Multimodal imaging reveals resilient memory networks in carriers of pathogenic ARID1B variants

Abstract

The phenotypic spectrum associated with pathogenic ARID1B variants is remarkably broad, ranging from classic Coffin–Siris syndrome to non-syndromic intellectual disability and autism spectrum disorders. While speech delay, motor impairments and learning difficulties are well documented, brain imaging investigations remain scarce in this population. We combined multimodal neuroimaging and neuropsychological assessments in 12 patients carrying pathogenic ARID1B variants (age = 13.8 ± 4.7 years) and 34 age-matched healthy controls. Whole-brain voxel-wise analyses included arterial spin labelling to measure cerebral blood flow (CBF) at rest and voxel-based morphometry to assess grey matter density. To investigate white matter abnormalities, we performed fixel-based analysis in a subgroup of 7 patients and 17 age-matched controls. While patients showed pronounced language, motor and social impairments, their memory performance ( + 5 SD) largely exceeded other cognitive and motor skills. Whole-brain voxel-wise analyses showed a significant bilateral increase of CBF at rest in several limbic structures, including hippocampi and visual areas. They also showed significant decrease in grey matter density and fibre density in the same limbic structures, language, motor and social circuits. This paradoxical coexistence of hyperperfusion and structural deficits within memory networks suggests functional resilience or compensatory mechanisms. These preserved visual and memory functions strongly contrasted with their impaired language, motor, and social abilities. Patients appeared to rely on visual cues and memory to compensate for deficits in verbal communication. These findings support the development of individualized and innovative interventions that build on preserved visual and memory abilities in children with pathogenic ARID1B variants.

Introduction

The ARID1B gene (also known as AT-rich interaction domain-containing protein 1B) encodes a chromatin remodelling protein belonging to the SWI/SNF family. The phenotypic spectrum associated with pathogenic ARID1B variants is remarkably broad, ranging from classic Coffin-Siris syndrome (CSS) to isolated autism spectrum disorders (ASD) or intellectual disability (ID), with or without dysmorphic features [1]. Pathogenic ARID1B variants are among the most common genetic causes of intellectual disability and autism spectrum disorders in large-scale exome sequencing studies [2, 3].

Most studies regarding cognitive dysfunctions have focused on speech delay, motor delay and learning difficulties, which affect up to 65.5% of patients [4]. Fine and gross motor impairments are observed in more than 90% of cases, and hyperactivity is reported in approximately 30% of affected individuals [1]. Conversely, while parents frequently mention the remarkable memory and limited attention span of patients, little interest has been paid to their attentional, neuro-visual and mnemonic skills, particularly regarding visuospatial memory.

Brain imaging investigations in patients presenting pathogenic ARID1B variants remain scarce. Anatomical anomalies have been described in 30% of patients [1], particularly corpus callosum anomalies and posterior fossa arachnoid cysts [1, 5, 6]. However, no studies using a multimodal neuroimaging approach in carriers of pathogenic ARID1B variants have been published to date, and the underpinnings of their cognitive dysfunctions and behavioural anomalies remain largely unexplored.

Recent advances in neuroimaging techniques provide valuable tools for investigating both cerebral perfusion and brain architecture. Arterial Spin Labelling (ASL) allows assessment of cerebral blood flow (CBF) at rest, while diffusion imaging and Fixel-Based Analysis (FBA) allows in-depth assessment of micro- and macrostructural properties of white matter (WM) [7]. Notably, FBA has been successfully used to identify axonal loss and fibre atrophy in several neurodevelopmental disorders [8].

In this study, we integrated multimodal neuroimaging techniques and dimensional view of cognition [9] in a series of patients carrying pathogenic ARID1B variants. We show that altered speech, motor and social circuitry contrasted with remarkably preserved memory skills and visual network function.

Materials and methods

Participants

Inclusion criteria were (i) known pathogenic ARID1B variant identified by whole exome sequencing or panel resequencing, (ii) availability of high-quality 3.0 T MRI with consistent protocols across analysable sequences (anatomical, ASL, and diffusion-weighted imaging) (iii) comprehensive cognitive and behavioural assessment using the Wechsler Intelligence Scales and the Developmental Neuropsychological Assessment (NEPSY-II), as well as the Autism Diagnostic Interview-Revised (ADI-R). The NEPSY-II was used to assess denomination, repetition, comprehension, figure copy, visuomotor precision, tapping, manual sequences and memory, while the Wechsler scales were used to assess implicit reasoning and visual exploration. Language and motor skills were not extensively assessed in this evaluation.

Four patients with known pathogenic ARID1B variant were excluded (three due to excessive agitation preventing MRI acquisition and one due to age over 25 years). A total of 12 patients ascertained between 2015 and 2024 were included in the study (5 males, 7 females; age = 13.8 ± 4.7 years; one patient did not complete the full clinical evaluation). None of the 12 patients included presented epilepsy.

The control group consisted of 34 age-matched healthy volunteers (14 males, 20 females; mean age = 13.6 ± 4.7 years), who underwent the same imaging sequences using the same 3.0 T MRI scanners. All parents of patients and healthy controls provided written informed consent. This prospective and monocentric study obtained the authorisation of a French ethical committee (24.00501.000307 2023-A02668-37).

Imaging acquisition

All images were acquired using two 3.0 T scanners in the Paediatric Radiology Centre at Necker Children’s Hospital in Paris. The scanners used the same 3.0 T magnet, with a system upgrade in 2023. The hardware was upgraded (“lift”) and used identical imaging protocols. The system was a General Electric Discovery MR750 3.0 T from 2015 to July 2023 and a Signa Premier 3.0 T from September 2023 to 2024.

The following sequences were obtained for all participants:

3D T1-weighted Fast Spoiled Gradient Echo sequence (FSPGR) (156 axial slices; repetition time (TR) = 6.9 ms; echo time (TE) = 3 ms; resolution in plane = 1 × 1 mm; slice thickness = 1 mm; flip angle (FA) = 13°), axial and coronal FSE T2-weighted imaging, and FLAIR sequences.

Arterial Spin Labelling (ASL) perfusion images were performed with a 3D pseudo continuous arterial spin labelling (3D pcASL) sequence with a fast spin echo acquisition and spiral K-space sampling. Imaging parameters included a repetition time (TR) of 4453 ms, echo time (TE) of 10.96 ms, and a post-labelling delay (PLD) of 1025 ms. The field of view (FOV) was 24 × 24 cm with a slice thickness of 4 mm, and the matrix size was 128 × 128, yielding an in-plane resolution of 1.875 × 1.875 mm.

Diffusion imaging sequences were acquired on the Signa Premier 3 T scanner using 60 uniformly distributed diffusion directions (b = 3000 s/mm²). Imaging parameters included an isotropic resolution with a voxel size of 0.859 × 0.859 mm, slice thickness of 2.3 mm, and a TR of 7000 ms with a TE of 75 ms. An additional reversed-phase encoding scan (b = 0) was performed to correct for susceptibility-induced distortions. The diffusion MRI acquisition lasted 7 min, while the reversed b = 0 scan was completed in 56 s. The diffusion sequence was performed on the same Signa Premier 3.0 T.

No participants were sedated during the MRI procedures, and the average scan time was 35 min.

Data preprocessing

3DT1 preprocessing

The images were preprocessed using Statistical Parametric Mapping 12 (SPM12, Wellcome Centre for Human Neuroimaging, London, UK, www.fil.ion.ucl.ac.uk/spm). Structural 3D T1-weighted FSPGR images were spatially normalised using the template from the Montreal Neurological Institute (MNI) and segmented into grey matter (GM), white matter (WM) and cerebrospinal fluid (CSF) using the CAT12 toolbox (http://www.neuro.uni-jena.de/cat/) and the tissue probability maps in SPM12. Unified segmentation enables spatial normalisation, tissue segmentation and bias correction within the same generative model [10]. Finally, the normalised, segmented and modulated GM images were smoothed using a Gaussian kernel of 10 mm full-width at half-maximum (FWHM). We also performed a quality check after preprocessing, and patients with failed segmentation due to significant motion or artefacts were excluded.

ASL preprocessing

All ASL-MRI images were first visually checked for major artefacts (e.g. motion) and then analysed using SPM12. The co-registration between the native ASL image and the native GM image was estimated for each participant. The combination of the deformation fields and the estimated co-registration was then applied to the ASL-MRI to put all images in the MNI space. The resulting images were checked for proper registration and smoothed with a Gaussian kernel of 6 mm FWHM. As with the 3DT1 images, a quality check was performed after preprocessing with exclusion of patients whose preprocessing failed due to important motion or artefacts.

Diffusion Imaging preprocessing

Preprocessing of the diffusion-weighted imaging (DWI) data was performed using command-line scripts via the MRtrix3 platform [11]. The preprocessing pipeline included several essential steps: denoising, correction of Gibbs ringing artefacts, and correction for distortion, susceptibility-induced contrast and head motion with the reversed b0 sequence. Additionally, bias field inhomogeneity was corrected. Quality control ensured effective corrections, with visual inspection confirming anatomical consistency and absence of artefacts.

Fixel-based analysis (FBA) was used to estimate both the cross section of individual fixels (fibre bundle cross-section [FC]) and the density of fibres within a fixel (FD) [12]. The product of these two measures (fibre density and bundle cross-section, or FDC) provides an estimation of the total number of fibres in a fibre bundle.

FBA was performed according to the recommended pipeline of MRtrix3 [11, 12]. Tissue response functions for WM and CSF were computed using the Dhollander algorithm [7], and an average response function was derived per tissue type across all participants. The diffusion MRI data, which included only two b-values (for WM and CSF), b0 and b3000, were upsampled to an isotropic voxel size of 1.3 mm³ to enhance spatial resolution. Fibre orientation distributions (FODs) were estimated using multishell multitissue constrained spherical deconvolution (CSD) with the group-averaged tissue response functions [13].

An unbiased WM FOD template was generated from all participants, to which individual FOD images were registered nonlinearly. The resulting transformed FODs were segmented to create fixels. FD was calculated as the integral of the warped FOD lobe corresponding to each fixel. FC was derived from the warp fields computed during the registration of individual FODs to the template space. The FC values were log-transformed for downstream analysis to centre the data around zero. The combined measure, FDC, was computed as the product of FD and FC [7].

Whole-brain probabilistic tractography was performed on the FOD template, enabling the selection of specific tracts of interest for further analysis. Spherical deconvolution-informed filtering of tractograms (SIFT2) was applied to reduce reconstruction bias and remove residual artefacts.

Statistical analyses

Differences between age and sex of patients and controls were analysed with R commander [14]. We performed a t-test to compare age variables and Fisher’s test to compare sex variables after checking the normality distribution of the variables using Shapiro-Wilk test.

Radiological analyses

All anatomical images (T1, T2, FLAIR) for both patients and healthy controls were visually examined by a senior paediatric neuroradiologist.

Group comparison of relative cerebral blood flow at rest

To compare rCBF at rest obtained from ASL images, between groups, we performed a whole-brain voxel-wise analysis within the framework of the general linear model (GLM) in SPM12. Age and scanners were considered as covariates in the statistical model. Voxels included in an explicit binary mask, corresponding to the cortical GM excluding the cerebellum were considered (due to the presence of posterior fossa arachnoid cyst). Statistical thresholds were set at the cluster-level at p < 0.05 corrected for multiple comparisons using family-wise error (FWE) correction.

Voxel based morphometry (VBM) group comparison

To investigate cortical GM differences between groups, a whole-brain voxel-wise analysis was performed within the framework of the GLM in SPM12. Age and scanners were considered as covariates in the statistical model. Analyses were performed on the smoothed, normalised GM images. The significance threshold was set at p < 0.05 FWE correction for multiple comparisons.

Region of interest (ROI) and hippocampus volume analysis

Bilateral hippocampal volumes were extracted using FreeSurfer v7.3.2 (http://surfer.nmr.mgh.harvard.edu) and exported in a table for analysis. Statistical analyses were conducted in R to compare hippocampal volumes between patients and healthy controls. We used a GLM to perform a linear regression on hippocampal volumes. The significance threshold was set at p < 0.05. Hippocampal volume comparisons between patients and controls were performed using analyses including intracranial volume as a covariate.

Statistical FBA

We conducted statistical analyses of the FBA metrics: FD, FC, and the combined measure FDC to compare patients with ARID1B to healthy controls across all fixels. These fixel-wise comparisons were implemented using a GLM, controlling for age. Two contrasts were tested: Patients with ARID1B < controls and Patients with ARID1B > controls. Connectivity-Based Fixel Enhancement was used to examine group differences in FD, FC, and FDC. Statistical significance was assessed via non-parametric permutation testing (5000 permutations) and fixels were considered significant at p < 0.05 after correction for multiple comparisons using FWE.

Results

For the ASL and VBM analyses, the final sample included 12 individuals with pathogenic ARID1B mutations (5 males; age = 13.8 ± 4.7 years) and 34 controls (14 males; age = 13.61 ± 4.70 years). The groups were comparable in age (t = 0.607; p = 0.53) and sex distribution (F = 1.184; p = 0.64; odds ratio = 1.186).

For the FBA analysis, the final sample included 24 individuals: seven with pathogenic ARID1B mutations (3 males; age = 15.5 ± 2.90 years) and 17 controls (10 males; age = 14.17 ± 4.12 years). The groups were comparable in age (t = 0.609; p = 0.549) and sex distribution (F = 1.175; p = 0.732; odds ratio = 1.177).

For the region of interest (ROI) and hippocampus volume post-hoc analysis, the final sample included 30 individuals: 10 with pathogenic ARID1B mutations (5 males; age = 13.83 ± 4.69 years) and 20 controls (14 males; age = 13,61 ± 4.70 years). The groups were comparable in age (t = 0.607; p = 0.537) and sex distribution (F = 1.184; p = 0.642; odds ratio = 1.186).

Clinical evaluation

A total of 11/12 patients underwent full clinical assessment (Supplementary Table). According to the Wechsler Intelligence Scale, IQ couldn’t be calculated as the cognitive capacities of our patients were heterogeneous, but 7/11 children presented important learning difficulties. Language impairment was present in 11/11 patients, ranging from non-verbal to mild articulation difficulties despite preserved comprehension and non-verbal reasoning. Fine motor difficulties including handwriting, drawing, and tool use were present in 6/11 patients, who were unable to copy or write. Two patients showed borderline performance due to poor graphomotor coordination (Supplementary Table).

According to the Autism Diagnostic Interview-Revised (ADI-R), all patients scored for ASD diagnosis. For the social interaction module and restrictive/repetitive behaviours/interest modules, all scores were considerably higher than the cut-off values. Verbal communication scores were hardly interpretable, as only 5/11 patients presented verbal language. For non-verbal communication scores, 5/11 did not reach cut-off values (Supplementary Table).

The NEPSY-II scale revealed intact short-term memory and performance, above the 50th percentile in delayed visual memory tasks in all patients. Notably, seven patients scored above the 75th percentile (Fig. 1). Significant discrepancies between verbal and visual abilities were observed, with verbal abilities being on average 1.5 SD lower than visual abilities (t = −3.61; p = 0.004). The NEPSY-II was used to assess language, copying abilities and memory, while the Wechsler scales were used to assess vocabulary, implicit reasoning and visual exploration.

Fig. 1. Scaled scores of the clinical profile of patients carrying ARID1B variants in expressive language, visual exploration, reasoning, motor skills and memory.

Vocabulary, implicit reasoning and visual exploration are assessed with Vocabulary, Matrices and Visual puzzles from the Wechsler scale. Copy, short-term and long-term memory are assessed with copy of figures and memory of figures from NEPSY-II. Scores above 5 show preserved functions, scores under 5 show deficits. Visual exploration showed preserved abilities, while short and long-term memory were enhanced compared to language and copy.

Moreover, memory performance exceeded other cognitive and motor skills by approximately 5 SD (t = 6.75; p = 0.00004), the latter being markedly altered (Fig. 1). For example, several families described the children as “navigators”, capable of accurately recalling detailed routes to familiar places. They also frequently mentioned an ability to recognise faces and people across time, remember objects they had seen only a few times several years earlier, or recall specific events and dates with an impressive accuracy. Preschool children could consistently identify bibs of all classmates without error every day, or successfully complete a Rummikub game by memorising and reproducing complex sequences. One parent reported making use of her child’s exceptional memory skills to support early reading through recognition of familiar written names.

Radiological analyses

Visual analysis of MRI brain scans by a senior paediatric neuroradiologist showed corpus callosum anomalies in all 12 patients, consisting of callosal dysmorphia rather than complete or partial agenesis, while no structural abnormalities of the fornix were identified. In addition, posterior fossa arachnoid cysts were present in 8/12 patients.

Group comparison of relative cerebral blood flow at rest

Whole-brain voxel-wise analyses showed a significantly decreased rCBF at rest in left and right supramarginal and angular gyri of patients compared to controls (p < 0.001, FWE, Fig. 2d; Table 1).

Fig. 2. Significant differences in rCBF and GM between in patients carrying pathogenic ARID1B variants compared to healthy controls.

A Significantly increased rCBF at rest in limbic structures and occipitotemporal regions in patients compared to controls. Left: SPM glass brain showing significantly increased rCBF at rest values (p < 0.05, FWE) in the bilateral hippocampi, bilateral parahippocampal gyri, amygdala, bilateral anterior thalami, bilateral posterior cingulate gyri, lingual gyri and fusiform gyri. Right: These regions are overlaid onto left and right sagittal, coronal and axial T1-weighted anatomical MRI template images. B Significantly increased GM density bilaterally in the posterior cingulate gyri in patients compared to controls. Left: SPM glass brain showing significantly increased GM values (p < 0.05, FWE) bilaterally in the posterior cingulate gyri in patients compared to controls. Right: These regions are overlaid onto left and right sagittal, coronal and axial T1-weighted anatomical MRI template images. C Axial, coronal and sagittal 3D views of the significant clusters from ASL and VBM analyses. 3D views of the clusters including the limbic system and lingual and fusiform gyri (in red) obtained from ASL analysis and the posterior cingulate gyri (in green) from the VBM analysis. D Significantly decreased rCBF at rest values (p < 0.05, FWE) in the bilateral supramarginal and angular gyri in patients compared to controls. Left: SPM glass brain showing significantly decreased rCBF at rest values (p < 0.05, FWE) in the supramarginal gyri and angular gyri. Right: These regions are overlaid onto left and right sagittal, coronal and axial T1-weighted anatomical MRI template images. E Significantly decreased GM density (p < 0.05, FWE) bilaterally in the limbic system, i.e. hippocampi and striatum, and also in the right superior temporal sulcus and temporal pole in patients compared to controls. Left and right: T1-weighted anatomical MRI template images showing significantly decreased GM density (p < 0.05, FWE) in superior temporal sulcus and temporal pole, hippocampi and striatum. F Axial, coronal and sagittal 3D views of the significant clusters from ASL and VBM analyses. 3D views of the clusters including the limbic system and basal ganglia (in blue) obtained from VBM analysis and the supramarginal and angular gyri (in purple) from the ASL analysis.

Table 1.

Whole-brain analysis comparing cerebral blood flow (rCBF) at rest and grey matter (GM) density between patients with pathogenic ARID1B variants and healthy controls.

Results	Localisation	Structures	FWE-corrected p value	T	Z	MNI coordinates
						x	y	z
Decreased rCBF	Inferior Parietal Lobule	Supramarginal gyri	0.001	4.49	4.04	−55 60	−42 −44	46 44
		Angular gyri	0.0001	5.96	5.04	−31 −27	−54 −52	42 42
Increased rCBF	Limbic system	Hippocampi	0.001	6.66	5.47	−27 26	−18 −18	−13 −13
		Parahippocampal gyri	0.006	5.86	4.98	−27 30	−26 −25	−15 −20
		Amygdala	0.007	5.95	5.04	−24 24	−6 −4	−12 −10
		Posterior cingulate gyri	0.002	6.41	5.33	−15 14	−45 −42	−2 0
	Thalamus	Thalamus	0.006	6.01	5.07	−16 15	−23 −23	3 2
	Occipital lobe	Lingual gyri	0.029	5.62	4.81	−15 16	−70 −70	−10 −10
	Temporal lobe	Fusiform gyri	0.024	5.54	4.77	−28 26	−56 −54	−10 −10
Decreased GM	Limbic system	Hippocampi	0.002	6.27	5.43	−28 30	−36 −15	−21 −22
		Amygdala	0.0001	6.99	5.9	−21 15	−6 −2	−6 −12
	Basal ganglia	Palidum	0.003	6.21	5.4	−21 −20	2 6	3 −4
		Putamen	0.001	6.48	5.57	−30 12	−14 8	−9 −8
	Temporal Lobe	Superior temporal sulcus	0.02	4.86	4.42	46	−10	−15
		Temporal pole	0.02	4.09	3.81	38	20	−10
Increased GM	Limbic system	Posterior cingulate gyri	0.0001	8.22	6.63	14 −15	−44 −45	8 9

The table displays significant clusters of decreased and increased rCBF and GM density in patients compared to controls, with corresponding anatomical localizations, statistical values (p-value, T-score, Z-score) and Montreal Neurological Institute (MNI) coordinates.

Conversely, whole-brain voxel-wise analyses showed a significantly increased rCBF at rest in left and right hippocampi (p = 0.001, FWE), parahippocampal gyri (p = 0.006, FWE), amygdala (p = 0.007, FWE), anterior thalami (p = 0.006, FWE), posterior cingulate gyrus (p = 0.002, FWE) and visual areas, namely lingual gyri (p = 0.029, FWE), and fusiform gyri (p = 0.024, FWE) of patients compared to controls (Fig. 2a; Table 1).

Voxel based morphometry group comparison

Whole-brain voxel-based morphometry (VBM) analyses showed a significantly decreased GM volume in left and right hippocampi (p = 0.002, FWE), amygdalae (p = 0.0001, FWE), pallidum (p = 0.003, FWE), and putamen (p = 0.001, FWE) of patients compared to controls. Additionally, decreased GM volume was found in the right superior temporal sulcus (p = 0.02, FWE) and right temporal pole (p = 0.02, FWE, Fig. 2e; Table 1).

Conversely, a significantly increased GM volume was observed in left and right posterior limbic system located in the posterior cingulate gyri (p = 0.0001, FWE) of patients compared to controls (Fig. 2b; Table 1).

Region of interest (ROI) and hippocampus volume analysis

Volumetric analyses focusing on whole hippocampi and their head, body, and tail revealed a significantly reduced bilateral hippocampal volume in ARID1B patients compared to controls (p < 0.01).

After inclusion of intracranial volume as a covariate, patients still exhibited significantly smaller hippocampal volumes compared with controls (p < 0.01).

FBA group comparison

FBA allows for in-depth assessment of micro- and macrostructural properties of WM. As previously mentioned, three key fixel-based metrics were examined here, namely FD FC and their product, FDC.

Whole-brain FBA revealed a significantly reduced left and right FD of patients compared to controls across multiple WM tracts, including the superior cerebellar peduncle, anterior commissure, fornix, corpus callosum and arcuate fasciculus (Table 2). Additionally, a significantly reduced FC was observed in left and right corticopontine tract, superior and middle cerebellar peduncles and arcuate fasciculus (Table 2). Finally, a significantly reduced FDC was observed in the corpus callosum and four main circuits, namely (i) the Papez circuit, with bilateral reduction in FDC affecting the hippocampofugal fibres of the fimbria, parahippocampal tracts and bilateral fornix; (ii) the entire corticopontine tract, particularly fibres originating from the pons to the primary cortices (Brodmann areas 4 and 6); (iii) the arcuate fasciculus, including both dorsal and ventral pathways; (iv) the cerebellotemporal pathways, comprising two tracts: the cerebellothalamocortical (CTC) tract, connecting the dentate nucleus to the red nucleus (via the superior cerebellar peduncle), the ventral thalamus (VA/VL) and finally the amygdala and temporal lobe (via the anterior commissure), and the cortico-ponto-cerebellar (CPC) tract, from the temporal lobe (Brodmann areas 44 and 45) to the pons through the retrolenticular part of the internal capsule and lateral cerebral peduncle, and then to the cerebellum via the middle cerebellar peduncle. Finally, no significant increase was observed in any FBA metrics in any tract in patients compared to healthy controls (Table 2, Fig. 3).

Table 2.

Whole-brain fixel-based analysis comparing patients with pathogenic ARID1B variants to healthy controls.

Tracts	FD	FC	FDC
Corpus callosum	X		X
Fornix	X		X
Fimbria			X
Hippocampofugal fibers			X
Anterior thalamus			X
Anterior comissure	X		X
Internal capsule			X
Acurate fasciculus	X	X	X
Corticospinal tract		X	X
Cortico-thalamic-cerebellar tract			X
Cortico-ponto-cerebellar tract			X
Superior cerebellar peduncle	X	X	X
Middle cerebellar peduncle	X	X	X

The table summarizes the involvement of specific white matter tracts based on fibre density (FD), fibre bundle cross-section (FC) and fibre density and bundle cross-section (FDC). An “X” indicates significant alterations in the respective measure for each tract. Key findings include widespread disruptions in FDC across multiple tracts, with selective involvement of FD and FC in specific pathways such as the superior cerebellar peduncle, arcuate fasciculus and anterior commissure. These results highlight the extensive impact of ARID1B-related neurodevelopmental disorders on both structural and functional neural connectivity.

Fig. 3. Fibre tract-specific reductions in patients carrying ARID1B variants from whole-brain FBA.

Streamline segments were cropped from the template tractogram to include only streamline points that correspond to significant fixels (FWE-corrected P-value < 0.05). Streamlines were coloured with colours representing the FOD in the patient group compared to the healthy control group (HC) for fibre density (FD), fibre bundle cross-section (FC) and FDC. Significant streamlines are displayed across axial slices of the population template map.

Discussion

In this study, we used a multimodal imaging approach to investigate brain anomalies in patients presenting pathogenic ARID1B variants and the underpinnings of their clinical and cognitive profiles. Cognitive assessments showed that patients exhibited a distinctive neurocognitive profile, characterised by a striking dissociation between preserved visuospatial memory and visual-spatial skills, and significant impairments in language, social interactions and fine motor coordination. Brain imaging showed anatomofunctional abnormalities mainly located in regions from the limbic, motor and language networks.

Owing to their remarkably preserved visual and visuospatial abilities, many patients seemed to rely on visual cues to compensate for their deficits in verbal communication. Clinical assessment revealed excellent facial and spatial recognition. Non-verbal children were reportedly able to master sign language and complex visual tools for communication. This clinical profile is consistent with our findings of an overt dissociation between verbal and visual-spatial abilities. Median performance scores in the visual domain far exceeded verbal performance across the cohort. Importantly, these preserved visual-spatial skills were supported by consistent functional neuroimaging findings. Indeed, ASL imaging revealed increased rCBF in the lingual and fusiform gyri, two crucial structures in visual recognition, facial processing, and spatial orientation [15]. This increased perfusion suggests sustained activity in visual circuits, consistent with the remarkable visual-spatial skills of the patients. ASL imaging also revealed significantly increased rCBF within key regions of the limbic circuit consistent with an efficiently engaged memory network, including the bilateral amygdala, hippocampi, parahippocampal gyri and posterior cingulate cortex.

Conversely to increased rCBF, we found a significant reduction of GM density, as well as white matter abnormalities, in several limbic regions. Indeed, structural MRI showed GM reduction in the hippocampus and amygdala bilaterally but an increased volume of the posterior limbic system located in the posterior cingulate cortex. Consistently, FBA identified widespread WM anomalies with significant reduction in FDC within the limbic circuit. This was particularly true for the bilateral fornix and hippocampofugal fibres of the fimbria, anterior thalamus and anterior commissure tracts of the patients. This paradoxical dissociation is suggestive of a form of functional resilience or compensatory mechanisms within the memory network, consistent with the clinical observation of preserved, and sometimes enhanced, visual memory skills of these patients.

Interestingly, Ellegood et al. [16] reported preserved memory performance and pronounced hippocampal structural anomalies in an ARID1B-haploinsufficient mouse model, suggestive of compensatory mechanisms similar to those reported here. Anatomical and microstructural studies of the mouse model showed reduced dendritic complexity and abnormal spine morphology in hippocampal neurons, mirroring the brain imaging findings observed in carriers of pathogenic ARID1B variants.

When considered as a whole, our results support the view of a resilient memory network, remaining functionally active—and even over-engaged—in carriers of pathogenic ARID1B variants, despite underlying GM and WM reduction. These findings highlight the remarkable robustness of their memory system, which stands out in contrast to the broader cognitive impairments observed in many other brain functions.

It remains open to question whether this peculiar pattern of dissociation between cognitive capacities is distinctive of pathogenic ARID1B variants. It is worth remembering that the dissociation reported here has not been reported in ASD, nor in carriers of pathogenic variants in the SHANK2, GRIN2A, or HTR2A genes [17]. Neither has this dissociation been reported in knockout mouse models for other Coffin-Siris disease genes (ARID2, SMARCB1, SMARCA4, SMARCE1, SOX11). As for now, this puzzling dissociation remains, to our knowledge, peculiar to carriers of pathogenic ARID1B variants.

Conversely, our abnormal brain imaging findings mirror the language deficits, motor difficulties and autistic traits that are reportedly common in ASD patients. With regard to language deficits, we found bilateral hypoperfusion in the inferior parietal lobule, particularly the angular and supramarginal gyri, two regions which are critical for phonological integration and lexical access [4, 18, 19]. Lesions in these areas are associated with alexia, agraphia, and elements of Gerstmann syndrome [20, 21]. These regions, which are part of Geschwind’s territory, connect dorsal and ventral segments of the arcuate fasciculus and play a central role in the phonological loop. The dissociation between poor spoken production and preserved comprehension in our patients likely reflects disruption in this network, consistent with expressive language deficits commonly reported in ARID1B-related disorders.

Similarly, regarding motor difficulties amongst the patients, FBA revealed significant reduction in FDC along the corticospinal tract, consistent with the known role of ARID1B in dendritic development and neuronal connectivity [22].

Regarding the autistic features that were identified in all patients, mostly in reciprocal social interactions and stereotyped behaviours, as per ADI-R scores, structural and functional imaging pointed to widespread disruptions in the social brain network. Notably, our results showed reduced GM density in the superior temporal sulcus (STS), a key region for processing social information [23, 24]. It is worth remembering that several patients presented with posterior fossa arachnoid cysts, possibly disrupting the cerebellocortical circuits involved in social cognition. It is tempting to speculate that the cerebellotemporal tract abnormalities evidenced by whole-brain FBA could contribute to cerebellar and temporal dysfunction. These findings are consistent with studies showing that posterior fossa arachnoid cysts are associated with temporal lobe hypoperfusion and altered social perception [25].

Taken together, these results suggest that while visual and memory networks are predominantly preserved in patients with ARID1B-related disorders, broader disruptions in motor, language, and social brain circuits contribute to the clinical severity of the disease. These should not be overlooked in clinical assessment and therapeutic planning.

Study limitations

This study has several limitations. First, the sample size is relatively small, owing to the scarcity of ARID1B-related disorders. Second, a selection bias may be present, as three children were too agitated to undergo MRI scanning, reflecting the inherent challenges of performing MRI in children with neurodevelopmental disorders. Third, this study lacks in-depth language and motor assessments. Indeed, our data do not allow distinguishing linguistic abnormalities from major oral-motor difficulties. Finally, two different 3.0 T MRI scanner systems were used to perform brain imaging. Both systems shared the same magnet, with the later system being installed through an upgrade (“lift”) of the original magnet. However, scanner type was included as a covariate in all statistical models as applicable. Future studies integrating quantitative behavioural scales and brain imaging data will strengthen the clinical relevance of our findings.

Perspectives and conclusion

This study provides new insights into the neurocognitive profile associated with pathogenic ARID1B variants, highlighting a consistent dissociation between preserved and impaired functions. Memory and visual processing emerged as relative strengths, both clinically and on imaging, with preserved or even increased functional activity in key limbic and occipitotemporal regions. In contrast, language, fine motor skills, and social interaction were consistently impaired and mirrored disruptions of the corresponding brain networks. These findings support the development of individualised interventions that would build on preserved visual and memory functions [26]. Future studies will hopefully confirm and extend these results in larger cohorts and clarify how these brain–behaviour patterns evolve over time in this patient population.

Author contributions

AF. included the patients, performed the neuropsychological assessments, conducted 3D T1 and ASL preprocessing, carried out the statistical analyses and drafted the original manuscript. KA. retrieved all genetic diagnoses, and contributed to MRI inclusion and data preprocessing. JB and LF. performed preprocessing and analyses of diffusion-weighted imaging data, including fixel-based analysis. LG. contributed to the identification of white matter tracts in fixel-based analysis results, as well as to the localization of clusters in grey matter and ASL analyses. AS. contributed to patient inclusion and manuscript revision. MR, JA, and CM. enabled the recruitment of ARID1B patients. VD-R and SC. reviewed all MRI data and contributed to manuscript writing. VC-D. authorized the collaboration with the medical genetics department and revised the manuscript. AM provided access to additional patients, contributed to data interpretation, and revised the manuscript. MZ. contributed to the conceptualization of the study and revised the manuscript. NB. supervised the entire project, from patient inclusion to manuscript preparation.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability

All the data are available are available from the corresponding author on request.

Competing interests

The authors declare no competing interests.

Ethics statement

All methods were performed in accordance with the relevant guidelines and regulations.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-026-04090-7.