Original Article

Molecular Psychiatry (2011) 16, 949–959; doi:10.1038/mp.2010.107; published online 16 November 2010

A genome-wide association study on common SNPs and rare CNVs in anorexia nervosa

K Wang1, H Zhang1, C S Bloss2, V Duvvuri3, W Kaye3, N J Schork2,4, W Berrettini5 and H Hakonarson1,6 the Price Foundation Collaborative Group7

  1. 1Center for Applied Genomics, Children’s Hospital of Philadelphia, Philadelphia, PA, USA
  2. 2The Scripps Translational Science Institute, Scripps Health, La Jolla, CA, USA
  3. 3Department of Psychiatry and Eating Disorders Treatment and Research Program, University of California San Diego, La Jolla, CA, USA
  4. 4Department of Molecular and Experimental Medicine, Scripps Research Institute, La Jolla, CA, USA
  5. 5Department of Psychiatry, University of Pennsylvania, Philadelphia, PA, USA
  6. 6Department of Pediatrics, University of Pennsylvania, Philadelphia, PA, USA

Correspondence: Dr H Hakonarson, Department of Pediatrics, University of Pennsylvania, Philadelphia, PA 19104, USA. E-mail: hakonarson@email.chop.edu or wadeb@mail.med.upenn.edu

7See Appendix.

Received 5 April 2010; Revised 3 August 2010; Accepted 22 September 2010; Published online 16 November 2010.



Anorexia nervosa (AN) is a mental illness with high mortality that most commonly afflicts adolescent female individuals. Clinical symptoms include chronic food refusal, weight loss and body image distortions. We carried out a genome-wide association study on 1033 AN cases and 3733 pediatric control subjects, all of whom were of European ancestry and were genotyped on the Illumina HumanHap610 platform (Illumina, San Diego, CA, USA). We confirmed that common single-nucleotide polymorphisms (SNPs) within OPRD1 (rs533123, P=0.0015) confer risk for AN, and obtained suggestive evidence that common SNPs near HTR1D (rs7532266, P=0.04) confer risk for restricting-type AN specifically. However, no SNPs reached genome-wide significance in our data, whereas top association signals were detected near ZNF804B, CSRP2BP, NTNG1, AKAP6 and CDH9. In parallel, we performed genome-wide analysis on copy number variations (CNVs) using the signal intensity data from the SNP arrays. We did not find evidence that AN cases have more CNVs than control subjects, nor do they have over-representation of rare or large CNVs. However, we identified several regions with rare CNVs that were only observed in AN cases, including a recurrent 13q12 deletion (1.5Mb) disrupting SCAS in two cases, and CNVs disrupting the CNTN6/CNTN4 region in several AN cases. In conclusion, our study suggests that both common SNPs and rare CNVs may confer genetic risk to AN. These results point to intriguing genes that await further validation in independent cohorts for confirmatory roles in AN.


anorexia; eating disorder; copy number variation; genome-wide association study; OPRD1; HTR1D



Anorexia nervosa (AN) is a syndrome characterized by chronic food refusal, weight loss, irrational fear of weight gain even when emaciated, and distortions about body weight and shape.1 Women are affected more often than men in an ~10:1 ratio, with ~0.9% of US women affected and a near uniform adolescent age at onset.2 There are several subtypes of AN.1 Those with the restricting subtype have severely diminished food intake, and those with the binge-eating/purging subtype additionally engage in binge-eating or purging behavior. Compared with other behavioral disorders, AN has a stereotypic presentation with respect to gender-specific risk, age of onset, symptoms, signs and disease course.2 AN is associated with significant morbidity and has the highest mortality rate among all psychiatric disorders,3 with recent estimates of a standardized mortality ratio as high as 6.2.4 Treatment for AN is challenging as most patients lack insight into their illness and are reluctant to undergo weight restoration.

Multiple pieces of evidence suggest a strong genetic component in susceptibility to AN. A high degree of familial aggregation is observed for AN, with a first-degree relative recurrence risk >10.5, 6 An increased rate of AN is also seen among first-degree relatives of bulimia nervosa probands, and vice versa.5 Twin studies have estimated AN heritability as ranging from 58 to 74% in two studies conducted in the United States,7 56% in a study conducted in Sweden7 and 75% in a study conducted in Denmark.8 The slight variability probably relates to differences in culture or ascertainment/diagnostic criteria.

Several studies have been conducted to investigate the genetic basis of AN.9, 10 A microsatellite linkage study mapped a locus on chromosome 1, which may harbor a risk allele for the most narrow diagnostic category of AN, the restricting-type AN (RAN).11 single-nucleotide polymorphisms (SNPs) genotyping in this region12 boosted the linkage signal and nominated susceptibility alleles of small effect, in two plausible candidate genes, namely OPRD1 (delta opioid receptor) and HTR1D (serotonin 1D receptor). The association signals on these candidate genes were later confirmed by an independent study.13 Linkage analyses using covariates, such as obsession-scale scores, identified multiple other genomic regions, which might harbor AN susceptibility alleles.14, 15 Several association studies on AN using candidate genes in selected pathways have also been conducted. In addition to studies on HTR1D and OPRD1, a study tested an SNP in the serotonin 2A receptor gene, but failed to find association.16 A few other studies have tested genes in the dopamine pathway, and detected evidence for association for DRD2 (dopamine D2),17 but not for COMT (catechol-O-methyltransferase).18 Positive evidence of association was also reported for BDNF (brain-derived neurotrophic factor),19, 20 but many other candidate gene association studies have generally failed to find significant association.21 More recently, a Japanese group conducted genome-wide investigation of AN by typing pooled DNA on microsatellite markers with follow-up SNP genotyping, and detected significant association on rs2048332 near SPATA17 on 1q41.22 In summary, despite many linkage studies, candidate gene association studies and small-scale genome-wide association studies, the genetic architecture underlying AN susceptibility remains largely unknown.

To identify AN susceptibility loci, we performed a genome-wide association study (GWAS) using high-density SNP arrays on a large AN cohort. Given the known contribution of copy number variations (CNVs) to neuropsychiatric disorders,23 we also performed a parallel study on CNVs, by taking advantage of the signal intensity data from SNP genotyping arrays. Here, we report the results from the analysis on both common SNPs and rare CNVs. Our study confirmed the potential role of known susceptibility genes, but also pointed to a few additional susceptibility genes and variants worthy of follow-up studies.


Materials and methods

Sample collection

The AN cases examined in this study were collected from several sources. A collaborative group to study AN genetics was formed ~15 years ago, including multiple AN clinical phenotypic experts.24 Methods, ascertainment sites and diagnostic instruments for this effort have been previously described.24 All cases met a lifetime diagnosis of DSM-IV AN, with or without amenorrhea, at least 3 years before study entry and by age 45. The amenorrhea criterion was waived because of its unreliability in retrospective assessment in females; data suggest that anorectic women with and without amenorrhea do not differ significantly.25 The threshold for low weight was defined as a body mass index at or below 18kgm–2, which corresponds to the fifth percentile body mass index values of the National Health and Nutrition Examination Survey epidemiological sample of females26 for the average age range (27–29 years) of the probands. Eating disorder symptoms and course of illness were assessed with the structured interview for anorexic and bulimic disorders.27 Psychiatric co-morbid disorders were assessed with semi-structured interviews and diagnoses were by DSM-IV criteria, using consensus best-estimate diagnostic procedures. Attitudes toward eating were quantified with the eating disorders inventory-2,28 whereas measures of the severity of obsessions and compulsions were quantified with the Yale–Brown Obsessive-Compulsive Scale.29 Personality constructs were measured by the temperament and character inventory,30 whereas perfectionism was assessed by the Multi-dimensional Perfectionism Scale.31 Potential probands were excluded if they had a history of severe CNS trauma, psychotic disorder or developmental disability, or if they had a medical or neurological condition that could confound the diagnosis of AN (such as type 1 diabetes). In total, we genotyped 1105 unrelated AN cases in this study, with the majority being female subjects. With stringent quality control measures (see below), a total of 1033 AN cases (1009 female subjects and 24 male subjects) were left in the GWAS analysis, including 39 without subtype classification. The distribution of the age of cases was 26.9±8.6 years (mean ± s.d.), whereas the age at first symptom was 15.1±3.1 years. The distribution of the lowest body mass index measure was 14.1±2.0kgm–2, and the Yale–Brown Obsessive-Compulsive Scale was 15.1±12.4.The disease subtypes were described in Supplementary Table 1.

The pediatric control group was recruited by the Children's Hospital of Philadelphia (CHOP) clinicians, nursing and medical assistant staff within the CHOP Health Care Network, which includes multiple primary care clinics and outpatient practices. The control subjects have linked electronic medical records in the clinical databases, and a coded version of these records can be accessed by researchers for genetic research purposes. At the time of the data analysis, the average age of the control subjects was 12.75 years with a s.d. of 4.2 years. One concern on using pediatric controls is the potential loss of power that is associated with the inability to exclude latent diagnoses of the phenotype of interest through intensive screening of controls; however, such loss of power is minimal when disease prevalence is below 1%,32 so our study is unlikely to be significantly affected by this design. We utilized multi-dimensional scaling (MDS) on both AN cases and control subjects, together with 11 HapMap3 populations, to ensure good matching of genetic backgrounds (Supplementary Figure 1). Only the subsets of control subjects with genetically inferred European ancestry were used in the subsequent association analysis (Supplementary Figure 2). All these control subjects were genotyped using the same Illumina Human610 SNP arrays (Illumina, San Diego, CA, USA) at the same genotyping facility by the same technical staff as the case subjects. To further address the concern of population stratification, we also performed association analysis by logistic regression using principal components as covariates, based on the MDS analysis of cases and control subjects of European ancestry. The results were still similar without loci reaching genome-wide significance, and the genomic control inflation factor remained at 1.08. These control subjects genotyped on Human610 arrays were not used in previous publications. The Research Ethics Board of CHOP and other participating centers approved the study, and written informed consent was obtained from all subjects or their parents.

SNP genotyping and quality control

All DNA samples were genotyped on the Illumina Human610-Quad version 1 SNP arrays with ~610000 markers (including ~20000 non-polymorphic markers), at the Center for Applied Genomics, CHOP. Standard data normalization procedures and canonical genotype clustering files provided by Illumina were used to process the genotyping signals.

Following genotyping, we removed samples with >5% missing genotypes, after excluding ~20000 non-polymorphic markers as they do not have genotypes. We used the PLINK software version 1.0633 for MDS on markers not in linkage disequilibrium (LD) to identify the ancestry origin (via the --mds-plot argument), and removed samples who were not of genetically inferred European ancestry. Furthermore, based on the whole-genome identity-by-descent estimate, we eliminated samples from pairs of samples showing cryptic relatedness (identity-by-descent score>0.2). A total of 1033 AN cases and 3733 control subjects were used in subsequent association tests on SNPs.

Association tests

For the genome-wide association analysis for SNPs, we utilized the PLINK software33 version 1.06, through Cochran–Armitage trend test. For chromosome X markers, the analysis was performed on female subjects only. Additionally, we explored association analysis on the subset of female subjects only. Furthermore, given the previous reports on the linkage signals specifically for the cases with RAN,11 we also examined association signals by performing the analysis on the subsets of RAN cases and control subjects. Additionally, we also explored restricting the analysis on AN cases with known family history, or AN cases with an age of onset before or equal to 16 years. We set a genome-wide significance threshold as P<1 × 10−8, considering the multiple types of association tests that have been examined.

CNV calling and analysis

The Log R ratio (LRR) and B allele frequency (BAF) measures for all markers for all samples were directly calculated and exported from the Illumina BeadStudio software. The CNV calls were generated using the PennCNV software (version 2009 Aug27),34 which utilizes an integrated hidden Markov model that incorporates multiple sources of information, including total signal intensity and allelic intensity ratio at each SNP marker, the distance between neighboring SNPs, the allele frequency of SNPs and the pedigree information where available. The default program parameters, library files and the genomic wave adjustment routine35 in detect_cnv.pl program were used in generating CNV calls. The scan_region.pl program in PennCNV was used to map called CNVs to specific genes and exons, using the RefSeq gene definitions. The genomic coordinates for the CNV calls were generated using the NCBI build 36 human genome assembly.

Several procedures were used for quality control on the CNV calls. We first excluded all samples with LRR_standard deviation (SD) measure >0.3, or with BAF_drift measure >0.01, or with number of CNV calls >75. Next, we excluded sparse CNV calls, that is, those CNV calls with average inter-marker distance >50kb (the average distance is ~5kb across the whole genome for the arrays that we used). Furthermore, we excluded all CNV calls whose >50% genomic span overlap with known immunoglobulin regions (chr22:20715572–21595082, chr14:105065301–106352275, chr2:88937989–89411302, chr14:21159897–22090937), as these CNVs may be a result of somatic changes. In addition, we excluded CNV calls whose >50% genomic span overlap with centromere (a list of genomic coordinates for centromere in human genome NCBI 36 build were given at the PennCNV website FAQ section). The final set of CNV calls were then used in the comparative analysis between AN cases and control subjects. Large CNV calls that were potentially interesting were each visually confirmed using a custom script in PennCNV that generates JPG image files for CNVs.



Genome-wide association of SNP data

We performed a GWAS in 1033 AN cases and 3773 pediatric control subjects, all of whom were genotyped by the Illumina Human610 SNP array with over 598000 SNP markers. We utilized MDS analysis on a subset of SNP markers not in LD, to ensure that only samples with genetically inferred European ancestry were used in association tests (Supplementary Figure 1) and to ensure that cases and control subjects were well matched in their genetic ancestry (Supplementary Figure 2). The Manhattan plot for the whole-genome association tests was given in Figure 1.

Figure 1.
Figure 1 - Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact help@nature.com or the author

The Manhattan plot of logarithm of P-values vs genomic coordinates for whole-genome single-nucleotide polymorphism (SNP) markers. GWAS, genome-wide association study.

Full figure and legend (165K)Download PowerPoint slide (294 KB)

None of the SNPs reached a stringent level of genome-wide significance, and the most significant marker is rs6959888 within ZNF804B on 7q21, with a P-value of 1.6 × 10−6 (Table 1). Its paralog ZNF804A was previously associated with schizophrenia,36 but the function for neither gene has been characterized. The list of 10 additional most associated loci with P<1 × 10−6 were annotated in Table 1. Of note, several loci, including APAK6, SSBP2, FAM155A, LPP2 and ELAVL2, harbor multiple SNPs with nominal significance levels. Performing logistic regression-based association tests incorporating principal components as covariates had minor effects on the results (Supplementary Table 2 and 3). For each of the most significant SNPs, we also examined the association statistics in (1) 1009 female cases and 1731 female control subjects (Supplementary Table 4); (2) 394 cases with RAN and 3773 control subjects (Supplementary Table 5); (3) 368 cases with a family history of eating disorders and 3773 control subjects (Supplementary Table 6) and (4) 489 cases with age of diagnosis less than or equal to16 years and 3773 control subjects (Supplementary Table 7). Considering the several types of association tests that we have performed, appropriate genome-wide significance threshold should be regarded as P<1 × 10−8. When examining female subjects only, most of the SNPs showed decreased significance, except an SNP located within the intergenic region between CDH10 and CDH9 (P=2.5 × 10−8). The region was previously associated with autism spectrum disorders,37 but the autism-associated SNP rs4307059 does not show evidence of association with AN (P=0.45). When comparing subjects with RAN and control subjects, none of the loci showed stronger evidence of association. When comparing cases with family history and control subjects, none of the loci showed stronger evidence of association. Similarly, when comparing subjects with early onset diagnosis and control subjects, we do not observe stronger evidence of association either. In summary, the list of loci in Table 1 did not reveal obvious candidate genes, but they represented a prioritized list of genes worthy of follow-up studies in additional independent cohorts.

Examination of previously reported candidate variants

We next examined whether we can confirm specific variants that were associated with AN in previously published studies. Two independent studies have shown that two variants located in OPRD1 and HTR1D on the 1p33–36 linkage region were associated with AN.12, 13 In the recent Brown et al.13 study, rs569356 within OPRD1 showed association with AN and RAN, whereas rs856510 and rs674386 (r2=1 with each other) within HTR1D showed weak evidence of association with RAN only. In our study, we examined the SNP rs533123 within OPRD1 (r2=0.95 with rs569356 and 4kb away), and confirmed its association and direction of effects with AN (P=1.5 × 10−3, odds ratio=1.2, minor allele frequency (MAF)=21.7% in cases, MAF=18.6% in controls). We also examined rs7532266 near HTR1D (r2=1 with rs674386 and 30kb away), but did not find evidence of association (P=0.44, MAF=31.1% in cases, MAF=32.0% in controls). As the Brown et al. study showed association between HTR1D and RAN, but not with the broader diagnosis of AN, we also compared the subset of RAN cases and control subjects. We detected weak association signals for rs7532266 (P=0.04, MAF=28.5% in RAN cases) with the allelic effects observed in the same direction as previous reports.13 There is no sample overlap between the Brown et al. and our study, and combining these studies by Fisher's combined method showed that OPRD1 is associated with AN (P=1.76 × 10−5), yet HTR1D is specifically associated with RAN (P=0.006).

We also examined the SNP rs2048332 on 1q41 reported in a Japanese study.22 This SNP was not placed on the Illumina Human610 array, so we examined rs6604568 instead, which has r2=0.84 with rs2048332 and is only 15.6kb away. The SNP rs6604568 was not associated with AN in our study (P=0.13, MAF=28.0% in cases, MAF=29.7% in controls). Comparing RAN cases and control subjects did not reveal any association either (P=0.62), suggesting that the previously reported association signals may be ethnicity specific. Furthermore, a CAG repeat polymorphism in KCNN3 has been associated with AN.38, 39 The CAG repeat was not annotated in HapMap samples, so we examined all SNPs within KCNN3. We identified 85 SNPs within KCNN3 and 6 SNPs near KCNN3 (0.3–27kb away), with the most significant SNP being rs906281 (P=0.002). However, the SNPs may not tag the CAG repeat polymorphism.

Genome-wide investigation of CNVs

Besides common SNPs, rare CNVs have been associated with multiple related neuropsychiatric disorders, including schizophrenia,40, 41 autism42 and bipolar disorder.43 Thus, we next examined the role of rare CNVs in predisposing to AN. Using signal intensity data from SNP arrays, we generated CNV calls for 1015 AN cases and 3532 control subjects who passed quality control measures by PennCNV.34 As increased ‘CNV load’ has been reported in schizophrenia cases,40 we first examined whether we can find more CNVs in AN cases than control subjects. We caution that these types of ‘load’ analysis may be highly susceptible to data biases caused by signal quality in genotyping experiments, so we investigated several different thresholds for CNV calling. When utilizing default PennCNV parameters for CNV calling (greater than or equal to3 markers), we observed slightly fewer CNVs in cases than controls (30.4 vs 31.2 per subject). When restricting the analysis to a set of more confident CNV calls (greater than or equal to10 markers, CNVgreater than or equal to100kb), we observed a similar number of CNVs in cases vs control subjects (3.0 vs 3.2 per subject). When examining large CNVs (greater than or equal to10 markers, CNVgreater than or equal to1Mb), similar observations were made (0.12 vs 0.12 per subject). No CNVs in cases or control subjects were >2Mb.

Given the similarity of total CNV loads in AN cases and control subjects, we next assessed whether AN cases tend to have more rare CNVs. For this analysis, we defined common CNV regions as those regions disrupted in >1% subjects, and then classified CNVs as ‘common’ if >50% of their genomic span overlap with a common CNV region, or ‘rare’ otherwise. In AN cases and control subjects, 22.7 and 22.7% of the CNVs were classified as rare CNVs, respectively. When examining rare CNVs>100kb, >500kb or >1Mb, we still could not obtain evidence that rare CNVs were over-represented in AN cases compared with control subjects. Therefore, the ‘genetic load’ of rare CNVs may have a less important function in AN susceptibility than in other psychiatric disorders.

Rare CNVs in AN cases point to specific regions and genes

As no CNV studies on AN have been published, we do not have any specific CNV findings to confirm. Instead, we first focused on a few CNVs implicated in other neuropsychiatric diseases such as schizophrenia, autism and epilepsy.40, 44, 45, 46, 47, 48 These include recurrent deletions/duplications on 1q21.1, 15q11.2, 15q13.3, 16p11.2, 16p13.1 and 22q11.2 as well as >100kb exonic deletions on NRXN1, all of which are very large and can be readily detected by the current SNP arrays. Given the small sample size, we did not expect to detect an association, and, therefore, presented descriptive results. In AN cases, we detected one 1.5Mb deletion and one duplication on 1q21.1, three 370kb deletions and 13 duplications on 15q11.2, one 800kb deletion and one duplication on 16p13.1, as well as one 530kb duplication on 16p11.2. These individuals are all female. Clinically, they do not appear to stand out relative to other cases with the exception that the case with the 1q21.1 deletion reported a lifetime minimum body mass index of 11.4, which is extremely low, even for individuals with AN. In comparison, in control subjects, we observed 16 deletions and 21 duplications on 15q11.2, one deletion on 15q13.3, two deletions and eight duplications on 16p13.1, one deletion and three duplications on 16p11.2. Although the 15q11.2 duplication showed some evidence of association (P=0.036, two-sided Fisher's exact test), it did not pass adjustment for multiple testing. These results were also summarized in Supplementary Table 8.

We next examined rare and large CNVs in AN cases to determine whether some of them tended to cluster in the same genomic region (Table 2). Interestingly, we identified a large and rare CNV on 13q12 (chr13:22426685–23795901, ~1.4Mb) that was present in two AN cases (Figure 2a). The CNVs were not detected in control subjects in our study. To identify the true population frequency of this particular CNV, we subsequently examined a larger data set comprising over 72918 subjects genotyped by the Illumina 550K or 610K arrays (all samples were genotyped at the CHOP), and identified 38 subjects carrying the 13q12 CNV (>80% base overlap), including 15 deletions and 23 duplications. Therefore, the population frequency of this deletion is ~1/5000 vs 1/500 for AN; however, this 10-fold enrichment could be inflated by winner's curse, and we note that the 95% confidence interval for odds ratio is 1.06–41.3. Several genes were enclosed in this CNV; among them, SCAS is highly expressed in neuronal tissues, and mutations in this gene result in autosomal recessive spastic ataxia.49 Additionally, one duplication (877kb) and one deletion (952kb) both disrupt the CNTN6 and CNTN4 genes and their intergenic regions (Figure 2b). Deletions or translocations at this region have been reported in autism, developmental delay and other diseases.50, 51 Considering that the case:control ratio in our data set is not 1:1, in addition to the analysis on case-specific CNVs, we also listed all large and rare (>500kb, <1%) CNVs in Table 2. Given the relatively small sample size, none of these regions would reach statistical significance. However, rare genetic variations with high penetrance may point to important candidate genes that are worthy of follow-up analysis.

Figure 2.
Figure 2 - Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact help@nature.com or the author

Genome Browser shots and visual validation on (a) 13q12 deletions and (b) deletions/duplications disrupting the CNTN4/CNTN6 region. The predicted copy number variation (CNV) region was enclosed within two gray vertical lines, whereas the red dots represent markers within the CNV. The Log R ratio (LRR) plot shows total signal intensity, and decreased or increased LRR suggests the presence of deletions or duplications. The B allele frequency (BAF) plot shows allelic intensity ratio, and lack of BAF around 0.5 suggests the presence of deletions, yet BAF clusters around 0.33 and 0.66 suggest the presence of duplications.

Full figure and legend (388K)Download PowerPoint slide (444 KB)



In this study, we performed genome-wide investigation of both SNPs and CNVs to identify genetic variants associated with AN. To our knowledge, this is the first study that utilized whole-genome and high-density genotyping technology to investigate the genetic basis of AN, and our analysis identified specific genes and regions that are worthy of additional studies in independent cohorts. However, while the CNV loci identified are compelling, we also wish to discuss several caveats related to the study design and the interpretation of the results.

First, despite the relatively large sample size (>1000 cases) compared with any previous association studies on AN, we were still unable to find genome-wide significant signals. This observation is similar to previous modest-size studies on bipolar disorder and schizophrenia, and may suggest that much larger sample sizes are needed to identify common variants with small effects for neuropsychiatric disorders. Alternatively, it may also suggest that rare variants probably have more significant functions for AN, and are hence difficult to detect without much larger sample sizes in GWAS. This observation motivates us to supplement common SNP analysis with rare CNV analysis. However, we were unable to garner support that rare CNVs tend to be enriched in cases compared with control subjects. While acknowledging that such ‘genetic load’ analysis is highly susceptible to noises in the data and differential biases in genotyping experiments, these results seem to suggest that rare variants, including recurrent CNVs and non-recurrent ones, appear to have a less important function, compared with other psychiatric illness such as schizophrenia and autism.40, 41, 44, 45

Second, compared with SNPs, given the rare nature of large CNVs that disrupt genes, it requires even larger sample sizes to establish the true association of CNVs and disease phenotypes. Therefore, although the genes identified by our rare CNV analysis (Table 2) are intriguing candidates, additional sample sets with whole-genome genotyping data are necessary to further validate these results and establish confirmatory associations. On the other hand, in this study, we did not specifically test for common CNVs, in part because common CNVs tend to be small (with increased false negative rates to detect by computer algorithms), and they are already well tagged by the SNP genotyping arrays.52

Finally, given that the phenotypic presentation of AN could differ between various subtypes, we have also performed subgroup-based analysis on multiple subgroups, including RAN, AN cases with family history, or AN cases with early age of diagnosis. However, none of these subtype analysis yielded genome-wide significant results. We note that this analysis further reduced sample size, so it has even smaller power to detect true associations. Nevertheless, it would be important and reasonable to explore whether specific association signals tend to be stronger or even tend to exist exclusively in specific subtypes of the diseases. For example, in our study, the OPRD1 association with AN was observed exclusively in the comparison between RAN and control subjects, and no evidence of association was found for the entire AN group.

We also find it important to discuss our results in comparison with other neuropsychiatric/neurodevelopmental diseases such as schizophrenia, bipolar disorder and autism spectrum disorders. Unlike other psychiatric disorders, the clinical diagnosis of AN is relatively straightforward, and it is much less likely that an incorrect diagnosis is given to AN cases. Additionally, the phenotypes are much less heterogeneous than affective disorders or autism spectrum disorders. Therefore, one would assume that phenotypic heterogeneity is less of a concern for AN. On the other hand, there are several unique challenges for the genetic analysis of AN. Compared with other psychiatric disorders, AN may be more likely to be related to cultural differences, which may complicate the genetic studies. The relative contributions of culture and biology to AN have been long debated in this field; future association analyses that incorporate cultural background or socioeconomic status may lead to improved power and precision to identify AN genes.

In conclusion, through genome-wide association analysis of both SNPs and CNVs, we have confirmed two previously published candidate genes, yet also identified additional genes that are likely to be involved in AN pathogenesis. These genes require testing in additional independent data sets. The study illustrates the importance of using whole-genome and unbiased genomic survey to identify disease susceptibility genes and mechanisms.


Conflict of interest

The authors declare no conflict of interest.



  1. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders. American Psychiatric Publishing, Inc, 2000.
  2. Hudson JI, Hiripi E, Pope Jr HG, Kessler RC. The prevalence and correlates of eating disorders in the National Comorbidity Survey Replication. Biol Psychiatry 2007; 61: 348–358. | Article | PubMed | ISI |
  3. Attia E. Anorexia nervosa: current status and future directions. Annu Rev Med 2010; 61: 425–435. | Article | PubMed |
  4. Papadopoulos FC, Ekbom A, Brandt L, Ekselius L. Excess mortality, causes of death and prognostic factors in anorexia nervosa. Br J Psychiatry 2009; 194: 10–17. | Article | PubMed |
  5. Strober M, Freeman R, Lampert C, Diamond J, Kaye W. Controlled family study of anorexia nervosa and bulimia nervosa: evidence of shared liability and transmission of partial syndromes. Am J Psychiatry 2000; 157: 393–401. | Article | PubMed | ISI | ChemPort |
  6. Lilenfeld LR, Kaye WH, Greeno CG, Merikangas KR, Plotnicov K, Pollice C et al. A controlled family study of anorexia nervosa and bulimia nervosa: psychiatric disorders in first-degree relatives and effects of proband comorbidity. Arch Gen Psychiatry 1998; 55: 603–610. | Article | PubMed | ISI | ChemPort |
  7. Bulik CM, Sullivan PF, Tozzi F, Furberg H, Lichtenstein P, Pedersen NL. Prevalence, heritability, and prospective risk factors for anorexia nervosa. Arch Gen Psychiatry 2006; 63: 305–312. | Article | PubMed | ISI |
  8. Kortegaard LS, Hoerder K, Joergensen J, Gillberg C, Kyvik KO. A preliminary population-based twin study of self-reported eating disorder. Psychol Med 2001; 31: 361–365. | Article | PubMed |
  9. Pinheiro AP, Root T, Bulik CM. The genetics of anorexia nervosa: current findings and future perspectives. Int J Child Adolesc Health 2009; 2: 153–164. | PubMed |
  10. Bulik CM, Slof-Op’t Landt MC, van Furth EF, Sullivan PF. The genetics of anorexia nervosa. Annu Rev Nutr 2007; 27: 263–275. | Article | PubMed | ISI | ChemPort |
  11. Grice DE, Halmi KA, Fichter MM, Strober M, Woodside DB, Treasure JT et al. Evidence for a susceptibility gene for anorexia nervosa on chromosome 1. Am J Hum Genet 2002; 70: 787–792. | Article | PubMed | ISI | ChemPort |
  12. Bergen AW, van den Bree MB, Yeager M, Welch R, Ganjei JK, Haque K et al. Candidate genes for anorexia nervosa in the 1p33-36 linkage region: serotonin 1D and delta opioid receptor loci exhibit significant association to anorexia nervosa. Mol Psychiatry 2003; 8: 397–406. | Article | PubMed | ISI | ChemPort |
  13. Brown KM, Bujac SR, Mann ET, Campbell DA, Stubbins MJ, Blundell JE. Further evidence of association of OPRD1 & HTR1D polymorphisms with susceptibility to anorexia nervosa. Biol Psychiatry 2007; 61: 367–373. | Article | PubMed |
  14. Devlin B, Bacanu SA, Klump KL, Bulik CM, Fichter MM, Halmi KA et al. Linkage analysis of anorexia nervosa incorporating behavioral covariates. Hum Mol Genet 2002; 11: 689–696. | Article | PubMed | ISI | ChemPort |
  15. Bacanu SA, Bulik CM, Klump KL, Fichter MM, Halmi KA, Keel P et al. Linkage analysis of anorexia and bulimia nervosa cohorts using selected behavioral phenotypes as quantitative traits or covariates. Am J Med Genet B Neuropsychiatr Genet 2005; 139B: 61–68.
  16. Gorwood P, Ades J, Bellodi L, Cellini E, Collier DA, Di Bella D et al. The 5-HT(2A) -1438G/A polymorphism in anorexia nervosa: a combined analysis of 316 trios from six European centres. Mol Psychiatry 2002; 7: 90–94. | Article | PubMed |
  17. Bergen AW, Yeager M, Welch RA, Haque K, Ganjei JK, van den Bree MB et al. Association of multiple DRD2 polymorphisms with anorexia nervosa. Neuropsychopharmacology 2005; 30: 1703–1710. | Article | PubMed | ChemPort |
  18. Gabrovsek M, Brecelj-Anderluh M, Bellodi L, Cellini E, Di Bella D, Estivill X et al. Combined family trio and case-control analysis of the COMT Val158Met polymorphism in European patients with anorexia nervosa. Am J Med Genet B Neuropsychiatr Genet 2004; 124B: 68–72.
  19. Ribases M, Gratacos M, Fernandez-Aranda F, Bellodi L, Boni C, Anderluh M et al. Association of BDNF with anorexia, bulimia and age of onset of weight loss in six European populations. Hum Mol Genet 2004; 13: 1205–1212. | Article | PubMed | ChemPort |
  20. Ribases M, Gratacos M, Fernandez-Aranda F, Bellodi L, Boni C, Anderluh M et al. Association of BDNF with restricting anorexia nervosa and minimum body mass index: a family-based association study of eight European populations. Eur J Hum Genet 2005; 13: 428–434. | Article | PubMed | ISI | ChemPort |
  21. Cellini E, Nacmias B, Brecelj-Anderluh M, Badia-Casanovas A, Bellodi L, Boni C et al. Case-control and combined family trios analysis of three polymorphisms in the ghrelin gene in European patients with anorexia and bulimia nervosa. Psychiatr Genet 2006; 16: 51–52. | Article | PubMed |
  22. Nakabayashi K, Komaki G, Tajima A, Ando T, Ishikawa M, Nomoto J et al. Identification of novel candidate loci for anorexia nervosa at 1q41 and 11q22 in Japanese by a genome-wide association analysis with microsatellite markers. J Hum Genet 2009; 54: 531–537. | Article | PubMed |
  23. Cook Jr EH, Scherer SW. Copy-number variations associated with neuropsychiatric conditions. Nature 2008; 455: 919–923. | Article | PubMed | ISI | ChemPort |
  24. Kaye WH, Lilenfeld LR, Berrettini WH, Strober M, Devlin B, Klump KL et al. A search for susceptibility loci for anorexia nervosa: methods and sample description. Biol Psychiatry 2000; 47: 794–803. | Article | PubMed | ISI |
  25. Gendall KA, Joyce PR, Carter FA, McIntosh VV, Jordan J, Bulik CM. The psychobiology and diagnostic significance of amenorrhea in patients with anorexia nervosa. Fertil Steril 2006; 85: 1531–1535. | Article | PubMed |
  26. Hebebrand J, Himmelmann GW, Heseker H, Schafer H, Remschmidt H. Use of percentiles for the body mass index in anorexia nervosa: diagnostic, epidemiological, and therapeutic considerations. Int J Eat Disord 1996; 19: 359–369. | Article | PubMed | ChemPort |
  27. Fichter MM, Herpertz S, Quadflieg N, Herpertz-Dahlmann B. Structured interview for anorexic and bulimic disorders for DSM-IV and ICD-10: updated (third) revision. Int J Eat Disord 1998; 24: 227–249. | Article | PubMed | ChemPort |
  28. Garner D. Eating Disorders Inventory-2: Professional Manual. Psychological Assessment Resources, Inc.: Odessa, FL, 1991.
  29. Goodman WK, Price LH, Rasmussen SA, Mazure C, Fleischmann RL, Hill CL et al. The yale-brown obsessive compulsive scale. I. Development, use, and reliability. Arch Gen Psychiatry 1989; 46: 1006–1011. | PubMed | ISI | ChemPort |
  30. Cloninger CR, Przybeck TR, Svrakic DM, Wetzel RD. The Temperament and Character Inventory (TCI): A Guide to Its Development and Use. Center for Psychobiology of Personality, Washington University: St Louis, MO, 1994.
  31. Frost R, Marten P, Lahart C, Rosenblate R. The dimensions of perfectionism. Cognit Ther Res 1990; 14: 449–468. | Article |
  32. McCarthy MI, Abecasis GR, Cardon LR, Goldstein DB, Little J, Ioannidis JP et al. Genome-wide association studies for complex traits: consensus, uncertainty and challenges. Nat Rev Genet 2008; 9: 356–369. | Article | PubMed | ISI | ChemPort |
  33. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, Bender D et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet 2007; 81: 559–575. | Article | PubMed | ISI | ChemPort |
  34. Wang K, Li M, Hadley D, Liu R, Glessner J, Grant SFA et al. PennCNV: an integrated hidden Markov model designed for high-resolution copy number variation detection in whole-genome SNP genotyping data. Genome Res 2007; 17: 1665–1674. | Article | PubMed | ISI | ChemPort |
  35. Diskin SJ, Li M, Hou C, Yang S, Glessner J, Hakonarson H et al. Adjustment of genomic waves in signal intensities from whole-genome SNP genotyping platforms. Nucleic Acids Res 2008; 36: e126. | Article | PubMed | ChemPort |
  36. O’Donovan MC, Craddock N, Norton N, Williams H, Peirce T, Moskvina V et al. Identification of loci associated with schizophrenia by genome-wide association and follow-up. Nat Genet 2008; 40: 1053–1055. | Article | PubMed | ISI | ChemPort |
  37. Wang K, Zhang H, Ma D, Bucan M, Glessner JT, Abrahams BS et al. Common genetic variants on 5p14.1 associate with autism spectrum disorders. Nature 2009; 459: 528–533. | Article | PubMed | ISI | ChemPort |
  38. Koronyo-Hamaoui M, Gak E, Stein D, Frisch A, Danziger Y, Leor S et al. CAG repeat polymorphism within the KCNN3 gene is a significant contributor to susceptibility to anorexia nervosa: a case-control study of female patients and several ethnic groups in the Israeli Jewish population. Am J Med Genet B Neuropsychiatr Genet 2004; 131B: 76–80.
  39. Koronyo-Hamaoui M, Danziger Y, Frisch A, Stein D, Leor S, Laufer N et al. Association between anorexia nervosa and the hsKCa3 gene: a family-based and case control study. Mol Psychiatry 2002; 7: 82–85. | Article | PubMed |
  40. International Schizophrenia Consortium. Rare chromosomal deletions and duplications increase risk of schizophrenia. Nature 2008; 455: 237–241. | Article | PubMed | ISI | ChemPort |
  41. Need AC, Ge D, Weale ME, Maia J, Feng S, Heinzen EL et al. A genome-wide investigation of SNPs and CNVs in schizophrenia. PLoS Genet 2009; 5: e1000373. | Article | PubMed | ChemPort |
  42. Glessner JT, Wang K, Cai G, Korvatska O, Kim CE, Wood S et al. Autism genome-wide copy number variation reveals ubiquitin and neuronal genes. Nature 2009; 459: 569–573. | Article | PubMed | ISI | ChemPort |
  43. Zhang D, Cheng L, Qian Y, Alliey-Rodriguez N, Kelsoe JR, Greenwood T et al. Singleton deletions throughout the genome increase risk of bipolar disorder. Mol Psychiatry 2009; 14: 376–380. | Article | PubMed | ISI | ChemPort |
  44. Stefansson H, Rujescu D, Cichon S, Pietilainen O, Ingason A, Steinberg S et al. Large recurrent microdeletions associated with schizophrenia. Nature 2008; 455: 232–236. | Article | PubMed | ISI | ChemPort |
  45. Weiss LA, Shen Y, Korn JM, Arking DE, Miller DT, Fossdal R et al. Association between microdeletion and microduplication at 16p11.2 and autism. N Engl J Med 2008; 358: 667–675. | Article | PubMed | ISI | ChemPort |
  46. Helbig I, Mefford HC, Sharp AJ, Guipponi M, Fichera M, Franke A et al. 15q13.3 microdeletions increase risk of idiopathic generalized epilepsy. Nat Genet 2009; 41: 160–162. | Article | PubMed | ISI | ChemPort |
  47. Rujescu D, Ingason A, Cichon S, Pietilainen OP, Barnes MR, Toulopoulou T et al. Disruption of the neurexin 1 gene is associated with schizophrenia. Hum Mol Genet 2009; 18: 988–996. | PubMed | ISI | ChemPort |
  48. Ingason A, Rujescu D, Cichon S, Sigurdsson E, Sigmundsson T, Pietilainen OP et al. Copy number variations of chromosome 16p13.1 region associated with schizophrenia. Mol Psychiatry 2009.
  49. Engert JC, Berube P, Mercier J, Dore C, Lepage P, Ge B et al. ARSACS, a spastic ataxia common in northeastern Quebec, is caused by mutations in a new gene encoding an 11.5-kb ORF. Nat Genet 2000; 24: 120–125. | Article | PubMed | ISI | ChemPort |
  50. Fernandez T, Morgan T, Davis N, Klin A, Morris A, Farhi A et al. Disruption of contactin 4 (CNTN4) results in developmental delay and other features of 3p deletion syndrome. Am J Hum Genet 2004; 74: 1286–1293. | Article | PubMed | ISI | ChemPort |
  51. Roohi J, Montagna C, Tegay DH, Palmer LE, DeVincent C, Pomeroy JC et al. Disruption of contactin 4 in three subjects with autism spectrum disorder. J Med Genet 2009; 46: 176–182. | Article | PubMed | ISI |
  52. Conrad DF, Pinto D, Redon R, Feuk L, Gokcumen O, Zhang Y et al. Origins and functional impact of copy number variation in the human genome. Nature 2009; 464: 704–712. | Article | PubMed | ISI | ChemPort |



Members of the Price Foundation Collaborative Group

Harry Brandt7, Steve Crawford7, Scott Crow8, Manfred M Fichter9, Katherine A Halmi10, Craig Johnson11, Allan S Kaplan12,13,14, Maria La Via15, James Mitchell16,17, Michael Strober18, Alessandro Rotondo19, Janet Treasure20, D Blake Woodside13,14,21, Cynthia M Bulik15,22, Pamela Keel21, Kelly L Klump23, Lisa Lilenfeld24, Laura M Thornton15, Kathy Plotnicov25, Andrew W Bergen26 and Pierre Magistretti27

7Department of Psychiatry, University of Maryland School of Medicine, Baltimore, MD, USA

8Department of Psychiatry, University of Minnesota, Minneapolis, MN, USA

9Roseneck Hospital for Behavioral Medicine, Prien, Germany and Department of Psychiatry, University of Munich (LMU), Munich, Germany

10New York Presbyterian Hospital-Westchester Division, Weill Medical College of Cornell, University, White Plains, NY, USA

11Eating Recovery Center, Denver, Colorado, USA

12Center for Addiction and Mental Health, Toronto, Canada

13Department of Psychiatry, Toronto General Hospital, University Health Network, Toronto, Canada

14Department of Psychiatry, University of Toronto, Toronto, Canada

15Department of Psychiatry, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA

16Neuropsychiatric Research Institute, Fargo, ND, USA

17Department of Clinical Neuroscience, University of North Dakota School of Medicine and Health Sciences, Grand Forks, ND, USA

18Department of Psychiatry and Biobehavioral Sciences, David Geffen School of Medicine, University of California at Los Angeles, Los Angeles, CA, USA

19Department of Psychiatry, Neurobiology, Pharmacology, and Biotechnology, University of Pisa, Pisa, Italy

20Dept Academic Psychiatry, Bermondsey Wing Guys Hospital, University of London, UK

21Department of Psychology, Florida State University, Tallahassee, FL, USA

22Department of Nutrition, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA

23Department of Psychology, Michigan State University, East Lansing, MI, USA

24Clinical Psychology Program, Argosy University, Washington, DC, USA

25Department of Psychiatry, University of Pittsburgh, Pittsburgh, PA, USA

26Center for Health Sciences, SRI International, Menlo Park, CA, USA

27Brain Mind Institute, EPFLCH-1015 Lausanne, Switzerland



We gratefully thank all the patients and their families who were enrolled in this study, as well as all the control subjects who donated blood samples to Children's Hospital of Philadelphia (CHOP) for genetic research purposes. We thank the Price Foundation for their support of the Collaborative Group effort that was responsible for recruitment of patients, collection of clinical information and provision of the DNA samples used in this study. The authors also thank the Klarman Family Foundation for supporting the study. We thank the technical staff at the Center for Applied Genomics at CHOP for producing the genotypes used for analyses and the nursing, medical assistant and medical staff for their invaluable help with sample recruitments. CTB and NJS are funded in part by the Scripps Translational Sciences Institute Clinical Translational Science Award [Grant Number U54 RR0252204-01]. All genome-wide genotyping was funded by an Institute Development Award to the Center for Applied Genomics from the CHOP.

Supplementary Information accompanies the paper on the Molecular Psychiatry website