Blood Pressure Changes and Chemical Constituents of Particulate Air Pollution: Results from the Healthy Volunteer Natural Relocation (HVNR) Study

Background: Elevated blood pressure (BP) has been associated with particulate matter (PM) air pollution, but associations with PM chemical constituents are still uncertain. Objectives: We investigated associations of BP with various chemical constituents of fine PM (PM2.5) during 460 repeated visits among a panel of 39 university students. Methods: Resting BP was measured using standardized methods before and after the university students relocated from a suburban campus to an urban campus with different air pollution contents in Beijing, China. Air pollution data were obtained from central monitors close to student residences. We used mixed-effects models to estimate associations of various PM2.5 constituents with systolic BP (SBP), diastolic BP (DBP), and pulse pressure. Results: An interquartile range increase of 51.2 μg/m3 in PM2.5 was associated with a 1.08-mmHg (95% CI: 0.17, 1.99) increase in SBP and a 0.96-mmHg (95% CI: 0.31, 1.61) increase in DBP on the following day. A subset of PM2.5 constituents, including carbonaceous fractions (organic carbon and elemental carbon), ions (chloride and fluoride), and metals/metalloid elements (nickel, zinc, magnesium, lead, and arsenic), were found to have robust positive associations with different BP variables, though robust negative associations of manganese, chromium, and molybdenum with SBP or DBP also were observed. Conclusions: Our results support relationships between specific PM2.5 constituents and BP. These findings have potential implications for the development of pollution abatement strategies that maximize public health benefits.


Research
Both elevated blood pressure (BP) and ambient particu late air pollu tion have been associated with increased cardio vascular morbidity and mortality (Blacher et al. 2000;Brook et al. 2010;Guo et al. 2010;Kan et al. 2007). An elevation in BP may be an important physio logi cal mechanism linking particu late air pollu tion and adverse cardio vascular outcomes (Brook and Rajagopalan 2009). Among the size fractions of ambient particu late matter (PM), PM with an aerodynamic diameter ≤ 2.5 µm (PM 2.5 ) has been associated with adverse cardio vascular effects (Brook et al. 2010;Kan et al. 2007). Controlled human and animal experiments have verified that exposure to PM 2.5 is capable of inducing elevated BP Chang et al. 2004;Urch et al. 2005;Zanobetti et al. 2004).
Ambient PM is a mixture of various chemical constitu ents, including carbonaceous fractions [organic carbon/elemental carbon (OC/EC)], ions, and transition metals. These constitu ents may have different effects on the cardio vascular system (Brook et al. 2010). Although several PM chemical constitu ents (e.g., carbonaceous fractions) have been associated with pro hyper tensive effects in different populations (Mordukhovich et al. 2009;Urch et al. 2005;Zanobetti et al. 2004), evidence for effects of specific PM chemical constitu ents on BP is still lacking. Specifically, trafficrelated PM may play a distinctive role in cardio vascular responses (Auchincloss et al. 2008;Delfino et al. 2010;Jia et al. 2012;Wu et al. 2010Wu et al. , 2011aWu et al. , 2011b. Thus, a hypothesis that a group of PM chemical constitu ents and related sources may confer greater PM cardio vascular toxicity is reasonable based on the existing literature. We conducted the Healthy Volunteer Natural Relocation (HVNR) study to examine the relation ship between various PM 2.5 chemical constitu ents and BP changes in a panel of healthy male university students in Beijing, China, before and after their relocation from a suburban campus to an urban campus with different PM air pollu tion constitu ents. We hypothesized that this relocation would substantially change the participants' exposures to ambient PM 2.5 and chemical constitu ents associated with local pollu tion sources, and thus facilitate an analy sis of relationships between PM 2.5 chemical constitu ents and BP in the study population.

Material and Methods
Study design. Beijing City covers an area of 16,410 km 2 , has nearly 20,000,000 inhabi tants, and is about 160 km from the near est coastline. More than 5,000,000 vehicles are its main source of urban air pollu tion. Our study population consisted of a panel of 41 male undergraduate college students from a university in Beijing [Beijing Institute of Technology (BIT)]. The BIT has two cam puses located in different areas of Beijing ( Figure 1). Study participants completed their first 2 years of undergraduate study (from autumn 2008 to summer 2010) at the BIT Liangxiang campus, which is located in a sub urban area (Fangshan District), then moved to the BIT main campus, which is located in an urban area (Haidian District) for their next 2 years of study (from autumn 2010 to summer 2012). The BIT Liangxiang campus is about 2 km from the nearest freeway. There were several active construction sites within 2 km of the campus during the study, in addi tion to some industrial facilities located within several kilometers of the campus. In contrast, the BIT main campus is located in the Beijing downtown area, along the northwest inner side of the third ring road that circles the city. There were no substantive construction activi ties or industrial facilities near the main cam pus during the study. We used the following inclusion criteria to select participants before the study began: male with a geographical origin other than Beijing, nonobese, no his tory of smoking, and without pulmonary, cardio vascular, and other chronic diseases. We used a selfadministered questionnaire to collect personal information, including name, age, and medical history/health status. We scheduled 12 biweekly study visits for each participant over the entire study, including 4 visits during each of the following three time periods: suburban period (22 April to 20 June 2010) at the BIT Liangxiang campus and urban period 1 (3 September to 8 November 2010) and urban period 2 (10 April to 12 June 2011) at the BIT main campus. The study was approved by the Institutional Review Board of Peking University Health Science Center, and informed consent was provided by each participant before the study began.
BP measurement. Study visits were sched uled between 1300 and 1500 hours on week days to minimize diurnal variation in BP. A trained technician in the hospital affiliated with the BIT Liangxiang campus or BIT main cam pus performed standardized resting BP mea surements during each study visit. Participants rested in a sitting position in a quiet room for at least 10 min before upper arm BP was measured using an Omron 705IT automated oscillometric monitor (HEM759E; Omron Healthcare Co. Ltd., Kyoto, Japan) at least three times with a 1min minimum interval between measurements. In most cases the sec ond and third sets of readings were averaged to calculate systolic BP (SBP) and diastolic BP (DBP) (Rioux et al. 2010). However, if the dif ference between SBP or DBP values of the sec ond and third measurements was > 5 mmHg, the BP was considered unstable, and another 1 to 3 measurements would be taken until the difference between the last two measurements was ≤ 5 mmHg. Under this condition, all read ings (from the second to the last measurement) within a 5mmHg range of difference were averaged to calculate the final BP values. Pulse pressure (PP) was calculated as the difference between the average SBP and DBP values. Weight and height were measured during the first study visit in each time period, and body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. Two volunteers who did not complete study visits after relocation to the urban campus were excluded from data analy sis, leaving a total of 39 participants.
Environmental data. Air pollu tion and weather data were measured using standard methods and quality controls at a central mon itoring site on the BIT Liangxiang campus (suburban period) or the BIT main campus (urban periods 1 and 2). The BIT Liangxiang monitoring site was on the roof of a three story building (about 10 m high) without any nearby structures that would obstruct air flow, and the BIT main campus monitoring site was on the roof of a fivestory building (about 15 m high) located within 200 m of the third ring road. The instruments and materials used for air monitoring included SKC sampling systems for PM 2.5 mass collection on Teflon filters and quartzfiber filters (SKC Inc., Eighty Four, PA, USA) and a digital dust monitor for realtime PM 2.5 concentration measurement (LD3K; Sibata Scientific Technology Inc., Tokyo, Japan); a model T15n enhanced carbon mon oxide (CO) measurer for realtime CO con centration measurement (Langan Products Inc., San Francisco, CA, USA); Ogawa pas sive samplers for nitrogen oxides and nitrogen dioxide (NO x and NO 2 , respectively) collec tion on cellulose fiber filters (Ogawa Air Inc., Osaka, Japan); and a HOBO Pro V2 logger for temperature and relative humidity mea surements (Onset Corp., Pocasset, MA, USA). Data on PM with an aerodynamic diameter of ≤ 10 µm (PM 10 ) were obtained from the near est city air monitoring stations (within 5 km of each campus) under the supervision of the Beijing Municipal Environmental Protection Bureau. Concentrations of coarse PM with an aerodynamic diameter of between 2.5 and 10 µm (PM 2.5-10 ) were calculated as the difference between the measured PM 10 and PM 2.5 concentrations.
Daily PM 2.5 mass concentrations were determined by standard weighing procedures before and after the sample collection . The PM 2.5 filters were analyzed in the laboratory for the following chemical constitu ents: OC and EC in quartzfiber fil ters by thermo/optical transmission method (Lab OC/EC analyzer; Sunset Laboratory Inc., Tigard, OR, USA); sulfate, nitrate, chloride, and fluoride in Teflon filters by ion chroma tography (model ICS2000; Dionex Corp., Sunnyvale, CA, USA); aluminum, calcium, sodium, potassium, magnesium, iron, and zinc in Teflon filters by inductively coupled plasma atomic emission spectrometry (model SPS8000; KCHG Co. Ltd., Beijing, China); and strontium, barium, lead, copper, titanium, nickel, molybdenum, cadmium, vanadium, chromium, manganese, arsenic, selenium, stannum, and antimony in Teflon filters by inductively coupled plasma mass spectrom etry (model ELAN DRC II; PerkinElmer Inc., Shelton, CT, USA). We also estimated the concentrations of three additional carbona ceous fractions [primary OC (POC), second ary OC (SOC), and particu late organic matter (POM)], as described in detail in Supplemental Material, p. 2 (http://dx.doi.org/10.1289/ ehp.1104812). PM 2.5 constitu ents were classi fied according to their chemical nature as car bonaceous fractions, negative ions, transition metals, crustal metals, or other metals/metal loid elements. NO x and NO 2 were collected on cellulose fiber filters and concentrations were determined using a spectrophotometer follow ing the manufacturer's specifications (Ogawa & Company USA, Inc., Pompano Beach, FL, USA). Nitric oxide (NO) concentrations were calculated as the difference between the NO x and NO 2 concentrations.
Statistical analy sis. We first used paired ttests to compare the mean BP changes between periods by subject, and then we used mixedeffects regression models in SAS ver sion 9.2 (SAS Institute Inc., Cary, NC, USA) to estimate associations between exposure varia bles and BP. Environmental data were matched with BP data for each subject before analy sis. The mixedeffects models included a  random intercept for each subject to account for withinsubject correlations due to repeated measurements. Base models included individ ual air pollu tants or PM 2.5 constitu ents, and were adjusted for age, BMI, temperature, and relative humidity as continuous varia bles, with linear and quadratic terms for temperature and relative humidity (Mordukhovich et al. 2009), and also adjusted for season, month, dayofweek, hourofday, and study site as binary or cate gorical varia bles. In addition, we included a dayofstudy varia ble and a squared dayofstudy varia ble in the models to account for secular trends in associations between air pollu tion and BP (Penttinen et al. 2001).
We used three kinds of models to investi gate the associations between exposure varia bles and BP after combining the data over the three time periods. First, we modeled individual air pollu tants or PM 2.5 constitu ents to estimate associations with BP, with adjustment for the potential confounders listed above. Second, we estimated associa tions of individual PM 2.5 constitu ents and BP with adjustment for total PM 2.5 . Third, we used a constitu ent residual model analy sis to address collinearity between total PM 2.5 and PM 2.5 constitu ents (Wu et al. 2011a). Specifically, we regressed daily concentrations of each PM 2.5 constitu ent on total PM 2.5 concentrations using a separate linear regres sion model for each time period to generate a constitu ent residual for each daily concentra tion value of the constitu ent. The constitu ent residual represented the proportion of the constitu ent that is uncorrelated with total PM 2.5 and therefore can be interpreted as a measure of the independent contribution of each constitu ent to associations with BP.
To estimate the cumulative effects of expo sure, we modeled the mean concentrations of exposure varia bles during the preceding 1-5 days before BP measurement (Hoffmann et al. 2012). We reported results using mean concentrations during the preceding 1-3 days because most associations with BP were observed with exposures during this time period. Results are expressed as absolute changes [in millimeters mercury (mmHg)] with 95% CIs for the BP varia bles associated with interquartile range (IQR) increases in air pollu tants and PM 2.5 constitu ents. The statis tical significant level was defined as p < 0.05 (two sided).

Results
The mean (range) age of the eligible study sub jects (n = 39) was 20.1 years (19-22 years), and their mean (range) BMI was 21.2 kg/m 2 (17.2-24.9 kg/m 2 ). Overall, 34 subjects completed all 12 biweekly visits, 4 completed 11 visits, and 1 completed 8 visits, resulting in a total of 460 visits. SBP and PP increased over the three time periods, whereas DBP levels remained rela tively stable (Table 1).
Most air pollu tants and PM 2.5 constitu ents showed substantial variation over the three time periods (Table 2). In particular, concentrations of gaseous air pollu tants (e.g., CO, NO x , and NO 2 ) and levels of several PM 2.5 carbonaceous fractions related to traffic (OC, EC, POC, and POM based on concen trations or proportions of PM 2.5 mass) were higher during the urban periods than in the suburban period.  SBP and PP levels during the urban periods were significantly higher than during the suburban period, and SBP and PP levels during the urban period 2 were also significantly higher than those during the urban period 1 (Table 3). There was no significant difference in DBP levels between any two periods. IQR increases in major air pollu tant concentrations (PM fractions, NO x , and NO 2 ) during the 1-3 days before study visits showed significant positive associations with SBP or DBP, with more consistent associations observed with DBP than SBP [see Supplemental Material, Table S1 (http://dx.doi.org/10.1289/ehp.1104812)].
An IQR increase (51.2 µg/m 3 ) in PM 2.5 during the preceding day was associated with a 1.08mmHg (95% CI: 0.17, 1.99) increase in SBP and a 0.96mmHg (95% CI: 0.31, 1.61) increase in DBP. There were no significant associations between the major air pollu tants and PP (data not shown).
We found significant associations between SBP or DBP and concentrations of several PM 2.5 constitu ents during the preceding day, though some associations that were statistically significant based on singleconstitu ent models were not significant after adjustment for PM 2.5 or based on constitu ent residual models (Figures 2-4). However, all three model estimates indicated significant positive associations between SBP and chloride, nickel, and strontium ( Figure 2); between DBP and OC, EC, POC, POM, chloride, fluoride, and lead ( Figure 3); and between PP and nickel and magnesium (Figure 4). In addition, all three model estimates indicated significant negative associations between SBP and manganese, chromium, and molybdenum, and between DBP and chromium and molybdenum. We also found significant associations between SBP and zinc during the preceding 3 days, and between PP and arsenic during the preceding 3 days based on three different models (data available upon request).

Discussion
We evaluated relationships between BP and various air pollu tants and PM 2.5 chemical constitu ents under natural exposure settings using a panel study design that repeatedly measured resting BP during three time peri ods with different ambient air pollu tion expo sures. The relocation of the study population from a suburban campus to an urban campus, which was a normal part of their university education, gave us the opportunity to study associations with specific PM 2.5 constitu ents that may have distinct cardio vascular effects ). We observed increases in SBP and PP after relocation from the sub urban campus to the urban campus located in a megacity with high air pollu tion levels, and we estimated consistent associations between BP measures and a subset of PM 2.5 chemical constitu ents.
Major sources of particu late air pollu tion in Beijing include road dust, motor vehicle exhaust, industry, incineration, and coal burning (Sun et al. 2004). During recent years, the number of motor vehicles in Beijing has increased rapidly and traffic emissions have become a dominant source of ambient air pollu tion (Stone 2008;Wang et al. 2009;Zhou et al. 2010). Evidence linking trafficrelated air pollu tion and cardio vascular outcomes has been growing, especially in urban areas where traffic emissions are one of the major pollu tion sources Delfino et al. 2010;Jia et al. 2012;Wu et al. 2010Wu et al. , 2011aWu et al. , 2011b. OC and EC (or its surrogate, black carbon) are two commonly used indicators of traffic emissions (Delfino et al. 2010;Sun et al. 2004;Wang et al. 2009). Shortterm exposures to these carbonaceous particles have been associated with pro hypertensive effects in patients with cardio vascular conditions (Delfino et al. 2010;Zanobetti et al. 2004). In our study, we found that PM 2.5 measured in the urban area contained larger proportions of carbonaceous fractions than PM 2.5 measured in the suburban area (Table 2). This suggests a greater contribution of traffic emissions  to the particu late air pollu tion in the urban area. We found robust positive associations between PM 2.5 carbonaceous fractions and DBP in our study participants. When OC was subclassified as POC or SOC, the association appeared to be specific to POC (Figure 3). POC is a representative indicator of particu late organics from fossil fuel combustion sources in the context of traffic related air pollu tion (Delfino et al. 2010). Therefore, our findings suggest that PM carbonaceous fractions related to traffic may play an important role in the pro hypertensive effects of PM air pollu tion. Several PM 2.5 metal constitu ents, includ ing nickel, zinc, magnesium, strontium, lead, and arsenic, had robust positive associations with SBP, DBP, or PP, whereas manganese, chromium, and molybdenum had robust nega tive associations with SBP or DBP. Among these constitu ents, nickel, zinc, manganese, chromium, and lead typically come from industrial emissions, including metallurgi cal processes, but some of these metals (e.g., zinc, nickel, manganese) may also come from traffic emissions (Loranger and Zayed 1995; Sun et al. 2004). Magnesium usually origi nates from mineral aerosols that may result from resuspended road dust and longrange transported dust (Sun et al. 2004), and arsenic is typically generated from coal burning (Xie et al. 2006). Relationships between BP and these metals/metalloid elements, especially the transition metals (e.g., nickel, zinc, manganese, cadmium), have been well demonstrated in previous toxicological studies in vivo (Fiorim et al. 2011;Kasten et al. 1994;Wang et al. 2002;Yanagisawa et al. 2004;Yang et al. 2007). When delivered to the airways, the transition metals could stimulate the produc tion of reactive oxygen species and then induce airway injury and inflammation, which are subsequently followed by a series of cardio pulmonary responses (GonzálezFlecha 2004). Epidemiologic evidence linking metals/metal loid elements and BP has also been growing. For example, chronic exposures to arsenic and lead have been associated with increased BP or higher hypertension prevalence in popula tions (Mordukhovich et al. 2012;NavasAcien et al. 2007). Specifically, we found robust neg ative associations between manganese, chro mium, and molybdenum and SBP or DBP. Environmental manganese is generally consid ered to reduce hypertension risk (Houtman 1996). A previous animal experiment found that infusing manganese into conscious, restrained rats resulted in a decrease in BP (Kasten et al. 1994), and results from a recent epidemiologic study found negative associa tions between chronic manganese exposure and BP in elderly men (Mordukhovich et al. 2012). However, our findings on the negative associations between BP and chromium and molybdenum are not supported by previous experimental and epidemiologic reports, and a causal explanation for these findings may require further investigation.
We found small but robust positive asso ciations between chloride and SBP and DBP, less consistent associations between fluo ride and DBP, and inconsistent associations between sulfate and nitrate and BP. Sulfate and nitrate are typical secondary pollu tants that constitute a significant proportion of the PM 2.5 mass. In urban areas far from coast lines, the major source for airborne chloride may be from burning polyvinylchloride plas tic in refuse dumps, and large amounts of pollu tants such as fluoride may also be emit ted when trash is being smashed or inciner ated (Sun et al. 2004). Experimental studies have demonstrated that chloride was critical in the development of hypertension (Kurtz and Morris 1984;Kurtz et al. 1987;Ziomber et al. 2008). That is, dietary intake of sev eral ions, including sodium, potassium, and chloride, were able to induce increased BP in rats (Ziomber et al. 2008), whereas sodium loading without chloride failed to increase BP in animals (Kurtz and Morris 1984) or men (Kurtz et al. 1987). Nevertheless, evi dence for the relationship between these air borne ions and BP is still rare and requires further investigation.
Associations with the different BP varia bles differed over the study. SBP and PP levels tended to increase over time, whereas DBP levels did not. Among the BP varia bles, PP has been regarded as a stronger predic tor of adverse cardiac outcomes, especially in hypertensive patients (Blacher et al. 2000). Several studies have investigated associations between air pollu tion and BP, but only a few of them have examined PP (Auchincloss et al. 2008;Rioux et al. 2010). Our findings also indicated wide variation among individual study participants, with SBP or PP increasing by > 10 mmHg after relocation to the urban campus in some cases. This suggests that sus ceptibility to the effects of air pollu tants may vary substantially among individuals in the general population.
The study has several strengths in addi tion to the natural relocation study design. Participants were young, healthy volunteers who were nonsmokers and free of any cardio vascular compromises. Therefore, confounding by factors such as age, smoking, disease status, medication use, or obesity was unlikely. We conducted the study in the spring and autumn seasons to avoid significant climate changes that might confound associations between air pollu tants and BP (Adamopoulos et al. 2010). We used constitu ent residual models Figure 3. Changes in DBP associated with IQR increases in PM 2.5 constitu ents at concentration during the preceding day before the BP measurement. Estimates are adjusted for age, BMI, season, month, day-ofstudy, squared day-of-study, day-of-week, hour-of-day, study site, temperature, and relative humidity in linear and quadratic terms. Data are presented as effect estimates ± 95% CIs. For constitu ent-PM 2.5 joint models, we used the main effect estimates of PM 2.5 constitu ents for result presentation.   Single-constituent residual model to estimate associations with PM 2.5 constitu ents, and we adjusted for many potential con founders. However, residual or unmeasured confounding cannot be excluded, and general izability to other populations may be limited. This study has several other limitations. First, we used ambient air pollu tion data from central monitoring sites rather than personal exposure measures. However, all participants lived in school dormitories within 300 m of the monitoring site for each campus, and they spent most of their time in naturally ventilated buildings near the monitoring site. Therefore, air pollution data from central monitoring sites could be used as good surro gates for their real exposures. Second, as in most epidemio logic studies, we were not able to determine whether the observed associations were due to the measured air pollu tion constitu ents or to other factors that varied along with constitu ent concentrations (e.g., pollu tion sources, seasons, locations, or even other pollu tants that might be related or correlated to the measured constitu ents). For example, both seasonal and regional variation in estimated effects of air pollu tion on cardio vascular outcomes have been reported by previous studies Choi et al. 2007). Changes in season or location may result in changes in air pollu tion sources and the constitu ents of air pollu tion, which in turn may influence the effects of air pollu tion on the cardio vascular system. As a result, we were not able to differentiate effects of season or location from potential effects of air pollu tion in the cur rent study. Third, other gaseous air pollu tants that may also contribute to the adverse cardio vascular outcomes (e.g., ozone and sulfur diox ide) were not measured in the present study. Finally, there is a potential confounding effect associated with the progression in university education on BP through changes in the partic ipants' stress levels across different periods that could not be excluded. However, in view that the estimated air pollu tion effects were quite strong after adjusting for various factors related to seasonal and regional factors, we believe this kind of confounding effect would not be able to change our findings materially.

Conclusions
Our findings suggest that specific PM 2.5 chemi cal constitu ents may be associated with BP in healthy adults. These findings also sug gest potential linkages between pollu tion sources and PMrelated cardio vascular effects. A better understanding of the responsible PM constitu ents and their sources could lead to more targeted and effective regulations (Brook et al. 2010). This is especially important for the most polluted regions and countries around the world. As the largest developing country, China is now facing the worst air pollu tion problem in the world along with its rapid economic expansion over the past decades (Kan et al. 2012). Our findings thus may have implications for the development of relevant pollu tion abatement strategies that maximize benefits to public health. Figure 4. Changes in PP associated with IQR increases in PM 2.5 constitu ents at concentration during the preceding day before the BP measurement. Estimates are adjusted for age, BMI, season, month, day-ofstudy, squared day-of-study, day-of-week, hour-of-day, study site, temperature, and relative humidity in linear and quadratic terms. Data are presented as effect estimates ± 95% CIs. For constitu ent-PM 2.5 joint models, we used the main effect estimates of PM 2.5 constitu ents for result presentation.   Single-constituent residual model