Genome-wide association analysis of dementia and its clinical endophenotypes reveal novel loci associated with Alzheimer's disease and three causality networks: The GR@ACE project

Abstract

Introduction

Large variability among Alzheimer's disease (AD) cases might impact genetic discoveries and complicate dissection of underlying biological pathways.

Methods

Genome Research at Fundacio ACE (GR@ACE) is a genome-wide study of dementia and its clinical endophenotypes, defined based on AD's clinical certainty and vascular burden. We assessed the impact of known AD loci across endophenotypes to generate loci categories. We incorporated gene coexpression data and conducted pathway analysis per category. Finally, to evaluate the effect of heterogeneity in genetic studies, GR@ACE series were meta-analyzed with additional genome-wide association study data sets.

Results

We classified known AD loci into three categories, which might reflect the disease clinical heterogeneity. Vascular processes were only detected as a causal mechanism in probable AD. The meta-analysis strategy revealed the ANKRD31-rs4704171 and NDUFAF6-rs10098778 and confirmed SCIMP-rs7225151 and CD33-rs3865444.

Discussion

The regulation of vasculature is a prominent causal component of probable AD. GR@ACE meta-analysis revealed novel AD genetic signals, strongly driven by the presence of clinical heterogeneity in the AD series.

Highlights

• We detected three gene categories operating differently in AD subgroups of patients.

• Vasculature regulation may be an essential part of the causative mechanism in pure AD.

• Two novel signals reached genome-wide significance in ANKRD31 and NDUFAF6loci.

• We confirmed SCIMP-rs7225151 and CD33-rs3865444 to be genome-wide significant.

• Genetic discoveries are strongly impacted by the presence of subgroups of AD.

1 Background

Dementia is an age-related clinical syndrome that devastates cognitive abilities and interferes in elderly people's daily activities. Although its incidence is decreasing due to improvements to public health systems and control of cardiovascular risk factors [1], its prevalence is steadily increasing due to rising life expectancy of human populations [2].

Dementia is linked to many underlying pathologies, with Alzheimer's disease (AD) being the most common condition. Clinical AD is a heterogeneous syndrome. Brain autopsies have shown that roughly 80% of clinical AD patients present with brain vascular pathology [3] in addition to the common neuropathological AD hallmarks: amyloidosis, neurofibrillary tangles, and cerebral amyloid angiopathy (CAA) [4]. In fact, brain vascular pathology has been shown to be an important risk factor for AD that accelerates cognitive decline [5] and lowers the threshold for clinical diagnosis of AD [6]. In that context, it has been suggested that dementia is represented by a gradient of neurodegenerative and vascular components [7], from pure AD forms, with a strong neurodegenerative component, to pure vascular dementia (VaD) cases and in-between mixed pathologies, representing the coexistence of both neurodegenerative and vascular components [7]. Despite that, whether there are differential biological routes operating under different levels of vascular burden in clinical AD patients remains mostly unknown.

In the search for the etiology of AD, genetic factors play a pivotal role. Two forms of the disease can be differentiated according to individual genetic background. The Mendelian form is an uncommon disorder that mainly affects families with early-onset AD (<65 years), whereas the polygenic form is a complex disorder mainly appearing in sporadic cases with late-onset AD (LOAD) (>65 years). Highly penetrant mutations detected in families with early-onset AD have been pinpointed to three genes: APP [8], PSEN1 [9], and PSEN2 [10], leading to the establishment of the amyloid hypothesis as a potential causal mechanism for the disease [11].

LOAD heritability falls in the range of 13%‒80% [12], [13]. Although APOE ε4 was the first to be discovered and still remains the strongest genetic risk factor for AD [14], almost 40 additional genetic variants have been identified [15]-[17] using genome-wide association studies (GWAS) and large sequencing projects. Among the biological pathways underlying genetic hits, the roles of the immune system, cholesterol, amyloid, and tau metabolism have been highlighted [18]. Despite these, current genetic findings account for 31% of LOAD heritability [19], and the biological picture of AD is still poorly understood. Several reasons can explain this. Among them, the presence of clinical and neuropathological heterogeneity between AD cases in genetic studies, as recently demonstrated [12], might compromise the power to detect genuine genetic associations and decrease the estimates of risk attributed to genetic variation.

The neuropathological variability in clinical AD cases comprises a wide spectrum, from those with concomitant vascular brain disease to those with a pure AD phenotype, as previously proposed by Viswanathan et al. [7]. This large heterogeneity might hamper the identification of functional categories of genes underlying differential biological routes to dementia and might impact AD genetic studies. To gain insight into the causality networks behind AD clinical subgroups and to explore their impact in large GWAS, we conducted the Genome Research at Fundacio ACE (GR@ACE) study. This is a GWAS of dementia and its clinical endophenotypes defined based on AD's clinical certainty and the burden of vascular comorbidity. The GR@ACE is a unique genomic resource comprising the largest number of dementia cases diagnosed in a single memory clinic to date. First, we determined whether we could identify categories of known LOAD genes linked to clinical subgroups of AD cases. Next, we explored whether these categories suggested different biological routes. Finally, to assess the impact of these clinical subgroups of AD cases in GWAS findings, we meta-analyzed the GR@ACE data with independent GWAS series.

2 Methods

2.1 Subjects

2.1.1 GR@ACE cohort and phenotype definitions

The GR@ACE study comprises 4120 AD cases and 3289 control individuals (Table 1). Cases were recruited from Fundació ACE, Institut Català de Neurociències Aplicades (Catalonia, Spain). Diagnoses were established by a multidisciplinary working group, including neurologists, neuropsychologists, and social workers, according to the Diagnostic and Statistical Manual of Mental Disorders–IV criteria for dementia and to the National Institute on Aging and Alzheimer's Association's (NIA-AA) 2011 guidelines for defining AD [20] (see Supplementary Material). In the present study, we considered AD cases as dementia individuals diagnosed with probable or possible AD at any moment of their clinical course.

Table 1. GR@ACE demographic characteristics and endophenotype definitions

Phenotype	Primary diagnostic	Secondary diagnostic	N	Mean age ± SD	Women %	APOE ε4 %
Controls	–	–	3289	54.3 ± 14.4	48.9	21.4
VaD^++	VaD	Pss AD	373	80.1 ± 5.5	54.9	25.0
VaD^+	VaD/Pss AD	VaD/Pss AD	1168	80.4 ± 6.3	65.0	32.8
AD	Pr/Pss AD at any time in medical history		4120	79.0 ± 7.5	69.6	40.1
AD^+	Pr/Pss AD	Pr/Pss AD	3797	79.2 ± 7.5	70.6	41.2
AD^++	Pr AD	Pr/Pss AD	2611	78.8 ± 7.9	72.8	44.6
AD^+++	Pr AD	Pr AD	1854	79.0 ± 8.0	74.6	47.0

• Abbreviations: AD, Alzheimer's disease; SD, standard deviation; VaD, vascular dementia; Pss AD, possible AD; Pr AD, probable AD.

We took advantage of this wide clinical definition to refine AD cases. Considering the dementia spectrum proposed by Viswanathan et al. [7], we classified Gr@ACE AD patients according to the degree of clinical certainty for AD phenotype and the presence of vascular comorbidity, from “pure” clinical AD cases to mixed and vascular enriched cases. This approach was feasible due to Fundació ACE's endorsement of a primary and a secondary etiologic diagnosis as well as routine follow-up evaluations [21] (Supplementary Methods). Using the entire clinical chart of each subject, we differentiated five clinical subgroups of patients representing the GR@ACE endophenotypes: (1) the AD⁺⁺⁺ endophenotype comprises individuals with a last clinical diagnosis of probable AD in both primary and secondary diagnoses (n = 1854); (2) the AD⁺⁺ includes individuals diagnosed with probable AD in the primary diagnosis and probable or possible AD in the secondary diagnosis (n = 2611); (3) the AD⁺ encompasses patients diagnosed with probable or possible AD either in the primary diagnosis or in the secondary diagnosis (n = 3797); (4) the VaD⁺ includes patients diagnosed with VaD or possible AD in the primary diagnosis (n = 1168); and (5) the VaD⁺⁺ comprises patients diagnosed with probable or possible vascular dementia in the primary diagnosis (n = 373) (Table 1) (Supplementary Fig. 1). Patients with VaD were defined according to the Neuroepidemiology Branch of the National Institute of Neurological Disorders and Stroke and the Association Internationale pour la Recherche et l'Enseignement en Neurosciences criteria [22].

Control individuals were recruited from three centers: Fundació ACE (Barcelona, Spain), Valme University Hospital (Seville, Spain), and the Spanish National DNA Bank Carlos III (University of Salamanca, Spain) (www.bancoadn.org). Written informed consent was obtained from all participants. The Ethics and Scientific Committees have approved this research protocol (Acta 25/2016. Ethics Committee. H. Clinic i Provincial, Barcelona, Spain).

2.1.2 Replication sample

With the objective to successfully replicate novel GWAS findings, we used an independent Spanish sample of 1943 AD cases (mean age = 79.2; standard deviation [SD] = 7.6; 66.3% women) and 3016 controls (mean age = 52.8; SD = 15.2; 46% women) presenting similar characteristics to the GR@ACE cohort and with available genetic data. All AD cases were examined at a single site, Fundació ACE, Institut Català de Neurociències Aplicades (Catalonia, Spain), and were assessed by applying the same criteria previously explained. The sample composition of dementia cases comprised 30.1% of AD⁺⁺⁺ (n = 584), 53.9% of AD⁺⁺ (n = 1048), 91.9% of AD⁺ (n = 1783), 23.0% of VaD⁺ (n = 447), and 4.6% of VaD⁺⁺ (n = 89) cases. Control individuals were selected from the Spanish population available at three centers, previously described.

2.2 GWAS genotyping, quality control, imputation, and statistical analysis

Participants were genotyped using the Axiom 815K Spanish Biobank Array (Thermo Fisher). Genotyping was performed in the Spanish National Center for Genotyping (CeGEN, Santiago de Compostela, Spain) (Supplementary Methods).

We removed samples with genotype call rates <97%, excess heterozygosity, duplicates, samples genetically related to other individuals in the cohort, or sample mix-up (PIHAT >0.1875). If a sex discrepancy was detected, the sample was removed unless the discrepancy was safely resolved. To detect population outliers of non-European ancestry (>6 SD from European population mean), principal component analysis was conducted using SMARTPCA from EIGENSOFT 6.1.4 (Fig. 1) (Supplementary Methods).

Figure 1

Results of genome-wide association analysis for the GR@ACE data set (n = 7409). (A) Principal component analysis; (B) principal component analysis centered in European population; (C) QQplot for the discovery model, adjusted for first four PCs; (D) Manhattan plot for genome-wide results. Abbreviations: AFR, African; AMR, Admixed American; EAS, East Asian; EUR, European; SAS, South Asian.

We excluded variants with a call rate <95% or that grossly deviated from Hardy-Weinberg equilibrium in controls (P value ≤ 1 × 10⁻⁶); we also excluded markers with a different missing rate between case and control (P value < 5 × 10⁻⁴ for the difference) or a minor allele frequency (MAF) < 0.01. Imputation was carried out using the Haplotype Reference Consortium panel in the Michigan Imputation Server (https://imputationserver.sph.umich.edu). Only common markers (MAF >0.01) with a high imputation quality (R² > 0.30) were selected for downstream analyses.

The GWAS was performed for GR@ACE as a whole and for each endophenotype, ([N _AD+++ = 5143]; [N _AD++ = 5900]; [N _AD+ = 7086]; [N _{AD dementia} = 7409]; [N _VaD+ = 4487]; and [N _VaD++ = 3662]). The GWAS was conducted for genotype dosages using an additive genetic model with PLINK 1.9. A model including the top four PCs as covariates was used for the discovery stage because it exhibited the lowest inflation and optimal power compared with alternative tested models (Supplementary Fig. 2). Further description is provided in Supplementary Methods. Power analysis was performed using QUANTO software v1.2.4 [23] to model the impact on statistical power at different MAFs and effect sizes in available case-control cohorts. Results were depicted using the ggplot2 package in R. Analyses were performed for a GWAS experiment that would meet criteria for genome-wide significance (P < 5 × 10⁻⁸) and for a replication experiment that would meet P < .05 (Supplementary Fig. 3). Calculations were performed considering a disease prevalence of 1.73%, according to registers in the Spanish population.

GR@ACE GWAS data have been deposited into the European Genome-phenome Archive (https://ega-archive.org), which is hosted by the European Bioinformatics Institute and the Center for Genomic Regulation, under the accession number EGAS00001003424.

2.3 Genetic exploration of GR@ACE clinical endophenotypes and enrichment analysis

With the objective to explore whether different biological routes operate under different levels of vascular burden in clinical AD patients, first, we classified GR@ACE dementia cases to cover the dementia spectrum previously proposed by Viswanathan et al. [7] (see Section 2.1). Second, we extracted the effect (odds ratio [OR]) for known LOAD genetic variants (MAF >1%). See included variants in Supplementary Table 1. Then, we explored whether previously identified LOAD variants were more strongly associated with a specific subgroup of AD patients. We quantified the strength of the association for each variant across endophenotypes, named henceforth global effect change. It was calculated as the absolute difference between variant OR for extreme endophenotypes (VaD⁺⁺ vs. AD⁺⁺⁺). According to the global effect change and direction of the enrichment, i.e., from VaD⁺⁺ to AD⁺⁺⁺ or from AD⁺⁺⁺ to VaD⁺⁺, we classified LOAD genetic variants into three categories. Thus, category A includes variants with an increase in the association effect from VaD⁺⁺ to AD⁺⁺⁺ endophenotypes and a global effect change >0.05, and category B includes variants with an increase in the association effect from AD⁺⁺⁺ to VaD⁺⁺ and a global effect change >0.05. Category C comprises variants not fulfilling criteria for categories A or B (Supplementary Table 1). Finally, we assessed the biological pathways underlying each category. We incorporated data from gene coexpression for each specific gene category using the GeneFriends tool (http://genefriends.org/) and performed pathway analysis of top coexpressed genes using the overrepresentation enrichment method in WebGestalt (http://www.webgestalt.org/option.php) (see Supplementary Methods). With the objective to further explore potential additional trends in category C, we performed a specific subanalysis in this category, widely described in Supplementary Methods. To validate previous gene classification, which strongly determines the pathway analysis results, we conducted a stringent subanalysis (see Supplementary Methods and Supplementary Fig. 4).

2.4 Meta-analysis: Data sets and association analysis

To explore the impact of the different clinical endophenotypes in GWAS findings, we combined the whole GR@ACE GWAS data set, which represents a dementia set of samples and its endophenotypes with (1) genotype-level data from nine additional GWAS series (N = 13,826), available through dbGaP (https://www.ncbi.nlm.nih.gov/gap) that we processed by applying identical quality control and imputation procedures to those described for the GR@ACE cohort (Supplementary Table 2) and (2) aggregated summary statistics available from the International Genomics Alzheimer's Project (IGAP) (http://web.pasteur-lille.fr/en/recherche/u744/igap/igap_download.php) [24], including IGAP stages I (N final = 61,571) and IGAP I and II (N final = 81,455) (Supplementary Methods). Meta-analyses were conducted using the inverse variant method in METAL software (https://genome.sph.umich.edu/wiki/METAL). The linkage disequilibrium (LD) score calculations, clumping, and conditional analysis are described in Supplementary Methods.

2.5 Replication of genome-wide significant findings

We then explored genome-wide significant (GWS) signals in an independent cohort of Spanish ancestry (N = 4959). We extracted variants of interest from GWAS data, which were genotyped and processed applying similar methods to those explained for the GR@ACE study (Section 2.3). Finally, meta-analysis including the discovery stage, named stage I, and the replica data set, stage II, was performed as previously described. Results were interpreted according to the American Statistical Association guidelines [25], [26].

2.6 Biological interpretation of meta-GWAS signals

Gene expression quantitative trait locus (eQTL) analysis was conducted to link meta-GWAS top signals to genes. Markers with moderate-to-high LD (r² ≥ 0.6) with the novel lead markers were identified using LDlink [27] for European population and were included in this analysis. We used brain (n = 11) and whole-blood (n = 1) tissues from the GTEx v7 repository (https://www.gtexportal.org/home) for mapping cis-eQTLs (Supplementary Table 3). As an extension of GTEx tissue eQTL mapping, we explored brain eQTLs for GWS genomic regions using additional databases available via Functional Mapping and Annotation of Genome-Wide Association Studies (FUMA) [28]. We also performed functional annotation for GWS markers, chromatin interaction, and gene-based analysis using similar criteria to those previously described by Jansen et al. [17] (see Supplementary Methods).

3 Results

3.1 GR@ACE genome-wide association study

After quality control and imputation, the GR@ACE study encompassed 7409 unrelated individuals from the Spanish population and 7.7 million variants (λ_GC = 1.03). The APOE-rs429358 marker was the only one to have a GWS association (OR = 2.27 [2.06–2.50]; P = 1.25 × 10⁻⁶²) (Fig. 1). Four additional LOAD variants displayed statistically significant evidence of replication (BIN1-rs6733839, MAPT-rs2732703, MS4A2-rs983392, and PICALM-rs10792832) and nine additional markers presented a consistent direction for the effect (Supplementary Table 1). MAPT marker association remains significant in APOE ε4 noncarriers, but the effect size was stable in both strata (Supplementary Table 4). GWAS of clinical endophenotypes showed that CNTNAP2-rs117834366 was associated with the VaD⁺⁺ endophenotype (Table 2). This marker is in complete linkage equilibrium with CNTNAP2-rs114360492 (r² = 0), previously reported in GWAS of AD by proxy [17]. See results in Supplementary Results (Supplementary Figs. 5 and 6; Supplementary Table 5).

Table 2. Association results for lead single-nucleotide polymorphisms reaching genome-wide significance

Marker	Near locus	Position Chr:bp	A1/A2	MAF	Stage I			Stage II (replica)		Stage I + II
					Discovery stage	OR (95% CI)	P	OR (95% CI)	P	OR (95% CI)	P
rs117834366^∗	CNTNAP2	7:147634891	A/G	0.011	GR@ACE VaD^++	6.03 (3.22–11.2)	1.91 × 10^−8	1.09 (0.66–1.78)	.73	2.08 (1.42–3.07)	1.79 × 10^−4
rs4704171^†	ANKRD31	5:74368254	C/T	0.123	GR@ACE + dbGaP	1.19 (1.12–1.27)	2.78 × 10^−8	1.10 (0.98–1.25)	.09	1.18 (1.11–1.24)	1.15 × 10^−8
rs10098778^†	TP53INP1/NDUFAF6	8:95992020	C/T	0.470	GR@ACE + IGAP I & II	0.94 (0.91–0.96)	2.54 × 10^−8	0.96 (0.89–1.05)	.40	0.94 (0.92–0.96)	2.32 × 10^−8
rs7225151^†	SCIMP	17:5137047	A/G	0.126	GR@ACE + IGAP I & II	1.10 (1.06–1.14)	8.98 × 10^−8	1.14 (1.00–1.30)	.04	1.10 (1.06–1.14)	1.38 × 10^−8
rs7225151^†	SCIMP	17:5137047	A/G	0.126	GR@ACE AD^+++ + IGAP I & II	1.11 (1.07–1.15)	1.12 × 10^−8	1.07 (0.88–1.30)	.49	1.11 (1.07–1.15)	9.45 × 10^−9

• NOTE. The threshold for genome-wide significance was 5 × 10⁻⁸. Position = GRCh37/hg19 coordinates Stage I corresponds to the discovery stage; stage II corresponds to the replica; stage I + II corresponds to the meta-analysis of the discovery and the replica. Stage I + II for rs117834366 corresponds to the meta-analysis of GR@ACE VaD⁺⁺ and the entire replica (N = 4959). Abbreviations: AD, Alzheimer's disease; VaD, vascular dementia; chr:bp, chromosome:base pair; A1, minor allele; A2, mayor allele; MAF, minor allele frequency obtained from the GR@ACE study (N = 7409); OR, odds ratio; CI, confidence interval.
• ∗ Imputed variant; rs117834366 r2 = 0.68.
• † Genotyped variant.

3.2 Genetic exploration of GR@ACE clinical endophenotypes and enrichment analysis

To explore whether clinical AD subgroups, representing GR@ACE endophenotypes, reflected variations in the underlying biological pathways driving dementia, we classified LOAD genetic variants into three categories. Category A comprised variants strongly related to the purest form of clinical AD (i.e., subjects with probable AD in primary and secondary diagnoses). The most prominent locus of this category was APOE-rs429358 (AD⁺⁺⁺ OR = 2.92 [2.60–3.27], P value = 9.26 × 10⁻⁷⁵; VaD⁺⁺ OR [95% confidence interval] = 1.27 [1.02–1.59], P value = .04). Other loci included in category A were CR1, BIN1, MEF2C, MS4A2, PICALM, MAPT, and CD33. In contrast, category B comprised variants with the strongest effect observed in subjects with AD mixed with vascular disease (SORL1, ADAM10, CASS4, ATP5H, and ACE) (Supplementary Table 1). Further description is provided in Supplementary Results. Category C comprised a group of variants with effects in all clinical endophenotypes (Fig. 2). Subanalysis for category C is shown in Supplementary Results and Supplementary Table 6.

Figure 2

(A) Enrichment trend per genetic marker and gene category across GR@ACE endophenotypes. (B) Graph centered in effect change range 0–0.5. Effect change per endophenotype = Variant Odds ratio–variant null effect; Enrichment trend per category was obtained applying a linear regression using ggplot2 in R. Abbreviations: AD, Alzheimer's disease; VaD, vascular dementia.

Next, we explored biological pathways for each gene category. Note that the regulation of vasculature development and blood vessel morphogenesis were only detected for genes in category A, which is more closely related to pure AD (P = 2.03 × 10⁻⁷, P = 1.90 × 10⁻⁶, respectively) (Table 3). Additional categories indicated immune system pathways (category B, P = 2.07 × 10⁻⁷; category C, P = 5.77 × 10⁻¹⁵) (Table 3). Finally, with the aim of validating previous results, we conducted a subanalysis by classifying LOAD genetic variants with more stringent classification criteria (Supplementary Methods). Again, APOE, CR1, MEF2C, MS4A2, and PICALM loci were found in category A; SORL1 and CASS4 were in category B; and additional AD loci were in category C (Supplementary Fig. 4). Regulation of vasculature development was exclusively identified as the top pathway in category A (P = 2.14 × 10⁻⁷) when we restricted the analysis to include those loci coexpressing with at least 4 LOAD genes (Supplementary Table 7). Further replication in cohorts with available neuropathological data would be recommended.

Table 3. Top ten biological pathways per gene category

Gene ontology pathway	Top 10 coregulated pathways for category A	P value
GO:1901342	Regulation of vasculature development	2.03 × 10^−7
GO:0060326	Cell chemotaxis	2.59 × 10^−7
GO:0048771	Tissue remodeling	6.77 × 10^−7
GO:0050865	Regulation of cell activation	1.14 × 10^−6
GO:0007159	Leukocyte cell-cell adhesion	1.21 × 10^−6
GO:0048514	Blood vessel morphogenesis	1.90 × 10^−6
GO:0003012	Muscle system process	2.54 × 10^−6
GO:0002764	Immune response-regulating signaling pathway	3.48 × 10^−6
GO:0032103	Positive regulation of response to external stimulus	3.91 × 10^−6
GO:0010959	Regulation of metal ion transport	4.36 × 10^−6

Gene ontology pathway	Top 10 coregulated pathways for category B	P value
GO:0009620	Response to fungus	2.02 × 10^−7
GO:0050886	Endocrine process	3.58 × 10^−7
GO:0002443	Leukocyte mediated immunity	5.47 × 10^−7
GO:0050865	Regulation of cell activation	1.52 × 10^−5
GO:0031349	Positive regulation of defense response	8.42 × 10^−5
GO:0032103	Positive regulation of response to external stimulus	1.00 × 10^−4
GO:0002250	Adaptive immune response	1.30 × 10^−4
GO:0098542	Defense response to other organism	2.00 × 10^−4
GO:1901568	Fatty acid derivative metabolic process	2.24 × 10^−4
GO:0050900	Leukocyte migration	2.57 × 10^−4

Gene ontology pathway	Top 10 coregulated pathways for category C	P value
GO:0007159	Leukocyte cell-cell adhesion	5.77 × 10^−15
GO:0050865	Regulation of cell activation	4.37 × 10^−14
GO:0002764	Immune response-regulating signaling pathway	1.33 × 10^−12
GO:0002253	Activation of immune response	3.96 × 10^−12
GO:0002443	Leukocyte mediated immunity	4.34 × 10^−12
GO:0002274	Myeloid leukocyte activation	7.78 × 10^−12
GO:0002250	Adaptive immune response	1.24 × 10^−11
GO:0002263	Cell activation involved in immune response	7.07 × 10^−11
GO:0022407	Regulation of cell-cell adhesion	5.40 × 10^−9
GO:0070661	Leukocyte proliferation	1.22 × 10^−8

3.3 Meta-analysis of GR@ACE study with other data sets

To assess the impact of sample composition in AD GWAS, and look for new AD loci, we first combined the GR@ACE data set with nine additional genomic databases that had genotypic level data available. Subtle genomic inflation was detected, mainly explained by polygenicity (λ_GC = 1.10; LD score intercept = 1.04). Five regions were associated with LOAD (Fig. 3); of these, four (APOE-rs429358, PICALM-rs10792832, MS4A2-rs983392, and BIN1-rs6733839) have been previously linked to AD (Supplementary Table 8), and one is a new GWAS finding (ANKDR31-rs4704171; OR = 1.19 [1.12–1.27]; P = 2.78 × 10⁻⁸) (Table 2).

Figure 3

A) Results of genome-wide association analysis for GR@ACE meta-analysis with nine additional databases (n = 21,235). (B) QQplot. (C) Associations of the region centered on rs4704171 located in the ANKRD31 locus and containing the HMGCR locus.

Then, we conducted a genome-wide meta-analysis combining the GR@ACE study with IGAP stage I. We identified 12 LOAD genomic regions reaching GWS. CD33-rs3865444, which did not reach GWS in the IGAP meta-analysis, was significantly associated with LOAD (OR = 0.92 [0.89–0.95]; P = 3.61 × 10⁻⁸) (Supplementary Fig. 7). Among the top suggestive signals, we detected HBEGF-rs4150233 (OR = 0.92 [0.90–0.95]; P = 5.10 × 10⁻⁸) previously identified by a transethnic GWAS [29].

Next, meta-analysis of the whole GR@ACE data set with IGAP I and II enabled the identification, for the first time, of NDUFAF6-rs10098778 as a GWS signal (OR = 0.94 [0.91–0.96]; P = 2.54 × 10⁻⁸). When we combined GR@ACE AD⁺⁺⁺ endophenotype with IGAP I and II, we also detected SCIMP-rs7225151 (OR = 1.11 [1.07–1.15]; P = 1.12 × 10⁻⁸) (Table 2) (Supplementary Fig. 8). It was previously reported as a genome-wide suggestive signal by IGAP [24]. Recently, SCIMP-rs113260531, which is in complete LD with our lead marker (r² = 1), was associated with AD [17].

3.4 Replication of genome-wide significant findings

Finally, we tested for replication of the new signals in an independent sample of 4959 Spanish individuals. The CNTNAP2-rs117834366, detected in the GWAS of GR@ACE VaD⁺⁺ endophenotype, had a P value of 0.79 with a similar effect direction to that reported previously but strongly deflated in the entire replica sample (OR = 1.09 [0.66–1.78]; P = .79) (Table 3). Analysis of the subspecific VaD phenotype in the replica (N = 89) would be highly inaccurate.

In the exploration of meta-GWAS findings, we observed that the ANKDR31-rs4704171-C marker increased the risk of AD (OR = 1.10 [0.98–1.25]; P = .09; power = 33%). Although the expected effect is in line with previous data, the precision of the estimate in this replica differs, ranging from a 2% decrease (a small negative association) to a 25% increase. Of note, the result emerging from the meta-analysis of the replica with the discovery sample (n = 26,194) is compatible with its potential role in dementia (OR = 1.18 [1.11–1.24]; P = 1.15 × 10⁻⁸) (Table 3). See forest plot in Supplementary Fig. 9.

We observed a similar effect direction in the NDUFAF6-rs10098778 marker to that reported in the discovery stage, with interval estimates ranging from a risk decrease of 11% to a risk increase of 5% (OR = 0.96 [0.89–1.05]; P = .40; power = 16%). This signal had a P = 2.32 × 10⁻⁸ in the final meta-analysis, including the whole GR@ACE data set, the replica, and IGAP I and II (n = 91,373) (Supplementary Fig. 9).

SCIMP-rs7225151 showed a risk effect in the whole replica. Limits of the interval were consistent with a positive association (OR = 1.14 [1.01–1.29]; P = .047; power = 39%; n cases = 1943). In the AD⁺⁺⁺ endophenotype, the marker presented the same positive risk effect direction (OR = 1.07 [0.88–1.30]), although had P = .49, which could be mainly explained by a reduction of the sample size (power = 19%; n cases = 584). Our results for both meta-analyses, the final meta-analysis of the whole GR@ACE data set (n = 91,373) and the final meta-analysis of GR@ACE AD⁺⁺⁺ endophenotype, (n = 85,055) were compatible with a potential effect of this marker in AD (Table 3). See forest plots in Supplementary Fig. 9.

3.5 Biological interpretation of meta-GWAS signals

To identify candidate genes and potential causal variants within novel meta-GWAS regions, we conducted cis-eQTL mapping. The rs2335107 marker located in the ANKRD31 locus (chr5:74,451,693) was associated with the cortical expression of the long noncoding RNA (lncRNA) CTD-2235C13.3 (P = 1.26 × 10⁻⁵). This variant is located 83.4 kb from the meta-GWAS lead single nucleotide polymorphism (rs4704171, chr5:74,368,254), and both are in complete LD (r² = 1). The CTD-2235C13.3 gene is located 1.6 kb from the HMGCR locus, and its function is unknown. The NDUFAF6 region mapped for NDUFAF6 RNA cortical expression (P = 5.56 × 10⁻⁶) and for TP53INP1 RNA blood expression (P = 1.17 × 10⁻¹⁰). Finally, rs73976325 (chr17:5,123,227), located in the SCIMP locus, to 13.8 kb from the meta-GWAS top signal (rs7225151, chr17:5,137,047), mapped to brain cis-acting eQTL for AC012146.1 lincRNA (P = 2.15 × 10⁻⁷). Two additional markers were pinpointed to blood eQTLs, SCIMP-rs6502851 (P = 3.89 × 10⁻⁰⁸) and RABEP-rs59277121 (P = 3.89 × 10⁻⁸) (Supplementary Table 3). In an additional prioritization strategy, combining information from positional mapping, eQTL, chromatin interaction, and gene-based genome-wide association analysis via FUMA, results pointed to ANKDR31 and POLK for the ANKDR31 genomic region, as well as NDUFAF6 and TP53INP1 for the NDUFAF6 region (Supplementary Table 9). Further description is provided in Supplementary Information and Supplementary Table 9.

4 Discussion

We present a comprehensive GWAS of AD dementia cases. This represents the first pilot study exploring the genetics and underlying biological pathways of subgroups of patients with AD, defined based on vascular burden. We showed differential biological routes underlying clinical endophenotypes and demonstrated how these differential subgroups of patients with AD impact GWAS discoveries. The GR@ACE study represents a unique genomic resource because all affected cases were diagnosed in a single memory clinic using the same screening and diagnostic techniques. This might limit potential sources of clinical variation between study participants, recently demonstrated in a large meta-GWAS [12].

Based on the increase in evidence suggesting that vascular brain pathology can act concomitantly with AD to produce a more rapid cognitive decline [5], we explored the effect of known LOAD loci across different levels of vascular burden in dementia patients using only clinical definitions. Our basic idea was to dissect, from a molecular point of view, the model previously proposed by Viswanathan et al. [7]. We observed three categories of loci, which might reflect the disease's clinical heterogeneity, from vascular and mixed forms to a “purer” AD phenotype. Intriguingly, we detected vascular processes to be the main causal mechanism in clinically pure AD and found the immune system pervasively detected across the three categories. Although both pathways have been previously associated with LOAD by network analysis [30], this is the first study to show that the association with the vascular system is conducted by AD-specific clinical subgroup.

It should be noted that the present study used clinical criteria to define the AD cases [20], but recently the classification of AD has evolved. In 2018, the NIA-AA proposed a novel research framework for the biological classification of AD based on the presence in vivo of biomarker evidences for amyloid (A), tau (T), and neurodegeneration (N), as surrogate of the neuropathological state of an individual [31]. The AT(N) biomarker system allows the classification of individuals into three categories: those with a normal biomarker profile, those with biomarkers compatible with AD change, and those with biomarkers compatible with non-AD pathological changes [31]. Using the NIA-AA approach, the generation of subgroups of AD patients considering vascular pathology would be unfeasible, as nowadays the ATN classification does not integrate measures of vascular dysfunction. Taking into account that most of dementia cases are caused by mixed pathologies, the current system seems deeply incomplete to study the probable interaction between neurodegeneration and vascular dysfunction. This idea has also been claimed by others [32], [33]. Thus, we encourage other groups to contrast the proposed loci classification, which was based on the GR@ACE clinical endophenotypes, but using well-powered GWAS cohorts with available neuropathological data.

Silent changes occur in brain microvasculature during AD progression. In fact, CAA is a well-recognized AD pathological feature characterized by the accumulation of amyloid proteins in the walls of small cerebral vessels. CAA has been proposed to compromise the perivascular drainage of Aβ from the brain to the peripheral system [34]. Almost all AD brains harbor CAA pathology to some extent, although in vivo most CAA cases remain undiagnosed, even using the validated Boston criteria [35]. Mendelian mutations of the APP gene have been found in CAA and AD [8], [36]. APOE ε4 and CR1 have been associated with an increased risk of CAA [37], [38]. In particular, distinct AD loci have been associated with capillary and noncapillary CAA [39]. Between them, APOE ε4 was strongly related to capillary CAA [39]. These links make it conceivable that a potential genetic overlap exists between CAA and AD and suggest that CAA pathology could represents an underlying process for AD. In that context, we think that intrinsic alterations to the vasculature could contribute to disease pathogenesis in more pure forms of AD, explaining our results. Conversely, in AD individuals with evident cerebrovascular lesions comprising mixed forms, the additional role of cardiovascular risk factors such as hypertension, atherosclerosis, or arteriosclerosis should be considered, as these could point to a systemic pathological state leading to vascular damage and dementia. This would agree with the limited genetic correlation between neurodegenerative and other neurologic disorders [12], as well as with the results obtained from heterochronic parabionts in aging models [40].

Understanding the role of vasculature pathology in AD seems to be a pertinent step. In that scenario, CAA would be a key AD hallmark. CAA represents the unique identified link between the vascular and amyloid hypotheses, but it has been completely neglected in the original hypothesis formulation.

From a clinical point of view, placing each patient somewhere along the disease spectrum proposed by Viswanathan et al. [7] is complex. A deep understanding of heterogeneity in AD seems necessary to design better genetic studies, which must drive the discovery of novel loci and, ultimately, innovative targets for AD therapies. In this study, we explored how clinical heterogeneity might impact GWAS findings by integrating distinct GWAS data sets with either the GR@ACE cohort as a whole or its endophenotypes. We found several new GWS signals that seem strongly dependent on the sample composition. For example, after combining IGAP stages I and II with the entire GR@ACE data set, we identified genetic signals in the NDUFAF6 genomic region but not in the SCIMP region. When this exercise was conducted using GR@ACE endophenotypes, the SCIMP signal was detected using the clinically “pure” AD GR@ACE endophenotype. It should be noted that the power to detect SCIMP signal in the meta-analysis with GRA@ACE dementia and GR@ACE AD⁺⁺⁺ was 75 and 70%, respectively. We tried to replicate this finding in a purer AD data set without clinical mixed dementia cases, but the available number of clinical AD⁺⁺⁺ cases might be compromising the statistical power to replicate (N cases = 584; power = 19%). Despite that, we think that using specific clinical subgroups of the AD population might empower genetic studies to detect genes associated with specific disease axes.

An alternative strategy is taking advantage of clinical heterogeneity. Specifically, heterogeneity might play a dual role in genetic studies. Although it might decrease the power to detect genes associated with more specific clinical subgroups, incorporating detailed clinical AD definitions can also promote identifying genes shared with other conditions or copathologies such as small vessel disease (SVD). In fact, this was the case for the ATP5H locus, which was previously found to be associated with AD [41] and more recently found in relation to SVD [42]. We think that the same could apply to the ANKRD31 finding. ANKRD31 encodes a protein containing ankyrin repeats, which is linked with neurodevelopmental disorders [43]. Of note, ANKRD31 GWAS signal mapped to the brain eQTL of a lncRNA, located 1.6 kb from the HMGCR locus and residing in the COL4A3BP gene. The HMGCR locus is one of the most important coregulators of cholesterol biosynthesis, and it is the therapeutic target of statins. The COL4A3BP gene is involved in lipid transport [44]. Several studies have linked HMGCR polymorphisms and AD risk or age at onset for AD [45], and the cholesterol pathway has been identified to be a biological route shared between AD and SVD. Interestingly, markers in the POLK locus, associated by gene-based genome-wide association analysis on this study and located in the same disequilibrium block of ANKDR31 (Fig. 3), jointly conferred risk for AD and plasma levels of LDL [46]. Considering prior findings, our results are consistent with this genomic region having a role in mixed dementia. The reported genetic signal should be considered a highly probable finding, although independent replications are still required.

In the present work, NDUFAF6 signals reached GWS for the first time. This finding presented the same effect direction in the independent sample (power = 16%) and remains as GWS after the final meta-analysis. Our lead GWAS marker is in high LD with NDUFAF6-rs4735340, the top suggestive signal reported by Kunkle et al. [18] at this region (r² = 0.95, for Utah residents of European ancestry population). Despite that, there are subtle differences in LD estimates for the Iberian population (r² = 0.87), suggesting that the genetic architecture of the Spanish population could be helping to pinpoint the region of interest. This region is in the close vicinity of TP53INP1, previously associated with AD by a gene-based approach [47]. TP53INP1 is involved in mitochondrial function. Considering these findings and results emerging from eQTL analysis, it would be feasible that a regulatory element for NUDFAF6 resides upstream of the TP53INP1 locus. We also detected that the SCIMP signal was mainly conducted by a specific group of AD cases. This signal was reported to be a suggestive signal by IGAP [24], and a proxy of it (SCIMP-rs113260531) reached GWS recently [17]. SCIMP regions has been involved in several eQTLs, from uncharacterized cortical lncRNA to blood eQTLs in SCIMP and RABEP1 loci, both associated with immune system function [48], [49]. The CD33 locus remains a controversial LOAD locus because large meta-GWAS were unable to replicate this signal [24], but here it reached GWS. We previously proposed that cryptic population substructure could explain the divergent observations for this locus [50].

Note that the lack of definitive neuropathological data for AD cases is a severe limitation of the present study. Clinical definitions have important uncertainties, and diagnosis misclassifications sometimes occur. Hence, some AD individuals included in enriched AD endophenotypes may present concomitant vascular brain disease. The generation of large histopathological GWAS cohorts with associated quantitative data on each pathological hallmark is the ultimate solution to tackling the intrinsic heterogeneity in AD. Unfortunately, there are few examples of neuropathological cohorts: only one GWAS has investigated the genetics of CAA, with APOE being the unique GWS signal [37]. In this study, a small number of AD cases evolved to vascular dementia during follow-up. Large cross-sectional clinical GWAS cannot control diagnostic changes occurring in clinical practice. Clinical diagnosis is a dynamic variable, so understanding the genetic profiles of subgroups of patients evolving to other pathologies would provide powerful information.

It should be considered that there is a limitation in reducing the sample size by splitting the cohort into different endodophenotypes, instead of combining them. Despite that, spanning the spectrum of dementia individuals to generate clinical endophenotypes provided us a versatile design, which let us explore the effect of heterogeneity in GWAS and replicate the main findings of pathway analysis using an alternative strategy. The limited number of VaD cases in subgroup analysis limits the accuracy of gene categorization and pathway analysis. Finally, the exact effector genes for LOAD genetic findings remain unclear. This is a severe limitation to pathway analysis that can only be circumvented by isolating the causative mutations. Independent replication will be needed to corroborate our new GWS signals. In that sense, the selection of specific patient groups might lead to successful replication studies.

The assessment of heterogeneity has important implications for gene discovery, the development of treatments, and their appropriate use in individual patients. The GR@ACE cohort provides useful genomic information, as it accounts for potential sources of variability and contains different subgroups of cases. This enabled us to analyze the LOAD genetic landscape in terms of clinical endophenotypes. Our efforts to disentangle the mechanistic pathways operating under clinical subgroups of patients revealed that vasculature regulation may be an essential part of the causative mechanism of LOAD. Finally, our exploration of AD genetics highlights the relevance of sample composition in genetic discoveries. Considering sample composition in the design of genetic studies might lead to the identification of genetic profiles, which can help clinicians distinguish subsets of patients within the disease spectrum and promote novel therapy targets for AD.

Research in Context

1. Systematic review: Recent literature search, conducted using PubMed database, revealed that the presence of Alzheimer's disease (AD) with vascular brain pathology is the most common form of dementia. Regardless, it is unclear whether subgroups of AD patients present differential genetic profiles or develop dementia through differential biological routes.

2. Interpretation: Our findings showed three gene categories operating differently across subgroups of patients with AD and highlighted the role of vascular pathways in pure forms of AD. We identified novel genome-wide significant signals, which seem strongly dependent on the AD subset used for meta-analysis.

3. Future directions: Our findings suggest the importance of investigating the role of cerebral amyloid angiopathy, a unique identified AD hallmark that connects both the vascular and amyloid hypotheses for AD. Accounting for sample composition in the design of genetic studies might lead to the identification of genetic profiles, which help clinicians to distinguish subsets of patients and establish novel therapy targets.

Acknowledgments

The authors would like to thank patients and controls who participated in this project. The Genome Research @ Fundació ACE project (GR@ACE) is supported by Fundación bancaria “La Caixa”, Grifols SA, Fundació ACE, and ISCIII (Ministry of Health, Spain). They also want to thank the private sponsors who support the basic and clinical projects of our institution (Piramal AG, Laboratorios Echevarne, Araclon Biotech S.A., and Fundació ACE). They are indebted to the Trinitat Port-Carbó legacy and her family for their support of Fundació ACE research programs. Fundació ACE is a participating center in the Dementia Genetics Spanish Consortium (DEGESCO). A.R. and M.B. receive support from the European Union/EFPIA Innovative Medicines Initiative Joint undertaking ADAPTED and MOPEAD projects (grant numbers 115975 and 115985, respectively). M.B. and A.R. are also supported by national grants PI13/02434, PI16/01861, and PI17/01474. Acción Estratégica en Salud is integrated into the Spanish National R + D + I Plan and funded by ISCIII (Instituto de Salud Carlos III)-Subdirección General de Evaluación and the Fondo Europeo de Desarrollo Regional (FEDER- “Una manera de Hacer Europa”). L.M.R. is supported by Consejería de Salud de la Junta de Andalucía (grant PI-0001/2017). Control samples and data from patients included in this study were provided in part by the National DNA Bank Carlos III (www.bancoadn.org, University of Salamanca, Spain) and Hospital Universitario Virgen de Valme (Sevilla, Spain); they were processed after standard operating procedures with the appropriate approval of the Ethical and Scientific Committee. The present work was performed as part of the Biochemistry, Molecular Biology, and Biomedicine doctoral program of S. Moreno-Grau at Universitat Autònoma de Barcelona (Barcelona, Spain).

Data collection and sharing for this project was partially funded by the Alzheimer's Disease Neuroimaging Initiative (ADNI) (National Institutes of Health grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12–2–0012). The ADNI is funded by the National Institute on Aging and the National Institute of Biomedical Imaging and Bioengineering, as well as through generous contributions from the following: AbbVie; the Alzheimer's Association; the Alzheimer's Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research provides funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study was coordinated by the Alzheimer's Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for NeuroImaging at the University of Southern California.

The AddNeuroMed data are from a public-private partnership supported by EFPIA companies and SMEs as part of InnoMed (Innovative Medicines in Europe), an integrated project funded by the European Union of the Sixth Framework program priority FP6–2004-LIFESCIHEALTH-5. Clinical leads responsible for data collection are Iwona Kłoszewska (Lodz), Simon Lovestone (London), Patrizia Mecocci (Perugia), Hilkka Soininen (Kuopio), Magda Tsolaki (Thessaloniki), and Bruno Vellas (Toulouse). Imaging leads are Andy Simmons (London), Lars-Olad Wahlund (Stockholm), and Christian Spenger (Zurich). Bioinformatics leads are Richard Dobson (London) and Stephen Newhouse (London).

Funding support for the Alzheimer's Disease Genetics Consortium (ADGC) was provided through the NIA Division of Neuroscience (U01-AG032984).

The genotypic and associated phenotypic data used in the study “Multi-Site Collaborative Study for Genotype-Phenotype Associations in Alzheimer's Disease (GenADA)” were provided by GlaxoSmithKline, R&D Limited. The data sets used for the analyses described in this manuscript were obtained from dbGaP at http://www.ncbi.nlm.nih.gov/gap through dbGaP accession number phs000219.

The Mayo Clinic Alzheimer's Disease Genetic Studies, led by Dr. Nilüfer Ertekin-Taner and Dr. Steven G. Younkin at the Mayo Clinic in Jacksonville, FL, used samples from the Mayo Clinic Study of Aging, the Mayo Clinic Alzheimer's Disease Research Center, and the Mayo Clinic Brain Bank. Data collection was supported through funding by NIA grants P50 AG016574, R01 AG032990, U01 AG046139, R01 AG018023, U01 AG006576, U01 AG006786, R01 AG025711, R01 AG017216, and R01 AG003949, NINDS grant R01 NS080820, the CurePSP Foundation, and support from the Mayo Foundation.

The Neocodex-Murcia study was funded by the Fundación Alzheimur (Murcia), the Ministerio de Educación y Ciencia (Gobierno de España), Corporación Tecnológica de Andalucía, Agencia IDEA (Consejería de Innovación, Junta de Andalucía), the Diabetes Research Laboratory, and the Biomedical Research Foundation. University Hospital Clínico San Carlos has been supported by CIBER de Diabetes y Enfermedades Metabólicas Asociadas (CIBERDEM); CIBERDEM is an ISCIII Project.

The ROS/MAP study data were provided by the Rush Alzheimer's Disease Center, Rush University Medical Center, Chicago. Data collection was supported through funding by NIA grants P30AG10161, R01AG15819, R01AG17917, R01AG30146, R01AG36836, U01AG32984 and U01AG46152, the Illinois Department of Public Health, and the Translational Genomics Research Institute.

The TGEN study was supported by Kronos Life Science Laboratories, the National Institute on Aging (Arizona Alzheimer's Disease Center grants P30 AG19610 and RO1 AG023193, the Mayo Clinic Alzheimer's Disease Center grant P50 AG16574, and the Intramural Research Program), the National Alzheimer's Coordinating Center (U01 AG016976), and the state of Arizona.

The results published here are in part based on the data obtained from the AMP-AD Knowledge Portal accessed at https://doi.org/10.7303/syn2580853.

The authors thank the International Genomics of Alzheimer's Project (IGAP) for providing summary result data for these analyses. The investigators within IGAP contributed to the design and implementation of IGAP and/or provided data but did not participate in analysis or writing of this report. IGAP was made possible by the generous participation of the control subjects, the patients, and their families. The i–Select chip was funded by the French National Foundation on Alzheimer's disease and related disorders. European Alzheimer's Disease Initiative (EADI) was supported by the LABEX (laboratory of excellence program investment for the future) DISTALZ grant, Inserm, Institut Pasteur de Lille, Université de Lille 2, and the Lille University Hospital. GERAD was supported by the Medical Research Council (grant no. 503480), Alzheimer's Research UK (grant no. 503176), the Wellcome Trust (grant no. 082604/2/07/Z), and German Federal Ministry of Education and Research (BMBF): Competence Network Dementia (CND) grant no. 01GI0102, 01GI0711, 01GI0420. CHARGE was partly supported by the NIH/NIA grant R01 AG033193 and the NIA AG081220 and AGES contract N01–AG–12100, the NHLBI grant R01 HL105756, the Icelandic Heart Association, and the Erasmus Medical Center and Erasmus University. ADGC was supported by the NIH/NIA grants: U01 AG032984, U24 AG021886, U01 AG016976, and the Alzheimer's Association grant ADGC–10–196728.

The GR@ACE study group: Abdelnour C^1,2, Aguilera N¹, Alarcon E^1,3, Alegret M^1,2, Benaque A¹, Boada M^1,2, Buendia M¹, Cañabate P^1,2, Carracedo A^4,5, Corbatón A⁶, de Rojas I¹, Diego S¹, Espinosa A^1,2, Gailhajenet A¹, García González P¹, Gil S¹, Guitart M¹, González Pérez A⁷, Hernández I^1,2, Ibarria, M¹, Lafuente A¹, Macias J⁸, Maroñas O⁴, Martín E¹, Martínez MT⁶, Marquié M¹, Mauleón A¹, Monté-Rubio G¹, Montrreal L¹, Moreno-Grau S^1,2, Moreno M¹, Orellana A¹, Ortega G^1,2, Pancho A¹, Pelejà E¹, Pérez-Cordon A¹, Pineda JA⁸, Preckler S¹, Quintela I³, Real LM^3,8, Rodríguez-Gómez O^1,2, Rosende-Roca M¹, Ruiz A^1,2, Ruiz S^1,2, Sáez ME⁷, Sanabria A^1,2, Santos-Santos MA¹, Serrano-Rios M⁶, Sotolongo-Grau O¹, Tárraga L^1,2, Valero S^1,2, Vargas L¹ (1. Research Center and Memory clinic Fundació ACE. Institut Català de Neurociències Aplicades. Universitat Internacional de Catalunya. Barcelona, Spain; 2. CIBERNED, Center for Networked Biomedical Research on Neurodegenerative Diseases, National Institute of Health Carlos III, Ministry of Economy and Competitiveness, Spain; 3. Dep. of Surgery, Biochemistry and Molecular Biology, School of Medicine. University of Málaga. Málaga, Spain; 4. Grupo de Medicina Xenómica, Centro Nacional de Genotipado (CEGEN-PRB3-ISCIII). Universidade de Santiago de Compostela, Santiago de Compostela, Spain; 5. Fundación Pública Galega de Medicina Xenómica- CIBERER-IDIS, Santiago de Compostela, Spain; 6. Centro de Investigación Biomédica en Red de Diabetes y Enfermedades Metabólicas Asociadas, CIBERDEM, Spain, Hospital Clínico San Carlos, Madrid, Spain; 7. CAEBI. Centro Andaluz de Estudios Bioinformáticos., Sevilla, Spain; 8. Unidad Clínica de Enfermedades Infecciosas y Microbiología. Hospital Universitario de Valme, Sevilla, Spain)

DEGESCO consortium: Adarmes-Gómez AD^1,2, Alarcón-Martín E^3,4, Álvarez I⁵, Álvarez V^6,7, Amer-Ferrer G⁸, Antequera M⁹, Antúnez C⁹, Baquero M¹⁰, Bernal M¹¹, Blesa R^2,12, Boada M^2,3, Buiza-Rueda D^1,2, Bullido MJ^2,13,14, Burguera JA¹⁰, Calero M^2,15,16, Carrillo F^1,2, Carrión-Claro M^1,2, Casajeros MJ¹⁷, Clarimón J^2,12, Cruz-Gamero JM⁴, de Pancorbo MM¹⁸, de Rojas I³, del Ser T¹⁴, Diez-Fairen M⁵, Fortea J^2,12, Franco E¹¹, Frank-García A^2,14,19, García-Alberca JM²⁰, Garcia Madrona S¹⁶, Garcia-Ribas G¹⁶, Gómez-Garre P^1,2, Hernández I^2,3, Hevilla S²⁰, Jesús S^1,2, Labrador Espinosa MA^1,2, Lage C^2,21, Legaz A⁹, Lleó A^2,12, López de Munáin A²², López-García S^2,21, Macias D^1,2, Manzanares S^9,23, Marín M¹¹, Marín-Muñoz J⁹, Marín T²⁰, Marquié M³, Martín Montes A^2,13,19, Martínez B⁹, Martínez C^7,24, Martínez V⁹, Martínez-Lage Álvarez P²⁵, Medina M^2,14, Mendioroz Iriarte M²⁶, Menéndez-González M^7,27, Mir P^1,2, Molinuevo JL²⁸, Monté-Rubio G³, Montrreal L³, Moreno-Grau S^2,3, Orellana A³, Pastor AB¹⁵, Pastor P⁵, Pérez Tur J^2,29,30, Periñán-Tocino T^1,2, Piñol Ripoll G^2,31, Rábano A^2,15,32, Real de Asúa D³³, Rodrigo S¹¹, Rodríguez-Rodríguez E^2,21, Royo JL⁴, Ruiz A^2,3, Sanchez del Valle Díaz R³⁴, Sánchez-Juan P^2,21, Sastre I^2,13, Sotolongo-Grau O3, Tárraga L^2,3, Valero S^2,3, Vicente MP⁹, Vivancos L⁹ (1. Unidad de Trastornos del Movimiento, Servicio de Neurología y Neurofisiología. Instituto de Biomedicina de Sevilla (IBiS), Hospital Universitario Virgen del Rocío/CSIC/Universidad de Sevilla, Seville, Spain; 2. CIBERNED, Network Center for Biomedical Research in Neurodegenerative Diseases, National Institute of Health Carlos III, Spain; 3. Research Center and Memory clinic Fundació ACE. Institut Català de Neurociències Aplicades. Universitat Internacional de Catalunya. Barcelona, Spain; 4. Dep. of Surgery, Biochemistry and Molecular Biology, School of Medicine. University of Málaga. Málaga, Spain; 5. Fundació per la Recerca Biomèdica i Social Mútua Terrassa, and Memory Disorders Unit, Department of Neurology, Hospital Universitari Mutua de Terrassa, University of Barcelona School of Medicine, Terrassa, Barcelona, Spain; 6. Laboratorio de Genética Hospital Universitario Central de Asturias, Oviedo; 7. Instituto de Investigación Biosanitaria del Principado de Asturias (ISPA); 8. Department of Neurology, Hospital Universitario Son Espases, Palma, Spain; 9. Unidad de Demencias. Hospital Clínico Universitario Virgen de la Arrixaca; 10. Servei de Neurologia, Hospital Universitari i Politècnic La Fe; 11. Unidad de Demencias, Servicio de Neurología y Neurofisiología. Instituto de Biomedicina de Sevilla (IBiS), Hospital Universitario Virgen del Rocío/CSIC/Universidad de Sevilla, Seville, Spain; 12. Memory Unit, Neurology Department and Sant Pau Biomedical Research Institute, Hospital de la Santa Creu i Sant Pau, Universitat Autònoma de Barcelona, Barcelona, Spain; 13. Centro de Biologia Molecular Severo Ochoa (C.S.I.C.-U.A.M.), Universidad Autonoma de Madrid, Madrid, Spain; 14. Instituto de Investigacion Sanitaria ‘Hospital la Paz’ (IdIPaz), Madrid, Spain; 15. CIEN Foundation, Queen Sofia Foundation Alzheimer Center, Madrid, Spain; 16. Instituto de Salud Carlos III (ISCIII), Madrid, Spain; 17. Hospital Universitario Ramón y Cajal; Madrid, Spain; 18. BIOMICs, País Vasco; Centro de Investigación Lascaray. Universidad del País Vasco UPV/EHU; 19. Neurology Service, Hospital Universitario La Paz (UAM), Madrid, Spain; 20. Alzheimer Research Center & Memory Clinic. Andalusian Institute for Neuroscience. Málaga, Spain; 21. Neurology Service, Marqués de Valdecilla University Hospital (University of Cantabria and IDIVAL), Santander, Spain; 22. Hospital Donostia de San Sebastían; 23. Fundación para la Formación e Investigación Sanitarias de la Región de Murcia; 24. Servicio de Neurología -Hospital de Cabueñes-Gijón; 25. Centro de Investigación y Terapias Avanzadas. Fundación CITA-alzheimer; 26. Navarrabiomed; 27. Servicio de Neurología -Hospital Universitario Central de Asturias, Oviedo; 28. Barcelona βeta Brain Research Center – Fundació Pasqual Maragall; 29. Unitat de Genètica Molecular. Institut de Biomedicina de València-CSIC; 30. Unidad Mixta de Neurologia Genètica. Instituto de Investigación Sanitaria La Fe; 31. Unitat Trastorns Cognitius, Hospital Universitari Santa Maria de Lleida, Institut de Recerca Biomédica de Lleida (IRBLLeida), Lleida, España. 32. BT-CIEN; 33. Hospital Universitario La Princesa, Madrid, Spain; 34. Hospital Clínic Barcelona).

Supplementary Data

Supplementary data related to this article can be found at https://doi.org/10.1016/j.jalz.2019.06.4950.