Introduction

Autism spectrum disorder (ASD) is a cluster of neurobiological developmental conditions clinically evident from early childhood. ASD is characterized by a multifactorial etiology, with genetic, prenatal, and postnatal environmental factors playing a role [67, 99]. Though individuals with ASD have a heterogeneous phenotype with symptom severity ranging from mild to severe [92], the disorder is defined by the presence of persistent deficits in social interaction and communication, repetitive-restricted patterns of behavior or interests [24]. Medical comorbidities often co-occur in ASD, including other psychiatric conditions (e.g., attention-deficit/hyperactivity, anxiety, conduct, obsessive–compulsive, and depressive disorders), genetic disorders (e.g., dystrophinopathies, Fragile X and Down syndromes), or neurological conditions (e.g., epilepsy, learning disabilities, cerebral palsy) [1, 50]. The prevalence of ASD is estimated to be about 7.6 in 1000 individuals worldwide [9], with a female-to-male ratio around 1:3 [66]. ASD is currently diagnosed after the integration of information across multiple contexts and the administration of an array of standardized assessments, which include interview-based instruments and observational tools [68].

Although there is a substantial amount of biomedical literature on these conditions, the autistic pathophysiology is still under investigation and the identification of ASD through sound biomarkers remains an open challenge [67, 99]. In particular, the technology and protocols of functional magnetic resonance imaging (fMRI) have seen three decades of intense development, providing an unprecedented tool for in-vivo assessment of the neurophysiological basis of the disorder. Resting-state fMRI (rs-fMRI), an established corpus of methods capable of detecting spontaneous low-frequency regional temporal correlations in the blood oxygen level-dependent (BOLD) signal [13], has shown that brain functional connectivity in ASD is atypical throughout development [40, 113]. Previous research has focused extensively on aberrations of long-range connectivity and demonstrated patterns of hyper-connectivity in independent cohorts of pediatric subjects with ASD [82, 108, 117]. In contrast, adults with ASD are characterized by long-range patterns of hypo-connectivity, or no differences, compared to typically developing controls (TDCs) [75, 82, 112].

Despite evidence from multidisciplinary efforts suggesting focal cytoarchitectonic disorganizations in the autistic cerebral cortex [25], the presence of short-range functional connectivity abnormalities is less robustly established. Several authors have speculated that neuronal activity may be locally over-connected (i.e., the general local over-connectivity theory) [10, 20, 74, 97, 126]. However, there is little empirical evidence to support this hypothesis, especially due to the lack of effective and reliable computational fMRI metrics. Moreover, electroencephalography (EEG) and magnetoencephalography (MEG) studies investigating local electrophysiological connectivity in autistic groups have reported conflicting results [18, 85], in both children [26, 36] and adult [79] groups.

In recent years, independent studies have investigated atypical local short-range connectivity in ASD using regional homogeneity (ReHo), a whole-brain rs-fMRI technique that uses Kendall’s coefficient of concordance to test the coherence of time series of the BOLD signal amplitude in small clusters of neighboring voxels [47, 137]. Although ReHo was originally developed for cluster purification in fMRI data [137], its voxel-wise nature and high test–retest reliability [141] have provided important insights into the spatial extent of local connectivity in ASD, whose changes seem to be mainly localized in brain areas associated with visual [22, 44, 51, 73, 80], default mode [22, 23, 32, 59, 73, 80], salience [23, 87], and sensorimotor [22, 87] networks. Nonetheless, these findings are highly inconsistent. For example, several studies reported mixed patterns of local hypo- and hyper-connectivity [22, 23, 51, 73, 80, 87], while other studies found only lower or higher ReHo in ASD compared to TDCs [32, 44, 59]. Moreover, contradictory effects have been reported for areas such as the cerebellum [22, 87], middle frontal gyrus [87, 103], and insula [22, 32]. Of note, ReHo studies in ASD have focused almost exclusively on pediatric cohorts (age \(\le\) 18 years). To our knowledge, only one study has examined ReHo changes in adults with ASD [22]. Unfortunately, this situation hinders a quantitative synthesis of local brain connectivity in ASD from a developmental life-long perspective [82, 113]. At the same time, however, it offers the opportunity to characterize short-range functional signatures in the critical time window in which ASD-related symptoms tend to emerge [96]. This may be important for the development of early neurobiology-based interventions, for example adopting meta-analytic identified areas for noninvasive brain stimulation [39, 65, 138].

In light of this context, this study aims to unravel for the first time, the most consistent and replicable patterns of ReHo changes in pediatric individuals with ASD and to test the hypothesis of general local over-connectivity in a data-driven manner. To this end, we used the Permutation-Subject Images version of Signed Differential Mapping (SDM-PSI) [4], one of the current coordinate-based meta-analytic methods that can provide a quantitative synthesis of neural changes across clinical groups [16, 56, 64, 107]. To further clarify the neurophysiological basis of ReHo aberrations, we examined the potential effects of clinical, sociodemographic, and methodological variables on published findings via voxel-wise meta-regression approach. Finally, we examined the large-scale network functional connectivity and psychological processes statistically related to the atypical ReHo clusters using the Neurosynth database [133], which allowed the interpretation of our findings from an observer-independent, unbiased perspective.

Methods

The design of the study adheres to the current best-practice rules for neuroimaging coordinate-based meta-analyses (CBMAs) [72, 78, 109] and to the quality criteria of the PRISMA statement [90] (Table S1).

Search strategy and data selection

A systematic literature search was performed in the PubMed database using the following combination of keywords: “autism” OR “autism spectrum disorder” OR “ASD” AND “regional homogeneity” OR “ReHo” OR “local connectivity”. Additionally, the reference lists of the included studies were manually checked and relevant reviews [40, 61] were inspected to identify articles that could have been missed during the dataset search. The final search was updated till January 2022, with no restrictions on publication year. For details see Table S2.

The identified articles were screened to verify their adherence with the following inclusion criteria: (1) to be an original article published in a peer-reviewed English-language journal; (2) to include one or more experiments investigating ReHo voxel-wise differences between subjects with ASD and TDCs at the whole-brain level; (3) to meet the diagnostic criteria of ASD based on the Autism Diagnostic Observation Schedule (ADOS) [69], Autism Diagnostic Interview-Revised (ADI-R) [70], Diagnostic and Statistical Manual of Mental Disorders (IV-R or 5 Edition) [5, 6], or the International Statistical Classification of Diseases and Related Health Problems 10th Revision [131]; (4) to include ASD and TDC subjects with an age at the scan session \(\le\) 18 years; (5) to report significant results and coordinates (x–y-z) of clusters of ReHo changes using the Talairach (TAL) or Montreal Neurological Institute (MNI) stereotactic spaces.

Articles were excluded if: (1) they were case-report, conference abstracts or reviews; (2) they focused on animal models; (3) they did not report a between-group comparison (i.e., longitudinal studies without TDC groups) [72]; (4) they had sample sizes with fewer than 7 participants per group; (5) they performed ReHo analysis on a restricted region of the brain (i.e., ROI analysis) [78]; (6) they used no resting-state fMRI data (e.g., ReHo data derived from task-fMRI); (7) they explicitly indicated, for subjects with ASD, a co-occurring chronic systemic medical illness (i.e., other known neurologic, psychiatric, or genetic disorders). This choice is consistent with the need to characterize homogeneous clinical samples [109].

Particular attention was paid to avoiding spurious results due to overlap in the clinical population, both between and within articles. In the case of multiple experiments included in a single article, only those reporting on independent clinical groups were considered. In the case of multiple articles published using the same clinical group (or part of it), only the earliest published data set was considered.

Data extraction

The articles were first extracted by one author (LD). The full-texts of the relevant articles were then independently evaluated by two authors (LD, MJ). Disagreements were resolved by consensus under the direction of the senior author (CF). Peak coordinates and related T values of abnormal ReHo clusters were extracted from all included experiments. When T-values were not provided, Z- or P values for significant clusters were converted to T values using the statistic converter utility of SDM (https://www.sdmproject.com/utilities/?show=Statistics).

Statistical methods

Coordinate-based meta-analysis

Quantitative synthesis was performed using the signed differential mapping-permutation of subject images (SDM-PSI) software package (v.6.21). SDM-PSI is a recent CBMA method for neuroimaging that allows meta-analytic evaluation of independent results from voxel-wise neuroimaging studies; it benefits from the use of standard effect-size calculation, anisotropic Gaussian kernel approach and meta-analytic random-effect models [95]. The novelty of the method is the use of standard univariate voxel-wise tests [130], which, instead of identifying spatial convergence of the alteration across experiments, detect the presence or absence of the effect for each brain voxel. Full details on SDM-PSI can be found in Albajes-Eizagirre et al. [4].

Here we briefly summarize the procedure. The bounds (lower and upper) of the possible effect sizes for all voxels were evaluated with multiple imputations. Maps of ReHo changes for each study were generated using the anisotropic Gaussian kernel, which assigns higher effect sizes to voxels that appear to be more correlated with peak coordinates. We then applied maximum likelihood techniques to determine the most likely effect size and its standard error. This imputed data set obtained from each study was meta-analyzed with a random-effects model and, then, the obtained data sets were combined using Rubin’s rules. Finally, we performed family-wise error correction for multiple comparisons and thresholded our meta-analysis employing the threshold-free cluster enhancement statistics.

Preprocessing and mean analysis default thresholds were therefore adopted (i.e., Functional MRI modality; SDM gray matter mask; anisotropy = 1; isotropic FWHM = 20 mm; voxel size = 2 mm; number of imputations = 50). Results were corrected for multiple comparisons (family-wise error rate; 1,000 permutation runs). As recently recommended [4], threshold-free cluster enhancement (TFCE) [104] was used in statistical thresholding, setting a TFCE-based FWER P \(\le\) 0.05 and a minimum cluster size of k = 10 voxels.

Heterogeneity and publication bias evaluation

The CBMA values of peak coordinates were extracted to direct heterogeneity statistics and publication bias analyses. We assessed heterogeneity between studies with the I2 statistic using a random-effects model, according to which an I2 < 50% is indicative of low heterogeneity [27]. We then performed funnel plots and Egger tests to estimate publication bias. An asymmetric plot and p < 0.05 were considered statistically significant.

Meta-regression analyses

The potential influences of clinical and methodological variables on ReHo findings were examined via meta-regression analysis. Heterogeneity between studies was explored for mean age at the scan session, gender distribution (i.e., percentage of female), cognitive functioning (i.e., FSIQ mean values), slice thickness and imaging smoothing level (i.e., FWHM), respectively. As suggested by the SDM team, a voxel-level threshold of Puncorrected \(\le\) 0.0005 was adopted to achieve an optimal balance between specificity and sensitivity [94].

Term associations and functional connectivity estimation

The probabilistic estimate of the association between voxels and significant psychological terms was derived from Neurosynth, a meta-analytic tool capable of retrieving results from more than 15,000 published fMRI studies using high-frequency keywords associated with fMRI voxel coordinates (https://github.com/neurosynth/neurosynth) [133]. This estimation is based on the probability that a certain psychological term is reported in association with the activation of a given voxel. The probabilistic estimate of Neurosynth can therefore be viewed as a quantitative indication of how activity in brain areas is functionally related to psychological processes. However, it is worth noting that Neurosynth does not distinguish between activations or deactivations of areas related to the term of interest. Even though Neurosynth reports more than 1,000 words, our focus was limited on cognitive and behavioral terms from the Cognitive Atlas [93], as recently suggested [37]. The resting-state functional connectivity was also estimated for each significant peak identified by our CBMA. We employed the 7-network atlas of Yeo et al. [135], who parceled the human cerebral cortex using rs-fMRI data from 1000 TDCs.

Results

The comprehensive literature search yielded 1190 articles. No additional articles were found in the reference lists of selected studies and relevant reviews. After title/abstract screening, 16 articles were evaluated at the full-text level, of which 8 were excluded based on our selective criteria. Four articles performed fMRI analyses that were not of interest [35, 42, 77, 121], 2 enrolled both pediatric and adult subjects in the experiment [23, 46], 1 performed an ROI analysis [48], 1 collected data while performing a visual task [103]. In total, 8 articles were included in the quantitative synthesis [22, 32, 44, 51, 59, 73, 80, 87], including 11 independent experimental datasets, 455 pediatric subjects with ASD (83 females and 372 males; mean age = 11.76 years), and 474 pediatric TDCs (110 females and 364 males; mean age = 11.94 years). No neurological, psychiatric, or genetic comorbidities were explicitly reported in the ASD groups of the selected ReHo experiments (details in Table S3). No article was excluded due to the presence of medical comorbidities in the ASD sample. The PRISMA flow diagram is shown in Fig. 1.

Fig. 1
figure 1

PRISMA flowchart for data selection in coordinate-based meta-analysis. ReHo regional homogeneity, ROI region of interest, T-fMRI task-based functional magnetic resonance imaging

Full size image

No significant differences were found between ASD and TDC groups with respect to age (mean confidence interval: − 0.179/0.090; Z = − 0.652, P = 0.514). For clinical-demographic and methodological details of included experiments, see also Table 1 and Table 2, respectively.

Table 1 Experiments included in the coordinate-based meta-analysis: demographic and clinical data
Full size table
Table 2 Experiments included in the coordinate-based meta-analysis: methodological data
Full size table

Our dataset included three articles analyzing pediatric subjects from the Autism Brain Imaging Data Exchange database [23]. Therefore, only ReHo experiments with no overlap in the clinical population (both between and within articles) were analyzed. Further details can be found in Table S4.

To note, 67 localized peaks of ReHo changes were found to involve cortical, subcortical and cerebellar regions. Both higher and lower ReHo in the pediatric ASD groups compared with the TDC groups were observed (see also Fig. 2 for the spatial distribution of coordinates included in the current study).

Fig. 2
figure 2

Anatomical distribution of stereotactic coordinates for each experiment included in the present coordinate-based meta-analysis. Nodes in blue reflect a significant regional homogeneity increase in pediatric subjects with autism spectrum disorder compared to typically developing controls. Nodes in red reflect a significant regional homogeneity decrease in pediatric subjects with autism spectrum disorder compared to typically developing controls. ASD autism spectrum disorder, TDC typically developing control, AG angular gyrus, Amy amygdala, CAL calcarine cortex, Caud caudate, Crus1 cerebellar crus I, Cun cuneus, FG fusiform gyrus, FP frontal pole, IFG inferior frontal gyrus, Ins insula, IOG inferior occipital gyrus, IPG inferior parietal gyrus, L-VI cerebellar lobule VI, L-VIII cerebellar lobule VIII, L-IX cerebellar lobule IX, LG lingual gyrus, LOG lateral occipital gyrus, MCC middle cingulate cortex, MCG middle cingulate gyrus, MFG middle frontal gyrus, MTG middle temporal gyrus, Nac nucleus accumbens, OFC orbitofrontal cortex, PCC posterior cingulate cortex, PCL paracentral lobule, Pcun precuneus, PL parietal lobe, PoCG posterior central gyrus, PrCG precentral gyrus, Put putamen, S1 primary somatosensory cortex, SFG superior frontal gyrus, SMG supramarginal gyrus, SOG superior occipital gyrus, STG superior temporal gyrus, SuMG superior medial gyrus, Thal thalamus, Verm verebellar vermis

Full size image

ReHo changes in pediatric ASD

Compared with TDCs, pediatric subjects with ASD showed three clusters of ReHo decrease involving both hemispheres with an altered total volume of 6,160 mm3 (Table 3 and Fig. 3). Local peaks were found in the: (cluster 1) right paracentral lobule (PCL; Brodmann area—BA 4), which is part of the largest cluster along with the left paracentral lobule and right supplementary motor area (SMA); (cluster 2) right superior frontal gyrus (SFG; medial orbital part), which extends to the left SFG and bilateral anterior cingulate cortex (ACC; BA 10); (cluster 3) left posterior cingulate cortex (PCC; BA 23), which extends to the left precuneus and bilateral median cingulate cortex (MCC).

Table 3 Clusters of regional homogeneity reduction in pediatric subjects with autism spectrum disorder compared with typically developing controls
Full size table
Fig. 3
figure 3

Brain clusters of regional homogeneity reduction in pediatric subjects with autism spectrum disorder compared to typically developing controls. Results are TFCE-based FWER corrected at 0.05. The PSI-SDM findings are visualized as hemispheric surfaces (3-D view) (upper panel) and coronal/axial/sagittal slices (2-D cortical and subcortical view) (lower panel). PCL paracentral lobule, PCC posterior cingulate cortex, SFG superior frontal gyrus

Full size image

Note that a single cluster of ReHo increase encompassing the calcarine cortex (BA 17) bilaterally was found at P = 0.005 (Table S5; Fig. S1); however, it did not survive at the TFCE-based FWE correction.

Analysis of heterogeneity and publication bias

Supplemental analyses revealed no significant heterogeneity of effect sizes in the current CBMA (I2 = 1.2 for peak 1; I2 = 1.0 peak 2; I2 = 5.7 peak 3). The results of Egger’s test and funnel plots revealed no obvious publication bias (P = 0.641 for peak 1; P = 0.755 peak 2; P = 0.474 peak 3).

Effects of clinical and methodological variables

Several moderators were examined to understand between-study heterogeneity on published ReHo findings. No linear associations were found with FSIQ, age, gender distribution, smoothing, and slice thickness at Puncorrected \(\le\) 0.0005.

Functional associations and connectivity

According to the term association and connectivity analyses of the Neurosynth database, the right PCL is functionally associated with psychological processes of pain and nociception (see also Table S6 for the concept definitions provided by the Cognitive Atlas). According to the rs-fMRI atlas of Yeo et al. [135], its reliable co-activation is with areas of the sensorimotor network (Fig. 4). The right SFG is associated with psychological terms of reward (evaluation) and autobiographical memory. The right PCC is associated with the Theory-of-Mind, mentalization, autobiographical memory, and empathy terms (Table S6). The large-scale functional connectivity of both nodes suggests a strong involvement of the default mode network [135] (Fig. 4).

Fig. 4
figure 4

Psychological term associations and large-scale functional connectivity that are preferentially related to the peaks of regional homogeneity reduction in pediatric subjects with autism spectrum disorder compared to typically developing controls. Findings were generated using the Neurosynth database. Font size of the terms represents their associated Z-scores. PCL paracentral lobule, SFG superior frontal gyrus, PCC posterior cingulate cortex

Full size image

Discussion

This study provides a unique quantitative synthesis of resting-state ReHo changes in pediatric individuals with ASD. Taking advantage of the current state-of-the-art methods in the field of CBMA, we revealed consistent patterns of local functional under-connectivity across included experiments, predominantly within the default mode and sensorimotor networks. These findings were highly reliable according to the heterogeneity, publication bias, and meta-regression analyses. Moreover, data-driven characterization of the identified areas revealed both sensory and high-level psychological associations that have been widely documented as deficient in the disorder. Altogether, these results are an important first step in resolving discrepancies in the ReHo literature of ASD and, at the same time, emphasize the necessity of reconsidering the theoretical hypothesis of the generalized local over-connectivity in pediatric autism.

An important finding of this investigation is that consistent patterns of local hypo-connectivity accumulate in the core components of the functionally defined DMN, namely the SFG (medial orbital part) and PCC [115]. This result accords well with the growing evidence for a significant contribution of DMN dysmaturation to the pathophysiology of ASD [38, 55, 61, 63, 88]. Both regions are implicated in the disorder, including aberrations in gray matter volume/concentration [63, 116], cortical thickness [12, 118], white matter connectivity [83, 122], intrinsic functional connectivity [71, 127], and task-related activation [49, 128]. It is worth noting that the long-range hypo-connectivity between the PCC and SFG is one of the most widely replicated findings in the fMRI literature on autism [14, 55] and is thought to underlie reduced integration of self- and other-referential processing in children with ASD [14]. These same sites are classified as hub nodes of the human connectome due to their high degree of anatomo-functional connections and pivotal role in maintaining global brain communication [81, 106, 140]. They also have been repeatedly found abnormal in resting-state activity in a wide range of clinical conditions, including depression [43], attention-deficit hyperactivity disorder [45], mild cognitive impairment [53], Parkinson’s disease [91], and schizophrenia [132]. Given these observations, our results are consistent with the cross-disorder dysconnectivity model proposed by Van den Heuvel and Sporns [119], which suggests a possible shared landscape of alteration in the functional connectome across multiple diseases, particularly affecting brain areas characterized by high topological centrality [21] and metabolic activity [110].

The lower regional homogeneity of the PCC is an interesting finding for future research on the role of this brain area in the disorder. This finding extends previous postmortem evidence of altered distribution of neurons within the PCC in pediatric individuals with ASD [30, 86] and is well consistent with the results of EEG research highlighting the PCC as one of the central nodes of disconnectivity in the autistic brain [125]. In this context, it is worth noting that this area is considered the main hub of the DMN due to its greatest functional centrality, which begins to emerge in 2-week-olds typically developing individuals [34]. Despite the broad consensus on its cognitive importance, the exact functional profile of the PCC has not been fully elucidated. The PCC is associated with self-awareness, detection of behaviorally relevant environmental changes, internally directed thought, and regulation of internal and external focuses of attention [58]. This picture is strongly consistent with our functional association analysis, in which high-level psychological terms were associated with the right PCC as well as with the typical clinical impairment of ASD. From a clinical perspective, dysfunction of the PCC has been associated with deficits in the Theory-of-Mind (ToM; i.e., the ability to infer people’s emotional states, thoughts or beliefs), mentalization (i.e., the ability to understand the mental state of others or oneself), and empathy (i.e., the ability to be sensitive to people’s feelings). The scientific literature showing that individuals with ASD have significant deficits in these interrelated functions is extensive and robust, e.g. [8, 17, 105]. From a neuroimaging perspective, a number of studies suggest that individuals with ASD exhibit abnormal activation and connectivity in certain DMN regions (i.e., PCC, precuneus, angular gyrus, and medial prefrontal cortex), thought to play a role in tasks related to ToM, empathy, and mentalizing [29, 31, 41, 49, 101, 128].

Another area that exhibited lower regional homogeneity was the SFG. Considering the conflicting results of the EEG and MRI literature on functional connectivity in ASD [85, 113, 123], the detection of short-range under-connectivity in this site is particularly noteworthy. Interestingly, in the mouse model with 16p11.2 deletion, one of the most common chromosomal copy number variations in ASD, the SFG was associated with disrupted functional connectivity [11]. In the same line of research, one study found that loss of the scaffolding protein SHANK3, which is commonly associated with ASD, can lead to impaired functional connectivity and abnormal neuroanatomical structures in prefrontal areas [89]. Although it is tempting to speculate that these genetic impairments might affect prefrontal functional connectivity in individuals with ASD, further multimodal research is needed. Moreover, recent research has revealed that thousands of genomic risk variants profoundly impact functional brain connectivity in a set of psychiatric conditions, including ASD [76].

The SFG disconnectivity may be predictive of deficits in social communication and interpersonal interaction, which, once again, is a distinctive feature of ASD [114]. Moreover, functional association analysis links this area and other frontal components especially to the processes of autobiographical memory and reward. There is evidence that the experience of recollecting personal events is reduced in ASD [19]. In particular, the deficit regards the ability of retrieving memories as well as of reconstructing autobiographical records. This, in turn, may lead to difficulties in simulating future scenarios, as this ability requires a system that can rely on past experiences, so as to identify and distinguish relevant elements that can be used to predict future outcomes [100]. The prefrontal cortex also plays an important role in modulating the reward process [15]; its dysfunction contributes to disrupting corticostriatal pathways, which may lead to the apparent difficulties in experiencing social reward that affect people with ASD [60]. Finally, growing evidence points out that changes of ReHo most likely induce alterations of remote functional connectivity [47]. Therefore, our findings of lower regional homogeneity of PCC and SFG further support the increasing body of experimental inquiries indicating aberrant functional network synchronization in ASD when frontal or parietal components are affected, as dysregulation of these areas may lead to reduced long-range coupling between prefrontal and parietal associative regions.

A consistent pattern of local hypo-connectivity was also revealed in a mainly right-sided cluster that included the PCL. The emerging picture of this U-shaped convolution, located between the precentral and postcentral gyri, provides an important framework for interpreting the clinical significance of our results. Indeed, this node of the sensorimotor network exhibits altered functional connectivity in ASD, particularly in the early stages of the disorder [33, 39, 57, 73, 80, 136]. Moreover, this area has been proposed as a site for noninvasive brain stimulation to alleviate sensory symptoms in pediatric individuals with ASD [39]. Previous studies have shown that this highly interconnected region is involved in pain-evoked activity, aversive emotional processing, and painful sensations [98, 102, 139]. Consistent with these reports, Neurosynth analysis highlights a statistical association of the right PCL to the psychological processes of pain and nociception. This finding, although indirect, is important because the role of aberrant processing and sensitivity in ASD is highly understudied, especially in children and adolescents [129, 134]. Still, considering that both sensory symptoms occur in approximately 69% of children with ASD [7], and that pain-related levels may be considered as a predictor of poor health outcome in adolescents with autism [111], there is a need for future fMRI research that directly addresses the neural pain signature in this disorder, for example, by adopting our coordinates as regions-of-interest.

Our findings are only partially consistent with those of previous CBMAs on rs-fMRI of ASD [54, 124]. In particular, we revealed hypo-connectivity at the level of the PCL and PCC clusters in line with Lau et al. [54], even though their result showed a more extensive volume of alteration and that only 3 out of the 11 experiments included here were previously meta-analyzed by the authors. In contrast to Wang et al. [124], we did not replicate the findings of hypo-connectivity at the level of the right middle temporal gyrus and bilateral cerebellar crus I, and hyper-connectivity in the left precentral gyrus, right inferior frontal gyrus and bilateral cerebellar lobule IX. In addition, we detected a cluster of hypo-connectivity in the SFG that was not found previously.

These discrepancies could be explained by several factors. First, previous investigations have adopted a multimodal resting-state perspective, that is, they have synthetized findings from ReHo, arterial spin labeling (ASL), independent component analysis (ICA), and amplitude of low-frequency fluctuations (ALFF) techniques. Of note, Wang et al. [124] also used coordinates from positron emission tomography (PET) and single-photon emission computed tomography (SPECT) experiments. Second, Lau et al. [54] and Wang et al. [124] used the activation likelihood estimation and effect-size version of SDM, respectively. These CBMA methods test for spatial convergence of alteration across coordinates. In contrast, our CBMA method performs standard univariate voxel-wise tests. From a methodological perspective, this means that SDM-PSI is able to overcome certain spatial drawbacks that can lead to either conservative or liberal results and reduce the statistical power of CBMA [2, 4]. Third, early CBMAs have summarized findings from a pediatric, adult, and mixed-age groups altogether. Although this is a valuable choice that provides reliable results in terms of generalization to the clinical condition of interest, it limits the accurate characterization of the autistic brain phenotype, which is known to differ across neurodevelopmental stages [14, 22, 52, 62, 82, 84, 113].

We found no significant clusters of local hyper-connectivity. This is an unexpected result considering that 10 of the 11 included experiments reported at least one coordinate of ReHo increase. Probably, a moot point is a calcarine cortex that, although locally hyperconnected in a number of studies [44, 51, 73, 80], did not survive our rigorous statistical thresholding procedure. One possible reason for this result is that the exact loci of local maximum differed considerably across samples. This and the fact that our design used x–y-z coordinates instead of three-dimensional parametric maps could explain why no effect was detected in this study. Another aspect could be related to the status of the participants’ eyes during MRI acquisition. As Nair et al. [80] have elegantly shown, ReHo measurements may be susceptible to eye openness/closure due to differential effects on local activity synchronization, especially in the visual-related regions. Unfortunately, the limited number of experiments, as well as the lack of specification of this variable in some studies, has hampered the ability to perform subanalyses on this topic. Therefore, further research is needed to better characterize this confounding factor.

Finally, it is worth noting that our findings do not support the currently prevailing theory of general local over-connectivity in ASD. Starting from the review of Courchesne and Pierce [20], several authors have proposed that the autistic brain, and in particular the frontal cortex, is characterized by excessive connectivity and disorganized heightened excitability. However, upon further review, we note that this hypothesis has survived over time via cross-citations [74, 97, 120, 126] and was based primarily on findings from anatomical and post-mortem microscopic studies due to the limited availability of rs-fMRI research data at the time of its first conception. Therefore, at least in pediatric individuals, ReHo research indicates that the notion of local over-connectivity needs to be reconsidered based on current data.

Limitations and future challenges

The current findings need to be contextualized with respect to some limitations. CBMA methods, by definition, have limited accuracy because they employ stereotactic coordinates instead of original statistical parametric maps of alteration. On the other hand, we need to observe that the high standardization of this method may limit the probability of spatial errors [28, 94]. Also, current findings are based on 11 experiments due to the limited availability of published research data. In performing the SDM-PSI analysis, we cannot rule out the possibility that considering a small set of experiments may bias the effect sizes slightly towards zero, even though a simulation using an algorithm with the maximum likelihood/multiple imputations has shown that this type of bias is almost negligible [3]. Meta-regression analyses revealed no apparent heterogeneity in demographic-clinical-methodological variables between experiments. These results are based only on a limited number of data and, therefore, should be taken with caution and should be confirmed with further studies. As highlighted throughout the text, the converging MRI-based literature suggests that functional brain connectivity in ASD should be characterized from a developmental perspective [22, 82, 113]. This study contributes in part to this view due to its cross-sectional nature and circumscribed focus on pediatric cohorts. Future ReHo research in adults with autism, as well as longitudinal studies examining the same individuals across the lifespan, is urgently needed to understand the precise developmental trajectory of local connectivity in ASD. Finally, one of the long-term goals of the fMRI connectivity approaches in ASD is to provide valuable insights for clinical practice. Therefore, future investigations could enroll autistic subjects with other common medical comorbidities to test in detail the impact of these co-occurrences on the brain landscape of people with ASD.

Conclusions

Our analysis of resting-state ReHo changes in pediatric individuals with ASD provides valuable insight into an area that remains poorly explored. Somewhat unexpectedly, no significant local hyper-connectivity was found, despite the existing hypothesis of generalized local over-connectivity. On the contrary, patterns of local functional hypo-connectivity were observed mainly in the bilateral PCC and SFG, as well as in the bilateral PCL. Functional characterization of these regions revealed associations with sensory and socio-executive domains known to be affected by ASD. Thus, our results provide an insightful step toward a better understanding of the complex pathophysiology of this multifaceted spectrum.