Abstract
Drug-induced gingival overgrowth (DIGO) is a multifactorial adverse effect associated with calcium channel blockers and antiepileptic drugs, yet genetic susceptibility-particularly in Asian populations-remains poorly defined. We performed whole-genome sequencing in 74 Thai individuals, including 36 DIGO cases and 38 drug-exposed nonresponder controls. A genome-wide association study was conducted with adjustment for amlodipine exposure, which showed a significant association with DIGO. Complementary analyses included gene-based testing using Multi-marker Analysis of GenoMic Annotation, fine-mapping with Sum of Single Effects, haplotype analysis, and receiver operating characteristic-based risk modeling. Although no single nucleotide polymorphisms (SNPs) reached genome-wide significance, we identified 350 lead SNPs across 34 genes showing strong associations with DIGO (p < 0.001). After multiple-testing correction, six genes/SNPs remained statistically significant (p < 0.05), representing the most robust findings. Haplotype analysis implicated TTC7B, RWDD1, TOM1L1, C1QL2, and BBS1 as DIGO risk genes. These genes, not previously linked to DIGO, are involved in cellular trafficking, phosphoinositide signaling, ciliary function, and protein regulation. Our findings indicate that DIGO susceptibility is driven by genetically determined cellular response pathways in the presence of amlodipine rather than by pharmacokinetic mechanisms. The absence of associations with CYP2C9 and HLA variants previously reported in other populations highlights the importance of ethnically diverse pharmacogenomic studies.
Supplementary Information
The online version contains supplementary material available at 10.1038/s41598-026-52779-y.
Keywords: Gingival overgrowth, Genetic susceptibility, Calcium channel blocker, Gingival hyperplasia, Gingival hypertrophy, Risk factors for gingival enlargement
Subject terms: Computational biology and bioinformatics, Diseases, Genetics, Medical research
Introduction
Gingival overgrowth is an enlargement of gingival tissue with altered morphology, arising from genetic predisposition, systemic conditions, or medication use1 (Fig. 1). Hereditary gingival fibromatosis (HGF; MIM 135300) is a rare autosomal dominant or recessive disorder linked to pathogenic variants in SOS1, REST, and ZNF862. Prevalence among exposed patients shows substantial variability across the three drug classes, ranging from 10 to 50% for phenytoin, 13–85% for cyclosporin, and 5–31% for amlodipine2–4. Not all patients exposed to the same drug and dose develop drug-induced gingival overgrowth (DIGO), however, indicating strong interindividual variability. A genetic basis is also supported by twin studies, which demonstrate differences in fibroblast proliferation between monozygotic and dizygotic pairs5.
Fig. 1.
Clinical features of drug-induced gingival overgrowth.
Genetic associations have implicated CYP2C9 (MIM 601130), which metabolizes all three drug classes, with reduced activity linked to higher drug levels and increased DIGO risk6. Immune variation also plays a role, with HLA-DR2 (MIM 142857) conferring susceptibility and HLA-DR1 (MIM 142860) appearing protective7. Fibrotic responses may further be influenced by genes regulating extracellular matrix turnover and epithelial-to-mesenchymal transition, including SPOCK1 (MIM 613795) and CTSL (MIM 116880)8.
This study is the first to apply whole-genome sequencing (WGS) to identify genetic determinants of DIGO, in contrast to prior studies that were limited to candidate-gene approaches and small cohorts9. WGS enables comprehensive, unbiased discovery of both common and rare variants, offering deeper insight into the genetic basis of gingival overgrowth. By integrating variant-, gene-, and haplotype-level analyses, we aimed to identify novel susceptibility loci and evaluate their contribution to DIGO risk prediction.
Materials and methods
Study population and data collection
Ethical approval for this study was obtained from the Human Experimentation Committee of the Faculty of Dentistry, Chiang Mai University (No. 17/2023). All experimental methods adhered to the relevant guidelines and regulations established by Chiang Mai University. We obtained written informed consent from all participants. We recruited 74 Thai patients from two hospitals who had received treatment for ≥ 3 months with a DIGO-associated drug (amlodipine, phenytoin, sodium valproate, carbamazepine, topiramate, lamotrigine, levetiracetam, phenobarbital, or benzodiazepines). Based on clinical presentation, the cohort was divided into 36 DIGO cases and 38 nonresponsive controls without gingival enlargement (Fig. 2). Demographic, clinical, and drug history data were systematically collected.
Fig. 2.
Flowchart summarizing patient recruitment, genetic analysis, statistical analysis, and key findings.
Whole-genome sequencing and quality control
Whole-genome sequencing (WGS) was performed by ThaiOmics (Chonburi, Thailand) on the DNBSEQ-T7 platform (MGI Tech, Shenzhen, China) at 30–40× coverage. Reads were aligned to GRCh38 using BWA-MEM10, and joint genotyping was done with GATK HaplotypeCaller11. Variants passing PLINK2 QC filters (genotype missingness < 5%, individual missingness < 5%, MAF > 5%)12 were retained.
Genome-wide association study (GWAS)
Previously reported rare coding variants predicted to be deleterious in established hereditary gingival fibromatosis genes (HP:0000169) were not identified in this cohort using whole-genome sequencing. We therefore conducted a genome-wide association study (GWAS) to evaluate common variant contributions to DIGO susceptibility. GWAS was conducted using PLINK2, employing logistic regression adjusted for amlodipine exposure, age, sex, and principal components to account for population stratification. Due to the limited sample size, a more relaxed p-value threshold was applied to capture candidate variants while still maintaining control over false-positive associations.
Gene-based association and fine-mapping
We conducted a two-stage analysis to enhance the detection and interpretation of association signals. To increase statistical power and interpretability, gene-based tests were performed using MAGMA13. MAGMA aggregates SNP-level associations into gene-level statistics while accounting for linkage disequilibrium (LD). To control for multiple testing, false discovery rate (FDR) adjustment using the Benjamini–Hochberg procedure was applied, with significance defined as FDR < 0.05. Genes with gene-level associations passing this threshold were subsequently fine-mapped using SuSiE14, which estimates the probability that each SNP is truly causal (posterior inclusion probability, PIP) and identifies credible SNP sets. This integrative approach enabled the prioritization of candidate variants for downstream analyses.
Haplotype analysis and risk prediction
For MAGMA-significant genes, haplotype blocks were defined using the PLINK2 LD algorithm and tested for DIGO association via logistic regression adjusting for amlodipine and demographics15. A predictive model incorporating haplotypes and amlodipine status was then evaluated using AUC16. Analyses were performed in PLINK2, MAGMA, SuSiE, and R programs17, with results considered hypothesis-generating pending independent validation.
Results
Study population and clinical characteristics
The demographic and clinical characteristics of the 74 patients are summarized in Table 1. The DIGO and nonresponsive control groups were well-matched, showing no significant differences in sex distribution, mean age, or drug exposure duration. Notably, amlodipine exposure was the only variable significantly associated with DIGO. Amlodipine use was over three times more prevalent in cases (34.2%) than in controls (10.5%; p = 0.013). Exposure rates for other common medications, including phenytoin, sodium valproate, and levetiracetam, did not differ significantly between the groups (Fig. 3a,b). The proportion of patients on monotherapy was identical in both cohorts.
Table 1.
Baseline patient demographics, drug exposure patterns, and polypharmacy distribution of DIGO cases and non-responsive controls. This table summarizes patient demographics, drug exposure patterns, and polypharmacy distribution, with p-values from Fisher’s exact test for categorical variables and t-test for continuous variables. Significant differences (p < 0.05) are highlighted for amlodipine exposure. p-values from Fisher’s exact test for categorical variables and t-test for continuous variables. Significant differences (p < 0.05) are highlighted for amlodipine exposure. All data are expressed to one significant digit.
| Demographics | DIGO cases (N = 36) | Control (N = 38) | p-value |
|---|---|---|---|
| Sex, female; N (%) | 17 (47.22%) | 19 (50%) | 0.8243 |
| Age (years); mean ± SD | 16.58 ± 19.32 | 19.01 ± 20.73 | 0.5171 |
| Drug exposure duration (months); Mean ± SD | 34.03 ± 70.23 | 70.27 ± 108.15 | 0.5188 |
| Drug exposure; N (% within group) | |||
| Amlodipine | 13 (34.21%) | 4 (10%) | 0.0125 |
| Phenytoin | 13 (34.21%) | 13 (32.5%) | 1.0000 |
| Sodium valproate | 11 (28.95%) | 8 (20%) | 0.4287 |
| Carbamazepine | 4 (10.53%) | 5 (12.5%) | 1.0000 |
| Topiramate | 6 (15.79%) | 6 (15%) | 1.0000 |
| Lamotrigine | 5 (13.16%) | 5 (12.5%) | 1.0000 |
| Levetiracetam | 12 (31.58%) | 11 (27.5%) | 0.8028 |
| Phenobarbitone | 4 (10.53%) | 10 (25%) | 0.1385 |
| Benzodiazepines | 2 (5.26%) | 5 (12.5%) | 0.4310 |
| Number of concurrent drugs; N (% within group) | |||
| Single drug | 19 (50%) | 19 (47.5%) | 1.0000 |
| Two drugs | 5 (13.16%) | 12 (30%) | 0.1005 |
| Three or more drugs | 12 (31.58%) | 7 (18.42%) | 0.1903 |
Fig. 3.
Drug exposure patterns in DIGO cases and controls. (A) Comparative bar chart of. specific drug exposure frequencies, highlighting elevated amlodipine use in cases with 95% confidence intervals. (B) Bar chart illustrating the distribution of concurrent drug use (single [1], two [2], or three [3] or more [> 3] drugs) between DIGO cases and controls.
Gene-based association and Fine-mapping identify six susceptibility loci
After adjusting for amlodipine exposure, the Genome-Wide Association Study (GWAS) detected no genome-wide significant variants (p < 5 × 10⁻⁸), consistent with limited cohort power (Fig. 4a and supplementary Table 1). We identified six DIGO-associated genes after False Discovery Rate (FDR) correction: BBS1, C1QL2, RSPH4A, RWDD1, TOM1L1, and TTC7B by using Gene-based testing with MAGMA (Fig. 4b and Supplementary Table 2).
Fig. 4.
Manhattan plots of single-variant and gene-based associations in drug-induced gingival overgrowth (DIGO) and Multivariate risk prediction model for DIGO using genetic haplotypes and amlodipine exposure. (A) Single-variant GWAS results, with –log10(p-values) on the y-axis and variants ordered by chromosomal position on the x-axis; top variants meeting a soft threshold (p < 0.001) are highlighted. (B) Gene-based associations from MAGMA analysis, with –log10(adjusted p-values) on the y-axis and genes ordered by chromosomal position on the x-axis; significant genes (FDR < 0.05), including BBS1, C1QL2, RSPH4A, RWDD1, TOM1L1, and TTC7B, are indicated.
BBS1 was suggested to be a leading candidate locus based upon fine-mapping (Figs. 5 and 6). The intronic SNP rs1791683 received the highest posterior inclusion probability (PIP = 1.0) in the SuSiE model; however, given the modest sample size and local linkage disequilibrium structure, this finding should be interpreted cautiously and considered hypothesis-generating pending independent replication. Other loci showed multiple suggestive non-coding variants, suggesting that altered regulatory mechanisms are consistent with risk. Several lead variants also differed markedly in frequency between gnomAD, a large-scale reference database that aggregates diverse human genomic data to provide allele frequency information for comparison with study populations, and Thai datasets (ThaiGER; https://thaiger.genomicsthailand.com), underscoring the value of diverse population studies (Table 2).
Fig. 5.
Fine mapping of candidate variants in C1QL2, RSPH4A, and RWDD1.
Fig. 6.
Fine mapping of candidate variants in BBS1, TTC7B, and TOM1L1.
Table 2.
Summary of GWAS and MAGMA results for significant genes.
| Gene | Genomic positions (GRCh38) rsID | Consequences | Allele frequencies | GWAS (AML_adjusted) | SuSiE PIP | LD- pruned status | MAGMA (17 genes)§ | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| gnomADg v.4.1 database | thADg database# | log (OR) (95% CI) | P-value | SNPs | p-value | FDR | |||||
| BBS1 |
chr11:66524089:A: G rs1791683 |
Intron variant | 0.65 | 0.57 | 8.78 (5.22,12.34) | 0.00081 | 100.0% | YES | 5 | 0.002 | 0.020 |
|
chr11:66532992:G: A rs1791686 |
3 prime UTR variant | 0.34 | 0.48 | − 8.05 (− 11.6,− 4.5) | 0.00206 | 33.13% | – | ||||
|
chr11:66514067:A: G rs1671062 |
Intron variant | 0.66 | 0.52 | − 8.05 (− 11.6,− 4.5) | 0.00206 | 33.13% | YES | ||||
|
chr11:66532044:A: G rs8432 |
3 prime UTR variant | 0.66 | 0.52 | − 8.05 (− 11.6,− 4.5) | 0.00206 | 33.13% | – | ||||
| C1QL2 |
chr2:119158737:T: A rs2121217 |
5 prime UTR variant | 0.20 | 0.58 | 6.79 (3.97,9.61) | 0.00106 | 22.81% | – | 2 | 0.001 | 0.020 |
|
chr2:119157933:G: T rs1317848 |
Synonymous variant | 0.20 | 0.58 | 6.79 (3.97,9.61) | 0.00106 | 22.81% | YES | ||||
| RSPH4A |
chr6:116633023:T: C rs784136 |
Downstream gene variant | 0.21 | 0.13 | 8.42 (5.21,11.62) | 0.00036 | 10.90% | – | 6 | 0.001 | 0.020 |
| RWDD1 |
chr6:116578159:C: G rs7757967 |
Intron variant | 0.81 | 0.87 | 8.76 (5.35,12.16) | 0.00047 | 21.14% | – | 8 | 0.001 | 0.020 |
|
chr6:116591572:A: C rs4142086 |
Intron variant | 0.81 | 0.87 | 8.76 (5.35,12.16) | 0.00047 | 21.14% | – | ||||
|
chr6:116597448:G: A rs1062353 |
3 prime UTR variant | 0.81 | 0.87 | 8.76 (5.35,12.16) | 0.00047 | 9.37% | – | ||||
|
chr6:116577253:A: C rs7752566 |
Intron variant | 0.81 | 0.87 | 8.76 (5.35,12.16) | 0.00047 | 21.14% | YES | ||||
|
chr6:116580477:G: A rs1321534 |
intron variant | 0.81 | 0.87 | 8.76 (5.35,12.16) | 0.00047 | 21.14% | – | ||||
|
chr6:116587423:G: A rs4946184 |
Intron variant | 0.81 | 0.87 | 8.76 (5.35,12.16) | 0.00047 | 21.14% | – | ||||
|
chr6:116598041:T: G rs6915260 |
Downstream gene variant | 0.81 | 0.87 | 8.76 (5.35,12.16) | 0.00047 | 9.37% | – | ||||
|
chr6:116602586:T: C rs6912594 |
Downstream gene variant | 0.81 | 0.87 | 8.76 (5.35,12.16) | 0.00047 | 9.37% | – | ||||
| TOM1L1 |
chr17:54923631:T: C rs7224810 |
Intron variant | 0.99 | 0.73 | 7.15 (3.77,10.53) | 0.00406 | 4.69% | – | 13 | 0.004 | 0.041 |
|
chr17:54921706:C: G rs2958910 |
Intron variant | 0.98 | 0.73 | 7.17 (3.84,10.5) | 0.00347 | 5.08% | YES | ||||
|
chr17:54909364:A: G rs4372751 |
Intron variant | 0.68 | 0.73 | 7.99 (4.59,11.39) | 0.00141 | 8.04% | YES | ||||
|
chr17:54946668:C: G rs2908862 |
Intron variant | 0.99 | 0.80 | 7.95 (4.53,11.38) | 0.00162 | 7.48% | YES | ||||
|
chr17:54911699:A: T rs7222890 |
Intron variant | 0.69 | 0.72 | 7.99 (4.59,11.39) | 0.00141 | 8.04% | – | ||||
|
chr17:54942935:A: G rs2958947 |
Intron variant | 0.98 | 0.73 | 7.15 (3.77,10.53) | 0.00406 | 4.69% | – | ||||
| TTC7B |
chr14:90719101:C: A rs10150862 |
Intron variant | 0.63 | 0.30 | 8.78 (5.59,11.97) | 0.00018 | 4.84% | – | 95 | 0.001 | 0.020 |
|
chr14:90675084:A: G rs12894664 |
Intron variant | 0.53 | 0.71 | 9.22 (5.9,12.54) | 0.00016 | 5.12% | YES | ||||
|
chr14:90675492:A: G rs753310 |
Intron variant | 0.53 | 0.71 | 10.32 (6.85,13.79) | 5.00E−05 | 6.55% | – | ||||
|
chr14:90814720:G: A rs1286305 |
Intron variant | 0.38 | 0.49 | 3.89 (1.51,6.26) | 0.02603 | 4.52% | – | ||||
|
chr14:90719154:C: T rs12896077 |
Intron variant | 0.38 | 0.27 | 8.93 (5.66,12.19) | 0.0002 | 5.33% | – | ||||
|
chr14:90813113:A: C rs1286306 |
Intron variant | 0.33 | 0.49 | 3.89 (1.51,6.26) | 0.02603 | 4.52% | YES | ||||
This table details variants within MAGMA-significant genes, including genomic positions, rsIDs, consequences, allele frequencies (gnomADg and thADg), GWAS statistics (log (OR), 95% CI, p-value), SuSiE posterior inclusion probabilities (PIP), LD-pruned status, and gene-level MAGMA p-values with FDR correction. Lead variants per gene are highlighted for clarity. Derived allele frequencies were obtained from the ThaiGeR database (https://thaiger.genomicsthailand.com). MAGMA p-values were FDR-corrected for multiple testing. Full details of the gene sets tested are provided in Supplementary Tables 1–2.
Haplotype-based risk prediction for DIGO
Haplotype blocks within six Multi-marker Analysis of GenoMic Annotation (MAGMA) identified genes were analyzed for DIGO risk, with five showing significant associations (Table 3):
Table 3.
Haplotype association analysis and risk prediction for DIGO susceptibility.
| Gene | Variant(s) | Haplotype/genotype | DIGO (Freq in DIGO) | Normal (Freq in normal) | AF/HF DIGO: normal | χ2 | χ2 P-value | Univariate P-value | Multivariate P-value |
|---|---|---|---|---|---|---|---|---|---|
| C1QL2 |
chr2:119157933:G: T (rs1317848) |
T | 32 (0.44) | 51 (0.67) | 0.39 | 7.71 | 0.0055 | Ref | NA |
| G | 40 (0.56) | 25 (0.33) | 0.62 | 0.00416 | 0.02782 | ||||
| RWDD1 |
chr6:116577253:A: C (rs7752566) |
C | 50 (0.69) | 70 (0.89) | 0.42 | 12.38 | 0.0004 | Ref | NA |
| A | 22 (0.31) | 6 (0.08) | 0.79 | 0.00168 | 0.01416 | ||||
| BBS1 |
chr11:66514067:A: G chr11:66524089:A: G (rs1671062 rs1791683) |
GG | 32 (0.44) | 38 (0.5) | 0.46 | 6.13 | 0.1054 | Ref | NA |
| GA | 0 (0) | 1 (0.01) | 0 | 0.98731 | |||||
| AG | 1 (0.01) | 7 (0.09) | 0.13 | 0.19719 | |||||
| AA | 39 (0.54) | 30 (0.33) | 0.57 | 0.18917 | |||||
| TTC7B |
chr14:90675084:A: G chr14:90813113:A: C (rs1286306 rs12894664) |
GA | 27 (0.38) | 37 (0.43) | 0.42 | 8.73 | 0.0332 | Ref | NA |
| GC | 20 (0.28) | 25 (0.26) | 0.44 | 0.42511 | |||||
| AC | 16 (0.22) | 13 (0.16) | 0.55 | 0.11491 | |||||
| AA | 9 (0.13) | 1 (0) | 0.9 | 0.007 | 0.01439 | ||||
| TOM1L1 |
chr17:54909364:A: G chr17:54921706:C: G chr17:54946668:C: G rs4372751 rs2958910 rs2908862 |
GGG | 50 (0.69) | 59 (0.62) | 0.46 | 9.98 | 0.0188 | Ref | NA |
| ACG | 1 (0.01) | 2 (0.01) | 0.33 | 0.5699 | |||||
| Other | 3 (0.04) | 9 (0.1) | 0.25 | 0.0873 | |||||
| ACC | 18 (0.25) | 6 (0.07) | 0.75 | 0.0069 | 0.14817 |
This table presents haplotype frequencies and statistical associations for five key genetic loci identified through MAGMA analysis. For each chromosome location, variant identifiers (rsIDs), haplotype/genotype combinations, frequencies in DIGO cases and controls, allele/haplotype frequency ratios, and statistical significance from chi-square tests, univariate analysis, and multivariate logistic regression are shown. Significant associations (p < 0.05) demonstrate differential haplotype distributions between cases and controls, with several haplotypes maintaining significance after multivariate adjustment for confounding factors. χ2 = chi-square test; Ref = reference haplotype. Significant associations (p < 0.05) are indicated for both univariate and multivariate logistic regression analyses.
C1QL2 (chr2, rs1317848): G allele enriched in cases (0.56 vs. 0.33), significant in both univariate (p = 0.0042) and multivariate (p = 0.028) analyses.
RWDD1 (chr6, rs7752566): Strongest signal; A allele enriched (0.31 vs. 0.08), highly significant in univariate (p = 0.0017) and multivariate (p = 0.014) analyses.
BBS1 (chr11, rs1671062–rs1791683): AA haplotype enriched (0.54 vs. 0.33) but not statistically significant.
TTC7B (chr14, rs1286306–rs12894664): AA haplotype strongly associated (0.13 vs. 0.01), significant in both univariate (p = 0.007) and multivariate (p = 0.014) analyses.
TOM1L1 (chr17, rs4372751–rs2908862): ACC haplotype enriched (0.25 vs. 0.07), significant in univariate (p = 0.0069) but not in multivariate analysis.
Comprehensive risk prediction model performance
We developed a multivariate logistic regression model combining significant haplotypes with amlodipine exposure to predict DIGO risk (Fig. 4a,b). Among genetic factors (Fig. 7a), the TTC7B chr14:AA haplotype showed the strongest effect (log OR 2.85, 95% CI 0.57–5.13), followed by RWDD1 chr6:A (log OR 1.47, 95% CI 0.30–2.65), C1QL2 chr2:G (log OR 1.10, 95% CI 0.12–2.08), and TOM1L1 chr17:ACC (log OR 1.02, 95% CI − 0.36–2.39). Amlodipine exposure itself remained an independent and strong predictor (log OR 2.11, 95% CI 0.63–3.59).
Fig. 7.
Multivariate risk prediction model for DIGO using genetic haplotypes and amlodipine exposure. (A) Forest plot displaying estimated beta coefficients (log odds ratios) and 95% confidence intervals. (B) ROC curves comparing predictive performance across three models: amlodipine exposure only (AUC = 0.628), genetic haplotypes only (AUC = 0.848), and combined genetic and pharmacological factors (AUC = 0.89).
Model performance was evaluated by Receiver Operating Characteristic (ROC) Curve analysis. Amlodipine exposure alone provided modest discrimination (Area Under the Curve = 0.628). Incorporating haplotypes markedly improved accuracy (Area Under the Curve = 0.848), and the combined model achieved excellent performance (Area Under the Curve = 0.89) (Fig. 7b). These findings indicate that genetic variation explains substantially more risk than drug exposure alone, and that integrating both yields clinically meaningful predictive accuracy. This combined approach may support personalized prescribing and risk stratification for DIGO.
Discussion
This study represents the first comprehensive pharmacogenomic investigation of DIGO in a Thai population, using WGS to elucidate novel genetic susceptibility factors. By integrating gene-based aggregation, haplotype analysis, and statistical fine-mapping, we aimed to mitigate some of the statistical limitations inherent in a modest-sized cohort. These approaches do not, however, eliminate the need for validation in larger, independent populations. Our findings reveal novel genetic pathways potentially underlying DIGO susceptibility and demonstrate that integrating these genetic factors can improve risk prediction beyond traditional clinical metrics alone.
Our analysis identified six genes significantly associated with DIGO (BBS1, C1QL2, RSPH4A, RWDD1, TOM1L1, and TTC7B), most of which harbored high-confidence regulatory variants rather than changes in coding regions. These genes converge on biological pathways including ciliary function (BBS1 and RSPH4A), protein regulation and signaling (RWDD1 and C1QL2), and cellular trafficking (TOM1L1 and TTC7B), suggesting that DIGO susceptibility arises from disrupted cellular response mechanisms rather than pharmacokinetic effects. Haplotype analyses confirmed population-specific risk variants, particularly in RWDD1 and TTC7B. Importantly, risk prediction modeling demonstrated that genetic haplotypes provided substantially greater predictive accuracy for DIGO (AUC = 0.848) than amlodipine exposure alone (AUC = 0.628), with the combined model achieving excellent performance (AUC = 0.89). These findings highlight the importance of gene-environment interactions in DIGO pathogenesis and demonstrate that genetic or ethnic background is a stronger determinant of risk than drug exposure alone.
In this study, amlodipine exposure emerged as a significant independent risk factor for DIGO. Amlodipine, a long-acting calcium channel blocker, reduces calcium influx into vascular smooth muscle cells18. Calcium ions are critical regulators of fibroblast activity, influencing cell growth, apoptosis, and extracellular matrix turnover19. At the gingival tissue level, amlodipine not only blocks L-type calcium channels but also perturbs calcium-dependent signaling in fibroblasts20. Such disruption may enhance fibroblast proliferation and fibrogenic activity, contributing to gingival enlargement.
Our genetic findings provide a mechanistic framework for this pharmacologic effect. BBS1 and RSPH4A, both linked to ciliary function, suggest that altered ciliary signaling may amplify fibroblast responsiveness to calcium imbalance. RWDD1 and C1QL2, associated with protein regulation and signaling, point to dysregulation of intracellular cascades that govern fibroblast activation and matrix remodeling. TOM1L1 and TTC7B, involved in cellular trafficking, may impair calcium handling and protein turnover in gingival connective tissue. Collectively, these pathways converge on fibroblast hyperactivation, extracellular matrix accumulation, and fibrotic remodeling-the hallmarks of DIGO pathogenesis. Importantly, while amlodipine exposure was more frequent among DIGO cases, multivariate analysis revealed that genetic factors (AUC = 0.848) provided substantially greater predictive power than drug exposure alone (AUC = 0.628). This finding challenges the traditional view of DIGO as primarily a dose-dependent adverse drug reaction and highlights individual genetic susceptibility as the predominant determinant of risk via pathways for ciliary function, cellular trafficking, and protein regulation.
The role of pharmacogenetics in DIGO has historically centered on drug metabolism. Variants in genes encoding cytochrome P450 enzymes such as CYP2C9 and CYP2C19 are associated with altered metabolism of phenytoin influencing DIGO susceptibility by modifying systemic drug levels6,21. Immune-related loci have also been implicated, with HLA-DR2 and HLA-B37 linked to increased risk of gingival overgrowth in patients receiving phenytoin or cyclosporine22.
In contrast, our comprehensive pharmacogenetic analysis based on metabolizer phenotypes did not identify associations with these previously reported candidate genes, including CYP2C9 and HLA variants (Supplementary Tables 3–5). This absence suggests that DIGO risk is not explained by single major-effect genes, but instead reflects a more complex genetic architecture, thereby reinforcing the need for unbiased genome-wide approaches to uncover novel risk loci.
Instead, our gene-based rare variant analysis revealed six novel susceptibility genes—TTC7B, RWDD1, TOM1L1, C1QL2, BBS1, and RSPH4A. These genes implicate biological pathways not previously linked to DIGO, including cellular trafficking, phosphoinositide signaling, ciliary function, and protein regulation. This shift from classical pharmacokinetic explanations toward cellular response pathways suggests that DIGO arises less from altered systemic drug levels and more from genetically determined differences in how gingival fibroblasts and surrounding tissues respond to drug exposure. This genetic framework helps explain why our predictive modeling showed that inherited haplotypes contribute more strongly to DIGO risk than pharmacological exposure alone, underscoring the predominance of gene-environment interactions in disease pathogenesis.
TTC7B showed the strongest association with DIGO (OR = 17.31, p = 0.007), with the chr14: AA haplotype yielding the highest effect size in our multivariate model. This represents a novel association, as TTC7B has not previously been implicated in gingival disorders. TTC7B encodes a scaffold protein containing a tetratricopeptide repeat (TPR) domain that maintains plasma membrane identity23. The scaffold protein stabilizes and recruits phosphatidylinositol 4-kinase alpha (PI4KA) to the membrane through interaction with the palmitoylated protein EFR3, forming a multiprotein complex24. This complex produces phosphatidylinositol 4-phosphate (PI4P), the precursor for PI(4,5)P2 and subsequently PI(3,4,5)P3 via PI3K activation25,26.
The PI3K/Akt pathway is a well-established driver of fibrotic remodeling in multiple organs, including lung and liver27,28, and has also been implicated in nifedipine-induced gingival overgrowth29. Interestingly, TTC7B exerts tissue-specific effects; in colon cancer, it inhibits cell proliferation through the PI3K/AKT-RXRA-FTO axis23, suggesting that downstream signaling outputs may depend on the cellular environment. Gingival fibroblasts appear uniquely sensitive to calcium dynamics: calcium influx enhances proliferative responses in gingival fibroblasts but not in pulmonary, dermal, or muscular fibroblasts, highlighting the presence of distinct calcium-sensing mechanisms in gingival tissue30,31. In this context, variants in TTC7B may disturb normal phosphoinositide signaling under calcium channel blockade, amplifying fibroblast proliferation and extracellular matrix deposition, thereby driving tissue-specific fibrotic responses characteristic of DIGO.
RWDD1, TOM1L1, C1QL2, and BBS1 haplotypes were significantly associated with DIGO risk, underscoring their potential roles in genetic susceptibility. Although RSPH4A showed a strong single-variant association, it was excluded from haplotype analysis because its variants were in strong linkage disequilibrium (r2 > 0.8) with those in RWDD1, and only the representative RWDD1 variant was retained after LD pruning, which removes highly correlated SNPs, keeping only representative variants (Fig. 8). Consequently, the robust haplotype-level associations observed for RWDD1 likely capture the underlying signal at the RSPH4A locus.
Fig. 8.
Linkage disequilibrium blocks of causal variants identified by fine mapping.
RWDD1 (OR = 4.96, p = 0.00168) encodes a transcriptional coactivator involved in hormone receptor signaling and extracellular matrix modulation32. The transcriptional coactivator may regulate AP-1-mediated TIMP-1 expression through interactions with DRG-2 and estrogen receptors33. Although its role in gingival tissue is unclear, RWDD1-mediated transcriptional coactivation could alter fibroblast activity under drug exposure by shifting MMP/TIMP balance-enhancing expression of matrix metalloproteinases (MMPs) while suppressing their inhibitors (TIMPs)-a mechanism previously implicated in drug-induced gingival overgrowth.
TOM1L1 (OR = 4.15, p = 0.0069) is a VHS domain–containing adaptor protein, which modulates Src family kinases (SFKs)34,35. It negatively regulates SFK-mediated DNA synthesis induced by PDGF36 and, together with the kinase Fyn, participates in EGFR-PI3K/Akt signaling37. Knockdown of TOM1L1 disrupts EGFR internalization38, suggesting that variants affecting its function could impair growth factor responses and promote abnormal gingival fibroblast proliferation.
BBS1 (OR = 2.95, p = 0.00206) points to ciliary dysfunction as a novel mechanism in DIGO. Primary cilia act as mechanosensors and regulate signaling pathways such as Hedgehog and WNT39. Defective ciliary signaling may disrupt how gingival cells integrate mechanical and growth signals, predisposing to fibroblast hyperactivation. Supporting this, gingival overgrowth has been sporadically reported in patients with Bardet–Biedl syndrome40.
C1QL2 (also known as nCLP2) encodes a secreted C1q-like protein expressed mainly in the central nervous system, where it maintains synaptic connections through multimerization and interaction with kainate receptors41. Although its function in gingival tissue remains unclear, the structural role of C1q proteins in extracellular matrix organization, where their collagen-like regions facilitate binding to core ECM proteins and promote scaffold stability, raises the possibility that C1QL2 variants may contribute to DIGO through altered matrix regulation in fibroblasts.
A key feature of our results is the striking difference in allele frequencies between Thai and global populations. According to gnomAD, C1QL2 variants were present at 0.58 in Thai individuals compared to 0.20 globally, underscoring why earlier candidate gene studies in other populations failed to identify these associations. Such population-specific variation highlights the need for diverse pharmacogenomic studies to ensure that risk prediction tools are clinically relevant.
Importantly, our integrated prediction model (AUC = 0.89) may have direct clinical utility. Studies need to be initiated to determine if pre-treatment genetic screening could stratify patients into high- and low-risk categories: those carrying multiple risk haplotypes could be offered alternative antihypertensives to prevent DIGO, while low-risk patients could safely continue amlodipine therapy. If confirmed, this proactive strategy would transform DIGO management from late-stage intervention to personalized prevention, representing a major step toward precision medicine in both dental and cardiovascular care.
From an implementation perspective, risk stratification does not require WGS in routine practice. Instead, a targeted genotyping panel of the top haplotypes (e.g., TTC7B, RWDD1, C1QL2, BBS1, TOM1L1) could provide rapid, low-cost testing before drug initiation. Such a panel could be integrated into existing pharmacogenomic workflows, similar to CYP2C9/VKORC1 testing in anticoagulant prescribing42. Over time, incorporation of DIGO risk variants into multi-gene pharmacogenomic platforms could support population-wide screening, enabling clinicians to personalize antihypertensive therapy while reducing adverse outcomes.
Limitations and future directions
The modest cohort size constrained statistical power, and the findings should, therefore, be interpreted as exploratory. This study leverages integrated bioinformatic approaches to generate novel, data-driven hypotheses regarding the genetic architecture of DIGO susceptibility. These hypotheses require confirmation in larger, independent cohorts. The absence of associations with previously reported candidates such as CYP2C9 and HLA may reflect population-specific risk factors in Thai individuals or methodological differences between candidate-gene and genome-wide approaches. Functional studies are necessary to elucidate the biological roles of the prioritized genes, particularly the potential link between TTC7B-mediated phosphoinositide signaling and calcium channel blocker-induced fibroblast activation. Replication across additional Asian and non-Asian populations will be essential to determine the robustness, generalizability, and translational relevance of these findings.
Conclusions
This study represents the first whole-genome sequencing–based investigation of drug-induced gingival overgrowth (DIGO) in a Thai population and suggests a novel genetic architecture underlying susceptibility. Through integrated genome-wide, gene-based, haplotype, and fine-mapping analyses, we identified six previously unreported candidate susceptibility genes—TTC7B, RWDD1, TOM1L1, C1QL2, BBS1, and RSPH4A-implicating pathways related to cellular trafficking, phosphoinositide signaling, ciliary function, and protein regulation. These findings broaden current understanding of DIGO beyond traditional pharmacokinetic and immune-related mechanisms and suggest that genetically mediated cellular response pathways may contribute to disease risk.
While amlodipine exposure remained an important clinical factor, incorporation of genetic haplotypes improved risk discrimination (combined model AUC = 0.89), supporting a role for gene–environment interaction in DIGO pathogenesis. The lack of association with previously reported candidates such as CYP2C9 and HLA, together with population-specific allele frequency differences, underscores the importance of conducting pharmacogenomic studies in diverse populations.
If validated in larger, independent cohorts, these findings may inform future risk stratification strategies and support the development of targeted genotyping approaches for personalized prevention. Overall, this work demonstrates the potential of genome-wide approaches to generate new hypotheses regarding the genetic basis of adverse drug reactions and provides a foundation for further mechanistic and translational studies.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgements
AcknowledgementsWe thank our patients and their families for their kind cooperation and for giving us the consent to use the medical and dental information for the benefit of other patients. This work was supported by Chiang Mai University, The Dental Association of Thailand, and the Genomics Thailand Research Grant of Health Systems Research Institute (P.K.).
Author contributions
T.K., K.K.C.K., T.J., R.W., and C.N. contributed to conception, design, data acquisition, data analysis and interpretation, and drafted the manuscript. V.N. and S.T. contributed to conception, design, data acquisition, data analysis and interpretation, drafted the manuscript, and critically revised manuscript. M.C.H. contributed to conception, drafted the manuscript, and critically revised manuscript. P.N.K. contributed to conception, design, data acquisition, data analysis and interpretation, supervised the project, drafted the manuscript, and critically revised manuscript. All authors gave their final approval and agreed to be accountable for all aspects of the work.
Funding
This work was supported by the Genomics Thailand initiative, under the pharmacogenomics focus, funded by the Health Systems Research Institute (P.K.), The Dental Association of Thailand (P.K.), and Chiang Mai University (P.K.).
Data availability
All results and data are available in the main manuscript or supplementary file. GWAS summary statistics, MAGMA statistics, and haplotype analysis data (RData format) generated in this study are available at 10.6084/m9.figshare.31835932.
Declarations
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Thunchanok Kiattiubolwong and Vorthunju Nakhonsri contributed equally to this work.
Contributor Information
Sissades Tongsima, Email: sissades.ton@biotec.or.th.
Piranit Nik Kantaputra, Email: dentaland17@gmail.com.
References
- 1.Straka, M., Varga, I., Erdelský, I., Straka-Trapezanlidis, M. & Krňoulová, J. Drug-induced gingival enlargement. Neuro Endocrinol. Lett.35, 567–576 (2014). [PubMed] [Google Scholar]
- 2.Allman, S. D., McWhorter, A. G. & Seale, N. S. Evaluation of cyclosporin-induced gingival overgrowth in the pediatric transplant patient. Pediatr. Dent.16, 36–40 (1994). [PubMed] [Google Scholar]
- 3.Angelopoulos, A. P. & Goaz, P. W. Incidence of diphenylhydantoin gingival hyperplasia. Oral Surg. Oral Med. Oral Pathol.34, 898–906. 10.1016/0030-4220(72)90228-9 (1972). [DOI] [PubMed] [Google Scholar]
- 4.Gopal, S. et al. Prevalence of gingival overgrowth induced by antihypertensive drugs: A hospital-based study. J. Indian Soc. Periodontol. 19, 308–311. 10.4103/0972-124x.153483 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Cockey, G., Boughnnan, J. & Hassell, T. Phenytoin response of gingival fibroblasts from human twins. J. Dent. Res.66, 320 (1987). [Google Scholar]
- 6.Babu, S. P., Ramesh, V., Samidorai, A. & Charles, N. S. Cytochrome P450 2C9 gene polymorphism in phenytoin induced gingival enlargement: A case report. J. Pharm. Bioallied Sci.5, 237–239. 10.4103/0975-7406.116828 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Radwan-Oczko, M., Boratyńska, M., Klinger, M. & Zietek, M. Risk factors of gingival overgrowth in kidney transplant recipients treated with cyclosporine A. Ann. Transpl.8, 57–62 (2003). [PubMed] [Google Scholar]
- 8.Imagawa, M. et al. Epithelial-to-mesenchymal transition, inflammation, subsequent collagen production, and reduced proteinase expression cooperatively contribute to cyclosporin-A-induced gingival overgrowth development. Front. Physiol.14, 1298813. 10.3389/fphys.2023.1298813 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Hwang, J. et al. Genetic analysis of hereditary gingival fibromatosis using whole exome sequencing and bioinformatics. Oral Dis.23, 102–109. 10.1111/odi.12583 (2017). [DOI] [PubMed] [Google Scholar]
- 10.Li, H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM. arXiv preprint arXiv:1303.3997 (2013).
- 11.McKenna, A. et al. The genome analysis toolkit: A MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res.20, 1297–1303. 10.1101/gr.107524.110 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Chang, C. C. et al. Second-generation PLINK: rising to the challenge of larger and richer datasets. Gigascience410.1186/s13742-015-0047-8 (2015). [DOI] [PMC free article] [PubMed]
- 13.de Leeuw, C. A., Mooij, J. M., Heskes, T. & Posthuma, D. MAGMA: Generalized gene-set analysis of GWAS data. PLoS Comput. Biol.11, e1004219. 10.1371/journal.pcbi.1004219 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Zou, Y., Carbonetto, P., Wang, G. & Stephens, M. Fine-mapping from summary data with the sum of single effects model. PLoS Genet.18, e1010299. 10.1371/journal.pgen.1010299 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Schaid, D. J. Evaluating associations of haplotypes with traits. Genet. Epidemiol.27, 348–364. 10.1002/gepi.20037 (2004). [DOI] [PubMed] [Google Scholar]
- 16.Janssens, A. C. et al. Predictive testing for complex diseases using multiple genes: Fact or fiction? Genet. Med.8, 395–400. 10.1097/01.gim.0000229689.18263.f4 (2006). [DOI] [PubMed] [Google Scholar]
- 17.Team, R. C. R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2023).
- 18.Godfraind, T. Discovery and development of calcium channel blockers. Front. Pharmacol.8, 286. 10.3389/fphar.2017.00286 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Roach, K. M. & Bradding, P. Ca2+ signalling in fibroblasts and the therapeutic potential of K(Ca)3.1 channel blockers in fibrotic diseases. Br. J. Pharmacol.177, 1003–1024. 10.1111/bph.14939 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Elghoulbzouri, H., Er-raji, S. & Ennibi, O. Periodontal management of amlodipine-induced gingival over growth: A 2 years follow-up case report. J. Med. Dent. Sci. Res.5, 1–5 (2018). [Google Scholar]
- 21.Charles, N. S. et al. Drug-induced gingival overgrowth: the genetic dimension. N. Am. J. Med. Sci.6, 478–480. 10.4103/1947-2714.141651 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Thomason, J. M. et al. Determinants of gingival overgrowth severity in organ transplant patients. An examination of the rôle of HLA phenotype. J. Clin. Periodontol. 23, 628–634. 10.1111/j.1600-051x.1996.tb00586.x (1996). [DOI] [PubMed] [Google Scholar]
- 23.Ren, Q. et al. TTC7B triggers the PI4KA-AKT1-RXRA-FTO axis and inhibits colon cancer cell proliferation by increasing RNA methylation. Int. J. Biol. Sci.21, 1127–1143. 10.7150/ijbs.102431 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Suresh, S. et al. Molecular basis for plasma membrane recruitment of PI4KA by EFR3. Sci. Adv.10, eadp6660. 10.1126/sciadv.adp6660 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Batrouni, A. G. & Baskin, J. M. The chemistry and biology of phosphatidylinositol 4-phosphate at the plasma membrane. Bioorg. Med. Chem.40, 116190. 10.1016/j.bmc.2021.116190 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Stojilkovic, S. S. & Balla, T. PI(4,5)P2-dependent and -independent roles of PI4P in the control of hormone secretion by pituitary cells. Front. Endocrinol. (Lausanne). 14, 1118744. 10.3389/fendo.2023.1118744 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Bhatt, J., Ghigo, A. & Hirsch, E. PI3K/Akt in IPF: Untangling fibrosis and charting therapies. Front. Immunol.16, 1549277. 10.3389/fimmu.2025.1549277 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Shamsan, E. et al. The role of PI3k/AKT signaling pathway in attenuating liver fibrosis: A comprehensive review. Front. Med. (Lausanne). 11, 1389329. 10.3389/fmed.2024.1389329 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Ju, Y., Huang, L., Wang, S. & Zhao, S. Transcriptional analysis reveals key genes in the pathogenesis of nifedipine-induced gingival overgrowth. Anal. Cell Pathol. (Amst)2020, 6128341. 10.1155/2020/6128341 (2020). [DOI] [PMC free article] [PubMed]
- 30.Hattori, T. & Wang, P. L. Elevation of cytosolic calcium level triggers calcium antagonist-induced gingival overgrowth. Eur. J. Pharmacol.583, 37–39. 10.1016/j.ejphar.2008.01.024 (2008). [DOI] [PubMed] [Google Scholar]
- 31.Minowa, E. et al. Enhancement of receptor-mediated calcium responses by phenytoin through the suppression of calcium excretion in human gingival fibroblasts. J. Periodontal Res.58, 274–282. 10.1111/jre.13089 (2023). [DOI] [PubMed] [Google Scholar]
- 32.Grötsch, H. et al. RWDD1 interacts with the ligand binding domain of the androgen receptor and acts as a coactivator of androgen-dependent transactivation. Mol. Cell. Endocrinol.358, 53–62. 10.1016/j.mce.2012.02.020 (2012). [DOI] [PubMed] [Google Scholar]
- 33.Tanabe, M., Fujiyama, S. & Horimoto, Y. Developmentally regulated GTP binding protein 2 (DRG2) and Nup107 are associated with epigenetic regulation via H2A. Z on promoter regions of specific genes in MCF7 breast cancer cells. J. Clin. Epigenet. 2, 21–29 (2016). [Google Scholar]
- 34.Chevalier, C. et al. TOM1L1 drives membrane delivery of MT1-MMP to promote ERBB2-induced breast cancer cell invasion. Nat. Commun.7, 10765. 10.1038/ncomms10765 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Seykora, J. T., Mei, L., Dotto, G. P. & Stein, P. L. Srcasm: A novel Src activating and signaling molecule. J. Biol. Chem.277, 2812–2822. 10.1074/jbc.M106813200 (2002). [DOI] [PubMed] [Google Scholar]
- 36.Franco, M. et al. The adaptor protein Tom1L1 is a negative regulator of Src mitogenic signaling induced by growth factors. Mol. Cell. Biol.26, 1932–1947. 10.1128/mcb.26.5.1932-1947.2006 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Cabral-Dias, R. et al. Fyn and TOM1L1 are recruited to clathrin-coated pits and regulate Akt signaling. J. Cell. Biol.22110.1083/jcb.201808181 (2022). [DOI] [PMC free article] [PubMed]
- 38.Liu, N. S., Loo, L. S., Loh, E., Seet, L. F. & Hong, W. Participation of Tom1L1 in EGF-stimulated endocytosis of EGF receptor. Embo J.28, 3485–3499. 10.1038/emboj.2009.282 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Hey, C. A. B. et al. BBS proteins affect ciliogenesis and are essential for hedgehog signaling, but not for formation of iPSC-derived RPE-65 expressing RPE-like cells. Int. J. Mol. Sci.2210.3390/ijms22031345 (2021). [DOI] [PMC free article] [PubMed]
- 40.Drugowick, R. M., Rós Gonçalves, D., Barrôso, L., Feres-Filho, A. S., Maia, L. C. & E. J. & Treatment of gingival overgrowth in a child with Bardet-Biedl syndrome. J. Periodontol. 78, 1159–1163. 10.1902/jop.2007.060378 (2007). [DOI] [PubMed] [Google Scholar]
- 41.Matsuda, K. et al. Transsynaptic modulation of kainate receptor functions by C1q-like proteins. Neuron90, 752–767. 10.1016/j.neuron.2016.04.001 (2016). [DOI] [PubMed] [Google Scholar]
- 42.Pirmohamed, M. et al. A randomized trial of genotype-guided dosing of warfarin. N Engl. J. Med.369, 2294–2303. 10.1056/NEJMoa1311386 (2013). [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
All results and data are available in the main manuscript or supplementary file. GWAS summary statistics, MAGMA statistics, and haplotype analysis data (RData format) generated in this study are available at 10.6084/m9.figshare.31835932.