Functional changes during visuo-spatial working memory in autism spectrum disorder: 2-year longitudinal functional magnetic resonance imaging study

Abstract

This study examined functional changes longitudinally over 2 years in neural correlates associated with working memory in youth with and without autism spectrum disorder, and the impact of increasing cognitive load. We used functional magnetic resonance imaging and a visuo-spatial 1-back task with four levels of difficulty. A total of 14 children with autism spectrum disorder and 15 typically developing children (ages 7–13) were included at baseline and followed up approximately 2 years later. Despite similar task performance between groups, differences were evident in the developmental trajectories of neural responses. Typically developing children showed greater load-dependent activation which intensified over time in the frontal, parietal and occipital lobes and the right fusiform gyrus, compared to those with autism spectrum disorder. Children with autism spectrum disorder showed minimal age-related changes in load-dependent activation, but greater longitudinal load-dependent deactivation in default mode network compared to typically developing children. Results suggest inadequate modulation of neural activity with increasing cognitive demands in children with autism spectrum disorder, which does not mature into adolescence, unlike their typically developing peers. Diminished ability for children with autism spectrum disorder to modulate neural activity during this period of maturation suggests that they may be more vulnerable to the increasing complexity of social and academic demands as they progress through adolescence than their peers.

Keywords

adolescents autism spectrum disorders cognitive load functional magnetic resonance imaging longitudinal school-age children working memory

Introduction

Autism spectrum disorder (ASD) is a heterogeneous neurodevelopmental disorder characterized by deficits in social communication and interactions, as well as restricted, repetitive patterns of behaviour and interests (American Psychiatric Association (APA), 2013). Individuals with ASD also often present with a number of cognitive, learning and executive function impairments (Hill, 2004b; Rosenthal et al., 2013). Key cognitive accounts of ASD include the theory of mind deficit hypothesis (Baron-Cohen et al., 2000), central coherence hypothesis (Happé, 2005), and the executive dysfunction hypothesis (Hill, 2004a; Joseph, 1999). Researchers have moved away from a model analysing a single, cognitive deficit as an explanatory account of ASD to a framework that supports multiple cognitive differences (Happé et al., 2006). While executive function impairments do not have a primary causal role in ASD, executive function problems have a significant impact on their developmental outcomes (Pellicano, 2012). Working memory is considered to be one of the core components of executive functions (Verté et al., 2006), and previous studies have reported working memory deficits in ASD relative to typically developing (TD) populations (Barendse et al., 2013). Furthermore, previous literature has identified the associations between working memory deficits and social function, adaptive behaviour and academic success in ASD (Alloway, 2009; Dennis et al., 2009; Engle et al., 1999; Gilotty et al., 2002; Killany et al., 2005; Richmond et al., 2013).

There is evidence for primarily visuo-spatial working memory impairment in ASD, whereas verbal working memory appears relatively intact in individuals (Boucher and Bowler, 2008; Minshew and Goldstein, 2001; Ozonoff and Strayer, 2001; Steele et al., 2007; Williams et al., 2005). Visual working memory is critical for recalling any type of visual information, including sequences of events, patterns and images, and has been shown to be a sensitive marker of developmental disability (Alloway et al., 2016). Visuo-spatial working memory, specifically, represents the ability to briefly maintain and manipulate spatial information (e.g. shapes and colours, as well as their locations) in mind. Thus, poor visuo-spatial working memory may have implications for how individuals with ASD process their social and academic environments. For instance, children with ASD may struggle to process visual cues (e.g. non-verbal language, facial expressions) during social interactions, making it challenging for them to carry on conversations and to relate to their peers, compared to TD individuals. Moreover, visuo-spatial working memory often operates as a ‘mental scrap paper’ in the classroom, and thus deficits can result in difficulties completing simple mental arithmetic, understanding concepts and organizing thoughts. These social and environmental demands become increasingly complex as children mature into adolescence, and consequently it is also important to consider the development of associated, underlying neuropsychological mechanisms, such as visuo-spatial working memory processing, that are vulnerable in ASD.

Cross-sectional neuropsychological studies of working memory in ASD reveal persistent impairments in visuo-spatial working memory throughout development (i.e. childhood, adolescence and adulthood; Luna et al., 2007). Using an oculomotor delayed response task, Luna et al. (2007) showed that although visuo-spatial working memory skills improved over time for both groups, a developmental delay of working memory was observed in ASD, extending into adulthood, and remained impaired compared to TD individuals. In addition, Rosenthal et al. (2013) showed significantly more parent-reported working memory impairments in older than younger children with ASD, which contrasts with the reported general improvement of working memory in everyday settings in TD children (Andersen et al., 2015). All of the above studies were cross-sectional; a few have used longitudinal designs to evaluate working memory development in ASD. A recent 2-year longitudinal study indicated that whereas verbal working memory continues to develop in typical and attention deficit hyperactivity disorder (ADHD) groups (ages 9–16), children with ASD displayed a developmental arrest (i.e. showed no improvement 2 years later; Andersen et al., 2015).

Despite some evidence for differential developmental trajectories of working memory in ASD compared to TD populations, to our knowledge there are no studies that examine the neural basis of this development. This is a critical gap in our knowledge, given that the frontal lobes, which support working memory (Greene et al., 2008), have a protracted maturation well into adolescence (Sowell et al., 2004), making the functions they support (e.g. working memory) vulnerable to developmental disturbances. Normative neurodevelopmental studies have found age-related increases in the intensity of activation of fronto-parietal clusters (Olesen et al., 2007) or only parietal regions (Spencer-Smith et al., 2013) during visuo-spatial working memory tasks. Other studies have reported developmental shifts of neural activation during a visual categorical n-back task from reliance on the dorsal visual stream in young children to reliance on the visual ventral pathway (involving prefrontal and inferior temporal regions) in adults (Ciesielski et al., 2006).

Although the developmental changes in working memory processing have not been studied in ASD, a few functional magnetic resonance imaging (fMRI) studies have examined the neural correlates of working memory function in either adolescents or adults with ASD separately, with only one study of pre-adolescent children by our group (Vogan et al., 2014). Overall, studies converge to identify a fronto-parietal visuo-spatial working memory network, involving a system of prefrontal, premotor, dorsal cingulate and posterior parietal activation, and adults with ASD show reduced activation in these areas during working memory tasks (Koshino et al., 2005; Luna et al., 2002). Using a working memory mental rotation task, Silk and colleagues (Silk et al., 2006) observed reduced cortical activation in frontal regions in adolescents with ASD compared to TD adolescents, including the anterior cingulate, dorsolateral prefrontal cortex (DLPFC) and caudate nucleus, despite no differences in task performance.

Some fMRI investigations have considered the impact of increasing cognitive load on neural systems underlying working memory, as manipulating load in executive function studies has shown to provide a more refined profile of abilities and impairments in ASD. A number of previous studies have failed to demonstrate performance differences between individuals with and without ASD on basic working memory tasks, but report that deficits emerge with increasing task complexity and/or working memory load in ASD (Koolen et al., 2014; Minshew and Goldstein, 2001; Rahko et al., 2016; Russo et al., 2007). Rahko et al. (2016) found that with a visuo-spatial working memory task contrasted against an attention task (e.g. 0-back), adolescents with ASD showed reduced modulation of brain activity across increasing task demands in the insula, somatosensory, motor and auditory cortices compared to TD children. Our group demonstrated inadequate task-load modulation in the precuneus, DLPFC and medial premotor cortex of youth with ASD during a visuo-spatial working memory paradigm with four levels of difficulty (Vogan et al., 2014). These findings suggest that neural functional capacity saturates with high information load in individuals with ASD, and they are unable to modulate the brain function according to increasing demands. However, to better understand neurodevelopmental trajectories of ASD compared to TD individuals, these patterns of brain activation should be studied longitudinally.

We also incorporated complexity into our variation of the ‘n-back’ protocol (the Colour Matching Task (CMT); Arsalidou et al., 2010), using six levels of difficulty. The CMT is unique in that it is a 1-back task, which systematically manipulates memory load while keeping other executive functions constant across all difficulty levels, allowing for a direct investigation of the influence of increasing demands on working memory. Increasing difficulty on typical n-back tasks requires the utilization of different mental strategies at each level (e.g. 0-back: recognition, 1-back: maintenance, 2-back: maintenance and monitoring). This manipulation increases executive function demand in addition to memory load, making it difficult to isolate neural responses specific to working memory processes (see Vogan et al., 2014 for a thorough description of the CMT). Our baseline study of the CMT demonstrated task-induced activation in working memory-related areas (e.g. frontal and parietal areas) and deactivation in the default mode network (DMN) with increasing load in TD youth (Vogan et al., 2014). However, this opposing system of cognitive processes was absent in those with ASD, who did not modulate neural activity in response to increasing task demands to the same extent as TD children.

The aim of this study was to identify functional changes longitudinally (over 2 years) in the neural correlates associated with increasing working memory load in children and adolescents with and without ASD. A 2-year follow-up period was chosen to provide enough time to observe memory change, while trying to limit participant attrition. We hypothesized that children with ASD would show less load-dependent activation than TD children in frontal and parietal regions, and these differences would become more marked into adolescence. As children mature, environmental demands become increasingly complex, prompting cognitive, social and affective maturation, as well as significant neuroanatomical changes (Paus, 2005; Sowell et al., 2001), making the period of early adolescence especially interesting to study. Not only is working memory fundamental to cognition (Engle et al., 1999), learning and academic achievement (Alloway, 2009), it also associated with many of the key characteristics of ASD, such as the social and communicative impairments (Gilotty et al., 2002; Kenworthy et al., 2009; Leung et al., 2016; Richmond et al., 2013). Understanding the evolution of neural development underlying working memory throughout childhood can provide valuable insights into the cognitive profile of ASD and localize the brain areas of vulnerability to developmental disturbances – both of which will help inform interventions for those with ASD.

Methods

Participants

A total of 83 participants were originally recruited for this study (2011–2013), consisting of 44 children with ASD and 39 TD children between the ages of 7 and 13 years. Participants were followed up 2 years later (9–15 years old). At each time point, the participants underwent neurocognitive assessments and neuroimaging. At the baseline assessment, 12 TD children and 23 children with ASD were excluded for excessive movement and inadequate fMRI task performance or protocol completion. Out of the remaining 48 participants, 16 were excluded from the study due to (1) not returning at the 2-year follow-up (e.g. braces; relocated; n = 9), (2) excessive head movement (n = 4) and (3) poor fMRI task performance (n = 3). To note, participants who failed to return for follow-up did not differ from those who did return in age (t₍₄₆₎ = 1.13, p = 0.26), sex (χ² = 0.22, p = 0.64) and baseline intelligence quotient (IQ; t₍₄₄₎ = 0.89, p = 0.38; missing IQ for two participants who did not return). There was also no significant difference in CMT performance at any level between participants who completed the follow-up and those who did not (Level 3: t₍₄₆₎ = 3.95, p = 0.70; Level 4: t₍₄₆₎ = 0.06, p = 0.96; Level 5: t₍₄₆₎ = 0.43, p = 0.67; Level 6: t₍₄₆₎ = 1.39, p = 0.17). After sex and age matching, the final sample consisted of 14 children with ASD (13 boys) and 15 TD children (10 boys). Although our rates of unusable data are higher than often reported in the literature, our task was long and complex, and thus was more taxing than those typically used with children, particularly clinical populations. Demographic data from both time points are included in Table 1.

Table 1.

Participant characteristics.

Variables	Time point	TD (N = 15)		ASD (N = 14)		Significance test
Variables	Time point	Range	M (SD)	Range	M (SD)	Significance test
Sex (M: F)			10:5		13:1	χ² = 0.08, p = 0.08
Age (years)	Baseline	7–14	11.7 (2.16)	7–13	10.9 (1.98)	t₍₂₇₎ = 1.05, p = 0.31
Age (years)	Follow-up	9–16	13.7 (2.17)	10–16	13.4 (1.82)	t₍₂₇₎ = 0.45, p = 0.65
IQ	Baseline	99–146	118.8 (11.60)	86–135	111.8 (17.21)	t₍₂₇₎ = 1.31, p = 0.20^*
IQ	Follow-up	104–136	116.5 (9.52)	86–127	107.5 (12.76)	t₍₂₇₎ = 2.17, p = 0.04^*
ADOS SA	Baseline			4–15	9.0 (3.06)
ADOS SA	Follow-up			3–21	10.6 (4.82)
ADOS RRB	Baseline			0–6	3.1 (2.0)
ADOS RRB	Follow-up			0–7	2.4 (1.70)
ADOS CSS	Baseline			4–9	7.1 (1.80)
ADOS CSS	Follow-up			3–10	7.3 (2.23)
Days between visits			760.8 (38.3)		791.8 (106.5)	t₍₂₇₎ = 1.03, p = 0.32

TD: typically developing; ASD: autism spectrum disorder; IQ: intelligence quotient; ADOS: Autism Diagnostic Observation Schedule; SA: Social Affect ADOS subdomain score; RRB: Restricted and Repetitive Behaviour ADOS subdomain score; CSS: calibrated severity score.

CSS ranges from 1 to 10 (10 = most severe).

There was no significant Group × Time interaction (p > 0.05) of IQ and thus developmental IQ trajectories were similar across groups.

All participants were free from any diagnosed psychiatric comorbidities, neurological disorders, medical illnesses, prematurity, uncorrected vision, colour blindness, as well as standard magnetic resonance imaging (MRI) contraindicators (e.g. ferromagnetic implants). A history of developmental delay, learning disability and ADHD was used to exclude TD children only; however, these factors were also not current primary diagnoses in any participants of the ASD group. Four children with ASD were each on one psychotropic medication at one time point (e.g. Biphentin, Strattera and Abilify).

Clinical diagnoses of ASD were confirmed in all cases, with a combination of expert clinical judgement, clinical records and the Autism Diagnostic Observation Schedule (ADOS; Lord et al., 2000) which was administered by a trained individual who maintains inter-rater research reliability. All participants possessed a full scale IQ estimate of 80 or above measured by the Wechsler Abbreviated Scale of Intelligence (WASI; Wechsler, 1999), two-subtest version. All children completed the Backwards Digit Recall, Listening Recall, Digit Recall, Mazes Memory and Block Recall subtest of the Working Memory Test Battery for Children (WMTB-C; Pickering and Gathercole, 2001) at both time points. See Supplementary Figure 1 for neuropsychological test data. Neuropsychological and clinical data were obtained on the same visit as scans, at both time points.

Children were recruited through community support centres, parent support groups, email lists, hospital ads and private schools. Informed consent, clinical and cognitive testing and MRI scanning were performed at the Hospital for Sick Children (SickKids) in Toronto, Ontario. Experimental procedures were approved by the Research Ethics Board at SickKids. All children provided informed assent, and the parents provided informed written consent.

fMRI visuo-spatial working memory paradigm (CMT)

The CMT is an n-back task, in which participants are instructed to attend to coloured figures of clowns presented in sequence, one at a time. Before scanning, the participants were trained and completed practice trials at both time points until they had an accuracy of ⩾80%. Participants were taught to ignore the face of the clown and colours that were irrelevant (blue and green) and focus on the relevant colours (pink, yellow, red, purple, orange, grey and brown). The number of ‘n’ relevant colours (i.e. load) in the clown figure was increased by one for every increase in difficulty level (D). See Figure 1(a) for an example of a sequence presented in D3 and D6. The CMT includes two functions that require mental attention: (1) participants must first identify the relevant colours within the clown figure and (2) second determine if the colours of the current clown match with colours of the previous clown, ignoring irrelevant colours (blue and green). Therefore, items with n (e.g. 3) relevant colours would have a difficulty level of n + 2 (e.g. 5), rather than difficulty level 3. After each stimulus, the participants indicated if the relevant colours in the figure were the same colours as in the immediately preceding figure (i.e. ‘1-back’), regardless of colour location or colour repetition (see Figure 1(a)). Participants responded by pushing a button for ‘same’ or ‘different’ using an MR-compatible button box with the right hand.

Figure 1.

Protocol description of the Colour Matching task (CMT). (a) Examples of a sequence presented in D3 and D6. There were six difficulty levels. The number of relevant colours (i.e. load) in the figures was increased by one colour for every increase in difficulty level. Difficulty = (# of colours) + 2. Children were taught to ignore the clown’s face, colour location, colour repetition and irrelevant colours (blue and green). After each stimulus, children indicated if the relevant colours in the figure were the same or different from the colours in the immediately preceding figure (i.e. 1-back). (b) CMT was a block design task, where each run consisted of six 32-s blocks (for each difficulty level) followed by (c) 20-s baseline blocks where clowns are presented in only blue and green (irrelevant colours).

Four runs were presented. Each run included six 32-s blocks, one for each of the six difficulty levels (see Figure 1(b)). The difficulty level was constant within each block, and all six difficulty levels were randomized within the runs. The same four runs were presented to the participants in the same order. Each task block contained eight stimuli, yielding 168 task trials in total and alternated with 20-s baseline blocks. For the baseline blocks, the clowns were coloured in blue and green only, and children were instructed to look at the clowns but not to respond (Figure 1(c)). In the task blocks, the participants had 3 s to view each stimulus and respond, followed by a fixation cross for a 1-s inter-stimulus interval. Total scan time for the fMRI protocol was approximately 22 min. Participants’ data were excluded from the analyses if they completed fewer than three runs.

Accuracy and response times were recorded during the fMRI task; items were correct if the child responded correctly within 3 s of stimulus presentation. Task performance was considered adequate if (a) participants reached ⩾60% accuracy (averaged across the four runs) on the two easiest difficulty levels and (b) participants completed two or more runs where at least 50% of the blocks were acceptable for performance (60% accuracy) and motion. Motion was considered acceptable if the participants moved ˂1.5 mm from the median head position in ⩾60% of the volumes per task block. The fMRI preprocessing section below describes the displacement calculations. Motion parameters were entered into the fMRI preprocessing pipeline. An accuracy criterion of 60% was chosen, as we could be sure that the participants were performing above chance (50%), but it was not too stringent for a clinical paediatric population.

Image acquisition

All imaging data were acquired using a 3-T Siemens Trio MRI scanner with a 12-channel head coil. Head stabilization and motion restriction were achieved with foam padding. A high-resolution T1-weighted three-dimensional (3D) magnetization-prepared rapid gradient-echo (MP-RAGE) structural scan (sagittal; field of view (FOV) = 192 mm × 240 mm × 256 mm; 1-mm isometric voxels; TR/TE/TI/FA = 2300/2.96/900/9) was used as an individual anatomical reference for the fMRI images. During structural image acquisition, the participants watched a movie of their choice using MR-compatible goggles and earphones. Functional images were acquired with single-shot echo planar imaging sequence (axial; FOV = 192 mm × 192 mm; Res = 64 × 64; 30 slices 5 mm thick; 3 mm× 3 mm × 5 mm voxels; TR/TE/FA = 2000/30/70). Visual stimuli for the functional task (CMT) were displayed on MR-compatible goggles. Stimuli were displayed and performance was recorded using the software Presentation (Neurobehavioral Systems Inc., Berkeley, CA, USA)

CMT behavioural data analyses

To be consistent with our previous studies (Vogan et al., 2014, 2016), only the first four difficulty levels (D3–D6) were analysed, as performance on D7 and D8 was at chance or only marginally above chance levels in the majority of the participants. Behavioural accuracy (% correct) and response times were calculated for each difficulty level, averaging across runs for each participant, at baseline and follow-up. Data were analysed using a 2-way mixed analysis of variance (ANOVA) for each time point, with group (ASD and TD) as a between-subject factor and difficulty level (D3, D4, D5 and D6) as a within-subject factor. Bonferroni post hoc pairwise comparisons were computed to explore the main effect of difficulty level for each group separately at each time point. All the reported significance levels were adjusted using Bonferroni correction in SPSS. This adjustment involves multiplying the significance levels by the number of pairwise comparisons performed (i.e. p values were multiplied by 6 when there were four difficulty levels to compare). In other words, pairwise comparisons had to be significant at the 0.05/6 = 0.00833 level to be significant at the 0.05 level under Bonferroni.

Behavioural data were also analysed using a supplemental 3-way mixed ANOVA, with group (ASD and TD) as a between-subject factor and difficulty level (D3, D4, D5, D6) and time (baseline and follow-up) as the within-subject factors. See Supplementary material for details of these analyses.

fMRI data analyses

Image preprocessing of functional data was performed using FMRIB’s Software Library (FSL) Version 5 (Worsley, 2001). The first three volumes of each run were discarded for scanner stabilization. Following slice timing and motion correction, the images were smoothed with a 6-mm full width at half maximum (FWHM) Gaussian filter, temporally filtered with a high-pass filter cut-off frequency of 0.01 Hz. To control for motion, MCFLIRT was used for volume alignment and the standard motion parameters found from MCFLIRT were also included as a covariate of no interest in the general linear model (GLM) of the standard FSL preprocessing. Maximum displacement (MD) was calculated using 3dvolreg from the AFNI toolbox (Cox, 1996). This calculates the MD (across the whole brain) of each volume to any single reference volume. This MD metric was used to flag volumes that had unacceptable motion, as described above. The average MD at each time point for subjects was used to explore differences in head motion across time. Results of a paired-samples t test showed that motion did not change across time for TD children (M_baseline = 0.35 mm, SD_baseline = 0.30 mm; M_follow-up = 0.27 mm, SD_follow-up = 0.21 mm; t₍₁₄₎ = 0.87, p = 0.40) or children with ASD (M_baseline = 0.65 mm, SD_baseline = 0.57 mm; M_follow-up = 0.50 mm, SD_follow-up = 0.50 mm; t₍₁₃₎ = 1.10, p = 0.29). The average MD was calculated across time for each subject to explore overall group differences in head motion. There were no significant differences in average head motion between children with (M = 0.58 mm, standard deviation (SD) = 0.46 mm) and without ASD (M = 0.31 mm, SD = 0.19 mm; t₍₂₇₎ = 1.98, p = 0.07). Although there was a trend for differences in head motion between groups, we used conservative methods to cope with it (e.g. including movement regressors in the analysis), and the average movement in both groups was minimal (i.e. less than 0.6 mm).

Both functional and T1 structural images were brain extracted. The functional data were registered to the corresponding T1 image using a 6-parameter linear transformation. Each subject’s T1 was registered to the MNI152 T1 template with a 12-parameter linear registration. These transformations were concatenated to transform functional data to standard space for higher level analyses.

First-level statistical analyses of blood oxygen level–dependent (BOLD) activity during CMT were conducted using FSL Expert Analysis Tool (FEAT; Woolrich et al., 2009). For each subject at each time point, the data were fit to a block design GLM convolved with a gamma function using the task parameters (D3–D6). Contrasts between D6 and D3 were calculated, and these results were averaged across runs for each subject at each time point in a second level analysis. Between-group comparisons were conducted using FMRIB’s Local Analysis of Mixed Effects type 1 (FLAME-1; Worsley, 2001) – whole-brain statistical analyses of BOLD activity during CMT. The longitudinal change in the D6 > D3 contrast was compared between participants with and without ASD using a Group × Time interaction. This method is described in the FEAT User Guide (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM) under ANOVA: 2-groups, 2-levels per subject (2-way Mixed Effect ANOVA). Significant interactions were reported using cluster-based thresholding determined by Z > |2.3| and a cluster-corrected significance threshold of p < 0.05. Central regions of interest (ROIs) were selected from areas exhibiting significant interactions. Average signal change was extracted from spherical ROIs (5 mm radius) and plotted for D6 > D3 and for each group at both baseline and follow-up time points.

Results

Behavioural data

There was no significant interaction between group and difficulty level on accuracy at baseline (F_(2.19,27) = 0.36, p = 0.72) or follow-up (F_(2.05,27) = 1.62, p = 0.21; Greenhouse–Geisser corrected degrees of freedom). There was also no significant effect of group on accuracy at baseline (F_(1,27) = 0.40, p = 0.54) or follow-up (F_(1,27) = 1.88, p = 0.18; Figure 2(a)). Therefore, comparisons of brain activity between TD children and children with ASD were made under comparable accuracy scores across all levels at both time points. There was a significant main effect of difficulty level on accuracy, with accuracy decreasing as a function of difficulty in both groups at baseline (F_(2.26,27) = 44.96, p < 0.001) and follow-up (F_(2.05,27) = 37.32, p < 0.001; Greenhouse–Geisser corrected degrees of freedom). At baseline, Bonferroni post hoc pairwise comparisons revealed that the accuracy decreased significantly between D4 and D5 (p_corr < 0.05) for TD children but not between any other consecutive difficulty levels. Children with ASD showed significant decreases in accuracy across all levels (p_corr < 0.05), except between D5 and D6. At follow-up, TD children showed no statistically significant decreases in performance between consecutive difficulty levels, whereas children with ASD showed a significant decline in performance between D4 and D5 (p_corr < 0.05).

Figure 2.

CMT behavioural performance: (a) mean proportion correct and (b) mean response times for D3–D6 at baseline and follow-up, and standard error bars. There were no main effects of group on accuracy or response time at either time point.

At both time points, response times (RTs) generally increased as a function of difficulty, with some exceptions (described below; see Figure 2(b)). There was no interaction between group and difficulty level on RTs at baseline (F_(2.07,27) = 2.74, p = 0.07) or follow-up (F_(1.80,27) = 1.77, p = 0.31; Greenhouse–Geisser corrected degrees of freedom). There was also no effect of group on RTs at baseline (F_(1,27) = 0.67, p = 0.42) or follow-up (F_(1,27) = 3.83, p = 0.06). However, as expected, there was a main effect of difficulty level on RTs at baseline (F_(2.07,27) = 44.73, p < 0.001) and follow-up (F_(1.80,27) = 51.01, p < 0.001; Greenhouse–Geisser corrected degrees of freedom). For TD children, Bonferroni post hoc pairwise comparisons at baseline showed significant increases in RTs between D3 and D4, and D4 and D5, but not between D5 and D6, whereas children with ASD showed only a significant increase in RTs between D3 and D4 (p_corr < 0.05). At follow-up, TD children showed increases in RTs between all consecutive difficulty levels, whereas children with ASD showed a significant increase only between D3 and D4, and D4 and D5 (p_corr < 0.05). The decrease in RTs at the most difficult level in the ASD group was not significant.

See Supplementary material for details of the 3-way mixed ANOVA; the results showed no significant three-way interactions in accuracy or RTs.

fMRI analyses

Group-level activation of D6 versus D3

Group-level activation during D6 versus D3 at baseline and follow-up are presented in Figure 3. In TD children, activations increased as a function of task difficulty (i.e. load; D6 > D3) in the right precuneus and left occipital cortex at baseline, and activations decreased as a function of load (i.e. D6 < D3) in typical DMN regions including the bilateral anterior medial frontal gyrus, posterior cingulate and angular gyri. D6 versus D3 contrasts of activation in TD children showed more widespread activity at follow-up compared to baseline. Specifically, at follow-up, activations increased as a function of load in the bilateral inferior frontal gyri, occipital cortex, lingual gyri, the precuneus extending into the superior parietal lobules, the anterior cingulate extending to the dorsal medial frontal gyrus and the right DLPFC. Consistent with baseline testing 2 years earlier, activation decreased with increasing load in the bilateral anterior medial frontal gyri, angular gyri and posterior cingulate.

Figure 3.

Group activation maps for the D6 versus D3 contrast during CMT at baseline and follow-up. Significant activations using cluster-based thresholding determined by Z > |2.3| and a corrected cluster significance threshold of p = 0.05. Areas in red depict regions of increasing activation from D3 to D6 (i.e. D6 > D3) and areas in blue depict regions of decreasing activation from D3 to D6 (i.e. D6 < D3).

Changes in D6 versus D3 activation patterns from baseline to follow-up were observed to be minimal in children with ASD. At baseline, activation increased as a function of increasing task difficulty in the right fusiform gyrus, bilateral middle occipital gyri and left precuneus in children with ASD, and the right insula showed decreasing activation as a function of load. At follow-up, children with ASD showed increasing activation with load in the bilateral fusiform gyri, middle occipital gyri and right precuneus, and decreasing activation in the bilateral posterior cingulate, anterior medial prefrontal gyri, angular gyri, middle temporal gyri and left DLPFC.

Longitudinal change in functional activation of D6 versus D3

A significant Group × Time interaction (Z > 2.3, p < 0.05, cluster-corrected) was observed in the D6 versus D3 contrast. Two different patterns were observed in regions exhibiting a significant interaction (see Figure 4). In regions such as the bilateral precuneus extending into the superior parietal lobules, right fusiform gyrus, left superior occipital gyrus/cuneal cortex and left angular gyrus, the TD group showed a greater positive change in BOLD in the D6 versus D3 contrast over time (baseline to 2-year follow-up) than children with ASD (see Table 2). In other words, in these regions, TD children showed increased recruitment as a function of load, and this load-dependent change in activation increased across time, compared to children with ASD. In regions such as the bilateral ventromedial prefrontal cortex and right parahippocampal gyrus, the D6 versus D3 contrast became increasingly more negative over time in children with ASD compared to TD children. In other words, children with ASD showed a decrease in BOLD signal with increasing load, and this pattern significantly increased across time compared to TD children. See Figure 5 for graphs of signal change in D3 and D6 at baseline and follow-up for children with and without ASD.

Figure 4.

Between-group comparisons in the longitudinal change in functional activation of D6 versus D3. Significant activations using cluster-based thresholding determined by Z > |2.3| and a corrected cluster significance threshold of p = 0.05. Areas in red depict regions that exhibited a significant Group × Time interaction. In the bilateral precuneus (Prec), superior parietal lobules (Sup Par Lob), right fusiform gyrus, left superior occipital gyrus (SupOcG)/cuneal cortex and left angular gyrus, the control group showed greater increase in the D6 versus D3 contrast over time (baseline to 2-year follow-up) than children with ASD. In areas circled in green, such as the bilateral ventromedial prefrontal cortex (vmPFC)/orbital frontal gyrus (Orb FG) and right parahippocampal gyrus (ParahipG), the D6 versus D3 contrast became increasingly more negative over time in children with ASD compared to control children.

Table 2.

Brain regions showing a significant Group × Time interaction of the D6 versus D3 contrast during CMT.

Voxels	MNI coordinates			Z value	P value	Hem.	Region
Voxels	x	y	z	Z value	P value	Hem.	Region
1208	−20	−62	44	2.21	8.29 × 10⁻⁶	L	Precuneus/superior parietal lobule^a
X	−18	−68	46	3.84		L	Precuneus^a
X	−22	−84	24	3.71		L	Superior occipital cortex^a
X	−14	−80	38	3.55		L	L inferior precuneus/cuneal cortex^a
X	−16	−78	24	3.47		L	Cuneal cortex^a
X	−28	−62	36	3.42		L	Angular gyrus^a
823	−6	60	2	4.11	3.23 × 10⁻⁴	L	Ventromedial prefrontal cortex^b
X	0	52	−6	3.87		L/R	Ventromedial prefrontal cortex^b
X	30	54	−12	3.56		R	Orbital frontal gyrus^b
562	36	−44	−14	3.65	5.42 × 10⁻³	R	Fusiform gyrus^a
X	34	−22	−18	3.49		R	Parahippocampal gyrus^b
495	26	−72	48	3.21	0.01	R	Superior parietal lobule^a
X	36	−72	36	3.19		R	Precuneus^a
X	28	−74	34	2.99		R	Precuneus^a

CMT: Colour Matching Task; MNI: Montreal Neurological Institute.

MNI coordinates represent the peak Z value of the cluster; X represents the peak local maxima within a cluster.

Regions where TD children showed a greater increase of activation in the D6 versus D3 contrast over time (baseline to 2-year follow-up) than children with ASD.

Regions where activation in the D6 versus D3 contrast became increasingly more negative (i.e. more deactivated) over time in children with ASD compared to TD children.

Figure 5.

Mean peak cluster percent signal change. Graphs of signal change during D3 and D6 at baseline and follow-up in areas that exhibited a significant Group × Time interaction of the D6 versus D3 contrast.

Discussion

This is the first longitudinal fMRI study to investigate functional changes in the brain during working memory processing in children and adolescents with ASD. Overall, we found a differential developmental trajectory for neural substrates underlying working memory over four levels of increasing difficulty between the groups. We found that TD children showed significantly greater longitudinal increased load-dependent activation in the parietal lobes (including the bilateral precuneus extending into the superior parietal lobules), right fusiform gyrus, left superior occipital gyrus/cuneal cortex and left angular gyrus compared to those with ASD. However, children with ASD showed significantly greater longitudinal load-dependent decreased activation in the bilateral ventromedial prefrontal cortex and right parahippocampal gyrus – areas associated with the DMN – compared to TD children. Despite differences in neural activation patterns, task performance and response time differences between the TD and ASD groups were absent at both time points. However, there was a trend (p < 0.06) for children with ASD to have minimally longer response times on the CMT than TD children at follow-up. Thus, although task performance differences between groups were not present, children with ASD took somewhat longer to respond to items, suggesting that they may have experienced more challenges than TD children. Results of this study extend previous child and adolescent fMRI studies of working memory in ASD (Rahko et al., 2016; Silk et al., 2006; Vogan et al., 2014) by adding longitudinal information on the development of functional neural networks underlying working memory.

The TD group was able to increase their brain activation as a function of working memory load in the parietal cortices, and this load-dependent activation intensified across development, whereas children with ASD displayed little developmental change. Previous (non-longitudinal) studies have found reduced task-load modulation in the parietal cortices in children with ASD compared to typical development (Rahko et al., 2016; Vogan et al., 2014), and this study extends this finding to show that impairment in ASD persists as children mature. Furthermore, behavioural studies exploring the maturation of executive functioning in ASD also demonstrate persistent impairments in working memory throughout development (Andersen et al., 2015; Luna et al., 2007). Other developmental cross-sectional fMRI studies of TD youth describe increasing parietal activity during visuo-spatial working memory with age (Ciesielski et al., 2006; Spencer-Smith et al., 2013); this developmental change was not evident in children with ASD in this study. In addition, whereas load-dependent activation was also observed in the frontal cortex for TD children at follow-up but not at baseline (Figure 3), children with ASD did not develop task modulation in the frontal regions across time. Surprisingly, there was no significant difference in frontal developmental trajectories (i.e. Group × Time interaction) between groups, likely due to variability in the ASD group.

The parietal region has been associated with increasing storage of visuo-spatial information critical for working memory capacity, namely, with the amount of information to be stored (Macoveanu et al., 2006). Findings of this study suggest that whereas TD children are able to modulate neural activity according to increasing demands as they mature, children with ASD do not appear to do this. Given similar task performance in ASD, it is important to entertain the idea that higher order processing (e.g. neural modulation) may not be necessary for ASD with tasks that can be processed efficiently using a perceptual processing approach due to enhanced perceptual functioning skills (Mottron et al., 2006). Enhanced perceptual processing, despite a deficiency in neural modulation, may be adequate for analysing visual details under highly controlled conditions created by laboratory-based tasks, such as the CMT. However, it may be less adaptive for more complex information processing, such as socialized or ‘real-world’ tasks. In other words, the observed neural deficits in ASD may have more of an impact on functioning in everyday life. Compelling research has demonstrated that executive functioning problems in everyday settings, including working memory, are observed in individuals with ASD, even when performance on laboratory tasks is intact (Kenworthy et al., 2008). This would have implications for working memory processing that takes place under real-world expectations, such as, for example, during social interactions, which require one to balance multiple demands, including decoding facial and non-verbal cues, listening and understanding verbal information and responding (i.e. increasing load; Baddeley et al.,1997). With insufficient neural modulation capacity, children with ASD may be more vulnerable to the increasing complexity of real-world demands as they progress into adolescence (e.g. the need to develop professional and romantic relationships) compared to their typical peers. It has also been suggested that hyper-perception (i.e. hyper-focusing on fragments of details so intensely) causes individuals with ASD to become oversaturated and overwhelmed with increasing cognitive demands (Markram and Markram, 2010). Therefore, impaired neural modulation of the parietal region likely has negative effects on a number of functions in ASD.

The visuo-spatial nature of the CMT and the gradual increase in the number of colours that needed to be processed should require increased involvement of regions associated with categorization (i.e. occipito-temporal) and spatial search (i.e. occipito-parietal areas) across cognitive load. The occipito-parietal pathway, referred to as the ‘dorsal stream’, includes the precuneus and is associated with encoding spatial properties, such as position, size and orientation. The occipital–temporal pathway, referred to as the ‘ventral stream’, includes the fusiform gyri and is specialized in encoding object properties such as colour, shape and texture (Pisella, 2016; Ungerleider and Mishkin, 1982). Previous fMRI studies have observed recruitment of the visual and dorsal streams during visuo-spatial working memory tasks in typical development (Ciesielski et al., 2006) and ASD (Vogan et al., 2014). In addition to activation differences in the parietal lobe, this study also observed a developmental impairment of task-load modulation in the right fusiform gyrus in children with ASD, compared to TD children. Both the dorsal and ventral streams are expected to activate and increase in activity as CMT presents a gradually increasing load of colours to be located and categorized. Findings suggest that as TD children mature, their dorsal and ventral neural pathways interact to modulate activity with increasing visuo-spatial demands, whereas children with ASD do not demonstrate this functional integration of networks. It is worth noting that dorsal stream deficits have been postulated in ASD (Pellicano and Gibson, 2008; Spencer et al., 2000) and other developmental disorders (Grinter et al., 2010). More recent literature highlights a developmental shift from early maturing dorsal (i.e. parietal) networks to ventral networks (i.e. fusiform) for working memory function in typical populations (Ciesielski et al., 2006) with the development of semantic/categorical processing across age. We found evidence of impaired function in both dorsal and ventral streams in ASD that persisted over the age range studied.

Another important finding of this study is the differential activity observed in areas of the DMN in response to increasing load over time between children with and without ASD, including the parahippocampal gyrus and ventromedial prefrontal cortex. This study found that children with ASD showed significantly greater load-dependent deactivation with age in DMN regions than TD children (i.e. deactivation in DMN as a function of difficulty intensified across age in children with ASD, more so than the TD group). Functional changes of the DMN are likely not observed in TD individuals because this group demonstrated adequate suppression of the DMN at baseline. As such, children with ASD are possibly showing delayed maturation of the DMN. Exploratory analysis comparing BOLD activation of the D6 versus D3 contrast of the TD group at baseline to the ASD group at follow-up revealed that there were no regions of significant differences, which supports the hypothesis of developmental delay. The DMN is a network of brain regions which are active during rest and suppressed (i.e. BOLD signal reduces) during cognitively demanding tasks (Raichle, 2015; Whitfield-Gabrieli and Ford, 2012) for more efficient cognitive processing, and thus DMN regions are expected to deactivate with increasing task difficulty during CMT. Previous studies have observed inadequate DMN suppression in response to working memory load increase in children and adolescents with ASD compared to the TD group (Rahko et al., 2016). Similarly, a number of neurodevelopmental disorders, including ASD, have been linked to abnormal or diminished DMN function in youth during rest (Assaf et al., 2010) and tasks (Fisher and Happé, 2005). Our study extends this literature, providing a longitudinal examination of task-related DMN function and suggests that this previously documented DMN impairment in children with ASD may become less marked with age, as they showed an emerging ability to suppress DMN areas with increasing task demands.

There were some limitations of this study to consider. To ensure that the observed neural activation occurred in response to the task, we only included individuals who were able to perform adequately; in doing so, our sample is not representative of lower functioning children with ASD (i.e. ASD with an intellectual disability). Future fMRI studies are required to better understand neural patterns during working memory function in individuals with ASD with intellectual deficits who often have very different developmental social, cognitive and academic outcomes; in this case, the use of simpler task may be adequate. Second, the results are limited by the small sample size, due to assessing a generally complex population and the longitudinal nature of the study design, which made our sample vulnerable to attrition. One common reason for ‘drop-out’ from baseline to follow-up in our sample was the high propensity of adolescents to undergo orthodontic work (i.e. metal braces), which is a contraindication for MRI scanning. Also, four participants were on medication at one time point in the study, and for ethical reasons we did not ask them to withhold their medications. Although medication did not have an impact on imaging findings in our baseline study of these data (Vogan et al., 2014), it is important to take this into consideration when interpreting the results of this study. Third, we utilized parametric statistical methods for clusterwise inferencing, which may increase false-positive rates (Eklund et al., 2016), and thus caution should be taken when interpreting the results. However, FSL’s FLAME-1 is known to reduce familywise error rates and returns only marginally higher false discovery rate compared to non-parametric methods (Ekland et al., 2016). Furthermore, as outlined in Table 2, the corrected p values of our clusters are far lower than the threshold required for significance (i.e. p < 8.29 × 10⁻⁶), with the exception of the right superior parietal lobule. Therefore, the acceptably conservative results provided by FLAME-1 compared to other neuroimaging analysis tools in combination with the extremely strong statistical evidence of our data lead us to conclude that concerns raised regarding parametric analysis of fMRI data do not compromise our findings. Finally, this study did not examine the correlational links between behavioural and imaging results; given the complexity of the experimental design and statistical analyses, we focussed on neuroimaging findings. Future research is needed to understand the brain–behaviour links underlying working memory processing. Overall, this is the first study of its kind; the repeated measures study design over a 2-year period in the same individuals adds considerable power, and our study makes a novel and valuable contribution to our knowledge of the neural systems underlying ASD and their development.

Conclusion and future directions

This study provides a clear neurodevelopmental profile of working memory impairment and abilities in ASD. Overall, the current results suggest inadequate modulation of neural activity during increased working memory load in the parietal cortex and fusiform gyrus in children with ASD that shows no significant maturation into adolescence, in contrast to TD peers. In addition, as observed by increased load-dependent suppression of DMN activity across time relative to typical children, our results may suggest that the previously reported DMN impairment in ASD may become less marked with age. The biological mechanisms underlying working memory development in ASD during this period of maturation reflect persistent impairment over time in the ability to modulate brain activity with progressively more complex cognitive demands. Children with ASD might, therefore, benefit from early intervention and accommodations to support working memory before the demands of their social and academic environment increase significantly in adolescence. This may include comprehensive instruction of sophisticated ‘chunking’ (i.e. organizational) strategies to facilitate enhanced recall of complex incoming information (Cowan, 2001) and applying various accommodations (e.g. reducing incoming information, simplifying instruction, increasing processing time, positive reinforcement) that support working memory capacity to various ASD interventions. Previous studies exploring working memory interventions for children with ASD have demonstrated mixed success (de Vries et al., 2015; Fisher and Happé, 2005; Solomon et al., 2004), and thus future research in this area is crucial. Furthermore, future fMRI studies should distinguish neural function during working memory processing beyond adolescence into adulthood and should also include lower functioning individuals with ASD, as neural underpinnings may vary substantially across the autism spectrum. It will also be particularly important for future work to explore the link between neurodevelopmental impairments underlying working memory processing and autistic symptomology and/or later developmental outcomes. Last, comparisons of neurodevelopment related to visuo-spatial working memory to other atypical populations who share similar cognitive but different clinical profiles, such as ADHD (Happé et al., 2006), will elucidate neural patterns that are unique to ASD, and is an important avenue for future studies.

Supplemental Material

AUT766572_Lay_Abstract – Supplemental material for Functional changes during visuo-spatial working memory in autism spectrum disorder: 2-year longitudinal functional magnetic resonance imaging study

Supplemental material, AUT766572_Lay_Abstract for Functional changes during visuo-spatial working memory in autism spectrum disorder: 2-year longitudinal functional magnetic resonance imaging study by Vanessa M Vogan, Benjamin R Morgan, Mary Lou Smith and Margot J Taylor in Autism

Supplemental Material

AUT766572_Supplementary_material – Supplemental material for Functional changes during visuo-spatial working memory in autism spectrum disorder: 2-year longitudinal functional magnetic resonance imaging study

Supplemental material, AUT766572_Supplementary_material for Functional changes during visuo-spatial working memory in autism spectrum disorder: 2-year longitudinal functional magnetic resonance imaging study by Vanessa M Vogan, Benjamin R Morgan, Mary Lou Smith and Margot J Taylor in Autism

Footnotes

Funding

This research was funded by Canadian Institutes of Health Research (CIHR; MOP-106582). Vanessa Vogan was supported through studentships, in part, by the Matching Funds Program, Hospital for Sick Children Foundation Student Scholarship Program and the Ontario Graduate Scholarship (OGS).

References

Alloway

(2009) Working memory, but not IQ, predicts subsequent learning in children with learning difficulties. European Journal of Psychological Assessment 25: 92–98.

Alloway

Seed

Tewolde

(2016) An investigation of cognitive overlap in working memory profiles in children with developmental disorders. International Journal of Educational Research 75: 1–6.

American Psychiatric Association (APA) (2013) Diagnostic and Statistical Manual of Mental Disorders. 5th edn. Arlington, VA: APA.

Andersen

Skogli

Hovik

et al . (2015) Working memory arrest in children with high-functioning autism compared to children with attention-deficit/hyperactivity disorder: results from a 2-year longitudinal study. Autism 19: 443–450.

Arsalidou

Pascual-Leone

Johnson

(2010) Misleading cues improve developmental assessment of working memory capacity: the color matching tasks. Cognitive Development 25: 262–277.

Assaf

Jagannathan

Calhoun

et al . (2010) Abnormal functional connectivity of default mode sub-networks in autism spectrum disorder patients. Neuroimage 53: 247–256.

Baddeley

Della Sala

Papagno

et al . (1997) Dual-task performance in dysexecutive and nondysexecutive patients with a frontal lesion. Neuropsychology 11: 187–194.

Barendse

Hendriks

Jansen

et al . (2013) Working memory deficits in high-functioning adolescents with autism spectrum disorders: neuropsychological and neuroimaging correlates. Journal of Neurodevelopmental Disorders 5: 14.

Baron-Cohen

Tager-Flusberg

Cohen

(2000) Understanding Other Minds: Perspectives from Develop-mental Cognitive Neuroscience. Oxford: Oxford University Press.

10.

Boucher

Bowler

(2008) Memory in Autism Theory and Evidence. New York: Cambridge University Press.

11.

Ciesielski

Lesnik

Savoy

et al . (2006) Developmental neural networks in children performing a Categorical N-Back Task. Neuroimage 33: 980–990.

12.

Cowan

(2001) The magical number 4 in short-term memory: a reconsideration of mental storage capacity. Behavioral and Brain Sciences 24: 87–185.

13.

Cox

(1996) AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Computer Biomedical Research 29: 162–173.

14.

de Vries

Prins

PJM

Schmand

et al . (2015) Working memory and cognitive flexibility-training for children with an autism spectrum disorder: a randomized controlled trial. Journal of Child Psychology and Psychiatry 56: 566–576.

15.

Dennis

Agostino

Roncadin

et al . (2009) Theory of mind depends on domain-general executive functions of working memory and cognitive inhibition in children with traumatic brain injury. Journal of Clinical and Experimental Neuropsychology 31: 835–847.

16.

Eklund

Nicholas

Knutsson

(2016) Cluster failure: why fMRI inferences for spatial extend have inflated false-positive rates. Proceedings of the National Academy of Sciences 113: 7900–7905.

17.

Engle

Tuholski

Laughlin

et al . (1999) Working memory, short-term memory, and general fluid intelligence: a latent-variable approach. Journal of Experimental Psychology: General 128: 309–331.

18.

Fisher

Happé

(2005) A training study of theory of mind and executive function in children with autistic spectrum disorders. Journal of Autism and Developmental Disorders 35: 757–771.

19.

Gilotty

Kenworthy

Sirian

et al . (2002) Adaptive skills and executive function in autism spectrum disorders. Child Neuropsychology 8: 241–248.

20.

Greene

Braet

Johnson

et al . (2008) Imaging the genetics of executive function. Biological Psychology 79: 30–42.

21.

Grinter

Maybery

Badcock

(2010) Vision in developmental disorders: is there a dorsal stream deficit? Brain Research Bulletin 82: 147–160.

22.

Happé

(2005) The weak central coherence account of autism. Handbook of Autism and Pervasive Developmental Disorders 3: 640–649.

23.

Happé

Booth

Charlton

et al . (2006) Executive function deficits in autism spectrum disorders and attention-deficit/hyperactivity disorder: examining profiles across domains and ages. Brain Cognition 61: 25–39.

24.

Happé

Ronald

Plomin

(2006) Time to give up on a single explanation for autism. Nature Neuroscience 9: 1218–1220.

25.

Hill

(2004a) Evaluating the theory of executive dysfunction in autism. Developmental Review 24: 189–233.

26.

Hill

(2004b) Executive dysfunction in autism. Trends in Cognitive Sciences 8: 26–32.

27.

Joseph

(1999) Neuropsychological frameworks for understanding autism. International Review of Psychiatry 11: 309–324.

28.

Kenworthy

Black

Harrison

et al . (2009) Are executive control functions related to autism symptoms in high-functioning children? Child Neuropsychology 15: 425–440.

29.

Kenworthy

Yerys

Anthony

et al . (2008) Understanding executive control in autism spectrum disorders in the lab and in the real world. Neuropsychological Review 18: 320–338.

30.

Killany

Moore

Rehbein

et al . (2005) Memory and executive functions in autism. In: Bauman

Kemper

(eds) The Neurobiology of Autism. 2nd edn. Baltimore, MD: Johns Hopkins University Press, 59–64.

31.

Koolen

Vissers

Egger

et al . (2014) How stimulus and task complexity affect monitoring in high-functioning adults with autism spectrum disorder. Journal of Autism and Developmental Disorders 44: 2499–2513.

32.

Koshino

Carpenter

Minshew

et al . (2005) Functional connectivity in an fMRI working memory task in high functioning autism. Neuroimage 24: 810–821.

33.

Leung

Vogan

Powell

et al . (2016) The role of executive function in social impairment in autism spectrum disorder. Child Neuropsychology 22: 226–344.

34.

Lord

Risi

Lambrecht

et al . (2000) The autism diagnostic observation schedule-generic: a standard measure of social and communication deficits associated with the spectrum of autism. Journal of Autism and Developmental Disorders 30: 205–223.

35.

Luna

Doll

Hegedus

et al . (2007) Maturation of executive function in autism. Biological Psychiatry 61: 474–481.

36.

Luna

Minshew

Garver

et al . (2002) Neocortical system abnormalities in autism: an fMRI study of spatial working memory. Neurology 59: 834–840.

37.

Macoveanu

Klingberg

Tegnér

(2006) A biophysical model of multiple-item working memory: a computational and neuroimaging study. Neuroscience 141: 1611–1618.

38.

Markram

(2010) The intense world theory – a unifying theory of the neurobiology of autism. Frontiers of Human Neuroscience 4: 1–29.

39.

Minshew

Goldstein

(2001) The pattern of intact and impaired memory functions in autism. Journal of Child Psychology and Psychiatry 42: 1095–1101.

40.

Mottron

Dawson

Soulières

et al . (2006) Enhanced perceptual functioning in autism: an update, and eight principles of autistic perception. Journal of Autism and Developmental Disorders 36: 27–43.

41.

Olesen

Macoveanu

Tegnér

et al . (2007) Brain activity related to working memory and distraction in children and adults. Cerebral Cortex 17: 1047–1054.

42.

Ozonoff

Strayer

(2001) Further evidence of intact working memory in autism. Journal of Autism and Developmental Disorders 31: 257–263.

43.

Paus

(2005) Mapping brain maturation and cognitive development during adolescence. Trends in Cognitive Science 9: 60–68.

44.

Pellicano

(2012) The development of executive function in autism. Autism Research and Treatment 2012: 146132.

45.

Pellicano

Gibson

(2008) Investigating the functional integrity of the dorsal visual pathway in autism and dyslexia. Neuropsychologia 46: 2593–2596.

46.

Pickering

Gathercole

(2001) Working Memory Test Battery for Children. London: Pearson Assessment.

47.

Pisella

(2016) Visual perception is dependent on visuospatial working memory and thus on the posterior parietal cortex. Annals of Physical Rehabilitation Medicine 60: 141–147.

48.

Rahko

Vuontela

Carlson

et al . (2016) Attention and working memory in adolescents with autism spectrum disorder: a functional MRI study. Child Psychiatry & Human Development 4: 503–517.

49.

Raichle

(2015) The brain’s default mode network. Annual Review of Neuroscience 38: 433–447.

50.

Richmond

Thorpe

Berryhill

et al . (2013) Individual differences in autistic trait load in the general population predict visual working memory performance. Quarterly Journal of Experimental Psychology 66: 1182–1195.

51.

Rosenthal

Wallace

Lawson

et al . (2013) Impairments in real-world executive function increase from childhood to adolescence in autism spectrum disorders. Neuropsychology 27: 13–18.

52.

Russo

Flanagan

Iarocci

et al . (2007) Deconstructing executive deficits among persons with autism: implications for cognitive neuroscience. Brain Cognition 65: 77–86.

53.

Silk

Rinehart

Bradshaw

et al . (2006) Visuospatial processing and the function of prefrontal-parietal networks in autism spectrum disorders: a functional MRI study. American Journal of Psychiatry 163: 1440–1443.

54.

Solomon

Goodlin-Jones

Anders

(2004) A social adjustment enhancement intervention for high functioning autism, Asperger’s syndrome, and pervasive developmental disorder NOS. Journal of Autism and Developmental Disorders 34: 649–668.

55.

Sowell

Thompson

Leonard

et al . (2004) Longitudinal mapping of cortical thickness and brain growth in normal children. Journal of Neuroscience 24: 8223–8231.

56.

Sowell

Thompson

Tessner

et al . (2001) Mapping continued brain growth and grey matter density reduction in dorsal frontal cortex: inverse relationships during postadolescent brain maturation. Journal of Neuroscience 15: 8819–8829.

57.

Spencer

O’Brien

Riggs

et al . (2000) Motion processing in autism: evidence for a dorsal stream deficiency. Cognitive Neuroscience and Neuropsychology 11: 2765–2766.

58.

Spencer-Smith

Ritter

Mürner-Lavanchy

et al . (2013) Age, sex, and performance influence the visuospatial working memory network in childhood. Developmental Neuropsychology 38: 236–255.

59.

Steele

Minshew

Luna

et al . (2007) Spatial working memory deficits in autism. Journal of Autism and Developmental Disorders 37: 605–612.

60.

Ungerleider

Mishkin

(1982) Two cortical visual systems. In: Ingle

Goodale

Mansfield

RJW

(eds) Analysis of Visual Behavior. Cambridge, MA: The MIT Press, 549–586.

61.

Verté

Geurts

Roeyers

et al . (2006) The relationship of working memory, inhibition, and response variability in child psychopathology. Journal of Neuroscience Methods 151: 5–14.

62.

Vogan

Morgan

Lee

et al . (2014) The neural correlates of visuospatial working memory in children with autism spectrum disorder: effects of cognitive load. Journal of Neurodevelopmental Disorders 6: 1.

63.

Vogan

Morgan

Powell

et al . (2016) The neurodevelopmental differences of increasing verbal working memory demand in children and adults. Developmental Cognitive Neuroscience 17: 19–27.

64.

Wechsler

(1999) Wechsler Abbreviated Scale of Intelligence (WASI). San Antonio, TX: Psychological Corporation.

65.

Whitfield-Gabrieli

Ford

(2012) Default mode network activity and connectivity in psychopathology. Annual Review of Clinical Psychology 8: 49–76.

66.

Williams

Goldstein

Carpenter

et al . (2005) Verbal and spatial working memory in autism. Journal of Autism and Developmental Disorders 35: 747.

67.

Woolrich

Jbabdi

Patenaude

et al . (2009) Bayesian analysis of neuroimaging data in FSL. Neuroimage 45: S173–S186.

68.

Worsley

(2001) Statistical analysis of activation images. In: Jezzard

Matthews

Smith

(eds) Functional MRI: An Introduction to Methods. Oxford: Oxford University Press, 251–270.

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

0.43 MB

0.27 MB