L-serine treatment in patients with GRIN-related encephalopathy: a phase 2A, non-randomized study

Abstract

GRIN-related disorders are rare developmental encephalopathies with variable manifestations and limited therapeutic options. Here, we present the first non-randomized, open-label, single-arm trial (NCT04646447) designed to evaluate the tolerability and efficacy of L-serine in children with GRIN genetic variants leading to loss-of-function.

In this phase 2A trial, patients aged 2–18 years with GRIN loss-of-function pathogenic variants received L-serine for 52 weeks. Primary end points included safety and efficacy by measuring changes in the Vineland Adaptive Behavior Scales, Bayley Scales, age-appropriate Wechsler Scales, Gross Motor Function-88, Sleep Disturbance Scale for Children, Pediatric Quality of Life Inventory, Child Behavior Checklist and the Caregiver-Teacher Report Form following 12 months of treatment. Secondary outcomes included seizure frequency and intensity reduction and EEG improvement. Assessments were performed 3 months and 1 day before starting treatment and 1, 3, 6 and 12 months after beginning the supplement.

Twenty-four participants were enrolled (13 males/11 females, mean age 9.8 years, SD 4.8), 23 of whom completed the study. Patients had GRIN2B, GRIN1 and GRIN2A variants (12, 6 and 5 cases, respectively). Their clinical phenotypes showed 91% had intellectual disability (61% severe), 83% had behavioural problems, 78% had movement disorders and 58% had epilepsy. Based on the Vineland Adaptive Behavior Composite standard scores, nine children were classified as mildly impaired (cut-off score > 55), whereas 14 were assigned to the clinically severe group.

An improvement was detected in the Daily Living Skills domain (P = 0035) from the Vineland Scales within the mild group. Expressive (P = 0.005), Personal (P = 0.003), Community (P = 0.009), Interpersonal (P = 0.005) and Fine Motor (P = 0.031) subdomains improved for the whole cohort, although improvement was mostly found in the mild group. The Growth Scale Values in the Cognitive subdomain of the Bayley-III Scale showed a significant improvement in the severe group (P = 0.016), with a mean increase of 21.6 points.

L-serine treatment was associated with significant improvement in the median Gross Motor Function-88 total score (P = 0.002) and the mean Pediatric Quality of Life total score (P = 0.00068), regardless of severity. L-serine normalized the EEG pattern in five children and the frequency of seizures in one clinically affected child. One patient discontinued treatment due to irritability and insomnia.

The trial provides evidence that L-serine is a safe treatment for children with GRIN loss-of-function variants, having the potential to improve adaptive behaviour, motor function and quality of life, with a better response to the treatment in mild phenotypes.

developmental and epileptic encephalopathy, neurodevelopmental disorders, N-methyl-D-aspartate receptor, L-serine, precision medicine

Introduction

Neurodevelopmental disorders (NDDs) result from a disrupted maturation of the nervous system leading to aberrant brain function.¹ Several genes regulating the glutamatergic synapse have been described as important causes of NDDs.²De novo pathogenic variants of GRIN genes have been defined as the genetic determinants of GRIN-related disorders (GRDs),^3-5 a group of rare paediatric encephalopathies with an estimated incidence ranging from 3.2 to 5.45 (depending on the GRIN gene) per 100 000 births in the US and over 500 reported cases worldwide.⁶GRIN genes (GRIN1, GRIN2A-D, GRIN3A-B) encode for the GluN subunits of the N-methyl-D-aspartate receptor (NMDAR) (GluN1, GluN2A-D and GluN3A-B, respectively). NMDARs are heterotetrametric channels resulting from the oligomerization of two obligatory GluN1 subunits and a combination of two additional subunits (GluN2A-D and GluN3A-B), whose distribution is finely regulated in a spatio-temporal manner.⁷ To date, 469 genetic variants of GRIN1, GRIN2A, GRIN2B and GRIN2D have been associated with GRD (GRIN variants database, May 2023; https://alf06.uab.es/grindb/home). Clinically, this neurodevelopmental condition is characterized by a spectrum of neurological and systemic alterations, including intellectual disability (ID), epilepsy, language impairment, autism spectrum disorder (ASD), movement and sleep disorders and gastrointestinal disturbances.^8-11

Several functional studies have been conducted to characterize the direct impact of de novo GRIN variant gene products on NMDAR activity. Thus, similar to other channelopathies, GRIN variants are currently classified into either gain- or loss-of-function (GoF/LoF), although a subset of complex GRIN variants exist.¹² The divergent consequences of GRIN LoF versus GoF variants necessitates precision medication strategies for normalization of NMDAR activity in GRD.^13-17 Theoretically and previously shown in proof-of-concept studies,^15,18 in the context of GRIN LoF variants, this restoration could be achieved through the supplementation of the NMDAR co-agonist L-serine or by the administration of NMDAR positive allosteric modulators. L-serine is a nutraceutical product already used in paediatric care and, in the context of GRD, it is hypothesized that its racemization into D-serine increases NMDAR activity,^18-21 thereby potentiating hypofunctional NMDARs, which in turn might partially rescue hypoglutamatergic function. We first reported the benefit of this strategy on motor and cognitive performance in an n-of-1 study with L-serine supplementation of a GRIN2B LoF individual.¹³ Following this proof-of-concept work, a couple of studies have been carried out in a small number of patients and following retrospective heterogeneous protocols of evaluation.^22,23 However, a prospective study within the context of a clinical trial and systematic analysis of L-serine supplementation schemes and outcome parameters has not been reported so far.

Here, we report the first phase 2A clinical trial, evaluating the tolerability and efficacy of L-serine in 24 patients with GRDs. More specifically, we aimed to determine whether L-serine: (i) is safe and well-tolerated; (ii) improves adaptive and cognitive function; (iii) improves motor function; (iv) improves quality of life; (v) improves EEG abnormalities; and (vi) reduces seizures. Additionally, the findings from this trial have been enlarged from harmonized data of nine additional patients following a similar protocol and referred to our worldwide clinician’s network.

Materials and methods

We conducted a phase 2A, multicentre, non-randomized, open-label clinical trial with L-serine for children with GRD (ClinicalTrials.gov Identifier NCT04646447). It was conducted at 11 sites in Spain (Supplementary Table 1). Hospital Sant Joan de Déu in Barcelona was the coordinating centre and in charge of study oversight. The study was conducted in accordance with the International Conference on Harmonization Guidelines for Good Clinical Practice and the Declaration of Helsinki, along with applicable national and local regulatory requirements. The protocol was approved by independent ethics committees or institutional review boards at each participating site, and written informed consent was provided for all patients by their guardians or legal representatives before screening or at the screening visit.

Patients

Recruitment was performed via identification of eligible patients by the principal investigators. Eligible patients were aged 2 to 18 years and harboured a pathogenic or likely pathogenic GRIN LoF variant, according to the guidelines of the American College of Medical Genetics and Genomics.²⁴ Prior to the inclusion of the patients, the functional stratification of likely LoF GRIN variants was performed. GRIN variants were annotated as LoF based on the presence of a reduction in surface expression and/or reduced NMDAR-mediated currents resulting from diminished ion channel conductance, open probability, co-agonist affinity and desensitization and/or deactivation rates.²⁵ Additional inclusion criteria required that patients and caregivers were willing and able (in the investigator’s opinion) to comply with all trial requirements (including the completion of all caregiver assessments by the same caregiver throughout the trial). Exclusion criteria comprised exposure to L-serine within 30 days prior to enrolment.

Functional stratification of GRIN variants

Patients carrying variants that produced a truncating or frameshift variant were classified as LoF variants, as these variants produce a haploinsufficiency.²⁶ Splice site variants and translocation that results in a loss of exons were also classified as LoF variants. For missense variants, the variants were classified according to previously described functional experiments developed by our laboratory.^26-28GRIN variants devoid of functional annotations (GluN2B-A549V, -T514A, -S555N, -V732E; GluN2A-T531M), were functionally stratified following a methodology previously reported.^26-28 Briefly, the GRIN variants were generated by oligonucleotide-directed mutagenesis of GRIN subunit-expressing plasmids and verified by Sanger sequencing. Cell-based assays were conducted in COS-7 and HEK-293T cell lines transiently transfected with wild-type or disease-associated constructs of the GluN subunits, for immunofluorescence and electrophysiological studies, respectively, as previously reported.²⁸

GRIN variants stratification was completed by in silico analysis of potential structural alterations of mutant NMDA receptors. Briefly, the structural model for the triheteromeric NMDA receptor (GluN1) 2-GluN2A-GluN2B was constructed based on 4PE5²⁹ and 5UOW³⁰ structures. Modeller 9.20³¹ was used to model the non-defined structural regions of the receptor and Scwrl4³² to position the non-determined side-chains of the NMDAR. The initial model was energy-minimized using GROMACS 5.³³ Variant GluN2B Ala734Val was introduced using PyMOL.³⁴

Procedures

The study duration per participant was 15 months in total. The first phase of the trial included a 3-month prospective observation period for collecting baseline data without treatment and a 12-month open-label treatment period. Patients who participated in the study underwent assessments at the beginning of the observation period (V0) and at 3 (V1-baseline visit), 4 (V2), 6 (V3), 9 (V4) and 15 months (V5).

L-serine (250 mg/kg/day) was orally administered in three divided doses during the first 2 weeks of the open-label period, and the dose was titrated at 3 weeks to the target dose of 500 mg/kg/day, with a maximum dose of 30 g/day for patients weighing ≥ 60 kg, which was the maintenance dose.

The proposed dosage was based on previous reports of its use for inborn errors of serine metabolism³⁵ and the hereditary sensory and autonomic neuropathy type 1 clinical trial,³⁶ as well as on previous observations showing that higher doses can cause reversible side effects, including nausea, vomiting, nystagmus and myoclonus.^37-39

The active drug was manufactured by Nutricia and presented in a powdered form of 100 g of the amino acid L-serine. Participants were instructed to return all empty and unused study medication containers at each visit. Drug compliance was assessed by tracking the number of unused study medication bottles and analysing the plasma L-serine levels.

Study objectives, clinical assessment and follow-up protocol

Primary outcomes

Safety and tolerability

Safety and tolerability were assessed based on adverse events and serious adverse events, together with changes in physical examination findings, clinically significant changes in laboratory parameters and the emergence of any new seizure type, if applicable. Blood samples were collected at the study sites for safety monitoring (including haematology, serum chemistry and coagulation). To monitor plasma amino acids, an ultra-performance liquid chromatography–tandem mass spectrometry procedure was applied using a Waters Xevo TQD triple-quadrupole mass spectrometer with positive electrospray ionization in multiple reaction monitoring mode, as previously reported.⁴⁰ Paediatric age-related reference values in plasma were established. Laboratory data were summarized and assessed at baseline and at 9 and 15 months.

Efficacy

Primary efficacy outcomes, which were evaluated after 12 months of L-serine supplementation and compared with baseline scores, included changes in functional adaptive behaviour using the Vineland Adaptive Behavior Scale–second edition (VABS-II), changes in developmental functioning assessed with the Bayley Scales of Infant Development–third edition (BSID-III) or age-appropriate Wechsler Scales [Wechsler Preschool and Primary Scale of Intelligence–fourth edition (WPPSI-IV) or Wechsler Intelligence Scale for Children–fifth edition (WISC-V)], scores on the Gross Motor Function Measure-88 (GMFM-88), scores on the Pediatric Quality of Life Inventory version-4.0 (PedsQL-v4.0), scores on the Sleep Disturbance Scale for Children (SDSC) and changes in behavioural, social and emotional symptoms measured by the Child Behavior Checklist (CBCL) and Caregiver-Teacher Report Form (C-TRF) from the Achenbach System of Empirically Based Assessment (ASEBA).

VABS-II assessments were administered to all children and developmental age equivalents were calculated. Based on the estimate of the developmental level as well as chronological age, the most appropriate measure of cognitive functioning was selected: BSID-III for children aged 42 months or younger and those with age equivalents ≤3 years and 6 months; WPPSI-IV for children with an estimated developmental level from ages 3 years and 6 months to 6 years; and WISC-V consistently for patients aged 6 to 16 years and 11 months during all subsequent study visits.

A detailed list of all tests and schedules are provided in Supplementary Fig. 1, Supplementary Table 2 and the Supplementary material ‘Methods’ section.

We included severity levels and age as factors in the analysis for the time evolution of the primary outcomes to control the effect that these variables may have on the efficacy of the treatment. VABS-II Adaptive Behavior Composite (ABC) standard score (SS) was used to establish the functional baseline level of children, with higher scores indicating better performance.⁴¹ For the purpose of the study, children with total scores of 55 or higher were classified in the mild impairment level group, whereas those with a total score below 55 were categorized in the severe group.

Neuropsychological and motor assessment was performed by a board-certified neuropsychologist and physiotherapist.

Secondary outcomes

Key secondary efficacy end points were to assess the changes from baseline to the end of treatment in frequency and severity of epileptic seizures and in video-EEG seizure burden and electrical pattern in all participants. Seizure frequency-related outcomes were assessed using daily seizure frequency and type entries made in a manual diary by the patient’s caregiver during the baseline and follow-up visits. The seizure classification was based on the International League Against Epilepsy (ILAE) guidelines.⁴² The video-EEG studies were 1-h wake/sleep studies.

Statistical analysis

Frequency tables were generated for the categorical variables. Descriptive statistics were calculated for numerical variables. Normality was assessed in numerical variables with quantile–quantile plots. In cases with two measures of a categorical variable, McNemar’s test was used to study the significance of the changes between both measures. For numerical variables with normal distribution, repeated measures ANOVA was used to study changes between the baseline measure and the last measure, including the severity as a between-cases factor and the age as a covariate. In case of a significant interaction between time and severity, post hoc analysis was performed with paired samples t-test. For numerical variables with no normal distribution, Wilcoxon’s signed rank test was used to study differences between the baseline measure and the last follow-up measure. Differences between severity groups were studied using the Wilcoxon’s signed rank test for both groups separately.

A sample size of 20 cases allowed us to detect effect sizes of 0.66 in bilateral tests to compare means with paired data, with a significance level of 0.05 and a statistical power of 80%. Finally, we chose a target enrolment of 24 patients.

Statistical analyses were performed using R version 4.3.0 with a P-value <0.05 considered to indicate statistical significance. P-values obtained are reported without adjustment for multiple comparisons given the limited power of the trial.

Patients assessed outside of the clinical trial

In addition to the clinical trial, we collected information from nine patients harbouring confirmed LoF GRIN variants and treated with L-serine, after approval from the respective ethics committees and informed consents from patients’ legal guardians. Data were shared by clinicians from different centres worldwide following a similar protocol, although not strictly the same in terms of tests and follow-up. Also, clinicians’ and parents/caregivers’ subjective impressions were recorded. Furthermore, the epilepsy diary and EEG study were completed by the nine participants.

Results

GRIN variants functional stratification and patient recruitment

The computational and experimental functional stratification of GRIN variants was employed to define the patient cohort. Using methods outlined in Santos-Gomez et al.,²⁶ nine GRD patients with (i) truncating/frameshift variants (six cases); (ii) chromosome translocation (one case); and (iii) splice site variants (two) were directly classified as GRIN LoF variants and recruited. A further 10 missense variants were classified as LoF, based on previous functional studies.^27,28 An additional five missense variants were identified as LoF based on experimental results obtained (Supplementary Fig. 2 and Supplementary Table 3).

Twenty-four patients were assessed and enrolled in the study in total. One patient withdrew from the trial 3 weeks after initiating medication due to side effects. Twenty-three patients were included in the final analysis. Recruitment was from July 2020 to May 2021, with the last patient’s follow-up concluded in December 2022.

Patient cohort

Demographics, genetics and major clinical symptoms of the 23 recruited patients included in the final analysis are detailed in Table 1, Supplementary Tables 4 and 5 and the Supplementary material ‘clinical information’ section. The participants’ mean age [± standard deviation (SD)] at recruitment was 10 ± 4.9 years (range 4–18 years) and 57% were male. Genetic variants comprised six GRIN1 missense variants, five GRIN2A (one missense, two splice site, one frameshift and one truncation) variants and 12 GRIN2B (eight missense, two truncations, one frameshift and one chromosome translocation) variants.

Table 1

Baseline demographics and disease characteristics

Characteristic	Total sample n = 23	Subtype variant
		GRIN1 (n = 6; 26%)	GRIN2A (n = 5; 21.8%)	GRIN2B (n = 12; 52.1%)
Mean age at enrolment, years (SD)	10 (4.9)	9.4 (5.6)	14 (3.2)	8.6 (4.5)
Mean age at diagnosis, years (SD)	8.1 (5)	7.2 (5.2)	11.3 (4.6)	7.1 (4.9)
Gender, n	F 10, M 13	F 2, M 4	F 2, M 3	F 6, M 6
Inheritance: de novo, n (%)/inherited, n (%)	20 (87%)/3 (13%)	6 (100%)/0 (0%)	2 (40%)/3 (60%)	12 (100%)/0 (0%)
Clinical findings (Human Phenotype Ontology term identifier) and personal history, n (%)
Neurodevelopment
Psychomotor retardation (HP:0001263)	20 (87%)	6 (100%)	2 (40%)	12 (100%)
Head control, acquired	22 (96%)	5 (83%)	5 (100%)	12 (100%)
Sitting, acquired	19 (83%)	3 (50%)	5 (100%)	11 (92%)
Walking alone, acquired	15 (65%)	3 (50%)	5 (100%)	7 (58%)
Cognition and language
ID (HP:0001249)	22 (96%)	6 (100%)	4 (80%)	12 (100%)
Severe-Profound ID (HP:0010864, HP:0002187)a	14 (61%)	6 (100%)	1 (20%)	7 (58%)
Moderate ID (HP:0002342)	4 (17%)	0 (0%)	1 (20%)	3 (25%)
Mild ID (HP:0001256)	4 (17%)	0 (0%)	2 (40%)	2 (17%)
Expressive language deficit (HP:0002474)	22 (96%)	6 (100%)	4 (80%)	12 (100%)
Non-verbal (HP:0001344)	7 (30%)	2 (33%)	0 (0%)	5 (42%)
Severe impairment (minimal utterances)	6 (26%)	3 (50%)	1 (20%)	2 (17%)
Moderate impairment (limited output)	3 (13%)	1 (17%)	0 (0%)	2 (17%)
Mild impairment	6 (26%)	0 (0%)	3 (60%)	3 (25%)
Motor function
Hypotonia (HP:0001252)	17 (74%)	5 (83%)	1 (20%)	12 (92%)
Spasticity (HP:0001257)	8 (35%)	2 (33%)	2 (40%)	4 (30%)
Gross motor function (GMFCS), grade IV-V	8 (33%)	3 (50%)	0 (0%)	5 (38%)
Movement disorders
Abnormality of movement (HP:0100022)	18 (78%)	5 (83%)	2 (40%)	11 (92%)
Dystonia (HP:0001332)	3 (13%)	1 (17%)	0 (0%)	2 (15%)
Dyskinesia (HP:0100660)	4 (17%)	1 (17%)	1 (20%)	2 (15%)
Choreiform movements (HP:0002072)	3 (13%)	1 (17%)	0 (0%)	2 (15%)
Hyperkinetic movements (HP:0002487)	7 (30%)	3 (50%)	0 (0%)	4 (33%)
Stereotypic movements (HP:0000733)	15 (65%)	3 (50%)	1 (20%)	11 (92%)
Behaviour disorders
Behaviour abnormalities	19 (83%)	4 (67%)	5 (100%)	10(83%)
Impulsivity (HP:0100710)	10 (43%)	1 (17%)	2 (40%)	7 (58%)
Inattention (HP:0000736)	17 (73%)	4 (67%)	5 (100%)	8 (67%)
Hyperactivity (HP:0000752)	8 (35%)	2 (33%)	1 (20%)	5 (42%)
Anxiety (HP:0000739)	7 (30%)	2 (33%)	2 (40%)	3 (25%)
ASD (HP:0000729)	8 (35%)	2 (33%)	2 (40%)	4 (33%)
Aggressive behaviour (HP:0000718)	3 (13%)	1 (17%)	1 (20%)	1 (8%)
Epilepsy
Ongoing or past epilepsy (HP:0001250)	13 (58%)	4 (67%)	5 (100%)	4 (33%)
Refractory epilepsy	3 (21.4%)	1 (25%)	1 (20%)	1 (20%)
Median age of epilepsy onset, years (range)	2 (0–7)	1.6 (0,17–7)	2.5 (1,4–4)	1 (0–7)
Other manifestations
Microcephaly (HP:0000252)	4 (17%)	2 (33%)	0 (0%)	2 (15%)
Sleep disorder (HP:0002360)	11 (48%)	3 (50%)	1 (20%)	7 (58%)
Strabismus (HP:0000486)	8 (35%)	3 (50%)	0 (0%)	5 (42%)
Oromotor dysfunction (HP:0011968)	7 (30%)	4 (67%)	0 (0%)	3 (23%)
Growth delay (HP:0001510)	5 (22%)	3 (50%)	0 (0%)	2 (15%)

ASD = autism spectrum disorder; AEDs = antiepileptic drugs; F = female; HP = human phenotype; ID = intellectual disability; M = male; SD = standard deviation.

^aChildren with Adaptive Behavior Composite standard scores below 55 were considered to have severe-profound intellectual disability.

Ninety-six per cent of individuals in the study displayed intellectual disability, ranging from severe-profound (ABC SS below 55; n = 14; 61%), moderate (n = 4; 17%) to mild (n = 4; 17%). In terms of psychomotor milestones, acquisition was delayed in 87% of the patients, with eight children (35%) being non-ambulant and 30% non-verbal. Movement disorders were present in 78% of the patients, with stereotypies being the most frequent feature, followed by hyperkinetic movements, dyskinesia, dystonia, chorea and oculogyric crisis. Heterogeneous behavioural traits were also displayed in 19 individuals (83%), including inattention (73%), impulsivity (43%), hyperactivity and autistic traits (35%, respectively), anxiety (30%), self-injurious (17%) and aggressive (13%) behaviours, as well as sleep disorder (48%). The study cohort also showed a history of epilepsy in 13 (58%) individuals in a GRIN variant-dependent manner, including four (67%) patients in the GRIN1 group, five (100%) patients in the GRIN2A group and four (33%) patients in the GRIN2B group. Ten patients had seizures controlled (no seizures in the preceding 12 months) at the time of trial initiation and three displayed drug-resistant epilepsy. EEG information was available in 20 individuals, including 16 individuals (80%) displaying epileptiform discharges (Supplementary material ‘clinical information’ section).

Primary outcomes of L-serine treatment

Safety evaluation and L-serine tolerance

All 24 patients were evaluated for safety and tolerability. Ninety-six per cent of participants completed the study, and no serious adverse events were recorded. Patient 24 (GRIN2B; p.Arg693Gly), severely affected at baseline, discontinued the treatment after 3 weeks of treatment at a dose of 500 mg/kg/day (30 g/day) due to worsening of behaviour characterized by irritability, self-aggression and insomnia as well as episodes of hyperventilation.

During the study, Patient 22 (GRIN2B; p.Asp732Val), without previously reported epileptic episodes, suffered a tonic-clonic seizure with abnormalities in EEG (normal at baseline). Nevertheless, L-serine supplementation was not withdrawn, since the patient showed an improvement in the different neuropsychological assessments. Levetiracetam was introduced, and the patient was seizure-free at the end of the trial. Patient 18 (GRIN2B; p.Gly689Ser) showed dyspepsia that was limited after transient L-serine dose reduction to 250 mg/kg/day for 2 weeks. Analysis of vital signs, physical examination findings and clinical laboratory measurements did not reveal adverse effects of L-serine.

Plasma L-serine

Plasma fasting L-serine levels were at a median (interquartile range, IQR) of 148.5 (53.2) μmol/l at baseline evaluation and 209 (96) μmol/l after 12 months of treatment (normal range: 92–197 μmol/l).

Efficacy: neuropsychological evaluation

Vineland Adaptive Behavior Scale–Second Edition

A total of 23 patients completed the five time point VABS-II evaluation, with a mean (±SD) ABC SS of 51 (±18.5) at baseline. Fourteen children were classified within the mild group and nine in the severe. Data for all VABS subdomains raw scores, domains and composite SS are reported in Table 2, Figs 1 and 2 and Supplementary Table 6.

Figure 1

Changes in functional adaptive behaviour as measured by the Vineland Adaptive Behavior Scale–second edition (VABS-II); domains and composite. A box plot representation of domains (Communication, Daily Living Skills and Socialization) and Adaptive Behavior Composite in standard scores (SS) at baseline (pale blue) and 12 months treatment (orange). For each analysis, three populations are presented: the whole study population (n = 23), the mild patients’ group (n = 9) and severe patients’ group (n = 14). A higher SS is indicative of improvement. The horizontal line in the box plot represents the median, and ‘x’ represents the mean. *P < 0.05, **P < 0.01, ***P < 0005.

Figure 2

Changes in functional adaptive behaviour as measured by the Vineland Adaptive Behavior Scale–second edition (VABS-II); subdomains. A box plot representation of subdomains in raw scores at baseline (pale blue) and 12 months treatment (orange) for (A) all patients, (B) the mild group and (C) the severe group. A higher raw score is indicative of improvement. The horizontal line in the box plot represents the median, and ‘x’ represents the mean.*P < 0.05, **P < 0.01, ***P < 0.005.

Table 2

Primary outcome measures

	Whole sample (n = 23)							Mild group (n = 9)					Severe group (n = 14)
	Baseline		12-month treatment					Baseline		12-month treatment			Baseline		12-month treatment
	Mean-Median	SD-IQR	Mean-Median	SD-IQR	P-valuec	P-value severity interactiond	P-value age interactione	Mean-Median	SD-IQR	Mean-Median	SD-IQR	P-valuef	Mean-Median	SD-IQR	Mean-Median	SD-IQR	P-valuef
VABS-II
Total score, SS (mean = 100 ± 15)
ABC	51	18.5	52.3	22.9	0.294	0.06	0.088	67.8	14.7	74.6	19.16	0.103	40.1	11.2	37.9	9.6	0.188
Domains, SS (mean = 100 ± 15)
Communication	51.4	18.3	52.1	19.8	0.460	0.301	0.245	68.4	14.2	72.1	15.51	0.288	40.4	10.4	39.4	8	0.496
Daily living skills	51.5	19.5	52	24	0.039	0.001	0.005	67.9	20.2	74.3	23.23	0.035	40.9	9.2	37.6	8.5	0.006
Socialization	57	18	58.3	22.5	0.220	0.142	0.079	72.4	15.7	79.9	18.11	0.193	47.1	11.2	44.4	11.3	0.209
Subdomains, RS
Receptive	17.8	11.2	19.3	12.2	0.091	0.118	Un	29	9	32.1	7,7	0.176	10.9	5.3	11	5.3	0.836
Expressiveb	20	61	25	68.5	0.005	NA	Un	89	32	93	29	0.039	15	5.8	15	6.5	0.075
Writtenb	0	16	0	16.5	0.05	NA	Un	23	15	25	18	0.089	0	0	0	0	1
Personalb	18	34.5	21	52.5	0.003	NA	Un	55	47	68	18	0.021	5.5	6.8	7	12.5	0.128
Domesticb	3	16	3	17	0.166	NA	Un	19	12	23	20	0.059	0	1.8	0	2	0.410
Communityb	2	22.5	2	27	0.009	NA	Un	25	30	40	41	0.028	0	1.7	0.5	2	0.071
Interpersonal relationships	31.1	17.5	34.6	22.1	0.005	0.004	Un	48.8	14.3	57.9	16.2	0.018	19.7	6	19.6	6.7	0.905
Copingb	6	19	6	33.5	0.307	NA	Un	25	26	50	27	0.123	3.5	4.8	1	5.5	0.481
Play	20	18.5	21.4	20.2	0.085	0.074	Un	39.3	14.7	43.1	14.7	0.123	7.5	4.3	7.4	4.7	0.673
Fine motor	27.4	22.6	29.7	21.9	0.031	0.87	Un	49.6	16.7	52.1	14.2	0.408	13.1	11.7	15.6	11.3	0.152
Gross motorb	46	21	55	60	0.196	NA	Un	67	15	72	14	0.183	10	39	10	33	0.751
GMFM-88
Total score (max. 100, %)b	76.3	74.9	78	74	0.002	NA	Un	98.9	3.4	100	1.1	0.043	27.7	44.5	33.3	54.1	0.016
PedsQLg
Total score (max. 100, %)	52.5	19.2	58.7	22.3	0.0007	0.026	Un	70.6	16.8	82.1	12.4	0.03	42	10.9	45	13.3	0.041

Data were tested for normal distribution with the Kolmogorov–Smirnov test and reported as mean ± standard deviation in the case of normally distributed data or median and interquartile range (IQR) otherwise. ABC = Adaptive Behavior Composite score; GMFM88 = Gross Motor Function Measure-88; NA = not applicable; RS = raw scores; SS = standard scores; Un = unassessed; PedsQL = Pediatric Quality of Life Inventory version-4.0; VABS-II = Vineland Adaptive Behavior Scale–Second Edition.

^aVariables with normal distribution.

^bVariables with non-normal distribution.

^cFor variables that exhibited normal distribution, P-value for the time factor in a repeated measures ANOVA with severity as the between-subjects factor; for non-normal variables, P-value of the Wilcoxon signed rank test comparing baseline and 12-month treatment measures.

^dP-value refers to the interaction between time and severity group for variables with normal distribution where the repeated measures ANOVA was applied.

^eP-value refers to the interaction between age and severity group for variables with normal distribution where the repeated measures ANOVA was applied (assessed in domains and composite scores on the VABS-II).

^fData were compared using the paired t-test for normally distributed data and the Wilcoxon signed rank test in case of non-normal distribution.

^gIn total n = 19 (7 mild and 12 severe).​​​​​​

We did not detect a significant effect of time for ABC SS values but an almost significant interaction between severity and time was detected [F(1,20) = 3.974, P = 0.06, η² = 0.017]. In the mild group, values improved (67.8 ± 14.7 at V1 versus 74.6 ± 19.16 at V5, P = 0.103), whereas, in the severe group, they worsened (40.1 ± 11.2 at V1 versus 37.9 ± 9.6 at V5, P = 0.188), suggesting that the observed interaction was due to a better time evolution in the mild group.

For the Daily Living Skills (DLS) SS, we detected a significant time effect [F(1,20) = 4.880, P = 0.039, η² = 0.006] and a significant interaction between time and severity [F(1,20) = 13.657, P = 0.001, η² = 0.015]. Post hoc comparisons within the mild severity group revealed a significant increase in mean DLS SS from baseline (68 ± 20.2) to 12 months treatment (74.3 ± 23.2) (P = 0.035), while the severe group had a lower DLS score at 12 months of treatment (37.6 ± 8.48) compared to baseline (40.9 ± 9.23) (P = 0.006). The post hoc results showed that the overall change detected was due to the improvement in the mild patients.

Regarding subdomains, we did find a significant time effect on expressive raw scores (median = 20, IQR = 61 at V1 versus median = 25, IQR = 68.5 at V5, P = 0.005), personal raw scores (median = 18, IQR = 34.5 at V1 versus median = 21, IQR = 52.5 at V5; P = 0.003), community raw scores (median = 2, P75 = 21 at V1 versus median = 2, P75 = 31 at V5; P = 0.009), interpersonal relationship raw scores (31.1 ± 17.5 at V1 versus 34.6 ± 22.1 at V5; P = 0.005) and fine motor raw scores (27.4 ± 22.6 at V1 versus 29.7 ± 21.9 at V5; P = 0.031) for the whole population.

Significant interaction between severity group and time was also found for interpersonal relationship raw scores [F(1,20) = 10.70, P = 0.004, η² = 0.047]. Post hoc comparisons revealed a significant increase in values within the mild group (48.8 ± 14.3 at V1 versus 57.9 ± 16.2 at V5, P = 0.018), while no significant differences were observed in the severe group (19.7 ± 6 at V1 versus 19.6 ± 6,7 at V5; P = 0.905).

Higher scores on the expressive raw scores at 12 months (median = 93, IQR = 29) compared to baseline (median = 89, IQR = 32) were found in the mild group (P = 0.038), as well on the personal raw scores (baseline median = 55 and median at 12 months = 68, P = 0.0209). Community subdomain also showed greater raw scores at 12 months of treatment in the mild group (P = 0.028), with a median baseline raw score of 25 points and 40 points at 12 months of treatment.

A time and age interaction was detected for DLS SS [F(1,20) = 9.842, P = 0.005, η² = 0.011]. Interpretation of the results showed that the improvement in DLS SS was larger for older patients.

BSID-III, WPPSI-IV, WISC-V, CBCL and TRF scores

After obtaining the developmental age equivalents using VABS-II, there were 15 patients who completed the BSID-III (Supplementary Table 7) and eight who completed the age-specific Wechsler Scales (Supplementary Table 8). One patient had to be excluded from the BSID-III analysis because we could not obtain data at V5 due to behavioural disturbances.

On the BSID-III, the baseline mean Growth Scale Value (GSV) scores for the cognitive subscale were 418.5 (±91.93) and 439.9 (±99.96) at V5. A paired t-test revealed a significant improvement in the GSV cognitive subscale post-intervention [t(13) = 2.77, P = 0.0158], with a mean increase of 21.6 points [95% confidence interval (CI) −37.99 to −4.73]. Additionally, a trend towards improvement in the GSV receptive subscale was observed at V5 compared to V1 [t(13) = 2.13, P = 0.052] (Table 3 and Supplementary Fig. 3).

Table 3

Primary outcome measures

BSID-III (n = 14)
Domains, GSV	Baseline	12-month treatment	P-value
Cognitivea	418.5 ± 91.93	439.86 ± 99.96	0.0158
Receptive communicationa	443.3 ± 96.51	457 ± 94.94	0.0525
Expressive communicationb	470 (78.75)	431 (84)	0.906
Fine motorb	422 (122.25)	446 (119.75)	0.701
Gross motorb	466 (142.75)	507 (144.25)	0.071

Tests were performed in 14 children. Data were tested for normal distribution with the Kolmogorov–Smirnov test and reported as mean ± standard deviation in the case of normally distributed data or median and interquartile range (IQR) otherwise. BSID-III = Bayley Scales of Infant Development–third edition; GSV = Growth Scale Values.

^aVariables with normal distribution.

^bVariables with non-normal distribution.

Regarding the age-specific Wechsler Scale, three and five patients completed the WPPSI-IV and WISC-V, respectively. The mean (±SD) full-scale intelligence quotient (FSIQ) was 59.5 ± 14.2. Overall, analysis of the WPPSI-IV and WISC-V FSIQ and five primary index scores in the eight patients detected no significant changes from baseline to post-treatment in any subtest and indicated that intra-patient variability was relatively low over this time period.

The results of the school-aged versions of the CBCL (n = 10) and TRF (n = 9) of the ASEBA, revealed no statistically relevant changes in the different subtests.

GMFM-88

Twenty-three patients completed GMFM-88 evaluation (Supplementary Table 9). Overall median GMFM-88 total scores increased from 76.3% (IQR = 21.9–96.7) at baseline to 78% (IQR = 25.5–99.4) at 12 months of treatment (P = 0.002). The mild group showed a median GMFM-88 score of 98.9% (IQR = 3.4) at V1, which improved to 100% at V5 (IQR = 1.1) (P = 0.03) and severe group had a median GMFM-88 score of 27.7% at baseline (IQR = 44.5), which increased to 33.3% (IQR = 54.1) at the end of the treatment (P = 0.041), (Fig. 3 and Table 2).

Figure 3

Changes in GMFM-88 and PedsQL total scores. (A) Box plot representation of GMFM-88 total score (%) in the entire sample, mild group and severe group at baseline (pale blue) and 12 months treatment (orange). (B) Box plot representation of PedsQL total score (%) in the entire sample, mild group and severe group at baseline (pale blue) and 12 months treatment (orange). A higher total score is indicative of improvement. The horizontal line in the box plot represents the median, and ‘x’ represents the mean. *P < 0.05, **P < 0.01, ***P < 0.005. GMFM88 = Gross Motor Function Measure-88; PedsQL = Pediatric Quality of Life Inventory version-4.0.

SDSC

The number of patients completing the SDSC evaluation was 21, with a mean (±SD) total score of 38.2 (±8.4) (Supplementary Table 10). The total score of the SDSC did not show a statistically significant change in the overall sample or related to severity group. When analysed by the six categories, we did not find significant differences in the proportion of abnormal scores between pre- and post-treatment.

PedsQL

Four patients had to be excluded from the analysis because the questionnaire at V5 was not completed by the same caregiver. In the end, a total of 19 two-point time PedsQL evaluations were analysed (Supplementary Table 9). Two-way repeated measures ANOVA showed a significant interaction between time factor and severity for PedsQL cognitive functioning total score [F(1,17) = 5.90, P = 0.026, η² = 0.027]. Post hoc comparisons demonstrated higher scores at 12 months treatment for both groups. Within the mild severity group, a significant increase in mean PedsQL total score was seen from baseline (70.64 ± 16.79) to 12 months treatment (82.14 ± 12.42) [t(6) = −2.82, 95%CI −21.48 to −1.52, P = 0.03] and within the severe group from 42.02 (±10.89) to 45.01 (±13.26) [t(11) = −2.31, 95%CI −5.84 to −0.14, P = 0.041] (Fig. 3 and Table 2).

Subjective observations from patient families and caregivers

Subjective observations from parents and caregivers were also collected, reporting additional improvements, especially in terms of children’s alertness, vigilance, concentration, attention span, reaction time and non-verbal communication. According to reports from teachers, improvements were observed in the quality of writing and drawing among individuals with a higher functional baseline level (Supplementary material ‘Methods’ section and Supplementary Fig. 4).

Secondary outcomes of L-serine treatment

Epilepsy and EEG

Epileptiform potentials on EEG were present in 16 of the 20 available. Clinical seizures were observed in three individuals with two of three presenting refractory epilepsy at baseline. Five of eighteen individuals (28%) had a complete resolution of epileptic activity after initiation of L-serine treatment. While EEG remained unchanged during treatment, one individual with refractory epilepsy decreased the frequency of seizures from one seizure/week to one seizure every 2 months.

One individual experienced a first tonic-clonic seizure during the study; no others experienced a worsening of EEG patterns or seizures.

Assessment of external patients treated with L-serine not included in the clinical trial

In parallel with the execution of the clinical trial, several worldwide clinicians followed the clinical protocol defined for the current clinical trial. There were nine children aged 3 to 24 years, who were carriers of eight GRIN2B and one GRIN1 LoF variants. Seven children had severe-profound and two mild-moderate intellectual disability. Genetic, phenotypic and treatment details are provided in Supplementary Table 11 and the Supplementary material ‘clinical information’ section.

Discussion

Here, we present the first non-randomized, open-label, single-arm, 15-month trial in 24 children with GRDs, designed to evaluate tolerability and efficacy of L-serine in patients with GRDs. Although, a randomized study with a blinded control or placebo group is considered the gold standard in clinical research, the decision to conduct a non-randomized study in the context of the rare but severe nature of neurodevelopmental disease without targeted therapeutic options was justified by ethical and humanitarian considerations.

While this study did not meet all the primary end points, this trial demonstrated that L-serine was well-tolerated and resulted in significant enhancements in motor function, quality of life and select VABS-II subdomains. Although the post hoc analyses should be interpreted with great caution due to the small sample size, the subgroup analyses pointed towards a better response to L-serine in terms of adaptive and developmental functioning in the mild group defined by ABC SS of ≥55. Additionally, children below 36 months of developmental equivalent age, as evaluated with BSID-III, showed an increased GSV on the cognitive domain.

Clinical evaluation of neurodevelopment in children with developmental delay represents a significant challenge. Traditional developmental tests are often normatively configured for neurotypical children and not wholly suited to the evaluation of children with developmental delay. Moreover, the lack of detailed natural history studies makes it difficult to select clinically meaningful primary trial end points. Consequently, with the absence of a GRD-adapted assessment scale, we employed an extensive panel of tests covering the assessment of cognition, motor function and quality of life as well as behaviour, social and emotional aspects.

From the clinical point of view and based on feedback from parents, it would be plausible that the response to treatment may differ among patients. Clinical severity appeared to be a crucial factor in these observations. To enable objective comparisons within the patient cohort, it was decided to establish a functional baseline level for the children using the ABC SS (cut-off of ABC ≥55). As a result, patients were stratified into mild or severe groups, and the analysis of the different primary objectives was conducted both for the whole population and separately, facilitating a more detailed analysis of the response to L-serine supplementation. Indeed, subgroup analysis showed a different treatment response between mild and severe patients, with better treatment effects in the milder patients. The responsiveness in patients with a milder phenotype may be related to a higher residual level of NMDAR activity in this group of patients, which can be further enhanced by L-serine supplementation. Another possibility is that patients with a milder phenotype may have less severe disruption of downstream signalling pathways,^43-45 making it easier to restore normal synaptic transmission with treatment.

Benefits found with L-serine were not consistently seen in all VABS-II domains. DLS improved for patients in the mild group exhibiting gains in personal and community subdomain scores. Improvement in the performance of everyday tasks can lead to a virtuous cycle of greater independence, self-sufficiency, socialization and inclusion, reducing the need for assistance, which has a positive impact on quality of life and the burden on caregivers. The expressive language subdomain showed a significant improvement in patients with higher adaptive level. A change in this domain is important because language and speech impairment are very prevalent in GRDs, affecting almost 100% of individuals with GRIN2A-related disorders.⁴⁶ The GluN2A subunit is expressed in cortical and subcortical structures involved in speech and language.^47,48 Dysfunctional NMDA receptor activity in the basal ganglia has been linked to impaired motor speech programming and execution, leading to dysarthria and speech apraxia.⁴⁹ Therefore, interventions aimed at improving communication skills hold potential clinical benefit.

Although mild patients had more relevant improvements, severe patients obtained better results on the cognitive domain (higher GSV) of the BSID-III at 12 months of treatment (mean change of 21 GSV in 1 year). The cognitive scale estimates general cognitive functioning based on non-verbal activities involving memory, problem-solving and manipulation.⁵⁰ Since a natural history study of GRDs is not yet available, we compared the rates of gain in developmental skills with another NDD. Sahwani and collaborators⁵¹ studied a cohort of 236 children with Angelman syndrome (AS), and they found that children with AS continued to gain skills over time at a rate of approximately 1–16 GSV/year at least until 12 years of age. This suggests that even children with severe NDD continue to learn new skills over time and, as slow as progression could be, might benefit from L-serine supplementation.

Besides the cognitive benefit, motor function was also improved by L-serine treatment, according to the GMFM-88 scale, and a relevant benefit was also detected along PedsQL evaluations. In particular, Patient 1 started to autonomously use public transportation and manage money, and Patient 5 found a job and started a ‘Middle’ degree in Visual Arts with no curricular adaptations. Subjectively, patients also showed improvement in conceptual abilities (including abstract thinking, symbolic play, imitation), non-verbal communication and social domains (including skills for establishing interpersonal relationships), which may also benefit quality of life, functionality and schooling achievements. Other observed improvements included better temper control, less hyperactivity and improvement in sustained attention. Besides the improvement of these CNS-related phenotypes, patients exhibited systemic improvements, notably related to decreased meteorism and an improvement in food tolerance.

L-serine had also an impact on epileptic activity. A significant proportion of patients (28%) demonstrated normalized EEG patterns and one patient showed a reduced frequency of seizures during the 12 months of treatment. Epilepsy is caused by an imbalance between excitatory and inhibitory brain activity. Glutamate is the major neurotransmitter for excitatory neuronal signalling, while GABA has an inhibitory function. Glutamate plays a role in both pre- and post-synaptic excitatory neurotransmission resulting in cellular and network hyperactivity and underlies the formation of epileptogenesis.⁵² While L-serine is an activator of glutamatergic transmission and theoretically should behave as a potentiator of epileptogenesis, it is interesting to observe the amelioration of the EEG and the epileptic symptoms in some of our patients and reported in a previous study.²³ This could be due to the fact that glutamate also potentiates the activity of inhibitory interneurons releasing GABA, which in turn activates astrocytic GABA_B receptors and induces an increased release of Ca²⁺ within astrocytes, triggering glutamate release.⁵³ Glutamate acts on presynaptic ionotropic glutamate receptors and augments the release of more GABA from the surrounding inhibitory neurons.⁵⁴

An important aspect of GRDs is the clinical spectrum observed in patients and, concomitantly, it is also important to address a potential correlation between the genetic variant and L-serine treatment efficacy. In our cohort, there was a tendency that patients that were responders carried GRIN2A genetic variants (in all five GRIN2A cases treated) or a missense/truncating/frameshift GRIN2B genetic variant that results in a reduction of NMDAR surface density. This is in accordance with the fact that these types of variants are clinically associated with milder clinical phenotypes.^26,55 Furthermore, the findings in GRIN2A individuals along the trial showed that, beyond the efficacy of L-serine reported previously for GRIN2B-LoF treatment, GRIN2A-LoF patients also exhibited an improvement, thus expanding the series of GRIN2 variant children treated to date.²³

Besides GRIN2A and GRIN2B variants, patients carrying GRIN1 missense variants responded less to L-serine treatment, which is in accordance with the reported high clinical severity of GRIN1 carriers amongst GRD individuals.²⁷ This is probably due to the obligatory presence of GluN1 subunits in all NMDAR subtypes. Accordingly, dominant negative effects of GRIN1 missense variants should be expected in all brain regions expressing NMDARs and along all developmental stages,^43,47,56,57 including very early periods, with wide detrimental consequences.

According to mutation severity, mild patients responded better than patients who were severe. Thus, beyond the genotype-phenotype correlation (under investigation by our and other groups), our trial pointed out an inverse correlation between clinical severity and efficacy of L-serine treatment. Nevertheless, it must be stated that this observation is based on a reduced sample size of the trial’s cohort, and additional cases will provide more data to directly predict the genotype-phenotype-clinical improvement correlation. In this sense, despite the lack of standardized evaluation in the nine external patients, some improvement was observed, mainly in verbal and non-verbal communication, as well as in attention, with scarce adverse events. These findings offer additional context for the data obtained within the trial.

A key finding of this study was the minimal side-effect profile among patients with GRD with high-dose L-serine, which contributed to the high compliance rates observed in this trial. We selected a dose of 500 mg/kg/day on the basis of recommended doses for inborn errors of metabolism of L-serine synthesis deficiency.³⁵ However, GRD patients do not have L-serine synthesis deficit, and serine levels are in the normal range. Therefore, L-serine supplementation could potentially raise serine levels beyond the normal range and have a detrimental effect. However, the absence of significant adverse events in our study and previous studies (with doses up to 850 mg/kg/day)¹⁴ suggests that even higher doses could be considered, although it is not clear whether there might be an impact on other metabolic pathways such as glycolysis, energy metabolism in general and the sphingolipids pathway.^58,59 Further investigation, including metabolomics will help in delineating both the optimal and minimally effective L-serine dosages as well as determine long-term effects.

Limitations

Our study had limitations, the most important being its single arm and non-blinded design. The absence of a placebo group did not allow the effect of the treatment to be analysed statistically, rendering the results somewhat less conclusive. Another limitation was the paucity of validated outcome measures and longitudinal natural history data specific to GRDs. Our study included a heterogeneous group with different genotypes and phenotypes, so the outcome variables had to be selected based on the most common symptoms of GRDs. Lastly, GRDs are very rare diseases, with only a few hundred patients diagnosed worldwide, which limited the recruitment capacity and resulted in a small sample size. Besides the study limitations, the optimal treatment dose of L-serine may require further examination.

Conclusions

The data reported here showed that L-serine administration is safe and well tolerated in GRD patients. Moreover, we showed that L-serine improved motor function, quality of life and adaptive behaviour in a subset of these children as well as a decrease in epileptic abnormalities on EEG and seizure frequency (in one child), suggesting that this amino acid may be an effective treatment for GRDs, irrespective of the exact GRIN gene that is mutated. Our findings also suggested that the response to treatment may differ according to the severity of the patients, particularly in terms of adaptive and developmental functioning, with a better response in mild phenotypes. Our research identified these differences and provides a better understanding of the specific outcomes associated with these two groups. By doing so, we hope to contribute to the development of more effective treatment strategies for GRD individuals with varying levels of functional impairment.

Based on our observations described herein, we propose L-serine as the first rational therapy for GRDs, addressing the metabolic effect of the underlying gene defect by modulating brain metabolism using strategies that have been largely applied in classic inborn errors of metabolism. It will be important to determine the optimal treatment dosage as well as the effects of long-term use in this patient population. Therefore, future larger double-blind, placebo-controlled trials to generate level one evidence for L-serine supplementation in patients with GRDs LoF variants is needed.

Data availability

Trial protocol and all data can be requested by any qualified researchers who engage in rigorous, independent scientific research and will be provided following review and approval of a research proposal and statistical analysis plan and execution of a data sharing agreement. Data requests can be submitted at any time and data will be accessible for 12 months, with possible extensions considered.

Acknowledgements

The authors want to express their gratitude to the GRD individuals and their families for their participation and support for this study. We also thank the scientific community for providing GRD clinical and genetic annotations.

Funding

The study was financed partially by the Nutricia Metabolics Research Fund. L-serine was manufactured and provided by Nutricia. N.J.-P. and A.G.-C. were supported by FI21/00073 ‘Instituto de Salud Carlos III (ISCIII)’ and ‘Fondo Europeo de desarrollo regional (FEDER)’; X.A. and M.O. were supported by ISCIII, co-funded by European Regional Development Fund (ERDF), a way to build Europe (grants PI19/00348 and PI22/00515); A.S. was supported by a Fundación Tatiana Pérez de Guzmán El Bueno PhD fellowship.

Competing interests

A.G.-C. has received honoraria for research support and lectures from PTC Therapeutics and Immedica, honoraria for lectures from Biomarin, Immedica, Eisai, Orchard Therapeutics, and Recordati Rare Diseases Foundation and is a co-founder of the Hospital Sant Joan de Déu start-up ‘Neuroprotect Life Sciences’. The other authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.

Author notes

Natalia Juliá-Palacios, Mireia Olivella, Xavier Altafaj and Ángeles García-Cazorla contributed equally to this work.