Spatiotemporal Distributions and Land-Use Regression Models of Ambient Bacteria and Endotoxins in the Greater Taipei Area

Ambient bacteria and endotoxins are components of bioaerosols, which are abundant in ambient air. Exposure to ambient bacteria and endotoxins has been associated with respiratory symptoms. We monitored the spatiotemporal distributions of ambient bioaerosols in the Greater Taipei area by using multisite sampling and developed regression models for ambient bacterial and endotoxin concentrations. Forty-four representative sampling sites were selected from the Greater Taipei area. Samples were collected in 4 seasons throughout a 1-y study period. Ambient bacteria were quantified using acridine orange staining, and endotoxins were analyzed using Limulus amebocyte lysate assays. Land-use types and major determinants of the bioaerosols were used to develop regression models. Ambient bacteria ranged from < limit of detection (LOD) to 1.68 × 10 cells/m, whereas endotoxins ranged from < LOD to 20.62 EU/m. Significant seasonal variations were observed for both bioaerosols, with the highest concentrations observed in spring. Regression analyses revealed temperature, relative humidity, and particulate matter as the major predictors of ambient bacteria and endotoxins. No land-use type was correlated with any of the bioaerosols. The number of schools and gas stations was positively associated with both bioaerosols. The leave-one-out cross-validation R of the final models for ambient bacteria and endotoxins were 0.11 and 0.31, respectively. The results of this study revealed high spatiotemporal variations in the distributions of ambient bacteria and endotoxins in the Greater Taipei area. Additional potential predictors should be included in future studies to develop better predictive models for ambient bacteria and endotoxins.


INTRODUCTION
Ambient bacteria are a component of bioaerosols, which are abundant in both indoor and outdoor environments.This group of bioaerosols comprises pathogenic, nonpathogenic, live, and dead cells along with their toxic products or components such as endotoxins (Macher, 1999;Douwes et al., 2003;Awad, 2007;Pommerville, 2011).Endotoxins, consisting of lipopolysaccharides (LPS), are the major components of the outer membranes of Gram-negative bacteria, which are released after cell lysis or during active cell growth.LPS are fairly heat stable and normally are not affected by usual sterilization processes (Milton 1999;Pommerville, 2011).Several studies have demonstrated that endotoxins can be detected in the ambient air by using various types of collection methods, such as filter cassettes connected to personal pumps, high volume samplers, and impingers (Macher, 1999;Su et al., 2002;Adhikari et al., 2011;Cheng et al., 2012).
Numerous studies have demonstrated the adverse health effects of exposure to these biological agents such as infections, allergies, respiratory diseases, and activation of the immune system (D'Agata et al., 1999;Braun-Fahrlander et al., 2002;Huang et al., 2002;Herr et al., 2003;Adhikari et al., 2011;Ege et al., 2011).Most of these studies have focused on occupational and indoor environments because humans spend a majority of their time in these environments.However, outdoor exposure is also crucial.Studies have indicated that bacterial and endotoxin concentrations vary among different types of outdoor environment, with considerable seasonal variations as well (Mueller-Anneling et al., 2004;Morgenstern et al., 2005;Bowers et al., 2010;Tager et al., 2010;Jeon et al., 2011;Bowers et al., 2012;Haas et al., 2013;Woo et al., 2013).
Recent studies have reported bacterial concentrations using total cell counts, culturable cells, and endotoxins in various fractions of the particulate matter (PM).Bowers et al., (2012) indicated that ambient bacteria constituted up to 22% of > 0.5 µm of near-surface aerosol particles.Several studies have detected varying concentrations of endotoxins in both PM with aerodynamic diameters ≤ 10 µm (PM 10 ) and ≤ 2.5 µm (PM 2.5 ).The varying results reported in various studies could be due to the varying sampling methods, meteorological factors, seasons, and potential spatiotemporal variations (Morgenstern et al., 2005;Nilsson et al., 2011;Cheng et al., 2012;Haas et al., 2013;Bari et al., 2014).However, certain studies have not reported significant associations between the concentrations of endotoxins and PM by using multiple regression analyses (Menetrez et al., 2007;Degobbi et al., 2011;Strak et al., 2011).
Although bacterial communities and endotoxin concentrations have been reported in several studies (Gorny and Dutkiewicz 2002;Park et al., 2002;Herr et al., 2003;Oppliger et al., 2005;Rusca et al., 2008;Tendal and Madsen, 2011), only a few studies have evaluated their relationships with the types of land use (Mueller-Anneling et al., 2004;Madsen, 2006;Bowers et al., 2010;Tager et al., 2010;Woo et al., 2013).In a study designed to investigate the risk factors of childhood asthma, air samples were collected from several locations throughout Fresno, a city in central California, USA; the study revealed that the endotoxin concentrations varied among various geographical regions.Higher concentrations were closely associated with cropland, pastureland, and animal feeding areas such as dairy farms (Tager et al., 2010).However, the types of land use did not prove as significant predictors of endotoxin concentrations in a study conducted in Munich, Germany, by using an inverse distance weighting method (IDW) to fit the regression model (Morgenstern et al., 2005).Regarding bacterial communities, Bowers et al. (2010) reported that bacterial cell counts were stable among various types of land use, but the biological ice nuclei concentrations were 2-and 8-times higher in agricultural areas compared with those in suburban areas and forests, respectively (Bowers et al., 2010).Furthermore, several environmental monitoring studies have reported high concentrations of bioaerosols, including bacteria, endotoxins, and pathogenic fungi, near biological sources, such as composting sites and animal farms (Pomorska et al., 2007;Thorne et al., 2009;Ko et al., 2010;Adhikari et al., 2011;Pankhurst et al., 2011b).
Because the knowledge of spatiotemporal variations of ambient bacteria and endotoxin is extremely limited, we conducted a 1-y study to evaluate the characteristics and determinants of bacteria and endotoxins throughout the Greater Taipei area.Multisite sampling was performed to evaluate various types of land use.In addition, we developed regression models to determine the distributions of ambient bacterial and endotoxin concentrations by applying landuse types and other crucial environmental factors.

Sampling Locations
The Greater Taipei area comprises the Taipei City and New Taipei City (formerly Taipei County).The area was divided into 5 homogenous areas (Fig. 1).In brief, the area was divided according to various social and environmental factors related to bioaerosol distributions (e.g., land-use information, meteorology, air pollutants, population density, and household density) by using principal component analysis and cluster analysis (Chen, 2011).To maximize collaboration, the sampling locations were primary and secondary schools in the Greater Taipei area.The potential sites were visited and determined to ensure an even distribution of all sampling sites, availability of space and power for sampling, and convenience of transportation.There are no specific rules to select the number of sampling sites.However, according to Hoek et al. (2008), a minimum of 40 sites are needed for LUR analyses in air pollution monitoring studies.Therefore, in total, 44 representative samplings sites were chosen for bioaerosol monitoring with 8 to 11 sampling sites in each homogeneous area.Fig. 2 shows the locations of all sampling sites.

Air Sampling
Air sampling was conducted in 4 seasons (fall, winter, spring, and summer) from November 2011 to August 2012.The sampling campaign for each season was completed in 2 wk, with 6 to 9 sites sampled each day.To adjust for temporal variations of different sampling dates, continuous sampling was conducted at a central monitoring site located on a rooftop platform of a 10-story building at Taipei Medical University.The sampling dates were selected to avoid major events in Taiwan, such as the Lunar New Year and Tomb-Sweeping Festival, which generate excessive particles that might interfere with results.
For air sampling, we used 37-mm polycarbonate (PC) filters with a 0.8-µm pore size in 3-piece plastic cassette holders, coupled with a personal pump (224-PCXR series, AirChek XR5000, SKC, Eighty Four, PA, USA) operated at a flow rate of 5 L/min.The air-sampling equipment was placed on the rooftop platform of each sampling location for a 24-h period.Pyrogen-free tubes were used to store the samples by applying an aseptic technique.The tubes were immediately returned to the laboratory, and the samples were stored at 4°C and subsequently extracted and analyzed within 1 wk.

Sample Analysis
The filter samples were extracted in 3 mL of extraction buffer (0.01% Tween 80 in pyrogen-free water), vortexed for 2 min, and sonicated for 15 min (Wang, 2012).The eluted samples were split into aliquots for subsequent analyses.
For analyzing total bacteria, 1 mL of the eluted sample was filtered through a 25-mm black PC filter with a 0.2μm pore size in a filter holder by using a vacuum pump.The PC filter was then stained with 1 mL of 0.1 mg/mL acridine orange for 10 min; the excess stain was removed, and the filter was mounted on a glass slide with a cover slip and sealed at the edges with nail polish (Burton, 2005).The slides were stored in slide boxes to avoid exposure to light until examination.A fluorescence microscope (IX81 Motorized Inverted Microscope, Olympus, New Orleans, LA, USA) was used to analyze ambient bacterial counts at a 400× magnification.Twenty fields were counted per slide, with 10 equidistant fields (0.5 mm apart) of 2 perpendicular diameters (Tsai, 2011;Wang, 2012).The ambient bacterial concentrations were calculated as follows: where C ambient bacteria (cells/m 3 ) is the total ambient bacterial concentration, N is the average bacterial count of 20 microscopic fields, R is the effective radius of the PC filter (12.5 mm), A is the microscopic field area (mm 2 ), V 1 is the extraction buffer volume (3 mL), V 2 is the eluted volume for analysis (1 mL), and air volume sampled (m 3 ) is the flow rate of the sampler (L/min) × sampling time (min) × 0.001 (m 3 /L).
A Limulus Amebocyte Lysate assay, CHROMO-LAL, was employed to quantify the endotoxin concentrations of each sample.The analytical protocol was performed according to manufacturer instructions (Associates of Cape Cod, E. Falmouth, MA, USA) (CAPE-COD 2007).Sample mixtures were analyzed and quantified using the Multiskan Ascent ELISA reader at 405 nm with the Multiscan Assent software (Thermo Fisher Scientific, Waltham MA, USA).A standard curve was created to calculate the endotoxin concentrations of each sample, with a coefficient of determination (R 2 ) ≥ 0.980.The final concentrations were reported as endotoxin units per air-sampling volume (EU/m 3 ).The endotoxin concentrations were calculated as follows:

Endotoxin = C × V/air volume sampled
(2) where Endotoxin (EU/m 3 ) is the total ambient endotoxin concentration, C (EU/mL) is the endotoxin concentration of eluted samples, V (mL) is the extraction buffer volume (3 mL), and air volume sampled (m 3 ) is the flow rate of the sampler (L/min) × sampling time (min) × 0.001 (m 3 /L).

Social and Environmental Data
Social factors included population density, household density, land-use type, and potential sources of bioaerosols, which were defined as points of interest (POIs).All social factor data were provided by the National Land Surveying and Mapping Center (NLSC, Taichung, Taiwan), except the POI data, which were obtained from Kingwaytek Technology (Taipei, Taiwan).Supplemental Table S1 lists the land-use types and POIs.
The environmental data, including terrain, vegetative cover, meteorological data, and atmospheric pollutants, were respectively obtained from the Aerial Survey Office, Forestry Bureau, Council of Agriculture (Taipei, Taiwan); NASA MODIS satellite image photos (http://modis.gsfc.nasa.gov);Central Weather Bureau (CWB, Taipei, Taiwan); and Environmental Protection Administration (EPA, Taipei, Taiwan).Fig. 2 lists the EPA and CWB monitoring sites.
The terrain data included slope, aspect, and land elevation.
The amount of vegetative cover was determined using the normalized difference vegetation index (NDVI).The meteorological factors included temperature (°C), relative humidity (RH, %), wind speed (m/s), and rainfall (mm).

Data Analysis
Microsoft EXCEL 2007, SigmaPlot (Version 12, Systat Software Inc., San Jose, CA, USA) and the SAS statistical package (Version 9.2, SAS Institute, Cary, NC, USA) were used for data processing and statistical analyses.The geographic information system (GIS) ArcGIS (Version 9.3, Esri, Redlands, CA, USA) was used for spatial analysis.Because our sampling locations differed from the CWB and EPA monitoring sites, a general Kriging interpolation method (Ordinary Kriging) in ArcGIS was used to estimate the meteorological and air pollution data at each sampling location.The values of the social parameters (e.g., population density) and environmental factors (e.g., NDVI) were calculated within a designated radius (100, 200, 400, 800, 1000, 1500, and 2000 m) of the sampling locations, called buffer zones.One-way analysis of variance (ANOVA) and the Kruskal-Wallis test were used to compare the mean values of the normally and non-normally distributed environmental factors, respectively, among the various seasons.Multiple regression analyses (SAS PROC MIXED procedure) were used to evaluate the relationships between the bioaerosol concentrations and the social and environmental factors.Moreover, the heterogeneous autoregressive covariance model (ARH [1]) was used to adjust for the autocorrelations resulting from repeated measurements.Each potential predictor variable was first examined using a univariate regression analysis.Variables with p ≤ 0.2 were applied for multivariate analyses.The final models included all the predictor variables with p < 0.05.Ambient bacterial and endotoxin concentrations were transformed using a base-10 logarithm to approximate normality in the regression analysis.To adjust for temporal variations during each sampling campaign, the concentrations at the central sampling site were used.Pseudo-R 2 statistics were computed from the variance component by using the SAS PROC REG procedure (Singer and Willett 2003).Land-use regression (LUR) models were evaluated using the leave-one-out cross-validation (LOOCV) method, and cross-validation R 2 (CV-R 2 ) values were calculated (Liu et al., 2008;Rivera et al., 2012).

RESULTS
Samples were collected in 4 seasons over a 1-y study period.Overall, 159 and 150 samples were collected for ambient bacterial analysis and endotoxin analysis, respectively.Table 1 shows the distributions of the meteorological and environmental parameters for each season, which differed significantly among the seasons (p < 0.05).The ambient bacterial concentrations varied from < limit of detection (LOD) to 1.68 × 10 5 cells/m 3 , revealing significant seasonal variations (p < 0.05) (Fig. 3).The median ambient bacteria concentration was highest in spring (3.4 × 10 4 cells/m 3 ), followed by winter (3.0 × 10 4 cells/m 3 ), fall (1.1 × 10 4 cells/m 3 ), and summer (9.7 × 10 3 cells/m 3 ).The ambient endotoxin concentrations ranged from < LOD to 20.62 EU/m 3 , revealing significant seasonal variations (p < 0.05),   as shown in Fig. 4. The median concentration was highest in spring (4.04 EU/m 3 ), followed by summer (1.69 EU/m 3 ), winter (0.78 EU/m 3 ), and fall (0.54 EU/ 3 ).Table 2 presents the multiple regression models for both ambient bacteria and endotoxins.The ambient bacterial concentrations were positively associated with RH and PM 2.5 .In addition, the endotoxin concentrations revealed positive associations with temperature, RH, O 3 , and PM 10 .Moreover, in these models, the number of schools and gas stations was positively associated with the ambient bacterial and endotoxin concentrations.However, no significant correlation was observed between land-use type and ambient bioaerosol concentrations.The cross-validation R 2 values of the final models for ambient bacteria and endotoxins were 0.11 and 0.31, respectively.
The spatial distributions of ambient bacteria and endotoxins in each season are presented in Figs. 5 and 6, respectively.High spatial variability was observed for ambient bacterial concentrations in all seasons except spring.Higher ambient bacterial concentrations were observed in the mountain areas of the northeastern and southwestern regions in fall, the northwestern region in winter, and the southern region in summer.In contrast to other seasons with obvious spatial variations, the overall bacterial concentrations in spring were relatively high and evenly distributed throughout the study area.
Regarding endotoxins (Fig. 6), higher concentrations were observed in the mountain areas in each season.In fall, higher concentrations were observed in the northern region and along the west coast to the southern region of Greater Taipei, which is also a mountain area.In winter, higher concentrations were observed in the mountain areas of the southern region and northeast coast.In spring and summer, the spatial distribution patterns were similar, with higher concentrations along the northern coast and southern mountain areas.Moreover, obvious spatial variations were observed in both of these seasons.

DISCUSSION
In this study, we performed multisite sampling over 1 y to examine the spatiotemporal distributions of bioaerosol concentrations in the Greater Taipei area.Instead of using a high-volume sampler at a few fixed sites, we used personal sampling pumps and filter cassettes to maximize the number of sampling sites with limited resources.We did not use impingers for sampling to avoid frequently refilling of the collection fluid.No strict rules were implemented to determine the appropriate number of sites for the LUR study.The number had to be appropriate according to the study domain, and the sampling sites had to represent the entire study area and be spatially distributed over the area (Hoek et al., 2008).In total, 44 sampling sites were located among 5 homogenous areas.In some areas, the sampling sites were not evenly distributed because of geographical limitations such as high mountains or forests.However, we assumed that the characteristics related to bioaerosol concentrations within each homogeneous area were similar.Thus, we selected 8 to 11 sites in each homogeneous area to represent the related characteristics.Although the sampling sites were not evenly distributed in some homogeneous areas, we presumed that these 44 sampling sites were representative of the entire Greater Taipei area.
Most ambient bacterial studies have focused on culturable species by using a culture-based method.The concentrations have been tended to be high because the sampling has been performed at the sources of bioaerosols, such as swine farms and composting sites, or during dust storm events (Pankhurst et al., 2011a, b;Raisi et al., 2012).However, culture-based methods can capture only culturable cells and exclude nonviable and non-culturable cells as well as their toxic  substances.The sampling method used in the present study can capture bacterial cells and their toxic substances.The staining method can stain all bacterial cells, dead or alive, thus adequately reflecting the actual bacterial concentrations in the ambient environment.However, this method cannot identify the component species in samples, in contrast to the molecular-based method (Jeon et al., 2011).
Compared with a similar study conducted in Colorado, USA, by Bowers et al. (2010), who evaluated the spatial variations of ambient bacterial concentrations among 3 land-use types, the bacterial concentrations determined in our study were approximately 10 to 100 times lower; this could have been due to the varying characteristics of the sampling sites.Our samplers were located on rooftops of buildings (3-5 stories tall) and collected bacteria in the ambient air.However, in Bowers et al., the samplers were located at 2.5 m above the ground at the sampling site, and the samplers could also collect the resuspended particles with soil bacteria attached.Therefore, their bacterial concentrations were higher (Bowers et al., 2010).Another study conducted by the same research group examined the seasonal variations in ambient bacterial concentrations at a high-elevation site (3220 m above sea level) in Colorado, USA, and revealed that the ambient bacterial concentrations were the highest in fall and spring (Bowers et al., 2012).However, our results indicated that the bacterial concentrations were higher in spring and winter.The differences between the results of these 2 studies could be attributed to the differences in their sampling locations and environmental conditions (Bowers et al., 2012).Our study area had a subtropical climate with a mild humid winter.By contrast, their sampling site (Bowers et al., 2012) was located on the mountain tops of Colorado, which are covered by snow throughout the year except in summer.Moreover, the local temperature and precipitation patterns significantly influence the ambient bacterial concentrations.In the present study, the endotoxin concentrations ranged < LOD to 20.62 EU/m 3 , with an overall mean and median of 2.75 and 1.27 EU/m 3 , respectively.This concentration range is comparable with those reported in other studies that have investigated ambient endotoxin concentrations in various countries (i.e., USA, Germany, Sweden, and Hong Kong) (Mueller-Anneling et al., 2004;Morgenstern et al., 2005;Tager et al., 2010;Nilsson et al., 2011;Cheng et al., 2012).However, the concentration range observed in the present study was lower than those reported in previous studies conducted in occupational settings, wherein the measurements were generally performed at the sources of endotoxins (Su et al., 2002;Madsen and Nielsen, 2010;Rimac et al., 2010).
After inclusion of all major potential variables in the multiple regression models, the final models revealed that meteorological and environmental parameters were positively correlated with both ambient bacterial and endotoxin concentrations.Compared with temperature, RH was more consistently correlated with both these bioaerosols.This significant correlation may be explained by water availability, which is crucial for bacterial growth.High water activity (a w ) is typically favorable for bacterial growth because the bacteria can absorb this water from their living substrates for metabolism, which also includes toxin production.In addition, the high RH may result in clumping of the cells, which possibly increases the cell survival (Marthi et al., 1990).However, the role of temperature is different.Different types of bacteria require different optimal temperature ranges to grow; thus, the relationships are nonlinear as explained by the growth-permissible temperature ranges (Heitzer et al., 1991;Macher, 1999;Pommerville, 2011).However, temperature still revealed a strong positive association with endotoxin concentrations, and this result is consistent with that reported in a study in Brazil that also indicated a positive association of temperature with endotoxin concentrations (Degobbi et al., 2011).
In the present study, the ambient bacterial and endotoxin concentrations revealed no significant correlation with the type of land use, although various types of land use were examined in our regression models.This result is consistent with that of a study conducted in Colorado, USA, wherein the bacterial concentrations were stable among 3 types of land use, but the compositions differed and were influenced by local terrestrial surfaces (Bowers et al., 2010).However, in our study, we did not identify bacterial taxa and thus could not further discuss this concern.Regarding endotoxins, studies conducted in California, USA, reported spatial variability of endotoxins throughout the study area.Endotoxin concentrations have been associated with agricultural and pastureland areas, particularly in the downwind direction (Mueller-Anneling et al., 2004;Tager et al., 2010).However, a low spatial variability was also reported in a study by Morgenstern et al.,(2005), which investigated the spatial distribution of endotoxins in Munich, Germany; this may be due to the similar characteristics of the sampling sites, which were all located in an urban area.Therefore, meteorological parameters exerted greater effects on ambient endotoxin concentrations than did land utilization.In addition, this study included potential sources of endotoxins, such as gardens, trash bins, and waste composting sites for analysis, revealing an association between the number of potential sources and the endotoxin concentrations.The most common potential sources are gardens and trash bins, which were present within 100 m of approximately 50% of the sampling sites (Morgenstern et al., 2005).The present study also included potential sources of endotoxins (POIs) in the regression models, which revealed an association between the number of schools and gas stations (POI5) and both ambient bacterial and endotoxin concentrations.This proves that bioaerosols are closely associated with human activities.Several human activities can resuspend settled PM into the ambient air, such as students playing on school playgrounds or customers driving in and out of gas stations all day.Moreover, studies have shown that bioaerosols can be detected from car tires, deposits, and airborne particles on the street (Miguel et al., 1996;Menetrez et al., 2007;Menetrez et al., 2009).
Furthermore, a positive correlation was observed between the ambient bacterial and endotoxin concentrations and several air pollutants.The size of a bacterium is approximately 0.5 to 5 µm in diameter (Pommerville, 2011) and is thus positively associated with PM 2.5 .Regarding endotoxins, because they are typically released after cell lysis or during active cell growth of Gram negative bacteria, they are likely to attach to larger particles in the ambient environment.Thus, a positive correlation was observed between endotoxin concentrations and PM 10 , which is consistent with the findings of several other studies (Mueller-Anneling et al., 2004;Nilsson et al., 2011).The positive association between O 3 and endotoxin concentrations is probably due to the covariation of various air pollutants.
Although several significant predictor variables were included in the final regression models, neither model provided strong predictive power.The cross validation R 2 values for ambient bacteria and endotoxins were 0.11 and 0.31, respectively.This indicates that in addition to the applied variables, there are still other variables that can influence ambient bacterial and endotoxin concentrations in the Greater Taipei area.These variables could be certain human activities, specific types of agriculture, plant, or building material, which warrants further investigation.
A high spatial variability was observed for ambient bacterial and endotoxin concentrations in the Greater Taipei area in most seasons (Figs. 5 and 6).According to the land utilization map (Fig. 2), the common areas (i.e., northern and southwestern) with high ambient bacterial and endotoxin concentrations were agricultural and forest lands.This finding is consistent with that of a study conducted in Fresno, California, USA, which revealed high endotoxin concentrations in agricultural areas, particularly in the downwind direction (Tager et al., 2010).The primary sources of these bacteria and endotoxin would be the surfaces of living plants, soil, and probably the ocean for the coastal areas (Otten and Burge 1999).However, the present study did not reveal and significant effects of these types of land use in the regression models.Therefore, additional studies must be conducted to determine all the factors responsible for high spatial variations.

CONCLUSION
We investigated the spatiotemporal distributions of ambient bacteria and endotoxins in the Greater Taipei area by using multisite sampling in 4 seasons.Significant spatial and temporal variations were observed in the ambient bacterial and endotoxin concentrations.Therefore, using data from a single monitoring station to estimate the exposure of residents in the area is inadequate.The regression analysis revealed a positive correlation between ambient bacterial concentrations and RH and PM 2.5 .Endotoxin concentrations revealed a positive association with temperature, RH, and PM 10 .Moreover, the number of schools and gas stations in the vicinity revealed positive associations with both ambient bacterial and endotoxin concentrations.However, the predictive power of the models is relatively low.Thus, additional potential variables should be included in future studies to develop superior predictive models for determining the distributions of ambient bacteria and endotoxins.

ACKNOWLEDGEMENT
This study was supported in part by Ministry of Science and Technology, Republic of China (NSC99-2221-E-038-005-MY2; MOST103-2119-M-038-001).The funding source had no role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; or in the decision to submit the manuscript for publication.The authors declare no competing financial interests.

APPENDIX A: SUPPLEMENTAL MATERIALS
Supplementary data associated with this article can be found in the online version at http://www.aaqr.org.

Fig. 1 .
Fig. 1.Map of Taiwan depicting the 5 homogenous areas throughout the Greater Taipei area.

Fig. 2 .
Fig. 2. The Greater Taipei area map with major land-use types, sampling locations, and monitoring sites of the Environmental Protection Administration (EPA) and Central Weather Bureau (CWB).

Fig. 3 .
Fig. 3. Seasonal distributions of total ambient bacterial concentrations in the Greater Taipei area (cells/m 3 ).The box plots indicate the mean (dotted line) and median values, including the 10 th , 25 th , 75 th , and 90 th percentiles, and outliers.

Fig. 4 .
Fig. 4. Seasonal distributions of ambient endotoxin concentrations in the Greater Taipei area (EU/m 3 ).The box plots indicate the mean (dotted line) and median values, including the 10 th , 25 th , 75 th , and 90 th percentiles, and outliers.

Fig. 5 .
Fig. 5. Spatial distributions of ambient bacteria in the 4 seasons in the Greater Taipei area.

Fig. 6 .
Fig. 6.Spatial distributions of endotoxins in the 4 seasons in the Greater Taipei area.

Table 1 .
Distributions of environmental parameters for each season.

Air Pollutants : mean (min, max)**
The summer sampling campaign was divided into 2 periods because of a typhoon event at the end of July 2012.** All meteorological factors and air pollutants exhibited statistically significant differences among the 4 seasons (p < 0.05) according to ANOVA or the Kruskal-Wallis test.

Table 2 .
Final LUR model for determining the distributions of ambient bacteria and endotoxins in the Greater Taipei area.