Default Mode Hypoconnectivity Underlies a Sex-Related Autism Spectrum

Background Females and males differ significantly in the prevalence and presentation of autism spectrum conditions. One theory of this effect postulates that autistic traits lie on a sex-related continuum in the general population, and autism represents the extreme male end of this spectrum. This theory predicts that any feature of autism in males should 1) be present in autistic females, 2) differentiate between the sexes in the typical population, and 3) correlate with autistic traits. We tested these three predictions for default mode network (DMN) hypoconnectivity during the resting state, one of the most robustly found neurobiological differences in autism. Methods We analyzed a primary dataset of adolescents (N = 121, 12–18 years of age) containing a relatively large number of females and a replication multisite dataset including children, adolescents, and adults (N = 980, 6–58 years of age). We quantified the average connectivity between DMN regions and tested for group differences and correlation with behavioral performance using robust regression. Results We found significant differences in DMN intraconnectivity between female controls and females with autism (p = .001 in the primary dataset; p = .009 in the replication dataset), and between female controls and male controls (p = .036 in the primary dataset; p = .002 in the replication dataset). We also found a significant correlation between DMN intraconnectivity and performance on a mentalizing task (p = .001) in the primary dataset. Conclusions Collectively, these findings provide the first evidence for DMN hypoconnectivity as a behaviorally relevant neuroimaging phenotype of the sex-related spectrum of autistic traits, of which autism represents the extreme case.


Participant Details Primary Dataset: Cambridge Family Study of Autism
For our first question, whether default mode network (DMN) connectivity constitutes a marker of autism in females, we included all of the medication-free adolescent girls diagnosed with an ASC (n = 16), typically-developing adolescent girls (n = 20), and unaffected female siblings (n = 30, 25 unrelated to ASC girls in our sample) whose data was collected as part of the Cambridge Family Study of Autism (CFSA) (1)(2)(3)(4)(5)(6)(7). All of the data and measures analyzed in this study were collected in these previous studies, which also confirmed the diagnostic status of the ASC group with gold-standard diagnostic tools (8; 9).
Siblings and controls both scored below the cut-off that differentiates them from people with ASC on the Autism-Spectrum Quotient (AQ) (10)

and the Social Communication
Questionnaire (SCQ) (11). Ethical approval for this study was granted by the Cambridgeshire 1 Research Ethics Committee. Demographic details are displayed in Table S1. We previously compared males with ASC, male siblings and controls in this dataset (6). In that study, we analyzed a subset of male participants, aiming to maximally match participant demographics. Here, we used a regression strategy to control for heterogeneity, and so were able to include all 20 typically-developing control males and 35 males with ASC to answer our second question on the presence of DMN hypoconnectivity in the typical population. Demographics and matching with control females are displayed in Table S2. Behavioral (mentalizing) and fMRI data was additionally collected for male siblings of people with autism (characteristics in Table S3). These were not relevant for the first two questions, but were included to assess the third question relating DMN intra-connectivity to task performance. This correlation remained significant in the absence of male siblings.

Positive Control Dataset: MR-IMPACT Adolescent Depression
The positive control dataset contained participants from a Cambridge study on depression (25). Males and females met diagnostic criteria for major depressive disorder as defined by DSM-IV (26). Patients with chronic and/or recurrent depressive episodes were included in the sample, as were those who were currently taking SSRI medication, but patients with drug/alcohol dependence, a learning disability, structural abnormalities of screening MRI scans, and/or importantly an ASC, were excluded from taking part. IQ data was not collected for all participants in the depressed groups. Ethical approval for this study was obtained from Cambridgeshire 2 Research Ethics Committee.

Replication Dataset
Full details of individual site acquisitions are provided in the original studies (12).

Control for Head Motion Artifacts
Head motion during acquisition can create artifacts of dysconnectivity in functional imaging datasets (27)(28)(29)(30). We conducted several quality control checks to examine the impact of motion on our functional connectivity (FC) estimates.
Firstly, we statistically compared motion parameters between groups by extracting six location parameters from the scans of each participant for each slice during the scan time-

Primary Dataset
We first tested the effect of motion on our data from the CFSA. Participants did not show gross movements in visual inspection of scans, and statistical analysis also showed no significant differences between the three female groups in mean motion (p = .126) or maximum motion (p = .264), and no differences in the other contrast of interest between control males and females for mean motion (p = .422) or maximum motion (p = .552). Figure   S1 shows the moving average of the correlations between edge weights and motion as a function of Euclidean distance between nodes. The correlation between FC and mean FD, and between FC and maximum FD, was close to zero and showed little distance dependence.
Both the slope and the magnitude of the correlation were not significantly different from the distribution obtained under the null hypothesis of no relation between FD and FC (p > .1).
These results indicate that our preprocessing pipeline was sufficient to remove global artifacts induced by motion from our connectivity estimates. Figure S1. Moving average of correlation between maximum framewise displacement (A) or mean framewise displacement (B) and functional connectivity against distance between nodes, for the primary dataset. The bold red lines reflect values from actual data, whilst straight red lines are fitted linear functions: gray lines were obtained by permuting movement values for participants.
We also examined the relationship between movement parameters and DMN intraconnectivity specifically, with all five subject groups included (n = 121). DMN intraconnectivity correlated with mean motion (r = − .290, p = .001) and maximum motion (r = − .319, p < .001). These results together indicate that DMN is more related to motion than overall functional connectivity patterns.

Replication Dataset
We repeated these steps for the independent data of 980 participants.  Figure S2). This suggests that there was still a relation between motion and FC, even after our preprocessing pipeline. As above, DMN intra-connectivity strongly correlated with mean movement.

Correlation Between DMN Connectivity and Motion
A correlation between motion and FC has been found in many studies. Although often considered artifactual, recent work has shown that, for the DMN in particular, weaker connectivity could also be a stable neurobiological trait that predisposes to movement (35).
These authors found DMN connectivity to be similar for high-and low-motion scans when these scans were from the same participant, but connectivity to differ when the scans were from distinct participants. This implies that the reduced DMN connectivity associated with movement is not the result of artifacts for the particular scan that contains high movement, but a neurobiological phenotype of certain individuals, shown in scans both when the individual moves much or little.
To further inspect whether the correlation between connectivity and motion may be particularly strong in the DMN in our data, we plotted the correlation values for edges within the DMN, and for all other edges. We found the DMN edges to be more strongly affected by motion than non-DMN edges ( Figure S3). Given the evidence by Zeng et al., this residual correlation between DMN and motion may be the result of an inherent trait. As our own dataset contained repeated measurements, each subject was scanned not just at rest but also during three tasks, we were able to perform a test along the lines of Zeng et al. We asked whether subject motion during tasks would predict DMN connectivity in rest, after correction for the motion in rest. If the correlation between DMN connectivity and motion were artifactual, no further information should be contained in the task motion for the rest connectivity. If DMN connectivity were a sign of a neurobiological trait that predisposes to movement, we should find that motion in task predicts rest connectivity. In other words, a subject that moves a lot in the tasks but happens to lie still during rest would still have low DMN connectivity during rest, and vice versa.
We tested whether motion in different scans predicted connectivity during rest using a partial correlation, which measures the degree of association between mean movement during task, x, and DMN connectivity, y, removing the effect of motion during rest, z. That is, we correlated 1) the residuals obtained after regressing x against z with 2) the residuals obtained after regressing y against z. We found a significant partial correlation of −0.19 (p = .03), corroborating the possibility that the correlation between DMN connectivity and motion may be of neurobiological interest, rather than an artifact.

Computation of the DMN Connectivity Metric
We defined connectivity between each pair of brain regions for each participant as the Pearson correlation of the voxel-average regional time series. We then thresholded the connectivity matrix for each participant, only keeping the 20% (0.2) strongest connections, in order to remove weak and spurious connections. We defined our measure of DMN intraconnectivity as the fraction of connections between DMN regions that survived the threshold, corrected for the fraction of such connections expected if the DMN were as strongly connected as any part of the brain. More formally, we defined our measure as, Where M DMN is the number of observed connections between DMN regions, N DMN is the number of DMN regions, and P threshold = 0.2 is the connectivity threshold. Below, we describe a sensitivity analysis that uses a measure that does not depend on a threshold.

Computation of Difference in Connectivity Between Groups
To get an idea of the magnitude of any difference in DMN connectivity found, we computed relative connectivity strength. For each individual in our dataset, we computed the predicted DMN intra-connectivity from our regression model. We computed the mean m i for each group of interest, as the average of the predicted values for each individual from that group.
We then computed the relative increase from group over group as (m im j ) ⁄ m j . We took the control males as our baseline group m j , and female controls, males with ASC and female siblings as the groups m i .

Robustness Analyses
To examine the robustness of our regression results, we computed additional models with important variations on our original methodology for our replication dataset. We aimed to test whether our original effects, particularly those relating to sex and ASC diagnosis, remained consistent or were due to the configuration of our tests. The analyses are described below, and the results are presented in Table S7. Note that the high intercept shown in this table simply means that the DMN is more strongly connected than randomly chosen brain regions. We highlighted the regressors of prime interest in bold. These are taken to the appropriate baseline, e.g. Males with ASC -Male controls refers to the beta associated with males with ASC, when male controls are taken as the baseline.

Exclusion of High-Motion Participants
To ensure results were not driven by a small group of subjects showing high motion, we excluded all participants with maximum FD larger than 2.5 mm, where FD was calculated as the summed absolute values of derivatives of the translational and rotational realignment estimates, after converting rotational estimates to displacement at 50 mm radius (31). 196 participants were discarded, leaving 784 (80%) participants in the analysis.

Additional Preprocessing: Scrubbing
Although our preprocessing pipeline was designed to remove artifacts caused by motion, we additionally tested whether our results are sensitive to the more radical technique of scrubbing. This technique consists of discarding data from volumes taken during excessive motion. We discarded all data points for all participants for which FD > 0.25, and excluded participants with fewer than 70%, or fewer than 100 data points remaining. This left 751 participants (77%).

Threshold-free Measure of Intra-DMN Connectivity
The measure of DMN connectivity we used in the main text is based on a threshold, aimed to reduce the effect of spurious correlations. To test for sensitivity to this approach, we reran our analysis, quantifying intra-DMN connectivity as the average correlation weight between all the nodes in the DMN, after subtracting the mean connectivity weight across the brain for each participant.
This alternative measure also gave us an alternative to calculating the group effect sizes. We again calculated the percentage decrease/increase of DMN intra-connectivity of ASC males (control females) with respect to control males, and found 17% (21%). This was comparable with the values found in the main text of 16% (27%).

Regressing Out Motion
Although it is clear that in-scanner motion is related to our connectivity measure (see section Motion), we here tested whether the group effect could be explained solely by movement.
We therefore included both maximum and mean movement as regressors.

Excluding Studies Previously Reporting DMN Differences
There are a number of studies in the literature that report DMN functional connectivity differences between subjects with ASC and controls. Three of the sites in the ABIDE dataset contain participants analyzed in these studies (UM, Stanford, Olin (15; 16; 22; 23)). To avoid potential for circularity, we repeated our analysis following complete removal of data from these three sites. Note that we had to remove all data from the three sites as we could not establish which of the participants included in ABIDE are those included in the published studies.

Different Age Groups
We investigated whether the effects were present throughout the lifespan. We repeated our robust regression analysis, for children (aged below 12), adolescent (aged 12-18) and adults (aged above 18). We reproduced results for children and adolescents, but not adults (Table   S8).

Positive Control: Depression
We performed two robust regression analyses to test the specificity of DMN intraconnectivity to ASC using the MR-IMPACT study data, combined with the pooled CFSA and ABIDE data. Firstly, we included age, sex, study and diagnosis as predictors. Secondly, we tested for specific effects in one of the sexes, by including an interaction effect between diagnosis and sex. Table S9 shows no significant effects of depression were found.

Further Analysis of Behavioral Data
We examined whether DMN intra-connectivity correlated with performance in the mentalizing and control task (making gender judgments), quantified as the percentage of incorrect responses. We employed robust regression with performance as the outcome variable, and DMN intra-connectivity as the predictor of interest. Other predictors included age, IQ, and the group of the participant (males with ASC, females with ASC, male siblings, female siblings, male controls and female controls). Thus, any correlation found demonstrates an effect beyond the effect of group. We repeated this analysis stratifying for sex. There were two clear outliers on the mentalizing task, showing error rates above 50%; we repeated the male-only analysis after discarding these outliers. Finally, we repeated the full analysis, including maximum and mean movement in the analysis.
Overall, we found a strong effect of DMN intra-connectivity and IQ for the mentalizing task, for both of the sexes (Table S10). No significant effects were found for the control task (Table S11). Motion did not correlate with performance in either task. Table S10. Results of analyses on performance on a mentalizing task. Reported are demographics and estimated effect sizes from regression modeling, multiplied by 10 2 to save space. Group effects are compared to males with ASC, to females with ASC for the female-only analysis, and to the male group for the three group-stratified analyses.