Here we meta-analyze ADHD data from Iceland and Norway (N = 8883 ADHD affected) (Methods and Supplementary Table 1). The Icelandic sample combines two ADHD study groups, a group of subjects diagnosed with ADHD and a group of subjects assumed to have ADHD based on prescription of ADHD medication. The subjects diagnosed with ADHD are on average 13.6 years younger than those prescribed medication for ADHD (mean age 30.3 and 43.9, respectively), and the combined ADHD sample has a male to female ratio of 3:2. The Norwegian ADHD sample, with a male to female ratio of 2:1, were from the Norwegian mother and child cohort study (MoBa)34,35, which includes children and adults, and from the Bergen adult ADHD study36.
We compiled a list of 19 neuropsychiatric CNVs that have been shown to confer risk of schizophrenia and/or ASD17,18,19,37,38,39 (Supplementary Table 2). All but the 2p16.3 deletions are recurrent and flanked by segmental duplications (Supplementary Fig. 1). All samples were genotyped using Illumina SNP arrays and the preselected neuropsychiatric CNVs were identified using the PennCNV algorithm and confirmed by visual inspection and segregation in pedigrees (Methods). Individually these CNVs are rare (0.0027–0.25% carrier frequency in the population), and we estimated that at 80% power a CNV with a frequency of 0.018% or greater was required to detect an association with an OR above 3.9 (Supplementary Fig. 2 and Methods). The individual associations were therefore restricted to CNVs with a population frequency of >0.018% in the combined Icelandic and Norwegian sample; deletions at 1q21.1 distal, 2p16.3 (NRXN1), 15q11.2, 15q13.3 (break point (BP)4 & BP4.5–BP5), 16p11.2 distal, 16p11.2 proximal, 16p12.1, 17p12 and 22q11.21 and duplications at 1q21.1 distal, 16p11.2 proximal, 16p13.11, 17q12 and 22q11.21.
Of the 14 CNVs tested, the two previously associated with ADHD23,26, 16p13.11 duplication and 22q11.21 deletion were replicated in the combined Icelandic and Norwegian sample (OR (95% CI) = 2.12 (1.31, 3.27), P = 0.0035 and OR (95% CI) = 10.73 (4.66, 23.15), P = 1.8 × 10−6, respectively; Cochran–Mantel–Haenszel χ2 test for count data and false discovery rate (FDR) adjusted P value); it should be noted that a part of the Icelandic sample was included in the original 16p13.11 duplication study23 (Fig. 1, Table 1 and Supplementary Table 3).
Previous reports have shown a higher frequency of ADHD in carriers of six (deletions at 1q21.1 distal, 15q11.2, 15q13.3 (BP4 & BP4.5–BP5) and 16p11.2 proximal and duplications at 1q21.1 distal and 16p11.2 proximal) of the remaining 12 CNVs, although not statistically tested27,28,29,32. We present evidence of significant association with ADHD for deletions at 15q11.2 and 15q13.3 (BP4 & BP4.5–BP5) and duplications at 1q21.1 distal and 16p11.2 proximal. The remaining six CNVs have not been associated with a diagnosis of ADHD before; the deletions at 2p16.3 (NRXN1), 16p11.2 distal, 16p12.1 and 17p12 and duplications at 17q12 and 22q11.21. Of those, the 2p16.3 (NRXN1) deletion and 22q11.21 duplication were significant in the combined sample (OR (95% CI) = 4.68 (1.82, 10.64), P = 0.0020 and OR (95% CI) = 2.24 (1.32, 3.63), P = 0.0042, respectively; Cochran–Mantel–Haenszel χ2 test for count data and FDR adjusted P value) (Fig. 1, Table 1 and Supplementary Table 3). Affected and control, carrier frequency for the five remaining, individually untested, CNVs are given in Fig. 1 and Supplementary Table 4.
Combining the 19 neuropsychiatric CNVs, we performed a Cochran–Mantel–Haenszel χ2 test for count data on the Icelandic and Norwegian ADHD samples to estimate the overall CNV burden. This revealed a significant association with ADHD (OR (95% CI) = 2.43 (2.05, 2.87), P = 1.6 × 10−21; counts adjusted for correction factor, see Methods). The combined neuropsychiatric CNVs have a carrier frequency of 2.15% in the Icelandic and Norwegian ADHD sample compared with 0.86% in the combined controls. The associations with ADHD in the Icelandic and Norwegian samples, separately, were similar (OR (95% CI) = 2.30 (1.86, 2.80), P = 1.4 × 10−11 and OR (95% CI) = 2.66 (2.04, 3.43), P = 7.7 × 10−12, respectively; Fisher’s exact test and corrected P values) (Table 2).
Removing CNVs individually associated with ADHD from the combined set, we found the remaining 11 CNVs (deletions at 1q21.1 distal, 3q29, 16p11.2 distal, 16p11.2 proximal, 16p12.1, 17p12 and 17q12 and duplications at 7q11.23, 7q36.3, 15q11.2–13.1, 17q12) still conferred a significant risk of ADHD (OR (95% CI) = 1.94 (1.33, 2.76), P = 6.0 × 10−4; Cochran–Mantel–Haenszel χ2 test for count data in the Icelandic and Norwegian sample counts adjusted with correction factors and combined). Although, this appears to be mostly accounted for by the six, relatively, more common CNVs tested, as when they are removed the OR is no longer significant for the five remaining CNVs (OR (95% CI) = 1.38 (0.27, 4.42), P = 0.49; Cochran–Mantel–Haenszel χ2 test for count data in the Icelandic and Norwegian sample counts adjusted with correction factors and combined) (Table 2).
We also explored what effect removing individuals with a diagnosis of ASD or schizophrenia, from the Icelandic sample of ADHD affected and controls, would have on the risk for ADHD conferred by the neuropsychiatric CNVs, and found a modest weakening of the signifcance as expected but a subtle increase in OR for nine out of 14 CNVs. However, the CNVs combined showed a similar, although less significant, effect (OR (95% CI) = 2.26 (1.81, 2.79), P = 1.2 × 10−11; Fisher’s exact test) (Supplementary Table 5).
For the Norwegian sample, the issue of ADHD comorbid with axis I to III disorders was addressed in a recent study where the authors found children with ADHD in MoBa were registered with fewer abnormal psychosocial situations (axis I disorders) compared with children in the general population42. Within the Bergen sample, none of the CNV carriers has a diagnosis of schizophrenia and only one is comorbid ADHD and ASD. It is, therefore, unlikely that the comorbid ASD or schizophrenia are responsible for the risk conferred by the neuropsychiatric CNVs present in the Norwegian sample.
We note that the OR for the neuropsychiatric CNVs appears to be higher in the sample of Icelandic subjects with an ADHD diagnosis than in the medication sample (OR (95% CI) = 3.49 (2.72, 4.42), P = 1.6 × 10−17 and OR (95% CI) = 1.25 (0.84, 1.79), P = 0.25, respectively; Fisher’s exact test and corrected P values. P value for difference = 2.4 × 10−5) (Supplementary Tables 6 and 7). A corresponding analysis in the Norwegian sample reveals a comparable, although nonsignificant (P = 0.10), trend where the OR is higher in children with ADHD than in adults with ADHD (OR (95% CI) = 2.69 (1.89, 3.78), P = 5.9 × 10−8 and OR (95% CI) = 1.56 (0.86, 2.63), P = 0.11, respectively; Fisher’s exact test and corrected P value) (Supplementary Tables 8 and 9).