Mutations in MAP3K7 that Alter the Activity of the TAK1 Signaling Complex Cause Frontometaphyseal Dysplasia

Abstract

Frontometaphyseal dysplasia (FMD) is a progressive sclerosing skeletal dysplasia affecting the long bones and skull. The cause of FMD in some individuals is gain-of-function mutations in FLNA, although how these mutations result in a hyperostotic phenotype remains unknown. Approximately one half of individuals with FMD have no identified mutation in FLNA and are phenotypically very similar to individuals with FLNA mutations, except for an increased tendency to form keloid scars. Using whole-exome sequencing and targeted Sanger sequencing in 19 FMD-affected individuals with no identifiable FLNA mutation, we identified mutations in two genes—MAP3K7, encoding transforming growth factor β (TGF-β)-activated kinase (TAK1), and TAB2, encoding TAK1-associated binding protein 2 (TAB2). Four mutations were found in MAP3K7, including one highly recurrent (n = 15) de novo mutation (c.1454C>T [ p.Pro485Leu]) proximal to the coiled-coil domain of TAK1 and three missense mutations affecting the kinase domain (c.208G>C [p.Glu70Gln], c.299T>A [p.Val100Glu], and c.502G>C [p.Gly168Arg]). Notably, the subjects with the latter three mutations had a milder FMD phenotype. An additional de novo mutation was found in TAB2 (c.1705G>A, p.Glu569Lys). The recurrent mutation does not destabilize TAK1, or impair its ability to homodimerize or bind TAB2, but it does increase TAK1 autophosphorylation and alter the activity of more than one signaling pathway regulated by the TAK1 kinase complex. These findings show that dysregulation of the TAK1 complex produces a close phenocopy of FMD caused by FLNA mutations. Furthermore, they suggest that the pathogenesis of some of the filaminopathies caused by FLNA mutations might be mediated by misregulation of signaling coordinated through the TAK1 signaling complex.

Introduction

Frontometaphyseal dysplasia (FMD [MIM: 305620]) is a progressive sclerosing skeletal dysplasia characterized by supraorbital hyperostosis, undermodeling of the long bones, small and large joint contractures, and other extraskeletal developmental abnormalities, most notably of the cardiorespiratory system and genitourinary tract.¹ An X-linked form of FMD has been clearly defined, and in this condition, males manifest most prominently, although females can express a milder phenotype that includes hyperostosis and limb anomalies.^{2, 3} The severity of the disorder in males can extend from normal stature with near-normal adaptive functioning to perinatal lethality, principally due to the consequences of the extraskeletal malformations.³ The hyperostosis observed in FMD is not associated with bone fragility, and the excess cortical bone has the histological appearance of mature lamellar bone.

Gain-of-function mutations in FLNA (MIM: 300017), located on the X chromosome and encoding the actin-binding protein filamin A, account for FMD in ∼50% of individuals.³ This form of the condition is allelic to a spectrum of related phenotypes known as the otopalatodigital spectrum disorders (OPDSDs).^{2, 4} The mutations leading to OPDSDs are principally missense and small in-frame deletions and insertions that are clustered throughout the coding regions of FLNA, some of which are highly recurrent.^{2, 5, 6} Many of these mutations, including some leading to FMD,³ result in the substitution of residues in the N-terminal actin-binding domain of the protein. The effect of FMD-associated mutations on the function of this domain has not been explicitly addressed, but a recurrent missense mutation leading to the allelic disorder otopalatodigital syndrome type 2 (OPD2 [MIM: 304120]) confers enhanced avidity for actin.⁷ How this activity results in malformations in addition to hyperostosis is unknown, but the mechanism could relate to one or more of the many and varied biochemical functions of filamin A that include engagement with integrin-mediated cell-cell adhesion,^{8, 9, 10} cytoskeletal remodeling,^{11, 12} cell spreading and migration,¹³ mechanotransduction,^{14, 15} and influencing many cell signaling pathways through physical interactions with a multiplicity of second messenger proteins.¹⁶

Around 50% of individuals with a diagnosis of FMD have no demonstrable mutation in FLNA.^{3, 17} These individuals are almost indistinguishable phenotypically from those with FLNA-mutation-positive FMD,^{3, 17} one notable difference being the preponderance for individuals without an identifiable FLNA mutation to develop keloid scarring. Unlike X-linked FMD, the severity of the phenotype does not vary between the sexes, although females are more likely to develop keloid, similar to the presentation of this sporadically arising cutaneous condition.¹⁷ The majority of individuals with the FMD phenotype not explained by a FLNA mutation are isolated cases in their families; only rare instances of vertical transmission of the phenotype have been reported.^{3, 17, 18}

Here, we analyze 19 individuals with FMD who do not have an identifiable FLNA mutation and find mutations in two genes encoding components of the transforming growth factor β (TGF-β)-activated kinase (TAK1) signaling complex in all subjects. Strikingly, 15 individuals have a recurrent mutation in MAP3K7 (MIM: 602614), which encodes for TAK1. We show that these mutations result in enhanced TAK1 autophosphorylation and alteration of MAPK and NF-κB signal transduction emanating from this protein complex. Three additional individuals have missense mutations in the MAP3K7 region that encodes the kinase domain. These mutations produce a notably milder phenotype, which suggests a domain-specific phenotype-genotype correlation in this form of FMD. These results place TAK1 as a key regulator of several cell signaling pathways that coordinately regulate osteogenesis.

Material and Methods

Subject Ascertainment and Ethical Approval

Individuals with FMD were recruited by physician-initiated referral and consented to participate under approved protocols MEC/08/08/094 and 13/STH/56 (Health and Disability Ethics Committee, New Zealand). Subjects were diagnosed with FMD on the basis of clinical and radiological assessment, according to published criteria.³ Sanger sequencing of FLNA did not detect any mutations. Where possible, DNA was also collected from parents and siblings, and familial relationships were confirmed as declared by the examination of the segregation of six unlinked microsatellite markers.

Whole-Exome Sequencing

Identification of candidate genetic variants was initially performed with exome sequence data from parent-proband trios for four simplex individuals (01, 02, 05, and 17). Additional Sanger and exome sequencing was carried out on samples from the remaining 15 subjects. Genomic DNA was extracted from peripheral blood according to standard protocols. Parental DNA was available from ten families.

For three trios (subjects 01, 02, and 05 and their parents), exome enrichment was performed with the Agilent SureSelect Human All Exon V4+UTRs capture kit, and paired-end sequencing (generating 100 base-pair reads) was performed on the Illumina HiSeq2000 sequencing platform. Sequencing data were aligned to the Ensembl Genome browser human genome assembly (GRCh37) with the Burrows-Wheeler Aligner (MEM algorithm) v.0.7.12. Realignment around indels, marking of duplicate reads, and recalibration of base quality scores, was undertaken with tools from Picard v.1.140 (Broad Institute) and the Genome Analysis Toolkit (GATK) v.3.4-46 (Broad Institute). Individual variant calling was undertaken with the GATK HaplotypeCaller and followed by multisample genotyping and variant quality score recalibration. Gene context annotation was added with SnpEff v.4.1L, and allele frequencies were obtained from 1000 Genomes Project phase 1, NHLBI GO Exome Sequencing Project ESP6500, and the Exome Aggregation Consortium (ExAC) via the GATK VariantAnnotator. Sequential filtering of the multisample VCF file was undertaken with GATK SelectVariants and SnpSift v.4.1L (SnpEff). Subsequently, this protocol was carried out on individual 14. For samples from subjects 17, 18, and 19 and their respective parents, exome enrichment was performed with an Illumina TruSeq Exome Enrichment kit. Massively parallel sequencing was performed with the Illumina HiSeq2000 platform, generating 100 base-pair paired-end reads. Demultiplexing, base calling, alignment, variant calling, and annotation of these trios was performed as described previously.¹⁹ For all individuals analyzed by exome sequencing, platform artifacts were removed and good quality, rare variants (minor allele frequency < 0.001) in coding regions (including splice sites and 5′ and 3′UTRs) were retained (Table S4). De novo variants (present in the affected child but not in either parent) were then identified for each individual. Candidate mutations, and their de novo status, were confirmed by Sanger sequencing in subjects and their parents. Samples from subject 12 and their parents were exome sequenced with a slightly different protocol and analytical pipeline that has been described previously.²⁰ Directed Sanger sequencing of the two candidate genes was performed in 12 additional individuals (for primer sequences, see Table S1).

cDNA Preparation and RT-PCR

Primary dermal fibroblasts from individuals with MAP3K7-mutation-positive FMD were cultured in DMEM (Gibco, Thermo Fisher), plus 10% fetal bovine serum (FBS; Moregate) and 1% streptomycin and penicillin (Gibco). RNA was extracted with Trizol reagent (Thermo Fisher) and the Nucleospin RNA kit (Machery-Nagel), according to kit protocols. cDNA was reverse-transcribed from total RNA with the SuperScriptIII kit (Thermo Fisher), according to the kit protocol. PCR amplification and Sanger sequencing was carried out with primers listed in Table S1.

Generation of Expression Constructs

The full coding sequence of MAP3K7 (variant A²¹ [GenBank: NM_003188.3]) was amplified from HEK293 cDNA with primer pair 1 (for all primer sequences, see Table S2) and cloned into pcDNA3.1 with the TOPO cloning kit (Thermo Fisher). This transcript was selected for these studies because (1) it represents the form that contains exon encoding sequences subject to mutation in the disorder under study, and (2) it represents the isoform studied by the vast majority of investigations into the properties of this gene. Primer pairs two, three, and four were used to sub-clone MAP3K7 into EcoRI- and XbaI-restriction-digested pCMV-Myc, pCMV-HA, and p3xFLAG-CMV with the Gibson assembly kit (NE Biolabs). TAB2 (GenBank: NM_015093.5) was amplified from HEK293 cDNA with primer pair four and cloned into pcDNA3.1 with TOPO cloning. Primer pair 5 was utilized for the sub-cloning of TAB2 into EcoRI- and XbaI-digested pCMV-HA with the Gibson assembly. TAB1 (pT7-FLAG) was a kind gift from K. Matsumoto (Nagoya University). Mutagenesis in MAP3K7 was carried out with overlapping amplicons containing the mutagenized base, followed by Gibson Assembly.

Co-immunoprecipitation

For competitive co-immunoprecipitation of TAK1 and TAB2, HEK293FT cells were cultured in DMEM (Gibco) plus 10% FBS (Moregate) and co-transfected with one of the following, using Lipofectamine2000 (Thermo Fisher) in a 24-well plate: (1) pCMV-HA TAB2 (800 ng), pCMV-MYC TAK1 (10 ng), and p3XFLAG TAK1 (10 ng), (2) pCMV-HA TAB2 (800 ng), pCMV-MYC TAK1 (10 ng), and p3XFLAG TAK1 p.Pro485Leu (10 ng), (3) pCMV-MYC TAK1 (10 ng) and p3XFLAG TAK1 (10 ng), or (4) pCMV-MYC TAK1 (10 ng) and p3XFLAG TAK1 p.Pro485Leu (10 ng). For analysis of TAK1 dimerization by competitive co-immunoprecipitation, HEK293FT cells were co-transfected with one of the following, using the same protocol: (1) pCMV-HA TAK1 (250 ng), pCMV-MYC TAK1 (250 ng), and p3XFLAG TAK1 (250 ng) or (2) pCMV-HA TAK1 (250 ng), pCMV-MYC TAK1 (250 ng), and p3XFLAG TAK1 p.Pro485Leu (250 ng). After 20 hr, cells were lysed in 1× PBS and 1% Triton X¹⁰⁰ with protease inhibitor (cOmplete mini, Roche). Lysates were clarified by centrifugation and 30 μL was retained for input samples. 2 μL of mouse anti-HA (Sigma Aldrich, H3663) was added to 150 μL supernatant, and samples were mixed at 4°C for 4 hr. 50 μl of washed Protein G Dynabeads (Thermo Fisher) were added to each sample and incubated for a further 30 min at 4°C, followed by three 1 mL washes in PBS. Protein was eluted in SDS sample buffer, and samples were resolved by 7% SDS-PAGE, followed by transfer to a nitrocellulose membrane. Membranes were blocked in 5% milk (PBS) then incubated with either mouse anti-HA (1:5,000, Sigma Aldrich, H3663), mouse anti-MYC (1:5,000, Clontech, 631206), or mouse anti-FLAG (1:7,000, Sigma Aldrich, F1804), followed by secondary goat anti-mouse IRDye 800CW (1:25,000, LI-COR, 926-32210). Membranes were scanned on a LI-COR Odyssey Clx, and bands were quantified with LI-COR Image Studio Lite software.

Luciferase Assays

HEK293 cells were cultured as previously described and transfected in a 24-well plate with 100 ng of plasmid DNA per well with Lipofectamine2000. Transfections were harvested 20 hr later in passive lysis buffer and assayed for Renilla and firefly luciferase activities with a Synergy2 multi-mode reader (BioTek) and injector module, which delivered the respective luciferase assay buffers (Promega). MAPK activity was assayed with the pFR-luc and pFA-ELK1 Pathdetect vectors (Agilent), which detect MAPK phosphorylation of the GAL4-ELK1 fusion protein and subsequent transcription from the pFR-luc reporter. NF-kB activity was measured from the reporter pNF-κBREluc (pGL4.32[luc2P/NF-kB-RE/Hygro]; Promega). DNA constructs were transfected with the following amounts (ng/well): pFRluc (10), pFA-ELK1 (10), pNF-κBREluc (5), pSVminRL (10–45), pTAK1 and mutants (10), pHA-TAB2 and mutant (40), and pCR3.1 (as empty vector) (10–50).

Western Blotting

For TAK1 stability, HEK293FT cells were transfected with 20 ng p3XFLAG TAK1 (wild-type [WT], p.Pro485Leu, or p.Gly168Arg) and treated with 85 μg/mL cycloheximide in DMSO or equivalent DMSO control after 24 hr. For quantification of phospho-TAK1, HEK293FT cells were transfected with 40 ng pCMV-HA TAB2 (WT or p.Glu569Lys), 20 ng pT7-FLAG TAB1, and 20 ng p3XFLAG TAK1 (WT and mutants) with Lipofectamine2000 in a 24-well plate. Transfections for phosphorylated p38 (p-p38) western blots were carried out as for luciferase assays. After 24 hr, cells were lysed in 1× PBS and 1% TritonX¹⁰⁰ with protease inhibitor; supernatant was cleared by centrifugation. Samples were denatured in 2.3% SDS and 1% DTT buffer. Proteins were resolved by electrophoresis on 8% polyacrylamide gels, followed by transfer to nitrocellulose membrane. Membranes were blocked in 5% milk (PBS) or Odyssey PBS buffer (LI-COR) and probed for 1 hr to overnight with anti-FLAG (F1804, Sigma Aldrich; 1:7000), anti-Myc (631206, Takara/Clontech; 1:5000), anti-phospho-TAK1 (Thr187, 1:1,500, Cell Signaling, CST4536), anti-p38 (1:2,000, Cell Signaling, CST9212), or anti-p-p38 (Thr180/Tyr182, CST9216, Cell Signaling Technologies; 1:4000). Detection was done with secondary goat anti-mouse IRDye 800CW (1:25,000, LI-COR, 926-32210) and goat anti-rabbit IRDye 700CW (1:25,000, LI-COR, 926-68071) antibodies as described above.

Statistical Analysis

Quantitative data from western blots and luciferase assays was imported into Excel and scaled and normalized to appropriate controls. Two-way, unpaired t tests were carried out, and critical p values were Bonferroni corrected (^∗p ≤ 0.05, ^∗∗p ≤ 0.01, ^∗∗∗p ≤ 0.001).

Results

Individuals with FMD with and without FLNA Mutations are Phenotypically Similar

We ascertained a cohort of 19 individuals with FMD without FLNA mutations. Clinically, they are very similar to FMD individuals with FLNA mutations, apart from an increased incidence of keloid formation^{17, 18} (Figure 1, Table 1). Core features of the phenotype include prominent supraorbital ridges, hyperostosis, downslanting palpebral fissures, hypertelorism, and a wide nasal bridge. The chin is typically small and pointed (Figure 1A). Progressive contractures of the joints are common, especially in the fingers and wrists (Figure 1B). Radiographs show a dense skull, especially the frontal bone and skull base (Figure 1C). The long bones of the hands and feet are undermodeled, frequently with sclerotic cortices (Figure 1D). Subjects often present with mild to severe scoliosis and have undermodeled, deformed ribs, sometimes with a “coat-hanger” configuration. The long bones have dense cortices and splaying of the metaphyses (Figure 1D). Three subjects had a somewhat milder skeletal phenotype (07, 18, and 19; Table 1), and individual 07 lacked a number of facial characteristics but still had prominent supraorbital ridges. Of the 19 individuals listed in Table 1, eight (two male, six female) exhibit keloid scarring that is occasionally progressive and severe and has the potential to form either spontaneously or after surgical trauma (Figure 1E).

Figure 1.

Clinical Images of Individuals with FMD

(A) Characteristic facial features, including supraorbital hyperostosis, downslanting palpebral fissures, hypertelorism, a broad nasal bridge, and a small pointed chin. Panels, from left to right, show subjects 14, 05, (both TAK1 p.Pro485Leu), and 17 (TAB2 p.Glu569Lys).

(B) Limb abnormalities include contractures of the metacarpophalangeal and interphalangeal joints (subjects 01 and 17) and valgus deformity of the foot (subject 14).

(C and D) Radiological assessment reveals (C) a dense, sclerotic skull, especially around the frontal bone and skull base (subjects 09 and 08) and (D) undermodeled phalanges, metatarsals, and metacarpals (subjects 01, 05, and 13) with occasional metacarpal synostosis (subject 01), scoliosis and undermodeled ribs (subject 08), and splaying of the metaphysis of the tibia and femur with bowing (subjects 01 and 02).

(E) Keloid scar formation in subjects 11 and 05.

Table 1.

Clinical and Radiographic Characteristics of Individuals with FMD

Subject	Sex	Clinical Features												Radiological Features
Supraorbital Ridges	Small Chin	Hearing Loss	Hypertelorism	Downslanting Palpebral Fissures	Wide Nasal Bridge	Cleft Palate/Bifid Uvulaa	Congenital Stridor/Subglottic Stenosis	Hydronephrosis	Scoliosis	Intellectual Disability	Keloid	Cervical Vertebral Fusion	Elbow Contractures/Dislocated Radial Head	Flared Metaphyses	Digital and Wrist Contractures	Under-modeled Phalanges	Broad Thumbs and Fingers
01	M	+	+	+	+	+	+	–	+	+	+	–	–	+	+	+	+	+	+
02	M	+	+	+	+	+	+	+	+	–	+	+	–	+	+	+	+	+	+
03	M	+	–	+	+	+	+	–	U	U	+	+	–	U	+	U	+	U	+
04	M	+	–	+	+	+	+	U	U	U	U	–	–	U	+	U	+	U	U
05	M	+	+	+	+	+	+	–	–	–	+	–	+	–	+	+	+	+	+
06	M	+	+	U	+	+	+	–	–	–	+	+	–	–	+	+	+	+	+
07	M	+	–	–	+	–	+	–	–	–	(+)	–	–	+	+	–	+	+	–
08b	M	+	+	+	+	–	+	–	+	–	+	–	+	+	+	+	+	+	+
09	F	+	+	+	+	+	+	+	–	–	–	–	+	+	+	+	+	+	+
10b	F	+	+	+	+	+	+	–	–	–	+	–	+	+	+	+	+	+	+
11b	F	+	–	+	+	+	+	–	+	–	+	–	+	+	+	+	+	+	+
12	F	+	+	+	+	+	+	+	+	–	–	(+)	–	U	+	U	+	U	+
13b	F	+	+	+	+	+	+	+	+	–	+	–	+	–	+	+	+	+	+
14	F	+	+	U	+	+	+	–	U	U	U	–	–	+	+	U	+	U	+
15	F	+	+	+	+	+	+	+	–	–	–	–	+	–	+	+	+	+	+
16b	F	+	+	+	+	+	+	–	–	+	+	–	+	+	+	+	+	+	+
17	F	+	+	+	+	+	+	–	–	–	+	–	–	–	+	+	+	+	+
18b	F	+	+	+	+	+	+	–	U	U	+	–	–	–	U	–	–	+	U
19	F	+	+	–	+	+	+	–	–	–	+	–	–		+	+	+	U	(+)

+, present; −, absent; (+), equivocal; U, unknown; M, male; F, female.

^aIncluding submucous cleft palate.

^bReported in Basart et al.¹⁷ or Morava et al.¹⁸.

Mutations in MAP3K7 and TAB2 Cause FMD

Whole-exome sequencing in family trios identified three individuals (01, 02, and 05) with the same de novo missense mutation, c.1454C>T, in MAP3K7 ([GenBank: NM_003188.3] Table 2). The subjects each had no more than three non-synonymous, validated de novo mutations in total across their respective exomes (Tables S3–S4). The recurrent mutation is predicted to substitute a proline to a leucine at position 485 (p.Pro485Leu) of the protein product of MAP3K7, TAK1. Subsequently, an additional individual (subject 12) was independently found, via exome analysis, to have the same de novo mutation.

Table 2.

Mutations in MAP3K7 and TAB2, Paternal Age at Birth, and Ancestry of Individuals with FMD

Subject	Gene	DNA	Protein	Inheritance	Paternal Age at Birth of Proband (Years)a	Ancestry
01	MAP3K7	c.1454C>T	p.Pro485Leu	de novo	33	Italian
02	MAP3K7	c.1454C>T	p.Pro485Leu	de novo	40	English
03	MAP3K7	c.1454C>T	p.Pro485Leu	U	36	Austrian
04	MAP3K7	c.1454C>T	p.Pro485Leu	U	38	Australian-European
05	MAP3K7	c.1454C>T	p.Pro485Leu	de novo	26	Scottish
06	MAP3K7	c.1454C>T	p.Pro485Leu	U	32	Swiss
08	MAP3K7	c.1454C>T	p.Pro485Leu	U	28	Korean
09	MAP3K7	c.1454C>T	p.Pro485Leu	U	18	Hispanic
10	MAP3K7	c.1454C>T	p.Pro485Leu	U	46	Turkish
11	MAP3K7	c.1454C>T	p.Pro485Leu	U	47	Mexican
12	MAP3K7	c.1454C>T	p.Pro485Leu	de novo	35	German
13	MAP3K7	c.1454C>T	p.Pro485Leu	U	26	English
14	MAP3K7	c.1454C>T	p.Pro485Leu	de novo	41	German
15	MAP3K7	c.1454C>T	p.Pro485Leu	U	30	Dutch
16	MAP3K7	c.1454C>T	p.Pro485Leu	de novo	40	Australian-European
18	MAP3K7	c.208G>C	p.Glu70Gln	dominant	U	French-Canadian
19	MAP3K7	c.299T>A	p.Val100Glu	dominant	U	Hungarian
07	MAP3K7	c.502G>C	p.Gly168Arg	de novo	28	Brazilian
17	TAB2	c.1705G>A	p.Glu569Lys	de novo	U	Australian-European

U, unknown.

^aMean, 34.4 and SD, 7.9.

Sequencing of another family trio (subject 17, Figure 1A, right panel) demonstrated a de novo missense variant, c.1705G>A, in TAB2 (MIM: 605101 [GenBank: NM_015093.5]), which encodes the binding partner of TAK1, TAK1-associated binding protein 2 (TAB2).²² This variant substitutes a glutamic acid to a lysine at position 569 (p.Glu569Lys) of TAB2 and was confirmed by Sanger sequencing. This individual had one other de novo variant in EFHC1: c.674C>G (p.Thr225Ser [GenBank: NM_018100.3]). Missense mutations in this region of EFHC1 are associated with juvenile absence or myoclonic epilepsy.^{23, 24} Because individual 17 has no history of seizures and because TAB2 is known to directly interact with and stabilize TAK1,²² the TAB2 mutation was prioritized for further analysis.

Further exome analysis and targeted Sanger sequencing of the exons of MAP3K7 and TAB2 in the remaining individuals with unsolved cases of FMD found that another 11 individuals have the recurrent c.1454C>T MAP3K7 mutation and an additional subject (07) was found to have novel de novo mutation in MAP3K7: c.502G>C, predicting the substitution p.Gly168Arg in TAK1 (Table 2). In addition, two unrelated individuals (18 and 19) who share the FMD phenotype with their respective mothers were found to have missense mutations in MAP3K7: c.208G>A and c.299T>A, predicting the substitutions p.Glu70Gln and p.Val100Glu, respectively. Both of these mutations were shown to have been inherited from the proband’s affected mother. In both instances, maternal grandparental samples were not available to ascertain if they had arisen de novo or not.

The phenotype of individuals with the recurrent c.1454C>T mutation is not noticeably different from that of subject 17, who has the missense mutation in TAB2 (Table 1; Figure 1A, right panel). The presentation of subject 07 with the p.Gly168Arg TAK1 substitution is, however, milder than that of the rest of the cohort. Individuals 18 and 19 (p.Glu70Gln and p.Val100Glu, respectively) have a typical facial appearance for FMD; however, the skeletal presentation is milder (Table 1). Only individuals with the TAK1 p.Pro485Leu substitution developed keloid scarring (Table 1).

Altogether, we found four variants in MAP3K7 to be causative of FMD in a total of 18 individuals. In the single remaining individual in this cohort, a missense mutation was identified in TAB2, which encodes for a protein that interacts directly with TAK1, the protein specified by MAP3K7.

TAK1 and TAB2 Mutations Predict Substitutions in Functionally Relevant Conserved Domains

Our genetic studies identified five different missense mutations affecting two proteins that form part of the TAK1 complex. MAP3K7 encodes TAK1, a MAP-3 kinase and the core enzymatic component of a multiprotein complex that is a hub for the control of many signaling pathways.²⁵ TAB2 encodes TAB2, a scaffolding protein necessary for the activation of some of signaling properties of TAK1,²² although it has not been shown to possess enzymatic activity itself.

TAK1 is comprised of an N-terminal kinase domain, a poorly characterized linker region, and a C-terminal coiled-coil domain^{25, 26} (Figure 2). Pro485 in TAK1, which is substituted to a leucine in 15 unrelated individuals with FMD, is phylogenetically highly conserved and the residue is invariant in both vertebrates and invertebrates, including C. elegans (genomic evolutionary rate profiling [GERP] score of 5.79, Figure 2A). The substitution occurs within the C terminus of TAK1 immediately N-terminal to the coiled-coil domain within a region that mediates interactions with TAB2²⁷ and that might also constitute a homodimerization interface²⁸ (Figure 2B). A crystal structure is available for the N-terminal kinase domain²⁹ but not for the linker region or C terminus, and therefore the structural consequences of substituting Pro485 cannot be ascertained. Variant effect prediction software (PolyPhen-2, MutationTaster, and SIFT) estimates that the recurrent p.Pro485Leu substitution is likely to be deleterious (PolyPhen score of 0.992, probably damaging; MutationTaster score of 0.999, disease causing; SIFT score of 0, damaging).

Figure 2.

Conservation and Structural Consequences of Substitutions in TAK1 and TAB2

(A) The residues mutated in FMD2 are conserved to C. elegans at TAK1^Pro485, to D. melanogaster at TAK1^Glu70 and TAK1^Gly168, and to zebrafish at TAK1^Val100 and TAB2^Glu569. Sequences were obtained from the Ensembl Genome browser and aligned with Clustal Omega (European Bioinformatics Institute).

(B) All five substitutions are found within regions of functional significance; p.Glu70Gln, p.Val100Glu, and p.Gly168Arg are located in the middle of the kinase domain (magenta), p.Pro485Leu is located within the TAB2 binding domain on TAK1 (light blue), and p.Gly569Lys is within the coiled-coil domain, of TAB2, which is within the region mapped as critical for TAK1 binding.²⁷

(C) Crystal structure for the N-terminal 303 amino acids of TAK1; Gly¹⁶⁸ is located within a tight β-hairpin turn. Substitution to an arginine at this site has the potential to disrupt the kinase domain. Glu70 is close to a peptide loop that sits over the active site of the TAK1 enzyme, and substitution to a glutamine is conservative. Val100 is located in a hydrophobic pocket along with Ile65, Phe74, Arg71, and Leu97, and substitution to a glutamic acid could be destabilizing to this region.

(D) TAK1^p.Pro485Leu and TAK1^p.Gly168Arg do not alter the stability of TAK1 when normalized to TAK1^WT up to 8 hr after cycloheximide treatment of transfected cells.

Error bars show SD.

Similarly, the p.Glu70Gln, p.Val100Glu, and p.Gly168Arg substitutions lie within a stretch of 120 almost completely conserved residues in seven vertebrates. A glutamine at position 70 is conserved to Drosophila (GERP score, 5.22), and a valine at position 100 is conserved in the seven vertebrate homologs analyzed (GERP score, 5.67; Figure 2A). Both are predicted to be disease causing and have PolyPhen scores of 0.995 and 0.951, respectively (probably damaging), MutationTaster scores of 0.999 (disease causing), and SIFT scores of 0 (damaging). A glycine at position 168 is conserved in Drosophila (GERP score, 5.18; Figure 2A). PolyPhen (0.802, possibly damaging), MutationTaster (0.999, disease causing), and SIFT (0, damaging) predict this substitution to be disease causing. These three substitutions lie within the conserved kinase domain of TAK1 (Figure 2B). Predictions based on the published crystal structure²⁹ (TAK1-TAB1 fusion protein [PDB: 2EVA]) of this region of the protein suggest that the p.Gly168Arg substitution would disrupt a tightly constrained β-hairpin turn that is required to order the active site of the kinase domain (Figure 2C). Glu70 is close to an important loop that sits over the active site of the TAK1 enzyme; however, substitution to a glutamine is conservative and it might establish similar interactions within the structure. Val100 is located in a hydrophobic pocket with Ile65, Phe74, Arg71, and Leu97, and substitution to a glutamic acid could be destabilizing to this region (Figure 2C).

Prediction of the effect of the TAB2 variant (p.Glu569Lys) is constrained by limited data on this protein. The substitution has a PolyPhen-2 score of 0.989 (probably damaging), a MutationTaster score of 0.999 (disease causing), and a SIFT score of 0 (damaging), suggesting the potential for pathogenicity. Glu569 lies in an area of high conservation in vertebrates (GERP score, 5.06; Figure 2A), but no tertiary structure for this protein is available. Similar to TAK1, TAB2 has a C-terminal coiled-coil domain within which the TAK1 binding domain resides.^{22, 27} The p.Glu569Lys substitution is located five residues N-terminal to this mapped TAK1 binding interface (residues 574–693, Figure 2B).

To evaluate the consequences of the recurrent MAP3K7 c.1454C>T mutation at the level of the transcript, cDNA was prepared from cultured primary fibroblasts obtained from subjects 02, 13, and 16 (MAP3K7 c.1454C>T) and RT-PCR followed by Sanger sequencing was performed. Sequence chromatograms demonstrated the persistence of the mutant allele in all instances and no additional RT-PCR products were observed (data not shown). An RNA source for subjects 07, 17, 18, and 19 was not available to test these alternative mutations in a similar manner. Together, these genetic, bioinformatic, and phylogenetic data constitute conclusive evidence that mutations in MAP3K7 cause frontometaphyseal dysplasia.

To explore the mechanism by which mutations confer this phenotype, we first considered a hypothesis that invokes haploinsufficiency for the pathogenesis of this form of FMD. Individuals with haploinsufficiency encompassing the MAP3K7 locus³⁰ and mice with conditional knockout of Map3k7 in the skeleton do not have an osteosclerotic phenotype,^{31, 32} suggesting that the phenotype arising from the MAP3K7 mutations described here is not a result of haploinsufficiency. To further test the hypothesis that the mutations in MAP3K7 do not exhibit their pathogenic effect by affecting the stability of TAK1, we performed quantitative western blots on lysates prepared from cycloheximide-treated cells transfected with expression constructs specifying either WT TAK1 (TAK1^WT) or TAK1 with the p.Pro485Leu (TAK1^p.Pro485Leu) or the p.Gly168Arg (TAK1^p.Gly168Arg) substitutions. There was no difference in protein stability between mutant TAK1 and WT protein up to 8 hr after cycloheximide treatment (Figure 2D), indicating that the pathogenic effect of these alleles is unlikely to relate to destabilization of the protein product. Similarly, individuals with haploinsufficiency of TAB2 have no skeletal dysplasia and instead present with cardiac malformations,³³ and Tab2-knockout mice die in utero of liver degeneration.³⁴ These data therefore suggest that the phenotype of the individual described here with a missense mutation at this locus is unlikely to arise from a haploinsufficient mechanism.

The p.Pro485Leu Substitution Does Not Affect TAK1/TAB2 Binding or TAK1 Homodimerization

A parsimonious hypothesis for the mechanism leading to FMD in these individuals would be that causative mutations disrupt the interaction between TAK1 and TAB2 given that both the recurrent mutation and the substitution in TAB2 lie within (or very close to) the interface formed by both proteins.²⁷ We used competitive co-immunoprecipitation to examine the relative TAB2 binding affinity of p.Pro485Leu in the presence of TAK1^WT by using a transient transfection protocol in HEK293FT cells. Two differentially tagged TAK1 proteins, with one construct specifying the p.Pro485Leu substitution and the other WT, were transfected alongside TAB2, and immunoprecipitation was performed to measure the relative ability of these proteins to bind TAB2. No significant difference is observed between TAK1^WT and TAK1^p.Pro485Leu in their ability to bind TAB2 (p = 0.59; unpaired t test, Figure 3A), indicating that this interaction is physically unimpaired by the presence of the p.Pro485Leu substitution in TAK1.

Figure 3.

The Recurrent TAK1^Pro485Leu Substitution Does Not Impair TAB2 Binding but Increases Autophosphorylation of TAK1

(A) Competitive co-immunoprecipitation between FLAG-TAK1^WT or FLAG-TAK1^p.Pro485Leu and HA-TAB2 shows no significant difference between TAK1^WT and TAK1^p.Pro485Leu in terms of their ability to bind TAB2 in the presence of TAK1^WT.

(B) When expressed in HEK293FT cells, TAK1^p.Pro485Leu and TAK1^p.Gly168Arg have significantly increased phosphorylation when expressed with TAB1 and TAB2^WT in comparison to TAK1^WT. In contrast, when TAK1^WT is expressed with TAB1 and TAB2^p.Glu569Lys, there is no increased phosphorylation of TAK1. Detection of FLAG was used for the quantification of TAK1.

Error bars show SD.

A second hypothesis relates to the possibility that TAK1 homodimerizes²⁸ and that this activity impacts regulation of the signaling functions of the TAK1 complex. The TAK1 homodimerization interface has been mapped within the coiled-coil domain to a region that is adjacent to, but not overlapping, Pro485.²⁸ We tested the capacity of TAK1^WT to dimerize with TAK1^p.Pro485Leu by using competitive co-immunoprecipitation. Again, no significant difference (p = 0.37; unpaired t test) is noted between the formation of TAK1^WT/WT homodimers and TAK1^{p.Pro485Leu/WT} heterodimers, suggesting that the p.Pro485Leu substitution does not impair this activity (Figure S2).

TAK1^p.Pro485Leu and TAK1^p.Gly168Arg Increase TAK1 Autophosphorylation

In the absence of a clear mechanism relating the p.Pro485Leu substitution to mapped protein-protein interactions within the C terminus of TAK1, we sought evidence that this substitution altered the kinase activity of this protein. Upon activation, TAK1 is sequentially autophosphorylated within its kinase domain,³⁵ starting at Ser192 and followed by Thr178, Thr187, and Thr184, which in turn triggers the phosphorylation of a number of downstream effectors.^{35, 36, 37} In addition to being scaffolded by TAB2, signaling through the TAK1 complex is also dependent on a second interacting factor, TAB1 (TAK1-associated binding protein 1 [MIM: 602615]).²⁶ To test whether the mutations leading to FMD affect TAK1 autophosphorylation, MAP3K7 constructs specifying TAK1^WT, TAK1^p.Pro485Leu, and TAK1^p.Gly168Arg were expressed in HEK293FT cells alongside constructs encoding its activating proteins TAB1 and TAB2. Under these conditions, quantitative western analysis revealed that both TAK1^p.Pro485Leu and TAK1^p.Gly168Arg are significantly more phosphorylated at Thr187 than TAK1^WT (p = 0.00014 and p = 0.00010 respectively; unpaired t tests; Figure 3B). Consistent with the established facultative requirement for the co-activator TAB1 in mediating TAK1 autophosphorylation, no difference in the phosphorylation of TAK1 mutants was observed when TAK1 was expressed with TAB2 alone (Figure S3A).

The p.Pro485Leu Substitution Alters Signaling Downstream of the TAK1 Complex

Enhanced autophosphorylation of TAK1 in the presence of substitutions that lead to FMD suggests that signaling pathways downstream of this complex should exhibit altered activity. Multiple signaling pathways are activated downstream of TAK1, including JNK,³⁸ p38 MAPK,³⁹ and NF-κB⁴⁰ (Figure 4A). We first employed a luciferase reporter assay (Pathdetect, Agilent) that measures a global readout for transcriptional activation mediated by several MAPK targets, including ERK, p38, and JNK. A significantly enhanced activation of the reporter is observed for TAK1^p.Pro485Leu in comparison to TAK1^WT (Figure 4B). Given that this reporter system detects activity mediated via a number of MAPK pathway effectors, and considering the documented specific role for p-p38 downstream of TAK1 in the maintenance of bone mineralization in the mouse,³¹ p-p38 was also assayed by western blot (Figure 4C). These data indicate that the TAK1^p.Pro485Leu construct results in a substantial increase in phosphorylation of p38 compared to that mediated by TAK1^WT. Notably however, neither MAPK signaling nor p-p38 are increased when TAK1^p.Gly168Arg is expressed under the same experimental conditions (Figures 4B and 4C), despite the significant increase in autophosphorylation noted in TAK1^p.Gly168Arg.

Figure 4.

Both MAPK and NF-κB Signaling Activity Are Altered by TAK1 and TAB2 Substitutions

(A) Activated TAK1 has the ability to activate numerous downstream pathways, including those leading to the activation of the kinases JNK and p38, and the pro-inflammatory transcription factor NF-κB.

(B and C) Luciferase reporter for MAPK activity (B) and an immunoblot for phosphorylated-p38 (C), demonstrating that both TAK1^p.Pro485Leu-TAB2^WT and TAK1^WT-TAB2^p.Glu569Lys combinations are associated with significantly more activated MAPK in comparison to combinations of WT constructs in HEK293 cells. TAK1^p.Gly168Arg-TAB2^WT is not significantly different from TAK1^WT-TAB2^WT.

(D) The activity of an NF-κB luciferase reporter showed that TAK1^p.Pro485Leu and TAK1^p.Gly168Arg have significantly reduced NF-κB activity in comparison to TAK1^WT, whereas TAB2^p.Glu569Lys was no different than TAK1^WT expressed with TAB2^WT.

Error bars show SD.

The second signaling pathway relevant to skeletogenesis that we evaluated was that mediated by NF-κB.^{32, 41} NF-κB is essential for RANK-L-mediated osteoclast differentiation,^{42, 43} whereas its activation inhibits osteoblastogenesis.⁴¹ We used a luciferase reporter sensitive to activation of this pathway and found that both the TAK1^p.Pro485Leu and the TAK1^p.Gly168Arg (Figure 4D) substitutions confer significantly reduced reporter activity in a transient transfection assay (Figure 4D). Together, these results indicate that signaling emanating from the TAK1 complex is altered by TAK1^p.Pro485Leu, but that the degree and quality of this disturbance is not exactly mirrored by the TAK1^p.Gly168Arg substitution even though both variants enhance the autophosphorylation of the enzyme and the phenotypic consequences in the subjects with these variants are similar.

The p.Glu569Lys Substitution in TAB2 Alters MAPK Signaling

The observation of a single instance of a de novo missense variant in TAB2 in an individual with FMD does not, in itself, constitute sufficiently strong evidence for pathogenicity. However, the location of this variant close to the mapped binding interface for TAB2 and TAK1 in an individual who demonstrates a phenotype that is indistinguishable from individuals with MAP3K7-associated FMD presents a strong a priori case for further investigation. We first questioned whether equivalent alterations to those observed in expression assays using the TAK1^p.Pro485Leu construct in signaling were conferred by the de novo mutation in TAB2. We found that the presence of the TAB2^p.Glu569Lys substitution alters neither TAK1 autophosphorylation (Figure 3B) nor NF-κB signaling activity (Figure 4D). There is, however, a clear increase in MAPK luciferase reporter activity and an increased p-p38/p38 ratio (Figures 4B and 4C), indicating that, despite a lack of enhanced autophosphorylation of TAK1, this variant in TAB2 has a strong effect on the activity of downstream components of the MAPK pathway.

Other Substituting Residues at TAK1 Pro485 Alter TAK1 Activity

Clinically, only the proline to leucine substitution is observed at position 485 in TAK1 in this cohort of individuals with FMD. Other mutations at this codon, however, might occur and lead to similar biochemical and clinical consequences. The likelihood of this might be dependent on the mode of gain of function and whether this primarily relates to the substitution of Pro485 and/or the identity of the substituting residue. We therefore examined the effect of other substitutions at the Pro485 codon to test the possibility that they might also confer a gain of function. Both p.Pro485Arg and p.Pro485Ala substitutions did enhance autophosphorylation (Figure S3B) and also had a variable effect on luciferase-based reporter assays for MAPK and NF-κB transcriptional targets. Substitution to an amino acid with a bulky and/or charged side chain (p.Pro485Arg) conferred more alteration in these signaling pathways than more conservative missense substitutions (e.g., p.Pro485Ala and p.Pro485Ser). Although TAK1^p.Pro485Arg shows no increase in MAPK activity (Figure S4A), it does show a significant increase in p-p38 when measured on immunoblot, in addition to significantly decreased NF-κB activity (Figures S4B and S4C). Substitutions to residues with smaller and less polar side chains (Ser and Ala) result in no difference in signaling, as assayed either through reporter assays or p38 phosphorylation, except for a small but significant increase in p-p38 with TAK1^p.Pro485Ser (Figure S4B). Overall, the changes in signaling function are smaller in magnitude to those conferred by the clinically observed substitution to leucine. We conclude that the identity of the substituting amino acid at this position does influence TAK1 function and that these substitutions also hold the potential to be disease causing.

Discussion

Until now, mutations in only one gene, FLNA, have been shown to cause FMD.⁴ Here, we have used a combination of exome and targeted Sanger sequencing to reveal four mutations in MAP3K7 to be the cause of FMD in 18 unrelated individuals and another variant in the gene encoding the interacting protein TAB2 to be likely causative of the condition in an additional individual. A specific mutation introducing a p.Pro485Leu substitution at the boundary of the C-terminal coiled-coil domain of the MAP-3 kinase, TAK1, was present in 15 unrelated individuals from diverse ethnic backgrounds. In all instances where parental samples were available (n = 6), this mutation had arisen de novo. A second de novo mutation predicting the substitution p.Gly168Arg was found in the highly conserved kinase domain of TAK1 in one individual. Two additional missense mutations leading to the substitutions p.Glu70Gln and p.Val100Glu, both inherited from a similarly affected parent, were found in another two subjects. We propose that the X-linked form of FMD be henceforth referred to as FMD1 to underscore its inheritance pattern and that the autosomal-dominant phenotype described here and caused by mutations in MAP3K7, the gene encoding TAK1, be denoted as FMD2. Although the de novo variant observed in TAB2 is an observation confined to a single individual, the functional relationship that this protein has with TAK1 and the biochemical activity it confers on the TAK1 signaling complex presents a strong, but still provisional, case that it also be considered causative of FMD. For this reason, we suggest that TAB2 will eventually be shown to represent a third causative locus and that this form of FMD will be denoted as FMD3.

Phenotypically, there might be some differences between FMD1, FMD2, and TAB2-related FMD. Subjects with substitutions in the kinase domain of TAK1 have a notably milder phenotype than those with TAK1 p.Pro485Leu substitutions or TAB2- or FLNA- associated FMD, suggesting a phenotype-genotype correlation. We noted that the individuals with the recurrent substitution in TAK1 have a higher incidence of cleft palate (n = 5/16, or 31% of subjects, compared to 9% reported in FMD1).^{3, 17, 44} We also previously reported a higher incidence of keloid scarring in individuals with FMD2 caused by TAK1^p.Pro485Leu.¹⁷ This is notable because keloid is a rare phenotypic manifestation in Mendelian disorders, and the data presented here could increase the understanding of the pathogenesis of spontaneous keloid formation. Keloid might only very occasionally appear as a feature of FMD1,³ but a recently described novel filaminopathy caused by a specific missense mutation in FLNA (c.4726G>A [p.Gly1576Arg] [GenBank: NM_001110556.1]) is also characterized by keloid scarring, joint contractures, and heart and kidney abnormalities. As such, it is reminiscent of FMD without its skeletal manifestations.^{45, 46} Therefore, it is possible that TAK1 and filamin A operate in the same pathway to promote keloid formation, a possibility consistent with data that indicate that activation of the TGF-β pathway, upstream of the MAP kinases p38, ERK, and JNK, has an important role in keloid pathogenesis. Small-molecule inhibition of p38, ERK, and JNK in cultured keloid fibroblasts impairs collagen accumulation and keloid development after TGFβ stimulation,⁴⁷ and therefore attention to therapeutic targets for keloid could usefully be focused on the pathway that links filamin A, TAK1, TGF-β, and MAP kinases.

FMD2 is primarily caused by TAK1^p.Pro485Leu in the cohort presented here (15/18 individuals). The finding of a highly recurrent mutation is consistent with a gain-of-function mechanism. Although instances of highly recurrent mutations can be the result of a paternal-age effect,^{48, 49} we only demonstrated a modest signal for this once we tabulated the age of the fathers at the birth of the TAK1^p.Pro485Leu individuals (Table 2).⁴⁸ A more salient observation is that the MAP3K7 mutation occurs at a hypermutable CpG dinucleotide and therefore the observation of a C>T transition is expected to be more common than other mutations at this site.^{50, 51} We predict that other substitutions at the Pro485 codon will exist and be associated with FMD phenotypes, but these have not been identified in this study, perhaps because of an ascertainment bias or because of the small number of individuals studied here. It might be relevant to note that a less significant effect on downstream signaling is observed when Pro485 is replaced with residues other than leucine, and so if individuals do exist with alternative residues replacing Pro485, they might present with a milder phenotype.

The structural and biochemical reasons for why TAK1^p.Pro485Leu is specifically implicated in this gain of function remain enigmatic. The TAK1^p.Pro485Leu substitution does not impair TAB2 binding, homodimerization, or TAK1 stability. We note that Pro485 lies within a consensus sequence for prolyl hydroxylation (LXXLAP) as represented in other proteins,^{52, 53, 54} but efforts to demonstrate that this site is subject to this form of modification by using mass spectroscopy, peptide substrates, and co-expression with prolyl hydroxylases failed to reveal any evidence to support this hypothesis (data not shown). Furthermore, substitution at this site with residues other than leucine did not invariably confer the same effect on autophosphorylation or signaling.

TAK1 is a MAP-3 kinase.³⁷ On activation, it binds its co-activators TAB1 and TAB2, leading to autophosphorylation, triggering downstream phosphorylation cascades.³⁷ The de novo TAK1 substitutions p.Pro485Leu and p.Gly168Arg significantly increase TAK1 autophosphorylation. The activated TAK1 complex is able to coordinate signaling in a number of different pathways, including the regulation of both osteoblast differentiation and activity.^{31, 55} The p38 pathway is especially critical because p38 phosphorylates and activates RUNX2, the master differentiation factor for osteoblasts, and DLX5, a transcription factor that stimulates the expression of the osteoblast genes IBSP (integrin-binding sialoprotein) and SP7 (osterix).^{31, 55} TAK1^p.Pro485Leu leads to increased phosphorylation of p38 and an enhanced ability to activate a generic MAPK transcriptional reporter. The TAK1^p.Gly168Arg substitutions did not show the same ability to activate downstream MAPK cascades; this could reflect the noticeably milder phenotype in this individual and the lack of sensitivity of our assays to detect subtle changes in signaling outputs. The TAB2 substitution p.Glu569Lys also increases MAPK signaling and enhances phosphorylation of p38, although these effects are not associated with activation of TAK1 via detectably enhanced autophosphorylation.

Not only is p38 essential for osteoblast differentiation,³¹ it is also critical for the proper differentiation of osteoclasts in response to RANK-L.⁵⁶ Therefore a gain of function in p38 signaling in both cell types is predicted to result in both increased osteoblast differentiation and activity and increased osteoclast differentiation. Further evidence that p38 is a central, but not necessarily exclusive, mediator of TAK1-directed osteogenesis, is the observation that the osteopenia observed in mice with Map3k7 deleted in the osteoblastic lineage is attributable to reduced p38 activation.³¹

The signaling output of TAK1 is complex and multifarious and also includes regulation of NF-κB.⁵⁷ Whereas MAPK signaling promotes the development of both the osteoblastic and osteoclastic lineage, NF-κB has opposing effects. Its activation by inflammatory cytokines is generally inhibitory for osteoblastogenesis^{41, 58} but essential for osteoclastogenesis stimulated by RANK and RANK-L.^{42, 43} Mice with an osteoclast-specific knockout of TAK1 have an osteopetrotic phenotype because of decreased NF-κB activation, demonstrating that TAK1 is necessary for proper osteoclast differentiation.³² Reduced NF-κB signaling could therefore promote hyperostosis by acting in concert in both of these cell types. Our data indicate that the TAK1^p.Pro485Leu and TAK1^p.Gly168Arg variants are associated with diminished NF-κB signaling, which is also the effect observed in an osteoclast-specific knockout of Map3k7 in mice. Hence, we conclude that activity conferred by TAK1 substitutions cannot be attributed to a simple biochemical gain-of-function mechanism across all affected pathways given that it is clear that the TAK1-complex variants alter multiple outputs, in different directions. This could be especially important for TAK1^p.Gly168Arg, which does not increase MAPK signaling output but does decrease NF-κB activity. This could still, therefore, disrupt the overall ratio of MAPK to NF-κB signaling and cause a milder FMD phenotype. Additionally, our in vitro data indicate that the TAB2^p.Glu569Lys substitution confers no difference in NF-κB signaling but still results in a florid FMD phenotype. Overall, this suggests that the combinatorial effect of these variants on all signaling pathways involved will likely only be definitively established once they can be evaluated in bone tissue from animal models of this disorder.

The TAB2^p.Glu569Lys substitution does not increase TAK1 autophosphorylation, nor does it decrease NF-κB signaling output. TAB2 has a number of TAK1-independent functions, and therefore the phenotype arising from this substitution could be attributed to these activities. For instance, TAB2 has been shown to localize in the nucleus with transcriptional repressors N-CoR and HDAC3, where it acts to repress gene targets of NF-κB; the complex is exported out of the nucleus upon stimulation of the TAK1 complex.⁵⁹ Additionally, TAB2 has been shown to bind Smad7 downstream of TGF-β activation, which blocks TAK1-TAB2-TRAF2 complex assembly.⁶⁰ Taken together, these data suggest that TAB2 acts to repress signaling mediated by TAK1, but it is unknown whether TAB2 might also act to promote MAPK functions, although our results indicate that the Glu569Lys substitution does confer such activity.

As well as influencing major developmental mechanisms in the skeleton, TAK1 has a headline role governing the innate immune response and inflammation.⁴⁰ The TAK1 complex is activated by inflammatory cytokines (e.g., TNF-α⁴⁰ and IL-1⁶¹) and triggers downstream signaling responses to these stimuli, such as activation of NF-κB.^{40, 62} We expect that individuals with gain-of-function TAK1 signaling might present with inflammatory phenotypes; however, this is not a clinically apparent aspect of the FMD2 phenotype. Focused study from this point could reveal hitherto unexpected subclinical anomalies in this disorder.

The TAK1 signaling complex is a broadly dispersed signaling hub in multiple tissues. Our observation that alteration of its signaling functions via a specific mutational mechanism hints at levels of regulation of this complex that remain uncharacterized. Although the phenotype of FMD extends to tissues other than bone, the pronounced effect that these mutations have on the mineralization of the skeleton suggests that understanding the function of this signaling hub in this tissue could produce new insights and therapeutic options for conditions characterized by osteopenia.

Published: July 14, 2016

Web Resources

• 1000 Genomes, http://www.1000genomes.org

• Burrows-Wheeler Aligner, http://bio-bwa.sourceforge.net/

• Clustal Omega, http://www.ebi.ac.uk/Tools/msa/clustalo/

• DECIPHER, http://decipher.sanger.ac.uk/

• Ensembl Genome Browser, http://www.ensembl.org/index.html

• ExAC Browser, http://exac.broadinstitute.org/

• GATK, https://www.broadinstitute.org/gatk/

• GenBank, http://www.ncbi.nlm.nih.gov/genbank/

• GERP, http://mendel.stanford.edu/SidowLab/downloads/gerp/

• MutationTaster, http://www.mutationtaster.org/

• NHLBI Exome Sequencing Project (ESP) Exome Variant Server, ESP6500, http://evs.gs.washington.edu/EVS/

• OMIM, http://www.omim.org/

• Picard, http://broadinstitute.github.io/picard/

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

• RCSB Protein Data Bank, http://www.rcsb.org/pdb/home/home.do

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