Preserved Global Resting-state Architecture
The whole-brain mean correlation matrices capture the entirety of intra-hemispheric resting-state data for the 42 age-matched controls and 26 child deletion carriers (Figure 1). The deletion carrier matrix has a similar overall network architecture compared to control participants. Both groups have positive within-network correlations (at the diagonals), and largely a similar pattern of positive and negative between-network correlations. Differences at p < 0.05 uncorrected are shown in the smaller bottom right matrix, although no single connection survived correction for multiple comparisons. Overall, this broad-scale view of intra-hemispheric functional connectivity demonstrates that 16p11.2 deletion carriers preserve the general organization of large-scale networks. In other words, the carrier group does not have a disorganized, unrecognizable connectivity matrix; nor is there any single large-scale network, such as the default network, that is wholly disrupted.
Degree Centrality for Local and Distant Connectivity
We then estimated the degree centrality of local and distant functional connectivity across the cortex for 16p11.2 deletion carriers. Typically, the human cortex shows a high degree of local connectivity at primary sensory cortices, medial prefrontal cortex and precuneus; whereas a high degree of distant connectivity is seen at association cortices, particularly at cortical hubs [24, 25]. The child control subjects replicated well the local and distant connectivity maps of degree centrality seen in a prior study using healthy adults , upon which our current methodology is based. Deletion carriers have a high degree of local links at primary cortices and the medial surface, and a high degree of distant links at heteromodal association areas similar to controls share the same general architecture as controls in terms of regions with a low or high degree centrality across the cortex (Figure 2A, D). Although regional differences do occur (shown later), at a broad level the maps are very similar. Deletion carriers have a high degree of local links at the medial surface and primary cortices, and high degree of distant links at heteromodal association areas similar to controls. Furthermore, these degree centrality maps do not demonstrate a uniform global decrease for all distant connections, or a uniform global increase in local connectivity across the cortex for deletion carriers. This assertion can be quantified by summating the total number of links across the cortex for each subject’s local map, and independently, their distant map. The group mean for the global sum of degree links is shown for local and distant connectivity (2B, E). No significant group differences are observed in local connectivity (t= -1.39, p= 0.18, Cohen’s d=0.34) or distant connectivity (t=1.21, p=0.23, Cohen’s d=0.30). To investigate further whether differences occurred at particular brain regions, the difference in degree centrality for controls minus deletion carriers is shown (2C, F) for local connectivity and distant connectivity as a t-statistic map (2C, F). A significantly greater degree of connections in controls is seen in red (p< 0.05), and a greater degree of connectivity in deletion carriers is seen in blue (p < 0.05). A decrease in the degree of local connectivity is seen in deletion carriers at the temporoparietal junction (TPJ), anterior medial prefrontal cortex (aMPFC), paracentral gyrus, and visual cortex. An increase in degree of local connectivity in deletion carriers is observed at lateral temporal cortex (LTC), the insula, posterior cingulate (PCC), anterior cingulate and the ventromedial prefrontal cortex. In regards to the distant map a decrease in the degree of distant connectivity in deletion carriers is observed at the inferior temporal gyrus, precentral gyrus, aMPFC, and cingulate sulcus; whereas an increase in the degree of connectivity is observed in deletion carriers at the superior temporal gyrus (STG), parietal operculum (OP), inferior and middle frontal gyrus (IFG, MFG), and the pre-Supplemental motor area (pre-SMA). Notice how these particular regions in blue in deletion carriers had a greater number higher degree of distant connections in deletion carriers despite an overall trend towards fewer long-range connections. In other words, both directions of change were observed (regions with increased degree in carriers and increased degree in controls). Regardless, no regions survived correction for multiple comparisons using FDR correction. These results were replicated using an independent methodology, FSL’s randomise function (v5.0.4). Statistical results were nearly identical between the two methods. Additionally, randomise can perform a threshold-free cluster enhancement of the data akin to a bootstrap as an alternative to FDR. No regions survived correction for multiple comparisons using either method. The left hemisphere is representative of the pattern seen in the right hemisphere, which similarly did not have any regions that survived FDR correction. Overall, quantitative differences in degree centrality were not a salient feature in differentiating the functional connectivity of control participants from 16p11.2 deletion carriers.
Degree of Differential Links Reveal Systems with Topographically Distinct Connectivity
Whole-brain correlation matrices of individuals can also discern unique connections in one group but not the other when analyzed link-by-link. The advantage of this kind of analysis is that it is a hypothesis-free, data-driven method for identifying distinct patterns of functional connectivity simultaneously across the cortex. As detailed in the methods section, if node i is connected to nodes a, f, p in one group but not the other, such that z-transformed-r is significantly different at those particular links, then a value of 3 will be displayed on the cortical map at node i (i.e, the 3 differential links i-a, i-f, i-p are counted while all of the shared connections are not (i-x, i-…n)). Only differential links (DL) significant at p<0.001 (two-tailed) are displayed (Figure 3). The results show that differential links are non-uniformly distributed across the cortex. These regions with a high concentration of DL are ‘hot spots’ of distinct functional connectivity. The auditory cortex, insula, middle-posterior cingulate and the paracentral gyrus had the greatest number of differential links for local connectivity (3A). For distant connectivity, the auditory cortex and insula were again present, but additionally the inferior parietal lobule (IPL), left MFG, right IFG, posterior mid-cingulate and posterior cingulate sulcus, and dorsal and anterior MPFC were regions with the greatest differential links (3B). Areas circumscribed in black denote DL that survived multiple comparisons from FDR correction.
Knowing that quantitative differences did not significantly differ between our groups in the number of connections at a given node (from the degree centrality analysis), a differential link should then represent a topographical displacement of a node to its connection (from iàj in controls to iàk in carriers); akin to a qualitative rather than a quantitative change. To visually demonstrate differences in topography, as a proof-of-principle, a seed was placed at these aforementioned hot spots (specifically regions with >50 DL at p < 0.001). The precise seed locations are displayed on the left hemisphere distant connectivity DL map (Figure 4, upper box), which was overlaid with the borders of the 7-network solution to better characterize network-level changes . The mean group maps for seed-based functional connectivity are displayed with control participants on the left, and 16p11.2 deletion carriers on the right. Statistical testing was intentionally not performed on this seed-based analysis to avoid circularity. The first three seed locations share an interwoven story (Figure 4A-C). First, the IPL shows correlation with the canonical default network (DN) in control participants (7-network, salmon) (4A). In deletion carriers the IPL does not appear to be functionally connected to the dorsal medial prefrontal cortex (dmPFC). Nor are there the expected negative correlations to the superior parietal lobule (SPL) of the dorsal attention network (dATN), or OP of the ventral attention/salience network (SN) visualized in controls (7-network, green and magenta respectively). Next, the auditory cortex, specifically at the secondary auditory cortex (or A2) is seeded (4B). The control group demonstrates the canonical pattern of connectivity restricted to nearby primary sensorimotor cortices (7-network, blue) with scant connectivity to the medial surface. The auditory seed in deletion carriers, however, shows aberrant connectivity to the dmPFC as well as robust connectivity to the other nodes of the default network (7-network, salmon). This is not an artifact of anatomical displacement of the seed location between groups, or from projection volume-space to surface-space (Supplemental Figure 2). Remarkably, these are the same regions that appeared to be missing from seeding the left IPL. Thus, two independent, large-scale resting-state networks are functionally correlated in deletion carriers, and these links between networks are not present in controls. Specifically, the secondary auditory cortex is functionally connected to the TPJ, lateral temporal cortex, temporal pole, IFG and MFG – which constitute the Dorsal Medial subsystem of the Default Network [34, 35] (Figure 4 lower box, in salmon). Finally, if that particular node at dmPFC is seeded (4C) deletion carriers do indeed show connectivity to the other regions of the DN, proving that the region is not functionally disconnected (or “underconnected”) from the rest of the cortex. It is not the case that the cortex itself at the dmPFC is problematic - as one would surmise if a study only performed a region of interest analysis in isolation at the IPL. It is also noteworthy that if this dataset had been approached with a traditional region-of-interest seed-based approach, one may incorrectly conclude that the dmPFC node is disconnected from the rest of the DN, and point to problematic cortex at that location. Alternatively, one could also conclude that the DN is “underconnected” between the IPL and the dmPFC. Both of these inferences are incorrect because we demonstrated that a seed placed at the dmPFC in deletion carriers continued to connect robustly to the other nodes of the dorsal medial subsystem of the DN as well as to the auditory cortex. It is conceivable that the very early functional magnetic resonance imaging (fMRI) studies of ASD fell into this methodological trap. Here the dmPFC is differentially connected to the auditory node. Additionally, this node of the dorsal medial subsystem has diminished connectivity to the posterior IPL and PCC of the DN core (Figure 4 lower box, in yellow). This suggests that the dorsal medial subsystem is no longer tightly integrated to the core DN. More recent higher-resolution, intra-individual analyses describe no DN Core subsystem at all, but two parallel interdigitated system that give the appearance of a DN Core when group averaged . In this re-interpretation the explanation becomes more simple - Network B of the DN is affected in 16p11.2 deletion carriers, while Network A of the DN is not. Collectively these three seed-based locations reveal a shift in topographical connectivity.
The second major finding in the left hemisphere involves the posterior insula (4D). Control participants show strong connectivity (z(r) ≥0.5) constrained locally to area circumscribed around the seed, which explains its preferential clustering in a winner-take-all parcellation to the somatomotor network (7-network, blue). Mild negative correlations are seen at parts of the dorsolateral prefrontal cortex of the frontoparietal control network (FPCN) (7-network, orange) and the SPL of the dATN (7-nework, green). This pattern of light blue is remarkably similar to the negative correlation pattern seen from seeding the auditory cortex in control participants in Figure 4A. Deletion carriers, in contrast, display strong connectivity extending into the middle insula and the OP – both nodes of the SN (7-network, magenta). One may recollect that enhanced functional connectivity between the Somatomotor and Salience networks was suggested by the correlation matrix in Figure 1. Negative correlations have also shifted to the default network, as one would expect with the salience network . The posterior insula is the primary sensory cortex for interoception (heat, pain, sensual, visceral and homeostatic sensations) [38-40]. The middle insula re-represents this construct with emotionally salient and environmental stimuli [41-44], and finally the anterior insula then integrates this percept with motivational, social and cognitive conditions  as a site of multimodal integration . Craig proposes a forward gradient starting with the interoceptive sensation x (posterior insula), to the subjective feeling of x (middle insula), and then the meta-cognitive self-awareness of x (anterior insula) . This model is also in line with the 7-network parcellation used in our analysis: posterior insula (somatomotor network, blue), middle insula (SN, magenta), and anterior insula (FPCN, orange).
Aberrant links in both cases, auditory-DN and posterior insula-SN, violate fundamental large-scale circuit properties by functionally connecting regions that specialize in local and hierarchical processing of sensation (somatomotor network) with regions that specialize in parallel-processing across widely-distributed networks (default and salience networks.).
Differential links also reveal shifts in topographic connectivity in the right hemisphere (Supplemental Figures 3 and 4).
Characterization of Group-Level Behavior
Hahamy and colleages  reported that high inter-subject variability in connectivity is the idiosyncratic defining feature of ASD, and the magnitude of change correlated with behavior. They used 5 sites from the ABIDE dataset, but 6 of the 20 sites in ABIDE collapsed data from ASD subtypes (autistic disorder, Asperger syndrome, pervasive development disorder-not otherwise specified) . Could this high inter-subject variance simply reflect the high degree of genetic heterogeneity? In contrast, 16p11.2 deletion carriers show stereotyped group-level behavior. A representative aberrant link connecting the auditory cortex with the TPJ of the default network can demonstrate this correlation with behavior at the group-level. The differential functional connectivity between carriers and controls in the left auditory node and left TPJ node is shown (Figure 5). The mean z-transformed Pearson’s r is displayed after regression of nuisance variables (age, sex, handedness, scanner site, and micromovements greater than 0.1 mm), which is referred to as the ‘adjusted’ value. In controls the adjusted mean z(r)= 0.06 ±0.24 and median z(r)= 0.03, while deletion carriers have an adjusted mean z(r)= 0.40 ±0.28, and median z(r)= 0.45 (5A). Statistical significance was not assessed between the two groups as to avoid circularity. The result is similar for the raw, uncorrected mean value in controls z(r)= 0.05 ±0.24 and deletion carriers z(r)= 0.43 ±0.29. The Shapiro-Wilk test confirms normality for the adj. left auditory-TPJ link (W= 0.941, p= 0.19), and Levene’s test confirms equal variance between groups (F= 0.80, p= 0.37). These results demonstrate that the difference at this link is not simply from high variance in deletion group, but rather a shift in the curve that represents a gain-of-function connection not present in controls. Next, a scatterplot of the left auditory–TPJ link z(r) is plotted against Vineland-II scores, a measure of adaptive social behavior (abnormal ≤ 85) (5B). The black line represents the best line of fit for the entire cohort (r= -0.46). Similarly, the correlation with SRS (a measure of broad social functioning, abnormal ≥60) is r= 0.42. The differential link is also plotted against CELF scores, a comprehensive battery of language ability (abnormal ≤ 85) (5C). The black line represents the best line of fit for the entire cohort (r= -0.39). This is a moderate group-level correlation between the connection strength of the left auditory cortex and the left TPJ with behavior – both to social impairment (Vineland-II) and language (CELF). The bolded triangles on the Vineland-II plot represent individuals with a DSM diagnosis of phonological processing disorder, and conversely on the CELF plot the bold triangles represents those who met diagnostic criteria for ASD (on either ADI or ADOS). Interestingly, at the level of greatest aberrant link strength (~0.8-1) only the subgroup of carriers with both language and social impairment (bolded triangles) are present. Of note, regression of group from the results removes the factor being studied. The aim of this study is to characterize group-level effects of the 16p11.2 CNV, induced from many other environmental, genetic and epigenetic factors that affect a particular individual.
The other main hubs of the default network showed connectivity similar to the auditory-TPJ node: left auditory-PCC (mean z(r): 0.07 ± 0.3 controls and 0.41 ± 0.3 in deletions, Vineland-II r= -0.46 and CELF r= -0.41), and left auditory-dMPFC (mean z(r) 0.12 ± 0.2 controls and 0.44 ± 0.2 in deletions, Vineland-II r=0.-37 and CELF r= -0.35). Interestingly, with the topographical shift of the auditory node away from the somatomotor network the auditory node is no longer connected to the posterior insula (mean z(r) 0.29 ± 0.2 controls and -0.02 ± 0.2 in deletions) with reversal in direction in correlation with group behavior (Vineland-II r= 0.39 and CELF r= 0.40). Finally, to reiterate that the core default network is intact, and not directly involved, the left dMPFC-PCC node is functionally connected in both groups (mean z(r): 0.44 ± 0.3 controls and 0.36 ± 0.3 in deletions) without correlation with group behavior (Vineland-II r= 0.13 and CELF r= 0.14).
Reduced Right-hemispheric Functional Lateralization
Functional lateralization was assessed by measuring the functional connectivity in the most lateralized connections of the cerebral cortex, which were determined in an independent sample of 100 participants without any history of psychiatric illness. The regions involved in the most left-lateralized connections were found almost exclusively in the default network (Figure 6A, left), whereas the regions in the most right-lateralized connections were found in the ventral attention, dorsal attention, frontoparietal control, and visual networks (Figure 6A, right). After determining the most lateralized connections in the independent sample, we then calculated the mean iLI for each subject in the 16p11.2 CNV sample. Group differences in mean left and right lateralization are shown (Figure 6B). The deletion carriers were less lateralized in the right hemisphere than age-matched controls even after controlling for age, sex, handedness, imaging site, and micromovements (t= -2.44, p= 0.02, Cohen’s d=0.69); however, the deletion carriers’ left iLI did not differ significantly from age-matched controls’ left iLI (t= 1.75, p= 0.09, Cohen’s d=0.52).
Reduced Left-hemispheric Laterality in Individuals Relates to IQ and Language
As follow up to the group difference in the deletion carriers, we tested whether any relationship existed between (1) left functional lateralization and language ability and (2) right functional lateralization and visuospatial ability (Figure 7). As hypothesized, language ability was negatively correlated with left iLI when controlling for sex, age, handedness, imaging site, and micro-movements (VIQ: t= -3.01, p= 0.004, R2= 0.14; CELF: t= -1.96, p= 0.05, R2= 0.07). In other words, as left lateralization increased (i.e., became more negative), language ability likewise increased. The relationship between left iLI and VIQ remained, even after controlling for group (t= -2.21, p= 0.03, R2= 0.08); however, the relationship between left iLI and CELF went away after controlling for group (t= -0.97, p= 0.34, R2= 0.02). No relationship was found between right iLI and visuospatial ability (NVIQ:t= 0.59, p= 0.56, R2= 0.01).