Single-cell transcriptomics identifies neural fate disruption and glial reprogramming caused by RARS2 deficiency

Abstract

Background

Pontocerebellar hypoplasia type 6 (PCH6) is caused by biallelic pathogenic variants in RARS2, encoding mitochondrial arginyl-tRNA synthetase. Although mitochondrial dysfunction is a recognised feature, how RARS2 deficiency disrupts neural lineage development remains unclear.

Methods

We generated rars2-deficient zebrafish using the clustered regularly interspaced short palindromic repeats/CRISPR-associated protein 9 (CRISPR/Cas9) system and performed single-cell RNA sequencing (scRNA-seq) at 48 hours postfertilisation, complemented by immunofluorescence, in situ hybridisation, behavioural assays and ultrastructural analyses. Neural lineage composition, developmental trajectories, intercellular communication and transcriptional programmes were systematically examined.

Results

rars2^−/− zebrafish displayed impaired survival, locomotor deficits, early mitochondrial ultrastructural damage and marked disruption of neurogenesis. scRNA-seq revealed reduced neuronal populations and expansion of neural progenitor and glial-like cells. Key neurogenic regulators (neurod4, her6 and pou3f1) were downregulated, whereas glial and stress-associated markers (hmgb1a, fabp7a and foxp1b) were upregulated. Developmental pathways including Notch and non-canonical Wnt were attenuated while extracellular matrix (ECM), adhesion and inflammatory programmes were activated. Additional trajectory-based analyses supported dysregulated lineage progression characterised by glial programme activation and impaired maintenance of neurogenic differentiation.

Conclusion

RARS2 deficiency disrupts mitochondrial integrity and reprograms neural lineage development through coordinated suppression of neurogenic transcriptional networks and activation of glial/ECM programmes. These findings provide mechanistic insight into loss-of-function RARS2 deficiency and highlight candidate molecular pathways for future therapeutic investigation.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• RARS2 variants cause pontocerebellar hypoplasia type 6 (PCH6), but how mitochondrial translation defects alter neural lineage specification at the cellular level remains largely unknown.

WHAT THIS STUDY ADDS

• Using a rars2-deficient zebrafish model combined with single-cell transcriptomics and histological validation, we show that RARS2 loss suppresses neurogenic transcription factors, disrupts neuronal differentiation, and is associated with glial and extracellular matrix-related transcriptional reprogramming driven by mitochondrial dysfunction.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• These findings provide mechanistic insight into PCH6 pathogenesis and identify potential molecular targets within neurogenic, glial and mitochondrial pathways that could inform future therapeutic strategies.

Introduction

Neurodevelopmental disorders represent a diverse group of conditions characterised by abnormal brain formation and neuronal maturation, often leading to intellectual disability, epilepsy and motor dysfunction. Disruption of neural lineage specification and glial differentiation has been recognised as a major pathological mechanism across multiple neurodevelopmental and neurodegenerative syndromes.^{1 2} In particular, mitochondrial dysfunction has emerged as a converging factor linking impaired bioenergetics, oxidative stress and cellular fate misregulation in the developing nervous system.^{3 4} Neurons are highly energy-dependent cells, and the maintenance of mitochondrial integrity is essential for synaptogenesis, axonal transport and neurotransmission.⁵ When mitochondrial translation or metabolic regulation is disrupted, developing neural circuits become vulnerable to stress and apoptosis, resulting in irreversible structural and functional deficits.⁶

Pontocerebellar hypoplasia (PCH) represents a rare but clinically and genetically heterogeneous group of neurodevelopmental disorders characterised by hypoplasia of the cerebellum and pons, accompanied by severe developmental delay, progressive microcephaly and variable motor and cognitive impairment.^{7 8} Among the known subtypes, PCH type 6 (PCH6) is uniquely associated with mutations in the RARS2 gene, which encodes the mitochondrial arginyl-tRNA synthetase.^9–11 This enzyme is indispensable for mitochondrial translation, catalysing the aminoacylation of tRNA-Arg, and its deficiency leads to incomplete mitochondrial protein synthesis, disrupted respiratory chain function and neuronal loss.^12–14 Clinically, patients with RARS2 mutations display early-onset encephalopathy, ataxia and pontocerebellar atrophy, often accompanied by elevated lactate levels and seizures.¹⁵ Despite these clinical characterisations, the molecular and cellular mechanisms by which RARS2 deficiency causes neural dysfunction remain poorly understood.

Previous studies have indicated that loss of mitochondrial aminoacyl-tRNA synthetases can trigger compensatory stress responses and transcriptional reprogramming in neurons.^{16 17} However, the specific impact of rars2 deficiency on cell fate determination—particularly the balance between neural progenitors, neurons and glial cells—has not been systematically investigated. Traditional bulk RNA sequencing approaches lack the resolution to capture the cellular heterogeneity underlying this process. Recent advances in single-cell RNA sequencing (scRNA-seq) have provided an unprecedented ability to dissect cell-type-specific transcriptional dynamics and to map developmental trajectories within complex neural tissues.^{18 19} This technique is especially powerful when integrated with in vivo genetic models that recapitulate human disease phenotypes.

The zebrafish (Danio rerio) has emerged as a valuable vertebrate model for studying neurodevelopmental disorders because of its genetic conservation with humans, optical transparency and suitability for live imaging.^{20 21} Its rapid embryonic development and amenability to genome editing allow precise modelling of mitochondrial and metabolic disorders, making it ideal for investigating early pathophysiological changes at single-cell resolution.

In this study, we used a zebrafish model of rars2 deficiency to explore how mitochondrial dysfunction disrupts neural development in PCH6. By combining single-cell transcriptomic profiling with immunofluorescence and ultrastructural validation, we aimed to uncover the cellular and molecular mechanisms linking mitochondrial translation defects to neural fate imbalance and glial activation. Through this approach, our goal was to provide a comprehensive framework for understanding how rars2-related mitochondrial dysfunction drives aberrant neurogenesis and to identify potential molecular targets for future therapeutic intervention.

Materials and methods

Clinical evaluation and genetic analysis

A preschool-aged proband with hypotonia, infantile-onset epilepsy, global developmental delay and PCH underwent clinical evaluation, brain MRI, electroencephalogram (EEG) assessment and longitudinal follow-up. Clinical data collected for phenotype characterisation included seizure history, EEG findings, treatment history, developmental assessment and neuroimaging results. Peripheral blood DNA from the proband and parents was subjected to whole-exome sequencing (WES) using the Nano WES Human Exome V1 kit (Berry Genomics) and Illumina NovaSeq 6000 sequencing. Reads were aligned to GRCh38 using the Burrows-Wheeler Aligner (BWA); variant calling was performed using the Genome Analysis Toolkit (GATK), and variants were annotated using ANNOtate VARiation (ANNOVAR). Variants were filtered by population frequency and predicted pathogenicity and interpreted according to American College of Medical Genetics and Genomics (ACMG)/Association for Molecular Pathology (AMP) guidelines. To determine whether the two exonic RARS2 variants might also affect pre-mRNA splicing, we performed in silico splice prediction using SpliceAI, dbscSNV and MaxEntScan/GeneBe. Candidate RARS2 variants were validated by Sanger sequencing.²² The primers were designed as follows: for c.772C>T, forward 5′-GCATTCTGTTGTCCTTACTC-3′ and reverse 5′-GTGTTCATGTGTGTGACAGG-3′; for c.436C>T, forward 5′-TTGGCAGGAGGCTCATAT-3′ and reverse 5′-GGCATGACCTAGTTGATTCT-3′. Ethical approval and written informed consent were obtained.

Structural modelling and molecular analysis

Evolutionary conservation of the affected residues was assessed across species using the Basic Local Alignment Search Tool (BLAST).²³ Wild-type and mutant RARS2 protein structures were modelled using AlphaFold2, visualised in PyMOL and analysed for local conformational alterations.²⁴

Zebrafish husbandry and generation of rars2-deficient lines

AB-strain zebrafish were maintained under standard conditions (28.5°C under a 14-hour light/10-hour dark cycle). rars2 mutants were generated using the clustered regularly interspaced short palindromic repeats/CRISPR-associated protein 9 (CRISPR/Cas9) genome-editing system to target exon regions; guide RNAs (gRNAs) were designed using SnapGene. Primers flanking the gRNA target sites of the rars2 gene were designed with SnapGene software, with the following sequences: rars2-772guide-F: 5′-ctgtgctttctgtccgtccatgt-3′, rars2-772guide-R: 5′-tatcgtgcatccctatcatggcc-3′, rars2-436guide-F: 5′-acactggtggagttcaggtcag-3′ and rars2-436guide-R: 5′-acgccgcatttctgttgcacag-3′.²⁵ Founder (F0) fish were screened by Sanger sequencing and stable F1/F2 carriers were established. All animal procedures were approved by the West China Second University Hospital Ethics Committee.

Behavioural assessment

Larval locomotion at 7 days postfertilisation (dpf) was recorded in 96-well plates for 60 min using EthoVision XT (Noldus). Adult zebrafish at 4 months postfertilisation (mpf) were assessed in a 30×30 cm tank under identical recording conditions. Quantitative parameters included total distance, mean speed, acceleration, active/resting duration and spatial orientation.

Transmission electron microscopy

Brain tissues from zebrafish embryos were fixed in glutaraldehyde, post-fixed in osmium tetroxide, dehydrated, embedded in resin and sectioned at ~70 nm. Ultrathin sections were stained with uranyl acetate/lead citrate and imaged using a HITACHI HT7700 transmission electron microscope to assess mitochondrial and synaptic ultrastructure.

Immunofluorescence staining

Whole-mount zebrafish embryos and larvae at 48 hours postfertilisation (hpf) and 4 dpf were subjected to immunofluorescence single- and double-labelling. Primary antibodies included antiacetylated α-tubulin (A24738, 1:50, Abclonal), glial fibrillary acidic protein (GFAP) (A19058, 1:50, Abclonal), ELAV-like RNA-binding proteins 3 (ELAVL3; HuC/D) (A6091, 1:50, Abclonal), subunit c of mitochondrial ATP synthase (SCMAS; A20960, 1:50, Abclonal), phospho-histone H3 (PH3) (AP0002, 1:50, Abclonal) and 5-bromo-2′-deoxyuridine (BrdU) (A1482, 1:50, Abclonal). Lysotracker-Red (100 nM, Yuanye, China) and acridine orange (7 µg/mL, Yuanye, China)²⁶ were applied for lysosome and apoptosis detection, respectively. Nuclei were counterstained with 4',6-diamidino-2-phenylindole. Images were acquired using a laser scanning confocal microscope (Olympus FV3000, Japan).

In situ hybridisation of zebrafish embryos

To assess gene expression changes associated with rars2 mutations; whole-mount in situ hybridisation (ISH) was performed in zebrafish embryos. The otx2b probe (~1 kb) was cloned into the pUC56 plasmid and linearised by PCR using universal primers ET-probe-F (5′-CGTTGTAAAACGACGGCCAGT-3′) and ET-probe-T7-R (5′-GAAATTAATACGACTCACTATAgggagacccCACAGGAAACAGCTATGACC-3′). Digoxigenin (DIG)-labelled RNA probes were synthesised in vitro and hybridised to embryos fixed at different developmental stages. Hybridisation signals were detected using an anti-DIG antibody (1:3000) followed by chromogenic staining, and embryos were examined under a stereomicroscope to assess early neural developmental patterns.²⁷

scRNA-seq

scRNA-seq was performed to investigate transcriptional alterations associated with rars2 deficiency in zebrafish embryonic development, with a focus on neural tissues.²⁸ Approximately 200 embryos per group (wild-type and rars2^–/–) at 48 hpf were dissociated into single cells using enzymatic/mechanical digestion and processed with the 10× Genomics Chromium Single Cell 3′ Kit (V.3.1). Libraries were sequenced on NovaSeq 6000. Raw data were processed with Cell Ranger (V.6.0.1) using the GRCz11 reference genome. Downstream analyses were performed in Seurat (V.4.1), including QC filtering, normalisation, principal component analysis, Uniform Manifold Approximation and Projection clustering and differential gene expression (Wilcoxon test). Cell types were annotated using canonical markers. Functional analyses included Gene Ontology (GO)/Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment, Gene Set Variation Analysis (GSVA), pseudotime (Monocle 3) and intercellular communication (CellChat). Raw and processed scRNA-seq data have been deposited in Gene Expression Omnibus under accession GSE329871.²⁹ Additional experimental and bioinformatic details are provided in the online supplemental methods.

Cell proliferation and apoptosis assays

Proliferating cells were labelled using phospho-histone H3 (PH3; M phase) and 5-bromo-2′-deoxyuridine (BrdU; S phase) at specified time points. Central nervous system (CNS) and notochord regions were quantified from confocal z-stacks. Apoptotic cells were detected by acridine orange staining, and lysosomal accumulation was assessed by LysoTracker Red fluorescence.

Statistical analysis

Statistical analyses were performed using GraphPad Prism 8. Data are presented as mean±SEM. Between-group comparisons were conducted using unpaired t-tests or analysis of variance, with significance thresholds of *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001; NS indicates no significant difference.

Results

Clinical and neuroimaging features defining a severe RARS2-related PCH

The proband (II-3) was a preschool-aged child from a non-consanguineous Chinese family who presented with infantile-onset epilepsy, hypotonia, global developmental delay and PCH (figure 1A). Seizures began at 5 months of age and were described clinically as recurrent limb convulsions/rigidity during infancy. According to longitudinal follow-up records, overt convulsive episodes were no longer clearly documented after approximately 7–8 months of age, although later follow-up still noted intermittent abnormal movements, gait instability and abnormal EEG activity. Sleep EEG showed diffuse background slowing and frequent epileptiform discharges involving the bilateral occipital and posterior temporal regions, more prominent during sleep, whereas no electroclinical seizure was captured during monitoring. The patient initially received traditional Chinese medicine for approximately 1 year during infancy and was later treated with lacosamide from 4 years 4 months of age, valproate oral solution at 4 years 11 months, and subsequent lacosamide dose adjustment at 8 years 4 months because of persistent abnormal movements and gait instability. However, according to the caregiver report, no definite clinical benefit was observed. Overall, the phenotype was most consistent with a history of infantile-onset epilepsy with bilateral motor seizures of uncertain onset in the setting of severe neurodevelopmental impairment. Brain MRI revealed cerebellar atrophy predominantly affecting the vermis and superior cerebellar hemispheres, with thinning of the superior cerebellar peduncles (figure 1B). These findings established a severe neurodevelopmental phenotype compatible with RARS2-related PCH6.

Figure 1. Clinical, radiological and molecular characterisation of a family with compound heterozygous RARS2 mutations associated with pontocerebellar hypoplasia type 6. (A) Family pedigree demonstrating a typical autosomal recessive inheritance pattern, with the proband (II-3) indicated by an arrow. (B) Brain MRI of the proband showing cerebellar atrophy predominantly affecting the vermis and superior cerebellar hemispheres, with thinning of the superior cerebellar peduncles. (C) Schematic diagram of the RARS2 gene indicating the locations of two missense variants, c.772C>T (p.R258C) and c.436C>T (p.R146C), within the arginyl-tRNA synthetase signature and ArgRS_core domains, respectively. (D) Multiple sequence alignment across vertebrate species demonstrating strong conservation of the affected residues R258 and R146. (E) Spatial distribution of the mutation sites within the three-dimensional model of the RARS2 protein, illustrating their proximity to the catalytic core region. (F) Structural representation of the local residue environment around p.R258C and p.R146C, highlighting altered side-chain orientations and potential disruption of intramolecular contacts. (G) AlphaFold-predicted tertiary structure models of WT and mutant RARS2 proteins showing conformational deviations consistent with impaired enzymatic stability and function. ArgRS, arginyl-tRNA synthetase; DALR, Asp-Ala-Leu-Arg anticodon-binding domain; Mut, mutant; WT, wild-type.

Identification and structural characterisation of compound heterozygous RARS2 variants

WES identified two novel compound heterozygous missense variants in RARS2 (NM_020320.5): c.772C>T (p.R258C), inherited from the mother (I-1) and c.436C>T (p.R146C), inherited from the father (I-2), consistent with autosomal recessive inheritance (figure 1A). The c.772C>T variant lies within the arginyl-tRNA synthetase signature domain and c.436C>T within the ArgRS_core domain (figure 1C). Both residues are highly conserved across vertebrate species (figure 1D), supporting functional importance. Sanger sequencing confirmed co-segregation of the two variants with the disease phenotype within the family (online supplemental figure 1).

Neither variant was present in major population databases and both were classified as pathogenic according to ACMG/AMP guidelines. We performed in silico splice prediction using SpliceAI, dbscSNV and MaxEntScan/GeneBe (online supplemental table 1). The c.436C>T (p.R146C) variant showed a borderline SpliceAI signal, with a maximum delta score of 0.22, exceeding the prespecified high-recall cut-off of 0.20 but remaining below the commonly used threshold of 0.50. In contrast, c.772C>T (p.R258C) showed a low SpliceAI maximum delta score of 0.03. dbscSNV scores were unavailable for both variants. MaxEntScan/GeneBe sliding-window analysis showed bidirectional local donor- and acceptor-like motif changes for both variants. For c.436C>T, the largest donor changes were Δ=−5.48 and +7.75, and the largest acceptor changes were Δ=−2.18 and +1.46. For c.772C>T, the GeneBe output displayed variant-proximal local donor Δscores of +5.79 (WT score: −20.47,Mut score: −14.68) and −4.88 (WT score: −15.32,Mut score: −20.20); the −4.9 value represented the largest local donor decrease, whereas the largest local donor increase among affected low-scoring donor-like sliding windows reached +7.75. The largest acceptor decrease and increase for c.772C>T were Δ = −2.50 and +1.71, respectively. Because all affected MaxEntScan/GeneBe windows had low or negative scores, these findings were interpreted as local splice-site-like motif perturbations rather than strong standalone evidence for splice-site disruption. Overall, the in silico analysis suggests a possible weak or borderline splice effect for c.436C>T, whereas c.772C>T showed no clear splice-altering signal by current computational prediction alone. AlphaFold-based structural modelling mapped p.R146C and p.R258C to the catalytic core of RARS2 (figure 1E). Local structural views showed altered side-chain orientations and intramolecular contacts around both residues (figure 1F), and superimposed wild-type versus mutant models demonstrated conformational rearrangements in the catalytic region (figure 1G), suggesting potential impairment of enzyme stability and substrate binding. Together with the genetic evidence, ACMG/AMP classification and structural modelling, these data support the pathogenicity of the compound heterozygous RARS2 variants and their causal role in the proband’s PCH6 phenotype.

Generation and characterisation of rars2-deficient zebrafish

For the disease model in vivo, we generated rars2 mutant zebrafish using CRISPR/Cas9-mediated editing. Indel mutations at the gRNA target sites produced frameshifts and premature stop codons predicted to abolish rars2 function (online supplemental figure 2A). Mosaic F0 founders were screened by sequencing and outcrossed to wild-type fish to generate stable F1 and F2 carriers, from which rars2^−/− lines were established.

rars2-deficient zebrafish show reduced survival and locomotor impairment

Continuous survival monitoring revealed a marked reduction in viability of rars2^−/− embryos compared with wild-type siblings over 20 dpf. Survival began to decline from 5 dpf and remained significantly lower throughout early development (figure 2A), indicating that rars2 deficiency severely compromises organismal fitness.

Figure 2. Survival rate and locomotor behaviour of rars2^−/− zebrafish (A) Kaplan-Meier survival analysis of WT and rars2^−/− zebrafish embryos over 20 days postfertilisation. (B) Representative locomotor trajectories of WT and rars2^−/− larvae at 7 days postfertilisation. (C) Quantitative analysis of locomotor parameters, including total distance, average speed, maximum acceleration, mobile time and resting time. Data are presented as mean±SEM; n>100 per group. Survival curves in panel A were compared using the log-rank (Mantel–Cox) test, whereas locomotor parameters in panel C were analysed using unpaired Student’s t-tests. WT, wild-type.

Locomotor trajectories at 7 dpf recorded by the EthoVision XT system showed that wild-type larvae displayed irregular but centrally distributed swim paths, whereas rars2^−/− larvae exhibited repetitive, restricted trajectories confined to the periphery of the arena (figure 2B). Quantitative analysis demonstrated significantly decreased total travel distance and lower mean swim speed in rars2^−/− larvae at 7 dpf (figure 2C). At 4 mpf, open-field testing showed that wild-type adults explored both central and peripheral regions, while rars2^−/− adults moved in a limited back-and-forth pattern along the tank edge (online supplemental figure 2B). Detailed behavioural parameters at 4 mpf confirmed persistent alterations in activity and resting frequency (online supplemental figure 2C).

Progressive ultrastructural neurodegeneration in rars2^−/− brains

Transmission electron microscopy revealed progressive neurostructural degeneration in rars2^−/− embryos. At 48 hpf, neuronal mitochondria were swollen with disrupted cristae and ruptured outer membranes (figure 3A). Lysosomes were increased and often fused with mitochondria, and synaptic vesicles were fewer and morphologically abnormal; synaptic clefts were widened and presynaptic/postsynaptic membranes partially dissolved (figure 3A). At 72 hpf, mitochondrial swelling, membrane rupture and synaptic disruption were more pronounced (online supplemental figure 3A), and by 5 dpf extensive myelin degeneration was evident, including thinning and delamination of myelin sheaths, vacuole-like changes within the myelin layers and focal axonal swelling or dissolution (online supplemental figure 3B). In some regions, axons appeared completely demyelinated. These ultrastructural changes demonstrate early and progressive mitochondrial and myelin pathology in rars2-deficient brains.

Figure 3. Neurodevelopmental abnormalities in rars2^−/− zebrafish revealed by ultrastructural, immunofluorescence and in situ analyses. (A) Representative transmission electron microscopy micrographs showing neuronal ultrastructure in WT and rars2^−/− embryos at 48 hpf. Red arrows indicate M; blue arrows indicate Sc; green arrows indicate Sv; purple arrows indicate Pom, and yellow arrows indicate L. n=3 embryos per group. (B) Immunofluorescence staining of acetylated α-tubulin in the brain regions of zebrafish embryos at 48 hpf. Internal subpanels A and B show WT and rars2^−/− zebrafish, respectively. (C) Immunofluorescence staining of GFAP in zebrafish embryos at 48 hpf. Arrows indicate GFAP-positive glial cell localisation, and dashed arrows mark regions of disrupted structural organisation. Internal subpanels A/A′ and B/B′ show WT and rars2^−/− embryos, respectively. (D) Immunofluorescence detection of SCMAS in zebrafish embryos at 48 hpf. Internal subpanels A/A′ and B/B′ show WT and rars2^−/− embryos, respectively. n=6 per group. (E) Whole-mount in situ hybridisation of otx2b at 24 hpf, showing diminished expression in rars2^−/− embryos compared with WT, indicating impaired neural progenitor development. (A and B): lateral views; (A′ and B′): dorsal views. (F) HuC/D immunostaining at 48 hpf demonstrating reduced and disordered neuronal distribution in rars2^−/− embryos relative to WT. Internal subpanels A/A′/A″ and B/B′/B″ show WT and rars2^−/− embryos, respectively. All embryos are oriented with the head to the left ; (G) Immunohistochemical staining of PH3 in zebrafish embryos at 48 hpf. Arrows indicate areas containing PH3-positive cells. (A and B) Posterior view with anterior to the left; (A′ and B′) lateral view with dorsal up. All images are Z-projections. (H) Representative confocal images of BrdU immunofluorescence in zebrafish embryos at 48 hpf. Internal subpanels A/A′ and B/B′ show WT and rars2^−/− embryos, respectively. A and B show the CNS region, whereas A′ and B′ show the corresponding notochord region. The white outlines indicate the CNS area, and arrows mark relative BrdU-positive proliferating cells. All embryos are oriented with the anterior to the left. Images are Z-projections. (I) Quantitative analysis of relative BrdU-positive area ratio in the CNS region at 48 hpf, showing a significant reduction in rars2^−/− embryos compared with WT (p=0.0017). (J) Quantitative analysis of relative BrdU-positive area ratio in the notochord region at 48 hpf. Each group includes at least five embryos per genotype (n≥5). Data are presented as mean±SEM, and p-values were determined using an unpaired t-test (**p<0.01). CNS, central nervous system; GFAP, glial fibrillary acidic protein; hpf, hours postfertilisation; Pem, presynaptic membrane; BrdU, 5-bromo-2′-deoxyuridine; PH3, phospho-histone H3; Pom, postsynaptic membrane; L, lysosomes; M, mitochondria; m, muscle fibres; N, nucleus; S, synapses; Sc, synaptic cleft; SCMAS, subunit c of mitochondrial ATP synthase; Sv, synaptic vesicles. Scale bars are indicated in the individual subpanels; WT, wild-type.

Disrupted axonal architecture and optic pathway development

Immunostaining for acetylated α-tubulin revealed well-organised axonal tracts and clearly defined brain regions in wild-type embryos at 48 hpf (figure 3B(A)). In contrast, rars2^−/− embryos showed fragmented and disorganised axonal tracts with blurred regional boundaries (figure 3B(B)). At 4 dpf, wild-type larvae maintained continuous axonal pathways and a well-defined optic nerve (online supplemental figure 4C), whereas rars2^−/− larvae showed underdeveloped and misaligned optic nerve fibres with regions of axonal discontinuity (online supplemental figure 4D). These findings indicate that rars2 loss impairs axonal patterning and optic pathway development.

Early disruption of glial networks

GFAP staining at 48 hpf revealed a continuous, well-organised network of glial fibres in wild-type embryos throughout the brain and notochord (figure 3C(A,A′)). In rars2^−/− embryos, GFAP signal intensity was reduced, and glial fibres appeared disordered and discontinuous (figure 3C(B,B′)). At 4 dpf, wild-type larvae exhibited strong GFAP expression along ventricular surfaces and diencephalic regions (online supplemental figure 4B,C), whereas rars2^−/− larvae showed fragmented, weakened GFAP staining with disrupted ventricular glial morphology (online supplemental figure 4B,D). These data indicate that rars2 deficiency perturbs glial network organisation early in development.

Abnormal mitochondrial protein accumulation and metabolic stress in neurons

To assess mitochondrial integrity, we examined SCMAS, the mitochondrial ATP synthase subunit c. At 48 hpf, wild-type embryos displayed weak, evenly distributed SCMAS signals in the forebrain and notochord (figure 3D(A,A′)). In rars2^−/− embryos, SCMAS fluorescence was markedly increased, with punctate aggregates along neuronal tracts in the brain and notochord (figure 3D(B,B′)). By 4 dpf, SCMAS remained low and diffuse in wild-type larvae but formed intense, widespread aggregates in the brain, optic vesicle and notochord of rars2^−/− larvae (online supplemental figure 4C,D). These findings implied impaired mitochondrial protein homeostasis, accompanied by mitochondrial structural stress in rars2-deficient neurons.

Impaired neuronal differentiation and progenitor development

HuC/D staining at 48 hpf showed abundant, well-organised postmitotic neurons throughout the brain and notochord of wild-type embryos (figure 3F(A and A′,A″)). rars2^−/− embryos showed reduced HuC/D signal and disrupted neuronal patterning, with notable neuronal loss in the forebrain (figure 3F(B and B′,B″)). At 4 dpf, HuC/D-positive neurons remained dense and orderly in wild-type larvae (online supplemental figure 4D(C)), whereas rars2^−/− larvae exhibited sparse, irregular neuronal distribution, particularly in brain regions (online supplemental figure 4D(D)), indicating progressive neuronal loss and disorganisation.

Expression of the neural progenitor marker otx2b was strong and well-defined in the forebrain of wild-type embryos at 24 hpf (figure 3E(A,A′)), but markedly reduced in rars2^−/− embryos (figure 3E(B,B′)). The combination of reduced otx2b and disorganised HuC/D patterns indicates that rars2 deficiency impairs both early progenitor development and subsequent neuronal differentiation.

Region-specific defects in cell proliferation

PH3 staining at 48 hpf showed numerous mitotic cells in the hindbrain and along the notochord in wild-type embryos (figure 3G(A,A′)), whereas rars2^−/− embryos had markedly fewer PH3-positive cells in both the brain and along the notochord (figure 3G(B,B′)). At 4 dpf, PH3-positive nuclei remained abundant in the optic tectum, forebrain and cerebellum of wild-type larvae (online supplemental figure 4E(C)), but were significantly reduced in rars2^−/− larvae (online supplemental figure 4E(D)), indicating persistent suppression of mitotic activity.

BrdU labelling revealed similar deficits. In the CNS region, rars2^−/− embryos had fewer relative BrdU-positive area ratio than wild-type. In the CNS region, rars2^−/− embryos showed reduced BrdU labelling at 24 hpf (online supplemental figure 4F) and 48 hpf (figure 3H);, and quantitative analysis confirmed significantly lower BrdU signal in the mutant CNS at 24 hpf (online supplemental figure 4G) and 48 hpf (figure 3I). In the notochord (somite 14–20), however, relative BrdU-positive area ratio did not differ significantly between genotypes at either timepoint (online supplemental figure 4H and figure 3J), indicating that rars2 deficiency primarily affects proliferation in brain progenitors.

Enhanced apoptosis and lysosomal dysfunction

Acridine orange and LysoTracker Red staining demonstrated increased apoptosis and lysosomal stress in rars2-deficient embryos. At 48 hpf, wild-type embryos showed localised AO fluorescence in the ventral forebrain and optic vesicle and evenly distributed LysoTracker signals (online supplemental figure 5A,B), with partial co-localisation. rars2^−/− embryos displayed markedly increased AO signal in the optic tectum and ventral head regions and aggregated LysoTracker fluorescence in forebrain and optic areas, with strong AO/LysoTracker co-localisation (online supplemental figure 5C,D). At 4 dpf, AO and LysoTracker signals remained low in wild-type larvae, whereas rars2^−/− larvae exhibited intense, overlapping signals in forebrain, midbrain, retina and yolk boundary (online supplemental figure 5E,F), indicating persistent lysosomal dysfunction and apoptosis in neural tissues.

Single-cell transcriptomic profiling reveals lineage imbalance and loss of neurogenic programmes

scRNA-seq of 48 hpf embryos passed stringent quality control, with comparable gene counts, Unique Molecular Identifier counts and mitochondrial/ribosomal content between wild-type and rars2^−/− samples (online supplemental figure 6A). Highly variable genes showed similar dispersion patterns across groups (online supplemental figure 6B), supporting downstream comparison.

Unsupervised clustering identified 23 major cell populations corresponding to neural, skeletal, muscular, epithelial, vascular and immune lineages (figure 4A and online supplemental table 2). Focusing on the neuroectoderm, secondary clustering resolved 16 neural subpopulations, including neural progenitors, retinal progenitors, radial glia, astrocytes/glia, inhibitory neurons and sensory neurons, each defined by specific marker combinations (figure 4B and online supplemental table 3). Cell-type composition analysis showed that neural progenitor and glial-like populations were expanded in rars2^−/− embryos, whereas differentiated neuronal populations were reduced (figure 4C). Retinal/neural progenitors were also enriched in mutants, suggesting delayed neuronal maturation.

Figure 4. Single-cell transcriptomic profiling reveals major cell populations and subtype-specific alterations in rars2^−/− zebrafish embryos. (A) UMAP visualisation of single-cell transcriptomes from WT and rars2^−/− zebrafish embryos. (B) Refined UMAP plot displaying neural lineage subclusters identified within the neuroectodermal population. (C) Comparative proportion analysis of each neural subpopulation between WT and rars2^−/− groups. (D) CellChat network analysis comparing intercellular communication patterns between WT and rars2^−/− embryos. Upper panels show the total number of interactions, and lower panels show interaction strength. (E) Differential pathway enrichment analysis highlighting communication networks significantly altered in rars2^−/− embryos. UMAP, Uniform Manifold Approximation and Projection; WT, wild-type.

Differential expression revealed downregulation of neurogenic transcription factors (eg, neurod4, neurog1, ascl1a, pou3f1 and her-family genes) and upregulation of glial/stress-associated genes (fabp7a, foxp1b, hmgb1a and sox2) in rars2-deficient neural clusters (figure 4C and online supplemental figures 9–11). GO and KEGG analyses showed attenuation of neurogenesis, axonogenesis, Notch signalling and spliceosome-related RNA processing, alongside enrichment of ECM organisation, ECM–receptor interaction and focal adhesion pathways (figure 5A,B, online supplemental figures 9,10 and online supplemental tables 8–11). These data indicate a coordinated shift from neuronal differentiation towards glial and structural programmes in rars2-deficient embryos.

Figure 5. Single-cell transcriptomic analyses reveal functional reprogramming in rars2^−/− embryos. (A) GO BP enrichment analysis between control and rars2^−/−. (B) KEGG pathway enrichment analysis between control and rars2^−/−. (C) Cluster-specific GSVA showing pathway changes across neural and glial populations. (D) Global GSVA reveals increased inflammatory and EMT pathways but reduced metabolic and mitochondrial activity in rars2^−/− embryos. (E) Cell cycle analysis of neural subpopulations in control and rars2^−/− embryos. EMT, epithelial-mesenchymal transition; GO BP, Gene Ontology Biological Process; GSVA, Gene Set Variation Analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Rewired intercellular signalling and developmental trajectories

CellChat analysis demonstrated that overall interaction topology among neural populations was preserved, but rars2^−/− embryos exhibited increased communication between astrocytes and neural progenitors and reduced signalling involving Notch⁺ inhibitory neurons (figure 4D and online supplemental tables 4 and 5). Pathway-level analysis revealed decreased Notch and non-canonical Wnt signalling, together with altered collagen-related communication and compensatory enhancement of adhesion-related pathways such as neural cell adhesion molecule (NCAM) and Eph receptor B (EPHB) signalling in mutants (figures 4E and 6A and online supplemental tables 6 and 7). These changes suggest weakened neurogenic signalling and increased ECM- and adhesion-mediated remodelling.

Figure 6. Pseudotime and BEAM analysis reveal transcriptional bifurcation and fate-specific gene programmes in rars2^−/− zebrafish embryos. (A) Comparison of major intercellular signalling pathways between WT and rars2^−/− embryos identified by CellChat analysis. (B) Pseudotime trajectory analysis of neural lineage cells. (B-1) trajectory coloured by pseudotime. (B-2) trajectory coloured by genotype (control and rars2^−/−). (B-3) trajectory coloured by neural subpopulation. (C) BEAM identifies dynamic gene modules along the bifurcated trajectories. BEAM, branched expression analysis modelling.

Pseudotime reconstruction suggested smooth progression from neural progenitors toward more differentiated neural states in wild-type embryos, whereas rars2-deficient cells were enriched in early and intermediate pseudotime regions (figure 6B). Branched expression analysis modelling (BEAM) revealed that one fate branch was enriched for neuronal specification genes, whereas the alternative branch showed preferential expression of glial and structural regulators such as hmgb1a and fabp7a (figure 6C and online supplemental figure 7). These trajectory differences reflected a global developmental shift or a more specific lineage abnormality; we performed additional trajectory-oriented analyses (online supplemental figure 8). Global pseudotime distribution between control and rars2-deficient cells did not show a statistically significant overall shift (p=0.07283), arguing against a simple uniform developmental delay. At the programme level, glial transcriptional programmes were globally increased in mutant cells, whereas neurogenic programmes showed increased heterogeneity rather than a consistent uniform decrease. Notably, within neural progenitor cells (cluster 2), both neurogenic and glial transcriptional programmes were significantly elevated in mutant cells, indicating a mixed or dysregulated lineage state rather than a simple binary neuron-to-glia fate switch. Consistently, pseudotime-dependent gene analysis showed that neurod4 exhibited a transient early-to-mid increase followed by decline, whereas fabp7a progressively increased towards later pseudotime stages.

Integrated pathway and cell-cycle changes link mitochondrial dysfunction to the clinical phenotype

Pathway enrichment confirmed that mitochondrial and translational pathways—oxidative phosphorylation, tricarboxylic acid cycle, aminoacyl-tRNA biosynthesis and ribosome function—were prominently altered in rars2-deficient embryos (figure 5A,B and online supplemental tables 8–11). GSVA demonstrated activation of inflammatory and stress-related signalling (eg, interleukin-6 (IL6)–Janus kinase–signal transducers and activators of transcription 3 (STAT3), tumour necrosis factor alpha–nuclear factor kappa-light-chain-enhancer of activated B cells, interferon responses) in astrocytes/glia and astrocyte-like clusters and suppression of oxidative phosphorylation and apoptosis pathways in inhibitory neurons (figure 5C,D and online supplemental figure 12). Cell-cycle analysis showed population-specific changes, including increased G1-phase accumulation in inhibitory neurons, a pronounced reduction in the G2/M fraction in Notch⁺ neural progenitors, and increased S/G2M fractions in astrocytic/glial populations (figure 5E).

Taken together, these multi-level data—from clinical observations and genetic findings to ultrastructural, histological and single-cell transcriptomic analyses—support a model in which RARS2 deficiency impairs mitochondrial translation, progenitor proliferation and neurogenic programmes and lineage progression, while promoting glial/structural programme activation, thereby producing the pontocerebellar hypoplasia and severe neurodevelopmental phenotype observed in the proband.

Discussion

PCH6 is an early-onset neurodevelopmental disorder, marked by underdevelopment of the pons and cerebellum, intractable seizures, profound developmental delay and high mortality. Variants in RARS2, which encodes mitochondrial arginyl-tRNA synthetase, are recognised as a genetic cause of PCH6, yet the mechanistic links between mitochondrial translation defects and neural development remain largely unexplored.^{30 31} Although recent clinical and metabolic profiling studies have expanded the phenotype of RARS2-related disorders, they largely focused on descriptive or biochemical aspects,^{32 33} leaving the cellular and transcriptional mechanisms of PCH6 pathogenesis unresolved. Our single-cell transcriptomic dissection of the rars2^−/− zebrafish embryos provides a high-resolution map of how mitochondrial translation defects disrupt neurodevelopmental programmes and rewire transcriptional networks. We acknowledge, however, that the current scRNA-seq dataset was generated from pooled embryos collected at a single developmental timepoint (48 hpf) and therefore provides a snapshot of transcriptional states rather than direct longitudinal lineage tracing. Accordingly, the pseudotime and branch analyses should be interpreted as computationally inferred lineage relationships, not as direct temporal observation of progressive developmental transitions. The zebrafish model used in this study represents a stable rars2 loss-of-function/null condition, whereas many affected individuals, including the proband reported here, carry missense variants that may retain partial residual function. Therefore, although this model is highly informative for defining core disease-relevant mechanisms, it may accentuate phenotypic severity and does not fully capture the allelic heterogeneity of human RARS2-related disease.

In contrast to previous studies that associated rars2 dysfunction with general mitochondrial insufficiency,¹² our pseudotime analysis(figure 6B,C) and GSVA (figure 5C,D) revealed that rars2 deficiency triggers a progressive redirection of neural progenitors away from normal neurogenic and axonogenic programmes towards ECM remodelling, cell adhesion and metabolic stress pathways. Unlike prior studies focusing solely on mitochondrial dysfunction,³⁴ our trajectory-oriented analyses support a model of dysregulated lineage progression and impaired lineage commitment, with neural progenitors increasingly expressing col1a2, matn1 and col11a1b, markers typical of mesenchymal and glial-like states (online supplemental figure 10A,B). This mis-specification is further supported by the strong upregulation of phosphoinositide 3-kinase–protein kinase B–mechanistic target of rapamycin and IL6–STAT3 signalling in astrocytic and glial clusters (figure 5D), suggesting that mitochondrial stress triggers a gliosis-like compensatory response similar to reactive astrogliosis observed in human PCH6 brains.³⁵ Electron microscopy in our model further confirmed extensive mitochondrial swelling and cristae disruption, morphologically supporting mitochondrial structural injury and cellular stress consistent with the single-cell findings. However, direct mitochondrial functional assays, such as ATP production or oxygen consumption measurements, were not performed in the present study. Therefore, the current data support mitochondrial structural dysfunction and stress but do not directly establish the magnitude of bioenergetic impairment.

The cell cycle analysis (figure 5E) demonstrated population-specific redistribution of cell-cycle phases, including increased G1-phase accumulation in inhibitory neurons and a reduced G2/M fraction in Notch⁺ neural progenitors, consistent with impaired proliferative dynamics. This aligns with our in situ hybridisation results showing decreased expression of neurod4 and pou3f1 in forebrain regions, together with enhanced fabp7a and gfap signals in ventricular zones, consistent with impaired neuronal differentiation accompanied by aberrant glial programme activation. These in situ and immunofluorescence results directly support our single-cell inference that rars2 loss promotes premature cell cycle exit and fate deviation.

A particularly novel layer of insight comes from our transcription-factor (TF) analysis (online supplemental figures 13 and 14). We identified downregulation of classical neurogenic TFs (neurod4, pou3f1 and her6) and activation of non-canonical regulators such as foxp1b, foxa1, hmgb1a and fabp7a. Previous clinical or genomic papers could not resolve this regulatory rewiring.¹⁰ Our results instead delineate a dual process of loss of neurogenic identity and gain of ECM/inflammatory programmes, indicating that rars2 deficiency disrupts chromatin accessibility and Notch-dependent lineage control at the transcriptional level (online supplemental figure 14C,D), consistent with enrichment results showing suppression of spliceosome and mRNA-processing pathways (online supplemental figure 10C). This regulatory breakdown provides the molecular context for the histological gliosis and astrocyte activation observed in human PCH6 tissue.

These findings should also be considered in the context of evidence that RARS2 pathogenicity may involve transcript-level mechanisms. de Valles-Ibáñez et al suggested that RARS2 is enriched for pathogenic variants that disrupt splicing, including variants previously annotated as missense changes,³⁶ and van Dijk et al showed that a RARS2 splice-site variant induced exon 4 skipping.³¹ Although our two exonic variants did not show strong computational evidence for a major splice-disruptive effect, the weak/borderline signal for c.436C>T and local MaxEntScan/GeneBe motif perturbations suggest that variant-level effects on RARS2 transcript processing cannot be fully excluded (online supplemental table 1). We therefore propose that patient variants may contribute to RARS2 transcript vulnerability, while rars2 deficiency secondarily disrupts spliceosome homeostasis and chromatin/lineage programmes, thereby amplifying neurodevelopmental pathology.

Among the newly identified factors, HMGB1A/HMGB3A, FABP7A, FOXP1B and NME2B.1 (online supplemental figure 15) stand out as potential molecular bridges linking mitochondrial dysfunction to fate reprogramming. While HMGB1A has been studied mainly in inflammation and cancer, our findings suggest a potential role in neuronal chromatin regulation and stress signalling, thereby adding a novel dimension in mitochondrial neuropathology. Likewise, FABP7A, a known astroglial marker, appears here as an associated marker of glial-biased lineage dysregulation, preceding overt gliosis. However, direct functional perturbation experiments to test the causal roles of individual regulators, such as neurod4, pou3f1, hmgb1a or fabp7a, were not performed in the present study. Therefore, these factors should be interpreted as associated markers or candidate mediators, rather than definitive drivers of pathology.

Importantly, our ultrastructural and histological findings are broadly consistent with the transcriptomic results. In rars2-deficient embryos, GFAP immunostaining revealed early disruption and disorganisation of glial networks rather than a normal continuous glial architecture, while HuC/D staining showed reduced and disordered neuronal distribution, supporting impaired neuronal differentiation and abnormal glial organisation. In parallel, SCMAS accumulation indicated disturbed mitochondrial protein homeostasis, and transmission electron microscopy demonstrated swollen mitochondria with disrupted cristae, reduced and abnormal synaptic vesicles, widened synaptic clefts and partial dissolution of presynaptic and postsynaptic membranes. Together, these observations support a model in which rars2 deficiency is associated with mitochondrial structural injury, synaptic abnormalities, impaired neuronal differentiation and disordered glial development, in line with the single-cell evidence for lineage dysregulation. Compared with prior RARS2 studies, our work provides a mechanistic framework for how severe rars2 loss-of-function may connect mitochondrial translation defects to transcriptional dysregulation and developmental instability. Moreover, we propose that similar pathways may underlie other mitochondrial aminoacyl-tRNA synthetase disorders, including DARS2- or RARS1-related leukoencephalopathies, raising the possibility that related pathways may also operate in other mitochondrial aminoacyl-tRNA synthetase disorders.³⁷

Collectively, our findings support a multi-level mechanistic model for RARS2-related PCH6 in which potential variant-level transcript vulnerability may converge with rars2 deficiency-driven mitochondrial translation failure to produce spliceosome/chromatin dysregulation, loss of neurogenic TFs, impaired lineage commitment with glial/structural programme activation and neuronal degeneration/gliosis (online supplemental figure 16). Compared with previous studies focusing solely on genetic or biochemical aspects, this work provides the first integrative cell-type-specific framework that connects mitochondrial dysfunction to transcriptional and developmental dysregulation in PCH6. The convergence of single-cell, immunohistochemical and ultrastructural evidence underscores the robustness of this model and reveals potential intervention points, such as targeting Notch/STAT3 reactivation or correcting splicing factor homeostasis. Future studies using patient-specific knock-in or hypomorphic zebrafish models will be important to determine how different RARS2 alleles influence phenotypic severity and neural lineage defects and to improve genotype-specific translational relevance. Formal rescue experiments in zebrafish will also be important to strengthen the causal link between rars2 deficiency and the developmental phenotypes observed in the present study. Future studies may explore whether metabolic modulation or mitochondrial quality control can rescue the observed lineage mis-specification, offering a rational foundation for therapy development in mitochondrial neurodevelopmental disorders.

Data availability statement

Data are available in a public, open access repository.