Introduction

The most common and most studied oral conditions to date include dental caries and periodontal disease, both bacteria-mediated infections1, and major causes of tooth loss2,3. Genetics has been suggested to play a role in the susceptibility to tooth loss due to both dental caries and periodontal disease4. Our efforts in the past years focused on the understanding of whether individuals’ genetics can influence both oral conditions such as tooth loss, and other systemic general health conditions, especially cancers5. Our long-term goal is to identify orofacial traits that can be markers for cancer risk and allow dental professionals to assist with cancer prevention and early cancer diagnosis by referring patients to more precise screenings.

Certain genetic pathways that regulate protein synthesis were found to influence multiple oral and systemic conditions, including metabolic disorders, type 2 diabetes and cancers6. Most interestingly, the endoplasmic reticulum stress (ER stress) pathway specifically, has been suggested to cause or aggravate cancers7, and we found it to be associated with oral health outcomes. We found that genetic variation in ERN1 (rs196929), an ER stress variant, was associated with dental caries and periodontal disease8. A later study from our group showed an excess of the less common homozygote of ERN1 rs196929 among relatives of individuals born with cleft lip and palate when they had positive family history of cancer. This pattern was similar for families that reported one type of cancer or multiple ones, or when cancer affecting females (breast or reproductive tract) or the structures of the gastro-intestinal tract were considered9. Moreover, a question that this previous study brought to light was whether this association could be replicated in individuals from a different geographic location and who were not born with cleft lip and palate.

In summary, our studies in the past years focused on the understanding of whether individuals’ genetics can influence both oral conditions such as tooth loss, dental caries and periodontitis and other systemic general health conditions, especially cancers5. We identified a knowledge gap regarding whether variation in the ER stress pathway genes play a role in cancer susceptibility the same way we found these genes to influence craniofacial conditions. With the hypothesis that the family history of cancer drove our previous results, this current study aims to test if ERN1 rs196929 variant is overrepresented in individuals with a diagnosis of cancer using a cohort of patients going for regular dental treatment.

Results

Our results show a statistically significant association between the rs196929 in ERN1, and all types of cancer combined in the genotypic (2 degrees of freedom) model (p = 0.0058) (Table 1). The heterozygous genotype (TC) was more frequent in individuals diagnosed with cancer as compared with matched controls (49.7% vs. 39.6%) whereas the presence of at least one copy of the allele T was more frequent in the individuals diagnosed with cancer compared to the matched controls (60% vs. 52.3%), with the results suggesting a dominant genetic model (OR = 1.37, 95% C.I. 1.06–1.79, p = 0.014) (Table 1).

Table 1 Results from the genotypic analysis of patients diagnosed with all types of cancer and the rs196929 variant (bold indicates statistically significant p-values under the threshold 0.05).
Full size table

Additionally, our results stratified by cancer type showed a statistically significant association between the rs196929 in ERN1 and skin and breast cancers in both the logistic regression and the conditional logistic regression (Tables 2 and 3). In the logistic regression, when we stratified the analysis by each type of cancer and adjusted by age, sex, ethnicity and smoking habits we found associations between the rs196929 in ERN1 and skin, breast and male’s reproductive system cancers (p < 0.05) (Table 2). When the false discovery rate was applied to adjust for multiple comparisons, only skin and breast cancer remained significantly associated with ERN1 (Table 2). The conditional logistic regression adjusted for smoking habits confirmed the association between the rs196929 in ERN1 and skin (OR = 2.07, 95% C.I. 1.27–3.37, p = 0.003) and breast cancer (OR = 1.83, 95% C.I. 1.13–2.99, p = 0.013) (Table 3).

Table 2 Logistic regression analysis of associations between the rs196929 and different types of cancer in patients with tooth loss (p-values below 0.05).
Full size table
Table 3 Conditional logistic regression analysis of associations between the rs196929 and different types of cancer in patients with tooth loss (p-values below 0.05).
Full size table

From 1,421 participants included in the study, 353 were smokers and within those 71 reported a diagnosis of at least one type of cancer. From the total patients with skin cancer, 12 reported having basal cell carcinoma (with 50% carrying one copy of the variant allele and 50% carrying two copies of the variant allele), 11 reported having melanoma (with 82% carrying two copies of the variant allele and 18% carrying one copy), 5 reported having squamous cell carcinoma (with 60% carrying two copies of the variant allele and 40% carrying one copy). The remaining patients did not know or did not report the type of skin cancer in the registry. Ninety four percent of the patients reporting a diagnosis of any type of skin cancer in the study carry at least one copy of the variant allele and all the patients who reported the type of skin cancer had at least one copy of the variant allele.

We included 104 subjects in the sensitivity analysis to determine the chronology of cancer diagnosis and tooth loss. We excluded 33 subjects because they did not have enough dental treatment history available in the records since they started treatment more recently, 113 because no information on cancer diagnosis timing was available and 62 subjects because they reported a cancer diagnosis prior to 2006 and we were not able to access their oral health before that period of time. From the total of 104 individuals included, 66 had tooth loss prior to cancer diagnosis, 13 did not, 53 had tooth loss post cancer diagnosis and 30 did not have a new occurrence of tooth loss post cancer diagnosis. In approximately 24% of these subjects missing data was present and we did not consider those for analysis. Our chi-square analysis showed a significant association in this subgroup indicating that one is almost 3 times more likely to lose teeth prior to a diagnosis of cancer (OR = 2.87, 95% C.I. 1.36–6.05, p = 0.004).

Discussion

Our results show a significant association between both skin and breast cancers and the rs196929 ERN1 marker in the logistic regression and the conditional logistic regression. The positive results for this genetic association and breast cancer are consistent with our previous study9, and the association with skin cancer is novel. Since statistical correction for multiple comparisons might be too stringent and lead to missing potential biologically relevant results10, we also report nominal associations (i.e., p values below 0.05). We found a nominal association between this same ERN1 marker and male’s reproductive system cancer, but this association should be interpreted with caution since there is a possibility that it is due to chance.

To date, very few studies have focused on the role of the ER stress pathway genes and skin cancer and, to the best of our knowledge, this is the first time that the rs196929 ERN1 marker was associated with cancers. One of these few studies, however, demonstrated that, in melanoma cells, intrinsic activation of the ER stress response is caused by increased outputs of protein synthesis driven by oncogenic activation of mitogen-activated protein kinase/extracellular signal-regulated kinase (MEK/ERK) and promotes proliferation and protects against apoptosis induced by acute ER stress11. This might explain the association between genetic variation in the ER stress pathway and skin cancer found in this present study, given that approximately 20% of patients who reported having skin cancer had melanoma and 82% of those carry two copies of the variant allele. The rs196929 is located in an intronic region of ERN1 and possible hypotheses on the contribution of this variant to the observed associations include that this might be a causative variant or a variant in linkage disequilibrium with another etiologically causative variant, although the exact mechanism of causation is yet to be determined.

Tooth loss has been associated with higher risk of cancers12,13,14,15,16, with factors such as behavior17, the oral microbiome18 and chronic inflammation19 being suggested as culprits for these associations. We have previously shown associations between genetic variation in ERN1 and oral phenotypes, including associations with the rs196929 marker and both periapical lesions and periodontal disease8. In this present study we selected from our dental registry only patients who had missing teeth, however, this inclusion criterion was very broad with 85% of the patients having lost at least one tooth in the registry. Only 11% of patients were diagnosed with cancer and never had tooth loss. The inclusion of subjects that seek oral care at the dental school could be perceived as a limitation because these subjects are not representative of the general population. However, this pre-selection of patients with tooth loss might eliminate the oral health confounders of associations between cancer and genetic variation in ERN1 such as our previously detected correlations with periodontal disease and periapical lesions. Another limitation was that we did not have data on patient’s diet, exercise habits, and alcohol consumption, which are variables that could impact the occurrence of cancers. Variables that we did not have data available included health behaviors, and sun exposure, which are especially important for the association with skin cancer. Furthermore, most of the subjects studied here were White, a group more susceptible to skin cancer20. In future studies, it would be interesting to obtain behavioral data on sun exposure and sunscreen use to correlate with oral care behavioral data and help explain the same genetic association we found for both tooth loss and skin cancer. Additionally, a future direction for these studies in the molecular level would be to determine the precise pathophysiology of these associated skin cancers for a better understanding of their etiology. Because in our study the patients self-reported their cancer diagnosis, and approximately 60% did not report the exact type of skin cancer diagnosis we were unable to precisely specify these associations. Thus, the use of self-reported data on cancer diagnosis is a limitation of this study. Although there is high accuracy in self-reporting diagnosis of cancer, especially for breast cancer21, we cannot discard potential issues related to recall bias of all cancer sites. Finally, because rarer cancer types have low numbers of cases in the registry, potential associations between those cancers might have not been identified due to lack of statistical power. Further, future studies that have larger sample sizes available should consider performing stratified analyses according to the reasons for tooth loss, especially for tooth loss due to caries and tooth loss due to periodontal disease which are the main contributors for tooth loss.

Strengths of our study included that we adjusted the analyses for other important variables such as smoking habits, age, sex and ethnicity. However, our results need to be confirmed with larger cohorts and different populations since the population that seeks care in our dental clinics, for the most part, lacks diversity because they are mostly White. Further, because we matched cases and controls for sex, age and ethnicity at the individual level, we performed a conditional logistic regression to confirm our results. The conditional logistic regression is more robust when individuals are matched because not only the matching process makes cases and controls similar for the variables of interest but also for the outcome status and this model corrects for this distortion22. The association between both skin and breast cancer and ERN1 genetic variation remained significant in both analyses and the remaining results were also not remarkably different. We also report the number of subjects who had tooth loss pre and post cancer diagnosis considering that this oral trait could be further investigated as a risk marker for cancer in prospective and larger studies. However, we understand that cancer can be a silent disease and the diagnosis can be established several years after cancer development. This is also the case with oral conditions that lead to tooth loss such as periodontal disease and dental caries that can take many years to develop. Nevertheless, determining this chronological sequence between diseases that lead to tooth loss and cancer development is challenging and our results should be taken cautiously.

In summary, we demonstrated that ERN1 genetic variation is associated with skin and breast cancer and potentially associated with male’s reproductive system cancers in a population that had tooth loss. We hope that our study helps inform future genetic and epidemiologic studies on cancers and the rs196929 ERN1 marker, as well as helps drive attention to the potential for oral health indicators of general health outcomes such as cancers. The early identification of genetic markers and/or oral conditions that indicate increased cancer risk could positively impact cancer outcomes and survival rates with timely implementation of preventive and diagnostic measures.

Methods

Subjects

Between 2006 and 2020 a total of 6,100 individuals who sought treatment at the dental clinics were recruited to participate in the Dental Registry and DNA Repository (DRDR) project at the University of Pittsburgh, School of Dental Medicine. From those, 1,671 individuals were selected in our previous study5 and included in this current study: 350 because they reported a diagnosis of cancer and 1,321 that match those individuals that had cancer, based on age, sex, and ethnicity. All patient data is collected prospectively, and the medical history assessment includes a question on whether the patients were ever diagnosed with cancer as well as with other systemic conditions. These questions are asked before the dental treatments start and the answers are included as part of the medical history.

We first analyzed the total sample for genetic association with the candidate gene and subsequently we performed analyses including only patients that had one or more teeth missing due to dental caries, periodontal disease, or periapical lesions. All patient evaluations and treatments are conducted at the clinics by dental students instructed by experienced professors. After excluding 250 individuals who did not have any teeth missing, a total of 1,421 subjects were included in the stratified analyses (Fig. 1). Three hundred and twelve self-reported a diagnosis of various cancer types (Table 4) and 1,109 reported never receiving a diagnosis of cancer reaching an approximate ratio of 1 case to 4 individuals for comparison. Table 5 shows the study sample demographics. Written informed consent was obtained from all participants and the project had University of Pittsburgh Institutional Review Board approval. All methods were performed in accordance with regulations, and we followed the strengthening the reporting of genetic association studies (STREGA) guidelines for this report.

Figure 1
figure 1

Overall study design.

Full size image
Table 4 Types of cancers in the study sample.
Full size table
Table 5 Study sample demographics.
Full size table

Genetic polymorphism and DNA extraction

We have selected the SNP rs196929 in ERN1 based on our previous study in which we tested 27 markers in eight genes of two cell regulatory pathways and five oral phenotypes8. Results showed that the SNP rs196929 in ERN1 associated with dental caries, periodontitis, and periapical lesions. All three conditions can lead to an extreme outcome of tooth loss.

Genomic DNA was extracted from salivary samples of the 1,421 individuals using established protocols23. Reactions were carried out using TaqMan chemistry in volumes of 3.0 μl in an ABI PRISM Sequence Detection System 7900, software version 1.7 (Applied Biosystems, Foster City, CA, USA). Genotypes were generated blindly to clinical diagnosis status. As a measure of quality control, we used positive and negative controls as well as replicates. The variant was consistent with Hardy–Weinberg equilibrium (chi-squared p-value = 0.99).

Genetic analyses

Association analyses were performed comparing genotypes and allele frequencies between the overall cancer group and the unaffected group as implemented in PLINK24. Models available included the allelic model that compares the frequencies of each allele in each group (D x d); assuming that D is the minor allele and d is the major allele. The genotypic model is an additive two degree of freedom model and compared the frequencies of each genotype in the groups (DD x Dd x dd), it considers that the combined effects of each allele equal the sum of their individual effects to the phenotype, the dominant model compared the two copies of the common alleles’ frequency versus the other combinations (dd x DD + Dd), and the recessive model compared the two copies of the rare alleles’ frequency versus the other combinations (DD x Dd + dd). p-values below 0.05 were considered statistically significant.

In order to identify the specific cancer types driving the associations, we ran both logistic regressions accounting for variables such as age, ethnicity, sex and smoking habits and conditional logistic regressions adjusting for smoking habits using the PheWAS package installed in R studio25. We defined smoking behavior as individuals who currently smoke every day and who have smoked at least 100 cigarettes in his or her lifetime. The diagnostic codes for types of cancer can be found at www.phewascatalog.org—the codes can be identified by either typing the correspondent ICD9 code or the phenotype of interest. This final file was then uploaded into R studio and used in the analyses. All analyses were performed using R studio version 1.4 and correction for multiple comparisons was performed using the false discovery rate to control the risk of false positive findings.

Sensitivity analysis for timing of cancer diagnosis and tooth loss

To determine the chronology of cancer diagnosis and tooth loss, the deidentified records were assessed for the occurrence of tooth loss in the periods pre and post cancer diagnosis. We searched the Dental Registry and DNA Repository records for subjects who had a history of dental treatments done in the past years to establish this timing and perform the sensitivity analysis. We created a two-by-two table and performed a chi-square test with an alpha set to 0.05.

Power calculations

Given N = 30 cases, N = 1250 controls, an assumed probability of exposure among control patients of 52%, and a correlation of exposure between matched individuals of 0.2, we are able to detect odds ratios as small as 3.11 with 80% power at the 5% significance level. With N = 60 cases, we can detect odds ratios as small as 2.20 with 80% power at the 5% significance level. Thus, we are adequately powered to detect odds ratios of approximately 2.20 to 3.11, which are considered medium-to-large effects per Cohen's criteria (note that OR = 1.42, 2.44, and 4.25 roughly correspond to Cohen's d = 0.20, 0.50, and 0.80, respectively).