Introduction

Panic disorder (PD) is characterized by recurrent and unexpected panic attacks, subsequent anticipatory anxiety and phobic avoidance, with the frequent development of agoraphobia.1 The lifetime prevalence of PD is 4.7% with female preponderance.2 PD frequently takes a chronic course, with many remissions and relapses.3 Genetic epidemiological researches including family and twin studies have shown that genetic as well as environmental factors have an important role in the pathogenesis of PD. First-degree relatives of proband with PD have a six-fold increased risk of developing the condition.4 Twin studies show that about 40% of liability toward PD relates to heritable factors.5 Although several linkage studies and many association studies on more than 350 candidate genes for PD have been concluded, the results were inconsistent, negative or not clearly replicated.6

Copy number variations (CNVs) are quantitative structural variants; they are deletions and duplications of DNA segments ranging from a kilobase to several megabases. CNVs cover 12% of the human genome,7 which is much more than single-nucleotide polymorphisms (SNPs). As recent studies have found evidence for the potential involvement of CNVs in neuropsychiatric conditions,8, 9, 10 we explored genome-wide CNV association with PD in a Japanese population.

Materials and methods

The following methods are shown using a simplified block diagram in Figure 1.

Figure 1
figure 1

Block diagram of materials and methods used for genome-wide rare and common CNV association analysis of panic disorder.

Sample recruitment

Participants were 2212 unrelated, ethnically Japanese sample of 595 PD cases (189 males and 406 females; age=38.9±10.9 years, mean±s.d.) and 1617 controls (706 males and 911 females; age=39.0±11.0 years, mean±s.d.) recruited in the vicinity of Tokyo, Nagoya or Niigata, Japan. These areas are known to belong to the main cluster of the Japanese population structure.11 The diagnosis was confirmed according to the Diagnostic and Statistical Manual, 4th edition criteria,1 using the Mini-International Neuropsychiatric Interview12 and clinical records. Details of the recruitment of cases13 and controls13, 14, 15, 16 are described elsewhere.

The aims of the present study were clearly explained to all subjects. They gave written, informed consent. The study was approved by Ethical Committees of the Faculties of Medicine at the University of Tokyo, Mie University and Niigata University, and conformed to the provisions of the Declaration of Helsinki.

CNV detection and determination

Genomic DNA was isolated from leukocytes in whole blood using the Wizard genomic DNA purification kit (Promega Corporation, Madison, WI, USA). SNPs and CNVs were detected using Genome-Wide Human SNP array 6.0 (Affymetrix, Santa Clara, CA, USA). The results of the SNP genome-wide association study are described elsewhere.17 CNVs were determined with Birdsuite software18 version 1.5.5 over the thresholds of logarithm of odds 10,18, 19 the number of probes in CNV 1018, 20, 21 and individual CNVs <300. A total of 906 600 SNP probes and 946 000 CNV non-polymorphic probes on the chip were genotyped at the University of Tokyo, Graduate School of Medicine, according to the manufacturer's protocol. PennCNV software22 version 1 May 2010 was also used as an alternative algorithm to Birdsuite, to establish stringent CNV data; common CNVs were evaluated if they showed concordance rates of >0.90 between the algorithms.18, 23

Quality control

A total of 157 subjects were excluded from further analysis, because they did not pass the quality control criteria for contrast quality control 0.4,24 sample-wise call rate 0.95,25 ambiguous gender, unexpected duplicates, cryptic relatives or population stratifications. Ambiguous gender was assessed by sex check function in PLINK software26 version 1.07. Unexpected duplicates and first- or second-degree cryptic relatives were assessed using the identity-by-state/identity-by-descent estimation function of PLINK to estimate the cryptic relatedness of each pair of subjects. To detect population outliers, the subjects were assessed through comparison with 45 JPT, 45 CHB, 45 CEU and 45 YRI HapMap27 subjects without relationships using the EIGENSTRAT software.28 Supplementary Figure 1 indicates the principal component analysis plots before and after filtering for quality control.

Filtering for quality control identified 2055 samples: 535 cases aged 38.2±11.7 years (174 males and 361 females) and 1520 controls aged 39.8±10.8 years (671 males and 849 females; Table 1).

Table 1 Subjects after quality controls

Statistical analysis

CNVs with segment sizes of <100 Kb were excluded by PLINK, because the platforms had limited power to detect smaller CNVs.19, 23, 29 CNVs were classified as rare (found in <1% of the total sample) or common (found in 5%). They were considered to be colocalized if they overlapped by at least 50% of their length, as implemented in PLINK. Common CNVs were also excluded when the lengths of the consensus regions shared by all overlapping CNV segments were equivalent to one base pair, which meant these regions included only one probe.

PLINK was used to perform global burden analysis for rare CNVs on the basis of 100 000 permutations and association analysis for common CNVs between cases and controls at each consensus region. Sex chromosomes were removed from the analysis, as they are incorrectly segmented by Birdsuite19 and PennCNV.21 Bonferroni-corrected P-values <0.5 were considered significant.

Results

The sample yielded 2039 rare CNVs (503 deletions and 1536 duplications) and 79 common CNVs (35 deletions and 44 duplications; Figure 1).

As for rare CNVs, neither the number of segments per person (RATE) nor the proportion of samples with one or more segments (PROP) showed significant differences between PD cases and controls (Table 2). The same analysis was repeated for CNVs showing values of <5% in the sample, and resulted in 550 deletions and 1646 duplications. The trends between cases and controls were virtually identical to those in Table 2 (data not presented).

Table 2 Global rare CNV burden analysis

Among common CNVs, 19 deletions and 7 duplications had concordance rates of >0.90 between Birdsuite and PennCNV. Five common duplications on the locus of 31 901 431–33 149 454 bp in chromosome 16 (16p11.2) showed corrected P-values of <0.05; the statistical power to detect these CNVs was estimated to be more than 0.8 for α=0.001 when the frequency of CNVs was 0.21, the penetrance of CNVs was 0.01 and the CNV relative risk was 1.4 for 535 cases and 1520 controls.30, 31, 32 Other common CNVs with concordance rates of >0.90 did not show P-values of <0.05 for either deletions or duplications (Supplementary Table 1).

Discussion

The present study explored genome-wide rare and common CNV associations between 535 PD cases and 1520 controls in a Japanese population. Rare CNVs exhibited no significant differences in global burden analysis. Individuals with PD did not have an increased genome-wide rare CNV burden. The number of rare CNVs was estimated to be smaller than that of the previous study19 on the basis of global burden analysis, because CNVs were considered to be colocalized and identical in the present study if they overlapped by at least 50% of their length.

The common CNVs found in this study had all been previously reported. Common duplications in 16p11.2 showed Bonferroni-corrected significance values of P<0.0019 (=0.05/(19 deletions and 7 duplications)) in association analysis. These included several genes of IGH, HERC2P4, TP53TG3, SLC6A8 and SLC6A10P in their union region for the total distance of 31 901 431–33 149 454 bp spanned by all duplication segments. No genes in their consensus regions were shared by all overlapping duplication segments (Table 3 and Supplementary Figure 2).

Table 3 Results of genome-wide association study for common CNVs in 16p11.2

Pseudogenes and non-coding RNAs (including microRNAs and small nucleolar RNAs (snoRNAs)) were sought in the union region using existing databases (Pseudogene.org,33 miRBase,34 snoRNABase35 and NONCODE36), as Salmena et al.37 recently emphasized their important roles in pathological conditions. Although no miRNAs or snoRNAs were found, 50 pseudogenes (listed in Supplementary Table 2) and 241 PIWI-interacting RNAs (piRNAs; shown in Supplementary Figure 2) were detected in the union region. No pseudogenes and three piRNAs (n138582, n119652 and n127652) were found in the consensus regions (Table 3). piRNAs are particularly abundant in the male germline and adult testes,38 and are involved in developmental regulation and genomic imprinting through DNA methylation.39, 40 PD development may be influenced by common duplications in 16p11.2 through piRNA regulation.

Table 4 shows information on subjects with significant common duplications in 16p11.2. No significant differences were observed in the mean age at the onset of PD, the frequency of cases, controls, comorbid agoraphobia or comorbid major depressive disorder among subjects with and without duplications, except for lower frequencies of comorbid bipolar disorders in all and male cases showing duplication with the consensus regions of 32 716 092–32 723 322 bp (χ2=5.91, df=1, P=0.02 in all; χ2=4.37, df=1, P=0.04 in males) and in all cases showing duplication with the consensus regions of 32 747 828–32 747 898 bp (χ2=5.61, df=1, P=0.02).

Table 4 Subjects harboring significant common duplications in 16p11.2 (16: 31 901 431–33 149 454 bp)

The pericentromeric region of 16p11.2 was reported to be rich in low copy repeats,41 causing non-allelic homologous recombination and the harboring of novel genomic disorders.42 It has been suggested that rare CNVs in the same region of 16p11.2 (but 2 Mb upstream of the region discussed here), are associated with developmental and neuropsychiatric disorders or dysmorphic manifestations such as mental retardation, autism spectrum disorders, attention-deficit/hyperactivity disorder, schizophrenia, abnormal head size or syringomyelia.20, 43, 44, 45 This is the first study to demonstrate significant associations of CNVs in 16p11.2 with PD.

Recently, the Wellcome Trust Case–Control Consortium identified biological artifacts of common CNVs typed on existing platforms that lead to false-positive associations.21 It might be appropriate to suggest that care should be taken when determining common CNVs using currently available SNP-based microarray and algorithm software, as differences in performance between algorithms have been reported.46, 47 In our study, we took a rigorous and conservative analytical approach to minimize false-positive findings. We set strict thresholds for logarithm of odds (10), the number of probes in CNV (10) and CNV segment sizes (100 Kb). In addition, we used two different algorithms to identify the CNV difference between cases and controls.

There are some limitations to our study. First, the case sample size (595 PD patients) might not be adequate in comparison with the control sample size (1617 subjects). A larger case sample is required to detect CNVs with lower P-values. Second, CNVs in sex chromosomes were not analyzed, because Birdsuite and PennCNV cannot determine them sufficiently. An improved algorithm is required for the determination of whole-genome CNVs, including those in sex chromosomes. Finally, our results may include false negatives due to the strict analytical parameters used. In particular, we focused on relatively large CNVs (>100 Kb) in this study. Given that the median size of CNVs in humans has been estimated as up to be 20 Kb,22, 48 a significant proportion of CNVs might not have been analyzed. In spite of these limitations, this is the first study to our knowledge that explores genome-wide CNV association with PD.

In conclusion, (i) individuals with PD did not exhibit an increased genome-wide rare CNV burden and (ii) common duplications in 16p11.2 were associated with PD in a Japanese population.