New missense variants in RELT causing hypomineralised amelogenesis imperfecta

Abstract Amelogenesis imperfecta (AI) is a heterogeneous group of genetic diseases characterised by dental enamel malformation. Pathogenic variants in at least 33 genes cause syndromic or non‐syndromic AI. Recently variants in RELT, encoding an orphan receptor in the tumour necrosis factor (TNF) superfamily, were found to cause recessive AI, as part of a syndrome encompassing small stature and severe childhood infections. Here we describe four additional families with autosomal recessive hypomineralised AI due to previously unreported homozygous mutations in RELT. Three families carried a homozygous missense variant in the fourth exon (c.164C>T, p.(T55I)) and a fourth family carried a homozygous missense variant in the 11th exon (c.1264C>T, p.(R422W)). We found no evidence of additional syndromic symptoms in affected individuals. Analyses of tooth microstructure with computerised tomography and scanning electron microscopy suggest a role for RELT in ameloblasts' coordination and interaction with the enamel matrix. Microsatellite genotyping in families segregating the T55I variant reveals a shared founder haplotype. These findings extend the RELT pathogenic variant spectrum, reveal a founder mutation in the UK Pakistani population and provide detailed analysis of human teeth affected by this hypomineralised phenotype, but do not support a possible syndromic presentation in all those with RELT‐variant associated AI.


| INTRODUCTION
Amelogenesis is the process by which dental enamel is formed during tooth development. It is characterised by three stages. 1 During the initial secretory phase, a protein-rich extracellular matrix (ECM) is secreted in a highly organised fashion by a monolayer of specialised cells known as ameloblasts that retreat away from a pre-formed dentine surface. There is then a transition phase when the immature enamel reaches its final thickness, ECM secretion stops, and ECM proteins begin to be degraded in a controlled way. This is followed by a Chris F. Inglehearn and Alan J. Mighell authors contributed equally to this work. maturation stage during which ameloblasts pump mineral ions into the ECM, forming hydroxyapatite crystals which are organised in an intricate decussating pattern to make up the enamel prisms (rods), while the ECM proteins continue to be removed to make room for crystal growth.
Defects in proteins involved in any of these stages and processes can cause amelogenesis imperfecta (AI). AI is the common endpoint for an assortment of heterogeneous genetic enamel defects, presenting with an isolated or syndromic phenotype and typically inherited as X-linked, dominant or recessive conditions. Thus far, variants in 19 genes have been implicated in non-syndromic AI. 2 A further 19 genes are listed in OMIM as being associated with syndromic forms of AI, with 5 implicated in both non-syndromic and syndromic disease. Prevalence varies between populations, with frequency estimates of between 1 in 14000 3 and 1 in 700. 4 AI has an adverse impact on affected individuals often starting in early childhood, with poor aesthetics and functional failure. Discoloured, weak enamel can quickly break down causing pain, infection and early tooth loss with subsequent malocclusion. This in turn leads to high levels of distress, social avoidance, discomfort, isolation and emotional problems. 5 Recently Kim and co-workers reported one missense and two nonsense variants in the gene encoding Receptor Expressed in Lymphoid Tissues (RELT, OMIM: 611211), causing hypoplastic AI. 6 Affected individuals were also of short stature and their medical history suggested increased susceptibility to childhood infection, potentially implying a syndromic form of AI. 6 By homology, the 430-amino acid RELT protein appears to be a receptor in the tumour necrosis factor (TNF) superfamily, which comprises 19 ligands (TNFs) and 29 receptors (TNFRs). 7 It consists of an extracellular region with two cysteine-rich domains homologous to other TNFR superfamily members and a predicted N-glycosylation site, a central transmembrane domain and a unique intracellular domain. 8 However, RELT is an orphan TNFR which does not appear to bind to any of the 19 known TNF ligands. 7 RELT has been shown to be abundant in haematopoietic tissues, where it is involved in the activation of the NF-κB pathway, inducing cell apoptosis by binding with TRAF1 (TNF receptor-associated factor 1). 8 Interestingly, RELT is unique among the TNFRs, as it does not have the characteristic death domain found in the intracellular region of the other TNFRs and, additionally, it only binds to TRAF1 and none of the other TRAF molecules, suggesting a non-canonical TNFR pathway of apoptosis. 8 It was also reported that the function of RELT might be dependent on the co-expression of its two homologues RELL1 and RELL2 (RELT-like 1 and 2), which are shown to induce apoptosis in the absence of trimeric TNF ligands. 9 More recently, RELT, along with RELL1 and RELL2, were shown to activate p38 MAPK induced apoptosis when overexpressed. 10 Prior to the study by Kim and colleagues, no connection had been made between RELT and any pathogenic phenotype in human or mouse models. Their study implicates RELT in the secretory stage of amelogenesis. Here we report two additional novel missense variants in the RELT gene that cause autosomal recessive hypomineralised AI in four families, confirming that mutations in RELT cause AI. Our findings extend the observed mutation spectrum, reveal a founder mutation common to three UK Pakistani families and provide detailed analysis of human teeth affected by this hypomineralised phenotype, but suggest that RELT variants are not always associated with a broader, syndromic phenotype.

| Whole exome sequencing and analysis
Three micrograms of genomic DNA, from the proband in each family, marked with an arrow on the respective pedigree ( Figure 1A-D), were subjected to whole exome sequencing (WES). Exome capture was performed using the Agilent SureSelect Human All Exon Enrichment System. Libraries were sequenced with a 150 bp paired-end protocol on an Illumina Hi-Seq 3000 sequencer. Sequences were aligned and variants filtered as described previously. 11

| Sanger sequencing and segregation
Potentially pathogenic variants identified by WES were confirmed by PCR and Sanger sequencing, with segregation with disease confirmed for all available family members. The PCR primers used are listed in Table S1. Sequencing reactions were performed using the BigDye Terminator v3.1 kit (Life Technologies, California), according to the manufacturer's instructions, and sequences were resolved on an ABI3130xl sequencer (Life Technologies). Sequence analysis was carried out using the SeqScape v2.5 suite (Life Technologies).

| Microsatellite analysis
Primer sequences for microsatellite markers were obtained from the UCSC genome browser and synthesised with 5 0 -HEX-tags (Table S1). PCR products were resolved on an ABI3130xl sequencer (Life Technologies) and sized relative to GeneScan 500 ROX (Life Technologies). Results were analysed using Genemapper v4.0 (Life Technologies).

| Protein tertiary structure prediction
The sequence of the human RELT protein was retrieved from Uniprot (https://www.uniprot.org/uniprot/Q969Z4) and a homology model was produced in silico, by selecting the structure with the lowest energy conformation and then performing Molecular Dynamics simulations to examine the effects of the variants on the structure.
Full methods used are detailed in Supporting Information section.

| Tooth phenotyping: computerised X-ray tomography
Teeth were obtained either after extraction for clinical reasons or as naturally shed primary teeth, from individuals with and without AI. These were analysed by micro-computerised X-ray tomography filters were used to prevent beam hardening artefacts. μCT images were reconstructed with Skyscan Recon software (Bruker). Hydroxyapatite mineral, of known densities (0.25, 0.75, and 2.9 g/cm 3 (Bruker)), was used to calibrate the images. Calibrated false colour maps of mineral density were generated from the calibrated μCT images using ImageJ 12 and the interactive 3D surface plot plugin (https://imagej.nih.gov/ij/plugins/surface-plot-3d.html). Videos were produced using CTVox software (Bruker).

| Tooth sectioning and SEM
Using an Accutom-5 cutter (Struers, Ballerup, Denmark) with a peripheral diamond cutting disc, cooled with minimal water, each tooth was cut across the bucco-lingual axis. After sectioning, the    Figure 1A, B, C and Table S4).
An adult L7 molar from the proband (IV:2) of family 3 and a deciduous incisor from the proband (II:2) of family 4 were analysed with μCT and then scanning electron microscopy (SEM), with appropriate controls. A section of the μCT scan for each tooth has been false colour calibrated for hydroxyapatite density and is shown in Figure S4.
Both control teeth display the expected prismatic enamel structure ( Figure 3A,B for the molar and Figure S5A,B for the incisor). Analysis of the affected L7 molar revealed two layers of enamel ( Figure 3C).
The outer enamel layer has enamel prisms with abnormalities ( Figure 3D), while the inner enamel layer is non-prismatic with an abnormal lamellal structure ( Figure 3E, F). The affected incisor of the proband (II:2) of Family 4 also has two distinctive layers of enamel ( Figure S5C). The inner layer shows the regular configuration of enamel prisms which then become disorganised and change orientation to become non-prismatic layered enamel ( Figure S5D-F). No enamel pits were identified. There is no indication in either tooth that the dentine or the dentine-enamel junction are affected.
The tertiary structure of RELT was predicted using homology models to structures of other TNFRs that are available in PDB. The resulting model is presented in Figure S6.

| DISCUSSION
In this study we report four families segregating autosomal recessive AI associated with two novel RELT variants. This confirms the finding of Kim and colleagues that mutations in RELT cause an AI phenotype, 6 and extends the observed spectrum of pathogenic variants, bringing the total to five (dna2.leeds.ac.uk/LOVD/genes/RELT). These include a variant in the splice acceptor site at the end of intron 3 which disrupts normal splicing, a 2 bp deletion and a frameshift variant in exon 10, as well as missense variants in exons 4 and 11. The fact that two of the three missense variants alter residue R422 could potentially imply some unique function in tooth formation for this site in the protein, which is disrupted by these variants. However, the high pathogenicity prediction scores for these variants and the presence of likely null mutations, together with a recessive pattern of inheritance, instead suggest a haploinsufficiency mechanism, with complete loss of RELT function in affected individuals. Additional reported mutations will further illuminate the likely disease mechanism underlying AI due to variants in RELT.
The four families presented have a shared enamel phenotype with normal or near-normal enamel volume present prior to tooth eruption. Post-eruptive changes are rapid and lead to enamel loss, particularly at sites subject to physical loading such as occlusal surfaces. This phenotype shares features with that reported for Relt −/− mice, which have hypomineralised, full thickness enamel with a typical arrangement of enamel prisms. 6 Although the difference in degree of mineralisation in AI and control teeth analysed in this study was modest, the overall features are most consistent with a hypomineralised form of AI. This highlights that "hypomineralised AI" is poorly defined and that the term does not take account of any variations in how the hydroxyapatite mineral is organised to give the resistance to physical attrition that characterises normal enamel. The enamel phenotype described here contrasts with the more severe phenotype of diminished enamel matrix formation leading to hypoplastic AI reported previously for RELT variants. 6 Additionally, there is variable but mild taurodontism involving permanent molar teeth, identified in the lower right second molar of family 3 ( Figure S1) and the second molar of family 4 ( Figure 1G) and consistent with the radiographs for family 3 reported by Kim and colleagues. 6 This radiographic feature is not as dramatic as that associated with AI due to DLX3 genetic variants. The evaluation of future RELT cases will clarify whether taurodontism is a consistent feature and whether this has any implications for clinical care, as well as help to determine the spectrum of enamel phenotypes due to RELT variations.
Early enamel attrition after eruption is probably a consequence of abnormal enamel structure even though enamel prisms are present.
The disorganised, non-prismatic layers of enamel noted in Figure 3 and Figure S5 resemble closely the enamel architecture in AI patients carrying heterozygous LAMB3 variants, as reported by Smith et al. 11 AI-causing variants in LAMB3 and other hemidesmosomal proteins are thought to result in defective attachment of ameloblasts to the enamel matrix. Teeth from families with AI due to either LAMB3 or RELT variants have layers of disorganised, non-prismatic enamel lamellae and areas where enamel is prismatic but abnormal. These similarities of expression staging and phenotype could suggest a possible common mechanism for these different forms of AI. Of note is the recent report that ADAM10, a protein known to release anchored proteins, is expressed during the secretory phase of amelogenesis (but not thereafter) and that ADAM 10 cleaves the extracellular domain of RELT. 15 Lamellal structures were not present in murine Relt −/− enamel, although there was an absence of the poorly mineralised spaces that would normally be expected at the dentino-enamel junction.
Kim and colleagues reported that both heterozygous parents of family 1, carrying the c.1169_1170delCT variant, presented with a mild AI phenotype, consisting of subtle, localised enamel surface roughness and vertical enamel layers. 6 This carrier phenotype was not noted in the other two families presented. Heterozygous members of the families reported in this study show no enamel abnormalities or F I G U R E 3 SEM photos of adult molars. A,B, Control L7 molar. C, Affected L7 molar from the proband (IV:2) of family 3, with the inserts showing the outer layer of enamel with normal looking prism organisation (D) that contrasts with abnormalities of the inner layer of stratified enamel (E) and the transitional phase (F) between the layers malformations, with the exception of the mother (I:2) in family 4, who shows lesions on the cusps of her molars and canines. These could constitute a mild carrier phenotype in I:2, but these findings are absent from the father (I:1), who is also a heterozygous carrier of the same variant. Further studies are therefore required to confirm or exclude the existence of a mild carrier phenotype for AI due to RELT variants.
The protein structure of RELT has not been experimentally observed, and its function and the specific ligand that binds to it have not yet been characterised. We attempted to predict the tertiary structure of RELT using homology recognition. The homology of RELT to existing models only spans the TNFR region which accounts for 26% of the protein sequence ( Figure S6). Mutating the predicted structure shows a probable decrease in protein stability for all of the variants implicated in AI causation. However, given the limited homology to other available models, the ab initio prediction of the structure of the entire peptide should be interpreted with caution, as should the predicted effect that the pathogenic variants will have on protein structure.
The previous report on biallelic RELT variants causing AI included evidence suggestive of a syndromic phenotype, including short stature and recurrent infantile infections. Kim and colleagues noted conservation of the RELT sequence, even in species that have no teeth, suggesting that this provided further support for probably syndromic consequences. However, sequence conservation across species need not imply conserved function, since proteins can have multiple functions, functions can change through time or the primary function can become obsolete. 16 Thus, proteins with an enamel-specific function might also have other roles in animals that have lost their teeth, such as pangolin (Manis javanica), platypus (Ornithorhynchus anatinus) and the baleen whales, for example, minke whale (Balaenoptera acutorostrata), or that do not have enamel, such as aardvark (Orycteropus afer), two-toed sloth (Choloepus hoffmanni) and other members of their families. For example, AMBN, another protein involved in the secretory stage of amelogenesis, 17 has been shown to participate in cell adhesion and proliferation, 18 and has roles in adipose cells 19 and in osteogenesis in mice. 20,21 Nevertheless, AMBN mutations appear to cause only nonsyndromic AI. Examination of patients from the four families recruited in this study and of their medical histories gave no indication of additional features. Our findings therefore do not support a clear syndromic phenotype for all RELT variants. As further cases are reported the extent of the syndromic phenotype should become clearer.
In conclusion, we confirm that biallelic variants in RELT cause hypomineralised AI characterised by normal or near normal enamel volumes, but leading to enamel that is prone to wear rapidly due to attrition. We find no evidence to support a clear syndromic element in the families reported here. This study extends the observed pathogenic variant spectrum and reveals a common founder mutation in UK Pakistani AI families. We also show how human enamel structure is affected, with regions of both layered enamel and enamel which retains some prismatic structure but is abnormally formed. We note similarities to the phenotype seen in teeth of patients carrying LAMB3 variants, indicating a possible connection between the role of RELT in amelogenesis and attachment of the ameloblasts to the enamel matrix.