The CBCL-JBD phenotype is a parent-rated measure based on a profile of CBCL scale-elevations that have been replicated in bipolar youth. The three scales that contribute to the overall score (ie Attention Problems, Aggressive Behavior and Anxious/Depressed) contain items that reflect symptoms of cognitive, behavioral and emotional dysregulation. We conducted a genomewide linkage scan of this measure in 154 families ascertained through ADHD sibling pairs. Based on a panel of 4885 SNP markers, our scan identified three regions of the genome (1p21.1, 6p21.3 and 8q21.13) that surpassed empirically-defined criteria for suggestive linkage. These results extend support for the suitability of the CBCL-JBD phenotype for molecular genetic analyses and raise the possibility that QTL under these peaks influence psychiatric symptomatology in youth.
Prior to this analysis, only one study had examined the CBCL-JBD phenotype using a genomewide approach. McGough et al.30
found a region of suggestive linkage on 2q23, a location at which genomewide significance had been achieved in a two-dimensional linkage scan of adult bipolar disorder in 52 families. Our current study did not yield a linkage peak in this region. Inconsistent patterns of replication are often found in linkage studies of complex phenotypes, most likely due to genetic heterogeneity and the low power of linkage studies to find genes of small effect.45
In the current case, it is possible that this lack of convergence is due to differences in methods or sample characteristics, e.g. the UCLA study required that two parents be available for study, used a different diagnostic interview (i.e. KSADS-PL), had a lower rate of BPD and larger percentage of non-Caucasian subjects and used a microsatellite-based genotyping platform with less dense coverage.
Although not overlapping with McGough et al.30
, the suggestive linkage peaks from the current scan include regions highlighted by prior genome scans of major psychiatric disorders. Like McGough et al., our data show convergence with risk loci identified by prior genomewide studies of BPD in adults. As shown in , our peaks on 6p21.3 and 8q21.13 were highlighted in a meta-analysis of genomewide expression studies of bipolar disorder. 46
Additionally, the region on 6p21 included a SNP ranked in the top 200 results from the STEP-UCL GWAS of BPD in adults47
, and the region on 8q21 overlapped with secondary-level associations in the bipolar sample from the WTCCC GWAS (1 × 10 -5 <p value <1× 10-4).48
Moreover, all three regions overlap with genomewide scans of autism and schizophrenia. Our region on 1p21.1 overlapped with the primary findings in two previous linkage studies of schizophrenia and one of autism. For example, in 236 affected sibling pair families from Japan49
, this region was the site of the primary (genomewide significant) linkage peak for schizophrenia in the study. In a smaller sample of 22 Canadian families with a heavy schizophrenia loading, this region harbored one of 18 peaks with a LOD score of >1.5 in.50
This region also included the peak marker in an initial genomewide scan of 90 multiplex autism families51
that showed allele sharing in an additional 49 families, though the combined sample fell short of genomewide significance.
Peaks on 6p21.3 and 8q21.13 also overlap with putative risk loci identified through genomewide scans of schizophrenia and autism. A meta-analysis of genomewide linkage scans from 200352
identified the 6p region as one of twelve (out of 120) bins considered strong schizophrenia candidate regions with P AvgRnk <.05 in weighted and unweighted analysis. Additionally, one of the top 25 SNPs from the first genomewide association scan of schizophrenia53
fell in this region, which also overlaps with 3 of 12 clusters of interest in the same study. The region identified on 8q21.13 overlapped with a secondary level association in the CATIE GWAS of schizophrenia.53
And, both the 6p and 8q regions were sites of autism-specific copy number variants.54
The current results are intriguing given data supporting a genetic overlap between schizophrenia and BPD,55
schizophrenia and autism 56
and growing evidence for a three-way genetic overlap between schizophrenia, BPD and autism.57
Although a small number of candidate genes that may influence this cluster of disorders have been explored in other regions (e.g. DISC1),58
more variants are expected to emerge to explain the familial and genetic overlap of these conditions. Our data raise the possibility that such variants lie within our highlighted regions and are indexed by the CBCL-JBD scale.
Determining precisely what construct is measured by the CBCL-JBD phenotype goes beyond the current study. Yet, convergence with genomewide scans of a range of conditions lends support to the hypothesis that the measure reflects a heritable trait relevant to a range of severe psychopathology rather than a specific measure of pediatric BPD per se. The items on the three scales that contribute to the CBCL-JBD profile reflect emotional and behavioral lability and distractability, i.e. items that index the capacity for self-regulation across a wide range of domains (i.e. cognitive, behavioral and affective). Further evidence for this conceptualization comes from Ayer and colleagues59
who found that the CBCL –JBD phenotype can be modeled as sharing a single latent trait with a different secondary CBCL scale purported to measure post-traumatic stress problems (PTSP). Like the CBCL-JBD phenotype, the PTSP scale is associated with suicidality and poor outcome and features a number of items overlapping with the CBCL-JBD that relate to self-regulation. Based on this analysis, the authors suggest both scales index a single dysregulatory syndrome.
The fact that the CBCL-JBD phenotype taps into a trait relevant to a range of psychiatric disorders may help to explain the profile’s lack of diagnostic specificity to juvenile-onset BPD in clinical studies.26
Yet, the empirical literature supports the relevance of the profile to the emotional and behavioral dysregulation that is characteristic of BPD in youth, thereby leaving open the possibility that genes that influence variation on the measure are relevant to this disorder. The profile of elevations that form the basis of the phenotype has been repeatedly replicated in affected children and adolescents.18
Faraone et al.19
also showed that the profile has excellent diagnostic efficiency for concurrent interview-based bipolar diagnoses in children and adolescents (AUC’s =.82 – .97) from an ADHD family study, and Diler et al.26
showed moderate diagnostic efficiency in retrospective ratings of individuals <12 years from a study of bipolar children (AUCs: .70–.77). A recent study21
further demonstrated good longitudinal prediction of BPD and associated features such as hospitalization, major depression and conduct disorder. Even in studies that have indicated poor diagnostic utility for the measure,22–24,26
the association of high CBCL-JBD scores with suicidality and hospitalization22–24,26
supports its relevance to a severe mood disorder.
Further work is needed to investigate susceptibility genes that may exist in these regions and their relationship to different forms of psychopathology. Given that genomewide association studies will have greater power and resolution than standard linkage scans to identify common variants of small effect, such studies provide the next logical step to follow-up these findings. Still, genomewide linkage scans such as this remain useful for identifying rare variants with strong effects in a small number of families or multiple mutations in the same gene that may differ across families. Thus, a lack of replication of the current results in a standard GWAS study of the CBCL-JBD phenotype would suggest that these other possibilities should be considered.
Such work is needed given how little is currently known about the genetic risk factors for mood disorders in children. As Mick and Faraone60
have recently reviewed, family studies and secondary analyses in adult bipolar linkage studies support the hypothesis that an early age of onset is associated with a greater genetic risk than adult-onset BPD. Nonetheless, no regions that have emerged in adult linkage studies of early-onset cases have been replicated or overlap with the current findings. A handful of candidate gene studies have found interesting associations with pediatric BPD for BDNF, SLC6A3 and GAD1, but contradictory findings and failures to replicate have not yielded definitive conclusions, and new candidates from the adult bipolar literature have yet to be tested in affected pediatric samples.
The current study should be considered in the context of its limitations. Our analyses, like those of McGough et al.30
, were conducted in families originally ascertained through ADHD sibling pairs. Such a sample is relevant to the etiology of pediatric BPD given evidence from multiple samples of shared familial influences on ADHD and the disorder 61
and the high rate of ADHD in youth with bipolar diagnoses. Additionally, the heritability of the phenotype in the current sample was comparable to the heritability estimated from large population-based twin studies. Moreover, the QTL identified in these two studies do not overlap with loci identified through genomewide studies of ADHD, with the exception of the finding on 6p21.3 which included one of the top 25 SNPs from a genomewide association scan of ADHD.62
Thus, the current findings do not appear solely driven by ADHD. Still, further studies are needed to replicate and extend our findings in non-ADHD samples.
A second limitation is the low rate at which subjects in the sample returned the CBCL questionnaires. Our analysis is based on only 54% of the original families in our linkage study. Yet the comparability of subject characteristics, including IQ and rates of comorbid diagnoses in this analysis, to those in the original study suggests that the current sample is not skewed in terms of high functioning families.
A small number of additional factors are worthy of consideration. For example, we note that the CBCL-JBD phenotype was positively skewed. However, the evidence for linkage at each marker was estimated through simulations (ie. we generated empirical p-values rather than rely on asymptotic theory). Thus, any potential bias introduced by normality deviations would have been accounted for (i.e. no increase in false-positive rates would be expected because of the normality deviation). We further note that this is the third paper based on this same linkage sample, following published papers examining the ADHD phenotype specifically32
and a multivariate analysis of neurocognitive traits and ADHD symptoms33
. Any single study can only realistically attempt to control its own study-wide error rate, and this and previous papers addressed specific, separate a priori hypotheses. Ultimately, though, issues of multiple testing should be acknowledged and our results require confirmation in independent samples. We also acknowledge that evidence for linkage to the identified regions is only suggestive. This finding, however, generates a very specific set of hypotheses (that loci regulating this phenotype are located in three defined chromosomal regions) that can be addressed by follow-up association studies.
Despite these concerns, our data raise the possibility that QTL on 1p21 6p21 and 8q21 contribute to variation on the CBCL-JBD phenotype. Further work is needed to confirm these findings, explore genes under these peaks and investigate the compelling hypothesis that risk variants in these regions contribute to severe, early-onset psychopathology and/or difficulties with self-regulation that overlap genetically with the shared risk for bipolar disorder, schizophrenia and autism.