INTRODUCTION

Adenosine is involved in numerous processes and pathways and plays an important role as homeostatic regulator and neuromodulator in the nervous system (Cunha, 2001; Fredholm et al, 2005). It exerts its functions via the G protein-coupled receptors A1, A2A, A2B, and A3, of which particularly the first one is widely distributed in the central nervous system with high expression rates in cortical, thalamic, hippocampal, and striatal neurons (Ribeiro et al, 2002; Wei et al, 2011). The adenosine A1 receptor (A1AR) is not only involved in physiologic conditions such as sleep, arousal, memory, cognition, and anxiety, but also in pathologic conditions like epilepsy, stroke, and neuroinflammation (Ribeiro et al, 2002; Wei et al, 2011). Given this broad impact, an increasing number of studies focused on imaging the distribution and availability of cerebral A1ARs in vivo by positron emission tomography (PET; for a recent review see Paul et al, 2011a) using [18F]CPFPX as radioligand (Holschbach et al, 2002; Bauer et al, 2003; Meyer et al, 2004, 2005). Thereby changes in the binding potential of [18F]CPFPX could be detected in different cerebral regions indicating, for example, pathologic conditions like glioma invasion (Bauer et al, 2005). Further, physiologic variation of A1AR binding was reported, such as an age-dependent A1AR loss (Meyer et al, 2007), an A1AR upregulation after prolonged wakefulness (Elmenhorst et al, 2007a) as well as an influence of caffeine on [18F]CPFPX binding (Elmenhorst et al, 2012).

Besides A1ARs, A2A adenosine receptors (A2AAR) are recognized as widely distributed in the brain (Gomes et al, 2011). Remarkably, in several brain regions—particularly in the striatum—A1ARs interact directly with A2AARs via coexpression and heteromerization. Thus, a functional cross-talk exists between both receptors regulating adenosinergic neuromodulation and glutamatergic neurotransmission (Cunha, 2001; Ferré et al, 2007; Casadó et al, 2010). A balanced activation of inhibitory A1ARs and facilitatory A2AARs seems important for physiological neuromodulation, whereas deviation appears in conjunction with brain disorders such as ischemia, epilepsy, Huntington’s disease, Alzheimer’s disease, schizophrenia, bipolar disorders, depression, and anxiety disorders (Gomes et al, 2011). Accordingly, genetic knockout studies underline the important role of A1AR together with A2AAR in the central nervous system and particularly in relation to different neurological and psychiatric disorders (Wei et al, 2011). It is unknown, however, to the best of our knowledge, if small genetic variants like single nucleotide polymorphisms (SNPs) might play a role in modulating the distribution and availability of the A1AR, as it is described to be the case for several other neurotransmitter systems such as the serotonergic, dopaminergic, and monoaminergic systems (Willeit and Praschak-Rieder 2010).

The human genes encoding the receptor types A1AR (ADORA1) and A2AAR (ADORA2A) both contain a set of SNPs, which have been intensely investigated over the years specifically in regard to pathophysiologic alterations. Association was reported, for example, with infarct size in patients with ischemic cardiomyopathy (Tang et al, 2007), post-traumatic seizures (Wagner et al, 2010), migraine with aura (Hohoff et al, 2007), schizophrenia (Gotoh et al, 2009), blood-injury phobia (Hohoff et al, 2009), panic disorder (Deckert et al, 1998; Hohoff et al, 2010), and autism spectrum disorder (Freitag et al, 2010). Aside from these findings studies in healthy subjects were conducted reporting association of several of these variants with amphetamine-induced anxiety (Hohoff et al, 2005), caffeine-induced anxiety (Alsene et al, 2003; Childs et al, 2008, Rogers et al, 2010), and caffeine effects on sleep (Rétey et al, 2007; Bodenmann et al, 2012). The strong overlap of physiologic and pathophysiologic conditions related to A1AR and A2AAR modulation on the one hand and to ADORA1 and ADORA2A polymorphisms on the other hand leads to the hypothesis that genetic variants in both receptor genes might have an impact on A1AR distribution and availability in different brain regions.

Therefore, the present pilot study investigated whether or not an influence of ADORA1 and ADORA2A SNPs exists on in vivo A1AR binding assessed with [18F]CPFPX PET in brains of healthy humans.

MATERIALS AND METHODS

Subjects

Altogether 28 normal control individuals were included in the study (mean age±SD 31.8±10.7, range 21–63, all males) after giving written informed consent for imaging procedure as well as for blood collection and genetic testing. Subjects completed standardized questionnaires regarding their medical history, current health status, and sleeping behavior before they underwent medical examination. Candidates were excluded if they fulfill any of the following criteria: neurological or psychiatric disorders, head trauma, systemic diseases, substance abuse, and drugs interfering with adenosine receptors (eg, theophylline). All procedures were approved by the Ethics Committee of the Medical Faculty of the University of Duesseldorf and the German Federal Office for Radiation Protection.

Participants were asked to avoid caffeine consumption 24 h prior to PET scanning. In order to provide a measure of caffeine consumption, we estimated the number of cups of caffeinated coffee per day. Other caffeinated drinks, such as caffeine-containing soft drinks (1 liter corresponding to 1.5 cups) and black tea (2 cups corresponding to 1 cup), were also taken into account (Mandel, 2002). Information on average sleep duration per night and the sleep duration of the night preceding PET scanning was obtained. Each subject underwent also a high-resolution three-dimensional T1-weighted magnetic resonance imaging (MRI; Magnetom VISION 1.5T, Siemens, Erlangen, Germany). MR images were used to exclude structural abnormalities and to define volume-of-interest (VOI).

PET Procedures

PET acquisitions were performed following a previously described protocol (Meyer et al, 2004), on a Siemens ECAT EXACT HR+ scanner (Siemens-CTI, Knoxville, TN, USA). For the analysis, only the first 60 min of dynamic acquisitions, starting with [18F]CPFPX injection, were used. [18F]CPFPX was synthesized as previously described (Holschbach et al, 2002). Experiments were performed at tracer doses, with a mean specific radioactivity at injection time was 161.9±81.0 GBq/μmol (range 46.4–333.2 GBq/μmol) and mean injected radioactivity was 288.8±26.9 MBq (224–317 MBq), resulting in a mass of injected CPFPX of 3.8±3.0 nmol on average (0.9–14.7 nmol). Blood samples were taken for blood volume correction of brain ROIs by subtracting 5% of the total blood activity from the time activity curves (TACs).

MR images were oriented to the anterior commissure/posterior commissure line using the VINCI software (Cologne, Germany). In order to coregister the PET to the MRI, a summed PET image (5.5–60 min p.i.) was used and the derived parameters applied to all individual frames of the dynamic acquisition.

For masking of gray matter in PET images the MRI was segmented (SPM5, Statistical Parametric Mapping, Wellcome Department of Cognitive Neurology, London, UK). Next, the coregistered MRI scan was used to delineate VOIs by an automatic procedure based on probability maps (PVElab-20080317; Svarer et al, 2005). Based on these VOIs, side-averaged, decay- and blood-volume-corrected TACs were extracted from the dynamic gray matter-masked PET sequences (PMOD, Version 2.6, PMOD Group, Zurich, Switzerland, www.pmod.com) for the following 19 cerebral regions: amygdala, anterior cingulate gyrus, caudate, dorsolateral prefrontal cortex, entorhinal cortex, hippocampus, insula, middle and inferior frontal gyrus, middle and inferior temporal gyrus, occipital cortex, orbitofrontal cortex, parietal cortex, posterior cingulate gyrus, putamen, sensorimotor cortex, superior frontal gyrus, superior temporal gyrus, thalamus, and ventrolateral prefrontal cortex. The binding potential BPND of [18F]CPFPX in cerebral tissue was assessed by Logan’s non-invasive graphical analysis (Logan et al, 1996) as previously described (Meyer et al, 2007). The cerebellum was used as a reference region in the present study because of the low abundance of A1ARs in this region in humans (Bauer et al, 2003). The slope was calculated from 20 min onwards, as previous analyses revealed that this time point corresponds to the time of linearization in [18F]CPFPX non-invasive graphical analyses (Meyer et al, 2007). For an illustration of main image analysis steps see Figure 1.

Figure 1
figure 1

The figure represents the main image analysis steps: the individual MRI (panel (a)) and the summed PET image (fused with the MRI in panel (b)) were aligned and coregistered. Gray matter-masked VOIs (overlaid on the MRI in panel (c)) were used to extract regional TACs, to which Logan’s non-invasive graphical analysis was applied to calculate regional BPND values (see Materials and Methods section for details).

PowerPoint slide

Genotyping

Blood samples were taken from all subjects for DNA isolation using standard methods (FlexiGene DNA Kit, Qiagen). Altogether 18 SNPs were selected to cover ADORA1 and ADORA2A based on tagging information (International HapMap Project, www.hapmap.org), functional potential (UCSC Genome Browser, http://genome.ucsc.edu) and own previous association studies (eg, Deckert et al, 1998; Alsene et al, 2003; Hohoff et al, 2005, 2007, 2009, 2010; Childs et al, 2008, Freitag et al, 2010; Rogers et al, 2010; Bodenmann et al, 2012), as illustrated in Figure 2. Genotypes were determined as described previously (refer Deckert et al, 1998; Freitag et al, 2010) and as summarized in Supplementary Table S1. To control for the risk of genotyping errors, about one-third of randomized subjects were genotyped additionally by independent assays (single-strand conformation analysis, direct sequencing, or custom TaqMan SNP genotyping assays by Life Technologies, Darmstadt, Germany) resulting in concordance rates of 100% (primer/probe sequences and assay conditions available on request). Overall genotyping resulted in completion rates of 100% and genotypes were assigned blind with respect to the phenotypic characteristics of the subjects.

Figure 2
figure 2

Organization of adenosine receptor genes ADORA1 (NM_000674, located on Chr. 1q32.1) and ADORA2A (NM_000675, located on Chr. 22q11.23) with position of selected SNPs and pairwise linkage disequilibrium (LD) structure. Gene bodies were illustrated by bold black horizontal lines on the 5′ to 3′ DNA sense (+) strand, both containing 5′ and 3′ untranslated exons/exon-parts displayed as flat black blocks. The two coding exons per gene are displayed as higher black blocks and SNP positions relative to genes and exons are indicated by arrows. In the LD plots shades of gray/numbers in boxes represent the extent of LD (darker gray/higher numbers=higher LD and black/empty box=complete LD) assessed with statistics D’ and R-squared together with the resulting haploblock assignment using the solid spine of LD method as implemented in Haploview.

PowerPoint slide

Statistical Analyses

Hardy–Weinberg equilibrium of individual SNPs as well as pairwise linkage disequilibrium (LD) of ADORA1 and ADORA2A SNPs, respectively, were assessed by Haploview version 4.1 (Barrett et al, 2005; www.broad.mit.edu/mpg/haploview). Genotype groups were formed for each SNP to increase statistical power: carriers of minor alleles of a given SNP (subjects homozygote or heterozygote for the rare allele) were combined to one group and further analyzed vs the major allele carriers of the same SNP (subjects homozygous for the frequent allele). Demographic and descriptive data such as age, coffee and/or cigarette consumption, sleep duration, injected activity, and specific activity at scantime (for a complete list see Supplementary Table S2) were tested for deviation from normal distribution (one-sample Kolmogorov–Smirnov and Shapiro–Wilk tests). Next, these data were checked for potential confounding effects in both categories, specific BPND data of all brain regions and genotype groups of all SNPs, by parametric (analysis of variance, Pearson’s correlation analysis) or nonparametric statistics (Pearson’s chi-square test, Fisher’s exact test, Mann–Whitney U-test, and Spearman’s rank correlation analysis) as appropriate. Exploratory analysis of genotype group effects of ADORA1 or ADORA2A SNPs on regional A1AR binding was performed using general linear models (GLM univariate analysis of variance) including genotype group (minor allele carriers vs major allele homozygotes) as independent factor, BPND data as dependent variables and demographic or descriptive data as confounder (factor or covariate) if relevant. Statistical tests were computed using SPSS Statistics software (version 22, IBM) with the alpha level set at P<0.05. Adjustment for multiple SNP testing was done in gene-wise manner (ADORA1: 7 SNPs; ADORA2A: 5 SNPs) using the conservative Bonferroni correction, which resulted in adjusted alpha levels of P<0.007 for ADORA1 SNPs and P<0.01 for ADORA2A SNPs. Exploratory correction for comparisons of multiple (19) brain regions was performed in a SNP-wise manner based on nominal P-values by controlling the false discovery rate. This followed the Benjamini–Hochberg procedure (Benjamini and Hochberg, 1995) resulting in adjusted alpha levels of P<0.0026 for the best (smallest) nominal P-value, P<0.0053 for the second best, P<0.0079 for the third best, and so forth.

RESULTS

Genotype frequencies of all ADORA1 and ADORA2A SNPs in the PET group of healthy volunteers conformed to Hardy–Weinberg equilibrium (Pall>0.1). Pairwise LD analysis revealed moderate to strong LD between most ADORA1 and ADORA2A variants (Figure 2). Complete LD was detected between ADORA1 SNPs rs9660662-rs11582098-rs722915-rs10920568 (each pairwise comparison: D’=1.0, r-squared=1.0), as well as ADORA2A SNPs rs3761422-rs5751876-rs35320474 (D’=1.0, r-squared=1.0) and rs2298383-rs4822492 (D’=1.0, r-squared=1.0). Therefore we dropped the redundant SNPs (rs9660662, rs11582098, rs722915, rs2298383, rs3761422, and rs35320474) leaving seven ADORA1 SNPs (rs1874142, rs10920568, rs12135643, rs3766566, rs17511192, rs6677137, and rs3753472) and five ADORA2A SNPs (rs5751862, rs5760405, rs2236624, rs5751876, and rs4822492) for further analysis.

Demographic and descriptive data analyses revealed an impact of age, sleep duration, and smoking on BPND data (Supplementary Table S2) and were therefore included as factor/covariate in all subsequent analyses. Analysis of ADORA1 and ADORA2A genotype group effects on regional [18F]CPFPX binding revealed association of several SNPs of both genes with a variety of different brain regions as summarized in Table 1 (for detailed data see Supplementary Table S3). Bonferroni adjustment for multiple SNP testing, however, suggested trend significance only for ADORA2A SNPs (rs5760405: Pcorr=0.063 and 0.070; rs2236624: Pcorr=0.080), but not for ADORA1 SNPs.

Table 1 Overview of SNP Effects on In Vivo Brain Regional A1AR Binding Presenting P-Values (uncorrected) in Total Sample of Healthy Subjects

Subsequent stratification to smoking further supported this difference between genes and revealed almost exclusively ADORA2A SNP effects on [18F]CPFPX binding in the subsample of non-smoking subjects (N=22) as overviewed in Table 2 (for detailed data see Supplementary Table S4). These ADORA2A effects overall expanded and strengthened compared to the total sample analysis, now including more brain regions as well as findings per SNP. Accordingly, Bonferroni adjustment for multiple SNP testing supported several of these findings, with associations withstanding correction and remaining significant: the effect of rs2236624 on the superior frontal gyrus (Pcorr=0.017) and dorsolateral prefrontal cortex (Pcorr=0.013) as well as of rs5751876 on the entorhinal cortex (Pcorr=0.006), hippocampus (Pcorr=0.040), and whole cortical BPND data (Pcorr=0.031); all illustrated also in Figure 3a–d).

Table 2 Overview of SNP Effects on In Vivo Brain A1AR Binding Presenting P-Values (uncorrected) in Subsample of Non-Smoking Subjects
Figure 3
figure 3

Boxplots showing ADORA2A genotype group effects on regional A1AR binding in the subsample of non-smoking subjects. General linear models with Bonferroni adjustment for multiple SNP testing revealed following effects: rs2236624 on superior frontal gyrus ((a); Pcorr=0.017) and dorsolateral prefrontal cortex ((b); Pcorr=0.013) as well as rs5751876 on entorhinal cortex ((c); Pcorr=0.006) and hippocampus ((d); Pcorr=0.04).

PowerPoint slide

DISCUSSION

The main finding of the present study is an association of SNPs of adenosine receptor genes with in vivo [18F]CPFPX binding to the A1AR in brains of healthy human subjects, an association that remained significant for several ADORA2A SNPs in the subsample of non-smoking subjects even after Bonferroni correction of multiple SNP testing. This is the first time, to the best of our knowledge, that variation of A1AR availability was shown associated with genetic factors at all. With it we could add another important factor for interindividual variability in A1AR availability beyond physiological or external factors such as age (Meyer et al, 2007), wakefulness (Elmenhorst et al, 2007a), or caffeine (Elmenhorst et al, 2012). Further, our finding is in line with recent findings of other genetic variants reported to play a role in availability of receptors of the serotonergic, dopaminergic, and monoaminergic systems (Willeit and Praschak-Rieder, 2010). Together these data support a broad-based impact of genetic factors on general receptor binding in the human brain.

Several brain regions were identified in our study as particularly prone to ADORA2A SNP effects: the superior frontal gyrus, the dorsolateral prefrontal cortex, the hippocampus, and the entorhinal cortex. They are important players in emotion processing and aberrations in volume, reactivity, or interconnectivity could be related to dysfunctions in emotion or mood regulation as shown to be the case in different psychiatric disorders like anxiety disorders, depression, bipolar disorders, and schizophrenia (Baiano et al, 2008; Moghaddam and Homayoun, 2008; Rich et al, 2011; Hamilton et al, 2012). Thus, our findings might hint to ADORA2A SNP effects playing an important role in emotion regulation.

In addition to the brain regions with highest sensitivity to SNP effects, also lower, statistically sub-threshold sensitivity to SNP effects was found in the remaining 15 investigated regions. Notably, therein the direction of SNP effects was highly consistent and in accordance to the highly sensitive regions. For example, in subjects homozygous for the rs5760405 C-allele, always higher [18F]CPFPX binding to the A1AR was detected across all 19 brain regions compared to T-allele carriers (CT or TT genotype). The same consistently higher A1AR availability was true for carriers of the rs5751876 T-allele (CT+TT) in all brain regions and for carriers of the rs2236624 T-allele (CT+TT) in nearly all regions (except caudate). This might indicate an underlying widespread genotype group effect on A1AR binding, which, however, overcomes the Bonferroni-corrected threshold for significance only in case of particular strong association with the above mentioned brain regions with importance for emotion regulation.

Our findings can be linked to studies reporting an association between ADORA2A SNPs and alterations in emotionality, especially altered anxiety. Some of the ADORA2A SNPs that turned out to be strongly involved in for instance anxious personality (Hohoff et al, 2010: rs5751862, rs5760405, and rs5751876) and panic disorder (Deckert et al, 1998; Hohoff et al, 2010: rs5751876), had also effects on A1AR binding in our present study. That is, rs5751862 GG homozygotes, rs5760405 CC homozygotes, and particularly carriers of the rs5751876 T-allele had higher A1AR availability in [18F]CPFPX PET. Thus, our data connected increased A1AR binding with reported risk alleles from the ADORA2A gene. This indirect impact of ADORA2A SNPs on A1AR binding was particularly strong in the subsample of non-smokers, a group without nicotine interference. Exploratory correction for multiple comparisons (19 brain regions) by controlling the false discovery rate supported at least the strongest ADORA2A SNP effects on A1AR binding (rs2236624: Pcorr=0.049 for superior frontal gyrus, Pcorr=0.032 for dorsolateral prefrontal cortex; and rs5751876: Pcorr=0.023 for entorhinal cortex, but only a trend for hippocampus with Pcorr=0.076).

Increased A1AR availability has been observed after different negative physiological conditions like prolonged wakefulness (Elmenhorst et al, 2007a) or acute exposure to ethanol (Paul et al, 2011b). Increased A1AR binding was also detected in the hippocampus of two different rat models of stress, using unpredictable chronic mild stress or chronic restraint stress (Crema et al, 2013). Crema et al discussed a possible relationship between the hypothalamic–pituitary–adrenal (HPA) axis and A1AR modulation, maybe via positive regulation by endogenous corticosteroids and suggested that an upregulation of A1AR could be involved in protecting hippocampal cells from insults induced by chronic stress. Our finding of increased A1AR availability in risk allele carriers is in line with these findings and suggestions. Stress response regulation including HPA axis activation is altered in panic disorder patients (recent review by Ising et al, 2012) and sleep deprivation occurs with negative mood and increased anxiety up to catastrophizing in them (Kahn et al, 2013). Notably, these consequences of sleep deprivation are greater after the presentation of a mild stressor, possibly due to a lowered psychological threshold to estimate a situation as stressful (Minkel et al, 2012). Increased sensitivity to mild stressors or stimuli combined with anxiety (anxiety sensitivity) was shown also as a risk factor in the development of anxiety disorders (Schmidt et al, 2006). It might correspond to a hyperarousal state that is assumed in ADORA2A risk allele carriers potentially due to transcriptional inhibition and decreased amounts of A2AAR molecules with functional consequences comparable to A2AAR blocking by receptor antagonists such as caffeine (Hohoff et al, 2010). As a consequence A1AR binding might be upregulated and serve as a compensatory mechanism, eg to protect brain cells from possibly negative effects of chronic stress (hyperarousal) as proposed by (Crema et al, 2013). This is in line with chronic caffeine intake leading to increased cerebral A1AR binding (Rudolphi et al, 1989; Johansson et al, 1993) and exhibiting neuroprotective effects (Rudolphi et al, 1989; Maia and de Mendonça 2002; Ascherio and Chen 2003).

However, at present no data are available on A2AAR receptor binding modulation by genetic variation in ADORA1 and ADORA2A genes. Increased A1AR binding might therefore not be the consequence of A2AAR availability and might alter the normally balanced system of inhibitory A1AR and facilitatory A2AAR necessary for physiological functioning as mentioned in the introduction. Thereby such misbalance might serve itself as an underlying risk factor contributing to the development of particularly anxiety disorders. Though intriguing, these considerations are highly speculative until studies on A2AAR binding and further studies with functional examinations are available. Obviously there are several limitations to our study. First of all, our study has a pilot character with exploratory analysis of SNP effects on A1AR binding in a sample of 28 individuals. Therefore statistical power is only moderate and might mask several associations between SNPs and brain regions. However, our results have to be considered with caution since we only performed correction for multiple SNP testing with no systematic correction for testing of multiple brain regions that might have turned out overly stringent in this exploratory study. Further, given the mutual dependence of brain regions as detected in our sample (Supplementary Table S5), future GLM multivariate procedures could provide an additional analysis tool of global SNP effects on brain A1AR binding. Such multivariate models allow simultaneous analysis of multiple dependent and highly inter-correlated variables, but need bigger sample sizes than 28 individuals, particularly if additional confounders have to be included as it is the case in our sample with age, sleep, and cigarette consume. Therefore replication studies in independent, larger samples of healthy subjects are strongly warranted. Further limitation might result from using the cerebellum as reference region. Former studies with the radiotracer [18F]CPFPX demonstrated that the outcome parameter BPND is stable and reliable. As the cerebellum contains a small portion of A1ARs (Bauer et al, 2003) there is a theoretical possibility that receptor polymorphisms might differently affect A1ARs in the reference region and the regions of interest, which would result in an over- or underestimation of BPND values. However, previous simulations showed that changes in the specific binding of [18F]CPFPX in the reference region (cerebellum) were alleviated in the resulting BPND (Elmenhorst et al, 2007b). Furthermore, additional comparisons of area under the curve values of normalized cerebellar TACs between major and minor allele carriers of all SNPs did not show any significant difference in A1AR BPND in any region. Another limitation might derive from the sample composition itself, as our sample included subjects ranging from 21 to 63 years and age turned out to confound brain-specific BPND data. An effect of aging on cerebral A1AR binding was observed also in a sample of comparable size (N=36) aged 22–74 years (Meyer et al, 2007). However, although we included age as covariate in the present sample analysis it should be considered for future recruitment of replication samples. Further, we aimed at a rather homogeneous sample for this pilot study (male European of young to medium age), which prevented analysis of potentially existing gender or race-/ethnicity-related effects. Similarly, we have no data on personality features like anxiety in our subjects that might be a possible confounding factor. A comparison of ADORA1 and ADORA2A risk-allele effect on A1AR binding in male and female individuals with increased anxiety or even panic disorder is therefore of particular interest in replication studies.

In summary, our results revealed variation of regional A1AR availability associated with ADORA1 and particularly ADORA2A SNPs, several of which had previously been shown to be associated with increased anxiety and panic disorder. Future studies are strongly warranted to replicate our pilot findings of increased A1AR in risk alleles carriers, to provide data on modification of A2AAR binding by variation in ADORA1 and ADORA2A genes and to evaluate the mechanistic considerations by adding studies in individuals with increased anxiety and panic disorder.

FUNDING AND DISCLOSURE

This work was supported by Deutsche Forschungsgemeinschaft (SFB TRR58 TPA08 and Z02). VA is a member of advisory boards and/or gave presentations for the following companies: AstraZeneca, Janssen-Organon, Lilly, Lundbeck, Servier, Otsuka, and Trommsdorff. He also receives funds from the German Ministry of Education and Research (BMBF) and from the European Union (EU-FP7). These cooperations have no relevance to the work described. JD has received speaker’s honoraria from Janssen, Bristol-Myers Squibb, Wyeth, Lundbeck, AstraZeneca, and Pfizer, and grant support from Medice; his research is funded by IZKF Würzburg, DFG, BMBF, EU, and Vogel Foundation. There are no competing financial interests in relation to the work described. The remaining authors declare no conflict of interest.