The Contribution of Increased Gamma Band Connectivity to Visual Non-Verbal Reasoning in Autistic Children: A MEG Study

Some individuals with autism spectrum (AS) perform better on visual reasoning tasks than would be predicted by their general cognitive performance. In individuals with AS, mechanisms in the brain’s visual area that underlie visual processing play a more prominent role in visual reasoning tasks than they do in normal individuals. In addition, increased connectivity with the visual area is thought to be one of the neural bases of autistic visual cognitive abilities. However, the contribution of such brain connectivity to visual cognitive abilities is not well understood, particularly in children. In this study, we investigated how functional connectivity between the visual areas and higher-order regions, which is reflected by alpha, beta and gamma band oscillations, contributes to the performance of visual reasoning tasks in typically developing (TD) (n = 18) children and AS children (n = 18). Brain activity was measured using a custom child-sized magneto-encephalograph. Imaginary coherence analysis was used as a proxy to estimate the functional connectivity between the occipital and other areas of the brain. Stronger connectivity from the occipital area, as evidenced by higher imaginary coherence in the gamma band, was associated with higher performance in the AS children only. We observed no significant correlation between the alpha or beta bands imaginary coherence and performance in the both groups. Alpha and beta bands reflect top-down pathways, while gamma band oscillations reflect a bottom-up influence. Therefore, our results suggest that visual reasoning in AS children is at least partially based on an enhanced reliance on visual perception and increased bottom-up connectivity from the visual areas.

Some individuals with autism spectrum (AS) perform better on visual reasoning tasks than would be predicted by their general cognitive performance. In individuals with AS, mechanisms in the brain's visual area that underlie visual processing play a more prominent role in visual reasoning tasks than they do in normal individuals. In addition, increased connectivity with the visual area is thought to be one of the neural bases of autistic visual cognitive abilities. However, the contribution of such brain connectivity to visual cognitive abilities is not well understood, particularly in children. In this study, we investigated how functional connectivity between the visual areas and higher-order regions, which is reflected by alpha, beta and gamma band oscillations, contributes to the performance of visual reasoning tasks in typically developing (TD) (n = 18) children and AS children (n = 18). Brain activity was measured using a custom child-sized magneto-encephalograph. Imaginary coherence analysis was used as a proxy to estimate the functional connectivity between the occipital and other areas of the brain. Stronger connectivity from the occipital area, as evidenced by higher imaginary coherence in the gamma band, was associated with higher performance in the AS children only. We observed no significant correlation between the alpha or beta bands imaginary coherence and performance in the both groups. Alpha and beta bands reflect top-down pathways, while gamma band oscillations reflect a bottom-up influence. Therefore, our results suggest that visual reasoning in AS children is at least partially based on an enhanced reliance on visual perception and increased bottom-up connectivity from the visual areas.

Introduction
interaction" [35,36]. Therefore, in the present study, we employed ImCoh (instead of conventional coherence) to evaluate inter-sensor connectivity, which included evaluating the sensors within short distances, as well as those at long distances.
The present study investigates cortical network connectivity in 4 -10-year-old children during passive visual information processing using custom child-sized MEG system (Fig 1A) in which sensors were located as close to the brain as possible for optimal recording in young children [37]. Alpha, beta, and gamma band oscillation were used as proxies for bottom-up and top-down information processing to assess their contribution to visual reasoning (e.g., mental rotation and matrix analogies tasks) in AS children [24].

Materials and Methods Participants
All participants were recruited from public nursery schools in Kanazawa city, Kanazawa University Hospital, and prefectural hospitals in Toyama. Children were diagnosed by a clinical psychiatrist and a clinical psychologist with more than five years of experience in AS using the Autism Diagnostic Observational Schedule-Generic (ADOS) [38], the Diagnostic Interview for Social and Communication Disorders (DISCO) [39], and the DSM-IV [40] criteria at the time the children participated in this study. AS children were included in this study when they fulfilled the diagnosis of childhood autism (66.7% of all included subjects), atypical autism (27.8%), or Asperger's syndrome (5.6%) by the DISCO. One patient did not meet the ADOS criteria for the autism spectrum; however, we included her because she fulfilled the criteria of atypical autism by the DISCO. The exclusion criteria included known hearing loss or intellectual disability. The children's cognitive skills were assessed by the Japanese version of the Kaufman Assessment Battery for Children (K-ABC) [41]. We excluded those subjects whose standardised score of the K-ABC (the average value of cognitive and achievement scales) was below 70 (i.e., 2 standard deviations below the mean).
Twenty-one AS children initially participated in the experiment. We excluded three children, one because their standardised score on the K-ABC scales was below 70 (i.e., intellectual disability), one because he could not follow our instructions, and one who had a reaction time over three standard deviations from the mean reaction time. We finally used data from 18 children for further analysis. The participants had a mean age of 83.9 months (range: 59-111). The mean standardised K-ABC score was 102.6 (range: 76.5-130.0). The TD group was matched to the AS group for chronological age, gender, and standardised K-ABC score ( Table 1). The performances (raw score) on all Kaufman Assessment Battery (K-ABC) subtests of the children with ASD and the TD young children are shown in the S1 Fig. All TD children were native Japanese and had no prior or current developmental, learning, or behavioral problems, as was reported on a questionnaire that was completed by their parents. All the children had normal visual acuity according to their available medical records, i.e., they had never been noted to have any problems with visual acuity at three-year-old routine health checkup, and they displayed no problems with visual acuity in their daily lives. Written informed consent (by the parents) was obtained prior to enrolment in the study. The Ethics Committee of Kanazawa University Hospital approved the methods and procedures, all of which were in accordance with the Declaration of Helsinki.

Visual reasoning task
The scores of the Matrix Analogies subtest in the K-ABC [41] represent visual reasoning ability. The Matrix Analogies test requires the selection of a picture or design that best completes a visual scene or pattern. The examiner conveys the instructions through gestures, and the child responds with hand movements as in Raven's Progressive Matrices test [42], a nonverbal measurement test.

Mental rotation task
Individuals' 3-D mental rotation ability was measured using part of the Purdue Spatial Visualization Tests: Visualization of Rotations (PSVT: R) [43][44][45]. We employed the first 10 questions of the mental rotation task. Participants were required to study how the object on the top line is rotated ( The experiment consisted of two exercises and 10 trials with a ca. 3-10 second break between the pictures based on the participant's desired timing. The period from the time the second picture was presented to the time they answered was defined as the reaction time. We calculated a value (efficiency score) to control for speed-accuracy trade-offs in the mental rotation outcome data by dividing the number of correct answers by the mean response time (RT) of all responses for each subject. For example, an average RT of 10 sec and six correct answers would yield an efficiency score value of 0.6 (/sec). A higher score indicates better performance. The visual tasks were generated using the software package SuperLab 4.0 (Cedrus, San Pedro, CA, USA). The number of correct answers and RT was measured and recorded by an examiner pressing a button. Magnetoencephalography recordings MEG data were recorded using a 151-channel SQUID (Superconducting Quantum Interference Device), whole-head coaxial gradiometer MEG system for children (PQ 1151R; Yokogawa/KIT, Kanazawa, Japan) ( Fig 1A) in a magnetically shielded room (Daido Steel, Nagoya, Japan). Staff members made contact with the children and played with them along with the parent(s)/caretaker(s). We confirmed the position of the head to be in the center of the MEG helmet by measuring the magnetic fields after passing currents through coils that were attached at three locations on the surface of the head as fiduciary points with respect to the landmarks (bilateral mastoid processes and nasion). MEG data were acquired using a sampling rate of 1000 Hz and filtered using a 200 Hz low-pass filter. The MEG session lasted six min. During MEG recording, one staff member (author YY) escorted each participant into the shielded room, which was decorated with colorful pictures of Japanese (cartoon) characters. During the MEG recording, the children passively viewed a video program projected onto a screen, selected according to the preference of each participant. This type of visual stimulation is an effective way to keep the young children's attention focused on the screen and their heads still during the MEG recording. It enabled us to obtain enough artifact-free segments (80-82 sec periods in total) for connectivity analysis. None of the participants endured high emotional tension or any other kind of discomfort during the measurements based on their evaluation.

Data analysis
Performance on the mental rotation task and the Matrix Analogies subtest in the TD and AS groups. The number of correct answers and reaction times between TD and AS children were compared using unpaired, two-tailed t-tests. The raw scores on the Matrix Analogies subtest in the TD and AS groups were also compared using unpaired, two-tailed t-tests. A Pearson's correlation was used to determine significant correlations between performance on the mental rotation task (the number of correct answers/reaction time of all responses) and the raw score on the Matrix Analogies subtest of the K-ABC in the TD and AS groups. The significance level was set at P < 0.05. Imaginary coherence analysis with MEG data. We employed ImCoh values between the occipital and other areas during passive exposure to a video program as a physiological index of the brain network. Offline analysis of the MEG data was performed using Brain Vision analyzer 2 (Brain Products GmbH, Gilching, Germany) software. In 151-channel SQUI, we excluded four sensors because they were malfunctioning for some subjects. We finally used 147 sensors for further analysis. The MEG data were resampled at 500 Hz. The data were split into 2 sec segments. Artifact-free segments were selected based on visual inspection. The process of eliminating contaminated data was performed blindly. Eighty sec periods were sufficiently long to obtain reproducible results for resting state power analyses [46][47][48]. In the present study, MEG spectra were calculated for 40 to 41 artifact-free segments (80-82 sec period) using fast Fourier transformation (FFT) with a spectral resolution of 0.5 Hz. Coherence (Cross-Spectrum/Autospectrum) is usually calculated following a Fourier transformation using the formula: In the second formula, totaling is carried out via segment number i. Calculation of the average also relates to segments with a fixed frequency, f, and a fixed channel, c. Using this methodology, values between 0 and 1 are obtained for each frequency and each channel. We employed imaginary coherence (ImCoh) analyses which were calculated using the square of the imaginary part of the cross-spectrum (imagCS) instead of the cross-spectrum for the coherence analysis. The formula is as follows: Artifact-free (e.g., muscle activities) segments were chosen by visual inspection. We selected two seed sensors corresponding to the right and left occipital area to minimize muscular noise contamination (Fig 2B and 2C). We calculated the ImCoh value between the seed sensor (selected for each right and left hemisphere) and the other 146 sensors (Fig 1B). We focused on the alpha (7-12 Hz), beta (13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29) and gamma  band ImCoh asa proxy for bottom-up and top-down functional connectivity between the occipital visual area and the other brain areas [24]. As shown in the S3 Fig, we determined the alpha range to be 7-12 Hz because the alpha peak frequencies that were observed in the current study were within 7-11 Hz. The frequency of the power supply for the MEG system was 60 Hz.
Statistical Analysis. Unpaired t tests were performed to compare the AS and TD groups for z-transformed ImCoh in the gamma band. A Pearson's correlation was used to determine significant correlations between the z-transformed ImCoh and performance on the mental Isocontour maps of the magnetic field after pattern reversal visual stimuli. The magnetic field strength is indicated by color, varying from blue (flux-in) to red (flux-out)) at the response peak 75 and 175 msec after pattern reversal. Two seed sensors are indicated by gray circles and the other 146 sensors are indicated by dots. In two seed sensors, signals from visual cortices (i.e., visually evoked response) were optimally recorded. (c) Contamination of muscular activity in resting state brain activity. Topography of gamma band (30-58 Hz) power during rest with (right) and without (left) visually confirmed muscle activity. Sensors located in the ventral area (right) are vulnerable to muscular noise. (d) Vulnerable areas to muscular noise are close to the muscles of head and neck (e.g. temporal muscles). Two seed sensors indicated by gray circles are located outside of these vulnerable areas to muscular noise. doi:10.1371/journal.pone.0163133.g002 rotation task (the number of correct answers/reaction time of all responses) or the raw score on the Matrix Analogies subtest of the K-ABC. We choose a significance level of 0.00034 (= 0.05 / 146) because of the multiple comparison of 146 connections between a seed sensor and other sensors. When significance was found with this conservative level, as a complementary approach, an alpha level of 0.05 was also used, with the risk of increasing the chance of Type I error.

Imaginary coherence analysis with MEG data in TD and AS children
There was no significant difference in alpha band (7)(8)(9)(10)(11)(12), beta band (13-29 Hz), or gamma band (30-58 Hz) ImCoh for any sensor pair between TD and AS children, using a conservative alpha level of 0.00034 (S4 Fig). Performance on a mental rotation task in TD and AS children There was no difference in the number of correct answers between TD and AS children (t = 0.18, P > 0.05; Fig 3A) nor in the efficiency index (i.e., number of correct answers/reaction time) (t = 0.38, P > 0.05; Fig 3D), whereas the response time was shorter for AS children than TD children both for all responses (t = 2.11, P = 0.042; Fig 3B) and for correct responses (t = 2.22, P = 0.034; Fig 3C). AS children responded faster than TD children for the mental rotation task.

Performance on "Matrix Analogies" in TD and AS children
The mean (± SD) raw score of the "Matrix Analogies" K-ABC subtest was 8.39 (± 5.16) and 9.39 (± 4.37) for TD and AS children, respectively. There was no difference for this score between TD and AS children (t = 0.63, P > 0.05).
Correlation between performance on a mental rotation task and on a "Matrix Analogies" subtest There was a significant correlation between performance on the mental rotation task (the number of correct answers/reaction time of all responses) and the raw score on the Matrix Analogies subtest of the K-ABC in AS children (r = 0.844, P < 0.001), whereas there was no significant correlation in TD children (r = 0.201, P > 0.05).

Performance on a mental rotation task and imaginary coherence in TD and AS children
There was no significant correlation between task performance and ImCoh for the alpha band (7-12 Hz) or beta band (13-29 Hz) for any sensor pair either in TD or AS children, using a conservative alpha level of 0.00034. For the gamma band (30-58 Hz), there was a significant positive correlation for one sensor pair in AS children using an alpha level of 0.00034 (Fig 4,  lower row; S5A Fig, r = 0.756, P = 0.0001), whereas there was no significant correlation in TD children. Using an alpha level of 0.05, there was a correlation between the performance of the mental rotation task and z-transformed gamma band ImCoh in AS children for 4 of 146 sensor pairs and 11 of 146 sensor pairs from the left and right occipital areas, respectively (Fig 4, lower  row). The number of sensor pairs for which significant correlations were found for the ImCoh from the right occipital area exceeded the number expected by chance (146 Ã 0.05 = 7.3).
We wanted to evaluate the possible existence of a gender and/or age effect on the significant relationship (r = 0.756, P = 0.0001) found in the gamma band ImCoh, so as a complementary analysis, we used multiple linear regression to predict the gamma band ImCoh (i.e., dependent variable) using performance on the mental rotation task, age and gender as predictors (i.e., three independent variables). We used a significance level of P < 0.05. In the multiple regression model, performance on the mental rotation task was a significant predictor of the gamma band ImCoh (n = 18, β = 0.606, P = 0.0063), whereas age (n = 18, β = 0.241, P > 0.05) and gender (n = 18, β = -0.111, P > 0.05) were not statistically significant predictors.

Performance on "Matrix Analogy" and imaginary coherence in TD and AS children
There was no significant correlation between the task performance and ImCoh for the alpha band (7-12 Hz) or beta band (13-29 Hz) for any sensor pair either in TD or AS children, using a conservative alpha level of 0.00034. For the gamma band (30-58 Hz), there was a significant Using an alpha level of 0.05, there was a correlation between the performance of the mental rotation task and z-transformed gamma band ImCoh in AS children, for 15 of 146 sensor pairs and 13 of 146 sensor pairs from the left and right occipital areas, respectively (Fig 5  (lower row)). The number of sensor pairs for which significant correlations were found for the ImCoh from the left and right occipital area exceeded the number expected by chance (146 Ã 0.05 = 7.3).
To evaluate the possible existence of a gender and/or age effect on the significant relationship (r = 0.792, P < 0.0001) found in the gamma band ImCoh, we used multiple linear regression (i.e., same methods as shown in the results of mental rotation task). Performance on the "Matrix Analogy" task was a significant predictor of the gamma band ImCoh (n = 18, β = 0.705, P = 0.0038), whereas age (n = 18, β = 0.055, P > 0.05) and gender (n = 18, β = -0.178, P > 0.05) were not statistically significant predictors.

Discussion
We measured brain activity during passive visual processing and its relation to visuo-spatial performance and visual reasoning using MEG. Our main finding is that the gamma band frequency correlates with visual reasoning ability in AS children only, as measured by two distinct tasks. This suggests that feedforward (but not feedback) functional connectivity may differently contribute to visual reasoning ability in AS children as alpha-beta and gamma band oscillations in visual areas are proxies for top-down and bottom-up influence, respectively.

Bottom-up connectivity and gamma band oscillation
The high temporal resolution of the MEG signal enables the investigation and quantification of various cortical rhythms [15,49] involved in brain activity, the inhibitory or excitatory nature of the neurons involved in the task, and the feedforward or feedback directionality of the connectivity. Recent animal studies demonstrated that the gamma-band is the main neurophysiological index of bottom-up connectivity in the visual cortex [50]. Gamma-mediated inter-areal influences are predominantly bottom-up in monkeys, i.e. directed from V1 to V4 visual areas [21]. Gamma synchronization emerges in V1 supragranular layers, and influences V4 through feedforward projections. In turn, gamma synchronization emerges in V4 supragranular layers and influences further downstream areas. Conversely, the top-down influence from V4 to V1 originates from deep V4 layers, where beta band oscillation is dominant, and is thus mediated to a much lesser extent through the gamma band. Gamma band connectivity is thus thought to reflect the strength of feedforward processing [24]. Our results suggest that this element contributes to a larger extent to visual reasoning tasks in AS children, despite similar feedforward connectivity during visual processing between AS and TD children. These results are consistent with those of another recent MEG study [51], which measured feedforward functional connectivity during somatosensory processing. The authors concluded that increased long-range feedforward connectivity and reduced long-range feedback connectivity are likely characteristic of AS (but see [52]). In our previous study, we found a significantly higher coherence value in children with AS (5-7 years old) between the right occipital-right temporal connection than for TD children, using conventional coherence analysis for the gamma band [32]. In the present study, we failed to demonstrate significant differences in ImCoh for gamma bands between TD and AS children. This may result from the lower statistical power in the present study (i.e., smaller sample size), and/or from differences in the methodology (i.e., coherence and imaginary coherence). In the present study, children with AS who had higher gamma band connectivity also performed better in visual reasoning tasks. This result suggests that the magnitude of feedforward connectivity associated with visual information represents a neurophysiological index of autistic visual strengths [10,53].

Bottom-up connectivity and visual reasoning tasks in AS
Mental rotation is a visual reasoning task. The response time for the mental rotation task was shorter for AS children than TD children, whereas the response accuracy was similar for both groups, consistent with previous studies conducted with children [54] and adults [7,[54][55][56][57]. One previous neuroimaging study demonstrated lower activation of the prefrontal cortex during mental rotation tasks in subjects with AS [7]. This was interpreted as there being a smaller contribution of top-down influence on perceptual processing in AS subjects.
The matrix analogies task, an autistic strength, is a reliable measure of fluid intelligence and general intellectual ability. In the current study, in AS children performance on the "Matrix Analogies" K-ABC subtest was significantly correlated with performance on the mental rotation task. Interestingly, in relation to our findings about performance on a mental rotation task, a previous study demonstrated that when they were solving Raven matrices, AS subjects performed with similar accuracy but had shorter response times than non-autistic controls [6]. Children and adults with AS perform better on matrix analogies tasks (such as RSPM) than would be predicted by their Wechsler intelligence quotient (IQ) [2]. In addition, they perform better than TD children matched for age and Wechsler IQ [58]. During the RSPM test, AS participants showed relatively increased task-related activity in extra-striate areas (BA18), and decreased activity in the lateral prefrontal cortex (BA9) and medial posterior parietal cortex (BA7) [6] than TD participants. Using psychophysiological interaction analyses for fMRI data during an RSPM test, a recent study revealed greater connectivity between the left inferior occipital gyrus and areas in the left superior frontal gyrus, right superior parietal lobe, right middle occipital gyrus, and right inferior temporal gyrus, which increases as a function of reasoning complexity [5] in AS subjects relative to non-autistic participants.
Altogether, these findings suggest that a combination of enhanced feedforward and atypical top-down influences is involved in autistic visual capacity. This is in agreement with the conclusion of a meta-analysis of fMRI studies examining how autistic people process complex visual material [9]. Enhanced activity in the visual associative region is reliably reported in fMRI investigations of visual reasoning and language tasks in AS children. Multiple psychological and neurophysiological studies suggests that increased bottom-up and reduced [59,60], or optional, top-down processing may characterize AS [12,61], and determine some aspects of the AS cognitive profile. An increased feedforward flow of information from visual areas has not only been reported in autistic visual [59,60,62,63] but also somatosensory [64] and auditory [65] processing. Previous studies also reported that the enhanced perceptual abilities in AS [12] and the atypical importance of complex visual perception in autistic intelligence [10] may be, at least partially, supported by higher feedforward connectivity from visual areas. Our results also support that higher bottom-up connectivity from the occipital area is associated with the higher visuo-spatial and nonverbal reasoning performance of children with AS.

Limitations
There are some limitations of this study. First, the differences in measured ImCoh may relate to the different spatiotemporal properties of the video stimuli during the periods corresponding to the selected artifact-free segments. Second, we did not evaluate the degree to which the subjects focused on the TV program that they chose. This limitation may be compensated for by the fact that all participants concentrated exclusively on the TV programs, which was confirmed by a video monitor system. Third, we eliminated any contaminated MEG data, including data that were obtained during obvious head movements. However, differences in fine head movements between the children with ASD and the TD children could have confounded the study results. Fourth, a majority of the children with ASD in the current study were subjects of typical level of intelligence who were able to remain stationary during the MEG measurements; therefore, the findings may not apply to children belonging to other subtypes within the autism spectrum. Fifth, given that young children were examined in the current study, we were unable to obtain structural brain information onto which we could superimpose the coordinate system of the source-estimated MEG signals. However, child-friendly, open-type MRI devices and source-space imaginary coherence analyses with beam-forming methods [35] have become a popular technique to estimate functional connectivity based on MEG/EEG. Using these methods in future studies will enable us to investigate source level brain networks in young children while they are conscious.

Conclusion
Stronger connectivity that originates in the occipital area and that is seen in gamma band oscillations (but not in the alpha and beta bands) was associated with higher performance in AS children. Given that alpha and beta bands reflect top-down pathways, and gamma band oscillations reflect a bottom-up influence, our results suggest that visual reasoning in AS children is, at least partially, based on an enhanced reliance on visual perception and an increased bottomup connectivity from the visual areas. A relatively higher ImCoh was observed for the sensors near the seed sensor relative to the sensors located a long distance from the seed sensor for both the TD children and the AS children (upper and middle row). Tmaps between the TD children (n = 18) and the AS children (n = 18) are in the lower row. There were no significant differences in the gamma band ImCohs for any sensor pair between the TD children and the AS children (i.e., P > 0.00034). ImCoh: imaginary coherence.