Blood Oxygen Level-Dependent Lag Patterns Differ Between Rest and Task Conditions,but Are Largely Typical in Autism

Abstract

Background/Introduction:

Autism spectrum disorder (ASD) is characterized by atypical functional connectivity (FC) within and between distributed brain networks. However, FC findings have often been inconsistent, possibly due to a focus on static FC rather than brain dynamics. Lagged connectivity analyses aim at evaluating temporal latency, and presumably neural propagation, between regions. This approach may, therefore, reveal a more detailed picture of network organization in ASD than traditional FC methods.

Methods:

The current study evaluated whole-brain lag patterns in adolescents with ASD (n = 28) and their typically developing peers (n = 22). Functional magnetic resonance imaging data were collected during rest and during a lexico-semantic decision task. Optimal lag was calculated for each pair of regions of interest by using cross-covariance, and mean latency projections were calculated for each region.

Results:

Latency projections did not regionally differ between groups, with the same regions emerging among the “earliest” and “latest.” Although many of the longest absolute latencies were preserved across resting-state and task conditions, lag patterns overall were affected by condition, as many regions shifted toward zero-lag during task performance. Lag structure was also strongly associated with literature-derived estimates of arterial transit time.

Discussion:

Results suggest that lag patterns are broadly typical in ASD but undergo changes during task performance. Moreover, lag patterns appear to reflect a combination of neural and vascular sources, which should be carefully considered when interpreting lagged FC.

Impact statement

Altered brain dynamics have been proposed in autism spectrum disorder (ASD). Lagged functional connectivity analysis uses cross-correlation between functional magnetic resonance imaging (fMRI) time series to determine regional latency. Few studies have examined blood oxygen level-dependent (BOLD) lag in ASD, and findings have been inconsistent. Using multi-echo fMRI data with improved artifact detection and removal, we find differences in lag structure between task and rest states, but not between adolescents with ASD and typically developing peers. Additional analyses exploring links with arterial transit time, however, highlight the impact of vascular organization on BOLD lag patterns and its potential to confound measures of neural dynamics.

Introduction

Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by deficits in social communication and restricted interests, repetitive behaviors, or unusual sensory responses (American Psychiatric Association, 2013). Recent prevalence estimates approach 2% (Maenner et al., 2020). However, neurobiological underpinnings of ASD remain incompletely understood.

Neuroimaging research in recent decades has led to a consensus that ASD is characterized by atypical communication among distributed brain networks (Di Martino et al., 2014), rather than focal abnormalities. However, ASD is highly heterogeneous (Geschwind and State, 2015; Lombardo et al., 2019; Olsson et al., 2015), and functional connectivity magnetic resonance imaging (fcMRI) studies have often reported discrepant or mixed findings of overconnectivity and underconnectivity involving various functional networks (Hull et al., 2016; Rane et al., 2015). Methodological variables likely explain some of these inconsistencies (Müller et al., 2011; Nair et al., 2014), but they cannot fully account for the wide range of findings.

Most fcMRI research has focused exclusively on static functional connectivity (FC), based on correlations between regional time series across an entire scan. Blood oxygen level-dependent (BOLD) activity in regions belonging to common functional networks tends to be highly correlated (Biswal et al., 1995; Fox and Raichle, 2007). However, static FC is limited by its inability to capture continuous changes in BOLD correlations that occur over seconds or minutes. Alternative, dynamic FC approaches, such as sliding window and clustering analyses, have demonstrated that the configuration of BOLD networks and the “strength” of individual connections vary on a timescale of seconds in typical development (Allen et al., 2014; Calhoun et al., 2014; Nomi et al., 2016). However, even the majority of dynamic FC studies assume that BOLD activity between related regions is synchronous (i.e., at zero-lag). Recent findings suggest that neural propagation (i.e., temporal latency between activity in a source region and a destination region) may be observable at a timescale of seconds via BOLD activity. For example, resting-state lag analysis (RSLA), first described by Mitra et al. (2014), determines the optimal latencies between pairs of time series by calculating lagged cross-covariance at several timepoints in either direction of zero-lag. Covariance estimates are then plotted at each lag, and parabolic interpolation is used to estimate timing of the peak covariance; this yields a pairwise lag estimate between each pair of brain regions examined. Finally, “latency projections” are calculated for each brain region as the mean lag with respect to all other regions. Pairwise lags describe the temporal relationship between any two regions, whereas averaged latency projections are useful in identifying brain areas that are consistently “early” or “late” and may therefore serve as general sources or sinks of neural activity.

Using a large sample of typically developing (TD) individuals, Mitra et al. (2014) demonstrated highly reproducible patterns of “lag structure” across seven independent subsamples, with consistently “early” and “late” nodes emerging in several functional networks. In other words, latency projections were negative for “early” regions and positive for “late” regions, with respect to the rest of the brain. In a small group of extensively sampled participants, this method revealed stable lag structure within individuals, with a number of early regions serving multiple functional networks showing similar latency across all subjects (Raut et al., 2019a).

Mitra et al. (2014) found that RSLA estimates varied with eye status (open vs. closed), after a button press task, and at different times of day (morning vs. evening); this was interpreted as support for a neural basis of the observed latency structure. Further, calcium/hemoglobin imaging and laminar recordings in a mouse model suggest that spontaneous infra-slow activity, most often associated with BOLD fluctuations (He et al., 2008; Pan et al., 2013), follows predictable propagation patterns that are distinct from those seen at higher frequencies (Mitra et al., 2018). However, primarily vascular, non-neuronal explanations for BOLD lag have also been proposed (Aso et al., 2017).

A limited number of studies have explored resting-state BOLD latency in ASD using various approaches. Using RSLA, Mitra et al. (2017) reported atypical lag structure in a small ASD sample. Specifically, they found that frontopolar cortex and putamen were significantly “earlier” in an ASD than a TD group, whereas occipital cortex was “later.” Further, in the ASD group, earlier frontopolar activity was associated with increased attention problems, and earlier putamen activity was associated with increased restricted and repetitive behaviors.

Another study compared resting-state FC between adults with and without ASD at different lags; broadly increased FC in ASD was discovered at greater lags (in both directions), despite an absence of group differences at zero-lag (King et al., 2018). This was accompanied by increased “sustained connectivity,” or FC duration, in the ASD group. Finally, a recent resting-state study using ultrafast magnetic resonance encephalography (MREG; repetition time [TR] = 100 msec) reported atypical latency patterns in ASD across multiple networks, with the ASD group generally showing shorter lags than the TD group (Raatikainen et al., 2020). This study used dynamic lag analysis, which assesses lag by comparing the timing of resting-state BOLD peaks across high-resolution time series.

Although these first lag studies show promise, there are several limitations to consider. First, the measurement of lagged cross-correlations is limited by the low temporal resolution of the BOLD signal. Second, latency analyses are particularly sensitive to data length (i.e., scan duration), motion artefact, physiological artefact, and censored timepoints, all of which can affect sampling error and reliability of estimates (Chen et al., 2020; Raut et al., 2019b; Smith et al., 2011); similar concerns have been raised about the effects of motion and physiological artefact on dynamic FC (Laumann et al., 2017; Nalci et al., 2019). Third, hemodynamic and neural contributions to the BOLD signal cannot be entirely separated. In other words, it is unclear as to what degree BOLD latency between two regions reflects directional neural communication, versus regional differences in vascular supply.

Therefore, group differences in vascular measures, such as arterial supply and cerebral blood flow (CBF), may confound BOLD lag estimates (Tong et al., 2019). Mitra et al. (2014) reported a lateness bias for regions near large venous sinuses in TD participants but did not find any relationship between latency and perfusion. These concerns merit particular attention in ASD studies, as localized differences in the hemodynamic response (Yan et al., 2018), as well as altered CBF (George et al., 1992; Jann et al., 2015; Starkstein et al., 2000; Yerys et al., 2018), have been documented in this population. To this end, the current study compared whole-brain lag maps with a publicly available map of vascular supply territories (including measures of cerebral perfusion and arterial transit time [ATT]) to further examine the associations between BOLD latency and blood supply.

The current study builds on the few previous reports of atypical lag structure in ASD by examining regional latencies of multi-echo simultaneous multi-slice (MESMS) BOLD data. These data offer higher temporal resolution than standard functional magnetic resonance imaging (fMRI) sequences (TR ≤1.25). Although one previous study has examined lag in ASD at even higher temporal resolution (Raatikainen et al., 2020), the MESMS protocol acquires data by using three distinct echo times (TE). This allows for multi-echo independent component analysis (ME-ICA) denoising, in which TE dependencies of ICA components are used to distinguish neural components from noise components. This method has shown superior artefact removal compared with standard denoising procedures (Kundu et al., 2012, 2013).

The previous literature has explored only resting-state lag structure in ASD, which relies on assumptions about the source and timing of neuronal events. In the absence of an explicit task, the degree to which synchronous BOLD activity reflects intrinsic connectivity versus state-specific processing remains ambiguous. Therefore, the current study assessed latency during rest as well as during a task, providing controlled data acquisition during which all participants are exposed to the same stimuli.

A complex lexicosemantic decision task was selected, due to its recruitment of numerous brain regions and networks, including primary visual cortex, motor regions, language networks, and executive regions involved in decision making, inhibition, and initiation. Therefore, this task was considered suitable for comparing BOLD latency differences between a relatively uncontrolled state (with spontaneous signal fluctuations) and a relatively controlled state (with task-driven fluctuations). Comparisons between resting-state and task conditions may shed light on the degree to which BOLD latency structure reflects a history of directional co-activation versus task-specific cognitive demands.

The current study aimed to clarify (1) differences in whole-brain lag structure in ASD compared with typical development, including between occipital, frontopolar, and subcortical areas previously implicated in ASD; (2) differences in lag structure in the presence versus absence of an explicit task; and (3) inter-individual variability of whole-brain lag structure. Secondarily, this study examined the associations between latency projections and literature-derived measures of cerebral perfusion and arterial blood supply, to assess the degree to which these variables may confound neural sources of lag in either group.

Materials and Methods

Participants

Seventy adolescents and young adults aged 12–21 years were enrolled and scanned as part of this study. Fourteen participants with ASD and six TD participants were excluded from analyses due to unacceptable data quality for one or more of the three functional scans. Criteria for exclusion were excessive motion (root mean squared displacement [RMSD] >0.15 mm), <60% accuracy on behavioral tasks, missing scans, preprocessing failure, or incidental magnetic resonance imaging (MRI) findings. This left a final sample of 50 participants (28 ASD, 22 TD) with high-quality data.

All ASD diagnoses were confirmed using the Autism Diagnostic Observation Schedule, Second Edition (ADOS-2) (Gotham et al., 2007), the Autism Diagnostic Interview–Revised (ADI-R) (Lord et al., 1994), and expert clinical judgment based on DSM-5 criteria (American Psychiatric Association, 2013). The Wechsler Abbreviated Scales of Intelligence-Second Edition (WASI-II) (Wechsler, 2011) and Social Responsiveness Scale, Second Edition (SRS-2) (Constantino and Gruber, 2012) parent report form were administered to individuals in both groups. Groups did not significantly differ with respect to age, handedness, gender, verbal IQ, nonverbal IQ, or in-scanner head motion (Table 1).

Table 1.

Sample Characteristics

	ASD (n = 28)	TD (n = 22)	Statistic	p
Sex	20 Male	17 Male	χ²(1) = 0.22	0.640
Handedness	27 Right	20 Right	χ²(1) = 0.67	0.415
Age	15.9 (2.2) [12.9–20.1]	15.3 (1.9) [12.4– 21.3]	t(48) = 1.01	0.319
RMSD Rest	0.07 (0.03) [0.01–0.13]	0.06 (0.02) [0.02–0.11]	t(48) = 1.07	0.290
RMSD Task	0.07 (0.02) [0.03–0.13]	0.06 (0.03) [0.03–0.15]	t(48) = 1.67	0.102
VIQ	106 (17) [68–134]	111 (13) [85–135]	t(48) = −1.27	0.210
NVIQ	108 (22) [54–156]	110 (12) [80–128]	t(48) = −0.33	0.746
Task accuracy (%)	88.0 (8.3) [70.7–99.0]	94.7 (3.3) [85.0–98.7]	t(48) = −3.63	<0.001
ADOS-2 SA	8.8 (3.0) [3–14]	—	—	—
ADOS-2 RRB	2.4 (1.9) [0–9]	—	—	—
ADOS-2 Total	10.8 (3.4) [6–20]	—	—	—

Relevant characteristics are reported for both groups as: mean (SD), [min–max]. ASD and TD groups only differed with respect to task accuracy, which was lower in the ASD group.

ADOS-2, Autism Diagnostic Observation Schedule, Second Edition; ASD, autism spectrum disorder; NVIQ, nonverbal IQ; RMSD, root mean squared displacement; RRB, Restricted and Repetitive Behaviors; SA, social affect; SD, standard deviation; TD, typically developing; VIQ, verbal IQ.

The RMSD was used as an estimate of head motion across all time points for both conditions (task and resting state). All included participants had low estimates of motion in both conditions (RMSD ≤0.15 mm). A mixed analysis of variance (ANOVA) found no effect of condition, diagnosis, or the condition × diagnosis interaction on RMSD (all p-values >0.12).

All TD participants were unmedicated. Eleven participants with ASD reported taking psychotropic medications, and co-occurring psychiatric diagnoses were documented in 10 participants with ASD. Medication status and co-occurring conditions were not documented for one individual in the ASD group. Details regarding medication usage and co-occurring diagnoses are reported in Supplementary Table S1.

Written informed consent was obtained from all participants and/or their caregivers, and written assent was obtained for participants under 18 years of age. All procedures were approved by the San Diego State University and University of California, San Diego Institutional Review Boards.

Data acquisition

Imaging data were acquired on a GE 3T MR750 scanner with a 32-channel head coil at the Center for Functional MRI (University of California, San Diego, CA). High-resolution structural images were acquired with a standard fast spoiled gradient echo T1-weighted sequence (TR: 8.136 msec, TE: 3.172 msec, flip angle: 8°; field of view (FOV) 256 mm; matrix: 256 × 192 1 mm³ resolution; 172 slices). Functional T2*-weighted images were acquired by using an accelerated MESMS echo-planar imaging (EPI) sequence (TR: 1250 msec; TE: 13.2, 30.3, and 47.4 msec; flip angle: 60°; 3 mm³ resolution; FOV: 216 mm; matrix: 72 × 72 with in-plane acceleration factor R: 2; in-plane resolution: 3 × 3; multiband acceleration factor R: 3; 54 slices).

Two single-echo field of view EPI sequences with the same parameters (TE: 30.3) and opposite phase-encoding directions were also acquired for later distortion correction (see the Data Processing section). The MESMS approach has been shown to improve the detection of functional networks and to substantially reduce motion and physiological artifact (see the Data Processing section for preprocessing details) (Kundu et al., 2012, 2013).

One 6-min 26-sec resting-state scan was obtained (309 volumes), followed by two task scans lasting 7 min each (340 volumes). The functional scan protocol for nine TD and two ASD participants differed slightly (TR = 1100 msec, 45 slices, 340 resting volumes, 6:14 resting scan duration, 386 task volumes, 7:00 duration for each of the two task scans). The mean temporal signal-to-noise ratio did not significantly differ between these participants and a sample of participants scanned with the standard protocol who were matched for motion and demographic variables [resting state: t(20) = −0.91, p = 0.37, task run 1: t(20) = −0.61, p = 0.55, task run 2: t(20) = 0.32, p = 0.75]. The first nine volumes of each scan were discarded to account for T1-equilibration effects.

During resting-state scans, participants were instructed to fixate on a cross projected onto the middle of a screen, viewed through a mirror in the bore, and to “Let your mind wander, relax, but please stay as still as you can. Do not fall asleep.” Compliance with instructions to remain still and awake was monitored via video recording.

During task scans, participants performed a lexical decision task. In this task, participants were asked to distinguish between animal words (e.g., “cat”), standard (nonanimal) words (e.g., “chair”), and pseudowords (e.g., “blont”). Null trials consisting of a fixation string (“xxxxxx”) were also included. Participants were asked to respond to standard words (90 trials) and animal words (30 trials) by using their left index and middle fingers, respectively, and to withhold responses to pseudowords (30 trials).

Stimuli were presented for 500 msec, followed by a 1500 msec fixation string to allow for response (total of 2s per trial). Conditions were matched for number of letters and syllables (p > 0.1). Animal and standard words were also matched for word frequency (Brysbaert and New, 2009; van Heuven et al., 2014) and age of acquisition (Kuperman et al., 2012), using publicly available corpora (p > 0.1).

Data processing

All structural and functional data were visually inspected for signal dropout, excessive motion artifact, and alignment errors at multiple preprocessing stages to ensure acceptable data quality. Functional images were processed using Analysis of Functional NeuroImages (AFNI) software (v17.2.07) (Cox, 1996) and FSL (v5.0) (Smith et al., 2004). Two spin-echo EPI acquisitions with opposite phase encoding directions were used to minimize susceptibility-induced distortions via FSL's TOPUP (Smith et al., 2004). Each functional volume was registered to the middle time point of the scan to adjust for motion via rigid-body realignment as implemented in AFNI.

Functional images were co-registered to the anatomical scan via FSL's linear image registration tool (FLIRT; Jenkison and Smith, 2001) and standardized to the atlas space of the Montreal Neurological Institute (MNI) template by using FSL's nonlinear registration tool (FNIRT). AFNI's 3dBlurToFWHM was used to smooth functional images to a Gaussian full width at half-maximum (FWHM) of 6 mm. Functional MRI time series were either bandpass filtered (0.008 < f < 0.08; resting-state scan) or highpass filtered (0.01 < f; task scans).

Denoising was conducted using ME-ICA to remove artefactual components (Kundu et al., 2013). Data from the three echoes were optimally combined to produce a single time series. Multi-echo weighted optimization and ME-ICA are described in detail by Olafsson et al. (2015; see appendix A) and are implemented by meica.py, which is publicly available at http://afni.nimh.nih.gov/afni No censoring was applied to preserve the continuity of time series for lag analysis.

Analyses

Regions of interest

An exploratory set of regions of interest (ROIs) covering all cortical and subcortical regions was created by combining 100 regions from a widely used functional cortical parcellation (Schaefer et al., 2018) with 14 regions from the Harvard-Oxford subcortical atlas (Bohland et al., 2009). Cortical regions were assigned to one of seven functional networks, as described by Yeo et al. (2011): visual, somatomotor, dorsal attention, ventral attention, limbic, frontoparietal control, and default-mode.

FC and lag analysis

For each participant, time series were extracted from all ROIs. Pearson correlations (transformed to Fisher's z) were calculated between time series for each ROI-ROI pairing, yielding a symmetrical FC matrix. FC at each ROI-ROI pair was compared between ASD and TD groups using independent-samples t-tests. The distribution of group differences in FC was evaluated using a one-sample t-test of all group difference t-scores, to examine whether, on average, group differences across all ROI pairs differed significantly from zero in either direction.

Significant findings were followed up using censored data to examine the potential effects of motion. Censoring was performed by removing volumes with >0.5 mm framewise displacement, as well as the three following volumes. If fewer than 18 usable timepoints fell between two censored timepoints, then all volumes in between were also censored.

Pairwise lag estimates and regional mean latency projections were calculated using the method developed and described in detail by Mitra et al. (2014), using in-house MATLAB code. Cross-covariance between each pair of timeseries was calculated at zero-lag, as well as at −4 to 4 TR lag (using MATLAB's xcov function). Cross-covariance was plotted at each lag by using piecewise cubic interpolation (MATLAB's pchip), and a local minimum or maximum was identified if possible. Parabolic interpolation was then used to estimate the optimal lag between two time series.

If a local minimum or maximum could not be identified within three TRs of zero-lag in either direction, the peak lag was considered unlikely to result from true neural signal (Raut et al., 2019a) and was discarded. The number of discarded peak lags was considerably larger during resting-state scans (ASD mean = 8.3%, TD mean = 8.8%) than task performance (ASD mean = 0.45%, TD mean = 0.39%), but it did not significantly differ between groups in either rest or task conditions (p = 0.81 and p = 0.75, respectively).

The procedure described earlier resulted in an anti-symmetric matrix of all pairwise lag values for each participant. “Latency projections” for each ROI, as defined by Mitra et al. (2014), were then calculated by averaging across all rows of the matrix, resulting in one mean lag value per column. In other words, latency projections represent the mean lag for a given ROI across all of its pairings with other ROIs.

To compare lag structure across conditions, average resting-state and task-related latency projections (across 114 ROIs) were correlated separately for ASD and TD groups. To compare lag structure between groups, group-averaged latency projections were correlated separately for resting-state and task conditions. These correlational analyses were also conducted for pairwise lag estimates.

Effects of diagnosis, condition, and their interaction on latency projections were evaluated using a mixed-design ANOVA for each ROI (using simple_mixed_anova in MATLAB) (Caplette, 2020), with diagnosis as a between-subjects variable and task condition as a within-subjects variable. Resulting p-values were adjusted to control for false discovery rate using the Benjamini-Hochberg linear step-up procedure (Benjamini and Hochberg, 1995).

Similarity analysis

Recent research suggests that the organization of functional networks may be more idiosyncratic in ASD (i.e., characterized by increased within-group variability) than in typical development (Hahamy et al., 2015; Nunes et al., 2019). Therefore, similarity analyses were conducted to examine the inter-individual variability of FC and latency projections for both resting-state and task conditions. For each group, all participants' pairwise FC estimates were correlated with all other participants' FC estimates.

The resulting correlation matrix was averaged across rows, yielding one similarity score per participant, which was then converted to a Fisher's z-score. These scores represent the average within-group similarity between each individual's whole-brain FC pattern and all other individuals' whole-brain FC patterns. A mixed ANOVA was conducted (as in lag analyses) to determine the effects of diagnosis, condition, and their interaction on within-group FC similarity. This similarity analysis procedure was repeated for latency projections.

All significant ANOVA effects were followed up with nonparametric permutation testing due to the nonindependence of similarity scores. In this procedure, p-values were calculated by comparing t-scores with a null T_max distribution (Chen et al., 2013) consisting of t-scores derived from 1000 random permutations of the grouping variable of interest (i.e., ASD or TD; task or rest).

Hemodynamic comparisons

To determine the possible vascular contributions to lag structure, secondary analyses were conducted comparing whole-brain latency projections with (1) CBF maps obtained from the Pediatric Template of Brain Perfusion (Avants et al., 2015), and (2) arterial territory maps and associated median ATT reported by Mutsaerts et al. (2015). For the first analysis, CBF was extracted from all ROIs and correlated with mean latency projections for each group. Correlations between CBF and latency were then calculated at the individual level, transformed to Fisher's z, and compared between groups using independent-samples t-tests.

For the second analysis, each cortical ROI was assigned to one of nine arterial flow territories from a publicly available map (Mutsaerts, 2020), based on descriptions by Tatu et al. (1998). Assignments were made according to spatial overlap between each ROI and flow territory. For every ROI, each voxel was assigned a number (1–9) corresponding to the flow territory that encompassed that voxel. This resulted in a vector of voxel locations for each ROI. The modal value of each vector then determined the territory assignment for the corresponding ROI (Supplementary Fig. S1). For 11 subcortical ROIs, the majority of voxels fell outside all 9 flow territories. Therefore, these ROIs were excluded from hemodynamic analyses. Mean latency was calculated for each territory and correlated with median ATT. The similarity of arterial flow patterns between groups was also examined by correlating ASD and TD mean latencies for all territories.

Results

FC analyses

A one-sample t-test of all group comparison t-scores (all ROI pairings in the matrix), examining the overall distribution of group differences in FC, revealed a broad shift toward overconnectivity in the ASD relative to the TD group in both resting-state [t(6440) = 101.42, p < 0.001] and task conditions [t(6440) = 158.64, p < 0.001]. On pairwise tests, resting-state overconnectivity (ASD > TD) between one pair of left visual ROIs [Vis 3-Vis 6, t(48) = 4.97, p < 0.001] survived correction for false-discovery rate (FDR). No other findings remained significant after FDR correction, given the large number of comparisons.

Moreover, there were no substantial changes in either resting-state [t(6440) = 95.34, p < 0.001] or task-based [t(6440) = 145.47, p < 0.001] findings after censoring volumes with excessive motion. Finally, there were no significant differences in resting-state or task-based FC between medicated and unmedicated participants with ASD, nor were there significant differences between participants with and without comorbid psychiatric diagnoses (all p-values >0.05).

Lag analyses

Latency projections revealed largely bilateral lag patterns across the brain in both groups (Fig. 1A, B), with estimates ranging from approximately −0.8 to +1.0 sec. Similar lag patterns were observed between resting-state and task conditions with respect to both latency projections (Rest vs. Task, ASD: r = 0.87, p < 0.001; TD: r = 0.78, p < 0.001) and pairwise lag matrices (Rest vs. Task, ASD: r = 0.78, p < 0.001; TD: r = 0.70, p < 0.001). ASD and TD participants also showed highly similar patterns of group-averaged latency projections (ASD vs. TD, resting-state: r = 0.90, p < 0.001; task: r = 0.95, p < 0.001), as well as pairwise lag matrices (ASD vs. TD, resting-state: r = 0.78 p < 0.001; task: r = 0.86, p < 0.001).

FIG. 1.

Pairwise lag matrices and latency projections. Pairwise lag matrices (left) and mean latency projections (right) are shown by group for the resting state (A) and task (B) conditions; colors represent lag in seconds. Differences between task and resting conditions for all participants combined are also presented (C). Only regions showing a significant main effect of condition on latency are shown on the right (FDR-corrected p < 0.05); colors represent mean difference in seconds. Pairwise lag matrices are organized by functional network. Each region's relationship to other regions is interpreted vertically, with “earlier” networks appearing as blue columns and “later” networks appearing as red columns. DA, dorsal attention; DMN, default mode network; FDR, false-discovery rate; FP, frontoparietal; Lim, limbic; SC, subcortical; SM, somatomotor; VA, ventral attention; Vis, Visual.

Several regions consistently emerged among the earliest and latest in both groups. In the resting-state, “early” regions included the putamen, inferior frontal gyrus, insula, lateral dorsal attention regions, and lateral somatomotor regions; whereas “late” regions included visual areas, thalamus, caudate, cingulate, and medial parietal cortex. During task performance, a similar pattern emerged with “early” activity in inferior frontal gyrus and insula, and “late” activity in visual and medial parietal regions; mean latencies were close to zero for other structures (Fig. 1B). In general, anterior, lateral structures tended to be early relative to posterior, medial structures, and this pattern affected more ROIs in the resting-state condition than the task condition.

Mixed ANOVAs for each ROI found no main effects of diagnosis on latency, and no interactions between diagnosis and task condition (all corrected p-values >0.05). However, 37 ROIs spanning most networks showed main effects of condition (all corrected p-values <0.05) (Fig. 1C and Supplementary Table S2), which consistently reflected reduced absolute latency in the task compared with the rest condition. In other words, many regions that were relatively early or late during the resting state showed latencies closer to zero during task performance.

A post hoc two-sample Kolmogorov-Smirnov test was conducted (using MATLAB's kstest2) to compare the distributions of mean pairwise lag values (6441 ROI pairs) in each condition. This test confirmed that group-averaged rest and task lag distributions differed from one another (D* = 0.19, p < 0.001), such that the Gaussian distribution of resting latencies was flatter and more spread out than the task-related distribution, which showed a large number of ROI-ROI latencies clustered closer to zero (Supplementary Fig. S2).

To ensure that lag differences between the task and resting-state conditions were not associated with acquisition length (i.e., one resting run vs. two concatenated task runs), this analysis was repeated using individual task runs. Results were similar to those reported above when run 1 only (D* = 0.19, p < 0.001) and run 2 only (D* = 0.17, p < 0.001) were used (Supplementary Fig. S2).

To account for the greater number of pairwise lags that could not be calculated at rest, the analysis was repeated after eliminating ROI-ROI pairs that were unusable in resting-state data from the task data. This procedure produced similar results to those described earlier (D* = 0.19, p < 0.001). Finally, independent-samples t-tests revealed no significant differences in resting-state or task-based latency between medicated and unmedicated participants with ASD, nor were there significant differences between participants with and without comorbid psychiatric diagnoses (all p-values >0.05).

Similarity analyses

A mixed ANOVA showed that within-group FC similarity was lower in the ASD relative to the TD group [ASD mean z = 0.50, TD mean z = 0.59, F(1, 48) = 65.28, p < 0.001], and that FC similarity was lower during the resting state than task performance [resting mean z = 0.46, task mean z = 0.62, F(1, 48) = 329.95, p < 0.001]. There was no interaction between diagnosis and condition. Post hoc permutation tests corroborated both the main effects of diagnosis [t(98) = −4.81, p < 0.001] and condition [t(49) = 18.51, p < 0.001] on FC similarity.

Similarity analyses of latency projections revealed reduced within-group similarity during the resting state compared with task performance [rest mean z = 0.28, task mean z = 0.44, F(1, 48) = 80.02, p < 0.001]. There was no main effect of diagnosis, nor was there a diagnosis × condition interaction. Post hoc permutation testing corroborated the main effect of condition [t(49) = 8.64, p < 0.001] on lag similarity. To account for differences in the amount of data (two task runs vs. one resting state run), this analysis was repeated using individual task runs. A main effect of condition (resting < task) was similarly observed for both run 1 [F(1, 48) = 12.52, p < 0.001] and run 2 [F(1, 48) = 12.27, p = 0.001], and these effects were corroborated using permutation testing [Run 1: t(49) = 5.31, p < 0.001; Run 2 t(49) = 5.32, p < 0.001].

Hemodynamic comparisons

CBF estimates from the Pediatric Template of Brain Perfusion (Avants et al., 2015) were not correlated with mean latency projections in either group, in either the resting-state (ASD: r = −0.12, p = 0.20; TD: r = −0.06, p = 0.56) or task conditions (ASD: r = −0.06, p = 0.53, TD: r = −0.10, p = 0.27). Correlations between CBF and latency did not significantly differ between groups for either the resting-state [t(48) = −0.91, p = 0.37] or task conditions [t(48) = 1.49, p = 0.14].

The mean latency of nine arterial flow territories was associated with the median ATT reported by Mutsaerts et al. (2015), with stronger relationships emerging during task performance (ASD: r = 0.79, p = 0.01; TD: r = 0.73, p = 0.03) than resting-state scans (ASD: r = 0.60, p = 0.09; TD: r = 0.46, p = 0.21) (Fig. 2A). Further, the ASD and TD groups showed extremely similar mean latency patterns across all flow territories in both resting-state (r = 0.94, p < 0.001) and task conditions (r = 0.99, p < 0.001) (Fig. 2B). Overall, regions supplied by the middle cerebral artery (MCA) were earlier, on average, than regions supplied by the anterior cerebral artery (ACA) and posterior cerebral artery (PCA). Regions supplied by the distal branches of the ACA and PCA were consistently latest (Fig. 2C).

FIG. 2.

Associations between lag and arterial supply. (A) Group-averaged correlations between mean latency and median ATT (from Mutsaerts et al., 2015) for each of the nine arterial flow territories are shown for the resting (top) and task (bottom) conditions. (B) Correlations between ASD group and TD group mean latency for each arterial territory are shown for each condition. (C) Mean latencies for the proximal, middle, and distal branches of the ACA, MCA, and PCA are shown for each group. ACA, anterior cerebral artery; ASD, autism spectrum disorder; ATT, arterial transit time; MCA, middle cerebral artery; PCA, posterior cerebral artery; TD, typically developing.

Discussion

Using high-resolution multi-echo fMRI data, this study found broadly similar whole-brain lag patterns in adolescents with ASD compared with a TD cohort. Latency projections differed in both groups as a function of task condition, with many regions shifting toward zero-lag during task performance. In other words, the Gaussian distribution of lags observed at rest showed reduced variance during task performance as more lags clustered around zero (Supplementary Fig. S2). Although (zero-lag) FC was more variable in the ASD group than the TD group, there were no group differences in the variability of lag patterns. However, both FC and latency projections showed more inter-individual variability during task performance than at rest. Finally, latency projections in both groups were strongly related to arterial blood supply and transit time.

Similar lag structure observed in both groups

With respect to resting-state lag structure, similar regions emerged among the “earliest” and “latest” in both ASD and TD groups. A general pattern emerged of earlier activity in lateral, anterior regions, compared with posterior, medial regions. In line with previous studies, lag structure was largely bilateral. The earliest regions during the resting state included inferior frontal gyrus, insula, lateral somatomotor and dorsal attention areas, putamen, and pallidum. Among the latest regions were visual cortex, cingulate, thalamus, caudate, and medial parietal areas.

Early cortical regions in the current study largely correspond to those identified by Raut et al. (2019a) using resting-state fMRI from 11 extensively sampled healthy adults; specifically, they reported early activity in anterior insula (left earlier than right, as in the current study), lateral premotor regions, and inferior frontal gyrus. Raut et al. found that early regions were identified more consistently across individuals than late regions, and they proposed that these presumed neural sources serve a wide range of tasks as part of a “multiple-demand” system (Duncan, 2010).

Lag patterns differ between resting-state and task performance

Average lag patterns for resting-state and task data were highly correlated, with similar patterns of “early” and “late” regions. However, absolute lag was reduced in many regions during task performance, resulting in differences in lag distributions between conditions. Compared with the resting-state, latency projections showed greater within-group similarity during task performance; this was also true of FC estimates. Finally, peak lag could not be calculated (presumably due to noise) in ∼8% of resting-state ROI-ROI pairs, but this occurred in <1% of ROI-ROI pairs during task performance. Altogether, this suggests that compared with the resting-state condition, latency patterns during an explicit task are less susceptible to noise, show greater consistency between individuals, and show greater distinction between few “early” (i.e., left anterior insula, inferior frontal) and “late” regions (visual, medial parietal) as other regions shift toward zero-lag.

Changes in the pairwise lag distribution during task performance may suggest that neural activity contributes to observed lag patterns, despite high correlations between resting-state and task-related latencies. One possible interpretation of this finding is that resting-state BOLD latency patterns partially reflect temporal relationships sculpted by past experience (Dosenbach et al., 2007), which are modified during task-related processing. However, the greater uncertainty of neural event timing for resting compared with task data may have contributed to the higher rate of unidentifiable pairwise peak lags, and thus a wider distribution of lags around zero. Further, shared input between regions during task performance has been shown to increase FC estimates (Cole et al., 2019), which could result in a lag distribution that is clustered around zero.

No regional lag differences between ASD and typical development

Across all cortical and subcortical ROIs, there was minimal evidence of regional group differences in latency during either rest or task performance. This contrasts with resting-state findings by Mitra et al. (2017), who reported group differences in frontopolar (late in TD, average in ASD), occipital (early in TD, late in ASD), and putamen latencies (average in TD, early in ASD). Interestingly, for these three regions, resting-state latencies found in both groups in the current study most closely resemble those reported for the ASD group by Mitra et al. (i.e., average latency in frontopolar cortex, early activity in the putamen, and late activity in visual cortex).

Partly divergent findings may relate to methodological differences, including eye status (open vs. closed), temporal resolution (TR = 1.25 vs. 2.2 sec), and denoising approach (ME-ICA vs. nuisance regressors), and differing ages of participants (adolescents vs. young adults). ME-ICA denoising uses data-driven methods to identify and remove non-BOLD artefacts from various sources, including motion, respiration, and pulse. Compared with standard denoising methods using nuisance and global signal regressors, this approach has demonstrated superior recovery of BOLD signals corresponding to known resting-state networks (Kundu et al., 2012, 2013; Olafsson et al., 2015). Differences in findings compared with previous lag studies may, therefore, be related to different methods of artifact removal.

In addition, eye status has been found to affect lag estimates in visual regions, with eyes-closed paradigms eliciting earlier activity in occipital cortex (Mitra et al., 2014). This may relate to extensive effects of eye status on (zero-lag) BOLD correlations within visual cortices observed in a large multisite ASD sample (Nair et al., 2018). Furthermore, group differences in FC likely vary across development in ASD (Henry et al., 2018; Nomi and Uddin, 2015; Uddin et al., 2013). Compared with the current study, mean age was c. 3 and 9 years higher in the samples studied by Mitra et al. (2017) and Raatikainen et al. (2020), respectively. Although it is currently unknown how lag patterns may differ through adolescence and young adulthood, maturational FC changes are another potential explanation for differences in results.

Finally, Raatikainen et al. (2020) found atypical lag between multiple resting-state networks using dynamic lag analysis (DLA) in ultra-fast MREG in adults with ASD. However, aside from potential maturational changes mentioned earlier, this study differed in multiple crucial respects from the current one (e.g., peak-to-peak DLA, use of ICA-based distributed resting-state networks as ROIs, exceptionally short TR), making it difficult to attribute differences in findings to a specific variable.

Although inter-individual variability for (zero-lag) FC patterns was greater in the ASD than the TD group, no group differences in variability were found for lag patterns. In TD individuals, FC similarity measures have sometimes been associated with motion artifact (Kopal et al., 2020; Raut et al., 2019b). However, in our sample, individual FC similarity scores across all participants were not significantly correlated with head motion (i.e., RMSD) in either the resting state (r = −0.21, p = 0.15) or the task (r = −0.08, p = 0.56) condition. Although our findings are therefore consistent with reports of “idiosyncrasy” of static FC (Hahamy et al., 2015; Nunes et al., 2019), they do not suggest that variability in lag patterns plays a major role in “idiosyncratic” FC in ASD.

Latency projections correspond to arterial flow territories

Mitra et al. (2014) reported a minimal association between CBF and latency projections. The current study corroborates this finding, with no evident relationship between latency projections and perfusion values from the Pediatric Template of Brain Perfusion (Avants et al., 2015). Mitra et al. (2014) also did not find a clear association between regional lag and standard vascular territories (Damasio, 1983). However, the current study determined that lateral regions supplied by the MCA tended to be earlier on average compared with medial regions supplied by the ACA and PCA, with the highest latencies in regions supplied by the distal branches of the ACA and PCA.

Furthermore, the current study is the first to use an independent measure of arterial blood flow (ATT) to assess potential vascular contributions to BOLD latencies, which may confound interpretation of lag patterns. Mean lag for each flow territory was positively related to literature-derived estimates of ATT (Mutsaerts et al., 2015), especially during task performance, with ATT accounting for >50% of the variance in lag patterns. However, ATT explained <30% of the variance in lag during the resting-state condition. The mean latency for each territory was also highly correlated between ASD and TD participants, suggesting that any effect of arterial supply on BOLD latency was consistent across diagnostic groups. Therefore, ATT appears to heavily impact lag patterns, and positive latencies in visual and medial parietal regions (supplied by the distal branches of the ACA and PCA) should be interpreted with particular caution.

Open questions and limitations

Overall, our findings suggest that BOLD latency reflects both neural and vascular contributions, with a major role of the latter. On one hand, while the overall pattern of lag structure was similar in resting-state and task conditions, the distribution of latencies differed between conditions. Task-based lag structure also showed more reliable peak latency detection and greater consistency between individuals compared with the resting state. Altogether, these changes associated with task performance suggest some neural contribution to lag structure.

However, relationships between latency and vascular supply (i.e., arterial flow territory, ATT) suggest that observed BOLD latencies are considerably influenced by hemodynamic factors. This is expected, as the BOLD signal is an indirect measure of neural activity that is based on neurovascular coupling and associated changes in magnetic resonance (Hillman, 2014). The current study is limited by its use of an approximate map of vascular territories, which were derived from arterial spin labeling (ASL) scans in a different sample of individuals (Mutsaerts et al., 2015).

In addition, this study used estimates of vascular territories, ATT, and CBF derived from samples that are not necessarily representative of our participants. Vascular territories and ATT estimates were calculated based on an elderly adult sample (Mutsaerts et al., 2015); although CBF estimates were drawn from a pediatric population (Avants et al., 2015), this sample did not include any individuals with ASD. Therefore, while the current study sheds light on a general relationship between vascular supply and BOLD latency, future research is necessary to more clearly separate the relative influences of neural and vascular contributions to the BOLD signal by combining fMRI with complementary imaging modalities, such as electroencephalography (EEG) or ASL, in a common sample of children with and without ASD.

Additionally, the current study did not directly explore regional differences in the hemodynamic response function (cf. Yan et al., 2018) across groups or task conditions. This alternative approach to characterizing hemodynamic variables may allow for a closer examination of both inter- and intra-individual differences. Future studies may also further explore associations between hemodynamic features and lag at the individual level, as opposed to the group level alone (as in the current study).

There are several other limitations to the current study. First, our sample was relatively small, which impacts sampling variability and therefore the precision of both FC and latency analyses. Although large samples can be obtained from publicly available datasets, such as the Autism Brain Imaging Database Exchange (Di Martino et al., 2014, 2017), our in-house data benefited from high temporal resolution MESMS acquisition, as well as combined availability of resting and task data in the same individuals.

In addition, our in-house sample provided data on psychotropic medication use and co-occurring conditions, both of which are common in the general ASD population (Hossain et al., 2020; Jobski et al., 2017) and were therefore not considered exclusionary criteria. However, the effects of medication (Linke et al., 2017), and multiple diagnoses (Mikita et al., 2016) on FC have been demonstrated, and they may have affected the findings reported in this study. Moreover, inclusionary age ranged from early adolescence through young adulthood. This developmental period is associated with changes in FC (Klapwijk et al., 2013; Solé-Padullés et al., 2016; Teeuw et al., 2019) that may affect BOLD lag patterns.

Finally, fMRI data are inherently limited by the slow nature of the BOLD signal. Coordinated brain dynamics take place across a range of frequencies, including high-frequency changes on a millisecond timescale that are not observable using fMRI. Multimodal imaging approaches, such as simultaneous EEG-fMRI, have shown great promise in early studies of lagged FC (Feige et al., 2017) and dynamic brain activity (Allen et al., 2017; Bridwell et al., 2013; Musso et al., 2010; Yuan et al., 2012). However, multimodal functional neuroimaging to date remains extremely limited in ASD (Mash et al., 2018).

Conclusions

Findings suggest that adolescents and young adults with ASD show typical patterns of BOLD latency across the brain during both resting-state and task performance. Although the earliest and latest regions were preserved across resting-state and task conditions, many areas showed reduced absolute latency during task performance, which may indicate neural contributions to lag patterns. However, latency projections in both groups were strongly associated with arterial supply and transit time. Our findings, which suggest that BOLD lag patterns reflect a combination of vascular and neural factors, extend the previous literature and provide a foundation for future research, exploring spatiotemporal BOLD dynamics in ASD.

Footnotes

Acknowledgments

The authors thank all the participants and their families who took part in this study. They also thank Kalekirstos Alemu for his role in participant recruitment and data management.

Authors' Contributions

All authors contributed substantially to this work. L.E.M. conceived of the study, developed methods, conducted analyses, and wrote the original draft of the article. A.C.L. contributed to the development of methods, implementation of analyses, and article revision. Y.G., M.W., and M.A.O. contributed to data collection and article revision. R.J.J.K. contributed to data collection, development of methods, project administration, and article revision. R.-A.M. provided laboratory space and resources, acquired funding, supervised the design and implementation of the study, and revised the article.

Author Disclosure Statement

The authors have no conflicts of interest to declare.

Funding Information

This work was supported by the National Institute of Mental Health R01 MH101173 (R.-A.M.), the National Science Foundation Graduate Research Fellowship 1321850 (L.E.M.), and the San Diego State University Graduate Fellowship (M.W.).

Supplementary Material

Supplementary Figure S1

Supplementary Figure S2

Supplementary Table S1

Supplementary Table S2

References

Allen

, Damaraju

, Eichele

, et al. 2017. EEG signatures of dynamic functional network connectivity states. Brain Topogr, 31:101–116.

Allen

, Damaraju

, Plis

, et al. 2014. Tracking whole-brain connectivity dynamics in the resting state. Cereb Cortex, 24:663–676.

American Psychiatric Association. 2013. Diagnostic and Statistical Manual of Mental Disorders: DSM-5. Arlington, VA: American Psychiatric Association; 5th ed. p. 947.

Aso

, Jiang

, Urayama

, et al. 2017. A resilient, non-neuronal source of the spatiotemporal lag structure detected by BOLD signal-based blood flow tracking. Front Neurosci, 11:256.

Avants

, Duda

, Kilroy

, et al. 2015. The pediatric template of brain perfusion. Sci Data, 2:150003.

Benjamini

, Hochberg

. 1995. Controlling the false discovery rate—a practical and powerful approach to multiple testing. J R Stat Soc B Stat Methodol, 57:289–300.

Biswal

, Yetkin

, Haughton

, et al. 1995. Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magn Reson Med, 34:537–541.

Bohland

, Bokil

, Allen

, et al. 2009. The brain atlas concordance problem: quantitative comparison of anatomical parcellations. PLoS One, 4:e7200.

Bridwell

, Wu

, Eichele

, et al. 2013. The spatiospectral characterization of brain networks: fusing concurrent EEG spectra and fMRI maps. Neuroimage, 69:101–111.

10.

Brysbaert

, New

. 2009. Moving beyond Kucera and Francis: a critical evaluation of current word frequency norms and the introduction of a new and improved word frequency measure for American English. Behav Res Methods, 41:977–990.

11.

Calhoun

, Miller

, Pearlson

, et al. 2014. The chronnectome: time-varying connectivity networks as the next frontier in fMRI data discovery. Neuron, 84:262–274.

12.

Caplette

. 2020. Simple RM/mixed ANOVA for any design: MATLAB Central File Exchange. www.mathworks.com/matlabcentral/fileexchange/64980-simple-rm-mixed-anova-for-any-design Last accessed June 30, 2021 .

13.

Chen

, Witte

, Heemsbergen

, et al. 2013. Multiple comparisons permutation test for image based data mining in radiotherapy. Radiat Oncol, 8:293.

14.

Chen

, Lewis

, Chang

, et al. 2020. Resting-state “physiological networks.”. Neuroimage, 213:116707.

15.

Cole

, Ito

, Schultz

, et al. 2019. Task activations produce spurious but systematic inflation of task functional connectivity estimates. Neuroimage, 189:1–18.

16.

Constantino

, Gruber

. 2012. Social Responsiveness Scale — Second Edition. Torrance, CA: Western Psychological. Services.

17.

Cox

. 1996. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res, 29:162–173.

18.

Damasio

. 1983. A computed tomographic guide to the identification of cerebral vascular territories. Arch Neurol, 40:138–142.

19.

Di Martino

, O'Connor

, Chen

, et al. 2017. Enhancing studies of the connectome in autism using the autism brain imaging data exchange II. Sci Data, 4:170010.

20.

Di Martino

, Yan

, Li

, et al. 2014. The autism brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain architecture in autism. Mol Psychiatry, 19:659–667.

21.

Dosenbach

, Fair

, Miezin

, et al. 2007. Distinct brain networks for adaptive and stable task control in humans. Proc Natl Acad Sci U S A, 104:11073–11078.

22.

Duncan

. 2010. The multiple-demand (MD) system of the primate brain: mental programs for intelligent behaviour. Trends Cogn Sci, 14:172–179.

23.

Feige

, Spiegelhalder

, Kiemen

, et al. 2017. Distinctive time-lagged resting-state networks revealed by simultaneous EEG-fMRI. Neuroimage, 145(Pt A):1–10.

24.

Fox

, Raichle

. 2007. Spontaneous fluctuations in brain activity observed with functional magnetic resonance imaging. Nat Rev Neurosci, 8:700–711.

25.

George

, Costa

, Kouris

, et al. 1992. Cerebral blood flow abnormalities in adults with infantile autism. J Nerv Ment Dis, 180:413–417.

26.

Geschwind

, State

. 2015. Gene hunting in autism spectrum disorder: on the path to precision medicine. Lancet Neurol, 14:1109–1120.

27.

Gotham

, Risi

, Pickles

, et al. 2007. The Autism Diagnostic Observation Schedule: revised algorithms for improved diagnostic validity. J Autism Dev Disord, 37:613–627.

28.

Hahamy

, Behrmann

, Malach

. 2015. The idiosyncratic brain: distortion of spontaneous connectivity patterns in autism spectrum disorder. Nat Neurosci, 18:302–309.

29.

, Snyder

, Zempel

, et al. 2008. Electrophysiological correlates of the brain's intrinsic large-scale functional architecture. Proc Natl Acad Sci U S A, 105:16039–16044.

30.

Henry

, Dichter

, Gates

. 2018. Age and gender effects on intrinsic connectivity in autism using functional integration and segregation. Biol Psychiatry Cogn Neurosci Neuroimaging, 3:414–422.

31.

Hillman

. 2014. Coupling mechanism and significance of the BOLD signal: a status report. Annu Rev Neurosci, 37:161–181.

32.

Hossain

, Khan

, Sultana

, et al. 2020. Prevalence of comorbid psychiatric disorders among people with autism spectrum disorder: an umbrella review of systematic reviews and meta-analyses. Psychiatry Res, 287:112922.

33.

Hull

, Jacokes

, Torgerson

, et al. 2016. Resting-state functional connectivity in autism spectrum disorders: a review. Front Psychiatry, 7:205.

34.

Jann

, Hernandez

, Beck-Pancer

, et al. 2015. Altered resting perfusion and functional connectivity of default mode network in youth with autism spectrum disorder. Brain and Behavior, 5:e00358.

35.

Jenkinson

, Smith

. 2001. A global optimisation method for robust affine registration of brain images. Med Image Anal, 5:143–156.

36.

Jobski

, Hofer

, Hoffmann

, et al. 2017. Use of psychotropic drugs in patients with autism spectrum disorders: a systematic review. Acta Psychiatr Scand, 135:8–28.

37.

King

, Prigge

MBD

, King

, et al. 2018. Evaluation of differences in temporal synchrony between brain regions in individuals with autism and typical development. JAMA Netw Open, 1:e184777.

38.

Klapwijk

, Goddings

A-L

, Burnett Heyes

, et al. 2013. Increased functional connectivity with puberty in the mentalising network involved in social emotion processing. Horm Behav, 64:314–322.

39.

Kopal

, Pidnebesna

, Tomeček

, et al. 2020. Typicality of functional connectivity robustly captures motion artifacts in rs-fMRI across datasets, atlases and preprocessing pipelines. Hum Brain Mapp, 41:5325–5340.

40.

Kundu

, Brenowitz

, Voon

, et al. 2013. Integrated strategy for improving functional connectivity mapping using multiecho fMRI. Proc Natl Acad Sci U S A, 110:16187–16192.

41.

Kundu

, Inati

, Evans

, et al. 2012. Differentiating BOLD and non-BOLD signals in fMRI time series using multi-echo EPI. Neuroimage, 60:1759–1770.

42.

Kuperman

, Stadthagen-Gonzalez

, Brysbaert

. 2012. Age-of-acquisition ratings for 30,000 English words. Behav Res Methods, 44:978–990.

43.

Laumann

, Snyder

, Mitra

, et al. 2017. On the stability of BOLD fMRI correlations. Cereb Cortex, 27:4719–4732.

44.

Linke

, Olson

, Gao

, et al. 2017. Psychotropic medication use in autism spectrum disorders may affect functional brain connectivity. Biol Psychiatry Cogn Neurosci Neuroimaging, 2:518–527.

45.

Lombardo

, Lai

, Baron-Cohen

. 2019. Big data approaches to decomposing heterogeneity across the autism spectrum. Mol Psychiatry, 24:1435–1450.

46.

Lord

, Rutter

, Couteur

. 1994. Autism Diagnostic Interview-Revised: a revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders. J Autism Dev Disord, 24:659–685.

47.

Maenner

, Shaw

, Baio

, et al. 2020. Prevalence of autism spectrum disorder among children aged 8 years—autism and developmental disabilities monitoring network, 11 sites, United States, 2016. MMWR Surveill Summ, 69:1–12.

48.

Mash

, Reiter

, Linke

, et al. 2018. Multimodal approaches to functional connectivity in autism spectrum disorders: an integrative perspective. Dev Neurobiol, 78:456–473.

49.

Mikita

, Simonoff

, Pine

, et al. 2016. Disentangling the autism-anxiety overlap: fMRI of reward processing in a community-based longitudinal study. Transl Psychiatry, 6:e845.

50.

Mitra

, Kraft

, Wright

, et al. 2018. Spontaneous infra-slow brain activity has unique spatiotemporal dynamics and laminar structure. Neuron, 98:297.e6–305.e6.

51.

Mitra

, Snyder

, Constantino

, et al. 2017. The lag structure of intrinsic activity is focally altered in high functioning adults with autism. Cereb Cortex, 27:1083–1093.

52.

Mitra

, Snyder

, Hacker

, et al. 2014. Lag structure in resting-state fMRI. J Neurophysiol, 111:2374–2391.

53.

Müller

, Shih

, Keehn

, et al. 2011. Underconnected, but how? A survey of functional connectivity MRI studies in autism spectrum disorders. Cereb Cortex, 21:2233–2243.

54.

Musso

, Brinkmeyer

, Mobascher

, et al. 2010. Spontaneous brain activity and EEG microstates. A novel EEG/fMRI analysis approach to explore resting-state networks. Neuroimage, 52:1149–1161.

55.

Mutsaerts

, van Dalen

, Heijtel

, et al. 2015. Cerebral perfusion measurements in elderly with hypertension using arterial spin labeling. PLoS One, 10:e0133717.

56.

Mutsaerts

. 2020. ATT based flow territories. https://figshare.com/articles/ATT_based_flow_territories_1488674 Last accessed March 23, 2020 .

57.

Nair

, Keown

, Datko

, et al. 2014. Impact of methodological variables on functional connectivity findings in autism spectrum disorders. Hum Brain Mapp, 35:4035–4048.

58.

Nair

, Jao Keehn

, Berkebile

, et al. 2018. Local resting state functional connectivity in autism: site and cohort variability and the effect of eye status. Brain Imaging Behav, 12:168–179.

59.

Nalci

, Rao

, Liu

. 2019. Nuisance effects and the limitations of nuisance regression in dynamic functional connectivity fMRI. Neuroimage, 184:1005–1031.

60.

Nomi

, Farrant

, Damaraju

, et al. 2016. Dynamic functional network connectivity reveals unique and overlapping profiles of insula subdivisions. Hum Brain Mapp, 37:1770–1787.

61.

Nomi

, Uddin

. 2015. Developmental changes in large-scale network connectivity in autism. Neuroimage Clin, 7:732–741.

62.

Nunes

, Peatfield

, Vakorin

, et al. 2019. Idiosyncratic organization of cortical networks in autism spectrum disorder. Neuroimage, 190:182–190.

63.

Olafsson

, Kundu

, Wong

, et al. 2015. Enhanced identification of BOLD-like components with multi-echo simultaneous multi-slice (MESMS) fMRI and multi-echo ICA. Neuroimage, 112:43–51.

64.

Olsson

, Westerlund

, Lundstrom

, et al. 2015. “Recovery” from the diagnosis of autism—and then?. Neuropsychiatr Dis Treat, 11:999–1005.

65.

Pan

, Thompson

, Magnuson

, et al. 2013. Infraslow LFP correlates to resting-state fMRI BOLD signals. Neuroimage, 74:288–297.

66.

Raatikainen

, Korhonen

, Borchardt

, et al. 2020. Dynamic lag analysis reveals atypical brain information flow in autism spectrum disorder. Autism Res, 13:244–258.

67.

Rane

, Cochran

, Hodge

, et al. 2015. Connectivity in autism: a review of MRI connectivity studies. Harv Rev Psychiatry, 23:223–244.

68.

Raut

, Mitra

, Marek

, et al. 2019a. Organization of propagated intrinsic brain activity in individual humans. Cerebral Cortex, 30:1716–1734.

69.

Raut

, Mitra

, Snyder

, et al. 2019b. On time delay estimation and sampling error in resting-state fMRI. Neuroimage, 194:211–227.

70.

Schaefer

, Kong

, Gordon

, et al. 2018. Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity MRI. Cereb Cortex, 28:3095–3114.

71.

Smith

, Jenkinson

, Woolrich

, et al. 2004. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage, 23(Suppl 1):S208–S219.

72.

Smith

, Miller

, Salimi-Khorshidi

, et al. 2011. Network modelling methods for FMRI. Neuroimage, 54:875–891.

73.

Solé-Padullés

, Castro-Fornieles

, de la Serna

, et al. 2016. Intrinsic connectivity networks from childhood to late adolescence: effects of age and sex. Dev Cogn Neurosci, 17:35–44.

74.

Starkstein

, Vazquez

, Vrancic

, et al. 2000. SPECT findings in mentally retarded autistic individuals. J Neuropsychiatry Clin Neurosci, 12:370–375.

75.

Tatu

, Moulin

, Bogousslavsky

, et al. 1998. Arterial territories of the human brain: cerebral hemispheres. Neurology, 50:1699–1708.

76.

Teeuw

, Brouwer

, Guimarães

JPOFT

, et al. 2019. Genetic and environmental influences on functional connectivity within and between canonical cortical resting-state networks throughout adolescent development in boys and girls. Neuroimage, 202:116073.

77.

Tong

, Yao

, Chen

, et al. 2019. The resting-state fMRI arterial signal predicts differential blood transit time through the brain. J Cereb Blood Flow Metab, 39:1148–1160.

78.

Uddin

, Supekar

, Menon

. 2013. Reconceptualizing functional brain connectivity in autism from a developmental perspective. Front Hum Neurosci, 7:458.

79.

van Heuven

, Mandera

, Keuleers

, et al. 2014. SUBTLEX-UK: a new and improved word frequency database for British English. Q J Exp Psychol (Hove), 67:1176–1190.

80.

Wechsler

. 2011. Wechsler Abbreviated Scale of Intelligence—Second Edition. San Antonio, TX: NCS. Pearson.

81.

Yan

, Rangaprakash

, Deshpande

. 2018. Aberrant hemodynamic responses in autism: implications for resting state fMRI functional connectivity studies. Neuroimage Clin, 19:320–330.

82.

Yeo

, Krienen

, Sepulcre

, et al. 2011. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol, 106:1125–1165.

83.

Yerys

, Herrington

, Bartley

, et al. 2018. Arterial spin labeling provides a reliable neurobiological marker of autism spectrum disorder. J Neurodev Disord, 10:32.

84.

Yuan

, Zotev

, Phillips

, et al. 2012. Spatiotemporal dynamics of the brain at rest—exploring EEG microstates as electrophysiological signatures of BOLD resting state networks. Neuroimage, 60:2062–2072.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

4.11 MB

6.52 MB

0.01 MB

0.02 MB