Abstract
Background:
The National Institute on Aging (NIA)/Alzheimer’s Association (AA) 2018 framework conceptualizes Alzheimer’s disease (AD) biologically. Evidence of brain amyloid by biomarkers defines AD pathologic change and the Alzheimer’s continuum. The presence of tau or neurodegeneration in the absence of amyloid defines non-AD pathologic change.
Objective:
To examine the relation of in vivo amyloid and neurodegeneration with verbal learning, one of the cognitive abilities affected early in AD, in late middle age.
Methods:
This was a cross-sectional study of amyloid and neurodegeneration biomarkers in a community-based cohort of 350 late-middle aged Hispanics without dementia (mean age: 64.15±3.34; 72.0%women). Amyloid (A) was measured as global standardized uptake value ratio (SUVR) with 18F-Florbetaben positron emission tomography (PET). Neurodegeneration (N) was ascertained as cortical thickness (CT) in AD signature areas using brain magnetic resonance imaging. We examined A/N continuously, categorically, by A/N profiles, and profile categories. The amyloid threshold for positivity was defined using the K means method. The CT threshold was defined as 2 standard deviations below the mean CT. Verbal learning was ascertained using total recall and delayed recall in the Buschke Selective Reminding test (SRT).
Results:
Higher cortical thickness was associated with higher performance in SRT delayed recall. Amyloid SUVR was not related to SRT performance. The low CT category was associated with lower performance in SRT delayed recall, while Amyloid categories were not related to any SRT score. The non-AD pathologic change group (A-N+) performed worse in SRT delayed recall compared to the Normal A/N profile group (A-N-).
Conclusion:
In late middle-aged Hispanics without dementia, non-AD pathologic change, but not the Alzheimer’s continuum, was related to verbal learning.
INTRODUCTION
Dementia due to Alzheimer’s disease (AD) affects 1 in every 10 individuals in the United States who is 65 years or older [1, 2], with a higher risk among Hispanics and non-Hispanic Blacks as compared with non-Hispanic Whites [2]. The dominating hypothesis in AD research is the amyloid hypo-thesis, which posits that brain amyloid-β (Aβ) accumulation accelerates tau accumulation, leading to neurodegeneration and cognitive impairment [3, 4]. Advances in AD biomarkers, particularly biomarkers of Aβ, tau, and neurodegeneration, have enabled their implementation to study the course of AD in observational studies and clinical trials, even among asymptomatic individuals and in non-clinical, community-based samples [5–7]. The 2018 National Institute on Aging (NIA)/Alzheimer’s Association (AA) research framework recommends conducting research using a biological definition of AD, focusing on biomarkers of Aβ, tau, and neurodegeneration, such as imaging biomarkers of these constructs, independent of cognitive measures [4]. In this framework, AD pathologic change is defined by the presence of Aβ, while the presence of tau or neurodegeneration in the absence of Aβ defines non-AD pathologic change. This approach is being implemented widely but is controversial, as these biomarkers may not consistently predict who becomes cognitively impaired [8]. Our objective was to examine the cross-sectional association of brain imaging biomarkers of Aβ and neurodegeneration with verbal learning, the cognitive ability affected earliest in AD [9], in a community based, Hispanic cohort of individuals who are mostly in the seventh decade of life, a critical period for the accumulation of Aβ and the AD cascade [7, 11]. This cohort has complete information on two of the constructs to the NIA/AA research framework, amyloid and neurodegeneration. We hypothesized that higher Aβ and neurodegeneration, individually and in aggregate would be related to lower verbal learning performance, the most prominent early manifestation of AD.
MATERIALS AND METHODS
Participants
This was a cross-sectional analysis of a cohort of 350 Hispanic middle-aged men and women in New York City participating in a study of AD biomarkers, recruited between March 1, 2016 and July 31, 2019 [12]. We targeted Hispanics because they are the most common ethnic group in the community surrounding Columbia University Irving Medical Center (CUIMC) [13] and because there is a paucity of AD biomarkers studies in Non-Whites [10]. Inclusion criteria included age between 55–69 years, willing and able to undergo phlebotomy, clinical and neuropsychological assessments, 3T brain magnetic resonance imaging (MRI), and positron emission tomography (PET) with injection of 18F-Florbetaben. Exclusion criteria included: diagnosis of dementia, cancer other than non-melanoma skin cancer, and MRI contraindications. We screened 659 participants; 114 (17.30%) declined participation, 178 (27.01%) were ineligible, 16 (2.43%) were eligible but did not complete study procedures, and 1 (0.15%) had incomplete APOE genotyping (see Supplementary Figure 1). The interval between Aβ PET and MRI was 15.79 ± 33.41 days. This study was approved by the Institutional Review Board and the Joint Radiation Safety Commission at CUIMC. All study participants provided written informed consent.
Independent variables
The main independent variables were brain Aβ burden (A) ascertained as global brain Aβ standardized uptake value ratio (SUVR) measured with 18F-Florbetaben PET, and neurodegeneration (N), measured as cortical thickness in areas affected by AD [9] obtained from 3T brain MRI. These A/N measures were examined using three approaches: continuously, categorically (high versus low), and in aggregate using the A/N profiles and profile categories adapted from the NIA/AA research framework [4] excluding tau. There are 4 possible A/N profiles (A-N-, A+N-, A-N+, A+ N+). These A/N profiles are further categorized into 3 profile categories: Normal A/N biomarkers (A-N-); “Alzheimer’s continuum” (any A+ combination); Non-AD pathologic change (A-N+). There are currently no gold standard approaches for defining high and low levels of these biomarkers [14]. We defined Aβ positivity using the K-means clustering method, which is a data fitting approach that identifies the partition between the 2 peaks in the Aβ SUVR distribution [15, 16]. The resulting Aβ SUVR threshold was 1.34. Neurodegeneration was defined with a cutoff at 2 standard deviations (SD) below the mean value of cortical thickness in AD signature regions [17], resulting in a threshold of 2.48 mm. As a sensitivity analysis we dichotomized measures of Aβ and neurodegeneration by a median split, which is an approach previously used by other studies, including the Atherosclerosis Risk in Communities (ARIC) Study [18]. For Aβ, the median Aβ SUVR cutoff was 1.13 and for neurodegeneration, the median cortical thickness cutoff was 2.69 mm.
Aβ PET
The dose of 18F-Florbetaben was 300 MBq (8.1mCi), maximum 30 mcg mass dose, administered as a single slow intravenous bolus. Images were acquired over 20 min starting 90 min after injection. Dynamic PET frames (4 scans) were aligned to the first frame using rigid-body registration and a static PET image was obtained by averaging the four registered frames. Freesurfer v6.0 (http://surfer.nmr.mgh.harvard.edu/) was used to define regions of interest (ROIs), as described previously [19]. The standardized uptake value (SUV), defined as the decay-corrected brain radioactivity concentration normalized for injected dose and body weight, was calculated in all ROIs and normalized to that of cerebellar gray matter to derive regional standardized uptake value ratios (SUVRs). Parametric SUVR images were created for voxel-wise analysis by dividing each voxel by the mean activity in the cerebellar gray matter. The primary Aβ variable of interest was global SUVR, calculated as the weighted average from lateral temporal cortex, parietal cortex, cingulate cortex, and frontal cortex.
MRI methods
The primary measure of neurodegeneration was cortical thickness obtained by averaging values from AD-related regions as specific patterns of cortical thinning are found in AD [9] derived with FreeSurfer. These regions included entorhinal cortex, parahippocampus, inferior parietal lobule, pars opercularis, pars orbitalis, pars triangularis, inferior temporal pole, supramarginal gyrus, superior parietal lobe, and superior frontal lobe.
Dependent variables
The primary dependent variable was a test of verbal learning. Verbal learning was ascertained using total recall and delayed recall in the Buschke Selective Reminding test (SRT) [20]. The SRT is a standard tool in the assessment of verbal memory and has been used as a sensitive longitudinal measure of changes in memory function. Several studies have reported on the predictive value of the SRT for dementia [21–23]. A validated Spanish version of the SRT was used for Spanish-speaking participants in this study [24]. Scores reflect words recalled, with higher scores indicating better verbal learning performance.
Cognitive classification
There were no participants with dementia in our sample. We classified participants into amnestic mild cognitive impairment (aMCI) and cognitively unimpaired depending on their performance on SRT delayed recall using norms adjusted for age, sex, education, and language from the community of Northern Manhattan [25]. Participants were classified as aMCI if their performance in the SRT delayed recall was worse than 1.5 standard deviations following these norms.
Covariates
Characteristics considered as potential covariates included age, sex, education in years, Hispanic subgroup, language of administration and APOE ɛ4 genotype. Hispanic subgroup was classified following the format of the 2010 Census by country or region of origin (e.g., Mexican, Puerto Rican, Cuban, Dominican) [26]. Participants were classified as APOE ɛ4 carriers if they were homozygous or heterozygous for APOE ɛ4. APOE ɛ4 genotyping was conducted by LGC genomics (Beverly, MA) using single nucleotide polymorphisms rs429358 and rs7412.
Statistical analyses
Global Aβ SUVR was not normally distributed. Global Aβ SUVR had a bimodal distribution as expected [27], and no transformation approximated a normal distribution. Bivariate comparisons across A/N categories were made using analysis of variance (ANOVA) for continuous variables. Differences in categorical variables were evaluated using chi-squared tests. The relationship between global brain Aβ SUVR, cortical thickness, and SRT scores were evaluated using multivariable linear regression. The relationship of A/N profiles and categories with SRT scores was examined as pairwise comparisons using multivariable linear regression models with the normal A/N biomarkers profile (A-/N-) as the reference. Model 1 was unadjusted, model 2 was adjusted for age, sex, education and language of administration. Statistical significance was considered at p < 0.05. All statistical analyses were performed using the Statistical Package for the Social Sciences (SPSS) 26 (SPSS, Chicago, IL).
RESULTS
Table 1 shows the characteristics of the total sample and across categories of Aβ, and cortical thickness. The mean age was 64.15±3.34; 72.0%were women, 85.7 %were Hispanic of Dominican origin; 35.40%were APOE ɛ4 carriers and 20.57%were classified as having aMCI. There were no significant differences in age, education, ethnicity, language of administration, cortical thickness, and the distribution cognitive categories (intact, aMCI) between Aβ categories. Those classified as in the high Aβ category were significantly more likely to be APOE ɛ4 carriers. There were no significant differences in age, education, ethnicity, APOE ɛ4 prevalence, language of administration, and Aβ SUVR between cortical thickness categories. There was a significantly higher proportion of women among the high cortical thickness category and higher prevalence of aMCI among the low cortical thickness category.
Characteristics of the entire sample and by categories (high, low) of global brain amyloid-β (SUVR) (determined by K-means clustering) and cortical thickness in Alzheimer’s disease signature regions (determined as 2 standard deviations of the mean)
SUVR, standardized uptake value ratio; SD, standard deviation: a There were 02 subjects with missing Selective Reminding Test data.
There were no differences in age, education, ethni-city, language of administration, and the distribution of aMCI among A/N profiles (Table 2). Participants in the A-N+ were less likely to be women. There was higher prevalence of APOE ɛ4 carriers among A+ N-profile and lower prevalence of aMCI among A-N- profile.
Characteristics of the entire sample and compared by amyloid (A) (determined by the k-means clustering) and neurodegeneration (N) (determined as 02 standard deviations of the mean) profiles
SUVR, standardized uptake value ratio; SD, standard deviation. a There were 02 subjects with missing Selective Reminding Test data. One subject only fell into the A+ N+ profile.
There were no differences in age, education, ethnicity, language of administration among A/N profile categories (Table 3). There was a higher prevalence of women and a higher proportion of APOE ɛ4 carriers in the “Alzheimer’s continuum” category. There was a lower prevalence of aMCI among normal A/N biomarkers category.
Characteristics of the entire sample by Amyloid/Neurodegeneration (A/N) profile category. Any combination with A+ is classified as in the Alzheimer’s continuum. A+ N- is classified as non-AD pathologic change. A-N- is classified as Normal A/N biomarkers
SUVR, standardized uptake value ratio; SD, standard deviation. aThere were 02 subjects with missing Selective Reminding Test data. Cutoffs were determined using the k-means for global brain Aβ SUVR and 02 standard deviations of the mean cortical thickness in AD signatures, for neurodegeneration.
Table 4 displays the bivariate correlations between age, education, SRT total recall, SRT delayed recall, Aβ SUVR, and cortical thickness. Age was negatively correlated with cortical thickness, SRT total recall, and SRT delayed recall. Although, these correlations were relatively weak. Education was moderately correlated with SRT total recall and SRT delayed recall. Cortical thickness was positively correlated with SRT total recall and SRT delayed recall. As expected, SRT total recall was highly correlated with SRT delayed recall. Aβ SUVR was not correlated to SRT scores.
Correlation matrix for age, education, global brain amyloid-β (Aβ) standardized uptake value ratio (SUVR), cortical thickness in Alzheimer’s disease signature regions, performance in Selective Reminding Test (SRT), total recall and delayed recall
*Significance level < 0.05; **Significance level < 0.01.
In multivariate analyses relating A/N biomarkers examined continuously with SRT scores (Table 5), we found that higher cortical thickness was significantly associated with higher performance in SRT delayed recall (β= 2.57; 95%CI = 0.37 to 4.78). Global Aβ SUVR was not associated with any SRT scores.
Coefficients, 95%confidence intervals (CI), and p values from linear regression models relating global brain amyloid-β standardized uptake value ratio (SUVR) and cortical thickness in Alzheimer’s disease signature regions to performance in Selective Reminding Test (SRT) total recall and delayed recall. Model 1 is unadjusted. Model 2 adjusted for age, sex, education, and language of administration
In analyses relating A/N categories (high/low) with SRT scores (Table 6), we found that low cortical thickness was significantly related to lower performance in SRT delayed recall (β= –1.25; 95%CI = –2.40, –0.10) as compared to high cortical thickness. No significant associations were found between Aβ categories and any of the SRT scores.
Coefficients, 95%confidence intervals (CI), and p values from linear regression models relating performance in Selective Reminding Test (SRT) total recall and delayed recall to categories of global brain amyloid-β standardized uptake value ratio (SUVR; determined by K-means clustering) and cortical thickness in Alzheimer’s disease signature regions (determined using 2 standard deviations below the mean cortical thickness). The low category is the reference for amyloid and the high category is the reference for cortical thickness. Model 1 is unadjusted. Model 2 adjusted for age, sex, education, and language of administration
Finally, Non-AD pathologic change (A-N+profile) was associated with lower performance in SRT delayed recall compared to the normal A/N biomarkers (A-N- profile) (β= –1.59; 95%CI = –2.76, –2.42) (Tables 7 8). No differences in SRT scores between the normal A/N biomarkers (A-N- profile) and the Alzheimer’s continuum (Table 8) were found. Only one participant was in the (A+ N+) profile.
Coefficients and p values from pairwise comparisons using linear regression models relating Amyloid/ Neurodegeneration (A/N) biomarker profiles to performance in the Selective Reminding Test (SRT) total recall and delayed recall. The A-N- biomarker profile is the reference for all analyses. Model 1 is unadjusted. Model 2 adjusted for age, sex, education, and language of administration. Cutoffs were determined using the k-means for global brain Aβ SUVR and 02 standard deviations below the mean cortical thickness in AD signatures, for neurodegeneration
Coefficients, 95%confidence intervals (CI), and p values from pairwise comparisons using linear regression models relating Amyloid/Neurodegeneration (A/N) profiles categories to performance in Selective Reminding Test (SRT) total recall and delayed recall. The normal A/N profile category was the reference in all analyses. Sample sizes were: Normal A/N biomarkers, n = 310; Alzheimer’s continuum, n = 28; Non-AD pathologic change, n = 12. Cutoffs were determined using the k-means for global brain Aβ SUVR and 2 standard deviations below the mean cortical thickness in AD signatures, for neurodegeneration
Sensitivity analyses
Results for sensitivity analyses defining cutoffs of Aβ and neurodegeneration by a median split are presented in Supplementary Tables 1 to 6. Aβ categories were not related to any of the SRT scores while low cortical thickness was related to worse performance in the SRT total recall (β= –1.94; 95%CI = –3.63, –0.24) and SRT delayed recall (β= –0.50; 95%CI = –0.94, –0.07) as compared to high cortical thickness.
DISCUSSION
Contrary to our hypothesis, only the measure of neurodegeneration and the category of non-AD pa-thological change were related to lower verbal learning performance. Aβ burden and the Alzheimer’s continuum were not related to verbal learning performance. These findings were consistent when examining biomarkers continuously and as categories.
Our results support the hypothesis that measures of neurodegeneration and non-AD pathological change, but not Aβ and Alzheimer’s continuum, correlate with verbal learning [28, 29], at least in late middle age. Consistent with our results, a recent cross-sectional study of 106 participants (mean age = 74.4) from the Swedish Biomarkers for Identifying Neurodegenerative Disorders Early and Reliably (Bio-FINDER) cohort [29] reported that lower cortical thickness was associated with lower cognition scores in prodromal AD and AD, but the effects of Aβ on cognition were weaker and less specific [29].
Neurodegeneration measured as cortical thickness in AD signature regions, as measured in this study, have been linked to poor memory performance [9, 17] and have been shown to predict progression to AD in cognitively normal (CN) individuals [30]. Other brain structures, such as hippocampal volume, have been also used as a surrogate of neurodegeneration in other studies and yielded similar results of inverse associations with memory performance [31–33].
Aβ on the other hand has been shown to correlate poorly with verbal learning in cognitively unimpaired subjects [34–36]. The reported association of Aβ and memory has been mostly observed in older adults or adults with mild cognitive impairment (MCI) or dementia [7, 37]. Younger subjects with parental history of dementia also showed a decline in memory with higher rates of Aβ [11]. Contrary to our findings, a recent study from the BioFINDER cohort of 300 cognitively unimpaired subjects found that memory was independently associated with Aβ in relatively younger participants (65.2–73.2 years), while atrophy was associated with memory in older subjects (73.3–88.4 years). This study, as compared with ours, used CSF levels of Aβ (Aβ42/40) as a measure of Aβ deposition while hippocampal volume was used as a measure of neurodegeneration. Neurodegeneration, measured as T-tau in CSF, was not associated with memory in the same study [38].
The finding that a marker of neurodegeneration, but not amyloid, was related to verbal learning in this relatively younger cohort merits further discussion. It is possible that non-AD pathological change is already present in late middle age before amyloid has had an effect on neurodegeneration, as would be expected following the amyloid hypothesis [3, 4]. However, it is also possible that cortical thickness in this stage of the lifespan is a measure of brain reserve and intelligence [39], and that the relation between cortical thickness and verbal learning in our study reflects pre-morbid brain structure and intelligence rather than neurodegeneration. Longitudinal follow-up of our cohort may allow us to disentangle these possibilities.
The main strength of our study is the availability of state-of-the-art biomarkers of Aβ and neurodegeneration in a community-based sample of middle-aged Hispanics, in which there is a paucity of information on in vivo AD neuropathology. In addition, we focus on late-middle age, a critical period for the accumulation of Aβ and the AD cascade. The main limitation of our study is its cross-sectional nature, which limits causal inferences. Lastly, we did not report on tau, another important hallmark of AD along with Aβ and neurodegeneration [4]. We are currently implementing tau PET in the same participants and will report results once we achieve a sample size comparable to the present study. Finally, another limitation is that we did not have a formal adjudication process for cognitive diagnoses, and we used an algorithm to diagnose aMCI retrospectively based on previously used norms for our measure of verbal learning.
In conclusion, in late-middle aged adults, neurodegeneration and non-AD-pathologic change, but not Aβ or the Alzheimer’s continuum is associated with worse verbal learning memory performance.
Footnotes
ACKNOWLEDGMENTS
Support for the reported work was provided by United States National Institutes of Health grants R01AG050440, RF1AG051556, and RF1AG051556-01S2. Partial support was also provided by grants K24AG045334, P30AG059303, and ULT1TR001873. P. Palta is supported by grant R00AG052380 from the National Institutes of Health.
