Modification Effect of PARP4 and ERCC1 Gene Polymorphisms on the Relationship between Particulate Matter Exposure and Fasting Glucose Level

Particulate matter (PM) has been linked to adverse health outcomes, including insulin resistance (IR). To evaluate the relationships between exposures to PM10, PM2.5–10, and PM2.5; the serum level of fasting glucose, a key IR indicator; and effects of polymorphisms of two repair genes (PARP4 and ERCC1) on these relations, PMs exposure data and blood samples for glucose measurement and genotyping were collected from 527 Korean elders. Daily average levels of PMs during 8 days, from 7 days before examination to the health examination day (from lag day 7 to lag day 0), were used for association analyses, and mean concentrations of PM10, PM2.5–10, and PM2.5 during the study period were 43.4 µg/m3, 19.9 µg/m3, and 23.6 µg/m3, respectively. All three PMs on lag day 4 (mean, 44.5 µg/m3 for PM10, 19.9 µg/m3 for PM2.5–10, and 24.3 µg/m3 for PM2.5) were most strongly associated with an increase in glucose level (percent change by inter-quartile range-change of PM: (β) = 1.4 and p = 0.0023 for PM10; β = 3.0 and p = 0.0010 for PM2.5–10; and β = 2.0 and p = 0.0134 for PM2.5). In particular, elders with PARP4 G-C-G or ERCC1 T-C haplotype were susceptible to PMs exposure in relation to glucose levels (PARP4 G-C-G: β = 2.6 and p = 0.0006 for PM10, β = 3.5 and p = 0.0009 for PM2.5–10, and β = 1.6 and p = 0.0020 for PM2.5; ERCC1 T-C: β = 2.2 and p = 0.0016 for PM10, β = 3.5 and p = 0.0003 for PM2.5–10, and β = 1.2 and p = 0.0158 for PM2.5). Our results indicated that genetic polymorphisms of PARP4 and ERCC1 could modify the relationship between PMs exposure and fasting glucose level in the elderly.


Introduction
Insulin resistance (IR), a prediabetic state of type 2 diabetes, is an important health issue with a rapidly increasing incidence worldwide [1,2]. Several studies reported positive associations between particulate matters (PMs) and serum levels of fasting glucose, an important IR indicator [3,4]. The adverse effect of PM 10 with a diameter of less than 10 µm on fasting glucose level was proven in the general population, regardless of short-term or long-term exposure [3,4]. Although short-term or long-term exposure to PM 2.5 with a diameter of less than 2.5 µm was also reported to induce IR in C57BL/6 mice and incident metabolic syndrome in KORA cohort [5,6], evidence for the effects of PM 2.5 and particularly PM 2.5-10 , known as coarse PM, on fasting glucose levels was insufficient.
Oxidative stress is considered to be a major biological mechanism underlying IR [7]. Although a variety of potential mechanisms of PMs inducing IR have been suggested, oxidative stress is still at the center of research attention [8][9][10][11]. Exposure to PM 2.5 in mice can cause vascular IR by inducing pulmonary oxidative stress [8,12]. Furthermore, the adverse effect of PM 2.5 on vascular IR can be reduced by removing superoxide from the lungs of mice [11,12]. The water-insoluble fraction of PM 10 can also induce oxidative stress in human lung epithelial A549 cells [13]. Because PMs contain pro-oxidant molecules, such as chromium, iron, nickel, and zinc, that can induce reactive oxygen species [10], oxidative stress is plausible as a major mechanism of PMs on IR. Because oxidative stress could induce oxidative DNA damages, cells need to up-regulate DNA repair genes to protect against oxidative DNA damage upon PMs exposure [14]. In fact, an epidemiologic study reported that PMs exposure could even induce oxidative stress-associated DNA damage in healthy young adults exposed to low concentrations of ambient PM 2.5-10 and PM 2.5 [15].
Poly(ADP-ribose) polymerase family member 4 (PARP4) and excision repair crosscomplementation 1 (ERCC1), respectively, repair DNA damage using base excision and nucleotide excision repair pathways to remove oxidized DNA bases or nucleotides [16,17]. These DNA damage repairs are considered to be very important processes because they can protect the human body against oxidative stress. It has been reported that PARP is activated by DNA strand breakage through the excessive accumulation of reactive oxygen species in relation to hyperglycemia [18]. Although PARP1 and PARP2, included in the same family as PARP4, were reported to affect glucose metabolism and insulin sensitivity [19], there is no evidence for the effect of PARP4 on blood glucose level. On the other hand, ERCC1 gene was reported to have an impact on glucose intolerance in a progeroid mouse model with ERCC1 deficiency, resulting in fat loss and IR by triggering an autoinflammatory response [20]. Previous evidence showed that genetic variations of PARP4 and ERCC1 could potentially affect IR differently by changing the capacity of corresponding enzymes encoded by them.
As population aging increases, we need to pay attention to the elderly, who are more vulnerable to chronic diseases than others [21]. With increasing age, the body's function as well as immune system become more sensitive and vulnerable [21]. The EpiAir study, an epidemiologic surveillance on air pollution and Italian health, indicated that elderly subjects were more vulnerable to exposure to particulate matters than to other pollutants [21]. Therefore, the objective of the present study was to evaluate relations between PMs (PM 10 , PM 2.5-10 , and PM 2.5 ) exposure and the serum level of fasting glucose among the elderly population considering modification by genetic polymorphisms of PARP4 and ERCC1.

Study Population and Sampling
The Korean Elderly Environmental Panel I study began from 2008. There were five repeated health examinations from start time to 2010 (twice in 2008, once in 2009, and twice in 2010) for 560 participants aged 60 or over recruited at a community welfare center for elders in Seoul, Korea. After excluding participants lacking blood samples and PM concentrations data, the final analysis included 527 participants. We asked participants to fast from midnight on the day before the examination. We collected their blood samples on the day of visit at around 10 A.M. Whole blood was centrifuged at 30 to 60 min after collection, and the serum and cellular layer were separated and stored in screw-cap tubes. All serum samples were frozen at −70 • C until analysis for glucose level measurement. The cellular layer was also stored at −70 • C for DNA preparation. To analyze urinary cotinine levels, spot urine samples were also collected and then stored at −20 • C until analysis. Information about basic demographics, including smoking status, was collected through an interview by trained staff.

PM Concentrations and Meteorological Factors
PM 10 and PM 2.5 concentrations were collected from the Korea National Institute of Environmental Research and Seoul Research Institute of Public Health and Environment, respectively, as day average levels during 8 days from seven days before to the health examination day (from lag day 7 to lag day 0) [4,22]. PM 2.5-10 concentration was calculated as the difference between PM 10 and PM 2.5 concentrations. Daily outdoor temperature and dewpoint of the day were obtained from the Korea Meteorological Administration. All PM concentrations and meteorological data were obtained from monitoring sites nearest to the residence of participants. Daily means were calculated and used as individual values corresponding to each participant.

Fasting Glucose Level Measurement
Serum level of fasting glucose was measured using a hexokinase method with a Pureauto S GLU kit (Daiichi Pure Chemicals, Tokyo, Japan) [4].

Cotinine Measurement
To determine exposure to tobacco smoke, urinary cotinine level was measured using an enzyme-linked immunosorbent assay (Cotinine Elisa; Bio-Quant, San Diego, CA, USA) following the manufacturer's procedure [4]. The lower limit of detection (LOD L ) of cotinine level was 1 µg/g, and the value under LOD L was assigned as 0.5 µg/g. A value greater than the upper LOD (LOD U = 10,000 µg/g) was assigned as 15,000 µg/g. Cotinine level was adjusted for urinary creatinine level in statistical analyses.

Genotyping
We listed all single-nucleotide polymorphisms (SNPs) on PARP4 and ERCC1 and examined their minor allele frequencies in Asian population using International HapMap data because low minor allele frequency may lead to null result; although, the SNP is meaningful for the risk of target outcome. After we searched for linkage structure using the Haploview to confirm which SNPs should be selected for haplotype construction, we finally selected three SNPs of PARP4 (rs12863638, rs3783073, and rs2275660) and two SNPs of ERCC1 (rs11615 and rs3212961) (Table A1). Table A1 shows detailed information for these five SNPs including rs number, Human Genome Variation Society (HGVS) name, chromosome number, their position, function, call rate, and accuracy. All five SNPs were genotyped using TaqMan method. In brief, a polymerase chain reaction (PCR) was carried out with a final volume of 5 µL, containing 10 ng of genomic DNA, 2.5 µL of 2X TaqMan Universal PCR Master Mix, 0.125 µL of 40X Assay Mix, and 1.25 µL of DNase-free water (Assay ID, C_9228399_10 for rs12863638; C_27515784_10 for rs3783073; C_15879320_10 for rs2275660; C_2532959_1 for rs11615; and C_25934767_10 for rs3212961). Thermal cycle conditions were: 50 • C for 2 min and 95 • C for 10 min, followed by 45 cycles at 95 • C for 15 s and 60 • C for 1 min. Dual 384-Well GeneAmp PCR System 9700 (ABI, Foster City, CA, USA) was used for PCR. Endpoint fluorescent readings were conducted using an ABI PRISM 7900 HT Sequence Detection System (ABI, Foster City, CA, USA). Five percent of samples were randomly chosen for repeated testing. Identical results were found with 100% accuracy rate (Table A1). We also tested the Hardy-Weinberg Equilibrium (HWE) for genotyped SNPs using chi-square test and found that all p-values were larger than 0.05, indicating that all SNPs were in HWE.

Haplotype Determination
Because adjacent SNPs in the same chromosome region can be inherited together in a haplotype, an analysis using multiple SNPs can simultaneously increase statistical sensitivity [23] or the power to detect loci relative to that of single SNPs [24]. Therefore, haplotypes composed of SNPs of each gene were made using PHASE program version 2.1 (http://stephenslab.uchicago.edu/phase/download.html, accessed on 14 May 2018). In brief, we deleted the data of people whose genotype was lacking for at least one SNP, and then composed haplotypes based on three SNPs of PARP4 (rs12863638, rs3783073, and rs2275660) and two SNPs of ERCC1 (rs11615 and rs3212961). The linkage disequilibrium (LD) between two SNPs of each gene was evaluated based on relative disequilibrium (D ) [25]. Statistical significance of LD was evaluated using Fisher's exact test.

Statistical Analysis
Baseline characteristics of participants were compared between males and females. Student's t-test was used for mean comparison and chi-square test was used for frequency comparison. Mean concentrations and selected percentiles of repeated measures, including fasting glucose and PMs, were calculated. For PMs, the day average from lag day 0 to lag day 7 was used. Original or log-transformed concentrations of PM 10 , PM 2.5-10 , and PM 2.5 were compared among lag days using analysis of variance (ANOVA). Relations among PM 10 , PM 2.5-10 , and PM 2.5 levels on each lag day were evaluated using Pearson's correlation. After fasting glucose levels were log-transformed for their normality, the relation of each PM with glucose level was estimated using mixed effect models since we repeated measurements for both PMs and glucose level. The relation of each PM with glucose level was also evaluated by genotypes and diplotypes of PARP4 and ERCC1 genes. In all models, changes in glucose level by inter-quartile range (IQR) changes in PM 10 , PM 2.5-10 , and PM 2.5 were evaluated after adjusting for age (year), sex (males vs. female), body mass index (BMI, kg/m 2 ), urinary cotinine level (µg/g creatinine), outdoor temperature ( • C), and dewpoint ( • C) of the day. Statistical significance was considered when p-value was lower than 0.05. All statistical analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA).

Baseline Characteristics of Study Participants
The study participants were 527 elders (Table 1). There were 136 (25.8%) males and 391 (74.2%) females, and the average visit number was 3.3. The mean age was 70.8 years for all participants, and most participants (n = 294, 55.8%) had BMI < 25. Frequencies of five SNPs were not significantly different between males and females (p > 0.05 for all five) ( Table 1).

Distribution of Repeated Measures during the Study Period
The average fasting glucose level was 5.5 mmol/L during the study period, while the mean concentrations of PM 10 , PM 2.5-10 , and PM 2.5 were 43.4 µg/m 3 , 19.9 µg/m 3 , and 23.6 µg/m 3 , respectively ( Table 2). The mean outdoor temperature and dewpoint of the day were 17.2 • C and 6.1 • C, respectively ( Table 2). The IQR was 22.7 µg/m 3 for PM 10 , 16.2 µg/m 3 for PM 2.5-10 , and 9.0 µg/m 3 for PM 2.5 ( Table 2).

PM Concentrations and Their Correlations
In the comparison of PM 10 , PM 2.5-10 , and PM 2.5 concentrations among lag days, PM 10 and PM 2.5-10 concentrations were significantly different among lag days for both original and log-transformed concentrations (p < 0.05 for all), while PM 2.5 concentrations were not significantly different among lag days; although, original values showed marginal significance (p = 0.0812) ( Figure 1). However, each PM 10 or PM 2.5-10 concentration was correlated among lag days (all p < 0.0001 with correlation coefficients ranging from 0.14 to 0.72 for PM 10 and from 0.24 to 0.66 for PM 2.5-10 ), while PM 2.5 concentration showed no significant correlations between levels on lag day 0 and levels on lag days 4 to 7 and between levels on lag day 1 and levels on lag days 6 and 7, although levels on other lag days were significantly correlated with each other (Table A2).
Since PM 2.5-10 and PM 2.5 were included in PM 10 , relations among PMs on the same day were also evaluated (Table A2). All three PMs were strongly correlated with each other with the maximum and minimum correlation coefficients of 0.92 and 0.54 (all p < 0.0001) (Table A2).

Association between PMs Exposure and Fasting Glucose Level
Relations between PM concentrations on eight lag days (from lag day 0 to lag day 7) and fasting glucose level were evaluated (Table 3). All three PMs significantly or marginally increased glucose levels from lag days 3 to 7, with the strongest effect on lag day 4, with mean concentrations of 44.5 µg/m 3 for PM 10 (Table 3). Because PMs on lag day 4 showed the strongest associations with glucose level based on the effect size, lag day 4 was chosen for further analyses.

Association between PMs Exposure and Fasting Glucose Level
Relations between PM concentrations on eight lag days (from lag day 0 to lag day 7) and fasting glucose level were evaluated ( Table 3). All three PMs significantly or marginally increased glucose levels from lag days 3 to 7, with the strongest effect on lag day 4, with mean concentrations of 44.5 µg/m 3 for PM10, 19.9 µg/m 3 for PM2.5-10, and 24.3 µg/m 3 for PM2.5 (percent change by IQR; change in PM: (β) = 1.4 and p = 0.0023 for PM10; β = 3.0 and p = 0.0010 for PM2.5-10; and β = 2.0 and p = 0.0134 for PM2.5) ( Table 3). Because PMs on lag day 4 showed the strongest associations with glucose level based on the effect size, lag day 4 was chosen for further analyses.

Association between PMs Exposure and Fasting Glucose Level by Genotypes or Diplotypes of Repair Genes
Relation between PMs exposure and glucose level was evaluated by genotypes of PARP4 and ERCC1. Participants with G allele for rs12863638, C allele for rs3783073, G allele for rs2275660, T allele for rs11615, and C allele for rs3212961 seemed to be susceptible to exposure to three PMs in relation to the increase in glucose level, although the

Association between PMs Exposure and Fasting Glucose Level by Genotypes or Diplotypes of Repair Genes
Relation between PMs exposure and glucose level was evaluated by genotypes of PARP4 and ERCC1. Participants with G allele for rs12863638, C allele for rs3783073, G allele for rs2275660, T allele for rs11615, and C allele for rs3212961 seemed to be susceptible to exposure to three PMs in relation to the increase in glucose level, although the ERCC1 TT genotype for rs11615 was marginally significant with the greatest effect size (Table 4).
All three SNPs of PARP4 and two SNPs of ERCC1 were in a strong LD with each other (p < 0.0001 for all relations). For PARP4, D was 0.95 between rs12863638 and rs3783073, 0.71 between rs12863638 and rs2275660, and 0.98 between rs3783073 and rs2275660. For ERCC1, D between rs11615 and rs3212961 was 0.89. Because the G allele for rs12863638, C allele for rs3783073, and G allele for rs2275660 of PARP4 and T allele for rs11615 and C allele for rs3212961 of ERCC1 showed an increasing trend for the susceptibility to PMs exposure and there were strong linkages among SNPs of same gene, we evaluated the relationship between PMs exposure and glucose level by combined diplotypes of PARP4 and ERCC1. First of all, we created the risky haplotypes of PARP4 (G-C-G haplotype composed of G allele for rs12863638, C allele for rs3783073, and G allele for rs2275660) and ERCC1 (T-C haplotype composed of T allele for rs11615 and C allele for rs3212961) and evaluated the effect of PMs exposure on glucose level in participants with or without risky haplotype (Table 4). For all three PMs, participants with the PARP4 G-C-G haplotype or ERCC1 T-C haplotype were found to be susceptible to PMs exposure based on their glucose levels (PARP4 G-C-G: β = 2.6 and p = 0.0006 for PM 10 , β = 3.5 and p = 0.0009 for PM 2.5-10 , and β = 1.6 and p = 0.0020 for PM 2.5 ; ERCC1 T-C: β = 2.2 and p = 0.0016 for PM 10 , β = 3.5 and p = 0.0003 for PM 2.5-10 , and β = 1.2 and p = 0.0158 for PM 2.5 ), while others (participants without PARP4 G-C-G haplotype or ERCC1 T-C haplotype) were not susceptible (p > 0.05 for all three PMs and both genes) ( Table 4). We also evaluated the interaction between PMs exposure and genotypes or diplotypes in relation to glucose level (Table 4). Two PMs (PM 10 and PM 2.5 ) showed significant interactions with PARP4 diplotypes for glucose level (p = 0.0262 for PM 10 and p = 0.0191 for PM 2.5 ), while PM 2.5 showed marginal significant interactions with PARP4 diplotypes (p = 0.0802) ( Table 4). However, ERCC1 diplotypes did not show interactions with all three PMs for glucose level (p > 0.05 for all three PMs) ( Table 4).
We also evaluated direct effects of genes on glucose level, but did not find significant effects of genes on glucose level (p > 0.05 for both genes) (Table A3).

Discussion
In the present study, all three PMs were strongly correlated with each other and PM 2.5 had a higher daily variability than the other PMs. Furthermore, participants with PARP4 G-C-G and ERCC1 T-C haplotypes were susceptible to PMs exposure in relation to fasting glucose level.
The World Health Organization (WHO) has suggested a daily average level of 45 µg/m 3 for PM 10 and 15 µg/m 3 for PM 2.5 in their Global Air Quality Guidelines [26]. Compared to the Global Air Quality Guidelines by WHO, our elderly population were exposed to a relatively and consistently high level of PMs, particularly PM 2.5 . Considering correlations among three PMs on a specific lag day or among several lag days for the same PM in our study, PMs were temporally variable, particularly PM 2.5 , although their levels were related to each other. In the present study, we chose lag day 4 for further analyses based on the strongest associations of PMs on lag day 4 with glucose level in the present study, and found the biggest change in glucose level after PM 2.5-10 exposure, followed by PM 2.5 and PM 10 . The biggest change in glucose level after PM 2.5-10 exposure was supported by previously reported studies. Liang et al. [27] showed that each increase of 10 µg/m 3 in 3-day moving averages of PM significantly increased the risk of outpatient visits of pneumonia, bronchiolitis, and asthma, regardless of PM size, with PM 2.5-10 showing the biggest effect size (4.36% of PM 10 , 9.19% of PM 2.5-10 , and 3.71% of PM 2.5 for outpatient visits of pneumonia; 3.12% of PM 10 , 9.13% of PM 2.5-10 , and 3.21% of PM 2.5 for bronchiolitis; and 3.33% of PM 10 , 11.69% of PM 2.5-10 , and 3.45% of PM 2.5 for asthma). Lei et al. [28] showed that PM 2.5-10 had stronger associations with the loss of lung function than PM 2.5 . The greatest effect size for PM 2.5-10 in our analyses could be attributable to the difference in the amount of lipopolysaccharide (LPS) among PMs. Biologic components intrinsic to PM such as LPS could directly activate Toll-like receptors, leading to inflammation [29]. In fact, LPS was reported to induce metabolic syndromes, such as IR [30,31]. Moreover, LPS was known to be dominant in urbanized environments in Asia [30]. The amount of LPS was reported to be higher in PM 10 than in PM 2.5 because of higher level of LPS in PM 2.5-10 [30]. Because the effect size of PM 10 could be offset by that of PM 2.5 in our study, the effect size of PM 2.5-10 on glucose level was the greatest. However, there is still a debate about which PM could play a role in oxidative-stress-related DNA damage. Feng et al. [32] indicated that DNA damage caused by fine particles, ranging from 0.43 to 2.1 µm in size, was greater than the damage caused by coarse particles, ranging from 4.7 to 10 µm in size. They also suggested that greater DNA damage in fine particles could be attributable to heavy metals enriched in fine particles [32]. Therefore, in the future, we need to clarify which PM size is more important for the prevention of oxidative-stress-related DNA damage as well as their biological functions and mechanisms.
DNA repair proteins such as PARP4 and ERCC1 can protect against metabolic dysfunction including IR [20,33]. Both PARP and ERCC family members were reported to be involved in the pathway of IR regulation through repairing oxidative DNA damage and inhibiting autoinflammatory response [20,33]. However, there is a lack of knowledge regarding the relation between PARP4, a particular member of PARP family, and fasting glucose level. Furthermore, the effect of PARP4 and ERCC1 variations on the relation between PMs exposure and glucose level has not yet been reported, although positive associations between DNA damage accumulation and the development of metabolic disorders were reported [33]. Because genetic polymorphisms of two repair genes, PARP4 and ERCC1, can affect DNA repair efficiency [34], their polymorphisms may modify the relation between PMs exposure and glucose level. In the present study, we found that participants with PARP4 G-C-G and ERCC1 T-C haplotypes were apparently susceptible to an increase in glucose level in relation to exposure to all three PMs, although the effect size was slightly different by PMs. Although no study reported the effects of PARP4 and ERCC1 polymorphisms on relation between PMs exposure and fasting glucose level, several studies supported the potential idea that those polymorphisms could affect the impact of PMs exposure on glucose level. For the ERCC1 gene, the defective ERCC1 gene could increase the incidence of vascular diseases [35], and genetic polymorphisms of ERCC1 could affect the efficiency of chemotherapy, as well as the susceptibility of a variety of diseases, including lung cancer and several cardiovascular diseases [36][37][38]. Particularly, the ERCC1 rs11615 T allele was found to be a risk factor for developing non-small cell lung cancer [37]. Because the lungs can be directly exposed to PMs [8,[11][12][13], the effects of these SNPs are plausible. In fact, the risk effect of rs11615 T allele [37] was consistent with our results, which we obtained with regard to both PMs and glucose level. Studies on ERCC1 rs3212961 were controversial, although the rs3212961 C allele was associated with a shorter overall survival in gastric cancer patients [39] and a higher risk for non-Hodgkin lymphoma development [40], which were consistent with results of the present study. However, we found no evidence for the relation between the three SNPs of the PARP4 gene examined in the present study and DNA damage or lung oxidation, although rs17080653, another intronic variant in PARP4, showed a protective effect on head and neck cancer [41] and inter-individual differences in DNA repair processes [42].
In the present study, we found interactions between PMs exposure and PARP4 genotypes or diplotypes in relation to glucose level. All three PMs, in particular PM 10 and PM 2.5 , showed interactions with PARP4 diplotypes, although the direct effect of PARP4 gene was not found. This meant the active functions of both environmental and genetic factors in relation with fasting glucose level, with an emphasis on the environmental factors in regulating genes. However, we did not find any significant interactions between PMs exposure and ERCC1 diplotypes, although ERCC1 rs11615 showed marginal interactions with PM 10 and PM 2.5 in relation with glucose. Therefore, in the future, we need to validate the interactive effect of the ERCC1 gene with PMs in relation to glucose level with a larger sample size.
To the best of our knowledge, the present study is the first to explore the effects of genetic modifications of PARP4 and ERCC1 on relations between three PM species and fasting blood glucose level, targeting elders who are susceptible to environmental pollutants. Although we used a panel study design, which could increase statistical power, our study also had limitations. First, we did not control for other air pollutants (O 3 or NO 2 ) and GST family genes polymorphisms, although we found adverse effects of O 3 and NO 2 exposures on the glucose level and modification of GSTM1 and T1, as well as an impact of P1 polymorphisms on the relation between PM 10 exposure and glucose level, in a previous study [4]. Because too many missing data could be produced if we matched O 3 and NO 2 to PM 10 and PM 2.5 on a daily basis, which could lead to non-significance due to small numbers of data used in the analyses, we did not control O 3 or NO 2 in our statistical models. This was the same for the GSTM1, T1, and P1 genes. In the future, we need to confirm whether the modification effect of PARP4 and ERCC1 polymorphisms on the relations between PMs exposure and glucose level still remains after controlling for these factors with a larger sample size. Second, in the present study, we only explored the acute effects of PMs on glucose level. Therefore, we need to clarify whether our genetic modification effect still remains after long-term exposure and short-term exposure. Third, in the future, we need to consider the chemical nature of the particles to confirm the causality of particles as well as genes in relation to glucose level.

Conclusions
In conclusion, all three PMs were strongly correlated with each other and PM 2.5 had higher daily variability than the others. Furthermore, elderly participants with PARP4 G-C-G and ERCC1 T-C haplotypes were susceptible to an increase in fasting glucose level after exposure to all three PMs, regardless of PM 10 , PM 2.5-10 , or PM 2.5 . These results should be confirmed in the future after considering other causal factors and confounders with a larger sample size. Informed Consent Statement: All participants provided written informed content.

Conflicts of Interest:
The authors declare no conflict of interest.