INTRODUCTION

Decision making depends on a complex set of processes that are orchestrated in various brain systems to find an optimal outcome (Elliott and Deakin, 2005; Paulus, 2007). Risk-taking refers to the willingness to accept a possible negative outcome in order to potentially achieve a desirable outcome, and typically involves assessing the relative probability of winning or losing against the values of the outcomes (Anderson et al, 2003; Cardinal, 2006). Subjects with depression or anxiety are less likely than controls to take risks to gain reward (Elliott et al, 1997; Forbes et al, 2006; Smoski et al, 2008; Paulus, 2007). In contrast, pathological gamblers are more risk-seeking (Holt et al, 2003). Another aspect of decision making is self-control, or the ability to delay gratification, which is more related to impulsiveness and is impaired in patients with substance-use disorder (Rogers et al, 1999; Bechara et al, 2002; Paulus, 2007).

Manipulation of serotonin (5-HT) appears to influence decision making. Acute tryptophan depletion (ATD) leads to impulsive choice and impaired decision making related to different magnitudes of gains, in both animals and humans (Rogers et al, 1999, 2003; Mobini et al, 2000; Walderhaug et al, 2007). Furthermore, a recent study showed that tryptophan supplementation significantly decreased loss-aversion (Murphy et al, 2008). Although 5-HT appears to influence choice involving delayed reinforcement or different magnitudes of gains, it remains unclear whether 5-HT affects choice related to the assessment of probability (Mobini et al, 2000; Anderson et al, 2003; Rogers et al, 2003; Cardinal, 2006), especially at a genetic level.

It has been reported that possession of the less active S allele of the 5-HT transporter 5HTTLPR polymorphism (SLC6A4 gene short/S–long/L promoter variant) increased sensitivity to the probability of winning in a ‘risky-choice task’ (Roiser et al, 2006) and promoted disadvantageous choices in the Iowa Gambling Task (IGT) probably due to lack of persistence (Must et al, 2007; Homberg et al, 2008) or to slower learning of advantageous decision making (Jollant et al, 2007). It has also been suggested that 5HTTLPR variations modulate sensitivity to punishment rather than reward, as the LL homozygous individuals are less sensitive to punishment-related information (Blair et al, 2008; Roiser et al, 2009). Jollant et al (2007) showed that genetic variations in the TPH1, TPH2 (tryptophan hydroxylase 1 and 2, respectively), and MAOA (monoamine oxydase A) genes that have been associated with suicide were also associated with poorer performance on the IGT in suicide attempters. However, all of the above-mentioned studies were carried out in small samples (less than 200 subjects) and require confirmation in larger samples.

In this study, the main candidate gene was the TPH2 that codes for the majority of TPH in the human brain (Walther et al, 2003; Zill et al, 2007a). Variations in the TPH2 gene have been repeatedly implicated in major depression disorder (MDD), and in suicidal behavior in MDD patients (Zill et al, 2004a, 2004b; Zhang et al, 2005; Zhou et al, 2005; Lopez et al, 2007a; Haghighi et al, 2008). To cover this gene, we used haplotype tagging SNPs (htSNPs) similar to the previous studies.

As the availability of synaptic 5-HT depends on other serotonergic genes, we also investigated the well-known 5HTTLPR polymorphism and a functional TPH1 gene variant (A218C, rs1800532). A218C has been shown to alter TPH immunoreactivity in postmortem brain samples of both suicide victims and controls (Ono et al, 2002) and has been associated with suicidal behavior (Li and He, 2006). It has been demonstrated that TPH1 is expressed in several human brain areas, although at lower levels than TPH2 in the raphe (Zill et al, 2007a). Finally, we genotyped for a functional variant in the 5-HTR1A receptor gene (C(-1019)G, rs6295) that modulates the expression of 5-HT1A autoreceptors in raphe cells and that controls 5-HT neuronal firing, and this influence synaptic 5-HT release (Lemonde et al, 2003). This functional polymorphism has previously been associated with depression-, suicide-, and anxiety-related traits (Lemonde et al, 2003; Strobel et al, 2003).

In order to investigate risky decision-making behavior in a large population, we developed a pencil-and-paper probabilistic gambling task administered remotely with questionnaires assessing personality data, current mood, and anxiety symptoms. On the basis of previous literature, we hypothesized that subjects with a self-reported history of depression or with a higher neuroticism, depression and/or anxiety scores would make safer (less risky) choices. Furthermore, we hypothesized that variants in the serotonergic genes and especially in the TPH2 gene that regulates the availability of synaptic 5-HT would influence risky decision making.

MATERIALS AND METHODS

Participants

This study was part of the EU funded NewMood (New Molecules for Mood Disorders) research program. We recruited participants aged 18–60 years from Greater Manchester, United Kingdom, through general practices, and a website (http://www.newmood.co.uk); details of our recruitment strategy and responses were published previously (Juhasz et al, 2009). The included participants were of Caucasian European origin and could have a self-reported history of depression or anxiety disorders; we excluded those reporting manic or hypomanic episodes, psychotic symptoms, or obsessive–compulsive disorder (all information based on background self-report questionnaire data). Thus, the population sample in this study consisted of 1188 participants who fulfilled the inclusion criteria, sent back the questionnaire with a completed gambling task and provided DNA. Population details are shown in Table 1. The study was approved by the local Ethics Committees and was carried out in accordance with the Declaration of Helsinki.

Table 1 Population Details for Those Who Fulfilled the Inclusion Criteria for the Study (n=1188)

Phenotype Assessments

The NewMood booklet consisted of brief or adapted versions of standard questionnaires that have been validated at a second level of the study in a subset of subjects (n=142) by full versions of the questionnaires and face-to-face interviews. The background questionnaire was adapted from a structured self-rating questionnaire (Hunt et al, 1999) that consists of 22 items and collects information about medical history, including personal psychiatric history (particularly depression), family psychiatric history, and socioeconomic background. Impulsivity was measured by the Impulsiveness, Venturesome and Empathy Questionnaire (IVE) Impulsiveness subscale (Eysenck and Eysenck, 1978). For the analysis, a continuous weighted dimension score (sum of item scores divided by the number of items completed) was calculated. The IVE Impulsivity score showed significant correlation with the NEO PI-R (Costa and McCrae, 1992) Impulsiveness (N5) facet score (Pearson's correlation: R=0.47, p<0.001, n=142) in our validation sample. To assess personality we used the Big Five Inventory (BFI-44) (John et al, 1991), and a continuous weighted dimension score was calculated for Neuroticism. The 53-item Brief Symptom Inventory (BSI) (Derogatis, 1993) was used to measure anxiety symptoms (using the anxiety subscale) and depressive symptoms (using the depression subscale plus additional items); a continuous weighted dimension score was calculated for Depression and Anxiety. Validation data of neuroticism and symptom scores were published previously (Juhasz et al, 2009).

The Recent Negative Life Events questionnaire was based on the validated List of Life Threatening Experiences (LTE) (Brugha et al, 1985; Rijsdijk et al, 2001) and the sum of life event items was used in the analysis. Questions based on the Childhood Trauma Questionnaire (CTQ) (Bernstein et al, 1994) were used to determine Childhood Adversity related to emotional and physical abuse, and emotional and physical neglect. An additional question asked about parental loss during childhood. The sum of item scores was used in the analysis. Summary scores from the full version of CTQ significantly correlated with scores derived from our short questionnaire (Pearson's correlation: R=0.75, p<0.001, n=142).

Gambling Task

This task was derived from that of Rachlin et al (1986). Subjects were presented with a picture of two ‘wheel of fortune’-like spinners each with a pointer at the top. Each spinner was divided into a white and a gray segment, and the subject was told to imagine they would ‘win’ if the spinner stopped with the white segment under the pointer, but not with the gray (Figure 1). Subjects were instructed to indicate which spinner they would choose to maximize their winnings. Dial ‘A’ provided a smaller (£1) but more likely ‘win,’ whereas dial ‘B’ gave a ‘win’ twice the amount but with a lower probability that was systematically varied. As our reward task was imaginary, subjects did not gain real monetary reward.

Figure 1
figure 1

Decision-making task sample trial.

The probability of winning on ‘A’ (pA) was set to p=0.95, 0.5, 0.33, and 0.25, whereas probability of winning on ‘B’ was varied from p=0.80 to 0.025 but was always lower than pA. As the probability of winning on ‘B’ increased, an indifference point is reached where the participant values both option equally, and may select ‘B’ as it offers the potential for a larger ‘win.’ Odds against winning (θ) were calculated as [1/p]−1, giving four θA values (0.05, 1, 2, 3). The odds against winning on ‘B’ (θB) at which the participant switched from dial ‘A’ to dial ‘B’ (the indifference point) was calculated for each value of ‘A’ by linear interpolation. To assess an individual's overall tendency to take a safer vs more risky choice, we plotted θB against θA and calculated a global area under the curve measure (AUCθ) using the trapezoid method requiring all four indifference points; a smaller value indicating safer choice. Using AUCθ gives weight to choices, in which wins are unlikely and is sensitive to how much the individual is willing to take a risk for a greater reward under conditions where both outcomes are uncertain (Ho et al, 1999).

Given the study's constraints, we were not able to determine directly whether participants understood the task and so we included two trials, in which one of the choices was so disadvantageous that its selection was irrational and if chosen the individual was excluded. We also required choices on ‘B’ to follow the same rank order as their corresponding value on ‘A’ (allowing one occurrence of the same value for contiguous choices on ‘B’). This excluded participants who, for example, chose a smaller indifference probability on ‘B’ for p=0.33 than for p=0.25 on ‘A.’ Using these rules, 87.1% (n=1035) of participants were included in the final analysis.

From these subjects, 72 completed the computer version of a similar decision-making task and 69 provided calculable data. During this task, the participants were shown two ‘wheel of fortune’ spinners—A and B—that had a green (win) and a red (lose) segment. Alternative A had a reward of 10p; alternative B had a reward of 20p but a lower probability chance of winning compared to A (2 blocks of 41 trials; θA: θ=0.00, 0.33, 1, and 3; θB: from θ=0.00 to ∞). Participants pressed A or B on the keyboard depending on which alternative they preferred. The outcome (which had been decided in a pseudorandom manner) was then shown to the subject. The reward was imaginary again, so the subjects did not get real monetary reward. A global AUCθ was calculated from these data similarly to the paper version of the task. Results from the two tasks showed significant correlation (Pearson's R=0.30, p=0.012) in this population.

Genotyping

Buccal mucosa cells were collected from the participants and genomic DNA was extracted according to a previously described method (Freeman et al, 2003).

HaploView software (http://www.broad.mit.edu/personal/jcbarret/haploview/) was used to identify htSNPs using the confidence interval method (Barrett et al, 2005; Gabriel et al, 2002). The tagging was based on the CEPH (Utah residents with ancestry from northern and western Europe) data that were available at the International HapMap Project Phase I. June 2005 release (http://www.hapmap.org), thus SNPs (functional or tagging) that were identified later were not included in this study. We also genotyped a further three functionally relevant serotonergic polymorphisms: 5HTTLPR, rs6295, and rs1800532.

The chosen SNPs were genotyped using the Sequenom® MassARRAY technology (Sequenom, San Diego, http://www.sequenom.com) using 25 ng of DNA. Determination of the 5HTTLPR genotype was performed in a Thermo-Fast® 384 PCR plate (ABgene Ltd, UK) on a PTC-220 Dyad TM thermocycler (Bio-Rad, Hercules, CA). Polymerase chain reaction products were analyzed using an ABI3100 Genetic Analyzer and GeneScan analysis software (Applied Biosystems, Netherlands). Genotyping was performed blinded with regard to phenotype. All laboratory work was performed under the ISO 9001:2000 quality management requirements. The genotyping methods in more details can be seen in Supplementary Document 1.

Statistical Analysis

Statistical analyses were performed using SPSS for Windows Statistical Analysis Software, Version 15.0. A p-value of <0.05 was adopted for all statistical testing, and all reported p-values were two-tailed.

We used the HaploView program to explore the haplotype structure of TPH2 gene (Gabriel et al, 2002; Barrett et al, 2005); Quanto 1.2 version (http://hydra.usc.edu/gxe) to calculate the power of the recruited populations; and PLINK v1.06 (http://pngu.mgh.harvard.edu/purcell/plink/) for testing association of different genetic models (dominant, recessive, and additive; linear regression model covariation with age and sex), interactions between SNPs (epistasis) and with sex, and for calculating Hardy–Weinberg analysis.

Haplotype analyses were performed with HelixTreeTM 6.4.1 (Golden Helix, USA) using haplotype trend regression. Only haplotypes with a frequency greater than 5% were used in the linear regression analysis, similar to the method described previously (Juhasz et al, 2009). In all cases, data were adjusted for age and sex. To reduce the influence of multiple testing, we used a permutation test, randomly grouping the sample 10 000 times.

For both single-marker association and haplotypic association analyses, we used false discovery rate calculation at a level of 5% (FDR; Qvalue: http://genomics.princeton.edu/storeylab/qvalue/) to adjust p-values according to the number of hypothesis tested (Storey and Tibshirani, 2003). We report q–value, which is a measure of significance of each test of many tests performed simultaneously.

RESULTS

Decision-Making Task

Women were less likely to make risky decisions than men, represented by lower AUCθ values (F=9.38, df=1,1032, p=0.002; AUCθ mean±SEM after covariation for age: males—10.42±0.36, females—9.09±0.24; explained variance 0.9%). Subjects with reported depression were also less likely to make risky decisions (F=4.83, df=1,1031, p=0.028; AUCθ mean±SEM after covariation for age and gender: no depression—9.99±0.30, depression—9.09±0.28). Higher neuroticism (Pearson's correlation R=−0.13, p<0.001), depression (Pearson correlation R=−0.13, p<0.001), and anxiety (Pearson's correlation R=−0.13, p<0.001) scores were also associated with risk aversion. These factors explained a further 1.6% variance in risk-taking behavior. However, impulsivity measured by IVE and risk taking measured by our task were independent of each other (Pearson's correlation R=−0.02, p=0.233). Both Childhood Adversity (Pearson's R=−0.06, p=0.039) and Recent Negative Life Events (Pearson's R=−0.06, p=0.042) significantly correlated with less risk taking, but after correction for depressive and anxiety symptoms this effect became nonsignificant (Childhood Adversity: Pearson's R=−0.02, p=0.629; Recent Negative Life Events: Pearson's R=−0.03, p=0.344).

After covariation for age and gender, there were no significant difference in neuroticism (p=0.45), impulsivity (p=0.20), symptom scores (depression, p=0.80; anxiety, p=0.82) or education level (p=0.24) between those who did not provide consistent task results (n=153) and who did (n=1035).

Using the computer version of the decision-making task, females were less likely to make risky decisions (F=3.78, df=1,66, p=0.056; AUCθ mean±SEM after covariation for age: males (n=30)—11.17±0.67, females (n=39)—9.44±0.59; explained variance 5.4%), although this effect missed the significance threshold. We did not find a significant difference between those with and without lifetime major depressive disorder according to SCID (F=0.91, df=1,64, p=0.344; AUCθ mean±SEM after covariation for age and gender: no psychiatric disorder (n=28)—9.66±0.71, lifetime MDD (n=40)—10.56±0.59). However, subjects with low symptom scores (MADRS7; n=53) tended to make more risky decisions than those who had high symptom scores (MADRS>7, n=16; F=3.42, df=1,65, p=0.069; AUCθ mean±SEM after covariation for age and gender: low symptom score—10.65±0.50, high symptom score—8.68±0.93; explained variance 2.8%).

Genotyping Results

HaploView analysis of the TPH2 gene—based on the HapMap Project Phase I. (June 2005) release CEPH population data—resulted in seven htSNPs that were genotyped in our study. Details about the TPH2 htSNPs and the genotyped three functional serotonergic polymorphisms can be seen in Table 2. Although no genomic control by random marker genotyping was performed in the study population, we attempted to reduce ethnic variation and stratification effects by including only independent Caucasian subjects of European origin in the analysis. Indeed, all of the genotyped polymorphisms were in Hardy–Weinberg equilibrium both in the total population and separately for those who reported depression and who did not (data not shown). TPH2 linkage disequilibrium (LD) block structure is similar in our population and in the CEPH population (data from the HapMap Project Phase II. (November 2008) release; see Figure 2). Although in this study we did not genotype functional SNPs in the TPH2 gene, we do show the LD data between our htSNPs and rs4570625 (−703G/A) (Chen et al, 2008) and rs7305115 (Lim et al, 2007) in the CEPH population in order to interpret our results. There are no available data in the CEPH population about other functional SNPs (rs11178997: −473T/A; rs11178998: 90A/G) (Chen et al, 2008; Lin et al, 2007; Scheuch et al, 2007).

Table 2 Summary of the Genotyped SNPs
Figure 2
figure 2

Linkage disequilibrium (LD) R2 data of the TPH2 gene in the CEPH population (HapMap Project Phase II. November 2008 release) and in our study. Blocks of LD have been identified on the basis of the method published by Gabriel et al (2002) using Haploview program.

Single-Marker Associations

We found significant association of the AUCθ and TPH2 htSNPs, and after FDR correction the association remained significant for rs6582078 and rs1352250 (Table 3). The minor alleles were positively associated with increased risk taking in all htSNPs. There was no significant marker-specific association between 5HTTLPR, rs6295 or rs1800532 and risk taking. None of the SNPs showed association with symptom or personality scores after correction for multiple testing (Supplementary Table S1).

Table 3 Single-Marker Association Data for Risk Taking (AUCθ) Using the Paper-and-Pencil Task (n=1035)

Because AUCθ showed significant difference between males and females, we tested the interaction of SNPs and sex. None of the SNPs showed significant interaction with sex on risk-taking AUCθ after correction for multiple testing (Supplementary Table S2a). Furthermore, TPH2 htSNPs did not show significant interaction (epistasis) with 5HTTLPR, rs6295, or rs1800532 on AUCθ (Supplementary Table S2b).

Assuming a continuous trait such as AUCθ data, using an independent-individual study design and assuming an additive linear model relating the phenotype to genotype, we had a 90–93% power in our population (N=1035 and N=1188) to detect a polymorphism (with a minor allele frequency 10%) that could explain 1% of the variance of the trait at the 5% two-tailed significance level. To detect gene × sex interaction in the same model as above and assuming that sex explains 5% variance, we had a 91–95% power in our population. Furthermore, for epistasis we had a 90–94% power in our population if the other gene also explains 1% variance and the gene × gene interaction explains further 1% variance.

Using the computer task data, single-marker associations showed similar results with higher effect sizes to the pencil-and-paper task results (see Table 4).

Table 4 Single-Marker Association Data For Risk Taking (AUCθ) Using the Computer Task (n=69)

TPH2 Haplotype Association With AUCθ

The overall genetic effect of TPH2 haplotypes for the AUCθ data was significant and remained significant after permutation tests and correction for multiple testing (Table 5). The TPH2 haplotypes explained 1.9% variance in risk taking. The haplotypic association did not change after adding BSI Depression and Anxiety scores (permutated FDR corrected q=0.018) or BFI neuroticism scores (permutated FDR corrected p=0.018) as covariates in the analysis. These results suggest that there is a genetic effect on this measure of decision making, which is independent of current mood and anxiety state or neuroticism. Indeed, the TPH2 gene showed no significant haplotypic association with BSI depression (permutated FDR corrected q=0.337), BSI anxiety (permutated FDR corrected q=0.172), BFI neuroticism (permutated FDR corrected q=0.314), or IVE impulsivity (permutated FDR corrected q=0.174) scores.

Table 5 TPH2 Haplotypic Regression Results For Risk Taking Measured by AUCθ

The most common haplotype T,C,T,C,G,T,A was found in 49% of study participants. Beta values suggest that carriers were relatively risk avoidant and made significantly less risky decisions than the other haplotype carriers (permutated FDR corrected q=0.022). This effect was also significant for the risk-taking AUCθ calculated from the computer task (β=−2.619, SE=1.271, t=−2.06, permutated p=0.044, explained variance 5.9%, n=69, frequency of the T,C,T,C,G,T,A haplotype=45%).

DISCUSSION

Using haplotypic association, we found significant differences in risk-taking behavior on a performance gambling task between the most prevalent haplotype and the other haplotypes of the TPH2 gene. Subjects carrying the most prevalent T,C,T,C,G,T,A haplotype had lower AUCθ parameters suggesting that they are less likely to make risky decisions. Thus, our finding implies involvement of the serotonergic system, namely, the TPH2 gene, in decision making through altering willingness to take risks for reward. In addition, we tested whether differences in risk-taking behavior measured by our task were influenced by other serotonergic genes. We genotyped a further three serotonergic functional polymorphisms (5HTTLPR in serotonin transporter, rs6295 in 5HTR1A receptor, and rs1800532 in TPH1 genes) but could not demonstrate significant main effects of these polymorphisms or interaction with the TPH2 htSNPs on risk taking. However, we could not exclude that these genes modulate other aspects of decision making that were not measured by our task, such as punishment- or feedback-related processing, as previous studies reported (Jollant et al, 2007; Must et al, 2007; Blair et al, 2008; Homberg et al, 2008).

Accumulating evidence suggest that the more common allele/haplotype variants in the TPH2 gene are associated with lower cerebrospinal fluid 5-hydroxyindoleacetic acid concentrations (Zhou et al, 2005) and with decreased TPH2 mRNA expression (Chen et al, 2008; Chen and Miller, 2008; Haghighi et al, 2008; Lim et al, 2007). Thus, a considerable proportion of the population, who carry the wild-type (most ancient) variation of the TPH2 gene, express TPH2 at low level that results in decreased serotonin availability at the raphe nucleus and its projections, and also show risk aversion on the basis of our results. This situation is somewhat parallel to the ATD, which has been shown to reduce 5-HT levels to a functionally significant degree. However, human studies have not demonstrated a consistent effect of reducing 5-HT function using dietary ATD on probabilistic choice behavior: Rogers et al (1999) reported that after ATD subjects chose the most likely option less often, while Talbot et al (2006) demonstrated the opposite and our group could not find any significant difference (Anderson et al, 2003). The most plausible explanation for these discrepancies is that these studies used different tasks to measure risk taking and probably were not adequately powered to detect the effect of ATD on probabilistic choice. Indeed, on the basis of our results, variability in serotonin synthesis caused by the TPH2 gene only explained 1.9% (paper-and-pencil task)—5.9% (computer task) of the variance in risk taking. Another possible explanation is that the genetic make-up of the participants interfered with the effect of the ATD, as has been suggested in relation of the 5HTTLPR variant (Blair et al, 2008).

Our results suggest that there is a biological distinction to be made between impulsive and safe/risky decision making as measured with our task. Although impulsivity has usually been associated with an increased tendency to take risks, such as in DSM-IV-defined impulse control disorders (American Psychiatric Association, 2000), it only appears to predict some risky behaviors, whereas others are more linked to neuroticism and extraversion (Cooper et al, 2000). Impulsivity may be more related to inability to delay gratification and Crean et al (2000) demonstrated that delay discounting has a closer relationship to self-reported impulsivity measurements than probability discounting. Low 5-HT function has been associated with increased impulsivity (eg Dolan et al, 2001) in contrast to our finding that the common TPH2 variant, likely to be associated with reduced 5-HT function, is associated with risk aversion rather than risk taking when faced with a probabilistic choice. This highlights the complex role of 5-HT in behavior and mood and the need to carefully specify neuropsychological function when investigating the neurobiology of behavior.

Our task has face validity when considered in relation to mood in our subjects. Depression is characterized by reduced motivation to obtain reward and reduced enjoyment of it (Elliott et al, 1997; Forbes et al, 2006), whereas anxiety is associated with intolerance of uncertainty (Paulus et al, 2003; Paulus, 2007). Hence, subjects with depression and anxiety, which are frequently comorbid disorders, would be expected to be less motivated, indeed averse, to take more risks for a greater uncertain reward compared to controls (Smoski et al, 2008). This is consistent with the findings in our sample, in which higher depression, anxiety, and neuroticism ratings were associated with a lower tendency to take risks (AUCθ) and explained 1.6%–2.8% variance in this phenotype. However, our study also demonstrated that the effect of TPH2 on risk taking is independent of depression and anxiety symptom scores and also independent of neuroticism, thus indicating a separate mode of action. Furthermore, we could not demonstrate any association between haplotype variants of the TPH2 gene and BSI depression, BSI anxiety, BFI neuroticism, or IVE impulsiveness scores. This is in agreement with recent studies which failed to find any association between the TPH2 gene and depression/suicidal behavior (Mann et al, 2008; Must et al, 2007; Zill et al, 2007b; De Luca et al, 2006; Lopez et al, 2007b), but in contrast to earlier findings reporting such an involvement (Zill et al, 2004a, 2004b; Zhou et al, 2005; Van Den Bogaert et al, 2006; Haghighi et al, 2008; Lopez et al, 2007a; Ke et al, 2006). Therefore, the interpretation we favor is that the variations in the TPH2 gene might be linked to an endophenotype influencing risk-taking behavior that could be an additive or in some cases a vulnerability factor, for psychiatric disorders, for example, affective disorders or disturbed mood in impulse control or substance misuse disorders. As previous studies have investigated severe clinical populations (hospitalized major depressive patients, completed suicides with MDD), it is likely they represented extreme cases in which TPH2 haplotypes may have contributed to the phenotypic picture through altered risk-taking behavior, potentially leading to patient selection bias and an apparent association between depression/suicide and TPH2 haplotypes. A modulatory rather than causative role of serotonin and TPH2 in psychiatric disorders was suggested by TPH2 and TPH1/TPH2 double knockout (KO) animal data. TPH2 KO animals did not show apparent neuronal developmental abnormalities (Gutknecht et al, 2008), and even double KOs showed only mildly increased anxiety-like behavior (Savelieva et al, 2008).

The main strength of our study is that this is the first attempt to investigate the role of serotonergic genes in decision-making behavior in a large sample using a performance measure plus personality and symptom questionnaires. Furthermore, we were able to validate the questionnaire findings in a subgroup using a more extensive assessment including questionnaire and observer ratings and a computerized task. We have also recently reported that there is no difference in risky choice behavior between receiving real and hypothetical rewards when using this computerized task (Hinvest and Anderson, 2009). The findings in the subgroup largely replicated the findings in the main sample. The computerized task showed only a modest, although significant, correlation with the pencil-and-paper task, which is likely to be explained by methodological differences (amount to be won and different probabilities) and sampling error, as each indifference point was only assessed on one occasion in the pencil-and-paper task but on two occasions in the computerized version. Differences in genetic effect sizes (small effect sizes for the paper-and-pencil task and medium-large effect sizes for the computerized task, see Tables 3 and 4) probably are related to these methodological differences as the computerized task is likely to be a more sensitive tool to measure decision-making differences.

Our TPH2 finding has limitations, as we did not genotype recently identified functional variants in this gene. Therefore, we compared the LD pattern in our population and in the CEPH population, using our htSNPs (based on the International HapMap Project Phase I. June 2005 release) and recently genotyped functional SNPs (CEPH population, HapMap Project Phase II. November 2008 release), to interpret our results (Figure 2). We can conclude that our study satisfactorily captured the possible influence of functional variants that resides in the region around introns 5–8 and previously have been implicated in regulation of splicing, and therefore might influence the level of the active enzyme (Haghighi et al, 2008; Lim et al, 2007). Nonetheless, we have limited information about the promoter region of the TPH2 gene, which exhibits only moderate LD with our two most significant SNPs (rs6582078 and rs1352250) (Figure 2b) (Chen et al, 2008; Lin et al, 2007; Scheuch et al, 2007). Associations have been demonstrated with the TPH2 promoter region and traits related to emotional dysregulation, such as harm avoidance (Gutknecht et al, 2007; Reuter et al, 2007; Strobel et al, 2007) and amygdala activation during emotional processing (Brown et al, 2005; Canli et al, 2005). However, we did not find any association between haplotype variants of the TPH2 gene and depression, anxiety, neuroticism, or impulsiveness scores that might be influenced by functional variations in the TPH2 promoter region. Using denser htSNP cover on the full length of the TPH2 gene and including independent replication populations into the study would help us to answer these questions and support our new findings. The lack of independent replication population in this study is a major limitation and warrants further research in this field.

We genotyped functional polymorphisms in other serotonergic genes but could not replicate previous positive findings (Ono et al, 2002; Lesch et al, 1996; Lemonde et al, 2003), despite our study being sufficiently powered to identify genetic effect that explain 1% variance in personality and symptom scores (Juhasz et al, 2009). Our findings are however consistent with the recent scientific literature, as a large meta-analysis showed that several studies have failed to show association between anxiety-related traits and 5HTTLPR, except neuroticism measured by NEO (Munafo et al, 2009): it has been argued that genetic effects on neuroticism might be complex and smaller than 1% per genetic variants (Shifman et al, 2008).

In conclusion, our study is the first to support a role for the TPH2 gene as a trait factor in determining risky choice behavior as tested in a probabilistic choice task. We also demonstrated that risk-taking behavior is modified by state-dependent elements, such depression and anxiety, independently of the TPH2 gene effect. Thus, this biological marker may contribute to the expression of psychiatric phenotypes and possibly to a vulnerability to develop specific disorders. Our results support the importance of identifying quantitative traits, biological markers, or endophenotypes that might be more informative for genetic studies than heterogenic, complex disorders such as depression or complex traits such as impulsivity.