Amyloid PET in sporadic early- versus late-onset Alzheimer’s disease: comparison of the LEADS and ADNI cohorts

Abstract

Objective

Early- (EOAD) and late-onset Alzheimer’s disease (LOAD) differ in many respects. Here, we address the issue of possible differences in fibrillar amyloid pathology as measured by PET, which remains unresolved due to the lack of large-scale comparative studies.

Methods

399 cognitively impaired participants younger than 65 years from the multicenter Longitudinal Early-onset Alzheimer’s Disease Study (LEADS) and 450 cognitively impaired participants older than 65 years from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) underwent clinical assessment, brain MRI, and amyloid PET and were included. We compared amyloid PET outcomes (positivity rate based on visual read and quantified tracer uptake expressed as Centiloids) between the two cohorts and studied their association with age, sex, APOE genotype, and cognition.

Results

Amyloid positivity rate was higher in LEADS (78%, 95% CI=74–82) than in ADNI (71%, 95% CI=67–75, p=0.02). Lower MMSE and APOE4 genotype increased odds of amyloid positivity in both cohorts. Visually positive scans had higher Centiloids in LEADS (EOAD, mean=95.3±26.1) than in ADNI (LOAD, mean=80.9±36.8, p<0.0001), predominantly in parietal cortex/precuneus, superior temporal, and frontal cortices. In amyloid-positive patients, i) Centiloids were higher in females in both cohorts, ii) APOE4 carriership was associated with lower Centiloids in EOAD, which was not observed in LOAD, iii) correlations between Centiloids and MMSE scores were significantly stronger in EOAD than in LOAD.

Interpretation

Differences in the burden of amyloid pathology may contribute to differences in clinical and anatomic patterns in sporadic EOAD and LOAD, and have implications for optimizing therapeutic strategies in each group.

Summary for Social Media If Published

The biological mechanisms underlying clinical difference between patients with early- (EOAD, age-at-onset<65) and late-onset Alzheimer’s disease (LOAD) are not clearly established. Previous studies comparing fibrillar amyloid pathology between EOAD and LOAD have yielded inconsistent results.

Here, we compared fibrillar amyloid pathology as measured by PET between patients clinically diagnosed with sporadic EOAD and LOAD, leveraging data from two large, well-characterized cohorts.

Amyloid positivity rate and amyloid tracer uptake were higher in patients under 65, suggesting differences in burden of fibrillar amyloid pathology between sporadic EOAD and LOAD. These results inform the design of future clinical trials and implementation of novel anti-amyloid therapies in patients with sporadic EOAD.

Graphical Abstract

INTRODUCTION

Sporadic early-onset Alzheimer’s disease (EOAD, age-at-onset<65) accounts for ~5% of all AD and has serious consequences on professional and family life. Because clinical presentation is often atypical (i.e., non-amnestic),^1,2 diagnosis is challenging and often delayed. As a result, patients with EOAD present at more advanced clinical and biological disease stages at the time of diagnosis, potentially limiting options for treatment and eligibility for therapeutic trials. A more aggressive disease course has been reported in EOAD,^3,4 as well as more marked cortical atrophy^5,6 and more severe cortical hypometabolism.⁷ Although these features of EOAD are widely reported, the biological mechanisms underlying the difference between sporadic EOAD and LOAD are not clearly established.

Amyloid PET can detect and quantify amyloid pathology in vivo by binding aggregated A𝜷 fibrils that make up neuritic plaques. When EOAD is clinically suspected, amyloid PET can establish the diagnosis and guide management.⁸ It may also be used for studying biological differences between EOAD and LOAD.

Neuropathological studies comparing EOAD and LOAD have generally reported higher tau burden in EOAD as well as higher amyloid load, although to a lesser degree.⁹ Given the amyloid cascade hypothesis, these differences, particularly in tau burden between EOAD and LOAD, can be expected to be partly related to quantitative or qualitative differences in amyloid pathology. Although there is no difference in CSF A𝜷42 levels,¹⁰ there is some suggestion from biofluid and pathological data that the amyloid itself may be different, with studies reporting lower CSF A𝜷43,¹¹ and presence of coarse-grained plaques in EOAD.¹²

While the results of PET studies are fairly consistent for tau pathology,^13,14 they are more nuanced for amyloid. The PET studies carried out to date have generally been small, single-site, and found minimal or no differences across EOAD and LOAD: most often, they show no difference in overall amyloid load,^4,15–18 although some show regional differences (higher tracer uptake in basal ganglia, thalamus, temporal cortex, parietal lobes, orbitofrontal regions, or occipital lobes in EOAD).^16,19–23 Large-scale studies comparing sporadic EOAD and LOAD are rare.

In this study, we compared amyloid PET results in the Longitudinal Early-Onset AD Study (LEADS) and the Alzheimer’s Disease Neuroimaging Initiative (ADNI), two large, multi-site observational studies of sporadic EOAD and LOAD, respectively. Comparing rates of amyloid PET positivity, baseline burden of amyloid, and correlations of amyloid with clinical measures between clinically-diagnosed sporadic EOAD and LOAD can inform clinical use of amyloid PET, implementation of newly approved amyloid therapies, and future trial design in each population. Our a priori hypotheses were: (i) rates of amyloid PET positivity will be higher in LEADS than in ADNI; (ii) amyloid PET binding at baseline will be higher in EOAD but show a similar distribution to LOAD; (iii) amyloid PET will show stronger correlations with clinical measures in EOAD than in LOAD.

METHODS

Study design and participants

We included participants from LEADS²⁴ and ADNI (https://adni.loni.usc.edu/). LEADS and ADNI protocols were harmonized to ensure compatibility of cognitive, biomarker, and neuroimaging data, with the intention of comparing sporadic EOAD and LOAD.

For LEADS participants, we included 399 patients whose baseline visits were completed between February 2018 and January 2024 and who (i) met the National Institute on Aging - Alzheimer’s Association (NIA-AA) criteria for mild cognitive impairment (MCI) due to AD²⁵ or dementia due to AD, including amnestic, dysexecutive, logopenic variant of primary progressive aphasia - PPA, and posterior cortical atrophy – PCA clinical AD variants;²⁶ (ii) had global Clinical Dementia Rating® (CDR) scores of 0.5 or 1; (iii) were 40–64 years of age at the time of consent; and (iv) had quality-controlled data in the National Alzheimer’s Coordinating Center’s (NACC) Uniform Data Set (UDS). Participants received a standardized baseline clinical assessment, which incorporated medical and family history, medication review, and medical/neurological examinations. They also underwent structural MRI and amyloid-PET imaging using ¹⁸F-Florbetaben (FBB) at baseline. Clinical diagnoses were made at each site through a multidisciplinary consensus process.²⁷ Written informed consent was obtained from all participants or their surrogates. The study was conducted under a central institutional review board based at Indiana University.

For ADNI participants, we included 450 patients from ADNI-2 and ADNI-3 whose baseline visits were between February 2011 and January 2020 with (i) amnestic cognitive impairment suggestive of AD; (ii) global CDR score of 0.5 or 1; (iii) estimated age of symptom onset > 65 (age of cognitive decline if available, or, if not, age at first inclusion in ADNI); (iv) baseline amyloid PET with available Centiloid (CL) values, indicating that the scan has been successfully processed with the ADNI pipeline. All ADNI participants underwent an extensive clinical and neuropsychological battery at a baseline visit upon their enrollment, similar to the LEADS neuropsychological battery which was designed to ensure comparability with ADNI. Within a span of 6 months, ADNI participants included in the current study underwent brain MRI and amyloid-PET imaging using either ¹⁸F-Florbetapir (FBP) or FBB, depending on the ADNI protocol. IRB approval was obtained locally for each ADNI site, and written informed consent was obtained from study participants or their authorized representatives (see supplementary methods for details).

Amyloid PET data acquisition and initial pre-processing

FBB PET (LEADS and ADNI) was acquired via intravenous injection of 8.0 ± 0.8 mCi, followed by acquisition of 4 × 5-minute PET frames at 90 to 110 minutes post-injection. FBP PET (ADNI only) was acquired via intravenous injection of 10.0 ± 1.0 mCi, followed by acquisition of 4 × 5-minute PET frames at 50 to 70 minutes post-injection.²⁸ PET frames acquired at each site were uploaded to the Laboratory of Neuroimaging (LONI) at the University of Southern California. Attenuation correction for each scan was performed using either CT or PET transmission data, and reconstruction used site-specific iterative algorithms developed for ADNI. Quality control and image standardization were performed at the University of Michigan: frames were realigned and averaged, set to a standard orientation, intensity-normalized, and smoothed to standard 8 mm isotropic resolution using procedures developed for ADNI and applied in LEADS. Resulting images were then re-uploaded to LONI. Further analysis of PET data was performed at the University of California, San Francisco (UCSF) Alzheimer’s Disease Research Center, based on these quality-controlled, pre-processed PET scans.

Determination of amyloid positivity

In all participants, amyloid PET clinical read was used to distinguish amyloid-positive and amyloid-negative patients via a process that was originally developed in LEADS.²⁹ The same approach was applied to ADNI scans, with a slight modification to accommodate for different radiotracers; LEADS scans were performed with FBB while ADNI participants received either FBB (n=112) or FBP (n=338). LEADS and ADNI scans were centrally analyzed at UCSF by combining expert visual read and amyloid-PET quantification to maximize confidence in the final scan interpretation. Briefly, each amyloid-PET scan was visually interpreted by a certified clinician according to published and FDA-approved criteria,^30,31 without access to the participant’s clinical information or PET quantification. In parallel, scans were pre-processed using a PET-only pipeline (used only to determine amyloid positivity),²⁹ which resulted in a global Standardized Uptake Value Ratio (SUVR_PET-only) quantification with tracer-specific positivity threshold values as follows: 1.18 for FBB [corresponding to 39.2 CL, as defined in LEADS] and 1.15 for FBP (PET-only SUVR value corresponding to 39.2 CL in a subset of ADNI participants who received FBP). If visual interpretation and quantification-based classification were incongruent, another visual read was performed by a second reader who was blind to the previous read and quantification. This second read was used as a tiebreaker to determine amyloid status (see supplementary methods for details).

MRI-based PET processing

Amyloid PET images were also processed with an MRI-based pipeline in order to extract a composite neocortical SUVR, which was then converted to CL using the equations developed in ADNI with identical image pre-processing.³² All results presented below are based on amyloid-PET measures derived from the MRI-based processing pipeline. Participants underwent a 3T brain MRI with sagittal 3D accelerated MPRAGE/IRSPGR T1-weighted sequence, which was processed with the FreeSurfer 7.1 processing stream (http://surfer.nmr.mgh.harvard.edu/). We visually inspected the FreeSurfer parcellation results to identify segmentation abnormalities and performed manual edits when necessary. PET images were co-registered to the participants’ respective T1 structural MR images using Statistical Parametric Mapping 12 (Wellcome Department of Imaging Neuroscience, Institute of Neurology, London, UK). FBB and FBP PET SUVR images were created using mean activity in FreeSurfer-parcellated whole cerebellum as the reference region. In addition to the composite neocortical CL value, mean SUVR values were extracted in native MRI space from ROIs defined in the Desikan-Killiany atlas labeled by FreeSurfer 7.1.

Although the MRI-based LEADS pipeline was designed to harmonize with ADNI to facilitate comparisons, we reprocessed ADNI scans at 8 mm resolution using the LEADS pipeline to exclude the possibility that differences in pipeline or smoothing could confound the results. We found that the CL values obtained with the two pipelines were very consistent, with an R² of 0.997 and a regression line almost identical to the identity line (intercept = −0.2, slope =1.01).

One ADNI participant was excluded because of FreeSurfer segmentation errors that could not be corrected and another because of artifacts on the amyloid scan that interfered with quantification.

Statistical analyses

Data were analyzed using R version 4.4.1 (R Core Team, 2017) and Stata v. 18.1 (StataCorp, College Station, Tx). Demographic and clinical data were summarized by means and standard deviation and analyzed via t-tests or one-way analysis of variance (ANOVA) for continuous variables. Categorical variables were reported as frequencies and percentages and compared using chi-squared tests. Logistic regressions in each cohort examined the relationship between amyloid positivity and age, sex, APOE genotype and MMSE score. We compared the amyloid load (CL values) between amyloid-positive and amyloid-negative patients across cohorts (LEADS and ADNI) via two-way ANOVA with an amyloid status-by-cohort interaction term. We performed sensitivity analyses by stratifying patients according to the CDR global score (0.5 or 1), and by removing LEADS patients with a clinical phenotype described as non-amnestic (PCA, PPA, other non-amnestic) to retain only those described as predominantly amnestic, who are thus more clinically comparable to ADNI patients (PCA and PPA are not represented in ADNI). In order to better characterize the distribution of CL values and to confirm the difference observed among amyloid-positive patients by defining amyloid positivity differently from the aforementioned clinical read, Gaussian mixture models (GMM) were fitted to CL values for each cohort (Mixtools R package), and data-driven CL cutoffs were calculated as mean + 2SD of the first (lower) Gaussian. Regression analyses were performed in the amyloid-positive subgroup (as defined by visual read) to determine the relationship between CL values and age, sex, APOE genotype (dichotomized and by copy number) and MMSE score in EOAD and in LOAD. The level of statistical significance was set at p < 0.05 for all analyses. P-values were adjusted using false discovery rate to account for multiple comparisons where necessary.

Finally, we analyzed regional differences in amyloid load between amyloid-positive participants from LEADS (EOAD) and ADNI (LOAD) in the ROIs defined in the Desikan-Killiany atlas. To do this, we only considered ADNI patients in whom FBB was used as the amyloid PET tracer. In contrast to CL values that measure global amyloid load and allow the merging of data acquired with different tracers, regional SUVR values cannot be directly compared across tracers. Even after excluding LEADS patients with a non-amnestic phenotype, a significant imbalance between the number of patients with FBB in LEADS and in ADNI existed (n=259 and n=62 respectively). To reduce the imbalance, we performed 2-for-1 matching using propensity scores to match each ADNI case to two LEADS cases by sex, CDR, and APOE status, in order to describe a subgroup of patients in LEADS who could be compared directly with those in ADNI. The matched groups were compared using ANOVA and we reported regional effect size (Cohen’s d). For these regional group comparisons, we applied a threshold of p < 0.01 after FDR correction (for 80 cortical and subcortical regions-of-interest) to define statistical significance.

RESULTS

General comparison of LEADS and ADNI cohorts

Demographic and clinical comparisons between LEADS and ADNI are shown in Table 1 and Supplementary Table S1. LEADS participants were by definition younger than ADNI participants. There was no statistically significant difference between cohorts in estimated duration of symptoms. There was a non-significant trend for higher proportion of females (p=0.07) and slightly lower mean years of education (p=0.002) in LEADS. There was no statistically significant difference between cohorts in APOE4 allele carriership. A study of the main demographic and clinical variables in relation to APOE4 status in each cohort showed that APOE4 carriership was associated with younger age in ADNI (p<0.001) but not in LEADS (where it was non-significantly associated with older age, p=0.25) (Supplementary Table S2). At baseline, LEADS participants were more clinically impaired in terms of both MMSE (p<0.001) and global CDR (p=0.002). In ADNI, participants scanned with FBP were older and more cognitively impaired than those scanned with FBB (Supplementary Table S3).

Table 1.

Main demographic and clinical characteristics of the participants included

	LEADS (n=399)	ADNI (n=450)	p value
Age at amyloid PET (years)	58.4 ± 4.4	77.2 ± 6.0	<0.001
Age at symptom onset (years)$	55.0 ± 4.9	73.6 ± 6.2	<0.001
Estimated duration of symptoms (years)$	3.5 ± 2.3	3.6 ± 3.2	0.54
Sex (F/M, %M)	198/201 (50%)	190/260 (58%)	0.07
Education (years)	15.5 ± 2.5	16.1 ± 2.6	0.002
Ethnicity: Not Hispanic or Latino/Hispanic or Latino/Unknown (% Not Hispanic or Latino)	385/14/0 (96%)	426/23/1 (95%)	0.37
Race: White/Black or African American/Asian/Native Hawaiian or Other Pacific Islander/More than one/Unknown (% White)	353/26/7/0/9/4 (88%)	407/29/5/1/6/2 (90%)	0.50
APOE4 n alleles (0/1/2, % E4 carriers)*	191/150/54 (52%)	216/172/57 (51%)	1.00
MMSE#	22.3 ± 5.3	25.8 ± 3.2	<0.001
CDR=0.5/CDR=1 (% CDR=1)	265/134 (34%)	347/103 (23%)	0.002
Amyloid tracer (florbetapir/florbetaben, % florbetaben)	0/399 (100%)	338/112 (25%)	-

Values represent mean ± standard deviation unless otherwise noted.

p values reflect results of independent samples t-tests between LEADS and ADNI participants for continuous variables, and Chi-Square for categorical variables (FDR correction for 10 comparisons).

^$5 missing data in LEADS

^*4 missing data in LEADS and 5 in ADNI.

^#3 missing data in LEADS

F = Female, M = Male, MMSE = Mini-Mental State Examination, CDR = Clinical Dementia Rating Global scale.

Amyloid positivity rates and predictors in each cohort

The rate of amyloid positivity (based on clinical read) was higher in LEADS (311/399, 78%, 95% CI=74–82) than ADNI (319/450, 71%, 95% CI=67–75, X²=5.2, p=0.02). In ADNI, 257/338 participants were amyloid-positive with FBP (76%) and 62/112 with FBB (55%). There was a higher proportion of APOE4 allele carriers among amyloid-positive participants, who also showed more marked cognitive impairment than their amyloid-negative counterparts in both LEADS and ADNI. EOAD vs. LOAD comparisons restricted to amyloid-positive participants in each cohort are shown in Supplementary Table S1 and generally mirrored group differences across the entire sample.

Logistic regressions, run separately in each cohort, found that lower MMSE (LEADS OR=0.79, 95% CI=0.73–0.85 and ADNI OR=0.78, 95% CI=0.71–0.86, p<0.001) and APOE4 genotype increased the odds of amyloid positivity in both cohorts, although the APOE4 effect was stronger in ADNI than in LEADS (ADNI OR=12.4, 95% CI=7.1–23 versus LEADS OR=2.5, 95% CI=1.49–4.38, p<0.001). There was no statistically significant association between age or sex and amyloid positivity in either cohort (Table 2).

Table 2.

Relations of amyloid positivity with age, sex, MMSE and APOE4 carriership in each cohort

	LEADS		ADNI
	OR [95%CI]	p value	OR [95%CI]	p value
Age	1.06 [0.99–1.12]	0.08	1 [0.97–1.05]	0.74
Sex (M)	0.68 [0.4–1.15]	0.15	0.88 [0.53–1.47]	0.63
MMSE	0.79 [0.73–0.85]	<0.001	0.78 [0.71–0.86]	<0.001
APOE4	2.54 [1.49–4.38]	0.001	12.4 [7.1–23]	<0.001

Logistic regressions in each cohort show the relations of amyloid positivity with age, sex, MMSE and APOE4 carriership (dichotomized as carrier/non-carrier). FDR correction for 4 comparisons.

OR = Odds ratio, CI = confidence interval, M = Male, MMSE = Mini-Mental State Examination.

Comparison of CL values between cohorts

CL values were bimodally distributed in both cohorts, but the separation between positive and negative scans was more prominent in LEADS. LEADS had fewer patients in an intermediate zone defined as CL values between 10 and 35, which corresponds to the transition from no pathology to subtle pathology, with values > 35 CL indicating established pathology (30/399 participants −7.5%- in LEADS and 54/450 −12%- in ADNI, p=0.04) (Fig 1A).³³ We found an interaction between amyloid status assessed by clinical read and the cohort (LEADS or ADNI) in their relation with CL values (p=0.016). Scans read clinically as positive had higher CL values in LEADS than in ADNI (mean CL = 95.3 vs 80.9, p<0.0001), whereas there was no statistically significant cohort difference in CL values for scans read clinically as negative (Fig 1B-C). ADNI mean CL values in scans read clinically as positive were similar in FBP and FBB scans, and both were significantly lower than LEADS scans read clinically as positive (all FBB, Supplementary Fig S1)

Figure 1:

Centiloid values in the two cohorts as a function of amyloid status (visual read) in all participants together with individual dots (A), represented in plots for amyloid-positive and amyloid-negative participants (B), in amyloid-positive participants only (EOAD and LOAD) with individual dots (C).

Sensitivity analyses

We studied the relation of CL values with amyloid status (clinical read) and cohort (LEADS or ADNI) by considering participants with CDR=0.5 and CDR=1 separately. The results were essentially unchanged, though the amyloid status-by-cohort interaction was only significant in the CDR=0.5 group. Results were also unchanged after removing LEADS participants with a non-amnestic clinical phenotype (PPA, PCA, or dysexecutive syndrome) (Supplementary Fig S2).

We repeated the analyses by defining amyloid status on the basis of data-driven CL cutoffs determined by GMM in each cohort rather than clinical reads (thresholds for positivity: 23.3 CL in LEADS, 25.2 CL in ADNI). Alternative amyloid positivity definitions (GMM-derived cut-off values for each cohort or single predefined values commonly reported in the literature: 25 CL and 18 CL) only had a minimal impact on positivity rates, which were significantly different between LEADS (77.9–81.0%) and ADNI (68.4–71.1%) across approaches (all p’s≤0.02) (Supplementary Table S4). The amyloid status-by-cohort interaction was no longer significant when defining amyloid positivity on the basis of GMM-derived cutoffs, although CL values remained higher in amyloid-positive patients in LEADS than ADNI (mean CL = 94.7 vs 84.5, p<0.0001) (Supplementary Fig S3).

We also studied the influence of the clinical site on amyloid PET outcomes. 123 ADNI participants were included from sites that also enrolled LEADS participants. Comparison of these patients with those enrolled from sites not participating in LEADS showed roughly similar rates of amyloid positivity (74% vs 70%, p=0.48) and similar CL values in either amyloid-positive or amyloid-negative patients (amyloid status-by-site interaction: p=0.61, with no main effect of site: p=0.74). A separate analysis performed in patients enrolled from sites shared by LEADS and ADNI also showed no statistically significant effect of the clinical site on the CL values. The difference in amyloid positivity rates between LEADS and ADNI may be partly driven by the trend towards a non-significantly lower rate in ADNI patients enrolled from sites not participating in LEADS (Supplementary Fig S4-5).

Regional analyses on the FBB sub-sample

120 amyloid-positive LEADS patients (EOAD) were matched on sex, global CDR, and APOE genotype to the 62 amyloid-positive patients from ADNI (LOAD) who underwent FBB PET (no available match for two patients). Region-of-interest analysis showed significantly higher SUVR values in EOAD patients, particularly in the parietal cortex/precuneus, superior temporal cortex, and frontal cortex, with no statistically significant difference in the sensorimotor and visual unimodal cortices, and in the basal ganglia (Fig 2, Supplementary Table S5).

Figure 2:

Regional comparison of the SUVR values between the 62 amyloid-positive ADNI participants (LOAD) scanned with florbetaben and 120 amyloid-positive LEADS participants (EOAD). LEADS participants had a predominantly amnestic clinical phenotype and were matched for sex, global CDR, and APOE genotype with ADNI participants (2 for 1). Mean SUVR values were extracted from 68 cortical and 12 subcortical ROIs from the Desikan-Killiany atlas. All comparisons are described in detail in Supplementary Table S2. We compared the actual SUVR values using ANOVA and calculated Cohen’s d values (Positive Cohen’s d values indicate higher SUVR values in EOAD than in LOAD), which are plotted in the figure for regions where the difference is significant (p<0.01). It should be noted that similar results were obtained when considering all LEADS participants, as well as other random samples within LEADS participants.

Relation between CL values and age, sex, APOE genotype, or MMSE in amyloid-positive patients

CL values, although correlated with age when all patients from the two cohorts were considered together (r=−0.18, p<0.0001), were not correlated with age within each cohort considered separately (r=−0.01, p=0.86 in EOAD, r=0.03, p=0.59 in LOAD) (Supplementary Fig S6).

CL values were higher in females in both cohorts (p=0.0005), with no statistically significant sex-by-cohort interaction (p=0.3) (Fig 3A).

Figure 3:

Representation of Centiloids by sex (A) or APOE genotype (dichotomized – B- or expressed as the number of E4 alleles -C) and cohort in amyloid-positive participants from LEADS (EOAD) and ADNI (LOAD).

The presence of at least one APOE ε4 allele was associated with lower CL values in EOAD, whereas the opposite trend was observed in LOAD (E4 carriership-by-cohort interaction: p<0.0001). Considering the number of E4 alleles, a dose effect was observed, with the lowest CL values in APOE4 homozygous EOAD participants, which was not the case in LOAD (Fig 3B-C). 5.1 or 7.3% of CL variance (depending on whether we consider the presence/absence or the number of E4 alleles) was explained by APOE genotype in EOAD, compared with 1.6% or 0.8% in LOAD. We also investigated the effect of family history of cognitive impairment on the relationship between CL values and APOE genotype in amyloid-positive EOAD patients (n=242 with available family history data) and found no statistically significant interaction between E4 carriership and a history of a first-degree relative with cognitive impairment (p=0.9) (Supplementary Fig S7).

CL values were negatively correlated with the MMSE score in EOAD (r=−0.18, p=0.001) and LOAD (r=−0.11, p=0.049); this relationship was more pronounced in EOAD (CL-by-cohort interaction: p=0.01) (Fig 4). Removing EOAD participants with a non-amnestic clinical phenotype did not lessen this relationship and, in fact, tended to amplify it (r=−0.25, p<0.0001) (Supplementary Fig S8).

Figure 4:

Correlation between Centiloids and MMSE score in amyloid-positive participants from LEADS (EOAD) and ADNI (LOAD).

DISCUSSION

This study sought to compare baseline amyloid PET findings in two large cohorts of clinically-diagnosed sporadic EOAD (LEADS) and LOAD (ADNI). We found higher rates of amyloid PET positivity and greater baseline amyloid PET uptake in LEADS than in ADNI. Among amyloid-positive patients, we also found stronger correlations between amyloid PET and global cognition in LEADS than ADNI, and a paradoxical effect whereby APOE4 genotype in LEADS was associated with lower CL values in a dose-dependent manner.

Consistent with previous literature,³⁴ we found slightly higher rates of amyloid positivity in young (LEADS) than in older (ADNI) patients with suspected AD. In both cohorts, a significant minority of participants presenting clinically with early-onset (22%) or late-onset (29%) AD were amyloid-negative, underscoring the diagnostic value of amyloid PET in both populations. In EOAD, the differential diagnosis more often includes frontotemporal dementia (FTD) and dementia with Lewy bodies (DLB), while amyloid-negative late-onset patients are more likely to have underlying Limbic-predominant Age-related TDP-43 Encephalopathy (LATE),³⁵ vascular pathology or DLB. Non-neurodegenerative etiologies of cognitive impairment should be considered at any age, although may be less frequent in older patients.³⁶ EOAD is identified as a high-utility indication for amyloid PET in both the original and recently updated Appropriate Use Criteria for amyloid PET^37,38 because amyloid is less prevalent in healthy individuals at age<65, which makes it easier to “rule-in” AD as a cause of impairment in younger patients on the basis of a positive amyloid scan. However, rates of amyloid-negative MCI/dementia mimicking AD clinically may be higher in older patients, and an important role of amyloid PET is also to exclude AD in the setting of a negative amyloid PET.^38,39 Based on significant rates of amyloid-negativity in ADNI, the Imaging Dementia-Evidence for Amyloid Scanning (IDEAS) study, clinical trials, and other observational studies, LOAD is also now considered an appropriate clinical indication for amyloid PET in updated Appropriate Use Criteria.³⁸

Previous PET studies comparing amyloid burden in EOAD and LOAD reported inconsistent results, with most often no significant difference found.^16,18,22,23 Our results are consistent with mildly greater amyloid load in EOAD, as has also been reported in autopsy studies. Previous studies may have failed to demonstrate this due to small effects and lack of power. Failure to differentiate between familial and sporadic EOAD in previous studies may have also confounded comparisons, since global tracer binding is lower in autosomal dominant AD (ADAD) than in sporadic EOAD,⁴⁰ perhaps due to the lower affinity of amyloid tracers for plaques found in genetic cases (e.g., cotton wool plaques).⁴¹ ADAD cases may have also driven the finding in some studies of elevated amyloid PET in the basal ganglia in EOAD, a pattern which has been mainly reported in ADAD but not in sporadic EOAD.¹⁶

The higher amyloid tracer uptake in EOAD than in LOAD could be explained by the fact that younger patients are diagnosed at a more advanced stage of amyloid pathology due to more frequent delayed diagnosis,² fewer co-pathologies, and higher cognitive and brain reserve, requiring more AD pathology to reach the same degree of cognitive impairment as older individuals. However, the estimated duration of symptoms was no longer in LEADS than ADNI (although this variable must be interpreted with caution as it is subject to bias based on patient and care partner recall, informant interview, atypical symptoms and delayed diagnosis, etc.). In addition, the main differences we observed in the regional analyses were in the temporal, parietal, and frontal associative regions, which are among the first to be affected by amyloid pathology in AD.⁴² If EOAD patients were at a more advanced stage of amyloid pathology, we might expect group differences to be mainly in the regions of late accumulation (sensorimotor and visual unimodal cortices), where tracer uptake can still increase, while the early regions are already saturated. This may suggest qualitative rather than purely quantitative differences in amyloid plaques between sporadic EOAD and LOAD. As suggested for ADAD versus sporadic EOAD, differences in Aβ fibril microstructure and available binding sites rather than Aβ pathology burden are conceivable.⁴⁰ They are suggested by the recent description of coarse-grained plaques in EOAD,¹² for which the affinity of amyloid tracers is not known, but could be different from that for neuritic plaques. These plaques are associated with intense neuroinflammation and vascular pathology, and they are predominantly found in frontal and parietal regions,¹² which are among those regions where we observed the greatest differences between EOAD and LOAD.

As previously reported,^43,44 we found a higher amyloid load in female patients in both LOAD and EOAD. A particularly interesting result, although counter-intuitive at first glance, is the lower amyloid load observed in EOAD APOE4 carriers^44,45 with a dose-dependent effect of the number of E4 alleles, which is not found in LOAD. This may be related to earlier diagnosis in APOE4 homozygotes due to the frequent presence of a family history of cognitive impairment, which makes families more aware of the disease. However, we found no difference in CL value in our EOAD patients according to the presence or absence of family history. It may alternatively be due to the association of APOE4 homozygosity with predominantly amnestic clinical phenotype,^46,47 contrary to the predominantly non-amnestic variants encountered in APOE4 non-carriers, which may be more delayed in diagnosis. This result could also reflect differential alteration of biological pathways other than amyloid pathology between EOAD and LOAD, which could contribute to heterogeneity in the expression of disease.⁴⁸ Non-Aβ pathways affected by the pleiotropic ApoE4 isoform include tau phosphorylation, glucose metabolism/insulin resistance, synaptic function, neuroinflammation via astrocyte and microglial function, and blood-brain barrier permeability.^45,49,50 It is likely that our sporadic EOAD cohort is enriched by novel, yet to be described genetic drivers, beyond APOE4.⁵¹ Future studies incorporating whole-genome sequencing, proteomic and transcriptomic analyses (e.g., in biofluids and brain tissue samples) are needed to further elucidate common and distinct biological pathways underlying EOAD and LOAD. Finally, we found a stronger association between amyloid pathology and global cognition in EOAD than in LOAD. This could be explained by the relatively lower degree of most co-pathologies in younger patients (with the exception of amygdala-predominant Lewy body disease and cerebral amyloid angiopathy, which have similar frequencies in EOAD and LOAD),⁵² leading to a more direct relationship between cognitive impairment and AD pathology. Across the AD continuum, the association between amyloid pathology and cognition appears to be mediated by tau pathology,^29,53 which was not studied here, but is likely to contribute to this result.

This study is the first to compare EOAD and LOAD in the era of amyloid lowering therapies, and our results have implications for clinical trials and implementation of these novel therapies. Our findings suggest that longer duration of anti-amyloid treatment may be required to achieve amyloid clearance in patients with EOAD, who start on average at higher CL values. Recent trials highlight that higher baseline amyloid levels, although associated with a greater amyloid reduction, led to a lower probability of achieving complete amyloid clearance, which may be related to clinical benefit.⁵⁴ The paradoxical effect of APOE genotype on CL values we found in EOAD is in line with the recently expressed idea that APOE4 homozygosity represents a distinct form of AD with its unique peak of effect in the 70s, which may manifest with pharmacogenomic differences in clinical response seen in clinical trials and with novel therapies.⁵⁵

Inherent differences in the design of LEADS and ADNI may have confounded our results. In recent years, ADNI has focused on the preclinical and early MCI stages of AD rather than the dementia stage, which explains a skew towards less impaired participants. In contrast, LEADS recruits EOAD participants with MCI and dementia, with a skew towards dementia, potentially exacerbated by the delay in clinical diagnosis of EOAD. Identifying individuals transitioning from preclinical AD to early MCI at age<65 is challenging given the low prevalence of disease in younger populations. This transition point is not captured in LEADS but is seen not infrequently in ADNI. Therefore, symptomatic EOAD-LEADS participants were overall more impaired at baseline than LOAD-ADNI participants, potentially contributing to the finding of higher amyloid at study entry. To address this confound, we stratified our analysis by global CDR. We also confirmed that defining amyloid positivity using a data-driven method (GMM) instead of clinical read did not affect our results. Additional differences between ADNI and LEADS include larger numbers of clinical sites in ADNI, two amyloid PET tracers in ADNI vs. one in LEADS, and inclusion of non-amnestic AD variants in LEADS. Sensitivity analyses addressing each of these issues suggested that our findings were not driven by differences in study design, and that differences in age were the most likely explanation.

This study has additional limitations. As mentioned above, although our sensitivity analyses are reassuring, certain differences may affect comparability between LEADS and ADNI. The use of two different amyloid tracers in ADNI limited the number of LOAD participants who underwent FBB-PET and could, therefore, be included in regional comparisons. Both LEADS and ADNI have limited racial and ethnic diversity, which makes it difficult to generalize the results to more representative populations. It will also be necessary to analyze longitudinal changes in amyloid tracer binding between the two cohorts, as well as the relationship with tau PET imaging.

Overall, the differences we found in amyloid PET positivity rates and amyloid tracer uptake between LEADS and ADNI inform the clinical use of amyloid PET and future trial design by highlighting specific characteristics of each population and the importance of considering the clinical/biological heterogeneity of AD. Longitudinal studies will clarify the process leading to an overall higher amyloid load in EOAD. Further work will be needed to elucidate possible biological differences in amyloid pathology between EOAD and LOAD.

Data Availability

The anonymized data that support the findings of this study are available on request from the LEADS data core (https://leads-study.medicine.iu.edu/researchers/leads-data-request-application/) and the ADNI website (https://adni.loni.usc.edu/).