Unique regional patterns of amyloid burden predict progression to prodromal and clinical stages of Alzheimer's disease.

Although beta-amyloid (Aβ) positivity has shown to be associated with higher risk of progression to Alzheimer's disease (AD) in mild cognitive impairment (MCI), information on the time to conversion to manifest dementia cannot be readily deduced from this binary classification. Here, we assessed if regional patterns of Aβ deposition measured with 18F-florbetapir may serve as biomarker for progression risk in Aβ-positive cognitively normal (CN) and MCI patients, including clinical follow-up data and cerebrospinal fluid (CSF) biomarkers. Voxel-wise group comparisons between age and sex-matched Aβ-positive groups (i.e., CN-stables [n = 38] vs. CN-to-MCI/AD progressors [n = 38], MCI-stables [n = 104] versus MCI-to-AD progressors [n = 104]) revealed higher Aβ burden in precuneus, subcortical, and parietal regions in CN-to-MCI/AD progressors and cingulate, temporal, and frontal regions in MCI-to-AD progressors. Importantly, these regional patterns predicted progression to advanced stages on the AD spectrum in the short and the long-term beyond global Aβ burden and CSF biomarkers. These results suggest that distinct regional patterns of Aβ burden are a valuable biomarker for risk of disease progression in CN and MCI.

Background

The earliest pathological hallmark of Alzheimer's disease (AD) is the accumulation of amyloid-beta (Aβ). Even though Aβ accumulation may be evident in individuals more than 20 years before progression to clinical dementia (Bateman et al., 2012; Rowe et al., 2010), no overt symptoms are present at this point. Since clinical trials targeting Aβ accumulation at clinical stages of the disease have mostly failed so far, the need for new treatment approaches and diagnostic tools to identify individuals at an early, preclinical stage of the disease trajectory is high.

Positron emission tomography (PET) Aβ-tracers selectively bind to Aβ plaques, allowing the visualization and quantification of Aβ burden in the brain non-invasively. The clinical standard currently is still a binary distinction of individual patients into Aβ-positive or negative based on the global Aβ burden. Although a binary distinction is imperative for the differential diagnosis, additional information could be extracted from Aβ PET images that may potentially inform disease trajectories. Already post-mortem pathologic studies showed that Aβ deposition follows a specific pattern of spread (Thal et al., 2002). However, the predictive power of these regional associations concerning disease progression from prodromal to clinical stages remains to be elucidated. Regional Aβ burden in the posterior cingulate, precuneus, and lateral parietal, as well as the banks of the superior temporal sulcus has been associated with longitudinal cognitive decline in amyloid-negative cognitively normal elderly adults (Guo et al., 2020; Farrell et al., 2018), emphasizing the contribution of regional Aβ to cognitive aging. Nevertheless, these studies do not allow inferences about the role of regional Aβ burden of disease progression in preclinical or prodromal populations. Global Aβ burden, on the other hand, has been shown to predict progression to AD in some cases (Ciarmiello et al., 2019; Jun et al., 2019), however, the time of progression to manifest dementia cannot be deduced from this binary classification or global measurement, so far. Nonetheless, this information would be essential from a clinical point of view, but also for the arrangement of patient's and care-taker's health-related matters.

We hypothesize that that specific regional patterns of Aβ deposition may constitute a predictive value for both Aβ-positive cognitively normal (CN) individuals and individuals with mild cognitive impairment (MCI) who subsequently advance on the continuum of clinical Alzheimer's disease (AD). Additionally, we propose that the regional susceptibility of Aβ to predict progression is different depending on the baseline diagnosis (i.e., CN or MCI). Lastly, we hypothesize that this potential identified pattern of regional Aβ may constitute a more predictive biomarker than global Aβ or CSF measures, and that they may be unique for disease progression stage.

Methods

Participants

Data used for this study were derived from the Alzheimer's Disease Neuroimaging Initiative (ADNI) (http://adni.loni.usc.edu/). Key screening criteria for inclusion were: (1) Aβ-positivity from baseline until follow-up (standardized uptake value ratio (SUVR) > 1.1, see Landau & Jagust (2015) for assessment of this measure), (2) availability of an 18F-Florpetapir PET scan at baseline, (3) a diagnosis of either cognitively normal or mild cognitive impairment at baseline, and (4) a follow-up diagnosis (at least 6 months from baseline) of either CN, MCI, or AD. The clinical diagnosis of Alzheimer's disease was based on the recommended diagnostic National Institute on Aging and Alzheimer's Association (NIA-AA) guidelines from 2011 (Albert et al., 2011; Jack Jr et al., 2011; McKhann et al., 2011; Sperling et al., 2011). This resulted in 284 participants, 168 males and 116 females (mean age 74.7 ± 6.5 years), out of which 4 groups were defined: (1) individuals who were cognitively normal at baseline and remained stable over a period of at least 6 months (CN-CN), (2) individuals who were cognitively normal and progressed to MCI or AD (CN-MCI/AD), (3) individuals with a baseline diagnosis of MCI, who remained stable over a time period of at least 6 months (MCI-MCI), and (4) individuals with a baseline diagnosis of MCI who progressed to AD (MCI-AD). The CN-CN and CN-MCI/AD groups and the MCI-MCI and MCI-AD groups were matched on age and sex, respectively. Furthermore, it was ensured that each progressed participant was matched with a participant who remained stable for at least the same amount of time their matched progressor stayed stable.

Positron emission tomography (PET)

All participants underwent a PET-scan with the 18F-florbetapir PET tracer. Participants received an average dosage of 370 MBq (10 mCi) of 18F-florbetapir with 20 minutes (4 × 5 minutes frames) acquisition at 50–70 minutes post-injection (for more information see http://adni.loni.usc.edu/methods/pet-analysis-method/pet-analysis/).

CSF measures

Most participants underwent a lumbar puncture to measure baseline amyloid-beta 1-42 peptide (CSF Aβ), total tau (CSF t-tau), and tau phosphorylated at the threonine 181 (CSF p-tau). Out of 286 participants, 228 had a CSF Aβ measure and 233 had a CSF t-tau as well as aCSF p-tau measure. Lumbar puncture was performed with a 20-24-gauge spinal needle as described in the ADNI procedures manual (http://www.adni-info.org/).

Imaging data pre-processing

PET data was pre-processed with Statistical Parametric Modeling 12 (SPM12, Wellcome Trust Center, www.fil.ion.ucl.ac.uk/spm) implemented in Matlab 2019b (Mathworks Inc., Sherborn, MA). All PET images were aligned to the anterior-posterior commissure and spatially normalized to the tissue probability map implemented in SPM12. Then, all images were smoothed with an 8 mm FWHM Gaussian filter, as this has been found to be an appropriate value to use (Tsutsui et al., 2018) and has been employed frequently in previous studies (Lin et al., 2016; Saint-Aubert et al., 2013; Teipel et al., 2015). Lastly, SUVRs for the images were computed by using the whole cerebellum as reference region.

A global Aβ score, previously published (details for pre-processing see Landau & Jagust, 2015), was used, which was derived by taking the fully processed 18F-florbetapir scans and co-registering and normalizing them to its corresponding MRI-image that was closest in time to that scan. All MRI images were skull-stripped, segmented, and delineated into cortical and subcortical regions. 18F-florbetapir SUVr means were extracted from grey matter in each subregion within 4 cortical regions (frontal, anterior/posterior cingulate, lateral parietal, and lateral temporal regions) using the whole cerebellum as reference region. Finally, a composite global Aβ SUVr score was computed based on the 4 cortical regions.

Statistical analysis

Non-parametric (i.e., Mann-Whitney-U-tests; Pearson-Chi-Square-tests) tests were performed to compare the groups on age, sex, years of education, Apolipoprotein E4 (APOE4) carriership, and baseline values of global Aβ burden, CSF Aβ, CSF t-tau, and CSF p-tau. To assess regional differences in Aβ burden between the stable and progressor groups, 2-sample t-tests were performed in SPM12 comparing the CN-CN group against the CN-MCI/AD group and the MCI-MCI group against the MCI-AD group, respectively. APOE4 carriership was entered as covariate. To reduce the number of voxel-wise comparisons, a brain mask with all regions of the AAL atlas, except for the cerebellum, was included in the analyses. Subsequently, significant regions were extracted and saved in a binarized format in MNI space. SUVRs of all participants were extracted from these regions and the overall mean across regions was computed. Binary logistic regression analyses were performed to quantify the effect of regional Aβ burden over all other variables of interest (age, sex, education, APOE4 carriership, global Aβ burden, and all 3 CSF measures) in the prediction of progression. For this, all predictor variables were entered simultaneously into the regression model together with regional Aβ burden.

To assess whether regional Aβ burden extracted from one of the respective groups (CN or MCI) is predictive of progression in the other group, additional logistic regression analyses were performed, including the mean regional SUVRs from the other respective groups and all other predictors from the previous analyses. Again, all predictor variables were entered in the regression model simultaneously. For the CN-CN versus CN-MCI/AD groups, 27 progressors and 28 stables were included in the analyses due to missing data points for some variables, while for the MCI-MCI versus MCI-AD groups, 74 progressors and 99 stables were included. Finally, multinomial logistic regression analyses were performed comparing fast against slow progressors in both the CN-MCI/AD group and the MCI-AD group, respectively (Appendix A1, Tables A1–3). All analyses were performed using SPSS 25.0.

Sensitivity analysis

To investigate whether the matching procedure based on follow-up timepoints (i.e., MCI-MCI with 2 years follow-up was matched to MCI-AD with 2 years follow-up) might have influenced our analyses, a sensitivity analysis was performed. For this, participants in the stable groups (CN-CN and MCI-MCI) were selected if they were stable for at least 4 years, while participants in the progressor groups (CN-MCI/AD and MCI-MCI) were selected if they progressed within 3 years of the 18F-Florpetapir PET baseline scan. The follow-up time periods were chosen to maximize time and group size. The approach mimicked the analysis described in 2.5 (Appendix B1, Table B1, Figure B1).

Table B1

Mean and standard deviations per group for the sensitivity analysis

Variable	CN-CN(N = 30)	CN-MCI/AD(N = 21)	MCI-MCI(N = 47)	MCI-AD(N = 47)
Age (years)	76.71 ± 4.85	77.94 ± 4.82	71.79 ± 5.64	72.00 ± 5.75
Sex (m/f)	17/13	11/10	27/20	27/20
APOE4 (±)	7/23	9/12	28/19	35/12
Education (years)	16.60 ±2.25	16.29 ± 2.89	15.87 ± 2.77	15.83 ± 2.75
Months stable	70.4 ± 17.18	20.00 ± 10.62	64.60 ± 13.60	24.26 ± 8.10
CSF Aβ	1001.95 ± 408.46 (24)	879.68 ± 382.40 (15)	805.31 ± 300.62 (38)	675.25 ± 178.64 (45)
CSF t-tau	310.09 ± 138.31 (24)	306.54 ± 102.79 (15)	275.09 ± 94.58 (42)	416.45 ± 138.36 (45)
CSF p-tau	29.88 ± 14.59 (24)	30.02 ± 11.86 (15)	26.23 ± 10.39 (42)	43.11 ± 15.56 (45)
Global Aβ	1.25 ± 0.20	1.55 ± 0.28	1.23 ± 0.16	1.49 ± 0.13

Values are given as mean ± standard deviation. Values in brackets are number of included datapoints if data was missing.

Key: APOE4, Apolipoprotein E4; CN-CN, cognitively normal stable; CN-MCI/AD, cognitively normal progressed to MCI or AD; MCI-MCI, mild cognitive impairment stable; MCI-AD, mild cognitive impairment progressed to AD.

Figure B1

Regional differences in Aβ burden for the sensitivity analysis. (A) Regional differences in Aβ burden for CN-MCI/AD (N = 30) > CN-CN (N = 21), and (B) MCI-AD (N = 47) > MCI-MCI (N = 47). Results are color-coded according to the t-value statistics and illustrated on an inflated representation of the brain using CAT12 toolbox.

Results

Group characteristics

Mean and standard deviations are reported (Table 1). At baseline, the CN-CN and CN-MCI/AD groups were of similar age, years of education, ratio of females to males, CSF Aβ, CSF t-tau, and CSF p-tau. The MCI-MCI and MCI-AD groups were of similar age, years of education, and ratio of females to males. The CN-MCI/AD group showed a higher global Aβ burden at baseline than the CN-CN group (p = 0.009). The CN-CN group showed an increased number of months of clinical stability (M 60.32, MCN-MCI/AD = 28.84, p < 0.001) and included more APOE4-negative individuals (p = 0.002) than the CN-MCI/AD group. Similarly, the MCI-MCI group included more APOE-negative individuals than the MCI-AD group (p = 0.026). The MCI-AD group showed a reduced number of months of clinical stability (M 43.15, MMCI-AD = 27.75, p < 0.001), a larger global Aβ burden (p < 0.001), CSF Aβ (p < 0.001), CSF t-tau (p < 0.001), and CSF p-tau (p < 0.001) than the MCI-MCI group (Table 2).

Table 1

Mean and standard deviations per group and variable

Variable	CN-CN(N = 38)	CN-MCI/AD(N = 38)	MCI-MCI(N = 104)	MCI-AD(N = 104)
Age (y)	78.05 ± 5.31	78.33 ± 5.38	73.57 ± 6.43	73.30 ± 6.53
Sex (m/f)	20/18	20/18	63/41	63/41
APOE4 (±)	8/30	21/17	63/41	78/26
Education (y)	16.71 ± 2.23	16.29 ± 2.89	16.24 ± 2.92	16.32 ± 2.63
Months stable	60.32 ± 25.03	28.84 ± 24.15	43.15 ± 23.43	27.75 ± 19.98
CSF Aβ	1018.78 ± 419.75 (28)	830.49 ± 354.30 (27)	800.86 ± 269.04 (74)	682.40 ± 184.59 (99)
CSF t-tau	306.33 ± 13.05 (28)	303.89 ± 91.11 (27)	287.59 ± 116.33 (79)	374.62 ± 137.40 (99)
CSF p-tau	29.54 ± 13.67 (28)	29.85 ± 10.45 (27)	27.42 ± 11.75 (79)	37.98 ± 15.92 (99)
Global Aβ	1.25 ± 0.19	1.34 ± 0.18	1.33 ± 0.18	1.44 ± 0.16

Values are given as mean ± standard deviation. Values in brackets are number of included data points if data was missing.

Key: APOE4, Apolipoprotein E4; CN-CN, cognitively normal stable; CN-MCI/AD, cognitively normal progressed to MCI or AD; MCI-MCI, mild cognitive impairment stable; MCI-AD, mild cognitive impairment progressed to AD.

Table 2

Group Differences between CN-CN versus CN-MCI/AD and MCI-MCI versus MCI-AD

	Mean rank	N	Mean rank	N	U / X2	Z	n²
	CN-CN		CN-MCI/AD
Months being stable	47.21	38	29.79	38	391.000b	-3.495	-0.83
Global Aβ	31.87	38	45.13	38	470.000a	-2.618	-0.90
APOE4 (±)	-	8/30	-	21/17	9.432a	-	-
	MCI-MCI		MCI-AD
Months being stable	125.06	104	83.94	104	3269.500b	-4.992	-0.88
Global Aβ	83.56	104	125.44	104	3230.000b	-5.018	-0.87
CSF Aβ	100.01	74	77.27	99	2700.000a	-2.955	-0.85
CSF t-tau	69.47	79	105.48	99	2328.500b	-4.632	-0.85
CSF p-tau	67.92	79	106.72	99	2205.500b	-4.992	-0.95
APOE4 (±)	-	63/41	-	78/26	4.954a	-	-

Mann-Whitney-U-Tests and Pearson-Chi Square-Tests for CN-CN versus CN-MCI/AD, and MCI-MCI versus MCI-AD groups. Data showing non-significant associations was omitted from the table.

Key: APOE4, Apolipoprotein E4; CN-CN, cognitively normal stable; CN-MCI/AD, cognitively normal progressed to MCI or AD; CSF, cerebrospinal fluid; MCI-MCI, mild cognitive impairment stable; MCI-AD, mild cognitive impairment progressed to AD; P-tau, Phospho-tau; T-tau, total-tau.

p < 0.05.

p < 0.001.

Regional differences in Aβ burden

The voxel-wise whole-brain analysis yielded higher regional Aβ burden in the left and right precuneus, right lingual and angular gyrus, putamen, caudate, pallidum, middle temporal gyrus, and superior temporal gyrus for the CN-MCI/AD group compared to the CN-CN group (p < 0.001, uncorrected, Fig. 1A). None of the clusters in this comparison survived correction for multiple comparisons. Regional Aβ burden was also higher in the left and right anterior cingulate gyrus, medial frontal cortex, precuneus, right transverse temporal gyrus, left middle and superior temporal gyrus, posterior and anterior insula, central operculum, and medial segment of the superior frontal gyrus for the MCI-AD group compared to the MCI-MCI group (FWE-corrected, Fig. 1B). Upon visual assessment, hemispheric asymmetry was observed, shifting from right (higher Aβ burden in the right hemisphere in CN) to left (higher Aβ burden in the left hemisphere in MCI) depending on the disease stage (i.e., CN or MCI).

Fig. 1

Regional differences in Aβ burden. (A) Regional differences in Aβ burden for CN-MCI/AD (N = 38) > CN-CN (N = 38), and (B) MCI-AD (N = 104) > MCI-MCI (N = 104). Results are color-coded according to the t-value statistics and illustrated on an inflated representation of the brain using CAT12 toolbox.

Predictive factors of progression

Results indicated that regional Aβ was predictive for progression from CN to MCI and AD, whereas all other predictors, including global Aβ burden, were not (X(9) = 21.029, p = 0.013). APOE4 carriership was trend significant in the model (p = 0.051). The model explained 42.4% of the variance and correctly classified 80.0% of cases (Table 3). For the full results table see Appendix C. Regional Aβ and increased age were particularly sensitive to predict progression in fast (≤ 36 months) compared to slow (>36 months) progressors. APOE4 carriership, on the other hand, was particularly sensitive to predict progression in slow compared to fast progressors (Appendix A1).

Table 3

Binary logistic regression analysis for variables predicting progression from CN to MCI or AD and from MCI to AD

	Predictor	β	SE β	Wald's X2	df	p	eβ
CN-CN versus CN-MCI/AD	Constant	-15.390	7.833	3.860	1	0.049	0.000
	Regional Aβ	5.013	1.865	7.221	1	0.007	150.333
MCI-MCI versus MCI-AD	Constant	-7.347	3.111	5.577	1	0.018	0.001
	CSF Aβ	-0.002	0.001	4.731	1	0.030	0.998
	Regional Aβ	9.834	3.551	7.670	1	0.006	18650.596
	Test			X2	df	p	Nagelk. R2
CN-CN versus CN-MCI/AD	Overall model evaluation			21.029	9	0.013	0.424
	Goodness-of-fit test			10.282	7	0.173	-
MCI-MCI versus MCI-AD	Overall model evaluation			51.686	9	0.000	0.347
	Goodness-of-fit test			6.333	8	0.610	-

All predictor variables that showed non-significant associations were omitted from the table.

CN-CN n = 27, CN-MCI/AD n = 28. MCI-MCI n = 99, MCI-AD n = 74.

Key: eβ, odds ratio. CI, confidence interval.

For the MCI-MCI versus MCI-AD comparison, the CSF Aβ score and the mean regional SUVR were significant in the model (X(9) = 51.686, p < 0.000), indicating that a higher CSF Aβ score and higher Aβ burden in the identified regions are predictive for progression from MCI to AD. The model explained 34.7% of variance and correctly classified 74.6% of cases (Table 3). Regional Aβ predicted conversion within 12 months after established Aβ-positivity, together with sex and APOE4 carriership (X(18) = 57.095, p < 0.001). None of the variables predicted slow conversion (Appendix A1).

Analyzing the unique regional pattern of Aβ burden as a predictor of disease progression

We evaluated if the mean SUVRs in the regional cluster identified in the CN-CN versus CN-MCI/AD comparison was also sensitive in the regression model to predict MCI-AD progression. This model only yielded a significant effect of CSF Aβ (X(9) = 43.022, p < 0.001), while all other predictors were not significant. When the reverse analysis was performed including the CN-CN and CN-MCI/AD groups with the mean SUVR of the identified regions from the MCI-MCI versus MCI-AD analysis, none of the variables were significant in the model. Results indicate that the identified regions of the 2 comparisons are both unique for predicting progression from and to specific disease stages. The sensitivity analysis only including individuals with at least 4 years of follow-up revealed a similar regional pattern, but the variance explained by this model increased by 20% (Appendix B1).

Discussion

The main results of this study showed that specific regional Aβ deposition patterns predict progression from CN to MCI or AD, as well as progression from MCI to AD. In addition to known predictors such as APOE4 carriership and age, the regional pattern of Aβ burden was predictive for more rapid progression from CN to MCI or AD within 3 years after established Aβ-positivity. Similarly, regional Aβ burden was predictive for fast conversion from MCI to AD within one year after established Aβ-positivity, together with sex and APOE4 carriership. Finally, Aβ regions identified in each group comparison were unique for the prediction of progression within each progressor group (i.e., CN to MCI/AD or MCI to AD) and were confirmed with an additional sensitivity analysis. Results demonstrate that regional Aβ constitutes a valuable predictor for progression to prodromal and clinical AD in Aβ-positive individuals, which had more predictive value than global Aβ or single CSF biomarkers. Findings are discussed in more detail in the context of (1) regional Aβ staging, (2) hemispheric asymmetry, and (3) the contribution of APOE4 carriership and CSF biomarkers.

Regional Aβ staging: Unique regional patterns of Aβ predict progression

Recent studies investigating regional Aβ burden explored its topographical pattern in vivo using Aβ-PET (Cho et al., 2016; Grothe et al., 2017; Mattsson et al., 2019; Sakr et al., 2019) and observed successive Aβ accumulation beginning in the precuneus, medial orbitofrontal and posterior cingulate cortices, spreading to core regions of the default mode network, associative neocortex, primary sensory-motor cortex, and medial temporal lobe, finally affecting the striatum. Here, we demonstrate that beyond global Aβ burden, Aβ accumulation in precuneus, lingual and angular gyrus, medial and superior temporal gyrus as well as subcortical regions was predictive of progression from a cognitively normal state to prodromal or clinical AD. Most of these regions are comprised in stage 1 of the staging model proposed by Mattsson and colleagues (Mattsson et al., 2019), suggesting that regions susceptible early in the process of Aβ deposition continue to accumulate Aβ and are indicative of disease progression. The identified regions and higher age were especially predictive for fast progression from CN to MCI or AD, that is, progression within the first 36 months after Aβ-positivity. The observed age effect is in line with previous studies reporting an increased risk of progression with advanced age (Corrada et al., 2010; Oulhaj et al., 2009). Furthermore, age-associated slowing of clearance mechanisms of Aβ has been suggested to be a factor of increased progression (Patterson et al., 2015). As our analyses have shown, the identified regions are unique for disease progression stage, despite some regional overlap (i.e., in the precuneus). The MCI-AD group showed a more widespread pattern of Aβ accumulation compared to the CN-MCI/AD group, demonstrating that regions such as the precuneus accumulate Aβ early and late during the disease progression, while regions such as the medial frontal cortex or anterior insula show increased Aβ accumulation only in advanced disease stages.

However, not all of the identified regions in our analyses matched the regional stage phase 1 reported by Mattsson's et al., (2019). These differences might be explained by different methodological approaches between the current study and the staging approach (Mattsson et al., 2019). Whereas Mattsson and colleagues (2019) used CSF and PET data to build composite regions of Aβ-staging, our study used different clinical diagnoses in addition to global measures of Aβ-positivity as main inclusion criteria (i.e., CN, MCI, AD) to identify regional differences. Overall, our study demonstrates the utility of regional Aβ deposition in the prediction of disease progression, similar to these previously presented staging approaches (Grothe et al., 2017; Mattsson et al., 2019). A study by Hanseeuw and colleagues (2018) included the striatum in their staging model and showed that the striatum is one of the last regions to accumulate Aβ while cortical regions start to accumulate Aβ first. In line with this, we were able to demonstrate that the striatum is a sensitive region for differentiating Aβ-positive CN-stables from CN-MCI/AD progressors. In later stages of the disease, other regions (i.e., anterior cingulate gyrus, medial frontal cortex, precuneus, transverse temporal gyrus, middle and superior temporal gyrus, posterior and anterior insula) seem to be more sensitive in differentiating Aβ-positive MCI-stables from MCI-AD progressors.

Our results suggest that a binary distinction in Aβ-positives and negatives might not be sufficient to inform on differential disease trajectories and therefore underscores the importance of considering additional regional susceptibilities in the clinical setting. Given our results, we propose that different magnitudes of Aβ burden in specific regions, as recently suggested (Bischof & Jacobs, 2019), could be utilized to examine the future course of the disease progression. Furthermore, our sensitivity analysis revealed mostly similar regional patterns of Aβ burden with our main analysis, indicating that the variability in follow-up length between progressors and non-progressors did not significantly influence the results.

Our results may have consequences for the clinical reading of Aβ-PET. In addition to assessing global Aβ burden visually, a regional reading approach using the identified cluster here, could potentially assist the clinician in evaluating a clinical prognosis for the patient. From a clinical point of view, prognosis and disease trajectories are the most frequent questions posed by caregivers and patients, therefore replicating these results is crucial from many perspectives.

Hemispheric asymmetry of Aβ deposition

Interestingly, when assessing the regional Aβ distribution patterns, a particular hemispheric dominance was observed dependent on progression type (i.e., CN-MCI/AD or MCI-AD). Here we showed that Aβ deposition appears to preferentially start in the right hemisphere in prodromal stages of the disease, ultimately spreading to regions in the left hemisphere with disease progression, suggesting that left-hemispheric Aβ accumulation is associated with cognitive decline. Most people show a left-hemispheric dominance, meaning that right-hemispheric pathology remains unnoticed for a longer time due to compensatory mechanisms of the left hemisphere. However, pathology does not have to start in the right hemisphere nor does right-hemispheric pathology reflect early disease stages. Left-hemispheric dominance of Aβ deposition may become symptomatic faster than right-hemispheric dominance. With this, it appears that left-hemispheric dominance of Aβ deposition is associated with advanced disease progression.

Aβ deposition follows a specific pattern of spread over time (Thal et al., 2002). To our knowledge, only few studies have reported a hemispheric asymmetry of Aβ deposition, among which a study by Frings and colleagues (Frings et al., 2015). They observed a leftward asymmetry of Aβ associated with more severe cognitive impairment in language performance, while there was a rightward asymmetry of Aβ associated with decline in tasks involving visuo-spatial perception. The previous inability to detect this relationship has often been ascribed to a plateau of Aβ deposition, which has been reached while degeneration and the progression of cognitive deficits. This is mostly true for studies using post-mortem Aβ pathology and in vivo cognition, while the current study investigated both in vivo Aβ pathology and cognition. Our results on hemispheric asymmetry of Aβ deposition specifically highlight the advantage of in vivo studies and suggest that hemispheric asymmetry of Aβ may be a valuable biomarker in the future that needs further validation.

Contribution of APOE4 carriership and CSF biomarkers

Even though APOE4 carriership was only trend significant in the CN-CN versus CN-MCI/AD analysis, it was a significant predictor for progression after 3 years of Aβ-positivity. It is thought that APOE4 carriership is particularly associated with the accumulation and initial spread of Aβ across cortical regions (Kanekiyo et al., 2014). However, once Aβ has accumulated across the brain, APOE4 carriership may be less relevant for cognitive decline, and tau pathology or neurodegeneration subsequently become more prevalent (Morris et al., 2010). Accordingly, APOE4 carriership predicted slow progression from CN to MCI/AD, hence progression after 36 months of Aβ-positivity. Since APOE4 carriership is a long-life risk factor for AD, potential compensatory mechanisms to Aβ accumulation may have developed earlier in life in APOE4-carriers compared to non-carriers, thus explaining the generally slower trajectories to advanced disease stages in this cohort. Current evidence on the relationship of differential trajectories of cognitive decline and APOE4 carriership is mixed. Some studies suggest APOE4 carriership accelerates cognitive decline (Craft et al., 1998; Hirono et al., 2003), while others suggest APOE4 carriership decelerates cognitive decline (Frisoni et al., 1995; Stern et al., 1997). Yet, others report no significant effect of APOE4 carriership on cognitive trajectories (Kleiman et al., 2006). However, most of the studies investigating this relationship showed different methodological approaches or included different populations (i.e., early vs. late-onset AD patients), which could explain the mixed result pattern. Notably, our results suggest that processes leading to the onset of AD may differ from those that determine its clinical onset. APOE4 carriers may have a greater risk for AD but show a different effect on processes determining the rate of progression than APOE4 non-carriers.

Interestingly, none of the CSF variables was significant in predicting progression from CN to MCI or AD. However, among the CSF biomarkers, CSF Aβ predicted progression from MCI to AD in both our main and sensitivity analyses. This suggests that in vivo assessments of regional Aβ may be a better predictor for particularly short-term progression than single CSF measures of Aβ or tau, once individuals have reached Aβ-positivity. CSF markers turn positive earlier in the disease cascade than in vivo PET imaging, limiting their value for short-term prediction of clinical progression (Fagan et al., 2006; Fagan et al., 2009). In contrast, ratios of specific CSF biomarkers (i.e., Aβ42/Aβ40 or Aβ42/p-tau) were successful in predicting progression from prodromal to clinical AD, potentially suggesting that ratio information rather than single parameters derived from CSF have more predictive value, specifically in Aβ-positive individuals (Ferreira et al., 2014). Despite CSF measures being more easily available than in vivo PET imaging, our results have shown that Aβ PET is a more informative measure for the prognostic prediction of progression to MCI or AD than single CSF measures, especially in already Aβ-positive individuals.

Limitations and future directions

Some limitations of this study need to be mentioned. First, due to the use of publicly available data from the ADNI database, the analyses were limited to patients for whom a complete dataset for all the assessed predictive variables were available. Nevertheless, for the longitudinal design of our approach, and the holistic biomarker model we interrogated, the number of subjects is sufficient, but warrants replication in a different cohort, as well as using different fluorine labelled tracers for Aβ quantification. Nevertheless, we had the privilege to use this unique dataset, containing a variety of longitudinal data, whereas other datasets still need to advance to show a comparable sample size. Secondly, we used the same cohort for the extraction of regional Aβ patterns and the investigation of the utility of these regions in predicting progression to advanced disease stages. Therefore, future studies may want to use these findings to validate the predictive value of the identified regions in a different cohort. Also, we only included single CSF measures instead of ratio information due to missing data from the ADNI database. Inclusion of these could have resulted in more conclusive results on the contribution of CSF biomarker and should be addressed in future work. Thirdly, results regarding the asymmetry pattern observed when comparing Fig. 1A and B should be taken with caution, since they are not based on longitudinal follow-up analyses in the same individuals. Even though our study points towards this hemispheric asymmetry, more longitudinal studies are needed to validate these results. Lastly, ADNI represents a trial population, including primarily amnestic MCI patients, rather than a clinical heterogeneous cohort. Thus, constructing models for Aβ PET in ADNI might have resulted in an overestimation of the effect of Aβ PET as found in amnestic MCI patients. However, especially amnestic MCI patients may benefit from these results, as clinicians may be able to provide disease prognoses by evaluation of Aβ burden in the identified regions.

The current recommendation of Aβ-PET imaging does not suggest a clinical value for Aβ-positivity in asymptomatic individuals. However, our results have shown that Aβ-PET imaging is able to provide information on the imminent cognitive decline and trajectory in the short-and long run in even asymptomatic individuals (i.e., CN). Results may aid in identifying those Aβ-positive healthy individuals who are at an increased risk for progression and could potentially lead to a revision of the current diagnostic guidelines of Aβ-PET imaging.

Conclusion

Regional Aβ burden appears to be the most sensitive prognostic biomarker of progression in comparison to multiple other biomarkers in both Aβ-positive CNs and MCIs. These findings suggest that regional Aβ burden may aid in the future to assess the short-term prognostic trajectories for imminent cognitive decline in preclinical and prodromal patients on the AD continuum. Especially Aβ-positive healthy individuals may benefit from these results, as current Aβ-PET guidelines do not include this cohort as a target population.

Verification

As corresponding author, I take full responsibility for the data, the analyses and interpretation, and the conduct of the research. I had full access to all of the data and have the right to publish any data. All co-authors have seen and agreed to conditions noted on the Author Agreement Form. As per ADNI protocols, all procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. More details can be found at adni.loni.usc.edu. (This manuscript does not contain any studies with human participants performed by any of the authors). The manuscript has not been previously published and is not under consideration for publication elsewhere. No submissions overlapping with the current work have previously been published.

Disclosure statement

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: JP, MCH, ED and GNB report no conflict of interest. TvE reports having received consulting and lecture fees from Lundbeck A/S, Lilly Germany, Shire Germany and research funding from the German Research Foundation (DFG), the Leibniz Association and the EU-joint program for neurodegenerative disease research (JPND). AD reports having received research support and speaker honoraria by Life Molecular Imaging, AVID/Lilly Radiopharmaceuticals, Siemens Healthineers, GE Healthcare.

Table A1

Group differences between fast and slow progressors

	Fast progressors	Slow progressors
	Mdn (SD)	Mdn (SD)	Test stat/p
CN-MCI/AD
Sex (m/f)	11/10	9/8	0.001/0.973b
APOE4 (±)	9/12	12/5	2.922/0.087b
Age	81.90 (4.81)	75.00 (4.22)	59.000/0.000a,c
Years of education	16.00 (2.89)	16.00 (2.96)	158.000/0.561c
Global Aβ	1.3 (0.16)	1.4 (0.20)	147.000/0.355c
CSF Aβ	793.30 (382.40)	747.00 (320.95)	75.500/0.486c
CSF t-tau	329.60 (102.79)	334.80 (78.43)	86.500/0.867c
CSF p-tau	33.37 (11.86)	30.32 (8.89)	85.500/0.829c
Regional Aβ	1.50 (0.24)	1.51 (0.21)	192.000/0.706c
MCI-AD
Sex (m/f)	20/15	24/11	0.979/0.322b
APOE4 (±)	31/4	22/13	6.293/0.012a,b
Age	74.30 (5.63)	73.40 (6.85)	590.000/0.792c
Years of education	16.00 (2.32)	16.00 (2.58)	532.000/0.334c
Global Aβ	1.48 (0.17)	1.38 (0.14)	424.000/0.027a,c
CSF Aβ	648.33 (189.51)	705.30 (207.16)	568.000/0.758c
CSF t-tau	357.15 (123.76)	319.00 (130.35)	425.500/0.127c
CSF p-tau	33.75 (14.25)	30.40 (15.52)	438.000/0.174c
Regional Aβ	1.48 (0.17)	1.39 (0.13)	401.000/0.013a,c

Data are presented as mean (standard deviation) for continuous variables or frequencies for categorical ones.

Key: Aβ, amyloid-beta; CSF, cerebrospinal fluid; P-tau, Phospho-tau; T-tau, Total-tau.

p < 0.05.

Test stats refer to Pearson's chi-square.

Test stats refer to Mann-Whitney's U.

Table A2

Final model summary of multinomial logistic regression analysis for variables predicting fast and slow progression in the CN-MCI/AD group

	Predictor	B	SE	Wald/p	Exp(B)	df	95% CI
Fast P.
	Age	.341	.134	6.531/0.011	1.407	1	1.083-1.828
	Regional Aβ	5.626	2.097	7.198/0.007	277.670	1	4.555-16926.470
	Intercept	-31.130	11.714	7.063/0.008	-	1	-
Slow P.
	APOE4 (0)	-2.693	1.160	5.389/0.020	0.068	1	0.007-0.657
	Intercept	-2.826	9.781	0.083/0.773	-	1	-
			Nagelk. R2	PAC	X2	p	Df
Overall model evaluation			0.572	74.5	38.068	18	0.004

Table A3

Final model summary of multinomial logistic regression analysis for variables predicting fast and slow progression in the MCI-AD group

	Predictor	B	SE	Wald/p	Exp(B)	df	95% CI
Fast P.
	Regional Aβ	8.287	4.314	3.691/0.005	500095.51	1	53.170–8924841576
	Sex (0)	-0.178	0.571	0.098/0.031	0.241	1	0.067–0.877
	APOE4 (0)	0.423	0.548	0.596/0.026	0.163	1	0.033–0.802
	Intercept	-17.864	5.306	11.335/0.001	-	1	-
			Nagelk. R2	PAC	X2	p	Df
Overall model evaluation			0.385	63.6	57.095	0.000	18

Table B2

Binary logistic regression analysis for variables predicting progression from CN to MCI or AD and from MCI to AD

	Predictor	β	SE β	Wald's X2	Df	P	eβ	95% CI eβ
CN-CN versus CN-MCI/AD	Constant	-30.138	13.859	4.729	1	0.030	0.000	-
	Regional Aβ	10.290	4.372	5.540	1	0.019	29432.965	5.594–154851292
	Age	0.390	0.172	5.169	1	0.023	1.477	1.055–2.068
MCI-MCI versus MCI-AD	Constant	-14.431	5.961	5.862	1	0.015	0.000	-
	CSF Aβ	-0.004	0.002	3.827	1	0.050	0.996	0.993–1.000
	Regional Aβ	17.406	7.546	5.321	1	0.021	36268952	13.694-9.606E+13
	Test			X2	df	p	Nagelk. R2	PAC
CN-CN versus CN-MCI/AD	Overall model evaluation			23.652	9	0.005	0.618	87.2
	Goodness-of-fit test			12.922	8	0.115	-	-
MCI-MCI versus MCI-AD	Overall model evaluation			52.746	9	0.000	0.629	86.7
	Goodness-of-fit test			2.468	8	0.963	-	-

All predictor variables that showed non-significant associations were omitted from the table.

Key: eβ, odds ratio. CI, confidence interval.

Table C1

Binary logistic regression analysis CN-CN versus CN-MCI/AD

		B	S.E.	Wald	df	p	Exp(B)
	Age	0.145	0.080	3.253	1	0.071	1.156
	Sex (1)	-1.089	0.902	1.458	1	0.227	0.337
	Education	-0.057	0.165	0.120	1	0.729	0.945
	APOE4 (1)	-1.556	0.797	3.811	1	0.051	0.211
	CSF amyloid	-0.002	0.001	1.739	1	0.187	0.998
	CSF t-tau	0.028	0.020	1.933	1	0.164	1.028
	CSF p-tau	-0.256	0.185	1.914	1	0.166	0.774
	Global amyloid	0.002	2.259	0.000	1	0.999	1.002
	Regional amyloid	5.013	1.865	7.221	1	0.007a	150.333
	Constant	-15.390	7.833	3.860	1	0.049	0.000

Sex coded as 1 = male and 2 = female; APOE4 carriership coded as 1 = negative and 2 = positive.

p < .005

Table C2

Binary logistic regression analysis MCI-MCI versus MCI-AD

	B	S.E.	Wald	df	p	Exp(B)
Age	0.031	0.033	0.875	1	0.350	1.031
Sex (1)	-0.675	0.438	2.375	1	0.123	0.509
Education	0.084	0.071	1.398	1	0.237	1.087
APOE4 (1)	-0.112	0.442	0.064	1	0.800	0.894
CSF amyloid	-6.409	3.425	3.501	1	0.061	0.002
CSF t-tau	-0.002	0.001	4.731	1	0.030a	0.998
CSF p-tau	-0.002	0.006	0.141	1	0.708	0.998
Global amyloid	0.065	0.058	1.248	1	0.264	1.067
Regional amyloid	9.834	3.551	7.670	1	0.006a	18650.596
Constant	-7.347	3.111	5.577	1	0.018	0.001

Sex coded as 1 = male and 2 = female; APOE4 carriership coded as 1 = negative and 2 = positive.

p < 0.005.