Loss-of-function mutations in UDP-Glucose 6-Dehydrogenase cause recessive developmental epileptic encephalopathy

Developmental epileptic encephalopathies are devastating disorders characterized by intractable epileptic seizures and developmental delay. Here, we report an allelic series of germline recessive mutations in UGDH in 36 cases from 25 families presenting with epileptic encephalopathy with developmental delay and hypotonia. UGDH encodes an oxidoreductase that converts UDP-glucose to UDP-glucuronic acid, a key component of specific proteoglycans and glycolipids. Consistent with being loss-of-function alleles, we show using patients’ primary fibroblasts and biochemical assays, that these mutations either impair UGDH stability, oligomerization, or enzymatic activity. In vitro, patient-derived cerebral organoids are smaller with a reduced number of proliferating neuronal progenitors while mutant ugdh zebrafish do not phenocopy the human disease. Our study defines UGDH as a key player for the production of extracellular matrix components that are essential for human brain development. Based on the incidence of variants observed, UGDH mutations are likely to be a frequent cause of recessive epileptic encephalopathy.

D evelopmental epileptic encephalopathies are a clinically and genetically heterogeneous group of devastating disorders characterized by severe epileptic seizures that are accompanied by developmental delay or regression 1 . In several cases, a genetic etiology has been identified 2 . Germline mutations in these genes lead to different pathophysiological defects 2 , including ion channel dysfunction, synaptic impairment, transporter defects and metabolic abnormalities, such as deficiencies in glycosylation pathways [3][4][5] . However, the genetic cause of many epileptic encephalopathies remains unknown.
Defects of glycosylation are causing more than 100 rare human genetic disorders, most of these affecting the central and/or peripheral nervous systems. Patients typically show developmental delay or intellectual disability, seizures, neuropathy, and metabolic abnormalities in multiple organ systems 3 . Adding the correct sugar chains (glycans) to proteins and lipids significantly impacts their function. UGDH (MIM603370) codes for an enzyme that converts UDP-glucose (UDP-Glc) to UDPglucuronic acid (UDP-GlcA) through the concomitant reduction of NAD + into NADH 6,7 . UDP-GlcA is not only needed for detoxification via glucuronidation, but is also an obligate precursor for the synthesis of glycosaminoglycans (GAGs), and therefore an important component of proteoglycans of the extracellular matrix.
In this study, we establish UGDH as a gene responsible for autosomal recessive developmental epileptic encephalopathy in humans. We catalog a series of 30 patients from 25 families with biallelic germline UGDH variants. Using patients' primary fibroblasts and biochemical assays, we demonstrate that these are loss-of-function alleles. While mutant ugdh zebrafish did not phenocopy the disease, we bring evidence that patient-derived cerebral organoids, which were smaller due to a reduced number of proliferating neuronal progenitors, can serve as an alternative disease-in-a-dish model for in vitro functional studies.

Results
Biallelic mutations in UGDH cause developmental epileptic encephalopathy. To identify the genetic cause of a developmental epileptic encephalopathy in a consanguineous Palestinian family with three affected siblings (Fig. 1a, F1), we performed exome sequencing on two affected siblings. No mutations in genes known to be associated with neurological disorders (either recessive or dominant) were found. As the consanguineous background and the pedigree suggested autosomal recessive inheritance, we focused on homozygous or compound heterozygous variants shared by the affected siblings. A rare homozygous variant c.131C > T in UDP-Glucose 6-Dehydrogenase (UGDH), which changes alanine into valine at position 44 of the UGDH protein, was the only segregating candidate variant. The UGDH p.A44V missense affects a highly conserved residue (Suppl. Fig. 1b and phyloP 100-way 8 score 9.43), is extremely rare in public databases (not present in EVS6500 9 , MAF of 0.0017% in ExAC 10 ) and is a good candidate according to in silico prediction scores (CADD score 11 of 33) (Suppl. Table 1). We then (i) screened the GENESIS 12 database for additional patients with recessive UGDH variants, (ii) contacted the EuroEPINOMICS RES Consortium, and (iii) searched with the help of Gene-Matcher 13 for additional families with germline UGDH mutations. We uncovered 27 additional patients from 24 families carrying either compound heterozygous or homozygous UGDH variants ( Fig. 1a and Suppl. Fig. 1a). All variants were absent or had an extremely low frequency (<0.01%) in the public databases ExAC/gnomAD 10 and EVS6500 (Suppl. Table 2). Nineteen of the 20 identified missense variants are in highly conserved residues (Suppl. Fig. 1b and phyloP 100-way between 3.81 and 9.43). The A44V variant, identified in the Palestinian index family, was also found in two additional families from Puerto Rico (F11) and from Spain (F13) indicative of independent but recurrent mutation in this residue. In ExAC the A44V variant is observed in African (MAF 0.0096%) and European (Non-Finish) populations (MAF 0.0015%), however, it is not present in the Greater Middle East Variome.
All 30 patients carrying biallelic mutations in UGDH presented with a common core phenotype consisting of marked developmental delay, epilepsy, mild dysmorphism, and motor disorder with axial hypotonia (Table 1 and Suppl. Data 1). Dysmorphic facial features such as short and flattened philtrum, outward protruding earlobes, ptosis, or blepharophimosis were mild but frequently present (Fig. 1b and Suppl. Data 1). Most patients have severe epilepsy ranging from neonatal onset developmental epileptic encephalopathy to infantile developmental epileptic encephalopathy (27 patients, 90%), of which 16 (53%) had infantile spasms (Table 1). Three patients have developmental encephalopathy, of which two had seizures in the setting of fever (F5-II:1 and F5-II:2). Only these two patients were seizure-free on sodium valproate. All other patients, except for one patient who seemed to benefit from ketogenic diet, did not respond to antiepileptic treatment. All patients had a severe motor disorder with axial hypotonia, while some patients presented with limb spasticity (43%), dystonia (17%), ataxia, chorea, and tremor, which were often present prior to onset of seizures. Twenty-four out of the 30 (80%) were noted to have swallowing difficulties and gastrostomy tubes were required for feeding in 12 infants. None but two patients (F5-II:1 and II:2) achieved sitting ability. A moderate to severe intellectual disability was observed in all patients. Three patients were deceased between 4 months and 6 years of age (Table 1 and Suppl. Data 1). Electroencephalography (EEG) was markedly abnormal with a burst suppression pattern in the neonatal period, hypsarrhythmia in affected children with infantile spasms, and focal and/or generalized spike-wave complexes in childhood (Suppl. Data 1). MRI revealed a spectrum of abnormalities with delayed myelination and enlarged ventricles probably due to cerebral and cerebellar atrophy in more severely affected patients without any signs of maldevelopment ( Fig. 1c and Suppl. Data 1).
UGDH mutations behave as hypomorphic alleles. The UGDH oxidoreductase consists of three distinct domains 14 : the NADbinding (N-terminal) and UDP-binding (C-terminal) domains, and an internal domain that bridges the two termini together 14 . The UGDH enzyme assembles into a disc-shaped double layer composed of a trimer of dimers 6 (Suppl. Fig. 2a, b). This hexameric structure is a prerequisite for proper UGDH enzymatic function 15 . The 23 germline mutations presented in this study are distributed throughout the UGDH gene and its encoded protein (Fig. 2a). One of the variants in Family 12 mutates the first nucleotide of exon 8 (c.907 G > A; p.Val303Ile, Fig. 2a, b), which is predicted to affect the splice donor site 16 . Three different nonsense mutations were found in a compound heterozygous state with a missense mutation ( Fig. 2a and Suppl. Table 2). All identified missense mutations are anticipated to be destabilizing according to DUET 17 (Suppl. Table 2, ΔΔG). The missense mutations in residues Y14, I42 and A44, which are close to the NAD-binding site ( Fig. 2c and Suppl. Fig. 2c) are expected to impair NAD + reduction. Alteration of residues in the central domain such as I255, G271, M306, and R317 are expected to affect homo-dimerization 18 (Fig. 2b and Suppl. Fig. 2d). The I116 residue (located in the NAD-binding domain), as well as the R393 and A410 residues (UDP-Glc binding domain) sit at the dimerdimer interface 19 (Fig. 2d), suggesting that these variants may ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14360-7 prevent UGDH from assembling into a functional hexameric enzyme.
To better understand the effect of the mutations on UGDH, we then derived and biobanked primary dermal fibroblasts from patients F3-II:1 (R393W/A410S), F4-II:1 (Y14C/S72P), F5-II:1 (A82T/A82T) and F6-II:1 (R65*/Y367C), and a non-affected parent F5-I:1 (WT/A82T). Endogenous UGDH messenger RNA (mRNA) levels were not significantly different in patients' primary cells as compared to control fibroblasts (Fig. 3a, (Fig. 3b, left panels). Patient's cells with the homozygous A82T mutation also exhibited a reduction in the synthesis of hyaluronic acid (HA), which requires UDP-glucuronate, a product of UGDH enzymatic activity (Fig. 3b, right panel). When produced in bacteria, mutant UGDH had altered stability, kinetic, and biochemical properties as compared to WT-UGDH. Compared to the wild-type enzyme, mutant A44V and A82T UGDH (mutations found at the homozygous state in the patients from Families 1 and 5, respectively) were more susceptible to partial proteolysis by trypsin (Fig. 3c). The stability of UGDH A44V could be partially rescued upon incubation with substrate, product, or cofactor, while the UGDH A82T remained strongly sensitive to proteolysis regardless of the presence of any cofactor or substrate (Fig. 3c). A thermal stability study showed that the melting temperature of UGDH A44V was significantly reduced relative to WT and could only partially be rescued upon addition of substrate, product, reduced or oxidized cofactor, or any combination thereof (Fig. 3d). Notably, UGDH A82T was so intrinsically unstable that a melting temperature was unable to be ascertained. By gel filtration chromatography, we investigated the effect of the A44V and A82T mutations on UGDH oligomerization. When compared to UGDH WT , UGDH Δ132 (an obligate hexamer 15 ), and UGDH T325D (an obligate dimer 15 ), we observed that UGDH A44V and UGDH A82T proteins were mainly eluted as dimer and monomer species, respectively, with virtually no stable hexameric population (Fig. 3e). This suggests that the A44V and A82T mutations may affect UGDH function by altering its capacity to form active hexamers. Finally, using equal amounts of recombinant enzyme, we determined that UGDH A44V and UGDH A82T were respectively 75 and 50% less efficient at reducing NAD + to NADH as compared to UGDH WT (Fig. 3f). Similarly, comparison of the steady state Michaelis-Menten kinetic constants (summarized in Table 2) showed that UGDH A44V V max was only~50% of the value of UGDH WT for both cofactor and substrate. In contrast, K m was not significantly different from UGDH WT for either cofactor or substrate, revealing that the mutation results in a reduced ability of the enzyme to catalyze the reaction, while still being able to associate with NAD + and UDP-Glc. Taken together, our biochemical findings indicate that these missense mutations mainly impact the enzymatic function of UGDH by altering its quaternary structure and/or directly impairing its oxidoreductive activity.
Patient-derived cerebral organoids partially phenocopy human disease. Our attempts to model this developmental epileptic encephalopathy using the existing zebrafish hypomorphic lossof-function ugdh (c.992 T > A; p.I331D) allele known as jekyll m151 [20][21][22] were unsuccessful (Suppl. Fig. 3). The behavioral activity of homozygous jekyll mutant larvae were recorded in presence or absence of the seizure-inducing drug pentylenetetrazol (PTZ) 23 . By quantitative PCR (qPCR), c-fos expression, which marks neural activity 23,24 , similarly increased in a dosedependent manner upon PTZ treatment in all larvae regardless of genotypes (Suppl. Fig. 3a-d). Homozygous mutant larvae did not show signs of increased c-fos expression at basal state, suggesting that fish depleted of Ugdh activity do not exhibit spontaneous seizure and are equally responsive to PTZ treatment. As noted by reviewers, ugdh mutant fish do not have fully-inflated swim bladders, which may contribute to their reduced locomotor activity and demise before 14-dpf (Suppl. Fig. 3e). These in vivo experiments suggest that zygotic ugdh depletion in zebrafish does not satisfactorily model the human disease. UGDH has been extensively studied in vertebrate model organisms where its complete knockout causes embryonic lethality around gastrulation 25,26 . To address its role in the context of central nervous system (CNS) development in humans, we attempted instead to model this disease in vitro by developing cerebral organoids 27,28 from several patients with compound heterozygous R65*/Y367C, Y14C/S72P or homozygous A82T mutations, and from a non-affected parent (WT/A82T). After 10 weeks of differentiation, the volume of cerebral organoids from patients with biallelic UGDH mutations was on average 50% smaller and showed rougher edges than that of WT or carrier WT/A82T cerebral organoids (Fig. 4a, b and Suppl. Fig. 4a). Quantitative reverse transcription PCR (RT-qPCR) analysis revealed decreased levels of the early and intermediate neuronal progenitors markers PAX6 and TBR2, respectively, while the levels of neuronal marker TUJ1 were unchanged ( Fig. 4c and a Pedigrees of 19 families segregating autosomal recessive developmental epileptic encephalopathy. Countries of origin are specified above each pedigree. Filled black symbols, affected individuals. Crossed symbols, deceased individual. Mutations in UGDH protein are presented below pedigrees. Homozygous mutations are presented in bold (m in the pedigrees). Compound heterozygous mutations are presented according to the parental origin of the mutation with a maternal origin in the first row (m1 in the pedigrees), and a paternal, de novo or unknown origin in the second row (m2 in the pedigrees). Healthy siblings that could be sequenced are heterozygous [F6-II:2 (p.Arg65*), F11-II:2 (p.Ala44Val), and F18-II:3 (p.Arg317Gln)]. b Facial photographs of 14 affected individuals with mild craniofacial dysmorphisms, including short and flattened philtrum, protruding earlobes, ptosis blepharophimosis, and epicanthic folds. c Spectrum of MRI findings in exemplary patients showing no evidence for maldevelopment but displaying variable abnormalities ranging from abnormal myelination and/or cerebral or cerebellar atrophy, to normal findings. Patient F5-II:2 presented with a normal MRI, including normal myelination at 2 years of age. In contrast, MRI of patient F3-II:1 revealed some myelination of cerebellar peduncles at 5 months (arrow) and no progress of myelination on follow-up at 15 months, indicative of hypomyelination. In addition, repeated MRI revealed enlarged posterior ventricles over time (arrow heads). MRI of patient F6-II:1 at 7 days of age also proved normal, the circle indicates onset of myelination in the Posterior Limb of the Internal Capsule (PLIC) according to age. Patient F7-II:1 showed mild cerebellar atrophy at 4 years of age. Patient F9-II:1 showed slightly delayed myelination on axial T2 and cerebellar atrophy on coronal and sagittal T1 images (stars). Patient F14-II:1 showed a diffuse cerebral atrophy, ventriculomegaly, thin corpus callosum, vermian, and lobar cerebellar atrophy, with normal brainstem, hyperintensity of cerebellar cortex in T2-weighted images (white square). Patient F15-II:1 presented with normal MRI at 5 months, but with severe diffuse atrophy, bilateral symmetrical hyperintensities of thalami and globus pallidus (white square) at 8 months old. In all pictures, MRI pulse sequences (T1, T2, and Flair) and image orientation (S: sagittal, A: axial and C: coronal) are indicated in the upper left corner. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14360-7 Suppl. Fig. 4b). Immunofluorescence revealed similar amounts of peripheral neurons marked by TUJ1, and astrocytes marked by GFAP while ventricular zones marked by SOX2-positive neuronal progenitors were appreciably less proliferative. This was evidenced by reduced PCNA staining in mutant cerebral organoids relative to WT and WT/A82T sections ( Fig. 4d and Suppl. Fig. 4c). These results argue that reduced UGDH activity is associated with impaired neuronal development in vitro, causing atrophy of patient-derived cerebral organoids. Even though our cerebral organoid data is congruent with our patients' phenotype and biochemistry data, replicative studies with additional WT and complete UGDH knockout lines are warranted in light of the known variability in induced pluripotent stem cells (iPSCs') response to differentiation protocols.
To understand whether mutations in UGDH directly affect neuronal function, we also differentiated WT, non-affected parent (WT/A82T), and patient (Y14C/S72P) iPSCs into neuroprecursor cells (NPCs), which were subsequently matured into neurons over a period of 21 days. Using a multi-electrode array (MEA) system, and in contrast to neurons mutant for CAMK2A 29 , no significant differences between controls and mutant UGDH neurons were recorded for either the total number of spontaneous spikes or the mean firing rate (Suppl. Fig. 4d). Altogether, these in vitro experiments suggest that while UGDH is not required for proper function of isolated neurons in culture, its absence significantly affects neuronal differentiation in cerebral organoids, which may provide a powerful platform to study the pathogenesis for the new disease in vitro.

Discussion
In this study, we described disease-causing mutations in UGDH in humans. These 23 coding variants represent an allelic series of germline mutations, which when inherited recessively are responsible for epileptic encephalopathy with variable degrees of developmental delay. We propose to name this novel Mendelian disease Jamuar Syndrome, a member of the early infantile epileptic encephalopathies (EIEE). The genetic, biochemical, cellular and developmental findings reveal that these UGDH germline mutations behave as loss-of-function alleles. This was confirmed in vitro using patient-derived cerebral organoids, which showed marked underdevelopment. In zebrafish, we found that hypomorphic ugdh I331D/I331D mutant larvae did not show signs of increased seizures at baseline or after PTZ treatment. The brainspecific UGDH phenotype in humans may come as a surprise since in Drosophila, zebrafish, and mouse, complete knockout of Ugdh cause early and lethal gastrulation defects by hindering FGF signaling 25,[30][31][32] . One potential explanation for this incongruity is that other proteoglycans not reliant on UGDH activity for the synthesis of UDP-GlcA or UDP-Xylose (UDP-Xyl) may be solicited and help bypass the need for UGDH during human gastrulation. Alternatively, a complete knockout of UGDH in humans may not be viable as we did not identify any homozygous or compound heterozygous truncating mutations. A search in ExAC for homozygous variants resulted in mostly synonymous or non-coding variants. Only two homozygous missense variants were detected, but no homozygous truncating mutations were seen (Suppl. Table 3). This suggests that the severity of the epileptic encephalopathy may correlate with the amount of residual UGDH activity, the extent of which may be sufficient to allow gastrulation to take place during early human embryonic stages but may be limiting for neuronal development thereafter.
As UDP-GlcA is the major product of the UGDH enzyme, it is possible that reduced levels of UDP-GlcA may trigger a cascade of secondary pathogenic events resulting in neurodevelopmental delay and encephalopathy. In support of this, is the recent demonstration that a homozygous loss-of-function mutation in the upstream enzyme UGP2 is also responsible for a severe form of developmental epileptic encephalopathy in humans 33 . UDP-GlcA is not only needed for detoxification via glucuronidation, but is also a key component of glycosaminoglycans (GAGs). UGDH deficiency might parralel other neurological diseases with defects in GAG synthesis, modification, and degradation. For example, EXTL3 and CHSY1 mutations, which affect heparan sulfate and chondroitin sulfate synthesis, respectively, cause developmental delay and intellectual disabilities 34,35 . Defects in heparan sulfate modification caused by NDST1 mutations are responsible for intellectual disability associated with epilepsy 36 . Moreover, mucopolysaccaridoses, diseases caused by defects in GAG degradation, affect cognitive development 37 . In addition, proteoglycans containing GlcA derived from UDP-GlcA are major components of the extracellular matrix (ECM) and key players in neuronal development and plasticity 38 , particularly in areas important for neuronal migration 39 . In human, various psychiatric and intellectual disorders are caused by mutations in genes involved in ECM homeostasis and may be driven by neuronal migration defects 38 . The central role of UDP-GlcA may open a window for early therapeutic interventions. In plants and lower animals, including zebrafish, UDP-GlcA can be synthesized by two alternative pathways. Apart from UGDH, UDP-GlcA can be generated via the myo-inositol oxygenation pathway from glucuronic acid by glucuronokinase and UDP-glucuronic acid pyrophosphorylase 40 . If a similar route exists in humans, supplementation of glucuronate may help to enhance this alternative pathway and increase levels of UDP-GlcA levels and its essential metabolites. To this date, however, the existence of human homologs of glucuronokinase and UDP-glucuronic acid pyrophosphorylase remains to be proven.
Conservative estimates of disease frequency resulting from germline UGDH mutations projects a prevalence of 1:14,000,000 to 1:2,000,000 (Suppl. Note 1). Considering that developmental epileptic encephalopathies are most commonly caused by de novo dominant mutations, this estimated prevalence seems relatively frequent. Fig. 3 Biallelic UGDH mutations behave as hypomorphic alleles. a RT-qPCR (top), western blotting (bottom), and b enzymatic activity, assessed by measuring NADH production (left panel) and quantification of HA (right panel), for endogenous UGDH using patient-derived primary fibroblasts. a, b Control (WT/WT), unaffected mother F5-I:1 (WT/A82T) and 4 (in a) or 3 (in b) different patients' fibroblasts (F5-II:1: A82T/A82T, F3-II:1: R393W/ A410S, F4-II:1: Y14C/S72P, and F6-II:1: R65*/Y367C). a (top) Endogenous UGDH mRNA levels are normalized to β-ACTIN and GAPDH. Fold change relative to control (WT/WT) is plotted. a (bottom) Western blot analysis for endogenous UGDH protein using cellular extracts. GAPDH is used as a loading control. b (left) UGDH enzymatic activity measured as the conversion of NAD + to NADH in whole-cell lysates. b (right) UGDH enzymatic activity measured as the HA production in conditioned media from primary fibroblast cultures. c Western blot analysis for UGDH sensitivity to limited proteolysis using purified WT and mutant (A44V and A82T) UGDH proteins in the absence or presence of its substrates and/or cofactors, as indicated. Results are representative of at least three experimental replicates. d Purified UGDH WT and A44V melting temperature (Tm) in the absence or presence of its substrates and/or cofactors, as indicated. Mean of three experiments ± S.D. is plotted for the T m of each enzyme. e Representative traces at λ = 280 nm of purified WT and mutant UGDH proteins fractionated by size exclusion chromatography. WT, obligate dimer Δ132 15 , obligate hexamer T325D 15 , A44V and A82T UGDH are plotted in the graph. Dashed lines correspond to the known hexamer, dimer and monomer peak elution times. f Purified WT, A44V, and A82T UGDH enzymatic activity measured as the conversion of NAD + to NADH. Asterisks indicate p-values of p < 0.05(*), p < 0.01(**), and p < 0.001 (***), NS: non-significant (p > 0.05) as determined by Student t-test. For gels and graphs source data, please refer to the source data files 1 and 2.

Methods
Ethical approval. Written informed consent was obtained from the parents of the underage patients for diagnostic procedures and next-generation sequencing, as well as for the publication of identifying facial images in Fig. 1b. The study has been approved by the local Institutional Review Board of the Medical Faculty of the University of Tübingen, Germany (vote 180/2010BO1).
Exome sequencing. To unravel the molecular cause of the disease exome sequencing were performed at different genetic institutes using next-generation sequencing techniques according to local standard protocols. Variants were confirmed via Sanger sequencing using standard methods and chemicals (primer sequences are available on request). Family 1: Exome sequencing for two affected siblings was performed on a HiSeq2500 System (Illumina, CA) after enrichment with SureSelectXT Human All Exon V5 (Agilent, Santa Clara CA). FASTQ files were imported into GENESIS (http://thegenesisprojectfoundation.org/) 12 for further analysis using a pipeline build on BWA 41 , Picard, and FreeBayes. Variants were filtered for changes that segregated in an autosomal recessive fashion and passed the following filter criteria: (i) frequency in public databases (ExAC 10 minor allele frequency (MAF) <0.1%), (ii) present in <5 families within GENESIS ( ∼ 4,300 exomes), (iii) conserved (PhyloP 100-way score >2 or PhastCons (100 vertebrate genomes) >0.75), (iv)  Families 2, 9, 10: Using genomic DNA from the proband and parents, the exonic regions and flanking splice junctions of the genome were captured using the Agilent SureSelect Human All Exon V4 (50 Mb) or the Clinical Research Exome kit (Agilent Technologies, Santa Clara, CA). Massively parallel (NextGen) sequencing was done on an Illumina system with 100 bp or greater paired-end reads. Reads were aligned to human genome build GRCh37/UCSC hg19, and analyzed for sequence variants using a custom-developed analysis tool. Additional sequencing technology and variant interpretation protocol has been previously described 42 . The general assertion criteria for variant classification are publicly available on the GeneDx ClinVar submission page (http://www.ncbi.nlm.nih.gov/clinvar/ submitters/26957/). After variant stratification based on population frequencies within an internal database and ExAC 10 , inheritance, in silico predictors such as Provean, Mutation Taster and CADD, GeneDX reported only the UGDH variants to be the best potentially pathogenic candidates and connected to this project via GeneMatcher entries.
Family 3: Samples of the oldest sibling and both parents were sequenced in context of the EUROCORES project EuroEPINOMICS-RES, for which the technical details have been reported before 43 . Briefly, the trio underwent exome sequencing at the Wellcome Trust Sanger Institute (Hinxton/Cambridge, UK). Capturing of the exome was performed using the SureSelect Human All Exon 50 Mb exome kit (Agilent). The enriched exome libraries were then sequenced on a HiSeq2000 platform (Illumina) as 75 bp paired-end reads. BWA was used to align the sequenced reads to the reference genome (hg19). De novo analysis of these data did not reveal any variants. As the younger sib later developed a similar disorder, exome sequencing was also performed locally on the second sibling: for library preparation, genomic DNA was sheared to the average size 150 bp (Covaris) and the genome libraries prepared using KAPA HTP Lib Prep Kit Illumina 96 rxns (07138008001). Exome capturing was performed using the SeqCap EZ Human Exome v3.0 capture system and the sample was sequenced on the NextSeq500 platform using NextSeq500 High-output V2 kit. Mapping to the human reference genome (Hg19) and variant calling were performed with the CLC Genomics Workbench. Subsequent annotation and filtering were executed with GenomeComb 44 (http://genomecomb.sourceforge.net/). Exome sequencing results of the trio and the second sibling were merged and reannotated and the family was reanalyzed as a quartet. Variants were filtered based on following quality parameters: coverage >7, quality >50 and not located in homopolymers >8 or tandem repeats. Only variants present with a frequency <0.5% in control population databases ExAC 10 and Exome Variant Server, seen <3 times in the local exome sequencing database, and with predicted impact on the encoded protein (missense, nonsense, frameshift, deletions, insertions and (essential) splice site) were retained for further analysis. Remaining variants were filtered under an autosomal recessive (homozygous or compound heterozygous) and x-linked hypothesis. In addition, heterozygous variants called in both siblings and absent in the parents were selected under a parental mosaics for which the ExAC 10 filter was set at <2 calls. This analysis revealed only this one compound heterozygous UGDH variants.
Family 4: Parent-proband trio exomes were prepared using the SureSelect Target Enrichment System (Agilent, Santa Clara CA) and sequenced on a HiSeq2000 System (Illumina, CA). Data processing, bioinformatics pipeline (for alignment, variant calling, annotation, and genetic model filtering), and analyses were previously described 45 . The compound heterozygous rare missense alterations in the UGDH gene, c.214T > C and c.41A > G, were interpreted as the only candidate genetic etiology.
Family 5: The exome library was prepared on an ION OneTouch System and sequenced on an Ion Proton instrument (Life Technologies, Carlsbad, CA, USA) using one ION PI chip. Sequence reads were aligned to the human GRCh37/hg19. Variants were filtered for common SNPs using the NCBI's "common and no known medical impacts" database (ClinVar), ExAC 10 , as well as an in-house database of 406 sequenced samples. Additional filters were applied to retain proband's exonic variants that were homozygous while heterozygous in both parents. Out of 5 homozygous variants, only one missense variant c.244G > A in the UGDH gene was found to segregate with the disease.
Family 6: Exome sequencing was performed on a NextSeq 500 System (Illumina, CA USA), with a 2 × 150 bp high-output sequencing kit after enrichment with Seq Cap EZ MedExome kit (Roche, Basel Switzerland). Sequence alignment, variant calling, and variant annotation was performed by Genosplice Technology (Paris France) with BWA 0.7.12, picard-tools-1.121, GenomeAnalysisTK-2014.3-17-g0583018 and SNPEff-4.2 with additional annotations from ClinVar and HGMD. The compound heterozygous UGDH variants were selected to be the most promising candidates and were thus submitted to GeneMatcher.
Families 7, 8, 16-25: Exome sequencing was performed essentially as described before 46 for families 7, 8, and 16-25. Target regions were enriched using the Agilent SureSelectXT Human All Exon 50 Mb Kit. Whole-exome sequencing was performed on the Illumina HiSeq platform (BGI, Copenhagen, Denmark) followed by data processing with BWA (read alignment,) and GATK (variant calling) software packages. Variants were annotated using an in-house developed pipeline. Prioritization of variants was done by an in-house designed "variant interface" and manual curation. As four families with similar phenotype shared the homozygous p.R317Q as best candidate, a GeneMatcher entry was made and the in-house database was systematically screened for other potentially pathogenic UGDH variants. This allowed the identification of families 18 to 25.
Family 11: The sequencing was performed at Claritas Genomics (Cambridge, USA). Extracted genomic DNA was amplified using the AmpliSeq system and sequenced using an IonTorrent Proton Instrument. Alignment and variant calling of the nuclear DNA was done on Proton data using Torrent Suite 4.4 Software. Nuclear variants were filtered for quality using a custom filtering tool. In addition, extracted genomic and mitochondrial DNA was also run on an Agilent Clinical Research Exome capture sequence and then sequenced using an Illumina NextSeq instrument. Alignment and variant calling on NextSeq data was performed by an implementation of GATK Best Practices Pipeline. Genomic DNA results from the two NGS runs on the proband were combined and annotated by a custom bioinformatics pipeline. Besides a heterozygous SLC6A5 missense variant inherited from the unaffected father, the compound heterozygous UGDH variants were the only candidates reported that had a minor allele frequency (MAF) of < or = 0.01% that passed the laboratory's quality metrics and were not de novo, X-linked or had biallelic variants.
Family 12: Isolated genomic DNA from peripheral blood leukocytes of proband and parents was captured with the Agilent Sure Select Clinical Research Exome (CRE) kit (v2). Sequencing was carried out with 150 bp paired-end reads on the Illumina HiSeq 4000. Reads alignments to the GRCh37/UCSC hg19 build were achieved using BWA (BWA-MEM v0.7.13). Variants were called using GATK (v3.7 (reference: http://www.broadinstitute.org/gatk/). Annotated Variants were filtered and prioritized using the Bench lab NGS v5.0.2 platform (Agilent technologies). The full exome analysis revealed the compound heterozygous variants in UGDH. The family was linked to this cohort via GeneMatcher.
Family 13: The exonic regions and flanking splice junctions of the genome were captured using proprietary GeneDx tools. Sequencing was performed on an Illumina system with 100 bp or greater paired-end reads. Reads were aligned to the human genome build GRCh37/UCSC hg19. A custom-developed analysis tool (Xome Analyzer) was used to call sequence variants. Sanger sequencing was used to confirm all potentially pathogenic variants identified in available family members. Additional variants not included in this report are available upon request.
compound heterozygous private missense variants in the UGDH gene, c.347T > C and c.1328G > A, were interpreted as the only candidate genetic etiology.
Family 15: genomic DNA from of the proband and parents were enriched for exonic sequences with the SureSelect Human All Exon 50 Mb V5 Kit (Agilent Technologies, Santa Clara, California, USA). The HiSeq2500 (Illumina, San Diego, California, USA) was used to generate 125-bp paired-end runs of sequences. Reads were aligned and variant called with DNAnexus (Palo Alto, California, USA) using the reference human genome assembly hg19 (GRCh37). A mean coverage of 104x was achieved for the proband. Data analysis was preformed using an in-house bioinformatics pipeline. The compound heterozygous UGDH variants were selected to be the most promising candidates and were thus submitted to GeneMatcher.
Brain magnetic resonance imaging. Magnetic resonance images (MRI) have been recorded on 1.5 or 3 Tesla scanners at the different clinical sites. Sagittal, transversal, and coronal images of the brain have been acquired with standard sequences, including T1, T2, and Flair images.
Cell culture. Primary dermal fibroblast cultures were established from skin biopsies obtained from individuals F3-II:1, F4-II:1, F5-I:1, and F5-II:1 according to standard procedures 55 . In brief, primary fibroblasts were derived from biopsy samples and cultured in Dulbecco's modified Eagle medium (DMEM; HyClone, SH30243.01) supplemented with 10% fetal bovine serum (Biological Industries) and 2 mM L-glutamine (Biological Industries). Written informed consent of healthy probands and parents of UGDH patients were received prior to biopsy according to the ethical approvals of the local Institutional Review Boards (IRB).
HA quantification. HA content was compared in the culture-conditioned media from fibroblasts expressing WT or mutant UGDH using a competitive binding assay as previously described 19 . Fibroblasts were grown in 10 cm plates, 1 mL of conditioned media was aspirated from technical replicates, and cells were counted. HA concentration was interpolated from a standard curve, normalized to cell number, and plotted as mean ± SD. Statistical significance was assessed by Student's t-test with at least three technical triplicates.
Specific activity measurement of UGDH. Fibroblasts expressing WT or mutant UGDH were assayed for UGDH-specific activity essentially as previously described 56 . Fibroblasts were cultured in 15 cm plates, washed three times with cold 1x PBS, and centrifuged at 1500 rpm for 5 min. Cells were resuspended in twice the pellet volume of Lysis Buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, and protease inhibitor cocktail). Samples were transferred to tubes with an equal volume of acid washed glass beads (Sigma) and lysed in the Bullet Blender 24 (Next Advance) at speed 8 for 3 min. The resulting supernatant was centrifuged at 13,000 rpm for 15 min to obtain final lysates. Enzymatic activity of the lysates (50 µg) was assayed with 1 mM UDP-glucose and 1 mM NAD + in the presence or absence of 1 mM UDP-xylose, a UGDH-specific inhibitor, and monitored for changes in NADH, A 340 . Reaction rates for samples containing UDP-Xyl were subtracted from samples without UDP-Xyl to obtain UGDH-specific activity reported as [NADH] in nmol min −1 mg −1 lysate as described above. Each fibroblast cell line analyzed contained three or more technical replicates for reactions with and without UDP-Xyl plotted as mean ± SD. Statistical significance was assessed by one-way ANOVA (Prism).
Generation and purification of UGDH point mutants. Point mutants of human UGDH were generated from the codon optimized E. coli expression construct, WT-UGDH pET28a, using polymerase chain reaction mutagenesis with appropriate primers as previously described 19,57 . Sequences were verified by Eurofins MWG Operon (Huntsville, AL). The UGDH mutant constructs were expressed in E. coli strain BL21(DE3) grown in 2xYT medium containing 50 mg L −1 kanamycin at 37°C. At an OD 600 of 0.6-0.8, protein expression was induced with the addition of IPTG at a final concentration of 0.5 mM, and cultures were incubated at 18°C overnight. Cells were harvested by centrifugation and lysed by sonication. All UGDH point mutants were expressed in the soluble fraction, and enzymes were purified by affinity chromatography using a HisTrap FF column (GE Healthcare) according to the manufacturer's protocol. The average protein yields were:~20 mg L −1 for UGDH WT and T325D,~1.5 mg L −1 for UGDH A82T, and~6 mg/L for UGDH A44V. Purified protein was dialyzed against 20 mM Tris-HCl pH 7.4 containing 1 mM dithiothreitol (DTT), concentrated, flash frozen in liquid nitrogen, and stored at -80°C.
Analytical gel filtration. Purified recombinant UGDH WT and all point mutants were analyzed by size exclusion chromatography as previously described 57 . All samples were centrifuged prior to loading. Each apoprotein sample was injected into a 250 µL loop and separated by FPLC in 1x PBS containing 1 mM DTT at a flow rate of 0.5 mL min −1 on a Superdex 300 10/200 GL gel filtration column (GE Healthcare). Elution was monitored by A 280 and plotted to compare alterations in oligomeric state.
Trypsin susceptibility assay. Purified recombinant WT-UGDH and all point mutants were assessed by limited trypsin proteolysis as previously described 19 . UGDH WT, A44V (10 µg), and A82T (14.2 µg) were digested with 10 ng trypsin in 1x PBS pH 7.4 for 2.5 h at room temperature in the absence or presence of 1 mM UDP-glucose, 1 mM UDP-glucuronate, 5 mM NAD + , 5 mM NADH, or combinations that yielded abortive and productive ternary complexes. Samples were analyzed by western blot probed for UGDH as previously described 58 .
Thermal stability measurement. Recombinant UGDH WT and A44V protein were assessed for thermal stability as previously described 19 with minor alterations. All samples of UGDH WT and A44V (~15 µg) were incubated in 1x PBS and Sypro Orange dye (Invitrogen; 1:500 dilution) in the absence or presence of 1 mM UDPglucose, 1 mM UDP-glucuronate, 5 mM NAD + , 5 mM NADH, or in combinations that yielded abortive and productive ternary complexes. Samples were handled at room temperature and transferred to an iCycler MyiQ thermocycler (Bio-Rad) for incremental thermal denaturation. T m was plotted as the mean ± SD for seven replicates. Statistical significance was assessed by two-way ANOVA (Prism).
Saturating enzymatic activity and kinetic characterization. Enzyme activity of recombinant UGDH WT and all point mutants was characterized as described previously 57 with minor alterations. Enzymatic activity was calculated by NADH turnover using the NADH extinction coefficient of 6220 M −1 cm −1 . UGDH A82T activity was converted to [NADH] in nmol min −1 mg −1 UGDH and subsequently normalized to the fractional purity of UGDH in the sample preparation. Samples were run in triplicate and statistical significance was determined using Student's t-test. Michaelis rate constants, K m and V max , were determined for UGDH WT and A44V as previously described using a 96-well plate assay to measure the change in NADH (A 340 ) with respect to both the substrate, UDP-glucose, and cofactor, NAD + .
Multi-electrode Array (MEA) recordings. Neurons on day 21 were dissociated and replated on 0.1 polyethylenimine (Sigma)-coated 48-well MEA plates (Axion Biosystems) in BrainPhys media supplemented with BDNF, GDNF, cAMP, and ascorbic acid as previously described 62 . Spontaneous neuronal activity was observed and recorded at 37 o C for 5 min every 2-3 days using the Maestro MEA System (Axion Biosystem).
Zebrafish strains and maintenance. All zebrafish husbandry procedures were performed in compliance with the Singapore National Advisory Committee on Laboratory Animal Research Guidelines from the Institutional Animal Care and Use Committee (IACUC, IACUC number 161172). The jek ugdh m151 mutant line 63 was used for this study and was maintained in the AB background. All embryos were raised in egg water (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl 2 , 0.33 mM MgSO 4 ).
Genomic DNA extraction. Genomic DNA from larvae used for the locomotion assay was extracted by larval fin clipping as previously described 64 . Briefly, microscopic caudal fin slices of 3-dpf larvae were sectioned under a stereomicroscope. The fins were then digested with proteinase K (Invitrogen, 25530049) overnight at 65°C, and the supernatant was used as a source of genomic DNA for genotyping. In 7-dpf larvae, heads were used for c-fos qPCR analysis, while the rest of the bodies was used for genotyping.
Genotyping of zebrafish. Genotyping was performed via nested PCR to amplify the exon 7 of ugdh, and restriction fragment length polymorphism as described previously 63 . The first PCR was performed with the following primers, 5′-TGTATGTGCAGGTGATTGACA-3′ and 5′-TGTAGGTCACAGGTTTTTGACA-3′. PCR products were cleaned-up by Exonuclease I (New England Biolabs, M0293L) and FastAP Thermosensitive Alkaline Phosphatase (Thermo Fisher, EF0651) treatment, and were then used as templates for a second PCR with the following primers: 5′-GACATGAATGAATATCAGAGAAAGAG-3′ and 5′-AGGAGAAACCCAACAACGC-3′. These PCR products were digested with the MluI enzyme (New England Biolabs, R3198L) that only cuts when the mutation is present.
Twenty-four-well locomotion and convulsion test assay. A Basler Ace (acA1300-200um; 1280 × 1024) camera was used to acquire videos of larval zebrafish at 50 fps. A custom hardware setup was designed to acquire full frame videos of 24-well flat bottom plates using a 25 mm lens attachment to the camera placed 65 cm above the plate. The 24-well plates were backlit using a white light LED lightbox that delivered uniform lighting across the entire field. 7-dpf larval fish were placed individually in each well in 500 µl of egg water and acclimated to the setup for 10 min. Then, 500 µl of either egg water (negative control) or PTZ 30 mM (15 mM final concentration, Sigma, P6500) was added to the well. Three videos of 2 min each were acquired for a duration of 10 min for each condition tested. Fish locomotion was tracked online during video recording on a custom written software in LabView (www.critta.org). Analysis of locomotion was automated and performed blind offline after each experiment using custom written Python scripts (available on Github: https://github.com/mechunderlyingbehavior/24-Well-Larval-Locomotion.git). In this assay, low and high speed were defined as average speeds of 0 to 8 mm/s and >8 mm s −1 , respectively, based on a recent publication measuring convulsive-induced speed changes in zebrafish larvae 65 .

Data availability
The data that support the findings of this study are available within the paper and its supplementary information files. All identified variants have been deposited in the ClinVar database (https://www.ncbi.nlm.nih.gov/clinvar/) under the name "UGDH001". Whole dataset from Family F3 was already published 66 and deposited in the European Genome-phenome Archive, accession numbers EGAS00001000190, EGAS00001000386, and EGAS00001000048. Whole datasets from Families F5 and F15 are available upon request from the corresponding authors. Consent restrictions preclude deposition of sequencing data of the other families, however, specific information (e.g., secondary variants etc., but not full datasets) can be obtained upon request from the corresponding authors. Lists of primers and antibodies are in Supplementary Tables 4 and 5, respectively. The source data underlying Fig. 3a-f, Fig. 4a, and c, Suppl. Fig. 3a-d, Suppl. Fig. 4a, b, and d are provided as a Source Data file 1 (for gels) and 2 (for graphs).

Code availability
The custom written Python scripts used to track fish locomotion is available following this link: https://github.com/mechunderlyingbehavior/24-Well-Larval-Locomotion.git.