Amelogenesis imperfecta (AI) is a heterogeneous group of genetic conditions characterized by defects in the formation of enamel in all teeth of both dentitions. In an effort to classify the disease, particular phenotypes have been defined but this approach can be confounded by mixed phenotypes (Figure 3). Hypoplastic AI describes thin but mineralized enamel, or in extreme cases, the complete absence of enamel, that results from failure during the secretory stage. Hypomineralized AI is caused by maturation stage failure, giving rise to enamel that is of full thickness but is weak and fails prematurely. The hypomineralized phenotype can be further subdivided into hypomaturation and hypocalcified AI. The former is caused by incomplete removal of protein from the enamel matrix and produces brittle enamel, while the latter is characterized by insufficient transport of calcium ions (Ca²⁺) into the developing enamel and produces soft enamel.

Phenotyping of teeth from AI patients is complicated by post-eruptive changes that occur during the time spent in the mouth. Obtaining unerupted genotyped human embryonic teeth would be difficult and ethically questionable, while the study of erupted teeth precludes the direct study of amelogenesis (though enamel composition and ultrastructure can provide some form of record of the events occurring during amelogenesis). Therefore, mouse models have proved invaluable to AI research. Murine phenotyping of AI disease models can utilize the continuously erupting incisor to view all stages of amelogenesis at once, or gain a snapshot of amelogenesis via analysis of embryonic/neonatal unerupted molar teeth. Figure 1 shows the histology of the murine incisor and Supplementary Table 1 summarizes a selection of the AI-relevant mouse models that have been well characterized to date. The table also highlights genes for which murine models have not yet been described or for which an abnormal dental phenotype has not been reported.

AI is reported to range in frequency in different populations from 1 in 700 to 1 in 14,000 (Witkop and Sauk, 1976; Backman and Holm, 1986; Crawford et al., 2007), and has a significant impact upon patients and healthcare provision (Coffield et al., 2005). AI enamel is abnormally thin, soft, fragile, pitted and/or discolored, causing patients severe embarrassment, eating difficulties and pain. It is also associated with negative social outcomes and poor aesthetics (Hashem et al., 2013). AI is very difficult to treat and there is a weak evidence base to inform clinical decision-making and management choices (Dashash et al., 2013). Interventions focus on aesthetics and maintaining occlusal height and tooth function, whilst maintaining the natural dentition for as long as possible. Diagnostically it is necessary to distinguish AI from more common enamel defects such as fluorosis, molar incisor hypomineralization (Gotler and Ratson, 2010) and those caused by time-limited events, such as serious systemic illness (Salanitri and Seow, 2013).

Amelogenin is a hydrophobic, proline and histidine rich protein, thought to act as an enamel matrix pH buffer (Guo et al., 2015) and as a scaffold for the spacing and growth of enamel crystallites (Chen et al., 2011). It is subject to extensive extracellular proteolytic processing following its secretion (Brookes et al., 1995). It is regarded as a tooth specific protein since it has not been detected elsewhere in human tissues (Chan et al., 2011). However, in murine tissues, it has been detected in dentine-forming, cementum-forming and bone-forming cells, as well as in the developing eye and brain (Fong and Hammarstrom, 2000; Janones et al., 2005; Haze et al., 2007).

AMELX mutations cause X-linked AI (MIM #301200) (Lagerstrom et al., 1991). Heterozygous mutations tend to present in female AI patients as stripes of normal and AI affected enamel due to lyonization (Berkman and Singer, 1971). In males, a copy of AMELX exists as AMELY on the Y chromosome, but AMELY transcription is around 10% of that of AMELX (Aldred et al., 1992; Salido et al., 1992) and hence cannot compensate for loss of AMELX expression. The AI phenotype in males is determined by the type and position of the mutation. Large deletions and N-terminal variants cause a hypomaturation AI defect with variable focal hypoplasia while mutations in the signal peptide and toward the C terminus cause smooth hypoplastic AI (Hart et al., 2002).

Over twenty AMELX mutations have been reported, including large deletions, frameshifts, nonsense and missense variants. For the majority of variants, pathology is thought to be due to loss of function, though an Amelx mutation in mice has been linked to toxic gain of function via activation of the pro-apoptotic UPR (Brookes et al., 2014). Altered splicing, due to variants such as the silent c.120T>C (NM_182680.1) change, has also been shown to result in enamel pathology (Cho et al., 2014). This particular variant prevents the excision of exon 4 from the majority of AMELX transcripts, thus preventing the formation of a miRNA from the normally excised exon 4 (Le et al., 2016). These examples show that further study, within the context of AMELX alternative splicing and its roles in signaling, is required to accurately define disease mechanisms.

Enamelin, the largest of the EMPs, is a tooth specific acidic protein expressed primarily by secretory stage ameloblasts (Hu and Yamakoshi, 2003). It is successively cleaved from its C terminus, resulting in numerous products (Fukae et al., 1993; Hu and Yamakoshi, 2003; Lu et al., 2008). Like amelogenin and ameloblastin, the uncleaved protein is found only within the newly secreted, outermost layer of the enamel matrix and is thought to be involved in enamel crystal extension (Hu et al., 2008). Some ENAM cleavage products, such as the 32 kDa fragment identified in pigs, have high affinity for HA crystals and accumulate within and between the enamel prisms (Hu and Yamakoshi, 2003).

The first mutation identified in ENAM caused an autosomal dominant AI with a severe, smooth hypoplastic phenotype (MIM #104500) as a result of a dominant-negative effect of aberrant splicing (Rajpar et al., 2001) and a milder, local hypoplastic phenotype (MIM #204650) caused by missense mutations (Mardh et al., 2002). Autosomal recessive inheritance has also been documented for ENAM mutations (Hart et al., 2003; Ozdemir et al., 2005a; Chan et al., 2010); homozygotes and heterozygotes present with a severe and a milder, local form, respectively (Ozdemir et al., 2005a). Severity based on zygosity is more often seen with nonsense or frameshift variants that escape nonsense mediated decay (NMD).

AMBN is rich in glycine, leucine and proline and, in addition to within the enamel matrix, localizes to the Tomes' processes and the DEJ (Krebsbach et al., 1996; MacDougall et al., 2000), but has also been detected in pre-odontoblasts, developing tooth roots and craniofacial bone (Fong et al., 1996, 1998; Spahr et al., 2006). AMBN is expressed throughout amelogenesis (Lee et al., 1996) but peaks during the secretory stage (Fukumoto et al., 2004). AMBN transcripts undergo alternative splicing to form two isoforms (Krebsbach et al., 1996; Hu et al., 1997). Porcine ameloblastin is extensively modified and is cleaved upon secretion by MMP20 to form a number of protein products that accumulate within different compartments of the enamel matrix. For example, N-terminal products accumulate between the enamel prisms throughout the matrix (Bartlett and Simmer, 1999). Although, it is known that AMBN can influence the differentiation and proliferation of ameloblasts (Fukumoto et al., 2004), it is also important in extracellular signaling to induce osteoblast differentiation (Iizuka et al., 2011), cell adhesion, via heparin and fibronectin (Beyeler et al., 2010), and mineralization (Yamakoshi et al., 2001; Zhang et al., 2011).

Only two mutations have been reported in AMBN in AI patients, both discovered through NGS. The first AMBN mutation reported, a large, in-frame deletion encompassing exon 6, segregated with recessive hypoplastic AI (MIM #616270) in a consanguineous Costa Rican family (Poulter et al., 2014c). Scanning electron microscopy showed both reduced mineral density and enamel thickness, mirroring the murine Ambn^{−5, 6/−5, 6} model. The second homozygous mutation, thought to alter splicing, was identified in one patient in a large cohort with oro-dental disease, using a targeted NGS assay (Prasad et al., 2016a).

Matrix metallopeptidases (MMPs) influence cell motility by regulating cell interactions and matrix degradation, crucial processes in many aspects of development (VanSaun and Matrisian, 2006). Like other MMPs, MMP20 is a zinc dependent endopeptidase that is secreted in an inactive precursor form that requires cleavage for its activation (Llano et al., 1997). MMP20 is secreted by ameloblasts concurrent with the EMPs, and is responsible for cleavage of EMP at specific residues, shortly after their secretion (Simmer and Hu, 2002). This generates products with specific, diverse roles during amelogenesis. MMP20 has been shown to be necessary for controlling HA crystal morphology (Prajapati et al., 2016) and through its action on amelogenin, may regulate mineralization (Kwak et al., 2016). MMP20 is also capable of cleaving the extracellular domains of cadherins that mediate cell-cell interactions as part of adherens junctions to allow ameloblast cell movement (Guan and Bartlett, 2013; Guan et al., 2016). This may affect amelogenesis since ameloblasts must move in synchronous groups in order to form typical enamel architecture. Since cadherins are linked to the actin cytoskeleton via catenins, cadherin cleavage releases β-catenin, which can act as a transcription factor and may be important for ameloblast differentiation (Bartlett et al., 2011; Guan et al., 2016).

Mutations in MMP20 lead to autosomal recessive hypomaturation AI (MIM #612529) (Kim et al., 2005). Eleven missense, nonsense, frameshift and splice site mutations have been described, all resulting in a similar phenotype (Ozdemir et al., 2005b; Papagerakis et al., 2008; Lee et al., 2010a; Gasse et al., 2013; Wang et al., 2013b). All five of the missense variants reported lie within either the catalytic peptidase domain or the hemopexin domain, which, through homology, is thought to influence substrate specificity or to bind inhibitors or activators of the pro-enzyme.

KLK4 encodes a serine protease that is expressed and secreted by ameloblasts in both the transition and maturation stages of amelogenesis (Hu et al., 2000, 2002; Simmer et al., 2009). Like MMP20, newly secreted KLK4 must be cleaved for its activation. In vitro experiments have shown that MMP20 can activate newly secreted KLK4 and that KLK4 can inactivate MMP20, potentially explaining the shift in proteinase activity during the transition stage (Yamakoshi et al., 2013).

KLK4 acts to further degrade the enamel proteins already cleaved by MMP20 during secretion and is capable of functioning over the wide pH range that occurs during maturation (Smith, 1998; Bartlett, 2013). Such activity aids removal of protein from the developing enamel by maturation stage ameloblasts, allowing the enamel crystallites to grow in width and thickness (Simmer et al., 2009; Bartlett, 2013).

KLK4 mutations cause autosomal recessive hypomaturation AI (Hart et al., 2004). All four KLK4 variants reported so far are either nonsense or frameshift mutations (Hart et al., 2004; Wright et al., 2011; Wang et al., 2013b; Smith et al., 2017a). However, only two of the four are predicted to lead to NMD. Of the two frameshift mutations affecting codons in the final exon, and therefore not expected to undergo NMD, one alters one of the three catalytic residues, p.S207, essential to the function of all kallikrein enzymes. This mutation has been shown to result in greatly reduced protein expression and proteolytic function (Seymen et al., 2015). Prior to the identification of the frameshift variant, c.632delT only three KLK4 variants in four families had been identified. However, the c.632delT variant, reported to occur at a frequency of 0.15% in the South Asian population, has been reported in five Pakistani families with hypomaturation AI and is predicted to disrupt three of six structurally important disulphide bonds (Smith et al., 2017a). Characterization of the human enamel phenotype revealed that overall the enamel was hypomineralized but that the deeper (inner) enamel was more seriously affected than the more superficial (outer) enamel (Smith et al., 2017a).

Integrin, β6 (ITGB6; MIM ^*147558) is a member of a large family of cell surface-adhesion receptors that mediate cell-cell and cell-ECM interactions by facilitating interaction with the cytoskeleton (Alberts et al., 2002). ITGB6 is predominantly found in epithelial cells and forms a heterodimer with integrin subunit alpha V (Busk et al., 1992; Breuss et al., 1993). Within the developing tooth, ITGB6 localizes predominantly in maturation stage ameloblasts (Wang et al., 2014b).

ITGB6 is known to bind to arginine-glycine-aspartic acid (RGD) motifs which are found in ECM proteins such as fibronectin, as well as the latency associated peptide of transforming growth factor-β1 (TGF-β1) (Breuss et al., 1993; Weinacker et al., 1994; Munger et al., 1999; Annes et al., 2002). Via this interaction and other mechanisms, ITGB6 is able to activate TGF-β1 (Munger et al., 1999).

An Itgb6 null mouse exhibited a hypomineralized AI phenotype, with the loss of any ordered enamel prism arrangement (Mohazab et al., 2013). Accumulation of amelogenin in the enamel matrix and the presence of enamel pits were also noted (Mohazab et al., 2013). Patients with mutations in ITGB6 have since been reported (MIM #616221) with either hypomineralized pitted enamel, similar to that reported in the Itgb6 null mouse (Poulter et al., 2014a), or hypoplastic enamel with a rough surface (Wang et al., 2014b), both recessively inherited. More recently, Ansar et al. (2015) reported a consanguineous family with a homozygous ITGB6 mutation with adolescent alopecia, intellectual disability and dentogingival abnormalities with rough, discolored enamel. However, it is unclear if these additional phenotypes result from the ITGB6 variant or are co-segregating, for example, due to an undetected copy number variant. The ITGB6 missense mutations identified so far lie within the β1 domain of the protein involved in binding to α integrin subunits, activity-modifying cations and ligands (Xiong et al., 2001, 2002). Wang et al. (2014b) also reported a patient with a homozygous ITGB6 nonsense mutation, but phenotyping of the enamel was complicated by the co-presence of a hemizygous Nance-Horan syndrome (congenital cataracts and dental anomalies) mutation (MIM ^*300457).

Laminin, alpha 3 (LAMA3; MIM ^*600805), laminin, beta 3 (LAMB3; MIM ^*150310) and laminin gamma 2 (LAMC2; MIM ^*150292) encode the three subunits of the heterotrimeric protein laminin 332 (LM332), which localizes to epithelial basement membranes of the ECM (Aberdam et al., 1994a). LM332 has a central role in the assembly and stability of hemidesmosomes, structures that mediate attachment between cells and the ECM (Nievers et al., 1999). Collagen type XVII, alpha-1 (COL17A1; MIM ^*113811) is a constituent of hemidesmosomes and is a ligand for LM332 (Nishie et al., 2011; Van den Bergh et al., 2011). Mutations in the genes encoding COL17A1 or the subunits of LM332 cause the autosomal recessive condition junctional epidermolysis bullosa (JEB; MIM #226700, #226650), in which failure to form hemidesmosomes between skin layers leads to extensive skin blistering (Aberdam et al., 1994b; Pulkkinen et al., 1994a,b; Kivirikko et al., 1995; McGrath et al., 1995). In addition, JEB patients often present with hypoplastic, pitted enamel (Wright et al., 1993), and it has been noted that heterozygous carriers of some mutations in these genes sometimes have AI in the absence of any skin phenotype (McGrath et al., 1996; Yuen et al., 2012; Kim et al., 2013; Poulter et al., 2014b).

LM332 has been implicated in ameloblast adhesion to the enamel surface via interaction with integrin alpha 6 beta 4 within hemidesmosomes; and in cell migration via binding of integrin alpha 3 beta 1 (Carter et al., 1991; Marchisio et al., 1993). Mature ameloblasts and Tomes' processes show particularly strong staining for mature LM332 protein (Yoshiba et al., 1998), which is thought to participate in the control of ameloblast differentiation and adhesion to the enamel matrix. This is supported by analysis of tooth buds from JEB patients, which show ameloblast disorganization and subsequent reduction in enamel volume (Brain and Wigglesworth, 1968). Enamel in JEB patients also has a number of changes in chemical composition, suggesting that mineral transport or ameloblast metabolism is affected (Kirkham et al., 2000).

No enamel phenotype has been reported for Lamb3 null mice since they die in early post-natal life (Kuster et al., 1997). In contrast, for Lama3 null mice, ameloblasts were smaller than those in wild-type (WT) mice, suggesting that LM332 is required for normal ameloblast differentiation (Ryan et al., 1999). Enamel deposition was described as abnormal and the enamel epithelium was disorganized (Ryan et al., 1999).

COL17A1 is expressed throughout enamel formation (Asaka et al., 2009). Col17^−/− mice have fewer hemidesmosomes than WT mice and exhibit thin, disorganized Tomes' processes (Asaka et al., 2009). This may be the result of alterations in ameloblast differentiation due to lack of contact with, and signals from, mesenchymal tissues. Mineralization is delayed in Col17^−/− mice. The enamel formed lacks the typical regular prism structure and the prisms themselves are malformed, a phenotype reminiscent of the Lama3^−/− mouse but not as severe (Asaka et al., 2009). Enamel proteins such as AMELX, AMBN, and ENAM are expressed by ameloblasts at a significantly lower level in Col17^−/− mice than WT, whereas expression of the pre-secretory protein tuftelin is increased, again suggesting that ameloblast differentiation is incomplete in Col17^−/− mice (Asaka et al., 2009).

Heterozygous carriers of some LAMA3, LAMB3, and COL17A1 mutations present with hypoplastic AI (MIM #104530) (Murrell et al., 2007; Pasmooij et al., 2007; Yuen et al., 2012; Kim et al., 2013; Poulter et al., 2014b). Initial reports of the tooth phenotype simply mentioned that the relatives of some JEB patients had poor enamel, without recognition that the phenotype was truly AI (Murrell et al., 2007; Pasmooij et al., 2007). It was not until later that families segregating AI with autosomal dominant inheritance, and without any family members with JEB, were recognized and the phenotype more accurately described as AI (Poulter et al., 2014b).

The enamel of LAMA3 and LAMB3 patients is similarly described as hypoplastic, with grooving and pitting often present (Kim et al., 2013; Lee et al., 2014; Poulter et al., 2014b). Carriers of the JEB-causing LAMA3 frameshift mutation c.488delG were found to have rough, pitted enamel due to haploinsufficiency of the protein (Yuen et al., 2012). One report of a patient carrying a LAMB3 mutation highlighted that the multi-cusped teeth were more severely affected by AI than the anterior teeth (Kim et al., 2016b), although further study is required to determine whether this is a general trend.

AI-causing mutations in LAMA3 and LAMB3 present somewhat of a dichotomy. LAMA3 variants that cause AI in heterozygous carriers, also cause JEB in biallelic individuals but the majority of AI-causing LAMB3 variants have not been reported in JEB patients. Most LAMB3 variants identified in AI patients are either frameshift or nonsense mutations predicted to escape NMD. The variants are consistent with a dominant gain of function disease mechanism, unlike the loss of function variants associated with recessively inherited JEB. However, Prasad et al. (2016a) did identify two AI patients carrying LAMB3 mutations that do not fit this pattern of pathogenesis. These mutations have also been identified in JEB patients as recurrent mutations at hypermutable CpG sites (Kivirikko et al., 1996). Nevertheless, the pathogenicity of these variants in AI remains to be confirmed since segregation of one of the variants with the dental phenotype was inconsistent, and for the other, no segregation was possible, since only one affected individual was recruited to the study.

COL17A1 mutation carriers with an AI phenotype harbor glycine substitutions, as well as nonsense, frameshift and splicing mutations (McGrath et al., 1996; Murrell et al., 2007). The nonsense and frameshift mutations identified would be expected to lead to NMD, suggesting that the cause of the phenotype is haploinsufficiency. The glycine substitutions are predicted to disrupt an extracellular collagenous triple helix, potentially affecting both the susceptibility of the protein to degradation and its secretion (McGrath et al., 1996).

No patients with AI and heterozygous LAMC2 mutations have yet been reported although it seems likely that these exist. In addition, other genes which are known to be involved in the etiology of JEB, such as integrin, alpha 6 (MIM ^*147556) (Ruzzi et al., 1997) and integrin, beta 4 (MIM ^*147557) (Vidal et al., 1995), may harbor heterozygous mutations that cause dental defects in the absence of skin blistering since homozygous patients with a number of different JEB sub-types present with enamel hypoplasia (Fine et al., 2008).

Amelotin (AMTN; MIM ^*610912) is a proline, leucine, threonine and glutamine rich protein secreted by transition and maturation stage ameloblasts (Iwasaki et al., 2005; Moffatt et al., 2006). The protein localizes to the ameloblast basal lamina where it known to bind to itself, to ODAM (odontogenic, ameloblast associated) and to SCPPPQ1 (secretory calcium-binding phosphoproteins proline-glutamine rich 1) (Holcroft and Ganss, 2011; Fouillen et al., 2017). It is hypothesized to form large aggregates (Holcroft and Ganss, 2011; Bartlett and Simmer, 2015). These aggregates are thought to mediate attachment between the maturation stage ameloblasts and the mineralizing enamel (Moffatt et al., 2014). AMTN is also expressed at the junctional epithelium, a structure partially formed from maturation stage ameloblasts that mediates the attachment of the gingiva to the tooth (Bosshardt and Lang, 2005; Moffatt et al., 2006).

Murine models of amelotin function include an Amel promoter driven Amtn overexpressing mouse (pAmel:Amtn^+/+) and a knockout (Amtn^−/−) model (Lacruz et al., 2012a; Nakayama et al., 2015). The pAmel:Amtn^+/+ model had thin brittle enamel with an irregular surface layer (Lacruz et al., 2012a). The Amtn^−/− model had mandibular incisors with a chalky appearance and enamel in which mineralization was delayed and organic material retained (Nakayama et al., 2015). The surface enamel easily chipped away and was found to be softer than the inner and middle enamel (Nakayama et al., 2015; Nunez et al., 2015). These phenotypes and an in vitro study that found that AMTN promotes HA precipitation, suggested a critical role for AMTN in the formation of compact surface aprismatic enamel during maturation (Abbarin et al., 2015). Therefore, AMTN could be bi-functional, with roles in both cell-matrix attachment and mineral nucleation.

In humans, only one AMTN mutation has been associated with AI; an in-frame deletion spanning exons 3 to 6 (Smith et al., 2016). The family exhibited hypomineralized AI with autosomal dominant inheritance. Phenotypic analysis of teeth revealed that the enamel was of a lower mineral density when compared to WT and the typical prismatic enamel structure was disturbed throughout the enamel layer.

Family with sequence similarity 83, member H (FAM83H; MIM ^*611927), is an intracellular protein with ubiquitous expression (Lee et al., 2008a). In oral tissues, the ameloblasts show the highest expression of FAM83H, especially in pre-secretory and secretory stages (Lee et al., 2008a). Lower expression is seen in maturation stage ameloblasts as well as in the odontoblasts and alveolar bone (Lee et al., 2008a).

Through homology, FAM83H was suggested to be involved in membrane vesicle trafficking or cytoskeletal reorganization (Foster and Xu, 2003; Ding et al., 2009). FAM83H has now been shown, via binding to casein kinase 1 (CK1), to regulate the organization of the keratin cytoskeleton and therefore also to be involved in desmosome formation (Kuga et al., 2016).

Human mutations identified in FAM83H cause autosomal dominant hypocalcified AI (MIM #130900) (Mendoza et al., 2007; Kim et al., 2008). Analysis of teeth from individuals with FAM83H mutations identified defects in enamel rods, especially at the DEJ, with increased organic content within the enamel (El-Sayed et al., 2010; Zhang et al., 2015a).

All of the mutations identified so far have been located in the final, largest exon and all but two are nonsense or frameshift variants predicted to lead to premature translation termination (Kim et al., 2008; Lee et al., 2008a, 2011; Ding et al., 2009; Hart et al., 2009; Hyun et al., 2009; Wright et al., 2009, 2011; El-Sayed et al., 2010; Chan et al., 2011). As terminating mutations in the last exon are generally not subject to NMD (Nagy and Maquat, 1998), the truncated products may cause AI through a dominant gain of function effect.

In vitro analysis of three of the reported AI-causing FAM83H mutations suggests that the mutations alter the localization of the FAM83H protein, leading to an increased concentration within the nucleus rather than its typical cytoplasmic location (Lee et al., 2011). Analysis of the shortest of the truncated mutant proteins reported, p.S287^* (NP_198488.3), revealed that it bound and inhibited CK1 (Kuga et al., 2016). All of the twenty-seven FAM83H mutations identified reside within the first 1343 bp of the final exon (up to Glu694) with the final 1460 bp devoid of known mutations, suggesting that this region may not be critical to FAM83H function. In addition, some degree of phenotypic variation has been reported in patients, with mutations predicted to produce longer truncation products having a milder phenotype confined to just the cervical areas of the teeth (Wright et al., 2009).

Interestingly, Fam83h knockout mice and a mouse overexpressing FAM83H do not show an AI enamel phenotype (Kweon et al., 2013) supporting a dominant negative effect as the disease mechanism.

WD repeat domain 72 (WDR72; MIM ^*613214) is believed to form a beta propeller structure (Jawad and Paoli, 2002; Valeyev et al., 2008) and is predicted by protein homology to be an intracellular vesicle coat protein (El-Sayed et al., 2009; Katsura et al., 2014). WDR72 expression is widespread, and is highest in bladder and kidney (Lee et al., 2010b). Immunolocalization of WDR72 in mouse incisors revealed more intense staining in maturation stage than secretory stage ameloblasts (El-Sayed et al., 2009), with a specific increase in expression noted to occur at the initiation of enamel maturation (Katsura et al., 2014).

Wdr72 null mouse models exhibit hypomaturation AI (Katsura et al., 2014; Wang et al., 2015). In one model, maturation stage ameloblasts were shorter compared with WT mice, whereas there was no difference for secretory stage ameloblasts (Katsura et al., 2014). Affected enamel appeared stained and opaque and had retained proteins within it, including amelogenin. In another null mouse model, attachment between the ameloblasts and the enamel matrix was found to be disrupted (Wang et al., 2015). Ruffle ended maturation stage ameloblasts, thought to be responsible for protein removal, did not appear to be present and uptake of processed enamel matrix proteins by maturation stage ameloblasts was affected (Wang et al., 2015). In addition, the localization of putative Ca²⁺ transporter SLC24A4 was altered in Wdr72^−/− mice.

Originally, three truncating WDR72 mutations were identified in patients with autosomal recessive hypomaturation AI (MIM #613211) (El-Sayed et al., 2009). Subsequently, other truncating mutations have been described, all causing an identical enamel phenotype and likely to be subject to NMD (Lee et al., 2010b; Chan et al., 2011; El-Sayed et al., 2011; Wright et al., 2011; Kuechler et al., 2012; Katsura et al., 2014). Patients with mutations affecting the region between the two beta propeller clusters have also been reported to exhibit hypodontia and delayed tooth eruption (Kuechler et al., 2012; Katsura et al., 2014). Additionally there have been reports of short stature in families with WDR72 variants; however, given the consanguineous nature of the majority of the families studied and the lack of adequate controls, it is difficult to directly associate the phenotype with WDR72 variants and to exclude the possibility that this is caused by an additional co-segregating variant.

Solute carrier family 24 (Sodium/potassium/calcium exchanger), member 4 (SLC24A4; MIM ^*609840) is one of a family of potassium dependent sodium/calcium exchangers. Members of this protein family share highly conserved hydrophobic regions, termed alpha-1 and alpha-2 repeats, which interact to form ion-binding pockets and lie within two clusters of five transmembrane helices (Iwamoto et al., 2000; Parry et al., 2013). SLC24A4 is highly expressed in a wide range of tissues including brain, aorta, lung and thymus (Li et al., 2002). Within the developing tooth, it is expressed by maturation stage ameloblasts and localizes to the membrane in contact with the developing enamel (Hu et al., 2012). In rat, Slc24a4 transcripts are highly upregulated during the shift from secretion to maturation in the enamel organ (Lacruz et al., 2012b) and it is suggested that SLC24A4 is responsible for the active transport of Ca²⁺ ions from ameloblasts into the enamel matrix during maturation (Wang et al., 2014a). In mice, SLC24A4 expression is restricted to ruffle-ended maturation stage ameloblasts (Bronckers et al., 2015).

Mutations in SLC24A4 cause a hypomaturation/hypomineralized AI phenotype with autosomal recessive inheritance (Parry et al., 2013; Seymen et al., 2014; Wang et al., 2014a; Herzog et al., 2015; Prasad et al., 2016a). Missense mutations affecting both alpha repeats and the cytoplasmic domain, as well as a nonsense mutation and a multi-exonic deletion, have been detected. Subsequent to the identification of SLC24A4 mutations in human AI, incisors from Slc24a4 null mice (Stephan et al., 2012) were examined using SEM and a similar phenotype was observed. The enamel was poorly mineralized and quickly wore away to expose the underlying dentine (Parry et al., 2013).

G protein-coupled receptor 68 (GPR68; MIM ^*601404) is a proton-sensing protein known to function in a wide range of cells including osteoblasts and osteocytes (Ludwig et al., 2003; Yang et al., 2006). Its activation leads to inositol phosphate formation and calcium release from intracellular stores (Ludwig et al., 2003). The protein contains seven transmembrane helices, with the histidine residues responsible for its pH sensing properties on the extracellular surface of the protein (Ludwig et al., 2003). GPR68 has been shown to be expressed in ameloblasts throughout all stages of amelogenesis, with strong expression at the ameloblast pole in contact with the enamel matrix (Parry et al., 2016). GPR68 activation is known to result in inositol phosphate formation that is associated with cytoplasmic re-organization and membrane ruffling (Honda et al., 1999; Czech, 2000; Parry et al., 2016). Therefore, Parry et al. (2016) proposed that GPR68 acts as a pH sensor in amelogenesis, directing ameloblasts to switch between the ruffle ended and smooth ended conformations during the maturation stage.

Three families with autosomal recessive hypomineralized AI have been reported with mutations in the single exon GPR68 gene (MIM #617217) (Parry et al., 2016). All are predicted to result in loss of function. Two of the families carried deletions expected to remove histidine residues shown to be crucial to the pH sensitivity or the structural integrity of the protein. The third carried a missense variant predicted to destabilize the second transmembrane helix.

In light of mutations in GPR68 leading to AI in humans, the incisors of Gpr68^−/− mice (Mogi et al., 2009) were assessed for enamel defects (Parry et al., 2016). A limited phenotype, of retarded formation of enamel with minor structural differences, was reported. This phenotype partially mirrors GRP68-associated AI in humans but may also reflect temporal differences in human and murine amelogenesis, the genetic strain of the Gpr68^−/− model or the possibility that mutant protein persists in cases of human AI.

The chromosome 4 open reading frame 26 (C4orf26; MIM ^*614829) gene encodes a proline rich protein containing a signal peptide, two highly conserved but unknown motifs and ten predicted phosphorylation sites (Parry et al., 2012). It has been suggested, based on its amino acid sequence, that C4orf26 belongs to the acidic phosphoprotein family of proteins. These are known to promote HA crystallization, and C4orf26 peptide with a phosphorylated C-terminus was shown to promote HA nucleation and support crystal growth in vitro (Parry et al., 2012). Analysis of C4orf26 expression in rat revealed transcripts in both secretory and maturation stage enamel organs but not in heart or kidney, and in a human cDNA panel lacking any dental tissues, expression was highest in placenta but was also evident in many other tissues (Parry et al., 2012).

Six mutations have been identified in C4orf26 in ten families with autosomal recessive hypomineralized AI (MIM #614832) (Parry et al., 2012; Prasad et al., 2016b). These include only nonsense, frameshift and splice site variants predicted to escape NMD. However, the resulting proteins are predicted to be non-functional. Enamel from affected individuals contained thin crystallites more typical of the early maturation stage of enamel development rather than of mature enamel. To date no mouse model has been characterized.

Acid phosphatase, testicular (ACPT; MIM ^*606362) is one of a group of enzymes capable of hydrolysing esters of orthophosphoric acid in acidic conditions (Romas et al., 1979). It is expressed by secretory, but not maturation stage, ameloblasts (Seymen et al., 2016) and it has been suggested to elicit odontoblast differentiation and mineralization by supplying phosphate during dentine formation (Choi et al., 2016). ACPT was first identified as a gene immediately centromeric to the kallikrein gene cluster within a region of chromosome 19q13.3-q13.4 that is differentially expressed in solid tumors (Yousef et al., 2001). ACPT is itself known to show lower expression in testicular cancer tissues compared to normal tissues and is regulated by steroid hormones (Yousef et al., 2001). Its chromosomal location, around 111 kb proximal to KLK4, is notable, although Acpt was first associated with amelogenesis through the differential analysis of ameloblast gene expression in rats (Lacruz et al., 2011). However, the function of ACPT during amelogenesis is currently unclear and no mouse model has been characterized to date.

ACPT is also expressed in the brain, where it is enriched at post-synaptic sites, and is thought to dephosphorylate the ErbB4 receptor (Fleisig et al., 2004). The ErbB4 receptor has important roles in neuronal differentiation and synaptogenesis, and ACPT acts as a tyrosine phosphatase to modulate signals mediated by ErbB4 that are important for neuronal development and synaptic plasticity (Fleisig et al., 2004).

Mutations in ACPT were recently identified in eight families with autosomal recessive hypoplastic AI (MIM #617297) (Seymen et al., 2016; Smith et al., 2017b). All seven variants identified are missense changes predicted to affect residues within the extracellular domain that makes up the majority (residues 29–390; NP_149059.1) of the 426 amino acid protein. Assessment of an affected tooth by X-ray computed tomography revealed that the enamel, where present, was around one tenth of the thickness of control enamel, but was well mineralized (Smith et al., 2017b). The authors found that the underlying dentine was hypermineralized compared to control, although no abnormalities were evident upon clinical examination (Smith et al., 2017b). Assessment of additional samples will be required to confirm this.

Family with sequence similarity 20, member A (FAM20A) is one of a family of three human homologs of the Drosophila four jointed (fj) protein kinase. Additional family member FAM20C is a Golgi casein kinase responsible for phosphorylating many secreted proteins involved in biomineralization (Tagliabracci et al., 2012). In vitro expression of FAM20A has shown that it is also located in the Golgi (Ishikawa et al., 2012; Wang et al., 2013a). FAM20A has been shown to control FAM20C localization and to potentiate its extracellular action in vitro (Ohyama et al., 2016). The protein is therefore designated a pseudokinase (Cui et al., 2015).

Mutations in FAM20A have been shown to cause autosomal recessive AI and gingival fibromatosis syndrome (AIGFS; O'Sullivan et al., 2011) and enamel renal syndrome (ERS; Jaureguiberry et al., 2012; Wang et al., 2013a). Both syndromes have since been recognized as phenotypic variants of the same condition and are now collectively termed ERS (MIM #204690) (de la Dure-Molla et al., 2014). Patients with FAM20A mutations have hypoplastic AI which, in extreme cases, may present as a complete absence of enamel (de la Dure-Molla et al., 2014) or characteristic “glassy” incisor teeth. There are a variety of associated ERS oral defects that may include delayed tooth eruption, hyperplastic dental follicles, pulp stones and gingival overgrowth with ectopic calcification (de la Dure-Molla et al., 2014). Calcification of other organs, most frequently nephrocalcinosis, is variably reported. This may be due to a combination of age, genetic modifiers, and exposure to Ca²⁺ channel blocking drugs (Poulter et al., 2015).

The symptoms associated with ERS suggest that FAM20A phosphorylates proteins involved in enamel formation and other aspects of tooth development, as well as those critical to Ca²⁺ regulation within the kidney. Analysis of developing murine oral tissues showed the presence of FAM20A within ameloblasts, odontoblasts and at consistently high levels within the oral epithelium (Wang et al., 2014c). Ameloblasts expressed FAM20A only during the secretory stage, consistent with the hypoplastic AI phenotype (Wang et al., 2014c). Analysis of teeth from Fam20a^−/− mice showed that the ameloblast layer was disorganized and detached from the DEJ, leading to pitted and thin enamel (Vogel et al., 2012). A similar phenotype was observed in an epithelial specific (K14-cre) Fam20a^−/− model, along with reduced levels of ENAM and MMP20 proteins in ameloblasts but increased levels of AMBN (Li et al., 2016). The authors suggest that reduced levels of ENAM may be due to a lack of phosphorylation by FAM20C due to the absence of FAM20A. However, the authors do not explain the increased and decreased levels of AMBN and MMP20, respectively.

Distal-less homeobox 3 (DLX3; MIM ^*600525) is one of a family of six DLX transcription factors essential to the development of placental, epidermal and ectodermal appendages (Beanan and Sargent, 2000). DLX3 is also expressed in skin, bone and dental tissues including both the odontoblasts and ameloblasts, with greatest expression in ameloblasts during the late secretory stage (Zhang et al., 2015b). Murine whisker and hair follicles strongly express DLX3 before birth but expression decreases after birth. DLX3 is expressed in the placenta during early embryonic development and Dlx3^−/− mice die around embryonic day 10 due to placental defects, whereas Dlx3^+/− mice appear phenotypically normal (Morasso et al., 1999). Although, the role of DLX3 has been studied in dentine, bone and hair, through the creation of conditional knockouts of Dlx3, (Hwang et al., 2008; Duverger et al., 2012, 2013) no study has reported the creation of a conditional knockout of Dlx3 in ameloblasts.

Mutations in DLX3 are known to cause the autosomal dominantly-inherited trichodentoosseous syndrome (TDO; MIM #190320) and amelogenesis imperfecta hypomaturation-hypoplastic type with taurodontism (AIHHT; MIM #104510). TDO is defined as AIHHT with additional features such as kinky hair, especially in infancy, and increased bone density/thickening of the craniofacial cortical bones (Lichtenstein and Warson, 1971). Other variably associated features include flattened fingernails and altered craniofacial morphology, including dolichocephaly and prognathism (Price et al., 1998).

A mutation was first associated with AIHHT in patients carrying a 2 bp deletion (c.561_562delCT; NM_005220.2) in DLX3 (Dong et al., 2005). Other families of different ethnicities have since been described with the same DLX3 mutation, potentially highlighting the position as a mutational hotspot. However, the phenotype varies from AI with only mild taurodontism (Kim et al., 2016a) to TDO (Lee et al., 2008b) suggesting that there may be wide phenotypic variation in presentation, even for identical DLX3 mutations.

DLX3, like all DLX proteins, contains a homeobox domain flanked by N- and C-terminal transactivation domains. The homeobox domain directly binds to target DNA sequences but the transactivation domains have also been shown to be crucial to this binding. DLX3 has been shown to bind to the enhancer regions of Amelx, Enam and Odam in a murine ameloblast cell lineage and to positively regulate their expression (Zhang et al., 2015b).

Only six DLX3 variants have been reported so far, including two frameshift variants predicted to escape NMD but to result in an altered C-terminal transactivation domain and four missense variants predicted to affect residues (spanning residues 133–182; NP_005211.1) within the central DNA binding homeodomain (residues 129–188). Disease has been postulated to result from haploinsufficiency (Nieminen et al., 2011) but other effects, including dominant negative through binding to WT DLX3 (Duverger et al., 2008), as well as gain of function, through interactions of the mutant protein with other transcription factors, have not been ruled out.