Alzheimer’s disease (AD) has both genetic and environmental risk factors. Gene-environment interaction may help explain some missing heritability. There is strong evidence for cigarette smoking as a risk factor for AD. To identify genetic-smoking-related associations with AD, we conducted a genome-wide association study (GWAS) assessing a SNP-smoking interaction and stratified analysis by smoking status. Lifetime smoking data were available and analyzed among 22,030 non-Hispanic White (NHW; 8,232 cases; 13,798 controls) and 3,126 African American (AFA; 921 cases; 2,205 controls) participants from the AD Genetic Consortium and the Framingham Heart Study. “Ever smoking” status was considered as a dichotomous exposure, defined by current smoking status or past history of smoking. Across 35 datasets, we conducted GWAS with two approaches: inclusion of a SNP-by-smoking interaction term and stratification by smoking status (12,080 smokers, 13,428 non-smokers). MAGEE was used to estimate SNP-by-smoking interaction effects and SAIGE was used to estimate SNP effects in stratified analysis. Age, sex, and principal components for population structure were included as covariates. METAL was used for inverse-variance weighted meta-analysis across datasets to estimate within- and cross-ancestry effects. The stratified analysis identified a genome-wide significant association among smokers in APAF1 on chromosome 12 (top SNP: rs12368451; smokers: MAF = 0.44, p = 2.2 × 10 -8 , OR = 1.20; non-smokers: MAF = 0.44, p = 0.97, OR = 1.00). Effects were present in both ancestry groups (NHW: MAF = 0.45, p = 6.1 × 10 -6 , OR = 1.16; AFA: MAF = 0.35, p = 6.6 × 10 -5 , OR = 1.46). APAF1 has been linked to gene-smoking interaction for non-AD related outcomes. A neighboring gene, ANKS1B , is highly expressed in the brain, interacts with amyloid-b precursor protein, and has shown GWAS signals for smoking initiation and cognitive ability. We also identified a genome-wide significant SNP-by-smoking interaction in the MIXL1/LIN9 region on chromosome 1 (top SNP: rs1091961, MAF = 0.35, p = 4.9 × 10 -8 , β SNP*smoking = 0.24; smokers: OR = 1.12, p = 0.0006; non-smokers: OR = 0.89, p = 0.0001). Within LIN9 , several SNPs in linkage disequilibrium with rs1091961 have shown sub-genome wide association with nicotine dependence. In this gene-smoking interaction and smoking-stratified GWAS of AD, we identified two promising loci. These findings highlight the strength of utilizing cross-ancestry datasets and considering both genetic and environmental factors together towards a personalized medicine approach to AD.

AbstractBackgroundGenome‐wide association studies (GWAS) have identified approximately 40 risk loci associated with Alzheimer’s disease (AD). However, only a fraction of heritability has been explained. In large meta‐analysis, smoking was associated with AD dementia. In this study, we sought to identify interactions between smoking and single‐nucleotide polymorphisms (SNPs) to further characterize AD genetic architecture.MethodWe used a subset of datasets of individuals of European ancestry from the Alzheimer’s Disease Genetic Consortium (ADGC) that had smoking information available from parent studies. The presence of any past or current smoking habit was considered to be positive smoking exposure. In each dataset, we ran a genome‐wide case‐control analysis including SNP and smoking main terms, and a SNP‐smoking interaction term. Models were adjusted for age, sex, and principal components of population substructure. Results from each dataset were meta‐analyzed using the inverse variance method. Pathway analysis of top SNPs was run using Ingenuity.ResultThe sample included 6,916 total individuals, including 2,862 AD cases and 4,054 controls. Although no interaction terms reached genome‐wide significance, there were two suggestive loci: top SNP rs3734416 [minor (C) allele frequency (MAF) = 0.27; interaction P = 7.2 × 10−6; OR for C allele in smokers = 1.23; OR for C allele in non‐smokers = 0.86] on chromosome 6 within the gene MTHFD1L and 95kb downstream of PLEKHG1, and top SNP rs693951 [MAF (G) = 0.27; interaction P = 1.3 × 10−6; OR for major A allele in smokers = 1.31; OR for A allele in non‐smokers = 0.83] on chromosome 8, 70kb upstream of ANGPT1. Pathway analysis implicated cellular development, cellular growth and proliferation and nervous system development and function. MTHFD1L has been implicated in candidate gene studies of AD and PLEKHG1 has been implicated in a GWAS of white matter hyperintensities. ANGPT1 has been implicated in recovery after ischemic stroke in candidate gene, transcriptomic and blood‐based biomarker studies.ConclusionWe found suggestive evidence of a smoking‐SNP interaction associated with AD risk at 2 loci previously linked with AD and cerebrovascular disease, but that were not identified in the largest AD GWAS to date. Expansion of the sample and replication are underway.

Background: There is considerable heterogeneity in clinical presentation among people with late-onset Alzheimer's disease (LOAD). We have categorized people with LOAD into subgroups based on relative impairments across cognitive domains. These 6 groups are people with no relatively impaired domains (AD-No Domains), 4 groups with one relatively impaired domain (AD-Memory, AD-Executive, AD-Language, and AD-Visuospatial), and a group with multiple relatively impaired domains (AD-Multiple Domains). Our previous analysis demonstrated that genetic factors vary across cognitively-defined LOAD groups. Objective: To determine whether risks associated with depression and traumatic brain injury with loss of consciousness (TBI) for cognitively defined LOAD subgroups are similar. Methods: We used cognitive data at LOAD diagnosis from three prospective cohort studies to determine cognitively-defined subgroups. We compared subgroups in endorsement of items from the Centers for Epidemiological Studies Depression (CES-D) scale and history of TBI. Results: Among 1,505 people with LOAD from the three studies, there were substantial differences across subgroups in total CES-D score, with lower scores (less depression) for people with AD with relative impairments in memory (AD-Memory) compared to those in other groups. Differences were noteworthy for the sleep-related item of the CES-D, as people with AD-Memory were less likely to report restless sleep than people in other groups. There were no differences in TBI history across groups. Conclusions: Differences in risk factor associations across subgroups such as differences in endorsement of depression symptoms and restless sleep provide support for the hypothesis that there are biologically coherent subgroups of AD.

This article was originally published under standard licence, but has now been made available under a [CC BY 4.0] license. The PDF and HTML versions of the paper have been modified accordingly.

We previously reported using cognitive profiles to define phenotypic subgroups among those with late-onset Alzheimer's dementia (LOAD) and found differences across subgroups in APOEgenotype, previously identified LOAD-related SNPs, and AD-related neuropathology, using one study's data. We used advanced psychometric methods applied to item-level cognitive data from people with LOAD from 5 studies to co-calibrate scores for memory, executive functioning, visuospatial performance, and language. We used our previously described framework to group people based on substantial (>0.80 SD) impairments compared to each person's average across domains. These procedures produce 6 groups: those with no domain with a substantial relative impairment; 4 groups (one per domain) with a single substantial relative impairment; and a group with multiple domains with substantial relative impairments. We compared prevalence of these six groups at the first visit with LOAD across studies, and evaluated associations with APOEgenotype and previously identified LOAD-related SNPs. We used regression approaches to define four continuous cognitively-defined LOAD endophenotypes: memory residuals corresponding to hippocampal impact, and residuals for the other three domains corresponding to regional cortical impact. We used PLINK to perform GWAS of each endophenotype. Substantial memory impairment was the most common single domain category at 16%, followed by executive. There was variation across studies in the pattern of domains with substantial impairment (Figure 1). In each study, the group with substantial relative memory impairment had >10% higher proportions with ≥1 APOE e4 allele compared to all people with LOAD (p<0.0001). LOAD-related SNPs replicated findings of substantial over-representation in particular subgroups (Table 1). GWAS of the four endophenotypes found genome-wide significant loci near APOE (rs429358, p=1.04x10-10) for the memory residual. Additional loci were suggestive for the executive residual (rs34877092 on chromosome 13 intergenic near STARD13, p=6.45x10-7), for the visuospatial residual (rs113310062 on chromosome 3 in EGFEM1P, p=1.70x10-7), and for the language residual (rs79360811 on chromosome 7 intergenic near ABCB5, p=2.32x10-6). These results from multiple studies show extensive replication of our single-study results and confirm that theory-driven cognitive phenotyping results in biologically coherent groups. Further research is warranted to elucidate the biological underpinnings of these subgroups. Prevalence of psychometrically defined subgroups overall and in each study. Summary of meta-analysis results for previously identified LOAD-related SNPs. Odds ratios (OR) are for each LOAD cognitive subgroup compared to controls, where bolding and size indicate statistical significance. Asterisks indicate heterogeneity p <0.05

In prior work we demonstrated genetic and neuropathology-based variation across cognitively defined Alzheimer's disease subgroups. Here we sought to determine whether patterns of vascular risk factors differed across cognitively defined subgroups. We used data from the Adult Changes in Thought (ACT) study. ACT is a prospective cohort study of Seattle-area Group Health members over age 65 and dementia free at enrollment. ACT follows people at 2-year intervals to identify incident dementia and AD. We used all available cognitive data at the time of AD diagnosis to determine scores for memory, visuospatial abilities, language, and executive functioning. We used these scores to determine each individual's average cognition and then domain-specific deficits below that individual average. We used these data to define six groups: no prominent domain, memory-, visuospatial-, language-, or executive functioning-prominent, and multiple domains (excluded here). We evaluated risk factors for these six groups from self-reported medical conditions including diabetes, stroke, and hypertension, and diagnosed atrial fibrillation or coronary artery disease. We used multinomial logistic regression models with the no prominent domain group as the reference. To account for multiple comparisons, we present tests of the null that each risk factor is unrelated to each subgroup, and an omnibus test of the association between each risk factor and any subgroup. Characteristics of participants in each AD subgroup are shown in Table 1. Of 825 cases with data for all four domains, nearly half had no prominent domain, a few had multiple prominent domains, and a single domain was prominent for the remainder. Memory-prominent domain was more common in people with diabetes (RR=1.97, 95% confidence interval [CI] 1.05, 3.72, p=0.04) and executive functioning prominent was less common (RR=0.47, 95% CI 0.16, 1.44, p=0.19), although the omnibus test did not reach statistical significance (p=0.08, Table 2). Despite low statistical power due to small group sizes, our results suggest possible differences in risk associated with diabetes. Subsequent work will refine these investigations using medications and evaluating glucose levels over time. These results support additional efforts to further understand cognitively-defined AD subgroups.

Preliminary work has revealed genetic heterogeneity among Alzheimer's disease (AD) subtypes defined by cognitive domain-specific impairments. There are currently no reports of risk factor associations for these AD subgroups. We used cognitive data from 825 Adult Changes in Thought (ACT) participants at the time they were identified with probable or possible AD to generate scores for memory, executive functioning/attention, language, and visuospatial ability. We determined individual mean scores across all domains, and identified specific impairments as >0.75 SD below each individual's mean domain score. Depressive symptom severity was measured with the CES-D, and study staff collected self-reported data on traumatic brain injury (TBI) exposure. We used multinomial logistic regression with the “no prominent domain group” designated as the reference category. We determined risk ratios for depression and TBI for each subgroup vs. the no prominent domain subgroup, and tested significance of any heterogeneity with an omnibus test. We controlled for age, sex, APOE genotype, and years of education. Characteristics of participants in each AD subgroup are shown in Table 1. Of 825 cases with data for all four domains, nearly half had no prominent domain, a few had multiple prominent domains, and a single domain was prominent for the remainder. Groups showed differences in APOE genotype, consistent with our previous reports. Risk factor results are shown in Table 2. Education's role as a risk factor varied across the five cognitively-defined AD subgroups (omnibus p=0.04). Education predicted development of memory prominent AD rather than AD with no prominent domain (risk ratio [RR] = 1.08 per year of education, 95% confidence interval (CI) 1.00, 1.16, p=0.05; omnibus p=0.04). Depression also varied across AD subgroups (omnibus p = 0.04). Higher CESD scores were associated with lower risk of developing memory prominent AD compared to no prominent domain AD (RR = 0.93 per point on the CESD, 95% CI 0.88, 0.98, p=0.01). We did not find differences in risk associated with TBI. Cognitively-defined AD subtypes show heterogeneity in their associations with depressive symptoms and education. It is possible that these risk factors are associated with biological differences across subgroups.

INTRODUCTION:There may be biologically relevant heterogeneity within typical late-onset Alzheimer's dementia. METHODS:We analyzed cognitive data from people with incident late-onset Alzheimer's dementia from a prospective cohort study. We determined individual averages across memory, visuospatial functioning, language, and executive functioning. We identified domains with substantial impairments relative to that average. We compared demographic, neuropathology, and genetic findings across groups defined by relative impairments. RESULTS:During 32,286 person-years of follow-up, 869 people developed Alzheimer's dementia. There were 393 (48%) with no domain with substantial relative impairments. Some participants had isolated relative impairments in memory (148, 18%), visuospatial functioning (117, 14%), language (71, 9%), and executive functioning (66, 8%). The group with isolated relative memory impairments had higher proportions with ≥ APOE ε4 allele, more extensive Alzheimer's-related neuropathology, and higher proportions with other Alzheimer's dementia genetic risk variants. DISCUSSION:A cognitive subgrouping strategy may identify biologically distinct subsets of people with Alzheimer's dementia.