* A. R. Hariri, Director, Developmental Imaging Genomics Program, Department of Psychiatry, University of Pittsburgh School of Medicine, Western Psychiatric Institute and Clinic, 3811 O'Hara Street, E-729, Pittsburgh, PA 15213–2593, USA. E-mail : firstname.lastname@example.org
Serotonin (5-hydroxytryptamine; 5-HT) is a potent modulator of the physiology and behavior involved in generating appropriate responses to environmental cues such as danger or threat. Furthermore, genetic variation in 5-HT subsystem genes can impact upon several dimensions of emotional behavior including neuroticism and psychopathology, but especially anxiety traits. Recently, functional neuroimaging has provided a dramatic illustration of how a promoter polymorphism in the human 5-HT transporter (5-HTT) gene, which has been weakly related to these behaviors, is strongly related to the engagement of neural systems, namely the amygdala, subserving emotional processes. In this commentary, we discuss how functional neuroimaging can be used to characterize the effects of polymorphisms in 5-HT subsystem genes on the response of neural circuits underlying the generation and regulation of mood and temperament as well as susceptibility to affective illness. We argue that in time, such knowledge will allow us to not only transcend phenomenological diagnosis and represent mechanisms of disease, but also identify at-risk individuals and biological pathways for the development of new treatments.
Converging evidence from animal and human studies using a myriad of experimental approaches has revealed that serotonin is a critical neurotransmitter in the generation and regulation of emotional behavior (Lucki 1998). Serotonergic neurotransmission has also been an efficacious target for the pharmacological treatment of mood disorders including depression, obsessive-compulsive disorder, anxiety and panic (Blier & de Montigny 1999). Moreover, genetic variation in several key 5-HT subsystems, presumably resulting in altered central serotonergic tone and neurotransmission, has been associated with various aspects of personality and temperament as well as susceptibility to affective illness (Murphy et al. 1998; Reif & Lesch 2003). Although many of these findings have led to novel insights about the neurobiology of complex behaviors and psychiatric disease, the enthusiasm for their collective results has been tempered by weak, inconsistent and failed attempts at replication.
The inability to substantiate these relationships through consistent replication in independent cohorts may simply reflect methodological issues such as inadequate control for non-genetic factors (e.g., age, sex and population stratification), insufficient power and/or inconsistency in the methods applied. Alternatively, and perhaps more importantly, such inconsistency may reflect the underlying biological nature of the relationship between allelic variants in serotonin genes, each of presumably small effect, and observable behaviors in the domain of mood and emotion that typically reflect complex functional interactions and emergent phenomena. Simply put, genes do not encode behaviors. Biology dictates that allelic variants will most likely have a functional impact on the cellular and molecular pathways associated with a gene. In turn, the resulting subtle cellular and molecular alterations may produce response biases at the systems level (i.e., in brain circuits that serve as obligatory intermediates of behavior), which ultimately may, or may not, impact upon overt behavior (Fig. 1). This inherently weak relationship between a genotype and a particular behavioral phenotype limits the power of genetic association in studies of human temperament and emotion.
Selection of appropriate phenotypes for association studies of complex behaviors is not only an issue for human research but also for animal models. For example, in a recent review of published studies involving 5-HT1A knockout mice and corticotropin-releasing hormone overexpressing mice, both of which have been considered as animal models of anxiety disorders, Groenink et al. (2003) concluded that physiological and not behavioral alterations are the most consistent findings in both animal models. This observation led them to speculate that ‘it may be possible that physiological processes are more sensitive to genetic alterations than behavioural responses’, and that ‘the behavioural repertoire of an animal to cope with an aversive situation may be more extensive, or the possibilities to adjust neuronal pathways underlying behavioural responses may be larger’ (Groenink et al. 2003). Given that the biological impact of a variation in a gene traverses an increasingly divergent path from cells to neural systems to behavior, this proposal hardly seems speculative.
Similarly, the response of brain regions subserving emotional processes in humans (e.g., amygdala, hippocampus, prefrontal cortex, anterior cingulate gyrus) may be more objectively measurable than the subjective experience of these same processes. Thus, functional polymorphisms in 5-HT genes weakly related to behaviors or psychiatric syndromes may be more strongly related to the integrity of these underlying neural systems. Functional neuroimaging (see Table 1 for summary of selected applications), by allowing for the rapid acquisition of hundreds of repeated measures of brain function within a single subject, offers a unique and potentially profound avenue to investigate gene effects by examining their influence on the activity of specific brain circuits during the processing of discrete stimuli or the performance of distinct behaviors. Moreover, the power afforded by these assays (i.e., both the statistical advantages of signal over-sampling and averaging and the biological advantages of prebehavioral phenotypes) may allow investigators to identify robust gene effects on brain information processing in considerably fewer subjects (tens vs. hundreds) than typical behavioral studies, and with much greater sensitivity (10–20 ×) than their behavioral counterparts. The most significant advances in our understanding of the relationships between genes, brain and behavior, however, will likely result from reciprocity between association studies using behavioral and neural phenotypes. Ideally, associations between gene variants and regional patterns of brain information processing will not only elucidate the biological mechanisms underlying previous links with behavior, but will also serve to direct attention to new behaviors that are mediated by brain systems influenced by the variants, and vice versa.
Table 1. : Neuroimaging applications to in vivo functional genomics
Functional magnetic resonance imaging (fMRI)
Task induced blood oxygen level dependent (BOLD) signal changes
Indirect assay of neuronal activity; temporal resolution > 1000 ms
Functional integrity of neural systems underlying distinct aspects of behavior
Positron emission tomography (PET)
Invasive; requires exposure to radioactivity; non-specificity of ligands; poor temporal resolution
Specific receptor and neurotransmitter binding levels; baseline measures of cerebral metabolism and blood flow
Electrical fields generated by neuronal activity
Only sensitive to cortical brain activity
Neural interactions with subsecond (10–100 ms) resolution
Magnetic fields generated by neuronal activity
Relatively poor source localization
Dynamic interactions of both cortical and subcortical circuits
The application of functional neuroimaging techniques to the identification of genetic effects on brain information processing, or what we have called imaging genomics, is at its core a form of genetic association analysis (Hariri & Weinberger 2003). However, rather than focusing on a behavioral phenotype or disease diagnosis, imaging genomics targets the physiological response of discrete brain circuits during specific forms of information processing (e.g., visual, auditory, cognitive, emotional). Ideally, imaging genomics studies will be targeted at genes with clearly defined functional polymorphisms that are associated with specific physiologic effects at the cellular level in distinct brain circuits. In the absence of detailed functional variants, genes of interest should be associated with identified single nucleotide polymorphisms (SNP) or other allele variants with likely functional implications involving circumscribed neuroanatomical systems.
As with all genetic association analyses, imaging genomics studies must account for systematic non-genetic differences between genotype groups that could either obscure or masquerade for a true gene effect. Such studies need to control for various demographics such as age, sex, IQ, population heterogeneity and substructure (‘stratification’) as well as environmental factors such as illness, injury or substance abuse. In addition, imaging genomics studies must consider variability in imaging task performance in interpreting potential gene effects because typical imaging signals (e.g., the blood oxygen level dependent (BOLD) response in functional magnetic resonance imaging (fMRI)) are linked pari passu with performance.
Furthermore, the experimental tasks employed in imaging genomics studies must maximize sensitivity and inferential value, as the interpretation of potential gene effects depends on the validity of the information processing paradigm. These tasks should consistently engage circumscribed brain circuits, produce robust signals and demonstrate variance across subjects in order to allow for the determination of gene effects. For these reasons, initial imaging genomics studies of specific allelic variants may benefit from traditional blocked design paradigms that maximize the detection efficacy of neuroimaging assays (Birn et al. 2002). Event-related fMRI designs, which allow for a more detailed estimation of hemodynamic onset, delay and amplitude for independent events, can be applied once genetic effects on specific brain regions have been identified using more traditional approaches.
Fear and the amygdala
The amygdala is a central brain structure in the generation of both normal and pathological emotional behavior, especially fear (LeDoux 1997). Furthermore, the amygdala is densely innervated by serotonergic neurons and 5-HT receptors are abundant throughout amygdala subnuclei (Azmitia & Gannon 1986). Thus, the activity of this subcortical region may be uniquely sensitive to alterations in serotonergic neurotransmission and any resulting variability in amygdala excitability is likely to contribute to individual differences in emergent phenomena such as mood and temperament.
A substantial number of functional neuroimaging studies have revealed that human facial expressions, especially those depicting fear and anger, are robust and consistent provocateurs of amygdala activation (Davis & Whalen 2001; Zald 2003). The consistency of amygdala engagement in response to fearful or threatening facial expressions likely reflects the intrinsic survival value of such stimuli (Darwin 1998), as conspecifics represent both the greatest threat to our safety (signaled by angry facial expressions) and the most valuable source of information about danger in our environment (signaled by fearful facial expressions). Functional neuroimaging studies of genetic variation in serotonergic neurotransmission, as highlighted below, can capitalize on this inherent sensitivity by utilizing angry and fearful facial expressions as primary stimuli, and thus maximize the ability to detect robust amygdala activation in all subjects and ensure that substantial variance in this response exists across subjects.
In addition, functional neuroimaging has provided important advances in our understanding of how different brain regions interact to regulate emotional behavior. Specifically, a series of landmark imaging studies have revealed that engagement of the prefrontal cortex, through a variety of cognitive tasks, results in the modulation, and possibly inhibition, of the amygdala (Beauregard et al. 2001; Hariri et al. 2000; Keightley et al. 2003; Lange et al. 2003; Nakamura et al. 1999; Narumoto et al. 2000). Collectively, these results suggest that the dynamic interactions of the amygdala and prefrontal cortex may be critical in regulating emotional behavior (Hariri et al. 2003). While fMRI, especially that utilizing event-related strategies, allows for a rough estimation of how these and other brain regions interact during the processing of emotional material, imaging techniques that offer far superior temporal resolution, such as electroencephalogram (EEG) and magneto-encephalography (MEG), will be needed to more carefully explore the dynamic interplay of these neural circuits and the effects that variations in 5-HT subsystem genes have on this functional connectivity.
Genetic variation in the serotonin transporter
The 5-HT transporter plays an important role in serotonergic neurotransmission by facilitating reuptake of 5-HT from the synaptic cleft. In 1996 a relatively common polymorphism was identified in the human 5-HTT gene (SLC6A4) located on chromosome 17q11.1-q12 (Heils et al. 1996). The polymorphism is a variable repeat sequence in the promoter region (5-HTTLPR) resulting in two common alleles: the short (s) variant comprised of 14 copies of a 20–23 base pair repeat unit, and the long (l) variant comprised of 16 copies. In populations of European ancestry, the frequency of the s allele is approximately 0.40, and the genotype frequencies are in Hardy–Weinberg equilibrium (l/l = 0.36, l/s = 0.48, s/s = 0.16). These relative allele frequencies, however, can vary substantially across populations (Gelernter et al. 1997).
Following the identification of this polymorphism, Lesch and colleagues demonstrated in vitro that the 5-HTTLPR alters both SLC6A4 transcription and the level of 5-HTT function (Lesch et al. 1996). Cultured human lymphoblast cell lines homozygous for the l allele have higher concentrations of 5-HTT mRNA and express nearly two-fold greater 5-HT re-uptake in comparison to cells possessing either one or two copies of the s allele. Subsequently, both in vivo imaging measures of radioligand binding to 5-HTT (Heinz et al. 2000) and postmortem calculation of 5-HTT density (Little et al. 1998) in humans reported nearly identical reductions in 5-HTT binding levels associated with the s allele as observed in vitro. These data are consistent with β-CIT SPECT studies in humans and non-human primates reporting an inverse relationship between 5-HTT availability and CSF concentrations of 5-hydroxyindoleacetic acid (5-HIAA), a 5-HT metabolite (Heinz et al. 1998; Heinz et al. 2002), and indicate that the 5-HTTLPR is functional and impacts on serotonergic neurotransmission.
In their initial study, Lesch and colleagues also demonstrated that individuals carrying the s allele are slightly more likely to display abnormal levels of anxiety in comparison to l/l homozygotes (Lesch et al. 1996). Since their original report, others have confirmed the association between the 5-HTTLPR s allele and heightened anxiety (Du et al. 2000; Katsuragi et al. 1999; Mazzanti et al. 1998; Melke et al. 2001), and have also demonstrated that individuals possessing the s allele more readily acquire conditioned fear responses (Garpenstrand et al. 2001) and develop affective illness (Lesch & Mossner 1998) in comparison to those homozygous for the l allele. Recent studies utilizing pharmacological challenge paradigms of the 5-HT system suggest that these differences in affect, mood and temperament may reflect 5-HTTLPR driven variation in 5-HTT expression and subsequent changes in synaptic concentrations of 5-HT (Moreno et al. 2002; Neumeister et al. 2002; Whale et al. 2000). Furthermore, reduced 5-HTT availability, which presumably exists in 5-HTTLPR s allele carriers, has been associated with mood disturbances including major depression (Malison et al. 1998) and the severity of depression and anxiety in various psychiatric disorders (Eggers et al. 2003; Heinz et al. 2002; Willeit et al. 2000).
Although the potential influence of genetic variation in 5-HTT function on human mood and temperament was bolstered by subsequent studies demonstrating increased anxiety-like behavior and abnormal fear conditioning in 5-HTT knockout mice (Holmes et al. in press), the underlying neurobiological correlates of this functional relationship remained unknown. Because the physiologic response of the amygdala during the processing of fearful stimuli may be more objectively measurable than the subjective experience of emotionality, the 5-HTTLPR may have a more obvious impact at the level of amygdala biology than at the level of individual responses to questionnaires or ratings of emotional symptoms.
In 2002, we utilized fMRI to directly explore the neural basis of the apparent relationship between the 5-HTTLPR and emotional behavior (Hariri et al. 2002b). Specifically, we hypothesized that 5-HTTLPR s allele carriers, who presumably have relatively lower 5-HTT function and higher synaptic concentrations of 5-HT (analogous to the 5-HTT knockout mice) and have been reported to be more anxious and fearful, would exhibit greater amygdala activity in response to fearful or threatening stimuli than those homozygous for the l allele, who presumably have lower levels of synaptic 5-HT and have been reported to be less anxious and fearful (analogous to the contrasting wild type mice).
In our initial study, subjects from two independent cohorts (n = 14 in each) were divided into equal groups based on their 5-HTTLPR genotype, with the groups matched for age, sex, IQ and task performance. During scanning, the subjects performed a simple perceptual processing task involving the matching of fearful and angry human facial expressions. Importantly, this task has been effective at consistently engaging the amygdala across multiple subject populations and experimental paradigms (Hariri et al. 2000; Hariri et al. 2002a; Hariri et al. 2002c; Tessitore et al. 2002).
Consistent with our hypothesis, we found that subjects carrying the less efficient 5-HTTLPR s allele exhibited significantly increased amygdala activity in comparison with subjects homozygous for the l allele (Hariri et al. 2002). In contrast, there were no significant group differences in subjective behavioral measures of anxiety-like or fear-related traits. In fact, the difference in amygdala activity between 5-HTTLPR genotype groups in this study was nearly fivefold, accounting for 20% of the total variance in the amygdala response during this experience, an effect size 10-fold greater than any previously reported behavioral associations. This finding suggests that the increased anxiety and fearfulness associated with individuals possessing the 5-HTTLPR s allele may reflect the hyper-responsiveness of their amygdala to relevant environmental stimuli. Most recently, we have replicated the finding of amygdala hyper-excitability in 5-HTTLPR s allele carriers in a third independent cohort of 42 healthy subjects (unpublished data). As with our initial report, 5-HTTLPR genotype in this new cohort accounted for nearly 22% of the total variance in amygdala activity in the absence of significant differences in subjective behavioral measures.
The results of these studies are striking not only because they provide the first evidence for a genetically driven difference in the response of brain regions underlying emotional behavior, but also because these differences at the neurobiological level were marked in relatively small sample populations in the absence of significant differences in behavioral measures. Moreover, the imaging genomics results provide an elucidation of a potential biological mechanism for the genetic association of the 5-HTTLPR with vague psychiatric disturbances, including various dimensions of anxiety and neuroticism. Future studies, in addition to verifying the importance of amygdala reactivity in mediating the potential behavioral impact of the 5-HTTLPR, should determine the specificity of this amygdala effect by utilizing additional biologically salient stimuli such as novel, neutral or ambiguous faces as well as non-face stimuli (e.g., complex visual scenes depicting fear or threat).
Mechanisms of 5-HTTLPR effects on the amygdala
While the finding of amygdala hyper-excitability in 5-HTTLPR s allele carriers using fMRI provides a potential breakthrough in our understanding of the neurobiological underpinnings of abnormal mood and affect associated with variation in 5-HT signaling, the intrinsic mechanisms by which this brain response bias emerges remain poorly understood. The application of additional functional neuroimaging modalities, most notably radioligand-specific positron emission tomography (PET), may provide a powerful tool for elucidating these pathways. For example, radioligands of increasing specificity and flexibility have been developed to probe both 5-HT synthesis (Hagberg et al. 2002) and 5-HTT availability (Meyer et al. 2001; Sandell et al. 2002; Szabo et al. 1999; Wilson et al. 2000), and could be utilized to determine and substantiate presumably differential transporter and 5-HT levels based on 5-HTTLPR genotype.
5-HT receptor subtype alterations
The 5-HT1A specific PET radioligand, [11C]WAY-100635 (Mathis et al. 1994), may be of particular interest in exploring the mechanism of 5-HTTLPR driven variation in amygdala activity. Studies in 5-HTT knockout mice have revealed that the elevated stress, fearfulness and anxiety common in these animals may be related to reductions in 5-HT1A autoreceptor density resulting in desensitized inhibitory feedback to ascending 5-HT neurons of the raphe (Li et al. 1999; Li et al. 2000). Similarly, 5-HTTLPR s allele carriers may also have relatively down regulated 5-HT1A receptors in comparison to l/l homozygote individuals, as a result of relatively greater lifelong synaptic 5-HT tone. Thus, reduced [11C]WAY-100635 binding in affective brain circuits (e.g., amygdala, prefrontal cortex) of 5-HTTLPR s allele carriers would provide compelling evidence that amygdala hyper-excitability in these subjects may reflect, at least in part, desensitized inhibitory feedback to 5-HT neurons.
In fact, PET studies have already demonstrated that elevated levels of brain 5-HT are associated with reduced 5-HT1A binding in rats (Hume et al. 2001; Zimmer et al. 2002), and that lower 5-HT1A binding is associated with alterations in mood and emotional behavior in human subjects (Parsey et al. 2002). While the specific contributions of pre- and postsynaptic 5-TH1A down-regulation remains unclear, these various findings are consistent with evidence in human subjects that 5-TH1A agonists have anxiolytic effects and that serotonin specific reuptake inhibitors (SSRIs) may augment 5-HT1A signaling, as well as with the heightened levels of anxiety common in 5-HT1A knockout mice (Gross et al. 2002).
Functional interactions between multiple gene variants may also contribute to the underlying mechanism of the amygdala hyper-excitability associated with the 5-HTTLPR s allele. Effects of gene/gene interactions on development, physiology and behavior have already been demonstrated in double and triple knockout mouse models (Salichon et al. 2001; Sora et al. 2001; Upton et al. 2002). For example, abnormal development of somatosensory projection pathways found in the 5-HTT and monoamine oxidase A (MAOA) knockout as well as the 5-HTT/MAOA double knockout is normalized in the 5-HTT/MAOA/5-HT1B receptor triple knockout, suggesting that this neurodevelopmental anomaly may be mediated by excess 5-HT neurotransmission through the 5-HT1B receptor (Salichon et al. 2001). In addition, while the dopamine transporter (DAT) knockout exhibits normal cocaine-reinforced place preference, the 5-HTT/DAT double knockout does not display this normal preference, demonstrating that intact 5-HT neurotransmission is critical for the reinforcing properties of cocaine (Sora et al. 2001).
5-HT uptake by dopamine neurons via the DAT has also been reported in the 5-HTT knockout mouse, indicating that intact monoamine transporters can provide compensatory monoamine reuptake (Zhou et al. 2002). The existence of a polymorphism in the human DAT gene, which has been associated with altered levels of transcription and in vivo availability of DAT (Bannon et al. 2001), raises the intriguing possibility that inheritance of the high-functioning DAT variant, by providing compensatory 5-HT uptake, may reduce synaptic 5-HT levels and normalize amygdala activity in 5-HTTLPR s allele carriers. By allowing for the characterization of robust gene effects on brain in relatively small samples, imaging genomics provides a powerful avenue to investigate how such gene/gene interactions may influence amygdala activity and subsequently, emotional behavior.
Recently, murine, primate and human studies have revealed profound gene/environment interactions on brain and behavior that are likely to play a major role in the functional outcome of genetic polymorphisms. For example, in 5-HT1A knockout mice, which manifest an anxiety phenotype as adults, conditional rescue of the 5-HT1A receptor during an early developmental window effectively blocks the emergence of the anxious phenotype in adulthood (Gross et al. 2002). Receptor expression induced at other times during life has no such effect. In rhesus macaques, an analogous variant to the human 5-HTTLPR s allele is associated with abnormal temperament and levels of 5-HIAA in peer-reared but not mother-reared monkeys (Bennett et al. 2002; Champoux et al. 2002). Finally, two recent behavioral association studies have reported that both the 5-HTTLPR s allele (Caspi et al. 2003) as well as a functional variant in the human MAOA gene (Caspi et al. 2002) are associated with disturbed emotional and social behavior but only in young adults who were abused as children. These studies highlight the important impact of early experience, environment and development on the functional consequences of genetic polymorphisms. Again, imaging genomics, by allowing for the identification of gene effects on brain in populations with well-characterized developmental histories or through prospective longitudinal studies, provides a tool with which these important interactions can be studied.
Other candidate 5-HT synthesis genes
Of the many allelic variants identified in serotonergic genes, those representing critical bottlenecks in serotonin synthesis, reuptake and metabolism may lead to the most dramatic alterations in serotonergic neurotransmission and consequently, the functional integrity of affective brain circuits (Table 2). For example, a promoter polymorphism in the human MAOA gene, responsible for the intracellular catabolism of 5-HT to 5-HIAA, has been associated with altered transcriptional activity and heightened levels of aggression and impulsivity in men (Caspi et al. 2002; Manuck et al. 2000; Manuck et al. 2002). Similarly, a polymorphism of putative functional significance in the human gene for aromatic L-amino acid decarboxylase (AADC), responsible for the conversion of 5-hydroxytryptophan to 5-HT, has been associated with bipolar disorder (Borglum et al. 1999). Finally, a frequent SNP in the human gene for tryptophan hydroxylase (TPH), the rate-limiting enzyme responsible for catalyzing the oxygenation of tryptophan to 5-hydroxytryptophan, has been associated with increased risk of suicide, impulsivity, aggression and alcoholism (Abbar et al. 2001 ; Nielsen et al. 1994; Nielsen et al. 1997; Nielsen et al. 1998; Nolan et al. 2000). However, recent work in the TPH knockout mouse has revealed the existence of a second TPH gene (TPH2), located on chromosome 12, that is more highly expressed in the murine brain, and which in fact may be responsible for 5-HT synthesis in the CNS (Walther et al. 2003). Such novel genetic findings illustrate the need to exercise caution in the selection of candidate genes as our knowledge of the genomic architecture continues to expand.
Because all of these candidate genes represent critical steps in 5-HT biosynthesis, genetic variation resulting in their altered activity could significantly impact central serotonergic tone and neurotransmission. These resulting alterations may impact the activity of brain regions, such as the amygdala, underlying the generation and regulation of emotional behavior. Thus, the effect of any specific polymorphism may be exaggerated or minimized based on variation background in other genes both within and beyond the 5-HT pathway. Imaging genomics, by virtue of its power to detect highly significant gene effects on brain information processing in relatively small samples, provides an ideal avenue through which a genetic profile, reflecting the additive, or perhaps even multiplicative, interactions of these and other 5-HT subsystem gene variants, underlying the diathesis of heightened anxiety, fearfulness and affective illness, can be identified.