Brain network dynamics in high-functioning individuals with autism

The indirect transition frequency was significantly reduced in ASD individuals compared to controls.

“whereas the indirect transition frequency was significantly reduced in the ASD individuals compared to the controls (χ2=283.6, P<10−5 in a χ2-test; Z=16.8, Puncorrected<10−5, PBonferroni<0.05 in a post hoc residual test; Fig. 3e)”

High-functioning adults with ASD show fewer neural transitions than neurotypical controls.

“high-functioning adults with ASD show fewer neural transitions due to an unstable intermediate state”

The strength of functional segregation between DMN/Visual and FPN/SAN/ATN/SMN/Auditory modules was significantly larger in ASD individuals.

“Although significant functional segregation was seen in both the TD and ASD groups (F1,96=72.3, P<10−5 as a main effect of FC types in a two-way factorial ANOVA; Fig. 7c), its strength was significantly larger in the ASD individuals (F1,96=5.6, P=0.020 as an interaction in a two-way factorial ANOVA; t48=2.3, P=0.02 in a post hoc two-sample t-test, P=0.03 in a post hoc permutation test, Cohen’s d=0”

The pairwise maximum entropy model accurately predicted empirical data in both TD and ASD groups (accuracy≥96.4%, r126≥0.98, P<10−5).

“This model could accurately predict the empirical data in both the TD and ASD groups (accuracy≥96.4%, r126≥0.98, P<10−5 in a test of no correlation; Fig. 1f).”

Energy values of local minima E and F were relatively high compared to the four other local minima.

“compared to these four local minima, energy values of the local minima E and F were relatively high”

In random-walk simulation, ASD brains showed significantly longer duration of the major states than TD brains.

“In this random-walk simulation, the ASD brains showed significantly longer duration of the major states than TD brains (t8735=3.9, P<10−4 in a two-sample t-test, P=0.0001 in a permutation test, Cohen’s d=3.1; Fig. 3g).”

In empirical fMRI data, indirect transition frequency was significantly lower in the ASD group.

“whereas the indirect transition frequency was significantly lower in the ASD group (t48=14.0, Puncorrected<10−5, PBonferroni<0.05 in a two-sample t-test, P=0.0001 in a permutation test, Cohen’s d=5.0).”

The two minor states showed significantly less appearance frequency in the ASD group than in the TD group.

“whereas the two minor states showed significantly less appearance frequency in the ASD group (t48>10.6, Puncorrected<10−13, PBonferroni<0.05 in two-sample t-tests, P=0.0001 in permutation tests, Cohen’s d≥2.4; Fig. 2e)”

Neurotypical brain activity frequently transits between two major brain states via an intermediate state.

“Whereas neurotypical brain activity frequently transits between two major brain states via an intermediate state”

In both TD and ASD groups, intermediate-state frequency was positively associated with indirect transitions.

“Partial correlation analyses showed that in both the TD (Fig. 3i) and ASD groups (Fig. 3j) the intermediate-state frequency was positively associated with the indirect transitions (rho>0.57, P<0.0044, df≥22 in tests of no correlation)”

In the TD group, FIQ scores were specifically correlated with indirect transition frequency.

“In the TD group, FIQ scores were specifically correlated with the indirect transition frequency (r24=0.46, Puncorrected=0.014, PBonferroni<0.05 in a test of no correlation; Fig. 4c).”

The intermediate-state frequency could predict ASD diagnosis with sensitivity=84%, specificity=85%.

“The intermediate-state frequency could predict the ASD diagnosis with relatively high accuracy (sensitivity=84%, specificity=85%; Fig. 9a).”

In the ASD group, FIQ scores were correlated with duration of the major states.

“Instead, their FIQ scores were correlated with the duration of the major states (r22=0.55, Puncorrected=0.004, PBonferroni<0.05 in a test of no correlation; right panel of Fig. 4c)”

In the TD group, duration of the major states was not significantly different between individuals with higher and lower ADOS scores.

“while the duration of the major states was not significantly different between the TD individuals with higher and lower ADOS scores (P>0.56).”

TD and ASD groups had energy landscapes with similar hierarchal structures consisting of the same six locally stable brain activity patterns (local minima A-F).

“We found that the TD and ASD groups had energy landscapes with similar hierarchal structures (Fig. 2a) consisting of the same six locally stable brain activity patterns (local minima A–F; Fig. 2b).”

The appearance frequency of the intermediate state was significantly smaller in the ASD group than in controls.

“As with the minor brain states (Fig. 2e), the appearance frequency of this intermediate state was significantly smaller in the ASD group than in the controls (t48=20.3, Puncorrected<10−5 in a two-sample t-test, P=0.0002 in a permutation test, Cohen’s d=3.5; Fig. 3d).”

Functional segregation strength was inversely associated with ADOS total scores.

“Moreover, this functional segregation strength was inversely associated with ADOS total scores (r22=–0.55, P=0.004 in a test of no correlation; Fig. 6f).”

The gap between within- and across-module FCs was significantly smaller in the ASD group compared to controls.

“However, the gap between the within- and across-module FCs was significantly smaller in the ASD group compared to the controls (F1,96=272.2, P<10−5 as an interaction in the two-way factorial ANOVA; t48=13.1, P<10−5 in a post hoc two-sample t-test, P=0.0001 in a post hoc permutation test, Cohen’s d=3.0; Fig. 6c)”

In the ASD group, indirect transition frequency was negatively correlated with ADOS total scores.

“In the ASD group, the indirect transition frequency was negatively correlated with ADOS total scores (r22=–0.47, Puncorrected=0.01, PBonferroni<0.05 in a test of no correlation; Fig. 4a)”

The direct transition frequency was not different between TD and ASD groups.

“The direct transition frequency was not different between the two groups”

In both TD and ASD groups, indirect transition frequency was inversely correlated with duration of major states.

“and the indirect transition frequency was inversely correlated with the duration of the major states (rho<–0.38, P<0.042, df≥22 in tests of no correlation).”

Brain-behaviour associations in ASD are related to functional segregation between brain networks.

“such brain–behaviour associations are related to functional segregation between brain networks”

In empirical data, the repetition length of the major states was significantly longer in ASD.

“This difference was reproduced in direct counting of the repetition length of the major states in the empirical data (t48=3.6, P=0.0008 in a two-sample t-test, P=0.0013 in a permutation test, Cohen’s d=1.0; Fig. 3h).”

There was no direct transition between the two minor states for both TD and ASD groups.

“we calculated the transition frequency between the four brain states (Fig. 3b), and found no direct transition between the two minor states for both the TD and ASD groups.”

In both TD and ASD groups, within-module FCs were significantly larger than across-module FCs.

“In both the TD and ASD groups, the within-module FCs were significantly larger than the across-module FCs (F1,96=690.9, P<10−5 as a main effect of FC types in a two-way factorial analysis of variance (ANOVA) with an (FC type: within/across-module) × (Group: TD/ASD) structure; Fig. 6c).”

Combining the two indices in a multivariate pattern analysis improved diagnosis accuracy (sensitivity=89%, specificity=93%).

“We could improve the diagnosis accuracy by combining the two indices in a multivariate pattern analysis method (sensitivity=89%, specificity=93%; Fig. 9c,d).”

In both TD and ASD groups, the three brain dynamic indices were not significantly correlated with ages.

“In both the TD and ASD groups, these three brain dynamic indices were not significantly correlated with the ages of the individuals (|r|≤0.18, P≥0.37, df≥22 in tests of no correlation; Supplementary Table 1).”

In neurotypical controls, IQ is correlated with neural transition frequency.

“in contrast to the controls whose IQ is correlated with the neural transition frequency”

The indirect transition frequency did not realize accurate prediction (sensitivity=68%, specificity=74%).

“In contrast, the indirect transition frequency did not realize such accurate prediction (sensitivity=68%, specificity=74%; Fig. 9b).”

The sizes of the minor states were significantly smaller in the ASD group than in controls.

“the sizes of the minor states were significantly smaller in the ASD group than in the controls (Z>3.1, Puncorrected<0.0019, PBonferroni<0.05 in post hoc residual tests)”

The distributions of brain state size were significantly different between TD and ASD groups (χ3=21.9, P<10−4).

“These distributions of the brain state size were significantly different between the two groups (χ3=21.9, P<10−4 in a χ2-test; Fig. 2c)”

Infrequent neural transitions predict the severity of autism.

“these infrequent transitions predict the severity of autism”

The two major brain states appeared more frequently in the ASD group than in the TD group.

“The two major brain states appeared more frequently in the ASD group than in the TD group (t48>7.8, Puncorrected<10−9, PBonferroni<0.05 in two-sample t-tests, P=0.0001 in permutation tests, Cohen’s d≥2.0; Fig. 2d)”

In the ASD group, duration of the major states did not show a significant correlation with ADOS total scores.

“whereas the duration of the major states did not show a significant correlation (r=–0.09).”

In the ASD group, association between indirect transition frequency and FIQ was not found.

“In contrast, such an association between the indirect transition frequency and FIQ was not found in the ASD group (r22=–0.25, P=0.21 in a test of no correlation; left panel of Fig. 4c).”

Atypically weaker functional segregation was correlated with intermediate-state frequency and indirect transition frequency.

“This atypically weaker functional segregation was correlated with the intermediate-state frequency (r22=0.68, P=0.0002 in a test of no correlation; Fig. 6d) and the indirect transition frequency (r22=0.42, P=0.03 in a test of no correlation; Fig. 6e).”

In individuals with ASD, IQ scores are predicted by the stability of their brain dynamics.

“IQ scores of individuals with ASD are instead predicted by the stability of their brain dynamics”

In the TD group, indirect transition frequency was significantly smaller in individuals with higher ADOS scores than in those with lower ADOS scores.

“Even in the TD group, the indirect transition frequency was significantly smaller in the individuals with higher ADOS scores (ADOS total=2–4) than in those with lower ADOS scores (ADOS total=0–1; t16=2.6, P=0.019 in a two-sample t-test; Supplementary Fig. 3)”

The brain-symptom association was not specific to either social or non-social core symptoms of autism.

“This brain–symptom association was not specific to either of the social or non-social core symptoms of autism (Supplementary Fig. 4).”

The two brain dynamic indices (intermediate-state frequency and indirect transition frequency) were clearly different between TD and ASD groups.

“We focused on these two indices for brain dynamics because they were clearly different between the TD and ASD groups (Fig. 3)”

Brain activity patterns of the other networks did not show such high similarity (<58%).

“whereas those of the other networks did not show such high similarity (<58%).”

In all seven networks, the similarity between binary network activity and neural activity of constituent ROIs was significantly high across participants (mean similarity ≥69%, P<10−4).

“In all the seven networks, the similarity between the binary network activity and neural activity of its constituent ROIs was significantly high across participant (mean similarity ≥69%, P<10−4 in a binominal test; Supplementary Fig. 14a).”

In both TD and ASD groups, functional segregation strength was not significantly correlated with global connectivity levels (|r|≤0.16, P≥0.42).

“Note that, in both TD and ASD groups, the functional segregation strength was not significantly correlated with global connectivity levels (|r|≤0.16, P≥0.42, Supplementary Table 4)”

Dividing the ATN into cingulo-opercular and dorsal/ventral attention systems yielded qualitatively the same energy landscapes with the same six local minima.

“even after we divided the ATN into cingulo-opercular network and the other two systems (dorsal/ventral attention networks), we observed qualitatively the same energy landscapes with the same six local minima (Supplementary Fig. 13c,d).”

High similarity was also observed at the voxel level across participants (mean similarity ≥66%, P<10−3).

“We can see such high similarity even at a voxel level across participant (mean similarity≥66%, P<10−3; Supplementary Fig. 14b).”

In binary form, brain activity patterns of the three attention networks were significantly similar to each other (≥71% in TD, ≥73% in ASD, P≤10−5).

“In fact, in the binary form, brain activity patterns of these three networks were significantly similar to each other (≥71% in TD, ≥73% in ASD, P≤10−5 in one-sample t-tests across participants; Supplementary Fig. 13a,b)”

Brains of ASD individuals have weaker functional segregation between the two brain modules.

“which suggests that the brains of ASD individuals have weaker functional segregation between the two brain modules.”

The findings were robust using two independent data sets collected in Indiana University and ETH Zürich.

“We confirmed the robustness of the findings using two independent data sets collected in Indiana University (Supplementary Table 2) and ETH Zürich (Supplementary Table 3).”

Local minima E and F were not so dominant and stable as the other four.

“which suggests that these two local minima were not so dominant and stable as the other four”

The current observations were robust across different definitions of brain networks.

“We also tested whether the current observations were robust across different definitions of brain networks.”

The fewer neural transitions in ASD are due to an unstable intermediate state.

“high-functioning adults with ASD show fewer neural transitions due to an unstable intermediate state”

The accurate model fitting and hierarchal structures of energy landscapes were preserved when the threshold for binarization of brain activity was changed.

“Such an accurate model fitting and the hierarchal structures of energy landscapes were preserved when we changed the threshold for the binarization of brain activity (Supplementary Fig. 2).”

The network activity patterns sufficiently represented the activity of cortical brain regions that the networks cover.

“These network activity patterns sufficiently represented the activity of cortical brain regions that the networks cover.”

These results justify merging the three networks into one system.

“These results are considered to justify our merging the three networks into one system.”

Another limitation concerns potential heterogeneity in the ASD group.

“Another limitation of our work concerns potential heterogeneity in the ASD group57.”

• cites [57] Jack, A. & Pelphrey, K. A. Annual research review: understudied populations within the autism spectrum—current trends and future directions in neuroimaging research. J. Child Psychol. Psychiatry 58, 411–435 (2017). sourceNot checked yet

  Not checked in the free run.

A previous study showed that FC is significantly different from Jij in terms of similarity to anatomical connections.

“In fact, our previous study showed that FC is significantly different from Jij in terms of similarity to anatomical connections32.”

• cites [32] Lord, C. et al. Autism diagnostic observation schedule: a standardized observation of communicative and social behavior. J. Autism Dev. Disord. 19, 185–212 (1989). sourceNot checked yet

  Not checked in the free run.

The dorsal/ventral attention networks and cingulo-opercular network are responsible for similar attention-related cognitive activity.

“these three networks are considered to be responsible for the similar attention-related cognitive activity30.”

• cites [30] Power, J. D. et al. Functional network organization of the human brain. Neuron 72, 665–678 (2011). sourceNot checked yet

  Not checked in the free run.

This study provides empirical evidence that autism can be characterized by atypical large-scale brain dynamics.

“This study provides empirical evidence for the concept that autism can be characterized by atypical large-scale brain dynamics11.”

• cites [11] Uhlhaas, P. J. & Singer, W. Neuronal dynamics and neuropsychiatric disorders: toward a translational paradigm for dysfunctional large-scale networks. Neuron 75, 963–980 (2012). sourceNot checked yet

  Not checked in the free run.

Observations about functional segregation cannot be explained simply by global network-wide connectivity.

“which suggests that observations about functional segregation cannot be explained simply by global network-wide connectivity.”

The current data size is not enough to perform energy-landscape analysis with nine factors.

“the current data size is not enough to perform energy-landscape analysis with nine factors”

The current data set size was too small to accurately perform energy-landscape analysis for 17 networks.

“This previous study also proposed a 17-network brain parcellation method, but we did not choose it because the size of the current data set was too small to accurately perform energy-landscape analysis for such 17 networks.”

The ASD participants were diagnosed by a clinical autism expert in accordance with ADOS and DSM-IV-TR.

“The ASD participants were diagnosed by a clinical autism expert in accordance with ADOS32 and Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (DSM-IV-TR).”

• cites [32] Lord, C. et al. Autism diagnostic observation schedule: a standardized observation of communicative and social behavior. J. Autism Dev. Disord. 19, 185–212 (1989). sourceNot checked yet

  Not checked in the free run.

Differences in brain dynamics between ASD and TD groups were confirmed with non-parametric permutation tests.

“Differences in brain dynamics between the ASD and TD groups were evaluated initially in parametric tests (here, t-tests) but also confirmed with non-parametric approaches using permutation tests with 104 random permutations.”