Introduction

Attention deficit hyperactivity disorder (ADHD) is a common neuropsychiatric condition, with a worldwide prevalence of 5 % in school-age children [1]. A substantial number of children continue to suffer even after they have reached adulthood. The detrimental effects of ADHD are all the more apparent when the extensive comorbidity with other psychiatric conditions is considered. The rate of comorbidity in ADHD is much higher than it is in other diagnoses. Antisocial behavior and substance use disorders are frequent comorbid conditions, and rates of mood and anxiety disorders reach almost 50 % [2]. Additionally, symptomatic overlap and extensive comorbidity with bipolar disorder have been intensively discussed [3]. Moreover, a large proportion of patients exhibiting substance abuse, mood or anxiety disorders, obesity, smoking, and illegal behavior also meet the diagnostic criteria for ADHD [4–6].

The extensive overlap between ADHD and such conditions suggests that certain features of ADHD may not be specific to ADHD but rather coextensive with a wider range of psychological and behavioral conditions. It has been proposed that ADHD may be a foundational disorder that substantially increases a person’s risk of additional psychiatric impairments during his or her lifetime [7]. Alternatively, some authors have argued that ADHD may be considered an extreme end of a continuum of traits common to the general population [8]. If this is true, then a broader conceptualization of the ADHD-related features that are common and continuous traits may offer a better target for genetic studies than would a specific diagnosis.

To study such traits, the target population needs to be extended to include the general population. Additionally, the candidate genes must be selected based on their implications for similarly common phenotypes. Disrupted circadian rhythm may be one such candidate phenotype. Indeed, this is very common in the general population, and its involvement in the core pathophysiology of ADHD has been suggested [9]. Patients with ADHD have various types of sleep disturbances [10] as well as altered circadian profiles of melatonin and cortisol secretions [11]. Furthermore, the degree of the disruption in circadian rhythm has a significant impact on ADHD symptom severity, daytime functioning, and health outcomes [12].

Diurnal preference, also called morning or evening preference, reflects individual differences in circadian topology [13]. Evening preference is related to various emotional and behavioral problems, stress response patterns, and specific personality traits. Interestingly, many of these are also associated with ADHD [14]. Subjective and objective measures of attention deficits [15, 16], impulsivity, and novelty seeking [17] are among the typical examples. It should be noted that comorbid conditions other than ADHD show strong relationships with evening preference. Evening preference has been shown to be more frequent than morning preference in depression [18, 19], substance use [20], and bipolar disorder [21]. These findings support the notion that the coextensive relationship between ADHD and other comorbid conditions may be mediated by disrupted circadian rhythm or a specific diurnal preference.

Diurnal preference is largely determined by stable genetic components [22]. Thus far, the so-called clock-related genes have been believed to be responsible. The clock-related genes and their products tightly regulate the human circadian rhythm [23]. The involvement of the CLOCK gene in determining individuals’ diurnal preference has been extensively studied. A study conducted by Katzenberg et al. [24] demonstrated that the rs1802160 polymorphism (T3111C) in the CLOCK gene is associated with evening preference; this finding was replicated in a Japanese population [25]. The relationship among the CLOCK gene, diurnal preference, and ADHD led to a hypothesis that changes in circadian rhythm induced by polymorphisms in the CLOCK gene may contribute to ADHD pathophysiology. An association between the CLOCK gene and ADHD has been demonstrated in German and Chinese populations [26, 27]. If this hypothesis were true, the CLOCK gene may link disrupted circadian rhythms and ADHD-related symptoms in a wide range of psychiatric conditions.

In this context, we aimed to investigate the genetic association between the CLOCK gene and ADHD-related features and to demonstrate the possible mediating role of evening preference in the relationship between these two factors. To this end, we typed the CLOCK gene polymorphisms in a sample of healthy adults and simultaneously measured diurnal preference and ADHD-related symptoms using the Composite Scale of Morningness (CSM) and the Wender Utah Rating Scale-25 (WURS-25), respectively. We expect that this study will help to clarify the possible mediating role of diurnal preference in the association between the CLOCK gene and ADHD-related symptoms.

Materials and methods

Study participants

Study participants were recruited from two different sources, referred to as subset #1 and subset #2 in the following discussion. Subset #1 included workers and students in university-affiliated general hospitals, and subset #2 consisted of workers in government offices. A description of the study’s purpose and procedure was advertised through posters and leaflets. All participants were provided with a detailed description of the study before written consent was obtained. Brief psychiatric interviews for screening purposes were performed by trained psychiatric nurses. Participants with a lifetime history of major psychiatric illness or brain trauma were excluded. Participants with a lifetime history of bipolar or unipolar mood disorder severe enough to require hospitalization were also excluded. Other psychiatric difficulties such as anxiety, minor depression, and substance problems were not considered as exclusion criteria.

Although no structured assessment tools were used, participants who reported severe attention or behavioral problems in childhood were excluded. A total of 1,289 participants were included. All participants were unrelated to one another, and all were ethnically Korean. Basic demographic data and information on current mood, sleep quality, and daily life stress were collected using self-administered questionnaires. This study was approved by the ethics committees of Eulji General Hospital and Eulji University Hospital.

Measurement of childhood ADHD features

ADHD-related features were measured with the WURS-25, which was developed for retrospective diagnosis of ADHD symptoms experienced before the age of 12 years [28]. Although the original WURS contained 61 items, the short version consisted of 25 items selected from the original version. The Korean version of the WURS-25 was developed and standardized using female Korean adults [29]. In this standardization study, the mean total score was 19.39 ± 14.21. A gender difference could not be ascertained because only female subjects participated in that study.

Measurement of circadian rhythm

The participants’ degree of evening preference was measured using the CSM, a 13-item questionnaire that assesses individual differences related to the time of day at which a person prefers to carry out various activities [30]. Higher scores represent greater morning preference. Previous studies have demonstrated good test–retest reliability and adequate external validity for the CSM [31]. The Korean version had also been standardized [32]. In this standardization study, the mean total scores were 34.00 ± 6.32 in male and 34.37 ± 5.47 in female. There was no significant difference between the two sexes.

Genotyping

DNA was extracted from blood samples using a DNA isolation kit (Roche, Mannheim, Germany). We selected five SNPs in the CLOCK gene based on the minor allele frequencies reported in Asians and the linkage disequilibrium pattern (Table 1). Relevant information was consulted in Entrez SNP of NCBI. Even though the number of SNPs was chosen to cover the whole CLOCK gene, we tried to include SNPs that were representative of the whole gene and selected the SNPs closest to the markers previously studied regarding psychiatric phenotypes. Genotyping was performed using the TaqMan method (Applied Biosystems, Foster City, CA, USA) [33]. The chromosomal location of SNPs was referenced from the UCSC Genome Browser on Human Assembly and Entrez SNP of NCBI in February, 2009 (GRCh37.p3/hg19).

Table 1 Five SNP markers in the CLOCK gene that were subjected to association analysis
Full size table

Statistical analysis

R statistical software for Windows version 2.15.1 was used for descriptive statistics, linear regression, and exploratory factor analysis [34]. Demographic variables such as age and the total scores of the CSM and WURS-25 were compared by sex using Student’s t-test. The influence of age and sex on CSM and WURS-25 total scores was analyzed with linear regression. Because the summary statistics for male and female participants were so different, the data set for each sex was analyzed separately from this point forward.

The exploratory factor analysis was performed to extract major factors representing the latent dimensions of the WURS-25. The principal component method was used for factor extraction. The number of factors was decided by eigenvalue criteria (≥1). The extracted factors were rotated by promax rotation with Kaiser normalization. After the rotation, the factor scores of each participant were calculated by regression. Partial correlation coefficients were obtained to examine the relationship among CSM and WURS-25 total scores and extracted factor scores on the WURS-25, controlling for age.

The genetic analysis was performed using PLINK software [35]. The linkage disequilibrium structure was investigated using the Haploview software [36], and separate haplotype blocks were determined based on confidence bounds on D′ [37]. The quantitative association analyses were performed as implemented in the PLINK software. The statistical framework was basically a linear regression analysis. The scores for each individual were entered as response variables, and all SNP markers or reconstructed haplotypes were entered as explanatory variables. Given that age had been shown to be a very strong confounding variable for both CSM and WURS-25 scores, age was included as a covariate in all subsequent analyses using the “—covar” option in the “—linear” command.

The sliding window approach was used for haplotype construction in the haplotype-wise association analysis. Of the five SNP markers in the CLOCK gene, four two-adjacent-marker haplotypes, three three-adjacent-marker haplotypes, two four-adjacent-marker haplotypes, and one five-marker haplotype were constructed (15 in total). Nonadjacent marker haplotypes were not considered. When performing haplotype-based association analyses, the “-hap-omnibus” option was used to simultaneously estimate all haplotype effects at a single locus.

All the reported p values were nominal p values without any adjustment. However, since multiple markers and phenotypes were analyzed simultaneously, the inflation of family-wise error rate was expected to be considerable. Meanwhile, strictly applying Bonferroni’s correction would be too conservative, because the tests involved were not mutually independent. In addition, since this study purported to be rather an exploratory study than a confirmatory study, controlling for type-II error was also essential. In order to compromise these dilemmas, we employed the false discovery rates (FDRs) method proposed by Storey [38]. FDR controls the expected proportion of null results that are falsely identified as significant in a set of predictions. In order to decide on the acceptable significance threshold, we calculated FDR at each level of significance level and set the significance level with the lowest FDR value as the significance threshold. FDR values were calculated with R-package “QVALUE.”

Results

Demographic variables and the influence of sex and age on CSM and WURS-25 scores

The number of male and female participants and their mean ages are shown in Table 2. About two-thirds of the participants were female; female participants were significantly younger than male participants (males: 27.37 ± 8.33 years, females: 23.70 ± 3.53 years, t = 10.988, p < 0.001). Average CSM total scores were significantly higher in male participants, suggesting a relative morning preference in this group (males: 32.25 ± 6.18, females: 29.74 ± 5.69, t = 7.415, p < 0.001). Additionally, average WURS-25 total scores were 19.49 ± 15.65 and 19.31 ± 14.23 in male and female participants, respectively, revealing no significant sex differences.

Table 2 Constitution of the study participants and the total CSM and WURS-25 scores in male and female participants
Full size table

As it has been shown that diurnal preference is highly dependent on age and sex [39], univariate linear regression was performed with sex as a fixed factor and age as a covariate to determine the influence of sex, age, and their interaction on CSM and WURS-25 scores. In the case of the CSM, not only were the two main effects of sex (F = 58.29, df = 1, p = 4.39 × 10−14) and age (F = 69.68, df = 1, p = 2.2 × 10−16) significant, but the interaction effect between them was also highly significant (F = 9.718, df = 1, p = 0.00187). This significant interaction was due to the fact that CSM scores were strongly correlated with age in male participants only (Pearson’s correlation coefficient R = 0.37, p = 2.2 × 10−16). In the case of WURS-25 scores, only the main effect of age (F = 12.44, df = 1, p = 0.000435) was significant; neither the main effect of sex (F = 0.0409, df = 1, p = 0.840) nor the interaction (F = 0.721, df = 1, p = 0.396) was significant.

Relationship between WURS-25 and CSM total scores

WURS-25 total scores were negatively correlated with CSM scores even after age was controlled as a confounding variable, suggesting the existence of more ADHD-related features in both male and female evening-preference individuals (partial correlation coefficient, male: R = −0.182, p = 5.93 × 10−5; female: R = −0.084, p = 0.0154).

Allele frequency, Hardy–Weinberg equilibrium, and linkage disequilibrium structure

All SNP markers from the CLOCK gene had minimum allele frequencies (MAFs) greater than 10 % and did not deviate from Hardy–Weinberg equilibrium (Table 1). All pair-wise D′ values among the markers were greater than 0.90. According to the criteria developed by Gabriel et al. [37], the data suggested that a single haplotype block covered 73 kb, which included rs3805148, rs12504300, rs4864542, and rs12649507. The remaining marker, rs1801260, was not included in this block.

Calculation of false discovery rates

In the overall analysis, the total number of markers considered was 15 (5 SNPs and their haplotypes), and the total number of phenotypes was 5 (2 for CSM, WURS total scores and 3 for factors of WURS). These tests were done separately in male and female, such that the grand total number of performed statistical tests was 150 (=15 × 5 × 2). Submitting the obtained p values to FDR calculation algorithm, we could obtain FDRs associated with each p values. The lowest FDR was 0.0695 and the associated nominal p value was 0.0180. Even though the lowest FDR (7 %) was well above the traditional acceptable risk of 5 %, we decided to use 0.0180 as a significance threshold for the following analyses.

Quantitative trait-association analysis: CSM and WURS-25 total scores

None of the markers of the CLOCK gene showed a significant association with CSM total scores in either male or female participants, neither allelewise nor haplotypewise (the lowest p value: p = 0.0194) (Table 3). On the other hand, marker rs1801260 showed a weakly significant association with WURS-25 total scores in male participants only (p = 0.0180). Additionally, all the haplotypes including rs1801260 showed similar significant associations with WURS-25 total scores (rs1801260–rs3805148: p = 0.0150; rs1801260–rs12504300: p = 0.0170; rs1801260–rs4864542: p = 0.0170; rs1801260–rs12649507: p = 0.0171).

Table 3 Results (nominal p-values) of the allele- and haplotype-based (sliding window approach) quantitative trait-association analysis results between (1) CSM and WURS-25 total scores and (2) five SNP markers in the CLOCK gene
Full size table

Extraction of factors from the WURS-25

To determine which components of the WURS-25 were responsible for the demonstrated association signal, exploratory factor analysis was performed to extract the relevant factors of the WURS-25. Three factors were extracted by the eigenvalue criteria. The combined extracted factors explained 55.22 % of the total variance (1st factor: 40.17 %, 2nd factor: 8.34 %, 3rd factor: 6.71 %). Comparable to similar factor analyses performed by other investigators and according to the loading pattern, three factors were tentatively named “impulsivity and defiant behavior” (1st factor: IMP), “inattention” (2nd factor: INATT), and “mood instability and anxiety” (3rd factor: MOOD) [40, 41].

The factor scores obtained for the WURS-25 were subjected to quantitative trait-association analysis using the same methodology as above. The allele-wise analysis revealed that none of the factors showed a significant association with rs1801260. In contrast, all markers except rs1801260 showed weak associations with factor INATT (rs3805148: p = 0.0147, rs12504300: p = 0.0168, rs4864542: p = 0.0168, rs12649507: p = 0.0152). In the haplotype-wise analysis, all of the constructed haplotypes showed weak associations with factor INATT at the significance threshold of p = 0.0180 (Table 4).

Table 4 Results (nominal p-values) of the allele- and haplotype-based (sliding window approach) quantitative trait-association analysis between WURS-25 factor scores and five SNP markers in the CLOCK gene
Full size table

Quantitative trait-association analysis of the WURS-25 when the CSM was controlled as a covariate

Of particular interest was the possibility that the demonstrated association between the CLOCK gene and the WURS-25 total score might be mediated by evening preference. To examine this possibility, the CSM total score was controlled as a covariate with age in the updated model, and the analysis was repeated (original analysis vs. CSM controlled analysis). Even though the allele-wise p value from the CSM controlled analysis (p = 0.0241) was somewhat higher than the original analysis (p = 0.0180), all the p values obtained from updated haplotype-wise analyses were lower than the original analyses (Table 5).

Table 5 Results (nominal p-values) of the allele- and haplotype-based (sliding window approach) quantitative trait-association analysis between WURS-25 total scores and rs1801260 in the CLOCK gene 1) before CSM had been controlled and 2) after CSM had been controlled as a covariate
Full size table

Discussion

We had initially proposed two linked hypotheses: (1) that genetic polymorphisms in the CLOCK gene may be associated with ADHD-related features in the general population, and (2) that this association may be mediated by individuals’ evening preference. To examine these hypotheses, we performed a quantitative trait-association study between polymorphic markers in the CLOCK gene and ADHD-related features in healthy adults, with diurnal preference controlled as a mediating variable.

We obtained the following results. First, we found a suggestive association between WURS-25 total scores and a single SNP marker in the 3′-UTR region of the CLOCK gene (rs1802160). Second, WURS-25 total scores were strongly correlated with CSM scores, with far more ADHD-related features found in evening-preference individuals. Third, the inclusion of the CSM score as a mediating variable did not much reduce the strength of the association between WURS-25 total scores and rs1801260. Consistent with this observation, no association was observed between the CSM score and the CLOCK gene. The latter two negative findings suggest that degree of evening preference does not mediate the association between the CLOCK gene and ADHD-related features. To summarize, our results support the first hypothesis, but not the second.

If our results are to be considered valid, our strategy of using WURS-25 as a measure of ADHD-related features should be justified beforehand. WURS was originally developed for diagnostic purposes. However, the limitations of WURS as a diagnostic tool have been repeatedly discussed. A correlation between WURS scores and an objective measure of attention has been difficult to demonstrate [42], and the sensitivity and specificity of the WURS with the standard cutoff point are not adequate to permit its use as a reliable diagnostic tool [43]. Rather, it has been more often used to measure general character traits common to a broad range of psychiatric diagnoses [44]. In the psychiatric literature, the WURS has been widely used to investigate the effect of ADH-related features on the severity and prognosis of other psychiatric and behavioral problems such as PTSD [45], eating disorder [46], and borderline personality [47]. Thus, it can be argued that our use of WURS-25 as a proxy measure of ADHD-related features conformed to previous literature.

Unlike the two previous reports demonstrating the association between rs1801260 in the CLOCK gene and ADHD in clinical samples [26, 27], we tried to demonstrate this in the general population. For this purpose, a broader concept of ADHD-related features was postulated, and such features were assumed to be continuously distributed among the general population. The boundary between attention deficits as a clinical symptom and as a normal variation cannot be clearly defined. This difficulty is reflected in the remarkable variation in ADHD prevalence depending on the phenotypic definition used (between 2.5 and 42.3 %) [48]. Additionally, if a biological substratum of the broader concept of ADHD-related features can be found, it will help us to understand the relationship between ADHD and other comorbid conditions. Although the association signal observed in this study was very weak, ours is the first attempt to ascertain the role of CLOCK gene polymorphisms in the variability of ADHD-related features in the general population.

Despite the fact that the WURS-25 includes several component factors, only the total score has been used in most previous research, and it has usually been equated with ADHD severity or nonspecific impulsivity [26]. Although nonspecific impulsivity actually explained the greatest proportion of the variance in our study, other factors also substantially contributed to the remaining variance. A three-factor solution has generally been accepted [40, 41]. The specific names of these factors may differ, but the underlying concepts are very similar. Recently, Joo et al. [41] named them IMP (impulsivity), INATT (inattention), and MOOD (mood instability).

In our results, although the strengths of the association signals were at most suggestive, some points warrant attention. First, all markers used except rs1801260 showed evidence of association with the INATT factor (p values: 0.0147–0.0168). Second, this finding was further supported by the fact that all the constructed haplotypes showed similar association with the INATT factor, but not with other factors (p values: 0.0150–0.0178) (Table 4). Although these findings cannot be regarded as rigorous evidence, they may indicate a possible direction for further research. It is generally accepted that although impulsivity is the most prominent symptom in ADHD, cognitive impairments are the most fundamental problems. Thus, the CLOCK gene’s contribution to ADHD-related features may operate via more fundamental cognitive impairments rather than via general impulsivity. To clarify this matter, further research focused on the relationship between cognitive impairments in ADHD patients and the CLOCK gene is needed.

According to the original guidelines developed by Baron and Kenny [49], if a variable functions as a mediator, then (1) it is simultaneously related to both independent and dependent variables, and (2) the strength of the relationship between the independent and dependent variables is greatly reduced when the effect of the variable is controlled. Therefore, to demonstrate that evening preference serves as a mediator, we would have to demonstrate both the relationship between the CLOCK gene and evening preference (“A” in Fig. 1) and that between evening preference and ADHD-related features (“B” in Fig. 1). Our result corroborates previous reports showing that evening preference is related to ADHD-related symptoms in both clinical and nonclinical samples [14, 16]. However, we could not confirm the previously reported association between rs1801260 and evening preference [24]. Although a few reports have successfully replicated this association [25], a greater number of similar studies have failed to do so [50–53]. Even a study done with Korean college students failed to find such an association [54]. Therefore, only relationship “B,” but not relationship “A,” could be demonstrated (Fig. 1).

Fig. 1
figure 1

Conceptual representation of the mediation model. A mediating relationship occurs when a mediator variable (evening preference in this study) can account for either all or some of the observed relationship between the independent and dependent variables (the CLOCK gene and ADHD-related features in this study)

Full size image

Furthermore, our results also failed to meet Baron and Kenny’s second requirement in that controlling for the CSM score had only a minimal effect on association strength; that is, the p value changed from 0.018 to 0.024. Undoubtedly, evening preference is just one tiny aspect of the full manifestation of circadian rhythm. Therefore, the finding that evening preference did not mediate the association does not lead to the conclusion that circadian rhythm disruption cannot be a mediating link. However, previous studies have reported that diurnal preference is correlated with a wide range of the manifestations of circadian rhythm, including the sleep–wake cycle, as well as with the circadian phase and the amplitude of the phase [55]. Diurnal preference has also been associated with various human behaviors. Not only sleep habits or disturbances but also emotional problems [56], psychopathology, and personality [57] are affected by diurnal preference. Thus, although diurnal preference does not represent the whole system of circadian rhythm manifestations, it may be the most versatile indicator of the circadian system. Therefore, the search for a mediating link between the CLOCK gene and ADHD-related features need not be limited to the circadian system. The so-called noncircadian effects of the CLOCK gene may also need to be considered [58].

Our study found several sex-specific results. The linear relationship between age and CSM scores was observed only in male subjects. Likewise, the association between the CLOCK gene and WURS-25 scores was found only in male subjects. Clearly, marked sex differences exist in the prevalence and symptom manifestation of ADHD. Studies of ADHD symptoms using the WURS-25 in healthy college students have also noted that male students scored higher than did females [59]. The two studies that demonstrated an association between the CLOCK gene and ADHD both used only male participants [26, 27]. It could be that the influence of the CLOCK gene on ADHD-related features has some sex-specific quality; however, it is also possible that the higher prevalence of attention problems in male participants renders detection of a weak association easier.

This study had several limitations. The most problematic was that we could not adequately control the inflation of family-wise error rate. The lowest false discovery rate of our study was about 7 %. It meant that if all the p values lower than 0.0180 were called significant then, there would be 7 % chance of falsely rejecting null hypotheses overall. If this risk cannot be deemed acceptable, then all the results in our study are just inconclusive findings. The only justification would be that this study was an explorative one aiming to prove the concept for future studies.

Another significant drawback was the fact that the participants were recruited from two different sources. Despite the difference in sex ratio and age distribution, we had to combine the two subsets of participants and analyze the whole set because the second subset was almost exclusively male. However, when we analyzed the first subset separately, the results were similar to those described above (data not shown). The generalizability of our results is also limited. The participants were almost exclusive in their 20s or 30s. The genetic composition of such a narrow age group could be affected by the cohort effect, and the character phenotypes could be influenced by a generation effect. Therefore, findings in a specific age group may not be valid in other age groups.

Presuming a direct link between a gene (CLOCK) and a phenotype (ADHD-related feature) mediated by another single phenotype (eveningness) would be a too simplistic approach to real phenomena. A great number of additional variables had to be accounted for to properly understand the given data. Especially among those variables, the sleep pattern and the seasonality have to be considered. Evening preference cannot substitute the actual sleep pattern. What determines the individual’s psychiatric difficulties may not be the evening preference, but the actual sleep-rhythm or sleep quality [60, 61]. Though we did not measure participants’ sleep patterns or qualities, it would be an essential step for the further studies. Meanwhile, the individual’s seasonality is another important variable that has to be considered. Seasonality refers to a seasonal variation in mood and behaviors and represents a longer biological rhythm than circadian. It was also found to be associated with mood disorders, ADHD, and cognitive function fluctuations [15, 62]. Since seasonality is found to be associated both with circadian preference and ADHD symptoms, it may be another candidate variable for mediating CLOCK gene and ADHD-related symptoms.

In conclusion, we found that a polymorphic marker (rs1801260) in the CLOCK gene showed male-specific associations with WURS-25 total scores in a large sample of healthy adults and that this association was not mediated by the degree of evening preference. Although the association signal identified in our study was very weak, this is the first study to ascertain the role of the CLOCK gene in the variation in ADHD-related features in a nonclinical sample. It is not yet clear what kinds of underlying mechanism are responsible for this association; however, the search for a mediating link may need to include the noncircadian effects of the CLOCK gene.