MRI Characterization of Structural Brain Abnormalities in NGLY1 Deficiency

ABSTRACT

Background and Objectives

NGLY1 (N‐Glycanase 1) deficiency is an ultrarare autosomal recessive disorder affecting ~165 patients worldwide, characterized by developmental delay, hyperkinetic movement disorders and shortened life expectancy. Despite its severe neurological manifestations, comprehensive neuroimaging characterization has been limited to case reports and small descriptive studies. Our objective was to investigate alterations in brain morphology in patients with NGLY1 deficiency and determine whether these metrics associate with clinical phenotypes.

Methods

This case series analysed real‐world MRI scans performed on 11 patients with molecularly diagnosed NGLY1 deficiency between 1999 and 2023 at sites across the globe. Ages ranged from 2 to 19 years at scan time (five female and six male). Cortical and subcortical morphology, including subcortical volume and cortical thickness, surface area, volume and curvature, was measured with three‐dimensional T1‐weighted magnetic resonance imaging (MRI) scans. Z scores were calculated using normative models from CentileBrain for patients > 3 years old or custom models for patients < 3 years old. Clinical phenotypes were matched to Human Phenotype Ontology codes.

Results

Sixteen scans from 11 patients met quality criteria for analysis. Both age groups (under and over 3 years old) showed significantly reduced subcortical volumes, particularly in bilateral thalamus and putamen. Younger patients demonstrated widespread reductions in cortical surface area, volume and curvature, indicating altered gyrification patterns. Older patients showed thinner dorsal and thicker ventral cortical regions with limited surface area reductions. Thalamic volume reduction in older patients correlated with gait disturbance, dysphagia and EEG abnormalities, with additional cortical associations with sleep and hearing abnormalities. Seizure presence in younger patients correlated with altered cortical thickness, surface area and curvature patterns.

Discussion

NGLY1 deficiency is associated with pervasive alterations in brain development affecting both subcortical and cortical morphology. Age‐dependent patterns of cortical alterations indicate disrupted neurodevelopmental trajectories that may reflect impaired neuronal migration and/or altered synaptic pruning. Correlations with clinical variables suggest that these measures may serve as useful biomarkers for tracking disease progression and/or treatment efficacy. These findings provide a comprehensive neuroimaging characterization of NGLY1 Deficiency and establish a foundation for understanding brain structure‐function relationships in this ultrarare disorder.

Structural MRI from 11 patients with NGLY1 deficiency, an ultrarare neurodevelopmental disorder, revealed reduced subcortical volume (thalamus and putamen) across ages and reduced cortical surface area in younger patients. These measures tracked with clinical features such as seizures and gait disturbance, supporting brain imaging as a candidate disease biomarker.

1. Introduction

NGLY1 (N‐Glycanase 1) deficiency is an ultrarare, debilitating autosomal recessive disorder with ~165 patients identified worldwide to date (Hasan and Warsi 2023; Grace Science Foundation 2025), although one study reports the true incidence may be closer to 12 births/year in the United States (Stanclift et al. 2022). It is caused by loss of function mutations in the NGLY1 gene that impair endoplasmic reticulum–associated degradation (ERAD) and the clearance of abnormal glycoproteins within the cytosol (Pandey et al. 2022). Patient phenotypes are heterogeneous, varying in prevalence and severity. The core phenotypes can be characterized by severe global developmental delay and/or intellectual disability, hyperkinetic movement disorder, transient elevation of transaminases, (hypo)alacrima and progressive, diffuse, length‐dependent sensorimotor polyneuropathy. Other, more diverse, phenotypes have been described including epilepsy, scoliosis, dysphagia, EEG abnormalities, sleep disturbances and microcephaly (Pandey et al. 2022). A recent natural history study by our group followed 29 patients over a few years and characterized clinical features and disease course, noting motor decline over time (Tong et al. 2023).

Neuroimaging findings in NGLY1 Deficiency have been limited to descriptive reports and a single small‐scale systematic study. Lam et al. (2017) collected brain magnetic resonance imaging in 11 patients ranging in age from 3–21 years old, describing delayed myelination in three of four youngest individuals (under 6 years old), cerebral atrophy in six patients (ranging from slight to moderate) and slight cerebellar atrophy in four patients, with atrophy correlating strongly with functional impairment as measured by the Nijmegen Pediatric CDG Severity scale and the Vineland Adaptive Behavior Scale. Additional case reports have described nonspecific findings including delayed myelination, atrophy of the corpus callosum, anterior commissure and/or cerebellum, prominent cisterna magna and enlarged ventricles (Kariminejad et al. 2021; Panneman et al. 2020; Abuduxikuer et al. 2020; Chang et al. 2019; Sonoda et al. 2024; Caglayan et al. 2015; Dabaj et al. 2021; Pekal et al. 2025). However, comprehensive analyses of brain morphometry have not been reported. Understanding the specific neuroanatomical profile of NGLY1 Deficiency may provide crucial insights into the pathophysiological mechanisms underlying the cognitive and motor impairments observed in affected individuals.

MRI offers a powerful, noninvasive approach to characterize brain morphology in rare genetic disorders. Advanced neuroimaging techniques, including volumetric analysis and cortical thickness measurements, can detect subtle structural alterations that may not be apparent on routine clinical imaging. In this study, real‐world MRI scans were collected from 11 patients, with 16 scans sufficient for analysis after review and processing, for assessment of the natural history and progression of brain morphology in NGLY1 Deficiency. Patient ages ranged from 2 to 19 years of age at the time of the MRI. Five of the patients were female and six were male. Given the central role of NGLY1 in cellular protein homeostasis and the neurodevelopmental phenotype associated with its deficiency, we hypothesised that affected individuals would demonstrate distinct patterns of regional brain volume reductions and altered cortical morphometry compared to typically developing controls.

2. Methods

2.1. Participants

This was a real‐world study in which the collected imaging was taken as part of the care of NGLY1 deficiency patients outside of clinical trials. Written consent was provided by parents. Participants were consented to share MRI under one of two protocols (Investigating the Biology and Natural History of the NGLY1 Related Disorder, #41906, or NGLY1 Deficiency: A Prospective Natural History Study, NCT03834987). Enrollment criteria included a molecular diagnosis of NGLY1 Deficiency (either through genetic testing or a combination of genetic testing and biomarker confirmation). Imaging was requested and transferred through secure transfer or by disc. Patient characteristics are shown in Table 1, and the Strobe Checklist is in the supplement.

TABLE 1.

Patient characteristics. Demographic and clinical information for the 11 patients, including sex, whether longitudinal MRI was available, age at scan(s) in years, gestational age (in weeks), birthweight (very low [< 1.5 kg], low [1.5–2.5 kg] and normal [> 2.5 kg]); whether patient was living as of 2023 or age at death (in years); and clinical phenotypes in HPO IDs: HP:0002015—dysphagia; HP:0002353—EEG abnormality; HP:0001288—gait disturbance; HP:0001290—generalized hypotonia; HP:0000364—hearing abnormality; HP:0002487—hyperkinetic movements; HP:0001250—seizure; HP:0002360—sleep abnormality; HP:0001337—tremor.

Patient	Sex	Longitudinal MRI	Age at scan(s) (years)	Gestational age	Birthweight	Living or age at death	Clinical phenotypes
MRI‐1	M	No	6.7	33	Very low	Living	HP:0002015, HP:0002353, HP:0001288, HP:0001290, HP:0000364, HP:0002487, HP:0001250, HP:0001337
MRI‐2	F	Yes	1.5|15.2|16.9	41	Normal	22	HP:0002015, HP:0002353, HP:0001288, HP:0001290, HP:0000364, HP:0002487, HP:0002360
MRI‐3	M	No	6.7	30	Normal	Living	HP:0002353, HP:0001288, HP:0001290, HP:0002487, HP:0001250, HP:0001337
MRI‐4	F	No	2	Not available	Normal	Living	HP:0001288, HP:0001290, HP:0002487, HP:0001250, HP:0002360
MRI‐5	F	No	19.2	40	Low	19.X	HP:0002015, HP:0001288, HP:0001290, HP:0002487, HP:0002360, HP:0001337
MRI‐6	M	Yes	0.7|3.9|5.2	39	Low	Living	HP:0001290
MRI‐7	F	Yes	12.2|12.5	40	Low	~13/14	HP:0002015, HP:0002353, HP:0001288, HP:0001290, HP:0001250, HP:0002360, HP:0001337
MRI‐8	M	No	1.4	36	Low	Living	HP:0002487, HP:0001250
MRI‐9	M	No	1.1	33	Not available	Living	HP:0001250
MRI‐10	M	No	2.3	39	Not available	Living	HP:0001290, HP:0002487
MRI‐11	F	No	4.9	36	Low	Living	HP:0000364, HP:0001250

2.2. MRI Collection

Participants were scanned at multiple sites around the world in the course of their medical care, and there was no consistent protocol used to collect data. In total, 41 scans from 16 patients were assessed for potential inclusion. Twenty‐eight were low resolution or had poor contrast or a cropped field of view. Three of these were rescued using SynthSeg, a tool from FreeSurfer to enhance the resolution of clinical‐grade scans and generate more accurate segmentations (Iglesias et al. 2023; Iglesias et al. 2021). Scan details for the final 16 acceptable for inclusion are provided in Table S1, including scan manufacturer and model, field strength and voxel size. Several example patient MRI images are provided in Figure S1. Half of the scans came from a 1.5‐T scanner and half were from a 3‐T scanner. Voxel resolution varied considerably, from 0.5‐mm isotropic to in‐plane resolution = 0.9 × 0.9 mm and slice thickness = 6.5 mm.

2.3. MRI Processing

All MRI scans from patients over 3 years old were processed through FreeSurfer Version 7.4 (Fischl 2012). Most participants had multiple T1‐weighted MRIs. All were processed and thoroughly visually quality checked. From the FreeSurfer aseg and aparc files, we extracted subcortical volume (7 regions bilateral), cortical surface area (31 regions bilateral) and cortical thickness (31 regions bilateral). The entorhinal cortex and parahippocampal gyrus were excluded from analyses, as the medial surfaces of the temporal lobe are frequently poorly segmented. FreeSurfer is not recommended for participants under 4 years old, so the infant‐specific version, infantFS, was used for younger patients (Zöllei et al. 2020). Image resolution varied, with multiple patients having voxel resolutions too large for good quality FreeSurfer segmentations; in these cases, SynthSeg was used to rescue these scans (Iglesias et al. 2023; Iglesias et al. 2021). SynthSeg is not available for infants, so analyses on patients < 3 years old were completed only on native resolution scans. These results are reported in the supplement.

2.4. Normative Modelling (Patients > 3 Years Old)

As there were no comparison participants scanned at each clinical site, we processed scans through CentileBrain, which compares values to a model trained on over 40,000 individuals between the ages of 3 and 90 years (Frangou et al. 2021; Dima et al. 2021; Ge et al. 2024). This yielded Z scores for each brain region, taking into account patient age and sex. Z scores were calculated for cortical thickness, surface area and subcortical volume.

2.5. Normative Modelling (Patients < 3 Years Old)

The normative models in CentileBrain are not valid for individuals under 3 years old, and there are no publicly available tools that provide a similar function; therefore, we created our own normative models using data accessible through the National Institute of Mental Health (NIMH) Data Archive or NDA. T1w MRIs were downloaded for all NDA participants between 6 and 36 months old for a total of 2539 files. These were filtered to exclude scans from clinical samples (e.g., autism spectrum disorders) and others that did not fit the parameters, resulting in 573 MRIs that were processed through infantFS. InfantFS failed for 25 of 573 scans. This output was thoroughly visually quality checked, leading to 131 scans excluded for overall poor‐quality segmentation. For the remaining 417 scans, we noted any regional failures, using processes outlined by the ENIGMA Consortium (Thompson et al. 2015; Thompson et al. 2022; Thompson et al. 2020). The most common areas of failure were the medial temporal lobes (as noted above) and the medial orbitofrontal cortex. For the scans for 6 and 7 month olds, the quality of the temporal lobe segmentation was poor in too many scans and therefore excluded. The number of control scans in the final model is shown in Table S2, broken down by age and sex. As above, we extracted subcortical volume, cortical thickness and cortical surface area from these scans, with the addition of cortical volume and curvature. For each sex, mean trajectories were modelled in R4.4.1 using basis splines with 3° of freedom to capture developmental patterns while preventing overfitting. Standard deviation was estimated by fitting a separate spline model (2° of freedom) to the absolute residuals from the mean model, with correction for the expected value of absolute deviations. This approach generated smooth, age‐specific percentile curves (± 1SD and ± 2SD) that account for both mean changes and variance heterogeneity across development. Growth curves for each morphological measure as shown in Figures S2–S8, with separate curves for males and females. Trajectories for temporal lobe regions start at 11 months due to the sparsity of high‐quality segmentations as noted above. NGLY1 patient metrics were converted to standardized Z scores by comparing observed values to the sex‐ and age‐specific normative mean and standard deviation derived from these models. Z scores were calculated for cortical thickness, surface area, curvature and volume, and subcortical volume. Cortical curvature and volume were not available through CentileBrain for the older patient group.

2.6. Additional Variables

Patient birthweight and gestational age were collected, with birthweight divided into very low (< 1.5 kg), low (1.5–2.5 kg) and normal (> 2.5 kg) categories. We additionally collected multiple phenotypes coded in human phenotype ontology (HPO) terminology. We examined the following HPO phenotypes: alacrima/hypolacrima (absence or reduced production of tears, HP:0000522), dysphagia (difficulty swallowing, HP:0002015), EEG abnormality (HP:0002353), gait disturbance (HP:0001288), generalized hypotonia (HP:0001290), hearing abnormality (HP:0000364), hyperkinetic movements (HP:0002487), seizure (HP:0001250), sleep disturbance (HP:0002360) and tremor (HP:0001337). Other HPO phenotypes were collected but not examined due to either no or limited variability across the samples (e.g., global developmental delay [HP:0001263] was present in all patients) or limited relevance to neuroimaging phenotypes (e.g., liver cirrhosis [HP:0001394]). Given the issues with data quality and the known motor phenotypes of NGLY1 deficiency, we considered whether the patients whose MRI data was of sufficient quality were less severely impacted than the general NGLY1 deficiency patient population. Comparing the incidence of each of these clinical variables between the sample with good quality MRI (N = 11) versus the full sample (N = 16), we did not see significant deviations in the expected counts (p = 0.995). This supports the assumption that the patients included in the following analyses are generally reflective of the larger patient population.

2.7. Statistical Analyses

We used a modified Bonferroni correction for multiple comparisons created by Li and Ji (2005), which yielded the effective number of independent variables (V _eff ) in our analysis as 12 and a significance threshold of p < 0.05/12 = 0.0042. A traditional Bonferroni correction was too conservative for our analysis because there are correlations between test statistics (i.e., between the 62 regions examined). Our primary analysis was to examine whether cortical and subcortical morphology differed between the NGLY1 deficiency patients and expected values. We conducted two‐sided T tests on the Z scores, testing whether they differed significantly from 0.

As secondary analyses, we examined correlations between brain morphometry and the clinical variables listed above. Given the small sample size, binary nature of clinical variables and exploratory nature of these analyses, Spearman rank correlations were computed to assess associations between clinical features and regional brain measures, as Spearman correlation is more robust to outliers and does not assume normality of the data.

A graphic explaining the cortical metrics examined is shown in Figure S9.

3. Results

All analyses were performed separately for the scans from older (> 3 years old) versus younger (< 3 years old) patients, with some longitudinal patients having scans included in both phases.

3.1. Primary Results

Z scores are shown for subcortical volume, cortical thickness, and cortical surface area (Figure 1) and displayed projected onto the brain in Figure 2. Of the 10 scans included in the older cohort, four were enhanced with SynthSeg. We reran all analyses in the older cohort excluding these scans (Figures S10–S13).

FIGURE 1.

Subcortical volume Z scores in patients. Boxplots display Z scores for each subcortical structure (n = 6 scans for 6 patients under 3 years old and 10 scans for 7 patients over 3 years old) relative to age‐ and sex‐matched normative data. Each region in the boxplot is represented by a different colour to allow easy comparison between panels. Box elements represent median, interquartile range (IQR) and whiskers extend to minimum and maximum. Symbols indicate significant deviations from normative means: ^# p < 0.05 after multiple comparisons correction (Li and Ji); *p < 0.05 uncorrected.

FIGURE 2.

Cortical thickness and surface area effects for older patients. The average Z scores for patients over 3 years old are displayed both in boxplot and projected onto the cortex for cortical thickness and surface area. Box elements (left panel) represent median, interquartile range (IQR) and whiskers extend to minimum and maximum. Symbols indicate significant deviations from normative means: ^# p < 0.05 after multiple comparisons correction (Li and Ji); *p < 0.05 uncorrected. Coloured boxes indicate lobe—dark blue = cingulate, light blue = frontal, middle two = insula, green = occipital, yellow = parietal, dark yellow = temporal. For the cortical projections (right panel), marginally significant regions (0.0042 < p < 0.05) are included to display trends. Colour corresponds to the average Z score across the group for a given region, with negative Z scores in blue and positive Z scores in red. Left in image is left in brain.

In the younger patients, we found significantly smaller volumes in the bilateral putamen and thalamus (t statistic = −5.2 – −6.6, p < 0.0042), along with marginally lower volume in the bilateral caudate (t = −3.8 – −4.1, 0.0042 < p < 0.05, Figure 1). We also found significantly lower volume bilaterally in older NGLY1 deficiency patients in the accumbens, hippocampus, pallidum, putamen and thalamus, and left caudate (t = −4.6 – −18.5), along with marginally lower volume in the right caudate (t = −3.2). Results in the subset of native resolution scans from the older patient group (n = 7) were generally consistent, although differences in the left amygdala, left putamen and right pallidum were no longer significant (Figure S10).

For cortical thickness, in younger patients, we found significantly comparatively thicker cortex than normative models in the right rostral middle frontal gyrus (t = 8.5), and marginally thicker cortex in the left pars opercularis, pars orbitalis and pars triangularis, along with the left cuneus, left superior parietal gyrus and left supramarginal gyrus (t = 3.1–8.5; Figure 1). We additionally found marginally thinner cortex in the bilateral isthmus of the cingulate (t = −2.7 to −3.1). In older patients, there was significantly comparatively thicker cortex than normative models in the left lateral occipital gyrus (t = 4.7), along with marginally thicker cortex in the left rostral anterior cingulate, lateral orbitofrontal gyrus and the right pars opercularis, insula and multiple regions of the occipital and temporal cortex (t = 2.3–3.5). In addition, we found marginally thinner cortex in the left isthmus of the cingulate, medial orbitofrontal gyrus, paracentral gyrus, pars triangularis, inferior parietal gyrus and postcentral gyrus, and the right supramarginal gyrus (t = −2.3 to −3.3). For the older patients, cortical thickness effects appear to cluster as thinner cortex in dorsal brain regions and thicker cortex in ventral brain regions. Results are largely consistent when limiting analyses in the older cohort to native resolution scans (Figure S11).

For surface area, we saw widespread areas of significantly lower surface area in the younger patients, including multiple sections of the cingulate gyrus, the bilateral insula and multiple gyri within the frontal and parietal cortices (t = −5.1 to −13.6), with additional areas of marginally lower surface area (t = −2.6 to −5.0). Differences in the older patients were less pronounced, with significantly lower surface area in the left temporal pole and right rostral anterior cingulate and fusiform gyrus (t = −3.9 to −4.9), and marginally lower surface area in the left lingual, pericalcarine and fusiform gyri, and the right caudal anterior cingulate, insula and postcentral gyrus (t = −2.3 to −3.8). Surface area was significantly higher in the left superior frontal gyrus (t = 5.6) and marginally higher in the right superior frontal gyrus, along with the right supramarginal gyrus (t = 3.0–3.2). Results were consistent in the native resolution only subset (Figure S12). All of the results comparing the native resolution subset to the full older patient sample are displayed in Figure S13.

Given the widespread differences in the younger patient cohort, we examined two additional cortical morphology measures: curvature and volume. Consistent with the effects for cortical thickness and surface area, we found significantly lower cortical curvature and volume in the NGLY1 Deficiency patients compared to typically developing infants and babies (complete results in Figure 3). There was significantly lower curvature in the right pars triangularis, rostral middle frontal, superior frontal and supramarginal gyrus, and the left insula (t = −2.6 to −4.4), along with marginally lower curvature across regions of the cingulate cortex, left frontal cortex, right insula, lateral occipital gyrus, left supramarginal gyrus, bilateral inferior temporal gyrus and right fusiform gyrus (t = −5.1 to −16.9). Volume was also significantly lower in the bilateral precentral gyrus, insula, precuneus, right cingulate cortex, fusiform gyrus, paracentral gyrus and pars opercularis (t = −5.1 to −9.2) and was marginally lower in the bilateral superior frontal gyrus, left caudal middle frontal, paracentral, pars opercularis, cingulate cortex, lateral occipital, superior parietal, fusiform, middle temporal gyrus and right inferior parietal gyrus (t = −2.6 to −4.8).

FIGURE 3.

Complete cortical morphology effects for younger patients. The average Z scores for patients under 3 years old are displayed both in boxplot and projected onto the cortex for cortical thickness, surface area, curvature and volume. Box elements (left panel) represent median, interquartile range (IQR) and whiskers extend to minimum and maximum. Symbols indicate significant deviations from normative means: ^# p < 0.05 after multiple comparisons correction (Li and Ji); *p < 0.05 uncorrected. Coloured boxes indicate lobe—dark blue = cingulate, light blue = frontal, middle two = insula, green = occipital, yellow = parietal, dark yellow = temporal. For the cortical projections (right panel), marginally significant regions (0.0042 < p < 0.05) are included to display trends. Colour corresponds to the average Z score across the group for a given region, with negative Z scores in blue and positive Z scores in red. Left in image is left in brain.

3.2. Clinical Correlations

Given the limitations of the SynthSeg data as mentioned above, we examined associations between cortical and subcortical morphology and clinical measures using the native resolution data only. Separate analyses were run for younger and older patients. In order to limit the number of statistical comparisons and to focus on clinically relevant associations, we limited the analysis to those regions that had p values below 0.10 in the primary results above. The top associations for younger and older patients are reported in Figures 4 and 5, and the rest of the results in Tables S3 and S4. Several brain regions showed identical correlation coefficients and p values with various clinical variables. This reflects the rank‐based nature of Spearman correlation: Although the absolute Z score values differed across regions, the relative ranking of subjects remained consistent, resulting in identical test statistics.

FIGURE 4.

Clinical associations, younger patients. Associations between regional cortical morphometry and seizure status in the younger patient group. Panels show cortical thickness, surface area and curvature for the regions most strongly associated with seizure (all ρ = 0.88, p = 0.021). Boxplots show differences in regional Z score based on the presence (blue) versus absence (grey) of seizure. Each panel is labelled with its region on the y axis. Statistical values for all associations are provided in Table S3. None of these associations survived correction for multiple comparisons (Li and Ji method); they should be interpreted as exploratory.

FIGURE 5.

Clinical associations, older patients. Associations between regional brain morphometry and clinical variables in the older patient group. Panels are grouped by measure, showing subcortical volume, cortical surface area and cortical thickness for the region‐measure pairs most strongly associated (all ρ = 0.87, p = 0.012). Boxplots show differences in regional Z score based on the presence (blue) versus absence (grey) of each clinical variable, indicated on the x axis; each panel is labelled with its region on the y axis. Statistical values for all associations are provided in Table S4. None of these associations survived correction for multiple comparisons (Li and Ji method); they should be interpreted as exploratory.

In the younger patients, the strongest associations (ρ = 0.88, p = 0.021) were all with seizure. Seizure was associated with thicker cortex in the right precuneus, right supramarginal gyrus and left rostral middle frontal gyrus; lower surface area in the right paracentral, right superior parietal and right lingual gyri, right precuneus and left supramarginal gyrus; and lower curvature in the left isthmus of the cingulate, right supramarginal gyrus, right insula, left pars opercularis and right rostral anterior cingulate. Additional associations at ρ = 0.83 (p = 0.042) with generalized hypotonia, gait disturbance, hyperkinetic movements, sleep abnormality and alacrima are reported in Table S3.

In the older patients, the strongest associations (ρ = 0.87, p = 0.012) centred on the thalamus. Smaller bilateral thalamic volume was associated with gait disturbance, dysphagia, EEG abnormality and alacrima. Other associations at this threshold included seizure with greater right insula surface area, hearing abnormality with smaller left temporal pole surface area and thinner left lateral occipital cortex, hyperkinetic movements with thicker left isthmus of the cingulate and sleep abnormality with smaller left lateral occipital surface area and larger left pallidum, right hippocampus and right supramarginal surface area. Weaker associations, including with tremor, are reported in Table S4.

None of these associations survived correction for multiple comparisons using the Li and Ji method (all adjusted p > 0.05; younger minimum adjusted p = 0.085, older minimum adjusted p = 0.094), and they should be interpreted as exploratory.

4. Discussion

Here, we describe alterations in the cortical and subcortical morphology of patients with NGLY1 deficiency, an ultrarare disorder that results in global developmental delay and is associated with a significantly shortened life expectancy (median age at death = 13 years old) (Stanclift et al. 2022). This comprehensive examination of brain structural abnormalities in this patient population to date reveals pervasive alterations in cortical and subcortical development. The most consistent finding across age ranges was smaller subcortical volumes, particularly for the thalamus but also including the caudate and putamen. This change in the thalamus may have a wide range of clinical consequences, given the broad functions (e.g., cognition, motor and sensory processing) and connectivity of the thalamus, as smaller thalamic volume in the older patients was associated with a number of clinical features of NGLY1 Deficiency. Effects in the cortex varied with age, with different patterns emerging in the younger versus older patient groups. Prior MRI studies in NGLY1 Deficiency have largely been case reports. Results from these have varied, with some having minimal if any structural abnormalities (Panneman et al. 2020; Abuduxikuer et al. 2020), whereas others have noted mild atrophy (Lam et al. 2017; Abuduxikuer et al. 2020; Chang et al. 2019; Sonoda et al. 2024) or thin spinal cord (Caglayan et al. 2015). One case report described more widespread atrophy along with white matter abnormalities including prominent perivascular spaces (Dabaj et al. 2021), whereas a relative expansion of fluid‐filled spaces both in and around the brain (Pekal et al. 2025).

In the younger patients, there was a trend towards thinner cortex in the cingulate and scattered areas of thicker cortex, including the right inferior frontal gyrus (pars opercularis, orbitalis and triangularis) and left parietal cortex. There were widespread areas of lower surface area in the younger patients, encompassing broad areas of the frontal, parietal, cingulate and insular cortex, with minimal effects in the temporal cortex and relative sparing of the occipital cortex. Additionally, cortical curvature and volume were lower across large areas of cortex. Lower curvature indicates less cortical gyrification, with wider, flatter gyri and shallower sulci. Taken together, these results fit with prior studies indicating that NGLY1 deficiency is associated with significant underdevelopment in brain structure.

Prior studies using patient stem cell‐derived organoids showed a reduction in neuronal and astrocytic protein markers (Abbott et al. 2023), suggesting that NGLY1 Deficiency may impair late‐stage neurogenesis and the subsequent migration of neurons to the upper cortical layers (Lin et al. 2022). Impairments in neurogenesis, neuronal differentiation and neuronal survival would have broad implications for brain structural development. Neuronal migration and layering are dependent on the guidance of radial glial cells (Llinares‐Benadero and Borrell 2019; de Vareilles et al. 2023), which also serve as progenitors for multiple neuronal populations, but the signalling necessary to maintain these is disrupted in cerebral organoids developed from NGLY1‐deficient stem cells (Lin et al. 2022). Differential expansion of the inner versus outer cortical layers contributes to cortical folding (Llinares‐Benadero and Borrell 2019), but upper‐layer neurons are affected in these organoids (Lin et al. 2022). Synaptic pruning is another key process shaping brain development, and it occurs at varying rates across the brain (Gogtay et al. 2004; Huttenlocher and Dabholkar 1997). Disruptions of protein degradation, proteostasis and proteasome function that result from NGLY1 Deficiency could impair the synaptic pruning process (Abbott et al. 2023). Taken together, there are multiple cellular mechanisms that are altered in NGLY1 Deficiency, and any or all of them could be responsible for the cortical and subcortical deficits we report.

In the older patients, there was a trend towards thinner cortex in dorsal brain regions such as the lateral parietal and occipital cortex, and thicker cortex in ventral regions such as the inferior temporal and occipital cortex. Examining gradients of cortical thickness over normal development, cortex is thickest in the dorsal frontal and temporal regions and is thinnest in the parietal and medial occipital cortex (Ball et al. 2020). Further, cortex thins fastest in the frontal pole and medial parietal cortex and thins slowest in the lateral parietal cortex during typical development. This could indicate an acceleration of the typical developmental pattern in affected patients. Thicker cortex has been reported in other developmental conditions, such as autism spectrum disorder (van Rooij et al. 2018), Down syndrome (Lee et al. 2016; Levman et al. 2019) and Williams syndrome (Thompson et al. 2005), where it is generally interpreted as reflecting altered or delayed cortical maturation rather than preserved tissue. The cerebellum, which is one of the primary areas of interest in NGLY1 deficiency given the loss of Purkinje cells, is highly connected to the cerebrum through the thalamus, the region where the most consistent deficits were observed (Yang, Torres‐Ramirez, et al. 2025; Stuut et al. 2021). Specific thalamic nuclei project to distinct cortical targets, so the variable effects on cortical morphology could be influenced by differential degeneration of different thalamic nuclei. Unfortunately, we did not have the resolution to examine effects in this level of detail, but future analyses may be able to collect more specific data. Interestingly, although younger patients showed widespread reductions in surface area across multiple brain regions compared to expected values, these differences were largely absent in the older patient cohort. This age‐related divergence in cortical phenotypes raises important interpretive questions. With limited longitudinal data available, we cannot definitively determine whether these varying effects reflect delayed neurodevelopment that partially normalizes over time or whether they result from survivor bias given the severe nature of this disorder. The latter interpretation is particularly relevant considering that the median lifespan of patients with NGLY1 Deficiency is 14.4 years (Morrison et al. 2026), with three patients in our cohort dying at ages 14, 19 and 22 years. Those who survive into adolescence and young adulthood may represent a phenotypically distinct subgroup with less severe cortical alterations, potentially explaining the more limited surface area reduction observed in the older cohort.

Both the younger and older patients displayed significantly lower subcortical volume than expected values, suggesting consistent subcortical underdevelopment in NGLY1 Deficiency. The subcortical regions that seemed to have the most robust effect sizes were the caudate, putamen and thalamus, which support a wide range of functions including motion, cognition and emotion. The subcortical volume reductions we observed, particularly in the thalamus, align with neuropathological findings reported in NGLY1 Deficiency. An autopsy case report of a 5‐year‐old with NGLY1 Deficiency demonstrated neuronal protein accumulations throughout subcortical brain regions and extensive cerebellar cell loss (Stuut et al. 2021). These protein accumulations likely result from the impaired proteostasis that characterizes NGLY1 Deficiency. Prior studies in mice have found poor brain development and accelerated neuronal degeneration, particularly in the thalamus (Asahina, Fujinawa, Fujihira, et al. 2021; Asahina, Fujinawa, Hirayama, et al. 2021; Asahina et al. 2020), whereas another recent study in rats reported evidence of widespread neuroinflammation, particularly in subcortical regions (Zhu et al. 2025). The correlation between our volumetric findings and these previously reported pathological changes suggests that the reduced subcortical volumes we detected may reflect underlying neuronal loss and dysfunction secondary to protein accumulation (Stuut et al. 2021).

Although none of the clinical associations survived correction for multiple comparisons, we discuss the uncorrected results given the exploratory nature of this study. When focusing on regions showing trend level alterations in the group analysis (p < 0.10, Tables S3 and S4), we found that thicker cortex, lower surface area and lower curvature were associated with the presence of seizures in the younger cohort. Other seizure disorders have some overlapping cortical patterns. For example, focal cortical dysplasia is a congenital brain malformation that leads to focal cortical thickening and often presents with severe, medication‐resistant seizures (Sisodiya et al. 2009), and childhood absence seizures are also associated with thicker cortex and lower volumes (Curwood et al. 2015; Tosun et al. 2011; Caplan et al. 2009). Treatment‐resistant seizures are common in NGLY1 deficiency, with only ⅓ of patients achieving full control of their seizures (Levy et al. 2022). In older patients, the most consistent finding was bilateral thalamic volume reduction associated with gait disturbance, dysphagia and EEG abnormality. A large study of patients with epilepsy (clustered into four epilepsy subgroups) reported smaller thalamic volume across all epilepsy subgroups compared to healthy controls (Whelan et al. 2018), indicating that the association between EEG abnormality and thalamic volume we observed is not specific to NGLY1 Deficiency but is consistent with thalamic involvement described broadly across epilepsy. However, because multiple clinical features co‐occurred in the same patients, our small sample size prevents us from determining whether these associations reflect overall disease severity or are specific to any one clinical variable.

Hyperkinetic movements are one of the core phenotypes of NGLY1 deficiency, but associations with neuroimaging metrics were limited in our sample. For both the younger and older patient groups, only a handful of regions showed associations, including several subregions of the cingulate along with scattered regions in the frontal, parietal and occipital cortex. In other hyperkinetic disorders, most notably Huntington's disease, degeneration is focused on the striatum (caudate and putamen), with cortical associations less common but focused on motor‐associated cortices and cingulate (Tabrizi et al. 2009). Among monogenic childhood‐onset hyperkinetic movement disorders, structural neuroimaging is frequently normal or shows only subtle changes (Pérez‐Dueñas et al. 2022). Where a neuroanatomical or molecular locus is defined, it typically involves the striatum or its dopaminergic input. Cortical involvement is uncommon, reported mainly in FOXG1‐related disease. In our cohort, subcortical volume loss was most pronounced in the thalamus, and hyperkinetic movements were not specifically associated with striatal or other subcortical volumes.

4.1. Limitations

This study has several important limitations. The small sample size (n = 11) limits statistical power and generalizability, though it reflects the ultrarare nature of NGLY1 Deficiency with ~165 documented cases worldwide. Data acquisition challenges introduced potential variability. Participants were each scanned at a different site using varying scanner models and protocols, with some undergoing lower resolution clinical‐quality scans rather than higher resolution research‐quality scans. This heterogeneity in scan resolution limited our ability to correlate neuroimaging findings with clinical phenotypes. Heterogeneity and small sample size additionally limited our ability to examine more detailed clinical and severity correlates. Age‐related methodological constraints further complicated analyses. We employed different processing pipelines for younger (< 3 years old) and older participants (> 3 years old) due to developmental differences in brain morphology and tissue contrast. Our normative reference models were not continuous across age groups, limiting longitudinal interpretations of Z scores. The normative model for younger participants was based on a smaller reference control sample (n = 417) and exhibited poor segmentation quality in temporal lobes for 6‐ and 7‐month‐old infants, leading us to exclude all normative data for the temporal lobes for the youngest participants. Finally, cerebellar analyses were not performed due to poor image quality in this region across multiple participants. Given pronounced motor impairments in NGLY1 deficiency, the cerebellum's role in motor coordination, and prior studies showing loss of Purkinje cells in NGLY1 deficiency (Yang, Dunn, et al. 2025), this represents an important gap that should be addressed in future studies with optimized cerebellar imaging protocols. Despite these limitations, this study provides valuable insights into the neuroanatomical correlates of this ultrarare disorder. Future multisite collaborative studies with standardized protocols, larger sample sizes and comprehensive phenotyping will be essential to validate and extend these findings.

This comprehensive neuroimaging characterization of NGLY1 deficiency reveals pervasive alterations in brain development that extend beyond the relatively nonspecific atrophy previously reported in case studies. These findings establish a neuroimaging phenotype characterized by consistent subcortical volume reductions across age groups, particularly affecting the thalamus, caudate and putamen, with age‐dependent patterns of cortical alterations including widespread surface area reductions and altered gyrification in younger patients. The correlations between cortical and subcortical morphological markers suggest that quantitative neuroimaging may serve as a valuable biomarker for disease progression and therapeutic monitoring in future clinical trials. Whereas our sample size was limited by the ultrarare nature of this disorder, and interpretation is complicated by potential survivor bias, these findings provide critical insights into the neurodevelopmental trajectory of NGLY1 deficiency. The distinct patterns of brain morphological changes identified may reflect the disrupted cellular processes inherent to this disorder, including impaired protein degradation, altered neuronal migration, impaired neuronal differentiation or survival, and abnormal synaptic development. As potential therapies for NGLY1 deficiency advance towards clinical trials, the neuroimaging metrics established here may prove essential for objectively measuring treatment response and understanding the relationship between brain structure and functional outcomes in this devastating paediatric disorder.

Funding

This work was supported by the Grace Science Foundation.

Conflicts of Interest

Emily L. Dennis and Kevin J. Lee are consultants for the Grace Science Foundation. William F. Mueller, Lei Zhu, Selina Dwight, Jennifer W. Cook and Gracie Morrison are or were employees of Grace Science, LLC.

Supporting information

Data S1: Supporting Information.

Table S1: Scan parameters.

Figure S1: Representative T1‐weighted MRI scans for selected patients. Axial, sagittal, and coronal views are shown for 4 patients. Left in image is right in brain.

Table S2: Data used for the infant normative database by participant age (in months) and sex. Data from 417 infants were pulled from the NIMH Data Archive (NDA). These include only participants designated as “typically developing” or “healthy control” by the contributing study and only those with scans that produced high quality FreeSurfer segmentations.

Figure S2: Normative trajectories of subcortical volume for ages 6–27 months. Normative trajectories created from 417 infants, separated by sex. Mean, 1 standard deviation (SD), and 2SD curves are shown for males (blue) and females (red).

Figure S3: Normative trajectories of cortical thickness (cingulate, insula, and occipital lobes) for ages 6–27 months. Normative trajectories created from 417 infants, separated by sex. Mean, 1 standard deviation (SD), and 2SD curves are shown for males (blue) and females (red).

Figure S4: Normative trajectories of cortical thickness (frontal lobe) for ages 6–27 months. Normative trajectories created from 417 infants, separated by sex. Mean, 1 standard deviation (SD), and 2SD curves are shown for males (blue) and females (red).

Figure S5: Normative trajectories of cortical thickness (parietal and temporal lobes) for ages 6–27 months. Normative trajectories created from 417 infants, separated by sex. Mean, 1 standard deviation (SD), and 2SD curves are shown for males (blue) and females (red).

Figure S6: Normative trajectories of cortical surface area (cingulate, insula, and occipital lobes) for ages 6–27 months. Normative trajectories created from 417 infants, separated by sex. Mean, 1 standard deviation (SD), and 2SD curves are shown for males (blue) and females (red).

Figure S7: Normative trajectories of cortical surface area (frontal lobe) for ages 6–27 months. Normative trajectories created from 417 infants, separated by sex. Mean, 1 standard deviation (SD), and 2SD curves are shown for males (blue) and females (red).

Figure S8: Normative trajectories of cortical surface area (parietal and temporal lobes) for ages 6–27 months. Normative trajectories created from 417 infants, separated by sex. Mean, 1 standard deviation (SD), and 2SD curves are shown for males (blue) and females (red).

Figure S9: Visualization of the cortical morphometry metrics examined in this paper. The pial boundary for an individual gyrus is outlined in the coronal pop out in white, with the boundary between gray and white matter outlined in black. The distance between these, shown in the red lines, is the cortical thickness. The surface area is the area of the gray matter surface, as shown in blue. Volume is the gray matter volume of a region, as shown in green. Curvature is local folding of the cortex and is defined by the radius of the sphere that fits the curve. Mean curvature is the average of the first two principal components of curvature for a region, which are orthogonal to each other.

Figure S10: Subcortical volume Z‐scores in older patients (real only and combined). Boxplots display Z‐scores for each subcortical structure (top: only native resolution scans, bottom: native and SynthSR enhanced scans) relative to age‐ and sex‐matched normative data. Box elements represent median, interquartile range (IQR), and whiskers extend to minimum and maximum excluding outliers; points beyond whiskers are outliers. Symbols indicate significant deviations from normative means: ^# p < 0.05 after multiple comparisons correction (Li & Ji); *p < 0.05 uncorrected.

Figure S11: Cortical thickness Z‐scores in older patients (real only and combined). Boxplots display Z‐scores for each cortical region (top: only native resolution scans, bottom: native and SynthSR enhanced scans) relative to age‐ and sex‐matched normative data. Box elements represent median, interquartile range (IQR), and whiskers extend to minimum and maximum excluding outliers; points beyond whiskers are outliers. Symbols indicate significant deviations from normative means: ^# p < 0.05 after multiple comparisons correction (Li & Ji); *p < 0.05 uncorrected.

Figure S12: Cortical surface area Z‐scores in older patients (real only and combined). Boxplots display Z‐scores for each cortical region (top: only native resolution scans, bottom: native and SynthSR enhanced scans) relative to age‐ and sex‐matched normative data. Box elements represent median, interquartile range (IQR), and whiskers extend to minimum and maximum excluding outliers; points beyond whiskers are outliers. Symbols indicate significant deviations from normative means: ^# p < 0.05 after multiple comparisons correction (Li & Ji); *p < 0.05 uncorrected.

Figure S13: Average Z‐scores for only native resolution (left) and the full sample (right) of older patients. Regions where the Z‐score differed from zero (p < 0.05, uncorrected). Marginally significant regions (0.0042 < p < 0.05) are included to display trends. Differences in cortical surface area (top), cortical thickness (middle), and subcortical volume are shown for the older patients only, with native resolution only scans in the left panel and native resolution (n = 6) plus enhanced clinical scans (n = 4) in the right panel. Color corresponds to the average Z‐score across the group for a given region, with negative Z‐scores in blue and positive Z‐scores in red. Left in image is left in brain.

Table S3: Complete clinical associations in the younger patient group. Among the regions where the mean Z‐score differed from 0 with p < 0.10, we examined associations with 10 clinical measures (presence or absence). For each pair, the number of patients included, Spearman ρ (rho), raw p, adjusted p using the Li & Ji method, and the effective number of tests calculated by the Li & Ji method.

Table S4: Complete clinical associations in the older patient group. Among the regions where the mean Z‐score differed from 0 with p < 0.10, we examined associations with 10 clinical measures (presence or absence). For each pair, the number of patients included, Spearman ρ (rho), raw p, adjusted p using the Li & Ji method, and the effective number of tests calculated by the Li & Ji method.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.