Introduction

Individuals suffering from anorexia nervosa (AN) perceive themselves as fat even when emaciated, and pursue thinness through food restriction (AN-R) or purging (AN-BP). ‘Anxiety’ genes likely contribute to the suggested genetic component in AN as sufferers are usually anxious prior to developing the disorder.1, 2 The ‘anxiety’ neurotransmitters serotonin (5-HT) and norepinephrine (NE) are cleared from the synaptic cleft by the 5-HT transporter (SERT/SLC6A4), the NE transporter (NET/SLC6A2), and monoamine oxidase A (MAOA) proteins.3 The genes that encode these proteins are consequently prime candidates for involvement in AN aetiology. The MAOA gene promoter contains a functional polymorphism known as the MAOA-uVNTR that has long (MAOA-L) and short (MAOA-S) variant forms due to a repeated 30-bp sequence. MAOA-L (4 repeats or rarely 3 repeats plus the first 18-bp of the 30-bp sequence) is transcribed more efficiently than MAOA-S (3 repeats of the 30-bp sequence). The transcriptional efficiency of another two rare alleles, 2 or 5 repeats, is unknown or debatable, respectively.4, 5 In male skin fibroblast cultures, the MAOA-L 4-repeat allele results in increased MAOA activity relative to MAOA-S.6 MAOA-L has suggestive association with panic disorder in females.5 The SERT gene promoter contains a functional polymorphism known as the 5-HTTLPR due to a 44-bp insertion (5-HTTLPR-L) or deletion (5-HTTLPR-S). The transcriptional activity of 5-HTTLPR-S is less than half that of 5-HTTLPR-L. Genotypes with one or two copies of 5-HTTLPR-S halve SERT uptake of 5-HT relative to the 5-HTTLPR-L/L genotype and appear associated with anxiety-related traits.7

In an earlier study, we reported association of AN-R with the long form of a polymorphism in a repetitive sequence known as the NETpPR in the NET gene promoter, the only published association study with this polymorphism to date.8 By itself (independently) the long form of the NETpPR polymorphism more than doubled risk for AN-R. Shortly after in a family-based study (ie parents and their children), we reported that MAOA-L is preferentially transmitted to children with AN-R if the long form of the NETpPR polymorphism is also transmitted.9 The contribution by MAO-L and the long form of the NETpPR polymorphism to increased risk of developing AN-R appeared to be additive, that is, the sum of their independent contributions, although the independent risk contributed by MAOA-L did not reach statistical significance in case control (using control data from the literature) or family-based analyses. The independent contribution by MAOA-L to increased risk of panic disorder is also thought to be small.5 To date no one has attempted a replication study to confirm or refute our findings regarding MAOA-L.9 Recently, we found no evidence for an independent contribution by the 5-HTTLPR to AN susceptibility in our family-based sample of at least a genotypic relative risk of 2.0 with 93% statistical power (odds ratio 1.0, 95% CI 0.7–1.5),10 which confirmed independent findings by others in three earlier studies.11, 12, 13 Data from the only other independent study of AN and the 5-HTTLPR suggested association of AN-BP with the 5-HTTLPR-S allele and 5-HTTLPR-S/S genotype (both P=0.02).14 Additionally, we found no suggestion of interaction between the NETpPR polymorphism and the 5-HTTLPR, that is, neither 5-HTTLPR allele appeared to increase risk of AN when 5-HTTLPR allele transmissions from 5-HTTLPR-S/L heterozygous parents were stratified based on the NETpPR genotypes of their children with AN.10 To date no one has attempted a replication study to confirm or refute our negative interaction findings between the NETpPR polymorphism and the 5-HTTLPR.10 In the current study, to further investigate our hypothesis that proteins which clear NE and 5-HT from the synaptic cleft modify susceptibility to develop AN, we assess potential interaction between the MAOA-uVNTR and 5-HTTLPR in 114 nuclear families. Findings from our analysis suggest that increased risk of AN due to the combined effects of these polymorphisms is greater than additive, meaning that these functional polymorphisms are interacting epistatically.

Materials and methods

Subjects

We recruited 114 consecutive consenting unrelated patients with AN (110 Caucasian:4 East Asian) and their parents (106 trios, eight duos) in Australia. Each trio consisted of one DSM-IV15 patient with AN plus both biological parents. Each duo consisted of one DSM-IV female patient with AN plus her biological mother. The trios and duos participated in our earlier studies in which they are described in detail.8, 9, 10 Briefly, the patients with AN consisted of 109 females (aged 16.8±3.5 years at assessment, onset of AN 2.0±2.0 years before assessment, minimum BMI measured in hospital after onset of AN: 14.6±1.7) and five males (aged 16.5±3.8 years, onset of AN 1.6±1.0 years before assessment, minimum BMI measured in hospital after onset: 14.7±1.2). Senior psychiatrists confirmed the diagnosis of DSM-IV AN blind to genotype using a structured clinical interview plus medical record data. We gained ethics approval from the appropriate institutional Committees on Human Experimentation in accord with the Helsinki Declaration of 1975. All participants gave written informed consent but parents signed for those aged under 14 years.

Molecular genetic methods

The PCR-based genotyping methods for detecting the MAOA-uVNTR and 5-HTTLPR alleles are described in our earlier reports.9, 10 The MAOA-uVNTR genotypes of the 114 patients with AN consisted of 11 MAOA-S/S, 43 MAOA-S/L, 55 MAOA-L/L, and five MAOA-L males. The MAOA-uVNTR genotypes of their mothers consisted of 16 MAOA-S/S, 55 MAOA-S/L, and 43 MAOA-L/L. The fathers' MAOA-uVNTR genotyping data could not be analysed (and are therefore not listed) as males have only one allele for X chromosome genes, for example the MAOA gene16 and therefore preferential transmission of alleles cannot be estimated. The father's MAOA-uVNTR genotype was used only to determine which allele the mother transmitted in the trios. We could deduce the mother's transmitted MAOA-uVNTR allele in all duos without knowledge of the father's genotype as no duo was composed of two MAOA-S/L heterozygotes.

5-HTTLPR genotyping data were available for the mother, father and their child (AN patient) in each of the 106 trios, and for the mother and her daughter (AN patient) in each of the eight duos. The 5-HTTLPR genotypes of the 114 patients with AN consisted of 24 5-HTTLPR-S/S (22 of these from the trios), 59 5-HTTLPR-S/L (57 of these from the trios), and 31 5-HTTLPR-L/L (27 of these from the trios). The 5-HTTLPR genotypes of the 114 mothers consisted of 29 5-HTTLPR-S/S (all from the trios), 51 5-HTTLPR-S/L (46 of these from the trios), and 34 5-HTTLPR-L/L (31 of these from the trios). The 5-HTTLPR genotypes of the 106 fathers in the trios consisted of 22 5-HTTLPR-S/S, 55 5-HTTLPR-S/L, and 29 5-HTTLPR-L/L. All and only transmissions from the 101 5-HTTLPR-S/L parents (46 mothers, 55 fathers) in the trios were included in Table 2 ‘Trio parents’ analysis. In Table 2 ‘Trio and duo mothers’ analysis, it was necessary to exclude 12 ‘trio’ mothers and 1 ‘duo’ mother as all mothers, fathers, and children in these trios, and the mother and daughter in the duo, consisted of 5-HTTLPR-S/L heterozygotes and so it was not possible to determine which allele was transmitted by the mother. Therefore all and only transmissions from 38 5-HTTLPR-S/L mothers were included in Table 2 ‘Trio and duo mothers’ analysis (ie 51 total 5-HTTLPR-S/L mothers minus 1 5-HTTLPR-S/L ‘duo’ mother minus 12 5-HTTLPR-S/L ‘trio’ mothers=38 mothers).

Table 2 Epistatic interaction analysis: parents' 5-HTTLPR TDT data stratified by MAOA-uVNTR genotypes of their AN children
Full size table

Statistical analysis

Logistic regression is useful for detecting epistatic interactions involving at least two independent variables when the distribution of responses on the dependent variable is expected to be nonlinear for at least one independent variable, although calculations become problematic when genotype combinations result in few data points. Novel methods employing mainly case control data are being developed.17, 18 We used the Armitage–Cochrane trend test to analyse Table 1 data because (a) there were two variables in our analyses including one ordinal independent variable (ie genotype) and a binomial dependent variable (ie allele transmitted), the binomial proportions across genotypes were linear, and we were posing the question ‘Is there a consistent trend towards decrease or increase in the difference in transmission of MAOA-L and MAOA-S alleles between genotype groups?’ The Armitage–Cochrane trend test is a statistically powerful test with 1 degree of freedom, which may be seen as both an advantage and disadvantage, the latter due to greater likelihood of a type 1 error. Consequently we supported our analysis of Table 1 with the Fisher–Freeman–Halton exact test for R × C contingency tables, a generalized version of the conservative Fisher's exact test. We used only the Fisher–Freeman–Halton exact test for Table 2 data analysis as binomial proportions across genotypes were not linear. We used SPSS Sample Power for K × C Tables to determine statistical power (two-tailed). We used two-sided P-values and a 0.05 level of significance.

Table 1 Epistatic interaction analysis: mothers' MAOA-uVNTR TDT data stratified by 5-HTTLPR genotypes of their AN children
Full size table

We used the following method to test the hypothesis of epistatic interaction being present: If no trend was detected (ie P≥0.05) then no evidence of epistasis was suggested. If a trend was detected (P<0.05) epistatic interaction was suggested as the 5-HTTLPR does not appear to contribute independently to risk of AN10, 11, 12, 13 and hence the trend could not be due to an additive effect. Synergistic or antagonistic epistatic interaction was suggested if the degree of risk was increased above or decreased below the risk of AN associated with MAOA-L alone, respectively.

In regard to simultaneous testing, the only analyses previously conducted on the sample in this report are those described in our studies involving three biallelic polymorphisms and two interactions between them.8, 9, 10 The current report analyses the third interaction.

Results

We tested the hypothesis that the MAOA-uVNTR and 5-HTTLPR interact epistatically in AN. We combined MAOA-uVNTR and 5-HTTLPR genotyping data from our earlier studies9, 10 with 5-HTTLPR genotyping data of the eight duos and MAOA-uVNTR genotyping data of 19 trios determined in the current study leading to an increase in MAOA-uVNTR and 5-HTTLPR genotyping data of 20 and 8%, respectively, over that reported in our earlier studies.9, 10 However, the odds ratio of 1.4 for MAOA-L (combined sample in Table 1) and odds ratio of 1.0 for 5-HTTLPR-S (combined sample in Table 2) remained the same as we reported earlier.9, 10 MAOA-uVNTR genotypes of patients and their mothers, and 5-HTTLPR genotypes of patients and parents were in Hardy–Weinberg equilibrium (data analysis not shown).

We tested for epistatic interaction between the 5-HTTLPR and MAOA-uVNTR through contingency table analysis by (a) stratifying allele transmissions from MAOA-S/L heterozygous mothers, that is, transmission disequilibrium test data19 based on the 5-HTTLPR genotypes of their affected children (Table 1), and (b) stratifying allele transmissions from 5-HTTLPR-S/L mothers and trio parents, that is, transmission disequilibrium test data19 based on the MAOA-uVNTR genotypes of their affected children (Table 2). Epistatic interaction is suggested from the data in Table 1 (Armitage–Cochrane χ2 trend test=9.457, 1 df, P=0.0021; Fisher–Freeman–Halton exact test P=0.0061). The power of the χ2 test of independence of the data in the contingency table is 80%. MAOA-L is preferentially transmitted from mothers to affected children having the 5-HTTLPR-S/S genotype (odds ratio: 11.0, 95% CI 1.6–474) but not to affected children having the 5-HTTLPR-L/L genotype (odds ratio: 0.4, 95% CI 0.1–1.6). AN-R data alone (n=98 nuclear families) produce a similar finding (Armitage–Cochrane χ2 trend test=7.477, 1 df, P=0.0063). However, the reciprocal analysis in Table 2 provides no evidence of epistatic interaction as neither 5-HTTLPR allele is preferentially transmitted when conditioned on the MAOA-uVNTR genotype of the affected child: Fisher–Freeman–Halton exact test P=0.238 for trio parents' data and P=0.619 for mothers' data alone in the entire AN sample. In support of our finding in Table 1, MAOA-uVNTR and 5-HTTLPR genotypes are independently distributed (Table 3) in mothers and AN daughters (Fisher–Freeman–Halton exact test P=0.271 and P=0.734, respectively) suggesting that observed nonrandom transmissions of MAOA-uVNTR alleles from mothers to daughters according to 5-HTTLPR genotypes of the latter do not just reflect a nonrandom distribution of 5-HTTLPR genotypes in our sample, independent of disease status.

Table 3 Joint distribution of MAOA-uVNTR and 5-HTTLPR genotypes of mothers and their AN daughters
Full size table

Discussion

Risk of AN appears to increase in a 5-HTTLPR-S/S individual who also receives a MAOA-L allele from an MAOA-S/L mother and may be eight times greater (odds ratio=11) than the risk imposed by MAOA-L alone (odds ratio=1.4) implying synergistic epistatic interaction between the MAOA-uVNTR and 5-HTTLPR. Epistatic interaction has not previously been observed between the MAOA-uVNTR and the 5-HTTLPR and cannot be due to proximity as these polymorphisms are on different chromosomes.16, 20 However, other functional sequence variation in linkage disequilibrium with either polymorphism may be contributing to AN susceptibility. In contrast, risk of AN appears not to increase in a MAOA-L/L individual who also receives a 5-HTTLPR-S allele from a 5-HTTLPR-S/L mother. We suggest two explanations for the discrepancy in the reciprocal analyses. Firstly, epistatic interaction suggested by Table 1 data may be a false positive. We measured preferential transmission of MAOA-uVNTR alleles from only mothers as the MAOA gene is on the X chromosome and consequently our sample size was restricted. Secondly, if there are two 5-HTTLPR-S alleles, for example, S1 and S2, the MAOA-uVNTR may interact with, for example, S1 (or another functional variant in linkage disequilibrium with S1) but not with S2. Preferential transmission of MAOA-L to 5-HTTLPR-S/S children is then due to 5-HTTLPR-S/S being 5-HTTLPR-S1/S1. In the reciprocal analysis, the interaction is not seen due to 5-HTTLPR-S/L mothers being S1/L and S2/L. Both S1 and S2 will be transmitted but if S2 is more common than S1 the epistatic interaction between the MAOA-uVNTR and 5-HTTLPR-S1/S1 suggested in Table 1 will be masked. If our findings in Tables 1 and 2 are independently replicated, a search should be conducted for 5-HTTLPR-S variants (including sequence variation in linkage disequilibrium with 5-HTTLPR-S) as three rare 5-HTTLPR-S variant alleles have been observed.21 A 5-HTTLPR-S variant may also form a haplotype with an allele of the intron 2 functional polymorphism22 (or another functional sequence variant nearby), which epistatically interacts with MAOA-L.

Until independently confirmed we cautiously interpret our earlier8, 9, 10 and current findings, but suggest that we may have identified three genes involved in susceptibility to AN, particularly AN-R: the MAOA, SERT, and NET genes, with the MAOA gene serving as the ‘hinge-pin’ as we have provided evidence in support of preferential transmission of MAOA-L in separate analyses with the NET and SERT genes but detected no interaction between the NET and SERT genes (at least involving the NET gene NETpPR polymorphism and SERT gene 5-HTTLPR). We could not globally assess interaction as the MAOA-uVNTR restricted our sample size (ie only maternal MAOA-uVNTR allele transmissions were informative) resulting in empty cells. Neither could we globally assess joint genotype distributions due to low frequency of genotypes homozygous for the short form of the NETpPR polymorphism which again resulted in empty cells. However, pairwise comparisons (Table 3, and data not shown) provide evidence to support independent distribution of NETpPR, MAOA-uVNTR, and 5-HTTLPR genotypes (all Fisher–Freeman–Halton exact test P-values>0.120 for parents and children), which suggests that nonrandom transmission of alleles in our sample does not reflect nonrandom distribution of genotypes independent of disease status.

Our findings may help elucidate abnormalities in long-term weight-recovered AN individuals whose biology likely approximates that of the premorbid state before starvation confounds observations. Preferential transmission of MAOA-L with consequent increased activity of MAOA can explain elevated CSF 5-hydroxyindoleacetic acid (5-HIAA) levels in recovered individuals23 as MAOA catalyses the breakdown of 5-HT to 5-HIAA. Preferential transmission of 5-HTTLPR-S/S can explain reduced binding of 5-HT at the 5-HT2a receptor in the mesial temporal cortex of persons who have recovered from AN-R.24 5-HTTLPR-S/S leads to reduced expression and reuptake of 5-HT by the SERT so that intrasynaptic 5-HT increases.7 5-HT2a receptors should downregulate in response to increased 5-HT leading to reduced binding at this receptor. Further evidence for increased intrasynaptic 5-HT is the reduced anxiety experienced by recovered AN patients after depleting the precursor of 5-HT, tryptophan.25 As the authors rightly state,25 tryptophan depletion decreases the blood tryptophan:tyrosine ratio thereby diminishing transport of tryptophan across the blood–brain barrier and consequently 5-HT synthesis. However, they do not state that tryptophan depletion also increases transport of tyrosine across the blood–brain barrier and enhances CSF levels of NE known to be low in individuals after recovery from AN.26 Preferential transmission of MAOA-L can explain low CSF NE in AN patients after recovery as MAOA breaks down NE. Catechol-o-methyltransferase (COMT) also breaks down NE and could contribute to low CSF NE in patients after recovery. However, only one of two studies involving AN and the Val158Met polymorphism in the COMT gene showed positive association with the allele that encodes a more active COMT variant.27 The other study found no association with either allele.28

Our findings may be important for treatment. As 5-HT appears high (shown by high CSF 5-HIAA) relative to NE in persons after recovery from AN, we propose that treatment be focused on balancing levels of 5-HT and NE in the brain of these individuals. If our cited and current findings are independently replicated, we plan a clinical trial in which L-tyrosine is administered to increase and decrease CSF tyrosine and tryptophan, respectively, through competition across the blood–brain barrier. CSF NE should rise and 5-HT should fall (relative to each other) due to increased and decreased synthesis, respectively. After a short ‘tyrosine-free’ period, drugs that increase intrasynaptic NE will be given. Additionally, we intend matching genotype with treatment response to try to predict drug efficacy in AN.