Multivariate permutation analysis associates multiple polymorphisms with subphenotypes of major depression

Abstract

Unipolar major depressive disorder (MDD) is a prevalent, disabling condition with multiple genetic and environmental factors impacting disease risk. The diagnosis of MDD relies on a cumulative measure derived from multiple trait dimensions and alone is limited in elucidating MDD genetic determinants. We and others have proposed that MDD may be better dissected using paradigms that assess how specific genes associate with component features of MDD. This within-disease design requires both a well-phenotyped cohort and a robust statistical approach that retains power with multiple tests of genetic association. In the present study, common polymorphic variants of genes related to central monoaminergic and cholinergic pathways that previous studies align with functional change in vitro or depression associations in vivo were genotyped in 110 individuals with unipolar MDD. Subphenotypic characteristics were examined using responses to individual items assessed with the Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders (DSM IV), the 17-item Hamilton Rating Scale for Depression (HAM-D) and the NEO Five Factor Inventory. Multivariate Permutation Testing (MPT) was used to infer genotype—phenotype relationships underlying dimensional findings within clinical categories. MPT analyses show significant associations of the norepinephrine transporter (NET, SLC6A2) -182 T/C (rs2242446) with recurrent depression [odds ratio, OR 5 4.15 (1.91–9.02)], NET -3081 A/T (rs28386840) with increase in appetite [OR 5 3.58 (1.53–8.39)] and the presynaptic choline transporter (CHT, SLC5A7) Ile89Val (rs1013940) with HAM-D-17 total score {i.e. overall depression severity [OR 5 2.74 (1.05–7.18)]}. These relationships illustrate an approach to the elucidation of gene influences on trait components of MDD and with replication, may help identify MDD sub-populations that can benefit from more targeted pharmacotherapy.

Introduction

Depression is one of the most common and disabling conditions worldwide (Murray & Lopez ^{1996, 1997}); yet, the mechanisms involved in its pathogenesis remain elusive. More than three decades of research suggests that the vulnerability to unipolar major depressive disorder (MDD) is significantly influenced by genetic factors (Balciuniene et al. 1998; Baron et al. 1990; van den Bree & Owen 2003; Goldin & Gershon 1983; Maier et al. 2003; Merikangas et al. 2002; Zubenko et al. 2002). Many studies have examined the relationships between a wide variety of genetic polymorphisms and depression (Malhi et al. 2000; Sullivan et al. 2000), for the most part treating depression as a unitary disorder. However, MDD is phenotypically rich, with widely varying symptom constructs including disturbances in sleep, appetite and motor function as well as differences in the course of depressive illness (e.g. recurrence and duration). Furthermore, positive familial correlation of depressive symptom clusters has been observed (Korszun et al. 2004), suggesting a tighter genotype/phenotype correlation for subphenotypic traits as opposed to a categorical MDD diagnosis.

Major depressive disorder recurrence, number of depressive episodes, suicidal ideation, severity of illness and abnormalities of sleep, appetite and weight changes also show familial association in twin studies of MDD (Kendler et al. ^{1992, 1999}). Moreover, specific gene variants associate with clinical features such as suicide attempts and seasonality (Arango et al. 2003; Arias et al. 2001; Joiner et al. 2003; Lin & Tsai 2004; Willeit et al. 2003) and associate with antidepressant response (Hahn & Blakely 2007). These results indicate that individual symptoms and clusters have significant familial associations that may be tracked to specific gene variation.

In this study, we examine 13 single nucleotide variants (SNPs) and variable nucleotide tandem repeats (VNTRs) in 110 MDD subjects for whom subphenotype information was gathered using results on individual items from structured clinical interviews to assess mood, cognition, psychopathology and personality. Genes targeted are derived from a large number of potential risk loci because of their critical position within monoaminergic signaling pathways where current psychopathological models attribute dysfunction and therapeutic drug action and because they are posited to be involved in fundamental aspects of affective and cognitive function that is disrupted in MDD. Importantly, the variants selected either have a reported functional impact at the gene, protein or signaling level and/or have been previously associated with MDD or its therapy. We discuss our findings in the context of how genetic variation can impact subphenotypes based on an understanding of how the gene participates in sustaining basic physiological mechanisms and behavior.

Materials and methods

Patients

Research was approved by the Vanderbilt University Institutional Review Board, which complies with the Code of Ethics of the World Medical Association (Declaration of Helsinki). Written informed consent was obtained from all participants prior to any research procedures. Subjects were drawn from the Adult Psychiatry Outpatient Clinic of the Department of Psychiatry, Vanderbilt University Medical Center. Participants were all physically healthy. Excluded Axis I and II conditions included any history of schizophrenia, bipolar disorder or current psychotic disorder; borderline, schizotypal or antisocial personality disorder or history of substance abuse or dependence within 6 months prior to the assessment.

Clinical measures

Patients were evaluated using the Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders (DSM IV) (First et al. 1996) and the Structured Clinical Interview for Axis II Personality Disorders (SCID) (First et al. 1997). Trained and experienced research staff conducted all interviews prior to treatment. In addition, a senior research diagnostician conducted a second, confirmatory interview. All Axis I and II disorders were assigned using DSM IV diagnostic criteria (American Psychiatric Association 1994). Clinical history (e.g. history of hospitalization, substance abuse, suicide attempts and depression subtype) was collected as part of the SCID. All participants were further assessed using the 17-item Hamilton Rating Scale for Depression (HAM-D) (Hamilton 1960). The HAM-D rates 17 depressive symptoms on Likert-type subscales (typically rating 0–2 or 0–4). Ratings are anchored to specific clinical descriptors, and reliability is routinely assessed at our site using standardized videotapes. This scale is widely used and is considered a valid and reliable instrument (Williams 1988). The version of the HAM-D used here incorporated items for reversed vegetative symptoms (hypersomnia, hyperphagia and weight gain) (Thase et al. 1991). Subjects were also assessed using the NEO Five Factor Inventory (Costa & McCrae 1995), a well-validated assessment of temperament. The NEO has been used extensively in the investigation of genomic contributors to both personality and depression vulnerability (Jorm et al. 2000). In particular, measures of neuroticism and extraversion have been consistently shown to represent vulnerability factors for depression. A complete list of clinical functioning measures is included in Table 1. Patient characteristic variables measured included age at intake, gender, race and ethnicity.

Table 1.

Clinical functioning variables

Variable	Type
History of suicide attempt (0 vs. ≥1)	Categorical
HAM-D-17 total score, intake	Continuous
Age at MDD onset	Continuous
Single vs. recurrent depression	Categorical
Chronic depression	Categorical
Dysthymia*	Categorical
Any comorbid anxiety disorder*	Categorical
Panic disorder*	Categorical
Posttraumatic stress disorder	Categorical
Social phobia*	Categorical
Generalized anxiety disorder†	Categorical
Specific phobia*	Categorical
Obsessive—compulsive disorder*	Categorical
Alcohol abuse or dependence*	Categorical
Drug abuse or dependence*	Categorical
HAM-D Item 1: depressed mood	Interval
HAM-D Item 2: guilt	Interval
HAM-D Item 3: suicide	Interval
HAM-D Item 4a: early insomnia	Interval
HAM-D Item 5a: middle insomnia	Interval
HAM-D Item 6a: late insomnia	Interval
HAM-D Item 4b: hypersomnia	Interval
HAM-D Item 5b: extra sleep hours	Interval
HAM-D Item 6b: daytime napping	Interval
HAM-D Item 7: work and activities (anhedonia)	Interval
HAM-D Item 8: motor retardation	Interval
HAM-D Item 9: agitation	Interval
HAM-D Item 10: psychic anxiety	Interval
HAM-D Item 11: somatic anxiety	Interval
HAM-D Item 12a: decreased appetite	Interval
HAM-D Item 12b: increased appetite	Interval
HAM-D Item 13: low energy	Interval
HAM-D Item 14: low libido	Interval
HAM-D Item 15: hypochondriasis	Interval
HAM-D Item 16a: weight loss	Interval
HAM-D Item 16b: weight gain	Interval
HAM-D Item 17: lack of insight†	Interval
HAM-D 17 total score, week 8	Continuous
Response or remission‡ at week 8	Ordinal
HAM-D score, week 16†	Continuous
Response or remission‡ at week 16†	Ordinal
Weight loss during this episode?	Categorical
Weight gain during this episode?	Categorical
NEO neuroticism score	Continuous
NEO extroversion score	Continuous
NEO openness score	Continuous
NEO agreeableness score	Continuous
NEO conscientiousness score	Continuous

^*Current or past history.

^†Not included in the analysis because of low frequency or missing data.

^‡Response or remission: intake vs. end HAM-D; response is ≥50% reduction and remission is HAM-D score <8.

Genetic analysis

A 10 ml blood sample was obtained from all participants for DNA extraction under approved Vanderbilt University Institutional Review Board protocol. The DNA was extracted by the Center for Human Genetics Research DNA Resources Core at Vanderbilt University Medical Center using the Puregene DNA isolation kit according to the manufacturer’s instructions (Gentra Systems Inc., Minneapolis, MN, USA). DNA was initially genotyped for 20 polymorphic variants including SNPs in the norepinephrine and choline transporters (NET and CHT, respectively), the serotonin 1A (5-HT_1A), 2A (5-HT_2A) and 2C (5-HT_2C) receptors, and VNTRs in monoamine oxidase A (MAO-A) and the serotonin and dopamine transporters (5-HTT and DAT, respectively) (Tables 2 and 3). All genotyping was performed blind to phenotype. All SNPs were subjected to one of the following quality control measures: SNP genotyping was repeated on a subset or the entire sample by the same or different method and/or a subset of the sample was subjected to dideoxy sequencing following other genotyping methods. Subsequently, DNAs confirmed by multiple methods were used in assays as sentinels for assay performance. Polymerase chain reaction (PCR) and genotyping conditions are provided in data S1. Primer sequences used in PCR and genotyping are available on request. Of the 20 genotyped polymorphic variants, 13 had sufficient minor allele frequencies to be included in the analyses (Tables 2 and 3).

Table 2.

Polymorphic variants genotyped

Gene symbol	Gene name	Polymorphic variant	dbSNP ID
SLC5A7	Choline transporter, CHT	CHT Ile89Val	rs1013940
SLC6A3	Dopamine transporter, DAT	DAT 3′ VNTR	rs28363170
MAOA	Monoamine oxidase A, MAO-A	MAO-A VNTR
SLC6A2	Norepinephrine transporter, NET	NET -3081 A/T	rs28386840
		NET -182 T/C	rs2242446
		NET 155 G/A	rs5569
SLC6A4	Serotonin transporter, 5-HTT	5-HTTLPR	rs4795541
HTR1A	Serotonin 1A receptor, 5-HT1A	5-HT1A -1019 C/G	rs6295
HTR2A	Serotonin 2A receptor, 5-HT2A	5-HT2A -1438 G/A	rs6311
		5-HT2A His452Tyr	rs6314
HTR2C	Serotonin 2C receptor 5-HT2C	5-HT2C -759 C/T	rs3813929
		5-HT2C -697 G/C	rs518147
		5-HT2C Cys23Ser	rs6318

Table 3.

Genotype and allele frequencies of polymorphisms

Polymorphism	Genotype	Frequency	Allele	Frequency	HWE P value
CHT Ile89Val	Ile/Ile	88 (0.80)	Ile	193 (0.88)	0.012
	Ile/Val	17 (0.16)	Val	27 (0.12)
	Val/Val	5 (0.05)
DAT 3′ VNTR	10/10	63 (0.57)	10	165 (0.75)	0.849
	10/9	39 (0.36)	9	55 (0.25)
	9/9	8 (0.06)
MAO-A VNTR	YL	28 (0.26)	L	98 (0.55)	0.820
	YS	15 (0.14)	S	79 (0.45)
	LL	17 (0.16)
	LS	36 (0.32)
	SS	14 (0.13)
NET -3081 A/T	AA	48 (0.44)	A	141 (0.64)	0.506
	AT	45 (0.41)	T	79 (0.36)
	TT	17 (0.15)
NET 155 G/A	GG	51 (0.46)	G	154 (0.70)	0.420
	GA	52 (0.47)	A	66 (0.30)
	AA	7 (0.06)
NET -182 T/C	TT	56 (0.51)	T	152 (0.69)	0.296
	TC	40 (0.36)	C	68 (0.31)
	CC	14 (0.13)
5-HTTLPR	ll	37 (0.34)	l	128 (0.58)	0.995
	ls	54 (0.49)	s	92 (0.42)
	ss	19 (0.17)
5-HT1A  -1019 C/G	CC	35 (0.32)	C	120 (0.55)	0.682
	CG	50 (0.46)	G	100 (0.45)
	GG	25 (0.23)
5-HT2A  -1438 G/A	GG	36 (0.33)	G	130 (0.59)	0.636
	GA	58 (0.53)	A	90 (0.41)
	AA	16 (0.15)
5-HT2A  His452Tyr	His/His	86 (0.78)	His	193 (0.88)	0.492
	His/Tyr	21 (0.19)	Tyr	27 (0.12)
	Tyr/Tyr	3 (0.03)
5-HT2C  -759 C/T	YC	35 (0.32)	G	150 (0.85)	0.400
	YT	8 (0.07)	T	27 (0.15)
	CC	48 (0.44)
	CT	19 (0.17)
	TT	0 (0.0)
5-HT2C  -697 G/C	YG	28 (0.27)	G	117 (0.71)	0.637
	YC	14 (0.14)	C	47 (0.29)
	GG	31 (0.30)
	GC	27 (0.26)
	CC	3 (0.3)
5-HT2C  Cys23Ser	YCys	35 (0.32)	Cys	155 (0.87)	0.587
	YSer	8 (0.07)	Ser	23 (0.13)
	CysCys	52 (0.47)
	CysSer	15 (0.14)
	SerSer	0 (0.0)

HWE, Harly-Weinberg Equilibrium.

Data analyses

Patient characteristics

Chi-squared tests were used to test for polymorphic variant differences in categorical patient characteristic variables. Independent t-tests were used for continuous patient characteristic variables. Significant findings were followed by within-group analyses. An alpha level of 0.05 was used for each analysis.

Multivariate permutation test analysis of clinical variables

The multivariate permutation test (MPT) was used to estimate statistical significance of association of genotype with clinical variables. Conducting multiple significance tests incrementally increases the family wise error (FWE) rate across all tests, with Type I errors contributing to well-known replication failures in psychiatric genetic reports. Bonferroni methods, commonly used to control FWE, are too conservative because independence of the outcome variables is assumed. The conservative correction of Type I error reduces statistical power, especially when the number of tests is large. In contrast, MPT methods do not assume independence but instead use the correlations among the outcome variables, resulting in increased statistical power. Multivariate permutation test methods (Westfall & Young 1993) have several other advantages, e.g. the data do not have to be normally distributed and the number of variables can exceed the number of subjects. Simulation studies have confirmed that statistical power is greater for MPT compared with Bonferroni methods, while controlling FWE rates to the specified alpha level (Blair et al. 1994; Troendle 1995; Yoder et al. 2004). As with most methods, there are also some limitations. Data must meet the assumption of exchangeability that under the null hypothesis, all possible permutations must be possible and valid. Power advantages over Bonferroni methods are lower when correlations among dependent variables are small. Finally, one recent report suggests that Type I error may be inflated if the distributional shapes of the groups are not the same (Huang et al. 2006). A number of issues may also contribute to Type II errors in our study, causing us to miss true associations. These include our relatively small sample size and the restricted range of some subphenotypic variables. For example, items on rating instruments such as the HAM-D-17 may not be sensitive enough to discriminate more subtle differences between patients. Nonetheless, MPT methods provide a flexible and powerful method for our efforts as they permit a large number of tests, while controlling FWE rates.

Statistical significance in MPT is determined by comparing the observed statistic to an empirical distribution of the test statistic, instead of the standard distribution typically used. The empirical distribution represents the probability of observing each statistical value under the null hypothesis of no genotype differences. The permutation is practically performed by randomly sorting genotype across subjects, while the phenotype variables remain unchanged. This randomization procedure effectively removes any natural association between genotype and phenotype, thus providing an estimate of the null hypothesis. With each permutation, the test statistic is recomputed for the phenotypes. The test statistics across all permutations form the empirical distribution, against which the observed test statistic is compared to determine the P value. A step-down procedure for testing each phenotype controls FWE for the entire set of phenotypes. Significance tests begin with the largest group difference. Following each significant result, all data for that phenotype are removed, and the empirical distribution is regenerated using the remaining data. Thus, the MPT procedure produces P values for each phenotype.

The SAS statistical software package (SAS Institute Inc., Cary, NC, USA) was used for all analyses with α = 0.05. PROC MULTTEST was used to perform stepwise MPTs of mean differences using a trend contrast with 20 000 permutations. Each genotype defined as a level in the trend analysis. For genes located on the X chromosome (HTR2C and MAO-A), hemizygous males were grouped with either major or minor homozygotes. For genes with homozygous minor allele genotype frequencies of less than five subjects, these genotypes were grouped with heterozygotes.

Odds ratio (OR) with confidence intervals were computed as a measure of effect size for all significant effects. For continuous variables, a median split was used to create the dichotomous variable used for the OR.

The distributions of major and minor alleles across clinical levels are also presented for descriptive purposes. For the allele frequencies, continuous variables were transformed into categorical variables using quartiles (0 = lowest 25%, 1 = middle 50% and 2 = highest 25%).

Results

A cohort of 110 patients who presented with MDD participated in this study. The mean age of the sample was 42.7 years (SD = 10.7). Gender distribution was 61% female and 39% male. Race was reported as 87% white, 6.5% African-American, 4.6% Asian, 1% Hispanic and 1% other. Subtype of depression was 18% atypical, 35% melancholic (or both melancholic and atypical) and 47% no subtype. Racial differences were observed with respect to the genotype distributions of two polymorphisms. Race was not evenly distributed across the DAT 3′ VNTR groups [χ²(2) = 13.53, P = 0.002]. Among white subjects, the distribution of DAT 3′ VNTR genotypes was 51% homozygous major, 41% heterozygous and 8% homozygous minor, whereas for non-white subjects, we observed 86% homozygous major and 14% homozygous minor with no heterozygotes. This finding is consistent with reported differences in DAT 3′ VNTR allele frequencies across ethnicities, in particular, a lower frequency of the 9-repeat allele in African-Americans (Hahn & Blakely 2002). There were also racial differences in the distribution of NET -3081 A/T [χ²(2) = 7.91, P = 0.02]. For white subjects, the distribution of genotypes was 47% homozygous major, 41% heterozygous and 12% homozygous minor, whereas for non-white subjects, the distribution was 21% homozygous major, 36% heterozygous and 43% homozygous minor.

Distribution differences were also observed for the DAT 3′ VNTR [χ²(4) = 11.29, P < 0.03] in relation to depression subtype. For subjects with melancholia, the distribution was 55% homozygous major and 45% heterozygous with no homozygous minor subjects. In contrast, subjects with atypical depression were distributed as 63% homozygous major, 21% heterozygous and 16% homozygous minor. For subjects who had neither melancholic nor atypical depression, the distribution was 52% homozygous major, 34% heterozygous and 14% homozygous minor. Finally, age differed by genotype for the 5-HTTLPR [F(1,108) = 4.55, P = 0.04] with subjects who were homozygous major (ll) being older than subjects with a minor (s) allele (mean = 45.76, SD = 10.68 and mean = 41.23, SD = 10.41, respectively).

Dose—response relationships between clinical variables and polymorphic variants were analyzed using MPTs. There were three significant findings: NET -182 T/C and recurrence of depression; NET -3081 A/T and increase in appetite and CHT Ile89Val and severity of depression as measured by the HAM-D-17 total score at intake (Table 4, in bold). Additionally, distributions of major and minor alleles across clinical functioning levels are presented to provide a comprehensive picture of the observed relationship between polymorphic variants and clinical function outcomes (Table S1).

Table 4.

Summary of significant dose—response relationships between polymorphic variants and clinical functioning variables

Gene	Polymorphic variant	Clinical functioning variable	Uncorrected P value	Exact P value	Dose response to minor allele
ChT	CHT Ile89Val	HAM-D-17 total score, intake	0.00	0.04	Increase
		HAM-D Item 11: somatic anxiety	0.01	0.30	Increase
		HAM-D Item 8: motor retardation	0.01	0.40	Increase
DAT	DAT 3′ VNTR	HAM-D Item 4a: early insomnia	0.03	0.70	Decrease
		HAM-D Item 4b: hypersomnia	0.01	0.39	Increase
		HAM-D Item 5b: extra sleep hours	0.01	0.22	Increase
		Weight gain during this episode?	0.02	0.58	Decrease
MAO-A	MAO-A VNTR	Dysthymia	0.03	0.73	Increase
		HAM-D Item 5b: extra sleep hours	0.01	0.33	Increase
NET	NET -3081 A/T	HAM-D Item 12b: increased appetite	0.00	0.01	Increase
		Dysthymia	0.05	0.80	Decrease
		HAM-D Item 12a: decreased appetite	0.00	0.09	Decrease
		HAM-D Item 15: hypochondriasis	0.02	0.42	Increase
		Conscientiousness	0.02	0.49	Increase
		Specific phobia	0.01	0.27	Increase
NET	NET -182 T/C	HAM-D Item 12b: increased appetite	0.01	0.39	Increase
		Single vs. recurrent	0.00	0.02	Decrease
		Specific phobia	0.01	0.56	Increase
NET	NET 155 G/A	Chronic depression	0.04	0.75	Increase
5-HT1A receptor	5-HT1A -1019 C/G	HAM-D Item 16b: weight gain	0.05	0.85	Decrease
		HAM-D Item 4b: hypersomnia	0.01	0.25	Decrease
		HAM-D Item 5b: extra sleep hours	0.00	0.11	Decrease
5-HT2A receptor	5-HT2A -1438 G/A	HAM-D Item 12b: increased appetite	0.01	0.44	Increase
5-HT2A receptor	5-HT2A His452Tyr	Weight gain during this episode?	0.03	0.67	Increase
		NEO neuroticism score	0.01	0.25	Decrease
5-HT2C receptor	5-HT2C -697 G/C	HAM-D Item 4a: early insomnia	0.03	0.67	Decrease
		HAM-D 11: somatic anxiety	0.01	0.35	Increase
		Drug abuse or dependence	0.01	0.43	Increase
5-HT2C receptor	5-HT2C -759 C/T	NEO neuroticism score	0.02	0.62	Increase
		Weight loss during this episode?	0.04	0.80	Increase
5-HT2C receptor	5-HT2C Cys23Ser	HAM-D Item 11: somatic anxiety	0.02	0.47	Increase
		HAM-D Item 4a: early insomnia	0.01	0.20	Decrease
		HAM-D Item 5a: middle insomnia	0.04	0.74	Decrease
5-HTT	5-HTTLPR	HAM-D Item 14: low libido	0.03	0.66	Decrease
		HAM-D Item 5a: middle insomnia	0.02	0.47	Decrease
		NEO extroversion score	0.03	0.63	Decrease
5-HTT	5-HTTLPR adjusted	HAM-D Item 2: guilt	0.04	0.80	Increase

Two of the significant outcomes of genotype association with clinical variables were found for the NET gene. First, the polymorphic variant NET -182 T/C was related to whether depression was a single episode or recurrent. There was a dose relationship between the number of T-containing genotypes and the presence of recurrent depression [P = 0.02; OR = 4.15 (1.91–9.02); Table 4]. Of the subjects who were homozygous major, 95% had recurrent depression compared with 82.5% of the heterozygous group and 57% of the homozygous minor group. This effect is also apparent in the distribution of alleles in which those patients without recurrent depression comprised 41% T and 59% C alleles, whereas the presence of recurrent depression coincided with 74% T and 26% C alleles (Table S1).

A second significant finding was that NET -3081 A/T was associated with increased appetite [P = 0.01; OR = 3.58 (1.53–8.39); Table 4]. This result was supported by an allele distribution in which patients without effects on appetite comprised 72% A and 28% T alleles, whereas those patients reporting obvious increased appetite were enriched for T alleles, 38% A vs. 62% T (Table S1).

We addressed the issue of association of both NET -3081 A/T and DAT 3′ VNTR with ethnicity. We performed an exploratory analysis on the white (the majority of our subjects) vs. non-white subgroups. For NET -3081, the original MPT finding of an association with NET -3081 and increased appetite was preserved for whites. Of the findings with significant uncorrected P values, one remained in the racial subgroup analyses; the association with decreased appetite remained significant (P = 0.01). Several new findings emerged in the subgroup analysis at the uncorrected level; for whites, having recurrent depression, chronic depression, weight gain and a history of drug use were related to NET -3081. For the DAT 3′ VNTR, the original findings were relatively stable in the racial subgroup analyses; all the originally significant (uncorrected) clinical outcome variables remained significant for the white subgroup.

A third significant finding of the MPT analysis of the full sample set was that the CHT Ile89Val polymorphic variant was related to the HAM-D-17 total score at intake or overall depression severity [P = 0.04; OR = 2.74 (1.05–7.18); Table 4]. The intake HAM-D-17 scores for the homozygous major and heterozygous groups were lower (mean = 21.94, SD = 3.34 and mean = 22.07, SD = 3.39, respectively) than the mean score for the homozygous minor group (mean = 28.00, SD = 4.18).

Finally, we addressed the presence within the 5-HTTLPR region of an A to G SNP (Nakamura et al. 2000). Recently, the coincident presence of both the l and G alleles was reported to result in levels of transcription similar to those generated by the s allele (Hu et al. 2006). To explore the possibility that l-G and s alleles could be functionally grouped, an adjusted 5-HTTLPR variable was created with s and l-G considered as a single allele. In the analysis with the original or adjusted variables, no significant associations were identified (Table S2).

Discussion

The present study examined 110 MDD subjects for association of polymorphisms in several candidate genes with subphenotypes consisting of dimensional variables related to symptom measures of MDD. Several significant differences arose in the initial analysis of demographic variables and genotype. Because our sample set contains mostly white subjects, but also contains subjects of other ethnicities, we employed uncorrected P values in testing racial distribution to maximize the likelihood that we would observe differences. Our intention was to err on the side of identifying any stratification that might affect our results. NET -3081 A/T was not evenly distributed across race, as we have previously noted (Kim et al. 2006). The DAT VNTR was also not equally distributed across race. Other demographic differences did not affect our major findings but show potentially intriguing relationships. For example, 5-HTTLPR genotype was unevenly distributed with respect to age at intake, with scontaining genotypes being younger than homozygous majors. Our finding is consistent with the s allele as a vulnerability factor for depression, possibly interacting with other factors to precipitate depression at an earlier age (Caspi et al. 2003). Finally, the DAT 3′ VNTR was unequally distributed across depression subtype, with homozygous majors (10/10) over-represented in the atypical depression subgroup. This is the first study to our knowledge that suggests a relationship between DAT 3′ VNTR and a particular subtype of depression and as such will be important to address in replication studies.

Assessment of a large number of depression subphenotypes poses challenges to conventional statistical association tests. The present study utilized an experimental design and MPT analytic methods that lend stringency and power to the examination of multiple SNPs and clinical variables (Blair et al. 1994; Troendle 1995; Yoder et al. 2004). Three highly significant associations were identified in the MPT analysis, two of which were for SNPs in the NET gene. First, NET -182 T/C was highly significantly associated with recurrent depression. This SNP has shown positive, negative or no association with depression and antidepressant response in previous reports (Inoue et al. 2004; Ryu et al. 2004; Yoshida et al. 2004; Zill et al. 2002). Earlier onset and recurrence of depression across the life span have been shown to have significant familiality (Levinson et al. 2003), and these data suggest that the NET -182 T/C SNP may contribute to this risk. The second highly significant association of a depression subphenotype with NET was that of NET -3081 A/T with increased appetite. NET -3081 A/T is located 3081-bp upstream of the translational start site and the presence of the minor allele, T, introduces a cis-acting element that binds the transcriptional repressor, Slug (Kim et al. 2006). This NET SNP may alter noradrenergic signaling to influence food consumption indirectly through sympathetic nervous system control of energy metabolism or, alternatively, through brainstem structures that link systemic glucose homeostasis to appetite (Ritter et al. 2006; Thomas & Palmiter 1997).

Strikingly, besides the association of NET -182 T/C with recurrent depression, additional relationships between NET and depression chronicity were observed. These effects were not significant with correction but may suggest trends for exploration in future research. Importantly, NET -182 T/C also showed a strong trend for association (P = 0.05 without correction) with chronic depression. The minor allele of 155 G/A was also associated with chronic depression. The synonymous SNP 155 G/A has met to date with mixed or negative results in studies of depression (Leszczynska-Rodziewicz et al. 2002; Owen et al. 1999; Yoshida et al. 2004). Thus, studies of NET 155 G/A may well benefit from replication efforts targeted to depression chronicity. Finally, in our analysis of NET -3081 A/T by ethnicity, both recurrent and chronic depression were associated with this SNP in the white subgroup. Some aspects of chronic depression suggest that it may represent a different phenotype than those of non-chronic depression; e.g. chronic depression is associated with an early age of onset (Gelenberg et al. 2006) and it may show significant familial association (Kendler et al. 1997; Klein et al. 1999). Taken together, both recurrent and chronic depression appear to be impacted by NET gene variation, and future studies implementing hypotheses to test these relationships may afford a path to more tailored pharmacotherapies, such as with NET-selective reuptake inhibitors.

The third highly significant association was that of CHT Ile89Val with depression severity. Although receiving less focus than monoaminergic systems, cholinergic signaling is clearly implicated in mood and mood disorders (Janowsky et al. 1983). The CHT Ile89Val variant has been reported to decrease transport activity (Okuda et al. 2002; B. English and R. D. Blakely, unpublished observations). Heterozygous CHT knockout mice exhibit reduced ACh and muscarinic receptor levels across multiple brain regions and show altered responses to muscarinic agents (Bazalakova et al. 2007). A hypofunctional CHT is thus expected to diminish central nervous system cholinergic tone and alter receptor sensitivity in humans and impact risk for features of MDD. Interestingly, the frequency of the Val allele was 12%, whereas other data indicate the frequency of this SNP to be 6% (Okuda et al. and English et al., unpublished observations). As our report is the first study to examine the Ile89Val SNP in MDD and we conducted a within-disease analysis, these findings will be important to replicate in other MDD samples and to compare in a case—control design. That this SNP associates with HAM-D-17 total score at intake, or overall depression severity, but not with a specific subphenotype supports an impact of altered cholinergic function on multiple traits within MDD. Indeed, other studies of genetic association and trials of antidepressant efficacy provide evidence of a role for cholinergic systems in major depression (Furey & Drevets 2006; Wang et al. 2004).

The significant findings from this study all had large effects, with allelic variation accounting for 11%–15% of the variability in clinical outcomes. This study was underpowered to find small- or medium-size effects. Power to detect effect sizes consistent with the uncorrected findings in Table 4 ranged from 5% to 32%. Sample sizes larger than 300 would have been necessary for these findings to survive correction for multiple tests.

Our assessment of genetic variation in MDD bears similarity to methods that seek to model the components of depression in animal models. These models often rely on analyses of gene and drug influences on features of behavior, such as shock avoidance or forced swim, that are reversed by antidepressants. Genetic manipulations that impact traits with parallels in human MDD, in essence, a subphenotype approach, drives current theories. For example, 5-HT2C receptor-null mice show sleep disruption (Frank et al. 2002) and 5-HTT knockout mice display altered dendritic morphology in stress-related brain regions (Wellman et al. 2007). Heterozygous knockout models are also useful in linking gene expression status to human hypofunction in specific signaling domains (Kalueff et al. 2007) and are important resources for modeling further specific trait attributes influenced by candidate genes that can be further explored in MDD studies.

Although depression encompasses genomic and phenotypic complexities, this study suggests that it may be possible to dissect genotype—subphenotype associations using sufficiently well-phenotyped populations, adequate sample size and improved methods of data analysis. The ability to enrich study populations by specific clinical characteristics may provide a way to improve both the detection of significant associations and their replication. Future studies with larger sample sizes should also be able to examine the impact of gene—gene interactions on the traits of depression. As these determinants are defined, they may offer additional tools to help devise more robust treatment strategies.