Epidemiologic evaluation of Nhanes for environmental Factors and periodontal disease

Periodontitis is a chronic inflammation that destroys periodontal tissues caused by the accumulation of bacterial biofilms that can be affected by environmental factors. This report describes an association study to evaluate the relationship of environmental factors to the expression of periodontitis using the National Health and Nutrition Examination Study (NHANES) from 1999–2004. A wide range of environmental variables (156) were assessed in patients categorized for periodontitis (n = 8884). Multiple statistical approaches were used to explore this dataset and identify environmental variable patterns that enhanced or lowered the prevalence of periodontitis. Our findings indicate an array of environmental variables were different in periodontitis in smokers, former smokers, or non-smokers, with a subset of specific environmental variables identified in each population subset. Discriminating environmental factors included blood levels of lead, phthalates, selected nutrients, and PCBs. Importantly, these factors were found to be coupled with more classical risk factors (i.e. age, gender, race/ethnicity) to create a model that indicated an increased disease prevalence of 2–4 fold across the sample population. Targeted environmental factors are statistically associated with the prevalence of periodontitis. Existing evidence suggests that these may contribute to altered gene expression and biologic processes that enhance inflammatory tissue destruction.


Results
The final statistical analysis was completed on 8,884 individuals who were >18 years old and had 16 or more teeth. Males comprised 48.4% of the sample. The majority of subjects were non-smokers (55.9%), and those with smoking experience were evenly distributed between former smokers (22.5%) and current smokers (21.6%). The ethnic distribution of the group was non-Hispanic white (48.5%), non-Hispanic black (18%), Mexican American (25.1%), other Hispanic (4.7%), and other race including multi-racial (3.7%). Approximately 72% of the sample population was older than 30 years of age. (Tables 1 and 2). The weighted prevalence of periodontitis was 8.1% across the entire >18 years of age population. When the periodontitis group defined by NHANES measures was compared to the subset of subjects considered periodontally healthy, individuals with periodontal disease were more likely to be male, older than 30 years of age, Mexican American, non-Hispanic black or Hispanic and current smoker compared to Non-Hispanic white and non-smoker (p < 0.001) (Tables 1 and 3). Using survey-weighted logistic regression, there were 44 environmental factors (cotinine, 1 dioxin, 4 heavy metals (lead levels in serum and in urine), 8 hydrocarbons, 8 nutrients, 18 PCBs and 3 volatile compounds) that resulted in adjusted odds ratio with p-values < 0.01 for disease versus health in this NHANES cohort (Table 4). When data was further stratified due to smoking status 8 environmental factors (1 heavy metal (lead in serum), and 7 PCBs) in current smokers, 9 factors (acrylamide, 1 heavy metal, 1 nutrient, and 6 PCBs) in former smokers, and 4 factors (2 heavy metals, 1 nutrients, and 1 organophosphate) in non-smokers had FDR values of less than 0.05 (Table 5).
In regression analyses considering each environmental factor separately, blood lead levels were consistently identified as a factor in both the overall and stratified analyses ( (Table 5). Among the 17 polychlorinated biphenyls (PCBs) found to be associated with periodontitis in the overall sample, 6 (i.e. PCB105, PCB157, PCB172, PCB177, PCB178, and PCB206) were also found to have estimated adjusted odds ratios ranging from 1.41 to 5.29. In addition, across these environmental variables, the adjusted OR estimates were lower in non-smokers compared to current and former smokers. The smoking population also demonstrated additional factors, including 6 PCBs (PCB66, PCB146, PCB167, PCD170, PCB183, PCB187) with adjusted OR estimates from 1.63-2.23. It might be expected that the relationship between the array of PCBs and periodontitis risk would be highly correlated. Of the 8 coplanar PCBs (28,66,74,105,118,156,157,167), we noted significant correlations among these ranging from 50-100% of the other coplanar agents. Similarly 70-90% of the 16 non-coplanar PCBs were significantly correlated within this category, while only 10 showed correlations with the coplanar congeners. Thus, there was some measure of independence in these relationships that would enable future more granular description of specific PCBs periodontitis risk and/or severity. Dioxins (PNCDD, TCDD) showed adjusted OR estimates of 1.66 and 1.81, and blood nutrients retinyl stearate and retinyl palmitate exhibited adjusted OR estimates from 1.32-1.35. In contrast, blood nutrients such as Vitamin D and cis-ß-carotene were estimated to be protective for periodontitis. Higher levels of Vitamin D estimated to decrease the odds of periodontitis by 39% and 24% in former and non-smoker groups, respectively ([a]OR = 0.61, 95% CI: (0.50, 0.74) for former smokers; [a]OR = 0.76, 95% CI: (0.67, 0.87) for non-smokers), and cis-ß-carotene was estimated to decreasing the odds of periodontitis by 22% in non-smokers ([a]OR = 0.78, 95% CI: (0.67, 0.92)) ( Table 5).
We subsequently employed Random Forests (RF) and Classification and Regression Tree (CART) analyses to identify and visualize relationships of critical demographic and environmental factors. Based upon the variables, which had high importance in the RF for each smoking status, a CART was performed separately for each of the smoking, former smoking, and non-smoking subsets. CART analysis on the smoking population, presents elevated blood lead levels as an initial discriminator, with age >35 yrs. stratifying patients with an  Table 3. Weighted directionality of gender, race and age to prevalence of periodontal disease.
www.nature.com/scientificreports www.nature.com/scientificreports/ approximate 4-fold prevalence of periodontitis. (Fig. 1) CART analysis on former smoker population, visualizes the factors classifying the disease risk in former smokers. In this case, race/ethnicity remained a critical factor. Those who reported race/ethnicity other than non-Hispanic white demonstrated increased disease prevalence; elevated blood lead levels and age >53, had an increased periodontitis prevalence of 37%. Within the subset of non-Hispanic white subjects and other race including multi-racial, a prevalence rate of 12% was observed in those All environmental variables were log-transformed (natural) and standardized, and the estimates should be interpreted on the same scale. Due to missingness in the data, the sample sizes were not the same for most of these analyses. www.nature.com/scientificreports www.nature.com/scientificreports/ with elevated blood lead levels. (Fig. 2) For non-smokers, which comprised 56% of the total population, multiple variables were identified to have relationships with periodontal disease status. Race-ethnicity and age were important distinguishing factors. The prevalence was low across those reporting a non-Hispanic white race, but even in this group subjects >44 years demonstrated an increased prevalence. The prevalence was further modified by elevated urine antimony that increased the observed prevalence of periodontitis to 33% from as low as 8% in the low urine antimony and high cis-ß-carotene group. For those with low levels of cis-ß-carotene, higher blood lead levels showed a higher prevalence of periodontitis of 18% compared to 11% for the lower blood lead group. (Fig. 3). This data analysis exercise represents an approach consistent with the current trend in precision health, in that identification of risk and use of model for early prediction of disease initiation/progression will be critical for future improvement of oral health and it impacts on systemic health in the population.

Discussion
The current paradigm of periodontitis is that it represents a dysregulation of the host response to a dysbiotic microbiome that occurs in a large portion of the global population. Substantial work is being conducted via the Human Microbiome Project 30 to discern not only the characteristics of the alterations in the disease microbiome, but also interrogating complex metagenomic datasets to assess functional changes in the microbial ecology associated with health and disease 31 . Additionally, a complementary research direction is attempting to document the role of individual genetic variation across the population that contributes to disease expression and severity 19 .  www.nature.com/scientificreports www.nature.com/scientificreports/ These studies have employed SNP analysis of specific targeted genes 19,22,32 , Genome-Wide Association Studies (GWAS) 24,33 and epigenetic analyses 26,34,35 to help elucidate the complex of factors that interact to create a disease susceptible host. This report describes an additional consideration in disease expression focused on the larger environmental variation to which individual members and subgroups of the U.S. population are exposed (i.e. exposome) as a potential direct contributor to the microbial dysbiosis 36 and/or a modifier of host responses through altered molecular pathways or modulation of genetic control of the disease 27,37 . The findings identified more classical factors (i.e. age, gender, race/ethnicity) in the disease model, but for the first time integrated a subset of environmental factors, both toxins and nutrients, that appear to substantially modify the prevalence of periodontitis in the population. The identification of the association of environmental toxins including lead, hydrocarbons, polychlorinated biphenyls, and nutrients such as retinyl stearate in models described an increase in the prevalence of disease. Thus, the findings support the potential for a role of these factors in modifying the challenge (i.e. bacterial biofilms) and/or host responses with a loss of homeostasis and tissue destruction.
The results demonstrated altered levels of various heavy metals, including lead, cadmium and antimony in periodontitis patients. A range of literature has shown the toxic properties of systemic elevations in heavy metals from environmental sources, including lead 38,39 . In particular, this toxin has been linked to substantial neurotoxicity and negative developmental processes in children 40,41 . This study identified, using CART analysis, an estimated threshold of >2.0 µg/dL that discriminated periodontitis from health in the adult population. While this level does not indicate the actual blood lead level across the periodontitis group, since CART attempts to fit the discrimination profile in the context of multiple variables, it was clear that in all subsets of smokers, former smokers and non-smokers that lead levels were elevated in periodontitis patients. An earlier evaluation of data from NHANES III (1988-94) demonstrated a significantly increased OR for periodontitis in both men and women with increased blood lead levels 29 . Reports examining various iterations of the Korean NHANES (KHANES) study demonstrated elevated lead, cadmium, or mercury in subjects with periodontitis, particularly related to smoking and in some instances gender associated similar to our data from NHANES [42][43][44][45] . An additional study reported that chronic occupational exposure of workers to lead resulted in significant changes in oral health and correlated with increasing blood lead levels 46 . Terrizzi et al., have reported that elevated lead levels under hypoxia induces alveolar bone resorption and periodontitis 47 . More recently they demonstrated that iNOS and PGE 2 levels are altered by lead and hypoxia as inflammatory responses that would contribute to damage of the periodontium 48 . Furthermore, the lead levels investigated in these previous studies generally targeted levels that have been shown in blood to have substantial neurotoxicity (≥10 µg/dL), although levels of >5 µg/dL are considered deleterious 49 . Further studies will be required to identify the relationship of blood lead levels to severity of the disease, age of onset, and response to therapy, as well as biologic studies determining the impact of these altered levels of lead on host responses, and even the microbial ecology related to the disease process.
Polychlorinated biphenyls (PCBs) were once widely deployed as dielectric and coolant fluids in electrical apparatus, carbonless copy paper, and in heat transfer fluids since they do not easily degrade. PCBs' environmental toxicity and classification as a persistent organic pollutant resulted their production and use of them being banned by the United States Congress in 1979. Coplanar PCBs, e.g. dioxin-like PCBs, since their structure is similar to dioxins, allows them to act as agonists of the aryl hydrocarbon receptor (AhR). They are considered as contributors to overall dioxin toxicity within the environment. The toxicity of PCBs varies considerably among various chemical structural iterations with the coplanar PCBs representing 12/209 possible PCB molecules (i.e. PCB 77,81,114,118,123,126,156,157,167,169,189) generally considered among the most toxic congeners with the majority of differences occurring in smokers and former smokers. Interestingly, the overall group of toxins included PCB105, PCB146, PCB172, PCB177, PCB178, PCB183, and PCB206, which are all members of the non-coplanar group of PCBs appeared to show the most frequent association with periodontitis. Elevated levels of non-coplanar PCBs, including PCB153, PCB170, PCB180 and PCB187 were detected in the blood of Canadian First Nations communities and were associated with elevated levels of an array of immune activation markers including IFNγ, IL-1ß, IL-8, IL-17A and TNFα 50 . Much of the molecular aspects of PCBs and host responses have focused on the coplanar, dioxin like congeners. The current study identified an array of PCBs that were increased across the periodontitis population. While some representative true dioxin molecules had increased OR for periodontitis these only were noted in smokers. No other reports are available identifying PCB levels and periodontitis in humans or animal models, nor focusing on biologic alterations in cells related to periodontal health and disease, thus, this family of exposome factors could present an important area for further investigation of disease variation and personalized documentation of disease features within the population. Finally, a single recent report demonstrated that PCB126 appeared to exacerbate periodontal disease in a susceptible species of mink 51 .
An interesting finding was the dichotomy between the effects of selected specific nutrients on the expression of periodontitis. Both carotenoids and Vitamin D levels had adjusted Odds Ratios which suggested that they were protecting against periodontitis. Carotenoids are organic pigments found in plants and some photosynthetic microorganisms and carotenoids from human diets are stored in the fatty tissues. There are over 600 known carotenoids classified as xanthophylls (β-cryptoxanthin, lutein, and zeaxanthin; non-vitamin A carotenoids) and carotenes (α-carotene, β-carotene, and lycopene). Generally, the health benefits of carotenoids are thought to be due to their role as antioxidants with dietary carotenoids proposed to interact with endogenous antioxidant enzymes to positively affect immunity 52 . Thus, various reports have shown that elevations in acute phase proteins are accompanied by low vitamin A levels 53 and that carotenoids significantly reduced proinflammatory cytokines, CRP, and other markers of inflammation in multiple tissues 54 . A study of inflammation in 60-70 year old men demonstrated an inverse relationship between elevated carotenoids and serum CRP levels 55 . Moreover, low blood levels of various carotenoids have been associated with an increased prevalence of periodontitis in 60-70 year old men 56 and carotenoid levels were related to positive outcomes of scaling and root planing with the relationship limited to non-smokers 57 . Thus, our data from a large population cohort is consistent with these findings and the support that increased availability of carotenoids appears to provide some level of protection from periodontitis. www.nature.com/scientificreports www.nature.com/scientificreports/ Vitamin D has received an increasingly detailed examination regarding its potential influence in periodontitis. Various reports have linked decreased serum or saliva vitamin D levels with tooth loss and periodontitis [58][59][60][61][62] including in smokers 63 , albeit not all studies are supportive since this was not observed in postmenopausal women 64 . Additionally, a gene polymorphism for vitamin D binding protein increases the risk for periodontitis 65 that appears exacerbated in smokers 25 . Our analysis of this nutrient was based upon examination of NHANES data, and demonstrated an estimated protective feature of this serum nutrient in periodontitis, specifically in non-smokers and former smokers. This type of finding is consistent with additional associational data from NHANES related to risk of cardiometabolic disease 66 , asthma 67 , and coronary heart disease and all-cause mortality 68 . Interestingly, a single recent report describes the interaction of an environmental exposure to phthalates may decrease blood levels of vitamin D 69 , an observation consistent with our results identifying "competing" impact of environmental toxins and nutrients on periodontitis as the clinical outcome.
In contrast, elevated levels of retinyl stearate and retinyl palmitate were each estimated to enhance the risk for periodontitis particularly in smokers. The retinoids comprise a class of compounds related to Vitamin A. These compounds have been used to regulate epithelial cell growth, as well as playing a role in vision, regulation of cell proliferation and differentiation, growth of bone tissue, immune functions, and even activation of tumor suppressor genes 70 . Our data demonstrated an increased OR for blood levels of retinyl stearate and retinyl palmitate in periodontitis. In serum, 56% of retinyl esters are retinyl stearate, 33% retinyl palmitate, and 5% retinyl oleate. Retinyl esters in humans are derived from animal sources and are hydrolyzed in the intestinal lumen to form retinol and fatty acids, such as retinyl palmitate or stearate. Enzymes in the intestinal lumen that hydrolyze dietary retinyl esters include cholesterol esterase from the pancreas and a retinyl ester hydrolase intrinsic to cells of the small intestine, which primarily acts on long-chain fatty acids, such as palmitate or stearate 70 . A single study has been reported regarding these compounds and periodontitis. Wang et al. 71 demonstrated that all-trans retinoic acid administration modulated the Th17/Treg balance and can modulate the expression of periodontitis in a murine model of P. gingivalis infection and provided protection against periodontitis with increased Treg activation and decreased Th17 functions. However, our data specifically related to endogenous levels of a specific retinoid, retinyl stearate, suggested an increased risk for periodontitis. This may relate to the more individualized functions of the various members of this family of dietary nutrients, and may highlight some unique features of the diet or intrinsic variation in the hydrolytic enzymes across the population that may link retinyl stearate and disease. Clearly additional studies will need to be conducted examining in more detail the clinical relationship with this compound, as well as its potential role in affecting an array of inflammatory responses that would be related to periodontitis.
This report describes an associational study of a large U.S. population sampled cross-sectionally during a 5 year interval via the NHANES project and demonstrated statistical associations of a subset of environmental challenges to the expression of periodontitis. A clear limitation in the approach is that the findings do not deliver any cause and effect relationship, and are affected by the lack of detailed clinical evaluation of periodontitis that is generally accepted within the field. However, the model developed identified an interaction of these exposome factors and more classical risk factors of age, gender, and race/ethnicity, thus providing some confidence that the findings are providing additional clues into population variation in disease expression. The model will also enable future testing with additional NHANES datasets, as well as the environmental features and categorization of disease. The individual exposome components that were identified can be further evaluated in more detailed clinical studies, and by implementing basic biologic studies of the host cells and microbiome components associated with health and disease to delineate modes of actions of these environmental factors that could contribute to the disease processes.

Materials and Methods
population data. The NHANES is a complex, multistage probability sample of non-institutionalized U.S. civilians and subsequently organized into 6 unique datasets derived from 2-year cycle population sampling (Centers for Disease Control and Prevention; National Center for Health Statistics). Each 2-year survey cycle examines a representative U.S sample of approximately 10,000 persons and collects health-related data. Full descriptions of the sample design for these NHANES datasets are publically available (https://www.cdc.gov/nchs/ nhanes/). These surveys, using the same methods, assessed the health status of a nationally representative sample of the civilian non-institutionalized US population, selected through a stratified multistage probability sampling design. In this study, periodontal examination data from three NHANES cohorts, 1999NHANES cohorts, -2000NHANES cohorts, , 2001NHANES cohorts, -2002NHANES cohorts, , 2003NHANES cohorts, -2004, were extracted and combined to comprise the study population. These data have been merged and processed and can be found at https://github.com/joshuawlambert/ PinarEtal2018/raw/master/data.zip. A unique identifier, SEQN https://wwwn.cdc.gov/Nchs/Nhanes/1999-2000/ DEMO.htm#SEQN for the NHANES participant from our years of study (1999)(2000)(2001)(2002)(2003)(2004) is included in these data.
Demographics. The demographic variables considered in this study included age, gender, race, socio-economic status, smoking status, and number of teeth. Racial-ethnic groups were summarized into five categories: Mexican American, Other Hispanic, Non-Hispanic White, Non-Hispanic Black, and Other Race. Socio-economic status, estimated using the poverty income ratio, was computed as the ratio of family/individual income to the appropriate federal poverty threshold. Smoking status, current smoker, former smoker, non-smoker, was derived from the two self-reported questions. Participants reported having historically smoked more than 100 cigarettes, but currently not smoking were defined as former smokers. Non-smokers were defined as reporting never smoking.
www.nature.com/scientificreports www.nature.com/scientificreports/ Clinical parameters. Periodontitis was defined as a minimum of 2 or more sites with clinical attachment loss (CAL) ≥3 mm and a periodontal pocket ≥4 mm as described by Eke et al. 72 . NHANES (1999-2004 used the partial-mouth periodontal examination (PMPE) protocol to sample teeth and sites. The PMPE protocols randomly selected two quadrants of the mouth and specified 2 to 3 sites per tooth for measurement of pocket depth, attachment loss, and bleed on probing. In 1999-2000, two sites per tooth (mid-facial and mesio-facial) were measured, while three sites per tooth (mid-facial, mesio-facial and distal) were measured in 2001-2002 and 2003-2004. Dentists trained in the survey examination protocol conducted the periodontal examinations collecting probing depth and attachment loss and bleeding on probing measurements 73-75 . environmental variables. The environmental factors were categorized into 15 classes based on NHANES categorization. Environmental variables measured in at least one of the three data cohorts (i.e. 1999-2004) were included in the study. A total of 156 environmental factors were measured in the NHANES data using blood and urine samples. These included chemical toxicants, pollutants, allergens, bacterial/viral organisms and nutrients. Environmental factors with laboratory measurements that had greater than 10% of the observations below a detection limit threshold defined by NHANES were omitted from analysis. The laboratory measurements using mass spectrometry and absorption spectroscopy demonstrated that the majority of the variables were detected in small ranges and were skewed and thus all 156 environmental variables were log-transformed (natural), standardized, and referred to as "processed". statistical approaches. Survey-weighted logistic regressions were performed for each of the processed environmental factors, adjusting for age, gender, ethnicity, socio-economic status, smoking status and number of teeth. The R package "survey" was used in R (Version 3.1.2) for the survey-weighted logistic regression. Weights were constructed in SAS (Version 9.4) using a 6 year weighting design from the NHANES variable WTMEC2YR73 (http://www.cdc.gov/nchs/tutorials/Nhanes/SurveyDesign/Weighting/Task2.htm). Survey weighted logistic regression seeks to minimize bias by weighting the samples to reflect the intended population. By doing this, better estimates of the standard error are obtained. The Odds Ratio estimates, Standard Errors, 95% CI, and FDRs were provided to demonstrate the association between the individual factors and periodontitis. These regressions were repeated by smoking status to examine potential associations within smoking categories.
Random forests (RF) and classification and regression trees (CART) were employed to investigate associations and potential interactions between environmental factors, demographic and socioeconomic characteristics, and periodontitis disease status for each smoking status 76 . Specifically, for each smoking status RF was used to identify important factors (main effects and interactions) and then a single CART was used to visually investigate these relationships. Variables which were in the top ten for variable importance, were subsequently used to build a CART model with minimum node size of 100 and Bonferroni test for the stopping criteria. These methods were selected because the data involved many potentially correlated environmental factors and had the ability to allow nonlinearities and interactions without modeling them explicitly 77 . These analyses were performed using the "party"website (Version 1.0-25) package in R (Version 3.1.2). Repository for the data, R code, and SAS code can be accessed at https://github.com/joshuawlambert/PinarEtal2018.