https://orcid.org/0000-0002-2408-2333
Figures
Abstract
Amelogenesis Imperfecta (AI) is a disorder of tooth development caused by mutations in genes involved in several stages of tooth enamel formation. Few proteins involved in tooth development or developmental anomalies are explored in detail. Knowledge of 3D protein structure is essential to studying protein function. However, crystallized complete protein structures related to teeth and oral development are rare in the Protein Data Bank. Computational approaches for automated protein structure prediction have become a popular alternative for generating protein 3D structures. In this study, we aimed to explore the potential of using computer-generated protein models to analyze mutations linked to AI. We took a systematic approach to identify, screen, and analyze AI-linked protein variants. Proteins with AI-linked mutations were identified from the NCBI and OMIM databases, followed by screening of sequences for intrinsically disordered regions (IDRs). The iterative threading assembly refinement (I-TASSER) server was used to generate homology models for the wildtype and mutant proteins. PyMOL was used to analyze and compare the 3D structures of the proteins. Nineteen human genes with AI-associated mutations were identified from NCBI and OMIM. We identified multiple AI-associated protein variants with structural differences compared to their wildtype form. The current evidence aligns with several of the structural alterations identified in our study. Our findings suggest the potential of utilizing computer-generated protein models to investigate disease-associated mutations. However, careful consideration of models, templates, and alignments over the regions of interest is necessary to predict any potential structural impact of a disease-causing protein variant.
Citation: Sharmin N, Yuan J, Chow AK (2025) Using computer-generated protein models to analyze mutations linked to Amelogenesis Imperfecta. PLoS One 20(6): e0326679. https://doi.org/10.1371/journal.pone.0326679
Editor: Sheikh Arslan Sehgal, Cholistan University of Veterinary and Animal Sciences, PAKISTAN
Received: November 29, 2024; Accepted: June 1, 2025; Published: June 26, 2025
Copyright: © 2025 Sharmin et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the paper and its Supporting Information file.
Funding: The author(s) received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
Introduction
The development of a tooth and its surrounding structure is a complex and highly regulated biological process [1]. Advancements in molecular biology have revealed strict genomic and proteomic control of odontogenesis, which guides the position, number, size, and shape of our teeth [2]. More than 200 genes and their protein products have been identified in tooth development [3,4]. Consequently, it is not surprising that many developmental anomalies are found in teeth and the craniofacial regions that are linked to genetic mutations. Amelogenesis imperfecta (AI) is one such disorder of tooth development caused by mutations in genes involved in several stages of tooth enamel formation [5,6]. Amelogenesis, the process of enamel formation, involves several steps highly regulated by multiple genes and protein-protein interactions [6,7].
Few proteins involved in tooth and facial development or linked to oral developmental anomalies have been studied in detail. Knowledge of three-dimensional (3D) protein structure is essential to studying protein function [8,9]. However, crystallized, complete protein structures related to teeth and oral development are rare in the Research Collaboratory for Structural Bioinformatics (RCSB) Protein Data Bank (PDB, www.rcsb.org/) [10]. Obtaining a 3D protein structure using X-ray crystallography or nuclear magnetic resonance (NMR) spectroscopy is expensive, time-consuming, and not possible with all types of proteins [11]. A computational approach for automated protein structure prediction is a solution to this problem. Advancements in protein sequence alignment, detecting distant homologues, and modeling of loops and side chains have contributed to the reliable prediction of protein structure [12]. AlphaFold, for example, is a database of predicted protein structures using an artificial intelligence system developed by DeepMind [12]. In recent years, generative models have also been developed to design protein structures [13]. One example is RFdiffusion, which was developed by fine-tuning the RoseTTAFold structure prediction network for protein structure denoising tasks [13]. While databases like AlphaFold do not currently allow users to input custom protein sequences for structure prediction, several homology modeling tools like MODELLER, and SWISS-MODEL are available to predict the structure of a given protein sequence by aligning it with known template structures [14,15]. Each tool has its own advantages and limitations, depending on factors such as ease of use, accuracy, and template availability [16]. Although there are some reports of computer-generated protein models in drug discoveries [17,18], their applications in the study of disease-causing mutations are rare. In this study, we aimed to take an in-silico approach to study the mutations identified in AI. Our research questions were:
- -. What types of protein mutations have been identified as being associated with AI?
- -. How accurate are computational protein modeling tools in predicting the structural changes of proteins caused by mutations associated with AI?
- -. What is the potential of using computer-generated protein models to analyze mutations linked to AI?
Materials and methods
A. Identifying mutations associated with amelogenesis imperfecta through database search
We took a systematic approach to search the genomic database of the National Library of Medicine (NCBI) with two terms, ‘Amelogenesis Imperfecta’ AND ‘Homo sapiens, to identify human genes and mutations involved in AI [19]. We further searched the same database with another two terms, ‘Amelogenesis’ AND ‘Homo sapiens.’ The two-way search aimed to identify all possible genes involved in AI, with and without a role in enamel formation. Non-protein-coding microRNAs and genes with no reported mutations were excluded. Additionally, the database of Online Mendelian Inheritance in Man (OMIM) was searched with the keyword “amelogenesis imperfecta” [20]. The finally included genes were those with AI-causing mutations, as documented in NCBI and OMIM. No additional database filters and pathogenicity thresholds were applied. The details of our search string and screening process are outlined in Fig 1.
B. Screening for intrinsically disordered proteins (IDPs)
In recent years, there has been a growing recognition that a large fraction of the human proteome is intrinsically disordered, meaning that these proteins carry out their biological functions without attaining a stable 3D structure under physiological conditions [21]. While intrinsically disordered proteins (IDPs) present unique challenges for predicting their 3D structures, several algorithms are available that can predict the disorder within a given protein sequence [22]. To screen our selected proteins for intrinsically disordered regions (IDR), D2P2 server was used, which provides IDP/IDR predictions made by 9 different predictors across 1765 complete genomes containing 10,429,761 sequences from 1256 distinct species [23].
C. Selecting protein modeler and evaluation tools for homology models
To select a protein modeler to produce reliable, full-length models for both the wildtype and mutant proteins, several test protein sequences, with and without homologous templates were ran in several modelers, including the iterative threading assembly refinement server (I-TASSER), SWISS-MODEL, Robetta, and PEP-FOLD3. SWISS- MODEL did not predict the structure of the template-independent regions of the custom protein sequence. The de novo protocol for Robetta is optimized for <120 residues single-domain proteins [24]. Structure prediction by PEP-FOLD was limited to only between 5 and 50 amino acid residues [25]. The I-TASSER was chosen over those modelers, considering the length and the quality of the output. Besides template dependence, I-TASSER applies ab initio structure modeling for protein targets in the regions with no or weakly homologous templates [26]. Several models from AlphaFold were used in this study to compare with the wildtype protein, but as AlphaFold does not allow users to input sequences, it was not used to create models for the mutant proteins [12]. Substitution/missense, and in-frame deletion mutations were modelled with I-TASSER.
I-TASSER returned five predicted full-length protein tertiary structures for each sequence input with corresponding C-scores, TM-scores, and the root-mean-square distance (RMSD) values. The C-score is calculated from the significance of threading template alignments and the convergence parameters of the structure assembly simulations. Higher values for C-scores represent higher confidence for the predicted protein model. TM-score and RMSD are estimated based on C-score and protein length following the correlation observed between these qualities [26]. For each of our target proteins, the best-predicted model was chosen based on the C-score, TM-score and RMSD.
To further evaluate the quality of the mutant and wildtype protein models generated by I-TASSER, the ProSA program (Protein Structure Analysis) and Ramachandran plot were used. ProSA-web analyzes a given protein structure and calculates a Z score (≤10), measuring the deviation of the total energy of the structure from an energy distribution derived from random conformations [27]. VADAR (Volume Area Dihedral Angle Reporter), a comprehensive web server, was used to generate the Ramachandran plot to get a visual assessment of protein structure quality [28].
D. Analysis of the protein structures using PyMOL
PyMOL was used to analyze and compare our wildtype and mutant 3D proteins [29]. To identify structural alterations in the protein’s backbone the wildtype protein models were superimposed on each of its varients using the sequence-independent ‘Super’ algorithm of PyMOL. The electrostatic potential surfaces of the wildtype and mutant proteins were also analyzed to determine any change in the surface charge between the mutant and the wildtype proteins.
Results
A. Proteins with mutations involved in amelogenesis imperfecta (AI)
Searching the database with ‘Amelogenesis Imperfecta’ AND ‘Homo sapiens’ and ‘Amelogenesis’ AND ‘Homo sapiens’ returned 37 and 51 genes, respectively. Four microRNAs were excluded from the list, leaving 44 unique protein-coding genes after the first round of screening. Searching the OMIM database returned 83 entries with “amelogenesis imperfecta”, of which 64 were removed for not being related to AI. After the 2nd round of screening and comparison between databases, as of March 2025, 19 genes were included in our study reported to have mutations in patients, documented in NCBI and in OMIM. For the finally included genes, data related to HGNC gene symbol, UniProtKB identifier, UniProtKB Annotation score cytogenetic location, protein and its function in tooth development, dental disease (OMIM phenotype), mutations (OMIM variants), and OMIM identifier were collected from NCBI, OMIM, and published literature. (Table 1).
The proteins encoded by these 19 genes are classified as enamel matrix proteins (amelogenin, ameloblastin, enamelin), enamel matrix proteases (matrix metallopeptidases, KLK4), cell-cell and cell-matrix adhesion proteins (integrin, laminin, amelotin, FAM83H), transport proteins (WD repeat domain 72, solute carrier family 24), proteins involved in pH sensing, crystal nucleation, and unknown functions (G-protein-coupled receptor 68, RELT TNF receptor, odontogenesis associated phosphoprotein, distal-less homeobox 3, acid phosphatase 4, FAM20A, Solute carrier family 10 member 7, Sp6 transcription factor) [6]. An analysis of 82 mutants are grouped as frameshift (20), truncation (26), substitution (24), splice-site mutation (7), and in-frame deletion/insertion (5). (Table 1).
Three truncation and 2 deletion mutations was identified as candidates for inducing Nonsense-mediated mRNA decay (NMD) (Table 2). NMD is a quality-control mechanism in eukaryotes that screens newly synthesized mRNAs and degrades the ones with premature termination codons to prevent the formation of disease-causing truncated proteins [47].
B. Proteins with Intrinsically disordered regions (IDRs)
Five proteins reported no IDRs. High (>50%) IDRs were identified in six proteins. Proteins with less than 30% IDRs were used for 3D structure prediction and further analysis (Table 3).
C. Homology models of wildtype and mutant proteins
I-TASSER successfully generated 3D models for mutant proteins. Ramachandran plot and Z-score from the ProSA program indicated good quality for most wildtype and mutant proteins generated by I-TASSER. Partial structural similarity was observed between some models of the wildtype proteins generated by I-TASSER and AlphaFold (S1 Table). When I-TASSER-generated structures were compared between the wildtype and the mutants, structural changes (RMSD >1.5) were observed for several mutants (S1 Table). For mutations that cause only minor RMSD changes, the mutated site is likely critical to the protein’s function, such as substrate binding, proteolytic cleavage, or interactions with other molecules [52].
To further evaluate the accuracy of homology models developed using I-TASSER, the modeled structure of FAM20A was compared with its closely related crystal structure available in PDB. Structure superimposition between the homology model of FAM20A and its partial crystal structure 5WRR from PBD showed good backbone alignment with a RMSD value of 0.291 (2522–2522 atoms), except for the N-terminal that was missing in 5WRR (Fig 2A). An RMSD value below 2 Å is considered a good alignment between two structures. I-TASSER is programmed to identify best-match templates from PDB using a meta-server threading approach. The output of I-TASSER showed that 5WRR was identified as a template for modeling the FAM20A sequence (Fig 2B). As PDB does not have closely related crystal structures for the other proteins included in our list, this step was performed only for FAM20A.
FAM20A (secretory pathway pseudokinase) has its crystal structure, 5WRR in PBD. Structure superimposition between FAM20A WT homology model and 5WRR showed good alignment between the two structures, except the N-terminal that was missing in 5WRR. The RMSD value for the alignment was 0.291 (2522 to 2522 atoms) (A). I-TASSER output for FAM20A homology model, showing 5WRR is used as template for modeling the FAM20A amino acid sequence (B).
Discussion
The lack of experimentally determined structures is one bottleneck in research focusing on proteins involved in dental anomalies and tooth development. In silico protein structure prediction depends on two facts: (i) the protein 3D structure is determined by its amino acid sequence, and (ii) the change in protein structure happens at a slower rate compared to the sequence during evolution [53]. In this study, I-TASSER was used to generate complete models for the wildtype and mutant proteins. Besides template dependence, I-TASSER also applies ab initio structure modeling for protein targets in the regions with weak homologous (<30% identity) templates, allowing us to generate a complete model for mutant proteins with sequence variation from their wildtype partners [26].
For a given protein sequence input, I-TASSER predicts the 3D structure and provides results along with key metrics such as C-scores, TM-scores, and RMSD values. The C-score reflects the quality of the predicted model and is derived from the significance of threading template alignments as well as the convergence of structure assembly simulations. It typically ranges from -5 to 2, with a higher C-score indicating a model with greater confidence and reliability [26].
RMSD (Root Mean Square Deviation) and TM-score (Template Modeling Score) are used to assess the similarity between protein structures, differing in their approach and sensitivity. RMSD measures the average distance between corresponding atoms. For proteins, like in our study, where the reliable template is not always available, I-TASSER calculates the RMSD of the predicted models relative to the native structures based on the C-score [26,53,54]. TM-score focuses on topological similarity. Protein pairs with a TM-score >0.5 are mostly in the same fold, while those with a TM-score <0.5 are mainly not in the same fold [54].
It is essential to note that the C-score and prediction confidence can vary along the length of the protein, particularly if the protein contains IDRs. To improve the reliability of the predicted protein, we screened the protein sequences for IDRs before using them as input in I-TASSER. Of the proteins we included in our study, 18% of the ODAPH sequence was identified as intrinsically disordered (Table 3), resulting in a homology model of the wildtype protein with the least reliable matrix (C-score = −4.30, TM-score = 0.26 ± 0.08, RMSD = 14.6 ± 3.7Å) (S1 Table)
Besides evaluating I-TASSER-generated models using a Ramachandran plot, RMSD, TM, and Z-scores, the wild-type models were also compared with those from AlphaFold. Similarity in the folding pattern was observed between the I-TASSER and AlphaFold models for most proteins, except for ODAPH, which is attributed to a lack of a faithful template for this protein (S1 Table). Furthermore, we identified 18% of the ODAPH sequence as intrinsically disordered (Table 3), resulting in a model with a relatively low C-score in I-TASSER.
Truncation mutations are the most frequently identified mutations associated with AI. However, many truncation mutations, along with certain frameshift mutations, are subject to NMD, which prevents the expression of the corresponding protein. As a result, our study excluded truncation mutations from further analysis and concentrated on missense and selected frameshift mutations. To identify possible structural changes in protein variants, the I-TASSER-generated wildtype and mutant protein models were superimposed in PyMOL. Considering sequence variability caused by several frame-shift mutations, a sequence-independent algorithm (Super) was employed for structure comparison, which utilizes a dynamic programming approach to identify the optimal structural match. The alignment RMSD of 1.5 or higher was identified as indicating structural changes between the wild-type and mutant proteins (S1 Table). Several of the structural variations observed in our study are found to be supported by experimental evidence. Wang et al., 2014 identified a double substitution mutation in ITGB6 in a patient with AI [55]. Our modeling showed no structural change in A143T and H275Q variant alone in ITGB6 (S1 Table); however, the double mutation altered a portion of the protein located away from the mutation site. A partial crystal structure of Integrin beta-6 (chains B, D in 4um8) is available in the Protein Data Bank. A structural comparison between the PDB, AlphaFold and I-TASSER model of ITGB6 showed modeling resemblance over the core part of the protein, containing the double mutation. Considering the location of the structural alteration in the ‘twilight zone,’ it is considered less reliable, and the functional effect of the double mutant is likely related to the pathogenic impact of the conserved function of A143 and H275. H275 in exon 6 is conserved among its vertebral orthologs mediating subunit interaction [55]. A143 in exon 4 is another evolutionary conserved amino acid located within a specific metal ion-dependent adhesion site. Mutations in H275 and A143 would likely prevent the subunit interaction and binding to extracellular matrix respectively, resulting in a loss of function [55].
Glutamic acid (E) at a well-conserved position of 133 in ACP4 is involved in homodimerization. Substitution of a negatively charged amino acid with a positively charged one (E133K) has shown minor structural alteration in our study (S1 Table). Aligned with our finding, mutations in this conserved region are suggested to alter size and charge, thus interfering with homodimerization [47]. Most other missense mutations in our list are in conserved regions and are likely to affect protein function without causing major structural changes. For example, two substitution mutations in H226 and H204 regions of MMP20 did not have an apparent structural change. It is thus unclear if these mutations can affect zinc-binding, crucial to MMP20 function [56].
With the completion of the human genome project and ongoing advances in sequencing technologies, millions of genetic variants have been discovered, many of which remain poorly understood. Computational protein modeling presents a powerful approach to help bridge this knowledge gap by facilitating the interpretation of clinically relevant, potentially pathogenic variants [57]. These models not only enable the study of previously uncharacterized proteins but also provide rapid insights crucial for addressing time-sensitive clinical scenarios [57].
We acknowledge several limitations of our study. First, the use of a single protein modeling tool, I-TASSER, may limit the generalizability of our findings. Additionally, the number of proteins and variants analyzed was constrained by the presence of IDRs in the protein sequences. Despite these limitations, our study highlights the potential of using computational models to investigate protein variants.It is essential to be aware that the modeller generated the RMDS and other scores for the entire protein sequence, which may not be accurate for the region of interest. A careful consideration of models, templates, alignment over the regions of interest and supporting literature is necessary to predict the possible structural impact of a variant causing human diseases like Amelogenesis Imperfecta.
Supporting information
S1 Table. Homology model of the proteins and mutants involved in Amelogenesis Imperfecta.
https://doi.org/10.1371/journal.pone.0326679.s001
(PDF)
References
- 1.
McKinney R, Olmo H. Developmental disturbances of the teeth, anomalies of structure. Treasure Island (FL): StatPearls Publishing; 2022.
- 2. Ghergie M, Cocîrla E, Lupan I, Kelemen BS, Popescu O. Genes and dental 6disorders. Clujul Med. 2013;86(3):196–9. pmid:26527946
- 3. Thesleff I, Pirinen S. Dental anomalies: genetics. eLS. 2001.
- 4. Klein ML, Nieminen P, Lammi L, Niebuhr E, Kreiborg S. Novel mutation of the initiation codon of PAX9 causes oligodontia. J Dent Res. 2005;84(1):43–7. pmid:15615874
- 5. Frazier-Bowers SA, Vora SR. Genetic Disorders of Dental Development: Tales from the Bony Crypt. Curr Osteoporos Rep. 2017;15(1):9–17.
- 6. Smith CEL, Poulter JA, Antanaviciute A, Kirkham J, Brookes SJ, Inglehearn CF, et al. Amelogenesis Imperfecta; Genes, Proteins, and Pathways. Front Physiol. 2017;8:435. pmid:28694781
- 7.
Ten Cate NA. Nanci A. Ten Cate’s oral histology: development, structure, and function. 9th ed. USA: Elseview; 2016.
- 8. Weber IT. Evaluation of homology modeling of HIV protease. Proteins. 1990;7(2):172–84. pmid:2158092
- 9. Blundell TL, Bedarkar S, Rinderknecht E, Humbel RE. Insulin-like growth factor: a model for tertiary structure accounting for immunoreactivity and receptor binding. Proc Natl Acad Sci U S A. 1978;75(1):180–4. pmid:272633
- 10. Berman HM, Battistuz T, Bhat TN, Bluhm WF, Bourne PE, Burkhardt K, et al. The Protein Data Bank. Acta Crystallogr D Biol Crystallogr. 2002;58(Pt 6 No 1):899–907. pmid:12037327
- 11. Johnson MS, Srinivasan N, Sowdhamini R, Blundell TL. Knowledge-based protein modeling. Crit Rev Biochem Mol Biol. 1994;29(1):1–68. pmid:8143488
- 12. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596(7873):583–9. pmid:34265844
- 13. Watson JL, Juergens D, Bennett NR, Trippe BL, Yim J, Eisenach HE, et al. De novo design of protein structure and function with RFdiffusion. Nature. 2023;620(7976):1089–100. pmid:37433327
- 14. Webb B, Sali A. Comparative Protein Structure Modeling Using MODELLER. Curr Protoc Bioinformatics. 2016;54:5.6.1-5.6.37. pmid:27322406
- 15. Schwede T, Kopp J, Guex N, Peitsch MC. SWISS-MODEL: An automated protein homology-modeling server. Nucleic Acids Res. 2003;31(13):3381–5. pmid:12824332
- 16. Wallner B, Elofsson A. All are not equal: a benchmark of different homology modeling programs. Protein Sci. 2005;14(5):1315–27. pmid:15840834
- 17. Fauman EB, Rai BK, Huang ES. Structure-based druggability assessment--identifying suitable targets for small molecule therapeutics. Curr Opin Chem Biol. 2011;15(4):463–8. pmid:21704549
- 18. Nisius B, Sha F, Gohlke H. Structure-based computational analysis of protein binding sites for function and druggability prediction. J Biotechnol. 2012;159(3):123–34. pmid:22197384
- 19. Benson DA, Cavanaugh M, Clark K, Karsch-Mizrachi I, Lipman DJ, Ostell J, et al. GenBank. Nucleic Acids Research. 2013;41(Database issue):D36-42.
- 20. Hamosh A, Scott AF, Amberger JS, Bocchini CA, McKusick VA. Online Mendelian Inheritance in Man (OMIM), a knowledgebase of human genes and genetic disorders. Nucleic Acids Res. 2005;33(Database issue):D514-7. pmid:15608251
- 21. Lermyte F. Roles, Characteristics, and Analysis of Intrinsically Disordered Proteins: A Minireview. Life (Basel). 2020;10(12):320. pmid:33266184
- 22. Liu Y, Wang X, Liu B. A comprehensive review and comparison of existing computational methods for intrinsically disordered protein and region prediction. Brief Bioinform. 2019;20(1):330–46. pmid:30657889
- 23. Oates ME, Romero P, Ishida T, Ghalwash M, Mizianty MJ, Xue B, et al. D2P2: database of disordered protein predictions. Nucleic Acids Research. 2012;41(D1):D508-16.
- 24. Kim DE, Chivian D, Baker D. Protein structure prediction and analysis using the Robetta server. Nucleic Acids Res. 2004;32(Web Server issue):W526-31. pmid:15215442
- 25. Lamiable A, Thévenet P, Rey J, Vavrusa M, Derreumaux P, Tufféry P. PEP-FOLD3: faster de novo structure prediction for linear peptides in solution and in complex. Nucleic Acids Res. 2016;44(W1):W449-54. pmid:27131374
- 26. Roy A, Kucukural A, Zhang Y. I-TASSER: a unified platform for automated protein structure and function prediction. Nat Protoc. 2010;5(4):725–38.
- 27. Wiederstein M, Sippl MJ. ProSA-web: interactive web service for the recognition of errors in three-dimensional structures of proteins. Nucleic Acids Res. 2007;35(Web Server issue):W407-10. pmid:17517781
- 28. Willard L, Ranjan A, Zhang H, Monzavi H, Boyko RF, Sykes BD, et al. VADAR: a web server for quantitative evaluation of protein structure quality. Nucleic Acids Res. 2003;31(13):3316–9. pmid:12824316
- 29. DeLano WL. The PyMOL molecular graphics system. 2002. http://www.pymol.org
- 30. Giovani PA, Salmon CR, Martins L, Paes Leme AF, Rebouças P, Puppin Rontani RM, et al. Secretome Profiling of Periodontal Ligament from Deciduous and Permanent Teeth Reveals a Distinct Expression Pattern of Laminin Chains. PLoS One. 2016;11(5):e0154957. pmid:27149379
- 31. Niwa T, Yamakoshi Y, Yamazaki H, Karakida T, Chiba R, Hu JC-C, et al. The dynamics of TGF-β in dental pulp, odontoblasts and dentin. Sci Rep. 2018;8(1):4450. pmid:29535349
- 32. Yu T, Klein OD. Molecular and cellular mechanisms of tooth development, homeostasis and repair. Development. 2020;147(2):dev184754. pmid:31980484
- 33. Parry DA, Brookes SJ, Logan CV, Poulter JA, El-Sayed W, Al-Bahlani S, et al. Mutations in C4orf26, encoding a peptide with in vitro hydroxyapatite crystal nucleation and growth activity, cause amelogenesis imperfecta. Am J Hum Genet. 2012;91(3):565–71. pmid:22901946
- 34. Laugel-Haushalter V, Bär S, Schaefer E, Stoetzel C, Geoffroy V, Alembik Y, et al. A New SLC10A7 Homozygous Missense Mutation Responsible for a Milder Phenotype of Skeletal Dysplasia With Amelogenesis Imperfecta. Front Genet. 2019;10:504. pmid:31191616
- 35. Yang M, Huang W, Yang F, Zhang T, Wang C, Song Y. Fam83h mutation inhibits the mineralization in ameloblasts by activating Wnt/β-catenin signaling pathway. Biochem Biophys Res Commun. 2018;501(1):206–11. pmid:29709481
- 36. Kim J-W, Zhang H, Seymen F, Koruyucu M, Hu Y, Kang J, et al. Mutations in RELT cause autosomal recessive amelogenesis imperfecta. Clin Genet. 2019;95(3):375–83. pmid:30506946
- 37. Llano E, Pendás AM, Knäuper V, Sorsa T, Salo T, Salido E, et al. Identification and structural and functional characterization of human enamelysin (MMP-20). Biochemistry. 1997;36(49):15101–8. pmid:9398237
- 38. Parry DA, Smith CEL, El-Sayed W, Poulter JA, Shore RC, Logan CV, et al. Mutations in the pH-Sensing G-protein-Coupled Receptor GPR68 Cause Amelogenesis Imperfecta. Am J Hum Genet. 2016;99(4):984–90. pmid:27693231
- 39. Wang S, Choi M, Richardson AS, Reid BM, Seymen F, Yildirim M, et al. STIM1 and SLC24A4 Are Critical for Enamel Maturation. J Dent Res. 2014;93(7 Suppl):94S-100S. pmid:24621671
- 40. Katsura KA, Horst JA, Chandra D, Le TQ, Nakano Y, Zhang Y, et al. WDR72 models of structure and function: a stage-specific regulator of enamel mineralization. Matrix Biol. 2014;38:48–58. pmid:25008349
- 41. Pang L, Zhang Z, Shen Y, Cheng Z, Gao X, Zhang B, et al. Mutant dlx3b disturbs normal tooth mineralization and bone formation in zebrafish. PeerJ. 2020;8:e8515. pmid:32117623
- 42. O’Sullivan J, Bitu CC, Daly SB, Urquhart JE, Barron MJ, Bhaskar SS, et al. Whole-Exome sequencing identifies FAM20A mutations as a cause of amelogenesis imperfecta and gingival hyperplasia syndrome. Am J Hum Genet. 2011;88(5):616–20. pmid:21549343
- 43. Choi H, Kim T-H, Yun C-Y, Kim J-W, Cho E-S. Testicular acid phosphatase induces odontoblast differentiation and mineralization. Cell Tissue Res. 2016;364(1):95–103. pmid:26547858
- 44. Núñez SM, Chun Y-HP, Ganss B, Hu Y, Richardson AS, Schmitz JE, et al. Maturation stage enamel malformations in Amtn and Klk4 null mice. Matrix Biol. 2016;52–54:219–33. pmid:26620968
- 45. Gibson CW, Yuan ZA, Hall B, Longenecker G, Chen E, Thyagarajan T, et al. Amelogenin-deficient mice display an amelogenesis imperfecta phenotype. J Biol Chem. 2001;276(34):31871–5. pmid:11406633
- 46. Scohy S, Gabant P, Van Reeth T, Hertveldt V, Drèze PL, Van Vooren P, et al. Identification of KLF13 and KLF14 (SP6), novel members of the SP/XKLF transcription factor family. Genomics. 2000;70(1):93–101. pmid:11087666
- 47. Kurosaki T, Maquat LE. Nonsense-mediated mRNA decay in humans at a glance. J Cell Sci. 2016;129(3):461–7. pmid:26787741
- 48. Mårdh CK, Bäckman B, Holmgren G, Hu JC-C, Simmer JP, Forsman-Semb K. A nonsense mutation in the enamelin gene causes local hypoplastic autosomal dominant amelogenesis imperfecta (AIH2). Hum Mol Genet. 2002;11(9):1069–74. pmid:11978766
- 49. Papagerakis P, Lin H-K, Lee KY, Hu Y, Simmer JP, Bartlett JD, et al. Premature stop codon in MMP20 causing amelogenesis imperfecta. J Dent Res. 2008;87(1):56–9. pmid:18096894
- 50. Cho SH, Seymen F, Lee KE, Lee SK, Kweon YS, Kim KJ, et al. Novel FAM20A mutations in hypoplastic amelogenesis imperfecta. Hum Mutat. 2012;33(1):91–4.
- 51. Aldred MJ, Crawford PJ, Roberts E, Thomas NS. Identification of a nonsense mutation in the amelogenin gene (AMELX) in a family with X-linked amelogenesis imperfecta (AIH1). Hum Genet. 1992;90(4):413–6. pmid:1483698
- 52. Carugo O, Pongor S. A normalized root-mean-square distance for comparing protein three-dimensional structures. Protein Sci. 2001;10(7):1470–3. pmid:11420449
- 53. Zhang Y, Skolnick J. Scoring function for automated assessment of protein structure template quality. Proteins. 2004;57(4):702–10. pmid:15476259
- 54. Xu J, Zhang Y. How significant is a protein structure similarity with TM-score = 0.5? Bioinformatics. 2010;26(7):889–95. pmid:20164152
- 55. Wang S-K, Choi M, Richardson AS, Reid BM, Lin BP, Wang SJ, et al. ITGB6 loss-of-function mutations cause autosomal recessive amelogenesis imperfecta. Hum Mol Genet. 2014;23(8):2157–63. pmid:24305999
- 56. Wang S-K, Hu Y, Simmer JP, Seymen F, Estrella NMRP, Pal S, et al. Novel KLK4 and MMP20 mutations discovered by whole-exome sequencing. J Dent Res. 2013;92(3):266–71. pmid:23355523
- 57. Prokop JW, Lazar J, Crapitto G, Smith DC, Worthey EA, Jacob HJ. Molecular modeling in the age of clinical genomics, the enterprise of the next generation. J Mol Model. 2017;23(3):75. pmid:28204942