amyloidosis is not proximate to cognition in the sequence of pathophysiological events in the AD pathway. To be sure, we have observed differences in the rate of amyloid accumulation (Jack et al., 2013) as a function of baseline PIB SUVR levels, but as shown here, it was neurodegeneration, and not β-amyloid, that changed in synchrony with declining clinical status.

APOE e4 carriage had a large effect on whether an individual had elevated β-amyloid, consistent with much prior work, eg (Morris et al., 2010; Vemuri et al., 2010). However, except for an association with greater metabolic and volumetric losses in the medial temporal ROI in the MCI A+N+ group, carriage of the APOE e4 allele did not impact MCI A+N− participants and had no impact on any CN group including the A+N+. To the extent that APOE e4 carriage is associated with the appearance of β-amyloid elevations roughly 7 years earlier than non-carriers (Jack et al., 2015), it seems plausible that the excess neurodegenerative changes in MT in MCI A+N+ APOE e4 carriers reflected their longer exposure to elevated β-amyloidosis and earlier neurodegenerative changes. That a role for APOE e4 carriage was not evident in the CN A+N+ group is somewhat enigmatic, but is consistent with prior studies showing a negligible effect of e4 carriage on hippocampal volume in normals, but a consistent effect in AD dementia (Jack et al., 1999; Lemaitre et al., 2005; Schuff et al., 2009).

Stratifying our groups by baseline neurodegeneration provided some insights into the A+N− groups. Those who were MCI A+N− entered the period of observation with less hypometabolism and volume loss (see Supplemental Figure 3 for scatterplots) than the MCI A+N+ group but showed generally similar rates of neurodegeneration. We are uncertain whether the MCI A+N− group represented persons in whom our neurodegenerative biomarkers failed to capture a salient amyloid (AD)-related process, or whether they were simply at an earlier stage of progression in the AD pathway. In contrast, the CN A+N− group showed less progression of neurodegeneration than the CN A+N+ group, again suggesting that the synergy between amyloidosis and neurodegeneration is critical.

Our method of defining neurodegeneration reflected a strategy that we have previously followed (Jack et al., 2012; Knopman et al., 2013; Jack et al., 2014; Knopman et al., 2015) that allowed abnormalities in either HVa or FDG SUVR as the definition. We have found that the modality or modalities used to define neurodegeneration had little impact on age-related prevalence of abnormality (Jack et al., 2015). We have previously shown that our approach to defining neurodegeneration with either HVa or FDG SUVR successfully predicts clinical progression from CN to MCI or dementia (Knopman et al., 2012)#2325] or from MCI to dementia (Petersen et al., 2013)2519]. Those observations were replicated in the current analyses, in which the CN A+N+ group experienced a 3-fold greater number of persons who progressed compared to the CN A+N− group, and the MCI A+N+ group experienced nearly a doubling of the number who progressed to dementia.

The joint presence of elevated β-amyloid and evidence for neurodegeneration as shown here was mirrored in clinical progression both from CN to MCI (Knopman et al., 2012; Roe et al., 2013; van Harten et al., 2013; Vos et al., 2013; Wirth et al., 2013; Mormino et al., 2014; Toledo et al., 2014) or from MCI to dementia (van Rossum et al., 2012; Petersen et al., 2013; Prestia et al., 2013; Caroli et al., 2015; Vos et al., 2015). It is hard to escape the conjecture that elevated β-amyloid levels plays a necessary role in promoting or accelerating neurodegeneration – once it reaches a certain threshold – during the CN and MCI stages of the AD process. The conjecture does not imply that β-amyloid initiates the process in sporadic AD, but rather that β-amyloid plays an essential role in promoting neurodegeneration at some point between the time that a person is cognitively normal into the symptomatic phase of AD. Recent work with mouse models of β-amyloidosis and tauopathy also support the role of β-amyloidosis as an accelerant of tauopathy and neurodegeneration (Pooler et al., 2015). In sporadic AD, the distinction between accelerant and initiator may seem subtle, but the implications are profound, and could account in part for disappointing therapeutic adventures with anti-amyloid agents to date

These analyses have some limitations because the groups were not matched demographically. The AD DEM group was slightly younger and slightly better educated than the CN and MCI A+N+ groups. Sex differences might reflect differential survival or sex-specific risk factors. Matching has some appeal but it induces other group differences that are artificial (eg, CN and MCI groups generally differ in age). Ideally the analyses here should also be performed with incident MCI and AD DEM cases to demonstrate more clearly the evolution of AD neurodegeneration on an individual level, but the rate of progression from CN to MCI is slow enough that we would need a decade, if not more, of longitudinal experience for sufficient numbers of originally CN persons to develop MCI. And over that length of time, attrition, comorbidity and mortality from other causes would greatly compromise analyses. We chose to perform these analyses by defining categorical clinical groups rather than using cognitive function as a continuous variable. There are several reasons for our choice. First, such an analysis can be mapped to clinical situations. Second, to the extent that the different sample sizes of CN, MCI and DEM groups do not match the distribution in the population, the proportions of participants at different levels of cognitive dysfunction could skew clinical-imaging relationships. Finally we used both FDG SUVR and GM volume to assess change. There were similar regional outcomes for both FDG SUVR and GM volume changes, but the two biomarkers of neurodegeneration were not measuring the same thing as shown by their modest correlations. In CN individuals such differences have been previously reported (La Joie et al., 2012). It was not our goal to determine if FDG PET or volumetric MR was superior to the other; rather we found that they both provided unique information.

Highlights.

• MCI A+N+ group had more metabolic and structural deterioration than CN A+N+.

• AD-DEM had more metabolic and structural deterioration than MCI A+N+.

• Specificity of amyloid related neurodegeneration for temporo-parietal isocortex

• Increasing rate of neurodegenerative change from CN A+N+ to MCI A+N+ to AD DEM.

Abbreviations

A−N−

amyloid not elevated and neurodegeneration absent

A+N−

amyloid elevated and neurodegeneration absent

A+N+

amyloid elevated and neurodegeneration present

A−N+

amyloid not elevated and neurodegeneration present.

APOE

apolipoprotein E

AD

Alzheimer’s disease

ADRC

Alzheimer’s Disease Research Center

CN

Clinically Normal

FOV

field of view

FDG PET

¹⁸fluorodeoxyglucose positron emission tomography

GM

grey matter

HVa

adjusted hippocampal volume

IQR

interquartile range

LP

lateral parietal

LT

lateral temporal

MCSA

Mayo Clinic Study of Aging

MPRAGE

magnetization-prepared radiofrequency pulses and rapid gradient echo

MP

medial parietal

MT

medial temporal

MR

magnetic resonance imaging

MCI

mild cognitive impairment

PiB PET

Pittsburgh compound B positron emission tomography

ROI

region of interest

SUVR

standardized uptake value ratio

TIV

total intracranial volume.