Two Rare AKAP9 Variants Are Associated with Alzheimer's Disease in African Americans

Alzheimer’s & Dementia 10 (2014) 609-618

Two rare AKAP9 variants are associated with Alzheimer’s disease in African Americans

Mark W. Loguea,b,1, Matthew Schua,1, Badri N. Vardarajana, John Farrella, David A. Bennettc, Joseph D. Buxbaumd, Goldie S. Byrde, Nilufer Ertekin-Tanerf, Denis Evansg, Tatiana Foroudh, Alison Goatei, Neill R. Graff-Radfordf, M. Ilyas Kambohj, Walter A. Kukullk, Jennifer J. Manlyl, Alzheimer Disease Genetics Consortium2, Jonathan L. Hainesm, Richard Mayeuxl, Margaret A. Pericak-Vancen, Gerard D. Schellenbergo, Kathryn L. Lunettab, Clinton T. Baldwina, M. Daniele Fallinp, Lindsay A. Farrera,b,q,r,s,* aDepartment of Medicine (Biomedical Genetics), Boston University School of Medicine, Boston, MA, USA bDepartment of Biostatistics, Boston University School of Public Health, Boston, MA, USA cRush Alzheimer’s Disease Center, Rush University Medical Center, Chicago, IL, USA dDepartments of Neuroscience and Genetics & Genomic Sciences, Mount Sinai School of Medicine, New York, NY, USA eDepartment of Biology, North Carolina A & T State University, Greensboro, NC, USA fDepartments of Neuroscience and Neurology, Mayo Clinic, Jacksonville, FL, USA gRush Institute for Healthy Aging, Department of Internal Medicine, Rush University Medical Center, Chicago, IL, USA hDepartment of Medical and Molecular Genetics, Indiana University, Indianapolis, IN, USA iDepartment of Psychiatry and Hope Center Program on Protein Aggregation and Neurodegeneration, Washington University School of Medicine, St. Louis, MI, USA jDepartment of Human Genetics and Alzheimer’s Disease Research Center, University of Pittsburgh, Pittsburgh, PA, USA kNational Alzheimer’s Coordinating Center and Department of Epidemiology, University of Washington, Seattle, WA, USA lDepartment of Neurology and the Taub Institute, Columbia University, New York, NY, USA mDepartment of Epidemiology and Biostatistics, Case Western Reserve University, Cleveland, OH, USA nThe John P. Hussman Institute for Human Genomics, University of Miami, Miami, FL, USA oDepartment of Pathology and Laboratory Medicine, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA pDepartment of Epidemiology, Johns Hopkins University School of Public Health, Baltimore, MD, USA qDepartment of Neurology, Boston University School of Medicine, Boston, MA, USA rDepartment of Ophthalmology, Boston University School of Medicine, Boston, MA, USA sDepartment of Epidemiology, Boston University School of Public Health, Boston, MA, USA

Abstract Background: Less is known about the genetic basis of Alzheimer’s disease (AD) in African Amer- icans (AAs) than in non-Hispanic whites. Methods: Whole exome sequencing (WES) was performed on seven AA AD cases. Disease association with potentially AD-related variants from WES was assessed in an AA discovery cohort of 422 cases and 394 controls. Replication was sought in an AA sample of 1037 cases and 1869 controls from the Alzheimer Disease Genetics Consortium (ADGC). Results: Forty-four single nucleotide polymorphisms (SNPs) from WES passed filtering criteria and were successfully genotyped. Nominally significant (P , .05) association to AD was observed with two African-descent specific AKAP9 SNPs in tight linkage disequilibrium: rs144662445 (P 5 .014) and rs149979685 (P 5.037). These associations were replicated in the ADGC sample (rs144662445: P 5 .0022, odds ratio [OR] 5 2.75; rs149979685: P 5 .0022, OR 5 3.61).

1These two authors contributed equally to this work. *Corresponding author. 2See Acknowledgments for additional ADGC members who contributed to E-mail address: [email protected] this study.

Keywords: Whole-exome sequencing; Late-onset Alzheimer’s disease; Rare variant; Genetic association; African American; AKAP9

1. Background design to detect highly penetrant disease variants entails sequencing individuals from families with Late onset Alzheimer’s disease (AD) is the most common multiple affected members who are more likely than form of dementia in the elderly. It has a substantial genetic sporadic cases to possess such variants. Potentially component, with heritability as high as 75% [1]. The apoli- disease-causing variants identified in familial cases can poprotein E (APOE) gene ε4 variant has been shown to be a then be genotyped in other family members or in a larger major component of this risk [2]; among non-Hispanics sample of unrelated individuals to evaluate association whites (NHW), ε4/ε4 homozygotes are estimated to have with disease. This study design was recently used in up to 15-fold increased odds of AD compared with those Caucasian samples to identify rare variants in the phos- with the most common ε3/ε3 genotype [3]. Genome-wide pholipase D3 (PLD3)gene[18]. Resequencing PLD3 association studies (GWAS) including very large NHW sam- exons in a larger number of cases and controls identified ples have identified genes harboring additional modest- additional AD-associated PLD3 variants including effect AD risk susceptibility loci (odds ratios between 1.1 Ala442Ala, which was subsequently shown to be nomi- and 1.3) [4–8]. nally significantly associated (P 5 .03) with AD in a Interestingly, even though the proportion of risk due to modest AA sample (130 cases, 172 controls). In this genetic influences is similar in African-Americans (AA) study, we performed whole-exome sequencing (WES) and NHWs [9], there is evidence that the genetic architecture of seven familial AA late-onset AD cases and subse- of the disease differs between these two populations. For quently tested association of AD risk with 44 rare vari- APOE ε example, the effect of 4 on AD risk is considerably ants selected on the basis of frequency, sequencing weaker in AAs than NHWs [3,10,11], although this quality, and relationship to previously implicated AD difference may be accounted for in part by the age- ε genes in 422 AA cases and 394 cognitively normal AA dependent effect of 4 [11]. Several genes associated with controls. We identified two African-descent specific ge- AD and AD-related hippocampal atrophy measured by mag- netic variants in the A kinase anchor protein 9 netic resonance imaging in NHWs also show evidence of as- (AKAP9) gene which are significantly enriched in AA sociation in AAs, however, most SNPs associated with AD cases versus controls, and subsequently replicated NHWs have not been replicated in AAs [12–14]. While it this association in a large independent AA cohort. The may be that the AA studies are underpowered compared studydesignissummarizedinFigure 1. with studies with NHWs, it also may be due to differences in the linkage disequilibrium (LD) structure between AAs and NHWs or population-specific causal variants underlying 2. Methods the associations. Interestingly, two GWAS replicated the as- 2.1. Study samples sociation with ABCA7 in AAs [12,14] and the risk associated with ABCA7 (odds ratio [OR] 5 1.79) approaches that of The discovery sample included AA participants of the APOE ε4 in AAs (OR 5 2.31) [14]. Nevertheless, a large Multi-Institutional Research on Alzheimer’s Genetic Epide- proportion of heritability in AAs remains unexplained. miology (MIRAGE) Study and the Genetic and Environ- Much of the heritability may be ascribed to uncommon mental Risk Factors for Alzheimer’s Disease among (frequency ,5%) or rare (frequency ,1%) variants which African Americans Study (GenerAAtions). Ascertainment would be undetectable in GWAS except in very large sam- methods, inclusion criteria, and genome-wide genotyping ples [15]. Causal high-penetrant rare variants in APP, in the MIRAGE and GenerAAtions cohorts are described PSEN1, and PSEN2 have been identified primarily in the in detail elsewhere [9,12]. The MIRAGE cohort consists of early onset familial form of the disease [16]. However, to nuclear families most of which are discordant sib pairs. date no causal variants in these genes have been reported There are also several affected sib pairs and larger in AAs. sibships. Seven MIRAGE-Study AD cases with multiple Next-generation sequencing methodologies have the affected family members, and thus more likely to be en- potential to identify rare or private risk variants many riched with rare AD-related variants, were selected for of which would be expected to exert moderate to high ef- WES. Descriptive statistics for these subjects are summa- fects on disease risk compared with most common rized in Supplementary Table 1. Four of these subjects disease-associated variants [15,17]. One efficient study were affected sib-pairs. The genotypes of single nucleotide M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 611

Fig. 1. The overall study design including a whole-exome sequencing (WES) of seven African-American (AA) late-onset AD cases, followed by variant prioritization and filtering, genotyping in a larger AA discovery sample, and replication in a large AA Alzheimer’s Disease Genetics Consortium (ADGC) cohort. polymorphisms (SNPs) identified as potentially AD related 2.2. Sequencing and genotyping in WES were genotyped in 218 cases (including the seven sequenced subjects) and 237 controls from the MIRAGE A detailed description of WES, SNP genotyping cohort and 223 cases and 190 controls from the GenerAA- methods and quality control, and Sanger-sequencing used tions cohort. Analyses were performed on a discovery cohort to confirm genotypes of the most significant SNPs from made up of the genotyped subjects excluding the seven the analysis of the discovery data is provided in sequenced AD cases and subjects less than 60 years at the Supplementary Methods. time of assessment. The discovery cohort includes 199 cases and 209 controls from the MIRAGE cohort and 223 cases 2.3. Variant filtering and 185 controls from the GenerAAtions cohort. Character- istics of the discovery cohort are summarized in Variants identified by WES and passing quality control Supplementary Table 1. The most significant variants were (QC)werefilteredfornovelty(absenceinthedbSNP genotyped in a replication cohort comprising a subset of database v. 132, using time since discovery as a proxy the Alzheimer’s Disease Genetics Consortium (ADGC) for low frequency because rare SNPs are more likely to AA sample, excluding the MIRAGE and GenerAAtions co- be included in later dbSNP revisions) and potential for horts, which contains 1037 cases and 1869 controls deleterious effect (must be a nonsynonymous base pair (Supplementary Table 2). The ADGC AA cohort is change). To filter out possible sequencing artifacts, we described in detail elsewhere [14]. Studies were performed additionally required that variants be observed in more with appropriate institutional review board approval and than one case, whether the variants occurred in siblings informed consent was obtained from all study participants. or unrelated subjects. The novel non-synonymous variants 612 M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 were selected for genotyping if they were in genes which 3. Results had been either previously-identified as associated with AD—that is, genes that are listed in the AlzGene data- A total of 88,867 variants were identified by WES. No base—or in genes belonging to networks with established coding APP, PSEN1, PSEN2, or PLD3 variants were identi- AD genes. We constructed gene networks using the IN- fied in the seven AA AD cases. After filtering, 63 SNPs from GENUITY Pathway Analysis (IPA) software. IPA ana- 58 genes were selected for further study and 44 of these lyses were seeded with genes emerging from WES and SNPs were successfully genotyped in the entire sample. established risk genes for early or late onset AD Detailed results of sequencing and variant filtering are pre- (APOE, BIN1, CLU, CR1, PICALM, MS4A6E, CD2AP, sented in the supplementary materials, including a complete CD33, ABCA7, EPHA1, SORL1, ACE, PSEN1, PSEN2, list of variants that were genotyped (Supplementary and APP) [2,4–8,19–25] to identify genes and the Table 4). Most genotyped SNPs were rare (29 SNPs had a , variants contained therein that might be related MAF 1%). Nominally significant results were obtained 5 5 physiologically to AD-related processes or are related to with AKAP9 SNPs rs144662445 (P .014, OR 8.4) and 5 genes with a known connection to AD. These networks rs149979685 (P .037; the OR for rs149979685 could not do not exclusively consist of genes involved in a single be estimated because the rare allele was not observed in AD-related biological function, but instead include genes any controls) (Table 1, Supplementary Table 5). The geno- that interact, either directly or indirectly, with known AD types for these AKAP9 SNPs were confirmed by Sanger genes and hence may implicate new disease-related path- sequencing. Five AD cases, in addition to the two cases ways. included in WES, were observed with rare alleles for both rs144662445 and rs149979685 (Supplementary Table 6). Five other subjects (four AD cases and one control) had 2.4. Statistical analysis the minor allele for rs144662445 but not rs149979685. Sub- Tests of association between genotyped SNPs and AD jects having at least one of these two rare variants are desig- 1 were performed using R [26]. Significance was determined nated AKAP9 . No subjects were observed having the using a one-sided Fisher’s exact test because rare SNPs minor allele for rs149979685 but not rs144662445. No can cause numerical stability issues in logistic models, and rare-allele homozygotes were observed for either SNP. because we specified a priori that our search was focused One control subject who was genotyped but not included on deleterious rare variants. The test of association within in the discovery sample due to age—a 54-year-old unaf- the discovery cohort may underestimate the strength of a fected sibling of an AD proband (whose age at onset was true association in the MIRAGE data set because many of 64)—had rare alleles for both rs144662445 and 1 the controls in this sample are unaffected siblings of AD rs149979685. The AKAP9 subjects are all unrelated ac- cases who would presumably be carriers of risk alleles at a cording to their self-reported data and identity by descent higher rate than unrelated controls. Although the MIRAGE (IBD) estimates calculated by PLINK [31] using genome- study ascertained primarily discordant sib-pairs, the sample wide SNP data. A PC analysis in EIGENSTRAT [27] indi- contained three affected sib-pairs including the two affected cated that these subjects were not members of a clearly sib-pairs in the WES sample. Both members of the sib-pair defined population subgroup within AAs (Supplementary not included in the WES sample were included in the discov- Figure 1). AKAP9 genotypes were available for relatives of 1 ery cohort. The possibility of inflated significance due to four AKAP9 AD cases. Two of these cases each had an un- non-independence of the samples is unlikely (see the gener- affected sibling who was AKAP9-. Each of the other two 1 alized estimating equation [GEE] analysis in the AKAP9 cases who were included in the WES discovery Supplementary Results). Principal components (PC) anal- sample had an AKAP9-affected sibling. ysis with EIGENSTRAT [27] based on all minor allele fre- Next, we sought replication of the association only quency (MAF) .5% markers from a whole-genome with the two AKAP9 SNPs in the ADGC AA sample. genotyping panel (described in [12]) found no evidence of AD-associated population substructure within the discovery . Table 1 sample (all P .05 on tests of association between AD and The most significant associations between AD and genotyped SNPs in the each of the first 10 PCs). In the replication sample, SNPs discovery cohort were analyzed for association with AD using a logistic Gene Position SNP ID Freq cases Freq controls P regression model adjusting for sex and study site. Methods for the haplotype, phylogenetic dendogram, and bio- AKAP9* 7:91709085 rs144662445 0.011 0.0013 .014* AKAP9* 7:91732110 rs149979685 0.0060 0.000 .037* informatic analyses (using PolyPhen-2 [28], Sorts Intolerant ATP2B4 1:203672867 rs145963279 0.023 0.013 .092 From Tolerant substitutions [SIFT] [29], and MutPred [30] LRP4 11:46911956 rs138878258 0.0072 0.0026 .17 software) are described in Supplementary Methods. Haplo- ITGA9 3:37774225 rs142726080 0.032 0.024 .20 type analyses were performed with PLINK [31] using both Abbreviations: AD, Alzheimer’s disease; Freq, frequency; SNP, single- WES variant data and tag SNPs from whole-genome geno- nucleotide polymorphism. typing. *Genotypes for AKAP9 SNPs confirmed via Sanger sequencing. M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 613

Valid genotypes for both SNPs were obtained in 1037 AD icant (P 5.071). No other coding variants in this region were cases and 1869 controls. In this sample, the average age observed in the two AKAP91 subjects with WES data at onset among AD cases was 80.1 and the average exam- through whom this association was ascertained. ination age for controls was 74.5. Average ages were Next, we performed phylogenetic dendogram analysis of similar among AKAP91 (79.7) and AKAP2 (80.1) cases the SNP information in the haplotype region among 11 of the (P 5 .83) and among AKAP91 (75.9) and AKAP2 246 African-descent subjects in 1000 Genomes for whom (74.5) controls (P 5 .47). The APOE ε4 allele frequency genotypes are available (including 88 Yorubans from was 33.7% in cases and 18.5% in controls in the replica- Nigeria, 97 Luhyans from Kenya, and 61 AAs from the tion sample. This does not differ between AKAP91 American Southwest) and who had the HAP0 haplotype as (38.8%) and AKAP92 (33.6%) cases (P 5 .59) or be- defined by the common SNPs. Three of these 11 individuals tween AKAP91 (12.5%) and AKAP92 (18.6%) controls (NA18499, NA19448, and NA20126) had minor alleles at (P 5 .50). A logistic regression model adjusting for age both rs144662445 and rs149979685, and one individual and study site demonstrated significant association of (NA20298) had only the rs144662445 variant. These AD risk with the minor alleles of both rs144662445 AKAP9 variants were observed in each of the three (P 5 .0022, OR 5 2.75) and rs149979685 (P 5 .0022, African-descent groups suggesting that the AKAP9 variants OR 5 3.61) (Table 2). Hence, the association observed are present in multiple African ancestral populations. Clus- in the discovery data was replicated in a larger indepen- ter analysis of the phased alleles for the 22 haplotypes dent sample. A logistic regression analysis of the subset further discriminated the HAP01 haplotypes from the of ADGC AA replication subjects who had both HAP02 haplotypes (Figure 2A). There is also a branch point AKAP9 SNP genotypes and genome-wide genotype data separating the HAP01 haplotype containing one AKAP9 (877 cases and 1641 controls) yielded nearly identical re- variant (designated HAP011) from the haplotype contain- sults for both SNPs when PCs were included ing both AKAP9 variants (HAP012). (rs144662445 P 5 .0034, OR 5 2.72 versus P 5 .0035, Basedontheseresults,weexaminedthe1000Ge- OR 5 2.72 with PCs; 149979685 P 5 .0064, nomes phased haplotypes for plausible exonic candidate OR 5 3.17 versus P 5 .0063, OR 5 3.19 with PCs), indi- AD-risk SNPs on the same haplotypes as rs144662445 cating that population substructure did not contribute to and rs149979685 by comparing the high-risk haplotypes the confirmation of the AKAP9 association in the replica- (HAP012 and HAP011) to the low-risk haplotype tion sample. (HAP02).Withintheregionfrom91.3Mbto92.1 To examine the possibility that an untested AD causal Mb—which extends well beyond the regions shared variant in AKAP9 or surrounding region could account for across those harboring AKAP9 mutations (Figure 2B)— the observed association, we compared haplotypes of the three subjects with HAP012 haplotypes shared 69 AKAP91 and AKAP92 individuals in the discovery sample. variants with MAF ,10% that were not observed in sub- The 50 end of AKAP9 overlaps a large LD block which in- jects with the HAP02 haplotype. Rs144662445 and cludes the neighboring genes CYP51A1, KRIT1, ANKIB1, rs149979685 are the only coding variants in this group and LRRD1 (see Supplementary Figure 2). Haplotype anal- of SNPs. Next, we compared the single copy of ysis of the 10 common (MAF .0.05) AKAP9 SNPs in the HAP011from1000Genomes(NA20298A)toHAP02. GWAS-genotyping array revealed that all of the AKAP91 In the same region, NA20298A contains 49 variants subjects harbored the same haplotype (designated HAP0). with MAF ,10%, which were not observed on the That is, the minor alleles (either as a singleton or together) (low-risk) HAP02 haplotypes, including three coding are present on the haplotype GGAAGGAAGC as defined variants: rs144662445 and two additional nonsynonymous by rs1859037, rs2299233, rs2282973, rs6465347, coding AKAP9 SNPs rs186619641 and rs201858518. rs2158138, rs733957, rs2079082, rs13239875, rs4265, and Rs186619641 and rs201858518 are unique to individual rs1063243 (Supplementary Table 7). The frequency of this NA20298 in the 1000 Genomes database. Using only haplotype is greater in AD cases (6.1%) compared with un- 1000 Genomes data we cannot determine whether these related controls (4.1%), but this difference is not quite signif- two additional nonsynonymous variants are shared by

Table 2 The association between AKAP9 variants and AD using a logistic regression adjusted for sex and study site in the replication cohort Frequency (proportion) of minor alleles SNP ID Cases Controls OR* 95% CI P rs144662445 0.011 (22y/2052) 0.0043 (16/3722) 2.75 1.38–5.61 .0022 rs149979685 0.0072 (15/1059) 0.0027 (10/3728) 3.61 1.51–9.00 .0022 Abbreviations: AD, Alzheimer’s disease; CI, conﬁdence interval; OR, odds ratio; SNP, single nucleotide polymorphism. *Effect size estimates, CI, and P values were derived from a logistic regression model adjusted for sex and study site. yIncludes one AD case who was homozygous for the minor allele. 614 M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 615

Table 3 Bioinformatic Examination of the AD-associated nonsynonymous AKAP9 variants Polyphen-2 SIFT MutPred SNP Score* Prediction Scorey Prediction Probability deleteriousz Actionable hypothesesz rs144662445 0.008 Benign 0.11 Tolerated 21% NA rs149979685 0.99 Probably damaging 0.01 Damaging 57% Loss of helix (P 5 .0033) Gain of loop (P 5 .024) Loss of phosphorylation (P 5 .029) Abbreviations: AD, Alzheimer’s disease; NA, not applicable; SNP, single nucleotide polymorphism; SIFT, Sorts Intolerant from Tolerant substitutions. *In Polyphen-2 higher scores indicate that a SNP is more likely to be damaging. yIn SIFT, lower scores indicate that a SNP is more likely to be deleterious. zIn MutPred, a higher probability indicates that a SNP is more likely to be damaging. Actionable hypotheses are predicted functional changes for which the probability deleterious is ,50% and for which the structural change P , .05.

all subjects in our study with HAP011, present in only a common background haplotype (HAP0) spanning AKAP9 portion of subjects with HAP011 haplotypes, or unique and the surrounding area. These SNPs were significantly en- to 1000 Genomes individual NA20298 who is the only riched in AD cases compared with cognitively normal con- AA subject in 1000 Genomes with one HAP011haplo- trols in a large African-American sample, and the type. We sought clarification in the ESP database which, associations were replicated in a larger independent AA based on allele counts of the AD-associated SNPs cohort. These missense variants are rare in African-descent rs144662445 and rs149979685, contains 3 copies of populations (ESP AA sample: MAFrs144662445 5 0.43% HAP011. These two exonic SNPs (rs186619641 and and MAFrs149979685 5 0.36%; 1000 Genomes AFR sample: rs201858518) were absent from the ESP database sug- MAFrs144662445 5 0.81% and MAFrs149979685 5 0.61%). gesting that the two AD-associated variants defining They are not present in approximately 4000 Caucasians in HAP011 and HAP012 are not surrogate markers for the ESP database or in currently available data for 379 Euro- rs186619641 and rs201858518. Therefore, based on our pean and 286 East-Asian subjects in the 1000 Genomes data- examination of HAP011andHAP012inpublicdata- base. Although the estimated OR’s are lower in the bases, we conclude that rs144662445 and rs149979685 replication sample compared with the discovery sample are the only plausible AD-related coding variants on these (as expected due to ‘winner’s curse’), the magnitude of ef- background haplotypes within AKAP9 or the surrounding fect of these variants on AD risk in the replication sample region. (estimated ORs of 2.75 and 3.61) is slightly larger than Bioinformatic analysis using several methods suggests that reported for APOE ε4 in AAs (OR 5 2.31) [14]. Hence, that rs149979685 (S3771L transcript Q99996-1) is likely as expected given their frequency, the identified AKAP9 risk to be functional whereas rs144662445 (I2558M transcript variants not likely to explain a great proportion of AD risk in Q99996-1) is not (Table 3). The MutPred analysis also indi- AAs. However, those inheriting at least one risk allele have a cated that the most likely impact of the rs149979685 variant substantially increased risk of AD. Bioinformatic analysis is loss of a helix (P 5.0033), but gain of a loop (P 5.024) or indicated that one of these SNPs (rs149979685) likely has loss of phosphorylation (P 5.029) are also possible. Howev- an adverse impact on the structure or post-translational er, because MutPred makes these predictions based on ho- modification of the AKAP protein. mology to known functional variants and the 3D structure We observed many differences in the 1000 Genomes data- of AKAP9 is yet unknown, it is unclear whether these struc- base between the HAP0 haplotype containing the AD- tures are present near rs149979685 or critical to the function associated AKAP9 SNPs and HAP0 haplotypes lacking both of AKAP9. The two NA20298-specific non-synonymous mutations, suggesting that the instances of the rare SNPs (rs186619641 and rs201858518) were also predicted rs149979685 and rs144662445 variants are likely identical to be benign according to Polyphen-2. by descent, although derived from a very old mutational event. This conclusion is supported by the observation of these vari- 4. Discussion ants in subjects of diverse African ancestry and by evidence in our study that subjects sharing these alleles do not appear to be In a small WES study of seven familial AD cases, we related or from a particular population subgroup based on identified two AKAP9 SNPs which co-occur on a single un- genome-wide measures of IBD sharing and PC analysis. = Fig. 2. Examination of the 1000 Genomes subjects with the haplotype containing the two identified AKAP9 variants including (A) a phylogenetic dendogram of haplotypes and (B) positions of infrequent single nucleotide polymorphisms (SNPs; 5% minor allele frequency) shared by haplotypes with 0, 1, and 2 variants. Key: HAP0: The background haplotype which contains the two identified AKAP9 variants. HAP01: the background variant with one or more of the two pu- tatively Alzheimer’s disease (AD)-associated variants; HAP02: the background haplotype without either AD-associated variants; HAP011 the background haplotype with one of the AD-associated AKAP9 variants; HAP012 the background haplotype with both AD-associated AKAP9 variants. 616 M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618

AKAP9, located on chromosome 7q21-22, encodes a cation and cell cycle progression [37]. Studies have also member of the A kinase anchoring protein (AKAP) fam- shown that AKAP450 is involved in microtubule anchoring ily. AKAPs bind or tether protein kinase A (PKA) and and organization at the centrosome [38] and necessary for other signaling molecules to relevant targets [32]. the initiation of new microtubule formation on the cis-side AKAP9 is differentially spliced into isoforms which of the golgi [39]. The non-synonymous change induced by localize to different cellular compartments [32] and are rs144662445 (I2558M transcript Q99996-1) is 5 amino acids involved in distinct biological processes. AKAP9 gene from the R2 binding site of AKAP450 (Figure 3). products include a short isoform called Yotiao and long Rs149979685 (S3771L transcript Q99996-1) is located isoforms (AKAP350, AKAP450, and CG-NAP) often within the pericentrin/AKAP450 centrosomal targeting referred to collectively as AKAP450 based on their mo- domain which is a highly conserved region near the 30 end lecular weight. of AKAP450 and is the region responsible for the localiza- AKAP9 was identified in a cDNA library of human brain tion of AKAP450 to the centrosome [40] (Figure 3). The where the Yotiao isoform is expressed in the hippocampus, role of AKAP450 in AD is unclear, however, it has func- cerebellum, and cerebral cortex [33]. A SNP in Yotiao has tional similarity with tau protein which is involved in micro- been identified as a cause of long-QT syndrome [34],a tubule stability and assembly and is a key constituent of congenital heart abnormality characterized by long-QT in- neurofibrillary tangles which accumulate in AD brain [41– tervals and arrhythmias. Disruption of the binding of PKA 44]. Reports also indicate that phosphorylated tau does not and AKAPs (including AKAP5 and Yotiao) at nerve termi- account for all of the microtubule loss and shortening nals in the hippocampus has been shown to interfere with observed in neurons of those with AD [45]. Thus, it is cellular mechanisms associated with spatial memory [35]. possible that functional defects in AKAP450 may contribute The locations of rs144662445 and rs149979685 are approx- to AD pathogenesis. imately 38 kb 30 and 61 kb 30, respectively, from the bound- These results should be interpreted cautiously. The bio- aries of the Yotiao coding region. Therefore, these variants informatics programs have predicted that rs149979685 has probably do not affect the Yotiao isoform, unless they are a possible function while rs144662445 does not. However, regulatory or simply in LD with unidentified non-coding without functional data and supporting molecular work, we risk variants. Hence it is more likely that the AD- cannot definitively demonstrate that rs149979685 is the associated AKAP9 variants affect the structure or function causal locus or whether rs144662445 is potentially causal of AKAP450. due to some other mechanism not accounted for in the soft- AKAP450 localizes to centrosomes of cells [36]. It is ex- ware used, or even whether another non-coding regulatory pressed in most tissues including brain. Disrupted binding of variant in LD with both rs144662445 and rs149979685 is AKAP450 to the centrosome interferes with centriole dupli- responsible for the observed association with AD.

Fig. 3. Annotated protein domain of AKAP9 canonical transcript Q99996-1 including Yotiao; AKAP450; the PKA binding sites; the pericentrin/AKAP450 centrosomal targeting (PACT) domain; Long QT syndrome amino-acid change; and the conservation of the two Alzheimer’s disease (AD)-associated variants rs144662445 (I2558M) and rs149979685 (S3771L). M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 617

However, examination of the region in 1000 Genomes and PLINK:http://pngu.mgh.harvard.edu/wpurcell/plink/;NH the ESP database indicates that such a variant is not likely LBI Grand Opportunity Exome Sequencing Project (ESP): to be a non-synonymous coding change. Resolution of this https://esp.gs.washington.edu/drupal/. question will require functional experimentation. Finally, we acknowledge that our findings do not represent the breadth of possible AD-related rare variation in African RESEARCH IN CONTEXT Americans because most such variants would not be present in the exons of seven individuals. Thus, for example, it is not surprising that we did not observe previously reported rare AD-risk variants in PLD3 [18].Itisalso 1. Systematic review: As members of the Alzheimer’s possible that several of the non-AKAP9 variants identified Disease Genetics Consortium, the authors have by WES and selected for genotyping are actually true risk been privy to much of the relevant literature as it de- loci which the discovery sample was underpowered to velops. PubMed and Google searches were used to confirm by association. Targeted resequencing in both Af- ensure that other relevant publications were not over- rican- and European-descent populations will be necessary looked. to identify the full complement of genetic variants which 2. Interpretation: This is the third report of rare Alz- will provide greater understanding of the role of AKAP9 heimer’s disease (AD) risk variants identified by in AD pathogenesis. whole-exome sequencing (WES) and the first report in African Americans (AAs). Moreover, the Acknowledgments estimated magnitude of effect of these variants is slightly larger than estimated risk due to APOE ε4in We gratefully acknowledge the assistance of Madison Steele AAs. This study indicates a potential new mecha- and Alison Peltz, participants in Boston University’s nism for AD. Research Internship in Science & Engineering (RISE) Pro- gram. 3. Future directions: One of the identified rare variants Additional Members of the Alzheimer’s Disease Genetics is predicted to be deleterious, but whether this variant Consortium Who Contributed to This Study: Lisa L. Barnes, is causal will need to be confirmed in functional PhD, Guiqing Cai, PhD, Laura B. Cantwell, MPH, Philip L. studies. Additionally, Sequencing in AA and other De Jager, MD, PhD, Rodney C. P. Go, PhD, Patrick Griffith, populations will be necessary to find the complete MD, Rosalyn Lang, PhD, Oscar L. Lopez, MD, Thomas O. set of AD-associated AKAP9 variants. Obisesan, MD, Towfique Raj, PhD. Funding Support: This work supported by NIA/NIH grants R01-AG09029, R01-AG025259, P30-AG13846 (Dr. Farrer); U01-AG032984, RC2-AG036528, U01-AG016976 (Dr. Ku- kull); U24-AG026395, U24-AG026390; R01-AG037212, References R37-AG015473 (Dr. Mayeux); U24-AG021886 (Dr. Foroud); R01-AG20688 (Dr. Fallin); P50-AG005133, AG041718, [1] Bienvenu OJ, Davydow DS, Kendler KS. Psychiatric ’diseases’ versus AG030653 (Dr. Kamboh); R01-AG019085 (Dr. Haines); behavioral disorders and degree of genetic influence. Psychol Med 2011;41:33–40. R01-AG1101, R01-AG030146, RC2-AG036650 (Dr. Evans); [2] Saunders AM, Strittmatter WJ, Schmechel D, George-Hyslop PH, P30-AG10161, R01-AG15819, R01-AG17917 (Dr. Bennett); Pericak-Vance MA, Joo SH, et al. Association of apolipoprotein E R01AG028786 (Dr. Manly); R01-AG22018, P30-AG10161 allele epsilon 4 with late-onset familial and sporadic Alzheimer’s dis- (Dr. Barnes); P50AG-16574, R01-032990, KL2-RR024151 ease. Neurology 1993;43:1467–72. (Dr. Ertekin-Taner); R01-AG027944, R01-AG028786 (Dr. [3] Farrer LA, Cupples LA, Haines JL, Hyman B, Kukull WA, Mayeux R, et al. Effects of age, sex, and ethnicity on the association between Pericak-Vance); P20-MD000546, R01-AG28786 (Dr. Byrd); apolipoprotein E genotype and Alzheimer disease. A meta-analysis. AG005138 (Dr. Buxbaum); P50-AG05681, P01-AG03991, APOE and Alzheimer Disease Meta Analysis Consortium. JAMA P01-AG026276 (Dr. Goate); and UO1-AG032984 (Dr. Schel- 1997;278:1349–56. lenberg). [4] Lambert JC, Heath S, Even G, Campion D, Sleegers K, Hiltunen M, Conflicts of Interest: The authors of this paper have no con- et al. Genome-wide association study identifies variants at CLU and CR1 associated with Alzheimer’s disease. Nat Genet 2009;41:1094–9. flicts of interest to report. [5] Seshadri S, Fitzpatrick AL, Ikram MA, DeStefano AL, Gudnason V, Web Resources: AlzGene database: http://www.alzgene.org; Boada M, et al. Genome-wide analysis of genetic loci associated INGENUITY Pathway Analysis: http://www.ingenuity. with Alzheimer disease. JAMA 2010;303:1832–40. com; Applied Biosystems’ Sequence Scanner Software [6] Harold D, Abraham R, Hollingworth P, Sims R, Gerrish A, v1.0: https://products.appliedbiosystems.com/ab/en/US/ Hamshere ML, et al. Genome-wide association study identifies vari- 5 5 ants at CLU and PICALM associated with Alzheimer’s disease. Nat adirect/ab?cmd catNavigate2&catID 600583; The Genet 2009;41:1088–93. R project for stastistical computing: http://www.r-project. [7] Naj AC, Jun G, Beecham GW, Wang LS, Vardarajan BN, Buros J, et al. org; 1000 Genomes Project: www.1000genomes.org; Common variants at MS4A4/MS4A6E, CD2AP, CD33 and EPHA1 618 M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618

are associated with late-onset Alzheimer’s disease. Nat Genet 2011; [26] R Development Core Team. R: A language and environment for statis- 43:436–41. tical computing. The R Foundation for Statistical Computing; Vienna, [8] Hollingworth P, Harold D, Sims R, Gerrish A, Lambert JC, Austria; 2011. Carrasquillo MM, et al. Common variants at ABCA7, MS4A6A/ [27] Price AL, Patterson NJ, Plenge RM, Weinblatt ME, Shadick NA, MS4A4E, EPHA1, CD33 and CD2AP are associated with Alzheimer’s Reich D. Principal components analysis corrects for stratification in disease. Nat Genet 2011;43:429–35. genome-wide association studies. Nat Genet 2006;38:904–9. [9] Green RC, Cupples LA, Go R, Benke KS, Edeki T, Griffith PA, et al. [28] Adzhubei IA, Schmidt S, Peshkin L, Ramensky VE, Gerasimova A, Risk of dementia among white and African American relatives of pa- Bork P, et al. A method and server for predicting damaging missense tients with Alzheimer disease. JAMA 2002;287:329–36. mutations. Nat Methods 2010;7:248–9. [10] Tang MX, Stern Y, Marder K, Bell K, Gurland B, Lantigua R, et al. [29] Ng PC, Henikoff S. Predicting deleterious amino acid substitutions. The APOE-epsilon4 allele and the risk of Alzheimer disease Genome Res 2001;11:863–74. among African Americans, whites, and Hispanics. JAMA 1998; [30] Li B, Krishnan VG, Mort ME, Xin F, Kamati KK, Cooper DN, et al. 279:751–5. Automated inference of molecular mechanisms of disease from amino [11] Graff-Radford NR, Green RC, Go RC, Hutton ML, Edeki T, acid substitutions. Bioinformatics 2009;25:2744–50. Bachman D, et al. Association between apolipoprotein E genotype [31] Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, Bender D, and Alzheimer disease in African American subjects. Arch Neurol et al. PLINK: a tool set for whole-genome association and population- 2002;59:594–600. based linkage analyses. Am J Hum Genet 2007;81:559–75. [12] Logue MW, Schu M, Vardarajan BN, Buros J, Green RC, Go RC, et al. [32] Wong W, Scott JD. AKAP signalling complexes: focal points in space A comprehensive genetic association study of Alzheimer disease in and time. Nat Rev Mol Cell Biol 2004;5:959–70. African Americans. Arch Neurol 2011;68:1569–79. [33] Lin JW, Wyszynski M, Madhavan R, Sealock R, Kim JU, Sheng M. [13] Melville SA, Buros J, Parrado AR, Vardarajan B, Logue MW, Shen L, Yotiao, a novel protein of neuromuscular junction and brain that inter- et al. Multiple loci influencing hippocampal degeneration identified by acts with specific splice variants of NMDA receptor subunit NR1. J genome scan. Ann Neurol 2012;72:65–75. Neurosci 1998;18:2017–27. [14] Reitz C, Jun G, Naj A, Rajbhandary R, Vardarajan BN, Wang LS, et al. [34] Chen L, Marquardt ML, Tester DJ, Sampson KJ, Ackerman MJ, Variants in the ATP-binding cassette transporter (ABCA7), apolipo- Kass RS. Mutation of an A-kinase-anchoring protein causes long- protein E 4,and the risk of late-onset Alzheimer disease in African QT syndrome. Proc Natl Acad Sci U S A 2007;104:20990–5. Americans. JAMA 2013;309:1483–92. [35] Nie T, McDonough CB, Huang T, Nguyen PV, Abel T. Genetic disrup- [15] Cirulli ET, Goldstein DB. Uncovering the roles of rare variants in com- tion of protein kinase A anchoring reveals a role for compartmental- mon disease through whole-genome sequencing. Nat Rev Genet 2010; ized kinase signaling in theta-burst long-term potentiation and 11:415–25. spatial memory. J Neurosci 2007;27:10278–88. [16] Bird TD. Genetic aspects of Alzheimer disease. Genet Med 2008; [36] Witczak O, Skalhegg BS, Keryer G, Bornens M, Tasken K, Jahnsen T, 10:231–9. et al. Cloning and characterization of a cDNA encoding an A-kinase [17] Schork NJ, Murray SS, Frazer KA, Topol EJ. Common vs. rare allele anchoring protein located in the centrosome, AKAP450. Embo J hypotheses for complex diseases. Curr Opin Genet Dev 2009; 1999;18:1858–68. 19:212–9. [37] Keryer G, Witczak O, Delouvee A, Kemmner WA, Rouillard D, [18] Cruchaga C, Karch CM, Jin SC, Benitez BA, Cai Y, Guerreiro R, et al. Tasken K, et al. Dissociating the centrosomal matrix protein Rare coding variants in the phospholipase D3 gene confer risk for Alz- AKAP450 from centrioles impairs centriole duplication and cell cycle heimer’s disease. Nature 2014;505:550–4. progression. Mol Biol Cell 2003;14:2436–46. [19] Rogaeva E, Meng Y, Lee JH, Gu Y, Kawarai T, Zou F, et al. The [38] Keryer G, Di Fiore B, Celati C, Lechtreck KF, Mogensen M, neuronal sortilin-related receptor SORL1 is genetically associated Delouvee A, et al. Part of Ran is associated with AKAP450 at the with Alzheimer disease. Nat Genet 2007;39:168–77. centrosome: involvement in microtubule-organizing activity. Mol [20] Reitz C, Cheng R, Rogaeva E, Lee JH, Tokuhiro S, Zou F, et al. Meta- Biol Cell 2003;14:4260–71. analysis of the association between variants in SORL1 and Alzheimer [39] Rivero S, Cardenas J, Bornens M, Rios RM. Microtubule nucleation at disease. Arch Neurol 2011;68:99–106. the cis-side of the Golgi apparatus requires AKAP450 and GM130. [21] Lehmann DJ, Cortina-Borja M, Warden DR, Smith AD, Sleegers K, Embo J 2009;28:1016–28. Prince JA, et al. Large meta-analysis establishes the ACE insertion- [40] Gillingham AK, Munro S. The PACT domain, a conserved centroso- deletion polymorphism as a marker of Alzheimer’s disease. Am J Epi- mal targeting motif in the coiled-coil proteins AKAP450 and pericen- demiol 2005;162:305–17. trin. EMBO Rep 2000;1:524–9. [22] Meng Y, Baldwin CT, Bowirrat A, Waraska K, Inzelberg R, [41] Grundke-Iqbal I, Iqbal K, Quinlan M, Tung YC, Zaidi MS, Friedland RP, et al. Association of polymorphisms in the Wisniewski HM. Microtubule-associated protein tau. A component Angiotensin-converting enzyme gene with Alzheimer disease in an Is- of Alzheimer paired helical filaments. J Biol Chem 1986;261:6084–9. raeli Arab community. Am J Hum Genet 2006;78:871–7. [42] Grundke-Iqbal I, Iqbal K, Tung YC, Quinlan M, Wisniewski HM, [23] Rogaev EI, Sherrington R, Rogaeva EA, Levesque G, Ikeda M, Binder LI. Abnormal phosphorylation of the microtubule-associated Liang Y, et al. Familial Alzheimer’s disease in kindreds with missense protein tau (tau) in Alzheimer cytoskeletal pathology. Proc Natl mutations in a gene on chromosome 1 related to the Alzheimer’s dis- Acad Sci U S A 1986;83:4913–7. ease type 3 gene. Nature 1995;376:775–8. [43] Iqbal K, Grundke-Iqbal I, Zaidi T, Merz PA, Wen GY, Shaikh SS, et al. [24] Sherrington R, Rogaev EI, Liang Y, Rogaeva EA, Levesque G, Defective brain microtubule assembly in Alzheimer’s disease. Lancet Ikeda M, et al. Cloning of a gene bearing missense mutations 1986;2:421–6. in early-onset familial Alzheimer’s disease. Nature 1995; [44] Yoshida H, Ihara Y. Tau in paired helical filaments is functionally 375:754–60. distinct from fetal tau: assembly incompetence of paired helical fila- [25] Goate A, Chartier-Harlin MC, Mullan M, Brown J, Crawford F, ment-tau. J Neurochem 1993;61:1183–6. Fidani L, et al. Segregation of a missense mutation in the amyloid pre- [45] Cash AD, Aliev G, Siedlak SL, Nunomura A, Fujioka H, Zhu X, et al. cursor protein gene with familial Alzheimer’s disease. Nature 1991; Microtubule reduction in Alzheimer’s disease and aging is indepen- 349:704–6. dent of tau filament formation. Am J Pathol 2003;162:1623–7. M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 618.e1

Supplementary online content sequencing were determined blind to the initial genotype calls and to AD status. Sanger sequencing was performed for all Two rare AKAP9 missense variants are associated with MIRAGE and GenerAAtions subjects which had been identi- Alzheimer disease in African Americans fied as having the minor allele at one of these SNPs (n 5 24). This included subjects whose genotypes had been excluded Supplementary methods from other analyses because of a high genotype-missingness rate. Additionally we genotyped any subject who was pre- Whole-exome sequencing and quality control dicted to have the background haplotype H0 but for whom ge- Whole-exome capture was performed using the Agilent notypes were missing or had been excluded due to high SureSelect Human All Exon kit. Sequencing was performed missing rate (n 5 13) and several other control subjects (4) using the Illumina GA IIx platform. The alignment of the who were identified as having H0 but whose genotype was 80 bp single-end reads was completed with the Burrows- the homozygous reference at each. In total, 41 samples Wheeler Aligner algorithm [1,2].Singlenucleotide were identified to be genotyped using Sanger sequencing. polymorphism (SNP) discovery and quality filtering was Two subjects had Sanger sequencing which failed for one or completed using a pipeline based on the Genome Analysis both SNPs and were excluded from further analysis. Toolkit [3,4]. Single nucleotide variants (SNVs) meeting initial quality control (QC) criteria had a minimum depth of Haplotype and bioinformatic analyses 10 reads, mapping quality scores of at least 30, and depth Haplotype inference, haplotype frequency estimation, and quality (QD) scores of at least 5. While these parameters haplotype association tests were performed using PLINK (v. aresomewhatliberal,onlySNVsobservedinmorethanone 1.07, October 2009) [6]. Haplotypes were generated using sequenced Alzheimer’s disease (AD)-case (whether they both rare-SNP genotypes determined in this study and geno- occurred within a sibling pair) were considered for further types for common SNPs obtained previously for the same in- analysis to filter out additional possible sequencing artifacts. dividuals [5]. Haplotypes were also generated for subjects This strategy likely increased the selection of AD-related var- included in release 1 of the 1000 Genomes Project using iants because sequencing was performed in cases only. PLINK [7]. Regional LD was assessed using Haploview [8] employing publically available HapMap2 reference panels Genotyping for the Yoruban (YRI) sample. Phylogenetic dendograms SNPs from whole-exome sequencing that were selected were generated from of 1000 Genomes Project phased haplo- for association testing were genotyped using KASP allele- types using R [9], first by measuring the similarity of the hap- specific PCR-based assays from LGC Genomics (Hoddes- lotypes using the dist function and then by performing don UK, Beverly MA) and an Applied Biosystems 7900 hierarchical clustering of the distances using the hclust func- Real-Time PCR System with 384-Well Block Module. Per- tion. Genomic positions were fixed according to the hg19 formance of the assays was validated by comparing geno- (February 2009) human genome assembly. The functionality types derived from sequencing and SNP genotyping for of associated SNPs was assessed using bioinformatic ap- the seven whole-exome sequenced individuals. Subjects proaches implemented in software packages including with a call rate of less than 90% were excluded from further PolyPhen-2 [10] which is a Bayesian probability function analysis, except for analyses using genotypes that were developed via a machine learning algorithm, SIFT which ex- confirmed using Sanger sequencing. Methods used to amines conservation of elements across similar sequences to generate genome-wide tag SNP data used for haplotype identify whether mutations are potentially deleterious [11], analysis of the Multi-Institutional Research on Alzheimer and MutPred which models the change induced by a mutation Genetic Epidemiology Study (MIRAGE) and Genetic and by comparing it to the original sequence and produces a prob- Environmental Risk Factors for Alzheimer Disease among ability of a change of structure or function [12]. African Americans Study (GenerAAtions) cohorts in PLINK and PC analysis in EIGENSTRAT are described in Replication analyses Logue et al. 2011 [5]. The replication data consisted of a subset of the Alz- heimer Disease Genetics Consortium African-American Sanger sequencing (AA) sample [13] excluding those subjects from the PCR was performed in a three step cycle repeated 35 times MIRAGE/GenerAAtions cohort. Specifically, genotypes using 100 mg of DNA and amplification primers using the were generated for rs144662445 and rs149979685 in avail- Qiagen Hot Start Tag polymerase. Custom primers were de- able AA samples from the Chicago Health and Aging Project signed using PrimerQuest software and supplied by Integrated [14,15], the four different sites of the African American DNA technology (see Supplementary Table 3 for details). Ge- Alzheimer’s Disease Genetics Study (Columbia University, notypes for which only the reverse or forward read was avail- University of Miami, North Carolina A&T State able were excluded. The same primers were used for both the University, and Vanderbilt) [16], a subset of unrelated indi- PCR and Sanger sequencing reactions. The results of the viduals from the National Institute on Aging Late-Onset experiment were examined using Applied Biosystems’ Alzheimer’s Disease study [17], the NIA Alzheimer’s Dis- Sequence Scanner Software v1.0. Genotypes from Sanger ease Centers [18], the Mayo Clinic [19], The Mount Sinai 618.e2 M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618

School of Medicine Brain Bank [20], the Rush University models, our initial screening analysis was limited to a Fisher’s Alzheimer’s disease Center [21–23], the University of exact test for association. However, after this initial screening, Miami/Vanderbilt University study [24–26], the University we explored the possibility of other sources of confounding of Pittsburgh [27], the Washington Heights Inwood bias. Because the more frequent AKAP9 variant Columbia Aging Project [28], and Washington University rs144662445 was observed in at least one control in the [29–32]. Genotyping of rs144662445 and rs149979685 MIRAGE/GenerAAtions data set, it was numerically possible was performed using TaqMan assays and an Applied to fit logistic GEE models [33,34] for this SNP. GEE allows Biosystems 7900 Real-Time PCR System with 384-Well for the inclusion of singletons and family data, is robust to Block Module. Only subjects for whom both SNPs were suc- misspecification of the correlation model, and was sufficient cessfully genotyped were included in the analysis. To be to prevent inflation on a genome-wide investigation of AD consistent with the discovery analysis, subjects were in these samples [5]. In a GEE model which included covari- excluded if their age was less than 60 years at diagnosis ates adjusting for age, sex, and study (MIRAGE/GenerAA- (cases) or at last examination (controls). Subjects who tions), the P-value for the rs144662445 covariate were diagnosed with other forms of dementia or mild cogni- approximated the P-value reported in Table 1 which was tive impairment were also excluded. The final sample size not adjusted for covariates (P 5 .022) and the OR estimate was 1037 cases and 1869 controls (Supplementary Table 2). was nearly identical (OR 5 8.3). Thus, age, non- independence of samples, sex, and study of origin do not Supplementary results appear to be confounding the association with rs144662445. These factors are also not likely confounding the results for Exome sequencing and variant filtering rs149979685 because these SNPs are in high LD. The success rate for aligning the 80 bp sequences from the seven subjects was 96.3% with 17.5x mean coverage Supplementary References within target regions. Before filtering, 88,867 SNPs were identified and the transition:transversion (TI/TV) ratio for [1] Li H, Durbin R. Fast and accurate short read alignment with Burrows- these SNPs was 2.8. A total of 81,335 SNPs after filtering Wheeler transform. Bioinformatics 2009;25:1754–60. for quality (read depth .10, QD Score .5, Map quality [2] Li H, Durbin R. Fast and accurate long-read alignment with Burrows- . Wheeler transform. Bioinformatics 2010;26:589–95. score 30). This number was reduced to 50,904 after [3] McKenna A, Hanna M, Banks E, Sivachenko A, Cibulskis K, filtering for those present in at least two subjects. The TI/ Kernytsky A, et al. The Genome Analysis Toolkit: a MapReduce TV ratio for these SNPs was 2.9, which approximates the ex- framework for analyzing next-generation DNA sequencing data. pected value for exonic SNPs. Filtering SNPs that were ab- Genome Res 2010;20:1297–303. sent in dbSNP build 132 yielded 1364 SNPs. Of these, 431 [4] DePristo MA, Banks E, Poplin R, Garimella KV, Maguire JR, Hartl C, et al. A framework for variation discovery and genotyping using next- were also non-synonymous. Only one of these SNPs is in a generation DNA sequencing data. Nat Genet 2011;43:491–8. gene with replicated association to AD: rs149704584 in [5] Logue MW, Schu M, Vardarajan BN, Buros J, Green RC, Go RC, et al. CD33. Of the 431 that passed cleaning and filtering criteria, A comprehensive genetic association study of Alzheimer disease in a total of 63 SNPs from 58 genes deemed most likely to be African Americans. Arch Neurol 2011;68:1569–79. functionally related to AD were genotyped and tested for as- [6] Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, Bender D, et al. PLINK: a tool set for whole-genome association and population- sociation. Twelve of these SNPs were in genes listed in Alz- based linkage analyses. Am J Hum Genet 2007;81:559–75. Gene. The remainder were from three Ingenuity Pathway [7] The 1000 Genomes Project Consortium. A map of human genome Analysis (IPA)-generated gene networks having genes variation from population-scale sequencing. Nature 2010; involved in a variety of biological functions including 467:1061–73. cellular assembly and organization, nervous system develop- [8] Barrett JC, Fry B, Maller J, Daly MJ. Haploview: analysis and visual- ization of LD and haplotype maps. Bioinformatics 2005;21:263–5. ment and function, carbohydrate metabolism, cell-to-cell [9] R Development Core Team. R: A language and environment for statis- signaling and interaction, hematological system develop- tical computing. The R Foundation for Statistical Computing; Vienna, ment and function, renal and urological system development Austria; 2011. and function, and cellular compromise (Supplementary [10] Adzhubei IA, Schmidt S, Peshkin L, Ramensky VE, Gerasimova A, Figure 3). Forty-six of these variants were identified previ- Bork P, et al. A method and server for predicting damaging missense mutations. Nat Methods 2010;7:248–9. ously (i.e., had assigned rsIDs in dbSNP build 135 even [11] Ng PC, Henikoff S. Predicting deleterious amino acid substitutions. though they were absent according to build 132) and had Genome Res 2001;11:863–74. population frequencies reported in the NHLBI Grand Op- [12] Li B, Krishnan VG, Mort ME, Xin F, Kamati KK, Cooper DN, et al. portunity Exome Sequencing Project. Forty-four of the 63 Automated inference of molecular mechanisms of disease from amino SNPs were successfully genotyped in the cohort of 441 cases acid substitutions. Bioinformatics 2009;25:2744–50. [13] Reitz C, Jun G, Naj A, Rajbhandary R, Vardarajan BN, Wang LS, et al. and 426 controls. Variants in the ATP-binding cassette transporter (ABCA7), apolipo- Generalized Estimating Equations (GEE) modeling of the protein E 4,and the risk of late-onset Alzheimer disease in African Americans. JAMA 2013;309:1483–92. AKAP9 variants [14] Bienias JL, Beckett LA, Bennett DA, Wilson RS, Evans DA. Design of Because of the numerical instability induced by examina- the Chicago Health and Aging Project (CHAP). J Alzheimers Dis tion of rare variants in the estimation of logistic regression 2003;5:349–55. M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 618.e3

[15] Evans DA, Bennett DA, Wilson RS, Bienias JL, Morris MC, risk locus for late-onset Alzheimer disease. Am J Hum Genet 2009; Scherr PA, et al. Incidence of Alzheimer disease in a biracial urban 84:35–43. community: relation to apolipoprotein E allele status. Arch Neurol [25] Edwards TL, Scott WK, Almonte C, Burt A, Powell EH, 2003;60:185–9. Beecham GW, et al. Genome-wide association study confirms SNPs [16] Meier IB, Manly JJ, Provenzano FA, Louie KS, Wasserman BT, in SNCA and the MAPT region as common risk factors for Parkinson Griffith EY, et al. White matter predictors of cognitive functioning disease. Ann Hum Genet 2010;74:97–109. in older adults. J Int Neuropsychol Soc 2012;18:414–27. [26] Scott WK, Nance MA, Watts RL, Hubble JP, Koller WC, Lyons K, [17] Lee JH, Cheng R, Graff-Radford N, Foroud T, Mayeux R. Analyses of et al. Complete genomic screen in Parkinson disease: evidence for the National Institute on Aging Late-Onset Alzheimer’s Disease Fam- multiple genes. JAMA 2001;286:2239–44. ily Study: implication of additional loci. Arch Neurol 2008; [27] Kamboh MI, Minster RL, Demirci FY, Ganguli M, Dekosky ST, 65:1518–26. Lopez OL, et al. Association of CLU and PICALM variants with Alz- [18] Jun G, Naj AC, Beecham GW, Wang LS, Buros J, Gallins PJ, et al. heimer’s disease. Neurobiol Aging 2012;33:518–21. Meta-analysis confirms CR1, CLU, and PICALM as Alzheimer dis- [28] Tang MX, Stern Y,Marder K, Bell K, Gurland B, Lantigua R, et al. The ease risk loci and reveals interactions with APOE genotypes. Arch APOE-epsilon4 allele and the risk of Alzheimer disease among Afri- Neurol 2010;67:1473–84. can Americans, whites, and Hispanics. JAMA 1998;279:751–5. [19] Carrasquillo MM, Zou F, Pankratz VS, Wilcox SL, Ma L, Walker LP, [29] Berg L, McKeel DW Jr, Miller JP, Storandt M, Rubin EH, Morris JC, et al. Genetic variation in PCDH11X is associated with susceptibility et al. Clinicopathologic studies in cognitively healthy aging and Alz- to late-onset Alzheimer’s disease. Nat Genet 2009;41:192–8. heimer’s disease: relation of histologic markers to dementia severity, [20] Haroutunian V, Perl DP, Purohit DP, Marin D, Khan K, Lantz M, et al. age, sex, and apolipoprotein E genotype. Arch Neurol 1998; Regional distribution of neuritic plaques in the nondemented elderly 55:326–35. and subjects with very mild Alzheimer disease. Arch Neurol 1998; [30] Morris JC, Roe CM, Xiong C, Fagan AM, Goate AM, Holtzman DM, 55:1185–91. et al. APOE predicts amyloid-beta but not tau Alzheimer pathology in [21] Bennett DA, Schneider JA, Bienias JL, Evans DA, Wilson RS. Mild cognitively normal aging. Ann Neurol 2010;67:122–31. cognitive impairment is related to Alzheimer disease pathology and [31] Morris JC, Weintraub S, Chui HC, Cummings J, Decarli C, Ferris S, cerebral infarctions. Neurology 2005;64:834–41. et al. The Uniform Data Set (UDS): clinical and cognitive variables [22] Bennett DA, Schneider JA, Buchman AS, Mendes de Leon C, and descriptive data from Alzheimer Disease Centers. Alzheimer Bienias JL, Wilson RS. The Rush Memory and Aging Project: study Dis Assoc Disord 2006;20:210–6. design and baseline characteristics of the study cohort. Neuroepidemi- [32] Storandt M, Grant EA, Miller JP, Morris JC. Longitudinal course and ology 2005;25:163–75. neuropathologic outcomes in original vs revised MCI and in pre-MCI. [23] Bennett DA, Wilson RS, Schneider JA, Evans DA, Beckett LA, Neurology 2006;67:467–73. Aggarwal NT, et al. Natural history of mild cognitive impairment in [33] Liang KY, Zeger SL. Longitudinal data analysis using linear models. older persons. Neurology 2002;59:198–205. Biometrika 1986;73:13–22. [24] Beecham GW, Martin ER, Li YJ, Slifer MA, Gilbert JR, Haines JL, [34] Zeger SL, Liang KY. Longitudinal data analysis for discrete and et al. Genome-wide association study implicates a chromosome 12 continuous outcomes. Biometrics 1986;42:121–30. 618.e4 M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618

Supplementary Figure 1. Plots of the ﬁrst 10 principal components of population substructure in the African American (AA) sample of 436 Alzheimer’s disease (AD) cases and 424 controls derived by EIGENSTRAT based on all MAF .5% markers from a whole-genome genotyping panel. The color of the plotting symbols indicates AKAP9 genotype: Black 5 subject lacks rs144662445 and rs149989685 variants; Red 5 subject has rs144662445 variant; Green 5 subject has both rs144662445 and rs149989685 variants. M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 618.e5

Supplementary Figure 2. Linkage disequilibrium (R2) plot of AKAP9 and surrounding region based on HapMap YRI reference panel of SNPs. 618.e6 M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618

Supplementary Figure 3. Networks derived by INGENUITY software containing previously-known Alzheimer’s disease (AD) genes (green) and genes containing non-synonymous SNPs identiﬁed through whole-exome sequencing (WES) (red). These networks are characterized by genes with biological functions related to: (A) cellular assembly and organization, nervous system development and function, and carbohydrate metabolism; (B) cell-to-cell signaling and interaction, cellular assembly and organization, and nervous system development and function; (C) hematological system development and function, renal and urological system development and function, and cellular compromise.

Supplementary Table 1 Characteristics of subjects from the discovery samples included in whole-exome sequencing (WES) and association studies Group N AD cases Controls Males Females Mean age (SD) APOE ε4 freq WES subjects 7 7 (100%) 0 (0%) 3 (43%) 4 (57%) 79.4 (5.2) 21% Analyzed MIRAGE subjects 408 199 (49%) 209 (51%) 121 (30%) 287 (70%) 75.1 (8.1) 34% Analyzed GenerAAtions subjects 408 223 (55%) 185 (45%) 173 (42%) 235 (58%) 79.5 (6.7) 29% Total discovery 816 422 (52%) 394 (48%) 294 (36%) 522 (64%) 77.3 (7.5) 32% Abbreviations: AD, Alzheimer disease; MIRAGE, Multi-Institutional Research on Alzheimer Genetic Epidemiology Study; GenerAAtions, Genetic and Environmental Risk Factors for Alzheimer Disease among African Americans Study. M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 618.e7

Supplementary Table 2 Characteristics of genotyped subjects from the ADGC AA replication cohort by study of origin Group N AD cases Controls Males Females Mean age (SD)* APOE ε4 freq ADC 24 4 (17%) 20 (83%) 8 (33%) 16 (67%) 73.6 (5.8) 14% CHAP 328 177 (54%) 151 (46%) 105 (32%) 223 (68%) 81.7 (6.6) 18% U. Miami 246 82 (33%) 164 (67%) 63 (26%) 183 (74%) 72.7 (8.4) 26% Washington U. 165 125 (76%) 40 (24%) 39 (24%) 126 (76%) 75.5 (7.6) 29% U. Pittsburgh 201 109 (54%) 92 (46%) 55 (27%) 146 (73%) 76.7 (7.4) 30% AAG - Columbia 242 20 (8%) 222 (92%) 44 (18%) 198 (82%) 69.8 (7.6) 21% AAG – Miami 199 40 (20%) 159 (80%) 46 (23%) 153 (77%) 71.7 (7.8) 25% AAG - NC A&T 226 61 (27%) 165 (73%) 68 (30%) 157 (69%) 71.3 (8.6) 27% AAG – Vandy 124 37 (30%) 87 (70%) 25 (20%) 99 (80%) 72.1 (8.8) 29% Mayo Clinic 308 80 (26%) 228 (74%) 68 (22%) 240 (78%) 78.2 (7.2) 24% MSSM 43 29 (67%) 14 (33%) 17 (40%) 26 (60%) 82.7 (10.0) 28% NIA-LOAD 15 6 (40%) 9 (60%) 1 (7%) 14 (93%) 72.5 (7.0) 25% ROS/MAP 130 28 (22%) 102 (78%) 54 (42%) 76 (58%) 81.0 (5.3) 18% Vanderbilt U. 34 9 (26%) 25 (74%) 7 (21%) 27 (79%) 70.4 (6.4) 24% WHICAP 621 230 (37%) 391 (63%) 159 (26%) 462 (74%) 80.8 (6.4) 19% Total 2906 1037 (36%) 1869 (64%) 759 (26%) 2146 (74%) 75.3 (8.5) 23% Abbreviations: AA, African-Americans; AD, Alzheimer disease; ADGC, Alzheimer Disease Genetics Consortium. *Age at evaluation or, alternately, age at diagnosis/death for samples and entire cohorts (MSSM) where age at evaluation was not available.

Supplementary Table 3 Primers used for Sanger Sequencing of the two AD-associated AKAP9 SNPs SNP Forward primer Reverse primer Rs144662445 5-GAG CAC TGT AAA GGA C-3 5-TGA CCC CAC CTC ATC TTC TCT TAG-3 Rs149979685 5-GTT ACT GGG GTT CCA GG-3 5-GAC ATG CTT CAG TGA TGG CAC-3 Abbreviation: SNP, single-nucleotide polymorphism. 618.e8 M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618

Supplementary Table 4 Characteristics of the SNPs identiﬁed for genotyping #WES Mapping Read Depth subjects ESP MAF:EA%/AA quality depth quality w. Alt. In Successfully Gene CHR POS rsID %* (MQ) (DP) (QD) Allele AlzGene Network genotyped NoC2L 1 888563 rs150127608 0.0116/0.7036 35.55 45 13.45 2 No N1 Yes NoC2L 1 892380 rs150615968 0.0/0.6809 36.91 212 11.93 2 No N1 No ISG15 1 949491 rs148041041 0.0116/0.1816 36.84 64 12.58 2 No N3 Yes AGRN 1 987191 rs139415524 0.0/0.4085 37 37 16.33 2 No N2 Yes FCRLA 1 161681237 rs144403707 0.0/0.1362 36.84 62 12.01 2 No N3 No FCRLA 1 161681238 rs144830049 0.0/0.1362 36.84 63 7.35 3 No N3 Yes ATP2B4 1 203672867 rs145963279 0.0/2.1108 37 141 13.52 2 No N2 Yes PLX02 1 208215558 rs144005934 0.0/0.0454 37 66 12.07 2 No N2 Yes AGT 1 230838903 rs143479528 0.0/0.522 37 90 15.6 2 Yes N1 Yes IFIH1 2 163137881 rs147000317 0.0/0.8171 37 180 14.52 2 No N3 Yes LRP2 2 170134388 rs145094511 0.0/0.0681 36.96 252 13.03 2 Yes N2 Yes ITGA9 3 37774225 rs142726080 0.0116/3.2683 37 46 16.67 2 No N2 Yes SCN5A 3 38646398 NA 0.0/0.0753 37 17 12.72 2 No N2 Yes NKTR 3 42678946 rs146815110 0.0/2.1335 36.88 340 12.91 2 No N11N3 Yes C3orf15 3 119456282 NA NA 37 219 18.39 2 No N1 No LRAT 4 155665876 rs147855559 0.0/0.0681 37 91 14.75 2 Yes N1 Yes HAND2 4 174448432 rs140861359 0.0/0.2724 37 88 18.8 2 No N1 Yes RIPK1 6 3105849 NA NA 37 164 16.92 2 No N3 No GPLD1 6 24436908 rs140754732 0.0/0.1362 37 123 12.76 2 No N1 Yes TAP2 6 32798457 rs150253319 0.1661/2.184 36.89 91 13.93 2 Yes – Yes TAP2 6 32802938 rs140654840 0.8859/2.1869 37 71 16.44 2 Yes – Yes FRK 6 116289888 rs143351471 0.0/0.0681 36.91 214 12.32 2 No N1 No TNFAIP3 6 138195991 rs146534657 0.1163/0.0908 36.91 109 18.32 2 No N3 Yes GRM1 6 146755454 rs144944927 0.0/2.5658 36.88 81 13.15 2 No N1 Yes GLI3 7 42007460 rs146130351 0.0/1.4299 37 177 15.72 2 No N1 Yes BLVRA 7 43832373 rs142897365 0.0116/0.7717 37 77 14.2 2 No N1 No SEMA3A 7 83823846 rs139951141 0.0/0.3404 36.89 283 12.47 2 No N2 No AKAP9 7 91709085 rs144662445 0.0/0.4312 36.92 131 13.06 2 No N1 Yes AKAP9 7 91732110 rs149979685 0.0/0.3631 37 166 12.71 2 No N1 Yes TRIP6 7 100466396 NA 0.0122/1.5259 37 21 15.15 2 No N3 Yes TRIB1 8 126448615 rs142512119 0.0/0.1362 37 176 14.05 2 No N2 No IF07 9 21202136 rs145794215 0.0116/2.6555 36.27 165 17.12 2 No N3 No MUSK 9 113530236 NA 0.0/0.1791 36.93 151 14.9 2 No N2 Yes RALGDS 9 135985773 rs140586035 0.0116/1.793 36.85 66 10.9 2 No N1 Yes ALOX5 10 45938883 NA NA 37 13 8.77 2 Yes – No TACR2 10 71167027 rs142415572 0.0116/5.9691 36.82 56 19.81 2 Yes – Yes SORCS3 10 106974224 NA 0.0/0.4539 36.94 177 12.15 2 Yes – Yes KNDC1 10 135015178 rs141463252 0.0/0.3404 36.68 32 11.99 2 No N1 Yes LRP4 11 46911956 rs138878258 0.1279/0.4771 37 57 12.14 2 No N2 Yes STX3 11 59562950 rs142968661 0.0466/2.3171 36.84 63 9.34 2 No N1 No BAD 11 64051762 NA NA 37 115 11.17 2 No N2 Yes SPTBN2 11 66475201 NA NA 37 67 7 2 No N1 Yes CEP290 12 88512485 rs147371999 0.0/1.8904 37 102 12.8 2 No N2 Yes ACACB 12 109660694 rs146002202 0.0116/0.8171 37 19 27.8 2 No N2 No ACACB 12 109683516 rs145398869 0.0/0.0908 36.91 107 13.17 2 No N2 Yes ATP2A2 12 110784163 NA 0.0/0.0227 37 64 22.08 2 No N2 No KL 13 33591385 NA 0.0/0.1235 37 13 22.79 2 No N2 Yes KIF26A 14 104642981 NA 0.0/0.5185 37 14 27.38 2 No N1 No HERC2 15 28518046 rs150363648 NA 35.25 51 7.8 5 No N2 No MEF2A 15 100211761 NA NA 36.81 52 5.49 2 Yes N2 No UBN1 16 4924905 rs146192777 0.0465/0.751 37 175 11.65 2 No N1 No PELP1 17 4579700 rs147186006 0.012/1.5853 37 59 17.67 2 No N1 Yes MYH13 17 10247309 rs144732640 0.0116/0.8852 35.61 165 13.12 2 Yes – Yes NoS2 17 26110055 rs149623743 0.0/0.0681 36.9 96 18.23 2 No N3 Yes IGFBP4 17 38612790 NA NA 37 64 13.24 2 No N2 No STXBP2 19 7712054 rs150174842 0.0117/0.7976 37 39 23.82 2 No N1 Yes MIA 19 41282931 rs144785745 0.0/0.227 36.79 49 18.95 2 No N2 No XRCC1 19 44047554 NA 0.0349/0.0 37 48 10.24 2 Yes – Yes CD33 19 51742799 rs149704584 0.0/0.0908 36.73 111 13.69 2 Yes N11N3 Yes (Continued) M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 618.e9

Supplementary Table 4 Characteristics of the SNPs identiﬁed for genotyping (Continued) #WES Mapping Read Depth subjects ESP MAF:EA%/AA quality depth quality w. Alt. In Successfully Gene CHR POS rsID %* (MQ) (DP) (QD) Allele AlzGene Network genotyped APOBEC3B 22 39382414 rs146055882 0.0467/13.1256 33.28 222 13.66 3 No N2 No CBX7 22 39530663 NA NA 37 10 12.27 2 No N1 No SBF1 22 50903029 rs149528827 0.0/2.1199 37 50 13.21 2 No N1 No CNGA2 X 150912171 rs150539917 0.0297/9.2568 37 116 14.1 2 No N1 Yes Abbreviations: AA, African-Americans; SNP, single-nucleotide polymorphism; CHR, chromosome; POS, position; MAF, minor allele frequency; EA, Eu- ropean American. *Allele frequency from NHLBI Grand Opportunity Exome Sequencing Project; (ESP) ESP6500 release/percentage in European Americans/African Amer- icans in the ESP database; Mapping Quality (MQ), Read Depth (DP), and Depth Quality (QD) as reported by the Genome Analysis Toolkit (GATK); NA: SNP not present in dbSNP v. 135 or in ESP. 618.e10 M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618

Supplementary Table 5 Association results for all genotyped SNPs including allele counts comparing all cases to all controls HWE # Case min. # Case maj. # Control min. # Control maj Fisher’s Gene CHR POS rsID P-value alleles alleles alleles alleles exact P NoC2L 1 888563 rs150127608 1 5 819 3 773 .396 ISG15 1 949491 rs148041041 1 0 832 0 784 1 AGRN 1 987191 rs139415524 1 2 824 4 776 .904 FCRLA 1 161681238 rs144830049 1 5 819 3 777 .393 ATP2B4 1 203672867 rs145963279 1 19 811 10 768 .092 PLX02 1 208215558 rs144005934 1 1 831 0 782 .5154 AGT 1 230838903 rs143479528 1 3 827 3 781 .683 IFIH1 2 163137881 rs147000317 1 7 819 4 778 .305 LRP2 2 170134388 rs145094511 1 0 830 0 784 1 ITGA9 3 37774225 rs142726080 1 26 804 18 760 .197 SCN5A 3 38646398 NA 1 0 828 2 782 1 NKTR 3 42678946 rs146815110 .26 9 823 21 761 .995 LRAT 4 155665876 rs147855559 1 0 832 0 782 1 HAND2 4 174448432 rs140861359 1 2 828 1 783 .521 GPLD1 6 24436908 rs140754732 1 1 831 1 777 .767 TAP2 6 32798457 rs150253319 1 19 811 14 770 .296 TAP2 6 32802938 rs140654840 .33 15 817 18 764 .812 TNFAIP3 6 138195991 rs146534657 1 0 832 0 782 1 GRM1 6 146755454 rs144944927 1 14 816 21 761 .939 GLI3 7 42007460 rs146130351 1 14 818 18 760 .861 AKAP9 7 91709085 rs144662445 1 9 835 1 785 .014 AKAP9 7 91732110 rs149979685 1 5 837 0 788 .0366 TRIP6 7 100466396 NA 1 18 810 15 767 .427 MUSK 9 113530236 NA 1 2 830 0 784 .265 RALGDS 9 135985773 rs140586035 1 15 807 13 763 .486 TACR2 10 71167027 rs142415572 .43 61 771 51 731 .294 SORCS3 10 106974224 NA 1 3 825 2 780 .527 KNDC1 10 135015178 rs141463252 1 6 824 3 777 .285 LRP4 11 46911956 rs138878258 1 6 826 2 778 .166 STX3 11 59562950 rs142968661 .67 18 818 17 711 .661 BAD 11 64051762 NA 1 0 832 0 784 1 SPTBN2 11 66475201 NA 1 0 832 0 780 1 CEP290 12 88512485 rs147371999 1 13 819 16 768 .819 ACACB 12 109683516 rs145398869 1 0 832 1 783 1 KL 13 33591385 NA 1 4 824 5 779 .773 KIF26A 14 104642981 NA .85 3 745 7 653 .965 PELP1 17 4579700 rs147186006 1 12 820 19 763 .948 MYH13 17 10247309 rs144732640 1 7 825 10 774 .864 NoS2 17 26110055 rs149623743 1 3 829 2 782 .528 IGFBP4 17 38612790 NA 1 0 838 0 722 1 STXBP2 19 7712054 rs150174842 1 6 824 6 776 .653 XRCC1 19 44047554 NA 1 0 832 0 782 1 CD33 19 51742799 rs149704584 1 1 827 0 782 .514 CNGA2 X 150912171 rs150539917 .21 74 756 67 709 .456 Abbreviations: SNP, single-nucleotide polymorphism; CHR, chromosome; POS, position; HWE, Hardy-Weinberg Equilibrium. M.W. Logue et al. / Alzheimer’s& Dementia 10 (2014) 609-618 618.e11

Supplementary Table 6 Characteristics of 13 AA Subjects in the MIRAGE/GenerAAtions sample with at least one AKAP9 variant in rs144662445 and rs149989685 Study Whole-exome sequenced Sex APOE Age* AD status rs144662445 rs149979685 GenerAAtions No Female 34 75 Control AG CC MIRAGE No Female 44 55y Control AG CT MIRAGE No Female 33 86 Case AG CC MIRAGE No Male 23 74 Case AG CC MIRAGE No Female 34 82 Case AG CC GenerAAtions No Female 33 88 Case AG CC MIRAGE No Female 33 81 Case AG CT MIRAGE Yes Female 34 82 Case AG CT MIRAGE Yes Female 44 84 Case AG CT MIRAGE No Male 33 78 Case AG CT GenerAAtions No Male 33 64 Case AG CT GenerAAtions No Female 33 71 Case AG CT GenerAAtions No Female 23 91 Case AG CT Abbreviations: AA, African-Americans; MIRAGE, Multi-Institutional Research on Alzheimer Genetic Epidemiology Study; GenerAAtions, Genetic and Environmental Risk Factors for Alzheimer Disease among African Americans Study; SNP, single-nucleotide polymorphism. *Age at assessment. yExcluded from analysis due to age ,60 at time of assessment.

Supplementary Table 7 Estimated frequencies of .1% MAF AKAP9 haplotypes based on SNPs from GWAS genotyping Haplotype* Frequency AGAGGAGAAC 0.425 GAAAAGAGGA 0.218 GGCAGGAGGA 0.129 GGAAGGAGGA 0.118 GGAAGGAAGCy 0.052 GGAGGAGAAC 0.045 Abbreviations: MAF, minor allele frequency; GWAS, Genome-wide association study; SNP, single-nucleotide polymorphism. *Alleles for SNPs rs1859037, rs2299233, rs2282973, rs6465347, rs2158138, rs733957, rs2079082, rs13239875, rs4265, and rs1063243, respectively. yHaplotype HAP0 containing rare variants at rs144662445 and rs149989685.