Genome scan reveals several loci associated with torus palatinus

Abstract

Objective

Torus Palatinus (TP) is a common trait with an unclear aetiology. Although prior studies suggest a hereditary component, the genetic factors that influence TP risk remain unknown. The purpose of this study is to identify genetic variants associated with TP.

Materials and Methods

We assessed the TP status of 829 individuals from various ancestral backgrounds using 3D palate scans. We then carried out a genome‐wide association study (GWAS) to identify common variants associated with TP. We also performed gene‐based tests across the exome to investigate the role of low‐frequency coding variants.

Results

Our GWAS did not identify any genome‐wide significant signals but identified suggestive associations including hits on chromosomes 2, 5 and 17 with p‐values less than 5 × 10⁻⁶. Candidate genes at these suggestive loci have been implicated in normal‐range craniofacial features, syndromes with facial and oral malformations, and bone density. We did not find evidence that low‐frequency coding variants influence TP risk. In addition, we failed to replicate associations identified in prior genetic studies of TP.

Conclusion

These findings suggest that multiple genes likely influence the development of TP. Independent replication will be required to confirm our suggestive associations.

1. INTRODUCTION

Torus palatinus (TP) is a midline bony oral exostosis located on the lingual surface of the hard palate. TP is common and shows a wide range of reported incidence (from less than 10% to over 60%), which varies by ancestry and sex. ¹ , ² Although generally considered benign when it occurs in isolation, TP can be mistaken for a tumour. ³ TP can also vary widely in size and shape. When very large, TP can interfere with normal oral function or the fit of oral prostheses and, in such cases, is sometimes removed or reduced surgically. ⁴

The aetiology of TP is unknown. While some studies have linked TP to biomechanical forces ⁵ or changes in bone density, ⁶ genetic factors are also believed to play an important role. Evidence to support a genetic aetiology includes familial recurrence patterns, differences in incidence by ancestry and the high rates of occurrence in certain genetic syndromes. Several family studies of TP have reported likely autosomal dominant inheritance with high penetrance, ⁷ , ⁸ , ⁹ suggesting monogenic or oligogenic inheritance. Moreover, recent studies have linked rare mutations in a handful of Wnt signalling pathway genes to TP, ¹⁰ , ¹¹ , ¹² , ¹³ which would seem to support this idea. However, TP is a relatively common trait in many populations, and such rare mutations would be expected to explain only a small fraction of hereditary TP cases. Complicating this picture, another recent association study concluded that TP genetics may be more complex, with multiple common genetic variants influencing risk. ¹⁴ However, because this was a very small study and the implicated genetic variants showed weak statistical evidence of association, the genetic basis of TP remains uncertain.

The goal of this study is to better understand the impact of genetic variants on TP risk. To assess the role of common variants, we conducted a genome‐wide scan of TP in 829 individuals. We then carried out gene‐based tests to evaluate the cumulative impact of low‐frequency coding variants. Finally, we tested previously associated genetic variants to determine if any replicate in our cohort.

2. MATERIALS AND METHODS

2.1. Study sample

The 829 individuals in this study were recruited as part of a larger multinational study of orofacial cleft aetiology, which involved the collection of maxillary impressions. To be selected for this analysis, participants had to be either an unaffected family member of an individual with a non‐syndromic cleft or an unaffected control with no prior family history of orofacial clefting. Participants were excluded if they were affected with an orofacial cleft, were from a family with a history of syndromic clefting, or indicated a history of surgical correction for TP. Recruitment sites included Pennsylvania, Puerto Rico, the Philippines, Colombia and Nigeria. Both self‐identified males (n = 353) and females (n = 476) were included. In addition, both adults and children were included, with ages ranging from 4 to 80 years (mean age was 26.4 years). This study was carried out in accordance with the World Medical Association Declaration of Helsinki. All participants provided written consent prior to participation in any research activities, and all study activities were approved by each participating institution's ethics oversight board.

2.2. Phenotyping approach

Maxillary impressions were obtained by trained dental personnel using standard hydrocolloid impression materials. These were then poured into plaster casts and digitised as 3D surface models by laser scanning (3Shape, Copenhagen, Denmark). The 3D maxillary models were visualised from multiple perspectives and evaluated for the presence or absence of TP by a single observer, who was blinded to the identity of each participant. We applied the same conservative criteria used in other recent reports, ² where an individual was determined to have TP if a clearly discernible and distinct elevation was present in the palatal midline. The reliability of TP phenotyping in this cohort has been evaluated by re‐rating the palate models of 20 randomly selected participants. ¹⁵ Three independent observers re‐rated the same 20 palate models, and the average intra‐observer Cohen's Kappa was 0.72, which indicates a substantial level of agreement. ¹⁶ An example scan with TP is shown in Figure 1.

FIGURE 1.

3D scan of a palatal cast showing a prominent torus palatinus.

2.3. Discovery GWAS approach

Participants were genotyped with the Infinium Global Diversity Array‐8 v1.0 Array with genome‐build GRCh37/hg19, which consists of 1,904,599 markers, including 1,894,665 genotyped single nucleotide polymorphisms (SNPs) and 9934 intensity‐only probes. Genotypes were lifted over to build GRCh38 and imputed using the TOPMed Imputation Server for all autosomal chromosomes and chromosome X. Standard data cleaning and imputation pipelines were implemented. Imputed genotypes with a certainty above 0.9 were included for variants that had high imputation quality (R ² > .8), no evidence of departure from Hardy‐Weinberg equilibrium (p > 1 × 10⁻⁴), and minor allele frequencies (MAF) > 3%, resulting in approximately 9.2 million SNPs available for analysis. We constructed ancestry principal components (PCs) based on PC analysis in a common set of LD‐pruned SNPs (Figure S1). We implemented the GWAS using the standard score test in GENESIS, ¹⁷ with sex, age and five ancestry PCs included as covariates and a genetic relatedness matrix to account for the phenotype correlation due to genetic sharing across the sample. We also ran models with and without cleft family history as a covariate to evaluate whether including the unaffected family members of individuals with an orofacial cleft had any effect on the outcome. GENESIS was used because a small number of our participants were related. Because we had unbalanced case and control sample sizes, we also ran the Saddle Point Approximation (SPA) version of the score test to verify the results. ¹⁸ The conventional p‐value threshold was used for genome‐wide statistical significance (p < 5 × 10⁻⁸). A p‐value threshold of p < 5 × 10⁻⁶ was used to identify suggestive associations.

2.4. Testing GWAS nominated variants in ancestral subsets

Due to the diversity present in our cohort, we wanted to evaluate how our GWAS‐nominated SNPs performed in specific ancestry groups. To obtain subgroups based on genetic ancestry, we performed supervised ancestry estimation using ADMIXTURE ¹⁹ to assign participants to their corresponding genetic ancestry cohorts based on maximum genetic ancestry proportions. This means that each participant was assigned to a designated genetic ancestry group based on their largest observed genetic ancestry proportion. A previously developed reference panel's genetic data ²⁰ was merged with the study data and quality controlled using BCFtools v1.18 ²¹ and PLINK v2.00a4LM, ²² respectively. Briefly, this reference panel consists of the sequenced genomes of anchor reference populations (Africa, East Asia, Europe, America, South Asia and Oceania) from the 1000 Genomes Project and the Human Genome Diversity Project. Quality control filters included bi‐allelic SNPs with less than 1% missingness, greater than 1 × 10⁻⁶ Hardy‐Weinberg equilibrium threshold, greater than 5% minor allele frequency and less than 0.1 LD r ² value. This analysis included 88,247 independent SNPs after quality control and resulted in relatively balanced African (n = 308), East Asian (n = 240) and European (n = 236) genetic ancestry cohorts but smaller genetic ancestry cohorts of American (n = 23) and South Asian (n = 22). All SNPs from discovery GWAS that surpassed our suggestive significance threshold of 5 × 10⁻⁶ and had a minor allele count of at least 10 were tested in each subgroup. The threshold for statistical significance for our subgroup analysis was determined by dividing 0.05 by the number of independent SNPs tested.

2.5. Gene‐based testing for low‐frequency coding SNPs

The Infinium Global Diversity Array‐8 v1.0 Array includes approximately 529,415 exonic variants. After applying standard data cleaning filters, enforcing a minor allele count threshold of at least three, a MAF threshold of ≤3% and excluding genes with fewer than two qualified variants, we performed gene‐based association testing on the subset of 793 unrelated individuals using 75,992 exonic variants located in 13,312 genes. The ≤3% MAF threshold was used to provide full coverage of low‐frequency variants not included in our GWAS. We ran CMC burden and SKAT permutation tests in the RVTESTS package ²³ with sex, age and five ancestry PCs included as covariates. We applied a Bonferroni adjusted p‐value threshold of p < 3.76 × 10⁻⁶ for statistical significance.

2.6. Testing candidate SNPs and regions from the literature

Nested within the GWAS, we performed in silico targeted testing of 15 common SNPs that showed prior suggestive evidence of association with TP. ¹⁴ A total of 14 of the 15 previously reported SNPs were available for testing through our GWAS SNP panel. Thus, we established a statistical threshold of p ≤ .0036 (0.05/14) as evidence of replication. We also evaluated a +/−250 kb window around several Wnt pathway genes (WLS, LRP4, LRP5, LRP6) and BMP4 for any evidence of an association signal. These genes were chosen because prior reports indicated rare mutations in familial cases of oral exostoses, including TP. ¹⁰ , ¹¹ , ¹² , ¹³

3. RESULTS

TP was present in 17.5% (n = 145) of our total sample. It occurred at a higher frequency in females (22.9%; n = 109) than in males (10.2%; n = 36), which is consistent with the literature.

While no signals exceeded the strict genome‐wide significance threshold (p < 5 × 10⁻⁸), our genome scan identified SNPs at three loci with p‐values less than 5 × 10⁻⁶. These included a peak on chromosome 2 (lead SNP: rs17013053; p = 8.28 × 10⁻⁷), a peak on chromosome 5 (lead SNP: rs4704136; p = 7.37 × 10⁻⁷) and peak on chromosome 17 (lead SNP: rs1859400; p = 9.51 × 10⁻⁷). These signals are shown in Figure 2. SNPs at 10 additional loci showed p‐values falling between 5 × 10⁻⁵ and 5 × 10⁻⁶. LocusZoom plots for these additional loci are included in Figure S2. Details of lead SNPs at all suggestive loci are included in Table 1, and a complete list of suggestive SNPs is provided in Table S1. When comparing our GWAS signals against published scRNAseq expression profiles from the posterior palate of E14.5 control mice, ²⁴ candidate genes at 10 of our 13 putative GWAS loci were present, with two genes (ADGRL3 and ANKRD11) showing strong expression.

FIGURE 2.

LocusZoom plots showing associations with TP at 2p12 (A) 5q13.3 (B) and 17q22 (C). The left y‐axis shows the log10‐transformed p‐values. The stippled line shows the conventional 5 × 10⁻⁸ p‐value threshold. Shading of the points represents the linkage disequilibrium (r ² based on the 1000 Genomes Project) between each SNP and the lead SNP, indicated by purple shading. The blue overlay shows the recombination rate (right y‐axis). Positions of genes are shown below the plot.

TABLE 1.

Lead SNPs at loci associated with torus palatinus.

Lead SNP	Chr:Pos	Allele	MAF	Type	Beta (SE)	p	Candidate Genes
rs55750412	1:239368030	T	0.16	I	0.90 (0.20)	4.09 × 10−6	FMN2
rs17013053	2:76715583	C	0.05	I	1.62 (0.33)	8.28 × 10−7	LRRTM4
rs150712164	3:82167060	C	0.03	I	1.88 (0.41)	4.12 × 10−6	GBE1
rs76123781	4:60358559	C	0.05	I	1.55 (0.33)	2.47 × 10−6	ADGRL3
rs4704136	5:74359470	A	0.26	I	−0.86 (0.17)	7.37 × 10−7	ENC1; HEXB; ARHGEF28
rs140849512	6:9181712	A	0.05	I	1.52 (0.33)	3.04 × 10−6	SLC35B3
rs17151837	7:128182807	C	0.04	I	1.63 (0.34)	2.24 × 10−6	PAX4; SND1
rs10117137	9:36960329	A	0.09	G	1.37 (0.29)	1.71 × 10−6	PAX5
rs17589032	11:37661871	T	0.04	I	1.61 (0.34)	2.23 × 10−6
rs4072344	16:89494024	T	0.07	G	−1.22 (0.26)	1.95 × 10−6	ANKRD11
rs1859400	17:58344423	G	0.11	G	1.25 (0.25)	9.51 × 10−7	SUPT4H1; MTMR4; TSPOAP1
rs72989043	19:5878269	T	0.05	G	1.52 (0.33)	3.42 × 10−6	RFX2
rs13043183	20:25294317	G	0.09	I	−1.59 (0.34)	3.81 × 10−6	GINS1

Abbreviations: MAF, minor allele frequency; Type: I, imputed; G, genotyped.

Both the standard score test and SPA test showed similar GWAS results. Likewise, including cleft family history as a covariate did not alter the results. These comparisons are shown in Figure S3. Thus, the GWAS results presented here are from the score test with age, sex and ancestry PCs as covariates.

Testing GWAS‐nominated SNPs in subgroups revealed some evidence of ancestry‐specific effects. For example, the association signal on chromosome five appears to be specific to individuals of East Asian ancestry. In many cases, though, these outcomes were driven by population differences in SNP informativeness (i.e., allele frequency differences). Moreover, due to low minor allele counts in some populations, many SNPs could not be effectively compared across groups. These results are shown in Table S2.

Gene‐based testing did not reveal any evidence that low‐frequency coding variants were significantly associated with TP risk. Four genes (AGER, STK33, ZSCAN22 and SIGLEC8) showed borderline associations, and these are presented in Figure S4.

We were able to test 14/15 previously reported common SNPs from the literature. None of the 14 tested SNPs showed any evidence of association in our study (Table S3). Moreover, there was no evidence of a signal near any of the previously implicated Wnt pathway genes (Figure S5).

4. DISCUSSION

Our GWAS identified several suggestive associations with TP, providing the strongest evidence to date that common variants impact oral exostoses. Although we lack functional data to identify functional SNPs or the genes they target, several potentially relevant candidate genes are located near our suggestive GWAS peaks. A handful of loci, for instance, are adjacent to genes previously implicated in facial morphology and dysmorphology. At locus 2p12, variants near LRRTM4 have been previously associated with measures of external mouth morphology. ²⁵ , ²⁶ , ²⁷ Incidentally, LRRTM4 is also a candidate gene for schizophrenia, which has been associated with a high rate of TP. ²⁸ At the 5p13.3 locus, variants near ENC1 have been associated with chin morphology, ²⁹ while variants near ARHGEF28 have been associated with cleft lip. ³⁰ At this same locus, variants near HEXB have been associated with external mouth morphology ²⁶ and damaging mutations in this gene have been reported to cause Sandhoff disease, which is characterised by craniofacial dysmorphology. ³¹ At locus 17q22, mutations in SUPT4H1 have been shown to result in Cockayne Syndrome B, which is characterised by facial and oral malformations. ³² While proximity of suggestive signals to genes previously implicated in facial morphology is notable, we warrant caution in interpreting these connections. Many gene products are multifunctional, and the etiological or mechanistic processes through which pathological variants in these genes lead to syndromes may not reflect the underlying biology of normal facial variation. That said, the relationship between TP and facial morphology is also supported by a recent study where we compared 3D maxillary arch shape between individuals with and without TP. ² This study found that those with a TP had a shorter and wider maxillary arch.

Genes at some TP loci have also been implicated in both syndromic and non‐syndromic forms of cleft palate. At locus 16q24.3, mutations in ANKRD11 have been reported to cause KGB Syndrome, which presents with craniofacial dysmorphology including cleft palate. ³³ Consequently, in mice, ankrd11 is expressed in the developing palate, and knock‐outs of this gene result in numerous craniofacial malformations, including cleft palate. ³⁴ At locus 7q32.1, variants near PAX4 have been reported to be associated with non‐syndromic cleft palate in a Chinese cohort. ³⁵ Because it is possible that the inclusion of unaffected relatives from families with a history of orofacial clefting in our TP study sample could drive these associations, we adjusted for family history status in our model and found that it had no effect on our association results. The more likely explanation is that the genetic overlap between TP and palatal anomalies is due to the role these genes play during palate formation.

Based on GWAS Catalogue results, genes at several of our reported TP loci have been previously associated with bone density; these include FMN2, GBE1, SND1 and RFX2. The link between TP and bone mineral density is supported by Belskey et al., ⁶ who reported that post‐menopausal women with a TP showed significantly higher bone mineral density than baseline measures from both peers and younger women. There was also a positive correlation between the size of the TP and measured bone mineral density, regardless of whether hormone replacement therapy was used. However, other studies do not support this relationship. ³⁶

We were unable to replicate any of the SNPs previously reported by Bezamat et al. ¹⁴ One explanation could be differences in the ancestral composition of our respective cohorts. The sample in Bezamat et al. was Filipino. While our sample also included individuals recruited from the Philippines, these individuals only represented a subset of our diverse cohort. Nevertheless, none of the 14 reported SNPs we tested showed low minor allele frequencies, suggesting that this was not a factor in the lack of replication. A more likely explanation is that the previously reported associations were false positives, as a stringent adjustment to control for type 1 error was not applied. We also failed to identify any associations between variants (common or low‐frequency) in or near genes involved in the Wnt pathway recently reported by Kantaputra and colleagues. ¹⁰ , ¹¹ , ¹² , ¹³ These reported mutations may simply be too rare to be picked up in our study.

Our findings, while certainly not definitive, suggest that TP development may be influenced by multiple genes. Our results do not provide support for a highly penetrant, autosomal dominant causal variant segregating in the majority of families with TP. If this was the case for a common trait like torus palatinus, then we would expect very robust GWAS signals, even with relatively modest sample sizes. The pattern of association we observed instead suggests a more complex genetic architecture. While this may be consistent with a polygenic inheritance pattern, it could also indicate genetic heterogeneity where different combinations of a more limited number of risk variants are implicated in different families. In such a scenario, any single risk variant would be expected to explain a small fraction of TP cases, making detection using GWAS more challenging.

Our study has several limitations that must be considered. First, our sample size is very small. This is the most likely explanation for our failure to identify signals surpassing strict genome‐wide significance. Nevertheless, it has been shown that GWAS signals in the suggestive range reflect biological enrichment. ³⁷ Our GWAS loci bear this out, implicating numerous genes with plausible roles in craniofacial and oral traits. In our phenotyping approach, we defined TP as a binary trait, similar to prior studies. There is always a degree of arbitrariness when defining traits that exhibit continuous variation as either present or absent. We attempted to be conservative in our phenotype definition to mitigate potential false positives. The downside with this approach is that more subtle variations of TP may have been missed. Measuring TP as a quantitative trait may be possible in future studies. Another important limitation is the lack of independent replication. This is the first study of TP to comprehensively survey the genome, and we are unaware of any comparably phenotyped cohorts that are currently available for either replication or meta‐analysis. As a result, the associations presented here should be interpreted cautiously and with appropriate scepticism.

AUTHOR CONTRIBUTIONS

Conceptualisation, SMW and JRS; methodology, MKL, AMES; JRS and SMW; formal analysis, MKL, NH and RMG; resources, CP, CJB, REL, CVR, CPRM, LMMU, WLA, AB, MLM, JRS, and SMW; writing – original draft preparation, MKL, NH and SMW; writing – review and editing, AMES, RMG, CP, CJB, REL, CVR, CPRM, LMMU, WLA, AB, MLM, JRS; visualisation, MKL, AMES; project administration, MLM, JRS and SMW; funding acquisition, MLM, JRS, SMW and CJB. All authors have read and agreed to the published version of the manuscript.

FUNDING INFORMATION

Funding for this work is provided in part by the National Institute for Dental and Craniofacial Research (http://www.nidcr.nih.gov/) through grants R01‐DE016148 (MM, SMW), R01‐DE032122 (JRS) X01‐HG011437 (JRS, MM), R00‐DE024571 (CJB), S21‐MD001830 (CJB) and U54‐GM133807 (CJB). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to declare.

ETHICS STATEMENT

Research activities were approved by the University of Pittsburgh's Human Research Protections Office (protocol: CR19080127). This study was carried out in accordance with the World Medical Association Declaration of Helsinki. All participants provided written consent prior to participation in any research activities.

Supporting information

Figure S1

Figure S2

Figure S3

Figure S4

Figure S5

Table S1

Table S2

Table S3

Lee MK, El Sergani AM, Herrick N, et al. Genome scan reveals several loci associated with torus palatinus. Orthod Craniofac Res. 2025;28:159‐165. doi: 10.1111/ocr.12857

DATA AVAILABILITY STATEMENT

Data used in this analysis are available through the NIH dbGaP controlled‐access repository (https://www.ncbi.nlm.nih.gov/gap/) at accession phs002815.v2.p1. Full GWAS summary statistics are available upon request.