Postoperative cerebellar mutism syndrome is an acquired autism-like network disturbance

Abstract

Postoperative cerebellar mutism syndrome (CMS),¹ often referred to as posterior fossa syndrome, is a serious complication that occurs in up to 30% of children undergoing surgery for posterior fossa tumors.² CMS remains an incompletely understood set of symptoms within a heterogeneous constellation of conditions that comprise the broader category of cerebellar cognitive affective syndromes (CCAS).^3,4 Although the hallmark of CMS is delayed-onset transient mutism, affected individuals may also manifest emotional lability and impaired socialization.⁵ Affected children commonly suffer from long-term impairments in speech and neuropsychological deficits^2,6,7 that require significant and protracted rehabilitation.^8,9

The substrates of CMS remain incompletely understood.⁴ Various presurgical risk factors have been reported,¹⁰ including younger age at the time of surgery,^11,12 larger tumor volume,^10,11,13 midline location,^10–12 medulloblastoma histology¹⁰ and molecular subtype.^11,12 Structurally, bilateral cerebellar hemispheric insults¹⁴ and injury to vermis and dentate nuclei^10,15–17 is strongly associated with the development of CMS. It is thought that CMS occurs as a result of bilateral damage to the proximal efferent cerebellar pathways^18,19 and dentato-rubro-thalamo-cortical tracts.^7,13,19 Due to cerebro-cerebellar diaschisis,¹⁹ injury to these fibers, which project to the red nucleus, thalamus, and cerebral cortex²⁰ is thought to impair remote supratentorial functions.^21–25 Recent evidence also highlights the significant role of the fastigial nuclei and their output projections in the development of CMS.²⁶ In children with CMS, cerebral abnormalities have been detected in the frontal lobe,^19,27–29 parietal cortex,²⁷ cingulate cortex,²⁹ temporal lobe,²⁸ brainstem,⁷ periaqueductal gray,³⁰ basal ganglia,²⁹ and thalamus.^18,29

The phenomenology of CMS overlaps with symptoms exhibited by children with Autism spectrum disorder (ASD).³¹ ASD is clinically diagnosed based on a set of defined behaviors and can be caused by heterogeneous neurodevelopmental conditions. One of its hallmarks is difficulty with communication and social interaction and another characteristic is a repetitive and restricted pattern of behaviors.³² Similar to CMS, social and communication deficits in ASD encompass social–emotional reciprocity, nonverbal communication behaviors and difficulty developing and maintaining relationships.³²

Disruption in cerebellar-cortical circuitry has long been associated with ASD. Cerebellar lesions in infancy are the leading noninheritable risk factor for ASD and ASD-like behaviors.^33,34 Furthermore, translational rodent models of ASD have shown that disrupted cerebellar Purkinje neuron function is sufficient to generate ASD-related social and repetitive behaviors,³⁵ which can be rescued by cerebellar stimulation.³⁵ A structural MRI analysis of 26 separate murine models of ASD showed consistent abnormal volume in the cerebellar cortex.³⁶ Furthermore, there may also be a developmental window during which acquired cerebellar lesions lead to ASD-like behaviors in children³⁴ and rodent models of ASD.³⁷ Younger children are particularly vulnerable to CMS,^11,12 a syndrome which rarely manifests after posterior fossa surgery in adulthood.²⁹

Recently, lesion network mapping (LNM) has enabled the identification of shared circuits that are disrupted by brain injuries occurring in different locations that give rise to similar symptoms.³⁸ By leveraging the human connectome, a map of human brain connectivity, large-scale disruptions in neural circuits caused by focal lesions can be evaluated and contrasted across different clinical populations. For example, heterogeneous lesions leading to speech and language deficits, with little spatial overlap, were found to share functional connectivity to specific brain regions involved in aphasia.³⁹ This approach may also yield novel therapeutic targets for treatment based on functional mapping of the core networks disrupted by lesions that lead to the expression of specific symptoms.⁴⁰

Given converging evidence for (i) a large-scale network basis for CMS, (ii) phenomenological overlap between CMS and ASD, and (iii) compelling evidence of cerebellar involvement in ASD-like behaviors, we sought to leverage LNM to identify brain networks associated with lesions causing CMS and assess their similarity to ASD. We hypothesized that acquired lesions in children who develop CMS are preferentially connected to specific large-scale networks that relate to ASD-associated behaviors. Furthermore, we hypothesized that these networks would be uniquely vulnerable during critical periods of neurodevelopment, reflecting the preponderance of CMS and ASD in pediatric populations. Taken together, the current work presents a novel connectomic-informed analysis of CMS and its associations with ASD, and as such, provides common and unique insights into both conditions.

Materials and Methods

Participant Cohort

Imaging and clinical data were collected from 90 children (age range 1–17 years) who underwent surgery for resection of posterior fossa medulloblastoma at the Hospital for Sick Children from 2010 to 2022. These data are part of a larger dataset previously analyzed to investigate the molecular correlates of CMS.¹¹ The cohort contained 55 males and 35 females. Postoperatively, a subset of children (n = 30) were diagnosed with CMS per the Delphi consensus conference criteria.¹ Specifically, the diagnosis of CMS was determined based on documentation of required mutism and at least 1 additional CMS symptom in the perioperative neuro-oncology and/or neurosurgery notes. Complete demographic data for all patients is available in Table 1. The study procedure was fully approved by an independent Research Ethics Board.

All children underwent pre and postoperative MRI with high spatial resolution 3D T1 (TR 4.95 ms, TE 2.3 ms, 384 × 383 matrix, 285 slices, 0.57 mm  × 0.57 mm  × 0.5 mm resolution) and T2-weighted fluid-attenuated inversion recovery (FLAIR) sequences (TR 1000 ms, TE 141 ms, TI 2850 ms, 432 × 432 matrix, 52 slices 0.51 mm  × 0.51 mm  × 3 mm). Cerebellar lesions were segmented from all postoperative scans (Figure 1A). Lesions were defined as hyperintense areas surrounding the surgical site on postoperative T2-FLAIR images that were not present preoperatively as a surrogate measure of postoperative tissue injury. Manual segmentation was performed by 4 reviewers with expertise in neuroimaging (H.S., K.M., N.M.W., and H.Y.), which yielded binary masks (Freesurfer, https://surfer.nmr.mgh.harvard.edu/).^41,42 Consensus on the margins of the lesion was reached between at least 2 reviewers. To understand whether these acute FLAIR lesions persisted, we also segmented delayed FLAIR hyperintensities on scans that were performed 4 weeks–6 months following surgery in a cohort of 82 patients. The concordance of these lesions with the acute-phase FLAIR lesions was then evaluated.

Figure 1.

Lesion localization. (A) Segmentation and registration of postoperative cerebellar lesions to MNI space. (B) Lesion voxels are common in children who developed CMS. The map on the left denotes the regional differences between CMS+ Lesions and CMS– lesions, with red-yellow indicating a higher presence of CMS+ lesions and blue indicating a higher presence of CMS– lesions. CMS+ lesions had greater odds of involving the vermis and paramidline cerebellar lobules. (C) CMS+ lesions showed a higher degree of overlap with each other than CMS– lesions

Open in new tabDownload slide

To ensure that our findings did not preclude areas of the cerebellum that were damaged or resected during surgery, we performed an additional segmentation to generate a set of lesion masks that included an estimation of the resected brain. This method involved normalization of cerebellar volumes to the SUIT template,⁴³ facilitating the delineation of cerebellar regions absent on individual MRI scans. This allowed for an estimation of the volume of absent cerebellar tissue, as previously described.⁴⁴

Lesion Odds Ratio

Individual lesions were transformed into common space in 2 steps. First, each individual brain was linearly (9 degrees of freedom) transformed into Montreal Neurological Institute (MNI) space (bestlinreg, https://www.mcgill.ca/bic/software/minc/minctoolkit). As the optimal alignment of the cerebellum was critical, an inclusion mask of the region around the cerebellum (anterior border:anterior pons; superior border:dorsal thalamus; full brain left–right) was used to calculate a full nonlinear transformation within the region using ANTs (https://github.com/ANTsX/ANTs).⁴⁵ Successful alignment of each individual cerebellum within the MNI152 space was visually confirmed.

The binarized individual lesion masks in MNI152 space were then used to create some maps in children who developed CMS (CMS+, n = 30) and those who did not (CMS–, n = 60). The spatial overlap between CMS+ and CMS– lesion volumes was examined using the Dice similarity index⁴⁶ (also known as the Sorensen–Dice coefficient) (Figure 1C). As in prior work,^47–49 a lesion odds ratio (LOR) was calculated and visualized on a flattened cerebellar surface⁵⁰ to describe the likelihood that involvement of a particular voxel is associated with the development of CMS. This was calculated as the ratio of the number of CMS+ lesions involving a voxel, compared to the number of CMS– lesions involving the same voxel:

where, N_p = number of CMS+ lesions; N_c = number of CMS– lesions; V_p = number of CMS+ lesions overlapping a particular voxel; V_c = number of CMS– lesions overlapping a particular voxel.

Lesion Network Mapping

To perform LNM, the volumes corresponding to cerebellar lesions were seeded into 3 different functional neuroimaging datasets (healthy adults, healthy children, and children with ASD). The first dataset (normative adult fMRI) was derived from over 1000 healthy subjects from the Brain Genomics Superstruct Project (https://dataverse.harvard.edu/dataverse/cohenlab).⁵¹ The second (normative pediatric fMRI) included 107 subjects from the Consortium for Reliability and Reproducibility (CoRR).⁵² The fMRI time-series were preprocessed, analyzed and assembled to be used as a ready-to-use connectome for seed-based connectivity analyses.^53,54 The majority of these subjects, n = 56, were below the age of 12 years, mean 12.11 years, and of whom 44 were male, and 63 were female. The third comprised a subset of pediatric participants of the Autism Brain Imaging Data Exchange (ABIDE)⁵⁵ dataset. The subset used included 204 participants with ASD (mean age = 12.54 years) and 223 controls (mean age = 12.50 years) of whom 350 were male, and 77 were female.

In the normative fMRI datasets (both pediatric and adult), we sought to identify the networks that preferentially connected to lesions in children who developed CMS (CMS+) compared to lesions in children who did not develop CMS (CMS–). As previously described, all binarized lesion masks, representing postoperative FLAIR changes in all children, were used as seed regions in the normative rsfMRI data^56,57 using custom MATLAB scripts (The MathWorks, Inc., Version R2018a. Natick, MA, USA). A connectivity r-map was calculated for each individual lesion mask (i.e. seed). This map shows average functional connectivity estimates (correlation coefficients) between the seed and every voxel in the brain based on low-frequency blood oxygen level-dependent (BOLD) signal fluctuations sampled across all fMRI acquisitions in a given dataset. Each r-map was then converted to a t-map (thresholded at t = 5.1, Bonferroni-corrected across the whole brain at p_cor < .05) to identify brain regions significantly functionally connected to the lesion mask. This stringent correction was performed to address possible differences in functional connectivity related to normal variations between connectomes.^58,59 The maps represented areas of the brain statistically significantly functionally connected to each lesion mask. These maps were subsequently binarized and summed for each group (CMS+ vs. CMS–). Similarly, to the computation of the LOR, we used these sum maps to compute connectivity odds ratios (CORs). The COR describes the relative likelihood of each voxel being connected to lesions that are and are not associated with CMS. Specifically, the COR of a voxel is the ratio of the number of CMS+ lesions that are connected to the voxel, to the number of CMS– lesions that are connected to the same voxel (Figure 2). This relationship is represented as follows:

Figure 2.

Connectivity of CMS+ lesions in adults and children. (A) Regions of the brain correlating to the location of the cerebellar lesion were seeded in adult and pediatric connectomes. In children, prefrontal connectivity was predominant, whereas parietal connectivity was predominant in adults. (B) CMS– regions were preferentially connected to widespread cortical regions related to thalamic and limbic brain regions

Open in new tabDownload slide

where N_p = number of CMS+ lesions; N_c = number of CMS– lesions; V_p = number of CMS+ lesions overlapping a particular voxel; V_c = number of CMS– lesions overlapping a particular voxel. The resulting map was thresholded only to show regions with a COR of 2 or higher.

ASD Cohort

In the dataset of children with ASD, we sought to independently reproduce the CMS+ associated network and test whether this circuitry could differentiate individuals with ASD from controls and correlate with different domains of symptoms exhibited in this population. Preprocessed data from the ABIDE connectomes initiative were utilized.⁵⁵ This is a large database of preprocessed and registered fMRI data in people with ASD and controls. Specifically, the CCS pipeline was used, with bandpass filtering but no global signal regression (“filt_noglobal”).⁵⁵ Following quality control, data from 427 participants (204 ASD and 223 healthy matched controls) were included. All included participants were under the age of 18. The primary reasons for subject exclusion were the exclusion of cerebellar volume from the masked fMRI data and registration errors.

To test the hypothesis that the CMS+ network was also related to communication deficits in individuals with ASD, a binary mask of the cerebellar lesion volumes common to at least 10 children who developed CMS were seeded in the ABIDE fMRI data. We selected a threshold of 10 children to ensure that the voxels included in the binary mask accurately represent lesions consistently observed in children who developed CMS. The decision to choose a threshold of 10 was based on a sensitivity analysis, which involved computing the centroid⁶⁰ of the CMS+ lesions over varying thresholds (Supplementary Figure S2).

First-level analysis was performed by correlating the mean time series of the CMS+ lesion volume with the time series of all voxels in the brain. This analysis was performed using FEAT Version 5.98, part of FSL (www.fmrib.ox.ac.uk/fsl). Time-series statistical analysis was carried out using FILM with local autocorrelation correction.⁶¹

Mixed-effects higher-level analysis was subsequently performed to contrast the first-level statistical parameter connectivity maps between ASD and controls and to identify association with ASD subscales using the Autism Diagnostic Observation Schedule (ADOS).⁶² This whole-brain analysis was performed using default parameters of FLAME-1 higher-level analysis in FSL. These algorithms rely on Gaussian random field theory for family-wise error rate (FWE) corrected voxel-wise and cluster-wise inference. Statistical images were thresholded using clusters determined by Z > 2.5 and a (corrected) cluster significance threshold of P < .05.⁶³

Multidimensional Associations Between CMS Network and ASD-Related Symptoms

The multidimensional relations between CMS+ network and ASD symptoms were also assessed using partial least squares analysis (PLS).⁶⁴ PLS is a statistical technique that decomposes correlations between a set of variables and extracts commonalities. Here, the clinical variables of interest were obtained from the ABIDE dataset and included ASD symptomology and intellectual quotients (full-scale IQ) from multiple assessment measures (n = 58).⁵⁵ Domain scores from the Gotham algorithm of the ADOS⁶⁵ (ADOS-Gotham; social affect, restricted/repetitive behaviors and total scores) and calibrated severity scores⁶⁶ were included. In addition, domain scores from ADOS module⁶² and the Autism Diagnostic Interview-Revised (ADI-R; restricted/repetitive behavior, communication, and social interaction)⁶⁷ were included.

The data were separated into predictor matrix X, representing the connectomic data and matrix Y, representing the ASD severity scores. The ABIDE preprocessed connectomes project also provides the average time series from various standard atlases,⁵⁵ including the Automated Anatomical Labeling (AAL) atlas.⁶⁸ For participants in the ASD cohort, we calculated the partial correlation coefficient between all the time series from all regions of the AAL and the CMS lesion time series.⁶⁹ Each row of X contained the subject-specific coefficient between the CMS lesion and 90 regions of the AAL atlas. The columns of Y included data-centered ASD symptom severity scores.

PLS was performed by calculating the covariance between the two sets of variables (X and Y), and then decomposing the resulting heterogeneous covariance matrix, R, through singular value decomposition (SVD).⁷⁰ The SVD produces latent variables (LVs), or components, explaining the greatest amount of correlation between X and Y. Significant LVs were identified with permutation sampling and the stability of the source saliences was tested by bootstrapping. The resampling distribution was used to calculate standard errors and 95% CI for the contributions of each covariate to the component.

Results

Lesions Associated with CMS Involve Midline Cerebellar Structures

Data from 90 children who underwent posterior fossa tumor surgery for medulloblastoma (30 of whom developed CMS) were successfully segmented and registered to a common brain atlas. The subtypes of these tumors were as follows: 13 (14%) WNT, 17 (19%) SHH, 24 (27%) Group 3, and 36 (40%) Group 4. There was no difference in the distribution of tumor types between the CMS+ and CMS– groups (P = .12). Twenty-nine (97%) tumors in the CMS+ group have a midline location compared to 50 (83%) tumors in the CMS– group (P = .14) (Table 1). Children who developed CMS (CMS+) demonstrated distinct postoperative cerebellar lesions compared to those who did not (CMS–, Figure 1A). Importantly, there was no significant difference in the volume of CMS+ lesions (mean 19 716 ± 14 823 mm³) compared to the volume of CMS– lesions (mean 19 938 ± 15 514 mm³; P = .84). There was, however, significantly greater spatial similarity in CMS+ lesions (mean Dice Index 0.24) compared to CMS– lesions, which were more spatially heterogenous (mean Dice Index 0.17; P < .01). (Figure 1C).

Lesion odds ratios were calculated at each cerebellar voxel to determine the likelihood that the involvement of a given voxel is associated with CMS. Lesions occurring in children who developed CMS preferentially involved the inferior vermis, and midline regions of bilateral lobules VIIb, VIIIa, VIIIb, and IX. Lesions in children who did not develop CMS were more likely to involve the more lateral aspects of Crus II, as well as midline lobules I–IV and VI (Figure 1B).

Taking into account estimated volumes of resected cerebellum did not change the spatial extent of the lesions (Supplementary Figure 3B). FLAIR lesions persisted in 72 of 82 children (88%) who also had delayed MRI imaging. The spatial localization of the delayed FLAIR images was similar to those defined immediately following surgery (Supplementary Figure S3A). However, the mean volume of the acute flair lesions was significantly larger (17 629 ± 14 121 mm³ vs. 4602 ± 6027 mm³, respectively, P-value < 0.00001).

Connectome Age Determines Neural Substrates of CMS+ Lesions

We subsequently performed LNM to identify large-scale networks selectively disrupted by lesions occurring in children who developed CMS. The CMS+ lesions demonstrated preferential connectivity to cerebello-rubro-thalamo-cortical circuitry (Figure 2A). In both the normative adult and pediatric connectome, notable associations (COR > 2.0) were identified between CMS+ lesions and the red nuclei (adult: COR left 3.1 right 4.1, pediatric: COR left 6.6, right 4.2), thalami (adult: COR left 3.5 right 2.1, pediatric: COR left 4.6, right 3.1), and limbic circuitry, including the parahippocampal gyri (adult: COR left 5.3, right 5.5, pediatric: COR left 4.3 right 4.2).

Beyond these brain regions, the remainder of connectivity patterns associated with CMS+ lesions diverged significantly between the pediatric and adult connectomes. Specifically, using the pediatric connectome, the prefrontal cortices, particularly the dorsolateral prefrontal cortex, were more strongly connected to CMS+ lesions compared to CMS– lesions (COR left 2.1–4.5, right 2.2–7.0). The pediatric CMS+ network also included the anterior cingulate cortex (COR left 2.5, right 3.0), precuneus (COR left 2.8, right 2.5) and right caudate (COR 4.5). In the adult connectome, the CMS+ lesions showed a preponderance of posterior-central connectivity, predominantly to the inferior parietal (COR left 2.6, right 2.9) and postcentral cortices (COR bilateral 3.3).

Conversely, regions corresponding to CMS– lesions were significantly more likely to be functionally connected to broader cerebellar regions (Figure 2B) (lobules IV–XIII, X and Crus I, COR left 1.6, right 4.6) as well as subcortical structures including the thalami bilaterally (COR left 2.1, right 2.4) and pallidum (COR left 2.1, right 4.9) in both the pediatric and adult datasets. CMS– lesions also encompassed more widespread cortical regions, including the precentral cortex (COR left 3.1, right 5.1), cingulum (COR left 4.2, right 4.4), precuneus (COR left 4.2, right 4.5), occipital cortex (COR left 1.6, right 3.2) and portions of the temporal lobe (COR left 1.8, right 5.1).

Connectivity Within the CMS Network is Associated With ASD Symptomatology

The CMS+ lesion mask was used as a seed in a whole-brain connectomic analysis in the ABIDE dataset of persons with ASD and age- and sex-matched controls (n = 427) (Table 2). Compared to controls, persons with ASD showed weaker patterns of functional connectivity between the CMS+ lesion, dorsolateral prefrontal cortex, anterior cingulate, and posterior cingulate cortices (P < .05, Figure 3A). With increasing connectomic age in this cohort, CMS+ lesions showed stronger connectivity to the brainstem, including the red nucleus, the thalami bilaterally, the parietal region, as well as the posterior cingulate and precuneus. Increasing age was also associated with reduced connectivity of the CMS+ lesion to the motor, supplementary motor regions, and inferior frontal cortices (P < .05, Figure 3B). Additional results examining the relationship between age and connectivity to these regions are provided in the Supplementary Materials (see Supplementary Figure S1 and Supplementary Table S1).

Figure 3.

Connectivity of CMS+ lesions in children with ASD and matched controls. (A) Children with ASD show weaker connectivity from the regions of the brain involved by the CMS+ Lesion volume to the anterior and posterior cingulate cortices. (B) With increasing age, there is stronger connectivity of the CMS+ Lesion volume to the brainstem, including the red nucleus, bilateral thalami, PCC and the precuneus. Increasing age is also associated with decreasing connectivity to widespread cortical areas including the inferior frontal lobe and motor areas

Open in new tabDownload slide

The connectivity of the CMS+ lesion mask was also associated with repetitive behaviors, communication and social deficits (n = 138, Figure 4A–C). Communication deficits were associated with reduced connectivity between the CMS+ lesion and dorsolateral prefrontal cortex, precuneus, and inferior parietal cortex. Social deficits were associated with weaker connectivity of the CMS+ lesions to the precuneus and ventromedial prefrontal cortex. Conversely, repetitive behaviors were associated with hyperconnectivity of the CMS+ lesion to the motor cortex, supplementary motor cortex, prefrontal cortex, and precuneus.

Figure 4.

Associations between ASD-related scores and CMS+ lesional volume connectivity. (A–C) Whole-brain voxel-wise connectivity of the CMS+ lesion volume is associated with communication, social deficits and repetitive behaviors in individuals with ASD. communication (A) and social deficits (B) show a decrease in connectivity in association with greater deficits whereas repetitive behaviors (C) show an increase in connectivity. (D) Multidimensional associations between ASD scores and imaging data reveal a significant latent variable with associations between communication deficits and CMS+ lesion volume connectivity to the prefrontal and parietal cortices.

Open in new tabDownload slide

Using PLS, multidimensional associations between ASD domain scores and connectomic data were identified in the subset of patients who had complete scores across several scales in the ABIDE dataset (n = 58). Two significant latent variables were identified. The first latent variable (P = .019) was characterized by more deficits across all ASD domains. The second latent variable (P = .022; Figure 4D) was specific to communication deficits. The latter was associated with weaker connectivity between the CMS+ lesion volume and prefrontal cortices and inferior parietal cortex and stronger connectivity with occipital and ventromedial prefrontal cortices.

Discussion

In this work, we identify convergent neural circuitry associated both with the development of postoperative CMS and specific ASD symptoms in children. We first applied LNM to identify neural circuitry preferentially connected to postoperative lesions following posterior fossa medulloblastoma tumor surgery. Importantly, we identified a large-scale circuit involving the prefrontal cortex associated with CMS in children which did not exist in the adult connectome. The same large-scale network disrupted in children was independently reproduced in ASD, and predicted communication and social deficits but not repetitive behaviors. These findings provide a novel network basis for the shared phenomenology between CMS and ASD and emphasize the critical role of cerebro-cerebellar circuitry in the development and function of human language and social behavior.

Convergent Cerebellar Involvement in CMS and ASD

The CMS+ lesions predominantly localize to the vermis and midline cerebellum (Figure 1B), which is concordant with recent lesion mapping studies.¹⁸ These findings strongly support the role these structures play in the emergence of CMS and communication deficits in ASD. Specifically, we identified a convergent neural circuit preferentially associated with CMS+ lesions that was associated with the severity of communication deficits in children with ASD.

Interestingly, cerebellar injury is the leading noninheritable risk factor for the development of ASD.³⁴ Specifically, the presence of cerebellar injuries at birth is associated with a 36-fold increase in ASD incidence³⁴ and over 90% of neuropathological studies in persons with ASD have shown a cerebellar abnormality.⁷¹ Hypoplasia of the cerebellum⁷² and cerebellar injuries in preterm infants³³ are associated with higher scores on ASD screening tests and nonmotor dysfunction in cognition, communication and behavior function. Perhaps most importantly, positive ASD screening in preterm children is almost exclusively associated with injuries to vermian and midline structures.^33,72 Meta-analyses of structural imaging have shown decreased volume of the posterior vermis (lobules VI–VII) across children with ASD, particularly under the age of two.⁷³ Decreased gray matter and lobar volumes within the cerebellar vermis are selectively associated with social and communication problems in this population.^74,75 In children with ASD, abnormal vermian recruitment is related to difficulty in processing irony⁷⁶ and facial expressions.⁷⁷ Additionally, alterations in lobule VIIIA are observed in language-impaired children with ASD.⁷⁸ In the context of our findings, these data therefore suggest disruption of a common network associated with vermian and midline cerebellar structures in both CMS and ASD.

Interestingly, animal models have also shown that cerebellum-specific insults are sufficient to generate an ASD-like phenotype. A Purkinje cell-specific knockout of the tuberous sclerosis gene led to ASD-like deficits, which were rescued by the mTOR inhibitor, rapamycin.⁷⁹ Furthermore, midline cerebellar lesions in juvenile rats result in subsequent deficits in social behavior and vocalization.^80,81

In children, the anatomical extent of inferior vermian transgression after surgery is associated with higher rates of CMS¹⁷ and the vermis-sparing telovelar approach for midline fourth ventricular tumors may yield lower rates of CMS.⁸² Given this, our findings provide data to support the use of surgical corridors such as these to in part mitigate the risk of CMS, particularly in younger children. However, Groups 3 and 4 medulloblastoma are associated with high rates of CMS^11,12 regardless of surgical approach.¹² This finding is thought to relate to their embryological origin at the rhombic lip,⁸³ which dissipates into the nodule of the inferior vermis. In our study, this heightened risk is evident as our patient population displayed a 33% incidence rate of CMS and was comprised of 67% Groups 3 and 4 tumors.

Cerebro-Cerebellar Circuitry in CMS and ASD

Our findings advance our understanding of large-scale circuitry disrupted in children with CMS. One of the leading theories of CMS relates to cerebellar diaschisis, whereby postoperative lesions lead to temporary functional depression of the reciprocal pathways that connect the cerebellum to the prefrontal, temporal, and parietal cortices. In patients with CMS, reduced metabolism in the vermis, thalami, and caudate nuclei has been reported.²⁹ While perfusion changes in the caudate and thalamus may resolve in the short term, prefrontal hypometabolism appears to be long-lasting.²⁹

We highlight the critical role of prefrontal and parietal cortices in the development of CMS. There is compelling evidence to suggest that cerebellar involvement in nonmotor functions is mediated by projections to the prefrontal and posterior parietal cortex.^22,84 In nonhuman primates, retrograde transneuronal viral studies have identified cerebellar outputs to the prefrontal cortex.⁸⁵ The significant contributions of the prefrontal cortex to the cortico-ponto-cerebellar system in humans may represent selective evolution toward higher cognitive skills, including social communication.⁸⁶ The dentate also projects to areas 9 m, 9 l, 46 d, 7 b, anterior intraparietal area, medial intraparietal area, and ventral lateral intraparietal area. Motor and nonmotor functions are spatially separated with distinct domains in the dentate nucleus.⁸⁷ The extent of the dentate nucleus dedicated to prefrontal and parietal outputs is comparable to that occupied by outputs to the motor cortex.⁸⁴

We have further shown that mutism in CMS may share a neural basis with communication deficits in individuals with ASD. Differences in cerebello-thalamo-cortical networks involving prefrontal regions have been previously implicated in ASD. Early studies involving positron emission topography showed that individuals with ASD had decreased right dentate nucleus activation concomitant with decreased left thalamic and frontal activation during both receptive and expressive language.⁸⁸ Compromise in white matter microstructure from the dentate nucleus to the thalamus correlates with the severity of language impairments in children with ASD under the age of 5 years.⁸⁹ In strong concordance with our findings, it has been reported that in low-functioning children with ASD, gray matter reduction in the vermis occurs in tandem with reductions within the inferior frontal, parietal, and mesial temporal structures.⁷⁵ Similarly, children with ASD associated with acquired cerebellar lesions demonstrate reduced white and gray matter volumes in the dorsolateral prefrontal, premotor, sensorimotor, and mid-temporal cortices, contralateral to the site of injury.⁹⁰ Furthermore, neurochemical markers have shown a significant relationship between neuronal integrity in the cerebellar vermis and the frontal lobe in children with ASD.⁹¹

CMS During Critical Windows for Network Development

Our findings provide a putative rationale for the preponderance of CMS in children.^11,12 Critically, we have shown that the neural networks connected to the same CMS+ lesion are different in the adult and pediatric connectomes. Disruption of the CMS network, which may exist only transiently in childhood, may therefore be necessary for the development of CMS symptoms. Indeed, there is substantial evidence that the cerebellum may act in early life to shape the function of other brain regions, especially those related to cognition and affect.³⁴ The developmental influence of the cerebellum on the maturation of distant brain networks has been previously termed “developmental diaschisis.”³⁴ Disruption of this circuitry during critical windows of network formation may be required to manifest CMS before reciprocal pathways for normal nonmotor learning are firmly established.

The red nuclei and thalami likely play a critical role in mediating developmental diaschisis as they demonstrate stronger connectivity with CMS+ lesion volumes as age increases (Figure 2A). Concurrently, frontal and motor regions of the brain show decreased connectivity with development. Disruption of prefrontal cerebro-cortical connectivity during a transient period of network emergence may result in CMS in younger children. In adults, the prefrontal contributions to the CMS network are less robust and therefore, the same anatomical lesion may not yield a comparable clinical deficit.

The reason why connectivity gradients increase between the cerebellum and parietal cortices over development is unknown. It has been reported that the inferior parietal lobule forms a critical component of the mirror neuron system, the dysfunction of which is implicated in ASD.⁹² With the development of the inferior parietal lobule within the mirror neuron system, increasing cerebellar connectivity gradients have been reported.⁹³ We speculate that these connectivity gradients may play a role in normative development during critical windows for the acquisition of social cognition and communication skills.

Interestingly, cognitive and social consequences of cerebellar injury show an opposite age dependence from motor consequences. Disruption of motor-related cerebellar regions may lead to ataxia, dysarthria, and other coordination problems, which improve with time. Cognitive and social functions of the cerebellum, however, are particularly vulnerable to early life perturbation and persist thereafter.³⁴ In preclinical studies, separate sensitive windows have been reported for motor and nonmotor symptoms of cerebellar injury,³⁷ which may explain the dominance of mutism as the hallmark of CMS. Similarly, there may be a sensitive window whereby acquired cerebellar lesions increase the risk of ASD.³⁴

We also acknowledge that CMS may occur in adults, albeit at lower rates.²⁹ Our group-level analyses highlight developmental network vulnerabilities that may predispose children to CMS. At the individual level, similar or different connectomic profiles may pose a risk to CMS throughout the lifespan. Greater efforts are needed to explore the network disturbances in adults with CMS and how they relate to or differ from CMS in childhood.

Shared Symptoms Converge on a Common Circuit

The current study highlights the similarity in brain circuitry associated with mutism in CMS and communication deficits in ASD. Other symptoms of ASD, including repetitive behaviors are likely involve a different neural basis, which is not shared with CMS. For example, we found that restricted/repetitive behaviors are associated with stronger connectivity of motor regions to the CMS+ lesion. Several studies have shown that the communication deficits and repetitive behaviors in ASD are dissociable.⁹⁴ Decreased gray matter and lobar volumes within the cerebellar vermis are selectively associated with social and communication problems in ASD, whereas larger vermian volumes seem to be associated with more severe restricted/repetitive behaviors.⁷⁴ Similarly, it is also likely that the other symptoms of CMS are mediated by different neural substrates. For example, apathy and behavioral dysregulation in children with CMS may relate to connectivity between the CMS+ lesion and mesial temporal structures.

Recently, an alternative theory was proposed by McAfee et al.³⁰ for the development of CMS, related to loss of cerebellar control over the periaqueductal gray (PAG) matter, affecting the expression of primitive reactions to stress, which includes suppression of vocalization. Our findings of a significant increase in midbrain–cerebellar connectivity with age further highlight the importance of viewing CMS through the lens of developing brain networks. We speculate that alterations in connectivity to midbrain structures, including PAG, may also contribute to age-dependent vulnerabilities in the incidence of CMS.

Future Directions and Limitations

The current work presents opportunities for neuromodulation in CMS with translational relevance to ASD. In contrast to communication deficits in children with ASD, the resulting mutism in CMS is often recoverable, suggesting the existence of compensatory cerebellar or extra-cerebellar mechanisms elsewhere in the brain. Analysis of these compensatory mechanisms may inform novel treatment targets for children with ASD. The findings of age-related strengthening of connectivity between the offending CMS lesion and red nucleus and thalami present a particularly compelling target for neuromodulation. Indeed, in preclinical models of ASD, central thalamic stimulation resulted in altered plasticity in corticolimbic and corticostriatal circuitry and improved social behaviors in rodents.⁹⁵

There are several limitations inherent to this work to acknowledge. We dichotomized children who did and did not develop CMS but did not measure the severity of their mutism, other associated symptoms, or extent of recovery. Future work linking large-scale network disturbances to these measures is warranted. Furthermore, we used the FLAIR sequence to index the offending CMS lesion volume, as these images were of adequate spatial resolution to define the lesion volume and subsequently coregister the images to a common atlas. Alternate sequences such as diffusion-weighted imaging could be leveraged to understand the effects of cytotoxic changes on large-scale brain networks. However, the resolution of the available diffusion imaging was insufficient to derive accurate segmentations from these images. Moreover, while normative data can be leveraged using LNM to understand affected brain circuitry, future studies using fMRI data from children with CMS could further validate these findings. Longitudinal functional imaging in particular would be valuable to assess compensatory mechanisms that may facilitate recovery. Our analysis also leverages the ABIDE database to draw comparisons between CMS and ASD. While the ABIDE dataset offers invaluable insights, it has an inherent limitation: it only includes patients aged 7 and older. This is most likely due to challenges in acquiring fMRI data from younger patients, who often require anesthesia. In contrast, our CMS dataset includes patients as young as one year old. The absence of younger ASD and control patients in ABIDE thus represents a limitation in our study. Finally, our study predominantly focuses on children with autism spectrum disorder (ASD), broadening the scope to encompass larger databases comprising varied disorders with overlapping symptomatology could provide deeper insights into shared neural circuits.

Conclusion

Using LNM, we identified a CMS network that is distinct to children and replicated in an independent population of individuals with ASD, which was related to communication and social deficits. Our findings highlight the role of cerebro-cerebellar connectivity in ASD-like behaviors and suggest a sensitive window for disruption of large-scale circuitry in the expression of CMS. Taken together, our findings provide evidence that mutism in CMS shares a common network with communication deficits in ASD, which is likely dysregulated during a sensitive window for development.

Conflict of interest statement

G.M.I has received consulting fees, and honoraria from LivaNova, Medtronic, and Synergia, unrelated to this work. G.M.I has also received a grant from LivaNova, also unrelated to this work. C.G. has received consulting fees and honoraria from Medtronic, and honoraria from Ipsen, unrelated to this work.

Funding

None declared.

Author Contributions

H.S. and B.R.M. performed data collection, data analysis, manuscript preparation and revisions. K.M. J.G., A.B., A.L., performed data analysis. N.M.W. and H.Y. performed data collection and data synthesis, and revisions. F.V.G., J.Y., J.Q., F.M. J.L., A.M.L., S.L., D.M., C.G., U.B., A.H., U.T., J.T.R, J.M.D, A.V.K, P.D, M.D.T, V.R, and M.C.D. aided in the interpretation of data, preparation of the manuscript and revisions. G.M.I contributed to the conception and design of the study, data analysis, manuscript preparation and revision, and provided relevant guidance.

Data Availability

Deidentified data and statistical analysis code available upon request

References

Author notes