The authors have declared that no competing interests exist.
Conceived and designed the experiments: RH TY. Performed the experiments: RH TY HO HW CK TO MT YT. Analyzed the data: RH TY. Contributed reagents/materials/analysis tools: RH TY. Wrote the paper: RH TY. Supervision: AI NK.
Individuals with autism spectrum condition (ASC) are known to excel in some perceptual cognitive tasks, but such developed functions have been often regarded as “islets of abilities” that do not significantly contribute to broader intellectual capacities. However, recent behavioral studies have reported that individuals with ASC have advantages for performing Raven's (Standard) Progressive Matrices (RPM/RSPM), a standard neuropsychological test for general fluid intelligence, raising the possibility that ASC′s cognitive strength can be utilized for more general purposes like novel problem solving. Here, the brain activity of 25 adults with high-functioning ASC and 26 matched normal controls (NC) was measured using functional magnetic resonance imaging (fMRI) to examine neural substrates of geometric reasoning during the engagement of a modified version of the RSPM test. Among the frontal and parietal brain regions involved in fluid intelligence, ASC showed larger activation in the left lateral occipitotemporal cortex (LOTC) during an analytic condition with moderate difficulty than NC. Activation in the left LOTC and ventrolateral prefrontal cortex (VLPFC) increased with task difficulty in NC, whereas such modulation of activity was absent in ASC. Furthermore, functional connectivity analysis revealed a significant reduction of activation coupling between the left inferior parietal cortex and the right anterior prefrontal cortex during both figural and analytic conditions in ASC. These results indicate altered pattern of functional specialization and integration in the neural system for geometric reasoning in ASC, which may explain its atypical cognitive pattern, including performance on the Raven's Matrices test.
The exact nature of autistic cognition and intelligence has been a subject of extensive controversy. Individuals with autism spectrum condition (ASC) sometimes perform better than neurotypical individuals on certain tasks, including visual search and the Block Design and Object Assembly subtests of the Wechsler Intelligence Scale (WAIS)
The RSPM test consists of figures of geometric matrices that can be categorized into two major classes of “figural” and “analytic” items
Neural substrates for general intelligence have been examined in healthy adult individuals by using functional imaging techniques
In the present study, brain activity of high-functioning ASC participants during the RSPM task was measured using fMRI to examine functional properties of their brain network of geometric analogical reasoning. A previous fMRI study reported that adult individuals with ASC showed enhanced RSPM task-related activity in the extrastriate visual cortex and reduced activity in the lateral frontal regions including the precentral gyrus
Twenty-five high-functioning adults with ASC and 26 normal control (NC) adults participated in this study (
The profile of the adult autism spectrum condition (ASC) participants in the present study is shown in red and that of the Asperger adults in a previous study (adapted from Soulieres et al., 2011) in blue. Error bars indicate the standard deviation of mean. INF, Information; SIM, Similarity; ARI, Arithmetic; VOC, Vocabulary; COM, Comprehension; PC, Picture Completion; COD, Digit Symbol-Coding; PA, Picture Arrangement; BD, Block Design; MA, Matrix.
Subject | NC | ASC |
Number | 26(4 female) | 25(3 female) |
Age | 32.2±7.7 | 30.7±7.78 |
Handedness |
86.4±39.9 | 55.7±72.4 |
WAIS full scale IQ |
103.9±11.4 | 106.9±15.9 |
WAIS verbal IQ |
105.5±11.7 | 113.5±15.2 |
WAIS performance IQ |
101.2±12.1 | 97.1±16.7 |
JART estimated IQ |
106.7±9.4 | |
AQ: total score |
16±6.4 | 35.7±5.6 |
Mean±standard deviation.
NC: Normal Control, ASC: Autism Spectrum Condition, WAIS: Wechsler Adult Intelligence Scale, JART: Japanese version of National Adult Reading Test, AQ: Autism Spectrum Quotient.
Assessed by the Edinburgh handedness inventory (Oldfield, 1971).
Data was collected from 21NC and 22 ASC participants.
WAIS- III or -R was performed by 15 NC and 25 ASC participants.
The JART score was collected from all NC participants.
The AQ was collected from 19 NC and 25 ASC participants.
A significant difference between NC and ASC (
ASC participants were recruited from outpatient units of the Karasuyama Hospital, Tokyo, Japan. The inclusion criteria were age between 18 and 50 years, no current use of psychotropic medication, and a formal diagnosis of pervasive developmental disorder (PDD) based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV). The exclusion criteria included a history of electroconvulsive therapy, alcohol or other drug abuse or dependence, or any neurologic illness affecting the central nervous system. PDD was diagnosed by a team of three experienced psychiatrists and one clinical psychologist on the basis of two detailed patient interviews of the patients regarding their development and behavior from infancy through adolescence [(1) developmental history; (2) present illness; (3) past history] and family history. The interviews were conducted independently by one psychiatrist and the clinical psychologist in the team. The patients were also asked to bring suitable informants who had known them in early childhood. At the end of the clinical interview, the psychiatrist diagnosed the patients according to the DSM-IV diagnostic criteria for PDD based on the consensus of the psychiatrists and the clinical psychologist. The diagnostic process required approximately 3 hours. These assessments resulted in all of the participants in the ASC group receiving clinical diagnoses of Asperger's disorder (n = 11), high-functioning autism (n = 11), or pervasive developmental disorder not otherwise specified (n = 3). Two of the psychiatrists and the psychologist confirmed that none of the participants met the DSM-IV criteria for any other psychiatric disorder (e.g., mood disorder, schizophrenia, anxiety disorder, or substance-related disorder). In addition, none of the NC subjects reported any severe medical problem or any neurological or psychiatric history. Moreover, the Mini-International Neuropsychiatric Interview was used to confirm that none of the normal controls met the diagnostic criteria for any psychiatric disorder. All of the participants had normal or corrected-to-normal vision. All the procedures of this study, including the method of obtaining consent, were approved by the Ethics Committee of the Faculty of Medicine of Showa University in accordance with the Declaration of Helsinki. Written informed consent was obtained from each participant after a full explanation of the purpose of the study was provided. Because participants were all high-functioning (IQ>80) adults without any other comorbidities, they were able to fully understand the content and nature of the study. Guardians' oral consents were not either documented or recorded because every patient was judged to possess the full ability to give consent on his/her own by his/her primary doctor (TY, HO, WH, or NK). Any concern regarding the possibility of reduced capacity to consent on his/her own was not voiced by either the ethics committee or patients' primary doctors. Every participant was assigned arbitrary identification number for this study such that all the data, including imaging, behavioral, and demographic data, was analyzed anonymously.
We measured brain activity during “figural,” “easy analytic,” and “difficult analytic” conditions using selected items of the RSPM set (
(A) “Figural” condition, (B) “easy analytic” condition, and (C) “difficult analytic” condition. Participants were asked to select one out of the six alternatives by pressing a button.
The RSPM task was performed in a block design. Blocks for easy and difficult analytic items alternated twice while blocks for figural served as a baseline, which resulted in nine blocks in total. Participants were instructed to select an answer among six response alternatives by pressing a button and were encouraged to deliberate thoroughly until they were reasonably certain about their answers. Duration of each task block was 30 seconds except for the last baseline block of 32 seconds. Participants proceeded in a self-paced manner during each block. Because the duration of each block was fixed, the last trial was usually terminated without button pressing. During a single fMRI session, NC and ASC individuals performed 27.84±8.27 (mean±standard deviation) and 27.70±8.53 trials for the figural condition, 6.27±2.42 and 6.88±2.05 trials for the easy analytic condition, and 2.61±1.60 and 3.45±2.18 trials for the difficult analytic condition, respectively. No significant group difference was observed between any of the conditions (
The RSPM figures were projected on an fMRI-compatible screen that was placed at the participant's feet. The participants were immobilized with padding and watched the screen through a mirror mounted on the headcoil. The subjects' responses were collected via a fiber-optic button box that was held by the participant while in the magnet. The button box was connected to an interface for the PC that allowed for the collection of response time and accuracy data (Current Designs, Philadelphia, PA). The presentation of the stimuli and collection of the responses were performed using Neurobehavioral System's Presentation software (Neurobehavioral Systems, Albany, CA).
Magnetic resonance (MR) images were acquired using a 1.5 T scanner with an 8-channel phased-array head coil (Signa Horizon; General Electric Medical Systems, Milwaukee, WI). Functional data were acquired using an echo planar imaging (EPI) sequence (27 axial slices, voxel size = 3.43×3.43×4 mm, TR = 2000 ms, TE = 40 ms, flip angle = 90°). During a single run, 136 volumes were acquired. A high-resolution structural T1 image was also acquired (gradient-echo; TR = 25 ms, TE = 9.2 ms, matrix size = 256×256×175, flip angle = 90°, voxel size = 0.975×0.975×1.4 mm) to allow for anatomical localization of activation.
Preprocessing and statistics of fMRI time-series data were performed using Statistical Parametric Mapping software (SPM5) (Wellcome Department of Cognitive Neurology, London, UK) running on MATLAB version R2009b (The Mathworks, Inc., Natick, MA). Functional images were realigned, normalized into the standard Montreal Neurological Institute (MNI) template, and smoothed using an 8 mm FWHM Gaussian smoothing kernel. The images were resampled to a resolution of 2×2×2 mm when they were normalized. In a first-level general linear model analysis for individual participants, task-related activation was modeled by boxcar functions for easy and difficult analytic blocks, each of which was convolved with the hemodynamic function. For each participant, the contrast images of easy vs. baseline, difficult vs. baseline, and analytic (easy+difficult) vs. baseline were generated. The contrast images were fed into a random-effects second-level analysis. Statistical threshold was set to
The comparison of analytic (easy+difficult) vs. baseline produced reliable clusters of activation in regions of the network of general fluid intelligence consisting of the bilateral VLPFC, premotor cortex (PMC), and IPC, as well as the area around the left LOTC
Motivated by the P-FIT model of intelligence
When any significant voxel cluster was found in the group comparison, we next examined activation coupling at each of figural (baseline), easy analytic, and difficult analytic conditions separately. For each condition, we first created r- and z-maps in each participant following the same procedures described above except that the number of data points was 60 (12 points×5 blocks), 24 (12 points×2 blocks), and 24 (12 points×2 blocks) for figural (baseline), easy analytical, and difficult analytical conditions, respectively. The first three data points were removed to exclude signal fluctuations due to block transition
The mean number of correct answers and response times (RT) are shown in
NC | ASC | |
Baseline | ||
Number of correct answers | 24.08±8.40 | 23.92±8.31 |
Response time (s) | 5.02±1.95 | 4.90±1.59 |
Easy analytic | ||
Number of correct answers | 4.35±2.45 | 5.17±2.16 |
Response time (s) | 8.27±3.80 | 7.21±2.45 |
Difficult analytic | ||
Number of correct answers | 0.88±0.95 | 0.83±0.87 |
Response time (s) | 14.74±5.14 | 13.46±5.47 |
Mean±standard deviation.
NC: Normal Control, ASC: Autism Spectrum Condition.
No significant main effect of Group either on the number of correct answers (
In the contrast of easy analytic vs. baseline, significant activation was observed in the bilateral IPC, left premotor cortex (PMC) and VLPFC, and a part of the left LOTC (junction of the left inferior temporal cortex and anterior LOC) in ASC, whereas NC showed no significant brain activation (
(a) Participants with autism spectrum condition (ASC) during the easy analytic condition. Normal control (NC) showed no significant activation during the easy analytic condition. (b) Direct comparison between NC and ASC (NC<ASC) during easy analytic condition. (c) NC during the difficult analytic condition. (d) ASC during the difficult analytic condition. (e) Combined analysis of the two groups during the two analytic conditions. Statistical threshold was set at
Left | Right | |||||||||
Region label | x | y | z | z value | Cluster size | x | y | z | z value | Cluster size |
A: NC No significant voxels | ||||||||||
B: ASC | ||||||||||
LOTC | −60 | −32 | −14 | 4.83 | 523 | |||||
IPC | −48 | −66 | 42 | 4.22 | 556 | 54 | −54 | 50 | 3.86 | 354 |
PMC | −44 | 14 | 54 | 4.17 | 187 | |||||
VLPFC | −42 | 50 | −6 | 4.12 | 106 | |||||
C: NC>ASC | No significant voxels | |||||||||
D: NC<ASC | ||||||||||
LOTC | −52 | −30 | −14 | 3.99 | 85 |
Extent threshold:
NC: Normal Control, ASC: Autism Spectrum Condition, LOTC: lateral occipitotemporal cortex.
IPC: inferior parietal cortex, PMC: premotor cortex, VLPFC: ventrolateral prefrontal cortex.
NC showed no significant brain activation during the easy analytic condition.
No areas showed larger activation for NC than ASC.
Left | Right | |||||||||
Region label | x | y | z | z value | Cluster size | x | y | z | z value | Cluster size |
A: NC | ||||||||||
IPC | −58 | −56 | 42 | 5.18 | 1301 | 52 | −74 | 32 | 5.34 | 720 |
VLPFC | −42 | 48 | −8 | 4.08 | 121 | 22 | 60 | 20 | 3.69 | 223 |
PMC | 24 | 24 | 62 | 3.61 | 79 | |||||
Precuneus | −6 | −62 | 34 | 3.99 | 236 | |||||
B: ASC | ||||||||||
IPC | −46 | −70 | 48 | 4.81 | 574 | 56 | −58 | 48 | 4.65 | 876 |
PMC | −20 | 20 | 62 | 4.02 | 447 | 32 | 26 | 52 | 4.30 | 777 |
VLPFC | 32 | 52 | −14 | 3.91 | 92 | |||||
C: NC>ASC | No significant voxels | |||||||||
D: NC<ASC | No significant voxels |
Extent threshold:
NC: Normal Control, ASC: Autism Spectrum Condition, IPC: inferior parietal cortex.
VLPFC: ventrolateral prefrontal cortex, PMC: premotor cortex.
No significant activation was identified by direct comparisons between the groups.
To identify locations of activation foci unbiased for both groups, we first performed the contrast of analytic (easy and difficult) vs. baseline by combining NC and ASC. We found reliable activation bilaterally in the IPC, PMC, and VLPFC and in the left LOTC (
(a) Left ventrolateral prefrontal cortex (VLPFC), (b) right VLPFC, (c) left premotor cortex (PMC), (d) right PMC, (e) left inferior parietal cortex (IPC), (f) right IPC, and (g) left lateral occipitotemporal cortex (LOTC). Error bars indicate the standard error of mean. †: A significant interaction of Group (normal contorol/autism spectrum condition)×Condition (easy/difficult) *: A significant main effect of Condition.
Left | Right | |||||||||
Region label | x | y | z | z value | Cluster size | x | y | z | z value | Cluster size |
A: NC | ||||||||||
IPC | −44 | −76 | 44 | 4.77 | 689 | 54 | −70 | 34 | 4.63 | 400 |
VLPFC | −42 | 48 | −10 | 3.69 | 130 | 34 | 34 | 44 | 3.42 | 140 |
PMC | −36 | 16 | 56 | 3.48 | 78 | |||||
B: ASC | ||||||||||
IPC | −48 | −64 | 40 | 5.03 | 894 | |||||
PMC | −36 | 12 | 52 | 4.78 | 698 | 36 | 28 | 56 | 4.35 | 299 |
LOTC | −62 | −28 | −16 | 4.35 | 480 | |||||
VLPFC | −38 | 52 | −2 | 3.91 | 114 | |||||
C: NC and ASC | ||||||||||
IPC/ | −50 | −68 | 44 | 6.62 | 2185 | 54 | −62 | 46 | 5.84 | 1700 |
LOTC | −64 | −28 | −16 | 4.65 | ||||||
PMC | −36 | 14 | 54 | 5.40 | 1301 | 38 | 28 | 54 | 4.66 | 1004 |
VLPFC | −42 | 48 | −6 | 4.99 | 875 | 38 | 50 | −10 | 4.42 | 291 |
Precuneus | 4 | −64 | 40 | 4.25 | 298 |
Coordinates are in the MNI space.
NC: Normal Control, ASC: Autism Spectrum Disorder, IPC: inferior parietal cortex.
PMC: premotor cortex, VLPFC: ventrolateral prefrontal cortex, LOTC: lateral occipitotemporal cortex.
Among the regions in the frontoparietal network for fluid general intelligence, only analysis using the left IPC as the seed revealed a cluster of voxels with significant group difference (see
(a) The location of the seed area in the left inferior parietal cortex (IPC). The cross hair shows the focus of the seed area. (b) The right anterior prefrontal cortex (aPFC) showing reduced connectivity with the left IPC in autism spectrum condition (ASC) compared with normal control (NC). The peak of the group difference is shown in the MNI space. The cluster size was 109 voxels.
Using the RSPM task, the present fMRI study explored functional characteristics of the neural substrates of geometric reasoning in ASC. Our findings revealed that, among nodes of the brain network of fluid intelligence, ASC patients showed larger activation in the left LOTC during the easy analytic task than NC, whereas both groups showed comparable activation during the difficult analytic task. ROI analysis confirmed interaction of Group and Condition, indicating that while activity increased with task difficulty in NC, such modulation of activity was absent in ASC. Furthermore, functional connectivity analysis revealed that ASC showed significant reduction of activation coupling of the left IPC with aPFC in the right hemisphere. These results indicate that neural substrates of geometric analogical reasoning are functionally altered in ASC in terms of both localized responsiveness of the left LOTC and cortical interaction within the large-scale frontoparietal network of fluid intelligence.
In the analysis of activation magnitude, the left LOTC was the only region in the network of fluid intelligence that showed a significant group difference in both voxel- and ROI-based analyses. While NC showed increased activity with task difficulty, ASC showed an already-elevated activation level during the easy analytic condition without further enhancement of activation. Such altered relationship between activation magnitude and task difficulty has been reported in several clinical conditions. In previous studies that parametrically examined neural response for working memory load schizophrenia, the peak of the inverted U-shaped curve of the load-response function was shifted to a lower load in patients compared with normal controls
Although the exact roles of the left LOTC in fluid intelligence remain inconclusive, several previous studies have reported involvement of this region in analogical reasoning using geometric pictures
Increased activation of the left LOTC in the easy geometric analogical reasoning is also consistent with the proposal of an enhanced perceptual functioning (EPF) model of ASC
Functional connectivity analysis revealed significant reduction of activation coupling between the left IPC and right aPFC during geometric reasoning in ASC. This finding is in contrast to the observation that localized activation of the left LOTC was enhanced in ASC at least during the easy analytic condition. A number of previous fMRI studies have indicated that the right aPFC is crucial for higher-level cognition, including reasoning and integration of diverse information
The first aim of the study was to identify activation for analogical reasoning by comparing analytic with figural problems. In this design, potential group differences in activation for figural processing, if any, cannot be examined. The design was based on the analysis of cognitive components that both analytic and figural problems would involve figural processing, but only analytic problems would demand higher processes of analogical reasoning, such as the induction of abstract relations, strategy shifting and planning, goal management, and executive control of these processes
Although the present study was motivated by the large discrepancy between RSPM and WAIS in ASC
To conclude, we found fMRI evidence that the neural system of geometric reasoning is functionally altered in high-functioning adults with ASC. The observations of an elevated level of activity in the left LOTC and a reduced activity coupling of the left IPC with the right aPFC support the view that the neural recruitment in ASC is biased toward the functionally specialized local regions rather than the distributed large-scale network. Such differential involvement of the brain system may help explain the uneven profile of neural substrates reflecting cognitive ability in ASC.
We would like to thank Drs. M. Mimura and N. Yahata for helpful discussions on the experimental design of this study.