De novo ARF3 variants cause neurodevelopmental disorder with brain abnormality

Abstract

An optimal Golgi transport system is important for mammalian cells. The adenosine diphosphate (ADP) ribosylation factors (ARF) are key proteins for regulating cargo sorting at the Golgi network. In this family, ARF3 mainly works at the trans-Golgi network (TGN), and no ARF3-related phenotypes have yet been described in humans. We here report the clinical and genetic evaluations of two unrelated children with de novo pathogenic variants in the ARF3 gene: c.200A > T (p.Asp67Val) and c.296G > T (p.Arg99Leu). Although the affected individuals presented commonly with developmental delay, epilepsy and brain abnormalities, there were differences in severity, clinical course and brain lesions. In vitro subcellular localization assays revealed that the p.Arg99Leu mutant localized to Golgi apparatus, similar to the wild-type, whereas the p.Asp67Val mutant tended to show a disperse cytosolic pattern together with abnormally dispersed Golgi localization, similar to that observed in a known dominant negative variant (p.Thr31Asn). Pull-down assays revealed that the p.Asp67Val had a loss-of-function effect and the p.Arg99Leu variant had increased binding of the adaptor protein, Golgi-localized, γ-adaptin ear-containing, ARF-binding protein 1 (GGA1), supporting the gain of function. Furthermore, in vivo studies revealed that p.Asp67Val transfection led to lethality in flies. In contrast, flies expressing p.Arg99Leu had abnormal rough eye, as observed in the gain-of-function variant p.Gln71Leu. These data indicate that two ARF3 variants, the possibly loss-of-function p.Asp67Val and the gain-of-function p.Arg99Leu, both impair the Golgi transport system. Therefore, it may not be unreasonable that they showed different clinical features like diffuse brain atrophy (p.Asp67Val) and cerebellar hypoplasia (p.Arg99Leu).

Introduction

The Golgi apparatus plays fundamental roles in cell homeostasis through the processing, sorting and transporting of newly synthesized proteins into targeted compartments, and the recycling of proteins to be returned (1,2). The trans-Golgi network (TGN) is regarded as a hub in which cargoes are sorted with distinct transport carriers for post-Golgi compartments. The key proteins involved in cargo sorting at the TGN are adenosine diphosphate (ADP) ribosylation factors (ARFs), cytosolic cargo adaptor proteins, coat proteins and accessory proteins (1). ARFs recruit these adaptor proteins to the TGN membranes and assemble the cargoes at the start of the transport process.

Accumulating evidence indicates that pathogenic variants in several genes encoding Golgi-associated proteins can cause microcephaly, white matter defects and intellectual disability, which are collectively known as ‘Golgipathies’ (3). To date, Ras-like proteins in brain (RAB; in the Ras sarcoma oncoproteins [RAS] superfamily) and RAB-associated proteins, some of which play roles in Golgi apparatus trafficking or morphology, have been well studied in Cohen syndrome (MIM: 216550) (VPA13B, MIM: 607817) (4), progressive cerebello-cerebral atrophy type 2 syndrome (MIM: 615851) (VPS53, MIM: 615850) (5) and Warburg Micro syndrome (MIM: 600118) (RAB3GAP1, MIM: 602536) (6). In addition, pathogenic variants in ARF1 were also reported (NM_001024226.2, MIM: 103180) associated with brain abnormality (7). However, it remains unknown whether abnormalities of ARF3, which belong to the RAS superfamily, also cause human diseases.

ARFs belong to a large superfamily known as the small GTP-binding protein (G protein) superfamily (8,9), which is divided into five superfamilies: RAS, RAB, Ras-like nuclear proteins (RAN), Ras homologous proteins (RHO) and ARF. Members of the ARF superfamily regulate various aspects of membrane traffic in mammalian tissue, including in neurons (1,2). There are five homologs of ARF family genes in humans, which are divided into three classes: class I: ARF1 and ARF3, class II: ARF4 and ARF5 and class III: ARF6. There is no ARF2 in humans. ARF3 localizes at the Golgi apparatus and is mainly involved in transportation at the TGN (10). ARF3 switches between an active GTP-bound form and an inactive GDP-bound form, which are regulated by guanine nucleotide exchange factors (GEFs) and GTPase-activating proteins (GAPs), respectively. An intrinsically weak GTP hydrolysis activity of small G proteins is stimulated by GAPs, which are important regulators of signaling pathways mediated by small G proteins (11). These switch systems are involved in various cellular signaling pathways. At the onset of transportation at the TGN, the GTP-bound active form of ARF3 recruits Golgi-localized, γ-adaptin ear-containing, ARF-binding protein1 (GGA1) to the TGN membrane, forming vesicle cargoes for trafficking (12,13).

In the present study, using trio-based whole-exome sequencing (WES) for two unrelated children, we identified two de novo variants in the ARF3 gene (MIM:103190), NM_001659.2: c.200A > T (p.Asp67Val) and c.296G > T (p.Arg99Leu). We then demonstrated some functional abnormalities of these ARF3 variants.

Results

Clinical features

Individuals 1 (Turkish) and 2 (Japanese) were unrelated affected females born to respective non-consanguineous parents (Fig. 1A and D). No family histories were noted in both families. Common clinical features between the two individuals included developmental delay, epilepsy and brain abnormalities. However, Individual 1 was more severely affected than Individual 2 (Table 1). In particular, Individual 1 showed progressive cerebral and brainstem atrophy and severe developmental delay, whereas Individual 2 showed non-progressive cerebellar hypoplasia. Common facial features, such as broad noses, full cheeks, short philtrum, strabismus and thin upper lips, were recognized between the two individuals (Fig. 1B, C, E and F).

Figure 1

The ARF3 variants and facial photographs of Individuals 1 and 2. (A and D) Familial pedigrees of the unrelated individuals 1 and 2. In both families, each individual had a de novo heterozygous variant (wild-type/mutant; WT/MT). Facial photographs of Individual 1 at the age of 2 months (B) and at 2 years (C) and of Individual 2 at 4 years (E) and at 14 years (F). (G) EEG in Individual 1 at 2 years of age showed spikes in the bilateral frontal and temporal regions. (H) EEG in Individual 2 at 12 years of age showed intermittent spikes in the parieto-occipital region and 2.5–3.5 Hz spike-and-wave discharges with right frontal predominance. (I) Electropherograms of ARF3 variants in the two families. The corresponding nucleotides and amino acid residues are shown under the electropherograms. (J) Schematic presentation of the pathogenic variants above the ARF3 gene structure. Upper shows gene structure of ARF3 with exon numbers, and lower indicates ARF3 protein, and functional domain (light blue) and motifs (green).

Table 1

Clinical features of affected individuals with pathogenic ARF3 variants

Cases	Individual 1	Individual 2
ARF3 variant	c.200A > T	c.296G > T
Protein alteration	p.Asp67Val	p.Arg99Leu
Sex	Female	Female
Ethnicity	Turkey	Japanese
Age at evaluation	4 years and 10 months	14 years
Consanguinity	−	−
Birth data
BW or HT below <5 percentile	−	−
Microcephaly	+ (−3.3 SD)	−
Onset
Age	Prenatal a	4 mo
Primary symptom	Microcephaly	Developmental delay
Clinical features
Developmental delay	Severe	Severe
DQ or IQ	N.A	IQ 23
Regression	−	−
Motor impairment	No independent sit	Independent walk
	No roll over	No run, no jump
Speech impairment	Utter no words	Utter some meaningful words
Hypotonia	−	−
Spasticity	+	−
Hyperreflexia	+	−
Cerebellar ataxia	−	−
Dysarthria	−	+
Epilepsy	+	+
Microcephaly	+ (Progressive, −5.7 SD)	−
Dysmorphic features
Broad nose	+	+
Full cheeks	+	+
Short philtrum	+	+
Strabismus	+ (Internal strabismus)	+ (External strabismus)
Thin upper lips	+	+
	Micrognathia and others b	Prognathia
Systemic involvement (metabolic, renal, liver,  and autonomic deregulation)	−	−
MRI findings
Cerebellum atrophy	+	−
Brainstem atrophy	+	−
Cerebellar hypoplasia	−	+
Outcome	Alive	Alive

^aMicrocephaly was found by ultrasound at a gestational age of 32 weeks.

^bFacial hypertrichosis, flesh ear lobe

Abbreviation, BW: body weight, HT: height, mo: months, N.A: not available

In Individual 1, microcephaly was suspected by intra-uterine ultrasonography at a gestational age of 32 weeks. At normal birth, her occipitofrontal circumference (OFC) was 30.0 cm (−3.3 standard deviation [SD]). At the age of 2 months, she visited our hospital because of microcephaly and had an OFC of 34 cm (−3.5 SD), with dysmorphic facial features including a broad nose, hypertrichosis, micrognathia, full cheeks and fleshy ear lobes. Blood and cerebrospinal fluid examinations were all normal. Brain magnetic resonance imaging (MRI) showed cerebral atrophy (Fig. 2A–C). At the age of 8 months, she had her first seizure and valproic acid (VPA) treatment was initiated. Her epilepsy remains controlled. Abnormal electroencephalogram (EEG) showed mild epileptic waves (Fig. 1G). Brain MRI at the age of 2 years and 10 months showed progressive cerebral, brainstem atrophy and delayed myelination with no cerebellar abnormalities (Fig. 2D–F). She had global developmental delay: head control at 3 months, sitting with support at 8 months, eye tracking and social smiling at 3 years. She is currently aged 4 years and 10 months, and has been seizure-free only by VPA for 1 year; however, progressive microcephaly [OFC of 41.5 cm (−5.7 SD)] is noted. Her latest height and weight at the age of 3 years and 9 months are 108 cm (+1.59 SD) and 13 kg (−1.47 SD), respectively, and there are no major organ malformations or endocrine dysfunctions. She cannot roll over, crawl or sit without support. She uses her hand to refuse feeding and utters no words.

Figure 2

Brains MRI of the two individuals. Brain MRI of Individual 1 (A–C at the age of 2 months and D–F at 2 years) and Individual 2 (G–I at 2 years and J–L at 6 years). T1-weighted images (A–F, H and K) and T2-weighted images (G, I, J, L) are shown. Individual 1 showed progressive cerebral and brainstem atrophy, and Individual 2 showed cerebellar hypoplasia.

Individual 2 was born via caesarean section because of abnormal rotation with no asphyxia. She showed developmental delay at the age of 7 months and was unable to roll over. Her developmental delay was examined at a hospital at the age of 1 year, but its causality was undetermined. At age 2 years and 5 months, she visited our hospital. She was able to stand with support (but not alone) but was unable to walk. She spoke no meaningful words and showed truncal hypotonia. EEG showed spike-and-slow waves at the right side of the frontal lobe. Furthermore, brain MRI revealed cerebellar hypoplasia and delayed myelination (Fig. 2G–I). At the age of 6 years, the cerebellar hypoplasia had not deteriorated, and myelination was complete (Fig. 2J–L). She showed global developmental delay: sitting at 8 months, crawling at 2 years, walking at 3 years, no running or jumping. She spoke meaningful words at 4 years. She had episodic intractable epilepsy: generalized onset atonic seizures at 6 years, which were well controlled by VPA; focal onset seizures weekly at 11 years; and generalized tonic–clonic seizures monthly at 12 years, for which levetiracetam was added. Her EEG at 12 years of age is presented in Figure 1H. She currently has seizures monthly, and perampanel has been added. At her current 15 years, her height and weight are 153.7 cm (−0.7 SD) and 51.1 kg (−0.1 SD), respectively, and there are no major organ malformations or endocrine dysfunctions. She can walk and eat by herself, but needs support for changing her clothes and egestion. She speaks less than 10 words and attends a special school. She shows no autistic behaviors. Her epilepsy is intractable, and her intelligence quotient was 23 at 12 years.

Genetic analysis

We performed trio-based WES with >91% coverage of coding sequences by >20× reads in all familial members tested, and identified the following de novo ARF3 variants: NM_001659.2: c.200A > T (p.Asp67Val) in Individual 1 and c.296G > T (p.Arg99Leu) in Individual 2 (Fig. 1). Sanger sequencing also confirmed the de novo occurrence of the variants in both families (Fig. 1I). Both variants were absent from the Human Genetic Variation Database, 575 in-house Japanese control genomes and gnomAD. Additionally, web-based tools, including SIFT (14), PolyPhen2 (15), Mutation Taster and CADD (16), predicted the two variants as disease-causing. (Table 2). No other candidate variants related to the disease were detected. The two ARF3 variants identified in the present study reside within the Arf domain, as predicted by Pfam, with unknown function (Fig. 1J). ARF3 has five guanine nucleotide binding motifs, which are highly preserved (17), and c.200A > T (p.Asp67Val) is located in one of them.

Table 2

Evaluation of novel ARF3 variants using web-based tools and databases

Variant	Annotation	SIFT	Poly-Phen2	Mutation Taster	CADD	HGVD	in-house	gnomAD
								Allele count	Allele frequency
Novel variants in this study
c.200A > T (p.Asp67Val)	Missense	0	0.997	Disease causing	32	0	0	0	0
c.296G > T (p.Arg99Leu)	Missense	0.006	0.913	Disease causing	34	0	0	0	0

Web-based prediction tools predicted that the two variants are disease-causing (SIFT: < 0.1, highly conserved position (14); PolyPhen2: > 0.85, probably damaging (15); CADD: > 23, judged as pathogenic (16)). The two variants were absent from the Human Genetic Variation Database (HGVD), an in-house database of 575 healthy Japanese controls and gnomAD. The URLs of the prediction tools and control databases are available in the Web Resources section.

ARF3 variants affecting protein functions

Previous studies have demonstrated that wild-type ARF3 localizes at the Golgi apparatus with crescent- or semilunar-like shapes, while a dominant negative variant (p.Thr31Asn) is redistributed from the Golgi to the cytoplasm (12,18,19). Moreover, overexpression of this variant results in dispersed Golgi morphology. To evaluate whether the two identified ARF3 variants were involved in Golgi functions, we initially investigated the effects of ARF3 overexpression on Golgi morphology. Immunofluorescence experiments using ARF-green fluorescent protein (GFP) and a Golgi matrix protein, GM130, as a Golgi marker revealed that wild-type ARF3 was localized to perinuclear regions with crescent-like shapes. Moreover, the regions highly overlapped with those of GM130, as previously reported (Fig. 3A). The p.Arg99Leu mutant had Golgi localization that was identical to that of the wild-type. In contrast, the p.Asp67Val mutant did not show a Golgi-like pattern. Furthermore, GM130 signals were dispersed throughout the cytoplasm, similar to that observed with the dominant negative variant p.Thr31Asn (12,19) (Fig. 3A, arrowheads). A quantitative analysis indicated that the p.Asp67Val variant showed a significantly different pattern of Golgi localization from that of the wild-type, similar to that of the dominant negative p.Thr31Asn variant (Fig. 3B).

Figure 3

Characterization of ARF3 variants in terms of subcellular localization and activity. (A) HeLa cells were transiently transfected with plasmids expressing GFP-tagged ARF3, and were then analyzed by immunofluorescence with anti-GM130 (a Golgi marker, orange) and anti-GFP (green) antibodies. 4′,6-Diamidino-2-phenylindole (DAPI; blue) was used as a DNA stain. Neither the p.Thr31Asn nor the p.Asp67Val variant showed Golgi localizations based on the dispersed localization of GM130 (white arrowheads). Cells that were not transfected with ARF3-GFP showed normal Golgi morphology (white arrows). Scale bars represent 10 μm. (B) Quantification of the Golgi morphology shown in (A). Golgi morphology was classified into three categories: normal (white bar), dispersed (black bar) and unclassified (gray bar). Data represent the mean ± SD from three independent experiments. Statistical comparisons were analyzed using Fisher’s exact test. ^***P < 0.001. n.s.: not significant. (C) Lysates from HEK293T cells transiently expressing V5-tagged ARF3 (input) were treated with or without GTP-γS, and were then incubated with Glutathione Sepharose beads preloaded with GST-GGA1-GAT fused protein. Input and pull-down fractions were analyzed by western blotting. Three independent experiments were performed. (D and E) HEK293T cells transfected with V5-tagged wild-type and p.Asp67Val ARF3 were treated with CHX (80 μg/mL). Lysates at the serial time points were processed for western blotting with the indicated antibodies. Protein band intensities were quantified, and the protein amount of wild-type at time zero (h = 0) was set as 100%. Data represent the mean ± SD from three independent experiments. Statistical comparisons were analyzed using the t-test. ^**P < 0.01, ^***P < 0.001.

The GAT domain of GGA1 interacts with the GTP-bound active form of ARF3. To evaluate the binding activities of wild-type and mutant ARF3 with GGA1, we performed a GST-GGA1 pull-down assay (12,13). GST-GGA1 fusion protein surely bound to p.Arg99Leu without GTP-γS, while GST-GGA1 fusion protein with GTP-γS clearly bound to both wild-type and p.Arg99Leu ARF3, but not to p.Asp67Val ARF3. This finding implies that the p.Arg99Leu ARF3 mutant has increased binding activity with GGA1, together with the wild-type. Thus, p.Arg99Leu is likely a gain-of-function variant (Fig. 3C, arrowhead). In this assay, we also noticed that p.Asp67Val had a lower band intensity and slightly smaller band size in the input fraction detected by the anti-V5 antibody (Fig. 3C, left). To confirm the transcript levels, we performed quantitative polymerase chain reaction (20) and detected similar levels of wild-type and p.Asp67Val transcripts in HEK293T cells with transient transfection of the ARF3 constructs (data not shown). This suggests that the p.Asp67Val protein may be more unstable than the wild-type protein. To address this possibility, we examined the protein stability of ARF3 after cycloheximide (CHX) treatment. As shown in Figure 3D and E, wild-type ARF3 was relatively stable during the 9-h time course, whereas the p.Asp67Val mutant showed rapid protein decay during the same time period, clearly indicating p.Asp67Val instability. A dominant negative variant, p.Thr31Asn, has been speculated based only on experimental data from ARF1 (21), with 96.1% amino acid identity to ARF3 (22); it leads to experimentally dispersed cytoplasmic distribution as well as dispersed Golgi morphology (12,18,19).

Structural considerations

To gain structural insights into the pathogenic mechanisms of the p.Asp67Val and p.Arg99Leu variants, we analyzed the reported co-crystal structure of the Arf3 p.Gln71Leu-GTP complex, where the catalytic Gln71 is mutated to capture the GTP-bound form (23). In this structure, Asp67 forms water-mediated interactions with the bound magnesium ion (Mg²⁺), which in turn is coordinated to the beta- and gamma-phosphate groups of the bound GTP (Fig. 4A). The p.Asp67Val substitution would disrupt this interaction, leading to the compromised binding of GTP or GDP. Notably, Thr31, whose asparagine substitution results in the dominant negative phenotype, also interacts with the magnesium ion (Fig. 4A). In a prototypical small G protein Ras (Fig. 4B) (24) the p.Ser17Asn substitution (corresponding to p.Thr31Asn of Arf3) also displays a dominant negative phenotype (25), and has defective magnesium ion binding (26). Thus, similar to Ras-p.Ser17Asn and Arf3-p.Thr31Asn, Arf3-p.Asp67Val may display a dominant negative phenotype or possibly loss-of-function phenotype caused by the compromised binding of GTP or GDP mediated by the magnesium ion. The variant disrupts the normal interaction between p.Asp67 and magnesium ion, which is required for nucleotide binding, and leads to decrease ARF3 activation. As ARF3 binds GTP and GDP in its large internal pocket, the nucleotide-free p.Asp67Val variant may be conformationally unstable and be degraded by cellular protein quality control systems. This may explain its low abundance shown by our cellular analysis (Fig. 3D).

Figure 4

Structural analysis of ARF3. Structures of Arf3 and related small G proteins. Hydrophilic interactions are indicated by yellow dotted lines. (A) A close-up view of the GTP-bound form of the Arf3 p.Q71L mutant. The ribbon model of Arf3 is colored green. The carbon atoms of Arf3 and GTP are colored green and white, respectively. (B) Left, a close-up view of the Ras-RasGAP complex bound with GDP-AlF₃. The ribbon model of Ras is colored green. The carbon atoms of Ras, RasGAP and GDP are colored green, cyan and white, respectively. For simplicity, only R789 (the catalytic arginine) of RasGAP is shown. The attacking water is shown as a red sphere. The nucleophilic attack is shown as a red arrow. Right, the Ras p.G12D substitution is modeled on the structure of the Ras-RasGAP complex. The sidechain atoms of G12D and Q61 are shown as spheres. The G12D sidechain (colored magenta) would sterically clash with Q61. (C) A close-up view of the Arf6-ArfGAP complex bound with GDP-AlF₃. The ribbon model of Arf6 is colored green. The carbon atoms of Arf6, ArfGAP and GDP are colored green, cyan and white, respectively. For simplicity, only R469 (the catalytic arginine) of ArfGAP is shown. The attacking water is shown as a red sphere. The nucleophilic attack is shown as a red arrow.

The possible gain-of-function phenotype of p.Arg99Leu suggests the impaired GTP hydrolysis activity of ARF3. In the Arf3 p.Gln71Leu-GTP complex, Arg99 interacts with Asp26 (Fig. 4A), which is strictly conserved among the ARF family (17). Notably, at the position corresponding to Arf3 Asp26, Ras has Gly12, which is one of the three residues that are frequently mutated in various cancers (27). Oncogenic Ras Gly12 variants, including p.Gly12Asp, are thought to sterically hinder the catalytic Ras Gln61 from adopting a conformation that stabilizes the attacking water for GTP hydrolysis (Fig. 4B) (24). In contrast, the side chain of Arf3 Asp26 points in a different direction via its interactions with Arg99, so the catalytic Gln71 is able to perform catalysis (Fig. 4A). In the co-crystal structure of the Arf6-ArfGAP-GTP analog complex (capturing the transition state of GTP hydrolysis) (28), Asp22 and Arg95 of Arf6 (corresponding to Asp26 and Arg99 of Arf3, respectively) also interact with each other, making room for the catalytic Arf6 Gln67 (corresponding to Gln71 of Arf3) to stabilize the attacking water (Fig. 4C). Thus, the substituted Arg99Leu residue of Arf3 would fail to stabilize the conformation of Asp26 to be compatible with catalysis, thereby resulting in impaired GTP hydrolysis activity, which was also observed for the Ras p.Gly12Asp mutant (Fig. 4B).

Toxicity evaluation of ARF3 variants in drosophila

To evaluate the effects of the p.Asp67Val and p.Arg99Leu mutations in ARF3 in vivo, we first generated transgenic flies expressing the recombinant wild-type ARF3, the reported dominant negative p.Thr31Asn, the reported constitutive active p.Gln71Leu, or the de novo p.Asp67Val and p.Arg99Leu mutations, each fused with GFP. Each variant was then expressed in a whole-body manner in Drosophila and the protein expressions were confirmed by western blotting. A single band was detected at 47 kD because ARF3 is about 20 kD and GFP is 26.9 kD (Fig. 5A). Only p.Asp67Val showed a band at a slightly smaller size, which was consistent with the results in HEK293T cells in vitro. Using these transgenic fruit flies, we then compared the toxicity of the ARF3 variants using the rough eye phenotype. The rough eye phenotype has been widely used to assess the toxicity of factors involved in neurodegenerative diseases, using the Gal4/upstream activating sequence (UAS) targeted gene expression system (29). We applied this approach by expressing each variant in an eye-specific manner using glass multiple reporter (GMR)-Gal4 (30), to evaluate the rough eye phenotype. The expression of wild-type ARF3 had no effect on the retina, but both p.Gln71Leu and p.Arg99Leu had significantly higher phenotypic scores, indicating increased disorder or symmetry changes in the sequence of the ommatidia, compared with controls. This indicates an increased severity of the rough eye phenotypes (Fig. 5B and C). There was a loss of retinal pigment with p.Gln71Leu, but not with p.Arg99Leu. Interestingly, the expression of eye-specific p.Thr31Asn and p.Asp67Val was lethal in flies. Because Drosophila can survive even when they lose their eyes, the GMR-Gal4 driver is often used to study the mutational effects of disease-related genes. Nevertheless, the lethality in these flies suggests that p.Thr31Asn and p.Asp67Val, which were expressed in very small amounts in tissues outside of the eye, may be very toxic. These variants were lethal even at a temperature of 20°C, when the expression level was lower than at 25°C (data not shown). These results suggest that p.Arg99Leu and p.Asp67Val have different phenotypes, and may thus have different mechanisms of toxicity. Together with the in vitro data, p.Arg99Leu showed a rough eye phenotype that was comparable to that of p.Gln71Leu, suggesting a potentially activated form, whereas p.Asp67Val was lethal, similar to p.Thr31Asn, suggesting a possibly dominant negative variant or loss-of-function.

Figure 5

Toxicity evaluation of ARF3 variants in Drosophila. (A) Western blotting confirmed the expression of the ARF3 variants. ARF3[WT]-GFP, ARF3[D67V]-GFP, ARF3[Q71L]-GFP, ARF3[R99L]-GFP and ARF3[T31N]-GFP were expressed in whole Drosophila bodies and detected with GFP antibodies. A single band of the ARF3-GFP recombinant proteins was detected at 47 kD. α-Tubulin was used as a loading control where indicated. (B) Representative bright-field microscope images of fly eyes displaying eye-specific ectopic expression of ARF3[WT]-GFP, ARF3[Q71L]-GFP and ARF3[R99L]-GFP using the GMR-Gal4 driver, reared at 25°C. Eyes of the ARF3[WT]-expressing control flies showed normal ommatidial organization, while the eyes of flies with expression of ARF3[Q71L] and ARF3[R99L] showed disruption in the morphology of the eye. Expression of ARF3 [T31N]-GFP and ARF3 [D67V] was lethal. (C) Quantification results of the phenotypic scores in ARF3[WT]-GFP (n = 10), ARF3[Q71L]-GFP (n = 10) and ARF3[R99L]-GFP (n = 10). Data represent the mean ± SD. Statistical comparisons were conducted using nonparametric ANOVA (Kruskal–Wallis test) followed by Dunn’s multiple comparisons test. n.s.: not significant. ^***P < 0.0001

Discussion

To date, ARF3-related phenotypes have never been reported in humans. In this study, we described two novel ARF3 variants, c.200A > T (p.Asp67Val) and c.296G > T (p.Arg99Leu), which were identified in two unrelated individuals with developmental delay, epilepsy and brain abnormalities. The clinical and genetic data as well as the results of in vitro and in vivo experiments indicate that these two pathogenic ARF3 variants may cause human diseases in different ways.

Clear clinical differences between the two individuals led to our hypothesis that the two variants might be functionally different. This type of functional diversity, in which different variants in the same gene have different functional effects related to different phenotypes, is well known (31,32). Our in vitro assay suggested that p.Asp67Val had a possibly loss-of-function effect and p.Arg99Leu had a gain-of-function effect. Based on the clinical severity observed in our individuals and the animal model, p.Asp67Val resulted in more severe effects. In evaluating the molecular pathomechanisms of small G proteins and their activity, three factors are important: the GTP hydrolysis rate, the GDP–GTP exchange rate and the affinities of the proteins for GDP or GTP (33). For example, the gain-of-function RAS-p.Gln61Leu variant maintains a constitutively active form by the impaired intrinsic hydrolysis of GTP (34), whereas the gain-of-function RAC1-p.Cys157Tyr variant (RAC1 belongs to the RHO superfamily) maintains an active form by the rapid transition from the GDP-bound state into the GTP-bound state (35). From our in silico structural considerations, the substituted p.Arg99Leu residue of ARF3 is likely to fail to stabilize the conformation of Asp26 that is compatible with catalysis, resulting in impaired GTP hydrolysis activity and gain-of-function activity in p.Arg99Leu ARF3. From our in silico structural considerations, similar to the dominant negative RAS-p.Ser17Asn (24–26) and ARF3-p.Thr31Asn (18,19,21), a defective interaction with the magnesium ion in the ARF3-p.Asp67Val substitution leads to the compromised binding of GTP or GDP, and may result in a loss-of-function or dominant negative phenotype. We have not been able to verify that these p.Thr31Asn and p.Asp67Val have an ability to decrease the activity of wild-type ARF3. In our subcellular localization assay, our variant p.Asp67Val showed similar properties with reported dominant-negative p.Thr31Asn (12,19), though other pull-down and protein stability assays possibly showed that p.Asp67Val has a loss-of-function effect. A reported pathogenic ARF1 variant (ARF1-p.Tyr35His) associated with brain abnormality also showed a loss-of-function effect by GST-GGA3 pull-down assay and was predicted to be deleterious because of its nearby localization to a nucleotide binding site (7). Thus, we could recognize the similarity in pathomechanism between ARF3-p.Asp67Val and ARF1-p.Tyr35His.

Our in vitro and vivo assays, p.Asp67Val showed a smaller band size than p.ARF3-WT-V5. Since the N-terminal is subject to post-translational modification, the C-terminal was tagged in this experiment. The post-translational modification (myristoylation) occurs in the second amino acid, glycine, which is essential for ARF3 to attach to the Golgi membrane. We supposed that the smaller size of p.Aps67Val may be related to N-terminal myristoylation and made p.Gly2Ala and p.Gly2del mutant constructs disrupting the myristoylation site of ARF3, though p.Gly2Ala and p.Gly2del showed the same band size with wild-type (data not shown), suggesting that the small size of p.Asp67Val might not be related to myristroylation.

Our data indicate that Individual 1, with ARF3-p.Asp67Val, fits the description of Golgipathies, which are caused by impaired Golgi trafficking and show progressive microcephaly and white matter defects. In Golgipathies, it is proposed that pathogenic variants in genes associated with altered Golgi trafficking in neurons and oligodendrocytes result in defective synaptic function and myelination raising progressive microcephaly and white matter defects (3). In mice and rats, brefeldin A, which inhibits GEFs, can be used to block ARFs. This leads to the disassembly of the Golgi apparatus, causing impaired Golgi trafficking and axonal growth retardation (36,37)^. Therefore, loss-of-function ARF3 variants, which may include p.Asp67Val, have deteriorating effects on brain growth and development.

Of note, Golgipathies are usually related to postnatal onset microcephaly (POM) rather than congenital microcephaly (3). However, Individual 1, with p.Asp67Val, showed congenital microcephaly. The main pathomechanisms of POM are likely related to abnormalities in myelination, synaptic pruning and gliogenesis, whereas those of congenital microcephaly are related to cortical neurogenesis, migration and gliogenesis. Therefore, Golgi trafficking associated with ARF3 might be essential for the formation of fetal and pediatric brain development.

In conclusion, we identified two pathogenic ARF3 missense variants causing neurodevelopmental disorders. It is possible that the loss-of-function c.200A > T (p.Asp67Val) has more severe effects than the gain-of-function c.296G > T (p.Arg99Leu) in humans and animals. Further studies that include more individuals with pathogenic ARF3 variants are needed to define more solid phenotype–genotype correlations.

Materials and Methods

Genetic analysis

This study was approved by the Institutional Review Board of the Yokohama City University Faculty of Medicine as previously described (38). Peripheral blood leukocytes were collected from the affected individuals and their parents after obtaining written informed consent. Detailed clinical information was obtained by the clinicians seeing the individuals. Trio-based whole exome sequencing (WES) was performed and data were analyzed in each family as previously described (38). Common single-nucleotide polymorphisms with minor allele frequencies (≥1%) in dbSNP 137 and variants that were observed in > 5 of our 575 in-house Japanese control exomes were excluded from the candidate variants. Among the remaining rare variants, we focused on amino acid-altering variants or variants upstream/downstream of exon–intron borders (±20 bp). Particular attention was paid to variants in known mutated genes that are associated with neurodevelopmental disorders. The candidate variants were validated by Sanger sequencing on an ABI Prism 3500xL autosequencer (Life Technologies, Carlsbad, CA, USA) using genomic DNA from the affected individuals and their parents.

Plasmid construction

We amplified the full-length cDNA of human ARF3 from HeLa cDNA libraries by PCR with gene-specific primers, which were introduced to include the attB Gateway Cloning sites. After agarose gel electrophoresis, objective bands were cut out and DNA was extracted from the agarose gel and cloned into a pDONOR vector using Gateway Cloning (Thermo Fisher Scientific, Waltham, MA, USA). To introduce a C-terminal green fluorescent protein (GFP)-tag or C-terminal V5-tag for ARF3, we used pcDNA-DEST47 and pcDNA-DEST40 vectors, respectively (Thermo Fisher Scientific). All variants were created using a KOD -Plus- Mutagenesis Kit (Toyobo, Osaka, Japan). For the glutathione S-transferase (GST)-pull-down assay, we constructed a GST-fused GAT domain of GGA1 as previously described (12). Like the ARF3 constructs, an objective cDNA fragment encoding GAT domain of Golgi-localized, γ-adaptin ear-containing, ARF-binding protein1 (GGA1) was cut out and subcloned into the pGEX-6P-1 vector (Cytiva, Tokyo, Japan). The GST-fusion proteins were expressed in E. coli BL21(DE3) and purified using Glutathione-Sepharose 4B beads (Cytiva) as previously described (39).

ARF3 cellular localization and Golgi morphology assays

HeLa cells were seeded at a density of 5 × 10⁴ in 24-well plates on 13 mm glass coverslips. After 24 h of culture, cells were transfected with the ARF3-GFP constructs using ViaFect Transfection Reagent (Promega, Madison, WI, USA). At 48 h after transfection, immunostaining was performed as described previously, with a slight modification (40). Briefly, cells were fixed with 2% paraformaldehyde, permeabilized with 0.1% Triton X-100 in phosphate-buffered saline, blocked with 10% normal goat serum and stained with rabbit anti-GFP (ab6556, Abcam, Cambridge, UK) and mouse anti-GM130 (610 822, BD Biosciences, San Jose, CA, USA) antibodies overnight at 4°C. This was followed by incubation with goat anti-rabbit Alexa Fluor 488 and goat anti-mouse Alexa Fluor 555 (both from Thermo Fisher Scientific). Cells were then mounted with ProLong™ Diamond Antifade Mountant with 4′,6-diamidino-2-phenylindole (DAPI; Thermo Fisher Scientific). Photos were captured on a Zeiss LSM 980, and the image files were processed using Zen software (Carl Zeiss Microscopy, Jena, Germany).

Pull-down assay

The GAT domain of GGA1 interacts with the GTP-bound active form of ARF3 (12,13). To evaluate the ability of wild-type and mutant ARF3 to bind to a GGA1 adaptor protein with/without GTP-γS, the GST-GGA1 pull-down assay was performed as previously described (12). HEK293T cells grown on a 6 cm dish were transfected with a C-terminal V5-tagged ARF3 expression vector (ARF3-V5) using X-tremeGENE 9 DNA transfection reagent (Roche, Basel, Switzerland). After 48 h, the cells were lysed in 0.3 mL of cell lysis buffer (25 mM HEPES/KOH pH 7.2, 1 mM ethylenediaminetetraacetic acid [EDTA], 100 mM NaCl, 1 mM MgCl₂, 0.1% Triton X-100) containing a complete EDTA-free protease inhibitor (Roche) and 1 mM dithiothreitol. They were then centrifuged at 14000 rpm (17 800 g) for 15 min at 4°C. The supernatant was added with or without 200 μM GTP-γS, and was then incubated for 30 min at 35°C. Next, the lysate was incubated with Glutathione-Sepharose 4B beads (Cytiva, Marlborough, MA, USA) preloaded with the GST-GGA1-GAT fused protein (25 μg) for 30 min at room temperature. The beads were then washed three times with the wash buffer (50 mM Tris-Cl, pH 7.5; 100 mM NaCl; 2 mM MgCl₂; 10% glycerol; and 1% NP-40). The bound materials were eluted by boiling in a sodium dodecyl sulphate-polyacrylamide gel electrophoresis sample buffer and subjected to western blot analysis.

ARF3 transcripts in transiently transfected cells

Wild-type and p.Asp67Val ARF3 constructs were transiently transfected into HEK293T cells using X-tremeGENE9 DNA transfection reagent (Roche). Forty-eight hours after transfection, total RNAs were extracted using the RNeasy Plus Mini Kit (Qiagen, Hilden, Germany). ARF3 and ACTB (as an internal control) mRNA was measured by quantitative PCR using iTaq (Bio-Rad, Hercules, CA, USA) and normalized with ACTB mRNA (20). Quantitative PCR was performed using a Rotor-Gene Q (Qiagen) and analyzed with the ΔΔCt method using Rotor-Gene 6000 Series software (Qiagen). Three independent experiments were performed. Information of primer sequences and PCR conditions is available on request.

Protein stability analysis

To evaluate protein stability, wild-type and p.Asp67Val ARF3 constructs were transiently transfected into HEK293T cells. Forty-eight hours after transfection, cells were treated with or without cycloheximide (CHX) at 80 μg/mL. At the timepoint of zero, cells with no treatment were collected with trypsin. Other cells were collected either 1, 3, 6 or 9 h after CHX treatment. All of the collected cells were lysed with lysis buffer, the same as for the preparation of lysate for the GST pull-down assay. After being centrifuged at 14 000 rpm (17 800 g) for 15 min at 4°C, the supernatants were subjected to western blot analysis.

Structural considerations

Structural considerations and figure preparations were performed using the program PyMOL (Schrödinger, Inc., New York, NY, USA). Although the structure of the Arf3 p.Gln71Leu-GTP complex has been reported (23), no structure of the Arf3-GAP complex was available. Thus, to gain insight into GTP hydrolysis mechanisms by Arf proteins, the structures of Arf6 (another Arf family protein) and Ras (a prototypical G protein) in their complexes with GAP proteins were also analyzed (24,28). The Mutagenesis Wizard of PyMOL was used to create the model structure of the Ras p.Gly12Asp mutant.

Fly strains and generation of the fly lines for expression of the ARF3 variants

Flies were maintained at 25°C on standard fly food. GMR-Gal4 (#1104), Tub-GAL80^ts (#7019) and Tub-Gal4 (#5138) were obtained from the BDRC (Bloomington, IN, USA). To express ARF3, vectors were constructed using Gateway technology (as mentioned in the plasmid construction section). All mutants were produced using KOD mutagenesis. They were then inserted into pUASTattB vectors (VectorBuilder Japan, Yokoohama, Japan). These plasmids were injected to embryos and inserted into the aTTP40 landing site (WellGenetics, Taipei, Taiwan).

Western blot analysis for the whole body of drosophila

Protein expression was temporarily carried out in the whole bodies by combining Tub-Gal4 and Tub-GAL80^ts. The flies crossed with each ARF3 variant were raised at the permissive temperature (20°C), and newly eclosed females were held at the restrictive temperature (29°C). After 3 days, the dissected brains (n = 10) were homogenized using a BioMasher II (Nippi, Tokyo, Japan) in lysis buffer (10 mM Tris–HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 2% DDM) supplemented with Protease Inhibitor Cocktail Set III (Calbiochem, La Jolla, CA, USA). rabbit anti-GFP (A11122, 1:5000; Thermo Fisher Scientific) and mouse anti-α-tubulin (T9026, 1:10000; Sigma-Aldrich, St. Louis, MO, USA) antibodies were used for western blotting.

Eye imaging using bright-field microscopy and the quantification of morphological defects in the eye

Each ARF3 variant was expressed by the GMR-Gal4 driver. One-day-old flies were imaged using an OM-D E-M5 (Olympus, Tokyo, Japan) connected to an Olympus SZX16 microscope with 8× magnification. Twenty to thirty photographs were taken with the focus shifted slightly. Each slice was depth-synthesized and then trimmed at the edges of the eyes using Photoshop CC 2017 (Adobe, San Jose, CA, USA). Finally, the phenotypic scores were calculated using Flynotyper (41). Experimental analyses were performed using Prism 9 (GraphPad Software Inc. San Diego, CA, USA). Statistical comparisons were conducted using nonparametric analysis of variance (Kruskal–Wallis test) followed by Dunn’s multiple comparisons test.

Web Resources

CADD, https://cadd.gs.washington.edu/

dbSNP, https://www.ncbi.nlm.nih.gov/snp/

Flynotyper http://flynotyper.sourceforge.net/

gnomAD, https://gnomad.broadinstitute.org/

Human Genetic Variation Database, http://www.hgvd.genome.med.kyoto-u.ac.jp/

Mutation Taster, http://www.mutationtaster.org/

Pfam, http://pfam.xfam.org/protein/Q06265

PolyPhen2, http://genetics.bwh.harvard.edu/pph2/

SIFT, https://sift.bii.a-star.edu.sg/

Acknowledgements

We thank the individuals and their families for their participation in this study. We thank Dr Kazuhisa Nakayama and Dr Hye-Won Shin at the Department of Physiological Chemistry, Graduate School of Pharmaceutical Sciences, Kyoto University for their valuable advice for the ARF3 in vitro studies. We thank N. Watanabe, T. Miyama, M. Sato, S. Sugimoto and K. Takabe for their excellent technical assistance. We also thank Bronwen Gardner, PhD, from Edanz Group (https://en-author-services.edanz.com/ac) for editing a draft of this manuscript. Conflict of Interest statement. The authors declare no competing interests.

Funding

Japan Agency for Medical Research and Development (AMED) (JP21ek0109486, JP21ek0109549, JP21ek0109493 to N.M.); Japan Society for the Promotion of Sciences (JSPS) KAKENHI (JP19H03621 to N.M., JP20K07907 to S.M., JP20K08164 to T.M., JP20K17936 to A.F., JP20K16932 to K.H., JP19K17865 to Y.U., JP20K17428 to N.T.; intramural grants (30-6, 30-7) of NCNP from the Ministry of Health, Labour and Welfare to N.M.; the Takeda Science Foundationto N.M., T.M. and N.M.; Yokohama Academic Foundation (776to M.S.).