CERT1 mutations perturb human development by disrupting sphingolipid homeostasis

Neural differentiation, synaptic transmission, and action potential propagation depend on membrane sphingolipids, whose metabolism is tightly regulated. Mutations in the ceramide transporter CERT (CERT1), which is involved in sphingolipid biosynthesis, are associated with intellectual disability, but the pathogenic mechanism remains obscure. Here, we characterize 31 individuals with de novo missense variants in CERT1. Several variants fall into a previously uncharacterized dimeric helical domain that enables CERT homeostatic inactivation, without which sphingolipid production goes unchecked. The clinical severity reflects the degree to which CERT autoregulation is disrupted, and inhibiting CERT pharmacologically corrects morphological and motor abnormalities in a Drosophila model of the disease, which we call ceramide transporter (CerTra) syndrome. These findings uncover a central role for CERT autoregulation in the control of sphingolipid biosynthetic flux, provide unexpected insight into the structural organization of CERT, and suggest a possible therapeutic approach for patients with CerTra syndrome.

degrees of intellectual disability and motor and speech delays; increased pain tolerance; and seizures. We investigated the effect of Cer transporter (CerTra) mutations on CERT function, regulation, and structure. We found that several CerTra mutations disrupt CERT autoregulation, leading to a gain in CERT activity, increased de novo sphingolipid synthesis in the ER, and skewed metabolic flux toward the production of potentially neurotoxic compounds. We found that CERT gain of function in Drosophila melanogaster led to head and brain size defects and impaired locomotor activity, which we corrected by pharmacological inhibition of CERT. Biochemical characterization of disease-causing CERT1 mutations led us to identify an unanticipated structured region within CERT that is essential to its autoregulation and to sphingolipid homeostasis.

Results
Characterization of CERT1-associated phenotypes Most of the CERT1 variants reported so far have been associated with intellectual disability (13) (Figure 1A, and Supplemental Table  1; supplemental material available online with this article; https:// doi.org/10.1172/JCI165019DS1), but the clinical phenotype has been characterized for only 1 of these individuals, who carries a p.S135P variant (13). We therefore sought CERT1 mutation carriers from multiple international disease consortia and databases (see Methods and Supplemental Methods). We identified 50 patients who carry a potentially pathogenic variant in CERT1 (45 in the coding sequence; hg19:NM_005713.3, Figure 1A, and Supplemental Tables 1 and 2) (17). We obtained thorough clinical histories for 31 patients: 27 from the present cohort, 3 who were initially included in genetic screening consortia for intellectual disability (8,10,12), and the patient who had been characterized clinically (13) (Supplemental Tables 2 and 3, and Supplemental Clinical Appendix for each patient). Family segregation confirmed that CERT1 variants occurred de novo in 93% (25 of 27) of patients. Biparental samples were not available for 4 of the patients (referred to herein as subject 1 [S1], S19, S20, and S24; see Supplemental Clinical Appen-

Introduction
Sphingolipids play numerous essential roles in membrane structure, signal transduction, and brain development and function (1)(2)(3). The central nervous system is particularly affected by disturbances in sphingolipid production or clearance: defective production can cause hereditary sensory neuropathy, spastic paraplegia, or infantile epilepsy syndrome, whereas toxic accumulation of sphingolipids underlies a number of devastating inborn errors of metabolism such as Gaucher disease, Farber disease, Niemann-Pick disease type A, Krabbe, Tay-Sachs, and Sandhoff diseases (1,4). Sphingolipid metabolic fluxes therefore must be tightly regulated through homeostatic circuits (5). A key checkpoint in sphingolipid biosynthesis occurs at contact sites between the endoplasmic reticulum (ER) and the trans Golgi membrane, where the ceramide transporter (CERT) transfers ceramide (Cer) from the ER to the trans Golgi for its conversion to sphingomyelin (SM) (6). When sufficient rates of SM production are reached, CERT is phosphorylated and undergoes a conformational change that renders it inactive (7).
Given the central position of CERT in sphingolipid metabolism, its malfunction should be detrimental to human health and particularly to neural function. Thus far, 1 case report and several case-control screening studies have described associations between variants in ceramide transporter 1 (CERT1), the gene that encodes CERT, and neurological abnormalities (8)(9)(10)(11)(12)(13)(14)(15)(16). Nonetheless, there has not been a systematic assessment of the mutational landscape of CERT1 in humans, and whether or how CERT1 mutations cause neurological disease remains to be proven.
In this study, we characterized 31 unrelated individuals with 22 distinct missense variants in CERT1, 18 of which have not to our knowledge been previously reported. These patients have a syndromic presentation characterized by infantile hypotonia; mild dysmorphologies (affecting the face, hands or feet); variable Neural differentiation, synaptic transmission, and action potential propagation depend on membrane sphingolipids, whose metabolism is tightly regulated. Mutations in the ceramide transporter CERT (CERT1), which is involved in sphingolipid biosynthesis, are associated with intellectual disability, but the pathogenic mechanism remains obscure. Here, we characterize 31 individuals with de novo missense variants in CERT1. Several variants fall into a previously uncharacterized dimeric helical domain that enables CERT homeostatic inactivation, without which sphingolipid production goes unchecked. The clinical severity reflects the degree to which CERT autoregulation is disrupted, and inhibiting CERT pharmacologically corrects morphological and motor abnormalities in a Drosophila model of the disease, which we call ceramide transporter (CerTra) syndrome. These findings uncover a central role for CERT autoregulation in the control of sphingolipid biosynthetic flux, provide unexpected insight into the structural organization of CERT, and suggest a possible therapeutic approach for patients with CerTra syndrome. The SRR is the target of protein kinase D (PKD) and casein kinase 1γ2 (CSNK1G2) phosphorylation. The FFAT (2 phenylalanines in an acidic tract) motif interacts with the integral membrane proteins VAP-A and VAP-B on the ER (70), and a C-terminal START-related domain extracts Cer from the ER membrane and delivers it to the trans Golgi (7). The schematic shows coding variants in CERT1 (NP_005704) in our cohort of 31 individuals above the gene diagram and other individuals identified from clinical databases (DECIPHER, version 9.31, ClinVar, and VKGL) below it (Supplemental Table 1). Colors indicate the age of onset; gray indicates that no information is available. The distribution of gnomAD singleton missense variants for healthy individuals is plotted below. (B) Range of severity in motor delays compared with the 75th percentile (light gray) and 25th percentile (dark gray), adapted from values published by the Denver Developmental Screening Test II. White and black circles indicate delayed sitting and walking ages, respectively; asterisks indicate that the individual needs sitting or walking support. White circles with an arrow indicate individuals who are currently immobile or have not yet developed independent walking. (C) Heatmap shows the degree of intellectual disability (ID), speech delay (SD), and motor delay (MD) of the patients bearing frequent CERT1 mutations. See Supplemental Table 2 for scores. C1, cluster 1; C2, cluster 2; C3, cluster 3; C4, cluster 4. used CERT1 isoform 1 (also known as CERT L ) for both aa numbering, overexpression, and structural experiments.
For the variants that recurred in multiple individuals, we were able to sketch out broad genotype-phenotype correlations. The most severe phenotypes -with congenital or perinatal onset, profound-to-severe intellectual disability, and the greatest motor delay -resulted from mutations at p.S132 and p.S135 (Figure 1, B and C, and Table 1). Individuals bearing mutations at p.S138, p.T166, or p.G243 tended to not have perinatal difficulties and achieved early developmental milestones (S8 did not show difficulties until 4 years of age) but then regressed or slowed in their development. Seizures probably contribute to the developmental delays: neither individual with a p.S138C mutation had seizures and have only moderate intellectual disability, but S13, the most severely affected of the p.T166A carriers, has an epileptic encephalopathy that apparently halted her development at 19 months of age. All variants associated with more severe phenotypes were predicted to be deleterious by Combined Annotation-Dependent Depletion, version 1.6 (CADD v1.6), Rare Exome Variant Ensemble Learner (REVEL), Mendelian Clinically Applicable Pathogenicity (M-CAP), and Eigen (Supplemental Table 4). Approximately half of the variants among the more moderately affected individuals had inconsistent pathogenicity predictions across algorithms. On average, variants among patients had significantly greater CADD v1.6, M-CAP, REVEL, and Eigen pathogenicity scores than did Genome Aggregation Database (gnomAD) singleton missense variants (P < 0.001, Mann-Whitney U test).
Variants p.D59E (15), p.T166A (19), and p.F182L (9) were previously identified in ASD cohorts (Supplemental Table 1). To determine the extent to which CERT1 mutations contribute to autism, revealed cognitive, motor, and speech delays of variable degrees of severity. Of the patients for whom we had birth information, only 4 were of average weight at birth; most were born slightly to significantly underweight. Similarly, only 4 of the individuals did not show some form of developmental delay by the end of the first year of life (4 of 26, 15%), with the latest onset being at age 4 years ( Figure 1, A-C). Fifteen (of 24) patients had neonatal feeding difficulties, often with hypotonia or failure to thrive. These were likely early manifestations of what would later become frank motor delays, affecting 26 of 29 patients ( Figure 1B). Intellectual disability ranged from mild to profound, as per the criteria of the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5) (18) (see Methods); the latter individuals are nonverbal, lack age-appropriate daily living skills, and require safety supervision. Neurobehavioral abnormalities frequently led to a diagnosis of autism spectrum disorder (ASD) (19 of 27, 70%); some patients displayed stereotypical hand movements (14 of 18), self-injurious behavior (9 of 19), high pain tolerance (9 of 18), disrupted sleep patterns (9 of 21), attention deficit-hyperactivity disorder (10 of 19), or aggression (6 of 20). Multiple seizure types were reported (16 of 29). Neuroimaging frequently revealed a thin corpus callosum, ventriculomegaly, delayed myelination, and cerebellar atrophy (Supplemental Figure 1B).
Subtle facial dysmorphisms included anteverted nares with a depressed or broad nasal bridge, enlarged earlobes, synophrys, micrognathia, dental anomalies (protruding incisors and diastema), and palatine ridges (Supplemental Figure 1A and Supplemental Figure 2, A-E). Anomalies affecting the hands, feet, or digits included third/fourth finger syndactyly, club foot, or hallux varus (sandal gaps); the first metatarsal also tended to be short, whereas the fifth fingers tended to be long (Supplemental Figure 2F).
The majority (27 of 31, 87%) of variants from our enrolled patients occurred in the region between the pleckstrin homology (PH) and C-terminal START-related (START) domain ( Figure 1A), as was the case for previously reported variants (13). These variants populated 4 distinct subregions, whereby the geometric mean distance (δg) between variants within each spatial group indicated a greater likelihood of clustering compared with random permutations ( Figure 1A). Human CERT1 produces at least 2 splicing variants that are both widely expressed (13). In this report, we have Table 1. Phenotype-genotype correlations for variants for which there was more than 1 study participant

Variant
Key features S132L (n = 4, S2-S5) Small at birth, perinatal difficulties, failure to thrive. Achieve sitting at 3-4 yr of age; may walk, supported, by 7 or 8 yr of age but may become immobile by late adolescence. Seizures severe enough to require hospitalization. Profound to severe ID. S135P (n = 2, S6-S7) Somewhat small at birth, perinatal difficulties, failure to thrive. Achieve sitting between 19 mo and 4 yr of age. Seizures may occur after infancy. Severe ID. S138C (n = 2, S8-S9) Normal to slightly low birth weight, may have feeding difficulties. Mild motor delay (may sit at 10 mo), mild speech delay. No seizures. Moderate ID.
T166A (n = 4, S10-S13) Normal birth weight. Mild motor delay at 1 year of age, mild speech delay (can use complete sentences at 5 years ). Seizures likely. May have moderate, mild, or no notable ID. Exception: Patient 13 has severe epileptic encephalopathy and severe ID.
G243R (n = 2, S14-S15) Normal birth weight, no neonatal difficulties, but 1 patient was noted to have developmental delay by the age of 4 mo (patient 14). Acquired growth delay; development slows or regresses, may become nonverbal. Seizures. Severe ID.
Although it is difficult to draw firm conclusions from 2 to 4 patients, we can sketch the main features that are associated with each variant thus far on the basis of available information. See the Clinical Appendix and Supplemental Table 2  Although we expected hypophosphorylation and increased membrane recruitment for CERT mutants within the SRR (cluster 1) (13,27), this was not intuitive for clusters 2 and 3. We therefore investigated the regulation of the 2 most frequent CERT1 mutations in these clusters, p.G243R, for which some biochemical characterization was already reported (30), and p.T166A. These 2 variants were effectively phosphorylated at p.S132 by PKD in both in vitro and cell-based assays (Supplemental Figure 3, D and E). Moreover, the p.S132 phosphomimetic mutant (HGS 130-132 DDD) (27) in the p.G243R and p.T166A backgrounds did not differ in either localization or SDS-PAGE migration (Supplemental Figure 3, F-H). This supports the idea that the cluster 2 and 3 mutations affect CERT regulation downstream of PKD phosphorylation. To test this possibility, we coexpressed CSNK1G2 or PP2Cε with the hypophosphorylated CERT mutants. Overexpression of CSNK1G2 led to reduced membrane association of most mutants (p.S132L served as a negative control) ( Figure 3, A-C), indicating that these can be phosphorylated and inactivated by CSNK1G2. Interestingly, although PP2Cε overexpression had little effect on WT CERT phosphorylation or localization, consistent with previous results (27), hypophosphorylated CERT mutants were sensitive to PP2Cε overexpression ( Figure 3, A-C).
Thus, CerTra-associated CERT1 mutations generally disturb CERT regulation and promote its membrane association. Mutations in clusters 1, 2, and 3 also reduce the pool of hyperphosphorylated/inactive CERT by influencing the CERT phosphorylation cycle that depends on the consecutive action of PKD and CSNK1G2 kinases and on the PP2Cε phosphatase.

CerTra mutations disrupt sphingolipid homeostasis
Sphingolipid biosynthesis starts at the ER with the generation of Cer (32)(33)(34). Cer is then transported to the Golgi complex by vesicular carriers for the production of glucosylceramide and downstream glycosphingolipids, or transferred directly to the trans Golgi by CERT for its conversion to SM (6) ( Figure 4A).
We profiled sphingolipids in CERT1-deficient (CERT1-KO) HeLa cells expressing 6 CERT mutants tagged with GFP. CERT1 KO causes a defect in Cer-to-SM conversion (35) that is expected to be reverted by the expression of active CERT. Overexpression of WT CERT in CERT1-KO cells, in fact, induced a 5-fold increase in SM levels coupled with an unexpected increase in Cer and complex glycosphingolipids ( Figure 4B and Supplemental Figure  4A). Overexpression of CERT mutants also increased SM and Cer levels but markedly reduced glycosphingolipid levels compared with WT levels. Unexpectedly, CERT overexpression increased dihydroceramide (dhCer) and dihydrosphingomyelin (dhSM), and this was even more pronounced with mutant overexpression ( Figure 4B and Supplemental Figure 4A). Variants associated with more severe manifestations in patients (i.e., at p.S132, p.S135, and p.G243) were associated with the greatest increases.
We observed that CERT1 overexpression produced an overall increase in sphingolipid levels ( Figure 4B). To determine whether this was due to increased synthesis, decreased catabolism, or both, we used a D 3 -15 N-serine labeling approach (see Methods). Lipids were extracted and subjected to chemical hydrolysis to release long-chain bases (LCBs) from sphingolipids and dihydrosphingolipids. These LCBs reflect total levels of steady-state we conducted a targeted de novo analysis interrogating the Simons Foundation Powering Autism Research (SPARK) initiative database, which includes genomic data from nearly 35,000 individuals with ASD (See Supplemental Methods). No de novo protein-altering variants were observed in the Simons exome cohort. Denovo-db (version 1.6.1) confirmed the negative findings for the exome cohort but identified 1 de novo genome missense mutation (pos=74721285, T>C, c.880A>G, p.(T294A), exon6) in 1 of 516 trios. This analysis, along with the fact that none of the individuals in our cohort was diagnosed with ASD as a primary condition, indicates that CERT1 variants are unlikely to be a significant contributor to autism but instead cause a recognizable neurodevelopmental syndrome distinct from ASD, which we will refer to as CerTra syndrome.
CerTra mutations disrupt CERT regulation CERT oscillates between cytosolic and membrane-associated forms (20), according to its activation state in the homeostatic cycle (7). Active, membrane-associated CERT provides Cer to SM synthase 1 (SMS1) for its conversion to SM, with concomitant production of diacylglycerol. The diacylglycerol produced in the SM synthase 1 reaction triggers a signaling response involving the recruitment of protein kinase D (PKD) to the trans Golgi (21-23). PKD initiates CERT phosphorylation at p.S132 (7); and CSNK1G2 (casein kinase 1 γ 1) then phosphorylates the rest of the SRR phosphosites (24)(25)(26)(27). Following phosphorylation, CERT undergoes a conformational change that leads to its cytosolic redistribution and inactivation (20,26,28); all this can then be reversed by the action of the ER phosphatase PP2Cε (protein phosphatase 2C epsilon) (29). Mutations leading to abnormally activated CERT redistribute the protein to Golgi membranes and cytosolic spots, while inactivating mutations redistribute it to the cytosol (30,31).
To determine whether disease-causing variants alter CERT localization, we used automated microscopy to capture GFPtagged WT and CerTra-associated CERT variants in transfected HeLa cells (see Supplemental Methods). Eighteen of the 22 CERT mutants exhibited a greater tendency than WT to associate with the perinuclear membranes and cytosolic spots (Figure 2, A and B). These included all variants mapping to clusters 1 to 4 ( Figure  1A), those located between clusters 3 and 4, the p.S93R in the PH domain, and the p.500L in the START domain. We observed a similar phenotype in patient-derived fibroblasts bearing the p.G243R mutation (Supplemental Figure 3, A-C). Four consecutive variants (p.R366T, p.I382V, p.E424G, p.A449V) did not show a different distribution compared with CERT WT.
We next evaluated the phosphorylation pattern of CERT WT and mutants by SDS-PAGE ( Figure 2C). WT CERT migrates on SDS gels as a doublet, with the fast-and slow-migrating bands corresponding to the active/hypophosphorylated and the inactive/ hyperphosphorylated forms of CERT, respectively (24). By contrast, all variants belonging to clusters 1, 2, and 3 showed hypophosphorylation, whereas mutations outside of these clusters did not. Mass spectrometry-based assessment of selected variants belonging to clusters 1, 2, and 3 showed that these CERT1 mutations did not affect CERT phosphorylation outside the SRR or SRR monophosphorylation (likely on p.S132), and suggests that CSNK1G2-mediated SSR phosphorylation is defective in these mutants (Supplemental Figure 3D). and newly synthesized sphingolipids, represented by sphingosine (SO) and sphingosine+3 (SO+3), respectively, while steady-state and newly synthesized dihydrosphingolipids are represented by sphinganine (SA) and sphinganine+3 (SA+3), respectively. Overexpression of CERT mutants had little effect on SO but increased SO+3 by 200%. All mutants had twice the amount of SA compared with WT CERT, but SA+3 levels rose at least 300% in the case of the mildest mutant, p.T166A, and considerably more for the other mutants ( Figure 4C).
These data indicate that CERT activity promotes de novo sphingolipid production, possibly by relaxing the Cer-dependent inhibition of serine palmitoyltransferase (SPT) (36) (Supplemental Figure 5A). To test this hypothesis, we repeated our metabolic labeling experiment in parental and CERT1-KO HeLa cells treated with the CERT inhibitor N-[(1R,3S)-3-hydroxy-1-(hydroxymethyl)-3-phenylpropyl]-dodecanamide (HPA-12) (37) or, as a control, with myriocin (Myr), an inhibitor of SPT and therefore of sphingolipid production (Supplemental Figure 4A). Treatment with HPA- 12 reduced sphingolipid synthesis more than 80% (similar to what was obtained by treating cells with Myr) in HeLa cells but had no effect on CERT1-KO cells, in which sphingolipid synthesis was already greatly reduced (Supplemental Figure 5B).
We confirmed the metabolic alterations induced by overexpression of CERT1 mutants in CERT1-KO cells in patient-derived fibroblasts bearing the p.G243R mutation and lymphoblasts bearing the p.T166A mutation. Patient cells produced substantially higher amounts of dhSM and dhCer than did their WT counterparts (Supplemental Figure 5, C and E). In contrast, inhibition of CERT activity with HPA-12 markedly reduced total dhCer and dhSM in both WT and patient-derived cells.
We observed a similar profile in steady-state sphingolipid levels and new synthesis when we examined patient cells after labeling with D 3 -15 N-serine (Supplemental Figure 5, D and F). SA+3 backbone levels were much higher in patient-derived cells than in controls. The increase in sphingolipid synthesis observed in patients' cells was blocked by the administration of HPA-12, confirming that these effects are indeed caused by altered CERT activity (Supplemental Figures 4 and 5).
These data indicate that CERT activity is inherently coupled to the homeostatic regulation of de novo sphingolipid synthesis and suggest that frequent CerTra mutations cause an increase in CERT activity ( Figure 4D).

CerTra mutations reveal an uncharacterized structural domain in CERT
The structures of the PH and START domains of CERT have been solved independently and as a complex (20,38,39), but little is known about the region between these 2 domains. It has been assumed to be largely unstructured except for a predicted coiled-coil segment that mediates CERT oligomerization upon UV irradiation (40,41). Given the appreciable effect of CERT mutations in clusters 2 and 3 ( Figure 1C), we decided to explore the structural context of these clusters. To this end, we subjected recombinant CERT to hydrogen-deuterium (H/D) exchange mass spectrometry (HDX-MS), which allows the study of protein dynamics and folding by assessing the solvent accessibility of protein surfaces (42,43).
As expected, the HDX-MS profile of CERT highlighted the presence of amides protected from H/D exchange at early time points of exchange at the N-and C-terminal regions, corresponding to the PH (aa 23-117) and START (aa 389-618) globular domains ( Figure 5, A and B). The areas from aa 187 to 218 and from aa 300 to 381 were highly accessible to solvent and  Figure 1). kDa, observed MW of 37 kDa) ( Figure 5E). When we evaluated the stoichiometry of recombinant full-length CERT with size-exclusion chromatography coupled to multiangle light scattering (SEC-MALS) (48), we found the absolute molar mass of the full-length CERT molecule to be, on average, 142.7 kDa, which, again, is compatible with a dimeric form (Supplemental Figure 6, C-F).
We therefore asked whether CERT dimerizes through its CCD domain. SEC-MALS analysis of the START domain alone (aa 389-618) indicated that it is a monomer, an oligomeric state also shared by the PH domain alone (26). By contrast, the construct encompassing the PH plus CCD domains (aa 1-341) was dimeric. Thus, although a recent study reported that CERT is organized as a homotrimer or a higher oligomer (49), our data strongly suggest that CERT dimerizes through its CCD (Supplemental Figure 6, D-F).
We then modeled the CCD dimer by rigid docking (see Supplemental Methods). Of the 6 best dimer models obtained, 5 had an antiparallel orientation (Supplemental Figure 6G, Supplemental Data Sets 1-7). Antiparallel orientation was also supported by coevolution restraints unexplained by the 3D model of the monomer ( Figure 5F). However, not all predicted contacts are explained by a unique conformation, perhaps because the dimer is highly therefore probably contained no or a very dynamic secondary structure. Strikingly, regions encompassing cluster 2 and 3 mutations (i.e., aa 152-187 and aa 240-300) were heavily protected from exchange, indicating secondary structure formation and/or their engagement in protein-protein interactions (i.e., CERT oligomerization).
by the CCD enables the conformational change that takes place upon CERT phosphorylation to deactivate the protein and signal that sufficient levels of SM production have been reached. Cer-Tra-causing CERT1 mutations in the CCD disturb this conformational change either by altering CCD stability (p.T166A) or CCD dynamics/interactions (p.G243R).
CERT gain of function alters sphingolipid metabolism, brain size, and locomotion in Drosophila melanogaster Altogether, our data indicate that CerTra mutations disrupt CERT autoregulation, resulting in an inappropriate gain of CERT function and excessive sphingolipid production. To test whether this is sufficient to induce neurological manifestations in an animal model, we increased the CERT1 dosage in D. melanogaster. The metabolic activity of D. melanogaster CERT (dCERT) is similar to that of mammalian CERT (50) and shares 43% sequence identity with human CERT, including all 6 of the most recurrent CerTra variants. We generated D. melanogaster strains on the w 1118 background (hereafter referred to as control [Ctrl]) bearing 2 extra copies of either dCERT WT (hereafter referred to as +WT) or dCERT p.S149L (hereafter referred to as +SL; equivalent to the most common and severe CerTra mutation, p.S132L) under the control of its endogenous promoter ( Figure 6A; see also Supplemental Methods).
Total dCERT mRNA levels were 50% higher than those in Ctrl flies in both +WT and +SL strains, with the increase being completely ascribable to the extra exogenous copies of dCERT ( Figure 6B). The increased gene dosage did not shorten the lifespan of transgenic Drosophila; in fact, both the male and female animals lived slightly longer (Supplemental Figure 7A). The transgenic strains had wing and abdomen sizes comparable to those of Ctrl flies (Supplemental Figure 7B), but smaller heads and brains (Figure 6, C-G, and Supplemental Figure 7, C and D). Feeding HPA-12 to +WT and +SL larvae (see Supplemental Methods) restored head size, suggesting that increased dCERT activity was responsible for this phenotype ( Figure 6F). The levels of Cer and dhCer were significantly and similarly increased in both +WT and +SL strains. Cer-phosphoethanolamine (CPE, the Drosophila equivalent of mammalian SM) (51) and dihydroceraminde (dhCPE) levels were also increased in both strains, but more substantially in the +SL strain ( Figure 6H and Supplemental Figure 7, E-G). These changes were similar to those observed in whole Drosophila bodies (Supplemental Figure 7E) and resembled those we observed in mammalian cells ( Figure  2B), suggesting that dCERT gain of function affects sphingolipid metabolism in flies in a manner similar to the effects of CerTra mutations in humans. It is worth noting that glycosphingolipid levels were not significantly altered in +WT or +SL strains, suggesting that changes in Cer, CPE, and their dihydro counterparts were sufficient to induce neurological phenotypes in these strains (Supplemental Figure 7E).
As most patients with CerTra experience motor delay, we monitored the locomotor activity of the +WT and +SL flies (52). Both +WT and +SL flies showed locomotor hypoactivity compared with Ctrl flies ( Figure 6I). Pretreatment with HPA-12 (10 μM) for 7 days, which reduced sphingolipid levels (Supplemental Figure 7H), rescued this phenotype in the transgenic lines but had little effect on Ctrl flies ( Figure 6J). Moreover, HPA-12 treatment had no effect on dynamic. Nonetheless, when CERT mutations were mapped onto the highest-scoring dimer model, they populated the interface between the H1 and H2 helices of the dimeric CCD, a region particularly protected from solvent as assessed by HDX-MS ( Figure  5G). Here, residues p.D240, p.G243, p.T247, p.T251, and p.G254 define the side of the H2 helix that interfaces with H1 in the segment between p.T166 and p.F182, whereas p.Y291 and p.T296 form the H2 helix in the other CCD monomer contacting the same H1 portion with a different orientation ( Figure 5G).On the basis of this new evidence, in addition to the existing x-ray structures and the CCD model, we propose the following 3D organization of the full-length CERT dimer: the dimer adopts a T-shaped conformation whereby the 2 PH domains are kept in close proximity at the extremity of the T stem, and the 2 START domains are at the extremities of the 2 T arms ( Figure 5H and Supplemental File 6). Upon SRR phosphorylation, CERT undergoes a conformational change, such that the PH domain first interacts with the phosphorylated SRR (26) and then forms a complex with the START domain (20). In the context of our structural model, this conformational change implies a rearrangement of the interface between H1 and H2 of the CCD, which may be impaired by CerTra mutations that map to this region ( Figure 5I).
We analyzed the effect of CERT mutations on the CCD structure and evaluated the melting curves of the synthetic peptides CERT 154-187 and purified CERT 151-309 WT and their mutant counterparts. We found that p.T166A decreased CERT H1 helicity and destabilized it, while p.G243R had negligible effects on CCD secondary structure and stability ( Figure 5J and Supplemental Figure 6H). Nonetheless, when we compared the HDX profile of recombinant full-length CERT WT with p.G243R, we observed diminished solvent accessibility for the p.G243R mutant in a region encompassing aa 249-256 ( Figure 5K), an H2 region coordinating with H1. This suggests that, rather than impairing CCD structure, p.G243R impairs CCD dynamics and possibly its interactions with other proteins. These results suggest a model where- leading to the transfer of a significant amount of dhCer to the trans Golgi and production of dhSM. Elevated dihydrosphingolipid levels cause neuropathology. For example, biallelic loss-of-function mutations in DEGS1, which catalyzes the final conversion of dhCer to Cer (Figure 2A), result in increased dihydrosphingolipid levels, causing a multisystem neurological disorder of both the central and peripheral nervous systems, characterized by hypomyelination and leukodystrophy (MIM #618404) (57,58). Recessively inherited loss-of-function mutations in alkaline ceramidase 3 (ACER3) also result in increased dihydrosphingolipid formation and a leukodystrophy phenotype (MIM #617762) (59). Impaired glycosphingolipid synthesis is a further cause of neuropathology: loss-of-function mutations in ST3GAL5 or B4GALNT1, which encode 2 key enzymes in ganglio-series glycosphingolipid synthesis (i.e., GM3 and GM2 synthases), cause neurodevelopmental regression (MIM #609056) and spastic paraplegia (MIM #609195), respectively (60)(61)(62)(63)(64)(65). Recently, a specific group of mutations in SPT subunits were reported to cause childhood-onset amyotrophic lateral sclerosis (66-68) and hereditary spastic paraplegia (69). All identified mutations disturbed the homeostatic control of de novo sphingolipid synthesis, resulting in greatly elevated sphingolipid levels.
Several CerTra mutations cause lipid metabolic derangements similar to those associated with the above-listed conditions, nonetheless, the molecular mechanisms of CERT dysregulation appear to be varied. Mutations in the SRR directly impair CERT-inactivating phosphorylation, whereas mutations in the CCD hinder the conformational change that follows this event. We have not directly addressed the effects of other mutations close to the FFAT motif or in the PH and START domains. These mutations likely affect CERT interaction with VAP proteins (70) and the final events of CERT inactivation that involve intramolecular interactions between the PH and START domains (24). Four consecutive variants (p.R366T, p.I381V, p.E424G, and p.A449V) do not affect CERT localization or the phosphorylation state. When analyzed in greater depth, these variants present some doubts about their actual pathogenicity: (a) each of them is associated with a single patient, and they are not part of a recognizable cluster of mutations; (b) p.I381V has been found in 1 healthy individual; (c) p.A449V has been found in an individual whose mother and father presented with intellectual disability independently of the mutational state of CERT1; (d) p.R366T has been found in an individual for whom the very little information we have would point toward a very mild phenotype. Further studies will be required to assess the actual role of these variants and, more in general, to untangle the cellular mechanisms by which sphingolipid metabolism disruption leads to CerTra syndrome. Nonetheless, the characterization of this rare disease entity and of the most frequent variants associated with it has already revealed unforeseen operating principles of sphingolipid homeostasis, led to the definition of a new structural region in CERT, and delineated a possible use of CERT inhibitors for the treatment of patients with CerTra.

Study participant identification and clinical characterization
Initial screening and identification of anonymized individuals harboring CERT1 variation was conducted using the following public data-the locomotion of a w 1118 derivative Drosophila strain bearing a large chromosomal defect in which the dCERT gene was deleted (hereafter referred to as KO) (Supplemental Figure 7I), suggesting that the HPA-12 effect on locomotion requires dCERT expression.
Although the +WT and +SL strains share motor hypoactivity, they do have distinct metabolic phenotypes ( Figure 6H), which we hypothesized should produce some difference in phenotype. Because patients with CerTra frequently have seizures, we decided to look for seizure-like reactions, such as the paralysis that seizure-susceptible flies often experience after vigorous mechanical stimulation (53,54). Our strains did not exhibit paralysis upon mechanical stimulation, but the +SL flies showed impaired negative geotaxis that could be corrected by HPA-12 treatment ( Figure  6K). Notably, negative geotaxis in nonstressed animals was indistinguishable across genotypes (Supplemental Figure 7J).
These data indicate that an increase in dCERT activity in Drosophila induced sphingolipid metabolic changes similar to those observed in CerTra syndrome, impaired locomotion, and caused central nervous system vulnerability to mechanical shock in the most severely affected line.

Discussion
In this study, we report that multiple de novo heterozygous variants in CERT1 caused an autosomal dominant developmental syndrome we named CerTra, which is characterized by various degrees of developmental delay, motor delay, cognitive impairment, behavioral abnormalities, and seizures. Several CerTra mutations reduce or abolish the capability of CERT to be inactivated in response to excessive SM production. This leads to uncontrolled Cer transfer out of the ER, with several consequences: (a) reduced Cer at the ER relaxes the homeostatic inhibition of SPT (55,56), resulting in increased de novo sphingolipid synthesis; (b) reduced Cer at the Golgi, the site for glucosylceramide synthesis, impairs glycosphingolipid production; and (c) excessive CERT activity likely competes with Cer desaturase activity at the ER, . Data shown are the mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, by 1-way ANOVA.
tion for variants i and j, respectively. A total of 1 × 10 8 permutations were performed, and statistical significance was determined (Fisher's exact test) on the basis of how many times the permuted mean distance was smaller or equal to the mean geometric distance observed in our cohort (i.e., significant clustering was determined if the permuted mean distance was larger than what was observed in our cohort).
Immunofluorescence staining and image analysis HeLa cells were grown on glass coverslips that were treated according to the experimental protocol, fixed with 4% paraformaldehyde for 15 minutes at room temperature (RT), and washed 3 times with PBS. After fixation, cells were blocked with 5% BSA and permeabilized with 0.5% saponin for 20 minutes at RT, followed by a 1-hour incubation with selected antibodies against the antigen of interest in the blocking solution. Cells were then washed 3 times with PBS and incubated with the appropriate isotype-matched, Alexa Fluor-conjugated secondary antibodies diluted in blocking solution for 30 minutes. After immunostaining, cells were washed 3 times in PBS and once in water to remove salts. After Hoechst staining for nuclei, the samples were mounted with Fluoromount-G (Southern Biotech, 0100-01) on glass microscope slides and analyzed under a Leica SP8 confocal microscope with a 63× oil objective (1.4 NA),a Zeiss LSM700 with a 40× oil objective (1.3 NA), or a Stellaris-5 with a 40× oil objective (1.3 NA). Optical confocal sections were taken at 1 Airy unit under nonsaturated conditions with a resolution of 1,024 × 1,024 pixels and a frame average of 2. Images were then processed using Fiji software (https:// imagej.net/Fiji) (80).

Statistics
Experimental design. For quantification of protein expression in patient-derived cell lines, values from at least 3 independent experiments were used. At every stage of the study, the experimenter was blinded to the identity of the control and patient-derived cell lines. For example, experimenter 1 made a list of samples and controls to be tested, and experimenter 2 randomized this list and relabeled the tubes; experimenter 2 was the only person with the key to identification of the samples. These samples were then distributed to experimenter 3 to culture the cells, and then to experimenter 1 to perform Western blot analysis, and finally, experimenters 1 and 4 analyzed the data. Only then was the key applied to identify the samples.
Software and statistical analysis. Statistical significance (P < 0.05) was analyzed using GraphPad Prism 8 (GraphPad Software), Excel (Microsoft), and R-4.2.3 (The R Project for Statistical Computing). Statistical details and the number of replicates for each experiment can be found in the figures and legends. The range of expression levels in Western blots was determined from at least 3 independent experiments. The Student's t test was used to compare the means between 2 groups, and ANOVA was used to compare the means among more than 2 groups.

Study approval
All study procedures were defined under protocol no. AAAS7401, which was approved by the IRB of Columbia University Irving Medical Center (VAG); the Cambridgeshire Research Ethics Committee for the whole of the United Kingdom; the IRB "Commissie Mens-gebondenOnderzoek Regio Arnhem-Nijmegen" under protocol no. 2011/188; the Hospital Universitario "Virgen de la Arrixaca," Wro-bases and tools: ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/), Decipher (71), GeneMatcher (72), and VKGL-NL Rotterdam (https:// www.vkgl.nl/nl/) (Supplemental Table 1). Each patient underwent a full clinical examination by a neurologist and/or medical geneticist. Clinical data were directly abstracted from medical records provided by the referring clinicians and included a behavioral assessment and EEGs. Developmental delay and cognitive ability tests were performed on patients using the following tests: Gesell Developmental Schedules, the Chinese National Health Commission Developmental Evaluation Scale, the Wechsler Intelligence Scale for Children, the Peabody Picture Vocabulary Test, 4th Edition (PPVT-4), and Bayley Scales of Infant Development II. When possible, a standardized assessment of impairment in conceptual, social, and practical domains for each patient was performed in accordance with the DSM-5 and noted to be mild, moderate, severe, or profound. The degrees of intellectual disability were classified according to the following verbal and nonverbal IQ scores: mild (IQ of 50-55 to ~70), moderate (IQ of 35-40 to 50-55), severe (IQ of 20-25 to [35][36][37][38][39][40], and profound (IQ below [20][21][22][23][24][25]. To characterize craniofacial/skeletal dysmorphia, we performed a deep phenotyping analysis for all of the patients whose families agreed to provide photographs. This was done by the team of Eva Bermejo-Sánchez at the Institute of Rare Disease Research (IIER) in Madrid, Spain. Dysmorphology analyses were performed by experimenters blinded to the variant detected for each patient. Each parameter was compared with age-, sex-, and ethnicity-matched healthy individuals. Briefly, a detailed reading of all the clinical reports was performed to extract a first list of dysmorphic features that was then used to evaluate all of the patients. All the available photographs were assessed on the basis of those features, giving a matrix of 96 rows that were scrutinized for all patients. Thus, the same traits were assessed by the same observer using homogeneous criteria for all the patients. See the Supplemental Clinical Appendix for all the features for each patient.

Pathogenicity analysis
All identified variants were analyzed in accordance with American College of Medical Genetics and Genomics (ACMG) guidelines for variant interpretation and classification (73). Minor allele frequencies (MAFs) of all CERT1 variants were obtained from gno-mAD (74), BRowse All Variants Online (BRAVO) (https://bravo.sph. umich.edu), and 1000 Genomes Project (https://www.genome.gov/ 27528684/1000-genomes-project). Functional annotation of variants was carried out with ANNOtate VARiation (ANNOVAR) (75) using M-CAP (76), REVEL (77), Eigen (78), and CADD (78) pathogenicity scores. As a general guideline, pathogenicity is predicted for variants with scores over 0.025 for M-CAP, over 0.5 for REVEL, over 0.5 for Eigen, and over 20 for CADD. Prediction scores for CerTra variants were compared with other missense variants found in singleton cases among the general population in gnomAD (i.e., missense variants reported in only 1 healthy individual in gnomAD).
The analysis of variant clustering was performed as previously described (17,79). In short, to determine whether the observed variants in patients (17 positions) cluster linearly more than expected compared with random permutations, we calculated the geometric mean distance (δg) between all missense variants on the CERT1 gene. We calculated δg by taking the mean distance normalized for the transcript length of CERT1 (hg19:NM_005713.3) over all variant combinations of x i and x j of the missense variants, where x represents the posi-The DEFIDIAG study is funded by grants from the French Ministry of Health in the framewok of the national French initiative for genomic medicine. The funders were not involved in the study design, data acquisition, analysis, or writing of the manuscript. Funding for the DECIPHER project was provided by Wellcome. Full study enrollment (i.e., acquisition of clinical data for further analysis) was conducted only after written, informed consent/assent was obtained for each study participant. The study also adhered to tenets outlined by the Declaration of Helsinki. All patient-related study procedures were conducted according to the respective ethics committees of each participating institution. Specific written, informed consent was also acquired for all participants whose photographs appear in the manuscript.

Data availability
This study made use of data generated by the DEFIDIAG study sponsored by the INSERM (France). A full list of centers that contributed to the generation of the data is available from https://defidiag. inserm.fr/ and via email from c16-110.coordo.isp@inserm.fr. This work also used Simons Simplex Collection (SSC) data including whole-exome sequencing (raw and processed) and associated phenotypic data obtained through Simons Foundation Autism Research Initiative (SFARI) Base (SFARI Request ID: 12030.1.1). This study makes use of data generated by the DECIPHER community. A full list of centers that contributed to the generation of the data is available at https://deciphergenomics.org/about/stats and via email from contact@deciphergenomics.org.