Spatiotemporal variability in the oxidative potential of ambient fine particulate matter in the Midwestern United States

We assessed the oxidative potential (OP) of both water-soluble and methanol-soluble fractions of ambient fine particulate matter (PM2.5) in the Midwestern United States. A large set of PM2.5 samples (N = 241) was collected from five sites set up in different environments, i.e., urban, rural, and roadside, in Illinois, Indiana, and Missouri during May 2018–May 2019. Five acellular OP endpoints, including the consumption rate of ascorbic acid and glutathione in a surrogate lung fluid (SLF) (OPAA and OPGSH, respectively), dithiothreitol (DTT) depletion rate (OPDTT), and qOH generation rate in SLF and DTT (OPOH−SLF and OPOH−DTT, respectively), were measured for all PM2.5 samples. PM2.5 mass concentrations in the Midwestern US as obtained from these samples were spatially homogeneously distributed, while most OP endpoints showed significant spatiotemporal heterogeneity. Seasonally, higher activities occurred in summer for most OP endpoints for both waterand methanolsoluble extracts. Spatially, the roadside site showed the highest activities for most OP endpoints in the water-soluble extracts, while only occasional peaks were observed at urban sites in the methanol-soluble OP. Most OP endpoints showed similar spatiotemporal trends between massand volume-normalized activities across different sites and seasons. Comparisons between two solvents (i.e., water and methanol) showed that methanol-soluble OP generally had higher activity levels than corresponding water-soluble OP. Site-to-site comparisons of OP showed stronger correlations for methanol-soluble OP compared to water-soluble OP, indicating a better extraction of water-insoluble redox-active compounds from various emission sources into methanol. We found a weak correlation and inconsistent slope values between PM2.5 mass and most OP endpoints. Moreover, the poor to moderate intercorrelations among different OP endpoints indicate different mechanisms of OP represented by these endpoints and thus demonstrate the rationale for analyzing multiple acellular endpoints for a better and more comprehensive assessment of OP.

The assessment of PM 2.5 -induced oxidative stress is conventionally carried out through biological tests, including both in vitro (Becker et al., 2005;Zhang et al., 2008;Oh et al., 2011;Yan et al., 2016;Abbas et al., 2016;Deng et al., 2013) and in vivo designs (Kleinman et al., 2005;Riva et al., 2011;Pei et al., 2016;Araujo et al., 2008;Xu et al., 2011;Sancini et al., 2014). Although these biological tests are highly relevant in terms of representing health effects in humans, the time-and labor-intensive protocols as well as the cost of experimental materials generally limit their application to only small sample sizes. Various acellular chemical assays which assess the OP by replicating intrinsic biological mechanisms were therefore developed as alternatives. These assays are generally divided into two categories. The OP analysis approaches in the first category directly probe the generation of ROS during redox cycling reactions in the presence of PM, such as the measurement of H 2 O 2 and q OH production in surrogate lung fluid (SLF) (Vidrio et al., 2009;Shen et al., 2011;Charrier et al., 2014;Ma et al., 2015), as well as H 2 O 2 and q OH production in dithiothreitol (DTT) Xiong et al., 2017;Chung et al., 2006;Kumagai et al., 2002). The assays in second category utilize the consumption of antioxidants such as AA (Visentin et al., 2016; and GSH (Künzli et al., 2006;Szigeti et al., 2016) or surrogates for cellular reductants such as DTT (Verma et al., 2014;Cho et al., 2005), as the OP indicator. Analyzing each PM sample for all of these chemical assays is also time-consuming. To address this concern, we have previously developed an automated OP analysis instrument named SAMERA -the Semi-Automated Multi-Endpoint ROS-activity Analyzer -that can measure the five most commonly used OP endpoints (i.e., the consumption rate of AA and GSH in SLF, OP AA , and OP GSH ; the consumption rate of DTT and OP DTT and the generation rate of q OH in SLF, DTT, OP OH−SLF , and OP OH−DTT ) for a PM extract in less than 3 h . Many of these acellular endpoints have been widely implemented by various researchers to assess the oxidative properties of PM. Calas et al. (2018) compared the responses of several OP endpoints (i.e., OP DTT , OP AA , OP GSH , and electron spin resonance -OP ESR ) on PM 10 samples, with N = 98, collected from Chamonix, France. Yang et al. (2014) also used four OP endpoints (OP AA , OP DTT , OP ESR , and reductive acridinium triggering -OP CRAT ) to investigate the effect of different ex-traction solvents and filter types on OP responses using the PM 2.5 samples (N = 20) collected from two cities (Rotterdam and Amsterdam) in the Netherlands. The comparison of OP AA , OP DTT , and OP GSH has been shown in two studies Gao et al., 2020a), both from the southeastern US. We are not aware of any study which has compared q OH generation in SLF or DTT with other endpoints based on antioxidant consumption (e.g., AA or GSH consumption). Clearly, studies systematically comparing the responses of these different endpoints on a large sample set collected from an extensive spatial scale, particularly in the United States, are very limited.
Although OP is proposed as an integrative PM 2.5 property, purportedly combining the individual and synergistic actions of its many active components, there have been limited attempts to integrate it into large-scale epidemiological studies. This is because, unlike other PM properties such as mass, sulfate, and nitrate, OP measurements in different geographical regions have been relatively sparse. Moreover, before integrating OP into epidemiological studies, it is important that we investigate the differences of its spatiotemporal distribution with other commonly measured PM properties such as mass. An understanding of the temporal variation of OP in a specific environment could be helpful in time series studies of short-term effects, while the spatial variation of OP can aid in studying the long-term health effects of PM 2.5 exposure among different regions (Yang et al., 2015a). Globally, spatiotemporal profiles of OP have been characterized for some geographical regions such as the Los Angeles Basin (Saffari et al., 2014(Saffari et al., , 2013, Denver (Zhang et al., 2008), and Atlanta Verma et al., 2014) in the US, Ontario (Canada) (Jeong et al., 2020;S. Weichenthal et al., 2019, France Calas et al., 2019;Weber et al., 2018Weber et al., , 2021, Italy (Cesari et al., 2019;Perrone et al., 2019;Pietrogrande et al., 2018), Athens in Greece (Paraskevopoulou et al., 2019), the Netherlands (Yang et al., 2015a, b), and some coastal cities near Bohai (Jinzhou, Tianjin and Yantai; Liu et al., 2018) and Beijing Liu et al., 2014) in China. Some of these studies have substantially contributed to enhancing our understanding of the role of OP in PM-induced health effects Tuet et al., 2016;Abrams et al., 2017;Yang et al., 2016;Bates et al., 2015). However, despite including many cities ranked high in terms of air pollution (e.g., Indianapolis -Rosenthal et al., 2008;Chicago -Dominici et al., 2003;St. Louis -Sarnat et al., 2015;Detroit -Zhou et al., 2011;Cincinnati -Kaufman et al., 2019;and Cleveland -Kumar et al., 2013), the midwestern region of the United States is an understudied region in terms of assessing the oxidative levels of ambient PM 2.5 .
Here, we investigate the detailed spatiotemporal profiles of ambient PM 2.5 mass concentrations and OP in the Midwestern United States. Simultaneous ambient PM 2.5 samples were collected from five different sites in the Midwestern US. The automated instrument SAMERA facilitated the measure-ment of OP on our large bulk of PM 2.5 samples (N = 241) collected from all the sites, which were extracted in both water and methanol separately. The goal of this analysis is to compare the spatiotemporal distribution of PM 2.5 OP with that of the mass concentrations. We also want to investigate if different measures of OP; i.e., OP AA , OP GSH , OP OH−SLF , OP DTT , and OP OH−DTT show different spatiotemporal trends or are correlated with each other. Correlations of OP with PM chemical composition and source apportionment analysis of PM 2.5 OP will be presented in our subsequent publications. Our paper presents the results from probably one of the most comprehensive OP analysis campaigns so far, combining five different acellular OP endpoints measured on both water-and organic-soluble extracts.
2 Experimental methods

Sampling campaign
Simultaneous sampling in five different sites spread across three states (i.e., Illinois, Indiana, and Missouri) was conducted every week for this project in the Midwestern US. The locations of the sampling sites are shown in Fig. 1. The Champaign (CMP) and Bondville (BON) sites are paired sites representing the urban (roadside) and rural environment of Champaign County, IL, respectively, while three major city sites (i.e., Chicago -CHI, Indianapolis -IND, and St. Louis -STL) are representatives of urban background regions in these respective cities.
CMP is located on top of a parking garage on the campus of the University of Illinois at Urbana-Champaign and is adjacent to a two-lane (both ways) road (i.e., University Avenue). This site is surrounded by the university facilities and is impacted by traffic emissions from the adjacent road. The site is about 1 km from downtown Champaign and is surrounded by dense housing and business development.
BON is a rural site, 15 km west of downtown Champaign, and is also a part of the IMPROVE (Interagency Monitoring of Protected Visual Environments) monitoring program. The station is managed by the Illinois State Water Survey and is surrounded by intensively managed agricultural fields. The major highways (I-57 and I-74) are at least 6 km north and east of this site, respectively.
The CHI site is located on a dormitory building -Carman Hall on the Illinois Institute of Technology (IIT) campus in Chicago, IL. This site is ∼ 500 m away from a two-way six-lane (including an emergency lane) interstate highway (I-90/94), 1.5 km west of Lake Michigan, and 5 km south of downtown Chicago. The highway I-90/94 has an annual average daily traffic flow of 300 000 vehicles per day, and heavy-duty vehicles account for ∼ 10 % of the traffic (Xiang et al., 2019). The site is situated in the mixed commercial and residential area of Chicago, and therefore emissions from traffic mixed with residential and commercial activities are expected.
The IND site is located inside the campus of the School of Public Health, Indiana University-Purdue University Indianapolis (IUPUI). This site is close to downtown Indianapolis (2 km southeast of the IND site) and a two-way four-lane interstate highway I-65 (1 km northeast of the IND site). The site is surrounded by miscellaneous facilities of IUPUI and Riley Hospital; therefore, the sources of ambient aerosols at the IND site may include vehicular emissions from the highway and emissions from residential and commercial activities related to miscellaneous university and hospital operations.
The STL site is located 3 km north of downtown St. Louis, MO. This site is 230 m west of the interstate I-44/70 and 1.2 km west of the Mississippi River. It is also surrounded by several industries for steel processing, zinc smelting, and copper production . Therefore, a significant portion of metals in PM at this site is assumed to be from industrial emissions. The urban activities in downtown St. Louis and traffic emissions from highway vehicles and river boating are also potential sources of PM 2.5 at this site.
The sampling period involved four seasons starting from 22 May 2018 to 30 May 2019. Integrated ambient PM 2.5 samples were collected simultaneously for 3 continuous days from all the sites. Each site was instrumented with a highvolume (hi-vol) air sampler equipped with a PM 2.5 inlet (flow rate = 1.13 m 3 min −1 ; Tisch Environmental; Cleves, OH). Both before and after the sampling campaign, we did a comparison of various samplers by running them in parallel to collect PM 2.5 samples and analyzing them for OP DTT (see Sect. S1 of the Supplement). All the samplers were equipped with a timer to enable automatic start of the sampling each Tuesday at 00:00 CT and turn-off each Friday at 00:00 CT (CT: Central Time). After the sampled filters were collected on Friday (before noon), new filters were loaded in the filter holder to start the next run of sampling. All five samplers were monthly-calibrated for the flow rate by using a variableflow calibration kit (Tisch Environmental), and the flow rate was measured every week before and after the sampling. We used quartz filters (Pall TissuquartzTM, 8 ×10 ) for collecting PM 2.5 . The filters were prebaked at 550 • C for 24 h before sampling. A total of 241 filters were collected during the whole campaign (44 from CHI, 47 from STL, 54 from IND, 51 from CMP, and 45 from BON). We also collected field blank filters (N = 10 from each site) once every 5 weeks by placing a blank quartz filter in the filter holder of the sampler for 1 h but without running the pump.
All filters were weighed before and after sampling using a lab-scale digital balance (0.2 mg readability, Sartorius A120S, Göttingen, Germany) for determining the PM 2.5 mass loading on each filter. Prior to each weighing, filters were equilibrated in a room with constant temperature (24 • C) and relative humidity (50 %) for 24 h. After sampling, the filters were individually wrapped in prebaked (550 • C) aluminum foils and stored in a freezer at −20 • C before analysis. More information on sampling, including the exact dates of sampling, is provided in Table S1 in the Supplement.

Sample extraction protocol
The sample extraction protocol for OP analysis was determined by the requirement to keep a relatively constant concentration of PM 2.5 in the liquid extracts. This is due to the nonlinear response of certain OP endpoints with PM 2.5 mass in the extracts (Charrier et al., 2016). Thus, the fraction of the filter and the volume of water used for extraction were varied depending on the PM 2.5 mass loading on each hi-vol filter. For the analyses of water-soluble OP, a few (usually three to five) circular sections (16-25 mm diameter) were punched from the filter and immersed into 15-20 mL of deionized Milli-Q water (DI, resistivity = 18.2 M cm −1 ). The volume of water was adjusted to achieve ∼ 100 µg of total PM 2.5 per milliliter of DI. The vials containing filter sections suspended in the DI were sonicated in an ultrasonic water bath for 1 h (Cole-Palmer, Vernon Hills, IL, US). These suspensions were then filtered through a 0.45 µm PTFE syringe filter to remove all water-insoluble components including filter fibers. 10.5 mL of these filtered extracts was separated and diluted with DI to 15 mL. These diluted extracts were then kept in the sample queue of SAMERA for OP analyses. SAM-ERA withdraws a different volume of these extracts into reaction vials (RVs) for each OP measurement, i.e., 3.5 mL for OP AA , OP GSH , and OP OH−SLF and 2.1 mL for OP DTT and OP OH−DTT measurements, all of which were further diluted to 5 mL in the RVs. Thus, the concentrations of PM 2.5 in RVs for SLF-based (i.e., OP AA , OP GSH , and OP OH−SLF ) and DTT-based (i.e., OP DTT and OP OH−DTT ) assays were maintained constant at 50 and 30 µg mL −1 (±1 %), respectively.
For methanol-soluble OP measurements, another fraction from each filter having the same area as used for the water-soluble PM 2.5 extraction was punched and extracted in 10 mL of methanol. After sonication for 1 h, the suspensions were filtered through the 0.45 µm PTFE syringe filter. The filtered extracts were then concentrated to less than 50 µL using a nitrogen dryer to evaporate methanol and were subsequently reconstituted in DI to the exact same volume as the water-soluble extracts. Reconstituted methanol extracts were vigorously shaken on an analog vortex mixer (VWR International, Batavia, IL, US) for at least 60 s at 3200 rpm to ensure a thorough flushing of the components probably deposited along the wall of the vials during evaporation. These methanol-soluble extracts were then analyzed for OP in the same way as water-soluble extracts.

OP analysis
OP activities of PM 2.5 extracts were analyzed using SAM-ERA. The setup and operation protocol of SAMERA have been discussed in detail in Yu et al. (2020). Briefly, the analysis of all OP endpoints for each extract was conducted in two stages: SLF-based endpoints were analyzed first, while DTTbased assays were conducted in the second stage. For measuring OP AA and OP GSH , 3.5 mL of the extract was mixed with 0.5 mL SLF and 1 mL of 0.5 M potassium phosphate buffer (K-PB) in an RV. SLF was made following the protocol of Yu et al. (2020), i.e., by mixing equal volumes (1 mL each) of four antioxidant stock solutions -20 mM AA, 10 mM GSH, 30 mM citric acid (CA), and 10 mM UA -and diluting the mixture with DI to 10 mL. Final concentrations of the antioxidants in the RV used for incubating the sample were 200 µM AA, 100 µM GSH, 300 µM CA, and 100 µM UA. At certain time intervals (i.e., 5, 24, 43, 62, and 81 min), two small aliquots of the reaction mixture were withdrawn and dispensed into two measurement vials (MV1 and MV2) separately. The mixture in MV1 was diluted by DI and was directly injected into a liquid waveguide capillary cell (LWCC-3100; World Precision Instruments, Inc., Sarasota, FL, USA) coupled to an online spectrophotometer (Ocean Optics, Inc., Dunedin, FL, USA), which measured the absorbance at 265 nm (signal from AA) and 600 nm (background) for determining the concentration of AA. 1.6 mL of o-phthalaldehyde (OPA) was added into the reaction mixture contained in MV2 to react with GSH, which forms a fluorescent product. The final mixture in MV2 was then pushed through a flow cell equipped in a Horiba Fluoromax-4 spectrofluorometer (Horiba Scientific, Edison, NJ, USA), and the fluorescence was measured at an excitation-emission wavelength of 310 / 427 nm. Simultaneously with the preparation of the reaction mixture for OP AA and OP GSH analyses, 3.5 mL of the extract was mixed with 0.5 mL SLF and 1 mL of 50 mM K-PB buffered disodium terephthalate (TPT) (pH = 7.4) in another RV2. TPT captures q OH generated in the reaction and forms another fluorescent product: 2-hydroxyterephthalic acid (2-OHTA). Small aliquots of this reaction mixture were withdrawn into MV2 at selected time intervals (10, 29, 48, 67, and 86 min), diluted by DI, and injected into the flow cell of the spectrofluorometer for measuring fluorescence at the same wavelengths as used for GSH measurement (i.e., 310 nm excitation and 427 nm emission). The concentration of 2-OHTA was determined by calibrating various concentrations (10-500 nM) of 2-OHTA standards, and the generation rate of q OH was determined as the formation rate of 2-OHTA divided by a yield factor (0.35) (Son et al., 2015).
Both RVs and MVs were flushed with DI after all SLF-based endpoints were analyzed, and DTT-based assays started immediately after this cleaning. Similar to the first step of the SLF assay, 2.1 mL of the diluted PM 2.5 extract was mixed with 1 mL of 50 mM TPT, 1.4 mL of DI, and 0.5 mL of 1 mM DTT in an RV. At certain time intervals (i.e., 5, 17, 29, 41, and 53 min), two small aliquots of this reaction mixture were withdrawn and diluted with DI in MV1 and MV2 separately for the measurement of DTT and q OH, respectively. DTNB was added into MV1 to capture residual DTT. The final mixture in MV1 was pushed through LWCC to measure the absorbance at 412 nm, while the mixture in MV2 was pushed through the flow cell of the spectrofluorometer for fluorescence measurement (310 nm excitation and 427 nm emission). The system was again cleaned by flushing DI to RVs, MVs, LWCC, and the flow cell of the spectrofluorometer for the next run. Once a week, we conducted a thorough cleaning of the entire system by replacing all chemicals and samples first with methanol, followed by DI, and running the program script 10 times with each solvent.

Quality control and quality assurance
One field blank filter extract along with a DI blank were used as the negative controls for each set of PM 2.5 samples analyzed in a batch (usually ∼ 10). Selected metals and organic compounds that are known to be sensitive for different OP endpoints, i.e., Cu(II) for OP AA and OP GSH , Fe(II) for OP OH−SLF , phenanthraquinone (PQ) for OP DTT , and 5hydroxy-1,4-naphthoquinone (5-H-1,4-NQ) for OP OH−DTT , were used as the positive control and were analyzed weekly with PM 2.5 samples to ensure the stability of SAMERA and correct for any possible drift.
The average and standard deviation of OP for negative and positive controls are shown in Table 1. Our previous study on the development of SAMERA  reported the values of OP for negative controls as 0.17 ± 0.07 µM min −1 for OP AA , 0.37 ± 0.06 µM min −1 for OP GSH , 4.57 ± 1.21 nM min −1 for OP OH−SLF , 0.65 ± 0.02 µM min −1 for OP DTT , and −0.38 ± 0.24 µM min −1 for OP OH−DTT , which are consistent with the values reported in Table 1. The precision of SAMERA was previously assessed using watersoluble extracts, and the coefficient of variations (CoVs) were reported to be less than 14 % (7.9 %-13.3 %) for all OP endpoints . We also assessed the precision using methanol-soluble extracts and found similar levels of CoVs, i.e., 8.9 %-14.5 % for all OP endpoints (see Table S2). Consistency of our current results for negative controls with those reported earlier, and the low CoVs obtained for the positive controls (1.1 %-11.8 %) and PM 2.5 extracts, ensured good quality assurance for the overall OP analysis. We blank-corrected all OP values of ambient samples by subtracting the averaged field blank measurements. After blank correction, the OP values below the detection limit were replaced with half of the detection limits for the corresponding OP endpoint. The mass-normalized (intrinsic, OPm) and volume-normalized (extrinsic, OPv) OP levels were obtained by dividing the blank-corrected OP activities by the extracted PM 2.5 mass (for OPm) and by the volume of air collected on the extracted fractions of filters (for OPv), respectively. The

Statistical analysis
To assess spatiotemporal variability in both OP and PM 2.5 mass, we compared their differences among all sites and seasons using a one-way analysis of variance (ANOVA) test, and different pairs (i.e., pairs of different sites or seasons) were compared by Fisher's least significant difference (LSD) post hoc test. The significant and highly significant differences were considered by one-way ANOVA when P < 0.05 and P < 0.01, respectively. Pearson's correlation coefficient (r) for single linear regression was computed to determine the correlation of OP between different sites, between water-soluble and methanol-soluble OP, and between OP and PM 2.5 , as well as the intercorrelation among different endpoints for each site. All PM 2.5 samples were assessed for spatiotemporal variability. However, since several OP endpoints (e.g., OP AA , OP GSH , and OP DTT ) were abnormally elevated in the week of 4 July (Independence Day celebration; discussed in Sect. 3.2), we removed this week's sample from our regression analysis to avoid any bias caused by this episodic event. Site-to-site comparisons were performed by calculating the coefficient of divergence (COD) of mass concentration and volume-normalized OP (i.e., OPv) for all site pairs as follows: where c ij and c ik are the PM 2.5 mass or OPv measured in the same week i at sites j and k, respectively, and N is the number of comparable sample pairs for sites j and k. COD ranges from 0 to 1. A larger COD (closer to 1) indicates more spatial heterogeneity between the sites, while a smaller COD (closer to 0) implies spatial homogeneity. A one-way ANOVA test was conducted in MATLAB R2019a, while other statistical analyses were carried out using Excel. 3 Results and discussion 3.1 PM 2.5 mass concentration Figure 2 shows the time series of 3 d averaged PM 2.5 mass concentration at five sampling sites, while the seasonal averages are shown in Table 2. The mass concentrations ranged from 2.0 to 21.7 µg m −3 across all sites, and the median was 11.0 µg m −3 . These results are comparable with the typical ranges of PM 2.5 in Midwestern US cities (2.1-48.6 µg m −3 ), e.g., St. Louis , Chicago (Milando et al., 2016), Detroit (Gildemeister et al., 2007), Bondville, and selected cities in Iowa (e.g., Cedar Rapids, Des Moines, and Davenport) (Kundu and Stone, 2014), as measured in several previous studies. Generally, the more urbanized sites of our study (i.e., CHI, STL, and IND) showed slightly higher mass concentrations (5.7-21.7 µg m −3 ; median: 11.8 µg m −3 ) compared to the smaller cities like CMP and its rural component (i.e., BON) (2.0-20.2 µg m −3 ; median: 9.2 µg m −3 ). The highest mass concentrations were recorded at CHI during winter (P < 0.01; Table S3) and STL during summer (P < 0.05), while BON exhibited the low- Summer 2018 11.2 ± 3.2 14.7 ± 3.4 11.9 ± 3.5 11.4 ± 3.9 10.4 ± 2.0 Fall 2018 10.9 ± 3.4 13.1 ± 3.7 11.5 ± 4.2 7.5 ± 4.3 9.7 ± 3.5 Winter 2018 14.6 ± 3.6 11.8 ± 2.8 11.0 ± 2.7 10.0 ± 3.0 8.6 ± 3.0 Spring 2019 12.6 ± 4.2 13.8 ± 4.0 12.2 ± 2.1 11.6 ± 3.1 9.2 ± 2.3 est concentrations in all seasons, except fall when the mass concentrations were lowest at CMP (P < 0.05). Other than these minor variations, the PM 2.5 mass concentrations are both spatially and temporally homogeneous in the Midwestern US with no significant seasonal differences (P > 0.05 at most sites).

Spatiotemporal variation in PM 2.5 OP
Time series of both mass-and volume-normalized OP (OPm and OPv, respectively) at all the sites are shown in Fig. 3 (water-soluble OP) and Fig. 4 (methanol-soluble OP). Seasonally averaged OPm and OPv of water-soluble and methanol-soluble PM 2.5 are also shown in Figs. 5 and 6, respectively. Differences in both OPm and OPv among different seasons or sites were determined by one-way ANOVA, and the results are listed in the Supplement in Table S4 (water-soluble OP) and Table S5 (methanol-soluble OP). Generally, OP showed much more spatiotemporal variability than the PM 2.5 mass in the Midwestern US.

Water-soluble PM 2.5 OP
Figures 3 and 5 (time series and seasonal averages of watersoluble OP) show significant spatial variability for SLFbased endpoints, particularly OP AA and OP GSH , in comparison to DTT-based OP (i.e., OP DTT and OP OH−DTT ) for both mass-and volume-normalized results. The highest OP AA and OP GSH activities (both mass-and volume-normalized) occurred at CMP (P < 0.01) in most seasons. OP OH−SLF was more spatially uniformly distributed than OP AA and OP GSH ; significantly higher OP OH−SLF m and OP OH−SLF v were observed at CMP only in summer and spring (P < 0.05).
For the DTT-based endpoints, OP DTT v was only marginally higher at CHI in winter and at CMP in summer and spring.
Other than that, no significant differences were observed for OP DTT v among various sites. The spatially uniform pattern for OP DTT v is consistent with Verma et al. (2014), who found limited spatial variation for OP DTT v in the southeastern US. In contrast, there was a significant variation in the OP DTT m, with elevated levels at CMP (P < 0.01) in all seasons. Interestingly, the OP OH−DTT endpoint showed more spatial variability and was generally lowest at CMP (P < 0.05) -the site which showed the highest levels for other OP endpoints. It implies that although OP DTT and OP OH−DTT endpoints are measured in the same DTT assay, different chemical com-ponents play differential roles in these endpoints. We found very similar spatial patterns of mass-and volume-normalized OP activities for most endpoints, indicating only a marginal role of PM 2.5 mass concentrations in causing the spatial variability in OP levels. Seasonally, the highest OP activities were generally observed in summer, while the lowest activities usually occurred in winter (Fig. 5). An exception to this trend was OP DTT , which exhibited limited temporal variation at most sites, with only slightly higher OP DTT observed in summer at BON (P < 0.05). The temporal uniformity of OP DTT in this study does not correspond to previous studies conducted in the southwestern and southeastern US. For the southeastern US, Verma et al. (2014) found significantly higher OP DTT v in winter (December 2012) compared to summer (June to August 2012), and this difference was even more pronounced in mass-normalized OP. Saffari et al. (2014) also observed higher OP DTT activities of quasi-ultrafine particles (PM 0.25 ) in the fall and winter seasons for the southwestern US (Los Angeles Basin) and attributed this trend to the partitioning of redox-active semi-volatile organic compounds to the particle phase in colder seasons. However, the trend of OP AA in our study is in agreement with another study in the southeastern US , which showed higher OP AA in warmer seasons (i.e., summer and fall) than in winter. The seasonal trends of mass-and volume-normalized activities were nearly identical for all endpoints, again indicating a marginal effect of PM 2.5 mass concentration on the temporal variation of OP.
Significant temporal variation was observed for CMP, with several spikes in the OP activities throughout the year, most prominently for OP AA (Fig. 3). These spikes might be attributed to the traffic, as CMP is the only site adjacent (< 10 m) to a major urban road and located on the roof of a parking garage. One of our previous studies, Wang et al. (2018), reported large variations in several redox-active metals (e.g., Cu, Fe, Mn, Pb, and Zn), which have been known to be related to vehicular emissions (Hulskotte et al., 2007;Garg et al., 2000;Gietl et al., 2010;Apeagyei et al., 2011;Councell et al., 2004), at the same CMP site. Since SLF-based endpoints have been shown to be highly sensitive to metals (Ayres et al., 2008;Calas et al., 2018;Fang et al., 2016;Moreno et al., 2017;Charrier and Anastasio, 2015;, the temporal variation in traffic intensity probably contributes to the spikes observed at CMP. The peaks in the week of 3 July were observed for multiple endpoints (e.g., OP AA , OP GSH , and OP DTT ) at most sites, which is attributed to the emissions from firecrackers at Independence Day (4 July) celebrations Puthussery et al., 2018).

Methanol-soluble PM 2.5 OP
Compared to water-soluble OP, most OP endpoints in the methanol-soluble extracts showed weaker seasonal variations (Figs. 4 and 6), as also confirmed by relatively lower F values (median of F = 1.61 shown in Table S5a compared to 2.71 for the water-soluble OP endpoints shown Table S4a). Similar to water-soluble OP, the highest activities for the methanol-soluble OP were generally observed in summer (Fig. 6). The spatial variations in OP were also weaker for the methanol-soluble extracts in comparison to water-soluble extracts (median of F = 1.96 shown in Table S5b compared to 4.52 for the water-soluble OP endpoints shown in Table S4b). However, some significantly higher OP levels were observed at certain sites in different seasons, e.g., OP AA v at CHI in winter and spring, OP GSH v at CHI and CMP during winter and spring, OP GSH m at CMP in all seasons, OP OH−SLF at CHI in summer and winter, and OP OH−DTT m and OP OH−DTT v at CHI in summer (P < 0.05).
Other than these few cases, the spatiotemporal trends were again largely similar between mass-and volume-normalized methanol-soluble OP activities.

Comparison of OP in the Midwestern US with previous investigations
A comparison of the ranges of OP endpoints measured in our study with those reported in previous studies is provided in Table S6. The purpose of this comparison is to validate our measurements and present a larger perspective on the general levels of OP in the Midwestern US in comparison to other regions of the world. For watersoluble PM 2.5 in our study, OP AA m ranged from 0.002 to 0.077 nmol min −1 µg −1 , which is within the ranges reported from previous studies conducted in Europe ( in previous studies (Mudway et al., 2005;Künzli et al., 2006;Godri et al., 2011). Similarly, the median of OP OH−SLF m (0.142 pmol min −1 µg −1 ) is comparable to the averages reported by Vidrio et al. (2009) and Ma et al. (2015) (0.092-0.253 pmol min −1 µg −1 ). The median of OP DTT m (0.014 nmol min −1 µg −1 ) of our samples is significantly lower than the medians or averages reported from most studies conducted in the US Charrier and Anastasio, 2012;Gao et al., 2020b;Hu et al., 2008; and Greece (0.019-0.041 nmol min −1 µg −1 ) (Paraskevopoulou et al., 2019) but is closer to the averages reported from studies conducted in Italy (0.010-0.012 nmol min −1 µg −1 ) (Cesari et al., 2019;Perrone et al., 2019). Similarly, the median of our OP DTT v (0.150 nmol min −1 m −3 ) is lower compared to several studies in the southeastern US and Europe (0.19-0.33 nmol min −1 m −3 ) Gao et al., 2017Gao et al., , 2020aParaskevopoulou et al., 2019;Perrone et al., 2019;Cesari et al., 2019) but closer to one study conducted in the southwestern US (0.14 nmol min −1 m −3 ) (Hu et al., 2008). The range of water-soluble OP OH−DTT v of our samples is quite large (0.004-3.565 pmol min −1 m −3 ); however, there are no previous data to compare to other than those reported in the studies conducted by our own group (Xiong et al., 2017;Yu et al., 2018), which were based on a much smaller sample size (N = 10) and limited spatial extent (single site), thus resulting in a much narrower range (0. , probably due to different measurement protocols (only AA in comparison to SLF in our approach). The medians of our methanol-soluble OP DTT m (0.021 nmol min −1 µg −1 ) and OP DTT v (0.234 nmol min −1 m −3 ) are slightly lower than the medians or averages reported in previous studies in the southeastern US (0.027-0.034 nmol min −1 µg −1 and 0.28-0.30 nmol min −1 m −3 , respectively, for OP DTT m and OP DTT v) (Verma et al., 2012;Gao et al., 2017Gao et al., , 2020b, which is consistent with the trend for water-soluble OP DTT (i.e., lower levels of our samples than previously reported at other sites).

Comparison of water-soluble and methanol-soluble OP
To assess the effect of solvent on the OP response, we computed the ratio of methanol-soluble OPv to water-soluble OPv (M / W OP ) for all samples and plotted it for the individual sites in Fig. 7. As shown in the figure, methanolsoluble extracts generally showed a greater response for most of the OP endpoints than the water-soluble extracts, with medians of M / W OP being either close to or greater than 1. The medians for M / W OP for OP GSH v and OP DTT v were closer to 1 at many sites, while they were significantly greater than 1 for the other three endpoints (OP AA v, OP OH−SLF v, and OP OH−DTT v). The only exception to this trend was for OP AA v at CMP, where significantly lower levels of methanol-soluble OP than water-soluble OP were observed (median of M / W OP = 0.7 for OP AA v at CMP). Our previous studies analyzing the chemical composition of PM collected at CMP have shown an elevated level of Cu (up to 60 ng m −3 ) at this site Puthussery et al., 2018) compared to the typical range (4-20 ng m −3 ) at most urban sites in the US (Buzcu-Guven et al., 2007;Kundu and Stone, 2014;Hammond et al., 2008;Baumann et al., 2008;Milando et al., 2016). Although watersoluble Cu has been shown to be the most important contributor to OP AA Ayres et al., 2008;Visentin et al., 2016), Lin and Yu (2020) reported a strong antagonistic interaction of Cu with imidazole and pyridine, both of which are alkaloid compounds (i.e., reduced organic nitrogen compounds), for oxidizing AA. The unprotonated nitrogen atom in alkaloids tends to chelate Cu, thus reducing its reactivity with AA. The antagonistic effects of Cu have been reported with other organic compounds (e.g., citric acid) as well . Thus, apparently lower levels of methanol-soluble OP AA compared to the water-soluble OP AA at CMP might be associated with the chelation of Cu by these alkaloids or other organic species, which could be more efficiently extracted in methanol.
The medians of M / W OP were very high (1.4-3.8) for both q OH-based endpoints (i.e., OP OH−SLF and OP OH−DTT v), indicating that methanol is able to more efficiently extract the Table 3. Pearson's correlation coefficient (r) between water-soluble and methanol-soluble OPv for different endpoints at five sampling sites. Correlations with r > 0.60 are shown in bold. Single asterisks * and double asterisks * * indicate significant (P < 0.05) and highly significant (P < 0.01) correlations, respectively. redox-active components driving the response of these OP endpoints. In addition to q OH-active organic species, e.g., quinones (Charrier and Anastasio, 2015;Xiong et al., 2017;Yu et al., 2018), which are more soluble in methanol, we suspect that one such component could be organic-complexed Fe. As a Fenton reagent, Fe can catalyze the transfer of electrons from H 2 O 2 to q OH (Held et al., 1996). The generation of q OH is further enhanced by the complexation of Fe with organic species Gonzalez et al., 2017;Xiong et al., 2017;Yu et al., 2018). In a previous study conducted at our CMP site, Wei et al. (2018) found a significant fraction of Fe complexed with hydrophobic organic species (28 ± 22 %). That study also reported a substantially higher ratio of Fe concentration in 50 % methanol to that in water (1.42±0.19), which showed some seasonality (1.97 ± 0.17 during winter and 1.33 ± 0.20 in summer). This seasonal pattern of Fe solubility in methanol versus water is consistent with the time series of M / W OP for OP OH−SLF v at most sites (showing higher values in winter than summer; Table S7), which further corroborated the fact that Fe complexed with the hydrophobic organic fraction of PM 2.5 could be majorly responsible for the OP OH−SLF v and OP OH−DTT v in the methanol extracts. However, detailed chemical characterization will be needed to confirm these hypotheses, which will be explored in our subsequent publications. We also calculated Pearson's r for the regression between respective water-soluble and methanol-soluble OP endpoints for individual sites, which are shown in Table 3. OP DTT v showed some good correlation between two extraction protocols (r = 0.43-0.74 except at STL), while correlations were generally poor (r < 0.60) for the other four endpoints (i.e., OP AA v, OP GSH v, OP OH−SLF v, and OP OH−DTT v). It indicates that the components driving the response of OP DTT could be more uniformly extracted in both water and methanol. However, there are additional water-insoluble species driving the response of OP AA v, OP GSH v, OP OH−SLF v, and OP OH−DTT v, which are more efficiently extracted in methanol than water.

Site-to-site comparison of OP and mass concentration of PM 2.5
To further evaluate the spatial trend of OP across the Midwestern US region, we calculated both COD and correlation coefficients (Pearson's r) for different site pairs, which are shown in Fig. 8 (mass concentrations and water-soluble OP of PM 2.5 ) and Fig. 9 (methanol-soluble PM 2.5 OP).
3.4.1 PM 2.5 mass concentration and water-soluble PM 2.5 OP PM 2.5 mass concentrations showed low levels of CODs (0.13-0.25, median: 0.20), confirming a spatially homogeneous distribution of PM 2.5 as indicated earlier (Fig. 8a). Conversely, we observed generally higher CODs (median = 0.27-0.43) for all water-soluble OPv endpoints ( Fig. 8b-f). Our results, showing stronger spatial variability in OP than PM mass, are largely in agreement with a recent study (Daellenbach et al., 2020) analyzing a comprehensive dataset for OP in Europe, which showed that both OPv (measured by DTT, 2 ,7 -dichlorofluorescin diacetate and AA assays) and PM 10 mass concentrations were elevated in urban environments (e.g., Paris and the Po Valley), but PM 10 was more regionally distributed than OPv. Interestingly, we found poor correlations for PM 2.5 among all site pairs (r < 0.60), except IND and BON (r = 0.63). It implies that despite a homogeneous spatial distribution, emission sources of the chemical species composing PM 2.5 are different at different sites. The correlations were also weak (r < 0.60 for most cases) for the OP endpoints showing high CODs, i.e., OP AA , OP GSH , OP OH−SLF , and OP OH−DTT , which indicates a more pronounced effect of local point sources on these OP endpoints compared to the regional sources. In contrast, OP DTT v showed stronger correlation (r = 0.48-0.76, median: 0.62) for most site pairs. Higher correlations for the DTT activity combined with lower CODs suggests that regional sources such as long-range transport or atmospheric processing could have a larger influence on OP DTT than local sources.

Methanol-soluble PM 2.5 OP
In comparison to water-soluble PM 2.5 OP, CODs for the methanol-soluble OP were generally lower (median: 0.21-0.35; Fig. 9), indicating higher spatial homogeneity of methanol-soluble PM chemical components that are sensitive to OP. Similar to water-soluble OP DTT v, the methanolsoluble OP DTT v showed the lowest COD (0.14-0.26, median: 0.21) among five endpoints (Fig. 9d), which was consistent with Gao et al. (2017), showing a rather low COD (less than 0.23) for both water-soluble and methanol-soluble OP DTT in the southeastern US. Overall, higher correlation coefficients were observed for the methanol-soluble OP (median: 0.41-0.67 for different endpoints) than the correspond- Figure 8. Coefficient of divergence (CoD) and Pearson's r for site-to-site comparison of (a) PM 2.5 mass and water-soluble OP activities: Single asterisks * and double asterisks * * on the bars of Pearson's r indicate significant (P < 0.05) and very significant (P < 0.01) correlations, respectively. Note: r values for the correlations of OP AA v between CHI and CMP and for the correlations of OP GSH v between IND and BON were negative (−0.14 and −0.06, respectively).
ing water-soluble endpoints (median: 0.13-0.62). The correlation coefficients were more elevated for certain endpoints such as OP AA v (r = 0.38-0.62, median: 0.46) and OP GSH v (r = 0.23-0.65, median: 0.41) than others. It is possible that methanol is able to extract more redox-active PM components coming from regional emission sources, e.g., biomass burning or secondary organic aerosols, present at these sites. The components originating from these common sources could mask the effect of other components originating from local sources having a narrower range of solubilities, thus yielding overall lower spatiotemporal variability and better correlation among different sites.

Correlations of OP with PM 2.5 mass concentration
Pearson's r and the slope for simple linear regression of volume-normalized OP activities versus PM 2.5 mass concentrations were computed for each individual site and are listed in Table 4. For both water-soluble and methanol-soluble OP, the endpoints of OP AA v, OP OH−SLF v, and OP OH−DTT v were poorly correlated with PM 2.5 mass (r < 0.60 in most cases), while OP GSH v and OP DTT v were moderately to strongly correlated with PM 2.5 mass (r = 0.38-0.73 for OP GSH v and 0.54-0.82 for OP DTT v, except at STL). The lower correla-tion of OP AA and higher correlation of OP DTT are consistent with multiple previous studies comparing these endpoints (Visentin et al., 2016;Yang et al., 2014;Janssen et al., 2014). Decent correlations for OP GSH v and OP DTT v showed that PM mass concentrations can drive these endpoints to some extent at a few locations. However, it is important to note that despite these good correlations, the slope of regression for OP vs. PM 2.5 mass varied a lot among five sampling sites (range for OP GSH v is 0.003-0.016 nmol min −1 µg −1 and 0.005-0.028 nmol min −1 µg −1 for OP DTT v), indicating substantial spatiotemporal heterogeneity in the intrinsic potency of the particles to generate ROS at these sites. This is further corroborated by the spatiotemporal variability of OP GSH m and OP DTT m at different sites, as shown in Figs. 5 and 6. Thus, PM 2.5 mass concentrations have only a limited role in determining the oxidative levels of the PM 2.5 at these sites, and OP seems to be largely driven by the PM chemical composition. Given that the current air quality standards across the world focus only on the mass concentration of PM 2.5 , these results point towards the inadequacy of this mass-centered approach.

Intercorrelation among different OP endpoints
We also calculated the correlation coefficient (Pearson's r) for all pairs of different OPv endpoints at each site, which are listed in Table 5. A high correlation coefficient indicates a common source (or a common pool of chemical components) driving the response of those OP endpoints. For watersoluble OP, the intercorrelations among different endpoints were generally poor at urban sites, i.e., CHI, STL, and IND (r < 0.60). Correlations were also poor for nearly all pairs of methanol-soluble OP at STL and IND, but CHI showed significantly elevated r values among different OP endpoints (r = 0.59-0.82). Compared to more urbanized sites, the correlations were generally higher at the local sites, i.e., CMP and BON, with r > 0.60 for many pairs of both water-soluble and methanol-soluble OPv. Since both of these sites are located in smaller cities, the sources of redox-active components probably have less complexity compared to the major city sites, which have multiple and more complex emission sources. As discussed in Sect. 3.2, CMP is largely impacted by vehicular emissions owing to its location adjacent to a major road. Similarly, BON, being a rural site, is largely impacted by agricultural emissions with a marginal impact from vehicular emissions and other sources such as long-range transport from surrounding cities (Kim et al., 2005;Buzcu-Guven et al., 2007). Thus, a lack of other major sources contributing to components, which can drive these endpoints in different directions through their interactions (i.e., synergistic or antagonistic), leads to the similarity of their responses and hence a good correlation among them at these two sites. Among all OP endpoints, OP OH−DTT v showed the poorest correlations with other endpoints except OP OH−SLF v, with which it was correlated at most sites (i.e., CHI, IND, CMP, and BON) for the methanol-soluble extracts (r = 0.66-0.84). Since both of these endpoints measure the rate of generation of q OH, it probably indicates a synergistic role of metals with organic compounds (e.g., Fe with humic-like substances -HULIS, as shown in many previous studies; Yu et al., 2018;Charrier and Anastasio, 2015;Gonzalez et al., 2017;Ma et al., 2015) in partly driving the response of both of these endpoints. Note that OP OH−DTT is a relatively newly developed assay, and there is hardly any previous literature on its comparison with other OP endpoints.
Overall, a poor to moderate and inconstant intercorrelation trend among different endpoints of both water-soluble and methanol-soluble OP at most sites indicates that all these assays could be less than ideal, and measuring a single endpoint is not enough to represent the overall OP activity. Although the OP endpoints used in our study have covered some of the well-known and important pathways of the in vivo oxidative Table 4. Pearson's r and slope for simple linear regression of water-soluble OPv versus PM 2.5 mass concentration at five sampling sites. Correlations with r > 0.60 are shown in bold. All slope values are in italic. Single asterisks * and double asterisks * * indicate significant (P < 0.05) and highly significant (P < 0.01) correlations, respectively. stress caused by PM 2.5 , there are other endpoints (e.g., consumption of cysteine, formation of H 2 O 2 ), and more assays can be developed in the future. We suggest that a collection of a diverse range of OP endpoints, measured separately as done in our study, could better capture the role of different PM components and their interactions via different pathways for driving the oxidative levels of the PM in a region. However, it should be noted that our study is not designed to assess and rank the biological relevance of these acellular endpoints, which will require an integration of these and possibly other novel assays involving different routes of oxidative stress in either toxicological or epidemiological studies.

Conclusion
We analyzed both water-soluble and methanol-soluble OP of ambient PM 2.5 in the Midwestern US using five different acellular endpoints, including OP AA , OP GSH , OP OH−SLF , OP DTT , and OP OH−DTT . The spatiotemporal profiles of all OP endpoints and PM 2.5 mass concentrations were investigated for a 1-year timescale from May 2018 to May 2019 using hi-vol filter samples collected from five Midwestern US sites located in urban, rural, and roadside environments. Compared to homogeneously distributed PM 2.5 mass, all OP endpoints showed significant spatiotemporal variations among different seasons and sites. Seasonally, most OP endpoints generally peaked in summer for both water-soluble and methanol-soluble OP. Spatially, the roadside site showed the highest OP levels for most OP endpoints in water-soluble extracts, while there were occasional peaks in methanolsoluble extracts at other urban sites. Our results showed very limited differences in the spatiotemporal profiles between OPm and OPv for most endpoints, indicating a marginal role of PM 2.5 mass in causing the spatiotemporal variability of OP.
Comparing the OP for water-and methanol-soluble extracts, we observed significantly higher OP levels in Table 5. Pearson's correlation coefficient (r) among various endpoints of OPv measured at five sampling sites. The values below the diagonal are for water-soluble OPv, while those above are for methanol-soluble OPv. Correlations with r > 0.60 are shown in bold. Single asterisks * and double asterisks * * indicate significant (P < 0.05) and highly significant (P < 0.01) correlations, respectively. methanol extracts than the corresponding water-soluble OP activities. This trend was much stronger for q OH generation endpoints (i.e., OP OH−SLF and OP OH−DTT ), indicating a substantial contribution of Fe and its organic complexes, which could be more efficiently extracted in methanol. In comparison to water-soluble OP, methanol-soluble OP showed lower spatial heterogeneity and higher intercorrelations among different endpoints, which is probably attributed to a more efficient extraction of water-insoluble redox-active species in methanol originating from various emission sources at different sites.
The correlations of OP with PM 2.5 mass showed a diverse range, with certain endpoints such as OP AA , OP OH−SLF , and OP OH−DTT showing a poor correlation, while other end-points (i.e., OP GSH and OP DTT ) showed a moderate to strong correlation. Despite these occasional strong correlations, the sensitivity of all OP endpoints to mass, indicated by the slope of OP vs. PM 2.5 mass and the intrinsic OP (OPm), varied substantially for all OP endpoints across different sites and seasons, showing only a marginal effect of mass concentrations in controlling the oxidative levels of PM 2.5 . Moreover, relatively poor and inconsistent correlations among different OP endpoints reflected different pathways of various ROSactive PM 2.5 components for exerting oxidative stress. Since our study cannot comment on the biological relevance of these different pathways, we recommend integrating all these and other assays in toxicological or epidemiological studies to assess their relative utilities.
Collectively, the results obtained through our study provide a strong rationale to recommend that the different endpoints of OP provide more useful additional information than the mass concentrations, which could be relevant to assess the public health impacts associated with ambient PM 2.5 . Our future studies will explore the contribution of different chemical components and their emission sources in determining the oxidative levels of ambient PM 2.5 in the Midwestern US.
Data availability. The data on OP and mass concentration of ambient PM 2.5 samples collected in the Midwestern US are available upon request from the corresponding author.
Author contributions. HY was responsible for collection of PM 2.5 samples, measurement of OP, data analysis, and paper organization and writing. JVP was responsible for collection of PM 2.5 samples as well as paper editing and revision. YW was responsible for collection of PM 2.5 samples, as well as paper editing and revision. VV was responsible for the conceptualization of study design and methodology, paper organization and editing, and overall project supervision.
Competing interests. The contact author has declared that neither they nor their co-authors have any competing interests.
Disclaimer. Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.