The sampling site for this study is located in Hongdong (36∘23′, 111∘40′ E) in Shanxi, China. To collect atmospheric particles of different sizes (0–10 µm), this study used a Thermo Anderson Mark II sampler to collect aerosol samples of nine sizes. The samples were collected on a prebaked quartz filter (450 ∘C, 4.5 h), and the sampling dates were as follows: in winter, 26 January to 4 February 2017, n=10; and in summer, 31 July to 24 August 2017, n=12. The samples were placed in a -20 ∘C refrigerator prior to analysis.

The EPR spectrometer (MS5000, Freiberg, Germany) is used to detect EPFRs in atmospheric samples. The filters were cut into thin strips (5mm×28mm) and put it into the sample tank of the quartz tissue cell (the size of the sample tank is 10mm×30mm). Then the quartz tissue cell with attached filter sample was placed in a resonant cavity and analyzed by an EPR spectrometer. The detection parameters were magnetic field strength, 335–342 mT; detection time, 60 s; modulation amplitude, 0.20 mT; number of detections, 1; and microwave intensity, 8.0 mW. Specific testing protocols have been described previously (Chen et al., 2018c).

The contents of organic carbon (OC) and elemental carbon (EC) in the filter samples were analyzed using a semicontinuous OC/EC analyzer (Model 4, Sunset Lab. Inc., Oregon, USA) with a NIOSH 5040 detection protocol (Lin et al., 2009).

The water-soluble organic carbon (WSOC) concentration was analyzed using an automatic TOC-LCPH analyzer (Shimadzu, Japan). The WSOC extraction was performed with ultrapure water under ultrasonication for 15 min, and all WSOC concentrations were blank-corrected. The concentration of OC in the MSM (methanol-soluble materials) was calculated as the difference between the OC and WSOC (water-soluble organic carbon) concentrations. This calculation assumes that all water-insoluble organic carbon (WISOC) in the aerosol can be extracted with MeOH, and the rationality of this assumption has been verified elsewhere (Mihara and Michihiro, 2013; Liu et al., 2013; Cheng et al., 2016; Chen et al., 2019a).

PAHs were detected using gas chromatography–mass spectrometry (GC-MS) on a GC7890B/MS5977A (Agilent Technologies, Clara, CA). Quartz-fiber filter samples (8 mm in diameter) were cut from each 25 mm quartz-fiber filter substrate used on the ELPI impactor stages using a stainless-steel round punch over a clean glass dish and loaded into the TD glass tube. Next, the TD glass tube was heated to 310 ∘C at a rate of 12 ∘C min-1 and thermally desorbed at 310 ∘C for 3 min. The desorbed organic compounds were trapped on the head of a GC column (DB-5MS: 5 % diphenyl–95 % dimethyl siloxane copolymer stationary phase, 0.25 mm i.d., 30 m length and 0.25 mm thickness). A total of 16 target PAHs were identified based on retention time and typical ion fragments of each PAH standard, including 16 EPA parent PAHs (p-PAHs). The method detection limits (MDLs) ranged from 0.2 pgmm-2 (Ace) to 0.6 pgmm-2 (Incdp). Naphthalene-D8, acenaphthene-D10, phenanthrene-D10, chrysene-D12 and perylene-D12 were used for the analytical recovery check. All compounds were recovered with a desorption recovery percentage of >90 %. Specific testing protocols have been described previously (Han et al., 2018; Song et al., 2020).

The concentration of metal elements in the samples was determined by a Thermo X2 series inductively coupled plasma mass spectrometer (ICP-MS, Thermo, USA). The metal elements analyzed in summer were Na, Mg, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Cd, Pb and Al, and those in winter were Al, Zn, V, Cr, Mn, Co, Ni, Cu, As, Se, Sr, Cd, Ba and Pb. The specific measurement method is based on the study of Qi et al. (2016).

The source and formation process of EPFRs in PM with different particle sizes were analyzed by nonnegative matrix factorization (NMF). The method is based on the study of Chen et al. (2016, 2019e). Briefly, NMF analysis of EPFR data, metal element contents, OC/EC contents and PAH contents was performed in MATLAB. The version of the NMF toolbox is 1.4 (https://sites.google.com/site/nmftool/, last access: 10 November 2017). A gradient-based multiplication algorithm was used to find a solution from multiple random starting values, and then the first algorithm was used to find the final solution based on the least-squares effective-set algorithm. To find a global solution, the model was run 100 times, each time with a different initial value. By comparing the 1–12=factor model (Fig. S4) with the residual of the spectral load, the 6-factor (summer) and 10-factor (winter) NMF models were finally selected.

To assess the health risks of EPFRs, this study divided the respiratory system into three parts based on the human breathing model: extrathoracic (ET) areas, including the anterior nasal cavity, posterior nasal cavity, oral cavity and throat; tracheobronchial (TB) areas, including the trachea, bronchi, bronchioles and terminal bronchi; and pulmonary (P) areas, including the alveolar ducts and alveoli. Then, the sedimentation rates of different particle sizes in different areas of the respiratory system were determined to calculate the exposure risk of EPFRs. Here, the human respiratory system particulate deposition model of Salma et al. (2002) was used, and the specific data can be found in Tables S3 and S4 in the Supplement.

In addition, the daily inhaled concentration of EPFRs into the concentration of free radicals in cigarettes was converted. The specific conversion method is as follows: 1Ncig=(CEPFRs⋅V)/(RCcig⋅Ctar), where Ncig represents the number of cigarettes (per person per day), CEPFRs (spins m-3) represents the atmospheric concentration of EPFRs in PM and V represents the amount of air inhaled by an adult per day (20 m3d-1) (Environmental Protection Agency, 1988). RCcig (4.75×1016 spins g-1) (Baum et al., 2003; Blakley et al., 2001; Pryor et al., 1983; Valavanidis and Haralambous, 2001) indicates the concentration of free radicals in cigarette tar, and Ctar (0.013 g per cigarette) indicates the amount of tar per cigarette (Gehling and Dellinger, 2013).

Figure 1a shows the concentration distribution of EPFRs with different particle sizes in different seasons. EPFRs were detected in the particles of each tested size (the EPR spectrum is shown in Fig. S1 in the Supplement), but their EPFR concentration levels were different. In summer, the concentration of EPFRs in fine particles (particle size <2.1 µm) is (3.2–8.1) ×1013 spins m-3, while the concentration of EPFRs in coarse particles (particle size >2.1 µm) is 1–2 orders of magnitude higher than that of fine particles, reaching values of (2.2–3.5) ×1014 spins m-3. Winter samples show completely different characteristics from summer samples. The concentration of EPFRs in fine particles (particle size <2.1 µm) is (1.8–3.6) ×1014 spins m-3, while the concentration of EPFRs in coarse particles (particle size >2.1 µm) is smaller than that of fine particles, with values of (1.0–2.1) ×1014 spins m-3. In addition, the concentration of EPFRs in particulates <0.43 µm in winter is very high, but it is very low in summer. According to the results of factor analysis in Sect. 3.2 of this study, this particulate matter is related to combustion, which indicates that coal combustion in winter may provide an important contribution to EPFRs. The EPFR concentration in the fine PM of Linfen reported above is equivalent to that in the fine PM of Xi'an, but it is 10 times smaller than that in the fine PM of Beijing (Yang et al., 2017; Chen et al., 2019b). Although the particle size distribution characteristics of EPFRs in winter and summer are different, their concentration levels are similar, which indicates that the EPFR concentration is not related to the PM concentration but is determined by the source characteristics. The source characteristics will be discussed in detail in the factor analysis section.

Figure 1b shows the concentration ratio of EPFRs in coarse and fine particles. The contribution of EPFRs in fine PM in summer is only 14.9 %, while in winter it is 58.5 %. The differences in EPFR concentrations with particle size may be related to the source of EPFRs. For example, coarse particles are often associated with dust sources and biogenic aerosols. In another study, the results have shown that dust particles contain large amounts of metallic EPFRs and that they can be transported over long distances (Chen et al., 2018b). EPFRs in fine particles may be mainly derived from the combustion process, such as traffic sources, which are considered to be an important source of EPFRs in atmospheric PM (Chen et al., 2019b). Due to winter heating in the Linfen area, the amount of coal burning increases sharply in this season. In 2017, the nonclean heating (coal-fired heating) rate of urban heating energy structures in Linfen was 40 % (data source: http://www.linfen.gov.cn/, last access: 28 May 2019). With the burning of coal, large amounts of EPFRs are produced, and in the summer, EPFRs emitted by burning coal should be much less than those emitted in winter. This can explain to a certain extent that the contribution of fine particles to summer EPFRs is small, and the contribution of winter EPFRs is very large.

The g factor obtained by using EPR to analyze the sample is an important parameter to distinguish the type of EPFR. It is the ratio of the electronic magnetic moment to its angular momentum (Shaltout et al., 2015; Arangio et al., 2016). The g factor of carbon-centered persistent free radicals is generally less than 2.003, the g factor of oxygen-centered persistent radicals is generally greater than 2.004 and the g factor of carbon-centered radicals with adjacent oxygen atoms is between 2.003 and 2.004 (Cruz et al., 2012). Figure 2a shows the g factor distribution characteristics of EPFRs in different particle sizes in summer and winter. The g factor of fine particles and coarse particles shows different characteristics. The g factor of EPFRs in fine particles (particle size <2.1 µm) ranges from 2.0034 to 2.0037, which may be from carbon-centered radicals with adjacent oxygen atoms. However, the g factor of EPFRs in coarse particles (particle size >2.1 µm) is significantly less than that of fine particles. The g factor ranges from 2.0031 to 2.0033, indicating that EPFRs in coarse particles are more carbon-centered than those in fine particles and are free of heteroatoms. As shown in Fig. 2b, the g factor varied differently depending on season. The g factor of summer PM showed a significant decreasing trend with increasing concentration, while the g factor of winter PM showed a significant increasing trend with increasing EPFR concentration. Oyana et al. (2017) studied EPFRs in the surface dust of leaves in the Memphis region of the United States and found that the concentration of EPFRs was positively correlated with the g factor, and they believed that this was related to the source of EPFRs. This phenomenon indicates that the sources and toxicity of EPFRs in winter and summer are different.

To explore the possible sources and formation process of EPFRs in atmospheric particles with different particle sizes, the NMF model was used to statistically analyze EPFRs, carbon components, PAHs and metal elements in samples. The factors obtained by the NMF model should reflect the different sources and generation process of EPFRs. As shown in Fig. 3a1 and b1, the three main contributing factors to EPFRs in summer and winter are shown (see Figs. S5, S6 for spectra of other factors), which explain 94.5 % and 83.8 % of the EPFR concentrations in summer and winter, respectively.

As shown in Fig. 3a1, the typical spectral characteristic of summer factor 1 is that it contains a small fraction of EC components and a large amount of OC components, which indicates that combustion may be the source associated with this factor. This factor has the highest loading of OC, especially WISOC; this fraction mainly contains macromolecular organic substances, which are considered to contribute to the main atmospheric particulate EPFRs and to be graphite oxide-like substances (Q. Chen et al., 2017, 2018a). Factor 2 is different from factor 1; factor 2 is more likely the combustion of fossil fuels, while factor 1 should be other combustion sources instead of burning coal, such as biomass combustion. The generation process is similar to a hybrid process, which includes the graphite oxide-like substances produced by incomplete combustion and the EPFRs formed by some metal oxides. The typical characteristic of factor 3 is that the contribution of metal elements is relatively high, while the contributions of EC and OC are very low. Metal elements such as Al, Ti, Mn and Co are typical crust elements, so this factor may represent dust sources (Pan et al., 2013; Srivastava et al., 2007; Trapp et al., 2010). The generation mechanism may be mainly due to the participation of metal oxides n the generation of EPFRs. The others are likely derived from the electroplating metallurgy industry (detailed in Sect. S1 in the Supplement). As shown in Fig. 3a2, the contribution ratios of different factors show that the contribution ratios of factor 1 and factor 2 are the highest, and factor 3 only has a small contribution, which indicates that combustion sources, especially incomplete combustion, are the main sources of EPFRs. The particle size distribution characteristics show that factor 1 is mainly distributed in particles larger than 2.1 µm, while factor 2 is mainly distributed in particles smaller than 0.43 µm.

The results of the factor analysis in winter are different from those in summer. As shown in Fig. 3b1, the typical spectral characteristic of factor 1 is that it contains a large amount of OC components and As and Se. As and Se are trace elements of coal combustion, as shown in many studies (Pan et al., 2013; Tian et al., 2010), so coal combustion may be the source represented by this factor. From the generation process viewpoint, the factor does not contain EC, but the content of OC is very high. In the particles with a particle size of less than 3.3, which is mainly present in factor 1, the concentration of OC is 16 times that of EC. So it may be mainly a graphite oxide-like substance formed by the agglomeration of gaseous volatile organic compounds (VOCs) generated during combustion. The typical spectral characteristics of factor 2 are due to a large amount of V and some Al, EC and OC. OC and EC are also typical combustion products. V is rich in fossil fuels, especially fuel oil (Karnae and John, 2011). Therefore, traffic is the source represented by this factor. The factor contains crust elements such as Al and Mn, so it is speculated that this factor may also include traffic-related dust. The typical spectral characteristics of factor 3 are similar to those of factor 1, and both contain relatively large amounts of As and Se, with the exception that factor 3 contains a large amount of EC, indicating that it is also mainly derived from incomplete combustion sources. The generation process of factor 3 should be different from factor 1, which may include both the graphite oxide-like material generated by fuel coking and the EPFRs generated by the metal oxide. The other factors are mainly atmospheric dust and electroplating or metallurgy (see Sect. S1). As shown in Fig. 3b2, factor 1 and factor 2 have the highest proportions, and factor 3 also has a small contribution, which indicates that winter is the same as summer, and combustion sources are the main source of EPFRs. The particle size distribution characteristics show that factor 1 is mainly distributed in particles with a size of 0.43–3.3 µm, while factor 2 is mainly distributed in particles larger than 3.3 µm.

Based on the above analysis, it can be found that combustion sources are the main sources of EPFRs, and EPFRs from these sources are mainly graphite oxide-like substances generated by the polymerization of organic matter or fuel coking. Studies have shown that graphene oxide can cause cell damage by generating ROS (Seabra et al., 2014). The surface of these compounds contains not only carbon atoms but also some heteroatoms, which leads to disorder and the presence of defects in the carbon-based structure (Lyu et al., 2018; Q. Chen et al., 2017; Mukome et al., 2013; Keiluweit et al., 2010). The dust source is also a source of important EPFRs identified in this study (with a contribution of approximately 10 %). It was shown in the above analysis that the concentration of EPFRs in coarse particles has a significant correlation with the concentration of metallic elements, particularly crustal elements. Some crustal elements, such as Al and Fe, not only have their own paramagnetism (Li et al., 2017; Yu et al., 2017; Nikitenko et al., 1992), but also interact with aromatic compounds attached to the surface of the particles to produce a stable single-electron structure.

To evaluate the health risks of EPFRs in PM with different particle sizes, this study evaluated the comprehensive exposure of EPFRs based on the deposition efficiency of PM with different particle sizes in different parts of the human body. The results are shown in Fig. 4a. The ET region is the region with the highest EPFR exposure, while the TB and P regions have relatively close EPFRs. This result shows that atmospheric EPFRs are the most harmful to the health of the human upper respiratory tract. Comparing the EPFR exposure in different seasons indicates that the exposure risk in the ET area in summer is significantly higher than that in winter. This difference occurs because the concentration of EPFRs in coarse particles is much higher than that of fine particles in summer, and the deposition efficiency of large particles in the ET area is generally higher. Fine particles are more efficiently deposited in the P region, leading to a higher risk of EPFR exposure in the P region in winter.

EPFRs were found early in cigarette tar and are considered one of the health risk factors in cigarette smoke (Lyons and Spence, 1960); thus, in this study, the exposure risks of EPFRs in particles deposited in the human body were converted to the equivalent number of cigarettes inhaled per adult per day. As shown in Fig. 4b, the ET area is the most contaminated area, with an average equivalence of 21 cigarettes (25 in summer and 16 in winter). The average values for the TB area (9 in summer and 7 in winter) and P area (7 in summer and 10 in winter) are eight. The results indicate that EPFRs pose significant health risks to human lungs in both winter and summer. Other similar studies, such as a study of the average amount of EPFRs in PM2.5 inhaled per person per day in Xi'an in 2017, found values equivalent to approximately 5 cigarettes (Chen et al., 2018a). Gehling and Dellinger (2013) found that EPFR exposure in PM2.5 is equivalent to approximately 0.3 cigarettes per person per day in St. Joaquin County, the location with the worst air pollution in the United States. The average exposure risk of EPFRs in fine particles in the Linfen area (approximately 13 cigarettes) was higher than those in these two studies. However, these previous studies only studied the exposure risk of EPFRs in fine particles. The results of this study indicate that the health risks of EPFRs are significantly increased when the particle size distribution of EPFRs is taken into account. Therefore, it is important to study the source characteristics and generation process of EPFRs with different particle sizes, which will be discussed in detail in the following paragraph.

This study calculated the proportion of EPFRs with different particle sizes in different parts of the respiratory system based on the deposition efficiency of particles with different particle sizes. As shown in Fig. 4c, in the ET region and the TB region, coarse particles are the dominant component in summer and winter. In particular, in summer, the proportion of EPFRs in coarse particles in these two regions exceeds 95 %. In the P region, there are significant differences between summer and winter. The P region in summer is still dominated by coarse particles, but its proportion is significantly lower than those in the ET and TB regions. In the P region in winter, fine particles are the dominant component (approximately 70 %). These distribution characteristics indicate different sources of EPFRs in different regions. As shown in Fig. 4d, in summer, combustion sources are the main source of EPFRs in the respiratory system. In winter, combustion and transportation sources contribute equally in the TB and ET regions, while in the alveoli, combustion sources are the main contributor. The ET region is the area with the highest risk of exposure to EPFRs (21 cigarettes). The generation process of these EPFRs is mainly attributable to graphene oxide-like substances. Studies have shown that graphene oxide is cytotoxic (Harmon et al., 2018). In the alveoli, the contribution of combustion sources is significantly increased (especially in winter). These EPFRs are mainly generated by the action of metal oxides and organic substances. Studies have shown that such EPFRs can generate ROS in the lung fluid environment (Khachatryan and Dellinger, 2011).