In this study, we sought out to test the hypothesis that genetic factors may influence antidepressant response to fluoxetine. The investigation focused on seven candidate genes in the serotonergic pathway involved in the synthesis, transport, recognition, and degradation of serotonin. Our clinical sample consisted of 96 subjects with unipolar major depression treated with fluoxetine with response variables assessed after a 12-week trial. Patient data were also collected to investigate the pattern of drug response. Using a high-throughput single-nucleotide polymorphism (SNP) genotyping platform and capillary electrophoresis, we genotyped patients at 110 SNPs and four repeat polymorphisms located in seven candidate genes (HTR1A, HTR2A, HTR2C, MAOA, SLC6A4, TPH1, and TPH2). Statistical tests performed included single-locus and haplotype association tests, and linkage disequilibrium (LD) estimation. Little evidence of population stratification was observed in the sample with 20 random SNPs using a genomic control procedure. Our most intriguing result involved three SNPs in the TPH1 gene and one SNP in the SLC6A4 gene, which show significant single-locus association when response to fluoxetine is compared to nonresponse (P=0.02–0.04). All odds ratios indicated an increased risk of not responding to fluoxetine. In the specific response vs nonspecific and nonresponse comparison, three SNPs in the TPH2 gene (P=0.02–0.04) were positively associated and one SNP in the HTR2A gene (P=0.02) was negatively associated. When comparing specific response to nonspecific response, we found significant negative associations in three SNPs in the HTR2A gene (P=0.001–0.03) and two SNPs in the MAOA gene (P=0.03–0.05). We observed variable, although strong LD, in each gene and unexpectedly low numbers of estimated haplotypes, formed from tagged SNPs. Significant haplotype associations were found in all but the HTR1A and HTR2C genes. Although these data should be interpreted cautiously due to the small sample size, these results implicate TPH1 and SLC6A4 in general response, and HTR2A, TPH2, and MAOA in the specificity of response to fluoxetine. Intriguingly, we observe that a number of the less frequent alleles of many of the SNP markers were associated with the nonresponse and nonspecific phenotypes.
Major depression has one of the highest lifetime incidence rates among psychiatric disorders.1, 2 Thus, antidepressant medications are among the most commonly prescribed pharmacological agents. However, despite recent advances in antidepressant pharmacotherapy, patient response rates can still be as low as 60% for the first drug administered.3 Interindividual variation in antidepressant efficacy is thought to be at least partly under a genetic control.4, 5, 6 Selective serotonin reuptake inhibitors (SSRIs), commonly prescribed medications, are believed to exert their antidepressant effect through inhibiting the serotonin transporter, which terminates serotonin (5-HT) transmission. A number of studies have tested the association between DNA variations in the serotonin transporter and response to various SSRIs.4 The majority of these studies rely on assaying a common insertion/deletion polymorphism in the SLC6A4 promoter that alters in vitro transcription of the gene.7 This 44 base pair insertion/deletion primarily comes in long (‘l’) and short (‘s’) forms, with the long form being more transcriptionally active. In Caucasian subjects taking SSRIs, the presence of the long allele has been associated with response in multiple studies,8, 9, 10 while in a Korean population, the short allele has been associated with a more favorable response.11 In a second Korean population, the long allele was associated with a better response.12 The reason for this discrepancy has not been addressed, but is likely to involve the unique population histories of the SLC6A4 locus between the European/US subjects and the Asian subjects. Furthermore, even in positive studies, this polymorphism does not explain all of the variance seen in response, suggesting the involvement of other variants within this gene or other genes.
As serotonin appears to play an important role in depression, we have chosen to pursue an analysis of a group of genes in the serotonin pathway. These genes are involved with the synthesis, signal transduction, transport, and catabolism of serotonin. Tryptophan hydroxylase (TPH1) catalyzes the rate-limiting step in the biosynthesis of 5-HT. Recently, Walther et al13 described a second isoform of TPH2, which was shown to be the major form expressed in the brain, with TPH1 being expressed mainly in the periphery. Serotonin receptors 1A (HTR1A), 2A (HTR2A), and 2C (HTR2C) are involved in the neurotransmission of 5-HT. The serotonin transporter (SLC6A4) is involved with clearing the synapse of 5-HT. Finally, monoamine oxidase A (MAOA) is one of the main enzymes involved in the degradation of 5-HT. Monoamine oxidase inhibitors have been used for decades to treat major depressive disorder. SSRIs have antagonistic properties at serotonin receptors, while the functional effect of SSRIs, increased synaptic 5-HT, has an indirect effect on HTR1A, HTR2A, and HTR2C function.14, 15, 16 SSRIs can modulate the expression of TPH mRNA, and inhibition of TPH activity has dramatic effects on brain serotonin levels.17, 18, 19 A number of studies have individually investigated the role of several of these genes in antidepressant response.20, 21, 22, 23
Given the small number of reports, as well as biological data implicating their role in serotonergic function, we investigated the association between treatment response to fluoxetine and a number of publicly available single-nucleotide polymorphisms (SNPs) in TPH1, TPH2, HTR1A, HTR2A, HTR2C, SLC6A4, and MAOA. We have used a well-phenotyped population of persons with unipolar major depression and carried out an analysis of association between these variants and response to fluoxetine. We attempted to limit phenotypic heterogeneity through the use of response pattern analysis that classifies patients as nonresponders, specific responders, or placebo responders to fluoxetine.24 We also utilized both single SNPs as well as haplotypes for detecting an association to SSRI response and a genomic control (GC) technique to correct for possible population stratification.
The study population consisted of 96 research subjects enrolled in an ongoing NIMH-funded protocol (PJM, principal investigator) to assess relapse following fluoxetine discontinuation in depressed subjects who had responded to fluoxetine. Inclusion in that clinical trial required subjects to be in a current episode of major depression, to be aged 18–65 years, and to give informed consent to be randomized to either fluoxetine continuation or to placebo substitution should they respond to acute treatment. There was no depression severity threshold for inclusion. The Structured Clinical Interview for DSM-IV Axis I Disorders—Patient Edition (SCID-I/P) was used to establish all psychiatric diagnoses.25 Subjects with any history of psychosis, mania, organic mental syndrome, a history of substance abuse or dependence active within the previous 6 months, with the exception of nicotine dependence, current bulimia nervosa, or unstable physical illnesses were excluded. Other Axis I comorbid disorders were not exclusionary. Medications known to cause or exacerbate depression, such as beta-blockers or corticosteroids, or to have significant antidepressant or anxiolytic properties, were exclusionary. Study subjects were included if they took occasional hyponotic medication of a nonbenzodiazepine type, oral contraceptives that were not temporally associated with onset or exacerbation of depression, or thyroid hormone replacement that was at a constant and effective dose for 3 months prior to the study. Concomitant medications such as diuretics and antihypertensives were permitted. Only data from the initial 12-week trial were used in these analyses as the number of subjects was considered to be inadequate to use follow-up data where subjects were randomized to fluoxetine or placebo and either maintained their response or relapsed.26
Patients were categorized as nonresponders, specific pattern responders, or placebo-pattern responders by pattern analysis after 12 weeks of open-label fluoxetine treatment.24 Fluoxetine daily dosage was 10 mg for 1 week, 20 mg daily for 3 weeks, 40 mg for 4 weeks, and 60 mg for the remaining 4weeks. Dosage increments were made only in those tolerating the medication well and insufficiently responsive to 40 mg. Response in any week was judged by the use of the Clinical Global Impression Improvement score, where a score of ‘much improved’ or ‘very much improved’ was required for response. This criterion was applied with the definition that ‘much improved’ characterized someone that the clinician believed was sufficiently improved that no change in treatment was warranted. This usually corresponded to a decrement of between 50 and 75% in baseline depression ratings on the Hamilton Depression Scale. Subjects responding at week 12, whose response began after the second week and were sustained until week 12, were considered ‘specific-pattern’ responders; subjects whose response began in weeks 1 or 2 but whose response was not sustained for all subsequent weeks until week 12 were considered ‘placebo-pattern’ or ‘nonspecific’ responders. In short, pattern analysis draws on the observation that a specific response to medication is associated with a delayed response to active medication that is persistent once achieved, while a nonspecific response has an earlier onset of response and/or lack of persistence in improvement after onset. This is based on the observation, replicated in several independent samples, that ‘placebo’ or ‘nonspecific’ patterns are found equally among patients on active drug and on placebo, while only specific patterns are significantly more common on active medication.26, 27 While pattern analysis clearly does not perfectly characterize placebo and active medication response, it is an attempt to deal with the problem that approximately half of all subjects (one-quarter in this study) who respond to medication treatment are having a placebo response.28 Informed consent was obtained from each research participant, as was IRB approval from the New York State Psychiatric Institute and the University of California at San Francisco. In this group, the average age was 37.1±11.6 years, and the male/female ratio was 47 to 49. There were 77 (80%) responders and 19 (20%) nonresponders to a 12-week trial of fluoxetine. The subjects were 78% Caucasian, 6% African-American, 7% Hispanic, 5% Asian, and 3% other. There were no significant differences in ethnicity (by exact test, P=0.07) or age (by t-test, P=0.19) between responders and nonresponders. Using a clinical pattern analysis paradigm (12), 20 of the 77 responders (26%) were determined to be ‘nonspecific’ responders to fluoxetine.
Four chimpanzee (Pan troglodytes) and one bonobo (Pan paniscus) genomic DNA samples were obtained from the Coriell Institute for use in determining the ancestral allele at candidate SNPs (Coriell Institute for Medical Research, Camden, NJ, USA).29
Patient genomic DNA was extracted from whole blood using a Puregene genomic DNA purification kit (Gentra Systems, Minneapolis, MN, USA). DNA was quantified using an ND-1000 spectrophotometer (NanoDrop Technologies, Rockland, DE, USA).
A total of 165 SNPs were identified from publicly available databases or the literature and were chosen with an unbiased approach in an effort to distribute them evenly across the seven candidate genes (see Supplementary Figure 1, web supporting information). SNPs in the 5′ and 3′ region flanking the gene were also included in this study. Of the 165, 110 were used for genotyping, as 32 were monomorphic in our population and an adequate SNP genotyping assay could not be developed for the final 23. Of the 110 SNPs used in this study, 66 were intronic, 10 were exonic (four nonsynonymous and six synonymous changes), and 34 were located in the 5′- and 3′-flanking regions. Fluorescence polarization detection of template-directed dye-terminator incorporation (FP-TDI) was used to genotype SNPs, as described elsewhere.30, 31 Briefly, the first step involves polymerase chain reactions (PCR) of 5 microliters (μl) containing 200 nM of the forward and reverse primers (Supplementary Table 1, web supporting information), 20 ng genomic DNA template, 50 μM dNTPs (Roche, Indianapolis, IN, USA), 1 M anhydrous betaine (Acros Organics, Geel, Belgium), 50 mM KCl, 20 mM Tris-HCl (pH 8.4), 2.5 mM MgCl2, and 0.25 U Platinum Taq DNA polymerase (Invitrogen, Carlsbad, CA, USA). All primers and TDI probes were designed using Primer3 software and were manufactured by Invitrogen (Carlsbad, CA, USA).32 Samples were cycled using a touchdown protocol at 94°C for 3 min, followed by seven cycles of 94°C for 30 s, 65–59°C for 30 s (decreased by 1°C intervals per cycle), and 72°C for 30 s, followed by 38 cycles of 90°C for 30 s, 58°C for 30 s, and 72°C for 30 s, with a final 10 min at 72°C. The reactions were performed on an Applied Biosystems GeneAmp PCR System 9700 (Foster City, CA, USA) using 384-well plates (MJ Research, Waltham, MA, USA). SLC6A4 SNPs were cycled on a DNA Engine Tetrad PTC-225 thermal cycler in 384-well plates (MJ Research, Waltham, MA, USA). For conditions for SLC6A4 SNPs 9 and 11, the magnesium concentration was reduced to 1.5 mM, and the following protocol was used: 94°C for 3 min, followed by 45 cycles of 90°C for 30 s, 55°C for 30 s, and 72°C for 30 s, with a final 10 min at 72°C. The excess primers and deoxynucleotides in the PCR reaction were then degraded by adding a 0.2 μl of 10 × PCR Clean-Up Reagent (containing a mixture of shrimp alkaline phosphatase and exonuclease I) and 1.8 μl of PCR Clean-Up Dilution Buffer to each 5 μl PCR reaction (Perkin-Elmer, Boston, MA, USA). The mixture was then incubated at 37°C for 60 min, followed by inactivation for 15 min at 80°C. The final step was the addition of a 13 μl solution containing a final concentration of 0.38 μM TDI probe (Supplementary Table 1, web supporting information), 2 μl of 10 × TDI Reaction Buffer, 0.5 μl of AcycloTerminator Mix (containing R110 and TAMRA-labeled AcycloTerminators, corresponding to the polymorphic base), and 0.025 μl of AcycloPol DNA polymerase (Perkin-Elmer). This mixture was cycled at 95°C for 2 min, followed by 25 cycles of 94°C for 15 s and 55°C for 30 s. Following template-directed incorporation, fluorescence polarization was read using a Victor2 1420 Multilabel Counter (Perkin-Elmer). For the SLC6A4 SNPs, the genotypes were read using a TECAN Ultra plate reader (TECAN-US, Research Triangle Park, NC, USA). Data output is expressed in dimensionless units, mP, as described previously.30
Repeat polymorphism genotyping
PCR amplification of the VNTR in the upstream regulatory region of MAOA was carried out using the primers listed in Supplementary Table 2 (web supporting information). Amplifications were performed in a final volume of 10 μl containing 20 ng of genomic DNA template, 50 μM dNTPs, 1 M anhydrous betaine, 50 mM KCl, 20 mM Tris-HCl (pH 8.4), 1.5 mM MgCl2, and 0.5 U Platinum Taq DNA polymerase. Samples were denatured at 95°C for 4 min, followed by 35 cycles of 95°C for 1 min, 62°C for 1 min, and 72°C for 1 min, with a final 10 min step at 72°C.33 PCR products were separated on an ABI Prism 3700 DNA Analyzer and alleles were scored using Genotyper 3.5 NT software (Applied Biosystems).
PCR amplification of the tandem repeat polymorphisms in the upstream regulatory region (5-HTTLPR) and intron 4 (intron 2 VNTR) of SLC6A4 was carried out with fluorescent dye-labeled primers listed in Supplementary Table 2 (web supporting information), as was amplification of a simple sequence repeat in intron 9 (Intron 7 (GAAA)n). Of note, marker names are keyed to traditional names for continuity with the literature, despite the misnaming due to the discovery of additional noncoding exons. For these polymorphisms, amplification was performed in a final volume of 5 μl containing 20 ng of genomic DNA template, 50 μM nucleotide mix (ie 50 μM each of dATP, dCTP, and dTTP, and 25 μM each of dGTP and 7-deaza-dGTP), 1 M anhydrous betaine, 5% DMSO, 50 mM KCl, 20 mM Tris-HCl (pH 8.4), 1.83 mM MgCl2, 200 nM primers, and 0.25 U Platinum Taq DNA polymerase. Samples were denatured at 95°C for 5 min, followed by 40 cycles of 95°C for 30 s, 61°C for 30 s, and 72°C for 1 min, with a final 6 min step at 72°C. For the intron 2 VNTR, amplification was performed in a final volume of 5 μl containing 20 ng of genomic DNA template, 50 μM dNTPs, 1 M anhydrous betaine, 5% DMSO, 50 mM KCl, 20 mM Tris-HCl (pH 8.4), 1.5 mM MgCl2, 300 nM primers, and 0.25 U Platinum Taq DNA polymerase. Samples were denatured at 95°C for 5 min, followed by 35 cycles of 95°C for 30 s, 56°C for 30 s, and 72°C for 40 s, with a final 6 min step at 72°C.33 Intron 7 (GAAA)n was amplified using the same conditions and cycling protocol described above for SNPs. PCR products for all three SLC6A4 repeat polymorphisms were separated on an ABI Prism 3100 DNA Analyzer and alleles were scored using Genotyper 3.5 NT software (Applied Biosystems).
In order to correct for any population stratification within the sample collection, the method of GC was used. A collection of 20 unlinked C/T SNPs were chosen randomly throughout the genome (Supplementary Table 3, web supporting information). They were chosen from a collection of 18 150 SNPs assayed using a pooled sequencing strategy by Dr Pui-Yan Kwok for the Allele Frequency Project of the SNP Consortium (http://snp.cshl.org/allele_frequency_project/). Candidate SNPs were then chosen by iterative elimination if the SNP: (a) failed in the three populations tested (Caucasian, African-American, and Asian); (b) was nonpolymorphic in Caucasians, the predominant group in this study; (c) was nonpolymorphic in the three populations; (d) had a minor allele frequency <0.3 in Caucasians; (e) failed in African-American population; and (f) was a non-C/T SNP. This yielded 1423 SNPs eligible for use. Of these, SNPs were chosen at random were analyzed bioinformatically to determine chromosomal location. This process was stopped when at least two SNPs were localized to each of the 22 autosomes. Then, the genome was essentially divided into 20 segments, and SNPs were chosen to fit into each of these segments. Naturally, very large chromosomes would be over-represented due to the proportion of the genome found on those chromosomes. SNPs near genes are over-represented, with 11 occurring in nongene regions and nine occurring in the introns of known genes. There were no exonic SNPs in this group. SNPs were limited to C/T SNPs based on laboratory convenience and uniformity. These SNPs were genotyped using FP-TDI, as described above.
Single point association tests were performed via logistic regression using the statistical package R 1.6.134 Alleles were coded as 0, 1, or 2 corresponding to the presence of 0, 1, or 2 copies of the rare allele. This coding scheme was chosen because of its robustness to departure from the true additive genetic model.35 For each SNP, three phenotypic comparisons were made based on the results from the response pattern analysis described in the sample description. The comparisons made were: (1) all responders (specific and nonspecific) vs nonresponders, (2) specific responders vs both nonspecific responders and nonresponders, and (3) specific responders vs nonspecific responders. Empirical P-values were obtained based on 100 000 simulations using the CLUMP computer program.36 This program holds the marginal allele frequencies constant and permutes the cell counts in a contingency table. For each SNP with nominally significant association, odds ratio estimates and 95% confidence intervals (CI) were computed, comparing carriers of the rare allele to noncarriers. Haplotypes using all markers from each gene were constructed and frequencies estimated using an expectation maximization algorithm in the Arlequin 2.0 program.37 For selection of haplotype ‘tag’ SNPs (htSNPs), haplotype frequency estimations for each gene region were entered into the SNPtagger program.38 A ‘coverage value’ of 80% was set in order to capture the major haplotype diversity while excluding the extremely rare haplotypes. Haplotype frequency differences were then tested for significance using the three response phenotype comparisons listed above with the CLUMP computer program.36
Hardy–Weinberg equilibrium was determined for each SNP using the Arlequin 2.0 program.37 Linkage disequilibrium (LD) across each candidate gene was assessed using the computer program GOLD.39 For MAOA and HTR2C, the X-linked option was chosen. We accounted for population stratification through the use of GC.40 This was determined by adjusting the single point χ2 statistic by a correction factor λ. Briefly, the χ2 statistic is generated for each GC SNP. These numbers are then averaged, generating λ. The χ2 for each candidate association test is computed and then divided by λ, approximating a χ2 with one degree of freedom. The significance level was set at P<0.05 after GC correction.
In our primary phenotypic comparison, response vs nonresponse to fluoxetine, three SNPs in the TPH1 gene were significantly associated (P<0.05) (Table 1). Odds ratio estimates indicated a protective effect, that is, carriers of the rare allele were less likely to respond to treatment. However, both of the previously studied A218C and A779C TPH1 SNPs22 did not reach significance in our study. A single SLC6A4 SNP showed nominally significant association with treatment response (P=0.037, OR=0.33, 95% CI=0.08–1.35, Table 1). None of the SNPs investigated in the TPH2, HTR1A, HTR2A, HTR2C, or MAOA gene region reached significance in this phenotypic comparison.
In an effort to utilize detailed clinical information in order to provide more precise phenotypic definitions, we attempted to further delineate treatment response phenotypes using clinical data. We performed two such alternative phenotypic comparisons involving subgroups of the patients by specificity of response type as classified through the use of pattern analysis. The first comparison is based on the hypothesis that specific responders differ genetically from all other subjects. The second comparison is a variation of this hypothesis: among responders, specific responders differ genetically from nonresponders. In the specific response vs all others (nonspecific response and nonresponse) comparison, one SNP in the HTR2A gene (OR=0.34, 95% CI=0.13–0.86) and three SNPs in the TPH2 gene led to significant associations (Table 1). None of the SNPs tested in the HTR1A, HTR2C, SLC6A4, TPH1, or MAOA gene regions yielded significant association with this phenotypic comparison.
In our third phenotypic comparison, specific response vs nonspecific response, both the HTR2A and the MAOA gene regions contained significantly associated SNPs. In the HTR2A gene region, three SNPs located at the 3′-end of the coding region showed significant negative associations (Table 1). While the MAOA-VNTR failed to show significance in this study, two SNPs in the MAOA gene did show significance. None of the SNPs studied in the HTR1A, HTR2C, SLC6A4, TPH1, or TPH2 gene regions had significant association using this phenotypic comparison.
To examine interaction of alleles from different SNPs within a gene, haplotypes (ie those including information from all markers at the locus) were inferred and tested for association using the three phenotypic comparisons (Table 2). Initially, all SNPs were considered and included in haplotype construction for each gene. Haplotypes for three genes (TPH2, SLC6A4, and HTR2A) were significantly associated with the specific response vs all others phenotype, and haplotypes for MAOA, SLC6A4, and HTR2A were associated with the specific responder vs nonspecific responder comparison. For example, full-length haplotypes constructed with all 17 HTR2A SNPs examined in this study and tested using the specific response vs all other subjects comparison showed significant association (P=0.011). When the phenotypic comparison was narrowed to specific responder vs nonspecific responder, HTR2A haplotypes were still associated with specific response (P=0.001). In general, positive haplotype association results followed the positive single-locus results; however, full-length haplotype testing of TPH1 failed to reach significance in the response vs nonresponse comparison.
Knowledge of extensive LD and the existence of haplotype blocks in the human genome suggest that a limited number of SNPs may capture a substantial amount of haplotypic diversity in a population.41 We sought to determine if a subset of the SNPs we have genotyped in our sample could be used to test for association between multimarker haplotypes and our clinical phenotypes. One approach to accomplish this is to use htSNPs, as identified using the SNPtagger program.38 We chose a coverage value of 80%, as we wanted to capture the most common haplotypes and a number of the less common haplotypes, but exclude the substantial number of haplotypes estimated to occur one or fewer times in our sample.
After htSNPs yielding at least 80% coverage of the haplotypic diversity were identified in each gene region, haplotypes were constructed from genotypes of this smaller set of SNPs using Arlequin and tested for association with the phenotype using the CLUMP program. In general, the use of htSNPs to construct haplotypes led to more significant associations than using all of the SNPs screened, with the exception of the SLC6A4 and MAOA genes (Table 2). We found that this procedure indicated that the majority of haplotypes could be captured with a limited number of SNPs. For example, we initially genotyped 19 SNPs in the TPH1 gene region, but after haplotype frequency estimations were calculated it was apparent that only two SNPs were needed to capture the four most common haplotypes, which accounted for 87% (167/192) of the haplotypes seen in our study population. The use of htSNPs also implicated the TPH1, TPH2, and HTR2A gene regions in the primary comparison, response vs nonresponse, which were not significant when all possible SNPs were used for haplotype construction. For instance, the TPH1 gene region, which contained six associated SNPs in the categorical response vs nonresponse comparison, was not significantly associated when all 19 SNPs were used to construct haplotypes (P=0.387). However, when only the two htSNPs were used to construct haplotypes, a positive association was seen (P=0.0006) (Table 2).
Assessment of population stratification using GC
There has been much debate about the role of population stratification in case–control studies with the attendant possibility of false-positive associations, and less discussed false-negative findings.42 A number of approaches have been proposed to estimate the role of stratification using the genome itself to determine population heterogeneity.43, 44 Here, we use the GC approach of Devlin, in which anonymous markers from across the genome are genotyped in cases and controls, and the association test statistic is rescaled based on the degree of observed stratification.40, 43 We chose 20 SNPs distributed across the genome, and genotyped our sample for those SNPs, as described in the Methods section. Allele frequency results for those SNPs are available on request. When comparing the response group vs nonresponse group, GC analysis produced λ of 1.21, indicating a need to adjust the P-values for slight population stratification. For the other two phenotypic comparisons made, specific response vs all others and specific response vs nonspecific response, GC analysis yielded a λ of <1.0, indicating that the patient populations were not significantly stratified given the limits of the small sample size.
Levels of LD were calculated for each gene region using genotypic data (Figure 1). As can be seen from the pairwise ∣D′∣ values shown in the figure, levels of LD were generally substantial, but variable between the gene regions screened. Although there is a trend for increased LD across smaller gene regions, this was not an absolute rule as one of the smaller gene regions (HTR2A—63 kb) showed significantly less LD than the largest gene region (HTR2C—395 kb).
SNPs examined in all genes this study, except for SLC6A4, were also genotyped in 5 primate genomic DNA samples, four unrelated chimpanzees and one bonobo, using the same primers and conditions as used for the human genotyping. Of the 93 SNPs investigated in total, six failed to produce readable genotype results. Four SNPs had at least one heterozygotic chimpanzee sample, indicating present-day variation within this species. There were 83 SNPs where the primate samples were all homozygous for the same allele, implicating it as the ancestral allele for humans (Supplementary Table 1, web supporting information). The most frequent SNP allele in our human population matched this presumed ancestral allele in 46 out of 83 cases (see Supplementary Figure 3, web supporting information).
There has been much recent interest in the use of genetic variants for the prediction of medication treatment response.45 This interest is quite strong for the psychopharmacologic treatment of psychiatric disorders, which in general are still treated empirically, with an unacceptable rate of failure for first-line treatments. The recent collection of large amounts of DNA variants in the form of SNPs and inexpensive genotyping technologies make more comprehensive analyses of genetic variation in genes of interest for treatment response feasible. We have utilized these tools in order to perform an in-depth analysis of common variants in seven genes involved in serotonin function. We have genotyped a well-characterized sample of patients with unipolar major depressive disorder at a number of SNPs within each gene, and tested for association with a treatment response phenotype using individual loci and multilocus haplotypes. Using convergent analytic strategies, we found association between several of these genes and antidepressant response phenotype. For our primary phenotype, categorical response vs nonresponse, several TPH1 SNPs were negatively associated with the phenotype. Odds ratios estimates indicated that carriers of the rare allele had a higher risk of not responding to treatment. This comparison also yielded a negative association with a single SLC6A4 SNP. The location of this SNP, some 200 bases 5′ to exon 1a, raises suspicion about a potential role in regulatory elements. However, without biological evidence, it is premature to speculate on the function of this SNP. Our results differ with regard to a number of other studies in which associations were found between response and 5-HTTLPR. It might be argued that perhaps our population differs from other Caucasian samples in which this association has been noted. However, for the SLC6A4 promoter polymorphism, we observed population allele frequencies (short=0.43, long=0.57) that were close to those found in other European American populations.8, 46 Our own observations also closely match those reported in a recent study of 847 Caucasians in New Zealand, which revealed an allele frequency for the long allele of 0.57.47 Likewise, the intron 2 VNTR allele frequencies (9=0.005, 10=0.390, 12=0.605) that we observed matched those seen in European Americans,48 although we did observe an under-representation of the rarer nine allele when compared to Gelernter et al.46 We observed association between our second phenotype, specific responders vs all other subjects, at two genes, HTR2A and TPH2. With this comparison, our data supported an increased risk of specific response of carriers of the less frequent TPH2 allele, but decreased risk for carriers of the minor HTR2A allele. Finally, we observed negative associations between SNPs coming from HTR2A and MAOA and our third phenotype, specific responders vs nonspecific responders. We are intrigued by the finding that in almost all cases in which we had significant association between a response phenotype and a SNP, it was carrier status of the minor allele that appeared to be associated with nonresponse or the nonspecific pattern of response. Does this suggest that the default genetic and biological substrate for fluoxetine treatment leads to specific response, while the presence of less common alleles somehow degrades response or changes the specificity of the response? Given the limitations of this study, this speculation is premature, but nevertheless a potentially exciting observation.
We have attempted to analyze multiple genetic components of the serotonin pathway for their role in antidepressant response. The use of both single-locus and haplotypic association is not novel, but it is still uncommon to see a coordinated investigation of multiple genes. It will be important to extend our findings by employing a joint analysis of the pathway, but such methods are currently in development or are untested.
Thus far, the vast majority of studies on serotonin pathway gene variants in antidepressant response have focused on single SNP associations. The use of haplotypes, or particular combinations of alleles observed in a population, has been shown both theoretically and empirically to be a powerful approach for the dissection of complex genetic traits.49, 50, 51, 52 We found haplotypes associated with our response phenotypes for several of the genes we investigated. Further, using a haplotype SNP tagging approach, we were able to reduce the number of SNPs required to represent the majority of haplotypes, and generally noted even greater association between these htSNP haplotypes and phenotype. It has become commonplace to note that limited haplotypic diversity exists in many regions of the genome, and the seven genes we investigate proved to be no exception. The use of multiple SNPs also allows for the estimation of LD within a given gene, which provides useful information for association mapping, particularly for the selection of markers. We were not surprised to find extensive LD across the genes under examination here.
Accurate assessment of clinical response is essential in pharmacogenetic studies, as there is a need to limit the amount of phenotypic heterogeneity. This is particularly true with antidepressant therapy, as placebo response rates can be as high as 60% for patients with major depressive disorder.53, 28 Previous studies with serotonin pathway gene variants and SSRI antidepressant response have failed to address these concerns. It is reasonable to hypothesize that genetic factors may influence placebo response to antidepressants, given recent evidence showing that subjects homozygous for the long allele of the 5-HTTLPR were more likely to respond placebo in a fluoxetine trial in 51 depressed subjects.54 In this study, we utilized a validated response pattern algorithm to classify patients as nonresponders, specific responders, or placebo (nonspecific) responders to fluoxetine.26, 27 We were thus able to obtain a more precise assessment of clinical response phenotype, and potentially identify phenocopies. The use of pattern analysis to define response type reduced the amount of phenotypic heterogeneity in our comparison groups and was integral in uncovering some of our findings. Indeed, had we only investigated the most obvious phenotypic comparison, responders vs nonresponders, we would have found negligible association between response and genotype. What is interesting in the current work is that we detected a number of differences between treatment populations using more precise phenotypic stratification, namely, by trying to separate specific responders from nonspecific responders. A number of our findings suggest that specific responders differ genetically from all others (ie from nonresponders and nonspecific responders) (HTR2A, SLC6A4, and TPH2). Similar findings indicate that even among responders, specific responders may differ at several candidate loci from nonspecific responders (HTR2A and MAOA). This suggests that there may be heritable differences underlying the nature of response to fluoxetine. This is interesting in the context of imaging data that shows differential brain glucose metabolism in persons with depression administered placebo or fluoxetine.55 Specifically, hospitalized males receiving placebo or fluoxetine for unipolar depression underwent positron emission tomography before and after 6 weeks of treatment. The authors found that placebo response was associated with specific regional changes in brain function. Remarkably, these regions showed overlap with changes seen in fluoxetine responders, but were not identical. Similar differences were seen using quantitative electroencephalography, suggesting different biological responses underlying specific response, nonspecific response, or nonresponse to antidepressant medication.56 Although our methodology for assessing the placebo phenotype differs from that of Rausch et al,55 it is interesting to note that our most positive findings for the serotonin transporter are seen when we attempted to dissect out placebo-responsive subjects in the work described here. Given the high placebo response rate for many antidepressants, it may prove necessary to control for nonspecific responses in pharmacogenetic studies on antidepressant response.
An important issue that has been overlooked in many case–control association studies of antidepressant response (or association studies in general) is population stratification, or differences in allele frequencies between populations that may lead to spurious associations. One method that has been used to deal with population stratification is the use of ethnically matched cases and controls. However, a significant amount of ‘cryptic stratification’ may still exist even after cases and control are carefully matched.57 We employed the method of GC to correct for any underlying stratification within our population. This method uses allele frequency data for a number of unlinked loci to assess quantitatively the degree of stratification present within the study population.58, 40 This produces a correction factor that is used to adjust the critical level of significance required for a positive association at the candidate SNPs, thereby reducing the rate of false positives.40 In general, we found little evidence for population stratification in our sample, suggesting that the genetic differences detected between our phenotypic groups are unlikely to be due to ethnic stratification. Although our GC markers suggest minimal stratification, it is also possible that the sample size may not be large enough to allow a more accurate estimation of stratification.
Genotyping of five primate samples gave insight into the evolutionary history of these polymorphic sites. Ancestral allele was inferred if the human SNP site was monomorphic across all non-human primate samples screened. A high fraction (37/83) of SNPs had discrepancy between the most common allele seen in humans and the most common allele from non-human primates. The correlation between human common allele state and ancestral allele state was similar to previous findings, indicating caution is in order when assigning ancestral allele state based exclusively on major human allele frequency (see Supplementary Figure 3, web supporting information).59 Ancestral allele state was not predictive of an association with fluoxetine response.
While this study produced several encouraging results, it was also subject to some limitations. The most obvious drawback of this study is the small sample size. This contributes to wide CIs and diminished power to detect associations between genotype and phenotype. For example, to detect a difference in allele frequencies of 0.2 between responders and nonresponders with α=0.05 and 1−β=0.8 would require almost a tripling of the total sample size.60 This is a critical limitation, as it suggests a much higher likelihood that our observations may contain more false-negative results than if we had a much larger sample size. We also believe that the highly characterized phenotypes used in this study helps to compensate for this limitation. Also, multiple statistical comparisons were made with the data generated from this study. However, at this point, it is unclear how to correct for multiple testing with data of this type, since the comparisons made were not independent. High levels of LD and low haplotype diversity indicate that the individual loci are related, and that using a standard Bonferroni correction would be overly conservative. Further research into this statistical problem needs to be performed to fully understand how to interpret data from association studies investigating large number of SNPs, and thus these results should be considered preliminary until replicated. A further limitation may involve our choice of D′ to estimate LD. Although D′ can be upwardly biased with small sample sizes, it is robust to differences in allele frequencies and our sample of 96 may be adequate to capture a reasonable estimate of LD in the gene regions studied.61 A final limitation might be our focus on a binary trait, antidepressant response, as our primary outcome variable. Continuous outcome variables (such as quantitative changes in symptoms) can often provide more information than a binary trait. However, the difficulty is in selecting which one to use. Trying them all would compromise our type I error rate significantly. We thus chose the most straightforward outcome variable for the present analyses.
In summary, we have used a number of innovative molecular, phenotypic, and statistical approaches to investigate the role of DNA variants in candidate genes for fluoxetine response and found several interesting associations. The TPH1 and SLC6A4 genes seem to be nominally associated with response to fluoxetine. The HTR2A, MAOA, and TPH2 genes appear to be involved in determining the specificity of response to fluoxetine. If confirmed in further studies, polymorphisms in these genes could be a useful tool for a genetically informed psychopharmacology of depression.
Kessler RC, McGonagle KA, Zhao S, Nelson CB, Hughes M, Eshleman S et al. Lifetime and 12-month prevalence of DSM-III-R psychiatric disorders in the United States. Results from the National Comorbidity Survey. Arch Gen Psychiatry 1994; 51: 8–19.
Kessler RC, Berglund P, Demler O, Jin R, Koretz D, Merikangas KR et al. The epidemiology of major depressive disorder: results from the National Comorbidity Survey Replication (NCS-R). JAMA 2003; 289: 3095–3105.
Sackeim HA . The definition and meaning of treatment-resistant depression. J Clin Psychiatry 2001; 62(Suppl 16): 10–17.
Serretti A, Lilli R, Smeraldi E . Pharmacogenetics in affective disorders. Eur J Pharmacol 2002; 438: 117–128.
Veenstra-VanderWeele J, Anderson GM, Cook Jr EH . Pharmacogenetics and the serotonin system: initial studies and future directions. Eur J Pharmacol 2000; 410: 165–181.
Lerer B, Macciardi F . Pharmacogenetics of antidepressant and mood-stabilizing drugs: a review of candidate–gene studies and future research directions. Int J Neuropsychopharmacol 2002; 5: 255–275.
Lesch K-P, Bengel D, Heils A, Sabol SZ, Greenberg BD, Petri S et al. Association of anxiety-related traits with a polymorphism in the serotonin transporter gene regulatory region. Science 1996; 274: 1527–1531.
Pollock BG, Ferrell RE, Mulsant BH, Mazumdar S, Miller M, Sweet RA et al. Allelic variation in the serotonin transporter promoter affects onset of paroxetine treatment response in late-life depression. Neuropsychopharmacology 2000; 23: 587–590.
Zanardi R, Serretti A, Rossini D, Franchini L, Cusin C, Lattuada E et al. Factors affecting fluvoxamine antidepressant activity: influence of pindolol and 5-HTTLPR in delusional and nondelusional depression. Biol Psychiatry 2001; 50: 323–330.
Durham LK, Webb SM, Milos PM, Clary CM, Seymour AB . The serotonin transporter polymorphism, 5HTTLPR, is associated with a faster response time to sertraline in an elderly population with major depressive disorder. Psychopharmacology (Berl) 2003 Epub: DOI: 10.1007/s00213-003-1562-3.
Kim DK, Lim SW, Lee S, Sohn SE, Kim S, Hahn CG et al. Serotonin transporter gene polymorphism and antidepressant response. NeuroReport 2000; 11: 215–219.
Yu YW, Tsai SJ, Chen TJ, Lin CH, Hong CJ . Association study of the serotonin transporter promoter polymorphism and symptomatology and antidepressant response in major depressive disorders. Mol Psychiatry 2002; 7: 1115–1119.
Walther DJ, Peter JU, Bashammakh S, Hortnagl H, Voits M, Fink H et al. Synthesis of serotonin by a second tryptophan hydroxylase isoform. Science 2003; 299: 76.
Sanchez C, Hyttel J . Comparison of the effects of antidepressants and their metabolites on reuptake of biogenic amines and on receptor binding. Cell Mol Neurobiol 1999; 19: 467–489.
Sargent P, Williamson DJ, Pearson G, Odontiadis J, Cowen PJ . Effect of paroxetine and nefazodone on 5-HT1A receptor sensitivity. Psychopharmacology (Berl) 1997; 132: 296–302.
Zanardi R, Artigas F, Moresco R, Colombo C, Messa C, Gobbo C et al. Increased 5-hydroxytryptamine-2 receptor binding in the frontal cortex of depressed patients responding to paroxetine treatment: a positron emission tomography scan study. J Clin Psychopharmacol 2001; 21: 53–58.
Kim SW, Park SY, Hwang O . Up-regulation of tryptophan hydroxylase expression and serotonin synthesis by sertraline. Mol Pharmacol 2002; 61: 778–785.
Stokes AH, Xu Y, Daunais JA, Tamir H, Gershon MD, Butkerait P et al. p-ethynylphenylalanine: a potent inhibitor of tryptophan hydroxylase. J Neurochem 2000; 74: 2067–2073.
Zimmer L, Luxen A, Giacomelli F, Pujol JF . Short- and long-term effects of p-ethynylphenylalanine on brain serotonin levels. Neurochem Res 2002; 27: 269–275.
Yoshida K, Naito S, Takahashi H, Sato K, Ito K, Kamata M et al. Monoamine oxidase: a gene polymorphism, tryptophan hydroxylase gene polymorphism and antidepressant response to fluvoxamine in Japanese patients with major depressive disorder. Prog Neuropsychopharmacol Biol Psychiatry 2002; 26: 1279–1283.
Cusin C, Serretti A, Zanardi R, Lattuada E, Rossini D, Lilli R et al. Influence of monoamine oxidase A and serotonin receptor 2A polymorphisms in SSRI antidepressant activity. Int J Neuropsychopharmacol 2002; 5: 27–35.
Serretti A, Zanardi R, Rossini D, Cusin C, Lilli R, Smeraldi E et al. Influence of tryptophan hydroxylase and serotonin transporter genes on fluvoxamine antidepressant activity. Mol Psychiatry 2001; 6: 586–592.
Serretti A, Zanardi R, Cusin C, Rossini D, Lorenzi C, Smeraldi E . Tryptophan hydroxylase gene associated with paroxetine antidepressant activity. Eur Neuropsychopharmacol 2001; 11: 375–380.
McGrath PJ, Stewart JW, Petkova E, Quitkin FM, Amsterdam JD, Fawcett J et al. Predictors of relapse during fluoxetine continuation or maintenance treatment of major depression. J Clin Psychiatry 2000; 61: 518–524.
First M, Spitzer R, Gibbon M, Williams JBW . Structured Clinical Interview for DSM-IV Axis I Disorders—Patient Edition (SCID-I/P), Version 2.0. 1995.
Stewart JW, Quitkin FM, McGrath PJ, Amsterdam J, Fava M, Fawcett J et al. Use of pattern analysis to predict differential relapse of remitted patients with major depression during 1 year of treatment with fluoxetine or placebo. Arch Gen Psychiatry 1998; 55: 334–343.
Quitkin FM, Rabkin JG, Ross D, Stewart JW . Identification of true drug response to antidepressants. Use of pattern analysis. Arch Gen Psychiatry 1984; 41: 782–786.
Quitkin FM . Placebos, drug effects, and study design: a clinician's guide. Am J Psychiatry 1999; 156: 829–836.
Iyengar S, Seaman M, Deinard AS, Rosenbaum HC, Sirugo G, Castiglione CM et al. Analyses of cross species polymerase chain reaction products to infer the ancestral state of human polymorphisms. DNA Sequence 1998; 8: 317–327.
Chen X, Levine L, Kwok PY . Fluorescence polarization in homogeneous nucleic acid analysis. Genome Res 1999; 9: 492–498.
Hamilton SP, Slager SL, Heiman GA, Deng Z, Haghighi F, Klein DF et al. Evidence for a susceptibility locus for panic disorder near the catechol-O-methyltransferase gene on chromosome 22. Biol Psychiatry 2002; 51: 591–601.
Rozen S, Skaletsky HJ . Primer3 on the WWW for general users and for biologist programmers. In: Krawetz S, Misener S. (eds). Bioinformatics Methods and Protocols. Human Press: Totowa, NJ, 2000; 365–386.
Melke J, Landen M, Baghei F, Rosmond R, Holm G, Bjorntorp P et al. Serotonin transporter gene polymorphisms are associated with anxiety-related personality traits in women. Am J Med Genet 2001; 105: 458–463.
Ihaka R, Gentleman R. R : A language for data analysis and graphics. J Comput Graph Stat 1996; 5: 299–314.
Freidlin B, Zheng G, Li Z, Gastwirth JL . Trend tests for case–control studies of genetic markers: power, sample size and robustness. Hum Hered 2002; 53: 146–152.
Sham PC, Curtis D . Monte Carlo tests for associations between disease and alleles at highly polymorphic loci. Ann Hum Genet 1995; 59(Part 1): 97–105.
Schneider S, Roessli D, Excoffier L . Arlequin: A Software for Population Genetics Data Analysis, Ver 2.000. Genetics and Biometry Lab, Department of Anthropology, University of Geneva. Geneva, Switz, 2000.
Ke X, Cardon LR . Efficient selective screening of haplotype tag SNPs. Bioinformatics 2003; 19: 287–288.
Abecasis GR, Cookson WO . GOLD—graphical overview of linkage disequilibrium. Bioinformatics 2000; 16: 182–183.
Bacanu SA, Devlin B, Roeder K . The power of genomic control. Am J Hum Genet 2000; 66: 1933–1944.
Johnson GC, Esposito L, Barratt BJ, Smith AN, Heward J, Di Genova G et al. Haplotype tagging for the identification of common disease genes. Nat Genet 2001; 29: 233–237.
Cardon LR, Palmer LJ . Population stratification and spurious allelic association. Lancet 2003; 361: 598–604.
Devlin B, Roeder K, Wasserman L . Genomic control, a new approach to genetic-based association studies. Theor Popul Biol 2001; 60: 155–166.
Pritchard JK, Rosenberg NA . Use of unlinked genetic markers to detect population stratification in association studies. Am J Hum Genet 1999; 65: 220–228.
Weinshilboum R . Inheritance and drug response. N Engl J Med 2003; 348: 529–537.
Gelernter J, Kranzler H, Coccaro EF, Siever LJ, New AS . Serotonin transporter protein gene polymorphism and personality measures in African American and European American subjects. Am J Psychiatry 1998; 155: 1332–1338.
Caspi A, Sugden K, Moffitt TE, Taylor A, Craig IW, Harrington H et al. Influence of life stress on depression: moderation by a polymorphism in the 5-HTT gene. Science 2003; 301: 386–389.
Gelernter J, Kranzler H, Cubells JF . Serotonin transporter protein (SLC6A4) allele and haplotype frequencies and linkage disequilibria in African- and European-American and Japanese populations and in alcohol-dependent subjects. Hum Genet 1997; 101: 243–246.
Gabriel SB, Schaffner Sf, Nguyen H, Moore JM, Roy J, Blumenstiel B et al. The structure of haplotype blocks in the human genome. Science 2002; 296: 2225–2229.
Rioux JD, Daly MJ, Silverberg MS, Lindblad K, Steinhart H, Cohen Z et al. Genetic variation in the 5q31 cytokine gene cluster confers susceptibility to Crohn disease. Nat Genet 2001; 29: 223–228.
Martin ER, Lai EH, Gilbert JR, Rogala AR, Afshari AJ, Riley J et al. SNPing away at complex diseases: analysis of single-nucleotide polymorphisms around APOE in Alzheimer disease. Am J Hum Genet 2000; 67: 383–394.
Zhang K, Calabrese P, Nordborg M, Sun F . Haplotype block structure and its applications to association studies: power and study designs. Am J Hum Genet 2002; 71: 1386–1394.
Bialik RJ, Ravindran AV, Bakish D, Lapierre YD . A comparison of placebo responders and nonresponders in subgroups of depressive disorder. J Psychiatry Neurosci 1995; 20: 265–270.
Rausch JL, Johnson ME, Fei YJ, Li JQ, Shendarkar N, Hobby HM et al. Initial conditions of serotonin transporter kinetics and genotype: influence on SSRI treatment trial outcome. Biol Psychiatry 2002; 51: 723–732.
Mayberg HS, Silva JA, Brannan SK, Tekell JL, Mahurin RK, McGinnis S et al. The functional neuroanatomy of the placebo effect. Am J Psychiatry 2002; 159: 728–737.
Leuchter AF, Cook IA, Witte EA, Morgan M, Abrams M . Changes in brain function of depressed subjects during treatment with placebo. Am J Psychiatry 2002; 159: 122–129.
Reich DE, Goldstein DB . Detecting association in a case–control study while correcting for population stratification. Genet Epidemiol 2001; 20: 4–16.
Devlin B, Roeder K, Bacanu SA . Unbiased methods for population-based association studies. Genet Epidemiol 2001; 21: 273–284.
Hacia JG, Fan JB, Ryder O, Jin L, Edgemon K, Ghandour G et al. Determination of ancestral alleles for human single-nucleotide polymorphisms using high-density oligonucleotide arrays. Nat Genet 1999; 22: 164–167.
Dupont WD, Plummer J . Power and sample size calculations *1: A review and computer program. Controlled Clin Trials 1990; 11: 116–128.
Weiss KM, Clark AG . Linkage disequilibrium and the mapping of complex human traits. Trends Genet 2002; 18: 19–24.
This paper is dedicated to the memory of Ira Herskowitz. We wish to thank the subjects of this study for their participation. Funding for this study was provided by NARSAD (SPH), HHMI (EJP), NIMH Grant # R10 MH56058 (PJM) and Grant CA 94919 (SLS) from the National Cancer Institute. We also wish to acknowledge Carmen Prieto, Maria Bautista, Manuel Abreu, and David Mayo for their expert technical assistance and the staff of the Depression Evaluation Service at the New York State Psychiatric Institute for assistance with patient recruitment and sample collection. We would also like to acknowledge TECAN-US for the use of an ULTRA plate reader.
About this article
Cite this article
Peters, E., Slager, S., McGrath, P. et al. Investigation of serotonin-related genes in antidepressant response. Mol Psychiatry 9, 879–889 (2004). https://doi.org/10.1038/sj.mp.4001502
- SNP, haplotype
- linkage disequilibrium
Serotonergic receptor gene polymorphism and response to selective serotonin reuptake inhibitors in ethnic Malay patients with first episode of major depressive disorder
The Pharmacogenomics Journal (2021)
Neuroscience Letters (2020)
Journal of Neural Transmission (2019)
Association Study Between Serotonin Transporter Gene and Fluoxetine Response in Mexican Patients With Major Depressive Disorder
Clinical Neuropharmacology (2019)