Background
The Alzheimer’s Disease Neuroimaging Initiative includes cognitive assessments at every study visit. We have had the opportunity to use modern psychometric approaches including ...confirmatory factor analysis (CFA) and item response theory (IRT) with these rich data. Here we discuss lessons learned from analyses of memory, executive functioning, language, and visuospatial functioning in ADNI.
Method
We analyzed data from all ADNI study waves. We considered granular data for each test. Our panel of experts considered each granular data element and assigned them to a single primary domain – memory, executive functioning, language, or visuospatial functioning – or to none of these. We then used CFA and IRT approaches to calibrate each domain separately. ADNI has extensive imaging and fluid biomarker data we used for validity assessments.
Result
The ADNI cognitive battery emphasizes memory assessment with multiple indicators. The assessment of language and executive functioning are less robust but are each measured with several indicators. The assessment of visuospatial functioning is much sparser. Assessment intensity is nicely reflected in measurement precision data, where standard errors of measurement are smallest for memory, intermediate for language and executive functioning, and largest for visuospatial functioning. Validity assessments were solid for all four domains. We have uploaded resulting composite scores (ADNI‐Mem, ADNI‐EF, ADNI‐Lan, and ADNI‐VS) to the Laboratory On Neuroimaging (LONI)‐hosted ADNI website.
Conclusion
The composite scores we have developed will be useful for investigators integrating domain‐specific cognitive data in their analyses of ADNI data. The measurement precision findings have implications for future study design; for example, more robust measurement of visuospatial functioning may be a wise investment. Measurement precision findings also have implications for analytic strategies. Standard typical analyses ignore measurement precision and treat observed scores as if they were measured without error. We are developing hierarchical Bayesian modeling strategies that can incorporate both point estimates for domain scores along with standard errors of measurement. These models will account for measurement error, ensuring that inferences are valid across domains with very different measurement properties. We will make this code widely available as well.
Abstract
Background
More than 75 common variant loci account for only a portion of the heritability for Alzheimer’s disease (AD). A more complete understanding of the genetic basis of AD can be ...deduced by exploring associations with AD‐related endophenotypes.
Method
We conducted genome‐wide scans for cognitive domain performance using harmonized and co‐calibrated scores derived by confirmatory factor analyses for executive function, language, and memory. We analyzed 103,796 longitudinal observations from 23,066 members of community‐based (FHS, ACT, ROSMAP) and clinic‐based (ADRCs, ADNI) cohorts using generalized linear mixed models including terms for SNP, age, SNP×age interaction, sex, education, and five ancestry principal components. Significance was determined based on a joint test of the SNP’s main effect and interaction with age. Results across datasets were combined using inverse‐variance meta‐analysis. Genome‐wide tests of pleiotropy for each domain pair as the outcome were performed using PLACO software.
Result
Individual domain and pleiotropy analyses revealed genome‐wide significant (GWS) associations with five established loci for AD and AD‐related disorders (BIN1, CR1, GRN, MS4A6A, APOE) and eight novel loci. ULK2 was associated with executive function in the community‐based cohorts (rs157405, P = 2.19×10
−9
). GWS associations for language were identified with CDK14 in the clinic‐based cohorts (rs705353, P = 1.73×10
−8
) and LINC02712 in the total sample (rs145012974, P = 3.66×10
−8
). GRN (rs5848, P = 4.21×10
−8
) and PURG (rs117523305, P = 1.73×10
−8
) were associated with memory in the total and community‐based cohorts, respectively. GWS pleiotropy was observed for language and memory with LOC107984373 (rs73005629, P = 3.12×10
−8
) in the clinic‐based cohorts, and with NCALD (rs56162098, P = 1.23×10
−9
) and PTPRD (rs145989094, P = 8.34×10
−9
) in the community‐based cohorts. GWS pleiotropy was also found for executive function and memory with OSGIN1 (rs12447050, P = 4.09×10
−8
) and PTPRD (rs145989094, P = 3.85×10
−8
) in the community‐based cohorts. Functional studies have previously linked AD to ULK2, NCALD, and PTPRD.
Conclusion
Our results provide some insight into genes associated with domain‐specific cognitive impairment and AD, as well as a conduit toward a syndrome‐specific precision medicine approach to AD. Increasing the size of datasets by applying the confirmatory factor analysis approach to derive harmonized measures of cognitive performance would likely enhance the discovery of additional genetic factors of cognitive decline leading to AD and related dementias.
Objective: To demonstrate measurement precision of cognitive domains in the Alzheimer's Disease Neuroimaging Initiative (ADNI) data set. Method: Participants with normal cognition (NC), mild ...cognitive impairment (MCI), and Alzheimer's disease (AD) were included from all ADNI waves. We used data from each person's last study visit to calibrate scores for memory, executive function, language, and visuospatial functioning. We extracted item information functions for each domain and used these to calculate standard errors of measurement. We derived scores for each domain for each diagnostic group and plotted standard errors of measurement for the observed range of scores. Results: Across all waves, there were 961 people with NC, 825 people with MCI, and 694 people with AD at their most recent study visit (data pulled February 25, 2019). Across ADNI's battery there were 34 memory items, 18 executive function items, 20 language items, and seven visuospatial items. Scores for each domain were highest on average for people with NC, intermediate for people with MCI, and lowest for people with AD, with most scores across all groups in the range of −1 to +1. Standard error of measurement in the range from −1 to +1 was highest for memory, intermediate for language and executive functioning, and lowest for visuospatial. Conclusion: Modern psychometric approaches provide tools to help understand measurement precision of the scales used in studies. In ADNI, there are important differences in measurement precision across cognitive domains.
Key Points
Question: How do ADNI's cognitive domains compare in terms of measurement precision? Findings: Memory is characterized by better measurement precision, and visuospatial by worse measurement precision, with intermediate values for language and executive function, in the range where scores were observed in ADNI. Importance: Measurement properties such as measurement precision may be useful in interpreting findings from ADNI, and may be useful in management of burden/precision trade-offs for researchers designing cognitive assessment approaches. Next Steps: Familiarity with measurement precision issues and metrics may be useful in understanding data from existing studies and in designing cognitive evaluation strategies for future studies.
Background
Modeling the dynamics of Alzheimer’s disease (AD) biomarkers over the entire continuum of AD progression is important, yet challenging due to limited resources to collect longitudinal ...biomarkers from the aging population with fully observed clinical spectrum of AD. This study proposed and applied a synchronized sigmoidal mixed‐effects model to characterize dynamics of longitudinal memory performance using data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). The model leveraged time to AD onset as the time scale and additionally allowed inclusion of participants without AD onset, drastically expanding future applications.
Method
ADNI participants with observed mild cognitive impairment (MCI) and/or AD onset (n = 312, mean (SD) baseline age 74.9 (6.44) years) were included (Table 1). A memory composite previously built in ADNI was leveraged for all analyses. A synchronized sigmoidal mixed‐effects model was constructed for dynamics of memory performance with parameters for initial memory level, magnitude of decline, and half‐life of decline. For participants with observed MCI but not AD onset, an additional parameter (t0
) quantifying the time from MCI onset to AD was incorporated (Figure 1). We considered random effects for all parameters and allowed t0
to vary by age at MCI onset (nonlinearly), sex, apolipoprotein E (APOE)‐ε4 status and their interactions.
Result
The mean initial harmonized memory score is 0.24 (95% CI: 0.17‐0.32). The mean decline in the harmonized memory score is 1.53 (95% CI: 1.43‐1.64). The mean time when the harmonized memory score declined by half is 0.57 years before AD onset (95% CI: 0.32‐0.82). Female is associated with faster progression from MCI onset to AD (p = 0.002). Age at MCI onset is nonlinearly associated with MCI‐to‐AD progression (p < 0.001) and APOE‐ε4 status interacts with age at MCI onset on MCI‐to‐AD progression (p = 0.002) (Figure 2).
Conclusion
The proposed synchronized sigmoidal mixed effect model can be used to characterize dynamics of AD biomarkers relative to AD onset using participants with and without AD onset. A model to estimate duration of MCI‐to‐AD progression can be simultaneously included for synchronization purpose, which identified gender, age at MCI onset and APOE‐ε4 status as factors associated with MCI‐to‐AD progression.
Background
The Australian Imaging, Biomarkers and Lifestyle (AIBL) Study is a prospective study collecting extensive cognitive, clinical, fluid, and imaging biomarkers data from older adults living ...in Australia. Integration of outcomes between large prospective studies of AD will provide greater precision in models of AD brain‐behavior relationships, so it is important to align composite scores for cognitive domains between such studies.
Methods
Detailed methods for AIBL, the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and the National Alzheimer’s Coordinating Center (NACC) have been published. Briefly, AIBL participants had cognition assessed with an extensive neuropsychological test battery alongside health and biomarker assessments at entry and each 18‐months thereafter. Granular‐level cognitive data were obtained and an expert panel of two neuropsychologists and a behavioral neurologist categorized each element as assessing memory, executive functioning, language, visuospatial, or none of these, exactly as we have done previously. We also identified elements we had previously calibrated from other studies; after careful quality control and confirmation these served as anchors enabling co‐calibration. We used confirmatory factor analysis bi‐factor models to calibrate the AIBL battery with other studies. We used those calibrations to obtain co‐calibrated scores for all AIBL participants at every study visit. Here we show descriptive statistics for baseline visits, separately by diagnosis (normal cognition, mild cognitive impairment (MCI), dementia) for two enrollment waves for AIBL as well as for each phase of ADNI and across the Uniform Data Set (UDS) 1 & 2 (UDS1/2) and UDS3 time periods for NACC.
Results
Box plots for memory, executive functioning, language, and visuospatial for people with normal cognition are in Figure 1, MCI in Figure 2, and dementia in Figure 3. These figures show there is substantial cognitive variation across waves within these disease stage groups and across studies.
Conclusion
Co‐calibrated neuropsychological domain scores provide a common metric for integrating cognitive data across studies. Co‐calibrated scores aggregated across large prospective AD studies such as AIBL, ADNI, and NACC provide a foundation for large‐scale models of the development of AD and can serve as phenotypes for genetics studies. Co‐calibrated scores are available from AIBL, ADNI, and from NACC.
Abstract
Background
A recently recognized subset of older individuals are an anomaly of cognitive decline; the “SuperAgers”, unsurprisingly named, achieve cognitive scores equivalent to much younger ...cognitively normal (CN) middle‐aged adults. Using longitudinal cognitive data harmonized across eight cohorts of aging and Alzheimer’s Dementia (AD), we investigated the genetic drivers of SuperAging.
Method
Harmonized memory, executive function, and language scores were estimated leveraging latent variable modeling and made available through the ADSP Phenotype Harmonization Consortium. SuperAgers (N = 1,095) were defined as individuals over 80 years with a mean sex‐adjusted memory score equal or exceeding CN individuals aged 50‐60, score within one age and sex‐adjusted standard deviation in the other two cognitive domains, and remain CN for all longitudinal visits. Young Cases (N = 1,906) were defined as individuals aged 50‐75 with a clinical diagnosis of AD. Old Controls (N = 3,247) were defined as CN individuals over 80, scoring within one age‐ and sex‐adjusted standard deviation in all three domains. We performed a GWAS on non‐Hispanic Whites using logistic regression comparing SuperAgers and their counterparts (Young Cases and Old Controls) with covaried adjustment for age, sex, education, and principal components for population substructure.
Result
Comparing SuperAgers with Young Cases (Figure 1), only variants in the well‐established APOE region were associated with genome‐wide significance (GWAS; P<510
−8
). Additionally, we observed a locus on chromosome 13 approach GWS (rs138699163, P = 6.5610
−8
). The locus centered on a relatively uncharacterized ncRNA, MIR4500. Analyses comparing SuperAgers to Old Controls did not find any GWS associations, with the strongest association observed at rs116535931 on chromosome 5 (P = 1.5210
−6
).
Conclusion
Our extreme‐phenotype GWAS comparing SuperAgers to Young Cases identified established and novel loci for AD. However, larger sample sizes may allow better characterization of the genetic architecture of SuperAging. Future analyses will extend to Case comparison groups to include Old Cases (age>80 years) and All Cases (age>50 years) and Control comparison groups to include Young Controls (age between 50‐60) and Agnostic Controls (age>50 years) with similar criteria.
Abstract
Background
Genetic analyses of cognitive endophenotypes have led to discoveries of novel loci contributing to Alzheimer’s disease (AD) risk. Sex differences are present in cognitive ...trajectories in aging and AD, and these may vary across cognitive domain. However, genetic drivers that may contribute to sex differences in cognitive trajectories have yet to be explored. Thus, we sought to investigate the sex‐specific genetic architecture of cognition.
Method
We leveraged 10 cohorts of cognitive aging and AD to complete this sex‐aware genetic study (N = 31,800; mean age = 73 yrs.; 55% female). Harmonized cognitive scores for memory, executive functioning, and language were derived using confirmatory factor analysis models. We calculated change in cognitive scores over time using a mixed effects model, to facilitate analysis on cognitive decline. We performed GWAS of baseline score and of estimated rate of decline in each domain and in each cohort separately among non‐Hispanic white (NHW) individuals, adjusting for baseline age and genetic principal components. Then we meta‐analyzed the results.
Result
In addition to the well‐characterized
APOE
locus, we identified a genome‐wide significant chromosome 2 locus that was associated with language decline among NHW women (rs13387871: MAF = 0.20; β
women
= 2.97×10
−3
; P
women
= 2.65×10
−9
), but not among male counterparts (β
men
= ‐3.14×10
−4
, P
men
= 0.60). This locus contains multiple eQTLs for
VRK2
, a serine/threonine kinase that has been previously linked to neuropsychiatric disorders, including schizophrenia. Furthermore, the top variant in this locus (rs13387871) was nominally significant for memory decline (β
women
= 1.74×10
−3
, P
women
= 0.01) and for executive functioning decline (β
women
= 7.58×10
−4
, P
women
= 0.02) in meta‐analyses among NHW women.
Conclusion
Our genetic analysis suggests that there may be some genetic drivers of language performance that differ by sex, and that these drivers may be shared to an extent across domains. Our future sex‐aware meta‐analyses will also include 1) non‐Hispanic black (NHB) within ancestry (N = 4,200), 2) cross‐ancestry (NHW + NHB), 3) diagnosis‐stratified, and 4) analysis of X‐chromosome. Planned follow‐up analyses will include gene‐set analyses, heritability tests, and genetic correlation tests. Through our preliminary analysis, we identified a promising locus for further exploration, and this is the first of many steps in elucidating the sex‐specific genetic architecture of cognition across ancestry groups.
Abstract
Background
Memory performance can serve as a strong endophenotype for Alzheimer’s disease (AD) that changes early and continues to decline with disease progression. Yet, the genetic ...architecture of memory is not well characterized in older adults. Here, we build on existing memory GWAS studies by performing predicted gene expression analysis (PrediXcan) among older (60+) individuals from four cohorts of aging and investigate specific gene‐tissue drivers of genetically regulated gene expression associated with memory performance.
Method
Tissue‐specific (49 tissues, 5455 genes) PrediXcan models were built following the method described in Gamazon et al. (Nature Genetics 2015) leveraging model weights derived in GTEx (v8 release, build 38). Baseline and longitudinal memory scores were harmonized leveraging cognitive item‐level data on 19,707 non‐Hispanic White participants from four cohort studies of aging and AD (mean age 75.6±7.7, 55% female) using confirmatory factor analyses models. PrediXcan analyses were run adjusting for age at baseline, sex, and 5 population PCs and then meta‐analyzed using a fixed effects model. Correction for multiple comparisons accounting for all gene‐tissue combinations (267,267) was completed with the false discovery rate procedure (fdr‐p<0.05). Sensitivity analyses excluded all non‐AD dementia and other comorbid conditions (N = 16,373; 57% females).
Result
As expected, several signals emerged from chromosome 19, including 48 gene‐tissue combinations near the APOE locus. Outside of APOE, we identified 20 gene‐tissue combinations from 17 known AD loci and three novel loci: higher predicted PUS7 expression in the caudate (β = ‐0.018, p = 0.034) and higher RP11‐18C3.1 expression in colon (β = ‐0.0165, fdr‐p = 0.034) related to faster cognitive decline, while higher predicted LRRC25 in the nucleus accumbens related to slower cognitive decline (β = 0.009, fdr‐p = 0.015). These signals remain comparable in sensitivity analysis.
Conclusion
We identified multiple candidates for future mechanistic analysis. LRRC25 is a particularly interesting candidate that is differentially expressed in the AD brain, regulates autophagy in myeloid cells, and is in a co‐expression network with other known AD genes in the immune pathway like MS4A4A (
https://agora.adknowledgeportal.org/
). Future work will test for replication of these effects and deconvolve genetically‐regulated versus measured gene expression effects in the AD brain.
Abstract
Background
Alzheimer’s disease (AD) is clinically characterized by disabling cognitive impairment, though substantial variability in cognitive symptoms and trajectories is observed in AD ...individuals. However, genetic predictors of domain‐specific cognitive performance remain undiscovered. We investigated cross‐sectional and longitudinal genetic architecture of harmonized memory, executive function, and language scores within and across ancestry groups.
Method
Using data from 7 cohorts of cognitive aging and AD, individuals >60 years at baseline were included (mean age at baseline = 71.2). Cognitive scores for memory, executive function, and language were harmonized using latent variable models. Slopes for cognitive scores were calculated for each domain with linear mixed‐effects models. GWAS was performed on each cognitive domain for individual cohorts, both at baseline and longitudinally. Models covaried for baseline age, sex, and the first three genetic principal components. Individual models were assessed among non‐Hispanic Whites (NHW) (N = 26,455), non‐Hispanic Blacks (NHB) (N = 3,410), and cross‐ancestry (NHW + NHB) (N = 29,865). Results were meta‐analyzed across cohorts.
Result
We identified six genetic loci showing a genome‐wide significant effect on cognition, in addition to well‐established associations between cognition and APOE: three loci in NHW, one locus in NHB, and two loci in cross‐ancestry results. In NHW, a chromosome 2 locus (rs6733839) near BIN1, a previously reported AD risk gene, was associated with longitudinal memory performance (MAF = 0.40, p = 3.36E‐08). Additionally, in NHW, two chromosome 2 loci (rs2940785 and rs2972059) were associated with memory decline (MAF = 0.05, = 3.92E‐09; MAF = 0.05, p = 5.06E‐09, respectively). Despite the small sample size, a chromosome 10 locus (rs77595416) was associated with longitudinal executive function in NHB (MAF = 0.01, p = 7.68E‐09). When analyzing cross‐ancestry results, two chromosome 2 loci near BIN1 (rs4663105 and rs6733839) were associated with memory decline (MAF = 0.44, p = 2.65E‐08; MAF = 0.40, p = 9.48E‐10, respectively).
Conclusion
We elucidate novel and replicate known genetic predictors of domain‐specific cognition in older adults. Furthermore, we show that genetic architecture of multiple cognitive domains in older adults differs by ancestry, highlighting SNPs observed in longitudinal memory (NHW and cross‐ancestry) and executive function (NHB). While replication is warranted, our results underscore the contribution of genetic predictors beyond APOE to cognitive decline and suggest the importance of ancestry‐specific analyses of cognition.
Abstract
Background
Apolipoprotein E4 (APOE4) is common in the population yet is the strongest genetic risk factor for late‐onset Alzheimer’s disease (AD). Here, we sought to identify genetic effects ...that differ by APOE4 genotype leveraging stratified and APOE interaction analyses. We hypothesized that we could identify novel genetic associations with longitudinal cognitive decline across three neuropsychological domains (memory, executive function, and language) that differ by APOE4 status. We leveraged a large, multi‐ancestry harmonized cognitive dataset (Nparticipants ∼ 23,000) from the AD Sequencing Project Phenotype Harmonization Consortium including more than 50,000 total longitudinal measures of cognition.
Method
This study included data from four cohort studies of aging and AD (NNHW = 20,117, NNHB = 2,631, Nobs = 22,748, 39% APOE4 carriers, 16% AD cases). Memory, executive function, and language composite scores were harmonized leveraging latent variable modeling. APOE4‐ stratified GWAS were performed on these phenotypes, controlled for age at baseline, sex, and genetic ancestry. APOE4 interaction models were leveraged to test for statistical differences based on markers identified in stratified discovery analyses. Post‐GWAS included gene tests with MAGMA and genetic correlation analyses with GNOVA.
Result
Among NHW, we identified an APOE4pos‐specific locus on chromosome 1 (rs7537669, βE4pos = ‐0.12, pE4pos = 2.2E‐08, βE4neg = ‐0.01, pE4neg = 0.17). This variant is with a strong eQTL for CD46, a regulatory element of the complement system. Among APOEneg NHW, we found TXNRD3 was associated with memory and a negative genetic correlation between amyotrophic lateral sclerosis (ALS) and memory performance that was not observed in APOE4pos individuals. Finally, we observed an APOE4neg association between CRELD and executive function among NHB individuals.
Conclusion
In the largest APOE4‐stratified GWAS of multi‐domain cognitive performance, we identified a number of novel genetic loci and genetic correlations that appear to act in an APOE4‐stratified manner. Given the known heterogeneity in clinical progression, age‐related risk, and response to therapeutics that has been reported, it will be important to disentangle molecular pathways that differ by APOE genotype to move towards precision interventions.