Pharmacogenomics in Alzheimer's disease: a genome-wide association study of response to cholinesterase inhibitors

Neurobiology of Aging 34 (2013) 1711.e7e1711.e13

Contents lists available at SciVerse ScienceDirect

Neurobiology of Aging journal homepage: www.elsevier.com/locate/neuaging

Pharmacogenomics in Alzheimer’s disease: a genome-wide association study of response to cholinesterase inhibitors Filippo Martinelli-Boneschi a, b, *, Giacomo Giacalone a, b, Giuseppe Magnani b, Gloria Biella c, Elisabetta Coppi b, Roberto Santangelo b, Paola Brambilla a, Federica Esposito a, b, Sara Lupoli d, Francesca Clerici e, Luisa Benussi f, Roberta Ghidoni f, Daniela Galimberti g, Rosanna Squitti h, Annamaria Confaloni i, Giuseppe Bruno j, Sabrina Pichler k, Manuel Mayhaus k, Matthias Riemenschneider k, Claudio Mariani e, Giancarlo Comi a, b, Elio Scarpini g, Giuliano Binetti f, Gianluigi Forloni c, Massimo Franceschi l, Diego Albani c, ** a

San Raffaele Scientific Institute, Division of Neuroscience, Laboratory of Genetics of Complex Neurological Disorders, Institute of Experimental Neurology (INSPE), Milan, Italy Memory Clinic, Department of Neurology, and Division of Neuroscience, San Raffaele Scientific Institute, Milan, Italy Department of Neuroscience, “Mario Negri” Institute for Pharmacological Research, Milan, Italy d Department of Health Sciences, University of Milan, Milan, Italy e Center for Research and Treatment of Cognitive Dysfunctions, Institute of Clinical Neurology, Department of Biomedical and Clinical Sciences, “Luigi Sacco” Hospital, University of Milan, Milan, Italy f Proteomics Unit and NeuroBioGen Lab-Memory Clinic, IRCCS Istituto Centro San Giovanni di Dio-Fatebenefratelli, Brescia, Italy g Department of Pathophysiology and Transplantation, "Dino Ferrari" Center, University of Milan and “Fondazione Cà Granda, IRCCS Ospedale Maggiore Policlinico,” Milan, Italy h Department of Neuroscience, AFaRdFatebenefratelli Hospital, Isola Tiberina, Rome, Italy i Department of Cell Biology and Neurosciences, Istituto Superiore di Sanità, Rome, Italy j Department of Neurology and Psychiatry, University of Rome “Sapienza,” Rome, Italy k Department of Psychiatry and Psychotherapy, Saarland University Hospital, Saarland University, Homburg, Germany l Department of Neurology, IRCCS Multimedica, Milan, Italy b c

a r t i c l e i n f o

a b s t r a c t

Article history: Received 20 October 2012 Received in revised form 4 December 2012 Accepted 18 December 2012 Available online 29 January 2013

We conducted a genome-wide association study in a cohort of 176 Italian Alzheimer’s disease (AD) patients with extreme phenotype of response to cholinesterase inhibitors. Patients were classified into responders in case of positive, stable, or
1 worsening of mini-mental state examination score and into nonresponders if >3 points worsening during a median follow-up of 0.85 years of treatment. Forty-eight single-nucleotide polymorphisms were selected for replication in 198 additional AD-treated patients. By using the dichotomous response trait and a quantitative trait approach (change of mini-mental state examination), a nominal replication and evidence of association when combining data were achieved for 2 singlenucleotide polymorphisms associated with response to treatment: rs6720975A (pcombined ¼ 2.9  105, beta regression coefficient: 1.61) and rs17798800A (pcombined ¼ 6.8  106, odds ratio ¼ 0.38, 95% confidence interval ¼ 0.25e0.58). Rs6720975 maps in the intronic region of PRKCE, a protein kinase involved in several cellular functions, whereas rs17798800 is intergenic and, according to expression quantitative trait locus (eQTL) analysis, it acts as a cis-regulator of NBEA, an A kinaseeanchoring protein playing a substantial role in the maturation of the nervous system. Despite its limitations, this project paves the way for the application of personalized medicine in AD patients and for collaborative efforts in this field. Ó 2013 Elsevier Inc. All rights reserved.

Keywords: Alzheimer’s disease Genetics Pharmacogenomics Acetylcholinesterase inhibitors Genome-wide association study

* Corresponding author at: San Raffaele Scientific Institute, Division of Neuroscience, Laboratory of Genetics of Complex Neurological Disorder, Institute of Experimental Neurology (INSPE), Via Olgettina 58, Milan 20123, Italy. Tel.: þ39 02 26437327; fax: þ39 02 26432277. ** Diego Albani, Unit of Genetics of Neurodegenerative Disorders, Istituto di Ricerche Farmacologiche “Mario Negri”, Via La Masa 19, 20156 Milan, Italy. Tel.: þ390239014594; fax: þ39023546277. E-mail addresses: martinelli.fi[email protected] (F. Martinelli-Boneschi), diego. [email protected] (D. Albani). 0197-4580/$ e see front matter Ó 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.neurobiolaging.2012.12.008

1. Introduction Alzheimer’s disease (AD) is the most frequent neurodegenerative disorder and one of the most common diseases in the industrialized world, affecting more than 33.9 million people worldwide, a figure predicted to triple over the next 40 years (Barnes and Yaffe, 2011). AD causes progressive loss of cognitive functions leading to dementia and death. Novel therapeutic approaches have emerged

1711.e8

F. Martinelli-Boneschi et al. / Neurobiology of Aging 34 (2013) 1711.e7e1711.e13

over the last years, but confirmation in clinical trials (Katsuno et al., 2012) has been very poor, and cholinesterase inhibitors (ChEI) are still the mainstay in the treatment of AD. These drugs act mainly by increasing cholinergic neurotransmission and raising the level of cerebral acetylcholine, but recent studies also suggest alternative modes of action including neuroprotective and immunomodulatory effects (Akaike et al., 2010; Noh et al., 2009; Reale et al., 2004). Evidence from therapeutic trials and clinical practice shows not only that AD patients with mild, moderate, or severe dementia treated with ChEIs improved in cognitive function after 6 months with an average of 1.4 points (Birks, 2006) on mini-mental state examination (MMSE), but also that the clinical response is variable and unpredictable (Birks, 2006). Therefore, clinical and genetic predictors of response to treatment are needed to help individualize therapy. So far, a fair amount of knowledge has been accumulated for clinical predictors, such as less cognitive impairment at drug start and the MMSE gain after 3 months of therapy (Calabria et al., 2009; Wallin et al., 2011; Wattmo et al., 2011). But less is known about AD-related genetic loci. Previous studies have applied a candidate-gene approach focusing on the apolipoprotein-E (APOE) epsilon 4 (e4) allele (Blesa et al., 2006; Choi et al., 2008), BCHE (Chianella et al., 2011; Scacchi et al., 2009), ACHE (Scacchi et al., 2009), CHAT (Harold et al., 2006; Scacchi et al., 2009), PON1 (Klimkowicz-Mrowiec et al., 2011; Pola et al., 2005), and CYP2D6, the key regulator of acetyl ChEI metabolism (Cacabelos, 2008; Chianella et al., 2011; Varsaldi et al., 2006). To our knowledge, this is the first genome-wide association study (GWAS) aimed at identifying common genetic variants predictive of response to ChEI in an Italian sample of AD patients. 2. Methods 2.1. Subjects From January 2002 to January 2009, 287 patients with a diagnosis of probable AD were recruited in 3 Italian Alzheimer units (Ospedale San Raffaele, IRCCS Multimedica, and IRCCS Ospedale Maggiore Policlinico) and retrospectively screened for enrollment in the GWA study. For the replication phase, an independent cohort of 252 patients with a diagnosis of probable AD were screened, coming from 5 Italian Alzheimer Units (IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli, IRCCS Ospedale Maggiore Policlinico, Fatebenefratelli Hospital Isola Tiberina, University of Rome “Sapienza,” and Luigi Sacco Hospital). Baseline demographic and clinical characteristics were recorded, and a follow-up neurologic visit was done. Cognitive evaluation was based on the MMSE, at the beginning of treatment and at the end of follow-up. The annualized difference between the MMSE scores at baseline and at follow-up was referred to as the delta MMSE. Inclusion criteria were (1) signed written informed consent for the genetic research study approved by the local ethics committees; (2) self-reported Italian origin up to grandparents; (3) diagnosis of probable AD according to the National Institute of Neurological and Communicative Disorders and StrokeeAlzheimer’s Disease and Related Disorders Association Work Group criteria (Blacker et al., 1994); (4) minimum 6 months of treatment with ChEIs (donepezil, rivastigmine, or galantamine) and no concomitant memantine; (5) MMSE done at baseline and at the end of follow-up; (6) followup lasting 6 and 18 months; and (7) baseline MMSE score between 8 and 28 inclusive. Concomitant pathologies, including hypertension, diabetes mellitus, and myocardial ischemia, were accepted if adequately controlled by pharmacotherapy. Patients were classified as responders (R) if the MMSE increased, remained stable, or the delta MMSE fell by 1 or less and as nonresponders (NR) in case of worsening of >3 points in delta MMSE.

We excluded 111 patients from the 287 in the discovery sample for the following reasons: 30 for missing phenotype data, 26 for follow-up duration (in 8 it was <6 months and in 18 it was >18 months), 2 for concomitant treatment with memantine, 1 for nonAD diagnosis, and 52 because they fell outside the defined responsiveness criteria. We, therefore, genotyped 176 patients in the discovery phase, of whom 94 R and 82 NR. In the replication phase, we excluded 54 patients from the 252 recruited for the following reasons: 12 for missing phenotype data, 2 for follow-up duration >18 months, 10 for concomitant treatment with memantine, 1 because of low baseline MMSE, and 29 because they fell outside the responsiveness criteria. A final list of 198 patients (109 R and 89 NR) were typed in the replication phase. 2.2. Genotyping The GWAS of the discovery sample was carried out using Human660W-Quad Genotyping BeadChips, according to the Illumina Infinium assay protocol (Illumina, San Diego, CA, USA) in the Laboratory of Genetics of Complex Neurological Disorders. A total of 657,366 single-nucleotide polymorphisms (SNPs) were genotyped across the entire genome. We used Genome Studio version 2011.1 software for quality controls (QC) of array data. Seven individuals were excluded: 4 because of genetic relatedness, assessed by pairwise identity by descent (IBD) estimation in PLINK (Purcell et al., 2007) using a relatedness measure (pi-hat) value >0.2; 1 was classified as an outlier using Eigenstrat software version 3.0, and 2 had a genotype call rate <95%. In all, 2517 SNPs were excluded for low genotype call rate (<90%) and 23,478 for a minor allele frequency <1%. After QC, 169 individuals (92 R and 77 NR) and 522,109 SNPs were left for statistical analysis. Genetic stratification was also explored, and the 2 principal components are plotted in Fig. 1. Using the first 30 eigenvectors, no significant differences were found between R and NR. After removal of the outlier, the population lambda inflation factor was 1.0 for R versus NR variable and 1.00095 for delta MMSE trait, confirming that our sample of Italian origin was not stratified. We genotyped 48 SNPs in the replication sample using a Sequenom MassArray platform (Sequenom, San Diego, CA, USA) in conjunction with the iPLEX assay (http://www.sequenom.com),

Fig. 1. Principal component analysis (PCA) of genome-wide association study data. This plot represents the first 2 eigenvectors of the PCA analysis applied to responders (circles) and nonresponders (squares) in the discovery sample.

F. Martinelli-Boneschi et al. / Neurobiology of Aging 34 (2013) 1711.e7e1711.e13

according to the manufacturer’s standard protocols. A MassArrayTyper version 3.4 was used to read the extended mass and genotype calls. Individuals with a genotype call rate <90% were removed (n ¼ 30). Four SNPs were excluded because of a genotype call rate <90%. After QC, we had 168 individuals (94 R and 74 NR) and 44 SNPs available for statistical analysis. 2.3. Statistical analysis The statistical power of the study was calculated using the Genetic Power Calculator software (http://pngu.mgh.harvard.edu/ wpurcell/gpc/) according to an allelic additive model, assuming a minor allele frequency of 0.3 and a frequency of the response trait of 30%. The GWAS (176 cases) was 80% powered to find an odds ratio (OR) of 1.90 with a p value of <5  104 and an OR of 3.55 with a genome-wide p value. For the quantitative trait, we used the Quanto software (http://hydra.usc.edu/gxe/) and the study was 80% powered to identify a beta coefficient of 2.5 with a p value <5  104 and a beta coefficient of 3.5 with a genome-wide significant p value. We used Plink version 1.07 software (Purcell et al., 2007) (http:// pngu.mgh.harvard.edu/wpurcell/plink/) for statistical analysis. In the GWAS, genotypic and allelic frequencies of each SNP were measured in R and NR, and the corresponding Pearson p values were computed using Fisher exact test. We also tested the influence of genetic variants on delta MMSE by calculating p values according to the Wald test statistics and estimating the beta regression coefficient for each SNP as implemented in Plink. We computed the OR and the upper and lower bound 95% confidence intervals (CIs) for the minor allele of each SNP. SNP sequences were mapped onto NCBI36/hg18 human genome assembly. The same statistical analyses were done in the replication cohort on selected SNPs. A replication was defined as the evidence of an association with a p value <0.05 and an effect in the same direction as the discovery phase using either the response dichotomous trait or the delta MMSE. We then did a meta-analysis of the allelic association in the discovery and replication phases using the dichotomous trait R versus NR and the quantitative trait delta MMSE. To test the homogeneity across samples, we used Cochran Q test and calculated I2. Analyses were restricted to SNPs with p < 0.10 at Cochran Q test and I2 <25% (Ioannidis et al., 2007). Combined analyses were done by a fixed-effects meta-analysis. To explore the putative functional role of rs17798800, we performed eQTL analysis by correlating genotype data to gene expression data using SNPExpress (Ge and Goldstein), which is a public database including genotypes (IlluminaHumanHap 550K

1711.e9

BeadChip) and gene expression (Affymetrix Human ST 1.0 chip) data on 93 brain samples (frontal cortex) and 80 peripheral blood mononuclear cells (PBMC) from neurologically healthy individuals. To discover cis-effects, we tested gene mapping within a 1-Mb window upstream and downstream of the SNP, according to GRCh37/hg19. To explore a trans-effect, genes with p < 0.05 in the linear regression analysis were used as input for Ingenuity Pathway Analysis (software version 8.8 (Ingenuity Systems, Inc., Redwood City, CA, USA; http://www.ingenuity.com/)): canonical pathway and transcription factor analyses were done. We tested for the association between SNPs and expression changes using a linear regression model. Cut-off p values were adjusted using the Bonferroni correction: 1.63  103 for cis-eQTL analysis and 2.8  106 for trans-eQTL analysis. 3. Results Table 1 summarizes the main clinical characteristics of AD patients enrolled in the discovery and replication samples. In the discovery sample, R and NR were matched apart from a higher frequency of rivastigmine-treated patients, a higher baseline MMSE and a shorter follow-up in NR than R. In the replication sample, there was a larger proportion of women and a shorter follow-up in the NR group. When comparing the 2 cohorts of patients, sources of heterogeneity were a female preponderance (p ¼ 0.01), a longer disease duration (p < 0.001), and a larger proportion of donepeziltreated patients (p < 0.001) in the replication cohort. None of the SNPs reached genome-wide significance in the GWAS study. Because this was conceived as a pilot study, we ranked the best 100 loci in the discovery sample that fulfilled the following criteria: p < 5  104 in R versus NR analysis and p < 5  103 in delta MMSE analysis or viceversa (Supplementary Table 1). In cases with 2 or more proxy SNPs per locus (defined as r2 > 0.2 using HapMap CEU 1000 Genome Project), we performed pruning and selected one. To validate the findings from the screening phase, 48 SNPs out of the best 100 were selected using biological criteria, cluster plot inspection, and the suitability of the context sequences for design of primers, and genotyped using the Sequenom technology (Supplementary Table 2). Rs6720975A (p value ¼ 2.6  102, beta regression coefficient: 1.33) and rs17798800A (p value ¼ 8.8  103, OR ¼ 0.44, 95% CI ¼ 0.24e0.82) were replicated in the second independent Italian sample with the same effect direction (Fig. 2B and C and Fig. 3B and C). Combined analysis was done using a meta-analytic approach with a fixed-effect model, and these 2 SNPs showed evidence of

Table 1 Clinical features of the discovery and replication cohorts p Valueb

Clinical features

Discovery sample (n ¼ 176)

Replication sample (n ¼ 198)

R, n ¼ 94

NR, n ¼ 82

p Valuea

R, n ¼ 109

NR, n ¼ 89

p Valuea

Women/men (%) Age at disease onset (y), mean  SD EOAD/LOAD Age at baseline (y), mean  SD) Disease duration (y), median (range) Donepezil (%) Rivastigmine (%) Galantamine (%) Duration of follow-up (y), median (range) Baseline MMSE score, median (range) Follow-up MMSE score, (median (range) ApoE number of e4 alleles 0/1/2 (%)

61/33 (64.9%) 72.3  6.9 15/79 74.6  7.0 2.1 (0.6e5.6) 64.9 17 18.1 0.8 (0.6e1.5) 20 (8.7e27) 21.4 (7.7e28.1) 57.4/31.9/10.6

59/23 (71.9%) 72.6  6.5 9/73 73.7  6.2 2 (0.4e5.1) 67.7 27.4 4.8 0.9 (0.5e1.5) 21.7 (12e28.3) 15.5 (0e23) 53.8/41/5.1

ns ns ns ns ns 0.021

79/30 (72.5%) 73.5  7.8 14.8/85.2 75.9  7.7 3 (1e17) 85.6 14.4 0 1 (0.5e1.5) 20 (9e27) 21 (9e30) 51.9/40.6/7.5

78/11 (87.6%) 72.1  7.1 14.9/85.1 75.5  7.3 3 (1e14) 90.2 8.5 1.2 1 (0.5e1.4) 20 (11e28) 13.3 (0e23.2) 44.8/46/9.2

0.007 ns ns ns ns ns

0.01 ns ns ns <0.001 <0.001

ns ns <0.001 ns

ns ns ns ns

ns 0.001 <0.001 ns

Demographic and clinical characteristics of cases and controls in the discovery and replication of cholinesterase inhibitoretreated Alzheimer’s disease samples. Key: ApoE, apolipoprotein-E; EOAD, early-onset Alzheimer’s disease; e4, epsilon 4; LOAD, late-onset Alzheimer’s disease; MMSE, mini-mental state examination; NR, nonresponders; ns: not significant; R, responders; SD, standard deviation. a p Value of comparison of clinical features of R and NR in the discovery and replication phases. b p Value of comparison of the discovery and replication samples.

1711.e10

F. Martinelli-Boneschi et al. / Neurobiology of Aging 34 (2013) 1711.e7e1711.e13

Fig. 2. Physical position, regional association plot, and forest plot of rs6720975. (A) The physical position of rs6720975, and the region spanning 500 kb around its position based on GRCh37/hg19 that includes the genes PRKCE, SRBD1, and EPAS1. (B) The regional association plot for the region spanning 500 kb around rs6720975; p values of genotyped singlenucleotide polymorphism (SNPs) are plotted as a negative logarithm against their physical position on the chromosome based on 1000 Genome pilot 1, as implemented in SNAP version 2.2 (www.broadinstitute.org/mpg/snap/index.php). Estimated recombination rates from the HapMap CEU population show the local linkage disequilibrium (LD) structure. The color of each SNP indicates the LD level with rs6720975 based on pairwise r2 values from HapMap CEU data. Three different symbols (red) represent rs6720975 association p values for delta mini-mental state examination (MMSE) trait in the discovery phase (square), the replication phase (triangle), and the meta-analysis (rhombus). (C) The forest plot of rs6720975 based on the results of discovery, replication, and meta-analysis for delta MMSE trait.

association: rs6720975A (pcombined ¼ 2.9  105, beta regression coefficient: 1.61) and rs17798800A (pcombined ¼ 6.8  106, OR ¼ 0.38, 95% CI ¼ 0.25e0.58) (Fig. 2B and C and Fig. 3B and C). No evidence of heterogeneity was found for either SNPs (Table 2). Analyses were adjusted for the clinical variables found to be unbalanced between R and NR: baseline MMSE and type of treatment in the discovery sample and sex in the replication sample. After adjustment, replication was confirmed for either SNPs (p values of 4.9  102 and 1.7  102 for rs6720975 and rs17798800), confirming an association of rs6720975A (pcombined ¼ 1.95  104, beta regression coefficient: 1.40) and rs17798800A (pcombined ¼ 5.19  105, OR ¼ 0.39, 95% CI ¼ 0.25e0.61). As regards the cis-eQTL analysis, we found 5 genes within a 1-Mb window upstream and downstream of rs17798800, but expression data were available for 3 of them (STARD13, RFC3, and NBEA). A significant association was found between tested SNP and NBEA gene (neurobeachin) expression levels in PBMCs (beta coefficient 16.08, p value 3  103) but not in brain samples (beta coefficient 60.53, p value 0.18) with same effect direction. Interestingly, an association with NBEA transcription levels with same direction in PBMCs was also found for rs492452 (beta coefficient 16.03, p value 5.6  103), a proxy SNP of rs17798800 associated with the response trait in the discovery phase. No correlation was found with RFC3 and STARD13. Trans-eQTL analysis did not detect any association beyond the accepted p value (p ¼ 2.8  106) throughout the brain and PBMC transcriptome in SNPExpress data set. The 2 top genes for trans-eQTL on brain samples were

EFNA2 (ephrin A2) (beta coefficient 12.94, p value 2.59  104) and CAMK2N2 (calcium/calmodulin-dependent protein kinase II inhibitor 2) (beta coefficient 12.45, p value 8.07  104). We decided to use a more liberal p value (p ¼ 0.05), which allowed the selection of 364 genes from trans-eQTL on brain samples analyzed using a pathway approach (Supplementary Table 3). The most enriched canonical pathways were IL4 signaling (adjusted p value using the BenjaminieHochberg correction: 2.65  103, ratio 0.11) and antigen presentation (adjusted p value: 2.65  103, ratio 0.15), whereas the most represented functional category was inflammatory response (p value 4.28  105, 55 molecules). As regards transcription factor analysis, 13 genes are known targets of CEBPB, and 8 of which parallel its known action according to beta coefficients (p value 7.68  103, z-score 3.017). 4. Discussion To our knowledge, this is the first GWAS aiming to identify common genetic variants that influence the response to ChEI drugs in AD patients. Previous studies focused on candidate genes including APOE (Blesa et al., 2006; Choi et al., 2008), BCHE (Chianella et al., 2011; Scacchi et al., 2009), ACHE (Scacchi et al., 2009), CHAT (Harold et al., 2006; Scacchi et al., 2009), and PON1 (Klimkowicz-Mrowiec et al., 2011; Pola et al., 2005) and reported conflicting results. A more consistent association with response trait was found for the rs1080985G allele at CYP2D6, the key regulator of ChEI metabolism, in a sample of 115 AD Italian patients and

F. Martinelli-Boneschi et al. / Neurobiology of Aging 34 (2013) 1711.e7e1711.e13

1711.e11

Fig. 3. Physical position, regional association plot, and forest plot of rs17798800. (A) The physical position of rs17798800 and the region spanning 1 MB around its position based on GRCh37/hg19 that includes the genes NBEA, STARD13, and RFC3. (B) The regional association plot of the region spanning 500 kb from rs17798800; p values of genotyped singlenucleotide polymorphism (SNPs) are plotted as a negative logarithm against their physical position on the chromosome based on 1000 Genome pilot 1, as implemented in SNAP version 2.2 (www.broadinstitute.org/mpg/snap/index.php). Estimated recombination rates from the HapMap CEU population show the local linkage disequilibrium (LD) structure. The color of each SNP indicates the LD level with rs17798800 based on pairwise r2 values from HapMap CEU data. Three different symbols (red) represent rs17798800 association p values for responder (R) versus nonresponder (NR) trait in the discovery phase (square), the replication phase (triangle), and the meta-analysis (rhombus). (C) The forest plot for rs17798800 based on the results of discovery, replication, and meta-analysis for R versus NR trait.

was recently confirmed by our collaborative effort in an independent cohort of 415 AD patients (Albani et al., 2012; Pilotto et al., 2009) of Italian origin. This SNP was not directly tagged by the array used in the present GWAS, and none of the proxy SNPs were among the best hits identified in this study. Several definitions of response to ChEI in AD have been proposed in clinical trials and pharmacogenetic studies (Burns et al., 2008), some relying on cognition and functional scales and others on cognition criteria only. The most stringent one is the NICE definition (http://www.nice.org.uk) (NHS, 2011), which considers as responders those patients with stability or improvement in cognition evaluated using the ADAS-Cog and the MMSE and an improvement in functional status evaluated by activities of daily living (ADLs) or instrumental activities of daily living (IADL) after a 6-month follow-up. However, the NICE definition, as already noted by Burns et al. (2008), might be too conservative, classifying as nonresponders also patients who show some clinical benefits (Burns et al., 2008) given that natural history studies indicate that untreated AD patients lose 2e4 points of MMSE per year on average (Courtney et al., 2004). In our study, we could not include functional status scales because they were available only for a subgroup of patients, and we used only MMSE for classification. Though this scale has clear limits, it still represents the most commonly used

tool to monitor treatment response and it is correlated with functional state (Rasovska and Rektorova, 2011). We decided to select and compare the population of ChEI-treated AD patients at both extremes of the response trait (individuals positive, stable, or with
1 MMSE decrease compared to those with >3 MMSE worsening) to boost the statistical power of the study as in other successful studies (Arking et al., 2006; Duncan et al., 2011). Selection criteria were defined before the clinical recruitment and led to the inclusion in our sample of the 36th and 59th percentiles of MMSE change distribution. We used quantitative trait approach by testing the influence of genetic variants on delta MMSE to further increase the statistical power (Yang et al., 2010). Sample size remains a clear limitation, but pharmacogenetic traits tend to have greater ORs than GWA signals (Motsinger-Reif et al., 2010). These results can simply reflect the low-hanging fruit because pharmacogenomic studies are less common than disease-trait studies, but we can also argue that pharmacogenetic traits could have largely evaded the rules of natural selection unlike other complex traits (Cirulli and Goldstein, 2010). Moreover, we are keen to share raw data and results with interested groups for using meta-analytic approaches to overcome the sample size limitation. Only 2 SNPs out of the 48 selected in the discovery phase were replicated in the second independent Italian sample. The sample

Key: Chr., chromosome; CI, confidence interval; MAF, minor allele frequency; MMSE, mini-mental state examination; NR, nonresponders; OR, odds ratio; R, responders; SNP, single-nucleotide polymorphism.

1.33 2.6  102 1.82 2.41  102 1.43 (0.92e2.21) 1.12  101 0.44 (0.24e0.82) 8.78  103 0.39 0.20 2.19 (1.4e3.42) 5.14  104 1.81 4.49  104 0.48 0.34 (0.19e0.6) 1.24  104 2.38 1.38  103 0.10 0.32 0.29 PRKCE 0.51 Intergenic 0.12 rs6720975 2p21 A rs17798800 13q13.2 A

Q p Value, delta MMSE Beta I2 Q Allelic p value, R vs. NR

Meta-analysis (n ¼ 337)

p Value, delta OR (95% CI) MMSE Allelic p value, Beta R vs. NR

Replication (n ¼ 168)

MAF, MAF, OR (95% CI) R NR p Value, delta MMSE Beta Allelic p value, R vs. NR

Discovery (n ¼ 169)

MAF, MAF, OR (95% CI) R NR Minor Locus allele Chr. SNP

Table 2 Results of the 2 replicated SNPs in the discovery sample, replication sample, and meta-analysis

1.76 (1.29e2.4) 3.87  104 0.18 45 1.61 2.9  105 0.54 0 0.38 (0.25e0.58) 6.84  106 0.54 0 2.12 8.23  105 0.6 0

F. Martinelli-Boneschi et al. / Neurobiology of Aging 34 (2013) 1711.e7e1711.e13

I2

1711.e12

size of the replication provided a statistical power of 91% to replicate the beta coefficient of rs6720975, and of 99% to replicate the OR of rs17798800 found in the discovery sample. Rs6720975 maps in the intronic region of PRKCE (Fig. 2A), which belongs to the protein kinase C family. PRKCE is widely expressed throughout the body with higher levels in the brain, particularly in the hippocampus (Shirai et al., 2008). Several studies have claimed a variety of potential functions for PRKCE; among these, long-term potentiation and long-term depression mechanisms (Shirai et al., 2008), regulation of CHAT activity (Dobransky et al., 2004), and regulation of beta-amyloid production (Shirai et al., 2008) are of potential interest in view of hypothetical connection with ChEI’s treatment effects and AD disease course. Further analyses are required to explore the functional role of this variant. Rs17798800 is an intergenic SNP (Fig. 3A), and the eQTL analyses showed a significant association with NBEA gene (neurobeachin) expression levels in PBMCs but not in brain samples, even if there was a trend in the same direction to reduced expression levels in individuals who carry the A allele, predictive of nonresponse to treatment. NBEA is an A kinaseeanchoring protein putatively involved in dendritic spine formation and synaptic function, particularly in the trafficking of pre- and postsynaptic components (Niesmann et al., 2011). Previous studies have shown that knockout of NBEA completely blocks synaptic transmission (Su et al., 2004), and heterozygous disruptions of the NBEA gene have been linked to idiopathic cases of nonfamilial autism (Castermans et al., 2003, 2010). The 2 SNPs could be markers of treatment response or of disease severity because these findings can be explained as a response to ChEI therapy or as part of AD natural history. Genomic regions around these SNPs have not been reported as associated with disease progression, but additional typing of these 2 variants in untreated or placebo-treated AD control groups would be important to clarify this. Independent and larger prospective studies are also needed to confirm the ability to predict the response to treatment based on these genetic variants. In conclusion, this study represents a first effort in the field of personalized treatment of AD and calls for leveraging existing data sets to boost the statistical power and to replicate our findings. Novel therapeutic approaches for AD have been emerging in the last few years, and the availability of genomic and nongenomic biomarkers should help achieve optimized, individually tailored management of the disease by facilitating the identification of patients more likely to respond to ChEI. This approach would surely have a profound impact not only on the individual level but also on a public health level considering the cost and widespread use of this drug category. Disclosure statement There are no conflicts of interest to report for the authors of this manuscript. All subjects in this study gave written informed consent consistent with the Declaration of Helsinki. The study was approved by the local Institutional Review Boards of the centers involved in the study. This study was supported by Fondazione Cariplo (grant number 2008e2359 awarded to D.A.). Acknowledgements We are grateful to Massimo Zambernardi for technical assistance to patients, and we thank AD patients for their participation in this study, which was supported by a grant from CARIPLO Foundation, Milan, Italy (number 2008e2359 awarded to D.A.). We are grateful to J.D. Baggott for English editing.

F. Martinelli-Boneschi et al. / Neurobiology of Aging 34 (2013) 1711.e7e1711.e13

Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.neurobiolaging.2012.12.008. References Akaike, A., Takada-Takatori, Y., Kume, T., Izumi, Y., 2010. Mechanisms of neuroprotective effects of nicotine and acetylcholinesterase inhibitors: role of alpha4 and alpha7 receptors in neuroprotection. J. Mol. Neurosci. 40, 211e216. Albani, D., Boneschi, F.M., Biella, G., Giacalone, G., Lupoli, S., Clerici, F., Benussi, L., Ghidoni, R., Galimberti, D., Squitti, R., Mariani, S., Confaloni, A., Bruno, G., Mariani, C., Scarpini, E., Binetti, G., Magnani, G., Franceschi, M., Forloni, G., 2012. Replication study to confirm the role of CYP2D6 polymorphism rs1080985 on Donepezil efficacy in Alzheimer’s disease patients. J. Alzheimers 30, 745e749. Arking, D.E., Pfeufer, A., Post, W., Kao, W.H., Newton-Cheh, C., Ikeda, M., West, K., Kashuk, C., Akyol, M., Perz, S., Jalilzadeh, S., Illig, T., Gieger, C., Guo, C.Y., Larson, M.G., Wichmann, H.E., Marban, E., O’Donnell, C.J., Hirschhorn, J.N., Kaab, S., Spooner, P.M., Meitinger, T., Chakravarti, A., 2006. A common genetic variant in the NOS1 regulator NOS1AP modulates cardiac repolarization. Nat. Genet. 38, 644e651. Barnes, D.E., Yaffe, K., 2011. The projected effect of risk factor reduction on Alzheimer’s disease prevalence. Lancet Neurol. 10, 819e828. Birks, J., 2006. Cholinesterase inhibitors for Alzheimer’s disease. Cochrane Database Syst. Rev. Issue 1. Art. No.: CD005593. Blacker, D., Albert, M.S., Bassett, S.S., Go, R.C., Harrell, L.E., Folstein, M.F., 1994. Reliability and validity of NINCDS-ADRDA criteria for Alzheimer’s disease. The National Institute of Mental Health Genetics Initiative. Arch. Neurol. 51, 1198e1204. Blesa, R., Aguilar, M., Casanova, J.P., Boada, M., Martinez, S., Alom, J., de la Hoz, C.H., Sancho, J., Fernandez, O., Gil-Neciga, E., Masso, J.F., 2006. Relationship between the efficacy of rivastigmine and apolipoprotein E (epsilon4) in patients with mild to moderately severe Alzheimer disease. Alzheimer Dis. Assoc. Disord. 20, 248e254. Burns, A., Yeates, A., Akintade, L., Del Valle, M., Zhang, R.Y., Schwam, E.M., Perdomo, C.A., 2008. Defining treatment response to donepezil in Alzheimer’s disease: responder analysis of patient-level data from randomized, placebocontrolled studies. Drugs Aging 25, 707e714. Cacabelos, R., 2008. Pharmacogenomics and therapeutic prospects in dementia. Eur. Arch. Psychiatry Clin. Neurosci. 258 (suppl 1), 28e47. Calabria, M., Geroldi, C., Lussignoli, G., Sabbatini, F., Zanetti, O., 2009. Efficacy of acetyl-cholinesterase-inhibitor (ACHEI) treatment in Alzheimer’s disease: a 21month follow-up "real world" study. Arch. Gerontol. Geriatr. 49, e6ee11. Castermans, D., Volders, K., Crepel, A., Backx, L., De Vos, R., Freson, K., Meulemans, S., Vermeesch, J.R., Schrander-Stumpel, C.T., De Rijk, P., Del-Favero, J., Van Geet, C., Van De Ven, W.J., Steyaert, J.G., Devriendt, K., Creemers, J.W., 2010. SCAMP5, NBEA and AMISYN: three candidate genes for autism involved in secretion of large dense-core vesicles. Hum. Mol. Genet. 19, 1368e1378. Castermans, D., Wilquet, V., Parthoens, E., Huysmans, C., Steyaert, J., Swinnen, L., Fryns, J.P., Van de Ven, W., Devriendt, K., 2003. The neurobeachin gene is disrupted by a translocation in a patient with idiopathic autism. J. Med. Genet. 40, 352e356. Chianella, C., Gragnaniello, D., Delser, P.M., Visentini, M.F., Sette, E., Tola, M.R., Barbujani, G., Fuselli, S., 2011. BCHE and CYP2D6 genetic variation in Alzheimer’s disease patients treated with cholinesterase inhibitors. Eur. J. Clin. Pharmacol. 67, 1147e1157. Choi, S.H., Kim, S.Y., Na, H.R., Kim, B.K., Yang, D.W., Kwon, J.C., Park, M.Y., 2008. Effect of ApoE genotype on response to donepezil in patients with Alzheimer’s disease. Dement. Geriatr. Cogn. Disord. 25, 445e450. Cirulli, E.T., Goldstein, D.B., 2010. Uncovering the roles of rare variants in common disease through whole-genome sequencing. Nat. Rev. Genet. 11, 415e425. Courtney, C., Farrell, D., Gray, R., Hills, R., Lynch, L., Sellwood, E., Edwards, S., Hardyman, W., Raftery, J., Crome, P., Lendon, C., Shaw, H., Bentham, P., 2004. Long-term donepezil treatment in 565 patients with Alzheimer’s disease (AD2000): randomised double-blind trial. Lancet 363, 2105e2115. Dobransky, T., Doherty-Kirby, A., Kim, A.R., Brewer, D., Lajoie, G., Rylett, R.J., 2004. Protein kinase C isoforms differentially phosphorylate human choline acetyltransferase regulating its catalytic activity. J. Biol. Chem. 279, 52059e52068. Duncan, E., Danoy, P., Kemp, J.P., Leo, P.J., McCloskey, E., Nicholson, G.C., Eastell, R., Prince, R.L., Eisman, J.A., Jones, G., Sambrook, P.N., Reid, I.R., Dennison, E.M., Wark, J., Richards, J.B., Uitterlinden, A.G., Spector, T.D., Esapa, C., Cox, R.D., Brown, S.D., Thakker, R.V., Addison, K.A., Bradbury, L.A., Center, J.R., Cooper, C., Cremin, C., Estrada, K., Felsenberg, D., Gluer, C.C., Hadler, J., Henry, M.J.,

1711.e13

Hofman, A., Kotowicz, M.A., Makovey, J., Nguyen, S.C., Nguyen, T.V., Pasco, J.A., Pryce, K., Reid, D.M., Rivadeneira, F., Roux, C., Stefansson, K., Styrkarsdottir, U., Thorleifsson, G., Tichawangana, R., Evans, D.M., Brown, M.A., 2011. Genomewide association study using extreme truncate selection identifies novel genes affecting bone mineral density and fracture risk. PLoS Genet. 7, e1001372. Ge D, Goldstein D.B. 2008. SNPExpress. Available at: http://compute1.lsrc.duke.edu/ softwares/SNPExpress/. Harold, D., Macgregor, S., Patterson, C.E., Hollingworth, P., Moore, P., Owen, M.J., Williams, J., O’Donovan, M., Passmore, P., McIlroy, S., Jones, L., 2006. A single nucleotide polymorphism in CHAT influences response to acetylcholinesterase inhibitors in Alzheimer’s disease. Pharmacogenet. Genomics 16, 75e77. Ioannidis, J.P., Patsopoulos, N.A., Evangelou, E., 2007. Uncertainty in heterogeneity estimates in meta-analyses. BMJ 335, 914e916. Katsuno, M., Tanaka, F., Sobue, G., 2012. Perspectives on molecular targeted therapies and clinical trials for neurodegenerative diseases. J. Neurol. Neurosurg. Psychiatry 83, 329e335. Klimkowicz-Mrowiec, A., Marona, M., Spisak, K., Jagiella, J., Wolkow, P., Szczudlik, A., Slowik, A., 2011. Paraoxonase 1 gene polymorphisms do not influence the response to treatment in Alzheimer’s disease. Dement. Geriatr. Cogn. Disord. 32, 26e31. Motsinger-Reif, A.A., Jorgenson, E., Relling, M.V., Kroetz, D.L., Weinshilboum, R., Cox, N.J., Roden, D.M., 2010. Genome-wide association studies in pharmacogenomics: successes and lessons. Pharmacogenet. Genomics. doi: 10.1097/FPC. 0b013e32833d7b45. NHS. 2011. National Institute for Health and Clinical Excellence. Available at: http:// www.nice.org.uk. Niesmann, K., Breuer, D., Brockhaus, J., Born, G., Wolff, I., Reissner, C., Kilimann, M.W., Rohlmann, A., Missler, M., 2011. Dendritic spine formation and synaptic function require neurobeachin. Nat. Commun. 2, 557. Noh, M.Y., Koh, S.H., Kim, Y., Kim, H.Y., Cho, G.W., Kim, S.H., 2009. Neuroprotective effects of donepezil through inhibition of GSK-3 activity in amyloid-betainduced neuronal cell death. J. Neurochem. 108, 1116e1125. Pilotto, A., Franceschi, M., D’Onofrio, G., Bizzarro, A., Mangialasche, F., Cascavilla, L., Paris, F., Matera, M.G., Pilotto, A., Daniele, A., Mecocci, P., Masullo, C., Dallapiccola, B., Seripa, D., 2009. Effect of a CYP2D6 polymorphism on the efficacy of donepezil in patients with Alzheimer disease. Neurology 73, 761e767. Pola, R., Flex, A., Ciaburri, M., Rovella, E., Valiani, A., Reali, G., Silveri, M.C., Bernabei, R., 2005. Responsiveness to cholinesterase inhibitors in Alzheimer’s disease: a possible role for the 192 Q/R polymorphism of the PON-1 gene. Neurosci. Lett. 382, 338e341. Purcell, S., Neale, B., Todd-Brown, K., Thomas, L., Ferreira, M.A., Bender, D., Maller, J., Sklar, P., de Bakker, P.I., Daly, M.J., Sham, P.C., 2007. PLINK: a tool set for wholegenome association and population-based linkage analyses. Am. J. Hum. Genet. 81, 559e575. Rasovska, H., Rektorova, I., 2011. Instrumental activities of daily living in Parkinson’s disease dementia as compared with Alzheimer’s disease: relationship to motor disability and cognitive deficits: a pilot study. J. Neurol. Sci. 310, 279e282. Reale, M., Iarlori, C., Gambi, F., Feliciani, C., Salone, A., Toma, L., DeLuca, G., Salvatore, M., Conti, P., Gambi, D., 2004. Treatment with an acetylcholinesterase inhibitor in Alzheimer patients modulates the expression and production of the pro-inflammatory and anti-inflammatory cytokines. J. Neuroimmunol. 148, 162e171. Scacchi, R., Gambina, G., Moretto, G., Corbo, R.M., 2009. Variability of AChE, BChE, and ChAT genes in the late-onset form of Alzheimer’s disease and relationships with response to treatment with Donepezil and Rivastigmine. Am. J. Med. Genet. B Neuropsychiatr. Genet. 150B, 502e507. Shirai, Y., Adachi, N., Saito, N., 2008. Protein kinase Cepsilon: function in neurons. FEBS J. 275, 3988e3994. Su, Y., Balice-Gordon, R.J., Hess, D.M., Landsman, D.S., Minarcik, J., Golden, J., Hurwitz, I., Liebhaber, S.A., Cooke, N.E., 2004. Neurobeachin is essential for neuromuscular synaptic transmission. J. Neurosci. 24, 3627e3636. Varsaldi, F., Miglio, G., Scordo, M.G., Dahl, M.L., Villa, L.M., Biolcati, A., Lombardi, G., 2006. Impact of the CYP2D6 polymorphism on steady-state plasma concentrations and clinical outcome of donepezil in Alzheimer’s disease patients. Eur. J. Clin. Pharmacol. 62, 721e726. Wallin, A.K., Wattmo, C., Minthon, L., 2011. Galantamine treatment in Alzheimer’s disease: response and long-term outcome in a routine clinical setting. Neuropsychiatr. Dis. Treat. 7, 565e576. Wattmo, C., Wallin, A.K., Londos, E., Minthon, L., 2011. Predictors of long-term cognitive outcome in Alzheimer’s disease. Alzheimers Res. Ther. 3, 23. Yang, J., Wray, N.R., Visscher, P.M., 2010. Comparing apples and oranges: equating the power of case-control and quantitative trait association studies. Genet. Epidemiol. 34, 254e257.