Molecular Psychiatry (2009) 14, 804–819 & 2009 Nature Publishing Group All rights reserved 1359-4184/09 $32.00 www.nature.com/mp ORIGINAL ARTICLE Association of the NPAS3 and five other loci with response to the antipsychotic iloperidone identified in a whole genome association study C Lavedan, L Licamele, S Volpi, J Hamilton, C Heaton, K Mack, R Lannan, A Thompson, CD Wolfgang and MH Polymeropoulos Vanda Pharmaceuticals Inc., Rockville, MD, USA

A whole genome association study was performed in a phase 3 clinical trial conducted to evaluate a novel antipsychotic, iloperidone, administered to treat patients with schizophrenia. Genotypes of 407 patients were analyzed for 334 563 single nucleotide polymorphisms (SNPs). SNPs associated with iloperidone efficacy were identified within the neuronal PAS domain 3 gene (NPAS3), close to a translocation breakpoint site previously observed in a family with schizophrenia. Five other loci were identified that include the XK, Kell blood group complex subunit-related family, member 4 gene (XKR4), the tenascin-R gene (TNR), the glutamate receptor, inotropic, AMPA 4 gene (GRIA4), the glial cell line-derived neurotrophic factor receptor-alpha2 gene (GFRA2), and the NUDT9P1 pseudogene located in the chromosomal region of the serotonin receptor 7 gene (HTR7). The study of these polymorphisms and may lead to a better understanding of the etiology of schizophrenia and of its treatment. These results provide new insight into response to iloperidone, developed with the ultimate goal of directing therapy to patients with the highest benefit-to-risk ratio. Molecular Psychiatry (2009) 14, 804–819; doi:10.1038/mp.2008.56; published online 3 June 2008 Keywords: iloperidone; antipsychotic; NPAS3; pharmacogenomics; schizophrenia

Introduction the most optimal drug and dosage with less trial and error. Schizophrenia is a psychotic disorder affecting Iloperidone is an investigational drug that has some approximately 1% of the population.1 It is character- of the receptor-binding characteristics of current ized by the presence of positive symptoms (unusual atypical antipsychotics, such as a high 5-HT2A/D2 thoughts or perceptions, including hallucinations and binding ratio, which predicts a tolerability profile delusions), negative symptoms (social withdrawal, better than that of typical antipsychotic agents.6–9 It lack of pleasure in everyday life) and impaired also has receptor-binding characteristics that distin- cognitive functions (verbal memory, information guish it from other atypical antipsychotics, including processing). A number of drugs have been approved high affinity for a2C-adrenergic receptors, which may to treat this chronic illness. However, patient re- result in antidepressant and anxiolytic activity and in sponse to treatment remains highly variable,2,3 and improved cognitive function.6,7 Additionally, it has a the discontinuation rate with antipsychotic treatment low affinity for histamine receptors, which may is high.4 No single antipsychotic agent offers optimal reduce the risk for sedation and weight gain.6,8 effect for every patient with schizophrenia. Few data Iloperidone treatment is associated with a low are available to guide clinicians and patients in the incidence of clinically notable weight gain, low rate selection of the most appropriate medication, and in of extrapyramidal symptoms and akathisia, and improving the treatment specificity for an individual minimal prolactin elevations.10–12 patient.5 Pharmacogenomics provides the opportu- Iloperidone is rapidly absorbed in humans and nity to discover genetic markers predictive of re- extensively metabolized in the liver via multiple sponse. Knowing how a patient with schizophrenia pathways, including pathways mediated by the might respond to a particular therapy based on his or cytochrome P450 enzymes CYP2D6 and CYP3A4. her genetic makeup may enable clinicians to select Furthermore, a pharmacokinetic–pharmacodynamic relationship for the efficacy of iloperidone has been established: statistically significantly greater propor- Correspondence: Dr C Lavedan, Vanda Pharmaceuticals Inc., 9605 tions of responders (X20% improvement from base- Medical Center Drive, Suite 300, Rockville, MD 20850, USA. line) in iloperidone-treated patients had average E-mail: [email protected] À1 Received 3 September 2007; revised 18 February 2008; accepted plasma concentrations X5ngml compared with 25 April 2008; published online 3 June 2008 those with < 5 ng mlÀ1.13 Genetic markers of iloperidone efficacy C Lavedan et al 805 A particular challenge in searching for genetic placebo. The primary efficacy variable was change markers of response to antipsychotic treatment, from baseline to the last scheduled observation especially in double-blind, placebo-controlled stu- in Positive and Negative Syndrome Scale total dies, is that patients are prone to drop out early if they (PANSS-T) score. Comparison of iloperidone re- experience unsatisfactory results before reaching sponse to placebo response was performed to ensure therapeutic steady state concentrations of the active the interpretation of the results and to comply with drug. Two methods have been used to deal with requirements of regulatory agencies. missing values from patient dropout. The first method, known as last observation carried forward PANSS measurements (LOCF), replaces a patient’s missing measurements at During the course of the clinical trial, the PANSS the end of the study with the last recorded measure- score was assessed at baseline and on days 7, 10, 14, ment. LOCF has been used extensively in regulatory 21 and 28 or at the last visit when early termination settings despite recognized biases. In particular, occurred ( < 36% of patients). The PANSS ratings LOCF, which assumes that response of a dropped- were only completed by a qualified MD, DO, PhD, out patient remains constant from the last observed PharmD, BA, BS or trained health care professional. value to the end of the trial, tends to bias efficacy Only those raters who had been qualified by United estimates. It also ignores trends over time and fails to BioSource Corporation (Bethesda, MD, USA) were account for missing-data uncertainty.14–17 The second allowed to rate in this study. Raters were to have a method, called mixed-effects model repeated mea- minimum of 1 year of experience using these scales or sures (MMRM) analysis, which facilitates unbiased comparable rating instruments in rating patients with estimates of postdropout behavior, was recently psychiatric disorders. Each site had to have at least developed to estimate efficacy parameters directly two certified raters for PANSS. Furthermore, inter- from incomplete data.18 rater reliability tests were undertaken. For prequali- We report here on the results of a whole-genome fication, raters were asked to view video recordings of association study (WGAS) of the efficacy response of patient interviews and complete rating of all items. At the novel atypical antipsychotic iloperidone in a an investigators’ meeting, all potential raters were phase 3 trial using both the LOCF and the MMRM further trained until acceptable results were achieved. methods. This clinical trial was a randomized, Interview skills were also assessed for all raters. double-blind, placebo-controlled and ziprasidone- Furthermore, while the study was in progress, controlled multicenter study conducted to evaluate refresher training was conducted, and new training the efficacy, safety, and tolerability of a 24 mg per day tapes were sent out to each site for rater reevaluation dose of iloperidone administered as 12 mg twice a day at appropriate intervals. All the new raters were for 28 days to patients with schizophrenia in acute required to be trained. exacerbation. We identified several single nucleotide polymorphisms (SNPs) associated with iloperidone Pharmacogenomics population efficacy. Further study of these polymorphisms and of Participation in the WGAS was optional, and written associated genes may lead to better understanding of informed consent was required. Blood samples were the etiology of schizophrenia and of patient responses collected, and quality genotype obtained (Quest to antipsychotic treatments. Diagnostics Laboratories, Van Nuys, CA, USA) from 426 of 457 patients who consented to the optional WGAS, including 218 patients administered iloper- Materials and methods idone, 103 treated with the active comparator and 105 Ethical conduct of the study who received placebo. The modified intent-to-treat The clinical study was conducted in accordance with (ITT) population contained all randomized patients the Declaration of Helsinki; with the US Code of who received at least one dose of double-blind study Federal Regulations governing the protection of hu- medication and from whom a baseline PANSS score man subjects (21 CFR 50), Institutional Review measurement was obtained and at least one postbase- Boards (21 CFR 56), and the obligations of clinical line PANSS efficacy measurement was obtained while investigators (21 CFR 312); and with the International on study medication. The ITT population used for the Conference on Harmonisation Guidance for Good analysis consisted of 407 patients, including 210, 98 Clinical Practice (Topic E6). and 99 patients who received iloperidone, ziprasi- done or placebo, respectively. More than 64% of Patients and study design patients completed the 28-day period. Patients 18–65 years of age with diagnosis of schizo- phrenia according to the Diagnostic and Statistical Sex and race of patients Manual of Mental Disorders, 4th edn1 were eligible to The iloperidone-treated patients of the ITT popula- participate in a randomized, double-blind, placebo- tion included 174 men (83%) and 36 women (17%). and ziprasidone-controlled, 28-day study of the Races were described as Asian (N = 17, 8%), black or efficacy and safety of iloperidone. Patients were African-American (N = 105, 50%), white (N = 80, randomly assigned to iloperidone 12 mg twice a day, 38%) or other (N = 8, 4%). This race assignment was ziprasidone 80 mg twice a day (active control) or consistent with the population structure inferred

Molecular Psychiatry Genetic markers of iloperidone efficacy C Lavedan et al 806 from the genotype data from more than 6000 random a one-way ANOVA of each SNP. Each genotype class loci tested as previously described.19 was tested for association with change in PANSS-T between baseline and day 28 against the class of the Statistical analysis other two genotypes combined; results from the ANOVA dictated the parsimonious genetic model to Two-phase approach. To independently discover be used for additional testing. MMRM analysis was and confirm biomarkers of iloperidone efficacy, we subsequently conducted using the parsimonious randomly separated patients who were administered genetic model of each SNP in the MMRM model, as iloperidone and who were genotyped for the used for the analysis of the clinical trial results.11 In WGAS into two groups, Group 1 (discovery phase) this MMRM analysis, only the best genetic model of and Group 2 (confirmatory phase). We used PROC each SNP, as determined in the ANOVA LOCF data SURVEYSELECT (SAS Institute, Cary, NC, USA) to analysis, was used. For each SNP, the null hypothesis stratify the groups to have approximately equal tested was that there was no difference in mean numbers of patients of each race and each sex. In PANSS-T change from baseline between patients of Group 1 (106 patients), each SNP was tested for different genotype classes. The PANSS-T score was potential association between genotype and change in considered as a continuous variable. Total scores were PANSS-T between patients in the low 30% (N = 30) assumed to be missing at random (MAR) and the and the high 30% (N = 32) extremes of the distribution change from baseline in PANSS-T rating was analyzed of day 28 LOCF data. The 30% threshold allowed the using a MMRM model that included the following selection of more than 30 samples at each extreme time points: day 7, day 10, day 14, day 21 and day 28 that represents patients with distinct efficacy (end of study or early termination). MMRM model phenotypes (mean change in PANSS-T À29.5 and was conducted in SAS using PROC MIXED through þ 9.2 for the low and high 30%, respectively). generalized least squares estimation. Models Calculation was done to estimate the power to included no intercept term, the three-way products detect a significant difference between distinct of an indicator for SNP (as dichotomized by genotype genotype classes for the two extremes of the class) with dummy indicators for each treatment distribution, based on the change in PANSS-T and group and dummy indicators for each time point standard deviation (s.d.) observed in the extremes of (days 7, 10, 14, 21 and 28) to allow arbitrary trends Group 1. The difference change in PANSS-T between across time, as well as covariates for centered baseline the two extremes (D) was 38.7 points, and the average response for each of the five time points and dummy s.d. was 9.9. We conducted a conservative power indicators for pooled center. Under this coding analysis assuming higher s.d. (15 and 20), and weaker scheme, the coefficient of each treatment group  differences in PANSS-T change (20 and 25 points). time point indicator  SNP was the estimated mean The lowest power calculated with D = 20 and s.d. = 20 change scores for that treatment group and time point was 99 and 78% for an a of 0.05 and 0.0005, and SNP, adjusted for heterogeneity at baseline and respectively. With D = 25, power was 96 and 100% heterogeneity among centers. Specified covariance for s.d. of 20 and 15, respectively, at a 0.0005, and still structure was implemented by REPEATED statement 88 and 100% at a 0.00005. Analysis was performed using Type = Unstructured with covariance parameters using a truncated nonparametric measure analysis of estimated by restricted maximum likelihood. variance (ANOVA), based on the Kruskal–Wallis test, with genotypes as the grouping variables. Based on a Statistical correction for multiple testing. In the significant omnibus test and ad hoc contrasts, the confirmatory step of the two-phase approach, results that produced the most parsimonious genetic Bonferroni corrections of a = 0.05/100 were applied. model (homozygous for one allele vs all other False discovery rate (FDR) corrections were also genotypes [for example, A/A vs A/B and B/B], or calculated using the Benjamini and Hochberg (BH) heterozygous vs homozygous genotypes [A/B vs A/A procedure.20 Within each of the two single-phase and B/B]) dictated the genetic model that was tested analyses, the BH procedure was used; SNPs with the in the confirmatory phase. For ease of global lowest P-values were analyzed without any specific description, given that each SNP might have a threshold. FDR corrections were obtained from PROC different parsimonious model, the genotypes of all MULTTEST (SAS Institute), which produces BH SNPs associated with positive iloperidone response adjustments to original analytical P-values. were referred to as genotype class C1, and the remaining genotypes for that SNP were referred to Genotyping methods as genotype class C2. If no parsimonious genetic model was obtained for a SNP, it was not tested DNA samples were genotyped for more than 500 000 further. Only the subset of 100 SNPs with the lowest SNPs using a microarray set (GeneChip Human P-values for the contrasts that defined C1 and C2 was Mapping 500 K Array Set; Affymetrix, Santa Clara, tested in the confirmatory phase (104 patients). CA, USA) according to the manufacturer’s instruc- tions. The set was composed of two arrays, each Single-phase approaches. Two separate methods capable of genotyping on average 250 000 SNPs were applied. LOCF data for PANSS-T were used in (B262 000 for Nsp arrays and 238 000 for Sty arrays).

Molecular Psychiatry Genetic markers of iloperidone efficacy C Lavedan et al 807 DNA samples were also genotyped for additional compared with the genotypes of all other samples to polymorphisms previously studied for association identify potential duplicate samples. If > 90% of the with schizophrenia or antipsychotic treatment3,21–28 genotypes were identical between two arrays, DNA but not represented on the microarray set. SNPs of was retested to confirm genotypes and, if necessary, interest were genotyped using either polymerase to eliminate duplicate samples. chain reaction (PCR) followed by restriction fragment length polymorphism detection or by allele-specific Lack of DNA contamination. Lack of DNA primer (ASP) PCR (see Table 1). Each PCR reaction contamination of a sample by another sample was contained 60 ng DNA, 1 Â PCR reaction buffer with assessed for each array by determining the overall

MgSO4 (0.5 mM), DNA polymerase (AmpliTaq; Ap- percentage of SNPs with heterozygote calls, as plied Biosystems, Foster City, CA, USA), 5 pmol calculated by the dynamic model-based algorithm. dNTP mix (Promega, Madison, WI, USA) and 10 pmol Genotype calls from arrays with < 30% heterozygote of each specific primer in a total volume of 25 ml. For calls were considered to have come from a pure DNA the ASP assays, HotStarTaq polymerase (Qiagen, sample. Only those arrays with < 30% heterozygote Valencia, CA, USA) was used, and two PCR reactions calls were used for further analysis. were performed for each sample, one for each allele. PCR conditions were as follows: 95 1C initial dena- Sex determination. For each DNA sample, sex was turation for 10 min, then 35 cycles of 30-s denatura- blindly determined by the BRLMM algorithm based tion at 95 1C, 30-s annealing (temperature specific to on the percentage of heterozygosity of the SNPs on the SNP), 1-min extension at 72 1C and finally 10-min X . The result was compared with the extension at 72 1C. PCR products were digested with expected sex. In case of divergent data, a PCR-based the appropriate restriction enzyme and separated on a assay for the amelogenin gene (AMELX) was 4% agarose gel against standard DNA ladders to performed on the original sample and on a new determine fragment sizes. DNA aliquot. A new genotyping experiment was performed with the Sty and Nsp arrays. Samples Genotyping data from individual PCR assays with incompatible results were excluded from the WGAS. Each genotype was determined independently by two scientists. Occasional absences of genotype call and Allele frequency. This project was focused on the most ambiguities were resolved by repeat genotyping of the common polymorphisms by selecting SNPs with minor DNA sample. allele frequency X10%. This 10% cutoff is commonly used in genome-wide studies31–33 as it reduces the Genotyping data from DNA arrays number of SNPs that cannot add information given the power of the study. It also raises the proportion of SNPs To ensure the integrity of the data collected from DNA with high informativity and increases the chance to arrays, the following quality control steps were taken. observe all three genotype for most SNPs.

Algorithms. Each microarray was analyzed with the SNP selection. A reference DNA (Human Genomic dynamic model-based genotyping algorithm29 and with DNA 103 Control; Affymetrix) with available genotypes the latest Bayesian robust linear model with Mahalanobis for SNPs on the microarray set was systematically tested (BRLMM) distance classifier with a confidence threshold in parallel with the patient samples. Eight DNA 103 of 0.5.30 Under these conditions, missing genotypes were replicates for the Sty and Nsp arrays were obtained and assumed MAR, and no imputation was made for any analyzed. Individual SNP assays were considered to genetic data. The BRLMM Analysis Tool 2.0 and the SNP make accurate and reproducible calls (concordant calls) Signal Tool 1.0.0.12. (Affymetrix) were used to analyze when the calls (A/A, A/B, B/B) were identical across the and visualize the distribution and separation of genotype eight replicates and identical to the reference calls calls of individual SNPs. provided by Affymetrix. Only SNPs with 100% con- cordance for DNA 103 were retained for WGAS. Five Call rate. Call rate was defined for a single array as SNPs that cross-hybridized with the Y chromosome and the percentage of SNPs that were called A/A, A/B or all SNPs on the X chromosome were not used. The total B/B by the BRLMM algorithm. Only arrays with number of SNPs analyzed in this WGAS was 334 563. X93% call rate were retained for further analysis. Results Concordance between arrays. Of the more than 500 000 SNPs genotyped, 50 were common to both The effectiveness of iloperidone for the treatment of the Sty and the Nsp arrays. Only arrays with > 90% schizophrenia was assessed using PANSS, a standar- concordance for these SNPs were used in further dized method commonly used in clinical studies of analysis. antipsychotics. Genetic association with the change in the PANSS-T between baseline and day 28 was Uniqueness of samples. The B250 000 SNP data tested for 334 563 SNPs. In an association study with obtained per sample on each array (Sty or Nsp) were such a large number of SNPs, the selection of the best

Molecular Psychiatry 808 oeua Psychiatry Molecular

Table 1 Genotyping assays for selected SNPs in candidate genes and for the six SNPs associated with iloperidone efficacy

Gene dbSNP Method Primer Sequences Temperature Enzyme Allele 1 Product (1) Allele 2 Product (2) (1C)

COMT rs4680 RFLPa F: GGA GCT GGG GGC CTA CTG TG 61 NlaIII G 114 þ 36 þ 35 A 96 þ 38 þ 34 þ 18 (Val158Met) R: GGC CCT TTT TCC AGG TCT GAC A DRD2 rs6275 RFLP F: GGA ATG GGA CCT TTC ACA GA 63 NcoI C 360 T 253 þ 107 (His313) R: GAG CTG GAG ATG GAG ATG CT DRD4 rs1800955 RFLPa F: GGA ATG GAG GAG GGA GCG GG 60 FspI C 562 þ 43 T 342 þ 220 þ 43 (À521C/T) R: CGC TCC ACC GTG AGC CCA GTA T GRM3 rs274622 RFLP F: CCG ACA ACC ATA GCA GGA CT 58 Sau96I C 124 þ 27 T 151 R: TCG CAT TCT CCC AAA TTT CT b a

HTR1A Not available ASP F1: GAA GAC CGA GTG TGT CTT CC 58 NA C NA G NA efficacy iloperidone of markers Genetic [C(À1019)G] F2: GAA GAC CGA GTG TGT CTT CG R: CTG AGG GAG TAA GGC TGG AC HTR2A rs6313 RFLPa F: TCT GCT ACA AGT TCT GGC TT 58 HpaII C 216 þ 125 T 341 (T102C) R: CTG CAG CTT TTT CTC TAG GG NOS1 rs3782221 ASP F1: CAC AGA CCC ACA GAA CCT GAG TA 55 NA A NA G NA F2: ACA GAC CCA CAG AAC CTG AGT G R: GCA CGG AGA AGA GTG AGA GG Lavedan C SYN2 rs795009 RFLP F: TCC CTG AGA TCG ATG TCC TC 58 Cac81 A 235 C 115 þ 80 R: GTC TGG GCA AGA CCC TTG TA SYN2 rs308961 ASP F: GTT CAC CGT GGA ATC TGA TG 56 NA A NA C NA al et R1: GGT TTC AGT TCG GAG ATA GGA G R2: GGT TTC AGT TCG GAG ATA GGA T NPAS3 rs11851892 RFLP F: CCC CAG CCC TAC CTA TCA ATT AC 55 BfuAI, BspMI G 208 þ 29 A 237 R: CTA CAT GCT GCA TTT TCA CCT G XKR4 rs9643483 RFLP F: ACA AAC CTG AAC ATC CTG CAC 55 BstEII G 196 þ 25 T 221 R: AGA TTT TTC TTT TTA ACT CTG GGT GAC TNR rs875326 RFLP F: GTA CCA GCG GCA TTG CAT AG 55 BbvI G 197 þ 32 A 229 R: AGC AGG ATA AGC AGG AAG CAG GRIA4 rs2513265 RFLP F: CTA AAG GCG AAA CAT GTG TAT G 55 BtsCI T 144 þ 78 A 95 þ 78 þ 49 R: CAT AGC AGC TGA CCA ACA TTC C GFRA2 rs7837682 RFLP F: CTG CCA CCC TAC TCA ACA CC 55 Taq aI A 205 G 74 þ 131 R: CAA TCT GCT TTC ACC TCA ACC C NUDT9P1 rs4528226 RFLP F: GAG CAG ATA AAT TTG CCT GG 55 MlyI T 193 þ 64 G 140 þ 64 þ 53 R: GTT GAA CCC AAA GAC ACC CCT G

Abbreviations: ASP, allele-specific primer; RFLP, restriction fragment length polymorphism; SNP, single nucleotide polymorphism. Gene names correspond to the NCBI official symbols. SNPs are identified by their unique dbSNP number (http://www.ncbi.nlm.nih.gov/projects/SNP/), when available. Their common designation in the literature is indicated between parentheses below the dbSNP number. Primer sequences are from 50 to 30; F and R refer to forward and reverse primers used in the PCR reaction, respectively (for ASP assays, 1 and 2 distinguish the allele- specific primers). Annealing temperature of each PCR reaction is indicated in degrees Celsius (1C). PCR product size after complete enzymatic digestion is shown for each allele in base pairs. aPrimer sequences were previously reported.25–28 bThe SNP tested within the HTR1A gene was previously referred to as C(–1019)G27 ; no dbSNP number is listed for this SNP. Genetic markers of iloperidone efficacy C Lavedan et al 809

Figure 1 Genome view of WGAS results for iloperidone response in the treatment of patients with schizophrenia. WGAS results of the single-phase approach analyses are shown per chromosome for MMRM (a) and LOCF (b) analyses. Each point shows the P-value (logarithmic scale on vertical axis) and the genomic location (horizontal axis) of one SNP for the association of change in PANSS-T score using the most parsimonious genetic model of that SNP. Stars designate selected SNPs described in the text. Abbreviations: LOCF, last observation carried forward; MMRM, mixed-effects model repeated measures.

candidates is based on extreme P-values. These may the lowest P-values were common between these two be more sensitive than usual to the particulars of the analyses. Six of these SNPs were also among the top 100 statistical method used. Thus, to select the most likely SNPs of the two-phase analysis. We report in more markers of the efficacy of iloperidone, we used detail on these six SNPs (rs11851892, rs9643483, several analytical methods and focused on those rs875326, rs2513265, rs7837682, rs4528226) and their SNPs with persistently low P-values across different association with iloperidone response (Table 2). Data methods. We performed three analyses. Analysis 1 from the different treatment groups were used in the was a two-phase approach in which DNA samples MMRM statistical model, which allowed us to compare were separated into two groups. The first group genotype effect between iloperidone treatment and consisted of 50% of the samples for a discovery placebo. Except when otherwise specified, results phase in which only the top 30% and the bottom 30% discussed herein refer to PANSS-T data and to P-values of the change in PANSS-T (LOCF data) were used, generated from the MMRM analysis. and the second was a hold-out group consisting of the other 50% for a confirmatory phase. Only the 100 NPAS3 and iloperidone response SNPs with the lowest P-values in the discovery phase SNP rs11851892 is located on chromosome 14q12-q13 were subsequently tested. A similar method was within intron 3 of the NPAS3 gene, which codes for successfully applied recently to identify a genetic the neuronal period/aryl hydrocarbon receptor/sin- variant that modulates cardiac repolarization.34 Anal- gle-minded (PAS) domain protein 3. Three other ysis 2 was an ANOVA of LOCF data for all iloperidone- SNPs within this gene were also significant at the treated patients in a single-phase approach. Analysis 3 P < 0.01 level (Figure 2a). The non-G/G rs11851892 was an MMRM analysis of all patients using the most genotype, which had a frequency of 31%, was present parsimonious genetic model of each SNP. The two in a group of iloperidone-treated patients who single-phase approaches (analyses 2 and 3) provided averaged > 20-point improvement from baseline in more power to detect significant associations by taking PANSS-T (–20.12 [±3.69]), 68% greater than the advantage of all available samples. mean response (–12.01 [±1.03]) of all iloperidone- Overall whole genome association results from the treated patients (Table 3).11 Patients who carried the LOCF and MMRM single-phase approaches are non-G/G genotype were approximately three times shown in Figure 1. Forty five of the 100 SNPs with more likely to experience 20% improvement than

Molecular Psychiatry 810 oeua Psychiatry Molecular

Table 2 SNPs significantly associated with PANSS-T in iloperidone-treated group

P Rank

SNP dbSNP Gene Location Position Geno- Disco- Confirma- BH- LOCFa BH- MMRMb BH- Discovery Confirma- ANOVA MMRM type very tory adjusted adjusted adjusted tory LOCF a b

P P P efficacy iloperidone of markers Genetic

SNP_A- rs11851892 NPAS3 14q12–q13 32921165 Non-G/G 0.000086 0.099 0.61 0.000028 0.86 0.000093 0.59 13 15 14 38 2048427 SNP_A- rs9643483 XKR4 8q12.1 55960228 Non-G/G 0.00013 0.066 0.55 0.000055 0.86 0.00017 0.59 28 12 45 70 2283283 SNP_A- rs875326 TNR 1q25.1 172021056 Non-A/G 0.00013 0.13 0.61 0.000040 0.86 0.000021 0.57 32 20 23 10

1973093 Lavedan C SNP_A- rs2513265 GRIA4 11q22 104929005 Non-T/T 0.00017 0.81 0.96 0.000042 0.86 0.000055 0.59 44 83 26 22 2284243 SNP_A- rs7837682 GFRA2 8p21.3 21329990 A/A 0.00021 0.13 0.61 0.000077 0.86 0.00018 0.59 62 21 60 73

2274533 al et SNP_A- rs4528226 NUDT9P1 10q23.31 92822226 G/T 0.00028 0.00041 0.041 0.00000011 0.12 0.00000078 0.26 76 1 1 1 2076797

Abbreviations: ANOVA, analysis of variance; LOCF, last-observation-carried-forward; MMRM, mixed-effects model repeated measures; SNP, single nucleotide polymorphism. SNPs are identified by their Affymetrix SNP nomenclature (SNP_A-) and their unique dbSNP number (http://www.ncbi.nlm.nih.gov/projects/SNP/). Each gene name corresponds to the National Center for Biotechnology Information (NCBI) official symbol of the gene closest to the SNP. The address of each SNP in the genome is indicated by the cytogenetic location and the physical position as listed in the NCBI Build 36.1, March 2006, as listed at http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db = snp. The genotype class associated with the better treatment response, as determined by the ANOVA test on LOCF data, is indicated. Allele designation refers to the Affymetrix annotation listed at http://www.ncbi.nlm.nih.gov/SNP/index.html. Rank refers to SNP in each analysis when ordered from lowest to highest P-value. P-values of the discovery phase were obtained from the truncated nonparametric measure ANOVA, based on the Kruskal–Wallis test with genotype as the grouping variable, using the LOCF data on the extremes of the distribution of 50% of the patients treated with iloperidone. P-values of the confirmatory phase were obtained using independent group t-tests of the change in PANSS-T on Day 28 LOCF data for patients in iloperidone Group 2, before and after correction for multiple testing (100 SNPs) using the FDR method (BH-adjusted P). aANOVA P-values were obtained with LOCF data of all iloperidone-treated patients before and after correction for multiple testing (BH-adjusted P). bP-values of the MMRM analysis were obtained between genotype classes among all iloperidone-treated patients before and after correction for multiple testing (BH-adjusted P). Genetic markers of iloperidone efficacy C Lavedan et al 811 patients who carried the G/G genotype (odds ratio smaller changes in PANSS-T than patients with (OR) = 2.74, P = 0.0011; Table 5). different genotypes (–3.30 [±3.93] vs –15.02 [±4.34], The NPAS3 gene has previously been linked to respectively) and than patients who received placebo schizophrenia. Kamnasaran et al.35 described a (Table 3). Patients who carried the non-G/G genotype mother and daughter with schizophrenia who were approximately 3.5 times more likely to experi- were carriers of a chromosome translocation, ence 20% improvement than patients who carried the t(9;14)(q34;q13). Two of the SNPs associated with G/G genotype (OR = 3.42, P = 0.0009; Table 5). change in PANSS-T in our study (rs11851892, SNP rs9643483 lies between the XK–Kell blood rs17583667) were localized within the same intron 3 group complex subunit-related family, member 4 of NPAS3 as the breakpoint observed in this family (XKR4) gene (B220 kb away) and the retinitis pig- (Figure 2a).35 mentosa 1 (RP1) gene (B250 kb away). No other SNPs in that region were significant at the P < 0.01 level XKR4 and iloperidone response (Figure 2b). SNP rs9643483 is located in 8q12.1. Its G/G genotype The XKR4 gene codes for a protein related to the XK was present in approximately 22% of iloperi- protein. Kell and XK are two distinct red blood cell done-treated patients, who, on average, experienced membrane that form the Kell blood group

Figure 2 Genetic association between change in PANSS-T score and SNPs in the NPAS3, XKR4, TNR, GRIA4, GFRA2 and NUDT9P1 gene regions. P-values (vertical axis) MMRM analysis are shown for the genomic regions containing NPAS3 (a), XKR4 (b), TNR (c), GRIA4 (d), GFRA2 (e) and NUDT9P1 (f) in order of the SNPs’ physical locations (horizontal axis). The direction of the maps relative to the centromere and to the telomere is indicated by horizontal arrows. SNPs with Pp0.01 are boxed. The approximate location in NPAS3 of the breakpoint described by Kamnasaran et al.35 in a family with schizophrenia is indicated by an asterisk (*). Abbreviations: cen, centromere; tel, telomere.

Molecular Psychiatry Genetic markers of iloperidone efficacy C Lavedan et al 812 Table 3 Genotype effect on iloperidone and placebo response

dbSNP Gene Genotype Iloperidone Placebo Iloperidone vs placebo Pa N Mean s.d. N Mean s.d.

rs11851892 NPAS3 A/A 9 À19.94 2.44 3 À8.97 3.33 A/G 56 À20.15 3.87 31 À11.10 3.50 G/G* 145 À9.20 4.25 63 À5.90 4.00 0.20 Non-G/Gw 65 À20.12 3.69 34 À10.91 3.49 0.0090 P 0.000093 0.34

rs9643483 XKR4 G/G* 47 À3.30 3.93 20 À11.21 3.77 0.12 G/T 96 À14.57 4.35 45 À6.33 4.21 T/T 67 À15.66 4.28 34 À7.19 4.00 Non-G/Gw 163 À15.02 4.34 79 À6.70 4.12 0.00043 P 0.00017 0.27

rs875326 TNR A/A 25 À14.99 4.27 10 À6.75 5.32 A/G* 76 À6.21 4.51 38 À7.27 3.51 0.72 G/G 105 À16.73 4.33 49 À7.81 4.40 Non-A/Gw 130 À16.39 4.35 59 À7.63 4.54 0.00061 P 0.000021 0.98

rs2513265 GRIA4 A/A 63 À15.30 4.46 33 À6.92 3.28 A/T 99 À15.17 4.16 50 À8.08 4.24 T/T* 48 À3.30 4.11 15 À4.84 4.03 0.86 Non-T/Tw 162 À15.22 4.26 83 À7.62 3.91 0.0010 P 0.000055 0.55

rs7837682 GFRA2 A/Aw 116 À15.95 4.23 59 À6.55 3.81 0.0011 A/G 64 À6.04 4.55 24 À7.97 4.28 G/G 18 À6.21 3.21 8 À9.79 3.97 Non-A/A* 82 À6.08 4.28 32 À8.43 4.22 0.50 P 0.00018 0.58

rs4528226 NUDT9P1 G/G 57 À7.14 4.10 28 À5.23 3.42 G/Tw 99 À18.73 3.99 42 À11.68 3.96 0.016 T/T 52 À6.04 4.45 28 À2.97 4.08 Non-G/T* 109 À6.61 4.29 56 À4.10 3.90 0.36 P 0.00000078 0.051

Abbreviation: SNP, single nucleotide polymorphism. SNPs are identified by their unique dbSNP number (http://www.ncbi.nlm.nih.gov/projects/SNP/). Each gene name corresponds to the NCBI official symbol of the gene closest to the SNP. The two genotypes classes used in the MMRM analysis are indicated as associated with lowest response (*) or highest response (w) to iloperidone treatment. Allele designation refers to the Affymetrix annotation, as listed at http://www.ncbi.nlm.nih.gov/SNP/index.html. Number of patients (N), mean of change in PANSS-T (Mean) and standard deviation (s.d.) of the mean shown are derived from the MMRM analysis. P-values are shown for the two genotype classes used in the analysis within each group, with the corresponding means in bold. aP-values of the effect of the two genotype classes tested in the MMRM analysis between the iloperidone and placebo groups are shown.

complex. XK is a putative membrane transporter in the RP1 gene have been reported to cause dominant whose substrate is unknown, but absence of the retinitis pigmentosa. The specific role of the protein protein is associated with McLeod syndrome.36 This encoded by this gene is still unknown, but it is rare, genetic, X-linked disease is characterized by red thought to be important in photoreceptor biology. The blood cell acanthocytosis and late-onset central association of retinitis pigmentosa with features of nervous system (CNS) and neuromuscular abnormal- psychosis has been reported in patients with Usher ities. CNS manifestations include cognitive impair- syndrome, but it is not well understood.38 ment, psychopathology, seizures and choreatic No reports have been published of an association movements. One report has mentioned schizophrenia between XKR4 or RP1 and schizophrenia or response as a manifestation of McLeod syndrome.37 Mutations to antipsychotic treatment.

Molecular Psychiatry Genetic markers of iloperidone efficacy C Lavedan et al 813 TNR and iloperidone response P = 0.00018) for patients with a different genotype Another SNP of interest is rs875326 on 1q25.1, (Table 3). More than 58% of iloperidone-treated located between the TNR (B2.5 kb away) and TNN patients who were homozygote A/A for rs7837682 genes (B170 kb away), which encode the tenascin-R were approximately three times more likely to and -N proteins, respectively. The genomic region experience 20% improvement than the patients who between TNR and TNN also contains the KIAA0040 carried a different genotype (OR = 2.95; P = 0.0004; locus of still unknown function (Figure 2c). The Table 5). rs875326 homozygote genotypes were associated with The closest known gene from rs7837682 is glial cell the highest response to iloperidone treatment, with a line-derived neurotrophic factor (GDNF) receptor-a2 mean change in PANSS-T of –16.39 (±4.35) com- (GFRA2), located approximately 360 kb away. Six pared with a mean change of –6.21 (±4.51; other SNPs downstream of this gene were also P = 0.000021) for patients with the heterozygote significant at the P < 0.01 level, including four SNPs genotype (Table 3). More than 63% of iloperidone- within approximately 15 kb (Figure 2e). GFRA2 codes treated patients homozygote for rs875326 were ap- for GDNF receptor-b (GDNFR-b). GDNF plays a key proximately 3.5 times more likely to experience 20% role in the control of neuron survival and differentia- improvement than those who carried the A/G geno- tion. Results from a study in an Italian population type (OR = 3.57, P < 0.0001; Table 5). suggest that a 30-UTR GDNF repeat polymorphism is Tenascin-R is an extracellular matrix protein ex- associated with schizophrenia.46 pressed primarily in the CNS and localized around motor neurons and on motor axons in the spinal cord, NUDT9P1 and iloperidone response cerebellum, hippocampus and olfactory bulb.39,40 It is SNP s4528226 was the only SNP that showed a a member of the tenascin gene family, which includes statistically significant association in the two-phase at least three other genes in humans—TNN on 1q, approach analysis at the P < 0.05 level after correction TNC on 9q33 and TNXB on 6p21.3. TNXB deficiency for multiple testing using either the Bonferroni leads to a clinically distinct, autosomal recessive form method or the FDR method (Table 2). This SNP also of Ehlers–Danlos syndrome, a severe connective showed the most significant statistical association tissue disorder. Two SNPs in the TNXB gene are when using all samples from the iloperidone-treated associated with schizophrenia in a British popula- group, with either the LOCF method or the MMRM tion.41 To date, no direct evidence has implicated method (Table 2). The rs4528226 G/T genotype was TNR or TNN in schizophrenia or in the response to associated with the highest response to iloperidone antipsychotic treatment. treatment, with a mean change in PANSS-T of –18.73 (±3.99) compared with a mean change of –6.61 (±4.29; GRIA4 and iloperidone response P = 0.00000078) for patients with a different genotype SNP rs2513265 is located on chromosome 11q22, (Table 3). More than 47% of iloperidone-treated patients approximately 58 kb from the 50 end of the glutamate heterozygote for rs4528226 were three times more likely receptor, inotropic, AMPA 4 (GRIA4) gene (Figure 2d). to experience a 20% improvement than homozygote The rs2513265 T/T genotype was present in approxi- patients (OR = 3.00, P = 0.0001; Table 5). mately 23% of iloperidone-treated patients who, on SNP rs4528226 is located on chromosome 10q23.31, average, experienced PANSS-T changes lower than close to the NUDT9P1 pseudogene, between the those of patients with a different genotype (–3.30 ankyrin repeat domain 1 (ANKRD1) gene and the [±4.11] vs –15.22 [±4.26], respectively, P = 0.000055) polycomb group ring finger 5 (PCGF5) gene, approxi- and comparable to those of patients receiving placebo mately 200 kb from the serotonin receptor 7 (HTR7) (Table 3). Patients who carried the non-T/T genotype gene (Figure 2f). No other SNPs in the same chromo- were approximately 2.5 times more likely to experi- somal region showed a statistically significant associa- ence 20% improvement than patients who carried the tion at the P < 0.01 level, including the closest flanking T/T genotype (OR = 2.43, P = 0.011; Table 5). Two SNPs located 3.5 and 6.3 kb away. The HTR7 gene, other SNPs, rs7930372 and rs4142579, located within which is considered a possible schizophrenia-suscept- approximately an 8-kb interval, were also significant ibility factor, has been associated with schizophrenia in at P < 0.01 (Figure 2d). Interestingly, some haplotypes a Japanese population.47 No link has been reported of the GRIA4 gene have shown a positive association between schizophrenia or the response to antipsychotic with schizophrenia in the Japanese population.42 treatment and NUDT9P1, ANKRD1 or PCGF5.

GFRA2 and iloperidone response Analysis of sex and race SNP rs7837682 is located in the 8p21.3 region, where SNP allele frequencies can vary between races, and the schizophrenia susceptibility locus referred to as recombination rates depend on sex; therefore, we SCZD6 has been mapped through a number of conducted ad hoc analyses of these potential con- positive genetic association studies.43–45 The founding factors. For all SNPs reported herein, the rs7837682 A/A genotype was associated with the association of a particular genotype with higher or highest response to iloperidone treatment, with a lower response was consistent between men and mean change in PANSS-T of –15.95 (±4.23) com- women (data not shown) and across races (Table 4), pared with a mean change of –6.08 (±4.28; indicating that the significant associations observed

Molecular Psychiatry Genetic markers of iloperidone efficacy C Lavedan et al 814 Table 4 Effect of genotypes in the PANSS-T score in the iloperidone-treated patients per race

dbSNP Gene Genotype Asian Black White Others

N Mean s.d. N Mean s.d. N Mean s.d. N Mean s.d.

rs11851892 NPAS3 A/A 0 2 À22.93 0.34 6 À18.57 1.66 1 À22.13 A/G 7 À20.44 3.63 26 À19.16 3.52 21 À21.21 4.04 2 À20.77 7.72 G/G* 10 À8.55 3.91 77 À8.37 4.14 53 À10.53 4.26 5 À9.30 4.15 Non-G/Gw 7 À20.44 3.63 28 À19.43 3.53 27 À20.62 3.79 3 À21.22 5.52 rs9643483 XKR4 G/G* 1 À7.76 36 À2.25 3.55 8 À5.49 2.55 2 À11.18 0.57 G/T 2 À11.70 2.55 53 À14.49 4.34 40 À14.84 4.50 1 À14.28 T/T 14 À14.82 4.08 16 À14.03 4.27 32 À17.25 4.07 5 À13.03 2.98 Non-G/Gw 16 À14.43 4.00 69 À14.38 4.30 72 À15.91 4.45 6 À13.24 2.72 rs875326 TNR A/A 1 À22.09 24 À14.69 4.09 0 0 A/G* 2 À6.81 3.67 46 À5.57 4.51 26 À7.09 4.48 2 À8.77 6.52 G/G 13 À15.60 3.93 34 À15.67 4.11 52 À17.76 4.42 6 À16.16 4.41 Non-A/Gw 14 À16.07 4.16 58 À15.27 4.09 52 À17.76 4.42 6 À16.16 4.41 rs2513265 GRIA4 A/T 7 À15.92 4.82 30 À15.44 3.83 23 À15.30 5.36 3 À12.49 1.98 A/T 8 À13.58 3.37 52 À14.10 4.11 36 À16.82 3.83 3 À18.26 4.04 T/T* 2 À5.05 3.39 23 À1.36 3.79 21 À4.99 3.56 2 À6.25 6.52 Non-T/Tw 15 À14.67 4.14 82 À14.59 4.04 59 À16.23 4.50 6 À15.38 4.25 rs7837682 GFRA2 A/Aw 15 À15.63 4.00 42 À14.95 4.26 52 À16.92 4.13 7 À15.44 4.55 A/G 2 À8.15 3.43 45 À5.38 4.31 16 À7.28 5.09 1 À11.95 G/G 0 13 À5.81 3.13 5 À7.23 3.55 0 Non-A/A* 2 À8.15 3.43 58 À5.48 4.06 21 À7.27 4.68 1 À11.95 rs4528226 NUDT9P1 G/G 7 À4.89 2.89 12 À6.54 5.07 34 À8.15 3.88 4 À4.30 1.46 G/Tw 9 À19.32 4.23 47 À18.19 3.53 41 À18.99 4.38 2 À23.27 3.49 T/T 0 46 À5.29 4.01 5 À11.34 3.65 1 À14.01 Non-G/T* 7 À4.89 2.89 58 À5.55 4.23 39 À8.56 3.96 5 À6.24 4.52

Abbreviation: SNP, single nucleotide polymorphism. SNPs are identified by their unique dbSNP number (http://www.ncbi.nlm.nih.gov/projects/SNP/). Each gene name corresponds to the NCBI official symbol of the gene closest to the SNP. The two genotypes classes used in the MMRM analysis are indicated as associated with lowest response (*) or highest response (w) to iloperidone treatment. Allele designation refers to the Affymetrix annotation, as listed at http:// www.ncbi.nlm.nih.gov/SNP/index.html. Number of patients (N), mean of change in PANSS-T (Mean) and standard deviation (s.d.) of the mean shown are derived from the MMRM analysis.

were not attributable to a bias resulting from the values of each selected marker (Table 5). SNP heterogeneity of the patient population studied. rs11851892 (NPAS3), which differentiated 31% of patients with superior iloperidone response, had the Analysis of PANSS subscales highest specificity (79%) and the strongest positive We also conducted an MMRM analysis with these six predictive value (65%). The two SNPs that differ- selected SNPs in the following PANSS subscales: entiated a small group ( < 23%) of iloperidone-treated positive symptom subscale (PANSS-P), negative patients with lower changes in PANSS-T, rs9643483 symptom subscale (PANSS-N) and general psycho- (XKR4) and rs2513265 (GRIA4), had the highest pathology subscale (PANSS-GP). For each SNP, a sensitivity (88 and 85%, respectively). The other strong genotype effect on the change from baseline at SNPs had intermediate values. day 28 was observed in the three subscales (Figure 3). The most statistically significant differences were observed for PANSS-T and PANSS-P with rs4528226 Validation of genotype calls (NUDT9P1; Figure 3f). The lowest P-values for Genotype calls generated by the microarray set were PANSS-N and PANSS-GP were obtained with also determined with PCR-based assays for the rs2513265 (GRIA4; Figure 3d) and rs875326 (TNR; iloperidone-treated patients. We observed an accu- Figure 3c), respectively. racy rate of 100% for rs9643483 (XKR4), rs875326 (TNR), rs2513265 (GRIA4) and rs4528226 Odds ratio, sensitivity, specificity and predictive values (NUDT9P1), 99.5% for rs11851892 (NPAS3), with of selected SNPs only one erroneous call, and an accuracy rate of 96% To better appreciate the potential clinical value of the for rs7837682 (GFRA2). Only 2 erroneous calls for markers in predicting response to iloperidone, we rs7837682 were between different genotype classes calculated OR, sensitivity, specificity and predictive (A/A vs non-A/A).

Molecular Psychiatry Genetic markers of iloperidone efficacy C Lavedan et al 815

Figure 3 Genotype effect on efficacy response measured on the PANSS-T scale and subscales. For each SNP (rs11851892 (a), rs9643483 (b), rs875326 (c), rs2513265 (d), rs7837682 (e), rs4528226 (f)) the improvement of symptoms was measured as a decrease in the respective scale between baseline and day 28. The improvement is here represented as a positive number on the y axis for each genotype class. Raw P-values from the MMRM analysis are shown below each graph.

Analysis of polymorphisms in candidate genes after 4 weeks of treatment, in the PANSS-T scale as Several SNPs previously studied for possible associa- well as in the positive, negative and general psycho- tion with schizophrenia or antipsychotic treatment pathology subscales (Table 3; Figure 3). response were genotyped for this WGAS using SNP rs11851892, within the NPAS3 gene, identified specifically designed assays (Table 1). They included a group of 31% of iloperidone-treated patients who SNPs in the COMT, DRD2, DRD4, GRM3, HTR1A, experienced greater improvement in PANSS-T. Two HTR2A, NOS1 and SYN2 genes.3,21–24 All other genes independent linkage studies have suggested that a of interest were tested only for the SNPs represented susceptibility locus for schizophrenia resides in on the microarray set. None of these SNPs showed chromosome 14q13.44,48 The NPAS3 gene, located on significant association with change in PANSS-T at the 14q12–q13, has already been linked to schizophrenia P < 0.01 level in both single-phase LOCF and MMRM in a family in which a mother and a daughter with the approaches. illness were carriers of a t(9;14)(q34;q13) transloca- tion with a breakpoint in intron 3.35 Of the NPAS3 Discussion SNPs significantly associated with change in PANSS-T in our study, two were in intron 3, including Through a WGAS performed in the context of a phase rs11851892, which had the most significant associa- 3 clinical trial that investigated the efficacy, safety tion for that region (P = 0.000028 for ANOVA LOCF; and tolerability of the novel atypical antipsychotic P = 0.000093 for MMRM). The NPAS3 protein belongs iloperidone, we identified several SNPs in six to a family of transcription factors that contain a basic chromosomal regions significantly associated with helix-loop-helix region followed by a PAS domain. efficacy response to iloperidone for the treatment of Kamnasaran et al.35 showed that the NPAS3 protein is schizophrenia. These SNPs showed a significant expressed in multiple adult brain tissues, including genotype effect on the improvement of symptoms, the hippocampus, thalamus and cortex. They

Molecular Psychiatry Genetic markers of iloperidone efficacy C Lavedan et al 816

þ proposed that haploinsufficiency of NPAS3 played a

ated role in schizophrenia in the family with the intron 3 PV breakpoint. Additional evidence supporting the À potential role of NPAS3 in schizophrenia comes from recent studies of mice. Pieper et al.49 showed that NPAS3-deficent mice manifest behavioral and ) for the genotype neuroanatomic abnormalities reminiscent of schizo- ymetrix annotation þ phrenia. Erbel-Sieler et al.50 demonstrated that NPAS3-homozygous adult mice had marked reduc- tions in reelin in the cortex, dentate gyrus and amygdala. Reelin is an extracellular matrix protein thought to control cell–cell interactions critical for cell positioning and neuronal migration during brain 51 Sensitivity Specificity PV development. In one genetic study, it was suggested that reelin is associated with treatment-resistant schizophrenia, and it has been shown that levels of reelin and its mRNA are significantly reduced in several brain areas of patients with undifferentiated 52

interval or paranoid schizophrenia. Our results provide the first evidence of a possible link between NPAS3 and the efficacy of an antipsychotic treatment. This finding in iloperidone-treated patients underscores the importance of understanding the role of this transcription factor in the etiology of schizophrenia and in the efficacy of antipsychotics. Of the other SNPs associated with change in

) for all other genotypes. PANSS-T in iloperidone-treated patients, several were

À of particular interest, including SNPs upstream of the GRIA4 gene. GRIA4 is a member of L-glutamate-gated ion channels, which mediate fast synaptic excitatory neurotransmission. This type of receptor is thought to 20% Odds ratio P 95% Confidence test modulate glutamatergic transmission in the brain by X

þ inhibiting glutamate release at the synapse. Although the specific AMPA4 glutamate receptor has not been extensively studied, an association between GRIA4 haplotypes and schizophrenia has been reported in a 42 test Japanese population. This finding was not seen in a À Chinese population,53 and no studies of other popula- tions have been reported. A potential role for glutamate in schizophrenia was first proposed based

test on the observation that the psychotomimetic agents

þ phencyclidine (PCP) and ketamine induce psychotic symptoms and neurocognitive disturbances similar to those of schizophrenia by blocking neurotransmission at NMDA-type glutamate receptors. Since then,

test evidence of glutamate dysregulation in schizophrenia À has been increasing.54 , positive predictive value.

þ Additionally, we found several SNPs associated with iloperidone efficacy upstream of the GFRA2 gene, located on 8p21.3. Numerous studies in different populations have indicated a genetic asso- G/T 71 38 38 61 3.00 0.0001 1.705–5.277 0.62 0.65 0.65 0.62 A/A 56 49 26 67 2.95 0.0004 1.627–5.33 0.72 0.53 0.68 0.58 Non-T/T 33 77 15 85 2.43 0.011 1.225–4.811 0.85 0.30 0.69 0.52 Non-A/G 54 53 22 77 3.57 < 0.0001 1.944–6.542 0.78 0.50 0.71 0.59 Non-G/G 35 75 12 88 3.42 0.0009 1.658–7.062 0.88 0.32 0.74 0.54 Non-G/G 87 23 58 42ciation 2.74 0.0011 between 1.492–5.028 schizophrenia 0.42 0.79 0.60and 0.65 chromosome 8p, particularly in the 8p21 region.43–45 This schizophre- nia susceptibility locus is referred to as SCZD6. Candidate genes in that region include neuregulin 1 NUDT9P1 GFRA2 GRIA4 TNR XKR4 NPAS3 (NRG1), adrenergic a-1A-receptor (ADRA1A) and frizzled homologue 3 (FZD3). Our findings suggest Odds ratios, sensitivity, specificity and predictive values of SNPs associated with change in PANSS-T in iloperidone-treated patients adding GFRA2 to the list of genes in that region, which may play a role in schizophrenia and anti- , negative predictive value; PV

À psychotic treatment. The GFRA2 protein is a cell- SNPs are identified byEach their gene unique name dbSNP correspondsThe number to genotype (http://www.ncbi.nlm.nih.gov/projects/SNP/). the class shown NCBIlisted here official is at symbol the http://www.ncbi.nlm.nih.gov/SNP/index.html. of genotype theThe class gene number associated of closest with patients better to withfrom efficacy the improvement the with in SNP. LOCF iloperidone PANSS-T data above treatment. as andclass Allele 100*[Change below designation associated from 20% baseline/(baseline—number refers is with of shown to better itemsPV per the comprising response genotype Aff the specified class. score)]. The in Genetic percentage test the of results ‘genotype’ improvement were in column defined PANSS-T was as and positive calcul as ( negative ( rs4528226 rs7837682 rs2513265 rs875326 rs9643483 Table 5 SNP Geners11851892 Genotype Improvement < 20% Improvement surface receptor for GDNF and neurturin. Several

Molecular Psychiatry Genetic markers of iloperidone efficacy C Lavedan et al 817 studies have linked GDNF to schizophrenia. In an of their association with iloperidone efficacy should animal study in which PCP was injected into rats to therefore be further considered as potential players in produce schizophrenia-like symptoms, Semba et al.55 the mechanism of schizophrenia. The six SNPs showed that PCP alters the function of GDNF in the identified here should be further explored in subse- ventral midbrain, where dopamine cell bodies are quent studies; this will provide the opportunity to localized. Our results suggest that new studies should validate not only their individual clinical utility but be undertaken to investigate the interplay of GDNF, also their potential combined value as predictors of neurturin and GFRA2 in the etiology of schizophrenia efficacy. and in the response to antipsychotic treatments. In conclusion, we have identified several genetic Several SNPs identified in our study lay in polymorphisms associated with the efficacy of iloper- intergenic regions, sometimes far away from any idone for the treatment of patients with schizophre- known gene. One such SNP is rs4528226, which nia. The allele frequency of the SNPs uncovered here showed the most statistically significant association and their individual effects on treatment response with change in PANSS-T in all analyses (Table 2; should be considered in the design and evaluation of Figures 1 and 2f). This SNP is located approximately subsequent clinical studies with iloperidone or with 200 kb upstream from the HTR7 gene. Interestingly, it other antipsychotic agents. Although our results has been shown that iloperidone has high affinity for highlight the genetic complexity of drug response, 56 the 5-HT7 receptor. The biological relevance of SNPs they also underscore the opportunity of this new kind located in intergenic DNA is currently difficult to of pharmacogenomic approach. In the near future, assess. Although most genetic research has histori- clinicians should be able to better estimate the cally focused on coding sequences of genes and individual risk-to-benefit ratios of various drugs nearby promoter regions, comparative genomics stu- based on each patient’s genetic makeup and, with dies are uncovering other functional genomic ele- more assurance to determine the best treatment ments that can be located in introns and in gene-poor options. ‘desert’ regions sometimes hundreds or thousands of kilobases away from the genes they regulate.57–59 It is Acknowledgments possible that the intergenic SNPs examined in our study are within regulatory elements, response ele- We thank all of the patients who participated in this ments, transposable elements, noncoding RNAs, cod- study. We also thank Michael Di Marino and Ingeborg ing sequences or splicing sites that have not yet been Holt for their assistance in the statistical analysis and identified. Alternatively, they may be in linkage Kelly Greer for her technical support. We are grateful disequilibrium with these elements or sequences. So to Dr Yoav Benjamini and Chip Clark for their critical far, no long-range cis-regulatory elements of HTR7 reviews of the paper. have been identified. Our results suggest that the genomic region of rs4528226 should be thoroughly Conflict of interests examined for the presence of such elements. All authors are employees of Vanda Pharmaceuticals We also analyzed nine SNPs in eight candidate Inc. and have no additional affiliations or financial genes because of their possible association with interests to disclose. schizophrenia. None of them show a significant association (at P < 0.01) with iloperidone response. However, as the SNP coverage was not complete, References these results do not exclude the possibility that 1 American Psychiatric Association. Diagnostic and Statistical these genes may be associated with the efficacy of Manual of Mental Disorders, 4th edn. American Psychiatric iloperidone. Association: Washington, DC, 2000. The etiology of schizophrenia remains largely 2 Bondy B, Spellmann I. Pharmacogenetics of antipsychotics: useful for the clinician? Curr Opin Psychiatry 2007; 20: 126–130. unknown, and it is still unclear to what extend 3 Reynolds GP, Templeman LA, Godlewska BR. Pharmacogenetics of genetic factors involved in drug response may play a schizophrenia. Expert Opin Pharmacother 2006; 7: 1429–1440. role in the predisposition to, or in the development of, 4 Lieberman JA, Stroup TS, McEvoy JP, Swartz MS, Rosenheck RA, the disease. It is expected that pathways involved in Perkins DO et al. Effectiveness of antipsychotic drugs in patients with chronic schizophrenia. N Engl J Med 2005; 353: 1209–1223. drug response may not necessarily overlap with 5 Kane JM, Malhotra A. The future of pharmacotherapy for disease mechanisms as exemplified by enzymes that schizophrenia. World Psychiatry 2003; 2: 81–86. metabolize multiple drugs across various indications. 6 Kalkman HO, Subramanian N, Hoyer D. Extended radioligand However, the study of polymorphisms in molecules binding profile of iloperidone: a broad spectrum dopamine/ believed to be involved in the etiology or the serotonin/norepinephrine receptor antagonist for the management of psychotic disorders. Neuropsychopharmacology 2001; 25: development of the disease as well as in its treatment, 904–914. suggests possible overlaps between pathways of 7 Kalkman HO, Feuerbach D, Lotscher E, Schoeffter P. Functional disease and drug response. In the case of schizo- characterization of the novel antipsychotic iloperidone at human phrenia, several dopamine and serotonin receptors D2, D3, alpha 2C, 5-HT6, and 5-HT1A receptors. Life Sci 2003; 73: have been implicated in both the mechanism of the 1151–1159. 60 8 Richelson E, Souder T. Binding of antipsychotic drugs to human disease and antipsychotic response. The poly- brain receptors focus on newer generation compounds. Life Sci morphisms and genes identified in our study because 2000; 68: 29–39.

Molecular Psychiatry Genetic markers of iloperidone efficacy C Lavedan et al 818 9 Ichikawa J, Meltzer HY. Relationship between dopaminergic and site. Available at: http://www.affymetrix.com/support/technical/ serotonergic neuronal activity in the frontal cortex and the action whitepapers/brlmm_whitepaper.pdf (accessed 26 July). of typical and atypical antipsychotic drugs. Eur Arch Psychiatry 31 Murabito JM, Rosenberg CL, Finger D, Kreger BE, Levy D, Splansky Clin Neurosci 1999; 249(Suppl 4): 90–98. GL et al. A genome-wide association study of breast and prostate 10 Jain KK. An assessment of iloperidone for the treatment of cancer in the NHLBI’s Framingham Heart Study. BMC Med Genet schizophrenia. Expert Opin Investig Drugs 2000; 9: 2935–2943. 2007; 8(Suppl 1): S6. 11 Cutler AJ, Kalali AH, Weiden P, Hamilton J, Wolfgang CD. Four- 32 Newton-Cheh C, Guo CY, Wang TJ, O’donnell CJ, Levy D, Larson week, double-blind, placebo- and ziprasidone-controlled trial of MG. Genome-wide association study of electrocardiographic and iloperidone in patients with acute exacerbations of schizophrenia. heart rate variability traits: the Framingham Heart Study. BMC J Clin Psychopharmacol 2008; 28(Suppl 1): S20–S28. Med Genet 2007; 8(Suppl 1): S7. 12 Weiden PJ, Cutler A, Polymeropoulos MH, Wolfgang CD. Safety 33 Gottlieb DJ, O’Connor GT, Wilk JB. Genome-wide association profile of iloperidone: a pooled analysis of 6-week acute-phase of sleep and circadian phenotypes. BMC Med Genet 2007; pivotal trials. J Clin Psychopharmacol 2008; 28(Suppl 1): S12–S19. 8(Suppl 1): S9. 13 Baroldi P, Wolfgang C. A Pharmacokinetic (PK)-Pharmacodynamic 34 Arking DE, Pfeufer A, Post W, Kao WHL, Newton-Cheh C, Ikeda M (PD) Relationship Exists for Efficacy of Iloperidone, a Novel et al. A common genetic variant in the NOS1 regulator NOS1AP Investigational Atypical Antipsychotic Agent. American Psychia- modulates cardiac repolarization. Nat Genet 2006; 38: 644–651. tric Association 2007 Meeting POSTER NR507—available at: 35 Kamnasaran D, Muir WJ, Ferguson-Smith MA, Cox DW. Disrup- http://www.psych.org/edu/other_res/lib_archives/archives/meet- tion of the neuronal PAS3 gene in a family affected with ings/AMN/2007nra.pdf. schizophrenia. J Med Genet 2003; 40: 325–332. 14 Verbeke G, Molenberghs G. Linear Mixed Models for Longitudinal 36 Calenda G, Peng J, Redman CM, Sha Q, Wu X, Lee S. Identification Data. Springer: New York, 2000. of two new members, XPLAC and XTES, of the XK family. Gene 15 Little RJA, Rubin DB. Statistical Analysis with Missing Data, 2nd 2006; 370: 6–16. edn. Wiley: New York, 2003. 37 Jung HH, Haker H. Schizophrenia as a manifestation of X-linked 16 Molenberghs G, Thijs H, Jansen I, Beunckens C, Kenward MG, McLeod-Neuroacanthocytosis syndrome. J Clin Psychiatry 2004; Mallinck-rodt C et al. Analyzing incomplete longitudinal clinical 65: 722–723. trial data. Biostatistics 2004; 5: 445–464. 38 Wu C-Y, Chiu C-C. Usher syndrome with psychotic symptoms: two 17 Cook RJ, Zeng L, Yi Y. Marginal analysis of incomplete long- cases in the same family. Psychiatry Clin Neurosci 2006; 60: 626–628. itudinal binary data: a cautionary note on LOCF imputation. 39 Carnemolla B, Leprini A, Borsi L, Querze G, Urbini S, Zardi L. Biometrics 2004; 60: 820–828. Human tenascin-R: complete primary structure, pre-mRNA alter- 18 Mallinckrodt CH, Clark WS, David SR. Accounting for dropout native splicing and gene localization on chromosome 1q23-q24. bias using mixed-effects models. J Biopharm Stat 2001; 11: 9–21. J Biol Chem 1996; 271: 8157–8160. 19 Pritchard JK, Stephens M, Donnelly P. Inference of population 40 Leprini A, Gherzi R, Siri A, Querze G, Viti F, Zardi L. The human structure using multilocus genotype data. Genetics 2000; 2: tenascin-R gene. J Biol Chem 1996; 271: 31251–31254. 945–959. 41 Wei J, Hemmings GP. TNXB locus may be a candidate gene 20 Benjamini Y, Hochberg Y. Controlling the false discovery rate: a predisposing to schizophrenia. Am J Med Genet B Neuropsychiatr practical and powerful approach to multiple testing. J R Stat Soc Genet 2004; 125B: 43–49. 1995; 57: 289–300. 42 Makino C, Fujii Y, Kikuta R, Hirata N, Tani A, Shibata A et al. 21 Kukreti R, Tripathi S, Bhatnagar P, Gupta S, Chauhan C, Positive association of the AMPA receptor subunit GluR4 gene Kubendran S et al. Association of DRD2 gene variant with (GRIA4) haplotype with schizophrenia: linkage disequilibrium schizophrenia. Neurosci Lett 2006; 392: 68–71. mapping using SNPs evenly distributed across the gene region. 22 Bishop JR, Ellingrod VL, Moline J, Miller D. Association between Am J Med Genet B Neuropsychiatr Genet 2003; 116B: 17–22. the polymorphic GRM3 gene and negative symptom improvement 43 Blaveri E, Kalsi G, Lawrence J, Quested D, Moorey H, Lamb G et al. during olanzapine treatment. Schizophr Res 2005; 77: 253–260. Genetic association studies of schizophrenia using the 8p21-22 23 Fallin MD, Lasseter VK, Avramopoulos D, Nicodemus KK, genes: prepronociceptin (PNOC), neuronal nicotinic choliner- Wolyniec PS, McGrath JA et al. Bipolar I disorder and schizo- gic receptor alpha polypeptide 2 (CHRNA2) and arylamine phrenia: a 440-single-nucleotide polymorphism screen of 64 N-acetyltransferase 1 (NAT1). Eur J Hum Genet 2001; 9: 469–472. candidate genes among Ashkenazi Jewish case-parent trios. Am J 44 Blouin JL, Dombroski BA, Nath SK, Lasseter VK, Wolyniec PS, Hum Genet 2005; 77: 918–936. Nestadt G et al. Schizophrenia susceptibility loci on 24 Lee HJ, Song JY, Kim JW, Jin S-Y, Hong MS, Park JK et al. 13q32 and 8p21. Nat Genet 1998; 20: 70–73. Association study of polymorphisms in synaptic vesicle-asso- 45 Stefansson H, Sigurdsson E, Steinthorsdottir V, Bjornsdottir S, ciated genes, SYN2 and CPLX2, with schizophrenia. Behav Brain Sigmundsson T, Ghosh S et al. Neuregulin 1 and susceptibility to Funct 2005; 1: 15. schizophrenia. Am J Hum Genet 2002; 71: 877–892. 25 Hamajima N, Matsuo K, Tajima K, Mizutani M, Iwata H, Iwase T et 46 Michelato A, Bonvicini C, Ventriglia M, Scassellati C, Randazzo R, 0 al. Limited association between a catechol-O-methyltransferase Bignotti S et al. 3 UTR (AGG)n repeat of glial cell line-derived (COMT) polymorphism and breast cancer risk in Japan. Int J Clin neurotrophic factor (GDNF) gene polymorphism in schizophrenia. Oncol 2001; 6: 13–18. Neurosci Lett 2004; 357: 235–237. 26 Ronai Z, Barta C, Guttman A, Lakatos K, Gervai J, Staub M et al. 47 Ikeda M, Iwata N, Kitajima T, Suzuki T, Yamanouchi Y, Kinoshita

Genotyping the –521C/T functional polymorphism in the promo- Y et al. Positive association of the serotonin 5-HT7 receptor gene ter region of dopamine D4 receptor (DRD4) gene. Electrophoresis with schizophrenia in a Japanese population. Neuropsychophar- 2001; 22: 1102–1105. macology 2006; 31: 866–871. 27 Huang YY, Battistuzzi C, Oquendo MA, Harkavy-Friedman J, 48 Moises HW, Yang L, Kristbjarnarson H, Wiese C, Byerley W,

Greenhill L, Zalsman G et al. Human 5-HT1A receptor C(–1019)G Macciardi F et al. An international two-stage genome-wide polymorphism and psychopathology. Int J Neuropsychopharma- search for schizophrenia susceptibility genes. Nat Genet 1995; col 2004; 7: 441–451. 11: 321–324. 28 Joober R, Benkelfat C, Brisebois K, Toulouse A, Turecki G, Lal S 49 Pieper AA, Wu X, Han TW, Estill SJ, Dang Q, Wu LC et al. The

et al. T102C polymorphism in the 5-HT2A gene and schizophrenia: neuronal PAS domain protein 3 transcription factor controls FGF- relation to phenotype and drug response variability. J Psychiatry mediated adult hippocampal neurogenesis in mice. Proc Natl Neurosci 1999; 24: 141–146. Acad Sci USA 2005; 102: 14052–14057. 29 Di X, Matsuzaki H, Webster TA, Hubbell E, Liu G, Dong S et al. 50 Erbel-Sieler C, Dudley C, Zhou Y, Wu X, Estill SJ, Han T et al. Dynamic model based algorithms for screening and genotyping Behavioral and regulatory abnormalities in mice deficient in the over 100 K SNPs on oligonucleotide microarrays. Bioinformatics NPAS1 and NPAS3 transcription factors. Proc Natl Acad Sci USA 2005; 21: 1958–1963. 2004; 101: 13648–13653. 30 BRLMM: an improved genotype calling method for the 51 Goldberger C, Gourion D, Leroy S, Schurhoff F, Bourdel M-C, GeneChips Human Mapping 500 K Array Set. Affymetrix Web Leboyer M et al. Population-based and family-based association

Molecular Psychiatry Genetic markers of iloperidone efficacy C Lavedan et al 819 study of 50UTR polymorphism of the reelin gene and schizo- 56 Kongsamut S, Roehr JE, Cai J, Hartman HB, Weissensee P, phrenia. Am J Med Genet B Neuropsychiatr Genet 2005; 137B: Kerman LL et al. Iloperidone binding to human and rat 51–55. dopamine and 5-HT receptors. Eur J Pharmacol 1996; 317: 52 Impagnatiello F, Guidotti AR, Pesold C, Dwivedi Y, Caruncho H, 417–423. Pisu MG et al. A decrease of reelin expression as a putative 57 Forton JT, Udalova IA, Campino S, Rockett KA, Hull J, vulnerability factor in schizophrenia. Proc Natl Acad Sci USA Kwiatkowski DP. Localization of a long-range cis-regulatory element 1998; 95: 15718–15723. of IL13 by allelic transcript ratio mapping. Genome Res 2007; 17: 53 Guo S, Shi Y, Zhao X, Duan S, Zhou J, Meng J et al. No genetic 82–87. association between polymorphisms in the AMPA receptor 58 Woolfe A, Goodson M, Goode DK, Snell P, McEwen GK, Vavouri T subunit GluR4 gene (GRIA4) and schizophrenia in the Chinese et al. Highly conserved non-coding sequences are associated with population. Neurosci Lett 2004; 369: 168–172. vertebrate development. PLoS Biol 2005; 3: e7. 54 Stone JM, Morrison PD, Pilowsky LS. Glutamate and dopamine 59 Washietl S, Hofacker IL, Lukasser M, Huttenhofer A, Stadler PF. dysregulation in schizophrenia—a synthesis and selective review. Mapping of conserved RNA secondary structures predicts thou- J Psychopharmacol 2007; 21: 440–452. sands of functional noncoding RNAs in the . Nat 55 Semba J, Akanuma N, Wakuta M, Tanaka N, Suhara T. Alterations Biotechnol 2005; 23: 1383–1390. in the expressions of mRNA for GDNF and its receptors in the 60 Arranz MJ, de Leon J. Pharmacogenetics and pharmacogenomics ventral midbrain of rats exposed to subchronic phencyclidine. of schizophrenia: a review of last decade of research. Mol Brain Res Mol Brain Res 2004; 124: 88–95. Psychiatry 2007; 8: 707–747.

Molecular Psychiatry