Genes, Brains, and Behavior: Imaging Genetics for Neuropsychiatric Disorders

Madness was not subject to systematic scientific scrutiny until the late 19th century, when two German-speaking psychiatrists developed diametrically opposed theories. Emil Kraepelin's opinion of madness was entirely biological. Sigmund Freud, on the other hand, formulated those as disorders of the mind principally related to early childhood experiences and psychic forces.¹ Kraepelin and Freud were both born within a few hundred kilometers from each other in 1856. But they never met—neither physically nor intellectually. Although Freud accepted that some sort of mental diseases could be physiological, and in his book of 1895 entitled Project for a Scientific Psychology, he proposed a program for a scientific psychology that would be firmly based in neurophysiology,² for decades the physiological and psychological theories of madness developed along parallel paths.¹

The diagnosis of madness did of course become much more differentiated. Indeed, by the 1970s, it was known that particular biochemical pathways in the brain underpinned some psychiatric diseases such as schizophrenia.³ However, it was only in the late 1990s that the accumulation of data obtained from brain sciences led to the question of whether the whole spectrum of mental disorders could be based on a biological framework.⁴

Despite this, psychiatric disorders are still primarily diagnosed on clinical grounds. According to the current version of the DSM-V,⁵ psychiatrists have to diagnose more than 300 different manifestations of mental disorders that they believe to represent the conditions of their patients by symptoms such as hallucination, delusion, or other fairly soft criteria. One seemingly intractable problem, for example, is that a main source of information about these symptoms may not be objective because they come from the patients themselves. However, many psychiatric illnesses can be reliably detected at the level of clinical signs and symptoms according to standard reference texts—most particularly the DSM and the ICD—for diagnosis that is based mainly on symptoms, yet these classifications often lack the objective tests to draw boundaries around a particular clinical state.^6–8 Therefore, the strength of the present phenomenological diagnostic guidelines derives from their reliability but not their validity.⁸

Since the publication of the sequence of the human genome in 2001,⁹ a rapid growth of different approaches to extract useful information from the genomic sequence have been evolved. These approaches included, but were not limited to, the analysis of genetic variation, gene expression, and gene products and their metabolic effects.¹⁰ As susceptibility for major psychiatric disorders show a strong degree of inheritance¹¹ and twin studies show that heritability for various aspects of cognition, temperament, and personality ranging from 40% to 70%,¹² these approaches trigger an enormous research effort to identify the biological bases of psychiatric disorders. Thus, many candidate genes have been investigated in relation to mental disorders such as schizophrenia or depression, but these investigations still remain at the theoretical level.^13,14 The two very big challenges are identifying the genes and identifying the phenotypes.¹⁵ Mental disorders are not associated by a single gene. Most types of mental disorders appear to be related to multiple genes that interact with each other and with the environment and vice versa. However, it is important to mention that, although genetic heterogeneity is generally accepted to be a complicating issue for psychiatric disorders, it is not an elusive problem for gene identification.¹⁶ Moreover, there has been relatively good progress on the identification of genetic biomarkers for other multifactorial diseases triggered by complex interaction of genes and environmental factors. This is mostly evident in cancer. For example the KRAS (Kirsten rat sarcoma viral oncogene homolog) mutation in colon and lung cancers, the EGFR (epidermal growth factor receptor) mutation in lung cancer, and the BRAF (v-raf murine sarcoma viral oncogene homolog B) mutation in colon cancer have led to the identification of molecular markers to be used for diagnosis.¹⁷

Conversely, although the new data from genome-wide association studies (GWASs) sound exciting,^18–22 genetic variants with a definitive effect on mental disorders such as schizophrenia phenotype are still unexplained.^23,24 In addition to the complexity of genetic factors, the complexity of psychiatric phenotypes and gene–environment interactions and the lack of biologically defined diagnostic criteria are among the reasons for this problem. Of these complexities, perhaps the most discriminative feature of psychiatric disorders compared with other multifactorial disorders is the complexity of psychiatric phenotypes, the phenotype bottleneck, reflected by the lack of biologically defined diagnostic criteria, which may manifest itself in poor results of genetic findings.

The Phenotype Bottleneck and Endophenotypes

The symptoms of mental diseases are usually behavioral complications such as visual or auditory hallucinations and disorganized speech or thought that are hard to study as precise, objective, and measurable phenotypes.²⁵ Also, some patients may exhibit a combination of such symptoms that cut across the existing disease boundaries defined by standard guidelines like the DSM. There is an overlap in genetic susceptibility across these disease boundaries. For example, schizophrenia, bipolar disorder, and schizoaffective disorders have been shown to share at least some genetic liability that cuts across the traditional binary classification of psychosis.²⁶ Moreover, for bipolar disorder, different twin studies have shown dramatically varying concordance rates depending on the diagnostic criteria, how the samples were ascertained, and how assessments were conducted.²⁷ Thus, one possible reason for difficulty in hunting susceptibility genes in psychiatry is that gene discovery studies have been searching for genetic effects on disease, but because disease diagnosis is not validated biologically, it does not provide a standard pool of subjects for top-down research. Nevertheless, the phenotype bottleneck is a well-known problem. Addressing this, the US National Institute of Mental Health (NIMH) has recently introduced its Research Domain Criteria Project, an alternative framework for research into psychiatric disorders.²⁸

Also, to be helpful, psychiatric phenotypes could be divided into very small traits in different domains, like cognition or emotion, so that they are specific enough to measure reliably and quantitatively. The nature of such traits—known as endophenotypes—may be electrophysiological, biochemical, endocrinological, or anatomical, but they must be heritable. The traditional endophenotype concept was proposed for schizophrenia genetics by Gottesmann and Shields in the early 1970s to facilitate gene discovery. An endophenotype was therefore expected to reduce genetic complexity and increase genetic effect size so that the complicated effects of gene interactions and distal genotype–phenotype associations could be sufficiently narrowed to clearly identify genetic risk factors.²⁹ Such requirements led to the establishment of certain criteria.²⁷ It is clear that only a few, if any, endophenotypes could actually fulfill these criteria, but a handful of robust endophenotypes that have been meeting these criteria have led to the identification of genomic sequences with disease relations. One example is the use of electrophysiological measurements as an endophenotype. Deficiencies in the electrophysiological response to auditory stimulation were used to identify an association of schizophrenia with the α7 nicotinic cholinergic receptors, a transmembrane protein of neurons.³⁰ Progressively, accumulating papers in the literature facilitate the standardization of the criteria, which increase the potential for the use of endophenotypes for psychiatric research²⁷ to identify new genes via this reversed genetics approach.¹³

Birth of Imaging Genetics: Candidate Gene Approach

In parallel with these developments, advanced brain imaging techniques have led to the emergence of a so-called forward genetics approach by which, instead of discovering genes, the already known genotypic feature, a candidate gene, is intended to be correlated to an intermediate phenotype. Imaging genetics or imaging genomics are increasingly used in this context, and the endophenotypes intended to be used for this purpose are usually called an intermediate phenotype—although the latter not necessarily being different than the former by means of essential selection criteria but differ in the purpose of its use. The intermediate phenotype strategy is based on the assumption that “gene effects at the level of the brain are a more direct effect of genetic variation than is complex behavior, and will show association in carriers of risk alleles even if the carriers show no clinical diagnostic characteristics.”¹³

Perhaps the most attractive examples for this trend come from the earlier attempts to identify the effect of single-nucleotide polymorphisms (SNPs) on psychiatric phenotypes determined by brain imaging. Because SNPs tend to be relatively stable genetically,³¹ they can serve as biological markers for pinpointing a mental disease in the human genome. Use of brain imaging proves to be especially powerful for addressing this. In this way, the effect of a known genetic diversity, which has been occurring naturally among human populations, is studied by brain imaging to determine whether one form of such a variation in human genes can cause a difference in the level of such psychiatric phenotypes and hence could make people more vulnerable to mental diseases. Of genetic functions influenced by SNPs, neurobiological studies increasingly validate neural correlates that are essentially involved in brain function during health and disease. These include a wide variety of protein groups encoded by these genes with diverse functions, ranging from neural growth and survival to signal transduction or, for example, brain-derived neurotrophic factor (BDNF),³² catechol-O-methyltransferase,³³ and serotonin transporter protein,³⁴ that are associated with psychotic or mood disorders. Using a candidate gene approach, such developments in understanding the biology of cellular mechanisms encoded by these genes consequently facilitated those imaging genetics studies that aid the correlation of genetic variation by psychiatric phenotypes. The rigorous results of preliminary studies already provide a proof of principle for an understanding of psychiatric phenotypes with biological underpinnings, although recent applications of GWASs to imaging genetics studies provided conflicting results. Regarding this, one group initially studied whether a SNP in the gene encoding for BDNF has an effect on the hippocampus and determined the genotype phenotype relations as a risk factor for a particular psychiatric condition.³⁵ It is already known that the size/morphology of the hippocampus varies among human populations and is estimated to be sufficiently heritable,³⁶ therefore suggesting this structure as an attractive candidate for genetic correlation based on the approach of an intermediate phenotype. The hippocampus is also known for being involved in the pathogenesis of several mental disorders including schizophrenia.^37–39 Reasonably, these data provide a strong basis for correlating changes seen in vivo to a potential genetic component when using a candidate gene approach. In this context, a widely expressed gene in the hippocampal formation encoding for a protein called BDNF, one of the neurotrophins that have been implicated in the neurobiology of schizophrenia, has been thought to be a good candidate. By coupling this information with an imaging genetics approach, Szeszko et al.³⁵ reported that a certain form of the BDNF gene variation (val66met) accounted for a greater proportion of the variance in the volume of the hippocampal formation (as assessed by MRI) in patients compared with healthy volunteers. Thus, the findings led to the idea that this effect (of genetic variance of BDNF) may be greater among patients compared with healthy volunteers through hippocampal pathogenesis and/or making the hippocampus more susceptible to environmental insults.³⁵ In contrast, recent data produced by a GWAS on hippocampal volumes failed to find the effects of BDNF val66met.⁴⁰

Conversely, such controversies are not always the case. A fruitful outcome is particularly possible when functional MRI (fMRI) is used to examine the neurobiological effect of a well-validated gene. Regarding this, the first study focusing on the ε4 allele of the apolipoprotein E gene (APOE-ε4) gene published in 2000 by Bookheimer et al.⁴¹ was followed by others (Table 1). The study of Bookheimer et al., which is also the first study of imaging genetics, provides evidence that patterns of brain activation during tasks requiring memory differ depending on the APOE-ε4 allele, the genetic risk of Alzheimer’s disease. Greater brain activity is detected in the hippocampus during memory tasks among APOE-ε4 carriers compared with noncarriers (Figure 1). Thus, the authors suggested that the hyperactivity could be a compensatory mechanism used to aid task performance. Several studies^42–44 have replicated these findings, whereas others have not.⁴⁵ However, factors such as the utilization of different cognitive tasks or the participants’ varying age across studies appear to be influential on this discrepancy.⁴⁶

In one of the earlier studies of imaging genetics, an extensive work using the candidate gene approach was reported by Meyer-Lindenberg et al.⁴⁷ They examined the brain activities of 126 healthy people with variants of the gene for catechol-O-metyltransferase, an important enzyme in the breakdown of neurotransmitters like dopamine. The gene is located on a region of chromosome 22 that has been linked to psychotic disorders such as bipolar disorder and schizophrenia. Dysfunction of the dorsolateral prefrontal cortex (DLPFC) has already been found using neuroimaging in schizophrenia. In this study, healthy people that carry variants of the catechol-O-metyltransferase gene associated with schizophrenia had less efficient processing in DLPFC during brain imaging, even when performance on the test was similar. This shows that disturbed processing in this region may mediate the genetic risk for schizophrenia associated with this gene by creating a vulnerable neural system that might decompensate under adverse environmental conditions. This particular function is one of many symptoms of schizophrenia, and a study showed that healthy people who can do this memory test without problems have lower activation of the DLPFC if they have the gene variant that is common among schizophrenics. They are compensating by using a different part of their brain, but it may also mean that they are at high risk of developing schizophrenia under particular environmental conditions. As suggested by Meyer-Lindenberg and Weinberger,¹³ this kind of study shows that genetic variance puts a person at higher risk by means of biological mechanisms converging to a specific brain circuitry: in this case, neostriatal–prefrontal and DLPFC–hippocampus circuitry that varies with the gene allele. Thus, there has been significant progress in describing the effect of common genetic polymorphisms on complex behavioral phenotypes and disease liability with the candidate gene strategy of imaging genetics. It seems like a well-replicated study of the candidate gene approach of imaging genetics might be particularly useful to address the neural system that is abnormal in a given illness.

However, the effect sizes of such studies appear to be a major source of discussion. For example, in 2002, Hariri et al.⁴⁸ published a paper in Science suggesting a strong association of the short allele of serotonin system–related genetic variant 5-HTTLPR (serotonin transporter–linked polymorphic region) with activity of the amygdala measured by fMRI. However, this has been strongly criticized due to a very high effect size (28%), which appears to conflict with the recent GWAS imaging data.⁴⁹

What can be deduced from this? Initial studies aimed to address the association of single neuroanatomical region with a few well-validated genetic variations in a small sample size. However, it appears that replication of initial studies of candidate genes is a must. A well-replicated study of such an approach might be useful to address the impact of genetic variation on a neural system that is abnormal in a given illness, despite the problem of false-positive findings.⁵⁰ One way to address this problem might be to create a genome-wide correction of associations with imaging phenotypes with sample sizes at least in the hundreds (this would be a noncandidate gene approach, which will be discussed later). However, statistical modeling is a serious challenge in any case: one has to be careful when considering the analysis of neuroimaging data because of the statistical modeling that can produce false-positive results. This is a well-known problem of multiple comparisons and selection biases that has long been debated.⁵¹ Moreover, analysis of neuroimaging data along with genome-wide data in large populations will naturally bring additional statistical challenges.

Stronger Candidates Supported by GWAS

Nevertheless, the candidate gene (or SNP) approach now has stronger candidates supported by recent GWAS data.^18–22 For example, the recently identified genome-wide significant genetic risk variants for schizophrenia²¹ and bipolar disorder²² naturally catalyze the efforts to explore their effects on cognition and brain structure. Reflecting this, the candidate gene approach has now gained a special momentum to find the neural correlates of genome-wide supported risk variants. Esslinger et al.⁵² found that the genome-wide supported risk variant for schizophrenia, rs1344706, specifically impacted on the connectivity between DLPFC and hippocampal formation, in parallel with data found in subjects with schizophrenia. In this study, 115 healthy participants were genotyped for rs1344706 and scanned by fMRI during an n-back task, related to heritable schizophrenia risk, DLPFC–hippocampal formation activity, and candidate gene variation. Task-related regional activation of DLPFC and hippocampal formation and coupling between these structures were assessed by functional connectivity analysis. Furthermore, the candidate gene approach with some convergent data from GWAS, expression studies, and animal models might be used to suggest a pathophysiological mechanism that may be accessible to drug targeting.⁵³ Thus, the current trend of the candidate gene studies of imaging genetics is to select a GWAS-supported risk variant and test it for a candidate phenotype (intermediate phenotype). This is the typical rationale of a candidate gene–candidate phenotype association. Elaborating on this, a candidate gene can be tested for a brain-wide phenotype. This new branch of candidate gene brain-wide associations^54,55 together with typical candidate gene–candidate phenotype associations is likely to be used for replication or validation of genetic effects (Table 1), although restriction of target genotypes and/or phenotypes may limit the capacity to identify important relationships.

Noncandidate Gene Approach

Another strategy of imaging genetics involve a noncandidate gene (or SNP) approach. Instead of replication/validation of a candidate gene association with a phenotype, it is a gene hunting strategy. It aims to associate either a candidate phenotype (measured by neuroimaging) with GWAS or adapts a hypothesis-free approach lacking any candidate gene or candidate phenotype: the genome-wide brain-wide association, provided by the comparison of the entire genome with the entire brain in large datasets. Table 1 provides examples for both of these strategies.

Several groups have already begun to use the former approach by which the volume of the hippocampus⁵⁶ or the caudate,⁵⁷ as well as activity measures of DLPFC during working memory,⁵⁸ hippocampal activity during episodic memory,⁵⁹ or amygdala reactivity during emotional faces task,⁶⁰ has been used as structural or functional phenotypes to compare across the whole genome. Typically, each genetic variant is independently tested for its association with the phenotype, which is called a mass univariate method. Such studies initially used the structural/functional brain measures as quantitative phenotypes obtained by MRI and compared them with 300,000–500,000 SNPs in relatively smaller numbers of subjects (60–400), whereas more recent publications have used much larger scales. Four papers^40,61–63 of such imaging GWASs published in Nature Genetics in 2012 demonstrate this trend. Regarding this, one study⁶¹ reported the genome-wide association meta-analyses and replication for mean bilateral hippocampal, total brain, and intracranial volumes from a large multinational consortium (N=21,151). Analysis of genome-wide SNP data and structural MRI data from healthy subjects suggests that (1) the intergenic rs7294919 variant, perhaps by regulating expression of the TESC gene encoding the protein called tescalcin, which interacts with the Na⁺/H⁺exchanger involved in the regulation of intracellular pH, cell volume, and cytoskeletal organization, was associated with decreased hippocampal volume (N=21,151); (2) rs10784502, located within HMGA2, encoding the high mobility group AT-hook 2 protein, was associated with intracranial volume (N=15,782); and (3) a suggestive association with total brain volume at rs10494373 within the DDR2 gene encoding discoidin domain-containing receptor 2 has been identified (N=6,500). Consequently, the effect of rs7294919 can now be the target of more focused studies either following the rationale of a candidate gene approach of imaging genetics, which has been published recently,⁶⁴ or by complementary studies such as mapping quantitative trait loci in the mammalian syntenic region, generating mouse knockouts of nearby genes, and investigating them. As a result, with the noncandidate gene approach, a synthesis of brain imaging with GWAS data can be used to discover new genetic variants.⁶⁵ However, the utility of such an approach depends on the replication of these data determined by the quality of imaging phenotypes, which is only partially achieved.^66,67

Genome-Wide Analysis Matched With Brain-Wide Scan

It has been suggested that the best strategy would involve the combination of genome-wide analysis with brain-wide scanning.⁶ Consequently, genome-wide whole brain approaches to investigate genetic effects on neuroimaging phenotypes for identifying quantitative trait loci have been described.^68–71 Technically speaking, these associations are quite challenging because of the multiple testing problem arising from possibly billions of statistical tests to be performed. The mass univariate method is frequently used for the analysis.^68,69 It is based on statistics at each single voxel, focused selectively on the highest association. However, it thus eliminates the possibility of “spatially extended signals that do not win any voxel.” Moreover, because of linkage disequilibrium, variants in the same haplotype block can be highly correlated.⁷¹ Addressing this, multivariate analysis based on principal component analysis has been proposed along with other multivariate techniques including penalized and sparse regression approaches with least absolute shrinkage and selection operator.⁷¹ Whereas principal component analysis may not be useful to identify the nonlinear interactions between genetic variants and associated phenotypes, some new multivariate techniques have been reported to have better statistical power. Accordingly, Ge et al.⁷¹ reported a multivariate method to compute the association tests between 448,294 SNPs and 18,043 genes in 31,662 voxels of the entire brain across 740 elderly subjects from the Alzheimer’s disease neuroimaging initiative. Structural MRI scans were analyzed using tensor-based morphometry to compute three-dimensional maps of regional brain volume differences compared with an average template image based on healthy elderly subjects. The subsequent analysis was based on a semiparametric regression model by which the complex interaction of SNPs is modeled nonparametrically using a least-squares kernel machines approach, whereas covariate effects, such as sex and age, are modeled parametrically (i.e., linearly). A number of genes were identified as having significant associations with volumetric changes. The most associated gene was GRIN2B, encoding the N-methyl-d-aspartate glutamate receptor NR2B subunit, and it affected both the parietal and temporal lobes in human brains, reflecting the successful replication of the results of a previous study.⁷⁰ Its role in Alzheimer’s disease has been widely acknowledged and studied, suggesting the validity of the approach. Figure 2 shows least-squares kernel machines results for the GRIN2B gene, which is associated with the parietal and temporal foci, both of which are known to be affected brain regions in Alzheimer’s disease. Table 1 lists some studies using mass univariate^68,69 or multivariate^70,71 analysis.

In this context, clearly increased sample sizes require multisite studies. Thus, there are several multicenter studies conducting genome-wide association meta-analyses. For example, the Enhancing Neuroimaging Genetics through Meta-Analysis Consortium,⁷² which analyzes brain measures and genotypes from multiple sites across the world to improve the power to detect genetic variants that influence the brain, has pooled genome-wide genotype and neuroimaging data to replicate and identify previously unknown variants explaining a small portion of the overall heritability of certain brain structure phenotypes. The consortium analyzed the MRI scans of the brain structures of 21,000 people including healthy individuals and individuals diagnosed with depression, anxiety, Alzheimer’s disease, and schizophrenia and compared them to genetic surveys linked to relevant diseases. Regarding this, the intergenic variant rs7294919 was associated with hippocampal volume, and rs10784502, located within the gene encoding for a transcriptional regulator called high mobility group AT-hook 2, was associated with intracranial volume.⁶¹

However, the need for increased sample sizes leading to multisite studies will likely cause additional confounds due to the variability of environmental conditions. Also, the small effect sizes and the extremely high numbers of statistical comparisons in this recent trend naturally bring additional statistical challenges. Addressing these, multivariate analysis by which models for voxel-wise genome-wide associations is proposed, but it requires further examination and replication.⁷¹ The utilization of different cognitive tasks and the participants’ varying age across studies appear to be highly influential on a given neuroimaging genetic study,⁴⁶ given the condition that replication is essential. In fact, replication is particularly important in the current climate, which questions the reliability of scientific findings in general^73,74 and neuroimaging^51,75 and imaging genetics⁴⁹ in particular.

Gene–Environment Interactions

A leverage of these advances will depend on understanding the genetic influences and their interactions with the environmental context within which they operate. It is clear that there are multiple genes, signaling pathways, and interactions besides other modulatory factors such as age and sex. Moreover, it is known that environmental factors are quite influential.¹⁴ Environmental factors are relevant even for highly heritable disorders such as schizophrenia, bipolar disorder, and autism and often have higher associated risk than common genetic variants for disorders such as anxiety and depressive disorders, with the exception of eating disorders.¹⁵ Thus, it has been a major challenge to elucidate the mechanisms underlying how genes and related neural processes operate in response to environmental adversity.

Perhaps the best demonstrative series of examples come from the studies of the human serotonin transporter gene variation called the serotonin transporter–linked polymorphic region (5-HTTLPR) and stress sensitivity.⁷⁶ As already supported by recent data obtained from a GWAS, an association between 5-HTTLPR and major depressive disorder has been repeatedly suggested.⁷⁷ It was already reported in 1996 that this polymorphism in the promoter region of the human serotonin transporter gene (SLC6A4; also known as 5-HTT) regulates gene expression.⁷⁸ Both human and animal research supports the evidence for the effect of environment on the stress sensitivity in relation to genetic variation of 5-HTT.^79,80 Following this, several other studies addressed the modulatory impact of the 5-HTTLPR genotype on the effect of environmental variables.^81–84 (Although some replications of these studies were not satisfactory, it is speculated that small effect size has created a statistical artifact.⁸⁵) Supporting this, a moderate effect of the 5-HTTLPR genotype in relation to early life stress and levels of rumination has been shown by brain imaging studies.⁸⁶ Taken together, these data represent a demonstrative example of how data from animal research, behavioral genetics, imaging genetic studies, and integrated environmental intervention are beginning to unravel the interaction of genetic variation and environmental factors. However, levels of rumination may not be a good candidate to measure the pure effect of environmental factors because resilience is likely to be affected by both genetics and environment. Moreover, the integration of environmental factors into imaging genetics in this study was obtained by the classification of subjects based on self-reports, which might threaten the validity of the research. One further step from here might be an analysis of the effect of genetic variation on the neural processing of variable psychosocial conditions. Indeed, a progress targeting this comes from the data developed by multinational teams with projects clustered in the form of European Network of Gene Environment Interactions. Among these efforts, the neuroimaging study conducted by Lederbogen et al.⁸⁷ identified distinct neural mechanisms for an established environmental risk factor: urbanicity. At the first stage, participants were scanned when they were taking cognitive tests that they were told to fail. Moreover, to increase the stress level, the participants were provided with negative feedback through headphones while they were checked for stress indicators such as high blood pressure. Although there was no significant difference in the mental health between the two groups of participants, the way these two groups dealt with the stress was different, and that difference is targeted to two regions: the amygdala and the perigenual anterior cingulate cortex. These two regions were associated with increased activity for urban upbringing.⁸⁷ Although this study is not devoid of its own limitations (as described in the original paper and thus will not be repeated here), it is a good example of exploring the effect of environmental risk factors (urbanicity) to analyze the neural processing of psychosocial conditions (stress). Thus, the integration of epidemiological associations to understand underlying neural mechanisms affected by environmental conditions seems feasible by the use of neuroimaging.

A further analysis of this type would be achieved by the refinement of psychosocial components in urban contexts besides to a careful consideration of the role of genetic variation. Two studies appear to fill these gaps. First, neuroimaging experiments, conducted by Zink et al. in 2008,⁸⁸ have already begun to unravel how unstable social status, as an environmental risk factor, is mediated by specific neural mechanisms. The study provided a profile of the neural correlates associated with processing social hierarchies in humans. Animal studies already showed that the more subordinate position in stable social hierarchies is associated with greater stress, but in dynamic unstable social hierarchies, the dominant position is exposed to most of the stressors because of the increased competition and instability.⁸⁹ Thus, to address the neural correlates of social hierarchies in humans, Zink et al. analyzed the brain activity patterns of 72 Caucasian, right-handed, healthy adults in response to different hierarchical settings (stable and unstable) during a task (e.g., multiple rounds of simple reaction time tasks or a simple visual discrimination task). The tasks allowed the players to perceive the hierarchy by the use of other players simulated in the computer screen. Interestingly, the social unstable hierarchical setting produced neural activations, which were not detected in other settings. These areas (thalamus, amygdala, and posterior cingulate) are those linked to social emotional processing and overlap with the findings of Lederbogen et al.⁸⁷ Second, another stream of research with a focus on oxytocin merges with these data. Research across species has demonstrated that oxytocin plays a key role in the regulation of social cognition and behavior. As reviewed by Meyer-Lindenberg and Tost,⁹⁰ it plays a crucial role in attachment, social exploration, and social recognition, as well as anxiety and stress-related behaviors. Among the relevant candidates for oxytocinergic signaling, the most extensively studied candidate is the gene coding for the oxytocin receptor.⁹⁰ Among socio-behavioral phenotypes, SNPs in oxytocin receptor have been associated with positive mood,^91,92 sensitivity to social support,⁹³ deficits in mother’s sensitivity to her children’s behavior,⁹⁴ and reward dependence.⁹⁵ Particularly, an SNP of unknown functionality in the third intron of oxytocin receptor has emerged as an interesting candidate: rs53576 (G/A). Psychiatric relevance of this genetic variant manifests itself by studies indicating that rs53576A is overtransmitted in some families to offspring with autism spectrum disorders.⁸⁸ To address the intermediate neural mechanisms of this phenomenon, Tost et al.⁹⁶ used multimodal neuroimaging in a large sample of healthy human subjects to identify structural and functional alterations in oxytocin receptor risk allele carriers and their link to temperament. They found activation and interregional coupling of the amygdala during the processing of emotionally salient social cues, depending on the oxytocin receptor rs53576 genotype. In addition, they reported structural alterations in key oxytocinergic regions, which predict the lower levels of reward dependence, specifically in male risk allele carriers. Thus, in this example, accumulating studies show that oxytocin receptor gene variation (rs53576) is associated with structural and functional alterations in limbic circuitry involving the amygdala, the hypothalamus, and the cingulate gyrus, suggesting that it influences social cognition and behavior by modulating neural circuits for processing of social information and negative affect. The findings presented here converge on a previously defined circuit, the amygdala cingulate circuit, modulated by the serotonin system–related gene variant, 5-HTTLPR, which is characterized as susceptibility to environmental adversity in relation to the risk of depression,^82,97,98 as already mentioned above. Taken together, these suggest that the amygdala cingulate circuit is a neural substrate, a form of neurobiological cluster, exposed to the effect of divergent molecular factors (i.e., of serotonergic and oxytocinergic systems impacting on environmental reactivity and acting as a functional unit that might play a major role for the pathophysiology of relevant psychiatric conditions). Thus, these data provide evidence for the utility of imaging genetics studies—when combined with data from different sources—in analyzing the effect of genetic variation and environmental factors on neural processing of social conditions as a susceptibility mechanism for psychiatric conditions, in this case via a mechanism dependent on the medial prefrontal cortex and amygdala.

Conclusions

Perhaps the best conclusion should start with a cancer analogy. Today, many biotech companies and start-ups already produce and sell genetic tests for molecular cancer diagnostics. Why is there no such progress in the psychiatric world? Naturally, one could argue that “cancer begins with a mutation within a single cell while diseases like schizophrenia likely involve a dysregulated neural network affected by multiple unidentified genes.” However, considering the fact that heritability of mental disorders such as schizophrenia¹¹ is higher than many multifactorial diseases such as cancer,⁹⁹ there must have been a better picture given the opportunities of genomic revolution. Despite advances in neuroscience and genetic research during the last two decades, the genetic or other biomarkers that reliably guide the diagnosis of psychiatric disorders are not satisfactory. Thus, psychiatry does not seem to benefit from the postgenomic era as desired. However, psychiatric circles need to demand it. It is a demand that can be addressed by doing things differently than before. This means a better consideration of what constitutes a phenotype. This means going beyond classical association studies by which heterogeneous patient groups selected by clinical symptoms are compared with controls. This means large-scale science. This means more interdisciplinary work by which multiple levels of analysis are achieved simultaneously. This is possible by using measures of neural activity, genetic variation, environmental context, and behavioral processes in the same experimental setup. As described above, imaging genetics have the potential to provide these opportunities. The candidate gene approach of imaging genetics can be used for validation/replication studies of genetic effects, whereas the hypothesis-free noncandidate gene approach appears to be a valuable tool for gene discovery. Large-scale research prompted by meta-analytic or multivariate imaging genomics initiatives and integration of environmental factors is another way to further the, thus far very limited, knowledge of mechanisms involved in the pathophysiology of neuropsychiatric disorders. Naturally, these developments should be considered in parallel with other large datasets. Thus, there is also a need for collecting systematic phenotype information in any given individual, at a series of different levels of environmental reactivity and resolution (molecules, cells, tissues, etc.) and then determining how these features can profitably be studied together. Hence, there is a long way to go. In this context, statistical modeling appears to be a major source of problem, along with the biases in meta-analysis and task paradigms that are conceptually linked to the genes under examination.

Nevertheless, the current progress refers to the emergence of a new period and thus holds the opportunities for the future that might lead to the identification of novel biological mechanisms underlying cognition and neuropsychiatric illness for better treatment opportunities.