Cerebellar mutism is linked to midbrain volatility and desynchronization from speech cortices

Abstract

Cerebellar mutism syndrome is a disorder of speech, movement and affect that can occur after tumour removal from the posterior fossa. Projections from the fastigial nuclei to the periaqueductal grey area were recently implicated in its pathogenesis, but the functional consequences of damaging these projections remain poorly understood.

Here, we examine functional MRI data from patients treated for medulloblastoma to identify functional changes in key brain areas that comprise the motor system for speech, which occur along the timeline of acute speech impairment in cerebellar mutism syndrome. One hundred and twenty-four participants, all with medulloblastoma, contributed to the study: 45 with cerebellar mutism syndrome, 11 patients with severe postoperative deficits other than mutism, and 68 without either (asymptomatic). We first performed a data-driven parcellation to spatially define functional nodes relevant to the cohort that align with brain regions critical for the motor control of speech. We then estimated functional connectivity between these nodes during the initial postoperative imaging sessions to identify functional deficits associated with the acute phase of the disorder. We further analysed how functional connectivity changed over time within a subset of participants that had suitable imaging acquired over the course of recovery. Signal dispersion was also measured in the periaqueductal grey area and red nuclei to estimate activity in midbrain regions considered key targets of the cerebellum with suspected involvement in cerebellar mutism pathogenesis.

We found evidence of periaqueductal grey dysfunction in the acute phase of the disorder, with abnormal volatility and desynchronization with neocortical language nodes. Functional connectivity with periaqueductal grey was restored in imaging sessions that occurred after speech recovery and was further shown to be increased with left dorsolateral prefrontal cortex. The amygdalae were also broadly hyperconnected with neocortical nodes in the acute phase. Stable connectivity differences between groups were broadly present throughout the cerebrum, and one of the most substantial differences—between Broca’s area and the supplementary motor area—was found to be inversely related to cerebellar outflow pathway damage in the mutism group.

These results reveal systemic changes in the speech motor system of patients with mutism, centred on limbic areas tasked with the control of phonation. These findings provide further support for the hypothesis that periaqueductal grey dysfunction (following cerebellar surgical injury) contributes to the transient postoperative non-verbal episode commonly observed in cerebellar mutism syndrome but highlights a potential role of intact cerebellocortical projections in chronic features of the disorder.

Introduction

Cerebellar mutism syndrome (CMS), also known as posterior fossa syndrome (PFS), is a disorder of speech production, volitional movement and emotional regulation that can emerge after surgery within the posterior fossa.¹ CMS is most often seen in young patients following removal of medulloblastoma,^2,3—an embryonal tumour of the cerebellum, and the most common malignant brain tumour in children.⁴ CMS occurs in roughly 25–35% of medulloblastoma patients^2,3,5 and is known to be triggered by injury to the cerebellar nuclei and/or the primary cerebellar outflow tracts, especially the proximal superior cerebellar peduncles and fastigial nuclei.^6–8 The acute symptoms of CMS appear within a few days of surgery and improve over the course of weeks or months, with speech being regained in some capacity by nearly all patients,^9,10 but long-term neurocognitive outcomes in CMS patients are generally worse than in those that did not experience CMS.¹¹

We recently showed that surgical damage to the fastigial nuclei and their projections to the periaqueductal grey area (PAG) is associated with postoperative mutism in medulloblastoma patients,⁶ and hypothesized that ‘core’, transient symptoms of CMS—including acute speech impairment, apraxia and emotional lability—may result from postoperative functional disturbance in cerebello-PAG circuits. Involvement of the PAG in speech impairment is an intriguing new possibility, as this evolutionarily primitive area appears to play a fundamental role in the control of vocalized sound in animals.^12,13 In small mammals such as rodents, the PAG serves to gate the expression of social vocalization^14,15 and voluntary movement,^16,17 a set of functions that is thought to limit conspicuous behaviour in the presence of potential predators.^17,18 In humans, where phonation is requisite for speech, the PAG functions as a neural pattern generator that activates respiratory and arytenoid muscles to drive breath support and vocal cord adduction for phonation.¹³ Vocal cord adduction is also necessary for effective swallowing,¹⁹ and is thought to be enacted by medullary nuclei downstream of the PAG.^20–22 Pertinently, as characterized in a previous article, the prevalence of dysphagia increases with the degree of speech impairment in CMS² (67.5% in patients with complete mutism, 25% in patients with partial mutism, and 5.2% in all other patients; see the ‘Materials and methods’ section for further diagnostic group details). Vocalization becomes meaningful speech via the modulation of vocal sounds by articulatory muscles of the mouth and throat. Volitional control of these muscles for expressive language is coordinated by cortical areas for speech like Broca’s area and the motor cortices for the face, mouth and throat.¹³

A review of the anatomy of speech motor system by Holstege and Subramanian¹³ highlights this functional segregation of phonation and articulation of speech, with coordinated activation of these functional pathways by medial frontal cortices. Crucially, this careful anatomic mapping offers us a framework for evaluating whether postoperative speech deficits are reflected in patterns of blood oxygen level-dependent (BOLD) functional MRI (fMRI) activity in either or both branches of this system. We recently suggested cerebello-PAG dysfunction as a likely mechanism for acute speech disruption in CMS, as opposed to dentato-rubro-thalamo-cortical diaschisis.^6,23

Here, we conducted an exploratory study to determine whether the PAG exhibits signs of dysfunction in patients with CMS, particularly in the form of BOLD volatility and functional connectivity (hereinafter connectivity) with other brain regions involved in speech production. To do so, we evaluated fMRI data gathered in a large prospective study of patients undergoing treatment for medulloblastoma. Using asymptomatic medulloblastoma patients as our control group, we evaluate the activity level within the PAG and its connectivity with other brain areas involved in speech during the initial acute phase of the disorder. Then, we incorporate imaging data acquired after speech recovery to extend the analysis to other brain areas involved in speech and motor planning, and to identify possible pathways involved in speech recovery.

Materials and methods

Participant data and diagnoses

Serial postoperative magnetic resonance images were acquired in 183 children and adolescents enrolled in a prospective study of patients undergoing treatment for medulloblastoma (SJMB12; NCT 01878617). All patients underwent a standardized postoperative neurological examination by a certified paediatric neurologist (R.B.K.) within 2 weeks of arrival at our facility. The interval between the last date of surgery and the neurological examination ranged from 8 to 276 days (median 26 days). The examination was described in detail recently² and included a questionnaire to capture possible history of mutism with resolution prior to examination. Each patient was assigned to one of the following diagnostic groups based on their examination: asymptomatic (patient experienced no remarkable speech, behavioural or motor deficits); complete mutism (patient experienced an episode of complete mutism, usually with significant motor and behavioural symptoms); partial mutism (patient experienced abnormal inhibited speech with inability to complete a three-word statement, usually with significant motor and behavioural symptoms); atypical PFS (patient experienced behavioural symptoms and ataxia in the absence of speech deficits); and ataxia only (patient experienced ataxia and absence of independent gait without speech deficit or behavioural symptoms). Imaging time points were defined based on time relative to enrolment in the trial. Structural and functional imaging was acquired upon arrival at our institution after tumour resection (0-month time point), after completion of craniospinal radiation therapy (3-month time point), and at 12-, 18- and 24-month time points for routine follow-up. All participants with structural and functional imaging of sufficient quality were used for independent component analysis and node discovery analysis. For activity and connectivity analyses, only image data acquired from patients with diagnoses of complete or partial mutism were considered as part of the ‘CMS’ group, as the symptoms in both of these groups are consistent with those in the consensus statement definition of paediatric postoperative cerebellar mutism syndrome put forth by the Posterior Fossa Society.⁹ The duration of speech disruption in the CMS group ranged from 21 to 312 days (median 71 days), based on records of loss and recovery of participant ability to complete a three-word statement. Since speech ability tends to improve rapidly around the time of the first three-word statement, imaging data from CMS group participants who reached this recovery benchmark over a week prior to initial scanning were only used to evaluate post-recovery midbrain activity and persistent connectivity. Time of the initial scan relative to three-word recovery in the CMS group ranged from −291 to 4 days (median −52 days). No MRI was acquired prior to symptom onset in the CMS group.

The asymptomatic medulloblastoma group served as our control. For consistency, all CMS-related analyses were further restricted to scans acquired under sedation with propofol, as this was necessary in most initial postoperative imaging sessions for patients with CMS (90.3% of participants with complete mutism and 70.0% of those with partial mutism). We restricted our analyses to scans acquired under sedation with propofol since propofol has been shown to affect BOLD measures,²⁴ but has minimal impact on cerebral haemodynamics,²⁵ and can be used for functional imaging in paediatric patients when sedation is required for diagnostic-quality MRI.²⁶ Additional exclusions were required for technical reasons such as insufficient field-of-view or imaging artefacts that obscured brain tissue. The total number of patients and scans used in each analysis is listed in Table 1. Diagnostic breakdown of the study dataset and participant subgroups for each analysis are shown in Supplementary Fig. 1A. No group differences in age were found after exclusions for image quality, diagnosis and use of propofol during acquisition were performed (Supplementary Fig. 1B).

Image acquisition

Up to five imaging sessions were conducted per participant, at time points of approximately 0, 3, 12, 18 and 24 months post-enrolment for oncologic care. An additional 6 min, 47 s resting state fMRI scan was acquired during each session using a 3 T Siemens Skyra or Siemens Prisma scanner with the following parameters: single shot T₂*-weighted EPI, repetition time = 2.26 s, echo time = 30 ms, field of view = 192 mm, matrix = 64 × 64, voxel size = 3.0 mm isotropic, bandwidth = 2055 Hz/pixel. A 3D T₁ MPRAGE with 1.0 mm isotropic resolution was also acquired during each imaging session. When necessary for the acquisition of adequate clinical imaging, participants were placed under anaesthesia via propofol. Depth of sedation was controlled such that participants continued spontaneous breathing, and supplemental oxygen was provided via nasal cannula throughout the scan.

Preprocessing and parcellation

Preprocessing was performed in SPM12 (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/), which included slice timing correction, motion correction, spatial normalization and smoothing. Motion-corrected EPI time-series data were co-registered to the native T₁ MPRAGE image for each session, and the T₁ image was normalized to an anatomic template. The same transformation was then applied to the EPI data for spatial normalization of the fMRI time series.

Following preprocessing and quality check, including a determination of whether normalization in participants with hydrocephalus was reasonable for the midbrain and cortical mantle, a spatiotemporal parcellation was created using independent component analysis. This data-driven approach was chosen to ensure an appropriate spatial parcellation scheme for the developmental range and clinical status of the medulloblastoma cohort. Fifty independent components were created using the GIFT toolbox (https://trendscenter.org/software/gift/). Upon visual inspection of the spatial maps, 6 of 50 components were denoted as pertaining to brain regions involved in speech production, including Broca’s area, the left supramarginal cortex (left SMC), putative somatomotor cortex for the face and mouth bilaterally (sM1), the amygdalae and PAG, the supplementary motor area (SMA), and the left dorsolateral prefrontal cortex (left DLPFC). Our independent component analysis did not identify a node that overlapped with the supracallosal medial frontal cortex (mFC) as highlighted by Holstege and Subramanian,¹³ so a region of interest (ROI) was manually created to estimate activity in this area [28 mm diameter spheroid, MNI coordinates: (0, 35, 23)]. Mean time courses were extracted from each of the ROIs.

Confounding signals from white matter and CSF were removed prior to further analysis. For white matter, the signal was averaged from four 6 mm diameter spheroids embedded in the parietal and frontal portions of the corona radiata bilaterally [MNI coordinates: (±28, −28, 29) and (±20, 38, 6), respectively]. For CSF a conservative mask was created within the lateral ventricles and mean time course was extracted. The time courses of speech-related ROIs were then denoised by taking the residuals after linear regression of the noise signals.

Activity and connectivity analyses

To assess the function of key midbrain nuclei in CMS, we used a measure of signal dispersion (the ratio of variance to the mean) for midbrain ROIs to estimate their activity.²⁷ A two-tailed t-test was used to compare activity between groups at each imaging time point. The change in activity level from ‘early’ to ‘late’ imaging sessions within participants (see trajectory analysis for definition of these terms) was also evaluated using a paired t-test. ROIs for the left and right red nuclei (RNs) were taken from a standard anatomic atlas and resampled for the fMRI space.²⁸

To identify potential changes in the function of the speech motor system during the acute phase of CMS, we evaluated connectivity within a network derived from the aforementioned review.¹³ The simplified network model included mFC, the amygdalae bilaterally, the PAG, Broca’s area, and left sM1²⁹ (Fig. 1A). Connectivity values were calculated as the time course correlation coefficient between ROIs. Connectivity values were calculated for node pairs with anatomical connections denoted in the model, and between PAG and all other nodes. For group comparison, connectivity values were z-transformed and a two-tailed t-test was conducted for each node-pair in the model. Owing to the exploratory nature of the study, no adjustments for multiple comparisons were performed.

Analysis of connectivity trajectory in CMS

As an addition and complement to examining group differences in connectivity during the acute phase of the disorder, we set out to identify functional group differences that evolved along the same time course as core CMS symptoms. To account for common effects of surgery, treatment and recovery on connectivity measures, the trajectory of connectivity in the asymptomatic medulloblastoma group served as a baseline upon which to compare the trajectory of the CMS group. The aim being to highlight transient differences in connectivity that resolve with speech recovery (and other core symptoms). We additionally sought late-developing differences, which could reflect compensatory changes for speech recovery or maladaptive changes related to long-term neurocognitive outcomes.

For this approach, we categorized scans as belonging to either an ‘early’ (initial postoperative imaging) or ‘late’ (12-month time point or later) epoch and evaluated how the group differences between CMS and asymptomatic medulloblastoma participants changed over time. Post-radiation therapy scans acquired at 3 months were not used for this analysis, as radiation therapy introduces an appreciable amount of variability into functional magnetic resonance measures of the cerebrum in our experience, and some CMS patients had not yet regained the capacity for speech. Participants with data missing from either epoch were excluded from the analysis, leaving 18 participants with CMS and 14 asymptomatic controls. Each connectivity measure from the CMS group was assigned a z-score based on the distribution of connectivity values from the control group for the corresponding epoch. A t-statistic and P-value were then calculated to compare trajectory differences.

For this analysis, we introduced additional nodes of interest, which are known to be involved in speech and/or motor planning. Additional nodes included the left SMC, the left DLPFC, the right sM1, and the SMA. Since pathways between these areas and the nodes of the speech motor system model are less clearly established, we chose to evaluate all possible ipsilateral and lateral-to-midline connections for new nodes.

For statistically significant results (alpha = 0.05), an attributability score was calculated to indicate the extent to which trajectory differences were attributable to early versus late group differences between CMS and asymptomatic participants, using Equation 1:

where CMS and Asymp denote the group mean connectivity value for the CMS and asymptomatic medulloblastoma groups, respectively, during the epoch denoted in subscript. Thus, a score of 0 reflects equal magnitude of group differences in early and late epochs, a score of 1 reflects group differences were exclusively present during late scans, and a score of −1 reflects the same but for early scans.

Finally, to detect group differences that were present across epochs, we evaluated the trajectory offset of the CMS group from the asymptomatic medulloblastoma group. The mean of the early and late z-score values was taken for each participant, and a t-statistic and P-value were then calculated to identify significant group differences that were immediate and persistent.

Analysis of lesion load along cerebellar outflow tract

To aid in the interpretation of connectivity results, we analysed the spatial distribution of surgically resected tissue in the CMS group. Our approach to lesion mapping has been described in detail previously.^6,30 To quantify lesion load to the superior cerebellar peduncles, we used a peer-reviewed probabilistic atlas of cerebellar white matter for anatomic reference³¹ with a 10% threshold applied. For each participant of interest, the proportion of superior cerebellar peduncle overlapping with surgical damage was calculated for each coronal slice moving forward from the centre of the dentate nuclei (MNI y coordinate = −57). One participant with damage within the dentate right dentate nucleus was excluded from this analysis. The maximum value from all coronal sections along the pathway was taken to represent lesion load for each participant.

Ethical approval

The protocol was approved by the St Jude Institutional Review Board and was performed in accordance with the ethical standards of the institution and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. All patients and/or their legal guardians provided written informed consent prior to participation in the study.

Results

Independent component analysis

We first performed an independent component analysis to generate a parcellation that is functionally relevant to the medulloblastoma cohort. In determining 50 independent components from the data, 27 components were deemed to be of likely BOLD origin and comprised a reasonable parcellation of cortical and subcortical areas. Peaks in the spatial projections of these components were considered viable nodes if they contained the global maximum that was at least 5 standard deviations (SD) above the mean, or a local maximum that was at least 10 SD above the mean. This resulted in the identification of 40 possible nodes. From this set, we selected nine nodes corresponding to areas known to be involved in speech to create ROI masks. Crucially, the PAG was highlighted with very high statistical confidence in one of the components (peak z-score = 17.79), indicating that it was feasible to obtain measurements in this area and that PAG BOLD fluctuations accounted for an appreciable amount of variance within the data. Notably, the PAG shared a component with the amygdalae, which further supports its functional relevance for the current purpose. Analysis nodes are visualized in Fig. 1.

Activity analysis

As an initial evaluation of cerebellar midbrain target function, we compared the relative BOLD activity in the PAG and RNs of patients with CMS to those who were asymptomatic (Fig. 1C). During the acute phase of the disorder, participants with CMS had a significantly higher degree of BOLD activity measured within the PAG and RNs bilaterally (PAG P = 0.0385, t = 2.1408; left RN P = 0.0102, t = 2.7207; right RN P = 0.0320, t = 2.2315). The magnitude of this group difference decreased over the first year and remained insignificant after 3 months. All nuclei showed initial hyperactivity when comparing CMS to asymptomatic medulloblastoma participants, but within-group analysis revealed that only PAG activity decreased significantly from the early to late imaging sessions in participants with CMS (Fig. 1C, P = 0.0267, paired t-test).

Acute changes in the speech motor system

To assess functional changes within the speech motor system during the acute phase of CMS, we looked for CMS-related connectivity differences between nodes within a simplified model of the speech motor system (Fig. 1D). Connectivity between Broca’s area and the PAG was decreased in patients with CMS (PAG-Broca P = 0.0042, t = −2.9730). Connectivity between the medial frontal cortex and PAG was significantly decreased in these patients as well, but to a lesser degree (P = 0.0149, t = −2.4990). Connectivity between the medial frontal cortex and left amygdala was increased in the CMS group (P = 0.0041, t = 3.0010). Other connectivity measures were not significantly altered in CMS participants.

Reactive and adaptive changes in the extended speech motor system

To highlight which functional changes accompany the transition between acute CMS and recovery, we compared the trajectory of connectivity in CMS and asymptomatic medulloblastoma participants (Fig. 2). We found that most of the significant changes were due to CMS-related hyperconnectivity in either the early or late scans, although the exceptions to this were notable. Connectivity differences that were strong during the acute phase but resolved with speech recovery were deemed ‘reactive’. We found that 8 of the 17 reactive hyperconnective changes were associated with the amygdalae, involving the sM1 bilaterally, Broca’s area, the L-SMC, the SMA and the mFC (sM1 P = 0.0091, t = 2.787; Broca P = 0.0099, t = 2.7314; L-SMC P = 0.0344, t = 2.2460; SMA P = 0.0034, t = 3.1508; mFC P < 0.001, t = 4.6564).

We further sought to identify functional changes characterized by increasing divergence of CMS patients from asymptomatic medulloblastoma patients following speech recovery. We deemed these changes ‘adaptive’ as they may reflect compensatory changes that support speech recovery, or reflect maladaptive changes related to long-term outcomes in CMS patients. Among the most interesting results identified in this analysis was adaptive hyperconnectivity between the PAG and the left DLPFC (P = 0.0071, t = −2.849). The sM1 bilaterally showed adaptive hyperconnectivity with the mFC (P = 0.0303, t = −2.258).

Lastly, we identified three significant trajectory differences attributable to group connectivity differences in both the early and late epochs. Most notably, connectivity between Broca’s area and PAG is weak during the acute phase of CMS (as demonstrated in Fig. 1D) but then becomes abnormally high in the later imaging time points (P < 0.001, t = −3.8854). The same was true for the PAG and mFC, but to a lesser degree (P = 0.0079, t = −2.9468). Conversely, Broca’s area showed early hyperconnectivity with the left DLPFC, but this appeared abnormally low in later imaging (P = 0.0055, t = 2.9672). All t- and P-values for this analysis are enumerated in Supplementary Tables 1 and 2, respectively.

Persistent differences in connectivity

While the primary aim of trajectory analysis is to identify connectivity patterns associated with transient mutism, proper interpretation of those results requires an understanding of potential CMS-associated changes that are both immediate and persistent (Fig. 3A). By comparing the offset in group trajectories, we found broad trends of persistent cortical hyperconnectivity affecting all nodes, especially the mFC, SMA, Broca’s area and the amygdalae (t-values and P-values given in Supplementary Tables 4 and 5, respectively). Connectivity between the amygdalae and PAG was uniquely reduced. To relate the strongest stable connectivity differences to permanent patterns of surgical damage in the cerebellum, we analysed the relationship between lesion load of the right superior cerebellar peduncle and connectivity between SMA and Broca’s area on an individual participant basis (Fig. 3B). We found that the abnormal signature of hyperconnectivity between these areas was reduced with increasing lesion load (Fig. 3C; R² = 0.5047, P = 0.0007). The participant excluded for surgical damage within the right dentate nucleus had a lower connectivity value than any of the participants with no damage to the superior cerebellar peduncle (z = 0.3102). All t- and P-values for this analysis are enumerated in Supplementary Tables 3 and 4, respectively.

Discussion

Acute midbrain dysfunction

The present study provides further evidence of PAG dysfunction as a neuronal driver of postoperative speech disruption in CMS. We found that the PAG was hyperactive and desynchronized from other critical speech areas during the acute phase of the disorder. PAG desynchronization was resolved after 12 months postoperative time, when each patient had regained some capacity for verbal expression.

We recently proposed surgical damage to the fastigial nuclei and their projections to the PAG as an aetiological trigger for speech disruption and apraxia in CMS.⁶ The acute functional changes observed in CMS patients could directly result from a loss of cerebellar control over PAG function, a projection that strongly affects the expression of primitive reactions to stress, including the suppression of vocalization and volitional movement.^14–17,32 Since this hypothesis contrasts with the conventional view that surgically disrupted cerebellar projections to cerebral cortex are the substrate of speech disruption, we also evaluated BOLD activity in the RNs, which relay information along this pathway. We found that activity measures in these nuclei were also elevated in the acute phase of the disorder, but their activity level did not change significantly with the recovery of speech faculties. Patients with CMS suffer from long term neurological and cognitive symptoms even after core symptoms improve, and this dissociation of effects in distinct cerebellar targets may reflect different neuronal mechanisms for acute versus chronic neurological symptoms, as well as multiple networks contributing to different aspects of CMS.

CMS connectivity changes and speech recovery

Analysis of connectivity trajectory revealed that synchronization of PAG with volitional speech cortices was disrupted in CMS participants during the acute phase of the disorder, but was restored or even increased with recovery [e.g. in Broca’s area: Fig. 2A(ii)]. The finding that PAG-Broca and PAG-mFC connectivity recovers with speech ability further supports the hypothesis that PAG dysfunction drives postoperative speech impairment. In general, all acute connectivity changes involved limbic structures, and the reactive changes seem to emphasize a dysregulation of synchrony between the two branches of the speech motor system. The only transient differences constrained to cortical nodes of the volitional motor system were adaptive, with somatomotor cortex exhibiting hyperconnectivity with mFC after 1 year.

All non-PAG nodes showed an initial hyperconnectivity with the amygdalae that partially resolved by 12 months postoperative time. The amygdalae have been implicated in many different cognitive and emotional processes beyond vocalization, and these involve its connections to dissociable (and in some cases, competitive) brain networks that can be identified in resting state fMRI.^33,34 The breadth of connectivity change is striking and could relate to any number of acute symptoms seen in CMS patients from emotional dysregulation to memory impairment.¹¹ In animals, a relatively direct functional relationship between the cerebellum and amygdalae has long been established,³⁵ and the anatomy supporting this functional relationship has come to light with advancements in viral cell tracers. Recent studies have identified disynaptic projections from cerebellum to the amygdala and hippocampus,^36,37 and suggested further modulation of limbic activity by the cerebellum was possible via projections to the basal forebrain.³⁸ Notably, all of these pathways to the amygdala have origins in the fastigial nuclei and traverse the superior cerebellar peduncle, where we found surgical damage to be closely linked to CMS.⁶ The relay of fastigial signals via specific thalamic nuclei suggests that the limbic cerebellum could play a role in the selective co-activation of amygdala with other structures, much in the same way the cerebro-cerebellum is thought to modulate cortical synchrony via the thalamus.³⁹ The apparent loss of selective connectivity between nodes of competing networks—such as left DLPFC for executive control and mFC for salience processing³³—with the surgical injury of limbic cerebellar circuits is consistent with this view of cerebello-thalamic function.

The left DLPFC was examined due to its involvement in language processing⁴⁰ and emotional regulation,⁴¹ and was found to exhibit a complex assortment of trajectory differences in CMS. Initially, left DLPFC showed enhanced connectivity with SMA, and was enmeshed in a strongly hyperconnected quartet of nodes including the mFC and amygdalae, and this partially resolved in late imaging time points. Between early and late scans, connectivity with Broca’s area shifted from hyper- to hypo-connected as well. Perhaps most intriguingly, the functional connection between left DLPFC and PAG was significantly increased in post-recovery imaging sessions, shifting from largely negative to largely positive correlation [Fig. 2A(iii)]. The role that this functional connection plays in expressive language is not well understood but may reflect increased demand on left DLPFC to coordinate respiration with the timing of other language processes in this case. As an essential component of breath support for vocalization, different subregions of the PAG exert different effects on the rate and depth of respiration,^13,42,43 and these subregions are thought to be influenced by various descending pathways.⁴⁴ In healthy subjects, neuronal and metabolic activity in the left prefrontal cortices exhibit a strong temporal relationship with respiration,⁴⁵ and this increases with the conscious awareness of regular breathing,^46,47 as well as the conscious effort to hold one’s breath.⁴⁸ DLPFC also exhibits functional connections with PAG related to pain⁴⁹ and threat processing,⁵⁰ including the threat of induced breathlessness⁴⁶ during which the cerebellar vermis is highly reactive.⁵¹ Thus, this acquired increase in connectivity may reflect changes associated with speech recovery where DLPFC projections to PAG are leveraged for greater control of respiration and the timing of phonation with volitional speech effort.

The broad shifts in connectivity of SMA is of interest due to its complex role in language and motor planning.⁵² Like mFC, the SMA is thought to play a role in the initiation of vocalization,⁵³ and has connections with primary motor cortices for the larynx, which are involved in vocal articulation.^54,55 Stimulation of SMA has been shown to elicit vocalizations in humans,⁵⁶ and strokes affecting left SMA can lead to transient mutism.⁵⁷ The SMA is also known to receive robust input from subcortical structures, including the dentate nuclei of the cerebellum,⁵⁸ which are thought to influence its interaction with other cortical regions for motor planning and speech production.^39,59 The hyperconnectivity observed here between SMA and mFC, Broca’s area and the amygdalae surely reflect the prominence of SMA in the hierarchy of speech motor planning. However, despite this evidence of SMA involvement in speech production and mutism pathogenesis generally, it remains an open question whether these changes are emblematic of neuronal dysfunction that contributes to mutism, reflective of general cerebello-cerebral dysfunction, or merely reactive to the mutism itself. We believe that PAG dysfunction remains a more plausible mechanism for mutism pathogenesis due to the transient nature of its dysfunction, the persistent nature of SMA connectivity (and red nuclei activity) changes, the monosynaptic proximity of PAG to high-risk surgical damage sites,^6,60 the spectrum of symptoms closely associated with mutism² and their alignment with PAG-mediated functions,^15,16,18 and the reasonable expectation that SMA hyperconnectivity and SMA injury would yield different behavioural consequences.

Mechanisms of cerebello-cerebral dysfunction

We showed previously that CMS participants from the current cohort did not have significantly common incidence of surgical damage in the dentate nuclei or within inferiolateral regions of the superior cerebellar peduncles that constitute their outflow pathways⁶ (Fig. 3B). A similar study also did not find significant differences in the amount of dentate outflow pathway damage between patients with CMS and patients with only cerebellar cognitive-affective symptoms.⁸ Evidence from the lesion analysis herein suggests that pathways for cerebellar feedback to neocortex remain largely intact per T₁ imaging in most of our CMS participants (Fig. 3B), and thus broad hyperconnectivity between SMA and other areas could be explained by abnormally persistent and/or non-specific cerebellar feedback. This is further supported by the finding that persistent hyperconnectivity between SMA and Broca’s area decreases as a greater proportion of the cerebellar outflow pathway is surgically damaged (Fig. 3C). Given the association of dentato-rubro surgical damage with cerebellar cognitive symptoms (but not mutism),⁸ the inverse relationship between superior cerebellar peduncle damage and SMA-Broca’s area hyperconnectivity, and the differences seen here between PAG and RN activity, these connectivity changes may be pointing to a mechanism of dysfunction within the functionally intact dentato-rubro-thalamo-cortical pathway that could contribute to long-term sequelae in the neurocognitive domain. It is likely that these cerebral connectivity changes are of cerebellar origin given the study design, however, what causes this reaction within the cerebellum in response to surgery and/or mutism remains as an important question.

Limitations

Functional imaging studies with prospectively gathered data focused on CMS aetiology are, to our knowledge, to date non-existent due to inherent difficulties. First, the rarity of CMS makes it difficult to power a study. Our institution sees a high volume of medulloblastoma patients even relative to other large paediatric hospitals, and these data were gathered over 9 years. The practical limitations on data acquisition make data-driven discovery of functional differences in CMS virtually impossible. Here, we addressed this challenge by focusing on areas involved in the production of speech rather than on all possible node-pairs from our data-driven parcellation. Statistics are reported individually without correction for multiple comparisons due to the exploratory nature of the study.

Diagnosis in any CMS cohort is a second major challenge, as symptom type and severity can vary widely between patients. This has been an issue in prior studies due to their retrospective nature and non-standardized neurological examination by non-neurologists. Clinical data on all participants of this study were prospectively collected by a neurologist for all medulloblastoma patients (irrespective of presumption of CMS) and CMS diagnosis and severity was based on degree of language and speech impairment, as described previously.² In-person examinations included evaluations for gait, stance, sitting balance, speech, dysarthria, dysphagia, ataxia, dysmetria, dysdiadochokinesia, ocular fixation, involuntary movement, irritability and apraxia. Here we reduce the complexity inherent to investigations of CMS-associated symptoms by focusing on the function of the speech motor system in participants with speech-related symptoms.

A clear limitation of this study was the need for propofol sedation during image acquisition. CMS tends to affect young children, as they are the most likely to develop molecular subtypes of medulloblastoma, which fill the fourth ventricle,⁶¹ and sedation is often needed to acquire diagnostic-quality imaging in young or uncooperative children free of motion-induced artefacts. We addressed this by only comparing data from CMS patients acquired with propofol sedation to data from asymptomatic patients who also required sedation. While sedation may reduce the ratio of neuronal signal-to-noise in the group being examined,⁶² there is no reason to anticipate differential effects of sedation in CMS and asymptomatic patients that would otherwise confound the current findings.

Conclusions

We used fMRI data acquired in a prospective study of medulloblastoma patients to investigate whether the midbrain PAG is linked to core symptoms of CMS, such as transient speech disruption. We showed that the PAG BOLD signal was abnormally volatile in CMS patient scans immediately following surgery, but this did not persist into later imaging sessions after speech faculties were recovered in all CMS participants. Additionally, we showed that PAG exhibited a deficit in synchrony with nodes of the volitional speech motor system in CMS patients, which also resolved with recovery. Having previously shown that surgical damage to cerebellar-PAG circuits is associated with CMS diagnosis,⁶ this functional imaging evidence offers further support to the theory that dysfunction of PAG survival circuits underlies some of the core symptoms of CMS.

Data availability

The data that support the findings of this study are available from the corresponding authors, M.A.S. and S.S.M., upon reasonable request.

Acknowledgements

We gratefully acknowledge the work of our Research Assistants within the St. Jude Department of Diagnostic Imaging, Kim Bailey-Johnson and Angela Edwards, for their contribution to data management and fMRI preprocessing. We gratefully acknowledge the guidance of Robert Ogg during the early stages of developing this project. We would like to thank Lisa Jacola, Chris Goode and Diana Storment for providing information about sedation procedure during image acquisition. S.S.M. would like to thank Gal Flam for sharing their expertise of anatomy and musculature for speech and swallowing.

Funding

This work was supported by the American Lebanese Syrian Associated Charities (ALSAC) and the National Cancer Institute grant (P30 CA021765) (Cancer Center Support Grant). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.

References