Visual body part representation of the lateral occipitotemporal cortex in individuals with autism spectrum disorder: A univariate and multivariate fMRI study Open Access

Human body parts such as faces and limbs convey crucial social information such as identity, emotions, and intentions. For instance, typically developing (TD) adults may easily discern the emotions of a person based on their face or other body parts, which are also rich sources of identity information (Bank et al., 2015; Hock et al., 2017). Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by social and communicative difficulties, restricted interests, and repetitive patterns of behavior (APA, 2013). Previous behavioral studies have shown that individuals with ASD exhibit difficulties in detecting intentions from bodily actions (Cossu et al., 2012) and in recognizing facial identity (Weigelt et al., 2012) and emotion (Yeung, 2022), which may be related to observed difficulties in social communication. Although these difficulties in recognizing emotions/identity could arise from differences in the neural representational body parts, the neural representation of the human body in the brains of individuals with ASD is currently not fully understood.

The lateral occipitotemporal cortex (LOTC) is a key node in a wider brain network that supports the recognition of body parts and has perceptual and social functions. The perceptual function of the LOTC is the recognition of certain body parts such as a hand or a foot. For instance, seminal functional magnetic resonance imaging (fMRI) studies have found that a part of the LOTC called the extrastriate body area (EBA) is more strongly activated when viewing non-face body parts than when viewing other objects such as tools (Downing et al., 2001, 2006). Contrarily, the LOTC is also involved to distinguish between self and others and to recognize the intentions of others during social interaction. These LOTC functions are implicated in action observation and execution (Astafiev et al., 2004; Gazzola & Keysers, 2009; Oosterhof et al., 2012), gestural interaction (Okamoto et al., 2014; Sasaki et al., 2012), and in understanding the meaning of actions (Kubiak & Króliczak, 2016; Okamoto et al., 2014; Wurm et al., 2016, 2017), which provides important information for interacting with another person.

A number of previous fMRI studies exploring the functions of the LOTC have shown differences in social functions between TD adults and those with ASD. For instance, adults with ASD showed lower LOTC activation when they were imitated by another person compared to TD adults (Okamoto et al., 2014). Furthermore, a meta-analysis revealed lower activation of the LOTC in individuals with ASD during action understanding and imitation tasks, compared to TD individuals (Yang & Hofmann, 2016). LOTC activation was lower in individuals with ASD than in TD individuals when hand perspective (first- and third-person perspective) was used to differentiate between self and other bodies (Okamoto et al., 2018). These findings suggest that individuals with ASD and those without ASD have differences in various social functions in the LOTC. In contrast, there is limited fMRI study of ASD individuals focusing on perceptual function in the LOTC. For instance, our previous fMRI studies using classical mass-univariate analyses have shown that mean responses to the presentation of bodies relative to other object in the LOTC are comparable between TD adults and adults with ASD (Okamoto et al., 2014, 2017, 2018). The study leads the hypothesis that individuals with and without ASD have similar perceptual functions of the LOTC.

However, more recent studies found different characteristics in perceptual function in the LOTC; the LOTC has different representation among each body part (Bracci et al., 2015; Orlov et al., 2010). Orlov et al. (2010) found a topographically ordered body part map with separate clusters, demonstrating an obvious preference for different visually presented body parts, such as the upper limbs or trunk. Bracci et al. (2015) further discovered body parts representation in the LOTC using representational similarity analysis (RSA). In their study, the organization of body part representation within the LOTC was characterized by functional-semantic qualities, forming three distinct groups: action effectors (hands, feet, arms, and legs), non-effector body parts (chest and waist), and facial parts (upper and lower faces). Okamoto et al. (2020) revealed that, although responses to whole-body images differ between TD children/adolescents and adults, body part representation in the LOTC of TD children follows a functional-semantic organization like TD adults. Furthermore, LOTC body part representation was associated with sensory characteristics in TD children/adolescents, which is frequently observed in individuals with ASD (Okamoto et al., 2020). If the functional organization of ASD individuals is different from that reported for TD individuals, this may partially explain observed differences in body recognition and social interaction between adults with ASD and TD adults. Alternatively, it is possible that individuals with ASD show similar body representation in the LOTC as TD individuals, and that higher-order functions such as self-other differentiation or social interaction in the LOTC might be associated with social difficulties in ASD. However, to the best of our knowledge, no study has tested the functional semantic organization of body parts within the LOTC of individuals with ASD.

In the present study, we conducted fMRI to investigate whether body part representation in the LOTC of adults with ASD is organized into the same three body part clusters found in TD adults (Bracci et al., 2015; Okamoto et al., 2020). As in our previous study (Okamoto et al., 2020), participants were shown photographs displaying a hand, foot, arm, leg, chest, waist, upper face (UF), lower face (LF), whole body, or chair and performed a 1-back task. RSA and classification-based multivariate pattern analysis (MVPA) were used to investigate body part representation in the LOTC after revealing LOTC activation using pictures of the whole body (relative to a chair). Here, we explored whether the functional semantic organization of body parts in the LOTC of adults with ASD is similar to that of TD adults or not. Furthermore, our previous study showed that sensory abnormalities are associated with differences in spatial LOTC organization in TD children/adolescents (Okamoto et al., 2020). Hence, we further explored whether spatial body part geometry in the LOTC was associated with several ASD-related individual traits during adulthood (both ASD and TD). As functional differences between ASD and TD are present during childhood but disappear in adulthood (Okamoto et al., 2017), the representation of body parts might be comparable between adults with and without ASD. Alternatively, it may be possible that body part representation in the LOTC is different between adults with and without ASD, although the overall activation of the human body is similar.

Twenty-three TD adults (mean age ± standard deviation (SD) [range]: 31.0 ± 10.2 [20–49]) and 23 adults with ASD (mean age ± SD [range]: 30.4 ± 5.31 [20–38]) participated in this study. All participants were diagnosed with ASD by a psychiatrist (H.K.) according to the DSM-5 criteria. General autistic traits, sensory processing characteristics, and intellectual abilities were measured in both TD participants and those with ASD. Specifically, general autistic traits were measured using the Social Responsiveness Scale (SRS) (Constantino & Gruber, 2005; Kamio et al., 2013) and the Japanese version of the Autism Spectrum Quotient (AQ) (Baron-Cohen et al., 2001; Wakabayashi et al., 2004). Sensory processing characteristics were measured using the Sensory Profile (SP) (Dunn, 1999; Ito et al., 2013). Intellectual ability was assessed using the Full-Scale Intelligence Quotient (FSIQ) of the Japanese version of the Wechsler Adult Intelligence Scale-Third Edition (WAIS) (Fujita et al., 2006). The results of the t-test on the FSIQ scores showed that there was no significant difference between the TD and ASD groups. Handedness was measured in all participants using the Edinburgh Handedness Inventory (Oldfield, 1971). The study protocol was approved by the Ethics Committees of the University of Fukui (Japan) and the Advanced Telecommunications Research Institute International (Japan). This study was conducted in accordance with the principles of the World Medical Association Declaration of Helsinki. Written informed consent was obtained from each participant after receiving a detailed explanation of the study.

Using a 3.0-T MR imager (SIGNA PET/MR; GE Healthcare), we obtained functional and structural volumes. Functional volumes were obtained using T2*-weighted gradient-echo echo-planar imaging (EPI) sequences. Each volume had 53 oblique slices that were each 3.0 mm thick with a 50% gap between them. With a flip angle (FA) of 90° and an echo time (TE) of 25 ms, the repetition time (TR) between two sequential acquisitions of the same slice was 3,000 ms. The field of view was 192 mm × 192 mm. The pixel sizes for the digital in-plane resolution were 3.0 mm, 3.0 mm, or 64 × 64 pixels. Three-dimensional fast-spoiled gradient-recalled acquisition (TR, 8.464 ms; TE, 3.248 ms; FA, 11°; 256 × 256 matrix; voxel size, 1 × 1 × 1 mm) was used to create a high-resolution anatomical image.

Presentation software (Neurobehavioral Systems) implemented on a Windows-based desktop computer was used to present visual stimuli and collect responses. Visual stimuli were presented on a liquid-crystal display monitor and viewed by the participants via a mirror attached to the head coil of an MRI scanner. Head motion was minimized by placing a comfortable but tight-fitting foam padding around the head of each participant.

The same task as in our previous study (Okamoto et al., 2020) (Fig. 1) was utilized in this study. Specifically, the participants completed four runs during which they looked at grayscale images comprising 10 different conditions (whole bodies, hands, feet, arms, legs, chests, waists, UFs, LFs, and chairs). Images of each condition, except that for the chair condition, included the entire bodies or body parts of 11 men, 11 women, 11 boys, and 11 girls. The chair conditions included 11 different types of chairs. Thus, 440 images were presented for each condition. All images were changed to greyscale with a white background color and a matrix size of 800 × 600 pixels using Adobe-Photoshop software (Adobe System Inc.). Each run consisted of 20 task blocks (10 conditions × 2 repetitions = 20 blocks). In each block, 12 pictures were presented for 500 ms, with a 500 ms interstimulus interval. Thus, each block lasted 12 s. The conditions were arranged in a pseudo-random order. A fixation-only baseline period was inserted before the 1^st block (27 s), after the 5^th, 10^th, and 15^th blocks (12 s), and after the 20^th block (15 s). Identical pictures were presented in sequence one per block, and participants were asked to complete a 1-back task in which they were required to press a button upon viewing two identical pictures in sequence.

Age, FSIQ, SRS, AQ, low registration score, sensory seeking score, sensory sensitivity score, and sensation avoidance score between the ASD and TD groups were compared using Welch’s t-test. All p-values were corrected using Bonferroni correction.

The correct response ratio, false alarm ratio, and reaction times for the target stimuli in the 1-back task, as well as number of responses for all stimuli were calculated for each participant. Correct response and false alarm ratios, and number of responses across all conditions were initially compared between TD and ASD groups using Welch’s t-test in R (https://www.r-project.org/). Additionally, we explored whether correct response ratios and reaction times differed by condition using a two-way analysis of variance (ANOVA) with the factors body parts (whole bodies, chairs, hands, feet, arms, legs, chests, waists, UFs, and LFs) and groups (TD and ASD) using R.

The first five volumes of each run were discarded due to unsteady magnetization. The remaining volumes were analyzed using Statistical Parametric Mapping software (SPM12; Wellcome Department of Imaging Neuroscience) (Friston et al., 1994) implemented in MATLAB 2021b (MathWorks). All functional images were initially realigned. The high-resolution anatomical images were then co-registered to the mean image of the realigned functional images and normalized to a tissue probability map that had already been fitted to the Montreal Neurological Institute (MNI) space via a segmentation–normalization procedure. The parameters from the segmentation-normalization process were then applied to all functional images, which were resampled to a final resolution of 2 × 2 × 2 mm³. A region of interest (ROI) analysis was performed using normalized unsmoothed images. To localize the whole-body sensitive region in the LOTC, normalized fMRI images were filtered using a Gaussian kernel of 8 mm (full width at half maximum) in the x-, y-, and z-axes.

The region that responded to viewing the whole body (i.e., the whole-body-sensitive region) in the LOTC was initially localized for each participant. We then conducted univariate and multivariate analyses within this whole-body-sensitive region to examine the activation patterns upon viewing different body parts. In both analyses, activation patterns in LOTC between TD and ASD participants were compared.

Classical mass-univariate analysis was conducted at two levels to define whole-body-sensitive ROIs in the LOTC. In the first-level single-subject analysis, a general linear model was fitted to the fMRI data of each participant (Friston et al., 1994; Worsley & Friston, 1995). The blood-oxygen level-dependent signal was modeled using boxcar functions convolved with a canonical hemodynamic response function. Each run included one regressor for each of the 10 conditions (whole bodies, chairs, hands, feet, arms, legs, chests, waists, UFs, and LFs) and six regressors of motion-related parameters (three displacements and three rotations obtained by the rigid-body realignment procedure). The time series of each voxel was high pass filtered at 1/128 Hz. Serial autocorrelation was estimated from a collection of pooled active voxels using the constrained maximum likelihood approach, assuming a first-order autoregressive model, and was used to whiten the data (Friston et al., 2002). Global signal scaling, such as scanner gain changes, is used to eliminate global confusion. The contrast estimates of whole bodies versus chairs for each participant were compared using linear contrasts.

Thereafter, a second-level group analysis was performed on contrast images (whole bodies vs. chairs) from the first-level analyses, and whole-body-sensitive regions (whole-body vs. chair) in each hemisphere were identified within each group. The resulting set of voxel values for each contrast constituted SPM{t}. The statistical threshold of SPM{t} was set at p < 0.05 to correct for multiple comparisons at the cluster level over the entire brain (family-wise error), with a height threshold of p < 0.001. In addition, to explore potential differences across the whole brain between TD and ASD, we performed a classical analysis using an uncorrected p-value threshold of 0.05, as well as a Bayesian 2nd-level analysis (Han & Park, 2018).

Subsequently, the activation of the whole-body sensitive region between TD participants and those with ASD was compared. We examined four variables using individual ROI analyses in the following order: ratio of participants exhibiting a category-sensitive reaction, extent of activation (i.e., total number of voxels over a threshold) for participants exhibiting this response, peak coordinate position for persons displaying activation, and activation pattern among four object categories at individual peak coordinates for those exhibiting this response. To avoid circular analysis, an 8-mm-diameter sphere centered on the coordinates of the original study was utilized (Bracci et al., 2015), which was converted from Talairach to MNI coordinates (Lancaster et al., 2007). The statistical threshold was set at p < 0.01, uncorrected for multiple comparisons (Okamoto et al., 2020). To compare the ratios of participants showing activation and the size of activation between TD individuals and those with ASD, a $χ2$ test and a two-sample t-test were conducted in R, respectively.

ROI analyses were based on the beta images (parameter estimates for each voxel) resulting from the single-subject mass-univariate whole-brain analysis performed on unsmoothed data. To rule out the possibility that group differences in ROI volumes mask any between-group differences in activation measures, the mean activation of the ASD and TD groups (contrast of whole bodies vs. chairs) was used.

Univariate regional average activation analysis: Mean beta estimates of eight body parts relative to baseline were extracted from the ROIs in each hemisphere for each participant to assess the overall activation intensity of the ROIs. R software was used to perform a two-way ANOVA on both body parts and groups.

Representational similarity analysis: The representational geometry of the neural population code was evaluated using correlation-based RSA (Bracci et al., 2015; Haxby et al., 2001; Kriegeskorte & Kievit, 2013). Within the whole-body sensitive region in each hemisphere, parameter estimates of eight body parts (hands, feet, arms, legs, chest, waist, UFs, and LFs) were compared with the fixation-only baseline of each voxel for each run. An eight-condition × eight-condition dissimilarity matrix (1 - r) for each ROI was built for each participant based on the correlation coefficient r between the parameter estimates of the odd and even runs. For each group, a mean dissimilarity matrix was generated, and multidimensional scaling (MDS) was used to visualize the similarity structures using the MATLAB function mdscale.

To examine whether the spatial activation patterns in the LOTC were organized according to the three categories (action effector body parts, non-effector body parts, and face parts) in each group, we utilized a hypothesis-driven approach based on previous studies (Bracci et al., 2015; Okamoto et al., 2020). For each group, an analysis of similarity (ANOSIM) (Clarke, 1993; Legendre & Legendre, 1998a, 1998b) was performed using the Fathom Toolbox (Jones, 2017) in MATLAB. ANOSIM is a hypothesis-driven non-parametric statistical test frequently employed in ecology to determine whether the similarity between categories is greater than or equal to the similarity within categories. According to ANOSIM, which provides a dissimilarity index (R-value) ranging from -1 to +1, a result near 1.0 indicates a strong grouping or high separation of body part categories. The statistical tests on the R values included a complete permutation test comparing the hypothesized model ([1] hand, arm, leg, and foot; [2] chest and waist; and [3] UF and LF) with 419 alternative models ([1] arm, leg, chest, and waist; [2] UF and LF; [3] foot, hand, and other possible patterns), totaling 420 patterns. This statistical method assesses the relevance of the hypothesized model compared to alternatives, making ANOSIM more suitable than approaches like hierarchical clustering for this study.

We further examined the comparability of spatial activation patterns in the LOTC between the TD and ASD groups. Initially, we utilized the Mantel test, which measures the correlation between two dissimilar matrices (Legendre & Legendre, 1998a, 1998b; Mantel, 1967). The Mantel test employs the Pearson product-moment correlation coefficient (R), which ranges from -1 to 1. An R-value of ±1 indicates a perfect positive or negative association, respectively. The null hypothesis (H0) assumes that there is no significant relationship between the two matrices, meaning the elements of one matrix are independent of those in the other. Under this hypothesis, any observed correlation is attributed to random chance. Conversely, the alternative hypothesis (H1) asserts that there is a significant correlation between the matrices, suggesting a systematic relationship between elements that cannot be explained by random variation alone. Therefore, rejecting H0 in favor of H1 implies that the two matrices are significantly similar.

Regarding individual variability, correlation coefficients were determined between the dissimilarity matrix of each participant and the mean dissimilarity matrix of their group, as well as between the dissimilarity matrix of each participant and the mean dissimilarity matrix of the other group. When correlating the dissimilarity matrix of a particular participant with those in the participant’s group, the group dissimilarity matrix was calculated by excluding the data for that participant. Four correlation coefficients (correlation between a participant with TD and the TD group, the correlation between a participant with TD and the ASD group, the correlation between a participant with ASD and the TD group, and the correlation between a participant with ASD and the ASD group) were calculated. The correlations were normalized using Fisher’s Z transformation. The type of correlation (within-group correlation/between-group correlation) and group (TD/ASD) were then compared using two-way ANOVA with R.

Finally, we investigated whether spatial body part geometry in the LOTC was associated with several ASD-related individual traits. First, the R-value of the ANOSIM for the LOTC ROIs of each participant was generated and utilized as a measure of spatial body part organization. For each hemisphere, a Spearman’s rho correlation analysis was performed between R values and scores measuring autistic traits (SRS and AQ total scores), sensory processing characteristics (SP low registration, SP sensory seeking, SP sensory sensitivity, and SP sensation avoidance), age, and intellectual ability (FSIQ). R software was used for this analysis, and Bonferroni correction was performed.

Classification-based multivariate pattern analysis: In addition to the RSA, the varied spatial activation patterns of the LOTC between categories (action effector [hands, limbs, arms, feet], faces [UFs, LFs], and non-effector body parts [chest, waist]) were confirmed using a support vector machine (SVM)-based classification analysis. For this, the Decoding Toolbox (TDT) (Hebart et al., 2014) implemented in SPM12 was utilized. The beta values for the feature vectors and a linear SVM classification with a leave-one-run-out cross-validation approach were used. More specifically, feature vectors were created in three of the four runs and tested in the remaining one. We repeated the process four times, changing the test run. TDT offers a template script for unbalanced data (i.e., decoding template unbalanced data), which was used during the cross-validation iteration because this study used an unbalanced design (i.e., the number of conditions varied within categories). In total, 100 bootstrap samples were used in the analysis. For the LOTC ROIs, the classification accuracy above chance level was determined to assess the classification performance. A one-sample t-test in R was used to determine if the decoding accuracies were higher than expected by chance (33.3%), and a two-sample t-test was used to compare the decoding accuracies between the two groups (TD and ASD). Furthermore, comprehensive permutation testing was used to confirm whether the spatial activation patterns in the LOTC could be classified into one of three classes in each group: action effector, face, or non-effector. The condition labels of each class (or category) were permuted to produce an empirical null distribution, which was used to estimate the likelihood of the original (hypothesized) labeling. For the three-class categorization, this resulted in a statistically significant p-value. All 420 feasible possibilities were investigated using a permutation process.

There were significant differences in SRS, AQ, and one sub-scale of sensory profile (Sensory seeking) between the TD and ASD groups, according to the two-sample Student’s t-test (Table 1). In contrast, there were no significant differences in age, FSIQ, and other three SP sub-scales between the TD and ASD groups (Table 1).

For the 1-back task, the overall percentage (%) of correct responses to target stimuli (correct response ratio) was significantly greater in the TD group than in the ASD group, according to the two-sample Welch’s t-test (t_34.15 = -2.92, p = 0.006, Cohen’s d = -0.861; Fig. 2A). In contrast, there were no group differences in the false alarm ratio (t_23.09 = -1.02, p = 0.318, Cohen’s d = -0.301; Fig. 2B) or response ratio for all stimuli (t_25.09 = -1.69, p = 0.104, Cohen’s d = -0.497; Fig. 2C).

Furthermore, we explored group differences in reaction time and correct response ratio using a two-way ANOVA on body parts and groups. The analysis of reaction time revealed a significant main effect of body part (F_1,396 = 3.74, p < 0.001, $ηp2$ = 0.078) (Supplementary Fig. S1A). In contrast, there was no significant main effect of group (F_1,396 = 1.21, p = 0.280, $ηp2$ = 0.027) and no interaction of group and body parts for reaction time (F_1,396 = 1.80, p = 0.066, $ηp2$ = 0.039). The posthoc analysis of body parts showed significant differences in reaction time between whole body and chests, legs, waists, feet, and arms, as well as between chairs and feet (whole body < chest, p < 0.001; whole body < leg, p < 0.001; whole body < waist, p = 0.002; chair < foot, p = 0.005; whole body < foot, p = 0.026; whole body < arm, p = 0.032). The two-way ANOVA on correct response ratio showed a significant main effect of group (F_1,396 = 10.38, p = 0.002, $ηp2$ = 0.191) (Supplementary Fig. S1B). However, there was no significant main effect of body part (F_1,396 = 1.20, p = 0.293, $ηp2$ = 0.027) and no interaction of group and body part in the correct response ratio (F_1,396 = 0.78, p = 0.639, $ηp2$ = 0.017). Collectively, the correct response rate of participants with ASD was lower than that of TD participants, regardless of the condition.

The head motion during scanning was initially confirmed in each group. The mean framewise displacement (FD) (Power et al., 2012) for each participant was calculated and compared between ASD and TD groups. There was no significant difference in head motion between ASD and TD groups (ASD group: mean ± SD = 0.102 ± 0.077, TD group: mean ± SD = 0.141 ± 0.139; t_34.34 = -1.17, p = 0.250, Cohen’s d = -0.345).

The whole-body-sensitive region in the LOTC was depicted for each group using the contrast of whole bodies versus chairs. We examined differences in activation between the ASD and TD groups. In the group analysis, activation was found in the bilateral LOTC in both the ASD and TD groups (Fig. 3A, B). We then compared the activation intensity between the ASD and TD groups; there was no significant difference. Figure 3C shows the mean activation of the ASD and TD groups, which were used for subsequent ROI analyses.

On the other hand, we checked the cluster groups that represent the difference of activities between TD and ASD when the uncorrected p-value threshold was set to 0.05 (Supplementary Fig. S2). Similarly, we checked the cluster groups that represent the difference of activities between TD and ASD when 2nd-level Bayesian analysis is performed (Effect size > 0.2) (Supplementary Fig. S3). Even when the threshold was lowered in these additional analyses, no notable differences emerged in the vicinity of the LOTC region.

For individual analyses, the size of whole-body sensitive activation and the ratio of participants showing whole-body sensitive regions in the LOTC between the ASD and TD groups were compared (Table 2). In the left hemisphere, 21 out of 23 participants in the TD group (91%) and 21 out of 23 participants in the ASD group (91%) exhibited activation; there were no group differences (χ₁₂ = 0, p = 1, Cramer’s V = 0). There was no significant difference in the activation size (t_32.41 = 1.27, p = 0.205, Cohen’s d = 0.340) using Welch’s t-test. In the right hemisphere, 23 out of 23 participants in the TD group (100%) and 22 out of 23 participants in the ASD group (96%) exhibited activation; there were no group differences (χ₁₂ = 0, p = 1, Cramer’s V = 0.149). In addition, there was no significant difference between the TD and ASD groups in terms of activation size (t_37.88 = 0.979, p = 0.334, Cohen’s d = 0.289; Table 2). Therefore, the activation size and ratio of participants in the whole-body sensitive region were highly similar in the ASD and TD groups.

To assess the overall activation intensity for each body part, the mean beta estimates of eight body parts (hands, feet, arms, legs, chest, waist, UFs, and LFs) compared with the baseline were extracted from each ROI (Fig. 4A). A two-way ANOVA on body part and group revealed a significant main effect of body parts (right: F_7,308 = 16.68, p < 0.001, $ηG2$ = 0.047; left: F_7,308 = 6.65, p < 0.001, $ηG2$ = 0.018). There was no significant main effect of group (right: F_1,44 = 1.33, p = 0.255, $ηG2$ = 0.026; left: F_1,44 = 0.01, p = 0.937, $ηp2$ = 0.001) or interaction of group and body part (right: F_7,308 = 0.54, p = 0.802, $ηp2$ = 0.002; left: F_7,308 = 0.09, p = 0.999, $ηp2$ = 0.0003). Post-hoc pairwise comparisons with Bonferroni correction revealed significant differences between the hand and other body parts, arm versus waist, leg versus waist, and foot versus waist for the right hemisphere, and hand versus foot, hand versus waist, hand versus chest, waist versus LF, chest versus LF, leg versus waist, UF versus LF, hand versus UF, and arm versus waist for the left hemisphere (all p values <0.05). Even when using percent signal change, similar results were obtained to those using the beta value (Supplementary Fig. S4, see Supplemental Information). Thus, the overall activation intensity of the bilateral LOTC when viewing each body part was comparable between TD individuals and those with ASD.

The representational geometry of body parts in the whole-body-sensitive regions was investigated (Fig. 4B, C). ANOSIM revealed that the spatial activation pattern in the whole-body sensitive region was well characterized by three distinct clusters in both the TD (R = 0.900, p = 0.005 for the right hemisphere; R = 0.675, p = 0.024 for the left hemisphere) and ASD groups (R = 0.925, p = 0.005 for the right hemisphere; R = 0.675, p = 0.017 for the left hemisphere). Therefore, the action effectors (hand, foot, arm, and leg), faces (upper face and lower face), and non-effector body parts (chest and waist) were clustered separately in the bilateral LOTC in the ASD and TD groups.

We then investigated if the dissimilarity matrices were comparable between the ASD and TD groups. The Mantel test revealed that the dissimilarity matrices of both groups were highly similar (R = 0.925, p < 0.001 for the right hemisphere: R = 0.892, p < 0.001 for the left hemisphere). Furthermore, we performed Bayesian linear regression analysis on the TD/ASD dissimilarity metrices body parts representations. The results showed extremely large Bayes factors (BF > 10⁴), providing decisive evidence that the slope (β) is non-zero in both the left and right LOTC. These results strongly support the alternative hypothesis and indicate positive similarity between TD and ASD body part representations in left/right LOTC (see Supplemental Information).

Among the correlation of each participant’s dissimilarity matrix and group dissimilarity matrix, a two-way ANOVA (correlation type [within-group correlation/between-group correlation] and group [ASD/TD]) on correlation coefficients in the left LOTC revealed no significant main effect of correlation type (F_1,44 = 0.58, p = 0.449, $ηp2$ = 0.013; Fig. 5A, B) or group (F_1,44 = 0.15, p = 0.704, $ηp2$ = 0.003; Fig. 5A, B) and no interaction of correlation type and group (F_1,44 = 0.20, p = 0.657, $ηp2$ = 0.005; Fig. 5A, B). In the right LOTC, there was no significant main effect of correlation type (F_1,44 = 3.67, p = 0.062, $ηp2$ = 0.077; Fig. 5C, D), main effect of group (F_1,44 = 0.24, p = 0.629, $ηp2$ = 0.005; Fig. 5C, D), or interaction of correlation type and group (F_1,44 = 0.72, p = 0.402, $ηp2$ = 0.016; Fig. 5C, D).

We further investigated the representational geometry (within-group value) in the LOTC in relation to individual ASD-related traits in the ASD and TD groups. No significant correlation was found for any of the individual trait measures in the TD group or for the integration of both the ASD and TD groups (Table 3).

In summary, the spatial representation of each body part in the whole-body sensitive region of the LOTC was organized into three clusters (action-effector body parts, faces, and non-effector body parts) in both groups and was highly similar between the ASD and TD groups. Furthermore, we did not observe any significant correlations between representational geometry and ASD-related traits.

Finally, the spatial activation patterns were confirmed to depend on the categories (i.e., action effector body parts, faces, and non-effector body parts) using a classification-based multivariate analysis. Classification accuracies were above chance level in both the ASD (right hemisphere: t₂₂ = 10.38, p < 0.001, Cohen’s d = 2.165; left hemisphere: t₂₂ = 12.31, p < 0.001, Cohen’s d = 2.570; Fig. 6A, C) and TD (right hemisphere: t₂₂ = 11.50, p < 0.001, Cohen’s d = 2.300; left hemisphere: t₂₂ = 14.73, p < 0.001, Cohen’s d = 3.072; Fig. 6A, C) groups, and there were no significant differences in classification accuracy between the groups (right hemisphere: t_42.38 = 0.73, p = 0.467, Cohen’s d = 0.216; left hemisphere: t_39.83 = 1.45, p = 0.154, Cohen’s d = 0.429). In both the ASD and TD groups, the spatial activation patterns in the LOTC were well characterized by three clusters representing action effectors, faces, and non-effector body parts (Fig. 6B, D). Thus, we confirm that body part representation in the bilateral LOTC was highly similar between the ASD and TD groups, further supporting the RSA results.

In this study, we investigated whether body part representations in the left/right LOTC of adults with ASD are organized into the same three body part clusters found in TD adults. In the behavioral results, adults with ASD showed lower behavioral performance in the 1-back task than TD adults, regardless of the stimulus category. Univariate analysis confirmed that the whole-body sensitive region in the bilateral LOTC was comparable between the ASD and TD groups. ANOSIM within the framework of RSA revealed that activity in the LOTC was significantly clustered into three categories: action effector body parts, face parts, and non-effector body parts. The Mantel test demonstrated that the dissimilarity maps of body parts in the LOTC were significantly similar between the TD and ASD groups. Furthermore, MVPA provided additional evidence of the shared representational structure of body parts in the LOTC between the groups. However, body part representations in the bilateral LOTC showed no relationship with individual traits in either group.

Behavioral analysis showed that the percentage of correct responses in the 1-back task was lower in the ASD group than in the TD group. It is unlikely that the low accuracy in the ASD group was attributable to a failure to complete the button-press task because the false alarm and response ratios for all stimuli were comparable across both groups. There is abundant evidence that individuals with ASD show reduced performance compared to TD individuals in tasks that require working memory (e.g., N-back tasks) (de Vries & Geurts, 2014; Geurts et al., 2004; Habib et al., 2019; Kenworthy et al., 2008; Williams et al., 2005, 2006). As our study did not show a significant interaction between stimulus condition and group, the lower accuracy observed in the ASD group may not be specific to the human body (faces or bodies). Therefore, the reduced response accuracy in our ASD group may indicate that the working memory of visual images, regardless of object type, in individuals with autism is lower than that in TD individuals, which is consistent with previous studies (de Vries & Geurts, 2014; Geurts et al., 2004; Habib et al., 2019; Kenworthy et al., 2008; Williams et al., 2005, 2006).

RSA revealed that the TD and ASD groups showed comparable spatial representations of each body part in the LOTC. In both groups, the spatial activity patterns of the bilateral LOTC were divided into three clusters: (1) action effector body parts (hands, feet, arms, and legs), (2) face parts (upper and lower face parts), and (3) non-effector body parts (chest and waist). Although several previous fMRI studies have addressed EBA activation in adults with ASD (Okamoto et al., 2017), to the best of our knowledge, no study has examined the body part representation in this region using RSA. Thus, the present study provides novel evidence that the body part representation in the LOTC of adults with ASD is similar to that in TD adults.

The main findings of this study were the significant division of representations of body parts using RSA into three groups and the significant similarity between representations of the TD and ASD groups. Univariate analysis revealed no significant differences in the bilateral whole-body-sensitive regions of the LOTC. However, using the Mantel test within the RSA framework, we found that body part representations in the LOTC were similar in TD adults and in those with ASD. Additionally, ANOSIM showed that, for both groups, bilateral LOTC spatial activation patterns were grouped into three clusters: (1) action effector body parts (hands, feet, arms, and legs), (2) face parts (upper and lower face), and (3) non-effector body parts (chest and waist). These results indicate that individuals with ASD show highly similar activation to TD adults in terms of responsiveness of their neuronal populations to the body as a whole and in their responsiveness to individual body parts. A previous study using a passive viewing task also revealed that adults with ASD showed a similar activation in the higher visual cortex (EBA for body parts and fusiform face area [FFA] for faces) to that of TD adults (Okamoto et al., 2017). However, when participants were imitated by someone else, the activation of the left EBA was diminished in adults with ASD (Okamoto et al., 2014), suggesting that the EBA plays a role in higher-order cognitive processes. This is further supported by studies showing a role for the EBA in the recognition of social contingency during reciprocal imitation and the sense of agency (David et al., 2007, 2009; Okamoto et al., 2014, 2017). Similarly, several fMRI studies in which participants were asked to perform a face-to-face task showed a reduced FFA activation in adults with ASD (Critchley et al., 2000; Humphreys et al., 2008; Kleinhans et al., 2010; Pierce et al., 2001; Piggot et al., 2004; Pinkham et al., 2008; Schultz et al., 2000). Emotional facial expressions are processed in a network between the “core system,” which includes the FFA, and the “extended system,” which includes the inferior parietal and frontal lobes associated with facial expression recognition, and the amygdala, insular cortex, and striatum associated with emotion recognition (Haxby & Gobbini, 2011). In contrast to previous studies, the body/face viewing task used in the present study did not require higher-order cognitive processes. The present results suggest that there is no difference in the lower-order processing of the whole body and the spatial representation of body parts between TD and ASD adults within the LOTC.

Unlike in adults, body part representation may differ between TD children and those with ASD. A previous study showed that adults with ASD showed similar activation in the EBA and FFA in TD adults, whereas children with ASD had atypical activation in these areas. For example, we revealed that the EBA was smaller in children with ASD than in TD children (Okamoto et al., 2017). Furthermore, when the amount and extent of neuronal activity in response to the human body increases from childhood to adulthood, the functional representation of these stimuli does not change in terms of spatial encoding (Ross et al., 2014). Thus, there may be differences in the representation of body parts in the LOTC between individuals with ASD and TD individuals until adolescence; however, these differences may disappear in adulthood. If this is the case, neural mechanisms underlying social difficulties are different between adults and children with ASD and initial difference of the processing of bodies in children with ASD might contribute to future social difficulties.

Our previous study showed an association between body-parts representation in the LOTC and the sensory characteristics (sensory avoidance tendency), which are frequently observed in ASD, in TD children and adolescents, although general autistic traits were not associated with the LOTC (Okamoto et al., 2020). The findings lead to the possibility that the pattern of body-parts representation in the LOTC leads to sensory-avoiding behaviors of children with ASD. In contrast, in the present study, we did not find any correlation between the spatial organization of body parts in the LOTC and individual traits in adults with or without ASD. Although some degree of sensory atypicality persists across ages in individuals with ASD (Leekam et al., 2007), atypicalities in the sensory profile of children with ASD tend to decrease with increasing age (Baranek et al., 2019; Ben-Sasson et al., 2009). Behavioral studies have shown that body and face recognition differences in ASD are lower in older individuals (Dowell et al., 2009; Serra et al., 1998; Strauss et al., 2012). Therefore, the observed absence of an association between body part representation and sensory characteristics in adults with autism might be due to this age-related decrease in sensory atypicality, which supports the idea that the difference in body part representation might be less in adulthood.

This study has five limitations. First, the sample size of the present study was limited. Recently, it has been claimed that studies with small sample sizes using resting-state fMRI and structural MRI have difficulties in detecting individual differences (Marek et al., 2022). However, Marek et al. argue for the importance of small-sample neuroimaging studies for clinical conditions using an induced-effects approach (such as task-related fMRI) (Marek et al., 2022). Task-related brain activation can improve the prediction of individual traits (Greene et al., 2018). However, a replication study with a larger sample could provide more robust knowledge. Second, as the participants were adults, it is unknown whether children with ASD showed similar or distinct results. Future studies involving a broader age range will provide important information for understanding the neural substrates of body perception from a developmental perspective. Third, although the ASD-related measurements differ from those of the TD group, it remains unclear where they fall within the autism spectrum. In a previous study, a meta-analysis of AQ and event-related potentials in the general population, the P3b amplitude of event-related potentials during visual stimulus tasks was higher in accordance with the characteristics of autism (Mazer et al., 2024). Furthermore, a meta-analysis of fMRI revealed that individuals with ASD have more activity in the extra-visual V2 cortex (BA18; equivalent to occipital cortex) during visual processing than typical controls (Jassim et al., 2021). As these are relatively simple tasks, we can provide a wider range of knowledge by conducting a meta-analysis including individuals with a wide range of ASD, such as those with intellectual disabilities. Fourth, in the experimental framework of this fMRI study, it is difficult to examine brain regions involved in higher-order processing of body parts, such as the middle temporal gyrus (MFG) and inferior frontal gyrus (IFG). In this study, we used a visual image localizer (body parts vs. chairs) to focus on the LOTC. In order to examine RSA in other brain regions, it is necessary to set up an appropriate localizer according to their brain functions. For example, IPL is suitable for comparison between action observation and execution (or resting) (de la Rosa et al., 2016; Koul et al., 2018). By setting appropriate localizers and examining higher brain regions, we can expect to gain further insights into the processing of body parts in the brain. Finally, while the Mantel test in this study revealed a statistically significant similarity in LOTC representations between TD and ASD groups, the absence of observed group differences means that concerns associated with a null effect—such as limited statistical power and uncertainties regarding the positioning of individuals with ASD along the spectrum—remain unresolved. In order to address this issue, as mentioned in the first limitation, it will be necessary in the future to consider whether the results of this study can be replicated by increasing the number of TD and ASD individuals further.

The present study demonstrated that the spatial extent of the whole-body-sensitive region in the bilateral LOTC was similar between the ASD and TD groups. In addition, body part representation in the LOTC was similar across TD and ASD adults and was organized into three distinct clusters (i.e., action effector body parts, non-effector body parts, and face parts) in both groups. Furthermore, we found no association between body part representation clustering in the LOTC and autism-related individual traits. Considering the univariate analysis of each body part, the visual processing of body parts in the LOTC was comparable between adults with and without autism. Alternatively, social difficulties in adults with ASD were not due to different body-part representation in the LOCT and might be associated with a higher level of cognition.

Participant data used in this study cannot be made available for public access and are only available on request from researcher and optout.

H.K., T.K., H.O., and Y.O. designed research; Y.O. and H.K. performed research; T.K. contributed unpublished reagents/analytic tools; Y.K. analyzed data; and Y.K., H.K., B.A.S., R.K., T.K., H.O., R.O., and Y.O. wrote the paper.

This work was supported by a Grant-in-Aid for Young Scientists from the Japan Society for the Promotion of Science [grant numbers: 17K17766 and 24H00179].

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Supplementary material for this article is available with the online version here: https://doi.org/10.1162/IMAG.a.24.