Mitochondrial FAD shortage in SLC25A32 deficiency affects folate-mediated one-carbon metabolism

The SLC25A32 dysfunction is associated with neural tube defects (NTDs) and exercise intolerance, but very little is known about disease-specific mechanisms due to a paucity of animal models. Here, we generated homozygous (Slc25a32Y174C/Y174C and Slc25a32K235R/K235R) and compound heterozygous (Slc25a32Y174C/K235R) knock-in mice by mimicking the missense mutations identified from our patient. A homozygous knock-out (Slc25a32−/−) mouse was also generated. The Slc25a32K235R/K235R and Slc25a32Y174C/K235R mice presented with mild motor impairment and recapitulated the biochemical disturbances of the patient. While Slc25a32−/− mice die in utero with NTDs. None of the Slc25a32 mutations hindered the mitochondrial uptake of folate. Instead, the mitochondrial uptake of flavin adenine dinucleotide (FAD) was specifically blocked by Slc25a32Y174C/K235R, Slc25a32K235R/K235R, and Slc25a32−/− mutations. A positive correlation between SLC25A32 dysfunction and flavoenzyme deficiency was observed. Besides the flavoenzymes involved in fatty acid β-oxidation and amino acid metabolism being impaired, Slc25a32−/− embryos also had a subunit of glycine cleavage system—dihydrolipoamide dehydrogenase damaged, resulting in glycine accumulation and glycine derived-formate reduction, which further disturbed folate-mediated one-carbon metabolism, leading to 5-methyltetrahydrofolate shortage and other folate intermediates accumulation. Maternal formate supplementation increased the 5-methyltetrahydrofolate levels and ameliorated the NTDs in Slc25a32−/− embryos. The Slc25a32K235R/K235R and Slc25a32Y174C/K235R mice had no glycine accumulation, but had another formate donor—dimethylglycine accumulated and formate deficiency. Meanwhile, they suffered from the absence of all folate intermediates in mitochondria. Formate supplementation increased the folate amounts, but this effect was not restricted to the Slc25a32 mutant mice only. In summary, we established novel animal models, which enabled us to understand the function of SLC25A32 better and to elucidate the role of SLC25A32 dysfunction in human disease development and progression.

The SLC25A32 gene locates on chromosome 8q22.3 and contains seven exons. It encodes a 35 kDa protein SLC25A32 [4] and is ubiquitously expressed in all tissues [5]. SLC25A32 was proposed as a mitochondrial FAD transporter according to its ability to functionally complement the FLX1-mutated yeast strain with mitochondrial FAD transport defect [6]. The pathogenesis of RREI was considered caused by the mitochondrial FAD transport defect, which reduced the activities of mitochondrial flavoproteins involved in fatty acid β-oxidation (FAO) and the respiratory chain [1]. However, earlier studies suggested that SLC25A32 was a mitochondrial folate transporter [4,7]. Variation in SLC25A32 SNP rs17803441 was found to be associated with lower 5-methyltetrahydrofolate (CH3-THF) concentrations in human red blood cells [8]. Recently, biallelic loss of function variants in the SLC25A32 gene had been identified in the human fetus with neural tube defects (NTDs), and Slc25a32 gene trapped knock-out mice also presented with cranial NTDs in the embryonic stage [9]. It is well known that the folate-mediated one-carbon metabolism (FOCM) abnormality is the most critical modifier of risk associated with neural tube closure [10,11].
In clinic, we identified a patient who carried novel SLC25A32 point mutations and presented with muscle weakness and intractable clubfeet. Neither riboflavin supplementation nor surgery blocked the development of clubfeet. Clubfoot is usually a secondary manifestation of spina bifida, distal arthrogryposis, congenital myotonic dystrophy, etc. Its etiopathogenesis is still unknown regardless of its long-lasting recognition [12], of which spina bifida is a type of NTD. The reported RREI patients [1][2][3] had similar clinical manifestations, biochemical features, and responses to riboflavin supplementation as MADD patients. However, clubfoot is not observed in MADD patients, although MADD patients with the most severe symptoms also had congenital anomalies, such as dysmorphic facial features, large cystic kidneys, hypospadias and chordee in males, and neuronal heterotopias [13]. The different phenotypes suggest distinct pathogenic mechanisms or common underlying pathogenesis for the two inborn errors. The complexity of the disease prompted us to reconsider the etiology and treatment strategy of RREI. In this study, we first reviewed the clinical history of the case and performed a morphology study and neuropsychological evaluation for the patient. By mimicking the mutations identified in our patient, we generated multiple mice models with variable degrees of SLC25A32 dysfunctions to study the pathophysiology of RREI. Our results advanced our understanding of the relationship between the Slc25a32 mutant and NTDs, and how SLC25A32 dysfunction affected mitochondrial flavoenzymes. New findings suggested potential therapeutic intervention for patients suffering from mitochondrial FAD deficiency.

Study approval
This study was performed in line with the principles of the Declaration of Helsinki. The ethics committee of Guangzhou Women and Children's Medical Center authorized this study, and the Institutional Animal Care and Use Committee of Guangzhou Medical University approved the animal studies. Informed consent was obtained from the patient and her guardian.

Generation of Slc25a32 mutant mice models
Human SLC25A32 c.521A > G (p.Y174C) and c.704A > G (p.K235R) are conserved to mouse Slc25a32 (NM_172402) c.521A > G (p.Y174C) and c.704A > G (p.K235R), respectively. CRISPR/Cas9 technology was utilized to generate the mouse mutants with the corresponding missense mutations. The sgRNAs used for generating the p.Y174C and p.K235R mouse mutants were TAT AAA TAT GAA GGT GTG CGTGG (matching the forward strand of Slc25a32), and ATA CGG GTA TGT TGC TGC TACGG (matching the reverse strand of Slc25a32), respectively (PAM sequence was underscored). The oligo donors for generating the p.Y174C and p.K235R mouse mutants were GGT GTT GCT AGC CCT TCA CAG AGA CAG TAT AAA GGA ATG TTT GAT GCA CTT GTG AAA ATA TGT AAA TAT GAA GGT GTG AGA GGA TTA TAC AAG GTA ACA AAT TAT CAT GAA TAT ATT TTG AGT ACT GGA AATC and ATC CTG CTT GTG TTC TCA TTT TAG AGT ACA GCA GAA TAC ATC TCT GTC GCA GCG CTA TCC AGA ATA TTT GCG GTA GCA GCA ACA TAC CCG TAT CAG GTT GTG AGA GCC CGC CTT CAG GAT CAG, respectively. The p.Y174C (c.521A > G, TAT to TGT) mutation in donor oligo was introduced into exon 4, and the p.K235R (c.704A > G, AAA to AGA) mutation in donor oligo was introduced into exon 6, by homology-directed repair. Silent mutations p.R180 = , CGT to AGA and c.714C > G, GCC to GCG were co-introduced for p.Y174C and p.K235R mutants, respectively, to prevent the binding and re-cutting of the sequence by sgRNA after homology-directed repair.
The mutant mice were crossed to wild-type C57BL/6 mice for at least three generations to eliminate potential off-targeting mutations. Potential off-targeting sites were predicted using CCTOP (https:// cctop. cos. uni-heide lberg. de/), and the whole Slc25a32 coding region and flanking sequences were sequenced for potential off-targeting mutations. Specifically, in the generation of the p.Y174C mouse mutant, we obtained a mutant mouse with an additional c.552G > A (AAG to AAA) mutation at the end of the fourth exon. This mutation interfered with the splicing of Slc25a32 and did not produce detectable normal transcripts. This mouse mutant was kept and used as a "knock-out" allele in this study. Genotyping was performed by sequencing the PCR amplicons of tail or embryo amnion, and the primers were: forward 5′-AAT ATG GAT TGC ATG AAA CAG TAC C-3′ and reverse 5′-TGT ACT CTG TAG TCT TGG ATG GGA A-3′.

Collection of tissues and embryos
Mice aged 2-3 months were killed by cervical dislocation, and skeletal muscle and brain tissues were isolated rapidly, then chilled in liquid nitrogen or minced into small pieces for mitochondria isolation. Embryos at embryonic day 11.5 (E11.5) were dissected from dams and chilled in liquid nitrogen or pooled by genotype for mitochondria isolation.

Isolation of mitochondria
Mitochondria used for this study were isolated from skeletal muscle using the differential centrifugation method, according to the published protocol [14]. Several E11.5 Slc25a32 −/− embryos were pooled as one sample for mitochondria isolation using the same protocol for isolating mice liver mitochondria [14]. The purity of mitochondria was examined by Janus green B staining, light microscopic examination, and immunoblot analyses of markers for individual subcellular organelles, as previously described [15].

Slc25a32 gene mRNA expression
Total mRNA was isolated from skeletal muscle or embryo using TRIzol ® Reagent (ambion ® , Life technologies, USA) according to the manufacturer's protocol. cDNA was synthesized by the PrimeScript™ RT Master Mix kit (TAKARA, Japan). For the mouse mutants with Slc25a32 missense mutations, real-time quantitative PCR (Q-PCR) was performed to determine the abundance of Slc25a32 mRNAs by using the primers 5′-ATG GGT GAC GAA AAC TCG CCTT-3′ and 5′-CGC ACC ATG TGA TGT TCC AAA-3′ on a LightCycler ® 480II system (Roche, Switzerland) with SYBR ® Premix Ex Taq™ II (TAKARA, Japan) reagent. GAPDH was used as the loading control. For the Slc25a32 −/− mice, primers located on exon 2 and 7 of Slc25a32 (5′-AAT ATG GAT TGC ATG AAA CAG TAC C-3′ and 5′-TCC ACC GAT GCC TTC TTT CC-3′) were used to amplify a 678-bp amplicon of the Slc25a32 coding region. The amplicons were then separated by gel electrophoresis and purified for DNA sequence analysis.

Determination of proteins by capillary immunoassay (simple western)
Total protein was extracted from skeletal muscle or embryos with RIPA lysis buffer (P0013C, Beyotime, China) according to the protocol from the manufacturer. The concentrations were determined using BCA Protein Assay Kit (BCA1, Sigma-Aldrich). A capillary-based automated Simple Western system (Wes, ProteinSimple, San Jose, CA, USA) measured the amounts of SLC25A32 and nine mitochondrial flavoenzymes in the lysates. The Wes 12-230 kDa separation module, 8 × 25 capillary cartridges (SM-W004), was used. Specifically, the lysate was diluted with sample buffer from the kit to 1.25 × of the desired final concentration, then mixed with the Master Mix from the kit. The mixture was denatured at 95 °C for 5 min. 3 μL of the denatured sample was loaded to a capillary. Primary antibodies were purchased and diluted with antibody dilution II solution from the kit before use  . Anti-GAPDH (AF7021, Affinity, OH, USA), antiα-tubulin (11224-1-AP, Proteintech), or anti-HSP60 (15282-1-AP, Proteintech) was chosen as loading control according to the molecular weight of the target protein. The target protein and its loading control were multiplexed in the same capillary by mixing the primary antibody with the loading control antibody. Goat anti-rabbit or anti-mouse IgG secondary antibody from the kit was used. The Wes system ran with the default program (Separation time 30 min, Separation voltage 375 V, Antibody diluent time 5 min, primary antibody incubation time 30 min, secondary antibody incubation time 30 min). Compass software (ProteinSimple) was used to visualize virtual gels. The relative protein quantification was obtained by normalizing the signal of the target protein against the signal of its loading control.

Metabolites analysis
Nitrogen chilled skeletal muscle and whole-brain tissues were first pulverized by CryoGrinder™ System and then homogenized in PBS by sonication. Embryos were homogenized in PBS by sonication directly. Homogenates were centrifuged to remove debris (4 °C, 15,000 g for 20 min), and the supernatants were used for metabolite analysis. An aliquot of each sample was removed for total protein quantification. Analysis of acylcarnitine, amino acids, organic acids, DMG, and formate was carried out according to the published protocols [19][20][21][22] using the LC-MS/MS system (Shimadzu Ultrahigh Pressure Nexera chromatograph system interfaced to an AB Sciex 3200 Q TRAP mass spectrometer (Foster City, CA, USA)), or a gas chromatography-mass spectrometry system (GC-MS, Agilent 5975GC-4890MS, USA) [23].

Mitochondrial formate synthesis assay
Measurement of mitochondrial synthesis of formate from glycine was carried out as described previously [27]. Briefly, mitochondria were resuspended in the incubation buffer (pH

Statistics
All results were presented as mean values ± standard error of the mean (SEM). Statistical analysis was performed using SPSS. The two-tailed unpaired Student's t test was used to compare two groups. P < 0.05 was considered significant.

An RREI case with novel SLC25A32 mutations
A 7-year-old female patient came to our department complaining of dyspnea, fatigue, and growth retardation (height 108 cm (−3 SD), weight 18 kg (− 2 SD to − 1 SD)). Blood gas analysis demonstrated that she suffered from acidosis with the pH at 7.276 (ref. 7.31-7.41) and lactic acid elevated to 10.7 mmoL/L (ref. 0.9-1.7 mmoL/L). Plasma metabolite analysis revealed a MADD-like biochemical profile with dimethylglycine (DMG), sarcosine, and multiple acylcarnitines accumulation (Fig. 1A). The patient was diagnosed with MADD and immediately received a low protein diet and oral supplementation of riboflavin and L-carnitine. After a 2-week treatment, the patient's dyspnea disappeared, and the biochemical abnormality ameliorated. From then on, the patient received riboflavin (30 mg P.O. tid) and L-carnitine (1 g P.O. qid) supplementation with a normal diet. Now, the patient is a 20-year-old with nearly normal growth (height 155 cm (-1SD); weight 58 kg (mean 51 kg)).
Before coming to our department, the patient was found to be unwilling to walk after suffering from influenza at 1.6-year-old. Her feet developed to equinovarus and talipes cavus progressively. The right foot was more severe than the left one initially. She underwent orthopedic surgery to correct her right leg at 5.5-year-old, and the surgery was successful (Fig. 1B). Unexpectedly, the deformity of her left leg deteriorated gradually even under riboflavin treatment, and she underwent her second orthopedic surgery at 9-yearold. However, this operation did not improve her symptom. Instead, it relapsed: the forefoot adducted, the talus broadened and was less convex, the the calcaneus became shorter, wider, and angulated medially, and the lateral angle of the talocalcaneal diminished (Fig. 1B). The left clubfoot relapsed after surgery. C Brain MRI revealed that the left lateral ventricle of the patient broadened slightly. D X-ray imaging showed that the left fibula of the patient was abnormally slim. E Compound heterozygous mutations in SLC25A32 were identified from the patient, and testing for the mutations in parental samples confirmed that the unaffected parents were heterozygous for the mutations. F Homology modeling analysis using the mitochondrial ADP/ATP carrier (PDB ID: 6GCI) as the template predicted that K235 was located at the substrate-binding pocket and the Y174 at the second matrix loops facing the mitochondrial matrix A morphology study was performed to investigate the etiology of lower extremity weakness and clubfeet of the patient. The magnetic resonance imaging (MRI) revealed an enlarged left lateral ventricle (Fig. 1C). Electromyography found no abnormality in the peroneal nerve, and the signal transmission between peroneal nerves to muscle was fluent. X-ray imaging of tibiofibular bones showed slender fibulae, and the left one was more severely affected than the right one, suggesting fibulae hypoplasia (Fig. 1D).
The patient then underwent a comprehensive neuropsychological evaluation at 15.8-year-old to identify whether neurologic impairment was involved. The gross motor function measure (GMFM) was carried out at first, and the results showed that the patient had scores in the D (standing) and E (walking, running, and jumping) domains at 90% and 60% of the total domain scores, respectively, indicating an inferior motor function in these domains. The intelligence test (Wechsler Intelligence Scale for Children IV) demonstrated that the patient's intelligence was at the borderline with a full-scale intelligence quotient of 71. In addition, her auditory-and visual-working memory function had mild to moderate defects with scores of 9 and 6, respectively. MADD is commonly caused by defects in electron transportation due to variants in the ETFA, ETFB, or ETFDH genes. However, Sanger sequencing found no mutation in these genes of the patient. Whole-exome sequencing was then performed, and compound heterozygous mutations in SLC25A32 were identified: c.521A > G (p.Y174C) and c.704A > G (p.K235R) (Fig. 1E), which were located at the fourth and sixth exon of SLC25A32, respectively. Testing the mutations in parental samples confirmed that the unaffected parents were heterozygous for the mutations. Neither variant was listed in the gnomAD database. Both affected residues were highly conserved across species (the p.Y174 conserved up to Drosophila melanogaster, and the p.K235 conserved up to Saccharomyces cerevisiae). Protein structure homology modeling analysis using the mitochondrial ADP/ ATP carrier (PDB ID: 6GCI) as the template predicted that K235 was located at the substrate-binding pocket and Y174 at the second matrix loops facing the mitochondrial matrix (Fig. 1F). Both variants were likely to damage the SLC25A32 protein by 4 of 4 in silico tools-Protein Variation Effect Analyzer (PROVEAN), Sorting Intolerant from Tolerant (SIFT), PolyPhen, and MutationTaster.
In summary, the phenotypes of our patient include riboflavin-responsive dyspnea, growth retardation, muscle weakness, and exercise intolerance, but riboflavinunresponsive clubfeet, hypoplasia of fibulae, and mild neuropsychological abnormality. We wondered whether mitochondrial folate transport deficiency was involved in RREI and whether the stubborn clubfeet resulted from folate transport defect. To answer these questions, we performed this study.

Various phenotypes were observed in mice carrying different Slc25a32 mutations.
T h e h o m o z y g o u s ( S l c 2 5 a 3 2 Y 1 7 4 C / Y 1 7 4 C a n d Slc25a32 K235R/K235 ) and compound heterozygous (Slc25a32 Y174C/K235R ) knock-in mice were born with the expected Mendelian ratio. No genotype effect was observed on birth weight, neonatal growth, or adult weight. Clubfoot was not observed in mice models after examining the gait of 40 Slc25a32 Y174C/Y174C , 67 Slc25a32 K235R/K235R , and 39 Slc25a32 Y174C/K235R mice. Q-PCR revealed the Slc25a32 Y174C/Y174C mice, but not the Slc25a32 K235R/K235R and Slc25a32 Y174C/K235R mice, showed significantly reduced Slc25a32 mRNA expression in muscle ( Fig. 2A). Consistently, the immunoblot assay showed reduced SLC25A32 protein amounts in Slc25a32 Y174C/Y174C muscle (Fig. 2B). To assess whether the mutant mice had motor impairment as our patient had, we tested the muscle strength, exercise coordination, and anti-fatigue capacity of mice by performing hanging wire test, rotarod test, and forced swimming test (FST), following the published protocols [28][29][30]. Adult mice aged 2-3 months (weighed 18.6 ± 1.6 g) were used. The times hanging on the wire were shorter for the Slc25a32 Y174C/K235R and Slc25a32 K235R/K235R mice than for their WT littermates, though the difference did not reach significance (Fig. 2C). The Slc25a32 K235R/K235R mice also performed worse in the rotarod test and FST than their WT littermates (Fig. 2D). The motor function of Slc25a32 Y174C/Y174C mice was comparable with their WT littermates.
No S l c 2 5 a 3 2 − / − p u p wa s b o r n f r o m t h e Slc25a32 +/− × Slc25a32 +/− mating. Embryos collected at the embryonic day of 11.5 (E11.5, the post-neurulation-stage in normal mouse embryos) showed that 89.5% (n = 94) of Slc25a32 −/− embryos had NTDs with exencephaly or craniorachischisis phenotypes, and the rest (n = 11) had normal neural tube closure but severe growth retardation (Fig. 2E). Sequencing of DNA amplicons of Slc25a32 −/− embryo amnions revealed that the point mutation (TAT to TGT) and the silent mutation (AAG to AAA) were co-introduced successfully in both strands of the Slc25a32 gene (Fig. 2F, G). The reverse transcription PCR (RT-PCR) using primers located on exon 2 and 7 of the Slc25a32 gene showed that a 678 bp cDNA fragment observed in WT embryo was absent in Slc25a32 −/− embryos. Instead, fragments at 517 bp and 431 bp were presented (Fig. 2H). Sequence analysis revealed that the 517 bp and 431 bp fragments were truncated transcripts that lacked exon 4 and both exon 3 and 4, respectively. Q-PCR using primers located on exon 4 and 5 showed no Slc25a32 mRNA expression in Slc25a32 −/− embryos (Fig. 2I).

The SLC25A32 is not a mitochondrial folate transporter, but plays an essential role in FOCM
To investigate whether folate transport deficiency was involved in RREI, we first analyzed the folate profile of mitochondria isolated from the skeletal muscle of mice. Fig. 2 Characterization of mouse models with Slc25a32 gene mutants. A Slc25a32 mRNA expression in skeletal muscle of Slc25a32 mutant mice (n = 3). A significant difference was observed between Slc25a32 +/+ and Slc25a32 Y174C/Y174C mice. B SLC25A32 protein in skeletal muscle of Slc25a32 mutant mice (n = 2). C The hanging wire test showed that the times hanging on the wire were shorter for the Slc25a32 Y174C/K235R and Slc25a32 K235R/K235R mice than for Slc25a32 +/+ mice, but the difference did not reach significance (n = 7-8). D The rotarod test and FST revealed that the Slc25a32 K235R/K235R mice performed worse than their WT littermates, but the difference did not reach significance (n = 8). E Various NTDs were observed in E11.5 Slc25a32 −/− embryos. a, No NTDs but growth retardation; b, c: exencephaly; d, e: craniorachischisis (scale bars represent 1 mm). F Generation of the Slc25a32 knock-out mouse model using the CRISPR/Cas9 system. Two mutation sites (c.12873 A > G, c.12904 G > A) in donor oligo were introduced into exon 4 by homology-directed repair. The primers used for RT-PCR and Q-PCR were indicated. G Sequencing of DNA amplicons of Slc25a32 +/+ and Slc25a32 −/− embryo amnions. The point mutation (TAT to TGT) and a silent mutation (AAG to AAA) were co-introduced in both strands of Slc25a32 gene for Slc25a32 −/− embryo. H RT-PCR for Slc25a32 mRNA expression in E11.5 embryos using primers F1-R1 showed that a 678 bp cDNA fragment observed in Slc25a32 +/+ embryo was absent in Slc25a32 −/− embryos. Instead, fragments at 517 bp and 431 bp were presented in Slc25a32 −/− embryos. I Q-PCR for Slc25a32 mRNA expression in E11.5 embryos using primers F2-R2 showed undetectable Slc25a32 transcripts in Slc25a32 −/− embryos (n = 3). n represents biologically independent replicates. Data are mean ± SEM. The P values were determined by an unpaired, twotailed Student's t test. ***P < 0.001 versus Slc25a32 +/+ 375 Page 8 of 18 The tetrahydrofolate (THF) and 5, 10-methenyl-tetrahydrofolate (CH-THF) were the primary folate intermediates in mitochondria, accounting for > 90% of the total folate. The Slc25a32 K235R/K235R mice, but not Slc25a32 Y174C/Y174C mice, had significantly lower levels of all folate intermediates than the WT mice (Fig. 3A). We further tested the mitochondrial folate uptake rate. The stable isotopelabeled [ 15 N]-FA and CHO-THF were used as substrates. The results showed that neither the uptake of [ 15 N]-FA nor the uptake of CHO-THF was hindered by any of the Slc25a32 mutants (Fig. 3B), which suggested that mitochondrial folates deficiency in Slc25a32 K235R/K235R mice was not caused by folate transport defect.
A previous study evidenced that maternal CH3-THF supplementation failed to rescue the NTDs of Slc25a32 knock-out embryos, but formate could [9]. We repeated the same experiment but used 0.1 mol/L sodium formate instead of calcium formate in the dam's drinking water. We collected 73 embryos from 9 litters, and genotyping revealed that there were 16 Slc25a32 +/+ , 41 Slc25a32 +/− , and 16 Slc25a32 −/− embryos, of which 87.5% of Slc25a32 −/− embryos completed neural tube closure (Fig. 3E). Analyzing the folate profile again showed that the levels of CH3-THF were significantly higher, and the levels of CH-THF were significantly lower in Slc25a32 −/− embryos without NTDs than in their WT siblings (Fig. 3F). The levels of DHF, THF, and CH2-THF decreased in Slc25a32 −/− embryos upon formate treatment, but they were still significantly higher when compared with the WT embryos. 12.5% of Slc25a32 −/− embryos still had NTDs after formate supplementation, whose folate profiles were similar to the Slc25a32 −/− embryos with NTDs before formate treatment.
These data proved that SLC25A32 was not involved in mitochondrial folate transport. However, Slc25a32 mutations would interrupt folate metabolism, leading to mitochondrial folate deficiency or whole embryo CH3-THF shortage.

Slc25a32 −/− mice embryos suffered from glycine cleavage defect due to deficiency in dihydrolipoamide dehydrogenase (DLDH)
Since formate supplementation successfully increased the CH3-THF levels and ameliorated the NTDs in Slc25a32 −/− embryos, we wondered whether the Slc25a32 −/− embryos suffered formate deficiency. We quantified the formate in whole embryos and found that the formate amounts in Slc25a32 −/− embryos were about 70% of controls (Fig. 4A). Formate is an intermediate of one-carbon (1C) metabolism and is produced through the mitochondrial metabolism of metabolic substrates, including serine, glycine, sarcosine, and dimethylglycine (DMG) [11]. Analysis of these metabolic substrates in embryo homogenates showed that the Slc25a32 −/− embryos had significantly increased glycine and serine contents, whereas Slc25a32 K235R/K235R had significantly increased serine but significantly decreased glycine contents (Fig. 4B), when compared with WT embryos. There was no significant difference between Slc25a32 mutant and WT embryos in sarcosine and DMG levels.
A more than threefold increase of glycine was specifically identified in the Slc25a32 −/− embryos, which led us to hypothesize that formate deficiency found in Slc25a32 −/− embryos resulted from glycine cleavage defect. To prove the hypothesis, we measured the mitochondria produced formate using a stable isotope-labeled [ 13 C 2 ]-glycine as the substrate. The amounts of [ 13 C]-formate produced by Slc25a32 −/− embryo mitochondria was about 10.6% of controls (Fig. 4C), uncovering a causative role of glycine oxidation deficit in Slc25a32 −/− embryos. We further measured the protein contents of the four subunits of glycine cleavage system (GCS). The results showed that the amounts of amino methyltransferase (AMT), glycine decarboxylase (GLDC), and hydrogen carrier protein (GCSH) in Slc25a32 −/− embryos were comparable to that in WT embryos, but the amounts of DLDH were significantly lower in Slc25a32 −/− embryos than in WT embryos (Fig. 4D). The enzymatic activity assay revealed that the mitochondrial DLDH activity in Slc25a32 −/− was about 23.9% of controls (Fig. 4E). DLDH is an oxidoreductase in the mitochondrial matrix, and its activity is dependent on the presence of both FAD and NAD [31].
The Slc25a32 mutations specifically blocked the mitochondrial uptake of FAD SLC25A32 had been identified as a mitochondrial FAD transporter based on the experiment that transfection of the human SLC25A32 gene functionally complemented the FLX1 mutant yeast strain with mitochondrial FAD transport defect [6]. We then hypothesized that DLDH deficiency in Slc25a32 −/− mitochondria resulted from the deprivation of FAD due to SLC25A32 dysfunction. Here, we first measured the mitochondrial contents of FAD and its precursors, FMN and Rf, in mitochondria. Compared with WT controls, the Slc25a32 Y174C/K235R and Slc25a32 K235R/K235R mice, but not the Slc25a32 Y174C/Y174C mice, had significantly lower FAD, FMN, and/or Rf contents (Fig. 5A). The Slc25a32 −/− embryos also suffered from mitochondrial FAD, FMN, and Rf shortages (Fig. 5B). By analyzing the mitochondrial uptake of stable isotope-labeled [ 13 C 5 ]-FAD and [ 13 C 4, 15 N 2 ]-Rf, we found the Slc25a32 K235R mutation specifically blocked the mitochondria uptake of [ 13 C 5 ]-FAD, as indicated by the significantly lower uptake amounts of [ 13 C 5 ]-FAD in Slc25a32 Y174C/K235R and Slc25a32 K235R/K235R mitochondria, when compared with controls (Fig. 5C). Instead, Slc25a32 mutations did not hinder the uptake of [ 13 C 4, 15 N 2 ]-Rf, and the Y174C mutation even promoted the uptake of Rf, which suggested that the mitochondrial Rf deficiency in Slc25a32 mutant mice derived from FAD shortage instead of Rf transport deficiency. Knock-out of Slc25a32 also disabled the mitochondria from transporting [ 13 C 5 ]-FAD, as indicated by the low abundance of [ 13 C 5 ]-FAD in the mitochondria and high abundance of [ 13 C 5 ]-FAD in the incubation buffer of Slc25a32 −/− embryos (Fig. 5D).
SBCAD, IVD, IBD, and GCDH are flavoenzymes involved in the metabolism of leucine, isoleucine, valine, tryptophan, hydroxylysine, and lysine, and SCAD, MCAD, and VLCAD are flavoenzymes participating in FAO. The enzymatic activity assays confirmed lower activities of all these flavoenzymes in Slc25a32 K235R/K235R mice than in WT controls (Fig. 7A-G). Decreased activities of GCDH and SCAD were also found in Slc25a32 Y174C/K235R mice. FAD supplementation in vitro completely restored the function of SBCAD (Fig. 7A) and markedly increased the activities of IVD and GCDH ( Fig. 7B and C), but only mildly improved the activities of SCAD, MCAD, VLCAD, and IBD ( Fig. 7D-G) in Slc25a32 K235R/K235R and Slc25a32 Y174C/K235R mice. The immunoblot assays further revealed lower amounts of these flavoenzymes in Slc25a32 K235R/K235R mice than in WT controls (Fig. 7H) and the reduction degrees correlated with their responses to FAD supplementation. We also measured the amounts of flavoenzymes involved in electron transportation and found that the amounts of ETFα, ETFβ, and ETFDH were lower in Slc25a32 K235R/K235R mice than in controls (Fig. 7H).

Effect of riboflavin and formate supplementation on Slc25a32 mutant mice
Since FAD could be synthesized from riboflavin in mitochondria [33], we tried riboflavin supplementation (1 g/kg riboflavin in diet) for Slc25a32 K235R/K235R and Slc25a32 Y174C/K235R mice. After 1 month of treatment, the mutant mice had no improvement in motor function. Further mitochondrial flavin analysis found no increase in riboflavin and FAD levels, suggesting that the orally supplemented riboflavin had not been effectively absorbed by mice. Since the Slc25a32 mutant mice suffered from mitochondrial folate shortage, we added 0.1 mol/L sodium formate to their drinking water to study whether formate could improve folate shortage. After a 3-week treatment, the WT mice had total folates increased from 5.33 ± 0.58 to 13.03 ± 0.41 pmol/ mg protein, and Slc25a32 K235R/K235R mice had total folates increased from 2.34 ± 0.70 to 6.08 ± 0.59 pmol/mg protein (Fig. 8). This folate elevation effect was not restricted to the Slc25a32 mutant mice, and the disparity in folate amounts between Slc25a32 K235R/K235R and WT mice still exist.

Discussion
In this work, we described a new case of RREI with SLC25A32 missense mutations and generated several mice models to characterize the molecular mechanisms underlying SLC25A32 mutations-induced RREI. Our study suggested that SLC25A32 was a mitochondrial FAD transporter instead of a folate transporter. SLC25A32 dysfunction blocked FAD entry into mitochondria and resulted in mitochondrial FAD deficiency. Two of the three mice models with Slc25a32 missense mutations, one possessed homozygous K235R mutations, and the other had compound heterozygous Y174C/K235R mutations, presented with mild motor impairment, which was consistent with the phenotype of our patient. Homology modeling analysis predicted that K235 was located at the substrate-binding pocket. Consistent with the prediction, the homozygous K235R mutations profoundly affected mitochondrial FAD uptake, resulting in severe metabolic disturbance. Y174 was predicted to be located at the second matrix loop of the Slc25a32 gene, facing the mitochondrial matrix. Homozygous Y174C mutation diminished Slc25a32 mRNA and protein expression, but it did not cause a significant defect in mitochondrial FAD transport. The compound heterozygous Y174C/K235R mutations, which mimicked the mutations of our patient, led to mild mitochondrial FAD uptake deficiency and metabolic disturbance.
A positive correlation between SLC25A32 dysfunction and mitochondrial flavoenzymes defect was observed by Fig. 6 Slc25a32 mutant mice recapitulated the MADD-like biochemical features of the RREI patient. A-C Acylcarnitine profiles of embryos, skeletal muscle, and brain tissues of mice (n = 5). Accumulation of multiple acylcarnitines and deficiency of free carnitine were found in Slc25a32 −/− embryos, Slc25a32 K235R/K235R , and Slc25a32 Y174C/K235R mice. D-E Glutaric acid and DMG amounts in skeletal muscle and brain tissues of Slc25a32 mutant mice (n = 5).
Glutaric acid and DMG accumulated in Slc25a32 Y174C/K235R and Slc25a32 K235R/K235R mice tissues. F Quantification of formate contents in skeletal muscle lysates of mice (n = 5). Slc25a32 K235R/K235R and Slc25a32 Y174C/K235R mice showed significantly reduced formate contents. n represents biologically independent replicates. Data are mean ± SEM. P values were determined by an unpaired, two-tailed Student's t test. *P < 0.05 versus Slc25a32 +/+ control studying the three mice models simultaneously. FAD is a redox cofactor ensuring the function of many flavoenzymes involved in fundamental biochemical processes [32]. Given the central importance of FAD, it is reasonable to expect a large global change in metabolites upon mitochondrial FAD deficiency. As expected, the metabolic analysis demonstrated that multiple biochemical processes related to ATP generation, including FAO, catabolism of multiple essential amino acids (leucine, isoleucine, valine, tryptophan, lysine), and degradation of choline were interrupted by SLC25A32 dysfunction (Fig. 9). Furthermore, enzymatic activity and immunoblot assays confirmed the impairment of flavoenzymes involved in these oxidation processes and revealed that the electron transport chain was also affected. Moreover, we found that the SLC25A32 dysfunction differentially affected different flavoenzymes. For example, it abolished the conversion of apoprotein to holoprotein of SBCAD, resulting in exogenous FAD-rescuable function deficiency in vitro; it also diminished the amounts of flavoproteins involved in FAO, leading to exogenous FAD-irretrievable activity. The latter further confirmed the role of FAD in flavoprotein folding and cellular stability [34,35]. This data explained the responses of plasma biochemical indicators of the patient to riboflavin supplementation: the indicators of SBCAD and IVD deficiencies, 2-methylbutyryl and isovaleryl carnitine, had levels normalized quickly, whereas the indicators of MCAD and VLCAD deficiencies, octanoyland myristoyl carnitine had the levels gradually normalized in more than 3 years. Fig. 7 Slc25a32 mutations affect multiple mitochondrial flavoenzymes. A-G Enzymatic activities of muscular SBCAD, IVD, SCAD, MCAD, VLCAD, GCDH, and IBD, in the reaction system with or without FAD supplementation (n = 5). Differences in activities between genotypes were noted for these enzymes. I Western blot analysis of nine mitochondrial flavoproteins in skeletal muscle tissues of Slc25a32 K235R/K235R mice compared to Slc25a32 +/+ mice (n = 3). HSP60 or GAPDH was used as a loading control for each protein. Significant differences in protein contents between genotypes were noted for IBD, SCAD, MCAD, VLCAD, and ETFα. n represents biologically independent replicates. Data are mean ± SEM. P values were determined by an unpaired, two-tailed Student's t test. *P < 0.05, **P < 0.01, ***P < 0.001 versus Slc25a32 +/+ control It has long been appreciated that SLC25A32 was a mammalian mitochondrial folate transporter [4,7]. In this study, our Slc25a32 −/− mice model presented with embryonic lethality and NTDs, which was consistent with the phenotype of Slc25a32 gene-trapped mice [9]. Given the known importance of FOCM in neural tube closure, the NTDs in Fig. 8 Effect of formate supplementation on mitochondrial folate profile of Slc25a32 K235R/K235R mice. The abundances of all folate intermediates were significantly lower in Slc25a32 K235R/K235R mice than in Slc25a32 +/+ mice (n = 5-6). n represents biologically independent replicates. Data are mean ± SEM. P values were determined by an unpaired, two-tailed Student's t test. ***P < 0.001 versus Slc25a32 +/+ control Fig. 9 Schematic diagram summarizing the principal metabolic pathways affected by SLC25A32 dysfunction. Enzymes involved in fatty acid β-oxidation, amino acids (glycine leucine, isoleucine, valine, tryptophan, lysine, and tryptophan) catabolism, and choline metabolism are FAD-dependent flavoproteins, which catalyze dehydrogenation reactions and transfer electrons to the mitochondrial respiratory chain through ETF and ETFDH sequentially. SLC25A32 dysfunction diminishes mitochondrial FAD contents, resulting in deficiencies of these metabolic pathways, which further affects ATP production and 1C unit generation, leading to impairment of muscle and brain function as well as neural tube closure the mice models suggested the involvement of FOCM in SLC25A32 dysfunction. By quantifying formate and all folate intermediate contents in embryo homogenates, we confirmed the disorder of FOCM in Slc25a32 −/− embryo. However, the mitochondrial folate uptake assay disclosed the intact mitochondrial folate transport capacity in Slc25a32 −/− embryo. Therefore, other factors should be considered as the cause of FOCM in Slc25a32 −/− embryo. Analysis of 1C donors revealed glycine accumulation in Slc25a32 −/− embryos. The glycine is the most critical 1C supplier in the neurulation stage of embryos [36]. It was oxidized to CO 2 , NH 3 , and a 1C unit in the form of CH2-THF (Fig. 9) under the function of strictly mitochondrial GCS. Glycine oxidation defects caused by dysfunctions of subunits of GCS, such as GLDC and AMT in humans and mice, had been evidenced with NTDs phenotype [37,38]. Our study further proved significantly reduced formate generation from glycine in Slc25a32 −/− embryo. Formate supplementation effectively increased the CH3-THF levels and ameliorated the NTDs in Slc25a32 −/− embryos, suggesting that the underlying NTDs were due to a lack of 1C units derived from glycine. The discovery of DLDH deficiency was facilitated by measuring protein contents of the four subunits of GCS, which revealed the protein amount of DLDH instead of protein amounts of AMT, GLDC, and GCSH diminished in Slc25a32 −/− embryos. SLC25A32 was a mitochondrial FAD transporter, according to our study. It was reasonable to propose that deletion of SLC25A32 made the FAD unavailable to DLDH, resulting in a deficiency of DLDH. Yeast with mitochondrial FAD transport defect also showed diminished activity of DLDH [39]. The correct folding of the DLDH depended on the presence of both FAD and NAD [31]. Previous studies have demonstrated that NAD deficiency inhibited the overall activity of GCS by competitive conjugation of apo-DLDH with other subunits of GCS [40]. In this study, the mildly decreased DLDH protein content (decreased 26.8%) and severely impaired GCS activity (reduced 89.4%) in Slc25a32 −/− embryos suggested that FAD deficiency should have the same effect as NAD deficiency on GCS.
Unlike the Slc25a32 knock-out mice embryos, the Slc25a32 knock-in mice models had no glycine accumulation, suggesting glycine oxidation was unhindered. However, they also suffered from formate deficiency (Fig. 6F) and DMG accumulation (Fig. 6D, E). DMG is an intermediate of choline metabolism, which converts to sarcosine and donates 1C units to the folates cycle under the function of FAD-dependent dimethylglycine dehydrogenase (DMGDH) (Fig. 9) [41]. In addition to DMG, other 1C donors were also affected by Slc25a32 missense mutations, as indicated by the elevated serine levels and decreased glycine levels in embryos (Fig. 4B). Moreover, high serine levels were also found in the adult Slc25a32 Y174C/K235R mice brain ((14.39 ± 0.65 vs. 16.64 ± 0.50) pmol/mg protein, WT vs. Slc25a32 Y174C/K235R ). The elevated serine and decreased glycine were probably due to an enhanced de novo synthesis of serine from the glycolytic intermediate 3-phosphoglycerate and enhanced utilization of glycine oxidation as an adaptive response to increase 1C donors [42]. The variation of glycine and serine in Slc25a32 knock-in mice embryos was similar to that in methylenetetrahydrofolate dehydrogenase 1-like gene (Mthfd1l)-null mice embryos [42], suggesting mitochondrial folate metabolism was affected in Slc25a32 knock-in mice. Quantification of folate intermediates confirmed the absence of mitochondrial folates (Fig. 3A). The data was consistent with the results obtained from glycine auxotrophy CHO-derived glyB cells [43] which possessed a G192E mutation in the SLC25A32 gene [7]. Mitochondrial folate deficiency might be caused by inhibition of folate production, enhancement of folate removal rate, or less efficient folate utilization [43]. A recent study disclosed that ablating formate generation in mitochondria resulted in folate degradation [44]. Therefore, it is very likely that the Slc25a32 missense mutations interrupted the choline metabolism, leading to a reduction of formate generation, which further enhanced folate removal and resulted in mitochondrial folate deficiency. Experiments to test this hypothesis are underway.
The relationships between mitochondrial folate deficiency and phenotypes of patients with RREI remain undefined. The mice models did not reproduce the phenotypes of clubfoot, fibulae hypoplasia, and lateral ventricle enlargement of our patient. However, previous studies already revealed a strong link between folate deficiency and clubfoot and ventriculomegaly in humans [45][46][47][48][49], which provided insights into the cause of clubfeet in our patient. In fact, the etiopathogenesis of congenital clubfoot is still unknown regardless of its long-lasting recognition. More recently, the clubfoot was proposed to be caused by a defect of both the radial and the longitudinal growth unevenly affecting the leg muscles with a consequent imbalance of the foot activators [50]. Interestingly, folate was found to play roles in skeletal muscle cell development and function [51].
Mitochondrial FAD deficiency could be caused not only by SLC25A32 dysfunction, but also by other inborn errors, such as electron transportation defects [52], riboflavin transport disorders [53], and FAD synthase deficiency [54]. Since SLC25A32 dysfunction-caused mitochondrial FAD deficiency resulted in folate shortage, other inborn errors would also cause similar metabolic abnormalities. Therefore, it is valuable to clarify the role of folate shortage in the morphological change of humans.
As summarized in Fig. 9, our study demonstrated that SLC25A32 was a mitochondrial FAD transporter instead of a folate transporter. SLC25A32 dysfunction blocked FAD entry into mitochondria, resulting in mitochondrial FAD shortage and impairment of mitochondrial acyl-CoA dehydrogenases and electron transfer chain, which further damaged muscle and brain function. SLC25A32 dysfunction also disrupted FOCM by diminishing the generation of 1C unit from mitochondria, leading to NTDs in embryos.