Microstructural cerebellar injury independently associated with processing speed in adult primary brain tumor patients: Implications for cognitive preservation

Abstract

Background

The cerebellum’s role in post-treatment neurocognitive decline is unexplored. We investigated associations between cerebellar microstructural integrity using quantitative neuroimaging biomarkers and neurocognition among primary brain tumor patients receiving partial-brain radiotherapy (RT).

Materials and Methods

On a prospective trial, 65 patients underwent volumetric brain MRI, diffusion tensor imaging, and memory, executive function, language, attention, and processing speed (PS) assessment pre-, 3, 6, and 12-months post-RT. Delis-Kaplan Executive Function System-Trail Making [DKEFS-TM] Visual Scanning and Number and Letter Sequencing and Wechsler-Adult Intelligence Scale-IV [WAIS] Coding evaluated PS. Cerebellar cortex and white matter (WM) and supratentorial structures subserving above cognitive domains were autosegmented. Volume was measured within each structure at each timepoint along with diffusion biomarkers (fractional anisotropy [FA] and mean diffusivity [MD]) in WM structures. Linear mixed-effects models assessed cerebellar biomarkers as predictors of neurocognitive scores. If associated, cerebellar biomarkers were evaluated as independent predictors of cognitive scores controlling for domain-specific supratentorial biomarkers.

Results

Left (p=0.04) and right (p<0.001) cerebellar WM volume declined significantly over time. Cerebellar biomarkers were not associated with memory, executive function, or language. Smaller left cerebellar cortex volume was associated with worse DKEFS-TM Number (p=0.01) and Letter (p=0.01) Sequencing scores. Smaller right cerebellar cortex volume correlated with worse DKEFS-TM Visual Scanning (p=0.02) and Number (p=0.03) and Letter (p=0.02) Sequencing scores. Greater right cerebellar WM MD, indicating white matter injury, was associated with worse DKEFS-TM Visual Scanning performance (p=0.03). Associations remained significant after controlling for corpus callosum and intrahemispheric WM injury biomarkers.

Conclusions

Injury to the cerebellum as measured with quantitative biomarkers correlates with worse post-RT PS, independent of corpus callosum and intrahemispheric WM damage. Efforts to preserve cerebellar integrity may preserve PS.

INTRODUCTION

Neurocognitive decline is an unfortunate consequence of primary and metastatic brain tumors and their treatment, which includes surgery, systemic therapy, and radiation therapy (RT)¹. As radiation dose is modifiable, brain tumor clinical trials have focused on avoidance of supratentorial structures such as the hippocampus and corpus callosum in an effort to preserve neurocognition^2,3.

Neuroanatomically, the cerebellum is commonly associated with motor function and coordination. However, there is also evidence that the cerebellum may play an important role in higher-order neurocognitive domains in healthy adults, including executive function, attention, processing speed (PS), memory, and language⁴. Neurocognitive deficits have also been documented among patients with posterior fossa tumors⁴, which comprise 20% of adult and 60% of childhood intracranial tumors⁴. Studies have identified radiation dose-dependent cerebellar atrophy in patients treated for gliomas⁵. In children, cerebellar RT dose has been shown to correlate with cognitive function after treatment for infratentorial ependymoma⁶. Nevertheless, radiation treatment planning largely ignores the potential cognitive sequelae of cerebellar injury. Current radiation dosing guidelines for brain tumors specify constraints to brain parenchyma (without further functional/neuroanatomic breakdown), brainstem, and optic pathway structures, but not the cerebellum⁷. Similarly, dose constraints to the temporal lobes, cochlea, and optic pathways are defined in the treatment of base of skull and head and neck tumors, while the cerebellum remains overlooked⁸.

Several small studies have identified worse cognitive function among pediatric brain tumor patients with greater cerebellar atrophy⁹ and white matter (WM) damage¹⁰, suggesting that cerebellar injury may contribute to post-treatment neurocognitive decline in this population. Yet, these studies were retrospective and cross-sectional, only assessed imaging and/or neurocognitive outcomes at a single timepoint, and did not analyze multiple neurocognitive domains. Moreover, they failed to investigate whether the relationship between cerebellar damage and cognitive function is independent of concurrent damage to supratentorial structures that subserve neurocognition such as the hippocampus¹¹, corpus callosum¹², or perisylvian WM¹³. Essentially, does the cerebellum play an independent role in higher order neurocognition? We aimed to explore changes in in vivo volumetric and diffusion tensor imaging (DTI) biomarkers of cerebellar integrity over time, the association between these biomarkers and RT dose, and the longitudinal relationship between these biomarkers and cognitive functioning among adult patients with primary brain tumors. We hypothesized that cerebellar cortical atrophy and WM injury (i.e., reduced fractional anisotropy [FA] and higher mean diffusivity [MD]) would correlate with worse domain-specific neurocognitive function at a given timepoint, independent of damage to supratentorial cognitive structures.

MATERIALS AND METHODS

Study Design

This prospective longitudinal observational clinical trial was approved by the UC San Diego Moores Cancer Center institutional review board and registered on clinicaltrials.gov. The trial examines quantitative neuroimaging biomarkers and neurocognitive functioning within the domains of memory, executive functioning, attention, PS, language, and fine motor skills among adult primary brain tumor patients before and after fractionated partial-brain RT. Participants underwent comprehensive neurocognitive testing, high-resolution 3D volumetric brain MRI, and diffusion tensor imaging (DTI) prior to RT and 3, 6, and 12 months post-RT.

Study Participants

All 72 patients enrolled in this prospective clinical trial provided written informed consent. Our inclusion criteria included age ≥ 18 years, Karnofsky performance status ≥70, life expectancy ≥1 year, and English proficiency. This analysis included 65 subjects with baseline neurocognitive assessments and imaging following the complete trial protocol. Of the 72 patients enrolled, 7 patients were excluded from analysis for the following reasons: no complete baseline neurocognitive testing (n=1); no complete trial protocol neuroimaging (n=4); no complete neurocognitive testing or neuroimaging per protocol (n=2). None of these patients had radiographic or clinical tumor progression during the study period.

Neurocognitive Assessment

All patients completed testing one-on-one with a clinical neuropsychologist. Neurocognitive assessment consisted of six tests (13 test indices) evaluating memory, executive function, attention, PS, and language. The Hopkins Verbal Learning Test Revised¹⁴ Total Recall and Delayed Recall and Brief Visuospatial Memory¹⁵ Total Recall and Delayed Recall subtests measured memory. The Delis-Kaplan Executive Function System (DKEFS)¹⁶ Verbal Fluency and Category Switching scores and Wisconsin Card Sorting Test¹⁷ Perseverative Errors subtests assessed executive function. The DKEFS Trail Making (DKEFS-TM) Visual Scanning, Number Sequencing, and Letter Sequencing and Wechsler-Adult Intelligence Scale-IV¹⁸ (WAIS) Coding subtests evaluated PS. The DKEFS-TM Visual Scanning subtest measured both PS and attention. The WAIS Digit Span Forward subtest measured attention. The DKEFS Category Fluency subtest assessed language. Higher raw scores on DKEFS-TM tests indicate worse performance. Higher raw scores on the remaining tests indicate better performance.

Imaging Acquisition and Processing

Methods for acquisition of high-resolution volumetric and diffusion-weighted MRIs in this study have been described in detail elsewhere^19,20. They are outlined in the Supplementary Methods. In-house algorithms were used to preprocess imaging data^21–23. Distortion correction was applied for anatomical imaging distortions due to gradient nonlinearities²⁴ and diffusion scan spatial and geometric distortions caused by susceptibility, eddy currents, and B₀ distortion²⁵. We derived FA and MD maps by fitting DWI data from b-values 0, 500, and 1500 s/mm² to a tensor. An ellipsoid defined by three perpendicular axes (eigenvectors) approximated the diffusion process within each voxel. FA is a unitless expression of the degree of diffusion directionality and ranges between 0 and 1; FA decreases with WM injury²⁶. MD is an average of the three eigenvalues expressed in µm²/ms representing the average mobility of water molecules; MD increases with WM injury²⁶. Volume is reported in cm³.

Region of Interest (ROI) Segmentation

Detailed ROI segmentation methods are outlined in the Supplementary Methods. Tissue impacted by tumor, surgical cavity, or edema was manually censored slice by slice on each image at each timepoint, blinded to ROI segmentation results. These voxels were excluded from the final ROI assessed at the same timepoint to prevent confounding by tumor and edema-related effects²⁷. Selected cerebellar ROIs included the cerebellar cortex and WM (Figure 1a). We also investigated supratentorial brain ROIs strongly linked to memory, executive function, attention, PS, and language (Supplementary Methods). These ROIs were chosen a priori based on neuroanatomic structure-function paradigms and our prior work^11,21,22,28. Supratentorial PS and attention ROIs included the posterior, mid-posterior, central, mid-anterior, and anterior corpus callosum (CC) along with the bilateral intrahemispheric WM (Figures 1b–c).

Figure 1.

Cerebellar and supratentorial processing speed- and attention-associated regions of interest

Cerebellar and supratentorial processing speed- and attention-associated ROIs assessed in this analysis. (A) Cerebellar cortex, dark blue; cerebellar WM, light blue. (B) Posterior CC, red; mid-posterior CC, orange; central CC, yellow; mid-anterior CC, green; anterior CC, blue. (C) right intrahemispheric WM, pink; left intrahemispheric WM, purple; CC, grey.

Abbreviations: ROIs, regions of interest; WM, white matter; CC, corpus callosum

Statistical Analysis

Linear mixed-effects (LME) analyses investigated change in cerebellar imaging biomarkers over time and the association between mean RT dose and cerebellar imaging biomarkers measured at each timepoint. LME models are well-suited for longitudinal analyses because they account for within-subject correlation between repeated measures and allow for incomplete outcome data²⁹. LME models contained volume, FA, or MD as the outcome, time as a main effect, and subject-specific random intercepts. Models evaluating the relationship between RT dose and imaging biomarkers additionally contained mean RT dose (Gy) as a main effect. All models estimating volume controlled for the percentage of the ROI censored at each timepoint to account for longitudinal censoring trends in volume measurements^22,23.

We then evaluated cerebellar imaging biomarkers as predictors of raw neurocognitive scores measured at the same timepoint. LME models contained each raw neurocognitive score as the outcome, imaging biomarkers and time as main effects, and subject-specific random intercepts.

We performed the following analysis to assess whether cerebellar injury was independently associated with our selected cognitive domains. For neurocognitive tests that were found to be associated with cerebellar imaging biomarkers, we used LME analysis to examine the association between these tests and damage to supratentorial structures empirically shown to be associated with each cognitive domain. If significantly associated with neurocognition (p<0.05), we controlled for each supratentorial structure biomarker in additional LME models evaluating the association between cerebellar biomarkers and cognitive scores assessed at the same timepoint. These models contained each raw neurocognitive score as the outcome, cerebellar and supratentorial structure FA, MD, or volume and time as main effects, and subject-specific random intercepts. We identified and removed outliers via Mahalanobis distance based on a chi-square distribution (assessed using p<0.001)³⁰. Statistical significance was set at α=0.05 for two-tailed tests. P-values were corrected for multiple comparisons using the false discovery rate³¹.

RESULTS

Patient Characteristics

Demographic and clinical characteristics are shown in Table 1. This cohort was mostly Non-Hispanic White (82%) and highly educated, with slightly more males (57%). Most (62%) patients had gliomas and three patients (5%) had cerebellar tumors.

Table 1.

Baseline patient, tumor, and treatment characteristics (n=65)

Characteristic	Patients, n (%) n = 65
Median age, years (range)	47 (20–75)
Sex
Male	37 (56.9)
Female	28 (43.1)
Ethnicity/Race
Hispanic	8 (12.3)
Non-Hispanic	57 (87.7)
White	53 (81.5)
Asian/Pacific Islander	3 (4.62)
Black	1 (1.54)
Highest education achieved, years (median, range)	16 (10–20)
Tumor diagnosis
Glioma	40 (61.5)
WHO Grade 1–2*	11 (16.9)
WHO Grade 3–4†	29 (44.6)
Meningioma	13 (20.0)
WHO Grade 1	10 (15.4)
WHO Grade 2	2 (3.10)
WHO Grade 3	1 (1.54)
Pituitary adenoma	5 (7.69)
Low-grade chondrosarcoma	1 (1.54)
Craniopharyngioma	3 (4.62)
Pineal tumor	1 (1.54)
Schwannoma	2 (3.08)
Tumor location
Frontal	19 (29.2)
Temporal	15 (23.1)
Suprasellar	10 (15.4)
Parietal	8 (12.3)
Base of skull	6 (9.23)
Cerebellar	3 (4.62)
Cavernous sinus	3 (4.62)
Pineal	1 (1.54)
Tumor side
Left	32 (49.2)
Right	26 (40.0)
Central	7 (10.8)
Radiotherapy type
IMRT/VMAT	49 (75.4)
Protons	16 (24.6)
Median RT dose, Gy/GyE (range)	59.4 (50.4–70.0)
Chemotherapy
Concurrent/adjuvant temozolomide	30 (46.2)
Concurrent/adjuvant temozolomide, and other‡	5 (7.69)
Adjuvant procarbazine, lomustine, and, vincristine	4 (6.15)
None	26 (40.0)
Surgery
Gross total resection	16 (24.6)
Subtotal resection	38 (58.5)
Biopsy	4 (6.15)
None	7 (10.8)
History of seizures during the study period	28 (43.1)
Antiepileptic drug use during the study period	37 (56.9)

^*WHO grade 1–2 gliomas included: grade 2 diffuse astrocytoma (isocitrate dehydrogenase 1 [IDH] wild-type [n=1], IDH mutated [n=1]; O[6]-methylguanine-DNA methyltransferase [MGMT] methylated [n=1], MGMT unmethylated [n=1]), grade I pilocytic astrocytoma (n=1), grade 2 IDH mutated 1p19q codeleted oligodendroglioma [n=6]

^†WHO grade 3–4 gliomas included: grade 4 glioblastoma multiforme (IDH wild-type [n=7], IDH mutated [n=2]; MGMT methylated [n=5], MGMT unmethylated [n=5]), grade 3 anaplastic astrocytoma (IDH wild-type [n=3], IDH mutated [n=9]), 1p19q codeleted oligodendroglioma (n=2).

^‡Other chemotherapy included: vaccine clinical trial (n=1), poly (ADP-ribose) polymerase inhibitor clinical trial (n=2), adjuvant lomustine (n=1), proteasome inhibitor clinical trial (n=1).

Abbreviations: IMRT/VMAT, intensity-modulated radiotherapy/volumetric modulated arc therapy; RT, radiotherapy; Gy, Gray; GyE, Gray equivalent dose

Neurocognitive Function and Cerebellar Biomarkers over Time

Memory, executive function, attention, PS, and language performance over the study period have been described in a subset of this cohort previously^11,21,22,28. Left (ß=−0.077 cm³/month, p=0.02) and right (ß=−0.182 cm³/month, p<0.001) cerebellar WM volumes declined significantly over time. (Supplementary Table 1).

Mean Dose as a Predictor of Biomarkers

Mean RT dose to each cerebellar ROI is shown in Supplementary Table 2. Mean RT dose to cerebellar ROIs was not associated with cerebellar imaging biomarkers (Supplementary Table 3).

Associations between Cerebellar Imaging Biomarkers and Neurocognitive Function

Cerebellar imaging biomarkers were not associated with memory, executive function, or language performance. Table 2 demonstrates ß and p-values from LME analysis of the association between selected cerebellar imaging biomarkers and PS assessed at the same timepoint. Greater left cerebellar cortex volume correlated with significantly better DKEFS-TM Number and Letter Sequencing performance at each timepoint. Higher right cerebellar cortex volume correlated with better DKEFS-TM Visual Scanning and Number and Letter Sequencing scores. Lower right cerebellar WM MD was significantly associated with better DKEFS-TM Visual Scanning performance.

Table 2.

Association between cerebellar imaging biomarkers and processing speed

Processing speed test	Left cerebellar cortex volume		Right cerebellar cortex volume		Right cerebellar WM MD
	ß Points/(month x cm3)	p-value	ß Points/(month x cm3)	p-value	ß Points/(month x µm2/ms)	p-value
DKEFS-TM Visual Scanning	−0.193	0.05	−0.237	0.02 *	28.8	0.03
DKEFS-TM Number Sequencing	−0.372	0.01 *	−0.321	0.03 *	34.1	0.10
DKEFS-TM Letter Sequencing	−0.458	0.01 *	−0.397	0.02 *	0.298	0.19
WAIS Coding	0.251	0.27	0.367	0.09	−7.45	0.77

Associations between processing speed tests and right and left cerebellar WM volume, left cerebellar WM FA and MD, and right cerebellar FA were not statistically significant (not shown). Higher scores on DKEFS-TM tests and lower scores on WAIS tests indicate better performance.

Abbreviations: DKFES-TM, Delis-Kaplan Executive Function System Trail Making Test; WAIS, Weschler Adult Intelligence Scale; MD, mean diffusivity; FA, fractional anisotropy; WM, white matter.

Bold indicates significant at P < 0.05 level.

^*P-values remained significant after correction for multiple comparisons.

Independent Association between Cerebellar Imaging Biomarkers and PS

LME analysis results of CC and intrahemispheric WM injury biomarkers as predictors of PS are shown in Supplementary Table 4. Table 3 demonstrates the ß and p-values from LME analysis of the association between cerebellar ROI biomarkers and PS after controlling for CC ROI biomarkers that were correlated with PS. Reduced right cerebellar cortex volume remained significantly associated with worse performance on DKEFS-TM Visual Scanning and Number Sequencing after controlling for posterior CC FA, DKEFS-TM Visual Scanning controlling for central CC volume, and DKEFS-TM Number and Letter Sequencing controlling for combined CC FA. Lower left cerebellar cortex volume remained significantly associated with poorer DKEFS-TM Visual Scanning and Number and Letter Sequencing scores accounting for posterior CC FA and DKEFS-TM Number Sequencing accounting for combined CC FA. Higher right cerebellum WM MD remained significantly correlated with DKEFS-TM Visual Scanning controlling for posterior CC FA and mid-anterior CC volume.

Table 3.

Association between cerebellar imaging biomarkers and processing speed independent of microstructural injury to corpus callosum

Cerebellar imaging biomarker	CC structure	Processing speed test	ß*	p-value
Right cerebellum cortex volume	Posterior CC FA	DKEFS-TM Visual Scanning	−0.250	0.01 †
		DKEFS-TM Number Sequencing	−0.347	0.02 †
	Central CC volume	DKEFS-TM Visual Scanning	−0.196	0.04
	Mid-anterior CC volume	DKEFS-TM Visual Scanning	−0.187	0.07
		DKEFS-TM Letter Sequencing	−0.310	0.08
	Combined CC FA	DKEFS-TM Number Sequencing	−0.389	0.01 †
		DKEFS-TM-Letter Sequencing	−0.436	0.01 †
Left cerebellum cortex volume	Posterior CC FA	DKEFS-TM Visual Scanning	−0.186	0.04 †
		DKEFS-TM Number Sequencing	−0.408	0.01 †
		DKEFS-TM Letter Sequencing	−0.469	0.01 †
	Mid-anterior CC volume	DKEFS-TM Visual Scanning	−0.121	0.22
	Central CC volume	DKEFS-TM Visual Scanning	−0.125	0.20
	Combined CC FA	DKEFS-TM Number Sequencing	−0.457	0.003 †
Right cerebellum WM MD	Posterior CC FA	DKEFS-TM Visual Scanning	30.4	0.02 †
	Central CC volume	DKEFS-TM Visual Scanning	17.1	0.23
	Mid-anterior CC volume	DKEFS-TM Visual Scanning	31.7	0.01 †

Association between cerebellar imaging biomarker (column 1) and processing speed tests (column 3), controlling for corpus callosum structures (column 2).

^*Units: volume, points/(month x cm³); MD, points/(month x µm²/ms).

Abbreviations: CC; corpus callosum; DKEFS-TM, Delis-Kaplan Executive Function System Trail Making Test; WAIS, Weschler Adult Intelligence Scale; MD, mean diffusivity; FA, fractional anisotropy; WM, white matter.

Bold indicates significant at P < 0.05 level.

^†P-values remained significant after correcting for multiple comparisons.

The relationship between cerebellar ROI biomarkers and PS after controlling for PS-associated intrahemispheric WM biomarkers is shown in Table 4. Reduced right cerebellar cortex volume remained significantly associated with DKEFS-TM Number and Letter Sequencing controlling for left intrahemispheric WM volume, DKEFS-TM Visual Scanning and Number and Letter Sequencing accounting for left intrahemispheric WM MD, DKEFS-TM Visual Scanning after controlling for right intrahemispheric MD, and DKEFS-TM Visual Scanning controlling for bilateral intrahemispheric MD. Lower left cerebellar cortex volume remained significantly correlated with DKEFS-TM Number Sequencing controlling for left intrahemispheric WM volume, DKEFS-TM Visual Scanning and Number Sequencing controlling for left intrahemispheric WM MD, DKEFS-TM Number Sequencing accounting for right intrahemispheric WM MD, and DKEFS-TM Visual Scanning and Number Sequencing controlling for bilateral intrahemispheric WM MD. The results of LME analysis of the independent association between cerebellar ROI biomarkers and PS are summarized in Figure 2.

Table 4.

Association between cerebellar imaging biomarkers and processing speed and attention independent of microstructural injury to intrahemispheric white matter

Cerebellar imaging biomarker	Intrahemispheric WM structure	Processing speed test	ß*	p-value
Right cerebellum cortex volume	Left intrahemispheric WM volume	DKEFS-TM Number Sequencing	−0.314	0.04 †
		DKEFS-TM Letter Sequencing	−0.394	0.02 †
	Left intrahemispheric WM MD	DKEFS-TM Visual Scanning	−0.218	0.02
		DKEFS-TM Number Sequencing	−0.293	0.04
		DKEFS-TM Letter Sequencing	−0.370	0.03
	Right intrahemispheric WM MD	DKEFS-TM Visual Scanning	−0.195	0.03
		DKEFS-TM Number Sequencing	81.3	0.05
	Bilateral intrahemispheric WM MD	DKEFS-TM Visual Scanning	−0.201	0.03
		DKEFS-TM Number Sequencing	115.3	0.06
Left cerebellum cortex volume	Left intrahemispheric WM volume	DKEFS-TM Number Sequencing	−0.338	0.03 †
	Left intrahemispheric WM MD	DKEFS-TM Visual Scanning	−0.192	0.04
		DKEFS-TM Number Sequencing	−0.406	0.01
	Right intrahemispheric WM MD	DKEFS-TM Visual Scanning	−0.178	0.05
		DKEFS-TM Number Sequencing	−0.372	0.01
	Bilateral intrahemispheric WM MD	DKEFS-TM Visual Scanning	−0.193	0.04
		DKEFS-TM Number Sequencing	−0.393	0.01 †
Right cerebellum WM MD	Left intrahemispheric WM MD	DKEFS-TM Visual Scanning	21.1	0.11
	Right intrahemispheric WM MD	DKEFS-TM Visual Scanning	14.5	0.27
	Bilateral intrahemispheric WM MD	DKEFS-TM Visual Scanning	15.0	0.26

Association between cerebellar imaging biomarker (column 1) and processing speed tests (column 3), controlling for intrahemispheric white matter structures (column 2).

^*Units: volume, points/(month x cm³); MD, points/(month x µm²/ms).

Abbreviations: DKEFS-TM, Delis-Kaplan Executive Function System Trail Making Test; WAIS, Weschler Adult Intelligence Scale; MD, mean diffusivity; FA, fractional anisotropy; WM, white matter.

Bold indicates significant at P < 0.05 level.

^†P-values remained significant after correcting for multiple comparisons.

Figure 2.

Cerebellar imaging biomarkers independently associated with processing speed

Blue squares indicate cerebellar cortex and WM imaging biomarkers (columns) found to be associated with processing speed independent of supratentorial WM structures known to subserve processing speed (rows). Grey squares represent cerebellar imaging biomarkers that were not associated with processing speed independent of supratentorial WM structures.

DISCUSSION

To our knowledge, this study represents the first prospective, longitudinal analysis of the relationship between post-treatment cerebellar damage using quantitative MRI and neurocognitive function in adult patients with brain tumors. We identified cerebellar cortex and WM biomarkers of injury independently associated with PS. Our results establish the cerebellum as an organ at risk of treatment-associated toxicity in adults and introduce a novel strategy for cognitive preservation within the field of neuro-oncology.

Our findings indicate that cerebellar damage is associated with worse PS in adult patients undergoing brain tumor treatment, implying that PS preservation efforts in this population should focus on minimizing cerebellar injury. Importantly, the link between microstructural cerebellar integrity and PS was independent of injury to supratentorial WM structures including the corpus callosum and intrahemispheric WM that are known to subserve PS²¹, as summarized in Figure 2. Neurocognitive decline has been recognized as one of the most significant sequelae of brain tumor treatment. This has given rise to the novel field of cognitive-sparing RT, which attempts to define RT dose constraints to brain structures critical to cognition. Hippocampal-sparing RT to preserve memory² is perhaps the most well-known of these efforts, but further studies avoiding additional brain structures including the corpus callosum³ and other critical WM tracts³² are currently underway. In adults, the entirety of this work has concentrated on the supratentorial brain, and the cerebellum’s potential relevance to neurocognition has largely been ignored. We demonstrate that the cerebellum is, in fact, an integral structure for PS, and that future cognitive-sparing treatment regimens in adults should consider the cerebellum. This work also has implications for cognitive sparing efforts in the treatment of base of skull tumors, head and neck cancer, and pediatric brain tumors, which most commonly arise from the posterior fossa.

The cerebellum’s contribution to PS has been previously confirmed by anatomical, functional MRI, and clinical evidence⁴ in various patient populations. To our knowledge, this study presents the first evidence of the cerebellum’s role in PS among adult patients with brain tumors. The cerebellum shares extensive connections with cerebral association areas, including the prefrontal and parietal association cortices³³, which synthesize sensory and motor information³⁴. Cerebellar injury may therefore compromise the rate at which these inputs are integrated and manifest as impaired PS. Indeed, cerebellar grey matter volume has been linked to PS in healthy children³⁵ and older adults³⁶, as well as in patients with multiple sclerosis³⁷. One prospective study of multi-domain cognitive functioning among children with medulloblastoma found the greatest deficits in PS at five years after diagnosis.³⁸ Our findings underscore the importance of further defining drivers of cerebellar injury to preserve PS in this population.

Interestingly, we did not find deficits in language, executive function, or memory among patients with greater cerebellar injury. Though the cerebellum is commonly associated with motor control and coordination, there is a breadth of literature documenting its role in a range of cognitive functions in healthy adults, including memory, attention, language, executive function, information PS, and affect³⁴. Our results imply that, of these cognitive domains, cerebellar injury may carry the greatest ramifications for PS and possibly attention in adult patients with brain tumors. Of note, we found that multiple cerebellar imaging biomarkers were associated with DKEFS-TM Visual Scanning performance, which requires participants to identify all the number “3s” on the page within 150 seconds. As this test is timed, it may measure both PS and attention³⁹, suggesting that cerebellar injury may also be linked to impaired attention among patients in our cohort. Brain tumors in the pediatric population, which are most often located in the posterior fossa, are certainly associated with attentional deficits similar to those seen in attention-deficit hyperactivity disorder⁴⁰. Nevertheless, we found that WAIS Digit Span Forward, which measures attention alone, did not correlate with any cerebellar imaging biomarkers. Cerebellar damage may therefore be associated with greater PS than attention compromise in adults.

This study identified more associations between PS cerebellar cortex versus WM microstructural integrity. Alternatively, our previous work²³ found that both cerebellar cortex and WM damage portended poorer fine motor skills in the same cohort. This suggests that cerebellar cortical, rather than WM, integrity may be more strongly linked to PS in this patient population. Cerebellar grey matter atrophy has been linked to worse cognitive performance in older adults and patients with autism and multiple sclerosis^35,37,41. Additionally, we found that PS was more often associated with injury to right-sided versus left-sided cerebellar ROIs. Similar patterns have been observed among patients with focal cerebellar lesions⁴² and Alzheimer’s disease⁴³. In a study investigating cerebellar functional asymmetry, Wang and colleagues found that the most strongly right lateralized cerebellar regions fell within areas linked to the cerebral association cortex⁴⁴. In conjunction with these results, our findings reflect potential functional differences between the cerebellar hemispheres, and may also imply greater vulnerability of right-sided cerebellar structures to pathological processes such as radiation-associated damage. Functional atlases of the cerebellum, such as those by King and colleagues⁴⁵, have clarified the role of individual cerebellar substructures in neurocognition and motor function among healthy patients. Further understanding of how these cerebellar subregions are affected by cancer treatment may inform the development of more intricate RT dose-avoidance strategies to preserve neurocognition.

We found that cerebellar WM volume declined throughout the study period. In contrast with prior studies⁵, this decline was not associated with mean RT dose. The majority of patients in our cohort had supratentorial tumors, and thus had lower dose RT exposure to the posterior fossa (Supplementary Table 2), which limits our ability to detect changes over a wide range of doses. Moreover, most patients in this cohort had surgery and/or chemotherapy in addition to RT, both of which may also compromise WM integrity⁴⁶. RT is known to drive WM injury via demyelination, neuroinflammation, and axonal and vascular damage⁴⁷. Higher posterior fossa RT doses have been associated with IQ decline in the pediatric population⁴. Merchant and colleagues further demonstrated that RT dosimetry can predict IQ after conformal RT in children with localized ependymoma⁴⁸. Such findings have established reduced-dose RT as the preferred treatment for malignant posterior fossa tumors in children⁴⁹. In adults, one study measured worse memory scores among patients receiving higher RT doses to the cerebellum⁵⁰. Nevertheless, cerebellar-sparing strategies have not been explored in adults. Ultimately, normal tissue complication probability analyses of cognitive function utilizing a wide range of cerebellar dosimetric and volumetric parameters (beyond mean dose) may guide the development of specific cerebellar dose constraints.

This study has several limitations. While our sample size was modest, this prospective cohort is larger than most other studies examining imaging biomarkers of neurocognition^51,52. Additionally, we collected multi-domain neurocognitive and imaging data for each patient at multiple timepoints. Our cohort included mostly White, highly educated, and English-speaking patients, which could limit generalizability. Our sample size limited our ability to investigate associations between cerebellar biomarkers and cognition independent of damage to multiple supratentorial WM structures concurrently. Nevertheless, our analysis included several key supratentorial PS-associated ROIs. We were also unable to concurrently control for the myriad of other clinical and patient variables that could impact neurocognitive function, given our modest sample size. We manually inspected all ROI segmentations for each patient at each timepoint to censor areas affected by tumor, surgical cavity, and edema^27,53. While this minimized confounding and errors in ROI definition, it limited our ability to measure cerebellar tissue effects of higher RT doses (e.g. high RT doses within tumor or edema). Future normal tissue complication probability (NTCP) analyses will help determine optimal dose-constraints for cerebellar structures in RT planning. Our cohort included patients with both benign tumors and gliomas as well as supra- and infratentorial tumors; thus, our findings are generalizable to a wide range of neuro-oncologic patients after treatment, including patients without cerebellar tumors.

CONCLUSIONS

We present the first prospective evidence of an association between cerebellar injury and PS. Notably, this relationship was independent of damage to the corpus callosum and intrahemispheric WM, both of which have been associated with PS. We also found that cerebellar WM volume declined in patients with primary brain tumors after treatment. Our results suggest the cerebellum should be an organ at risk in neuro-oncologic treatment planning and support further exploration of the cerebellum’s response to RT. This work also helps frame counseling and discussions with brain tumor patients about expected post-treatment outcomes, especially given that the cerebellum has not been commonly implicated in higher order neurocognition. Our findings also imply that efforts to preserve processing speed should attempt to minimize cerebellar damage, whether this is from radiation, chemotherapy, surgery, or other causes. Cerebellar-sparing treatment strategies may contribute to PS preservation in patients with primary and metastatic brain tumors and ultimately improve quality of life.

Data sharing:

Anonymized data generated in this study are available upon request to the corresponding author.