Three circadian clock genes Per2, Arntl, and Npas2 contribute to winter depression

Abstract

Background. Multiple lines of evidence suggest that the circadian clock contributes to the pathogenesis of winter depression or seasonal affective disorder (SAD). We hypothesized that sequence variations in three genes, including Per2, Arntl, and Npas2, which form a functional unit at the core of the circadian clock, predispose to winter depression.

Methods. In silico analysis of the biological effects of allelic differences suggested the target single‐nucleotide polymorphisms (SNPs) to be analyzed in a sample of 189 patients and 189 matched controls. The most relevant SNP in each gene was identified for the interaction analysis and included in the multivariate assessment of the combined effects of all three SNPs on the disease risk.

Results. SAD was associated with variations in each of the three genes in gene‐wise logistic regression analysis. In combination analysis of variations of Per2, Arntl, and Npas2, we found additive effects and identified a genetic risk profile for the disorder. Carriers of the risk genotype combination had the odds ratio of 4.43 of developing SAD as compared with the remaining genotypes, and of 10.67 as compared with the most protective genotype combination.

Conclusion. Variations in the three circadian clock genes Per2, Arntl, and Npas2 are associated with the disease, supporting the hypothesis that the circadian clock mechanisms contribute to winter depression.

• Circadian
• depressive
• in silico
• seasonal affective disorder
• single‐nucleotide polymorphism

View correction statement:

ERRATUM

Introduction

Both genetic variants and shared environmental factors contribute to the phenotypes of mood disorders 1. Both are present in seasonal affective disorder (SAD) or winter depression, which is defined by recurrences of a major depressive episode in the fall 2 together with circadian abnormalities 3 and full remissions the following spring (for review, see 4). No psychosocial factor accounts for these episodes in these patients. Patients with SAD tend to have atypical depression including reversed diurnal mood variation, prolonged sleep duration, increased appetite and weight gain 5, along with abnormalities in their circadian clockwork such as phase delays 6, lowered amplitudes 7, or reset errors 8.

Key components of the circadian clock include Per‐Arnt‐Sim homology domain (PAS) proteins. They take part in extensive transcriptional regulation, regulate biological responses to light, and can mediate a number of biochemical functions needed for developmental and physiological events 9. Usually, PAS proteins also harbor a basic‐helix‐loop‐helix domain. This domain supports dimerization and provides a platform for specific DNA contacts within target enhancer elements 10.

Interacting molecules of the circadian clock include the PAS proteins PER2 (period homolog 2), ARNTL (aryl hydrocarbon receptor nuclear translocator‐like, also known as BMAL1, brain and muscle ARNT‐like protein 1), and NPAS2 (neuronal PAS domain protein 2). NPAS2 binds DNA as a dimeric partner of ARNTL. PER2 is a positive regulator of the Arntl 11 and Npas2 genes and the principal regulator of NPAS2, as well as stimulates the ARNTL‐NPAS2 transcriptional activity 12.

Mutations in or deletions of a single circadian clock gene cause alterations to the circadian rhythms, rest‐activity cycles and sleep patterns 13, including the Per2 14–16, Arntl 17 and Npas2 18,19 genes. Many genes contribute to the phenotype of complex disorders, such as SAD, but since the circadian clock genes in particular are relevant 20,21, we focused on these.

Because the Per2, Arntl, and Npas2 genes form a key functional unit at the core of the circadian clock 12, we selected them and hypothesized that variations in these three genes together were associated with the disorder, and analyzed their combined effect. Herein, we emphasized Per2, because activation of Per2 expression is dependent on light exposure and responsible for reset of the circadian clock 22.

Supplementary information can be found in the online version of this paper.

Key messages

• The Per2, Arntl, and Npas2 genes form a functional unit for the circadian clock.

• Winter depression was associated with variations in each of the three genes.

• These three circadian clock genes contribute to the pathogenesis of winter depression.

Materials and methods

Diagnostic assessment

All patients (n = 189) were recruited from outpatient services and met the diagnostic criteria for major depressive or bipolar disorder with the seasonal (winter) pattern, according to DSM‐IV 23. Sex distribution was 156 women and 33 men, which agrees with earlier SAD samples. Affected individuals were unrelated Caucasians originating from Sweden (118 patients), Finland (41 patients), Austria (19 patients), and Germany (11 patients).

Representative unrelated, normal controls (n = 189) were matched with cases by the ethnicity, sex, age in a range of 5 years, and nationality. They were recruited as healthy volunteers from studies run by the institutes, and structured clinical interviews were conducted to exclude participants with any current or past psychotic or mood disorders. Control individuals were thus representative of the population at risk of becoming a case, selected independently of the exposure and free of the outcome, as required for a case‐control study.

Ethics

The study was approved by the local ethics committees, and all the participants signed informed consent after the protocol had been fully explained.

Self‐report questionnaires

A total of 108 patients and 98 controls completed the Seasonal Pattern Assessment Questionnaire (SPAQ) to assess the seasonal variation in the length of sleep, social activity, mood, weight, appetite, and energy level 24. The sum of the six items yields the Global Seasonality Score (GSS), which is a measure of changes in physiological and social activities within the seasons. It ranges from 0 to 24. The mean GSS was 13.9, with standard deviation (SD) of 2.9 for the patients and 3.7 (SD = 2.6) for the controls. Data derived from this questionnaire were used for description only.

A total of 74 patients and 46 controls of Swedish or Finnish origin completed the Morningness Eveningness Questionnaire (MEQ) to assess the daily pattern of activities 25. The sum of the questionnaire yields the Morningness Eveningness Score (MES), which is a measure of preference for physiological and social activities in the early or late parts of the day, and is well correlated with the intrinsic circadian period 26. It ranges from 16 to 86. The mean MES was 54.1 (SD = 5.8) for the patients, when not depressed, and 52.8 (SD = 4.0) for the controls.

SNP selection

Three circadian clock genes (Per2, Arntl, Npas2), which form a functional unit, were analyzed for potentially functional genetic variations (see Figure 1). To identify combined gene effects, we selected potentially functional single‐nucleotide polymorphisms (SNPs) in each of the three genes. In silico analysis of the biological effects of allelic differences suggested that the selection of target SNPs of each gene was guided by assessing potential functionality 27,28. SNPs in the three genes (Per2, Arntl, Npas2) were identified with the use of public databases 29,30 and Celera Human SNP Database (http://www.celera.com; CV2153849), and with the sequencing effort (10870 located at NT_005120.15 position 5118260 in the sequence cagatgaaaatgcccaagact). Selection of SNPs was guided by assessing the potential functionality in silico, linkage disequilibrium (LD) structure, and equal distribution along the genes. All nonsynonymous substitutions identified were included in the study, except the Per2 SNP rs934945 leading to a Gly/Glu amino acid exchange in exon 17 because of its complete LD with rs13033501 in intron 4 (chi‐square = 27.4, P<0.00001, D' = 1.0). Polymorphisms of very small minor allele frequency (<1%) were not considered.

Figure 1. Gene regions and single‐nucleotide polymorphisms (SNPs) analyzed.

Regulatory target polymorphisms of the three genes were selected as follows. In the case of variants located in intronic regions, the function was assessed using in silico analysis of transcription factor binding sites. Both possible alleles of each SNP were tested for their binding capability to human transcription factors. Options employed for the transcription factor binding search using TESS 31 were 21 bases of genomic sequence around each SNP (10 bases on either side of the SNP) and string‐based search query with default settings. Recent findings show that transcription factor binding sites may be commonly located in introns or other noncoding regions 32.

We checked three experimentally verified functions of silent changes: First, synonymous substitutions can alter base pairing structures in transcribed sequences which have functional effects on the expression 33,34 and can be detected using the Mfold program 35. Second, the rate of protein synthesis depends on the rate of elongation of the growing peptide chain 36. Synonymous codons that are more efficiently translated by more abundant tRNAs lead to increased synthesis of the protein. A relationship between synonymous codon usage and protein structure of eukaryote‐specific sequential, co‐translational folding may exist 37,38. For the human genome, the relative frequency of each codon among synonymous codons is known (http://bioinformatics.weizmann.ac.il/databases/codon/). Third, exonic splicing enhancers serve as binding sites for specific serine/arginine‐rich proteins, highly conserved splicing factors characterized by one or two RNA recognition motifs, and by a distinctive C‐terminal domain 39. These proteins can promote the exon definition by recruiting the splicing machinery through their domain directly or by antagonizing the action of nearby silencer elements 40,41.

3' untranslated region (UTR) functional element patterns were searched using UTRdb 42,43. Prediction of further SNPs in the respective genes (r² = 0.8, logarithm of the odds LOD ⩾3.0) was performed independent of the haplotype block structure in pair‐wise tagging mode 44.

Genotype analysis

Four tagging SNPs in the Per2 gene were analyzed using the TaqMan MGB biallelic discrimination system. Probes and primers were ordered from and automatically designed by Applied Biosystems using the Assay‐by‐Design product. Polymerase chain reactions were performed in Biometra T1 thermocyclers, and fluorescence results were determined with the use of an ABI Prism 7900HT sequence detector end‐point read. Process and genotyping data were exported into an internal LIM System. Genotyping of four SNPs in the Arntl gene and five SNPs in the Npas2 gene was run by real‐time pyrophosphate DNA sequencing, according to standard protocols provided by the manufacturer (Pyrosequencing AB, Uppsala, Sweden). One of the polymerase chain reaction primers in each pair was 5'‐biotinylated and after denaturing the single stranded DNA was separated using streptavidin‐coated magnetic beads (Dynabeads M‐280 Streptavidin from Dynal Biotech, Oslo, Norway) and sequenced using specific sequencing primers. The data concerning the Npas2 SNP S471L as analyzed earlier by us were available from these samples and included in the analysis 45.

Statistics

Assuming a scenario in which the predisposing variant or its proxy in complete LD was typed 46, our sample size was big enough to detect a significant association with the disorder (power of 80%), with the frequency of 11% for a susceptibility allele in controls (http://pngu.mgh.harvard.edu/˜purcell/gpc/).

In order to find the most relevant SNP in each of the three genes, we used an ordinary logistic regression model with step‐wise forward selection of predictors for any of the three genes separately. Each SNP in the gene was represented by a set of two indicator variables as predictors. SNPs to be included in a multivariate model were selected for the interaction analysis in a univariate framework applying testing procedures for the comparison of two trinomial distributions from which independent samples have been taken.

For the multivariate assessment of the simultaneous effects of all three SNPs on the disease risk, the frequencies of the three genotypes per SNP were reduced to binomial observations by collapsing those two categories which did not show markedly increased frequencies among cases as compared with controls. For each combination of these dichotomized SNP‐based genotypes, a dummy indicator was defined and incorporated as a covariate in a logistic regression model. In order to avoid redundancy, the genotype combination associated with the lowest observed risk was taken as baseline without explicit representation in the model.

Results for the model estimation were given as the odds ratio (OR) and their asymptotic 95% confidence intervals (CI). To compare the maximum risk genotype combination of the three genes against the whole remainder as well as against the minimum risk genotype combination, Fisher's exact test for 2×2 contingency tables was applied. The program used for that purpose (proc freq of the SAS system, version 9.1) also provided exact confidence limits to the odds ratios obtained from the tables under analysis.

Statistical testing for a difference in the frequency of the disorder between the genotype‐defined groups was calculated for each locus separately using Pearson's chi‐square test. Cochran‐Armitage tests for trend were computed in order to detect linear trends in the association between the specific single SNPs and the status. All results were controlled for potentially modifying effects of the country of origin and sex on the results obtained using stratified univariate analyses. These covariates had no significant influence on the results.

Results

In silico analysis of the biological effects of allelic differences provided the rationale for the selection of target SNPs of each gene. First, in the Per2 gene, intronic allelic differences in transcription factor binding sites were found for the SNP 10870, rs2304674 and rs10201361 (see Supplementary information). Information about LD led to the inclusion of rs13033501. Second, in the Arntl gene, intronic allelic differences in transcription factor binding sites were found for rs3789327 and rs2290035. LD information led to the inclusion of rs2279287 (rs2279287/rs3789327: r² = 0.007, LOD of 0.28) and rs969485. Third, in the Npas2 gene, the synonymous exchange of rs1053091 alters the relative frequency of the codon among synonymous codons for alanine. Whereas the G‐allele (triplet GCG) is characterized by a relative frequency of 0.10, the relative frequency of the A‐allele is 0.22, suggesting greater translation efficiency by a factor of up to 2.2 in mRNAs having the A‐allele as compared with the G‐allele. We found a possible effect of rs1053091 (silent) and rs2305158 (located in the 3'‐UTR) on predicted mRNA structure, resulting in paired/unpaired condition of the SNP position. Allelic changes in formation/destruction of exonic splicing enhancer motives were found for rs1562313 and rs2305160. Both the SNP S471L and rs2305160 are nonsynonymous substitutions in the coding region of the gene.

The minor allele frequencies and genotypes among the patients and controls are presented in Tables I and II. Next, we assessed the individual contribution of the genetic variations of each gene to the phenotype, using a step‐wise regression analysis with forward selection. This gene‐specific analysis identified the Per2 SNP 10870, Arntl SNP rs2290035 and Npas2 SNP S471L as the only genetic variations having a significant association with SAD (Table III).

Table I. Minor allele frequencies.

Chromosome number	Gene name	Single‐nucleotide polymorphism (SNP)			Alleles^a	Minor allele frequency
		Name	Id	Position	Name	Controls	Patients
2	PER2	10870	10870	238967505	G (A)	0.11	0.16
2	PER2	10562	rs2304674	238963904	C (T)	0.31	0.34
2	PER2	10563	rs13033501	238963860	A (G)	0.12	0.1
2	PER2	10565	rs10201361	238955905	G (C)	0.01	0.02
2	NPAS2	Y353Y	rs1562313	101045973	A (G)	0.15	0.18
2	NPAS2	A394T	rs2305160	101049822	A (G)	0.38	0.33
2	NPAS2	S471L	S471L	101052708	A (G)	0.2	0.2
2	NPAS2	A640A	rs1053091	101065328	T (C)	0.04	0.05
11	ARNTL		rs2279287	13255061	A (G)	0.3	0.28
11	ARNTL		rs3789327	13341892	G (A)	0.42	0.48
11	ARNTL		rs969485	13359619	G (A)	0.25	0.24
11	ARNTL		rs2290035	13364347	A (T)	0.43	0.42

^a Minor allele (major allele in parentheses).

Table II. Genotype distributions.

Single‐nucleotide polymorphism (SNP)	Controls			Patients
	Alleles			Alleles
	11 n (%)	12 n (%)	22 n (%)	11 n (%)	12 n (%)	22 n (%)
PER2:
10870	143 (79.44)	36 (20.00)	1 (0.56)	128 (69.57)	52 (28.26)	4 (2.17)
rs2304674	82 (46.33)	81 (45.76)	14 (7.91)	81 (44.51)	79 (43.41)	22 (12.09)
rs13033501	139 (78.09)	35 (19.66)	4 (2.25)	148 (80.87)	33 (18.03)	2 (1.09)
rs10201361	176 (97.24)	5 (2.76)	0	180 (96.77)	6 (3.23)	0
ARNTL:
rs2279287	80 (51.95)	56 (36.36)	18 (11.69)	85 (53.46)	59 (37.11)	15 (9.43)
rs3789327	67 (35.83)	84 (44.92)	36 (19.25)	53 (29.28)	84 (46.41)	44 (24.31)
rs969485	101 (56.42)	65 (36.31)	13 (7.26)	95 (55.23)	71 (41.28)	6 (3.49)
rs2290035	65 (34.39)	84 (44.44)	40 (21.16)	56 (29.63)	108 (57.14)	25 (13.23)
NPAS2:
rs1562313	114 (71.70)	41 (25.79)	4 (2.52)	105 (66.04)	50 (31.45)	4 (2.52)
rs2305160	57 (35.85)	84 (52.83)	18 (11.32)	75 (47.17)	63 (39.62)	21 (13.21)
S471L	95 (59.75)	63 (39.62)	1 (0.63)	105 (66.04)	45 (28.30)	9 (5.66)
rs1053091	146 (91.82)	12 (7.55)	1 (0.63)	144 (90.57)	13 (8.18)	2 (1.26)

Table III. Step‐wise logistic regression analysis of the single‐nucleotide polymorphisms (SNPs).

Gene	Patients	Controls	Significance (model)	SNP	Significance (effect)	Odds ratio
	n	n	P‐value		P‐value	OR	95% CI
PER2	177	173	0.03	10870	0.03	1.66	1.06–2.60
ARNTL	137	136	0.01	rs2290035	0.02	–	–
				A/A vs. T/T	0.19	0.62	0.31–1.27
				T/A vs. T/T	0.09	1.60	0.93–2.74
NPAS2	159	159	0.004	S471L	0.02	–	–
				A/A vs. G/G	0.05	8.14	1.01–65.48
				G/A vs. G/G	0.07	0.65	0.40–1.04

Finally, to identify groups with the highest and lowest genetic risk for the disorder, the combined effect of these genotypes was assessed. The genotype combination carrying the highest risk was the one for which the individual is either heterozygous (A/G) or homozygous (G/G) for Per2 SNP 10870 plus heterozygous for Arntl SNP rs2290035 plus homozygous for Npas2 SNP S471L. For this genotype combination, the OR of having SAD was 10.67 as compared with the group carrying the lowest risk (Table IV). When the group with the highest risk was compared with all other remaining genotypes, the OR was 4.43.

Table IV. Odds ratios of the risk versus the protective and of the risk versus the remaining genotype combinations.

	Genotype combination^a		Significance	Odds ratio
	Risk	Protective	Exact P‐value, 2‐sided	OR	Exact 95% CI
	n	n	0.003	10.67	1.74–76.50
Controls	6	8
Patients	24	3
	Genotype combination		Significance	Odds ratio
	Risk	Remaining	Exact P‐value, 2‐sided	OR	Exact 95% CI
	n	n	0.0004	4.43	1.69–13.62
Controls	6	144
Patients	24	130

^a The risk genotype combination is 10870 (A/G or G/G) plus rs2290035 (T/A) plus S471L (Leu/Leu or Ser/Ser), whereas the protective genotype combination is 10870 (A/A) plus rs2290035 (T/T or A/A) plus S471L (Leu/Ser).

Homozygosity for S471L had the greatest influence on the odds, as otherwise the first‐rank point estimate was reduced to 2.29. The point estimate was reduced to 3.15 with homozygosity for rs2290035 and to 3.17 with homozygosity (A/A) for 10870 (data not shown). In order to detect a possible interaction of the SNPs analyzed, the main effects were compared with a saturated model. No evidence for interaction was observed, suggesting additive combined gene effects (Pearson goodness of fit test: P = 0.2).

We found a significant post‐hoc association of preference to activities in the morning hours with the risk genotype of Per2 SNP 10870 in the patients (Armitage trend test, one‐sided as to test whether the preference was to the evening hours: z = −1.8, P = 0.03) but not controls. There was no significant association between genotypes of Arntl or Npas2 and MES.

Discussion

We hypothesized that the circadian clock is a key to the pathogenesis of SAD. A rationale for the present study was provided by two findings. First, the Per2, Arntl, and Npas2 genes form a key functional unit at the core of the circadian clock 12. Transcriptional networks which include the circadian clock and have control of downstream pathways by circadian oscillators implicate further the potential of circadian clock genes to the pathogenesis of SAD 47. Second, our earlier findings that SAD was associated with a functional polymorphism in the Npas2 gene 45, but not with those involved in serotonin‐related metabolism or transmission 48,49, gave further support to the hypothesis.

In our current study, gene‐wise logistic regression analysis indicated that the disorder was associated with variations in each of the three genes. Moreover, in an analysis of the combined effect of the three genes, we demonstrated additive effects and identified a genetic risk profile for the disorder. Carriers of the risk genotype combination had a four‐fold risk of having the disorder as compared with the remaining genotypes, and a ten‐fold risk as compared with those carrying the protective genotype combination. This analytical strategy follows recent statistical studies recommending a two‐step approach in which single‐locus effects are first detected using liberal statistical criteria, followed by a search for possible interactions among the detected loci 50. These risk and protective genotype combinations need to be verified first and thereafter characterized further in terms of the phenotype. Having this goal in mind, we have started the analysis of a sample of approximately 6000 individuals being representative of the Finnish population aged over 30.

Herein, we argue that these three genes together are associated with SAD and may be important to its pathogenesis, and we present a hypothetical mechanism of action for the observed effect for one of these genes. Although we have no functional evidence yet, our analysis of a functional triad of three key circadian clock genes now points at the fact that the circadian clock is involved in the etiology or pathogenesis of winter depression. Because approximately 10% of all mood disorders follow a seasonal pattern 51, winter depression can be seen as a model for the molecular mechanisms in depressive or bipolar disorders.

Our findings of the association of SAD and the circadian phenotype in the patients with Per2 SNP 10870 indicate a contribution of this gene to mood disorders, in addition to its role in regulation of alcohol intake as reported earlier 52,53. The stretch of DNA in which this SNP is located is exceptionally rich in hypothetical allelic transcription factor binding site changes (GGGCGT for SP1 and GCGTTT for MYB regarding the G‐allele; GGGaATTTTC (lower‐case letters indicate mismatches to the binding site as listed in the Transfac‐database) for NFKB1 regarding the A‐allele). This polymorphism may thus be an intronic transcriptional enhancer element, whose allelic constitution influences expression 52. Binding motifs for the transcription factor SP1 were described to serve as long distance activator elements 54,55 and may activate transcription by the mechanism of DNA loop formation 56,57. The rare risk allele of the SNP, i.e. G, may create a binding site for SP1, which is not present in most of the normal population, as indicated by the low abundance in the controls.

This SP1 binding site, which is hypothesized to confer the risk of SAD, increases transcript levels of the Per2 gene. Expression of the Per2 gene oscillates endogenously with a circadian rhythm in the suprachiasmatic nuclei, but is responsive to the light‐dark transitions. Key structures of the limbic system generate oscillations in Per2 expression 58 and are strategically positioned to modulate not only circadian rhythms downstream from the principal circadian clock, but also emotional reactions and mood.

PER2 is known to be a transcriptional repressor of the E‐box, which is an important node of the entire transcriptional network of endogenous clocks 59 and sets the pace for circadian oscillations 60. Impairment in E‐box or prime E‐box regulation may therefore greatly affect the circadian system. Supportive evidence for the hypothesis of a quantitative action of Per2 transcription was provided by recent experimentations which showed that fibroblast cell lines overexpressing mPer2 mRNA levels severely impaired the expression of circadian clock genes 61. Our finding of the increased morning activity in carriers of the risk genotype of SNP 10870 further substantiates the effect of the SNP on the circadian clock and points towards a causal role of the Per2 gene in SAD.

ARNTL as a partner of heterodimers with either NPAS2 or CLOCK (circadian locomoter output cycles kaput homolog) drives transcription from elements in promoters of the responsive genes included in the circadian pacemaker system, such as Per2 62. The SNP rs2290035 in the Arntl gene is located in the last intron. It alters a SP1 binding site and may thereby regulate the circadian activities of ARNTL 63.

Expression of the Arntl gene is activated directly by RORA 64, a retinoic acid receptor‐related orphan receptor, which is a key component of the circadian clock 65. The pineal gland hormone melatonin is a ligand for RORA 66, and therefore a signal of day length may have an additional effect on mood, as routinely seen in patients with SAD in specific 67. Availability of the ligands for RORA and subsequent drive for Arntl expression might thereby link to the interaction of depressive and seasonal components in patients with SAD, providing a further point of view on the dual vulnerability or two‐trait hypothesis.

Abnormalities of the circadian clockwork may indicate a direct causative role of circadian clock genes in the pathogenesis. In addition, the Per2‐mediated regulation of synaptic concentrations of glutamate, which influence alcohol intake, suggests a mechanism of action for regulation of a secondary set of genes. Our work suggests that the strategy applied herein for analysis is useful as a template for subsequent studies. It also shows that SAD is a feasible model for studying the biologic basis of common mood disorders in which both genetic and shared environmental components are present. Genes generating the ultradian (shorter than circadian) metabolic cycles 68,69 that may take over and drive the intrinsic clockwork, those reacting to light exposure upstream from the core of the circadian clock 70,71, and those translating the cellular timekeeping into behavior downstream from the core of the circadian clock 72,73 will be of high interest in future studies. In patients with SAD, there is increased resting metabolic rate 74, abnormalities in response to light 75, and the default‐like prolactin secretion pattern 76, all giving further support to the view that abnormalities in the circadian clock contribute to the pathogenesis of winter depression 77.

In summary, we found that SAD was associated with a potentially functional polymorphism in each of the three circadian clock genes. The proteins encoded by these genes are known to form a functional unit within the circadian clock, and there is a substantial circadian component in SAD. Our findings now point at the possibility that these circadian clock genes are a key to the pathogenesis of SAD.

Acknowledgements

The study was supported in part by grants from Academy of Finland (201097 and 210262) and The Finnish Medical Foundation to Dr Partonen, and by grants from the SFB 636, the NGFN (01GS0117), and the MWK Baden Württemberg (SUFO‐Projekt 12) to Dr Schumann.