Biallelic Truncating Mutations in FMN2, Encoding the Actin-Regulatory Protein Formin 2, Cause Nonsyndromic Autosomal-Recessive Intellectual Disability

Abstract

Dendritic spines represent the major site of neuronal activity in the brain; they serve as the receiving point for neurotransmitters and undergo rapid activity-dependent morphological changes that correlate with learning and memory. Using a combination of homozygosity mapping and next-generation sequencing in two consanguineous families affected by nonsyndromic autosomal-recessive intellectual disability, we identified truncating mutations in formin 2 (FMN2), encoding a protein that belongs to the formin family of actin cytoskeleton nucleation factors and is highly expressed in the maturing brain. We found that FMN2 localizes to punctae along dendrites and that germline inactivation of mouse Fmn2 resulted in animals with decreased spine density; such mice were previously demonstrated to have a conditioned fear-learning defect. Furthermore, patient neural cells derived from induced pluripotent stem cells showed correlated decreased synaptic density. Thus, FMN2 mutations link intellectual disability either directly or indirectly to the regulation of actin-mediated synaptic spine density.

Main Text

Intellectual disability (ID), or early cognitive impairment, is a common neurodevelopmental disorder that is estimated to affect >1% of the population.¹ Globally, ID remains a healthcare challenge as well as a socioeconomic burden in both developing and developed countries, and it has immense effects on individuals and their families. Although ID is defined by cognitive deficits (IQ < 70) and limitations in adaptive behaviors (ICD-10), cases can also be classified according to accompanying radiologic, morphologic, and/or metabolic syndromes. When classification is made on the basis of such accompanying syndromes, the disorder is known as syndromic ID (S-ID); in the absence of other accompanying syndromes, it is known as nonsyndromic ID (NS-ID [MIM 249500]). Genetic heterogeneity, mode-of-inheritance heterogeneity, and gene-gene or gene-environment interactions in NS-ID have made identification of causes difficult, but more than 50 genes are now known to be mutated in a biallelic fashion in autosomal-recessive ID (AR-ID).^2,3

Accumulating evidence suggests that impaired synapse formation and synaptic plasticity might underlie ID: neuropathological examination of brains from affected individuals consistently shows abnormalities in the dendritic and synaptic organization of the cortex.⁴ Furthermore, many of the NS-ID encoded proteins localize at or near neuronal synapses, and most animal models show defective synaptic structure or function. For instance, synaptic spine morphological defects are linked with inactivation of murine Fmr1.⁵

Synaptogenesis initiates during human fetal life, peaks in childhood, reaches a steady state in adolescence⁶ and for some brain regions such as the hippocampus, continues into adulthood. Modeling neurodevelopmental disease with human induced pluripotent stem cells (hIPSCs) has shown neuronal structural defects, including alterations in synaptic density. For example, hIPSC ID models, including those involving CDKL5 (MIM 300203) mutations, MECP2 (MIM 300005) mutations, and Down syndrome (MIM 190685), show reduced synaptic density,^7–9 consistent with defects in synapses as an underlying substrate.

In this study, we performed homozygosity mapping and linkage analysis, followed by next-generation sequencing in two independent, consanguineous ID-affected families from Pakistan and Egypt. Participants were part of larger cohorts originating from the Consortium for the Study of Autosomal-Recessive Intellectual Disabilities (CARID). These cohorts consisted of approximately 2,000 families, most with documented parental consanguinity and two or more affected members with moderate to severe NS-ID, presumed recessive in nature; exome sequencing was performed or is being performed on most of these families. Participants with major dysmorphism, structural brain defects, syndromic features, metabolic derangements, or demonstrable cytogenetic abnormalities were excluded. The study was approved by the ethics committees of the participating institutions, the procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national), and proper informed consent was obtained.

Extracted blood-derived DNA was subjected to a 5K SNP array with the Illumina Linkage IVb mapping panel for all informative members and analyzed with easyLinkage-Plus software for calculation of multipoint LOD scores (for ARID-628); the Affymetrix 500K Nsp GeneChip SNP mapping panel was used for the three affected offspring and one healthy offspring, and DNA was analyzed for homozygosity with dCHIP and HomozygosityMapper (for ARID-49, Figures 1A and 1D).^10–12 ARID-628 demonstrated the major linkage peak at chr1: q43–44, delineated by rs1481141–rs6426327 (or chr1: 238,408,184–246,089,811) (hg19), with a multipoint pLOD of 2.30, which is similar to the predicted pLOD score¹³ (Figure S1 in the Supplemental Data available with this article online). Several narrower peaks of similar significance were located on other autosomes. ARID-49 demonstrated two blocks of homozygosity, the major one delineated by rs6702864–rs12137158 (or chr1: 230,161,767–240,660,818) and another delineated by rs1019258–rs7588049 (or chr2: 178,012,446–178,390,778) (Figure S2). There was no shared haplotype between the two families, and although the phenotypes were similarly severe with respect to ID, there were some apparent differences. Nevertheless, the two families shared a minimal candidate interval defined as chr1: 238,408,184–240,660,818, or 2.25 Mb, containing three genes, CHRM2, FMN2, and GREM2 (MIM 118493, 606373, and 608832, respectively), and the long intergenic non-protein-coding RNA LINC01139.

Figure 1.

ARID-628 and ARID-49 with Homozygous Inactivating Mutations in FMN2

(A and D) Consanguineous parents and multiple affected individuals suggest recessive inheritance.

(B and E) Homozygous exome haplotypes containing FMN2 in telomeric chromosome 1 (arrows).

(C and F) ARID-628 homozygous chr1:240,256,799insG (arrow) and ARID-49 showing major-allele control TTCAAACGA, minor-allele control (MA) TTCAAACCA, and diseased TTCAACGA (arrow) chr1:240,370,627delA.

(G) Genomic organization and location of mutations for chr1q43–44, including FMN2, from the UCSC Genome Browser; mutations are not predicted to interfere with shorter transcripts.

(H) FMN2 encodes a 1,726 residue protein with DEP, FH1, and FH2 domains, several putative sites of phosphorylation (small vertical ticks), and an FMN2-SPIRE1-interacting domain (FSI).

(I) RT-PCR of FMN2 primary fibroblasts shows that products are absent in IV-4 but present in the control (C), III-2, and GAPDH fibroblasts, suggesting nonsense-mediated decay. Primer sequencing: FMN2-F, 5′-CCGTCTCAGTCCCCTAATCA-3′; FMN2-R, 5′-ATCCGGGAGCAAAACTTCTC-3′; GAPDH-F, 5′-ATCCCATCACCATCTTCCAG-3′; and GAPDH-R, 5′-CCATCACGCCACAGTTTCC-3′.

(J) An immunoblot of FMN2 in primary fibroblasts shows an absence of protein in IV-4 but presence in the control, III-2. GAPDH was used as a loading control. Antibodies: FMN2, Sigma HPA004937, 1:1,000; and GAPDH, Millipore mab374, 1:5,000.

We generated whole-exome sequence for two affected individuals from ARID-628 and for one individual from ARID-49 by using the Agilent SureSelect Human All Exome 50 Mb kit. We performed sequencing with an Illumina HiSeq2000 or ABI SOLiD 5500 instrument, resulting in ∼94% recovery at >10× coverage. GATK was used for SNP and INDEL variant identification, and SeattleSeq was used for annotation; variants were then filtered with >1% allele frequency according to presence in dbSNP, the Exome Variant Server, or in our in-house exome data set of more than 2,000 individuals. In addition, they were filtered if annotated as benign according to functional prediction scores (PolyPhen, Grantham, Phastcon, GERP).^14,15 We analyzed variants from both families with HomozygosityMapper to confirm linkage intervals (Figures 1B and 1E).¹⁶ We tested segregation of all exonic homozygous rare variants predicted to be deleterious, and we only report the variant as causative if it is the single segregating variant. Affected members of both families harbored a homozygous predicted frameshift variant in FMN2 (MIM 606373), encoding formin 2 (FMN2) (Figures 1C and 1F). None of the other variants considered as candidates showed patterns consistent with a recessive mode of inheritance for affectation status (Tables S1 and S2).

ARID-628 was recruited from Lower Egypt as a consanguineous pedigree with two affected members, aged 16 and 14 years (Table 1). The children were delivered at term but experienced delayed milestones, primarily in cognition and speech, whereas gross-motor and fine-motor development were not as dramatically affected. IQ was 50–60, tested in the teenage years, and both children remain illiterate past adolescence and speak in only 1–2 word phrases. Independent gait was obtained at 2 years. Rare partial complex seizures developed at age 10 years, controlled with carbamazepine. The older individual has a history of asymptomatic mitral valve prolapse. ARID-49 was recruited from a remote village in Azad Jammu and Kashmir, Pakistan as a consanguineous pedigree with three affected members, aged 30, 25, and 17 years, and three healthy siblings. All were born after normal pregnancies. Individual IV-1 was independent in self-care, and speech was well developed, but he was unable to read or write and has never been gainfully employed. He frequently disappeared from home for days at a time, wandered in the local area, and functioned in the mildly intellectually disabled range. The other two affected members never developed speech beyond 2 or 3 word phrases, nor reading or writing, and were unable to care for themselves. There were no hearing or visual impairments evident and no history of seizures. For both families, no dysmorphic features were noted, and occipital frontal head circumference was within the normal range. MRI and EEG analyses were normal in one affected member.

Table 1.

Clinical Description of Affected Individuals from Families ARID-628 and ARID-49

	ARID-628-IV-4	ARID-628-IV-5	ARID-49-IV-1	ARID-49-IV-2	ARID-49-IV-3
Family Information
Consanguineous	+	+	+	+	+
Multiplex	+	+	+	+	+
Number of affected individuals	2	2	3	3	3
Extended family history of ID and/or epilepsy	+	+	-	-	-
Clinical Information
Age at examination	16 years	14 years	30 years	25 years	17 years
Sex	female	male	male	female	female
Intellectual disability	+	+	+	+	+
Speech	1–2 words	1–2 words	full sentences	2–3 words	single words
Hearing	normal	normal	normal	normal	normal
Daily activities (eating, toilet)	dependent	dependent	Independent	dependent	dependent
Dysmorphic facies	-	-	-	-	-
Hypotonia	+	+	-	-	-
Microcephaly	-	-	-	-	-
Seizures (partial complex)	controlled	controlled	-	-	-
Electroencephalogram	normal	N/A	N/A	N/A	N/A
Medication	Carbamazepine	Carbamazepine
Spasticity	-	-	-	-	-
Visual abnormalities	-	-	-	-	-
Head circumference	52.4 cm	50.3 cm	51.7 cm	51.6 cm	50.4 cm
Other systemic features	mitral valve prolapse	-	-	-	-
Brain MRI	normal	N/A	N/A	N/A	normal
Homozygous mutation (hg19 coordinates)	chr1:240,256,799insG	chr1:240,256,799insG	chr1:240,370,627delA	chr1:240,370,627delA	chr1:240,370,627delA
cDNA mutation (NM_020066.4)	c.1394_1395insC	c.1394_1395insC	c.2515_2515delA	c.2515_2515delA	c.2515_2515delA
Protein alteration (NP_064450.3)	p.Ala466Glyfs∗483	p.Ala466Glyfs∗483	p.Thr839Argfs∗848	p.Thr839Argfs∗848	p.Thr839Argfs∗848

N/A indicates “not assessed.”

The putative causative variant in ARID-628 was in the first exon and was annotated in hg19 as chr1: 240,256,799insG, a single G nucleotide insertion, predicting a frameshift at amino acid 466 of 1,726 residues (c.1394_1395insC [p.Ala466Glyfs^∗483], RefSeq transcript NM_020066.4). The variant in ARID-49 was in the fifth constitutive exon, chr1:240,370,627delA, a single A nucleotide deletion, predicting a frameshift at amino acid 839 (c.2515_2515delA [p.Thr839Argfs^∗848], RefSeq transcript NM_020066.4). We confirmed both variants with Sanger sequencing, and additionally confirmed that the variants segregated according to a strict recessive model with full penetrance. The ARID-49 variant occurred on the minor allele of a common synonymous SNP, rs10926166, and thus sequences were compared with two different healthy individuals, one displaying the major and one displaying the minor allele (Figures 1C and 1F). Neither frameshifting variant was predicted to affect several shorter FMN2 transcript variants (from mRNAs AK297755 and AK298141, derived from coronary artery smooth muscle cells and synoviocytes, respectively (Figure 1G), when alternative start sites were used. The variants were not identified in any other individuals from more than 3,000 whole-exome sequences, in 200 healthy control geographically matched individuals, or in any public database. There were no other potentially deleterious FMN2 variants in the CARID database or in private databases consisting of more than 4,000 intellectually disabled individuals who underwent exome sequencing.

The NHLBI study identified several different potentially deleterious variants in the database of >4,000 individuals, but none argued strongly against FMN2 as a potential cause of ID in these families. The rs375865107 variant predicts p.Trp601^∗ and was found to be heterozygous in one of 6,502 individuals, and the chr1:240286646_240286647insT (RefSeq accession number NM_020066.4; c.1782+1_1782+2insT) variant was found to be heterozygous in one of 6,295 individuals; probably both variants are valid. However, the chr1:240371554_240371558del5 (RefSeq NM_020066.4; c.3442_3446del5) variant, predicting p.Arg1148Glyfs^∗103, and the chr1:240371609_240371610insT (RefSeq NM_020066.4; c.3497_3498insT) variant, predicting p.Leu1168Serfs^∗85, have average read depth below 10 and were far outside of Hardy-Weinberg equilibrium (p value < 1 × 10⁻³¹¹), and the latter occurred in a stretch of homocytosines. None were found in any other public databases (or in our internal database), suggesting sequencing errors. Finally, several variants, including rs141879002, rs370558399, rs142397272, rs147210965, rs199721402, rs146873580, rs374743076, rs150801382, and rs183078344, modify well-conserved residues, but none are homozygous in any retrievable database.

To investigate the effect of the mutation on mRNA and protein, we tested expression of FMN2 in control lymphocytes but were unable to detect any expression, precluding analysis from family ARID-49, where only lymphocytes were available. Thus, skin biopsy was obtained from family ARID-628, RNA was reverse transcribed, and primers were designed to span coding exons. FMN2 encodes a 1,726 amino acid polypeptide as the major isoform (Figure 1H). We amplified a 398 bp product in control and III-2 individuals but no product in affected member IV-4, consistent with mRNA nonsense mediated decay, whereas GAPDH amplified a 383 bp product similarly in the three individuals (Figure 1I). Total cellular protein from confluent fibroblasts was analyzed by SDS-PAGE and blotted with a commercial FMN2 antibody, yielding a 195 Kd protein in control and III-2 individuals but no product in affected member IV-4, whereas GAPDH yielded similar band intensities in all (Figure 1J). We conclude that the ARID-628 mutation results in the absence of detectable full-length protein in cells from the affected individual.

FMN2 is one of 15 formin proteins identified in humans. Its identification is based upon the presence of a catalytic actin-nucleating and polymerizing formin homology 2 (FH2) domain.¹⁷ FMN2 additionally encodes a DEP (Dishevelled, Egl-10, and Pleckstrin) domain, suggesting a role in G-protein-coupled receptor signaling; a proline-rich FH1 domain; a formin-Spire1 interaction (FSI) domain¹⁸ involved in actin nucleation; and several phosphorylation sites, which probably regulate its activity. FMN2 has 18 exons, encodes three major isoforms, and is predicted to have important roles in actin cytoskeletal organization and the establishment of cellular polarity. Although FMN2 is expressed predominantly in the developing and mature brain,¹⁹ Fmn2 knockout mice (Neo cassette replacing 433 amino acids at the FH1 domain, Jackson labs, stock #016264, maintained on a pure 129S6/SvEvTac background) survive without major defects and have neither gross nor histological brain differences when they are compared with littermates.¹⁹ Fmn2^−/− germline female mutant mice show decreased fertility and abnormal positioning of the metaphase spindle and formation of the first polar body during oogenesis, but males show no corresponding defect.²⁰ Alhough FMN2 overexpression in NIH 3T3 cells showed dramatic disruption of the actin and microtubule network,²¹ we noted no major cellular morphology defect in FMN2 mutant fibroblasts under basal growth conditions.

Fmn2 was identified a “learning-regulated gene” as part of a screen for genes that were upregulated and for which H4K12 acetylation of the coding region increased upon fear conditioning in 3-month-old but not 16-month-old mice. Furthermore, Fmn2^−/− 8-month-old but not 3-month-old mice showed impaired associative learning, whereas pain sensation, explorative behavior, and basal anxiety were similar among groups.²² Because fear conditioning is hippocampal dependent, we studied FMN2 in the hippocampus at P14 in home-cage-reared mice and found that the protein within the granule neuron dendritic tree colocalized with the actin cytoskeleton, on the basis of phalloidin colocalization; as a control, we found no staining in Fmn2^−/− mice (Figure 2A). Hippocampal neuroanatomy was grossly intact, arguing against a role for Fmn2 in neuronal migration or differentiation.

Figure 2.

FMN2 Is Required for Dendritic-Spine Morphogenesis in Mice

(A) P14 hippocampus from Fmn2^+/+ and Fmn2^−/−. Immunofluorescence showed FMN2 staining in the granule neuron dendritic region (the highlighted section is the CA2 region) of Fmn2^+/+ mice but no staining in knockout mice. Phalloidin staining was not overtly different between genotypes. The scale bar represents 150 μm.

(B) FMN2 punctae localize along dendrites of cultured murine primary hippocampal neurons at 21 days in vitro. Staining was adjacent to both post- (PSD-95) and pre-synaptic (VGLUT1 and SYNAPSIN) markers but did not fully colocalize. The scale bar represents 5 μm. FMN2: Sigma HPA004937, 1:200. PSD95 Thermo MA1-045, VGLUT1 Synaptic Systems, Inc. #135302, SYNAPSIN Cell Signaling #5297.

(C and D) Golgi-stained images of Fmn2^+/+ and Fmn2^−/− are shown at low and high magnification (FDneurotech Rapid Golgi Stain method). Neuronal morphology was indistinguishable, but spine density was reduced in knockout neurons when Sholl’s method for dendritic arborization was used. The scale bars in (D) and (E) represent 50 μm; in (D′) and (E′) it represents 20 μm.

(E) Quantification of spine density per micron of dendrite of hippocampal granule neuron from the CA2 region from three mice of each genotype, two sections per animal, >1,000 spines per animal. ^∗p < 0.001 by Student’s t test. The error bar shows SEM (0.80 ± 0.05 versus 0.54 ± 0.04 spines/μm, n = 3 mice of each genotype).

Most CNS excitatory synapses are located on small membrane protrusions called dendritic spines, in which a dynamic F-actin matrix is linked to postsynaptic-density scaffold proteins, signaling proteins, and NMDA and AMPA receptors. Actin dynamics are of great importance for synaptic plasticity and memory formation, and formins have been proposed as candidates for regulating the nucleation and polymerization of actin filaments in dendritic filopodia. In fact, the formin member mDia2 localizes to hippocampal synaptic spines, and knockdown results in altered spine morphology.²³ In cultured murine primary hippocampal neurons, we found partially adjacent and overlapping protein localization between FMN2 and the postsynaptic-density marker PSD95 and the presynaptic markers VGLUT1 and SYNAPSIN, suggesting that FMN2 also localizes to intracellular sites nearby spines (Figure 2B), although determining precise subspine compartmental localization will require additional work.

To determine whether Fmn2 is required for the morphology of hippocampal dendritic spines, we performed Golgi staining in P47 wild-type mice and their Fmn2^−/− littermates, then measured spine density. We found that overall neuronal architecture was intact and that there were no notable differences in dendritic branching or length. However, dendritic spines were reduced in density by 32% on mutant hippocampal dentate granule neurons (Figures 2C–2E). We conclude that hippocampal dendritic-spine density is reduced in Fmn2^−/− mice.

To determine whether this defect in synaptic density occurred in human cells, we reprogrammed fibroblasts from one diseased and one healthy member of ARID-628 and a healthy age- and sex-matched control by using integration-free episomal methods,²⁴ excluded gross chromosomal postreprogramming rearrangements, and generated neural cells by using a dual-SMAD inhibition protocol.²⁵ There was no defect noted in the efficiency of reprogramming or in the formation of neural rosettes, neural precursor cells (NPCs), or mature neuronal cells (Figure S3A), nor was any defect noted in the differentiation of hIPSCs into mesodermal or endodermal lineages (Figure S3B), suggesting that the hIPSCs reprogrammed fully and generated healthy neural cells. In neurons cultured for 4 weeks, there were no defects in survival, differentiation, dendritic complexity, or nuclear morphology among the three lines. We subsequently stained neural cells for F-actin and VGlut1 in order to identify excitatory synapses (Figure S4), then segmented punctae to determine synaptic density (Figure 3A). We found that synaptic density in diseased cells was reduced by 43% in comparison with healthy controls (Figure 3B), suggesting reduced synaptic density in human IPSC-derived neural cells in the absence of FMN2.

Figure 3.

Reduced Synaptic Density in Human FMN2 Mutant IPSC-Derived Neural Cells

(A) FMN2 mutant cells exhibit reduced density of excitatory synapses as assessed by VGlut1 (green) staining of neural cells (see also Figure S4). F-actin was visualized via phalloidin staining (red).

(B) Quantification of excitatory synapses per square micron showed reduced density in cells from affected members specifically under the Imaris MeasurementPro module. ^∗p < 0.01 by Student’s t test. The error bar represents SEM (0.83 ± 0.085 versus 0.81 ± 0.065 versus 0.47 ± 0.110 in the control individual, unaffected individual, and affected individual, respectively; n = 10 dendrites segmented from each of four neurons from two independent cultures per individual).

In summary, we report homozygous truncating mutations in FMN2 as a cause of autosomal-recessive nonsyndromic ID. The work is corroborated by an absence of protein in cells from affected members, by study of expression and localization of the endogenous gene and protein in mice, by synaptic spine defects in knockout mice, and by synaptic density defects in mutant human neural cells. We suggest that FMN2 might be required for regulation of the actin cytoskeleton during spine development, maturation, or remodeling. Previously, two individuals with de novo interstitial deletions involving FMN2, among other genes, demonstrated intellectual disability,^26,27 suggesting that haploinsufficiency, perhaps in combination with other gene deficiencies, might similarly correlate with ID.

Mutations in other FH2-domain-containing proteins have been implicated in human disease: for example, mutations in INF2 (MIM 610982) cause autosomal-dominant intermediate Charcot-Marie-Tooth disease type E (MIM 614455) and dominant focal segmental Glomerulosclerosis type 5 (MIM 613237), and mutations in DIAPH2 (MIM 300108) cause X-linked premature ovarian failure (MIM 200511). Recently, a biallelic null mutation in DIAPH1 (MIM 602121), encoding a diaphanous-related formin-family-of-Rho effector protein, was identified in a family affected by microcephaly.²⁸ FMN1 (MIM 136535) shows embryonic expression,²⁹ targeted deletion in mouse shows oligodactylism and alters BMP signaling,³⁰ and a genomic rearrangement including FMN1 shows nonsyndromic oligosyndactyly.³¹ FMN2 now joins a group of genes implicated in neuronal functions mediating higher cognition in humans.

Web Resources

The URLs for data presented herein are as follows:

• NHLBI Exome Sequencing Project (ESP) Exome Variant Server, http://evs.gs.washington.edu/EVS/

• Online Mendelian Inheritance in Man (OMIM), http://www.omim.org/

• PolyPhen-2, www.genetics.bwh.harvard.edu/pph2/

• Quantsmooth, http://www.bioconductor.org/packages/release/bioc/html/quantsmooth.html

• RefSeq, http://www.ncbi.nlm.nih.gov/RefSeq

• SeattleSeq, http://snp.gs.washington.edu/SeattleSeqAnnotation137/

Accession Numbers

RefSeq NCBI transcript FMN2: NM_020066.4

Exome data for ARID-628 have been deposited into dbGaP (phs000288).