GWAS of Dental Caries Patterns in the Permanent Dentition

Abstract

The importance of susceptibility genes in the risk for dental caries has been clearly established. While many candidate caries genes have been proposed, to date, few of them have been rigorously validated through observational and experimental studies. Moreover, most genetic epidemiological studies have analyzed global caries phenotypes that ignore the possibility that genes may exert differential effects across tooth surfaces of the dentition. Therefore, we performed genome-wide association studies (GWAS) of 5 novel dental caries phenotypes (developed by clustering the permanent dentition into categories of tooth surfaces based on co-occurrence of caries) to nominate new candidate caries genes. GWAS was performed in 920 self-reported white participants, aged 18 to 75 years, with genotype data on 518,997 genetic variants. We identified a significant genetic association between dental caries of the anterior mandibular teeth and LYZL2 (p value = 9e-9), which codes a bacteriolytic agent thought to be involved in host defense. We also identified a significant genetic association between caries of the mid- dentition tooth surfaces and AJAP1 (p value = 2e-8), a gene possibly involved in tooth development. Suggestive genetic associations were also observed for ABCG2, PKD2, the dentin/bone SCPP sub-family, EDNRA, TJFBR1, NKX2-3, IFT88, TWSG1, IL17D, and SMAD7 (p values < 7e-6). We nominate these novel genes for future study.

Introduction

Dental caries is the most common chronic disease worldwide and affects approximately 90% of adults in the US (Beltran-Aguilar et al., 2005). Many factors contribute to disease, including environmental, endogenous, and behavioral factors such as diet, bacterial flora, fluoride exposures, oral hygiene, salivary flow and composition, morphological and positional characteristics of the dentition, host immune response, and preventive interventions. Genes and gene-by-environmental interactions are widely acknowledged to contribute to caries susceptibility through mechanisms affecting these and other unknown risk factors. The heritability of dental caries is high (30-55%) (Boraas et al., 1988; Shaffer et al., 2012a), and previous studies, including a genome-wide association study (GWAS) (Shaffer et al., 2011) have sought to identify the specific genetic factors involved. However, to date, the causal roles of caries genes have not been rigorously established. The vast majority of genetic variants affecting disease remain unknown.

To date, a potential limitation of most genetic epidemiological studies of dental caries is the global phenotype definition. Most genetics studies have used a single measure of decay, such as a binary (yes/no) affection status or the DMFT/S index (calculated as the sum of decayed, missing due to decay, or filled/restored teeth/surfaces). As global measures of decay, phenotypes such as these ignore the fact that tooth surfaces across the dentition exhibit differences in susceptibility to dental caries and are differentially affected by risk factors. By ignoring the patterns of decay, global caries measures may be poor phenotypes for identifying genetic variants affecting dental caries, which are likely to exert individually weak and differential effects across the various surfaces of the permanent dentition. In a related study presented in this issue of the Journal, we describe the use of hierarchical clustering to generate novel caries phenotypes that capture the non-uniform risk of caries across tooth surfaces of the dentition (Shaffer et al., 2013). In this study, we performed GWAS for these novel dental caries phenotypes to nominate candidate caries genes for future observational and experimental studies.

Methods

Sample Recruitment

The sample was recruited as part of an initiative by the Center for Oral Health Research in Appalachia (COHRA) to study the factors affecting oral health in a rural, underserved population. Details regarding recruitment and data collection for this sample have been described previously (Polk et al., 2008) and are summarized in a separate article (Shaffer et al., 2013). Nine hundred twenty self-reported whites, aged 18 to 75 yrs, with dental caries assessments and genome-wide genotype data were included in this study.

Novel Caries Phenotypes

To study the genetic factors influencing dental caries, which we hypothesized may exert differential effects across tooth surfaces of the permanent dentition, we first developed novel caries phenotypes that capture the patterns of decay. Development of these novel phenotypes is fully described in a separate article (Shaffer et al., 2013). In summary, we used hierarchical clustering to group tooth surfaces based on co-occurrence of dental caries. We hypothesized that tooth surfaces that co-vary with respect to dental caries may be similarly influenced by genetic risk factors. Cluster analysis yielded 5 stable clusters (Table 1), one of which, the mid-dentition surfaces, could be further partitioned into sub-clusters comprised of maxillary and mandibular tooth surfaces. For each cluster (or sub-cluster), we generated the “partial DMFS index”, which is the count of carious surfaces (i.e., non-cavitated or cavitated decay, missing due to decay, or restored) for a given cluster. We denote these novel phenotypes as DMFS1, DMFS2, DMFS3, DMFS4, and DMFS5 (which is sub-divided into DMFS5_max and DMFS5_mand).

Table 1.

Novel Dental Caries Phenotypes Generated via Cluster Analysis on Tooth-surface-level Caries Data

		Partial DMFS		Heritability			GWAS Results (# loci)
Phenotype	Tooth Surfaces Included	Mean	SD	h2	p value	λ	Significanta	Suggestiveb
DMFS1	pit and fissure molar	6.37	3.75	0.27	0.057	1.022	-	-
DMFS2	mandibular anterior (incisors, canines, first pre-molar)	0.66	2.07	0.54	0.003	1.030	1	8
DMFS3	posterior (molars and maxillary pre-molars excluding molar pit and fissure)	7.78	8.84	0.43	0.004	1.053	0	5
DMFS4	maxillary anterior (incisors)	2.42	3.98	0.00	0.500	0.997	-	-
DMFS5	mid-dentition (pre-molars and canines)	3.70	6.01	0.40	0.008	1.047	0	10
DMFS5max	maxillary mid-dentition	1.86	3.73	0.34	0.041	1.041	1	6
DMFS5mand	mandibular mid-dentition	1.84	2.97	0.31	0.016	1.038	0	10

^ap value threshold for genome-wide significant associations = 10^−7.3λ.

^bp value threshold for suggestive associations = 10^−5λ.

Mean = mean number of carious surfaces within a cluster.

SD = standard deviation.

h² = heritability estimate.

p value = significance of test that heritability estimate differs from zero; bold indicates p value < 0.05.

λ = genomic inflation factor.

Genotyping

Genotyping was performed at the Center of Inherited Disease Research (CIDR) of Johns Hopkins University. Data cleaning and quality assurance procedures were conducted in conjunction with the CIDR data cleaning center at the University of Washington. The Illumina Human610-Quadv1_B BeadChip (Illumina, Inc., San Diego, CA, USA) and Illumina Infinium II assay protocol were used. In general, genotyping quality was excellent; 518,997 genetic markers (single-nucleotide polymorphisms; SNPs) passing quality control and analysis filters (participant call rates > 90%; SNP call rates > 99%; Hardy-Weinberg p values > 0.0001; minor allele frequency > 0.02) were analyzed in this study. Details regarding genotyping and data cleaning are publicly available (www.genevastudy.org).

Statistical Analysis

GWAS were performed for each heritable dental caries phenotype (i.e., DMFS2, DMFS3, DMFS5, DMFS5_max, DMFS5_mand). Genetic association for each of the 518,997 SNPs was tested by linear regression in PLINK (Purcell et al., 2007), adjusted for the effects of sex, age, and age². The genomic inflation factor, λ, and Manhattan plots were generated in R (R Foundation for Statistical Computing, Vienna, AU). Given the issue of multiple comparisons, conservative p value thresholds for statistical significance were set equal to 5e-8 after adjustment for genomic inflation (i.e., p ≤ 10^−7.3λ). The p value thresholds for suggestive significance were set equal to 1e-5 after adjustment for genomic inflation (i.e., p ≤ 10^−5λ). Under the GWAS approach, associated markers are not expected to be causal, but are assumed to be physically proximal and in linkage disequilibrium (i.e., correlated due to ancestry) with unobserved causal variants. Moreover, where a causal variant may be situated with respect to the gene it influences is currently unknown, but, typically, is thought to be also physically proximal. Therefore, we explored and report the known biological functions of genes near our significant and suggestive hits for plausible roles in dental caries.

Results

The study sample included 920 self-reported whites, aged 18 to 75 yrs, with available data for caries phenotypes, covariates (i.e., age, age², and sex), and genetic markers. Novel dental caries phenotypes reflecting decay in different categories of tooth surfaces were generated by cluster analysis. A separate article describes the development and heritability estimation of the phenotypes used herein (Shaffer et al., 2013). Descriptive statistics for novel caries phenotypes are shown in Table 1. The genomic inflation factor (λ) ranged from 1.00 to 1.05, indicating minimal to minor inflation of p values compared with that expected by chance alone. Thresholds for genome-wide and suggestive statistical significance were adjusted for the genomic inflation factor observed for each GWAS. Manhattan plots illustrating the GWAS results for the 5 heritable dental caries phenotypes (i.e., DMFS2, DMFS3, DMFS5, DMFS5_max, and DMFS5_mand) are shown in the Fig.

Figure.

Manhattan plots showing GWAS results for (A) DMFS2, (B) DMFS3, (C) DMFS5, (D) DMFS5_max, and (E) DMFS5_mand. Solid lines represent thresholds for genome-wide significance (p value < 10^−7.3λ). Dotted lines represent thresholds for suggestive significance (p value < 10^−5λ).

Two loci exceeded the threshold for genome-wide significance (Table 2). The most significant association that we observed was for DMFS2 (mandibular anterior tooth surfaces) and LYZL2 on chromosome 10p11.23 (rs399593; p value = 9.4e-9). Though very little is known about LYZL2 specifically, it belongs to the family of c-type lysosomes, which are well-recognized bacteriolytic factors of host defense (Zhang et al., 2005). It is currently unknown whether LYZL2 affects dental caries, although its putative antibacterial function suggests a biologically plausible mechanism through which LYZL2 may influence cariogenesis. The second most significant association that we observed was for DMFS5_max and the upstream un-translated region of AJAP1 on chromosome 1p (rs3896439; p value = 2.4e-8). The protein product of AJAP1, SHREW1, interacts with basigin, a mediator of matrix metalloprotease (MMP) activity (Schreiner et al., 2007) involved in tooth development in rat (Schwab et al., 2007) and mouse models (Kumamoto and Ooya, 2006). This association was also observed for DMFS5 at the suggestive level of significance (rs3896439; p value = 1.1e-6). Likewise, associated genetic variants upstream of AJAP1 were among the top hits in an independent GWAS of dental caries in COHRA children (unpublished observations). These lines of evidence suggest that AJAP1 may influence dental caries, potentially through the regulation tooth development.

Table 2.

Significant (p value and gene symbol in bold) and Suggestive Associated Loci

Chr.	Phenotype(s)	Top SNP	BP	p value	Gene(s)	Corroborative Evidence/Plausible Role in Dental Caries
1p36	DMFS5max, DMFS5	rs3896439	4568530	2E-8	AJAP1	Interacts with basagin (Schreiner et al., 2007), a mediator of MMP activity during tooth development (Kumamoto and Ooya, 2006; Schwab et al., 2007)
1p36	DMFS5mand	rs9308447	9354977	5E-7	SPSB1	Unknown
1p22	DMFS5	rs1750491	84998887	2E-6	LPAR3	Unknown
1p21	DMFS3, DMFS5, DMFS5max	rs11166135	99121424	1E-6	LPPR5	Unknown
1q32	DMFS5, DMFS5max	rs7552806	203050641	3E-6	NFASC	Unknown
2p24	DMFS5, DMFS5mand	rs10180496	12886351	1E-7	TRIB2	Unknown
3q21	DMFS2	rs9810890	130135243	6E-6	ACAD9	Unknown
4p15	DMFS5max	rs2531154	15627422	9E-7	PROM1	Unknown
4q22	DMFS5mand, DMFS5	rs3114018	89283605	7E-8	ABCG2	Transporter protein and stem-cell marker expressed in human dental pulp (Honda et al., 2007) and in developing mouse incisor (Li et al., 2011); increased expression in ameloblastic tumors suggests that ABCG2 may regulate the maintenance of odontogenic tissues (Kumamoto and Ohki, 2010)
					PKD2	Pkd2 mutations in mice caused dental loss, root fractures (unpublished observations presented at IADR 2012 by Khonsari, http://iadr.confex.com/iadr/2012rio/webprogram/Paper160464.html)
					dentin/bone sub-family SCPP (SPP1, MEPE, IBSP, DMP1, and DSPP)	SCPP sub-family consisting of paralogous genes coding extracellular matrix proteins of dentin/bone; shares homology with enamel SCPP sub-family (Kawasaki and Weiss, 2008)
4q31	DMFS3	rs11100904	147123754	2E-6	ZNF827	Unknown
4q31	DMFS2, DMFS5mand	rs1429138	148501792	2E-7	EDNRA	Signaling gene expressed during early craniofacial development; mice knock-outs exhibit severe craniofacial defects (Ruest et al., 2004)
6p25	DMFS3	rs2476842	511741	3E-6	EXOC2	Unknown
6q22	DMFS2	rs1204798	116650540	2E-6	NT5DC1 (6q22.1)	Unknown; modest genetic association (p value = 0.02) in a candidate gene study following up linkage signal (Vieira et al., 2008)
7q11	DMFS2	rs848452	77434748	2E-6	PHTF2	Unknown
7q22	DMFS5mand	rs10242311	105060956	6E-7	ATXN7L1	Unknown
8p22	DMFS5max, DMFS5	rs10111661	20380853	2E-6	LZTS1	Unknown
9q22	DMFS5	rs649057	101274144	5E-6	300 Kb from TGFBR1	Strongly expressed in ameloblasts; promotes MMP20 expression during amelogenesis (Gao et al., 2009)
					300 Kb from NR4A3	Transcription factor up-regulated in dental follicle cells during osteogenic differentiation (Morsczeck et al., 2009)
10p14	DMFS5mand	rs11256676	10641866	5E-6	SFTA1P	Unknown
10p11	DMFS2	rs399593	30952036	9E-9	LYZL2	Bacteriolytic factor thought to be involved in host defense (Zhang et al., 2005)
10q22	DMFS5	rs2441755	67835273	4E-6	CTNNA3	Unknown
10q24	DMFS5mand, DMFS2	rs7078219	101264355	6E-6	NKX2-3	Involved in salivary gland and tooth morphogenesis (Biben et al., 2002)
13q12	DMFS2	rs735539	20178034	4E-6	IFT88	Mutation in IFT88 leads to increased SHH signaling during development, resulting in ectopic extra molars (Ohazama et al., 2009)
					IL17D	Cytokine that enhances pro-inflammatory response during bacterial infection via up-regulation of TLR4 (Guzzo et al., 2012)
13q21	DMFS3	rs2875517	66594146	3E-6	PCDH9	Unknown
13q33	DMFS2	rs17485138	108300850	7E-6	MYO16	Unknown
17q11	DMFS5mand	rs12602978	22591207	7E-7	WSB1	Unknown
18p11	DMFS2	rs2864527	9278456	4E-6	TWSG1	Regulates BMP signaling in the mandibular arch; mouse knock-outs exhibit craniofacial defects and salivary gland dysmorphogenesis (Melnick et al., 2006; MacKenzie et al., 2009)
18q21	DMFS5max	rs357894	44833968	4E-6	SMAD7	Regulator of TGF-beta mediated tooth development (Xu et al., 2003); expressed in human tooth bud (Bao et al., 2003); regulator of enamel deposition (Klopcic et al., 2007)
18q23	DMFS3	rs13381277	72447598	4E-6	ZNF516	Unknown
19p12	DMFS5mand	rs931608	22405962	1E-6	ZNF98	Unknown
21q21	DMFS5mand	rs2829459	25201222	3E-6	(gene desert)	Unknown
Xq26	DMFS5, DMFS5max	rs3788848	129027493	3E-6	BCORL1	Shares homology with BCOR, a gene affecting tooth development (Fan et al., 2009); association observed in GWAS of related dental caries phenotype in this cohort

In addition to the 2 genome-wide significant associations described above, we also observed several suggestive association signals. The strongest of these was observed for DMFS5_mand and a locus on chromosome 4q22.1 harboring several tooth-related genes. The top SNP in this region (rs3114018; p value = 7.478e-8) was in ABCG2, a transporter protein and stem-cell marker expressed in human dental pulp (Honda et al., 2007) and in developing mouse incisors (Li et al., 2011). Increased expression in ameloblastic tumors suggests that ABCG2 may regulate the maintenance of odontogenic tissues (Kumamoto and Ohki, 2010). The adjacent gene, PKD2, has also been implicated in craniofacial development: PKD2 mutations in mice cause dental loss, root fractures, and other craniofacial anomalies, and are associated with facial asymmetry in humans (Khonsari; unpublished observations presented at IADR 2012, http://iadr.confex.com/iadr/2012rio/webprogram/Paper160464). The associated locus also includes the 5 paralogous genes of the dentin/bone SCPP sub-family (i.e., SPP1, MEPE, IBSP, DMP1, and DSPP), which are extracellular matrix proteins involved in tooth (and bone) formation (Kawasaki and Weiss, 2008). Genetic variation at this locus may affect one or more of these genes related to tooth development, which, in turn, may affect susceptibility to dental caries.

Many additional suggestive associations were observed, including several in or near genes with biologically plausible roles in caries susceptibility, as summarized in Table 2. Although none of these genes has previously been implicated in dental caries, their putative roles in tooth, salivary gland, and craniofacial development, and in response to bacterial infection, suggest plausible mechanisms through which they may influence cariogenesis. Other suggestive associations were observed for genetic loci that have no obvious biological functions related to dental caries.

Discussion

Dental caries is an extremely multi-factorial disease, whereby genes operating through numerous biological avenues may influence susceptibility. In this study, we performed genome-wide association scans for novel dental caries phenotypes developed under the hypothesis that genetic variants (as is the case with environmental risk factors [Shaffer et al., 2012b]) may differentially affect caries risk across tooth surfaces of the permanent dentition. Two genome-wide significant loci were observed: LYZL2 for mandibular anterior surfaces (i.e., surfaces of the incisors, canines, and premolars), and AJAP1 for mid-dentition surfaces (i.e., surfaces of the premolars and maxillary canines). Numerous suggestive loci were also observed.

A theme among the associated loci was genes thought to be involved in craniofacial, salivary gland, and tooth development, including AJAP1, ABCG2, PKD2, the dentin/bone SCPP sub-family, EDNRA, TJFBR1, NKX2-3, IFT88, TWSG1, and SMAD7. Developmental genes such as these add to the growing list of host genes, including taste preference, enamel-related, and bactericidal genes, that have been chosen for investigation. In light of the observational study design, we cautiously interpret our results to nominate these loci as putative caries genes, pending confirmation in additional epidemiological and experimental studies. More work, including functional studies, will be required to determine whether associated loci may be useful for disease prediction, diagnosis, or treatment. Either way, the novel associations identified here may lead to better understanding of the biological mechanisms influencing cariogenesis.

For the most part, the significant and suggestive associations identified in this study did not overlap with loci identified via GWAS of global caries phenotypes of the permanent dentition in this sample (unpublished observations) or global phenotypes of the primary dentition in a related sample of children from the same population (Shaffer et al., 2011). This observation suggests that data-driven caries phenotypes that attempt to capture biologically informative patterns of tooth decay, such as those generated for this study, may be useful for identifying and understanding the genetic contributions to dental caries. Indeed, the X-linked association between DMFS5 and BCORL1 was also observed in a GWAS of smooth-surface caries, a similar caries phenotype defined based on a priori tooth-surface classifications (unpublished observations).

In the interpretation of our results, a few design aspects of the GWAS approach warrant discussion. First, we view GWAS as a discovery-driven approach, with the goal of generating new hypotheses (as opposed to the paradigm of testing hypotheses). Therefore, in addition to scrutinizing our genome-wide significant hits (which met a very strict significance threshold, even considering the multiple testing), we also examined and report suggestive associations. While the genome-wide significant associations are extremely unlikely to occur by chance alone, associations meeting suggestive significance may occur by chance (though likely not as many as we observed). Therefore, we expect that some of the suggestive associations observed in this study may be spurious, whereas others are likely true reflections of genetic risk. Ideally, these associations should be replicated in an independent sample; however, we analyzed novel caries phenotypes developed specifically for this GWAS from tooth-surface-level caries data not currently available in any other adult sample. Therefore, replication in an independent sample is not currently possible. This limitation is partly countered by the corroborating evidence of gene functions connecting many of our associated loci to tooth-related biological processes. In short, many of our associations make sense, and therefore, we propose prioritizing them among the growing list of caries candidate genes. Additional work is needed to explore the independent or interacting effects of these candidate genes with known environmental risk factors, such as fluoride exposures, dietary and oral hygiene behaviors, and demographics.

A major strength and innovation of this study was the use of novel caries phenotypes, which were developed by hierarchical clustering analysis on tooth-surface-level data. We have previously shown that grouping surfaces based on co-occurrence of caries resulted in sensible categories, which were reproducible in NHANES 1999-2000, an independent national cohort (Shaffer et al., 2013). Some of the novel caries phenotypes were heritable and/or exhibited significant correlations with putative environmental predictors such as age, sex, educational attainment, and toothbrushing behaviors (Shaffer et al., 2013). Here, we showed that these novel caries outcomes were also useful for identifying genetic loci, which may exert differential effects across categories of tooth surfaces. For example, bacteriolytic LYZL2 and pro-inflammatory IL17D, both genes involved in host response to bacterial infection, were implicated in DMFS2, which reflects caries of the mandibular anterior tooth surfaces (i.e., incisors, canines, and first premolars). These surfaces are similar in their positions in the mouth and their comparatively low prevalence of caries. In contrast, several tooth development genes were implicated for DMFS5 (and sub-clusters DMFS5_max and DMFS5_mand), reflecting caries of the mid-dentition surfaces (i.e., pre-molars and canines, which exhibit moderate caries prevalence). DMFS3, which reflects posterior surfaces excluding the molar pit and fissure surfaces, showed associations with a number of genetic loci, though no obvious candidate genes were identified. In contrast, DMFS1 (reflecting molar pit and fissure surfaces, which exhibit the greatest caries prevalence) and DMFS4 (reflecting maxillary incisors, which exhibit moderate caries prevalence) were not significantly heritable, and therefore, GWAS for these outcomes were not considered. These two categories of tooth surfaces are likely strongly influenced by important environmental factors (e.g., sealants and dietary choices) that may render them less informative for discovering genetic risk factors. Overall, we take these results to support our premise that susceptibility genes exert their effects differentially across the permanent dentition and that modeling tooth-surface-level caries data can benefit gene-mapping efforts.

In conclusion, we identified many novel genetic associations for dental caries and nominated several new caries genes. Moreover, we observed that different genetic variants were associated with caries across the different categories of tooth surfaces, which supports our central hypothesis. This study demonstrated the utility of novel dental caries phenotypes for identifying genetic risk factors, which we believe may prove useful for understanding the biological mechanisms predisposing to dental caries, and, ultimately, may lead to improvements in disease prevention, detection, and treatment.