Association of Cardiovascular Responses in Mice with Source-apportioned PM2.5 Air Pollution in Beijing

In this study, factor analysis and mass regression were used to identify four fine particulate matter sources and estimate their contributions to the ambient air pollution in Beijing. The identified sources were traffic re-suspended soil, mixed industrial sources, oil combustion, and secondary sulfate. The estimated source contributions were then introduced into two models as exposure variables to explore the relationships between cardiovascular responses in mice and PM exposures. We observed that PM2.5 has a small negative acute effect on heart rate, but the individual source factors showed much more significant effects. Traffic re-suspended soil had the most significant effect on heart rate, with a positive contribution on the day of exposure and a negative one on day lag 1. Acute heart rate variability outcomes were better explained by the total PM2.5 than by the source components. Chronic effects were observed as a decreased heart rate but an increased number of heart rate variability outcomes.


INTRODUCTION
The economic development and urbanization of Asia as whole, and China in particular, during past decades were accompanied by unprecedented growth in energy consumption, traffic, and pollution, thus creating a need for systematic assessment of the health effects of air pollution on populations in that region (HEI Report 18,20).Exposures to ambient particulate matter (PM), especially PM 2.5 (fine particles with an aerodynamic diameter smaller than 2.5 µm) have been shown to be associated with increased cardiovascular mortality and morbidity in extensive epidemiologic studies (e.g., Pope et al., 2004, Brook et al., 2010).The WHO and many other regulating bodies, have identified and implicated air pollution to be the most crucial environmental factor in human health evaluations.Globally, an estimated 2.5 million yearly deaths have been attributed to PM 2.5 and PM 10 exposure (Chuang et al., 2011;Shah et al., 2013).Premature mortality, increased emergency room and hospital admissions, as well as respiratory and cardiovascular disease exacerbations have been linked to ambient pollution exposure worldwide (Halonen et al., 2009;Mallone et al., 2011;Perez et al., 2012).Our prior subchronic inhalation studies using concentrated ambient PM 2.5 found acute and persistent changes in heart rate (HR) (Hwang et al., 2005) and heart rate variability (HRV) (Chen and Hwang 2005), enhancement of aortic plaque size (Chen and Nadziejko, 2005;Sun et al., 2005), and changes in brain cell distribution and function (Veronesi et al., 2005).
A recent analysis of ambient air PM data identified Beijing as among the most polluted cities in the world (WHO Database, 2016).Annually, Beijing's air pollution problem usually worsens mid-September with the onset of autumn and winter, as demand for heating increases and coal-fired power plants increase their production.The concerns about the health of athletes and international visitors to the 2008 Olympic and Paralympic Games in Beijing, China, (held August 8-24 and September 6-16, respectively) prompted the Beijing municipal government to mitigate the ambient air pollution by relocating, limiting, or temporarily closing highly polluting, energy-intensive facilities in and around the city.Additionally, motor vehicle usage was reduced by 60% by restricting operation to only odd or even days (Wang et al., 2009).Proven effective, these drastic anti-pollution measures have been employed in past decade to produce "clear skies" for the foreign state officials' visits and, most recently, the party congress (Zhou, 2017).
These Olympics-related air quality interventions, albeit temporary, encouraged numerous investigations on air pollution and its biological effects before, during and after the Games, and provided a unique opportunity to assess the effect of the temporary reduction in PM 2.5 on cardiovascular responses.Changes in air pollution levels were shown to be associated with acute changes in biomarkers of inflammation and thrombosis (Rich et al., 2012), inflammatory and oxidative stress biomarkers (Huang et al., 2012) in healthy young adults, heart rate variability in young healthy taxi drivers (Wu et al., 2011) and healthy elderly adults (Jia et al., 2011), as well as a decrease in outpatient visits for asthma in adults (Li et al., 2010).Furthermore, this temporary improved air quality showed a beneficial effect on decreasing the systemic inflammation in asthmatic patients for post 2-month follow-up (Gao et al., 2017).In our parallel study, Xu et al. (2012) also reported reduced overall inflammatory response in exposed mice.However, none of these studies had identified the source profiles that are most important in eliciting these adverse health effects.
In several past studies, source-apportionment has been used to control for specific influences other than PM 2.5 on the heterogeneity of responses (Ito et al., 2004).Temporal and spatial variability of PM 2.5 appear to account for specific inter-regional differences (Bell et al., 2009).Based on toxicological investigations, biological mechanisms, such as inflammation and oxidative stress have been suggested to account for heterogeneity of response (Brook et al., 2010).Some studies even incorporated source-apportionment analyses into exposure assessments of cardiac function changes in mice (Lippmann et al., 2005), but few such studies have been conducted in Beijing.
To further elucidate the biological mechanisms that account for these effects, several studies have involved animal models, but few have considered source identification and apportionment, as we present here.The purpose of this study was to test the hypothesis that alterations in cardiac parameters are closely related to changes in both ambient PM concentration and composition as a result of traffic and other source reductions during the 2008 Olympic Games.If PM 2.5 toxicity could be determined based on specific sources, the regulation of PM pollution could be more effective.

Inhalation Exposure
The inhalation exposure facility was located on the Peking University Health Science Center campus (39°58ʹ52ʺN, 116°20ʹ56ʺE), which is 5 km west of the main Olympics venues.The building was a one-story, flat-roof structure, with the air intake inlet 10 m above ground.The incoming air was passed through the PM 2.5 cyclone and then split into two separate streams into 1 m 3 stainless steel chambers at a flow rate of 150 liters per minute for: 1) the animal chambers with PM 2.5 exposure; and 2) filtered air for the control group.Additional information on the exposure and sample collection for the markers of inflammation is in Sun et al. (2012).Particle collection experimental design and evaluation are described in Maciejczyk et al. (2005).Although the traffic control measures were implemented from July 1 till September 20, 2008, in the metropolitan area of Beijing, before and during the Olympics games, we did not start our study until July 24 due to delays in transporting equipment and animal acquisition.The study was conducted July 24 through December 10, 2008.
We used C57/B6 male mice, 3-month old at baseline, fed with normal chow (4% fat, w/w, 8% calories from fat), implanted with Data Sciences International (DSI) telemetry transmitters.Control and exposure animal groups (n = 8 each) were housed in adjacent chambers 24 hr/d, 7 d/wk during the study period.Both groups were exposed to filtered air only prior to the Olympics, during July 24-29, to establish the telemetry outcome baseline.
The outdoor monitoring included semi-continuous measurements of ambient black carbon (BC) (as a surrogate index of elemental carbon (EC) concentrations) recorded every 5 min throughout the study using an Aethalometer (Thermo Electron).We previously reported a good correlation between the BC concentrations measured by Aethalometer and EC measured by Sunset Lab on quartz filters (r 2 = 0.67), with Aethalometer's concentration being 2.8 times higher than those of EC as measured by Sunset Lab (Chen et al., 2010).
The PM 2.5 concentrations in the exposure chamber were measured continuously with a Personal DataRAM.Daily 24-hr PM 2.5 samples were collected on Teflon filters (GelmanTeflo, 37 mm, 0.2 µm pore) for gravimetric and elemental analyses.Filter samples were stored at constant temperature and relative humidity (21 ± 0.5°C, 40 ± 5% RH) until analyzed.The description for gravimetric and elemental analysis via energy dispersive X-ray fluorescence, ED-XRF, of all PM samples is provided in Maciejczyk et al. (2005).Briefly, filter masses were measured on a microbalance (model MT5, Mettler-Toledo Inc., Highstown, NJ) before and after sampling.Samples were then analyzed for 33 elements by nondestructive XRF (model EX-6600-AF, Jordan Valley) using five secondary fluorescers (Si, Ti, Fe, Ge, and Mo), and spectral software XRF2000v3.1 (U.S. EPA and ManTech Environmental Technology, Inc.) Concentrations of elements above detection limit, defined as 3 times of the uncertainty of the measurements (3σ), are reported in Table 1.
The NO x concentrations reported by a central monitoring station were obtained through a Central Monitoring website, and its average 24-hr concentration was used in source apportionment.

Measurement of Cardiovascular Responses
Implanted DSI telemetry transmitters were used to collect data for 30 sec out of every 5 minute interval from  Hwang et al. (2005), that we previously applied to determine changes in cardio function variables in mice that were exposed to concentrated air pollution in New York (Chen and Hwang 2005;Lippmann et al., 2005).The average HR and five HRV parameters were calculated for each mouse in each 5-minute interval.These parameters were SDNN (standard deviation in the occurrence of two consecutive R waves in the ECG waveform), RMSSD (the square root of the mean squared differences of successive RR intervals), LF (low frequency of the beats), HF (high frequency of the beats) and LF/HF ratio.

Source Identification and Apportionment
Several multivariate methods have been developed and are widely used in source apportionment of air pollution for the purpose of epidemiology (Thurston et al., 2005;Hopke et al., 2006;Heal et al., 2012).Traditional Factor Analysis (FA) and Principal Component Analysis (PCA) are accepted useful methods for identifying source components contributing to the particulate matter mass with no a priori knowledge of the number of sources or source profiles (Thurston et al., 2011).Typically, factor analysis starts from a correlation matrix of input variables (here, trace element concentrations), then calculates the loadings of each observed variable on underlying common shared factors (e.g., pollution source categories), allowing computation of corresponding factor scores that are related to factor (source category) impacts on each day.In this study, ambient daily concentrations of Na, Al, Si, S, Cl, K, Ca, Ti, V, Mn, Fe, Ni, Cu, Zn, As, Se, Br, Pb, black carbon (BC) and gaseous oxides of nitrogen (NO x ) were used as independent variables for PM 2.5 factor analysis modeling using SAS version 9.2 (SAS Institute Inc., Cary, NC).The selection of species was based on existing knowledge of elemental tracers as well as XRF detectability.
Following FA, fine mass was apportioned to each of these identified sources through a mass regression model, Absolute Principal Component Analysis (APCA) (Thurston and Spengler, 1985).APCA uses factor-scoring analysis to score an "extra" day wherein all the concentrations of input variables are zero.These "zero pollution" day scores are then subtracted from observed factor scores (derived from prior step) to estimate absolute PC scores (APCS) for each day for each factor.This step is actually a process of inverse standardization because the initial scoring coefficients follow normal distribution.In the next step, regressing daily fine particles mass on the above APCS gives estimates of the coefficients which convert the APCS into mass contribution from each pollution source, as follows: where M k is the daily fine particles mass (in µg m -3 ) during observation k; APCS jk is the estimated absolute PC score; β j APCS jk is the mass contribution on observation k (in µg m -3 ) made by the pollution sources identified with component j.
As the result of the combined FA and APCA, we identified fine particles pollution sources in Beijing during this period, and daily mass contribution estimates for each of the FA identified source categories to the PM 2.5 .

Statistical Analysis
For each physiological outcome of HR, SDNN, RMSSD, LF, HF and LF/HF, the daily averages of each mouse at selected time periods 9:00-13:00, 13:00-17:00, 17:00-21:00, 21:00-1:00, 1:00-5:00, and 5:00-9:00 were calculated, separately, based on the recorded 5-min physiological outcome measurements at each time period.For each time period, the mean of the daily average outcome of an animal across July 24-29 was calculated and treated as the outcome baseline at that time period for the mouse.For each mouse, the outcome baseline is subtracted from all the daily average outcomes at the same time period from July 30 to the end of the study.These daily baseline subtracted outcomes, called "daily outcomes" hereafter, were the data for statistical modeling.
In notation, let y ijt be the t th daily outcome at a time period for the j th mouse in the i th group, where i = 1 (control) and 2 (exposure), j = 1, …, 8 for the control group and j = 1, …, 7 for the exposure group, and t = 1, …, 99 corresponding to July 30-November 5. Note that one mouse in the exposure group died in mid-August and is therefore excluded from the effect modeling.
For each time period, the daily outcomes were fitted to the mixed-effects model: where a ij is normally distributed with mean 0 and a constant variance representing outcome deviation of the j th mouse in the i th group from the overall mean outcome due to sampling effects, and the error term ε itj was assumed an autoregressive process of order 1 for all the mice to account for the autocorrelations of repeated measurements from each same animal.The exposed mean-subtracted concentration from the k th source at lagged h days of the t th day was denoted by C k,t-h .The polynomial function, f(t), was used to model overall trend of the outcomes among all the mice across the 99 days of exposure experiment.The polynomial function, g(t), represented chronic exposure effect patterns across the 99 days of experiment.The parameters β k0 , β k1 , β k2 were the acute effects due to exposure on current day, prior day and two days from the k th pollution source, respectively.Model selection method of Akaike information criterion (AIC) was applied to determine the degrees of the polynomial functions.The degrees of f(t) are first determined using the daily outcomes of control group in the mixedeffects model without exposure terms.The degrees of g(t) are then determined in the mixed-effects model with the chosen degrees of f(t).

PM Mass and Elemental Concentrations
In total, 134 PM 2.5 samples with 18 variables were used for modeling.The mean concentrations of elements with standard deviations, and total mass are shown in Table 1.The mean PM 2.5 mass concentration was 79.1 ± 59.2 µg m -3 , where the large standard deviation indicated high temporal variability ranging from 5.4 to 334 µg m -3 .These levels were compatible to those reported by others, during Olympic 45.2 ± 27.0, after Olympics, 80.4 ± 72.5 (Wu et al., 2010) indicating minimum particles loss in the exposure system.As shown in Table 1, the largest elemental contributors to the PM 2.5 mass were S and Si, followed by Cl and K.The sum of the elements that were measured accounted for 19% of PM 2.5 mass.Reconstructed mass was computed from the elemental data for comparison with the measured mass concentrations for the same days.This was accomplished by using the following equations that account for the sum of typical PM components observed in the ambient atmosphere: ammonium sulfate, soil, and black carbon (Malm et al., 1994;IMPROVE, 2000): As expected, due to the lack of nitrate and organic carbon concentrations, and the possibility of unaccounted mass not represented by the equations, the reconstructed average mass concentrations are lower than those that were measured: 31.5 µg m -3 for PM 2.5 (accounting for 45.6% of measured mass).Although we measured gaseous NO x data, we chose not to include it into the reconstructed mass calculations due to insufficient evidence of complete conversion of NO x into ammonium nitrate in the Beijing environment.In fact, it has been noted that while the concentration of NO 2 has been decreasing in 2000-2008, nitrate concentrations have increased (Lang et al., 2017).Conservatively, using previously measured nitrate values 10-14 µg m -3 for 2003-2007 (Lang et al., 2017) in our analysis, the nitrate would have accounted for 14-19% of PM 2.5 mass.Our sampling also did not include organic carbon (OC), which is commonly attributed to both primary emissions and secondary photochemical conversions of VOC, both mainly from motor vehicles (Zheng et al., 2015).An extensive review of Beijing data by Lang et al. (2017) reports that the concentration of OC had almost no change from 2000 till 2006, with an average concentration reported at around 23.0 µg m -3 , which then began an annual decline of 1.0 µg m -3 per year from 2006 till 2016.In this estimate, OC would have accounted for 28% of PM 2.5 mass.The difference between the actual measured PM 2.5 and the reconstructed mass would therefore include ammonium ion, nitrate, organic carbon particulates, and any other PM component (such as biological aerosol) that had not been accounted for.
To investigate the effect of the pollution mitigation actions implemented during the Olympics, we separated the elemental concentrations by date (before and after September 20, when the Olympics games and the traffic control in Beijing ended).The elemental ratios are shown in Table 1 and are discussed in source apportionment section.

Sources of Beijing PM 2.5 Aerosol-Factor Analysis
We used factor analysis with Varimax rotation to identify four sources for PM 2.5 .The factor loadings for each of the source categories are shown in Fig. 1.We let a pattern be limited to those variables with 40% or more of their variation involved in a pattern.Factors were interpreted by comparing the factor loadings for all factors and variables.In our modeling, we also considered the communalities for elements, which indicate the proportion of a variable's total variation that is involved in the patterns, and percent total variance for each factor.The contributions of each source to overall mass were computed by each factor score regression onto daily mass, and the averages of daily mass contributions per source are reported with each source description.We also further separated these mass contributions by pollution mitigation date and looked for the significance of the difference (shown in Fig. 2).
Four source-related factors were identified by the FA for PM 2.5 , and they explained 87.7% of total variance of the elemental data set, and 84.1% of mass.The communalities for most species were high (> 0.80), indicating that these element concentrations can be predicted from identified sources, and the four factors identified were satisfactory.
The most explanatory factor in our FA, the first factor source, had the largest mean mass contribution 20.1 µg m -3 , or 30.5% of total mass.This factor explained most of the total variance (39.7%) and was heavily loaded with Al, Ti, Ca and Fe, typical elements in soil.However, we suspect a strong contribution of traffic tailpipe and road dust emissions due to co-presence on this factor of BC and NO x from auto exhaust, Mn and Zn from tire wear, and Cu and Zn from engine oil lubricant and auto exhaust (Pacyna, 1986;Madanhire and Mbohwa, 2016).Thus, this source was classified as traffic re-suspended soil source category.Atmospheric Pb and Br historically have been attributed to traffic emissions as these elements were bound to fuel additives.Despite the worldwide abatement measures for fuel additives, both Pb and Br are still present in both urban and rural environments around the world.For example, Pb and Br in the daily TSP and PM 10 samples from 1998-2000 collected in Germany were correlated, however not strongly with r(Pb, Br) = 0.56 (Lammel et al., 2002).Similarly, r = 0.61 was found in rural NY samples (Maciejczyk and Chen 2005;Maciejczyk et al., 2010), and Jeddah, Saudi Arabia, PM 2.5 r(Pb, Br) = 0.56 (Khodeir et al., 2012).Here, r(Pb, Br) of 0.93 clearly indicated that although Pb has been banned in gasoline in China since 1997, the levels of lead in the street dust of the urban areas can still reflect the significant degree of lead contamination from the past.Additional vehicular Pb emissions are also caused by engine wear (Smichowskiet et al., 2008).Since traffic resuspended soil is a local source, pollution mitigation actions in Beijing had the highest impact to this factor.Elements associated with this source category had among the highest ratios of concentrations before and after the Olympics, as shown in Table 1.The average daily mass   contribution from traffic re-suspended soil was 12.4 µg m -3 before September 20, and 29.8 µg m -3 after (p < 0.001 between the two groups), as shown in Fig. 2. The second PM 2.5 source-related category factor, rich in Cl, As, Br, Se, K and metals Pb, Zn, Cu, presents as a mixture of several industrial combustion sources-coal combustion, biomass burning, industrial processes and incineration.We did not have sufficient data to permit further division of this source category grouping.Mass contribution from this source was 16.8 µg m -3 (or 25.6% of PM 2.5 mass), and explained 27.7% of total variance.Previous apportionment of annual data in Beijing by Yu et al. (2013) identified fossil fuel combustion inclusive of coal and oil, based on tracers Cl, Se, V, Ni, As and Pb.In our source apportionment, we were able to separate oil combustion as Factor 3, as discussed later.Here, the second source category factor apparently includes coal combustion based on presence of Se, a tracer for local coal firing (Ondov and Wexler, 1998); and Cl, an elemental tracer for coal combustion in Beijing previously identified by Duan et al. (2006), andYu et al. (2013).Elevated Pb, Se and As were also previously identified near a coal fired thermal power plant (Jaysekher, 2009), and are highly enriched in coal emissions (NAS, 1980).Elements Zn, Pb and K are traditional markers for incinerators (Gordon, 1988) when present, and Cl in fine particulates possibly comes from burning plastics (Maciejczyk et al., 2012).Additionally, K was attributed to biomass burning (Yu et al., 2013).Interestingly, while the means of samples grouped before and after September 20 looked markedly different (11.4 and 23.6 µg m -3 ), the overall difference was only marginally significant (p = 0.0662) as shown in Fig. 2.
The third PM 2.5 factor was rich in V and Ni, and was assigned as emissions of oil combustion (Gordon, 1988) from the local oil-fired power plants, as well as home and commercial building boilers.Mass contribution from this source was 4.5 µg m -3 (or 6.8% of PM 2.5 mass), and explained 10.8% of total variance.Overall, this source contribution had little temporal variation.The differences in before and after September 20 ratios of about 2 (Table 1) are non-significant since the standard deviations are large.The average Ni concentration (7.1 ng m -3 ) was much higher than the concentration of V (2.4 ng m -3 ), consistent with the report by Peltier and Lippmann (2010) that attributed a high Ni/V ratio to emissions from residential and commercial buildings oil boilers within the northern areas of New York City.Here, the ratio of Ni/V was on average 1.4, but with no identified seasonal pattern since we did not have a full seasonal data set.In fact, the PM 2.5 contribution of this source factor was remarkably consistent over time and the difference of average mass contribution before and after the Sept. 20 was non-significant, as shown on Fig. 2.
We assigned factor rich in S and Se to secondary sulfur aerosol.Sulfur is emitted from industrial coal burning and coal-fired power plants as sulfur dioxide (SO 2 ), which undergoes the photochemical oxidation and conversion to sulfate secondary aerosol over long-range regional transport.Comparison of particulate data at a two Beijing sites in the summer of 2006 by Guo et al. (2010) found that almost 90% of fine sulfates were from regional contributions.As stated earlier, Se is mainly a coal combustion emission, and the average Chinese coal is highly enriched in Se (Zhao et al., 2002, Wang et al., 2010).The local 24-hr sulfur dioxide concentrations did not correlate with either sulfate or selenium, which suggests these elements were largely derived from more distant upwind sources that clearly were not affected by the temporary air quality restrictions.This source category had an average mean contribution to overall mass of 13.9 µg m -3 (or 21.1%).This is comparable with the following 2010 year study by Yu et al. (2013) that measured annual secondary sulfate contribution of 13.8 µg m -3 (or 26.5% of the annual PM 2.5 mass).Sulfur-rich secondary aerosol generally showed a seasonal increase over the summer months (with p-value < 0.01) when the photochemical activity is the highest in comparison to the post-Sept.20, as shown on Fig. 2. Indeed, within the duration of the study, the top three regional sulfate contributions to mass were during the warmer months of July and August, consistent with their having transported secondary aerosol origins.

Associations of Mass Concentrations and Source Categories with Cardiac Function Parameters
For each of the six health outcome variables (HR, SDNN, RMSSD, LF, HF and LF/HF) we modeled daily outcomes at six-time periods, separately, with daily PM 2.5 concentrations (single pollutant model) and the four source scores (multiple pollutants model).In total, we have fitted 72 models (6 outcomes for 6-time periods and 2 types of pollutant variables).In each model, we were interested in the acute effects of current day exposure, delayed acute effects one or two days after exposure, and trends for chronic exposure.
For the examination of acute PM 2.5 effects on the 6 health outcomes, we have obtained 108 acute effects estimates (6 outcomes × 6-time periods × 3 time lags).Similarly, the estimated acute effects of the four pollution sources were investigated.We have summarized these estimated acute effects by the total number of significance (p < 0. 05), and results for effects onto HR and SDNN are shown in Figs. 3, 4, and 5.
In comparison to the overall small negative effect of PM 2.5 on HR across all lag days (Fig. 3), several individual factors had both noticeable positive or negative effect on HR response (Figs. 3 and 4).For example, the largest effect was due to traffic re-suspended soil, both on day 0 (positive response) and day 1 and 2 (negative response).

Oil combustion
Secondary sulfate 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 9:00-13:00 13:00-17:00 17:00-21:00 21:00-1:00 1:00-5:00 5:00-9:00 Oil combustion had a positive correlation on day 0 and 1, and negative on day 2. Secondary sulfate, on the other hand, had a negative response on day 0 and positive on day 1.Overall, we concluded that the changes in HR were better explained by the individual PM source factors than by total PM 2.5 .The opposite was observed when considering the HR variability parameters.As seen on Fig. 5, we have selected SDNN to illustrate this point.The significant changes in SDNN are explained by PM 2.5 across all sampling time periods, while individual factors did not have much significant effect.Similar results were observed for RMSSD, LF, HF and LF/HF (data not shown), where changes in HR variability parameters could not be explained by changes in modeled source factors alone.Overall, although the four pollution source categories had different acute effects on the 6 outcomes at different time periods and time lags, we found only some patterns from the results.For example, similarly to the single pollutant model, most of the significant effects were lagged by 2 days, especially for the oil combustion source category, but not for the secondary sulfate category.Interestingly, traffic re-suspended soil source category showed the largest number (47) of significant coefficients, and also the largest numbers for each cardiac outcome.Regardless of lags, most effects occurred during 13:00-17:00 and then 1:00-5:00 time period.
We also have analyzed the six health outcome variables (HR, SDNN, RMSSD, LF, HF and LF/HF) at six-time periods, separately, for examining chronic trends, especially in relationship to the industry and traffic control measures that were implemented from July 1 till September 20, 2008, in the metropolitan area of Beijing, before and during the Olympic games.These estimated chronic effects were very similar between the two models.The estimated chronic effects on HR trends are summarized in Fig. 6, where the points of crude effects are differences between mean baseline Fig. 6.Effects on heart rate for all exposure day, separated into time intervals.Each circle is an average difference between mean baseline subtracted HR in exposure group and control group; solid line is estimated chronic trend; dashed line is its 95% C.I.
subtracted HR in the exposure group and the control group across the experiment.We find the trends of HR changes caused by the PM 2.5 exposure over days, although not significant at some of the time periods, generally increasing until the end of September, and then followed by steady decline.The estimated chronic effects on SDNN shown in Fig. 7 reveals a decreasing trend from starting exposure to around the end of September, although not significantly.The same effect patterns are found for the other HRV parameters of RMSSD, LF, HF.However, shown in Fig. 8, the effects on LF/HF show a clearly increasing trend till the end of September and slightly decreasing to the end of the experiment.As shown in Fig. 2, the only significant changes in pollution categories were the increase in traffic re-suspended soil as the traffic restrictions were removed, and the decrease in secondary sulfate due to reduced photochemical processes.The effects on changes in HR variability parameters were opposite-after a steady decline through the end of September, there was a moderate increase in HRV outcomes in almost all time intervals.

Possible Mechanisms and Limitations
In this study, we used real-time assessment of the health effects of "real world" exposure to PM 2.5 during and after the Olympic Games.Even with significant air quality improvement in Beijing during the 2008 Olympic Games, we found statistically significant impact of PM 2.5 on cardiac autonomic functions as reflected in changes in HR and HRV parameters.These changes were associated with traffic related components of PM 2.5 with various lag structures.Upon sacrifice at the end of exposure, we also evaluated the relationships between PM 2.5 exposure levels and systemic and tissue inflammation with three major inflammatory cell lineages: neutrophil, monocyte/macrophage, and lymphocyte (Xu et al., 2012).The main findings of that investigation were that increased PM 2.5 exposure was Fig. 7. Effects on log SDNN for all exposure day, separated into time intervals.Each circle is an average difference between mean baseline subtracted log SDNN in exposure group and control group; solid line is estimated chronic trend; dashed line is its 95% C.I. Fig. 8. Effects on LF/HF for all exposure day, separated into time intervals.Each circle is an average difference between mean baseline subtracted LF/HF in exposure group and control group; solid line is estimated chronic trend; dashed line is its 95% C.I. associated with increases of the systemic chemokine levels of MCP-1 and IL-6, and with induced infiltration of macrophages and neutrophils in the adipose tissue, lung, spleen, and thymus.T lymphocytes were also observed in these tissues other than lung in relation to PM 2.5 exposure.Furthermore, air quality improvement during the Olympic Games was significantly associated with reduced overall inflammatory responses in the mice, especially in the lung, visceral adipose tissue, spleen, and thymus.However, whether these systemic inflammatory responses were directly associated with changes in autonomic function described here is not clear.Further studies will be needed to elucidate these mechanisms.
While our study produced important new insights showing effective air pollution control could have significant impact on health effects, there were several limitations.First, using "real world" exposure scenario precludes more comprehensive mechanistic investigation, which requires larger animal cohorts.Second, the distance of the exposure location to the Olympic Games Stadium as well as a single point monitoring may not represent the PM 2.5 characteristic of the entire Beijing metropolitan area.Third, although there was a filtered air control group for each time period that controlled for the co-presence of pollutant gases, the potential for the effects of differences between the during and after periods in seasonal, temperature, bioactive (pollens, mold, etc.) agents, and the possible impacts of the preexposure environment of the animals could not be studied in this project due to technical limitations.

CONCLUSION
In this study, factor analysis and mass regression were used to identify four fine particulate matter sources and estimate their contributions to the ambient air pollution in Beijing.We found that the traffic restrictions imposed during the Olympics did in fact significantly decrease the PM 2.5 mass contribution from the traffic re-suspended soil dust category but did not have much effect on the other three source categories.The significant decrease in secondary sulfate pollution was attributed to the decreased photochemical processes.We correlated the changes in PM 2.5 and individual source categories to the six cardio variables in mice.We observed the chronic effect trend of increase in HR during the cleaner air Olympic period, followed by the decrease as regulations were removed.The effect on the HRV parameters was the opposite.
The estimated mass contributions of each identified source category were then introduced into two models as an exposure variable to see whether the relationship between animal cardiovascular responses and PM varied with the activity of each PM source category.The acute effects of each pollutant were also investigated, including on the two lag days.Although the overall PM mass had a negative effect on HR, interestingly, the traffic related and mixed industrial categories increased HR on the day of exposure but had significant negative contributions on day lag 1 and 2, whereas oil combustion and secondary sulfate had a positive effect on day lag 2.
Clearly, these effects on cardio health parameters occurred during the exposure and had different temporal lags with regard to the PM and its varying composition.Although we have not investigated the biological mechanism underlying these effects, our study further emphasizes the importance of looking into correlations between sources of pollution and health outcomes.

Fig. 4 .
Fig. 4. Daily averaged beta values for HR for each factor by lag day.The 4-hour response values were averaged for each day, weighted by -log(p).

Fig. 5 .
Fig. 5. Significance of the estimated acute effects of pollutants onto HR variability parameter SDNN.The -log(p value) was averaged by time intervals and day lags (0, 1, or 2).The dashed line is the threshold of significance p = 0.05.

Table 1 .
Daily 24-hr concentrations (ng m -3 ) for the time periods of before and after Sept 20.