**3. Results**

### *3.1. Air Pollution Exposure over Seasons*

Table 1 compares particulate air pollution exposure between spring and summer. Independent two-sample t tests were conducted on PM2.5, BC and other measurements from passive sampling.

From Table 1, PM2.5 concentrations were significantly lower in the summer season than in the spring season, a generally expected finding given the generally cleaner conditions in summer at this location. While the mean PM2.5 concentration was larger by a factor of 2.7 during spring compared to that in summer, the mean BC concentration was larger by a factor of 1.1 during spring compared to that in summer. Measurements were made from busy roads and roadsides, and the BC sources at these sites were related mainly with vehicular exhaust. Because traffic activities were not expected to be different during two seasons, the insignificant difference in BC concentrations suggested the importance of traffic-related PM sources in both seasons. The increased level of PM2.5 from summer to spring was due to the higher regional pollution in spring than in summer. There were also additional sources such as seasonal operation of brick kilns and refuse burning during the spring season. In contrast to particle pollution, nitrogen oxides, sulfur dioxide, and ozone measurements obtained from passive sampling were not significantly different between the two seasons. Although the PM2.5 level was shown to be associated with the season, they did not present highly overlapping power in explaining the biomarker response variables in the linear mixed models. It was shown that in the full model with the six independent variables, the variance inflation factors (VIFs) of PM2.5 and season, which measure the dependency among variables, were both below 4, where the multicollinearity issue usually arises when any of the VIFs exceeds 5 or 10. More model results about the effect of PM2.5 and season will be discussed in Section 3.3.

### *3.2. Biomarkers over Seasons*

The comparison results of biomarker concentrations between spring and summer of 2014 are shown in Table 2. Three statistical tests were considered. The independent *t*-test performed a two-sample *t*-test assuming independence of the volunteer samples in the two seasons. However, there were repeated subjects who participated in the samples from both seasons. Unfortunately, we were not able to follow all subjects during both seasons. In fact, during the second phase of study in summer, more than half of subjects had been transferred to other locations. Only 13 subjects were repeated from the respective sites in both seasons. Out of these 13 repeated subjects, serum biomarker data were available for only 12 subjects. Hence, dependent samples only counted these 12 subjects. Then, a parametric two-sample *t*-test and a nonparametric Wilcox test were both performed on the dependent samples. All these three tests using independent or dependent samples were two-sided and conducted on the data of each individual's average biomarker concentrations. From Table 2, out of 14 biomarkers, 10 of them showed significant increases, at the significance level of 0.05, in summer than in spring (consistent across all three tests), whereas VCAM-1 exhibited no significant differences in any of the three tests. Three biomarkers, CRP, ICAM-1, and IFN-γ, showed significant differences only partially from one or two of the tests. Of these ten biomarkers with consistently significant results, eight of them were very significant with *p*-values below 0.001 while assuming independent samples and five of them were very significant using dependent samples. The top three biomarkers with the overall most significant differences were IL-2, IL-4, and IL-6.

### *3.3. Effects of Air Pollution and Various Factors on Biomarkers*

In the previous section, test results appeared to show that season had a strong impact on biomarker concentrations. However, those results did not take into account the potential effects of other factors, along with season, on biomarkers. Hence, linear mixed regression models were run to investigate the effect of six variables simultaneously on biomarker levels: PM2.5, BC, season, smoking habit, wearing mask, and gender. Table 3 lists the estimated coefficients of these variables as fixed effects in the linear mixed models for each biomarker. The results showed that PM2.5 exposure was positively associated with three biomarkers, i.e., CRP, SAA, and IL-10, and negatively associated with four biomarkers, i.e., IL-1β, IL-12, IL-13, and IFN-γ, while holding all the other factors in the models constant. BC did not have an effect on most of biomarkers except two biomarkers, where it was negatively correlated with two biomarkers, i.e., Il-10 and TNFα, but the effect was very small (Table 3). As expected, season had

a strong effect on biomarker levels and it was the only factor that showed a significant effect for all the biomarkers except IFN-γ. Consistently, the concentration levels across all these biomarkers tended to be higher in summer than in spring.

Subject gender also had an effect on eight biomarkers: SAA, IL-2, IL-4, IL-6, IL-10, IL-12, IL-13, and IFN-γ. For these eight biomarkers, females tended to have higher concentration levels than males. Note that there were 7 female subjects out of 53 subjects. The effects of season and gender are illustrated in Figure 2 in the model for IL-6. In addition, it was found that wearing a protective facemask had a positive effect on IL-10, IL-12, and IL-13. The smoking habit was positively associated only with TNF-α with a *p*-value of 0.02.

**Figure 2.** Predicted log (IL-6) concentration (with confidence intervals) versus season (Su: summer; Sp: spring) and gender (F: female; M: male).

Since the PM2.5 concentration did not exhibit a significant effect on more than half of the biomarkers from the linear mixed models, we considered to further investigate the PM2.5 chemical composition (elemental concentrations) and their relations with season on biomarkers. The results were plotted in Figure 3. It can be seen that the contributions of six elements—aluminum (Al), silica (Si), potassium (K), iron (Fe), nickel (Ni), and zinc (Zn)—were enhanced in summer than in spring. This showed that the variation among biomarker levels may be related to particulate composition more than the total particulate mass alone. Some of these elements such as Fe, Ni, and Zn are also related with traffic emissions [51], suggesting also the changes in traffic patterns in two seasons. However, there was no significant difference in BC concentrations between two seasons. Due to a large amount of missing values in element contributions, they were not included as explanatory variables in the linear mixed models of this study, but they could be valuable factors for biomarkers to be investigated in future studies. Besides these six elements, other elements were also measured on collected filters by X-ray fluorescence spectroscopy, but they were not enhanced during summer than during spring [40].

**Figure 3.** Normalized concentrations of elements (μg/m<sup>3</sup> of elements normalized by μg/m<sup>3</sup> of PM2.5) during spring and summer seasons, 2014.
