We first scanned 32 children (16 typically developing and 16 with autism) (table S1) using functional magnetic resonance imaging (fMRI) during a variation of a reward-guided implicit learning task known to engage frontostriatal circuits (28
), regions that overlap areas of CNTNAP2
gene expression during development (7
). Genotype frequency was similar between the typically developing and the autism groups, and data were collapsed across groups to investigate the specific effects of the risk variant on brain function. Comparisons between risk allele carriers (n
= 9) and nonrisk allele carriers (n
= 23) showed significantly reduced activity in the medial PFC (mPFC) during reward feedback processing in the nonrisk group (P
=0.001) (). This finding was robust when examined within diagnostic groups (fig. S1), supporting a dominant effect of the rs2710102 risk allele on brain function. This finding is consistent with previous studies that have demonstrated a dominant effect of the autism and SLI CNTNAP2
risk allele on language endophenotypes, such that heterozygous individuals perform similarly to homozygous risk carriers (1
). The mPFC is part of the default mode network (29
), which is typically more active during resting baseline than during externally directed attention. Thus, the risk group shows less of the normally observed decrease in medial prefrontal activity than does the nonrisk group during a task requiring externally directed attention (fig. S2).
Fig. 1 Comparison between risk and nonrisk carriers of the autism-associated CNTNAP2 allele (rs2710102) during reward processing. (Top) Significantly reduced activation in mPFC in the nonrisk individuals compared to risk carriers (Z > 2.3, P < (more ...)
On the basis of the known expression patterns for CNTNAP2
in the frontal cortex and its role in synaptic transmission, we hypothesized that frontal functional connectivity may serve as a biologically relevant intermediate phenotype for an association study with CNTNAP2
. Recently, it has been established that functional connectivity within the default mode is under genetic control, with a family-based estimated heritability of 0.42 (30
). Abnormal patterns of functional connectivity have also been observed in a number of neurodevelopmental disorders with large genetic determinants, including both autism and schizophrenia [see (31
) for review]. Therefore, we tested functional connectivity networks for association with our risk allele. To investigate whether this genetic risk allele modulates functional interactions between frontal systems and more posterior cortical regions, irrespective of the functional demands of the task, we performed functional connectivity analysis on residual time series after accounting for the task effects (functional connectivity MRI; see Supplementary Material). This analysis revealed a left-lateralized network composed of the left IFG, insula, anterior temporal pole, superior temporal gyrus, and angular gyrus in the nonrisk group (, first row, and table S2). Conversely, the risk carriers had more widespread and bilateral connectivity throughout the frontal cortex and anterior temporal poles (, second row, and table S2). Both groups show some evidence of connectivity with medial posterior regions such as the posterior cingulate. In contrast to the diffuse bilateral network in the risk carriers, the discrete left-lateralized frontotemporal network in the nonrisk group overlaps with regions known to be important in language processing, such as the IFG and superior temporal gyrus, and is particularly interesting given the previous association of CNTNAP2
with language abilities and the expression pattern of CNTNAP2
Fig. 2 Functional connectivity with the mPFC is associated with CNTNAP2. (A) Yellow circles highlight the discrete left-lateralized mPFC functional connectivity network observed in the nonrisk group, in contrast to the more distributed, bilateral network observed (more ...)
To confirm the diagnosis-independent relationship between CNTNAP2 genotype and functional connectivity, we examined this genotype-phenotype relationship in a separate cohort restricted to 39 typically developing children (16 female/23 male; 10 nonrisk/29 risk) scanned during a different language-learning paradigm (see Supplementary Material for details). Supporting a role for CNTNAP2 in structuring frontal connectivity, the same networks were observed (, third and fourth rows; see rows 5 and 6 for combined analyses), replicating findings obtained in our discovery cohort. In both the discovery and the replication samples, the differences in connectivity patterns between risk and nonrisk allele carriers were statistically significant, such that nonrisk showed stronger long-range anterior-posterior connectivity between the mPFC and the medial occipital and ventral temporal cortices (Pdiscovery = 2.44 × 10−6; Preplication = 5.47 × 10−9; , left), whereas the risk carriers had stronger local connectivity between the mPFC and the right frontal cortex (Pdiscovery = 4.21 × 10−4; Preplication = 1.01 × 10−4; , right). These results were consistent when examined in males only, indicating that the results were not confounded by gender (table S4). Both risk and nonrisk subjects from the discovery and replication samples demonstrated similar negative correlations with mPFC activity in ventral visual cortices, superior parietal lobe, and cerebellum (table S5). To confirm replication in the exact regions identified as significantly different between risk and nonrisk carriers in the discovery sample, we next conducted a region-of-interest (ROI) analysis based on these clusters. We were able to replicate the CNTNAP2 risk allele effects in both the right middle frontal gyrus for risk > nonrisk (P = 0.0081) and the left intracalcarine cortex for the contrasting nonrisk > risk (P = 0.0003) (table S6).
To verify that our replication and discovery samples were not significantly different, we ran between-group comparisons, restricting our investigation to all significant voxels identified in the analysis of the discovery sample. We found no significant differences between the discovery and the replication samples for risk carriers. In the nonrisk carriers, we detected a small cluster in the right orbitofrontal cortex (3 voxels; x, y, z = 28, 28, −16; Z =4.94, P < 0.05 corrected) as significantly different between discovery > replication samples and a small cluster in the left paracingulate cortex (23 voxels; x, y, z = −8, 54, 12; Z = 5.84, P <0.05 corrected) for the contrast replication > discovery. These clusters did not overlap with regions of significantly different functional connectivity observed in between-group comparisons for risk and nonrisk carriers in either discovery or replication samples. Thus, these minor differences did not influence our main results of increased local frontal connectivity and reduced long-range connectivity in risk allele carriers.