Human Genetics

, Volume 120, Issue 6, pp 889–906

Evidence for statistical epistasis between catechol-O-methyltransferase (COMT) and polymorphisms in RGS4, G72 (DAOA), GRM3, and DISC1: influence on risk of schizophrenia

Authors

  • Kristin K. Nicodemus
    • Clinical Brain Disorders Branch, National Institute of Mental HealthNational Institute of Health
    • Department of EpidemiologyJohns Hopkins Bloomberg School of Public Health
  • Bhaskar S. Kolachana
    • Clinical Brain Disorders Branch, National Institute of Mental HealthNational Institute of Health
  • Radhakrishna Vakkalanka
    • Clinical Brain Disorders Branch, National Institute of Mental HealthNational Institute of Health
  • Richard E. Straub
    • Clinical Brain Disorders Branch, National Institute of Mental HealthNational Institute of Health
  • Ina Giegling
    • Molecular and Clinical Neurobiology, Department of PsychiatryLudwig Maximilians University
  • Michael F. Egan
    • Clinical Brain Disorders Branch, National Institute of Mental HealthNational Institute of Health
  • Dan Rujescu
    • Molecular and Clinical Neurobiology, Department of PsychiatryLudwig Maximilians University
    • Clinical Brain Disorders Branch, National Institute of Mental HealthNational Institute of Health
    • Genes, Cognition and Psychosis ProgramIRP, NIMH, NIH
Original Investigation

DOI: 10.1007/s00439-006-0257-3

Cite this article as:
Nicodemus, K.K., Kolachana, B.S., Vakkalanka, R. et al. Hum Genet (2007) 120: 889. doi:10.1007/s00439-006-0257-3

Abstract

Catechol-O-methyltransferase (COMT) regulates dopamine degradation and is located in a genomic region that is deleted in a syndrome associated with psychosis, making it a promising candidate gene for schizophrenia. COMT also has been shown to influence prefrontal cortex processing efficiency. Prefrontal processing dysfunction is a common finding in schizophrenia, and a background of inefficient processing may modulate the effect of other candidate genes. Using the NIMH sibling study (SS), a non-independent case-control set, and an independent German (G) case-control set, we performed conditional/unconditional logistic regression to test for epistasis between SNPs in COMT (rs2097603, Val158Met (rs4680), rs165599) and polymorphisms in other schizophrenia susceptibility genes. Evidence for interaction was evaluated using a likelihood ratio test (LRT) between nested models. SNPs in RGS4, G72, GRM3, and DISC1 showed evidence for significant statistical epistasis with COMT. A striking result was found in RGS4: three of five SNPs showed a significant increase in risk [LRT P-values: 90387 = 0.05 (SS); SNP4 = 0.02 (SS), 0.02 (G); SNP18 = 0.04 (SS), 0.008 (G)] in interaction with COMT; main effects for RGS4 SNPs were null. Significant results for SNP4 and SNP18 were also found in the German study. We were able to detect statistical interaction between COMT and polymorphisms in candidate genes for schizophrenia, many of which had no significant main effect. In addition, we were able to replicate other studies, including allelic directionality. The use of epistatic models may improve replication of psychiatric candidate gene studies.

Introduction

Schizophrenia (SCZD [MIM 181500]) is a complex genetic disorder and genetic predisposition is likely to be determined in a complex network of interactions between genes and/or genes and environmental risk factors. Risch (1990) estimated 2–3 interacting loci affected the risk of schizophrenia in Western European populations; however, an earlier segregation analysis suggested either a multiple loci (polygenic) model or a model with a major single locus interacting with a polygenic component fit the data well (Risch 1990; Risch et al. 1984).

A number of association studies endeavoring to discover schizophrenia risk loci have been conducted, and a few genes have had some success in being replicated in independent samples (Harrison et al. 2005). However, no genetic association has been consistently replicated, likely because of issues of power, but also likely because of genetic, allelic and locus heterogeneity and the possibility that epistatic interactions may vary across samples and exaggerate or diminish the effect size of individual genotypes. One of the few genes associated with schizophrenia that has been shown to involve a functional variant is catechol-O-methyltransferase (COMT) (Kunugi et al. 1997; Li et al. 2000; Egan et al. 2001; Shifman et al. 2002). COMT is a major enzyme involved with the metabolism of dopamine, especially cortical dopamine (Tunbridge et al. 2006). Dopamine dysregulation has long been implicated as a potential pathogenic factor in schizophrenia, and recent data suggests that cortical DA may be especially important (Winterer et al. 2004). COMT is physically located on chromosome 22q11.2, within a region that is deleted in velocardiofacial syndrome (VCFS [MIM 192430]); VCFS is associated with high rates of psychotic illness (Shprintzen et al. 1981) and COMT genotypes may predict the emergence of psychosis in VCFS patients (Gothelf et al. 2005). A polymorphism at codon 158 causes a missense mutation that substitutes methionine for valine in the COMT protein (Lotta et al. 1995). The resulting Met enzyme has significantly less affinity for its substrate and thus lowers enzymatic activity (Lotta et al. 1995; Chen et al. 2004). A meta-analysis of family-based and case-control studies examining the association between the COMT Val158Met polymorphism and schizophrenia concluded the Val allele was associated with risk of schizophrenia in Caucasian family-based studies; however, this significant association was not observed for the Caucasian case-control studies, possibly because of undetected population stratification (Glatt et al. 2003). A more recent meta-analysis of case control studies found a significant but weak effect of COMT in European samples (Fan et al. 2005). Other studies have suggested that the Val/Met polymorphism may not capture the relevant genetic variation in COMT (Meyer-Lindenberg et al. 2006; Shifman et al. 2004; Handoko et al. 2005, Tunbridge et al. 2006). Studies of COMT enzyme activity (Chen et al. 2004) and of expression of COMT mRNA (Bray et al. 2003) have suggested that there may be at least three functional sites in COMT, two in regulatory regions in addition to the Val/Met coding variant. We have recently proposed that inconsistencies in the clinical association literature may be explained by different combinations of alleles at these functional loci in different population samples (Meyer-Lindenberg et al. 2006).

A hallmark feature of schizophrenia is a deficit in executive cognition, implicating the prefrontal cortex (Weinberger et al. 2001). COMT has been shown to influence executive cognition and prefrontal cortex (PFC) processing (Meyer-Lindenberg et al. 2006; Egan et al. 2001; Goldberg et al. 2003; Bruder et al. 2005; Mattay et al. 2003). However, the effect of polymorphisms in COMT on prefrontal cortex function has been shown to be complex. During neuroimaging of healthy subjects performing a working-memory task, Meyer-Lindenberg et al. (2006) found a 3-SNP haplotype of rs2097603, Val158Met, and rs165599 that was highly significantly associated with PFC efficiency; haplotypes A-Val-G and G-Val-A showed especially inefficient PFC function and haplotypes A-Met-A and G-Met-A showed especially efficient PFC function (Meyer-Lindenberg et al. 2006). Therefore, the information contained by a single SNP genotype may not be sufficient to detect information about COMT function in the PFC, which may ultimately represent the critical biological factor of impact on illness susceptibility. The association between dopamine signaling and PFC function during working memory tasks has been shown to follow an inverted-U shaped curve: too much or two little dopaminergic activity both lead to poorer working memory processing ability (Williams et al. 1995, Mattay et al. 2003). The relevance of this inverted U shaped curve to cortical function and in humans its relationship to COMT genotype has been demonstrated in several recent studies (e.g., Mattay et al. 2003; Meyer-Lindenberg et al. 2005, Gothelf et al. 2005). There is also evidence that other putative schizophrenia susceptibility genes may impact on the biology of prefrontal cortex, including GRM3 (Egan et al. 2003), DISC1 (Cannon et al. 2005), RGS4 (Prasad et al. 2005), and DAOA (Goldberg et al. 2006). Therefore, it might be hypothesized that COMT would interact with these genes to modulate their association with schizophrenia.

Previous attempts to detect epistasis contributing to risk for schizophrenia have been infrequent and inconclusive. An interaction between markers on 8p21 and 14q was reported using an affected sib-pair multilocus linkage analysis method (Chiu et al. 2002), although an earlier study showed no evidence for interaction between markers on 8p21 and 13q32 (Blouin et al. 1998). Additional linkage analyses of epistasis between marker loci on chromosomes 2, 4, 5, 6, 8, and 10 revealed no significant interactions (Sullivan et al. 2001). Association-based epistasis influencing risk for schizophrenia has been reported for NRG1 and its receptor ErbB4 (Norton et al. 2006), for the tryptophan hydroxylase gene (TPH) and the serotonin transporter gene (5-HTTLPR) (Chotai et al. 2004), for two nicotinic receptor subunits (CHRNA4 and CHRNB2) (De Luca et al. 2006) and for G72 and its interaction partner DAAO (Chumakov et al. 2002). No replication of any of these interactions has yet been reported, although the lack of replication may be likely to be attributable to small sample sizes for many studies.

Given the abovementioned estimates of oligo- or polygenic etiology of schizophrenia and the high prior probability that COMT influences the efficiency of prefrontal cortex processing and thereby the biology of susceptibility to schizophrenia, we conducted both a family-based and a case-control study assessing the evidence for statistical epistasis between COMT genotypes/haplotypes and SNPs in other schizophrenia candidate genes with at least moderate prior probability of association. Of genotypes available in our samples, we selected genes that have been reported to be positive in at least three independent studies as of 2005, adapting the critieria of Lohmueller et al. (2003) for gene selection. Although statistical evidence does not prove or disprove biologic interaction, it can indicate areas for future advancement in determining the underlying etiology of schizophrenia. Since it has been shown that Val158Met allele frequencies vary across populations (Palmatier et al. 1999; DeMille et al. 2002), if epistasis is observed between COMT and other candidate genes, it is possible that main effects of single candidate genes may be undetected unless COMT background is also taken into account through appropriate study designs or statistical models.

Materials and methods

Study population

The NIMH sibling study

Subjects were ascertained as part of the Clinical Brain Disorders Branch/National Institute of Mental Health Sibling Study (Egan et al. 2000). To qualify for participation, a potential proband had to meet DSM-IV criteria for a broad diagnosis category consisting of the following diagnoses: schizophrenia, schizoaffective disorder, simple schizophrenia, psychosis NOS, delusional disorder, schizotypal, schizoid, or paranoid personality disorder. Control individuals were ascertained from the NIH normal volunteer office and were required to not be diagnosed with a psychiatric disorder; in addition, a further requirement extended this to include the control’s first-degree relatives. Inclusion criteria for controls and probands included: between 18 and 60 years, no significant medical problems or history of head trauma, no recent history of drug and/or alcohol abuse, and IQ (for affected individuals, premorbid IQ) of above 70. All participants had to be able to give informed consent and signed an informed consent form. A research psychiatrist screened and interviewed all participants using the structured clinical interview (First et al. 1996a; First et al. 1996b). All individuals self-identified as Caucasian. Nuclear families were used for family-based analysis; the proband from each family plus unrelated controls were used as a non-independent case-control sample.

The German sample

Individuals with schizophrenia were ascertained from the Munich area in Germany and all were self-identified Caucasian. Case participants had a DSM-IV and ICD-10 diagnosis of schizophrenia. Detailed medical and psychiatric histories were collected for cases, including a clinical interview using the structured clinical interview for DSM-IV, (SCID) to evaluate lifetime Axis I and II diagnosis. Four physicians and one psychologist rated the SCID interviews and all measurements were double-rated by a senior researcher. Exclusion criteria included medical conditions that may affect the brain. All case participants were outpatients or stable in-patients.

Unrelated control volunteers were randomly selected from the general population of Munich, Germany, and contacted by mail. All individuals self-identified as Caucasian. To exclude participants who had psychiatric disorders or who had first-degree relatives with psychiatric disorders, a three-stage screening process was employed. First, a phone screening of potential controls was conducted. Second, detailed medical and psychiatric histories were assessed for the volunteers and their first-degree relatives by using systematic forms. If individuals were eligible after the screenings, they were invited to participate in the study and were administered the structured clinical interview for DSM-IV (SCID) to exclude lifetime Axis I and II psychiatric disorders. A questionnaire to ascertain any psychiatric diagnoses among their first-degree relatives was also administered. Exclusion criteria at the third stage included individuals with relevant somatic diseases or a lifetime history of any Axis I or II psychiatric disorder, or if the individual reported a first-degree relative a history of mental disorders.

Genotyping and SNP selection

In both datasets, blood was collected and DNA was extracted using standard methods. Genotypes for all SNPs were obtained using the Taqman 5′-exonuclease allelic discrimination assay (McGuigan et al. 2002). Quality control was conducted by regenotyping selected SNPs and showed a discordance rate of 0.4%. Both the sibling study and the German study were genotyped in the same laboratory; however, only SNPs exhibiting interaction with COMT on risk of schizophrenia in the sibling study were genotyped in the German sample.

The selection of SNPs in RGS4, DISC1, GRM3, NRG1, BDNF, and G72 (DAOA) was based on prior marginally significant (P-value < 0.10) association in our single-gene association studies or other published studies. We considered significant and marginally significant associations because the independent effect of single genes on risk in complex disorders is thought to be modest. With the exception of the 4 SNPs in the NRG1 ICE haplotype and the BDNF Val/Met variant, evidence for prior association for selected SNPs (N = 25) is given in Table 1. A study was considered to show association with a SNP if the SNP itself, or a haplotype containing the SNP, was reported to have a P-value of less than 0.10 in either case-control or family-based analyses based on a broad diagnosis category. An additional requirement was that the SNP had been previously genotyped in our laboratory. Note that for the four SNPs in NRG1, prior reported haplotype analysis was considered significant only if it did not include the microsatellite markers from the haplotype reported by Stefansson et al. (2002). Note that each study in Table 1 is independent of one another and all are independent of our SS and German samples, with the exception of Goldberg et al. (2006), Egan et al. (2004) and Callicott et al. (2005).
Table 1

Candidate genes with prior evidence for association with schizophrenia

GENE

SNP

Allele associated

Reporting association (P < 0.10)

Reporting no association (P < 0.10)

RGS4

rs1507754 (90387)

A

Chowdari (2002; Pittsburgh, N families = 93)

Sobell (2005; N cases = 568, N controls = 689), sibling study (unpublished data; N families = 296, N cases = 296, N controls = 370)

RGS4

rs1507754 (90387)

G

Chowdari (2002; GI families, N = 39)

Sobell (2005; N cases = 568, N controls = 689), sibling study (unpublished data; N families = 296, N cases = 296, N controls = 370)

RGS4

rs10917670 (SNP1)

G

Chowdari (2002; Pittsburgh, N families = 93), Chen, X (2004; N families = 274), Morris (2004; N cases = 249, N controls = 231), Williams (2004; N cases = 709, N controls = 710), Zhang (2005; Scottish, N cases = 580, N controls = 620)

Sobell (2005; N cases = 568, N controls = 689), Zhang (2005; Chinese, N families = 322), sibling study (unpublished data; N families = 296, N cases = 296, N controls = 370)

RGS4

rs10917670 (SNP1)

A

Chowdari (2002; GI families, N = 39)

Sobell (2005; N cases = 568, N controls = 689), Zhang (2005; Chinese, N families = 322), sibling study (unpublished data; N families = 296, N cases = 296, N controls = 370)

RGS4

rs951436 (SNP4)

G

Chowdari (2002; Pittsburgh, N families = 93, GI families, N = 39), Chen, X (2004; N families = 274), Morris (2004; N cases = 249, N controls = 231)

Sobell (2005; N cases = 568, N controls = 689), Zhang (2005; Chinese, N families = 322), sibling study (unpublished data; N families = 296, N cases = 296, N controls = 370)

RGS4

rs951436 (SNP4)

T

Chowdari (2002; GI families, N = 39), Williams (2004; N cases = 709, N controls = 710), Zhang (2005; Scottish, N cases = 580, N controls = 620)

Sobell (2005; N cases = 568, N controls = 689), Zhang (2005; Chinese, N families = 322), sibling study (unpublished data; N families = 296, N cases = 296, N controls = 370)

RGS4

rs951439 (SNP7)

G

Chowdari (2002; Pittsburgh, N families = 93), Chen, X (2004; N families = 274), Morris (2004; N cases = 249, N controls = 231), Corderiro (2005; N cases = 271, N controls = 576, N families = 49), Zhang (2005; Scottish, N cases = 580, N controls = 620)

Sobell (2005; N cases = 568, N controls = 689), Zhang (2005; Chinese, N families = 322), sibling study (unpublished data; N families = 296, N cases = 296, N controls = 370)

RGS4

rs951439 (SNP7)

A

Chowdari (2002; GI families, N = 39)

Sobell (2005; N cases = 568, N controls = 689), Zhang (2005; Chinese, N families = 322), sibling study (unpublished data; N families = 296, N cases = 296, N controls = 370)

RGS4

rs2661319 (SNP18)

G

Chowdari (2002; GI families, N = 39), Chen, X (2004; N families = 274), Morris (2004; N cases = 249, N controls = 231), Corderiro (2005; N cases = 271, N controls = 576, N families = 49) , Zhang (2005; Scottish, N cases = 580, N controls = 620)

Sobell (2005; N cases = 568, N controls = 689), Zhang (2005; Chinese, N families = 322), sibling study (unpublished data; N families = 296, N cases = 296, N controls = 370)

RGS4

rs2661319 (SNP18)

A

Chowdari (2002; GI families, N = 39), Williams (2004; N cases = 709, N controls = 710)

Sobell (2005; N cases = 568, N controls = 689), Zhang (2005; Chinese, N families = 322), sibling study (unpublished data; N families = 296, N cases = 296, N controls = 370)

G72

rs746187 (M7)

T

Goldberg (2006; N families = 230)

Chumakov (2002; French Canadian, N cases = 213, N controls = 241; Russian, N cases = 183, N controls = 183)

G72

rs3916967 (M14)

A

Chumakov (2002; French Canadian, N cases = 213, N controls = 241)

Chumakov (2002; Russian, N cases = 183, N controls = 183), Wang (2004; N cases = 537, N controls = 538), Mulle (2005; N families = 159), Korostishevsky (2004; N cases = 60, N controls = 130), Goldberg (2006; N families = 230)

G72

rs3916967 (M14)

G

Zou (2005; N families = 233)

Wang (2004; N cases = 537, N controls = 538), Mulle (2005; N families = 159), Korostishevsky (2004; N cases = 60, N controls = 130), Goldberg (2006; N families = 230)

G72

rs2391191 (M15)

G

Chumakov (2002; French Canadian, N cases = 213, N controls = 241)

Mulle (2005; N families = 159), Korostishevsky (2004; N cases = 60, N controls = 130), Goldberg (2006; N families = 230)

G72

rs2391191 (M15)

A

Schumacher (2004; N cases = 299, N controls = 300), Wang (2004; N cases = 537, N controls = 538), Zou (2005; N families = 233)

Mulle (2005; N families = 159), Korostishevsky (2004; N cases = 60, N controls = 130), Goldberg (2006; N families = 230)

G72

rs778293 (M22)

A

Chumakov (2002; French Canadian, N cases = 213, N controls = 241), Korostishevsky (2004; N cases = 60, N controls = 130)

Chumakov (2002; Russian, N cases = 183, N controls = 183), Mulle (2005; N families = 159), Goldberg (2006; N families = 230)

G72

rs3918342 (M23)

T

Chumakov (2002; French Canadian, N cases = 213, N controls = 241; Russian, N cases = 183, N controls = 183), Korostishevsky (2004; N cases = 60, N controls = 130)

Mulle (2005; N families = 159), Goldberg (2006; N families = 230)

G72

rs3918342 (M23)

C

Schumacher (2004; N cases = 299, N controls = 300)

Mulle (2005; N families = 159), Goldberg (2006; N families = 230)

G72

rs1421292 (M24)

T

Chumakov (2002; French Canadian, N cases = 213, N controls = 241; Russian, N cases = 183, N controls = 183), Schumacher (2004; N cases = 299, N controls = 300), Goldberg (2006; N families = 230)

Mulle (2005; N families = 159)

GRM3

rs187993

G

Egan (2004; sibling study, N families = 217)

None

GRM3

rs187993

T

Norton (2005; N cases = 674, N controls = 416)

None

GRM3

rs917071

C

Egan (2004; sibling study, N families = 217 & GI Caucasians, N families = 67)

Norton (2005; N cases = 674, N controls = 416)

GRM3

rs6465084

A

Egan (2004; sibling study, N families = 217)

Norton (2005; N cases = 674, N controls = 416)

GRM3

rs2228595

C

Egan (2004; sibling study, N families = 217)

Norton (2005; N cases = 674, N controls = 416)

GRM3

rs2228595

T

Marti (2002; N cases = 265, N controls = 227), Egan (2004; sibling study, N families = 217)

Norton (2005; N cases = 674, N controls = 416)

GRM3

rs1468412

T

Chen (2005; N cases = 752, N controls = 752)

Norton (2005; N cases = 674, N controls = 416)

GRM3

rs1468412

A

Fujii (2003; N cases = 100, N controls = 100), Egan (2004; GI African Americans, N families = 51)

Norton (2005; N cases = 674, N controls = 416)

DISC1

rs1572899

Allele not specified

Hodgkinson (2004; N cases = 258, N controls = 217)

Callicott (2005; SS N families = 252, GI Caucasian N families = 67; GI African American N families = 51), Zhang (2005; N cases = 338, N controls = 338)

DISC1

rs999710

Allele not specified

Hodgkinson (2004; N cases = 258, N controls = 217)

Hennah (2003; N families = 458), Callicott (2005; SS N families = 252, GI Caucasian N families = 67; GI African American N families = 51), Thomson (2005; N cases = 394, N controls = 478), Zhang (2005; N cases = 338, N controls = 338)

DISC1

rs7543610 hCV219779

C

Callicott (2005; SS N families = 252)

Callicott (2005; GI Caucasian N families = 67; GI African American N families = 51)

DISC1

rs821597 hCV1433188

G

Callicott (2005; SS N families = 252)

Hodgkinson (2004; N cases = 258, N controls = 217), Callicott (2005; GI Caucasian N families = 67; GI African American N families = 51), Thomson (2005; N cases = 394, N controls = 478)

DISC1

rs821616 hCV1433135

A

Callicott (2005; SS N families = 252), Zhang (2005; N cases = 338, N controls = 338)

Hennah (2003; N families = 458), Hodgkinson (2004; N cases = 258, N controls = 217), Callicott (2005; GI caucasian N families = 67; GI African American N families = 51), Thomson (2005; N cases = 394, N controls = 478)

Statistical methods

Hardy Weinberg equilibrium (HWE) was tested using exact methods. Linkage disequilibrium between SNPs in the same gene for SS controls was measured using r2 and D′. Main effects analyses of COMT single SNPs and haplotypes were conducted using unconditional logistic regression in the case-control samples and using the family based association test (FBAT) in nuclear families (Horvath et al. 2001). For main effects diplotype-based analyses in case-control sets, diplotypes were inferred using phase v.2.1 (Stephens et al. 2001, 2003). Comparison of continuous covariates (e.g., age at examination) between cases and controls were conducted using t-tests. Comparison of categorical covariates (e.g., sex) between cases and controls were conducted using χ2 tests or exact tests, when appropriate, to accommodate small cell counts. Conditional logistic regression, where the “cases” are the combination of alleles/genotypes transmitted to the proband and the “pseudocontrols” are those that could have been but were not transmitted to the proband, was conducted for family-based data to assess epistasis between genotypes at COMT and candidate genes and affection status (Cordell et al. 2004). A simple main effects conditional logistic regression model of this type, used here as an example of the method only and not necessarily as the actual model employed during analysis, where the 2 allele is considered dominant, can be expressed as:
$$ \ln {\left( {\frac{{P{\text{(Case $|$ FamilySet)}}}} {{1 - P{\text{(Case $|$ FamilySet)}}}}} \right)} = \upalpha + \upbeta1{\text{(SNP}}1\_2{\text{carrier)}} $$
(1)
For simplicity, assume the second SNP in an interaction model also has a dominant minor allele, and expressed as:
$$ \ln {\left( {\frac{{P{\text{(Case $|$ FamilySet)}}}} {{1 - P{\text{(Case $|$ FamilySet)}}}}} \right)} = \upalpha + \upbeta1{\text{(SNP}}1\_2{\text{carrier)}} + \upbeta2{\text{(SNP}}1\_2{\text{carrier}}*{\text{SNP}}2\_2{\text{carrier)}} $$
(2)
No main effect term for SNP2 is included in the interaction model because, within a set (family), all “cases” and “pseudocontrols” have the same value for SNP2. A main assumption of this method is independence between SNPs at genes in the general population; when this assumption is broken it introduces bias in the parameter estimates. To empirically rule out non-independence in the general population we tested for correlation between the SNPs in COMT and SNPs in other candidate genes in our normal control sample for all significant interactions reported using the conditional logistic regression. No significant or marginally significant (P ≤ 0.10) correlations were found (Val158Met-rs951436 P-value = 0.88; Val158Met-rs2661319 P-value = 0.99; rs165599-rs746187 P-value = 0.90; Val158Met-rs3916967 P-value = 0.63; Val158Met-rs2391191 P-value = 0.81; rs2097603-rs778293 P-value = 0.56; Val158Met-rs999710 P-value = 0.79). Unconditional logistic regression was used to assess epistasis between COMT genotypes/haplotypes and genotypes in candidate genes and affection status for case-control data. For both conditional and unconditional logistic regression, to determine evidence for epistasis, a likelihood ratio test (LRT) was conducted comparing nested models: using the example given above, the LRT can be expressed as 2(lnL[1]–lnL[2]) ∼ χ2 with degrees of freedom (df) equal to the difference in coefficients estimated between models (in our example, 1). Whenever possible, a general model using 2 df was used that does not impose a genetic model (additive, recessive, dominant) onto the data.

Although the alpha level to determine “significant” interaction using the LRT was set at 0.05, we included in tables all models with P-values < 0.10. ORs less than 1 are shown when the small number of individuals in the opposite genotype group prevent recoding variables as “risk” in the statistical model. P-values have not been adjusted for multiple testing for any sample. Power calculations were performed using the ggipower command (Self et al. 1992; Brown et al. 1999; Longmate 2001). Unless specified, all analysis was conducted using STATA, version 8.2 (College Station, TX).

Detection of within-population substructure is an area of active research in our laboratory, and we are currently developing and evaluating methods and SNP panels to detect population stratification in persons of European descent. At the present time we do not have conclusive evidence for or against population stratification in our sibling study sample; the German Sample has not yet been tested. Population stratification is not of concern in the family-based analyses.

Results

In the sibling study case-control sample, the number of cases was 296 and the number of controls was 370; for the family-based sample the number of families was 296. The German sample consisted of 501 cases and 627 controls. The proportion of males was higher in the cases (78.6%) than in the controls (48.6%) (P-value < 0.0001) in the sibling study and also in the German study (male cases = 65.5%, male controls = 47.7%; P-value < 0.0001); however, the age at examination was not statistically different in cases than in controls in the sibling study (P-value = 0.11). Therefore, only sex was retained in the unconditional logistic regression models to control for potential confounding.

Main effect of COMT genotypes and haplotypes

We tested whether genotypic or haplotypic association between COMT polymorphisms and schizophrenia existed in the sibling study’s family study, case-control study, or the German case-control study. Allele and haplotype frequencies are given in Table 2 for the sibling study case-control set and the German case-control set. No polymorphism in either the sibling study or the German study significantly varied from Hardy-Weinberg expectation in cases or controls (data not shown). Slight differences in allele and haplotype frequencies were present between the sibling study Caucasians and German Caucasians, although no frequency difference was greater than 10%. All single SNP association results in all samples were not significant (data not shown). Haplotype analysis was performed by grouping 3 SNP haplotypes of rs2097603, Val158Met, and rs165599 by prefrontal cortex efficiency (Meyer-Lindenberg et al. 2006). The 3-SNP haplotype was selected over the 2-SNP haplotype described by Meyer-Lindenberg at al. because the 3-SNP haplotype showed the strongest association with prefrontal function (Meyer-Lindenberg et al. 2006). Homozygotes for the inefficient haplotypes (A-Val-G or G-Val-A) and homozygotes for the efficient haplotypes (G-Met-A or A-Met-A) were considered versus a reference group consisting of all other haplotype combinations. Therefore, the reference group consisted of individuals with expected average prefrontal cortex efficiency. In the sibling study case-control sample, a nonsignificant increase in risk of schizophrenia was found for individuals homozygous for the inefficient 3-SNP haplotype (OR = 1.90; 95% Confidence Interval (CI) 0.98, 3.67; P-value = 0.056) vs. those not homozygous for either efficient or inefficient haplotypes, although a trend toward increased risk was observed. No difference in risk was observed for individuals homozygous for the efficient haplotype [OR = 1.14; 95% CI 0.66, 1.94; P-value = 0.64] versus individuals with expected average prefrontal cortex efficiency. These data provide modest support for the prior prediction that association with COMT may depend on diplotypic combinations of alleles at three functional sites in the gene, and not on a specific single SNP allele or specific haplotype. No evidence for association with schizophrenia was found in the German case-control set using COMT haplotypes, although frequencies of other functional polymorphisms among COMT Val/Val homozygotes varied across samples and case/control status (Table 3). Diplotype association analyses and diplotype-by-candidate gene interaction analyses were only conducted in the sibling study case-control and the German case-control studies.
Table 2

COMT Minor Allele and Haplotype Frequencies

SNP/Haplotype

Allele/Haplotype

SS controlsa

SS cases

German controls

German cases

rs2097603A/G

G

0.42

0.40

0.43

0.41

V158M

M

0.49

0.48

0.51

0.52

rs165599A/G

G

0.33

0.35

0.32

0.33

rs2097603A/G V158M

Efficient

0.15

0.17

0.19

0.20

rs165599A/G

Inefficient

0.078

0.14b

0.09

0.10

aSS Sibling study

bLogistic regression β coefficient P-value = 0.056

Table 3

Frequencies of COMT functional polymorphisms in COMT val homozygotes across caucasian samples

Sample

Polymorphism

Affection status

Frequency 1/1

Frequency 1/2

Frequency 2/2

Sibling study

rs2097603

Case

0.1379

0.4828

0.3793

Control

0.0923

0.5231

0.3846

German

rs2097603

Case

0.1111

0.5317

0.3571

Control

0.1208

0.4966

0.3826

Sibling study

rs165599

Case

0.7755

0.2041

0.0204

Control

0.7333

0.2667

0.0000

German

rs165599

Case

0.7752

0.2093

0.0155

Control

0.7908

0.1961

0.0131

COMT-candidate gene interaction model results

All polymorphisms found to have significant or marginally significant interaction with COMT SNPs or haplotypes displayed frequencies statistically not different from those expected under Hardy Weinberg equilibrium in cases and controls, with the exception of rs3916967 (G72/G30) in cases (Table 4). We were unable to detect statistical interaction between SNPs/haplotypes in COMT and the selected SNPs in NRG1 or the Val66Met variant of BDNF in any of our samples (data not shown).
Table 4

COMT-candidate gene interaction results

Gene

SNP

COMT MARKER

SAMPLEa

MAF cases (n cases)b

MAF controls (n controls)

Exact HWE cases

Exact HWE controls

COMT genotype or Haplotype

Candidate gene SNP genotype

OR

95% CI

OR P-value

LRT P-value

RGS4

90387 (Chowdari)

rs2097603-V158M-rs165599

SS C-C

0.36 (251)

0.43 (217)

0.68

0.61

G-Met-A/ A-Met-A

G (1) carrier

2.59

(1.07, 6.28)

0.035

0.051

RGS4

90387 (Chowdari)

rs2097603-V158M-rs165599

SS C-C

c

A-Val-G/ G-Val-A

G (1) carrier

3.26

(1.19, 8.93)

0.022

RGS4

rs951436 (SNP 4)

V158M

SS Fam

0.46 (124)

0.37

Val/Met

G/G (2/2)

3.91

(1.55, 9.87)

0.004

0.021

RGS4

rs951436 (SNP 4)

rs165599

German C-C

0.50 (485)

0.46 (616)

0.96

0.11

G/G

G/G (2/2)

3.67

(1.37, 9.81)

0.01

0.023

RGS4

rs2661319 (SNP 18)

V158M

SS Fam

0.49 (142)

0.62

Val/Met

G/G (2/2)

2.49

(1.12, 5.56)

0.026

0.044

RGS4

rs2661319 (SNP 18)

rs165599

German C-C

0.51 (487)

0.49 (608)

0.86

0.47

G/G

G/G (2/2)

3.56

(1.42, 8.93)

0.007

0.0075

G72/G30

rs746187 (M07)

rs165599

SS Fam

0.47 (118)

0.58

G/G

A/G (1/2)

4.98

(0.80, 31.20)

0.086

0.069

G72/G30

rs746187 (M07)

rs165599

SS Fam

A/G

A/G (1/2)

2.98

(1.38, 6.42)

0.005

G72/G30

rs3916967 (M14)

V158M

SS Fam

0.39 (131)

0.02

Val/Met

G (2) carrier

2.04

(1.10, 3.81)

0.024

0.068

G72/G30

rs2391191 (M15)

V158M

SS Fam

0.35 (99)

0.83

Val/Val

G (2) carrier

0.27

(0.07, 0.98)

0.046

0.062

G72/G30

rs778293 (M22)

rs2097603

SS Fam

0.37 (127)

1.00

A/G

G/G (2/2)

0.24

(0.068, 0.82)

0.022

0.071

G72/G30

rs3918342 (M23)

rs2097603

SS C-C

0.44 (260)

0.49 (352)

0.17

0.92

A/G

C (1) carrier

1.89

(0.92, 3.86)

0.083

0.028

G72/G30

rs3918342 (M23)

rs2097603-V158M-rs165599

SS C-C

G-Met-A/ A-Met-A

C (1) carrier

4.90

(1.44, 16.75)

0.011

0.019

G72/G30

rs3918342 (M23)

rs2097603-V158M-rs165599

SS C-C

A-Val-G/ G-Val-A

C (1) carrier

4.17

(0.97, 17.85)

0.054

G72/G30

rs1421292 (M24)

rs2097603-V158M-rs165599

SS C-C

0.46 (259)

0.43 (308)

0.71

0.82

A-Val-G/ G-Val-A

T (2) carrier

9.10

(1.37, 60.47)

0.022

0.018

GRM3

rs187993

V158M

SS C-C

0.34 (278)

0.29 (347)

0.35

0.6

Val/Val

G (2) carrier

1.94

(1.07, 3.52)

0.03

0.091

GRM3

rs1468412

V158M

SS C-C

0.29 (273)

0.29 (341)

0.46

0.9

Val/Val

T (2) carrier

0.51

(0.27, 0.95)

0.03

0.09

GRM3

rs1468412

rs2097603-V158M-rs165599

German C-C

0.24 (496)

0.28 (605)

0.54

0.61

A-Val-G/ G-Val-A

A/A (1/1)

2.18

(1.18, 4.01)

0.012

0.092

GRM3

rs6465084

rs2097603-V158M-rs165599

SS C-C

0.27 (254)

0.25 (294)

0.20

0.76

A-Val-G/ G-Val-A

G (2) carrier

3.06

(1.09, 8.57)

0.034

0.047

DISC1

rs7546310

V158M

SS C-C

0.50 (263)

0.48 (302)

0.22

0.64

Val/Met

C (1) carrier

2.77

(1.09, 7.02)

0.032

0.013

DISC1

rs7546310

V158M

SS C-C

Met/Met

C (1) carrier

14.63

(2.03, 105.5)

0.008

DISC1

rs999710

V158M

SS Fam

0.39 (130)

0.46

Val/Met

A (2) carrier

2.46

(1.27, 4.78)

0.008

0.093

aSS C-C Sibling study case-control, SS fam sibling study families, German C-C German case-control

bFor family-based analyses, the MAF and HWE are for probands; the number in parentheses is the number of informative families

cIndicates cell is empty because information is contained in the above cell

RGS4

Three of five SNPs selected in RGS4 showed significant association in interaction with COMT genotypes or haplotypes in the sibling study, although single SNP main effects for RGS4 in the sibling study’s family and case-control studies were all null (data not shown) (Talkowski et al. 2005) (Table 4). In addition, two SNPs (rs951436/SNP4, rs2661319/SNP18) were found to have significant interaction with COMT SNPs or haplotypes in the German case-control study. Pair-wise linkage disequilibrium measures in sibling study controls revealed relatively weak LD between 90387-SNP4 (r2 = 0.23; D′ = 0.61) and 90387-SNP18 (r2 = 0.14; D′ = 0.44), but stronger LD between SNP4-SNP18 (r2 = 0.77; D′ = 0.93). In the sibling study’s case-control analysis, a marginally significant increase in risk of schizophrenia was found in G allele carriers at the Chowdari et al. (2002) SNP 90387, in combination with both the efficient (G-Met-A/A-Met-A) and inefficient (A-Val-G/G-Val-A) 3-SNP haplotype (LRT P-value = 0.051). The 90387 G allele carriers who also were homozygous for the efficient haplotype had a 2.59-fold (95% CI 1.07, 6.29; P-value = 0.035) increased risk of schizophrenia versus individuals with at least one average efficiency haplotype and A/A 90387 genotype. Similarly, individuals carrying a G allele at SNP 90387 and who were homozygous for the inefficient haplotype showed a 3.26-fold (95% CI 1.19, 8.93; P-value = 0.022) increase in risk versus individuals not homozygous for the efficient or inefficient haplotypes and who were A/A at 90387. The G/G genotype at SNP4 and SNP18 conferred risk in the sibling study family-based set for Val158Met genotype carriers (LRT P-value = 0.021 and LRT P-value = 0.044, respectively). In the sibling study families, the G/G genotype at SNP4 was preferentially transmitted to affected offspring who were heterozygous at COMT Val158Met (OR = 3.91; 95% CI 1.55, 9.87; P-value = 0.004); this was also found at SNP18 (OR = 2.49; 95% CI 1.12, 5.56; P-value = 0.026). In the German case-control study, SNP4 G/G genotype and rs165599 G/G (LRT P-value = 0.023) showed evidence for statistical interaction, whereas at SNP18 G/G and rs165599 G/G showed strong evidence for interaction (LRT P-value = 0.0075). Individuals with G/G genotypes at COMT rs165599 and RGS4 SNP18 showed 3.56-fold (95% CI 1.42, 8.93; P-value = 0.007) higher risk of schizophrenia versus those carrying A alleles at either locus.

G72 (DAOA)

Six polymorphisms in G72 with prior positive association were found to be associated with schizophrenia status once COMT genotype or haplotype was considered in interaction models (Table 4). Linkage disequilibrium between G72 SNPs in the sibling study’s controls was generally weak (r2 < 0.02 and D′ < 0.15 for all pairwise combinations) except between M14-M15 (r= 0.81; D′ = 0.94), M22-M23 (r2 = 0.22; D′ = 0.57); M22-M24 (r2 = 0.43; D = 0.87) and M23-M24 (r2 = 0.39; D′ = 0.68). We found marginal evidence for interaction between M07 and COMT rs165599 (LRT P-value = 0.059) in the SS family-based group. Affected probands who were either rs165599 A/G or G/G genotype were preferentially transmitted the heterozygous A/G genotype at M07 (OR M07 A/G and rs165599 G/G = 4.98; 95% CI 0.80, 31.2; P-value = 0.086; OR M07 A/G and rs165599 A/G = 2.98; 95% CI 1.38, 6.42; P-value = 0.005). Marginal evidence for interaction was found at M14 (LRT P-value = 0.068) and M15 (LRT P-value = 0.062) in the SS family set. At M14 the G allele was preferentially transmitted to individuals carrying COMT 158Val/Met genotypes (OR = 2.04; 95% CI 1.10, 3.81; P-value = 0.024) whereas at M15 the G allele was under-transmitted to probands carrying COMT 158Val/Val genotypes (OR = 0.27; 95% CI 0.07, 0.98; P-value = 0.046). Only marginal evidence for interaction was observed between COMT rs2097603 A/G and M22 G/G (LRT P-value = 0.071; OR = 0.24; 95% CI 0.068, 0.82; P-value = 0.022). However, in the SS case-control sample, significant evidence for interaction (LRT P-value = 0.028) was revealed between rs2097603 heterozygotes and C allele carriers at M23 (OR = 1.89; 95% CI 0.92, 3.86, P-value = 0.083). We also observed a significant interaction between COMT haplotype and C allele carriers at M23 (LRT P-value = 0.019), with higher risk for the C allele carriers in the inefficient haplotype homozygotes (OR = 4.17; 95% CI 0.97, 17.85; P-value 0.054) and the efficient haplotype homozygotes (OR = 4.90; 95% CI 1.44, 16.75; P-value 0.011). At M24 the T carriers who were homozygous for the inefficient haplotype showed a 9.1-fold higher risk (95% CI 1.37, 60.47; P-value = 0.022; LRT P-value = 0.018) versus those who carried at least one copy of an average efficiency haplotype and who were A/A at M24. No epistasis was detected between G72 SNPs and COMT genotypes or haplotypes in the German sample, although only 3 SNPs (M14, M23, M24) with the strongest evidence for association were genotyped.

GRM3

Linkage disequilibrium between rs187993 and rs1468412 was relatively weak (r2 = 0.05; D′ = 0.52), also between rs187993 and rs6465084 (r2 = 0.06; D′ = 0.60), but was stronger between rs1468412 and rs6465084 (r2 = 0.61; D′ = 0.82). Three SNPs in GRM3 displayed at minimum marginal evidence for epistasis with COMT genotypes or haplotypes (Table 4). In the SS case-control set, SNP rs187993 G allele carriers who were COMT Val158Met Val homozygotes showed marginally significant increase in risk of schizophrenia (OR = 1.94; 95% CI 1.07, 3.52; P-value = 0.03; LRT P-value = 0.091) versus those who were T/T genotype at rs187993 and carried a Met allele at Val158Met. T allele carriers at rs1468412 who were also homozygous for the Val allele at Val158Met were marginally lower risk (OR = 0.51; 95% CI 0.27, 0.95; P-value = 0.03; LRT P-value = 0.093) for schizophrenia versus individuals who were A/A genotype at rs1468412 and who carried a Met allele at Val158Met. This was partially replicated in the German case-control sample: homozygotes for the inefficient haplotype (also containing Val/Val at Val158Met) were at marginally significantly increased risk (OR = 2.18; 95% CI 1.18, 4.01; P-value = 0.012; LRT P-value = 0.092) if they also were A/A genotype at rs1468412 versus individuals who carried at least one average efficiency haplotype and who were T carriers at rs1468412. Note the T allele shows protection and the A allele confers risk for schizophrenia, and the odds ratios correspond to this interpretation. The SS case-control sample also indicated evidence for interaction (LRT = 0.047) between homozygotes for the inefficient haplotype and G carriers at rs6465084; with individuals having both risk factors showing a 3.06-fold increased risk for schizophrenia (95% CI 1.09, 8.57; P-value = 0.034) versus those carrying at least one average efficiency haplotype and A/A genotype at rs6465084.

DISC1

Linkage disequilibrium between rs7546310 and rs999710 was weak in the SS controls (r2 = 0.04; D′ = 0.30). We detected significant evidence for interaction between DISC1 rs7546310 and Val158Met (LRT P-value = 0.013) in the SS case-control study and marginal evidence in SS family study (LRT P-value = 0.093) for interaction between rs999710 and Val158Met (Table 4). In the SS case-control set, an increasing risk was observed for individuals with greater numbers of Val158Met Met alleles who were also C carriers at rs7546310. Individuals who were COMT Val/Met and C allele carriers showed a 2.77-fold increased risk (95% CI 1.09, 7.02; P-value = 0.032) and individuals with 2 COMT Met alleles showed a 14.63-fold increased risk (95% CI 2.03, 105.5; P-value = 0.008) versus individuals who were COMT Val/Val and A/A at rs7546310. A marginally significant association was also observed between over-transmission of the A allele at rs999710 to probands among COMT Val158Met genotype sets (OR = 2.46; 95% CI 1.27, 4.78; P-value = 0.008; LRT P-value = 0.093).

Discussion

We were able to detect statistical association between COMT genotypes or haplotypes and SNPs in RGS4, G72, DISC1, and GRM3, most of which were not detectable using a single candidate gene analysis. In addition, we were able to replicate, at the gene level, interactions in RGS4 and GRM3 with COMT in an independent case-control sample of Caucasians from Germany. In other studies from our group, we have found statistical evidence of epistasis between COMT and and GAD1 (Straub et al. in press) and between COMT and DTNBP1 (Straub RE, Egan MF, Weinberger DR, unpublished data). We were not able to detect evidence for interaction between four SNPs in NRG1 or the BDNF Val66Met polymorphism with COMT genotypes or haplotypes. Power calculations suggest we had 53.5% power to detect SNP–SNP interaction in the SS case-control data and 67.1% power in the SS family-based set. These calculations are based on population disease frequency of 1%, both SNPs having 0.50 MAF, a dominant main effect at each SNP of 1.25, and in increase in risk for interaction for 2 allele carriers at both SNPs of 2.0, setting α = 0.05 and using 300 cases and controls for the case-control calculation and 300 nuclear families for the family-based calculation. Because we lacked sufficient statistical power to detect interaction even under ideal conditions of equal MAFs, a dominant model, and a two-fold increase in interaction risk, it is surprising that approximately half of all SNPs tested showed significant or marginally significant evidence for interaction with COMT. Although we performed many statistical tests, the true number of “independent” tests is far less than the total number of tests because several levels of correlation exist between each of the hypotheses tested: the 3 COMT SNPs are correlated with one another and with the COMT haplotype, within RGS4, G72 and GRM3 strong correlation (r2 > 0.50 and D′ > 0.80) was observed between several SNPs, and the SS case-control study is not independent from the family-based study because the cases are the same in both groups. Although it is not currently possible to calculate the number of independent tests to perform correction for multiple testing, we believe it is statistically unlikely to have found evidence for epistasis in four of the six previously associated genes tested given that we lacked sufficient statistical power to detect the low–moderate levels of interaction observed. However, it is very possible that any number of the results reported are spurious and all interactions will need replication by indepdendent researchers. We also find it encouraging that nearly all of our interaction results agree with the risk alleles/genotypes previously reported by independent research groups. In addition, related research groups focusing on neuroimaging and brain expression have used our statistical findings to generate hypothesis-testing experiments, a crucial factor in determining whether statistical interaction reflects interaction at the biological level. These groups have reported effects of COMT on RGS4 mRNA expression in human brain (Lipska et al. 2006), an interaction of COMT and RGS4 on cortcial function evaluated with fMRI (Buckholtz et al. submitted), and an interaction of COMT and GRM3 on prefrontal physiological efficiency (Tan et al. submitted). It is our hope that other researchers will endeavor to do the same. However, we note again that only independent replication will strengthen confidence in our findings.

RGS4

Association was detected between three SNPs in RGS4 only after considering them in interaction with COMT genotypes or haplotypes. SNP 90387 A/A was found to be marginally significantly associated with schizophrenia in individual homozygous for either the efficient or inefficient 3-SNP COMT haplotype, which is consistent with findings reported for 90387 in both the Pittsburgh Schizophrenia and NIMH family samples (Chowdari et al. 2002); however, Chowdari et al. (2002) also reported haplotypes carrying the G allele were significantly associated with schizophrenia in the NIMH family sample. It is unclear why the 90387 A/A genotype increases risk for both the efficient and inefficient 3-SNP COMT haplotype homozygote groups. However, the homozygotes for the efficient haplotype did not show a decrease in risk of schizophrenia in the main effects model (Table 2), so it is possible this haplotype is not protective against the development of schizophrenia. In fact, in both the sibling study and German case-control samples we found cases slightly enriched for the efficient haplotype. Alternatively, the 90387 A/A genotype may interact with both efficient and inefficient COMT background to increase risk for schizophrenia. We have argued that the critical issue involved in risk conveyed by COMT on schizophrenia is the effect that COMT has on prefrontal cortical function. This effect is predicted by an inverted U-shaped curve that characterizes the effect of dopamine signaling on prefrontal function; the extremes of COMT activity related to homozygosity at either maximally efficient or inefficient portions of the curve could both increase risk (Meyer-Lindenberg et al. 2006; Mattay et al. 2003). At both SNP4 and SNP18 the G/G genotype was found to increase risk of schizophrenia approximately 2.5 to 4-fold, which is in agreement with Chowdari et al. (2002), Chen et al. (2004), Morris et al. (2004) and Williams et al. (2004) for SNP4 and with Chowdari et al. (2002), Chen et al. (2004), Morris et al. (2004), Corderiro et al. (2005) and Zhang et al. (2005) for SNP18. Although the opposite allele was found to be positively associated in three studies for SNP4 (Chowdari et al. 2002; Williams et al. 2004; Cordeiro et al. 2005), the majority of the studies for SNP18, including ours, are in agreement that the G allele confers risk for schizophrenia, although at least one study did not find association between SNPs 1, 4, 7, and 18 and schizophrenia (Sobell et al. 2005). Moderately strong LD exists between these SNPs, so these results cannot be considered completely independent.

G72 (DAOA)

Our previous work in this gene had reported the Chumakov et al. (2002) M07 (rs746187) A allele and M24 (rs1421292) T allele were marginally associated with schizophrenia and/or cognitive function (Goldberg et al. 2006). Since the findings presented here show excess transmission of the heterozygous form of M07, we cannot say whether this indicates agreement with earlier work. Both of the possible alleles at markers M14 and M15 have been reported to be associated with schizophrenia in previous studies (Chumakov et al. 2002; Schumacher et al. 2004; Wang et al. 2004; Zou et al. 2005). Our results support Wang et al. (2004) and Zhou et al. (2005) at M14 and Schumacher et al. (2004), Wang et al. (2004) and Zou et al. (2005) at M15, who also found G at M14 and A at M15 to be risk alleles for schizophrenia. They are not in agreement with Chumakov et al. (2002) who reported the opposite alleles to be associated at M14 and M15. This lack of agreement could possibly be due to a different ancestral chromosome harboring a causal variant in the French Canadian sample described by Chumakov et al. (2002) given that French Canadians can be considered a population isolate. However, the more likely reason for disagreement is false positive results in one or more studies. SNPs M22, M23, and M24 are in strong LD with one another and should be considered together. We found marginal evidence for interaction between COMT SNPs and M22 in the SS family sample and significant evidence for interaction between COMT SNPs and M23 and M24 in the SS case-control sample. Although these samples are not completely independent, we consider this semi-replication at the gene level. We were not able to replicate our findings in the German case-control sample, although only 3 SNPs (M14, M23, M24) with the strongest evidence for association were genotyped. In the SS case-control sample, M22 G/G genotype was found to be weakly preferentially undertransmitted within families with rs2097603 heterozygous backgrounds. This finding is in agreement with both Chumakov et al. (2002) and Korostishevsky et al. (2004) who found positive association between the A allele and risk of schizophrenia. COMT rs2097603 heterozygotes who carried a C allele at M23 showed a 1.89-fold (95% CI 0.92, 3.86) higher risk of schizophrenia than those homozygous for the A allele at rs2097603 and T/T at M23. The interaction between the rs2097603 heterozygote and C carriers at M23 may be explained by the significant (and much stronger) evidence for interaction between the inefficient and efficient haplotype homozygotes and C allele carriers at M23, of which part of each COMT haplotype group has an A at rs2097603 and the remainder has a G; the heterozygous form of rs2097603 may be found in either homozygous haplotype group. Interestingly, both inefficient and efficient COMT haplotype homozygotes show a greater than four-fold increase in risk if they are also C allele carriers at M23 versus individuals T/T at M23 and who carry at least one average efficiency haplotype. These results are not in accordance with Chumakov et al. (2005) or Korostishevsky et al. (2004), but are in agreement with Schumacher et al. (2004). Even with the moderately strong LD between M23-M24, only inefficient COMT haplotype homozygotes are at increased risk for schizophrenia if they are also T allele carriers at M24, which is in agreement with our earlier work on cognitive function (Goldberg et al. 2006). Although based on small numbers, inefficient COMT haplotype homozygotes that also harbor a T allele at M24 were found to have a nine-fold increased risk (OR = 9.1; 95% CI 1.37, 60.47; P-value = 0.022; LRT P-value = 0.018) of schizophrenia versus individuals carrying at least one average efficiency haplotype and who are A/A at M24. This point estimate is very imprecise; however, we are encouraged by other reports of the T allele conferring risk for schizophrenia at this locus (Chumakov et al. 2002; Schumacher et al. 2004), although at least one study has reported no association between M24 and schizophrenia (Mulle et al. 2005).

GRM3

The evidence for association in single candidate gene studies of GRM3 has been more controversial than for RGS4 and G72/G30 (Table 1). We previously reported significant association between rs187993 G allele, rs917071 C allele, rs6465084 A allele, and rs2228595 C (sibling study) and T (NIMH study) alleles (Egan et al. 2004). Of the five SNPs selected for study, four of them have had reports of positive association between both possible alleles and schizophrenia or reported no significant association (Egan et al. 2004; Fujii et al. 2003; Marti et al. 2002; Chen et al. 2005; Norton et al. 2005). The present study reports marginal evidence for interaction between COMT alleles or haplotypes and SNPs in GRM3. Consistent with our previous work (Egan et al. 2004), SNP rs187993 G allele carriers were found to have higher risk of schizophrenia; however, this association is strengthened in individuals who are also homozygous for the Val form of the Val158Met COMT polymorphism. In accordance with Egan et al. (2004) and Fujii et al. (2003), after considering COMT Val158Met in interaction with rs1468412, we find the T allele is protective for schizophrenia status among individuals homozygous for the Val allele at the COMT Val158Met SNP. This finding of marginally significant interaction between Val/Val genotype at COMT Val158Met and A/A genotype at rs1468412 was semi-replicated in the German sample; however, in the German sample the COMT risk group was homozygous for the inefficient haplotype (A-Val-G or G-Val-A), who are also Val/Val at COMT Val158Met. However, at rs6465084 our results are not consistent with those previously reported by Egan et al. (2004), who reported the A allele positively associated with schizophrenia in SS family-based analyses. We found, after assessing interaction with COMT haplotypes, the G carrier at rs6465084 who was homozygous for the inefficient haplotype displayed a significantly increased risk for schizophrenia in the SS case-control sample, but found no evidence for this interaction in the SS family-based set. Likely reasons for this discrepancy are many, and include false positives due to multiple testing, although the detection of true epistasis cannot be dismissed.

DISC1

Callicott et al. (2005) reported marginally significant association between the C allele of rs7546310 and schizophrenia in the SS family set. We find significant evidence for interaction between the same allele of rs7546310 and COMT Val158Met (LRT = 0.01). Although the point estimates are imprecise, it appears that the effect of the C allele at rs7546310 is dependent on the number of COMT Met alleles, with increased risk corresponding to increased copies of the Met allele. This is especially interesting given that the Val allele, not the Met allele, of COMT has been reported to be the risk allele for schizophrenia and is on both inefficient haplotypes. The network of genes sufficient to increase risk of schizophrenia may vary across COMT background. Indeed, the Met allele, which is associated with more efficient prefrontal function during cognitive processing, shows a more abnormal brain effect during emotion processing in the hippocampal formation (Smolka et al. 2005; Drabant et al. in press). This is of interest because Callicott et al. (2005) found evidence that the DISC1 rs7546310 allele is associated with abnormal hippocampal function. We also found marginal evidence for overtransmission of the A allele in case-pseudocontrol sets having COMT Val/Met genotype. Hodgkinson et al. (2004) reported significant association with the same SNP, but the positively associated allele was not listed, so it is unclear if we have replicated their findings. We previously found no association between this SNP and schizophrenia in the SS family sample (2005). Unfortunately, we were unable to detect epistasis between COMT SNPs or haplotypes and the coding variant rs821616, although we previously reported association with this SNP (2005) along with Zhang et al. (2005), but others have reported no association between rs821616 and schizophrenia (Hennah et al. 2003; Thompson et al. 2005).

Given that risk of schizophrenia is thought to be the product of several genes, each with very modest effect across heterogeneous samples, and that detection of statistical interaction requires large sample sizes, we decided to allow for a balance between detection of false positives and rejection of true positives. Power calculations showed only moderate power to detect SNP–SNP interactions given our sample size under reasonable conditions. The replication of our findings of statistical epistasis in independent samples is crucial to determining whether epistasis between COMT and other schizophrenia candidate genes is an important factor in determining risk for schizophrenia. In addition, models that show interaction between heterozygotes at COMT may reflect the effect of the 3-SNP haplotype efficiency: for example, the COMT inefficient haplotype (A-Val-G and G-Val-A) is only constant at Val158Met. Therefore, either A or G alleles at COMT rs2097603 and rs165599 could possibly show association in isolation from the haplotype, dependent on how enriched the population under study is for a particular haplotype. We demonstrated that the distribution of rs2097603 and rs165599 genotypes carried by COMT Val homozygotes varied by both case status and Caucasian population under study, which adds strength to the argument that haplotype distributions may vary across populations. Alternatively, heterozygote interactions could reflect lack of power due to a small sample size. Another potential drawback of this study is that we did not take into account uncertainty in haplotype assignment for the analyses that considered COMT haplotype-by-candidate gene SNP interactions; ignoring the uncertainty can introduce bias (Li et al. 2003; Zhao et al. 2003). However, by restricting our analyses to those individuals homozygous for the efficient or inefficient haplotypes, the majority of individuals classified as “efficient” or “inefficient” were homozygous for a single haplotype, and thus phase was unambiguous.

We were able to partially replicate results for SNPs in RGS4, GRM3, and DISC1 in an independent German case-control sample. In each case of independent replication the same candidate gene genotype was found to increase risk; however, the COMT genotype or haplotype involved in epistasis varied between the sibling study and German samples. We find it encouraging that our interaction results for SNPs in candidate genes replicate allelic directionality reports of single-gene association by independent research groups in nearly every circumstance. In addition, frequency differences were found between the sibling study and German Caucasians for COMT genotypes and haplotypes (Table 2), although these differences were not large. One possibility for incomplete replication is that, although the interaction between COMT and each candidate gene may be important in determining schizophrenia risk, a slightly different constellation of COMT haplotypes responsible for determining risk exists in our two Caucasian populations. We have found some support for this possibility, as allele frequencies of the noncoding SNPs in COMT varied in homozygote backgrounds at Val/Met across the samples (Table 3). These differences may influence interactions if the critical factor is the biology of the gene as determined by the combination of functional alleles at the three functional loci (Meyer-Lindenberg et al. 2006). Another possibility is that our sample sizes were too small to definitively ascertain whether statistical interaction exists between COMT and other candidate genes, as detection of epistasis is dependent on penetrance and effect size for each COMT-candidate gene combination, which may vary across populations. Notwithstanding these methodological limitations, there is biologic plausibility to our finding of an impact of COMT on the risk profile of several schizophrenia susceptibility genes. COMT impacts on the signal to noise of cortical function, a putatively critical parameter in the pathophysiology of schizophrenia (Winterer et al. 2004). To the extent that other susceptibility factors, both genetic and environmental (Caspi et al. 2005), also increase risk for schizophrenia by impacting on cortical development and function, these effects might be expected to be exaggerated by the effects of COMT.

In conclusion, we were able to replicate reports from independent research groups for SNPs in RGS4, G72, and DISC1 that were not detectable without jointly considering genotypes or haplotypes in COMT. Replication in independent samples is necessary. Candidate genes for complex disorders are very unlikely to act independently of a wide net of interactions. We hope to encourage others to consider assessment of statistical and biological epistasis as important as the main effects of candidate genes.

Acknowledgments

GlaxoSmithKline supported the recruitment of the German patients.

Copyright information

© Springer-Verlag 2006