Trace metals and magnetic particles in PM2.5: Magnetic identification and its implications

Magnetic measurement was combined with geochemical analysis to investigate the trace metal pollution of PM2.5. The study was carried out in Nanjing, China, where the average PM2.5 concentrations in summer and winter in 2013–2014 were 66.37 and 96.92 μg/m3, respectively. The dominant magnetic mineral in PM2.5 had a low-coercivity pseudo-single domain and consisted of magnetite and hematite. Iron-oxide magnetic particles comprised spherical as well as angular particles. Stable Pb isotopic ratio determinations showed that Pb in summer samples derived from coal emissions while the main sources of winter samples were smelting industry and coal emissions. The magnetic properties of the particles correlated strongly with trace metals derived from anthropogenic activities, such as industrial emission, coal combustion, and traffic vehicle activities, but poorly with those derived from natural sources. In the multiple linear regression analysis, Cr and Fe had higher correlation coefficients (training R > 0.7) in contrast to the low training R of As, Cd, Ni, Sr, and Ti (<0.5) determined using the PM2.5 concentrations and magnetic parameter values as the decision variables. Our results support the use of environmental magnetism determinations as a simple and fast method to assess trace metals in urban particulate matter.

Trace metal concentrations in PM 2.5 . The total trace metal concentrations in the PM 2.5 samples are summarized in Table 1. The mean trace metal concentrations, expressed as volume-related contents in air (in ng/m 3 ), generally decreased in the order Fe > Zn > Pb > Ti ≈ Mn > Cu > Cr > Ni ≈ Sr > As > Cd ≈ V > Co, consistent with the results of a study by Li et al. 11 . The mean As and Ni concentrations were significantly higher in the winter samples (As: 16.87 ± 3.409 ng/m 3 ; Ni: 29.14 ± 7.868 ng/m 3 ) than in the summer samples (9.633 ± 1.986 ng/m 3 , P = 0.000, and 8.637 ± 2.992 ng/m 3 , P = 0.000, respectively) based on PM 2.5 air volume (unit: ng/m 3 ), but for the other trace metals the concentrations did not differ between summer and winter (Table 1). There were no clear differences in the trace metal concentrations of the daytime and nighttime samples (P > 0.05).
The As concentrations in all of the PM 2.5 samples were above the NAAQS (GB3095-2012: 6 ng/m 3 ) and the limit set by the World Health Organization (WHO: 6.6 ng/m 3 ). The Cd concentrations in a few winter PM 2.5 samples exceeded the NAAQS (GB3095-2012) and WHO limits (5 ng/m 3 ). The Ni concentrations in 55% of the winter PM 2.5 samples also exceeded the WHO limit (25 ng/m 3 ), whereas those in all of the summer PM 2.5 samples were below the WHO limit. The Mn concentrations in both the winter and the summer PM 2.5 samples were below the WHO limits (150 ng/m 3 for Mn). The Pb concentrations in of all the PM 2.5 samples were below the NAAQS (GB3095-2012) and WHO [500 ng/m 3 in total suspended particles (TSP)] limits, but the Pb concentrations in 20% of the winter PM 2.5 samples were above the US NAAQS limit (150 ng/m 3 in TSP). Although some of the trace metal concentrations were lower than the relevant limits for coarse PM, they could cause adverse effects when present in PM 2.5 . Whether this is indeed the case remains to be investigated in further research 13,41 .
The mass-related trace metal concentrations (μg/g) in the PM 2.5 samples are shown in Table 1. The As, Fe, Ni, Pb, and Zn concentrations were slightly higher in the summer than in the winter samples based on PM 2.5 particle Element All samples Summer Winter Background value (μg/g) Pb ratios in the winter and summer samples were not significantly different. The ratios in the PM 2.5 samples were higher than those of background soil in Nanjing, indicating that the Pb in the PM 2.5 samples mainly came from anthropogenic sources, consistent with the conclusions drawn from the EF value of Pb (Table 1). The Pb isotope ratios in most of the PM 2.5 samples were between those of coal combustion ash and metallurgical industry dust ( Fig. 1), implicating these sources in the Pb pollution in Nanjing, as suggested in previous studies 41,46 . According to the Nanjing Statistical Yearbook 2012, 80.5% of the energy consumed in Nanjing is produced by burning coal 44 . The Pb isotope ratios were similar to those in metallurgical industry dust in only a few of the summer PM 2.5 samples; thus smelting industry activities were not the main sources of Pb in the summer sampling period. This can be explained by the fact that many smelters had been shut down before the Youth Olympic Games were held, in 2014. The Pb isotope ratios for several PM 2.5 samples were similar to those of cement dust, consistent with the location of the sampling site in a new urban development zone in which construction work was taking place. The Pb isotope ratios were distinctly lower in the PM 2.5 samples than in exhaust emissions from vehicles using unleaded gasoline. However, Pb is used as a tracer of non-exhaust traffic emissions 43 , and the stable Pb isotope ratios for dust emitted on trunk roads overlapped with those of some of the PM 2.5 samples, indicating that traffic contributes to Pb pollution in Nanjing.

Content in mass (μg/g) EF
Magnetic properties of PM 2.5 . The means, standard deviations, and ranges of the magnetic parameters for the summer and winter samples are shown in Table 2. The low frequency susceptibility (χ LF ) and saturation isothermal remanent magnetization (SIRM) generally reflect the concentrations of magnetic minerals, especially ferromagnetic minerals (e.g., magnetite) 47 . Unlike SIRM, χ LF is influenced by paramagnetic and diamagnetic minerals 47 , attributable to the seasonally different sources of PM 2.5 and seasonally different magnetic mineral species found in PM 2.5 . The mean values of χ LF , 'hard' isothermal remanent magnetization (HIRM), and anhysteretic remanent magnetization susceptibility (χ ARM ) were slightly higher in the winter than in the summer samples. However, the differences between the summer and winter χ LF , HIRM, χ ARM , and SIRM values were not significant.
The SIRM and χ LF values of all PM 2.5 samples correlated linearly (R 2 = 0.9364), as shown in Fig. 2a, and indicated that the dominant magnetic minerals in the PM 2.5 samples were ferrimagnetic 48 . The χ ARM may constrain the sizes of ferrimagnetic grains because of its sensitivity to single-domain and pseudo-single-domain magnetic particles 49 . The significant correlation between the χ ARM and χ LF values (R 2 = 0.958, Fig. 2b) suggested that the significant increases in the χ LF values were closely related to coarse ferrimagnetic phases. This was confirmed by the Day plot (Fig. 3a), which showed that either a pseudo-single-domain mineral or a mixture of single-domain and multiple-domain magnetite dominated the PM 2.5 samples collected in both seasons. The results in a study of PM 2.5 in Beijing were similar 34 .
The temperature-dependent susceptibility (χ−T) cycles and hysteresis loops for representative winter and summer samples are shown in Fig. 3. The χ−T curves were used to identify the Curie temperature for typical PM 2.5 samples in summer and winter (Fig. 3b,c). A Curie temperature of ~580 °C was determined for the PM 2.5 samples based on χ−T measurements. This value was compatible with the results expected for magnetite. The χ value decreased slightly between 600 and 700 °C, revealing the presence of hematite. The χ value increased before a temperature of 300 °C was reached, possibly because fine-grained ferrimagnetic particles gradually became unblocked. A decrease in the χ value after 300 °C is generally interpreted as reflecting the conversion of maghemite to hematite 50 . The hysteresis loops for the PM 2.5 sample in a field of ~300 mT were thin, closed, and approached magnetic saturation (Fig. 3d,e), indicative of the presence of magnetic minerals in the samples. The S-ratios were high (96.33-98.07% in summer and 95.43-98.32% in winter), which confirmed the presence of soft, low-coercivity magnetite-type ferromagnetic minerals in the PM 2.5 samples 47 . The L-ratio values were low and relatively stable (17.1-25.5% in summer and 14.6-30.4% in winter), which also indicated that soft ferromagnetic minerals dominated the magnetic remanence ( Table 2).
The optical characteristics of representative PM 2.5 samples were determined using a FEI Quanta 250 FEG scanning electron microscope (Thermo Fisher Scientific, Waltham, MA, USA) with an AZtec X-Max 80 energy dispersive X-ray spectroscopy (EDX) system (Oxford Instruments, Abingdon, UK). Most of the magnetic iron oxide particles were either spherical (Fig. 4a,c) or angular (Fig. 4b). EDX analysis revealed that the Fe content of these magnetic particles was 29.2-85.6%. The spherical particles were probably produced during the combustion  Relationship between magnetic properties and trace metal concentrations. The mass concentrations of trace metals can reflect the variations in particle sources and allow particle toxicities to be assessed accurately. In efforts to control atmospheric pollution, it is essential to obtain trace metal mass concentration data. We therefore used trace metal mass concentrations (in μg/g) to assess the relationships between the magnetic properties and trace metal contents of the samples. As shown in Supplementary Table S2,  The Ni concentration correlated positively and significantly with the χ ARM /SIRM ratio but negatively with the SIRM/χ LF ratio. The correlation coefficients were higher for the relationships between the SIRM values and the Fe and Pb concentrations than for the relationships between the χ LF value and the concentrations of these same metals. This could be explained by the influence on the SIRM of the ferrimagnetic and incomplete anti-ferromagnetic mineral concentrations, with the contributions of the latter under-represented and the contributions of pedogenic superparamagnetic particles not taken into account 26,51 . This finding agreed with the results of previous studies and suggested that, as an indicator of the presence of trace metals, the SIRM value is more effective than the χ LF value 17,52,53 . The PM 2.5 concentration correlated positively with the χ ARM /χ LF ratio and the L-ratio but negatively with the S-ratio.
The interrelationships of the PM 2.5 concentrations, trace metal contents, and magnetic parameters were further examined in a multiple regression analysis. As shown in Table 3, the training R values of Cr and Fe were > 0.7, in contrast to the low correlation coefficients of As, Cd, Ni, Sr, and Ti (R < 0.5). The test R values of Fe, Pb, Sr, Mn, and Cd were relatively higher (R > 0.5). The pollution levels for most of the trace metals (except Co, the concentration of which did not correlate significantly with either the PM 2.5 concentration or the magnetic parameters) could be linearly determined from the PM 2.5 concentration and the magnetic parameters.
The relationships between the PM 2.5 concentrations, trace metal concentrations, and magnetic parameters were analyzed in a principal components analysis. The results are shown in Table 4    and χ ARM values, the L-ratio, and the S-ratio and explained 22.97% of the total variance. Only the S-ratio had a negative factor 2 value. Factor 3 explained 10.94% of the total variance and was dominated by the Ni concentration and the χ ARM /SIRM and SIRM/χ LF ratios. The Ni concentration and the χ ARM /SIRM ratio had positive factor 3 values, but the SIRM/χ LF ratio had a negative factor 3 value. The χ ARM /SIRM and SIRM/χ LF ratios are mainly related to the magnetic mineralogy and grain size. Factor 4 explained 8.09% of the total variance and was dominated by the L-ratio.
Environmental implications. The use of magnetic measurements to determine trace metal contamination assumes a relationship between the magnetic parameters and the trace metal concentrations. Specifically, the two most important factors are the relationships between the magnetic parameters and the Fe concentration and between the Fe concentration and the concentrations of the other trace metals 17 . In our study, the Fe concentration correlated strongly with both the concentrations of most of the anthropogenic trace metals (e.g., Cd, Mn, and Pb) and the magnetic parameters, that is, the χ LF , SIRM, HIRM, and χ ARM values, indicating that the magnetic parameters could be used to assess trace metal pollution in PM 2.5 . However, Ni, Co, and V have multiple sources such that their concentrations correlated poorly with the magnetic parameters. According to these results, the link between magnetic properties and trace metal concentrations mainly depends on the sources of the trace metals and the magnetic minerals. In previous studies, the relationships between magnetic properties and trace metal concentrations were typically investigated using the χ LF and SIRM values 26 . Our study showed that the χ ARM /SIRM and SIRM/χ LF ratios correlated strongly with the Ni and therefore that comprehensive magnetic measurements should be performed before trace metal contamination is assessed.
Conclusions. PM 2.5 samples were collected in Nanjing in summer and winter. The PM 2.5 concentrations in almost all of the winter samples and a few of the summer samples exceeded the NAAQS. The mean As and Ni concentrations were significantly higher in the winter than in the summer samples (P < 0.05). The As concentrations in all of the PM 2.5 samples were higher than the limits set by the NAAQS and WHO. The Cd, Ni, and Pb concentrations in a few of the winter PM 2.5 samples exceeded the NAAQS and WHO limits. These findings led us to conclude that some trace metals (As, Cd, Cr, Cu, Ni, Pb, and Zn) in our PM 2.5 samples came from anthropogenic sources but that Co, Sr, and V (EF < 10) predominantly came from natural sources. The stable Pb isotope ratios pointed to coal combustion emissions as the main source of Pb in the PM 2.5 from the summer samples, and to smelting and coal combustion emissions as the main sources of Pb in the winter PM 2.5 samples.  Detailed analyses of the magnetic properties of the PM 2.5 revealed that the dominant magnetic minerals were pseudo-single-domain magnetite in soft, low-coercivity particles. The magnetic minerals in the PM 2.5 mainly consisted of magnetite and hematite. Scanning electron microscopy-EDX analysis showed that most of the iron oxide magnetic particles were spherical and angular particles with a high Fe content. The magnetic properties (χ LF , SIRM, HIRM, and χ ARM values) correlated strongly with the concentrations of trace metals derived from anthropogenic activities (Cd, Fe, Mn, and Pb) but poorly with the concentrations of trace metals derived from natural processes. Multiple linear regression models for the Cr and Fe had high correlation coefficients (training R > 0.7), and those for the As, Cd, Ni, Sr, and Ti low correlation coefficients (training R < 0.5), using the PM 2.5 concentrations and magnetic parameters as decision variables. Taken together, these results demonstrate the potential of magnetic measurements as efficient proxies to assess trace metal pollution in urban PM 2.5 .

Methods
Sample collection. Nanjing City (118° 46′ E, 32° 03′ N) is the capital of Jiangsu Province and the second largest city in the Yangtze River delta region. It covers approximately 6587 km 2 and has a population of 8.2 million. Nanjing has a northern subtropical monsoon climate, and the wind predominantly comes from the southeast in summer and the northwest in winter. There are five dominant industries in Nanjing, the automobile, electric power production, electronics, petrochemical, and steel industries. In addition, Nanjing is the main transport hub in eastern China. PM 2.5 samples were collected in Aoti, in the center of the new urban districts of Nanjing (see Supplementary Fig. S3).
One sample was collected during the day (from 08:00 to 18:00) and another during the night (from 19:00 to 07:00) on each consecutive clear day between 8 and 23 July 2013 (i.e., in summer) and between 7 and 17 January 2014 (i.e., in winter). The samples were collected using a high-volume air sampler (TE6070, Tisch Environmental, Cleves, OH, USA) and at a flow rate of 1.13 m 3 /min. Each sample was collected on a Whatman quartz microfiber filter measuring 203 mm × 254 mm (GE Healthcare Bio-Sciences, Pittsburgh, PA, USA). No samples were collected when the weather was wet and windy (which can occur in both summer and winter in Nanjing) to ensure that the samples were representative. The sampler was placed on the rooftop of an open three-story building (Fig. 1). A total of 42 PM 2.5 samples were collected. Each filter was conditioned for 48 h in a desiccator at 25 °C and 40% relative humidity before and after sampling, and then weighed using a microbalance (Mettler-Toledo, Greifensee, Switzerland).
It should be noted that pollution controls implemented to improve air quality for the 2014 Youth Olympic Games in Nanjing (16-28 August 2014) decreased the PM 2.5 concentrations in the atmosphere. Before the Youth Olympic Games started, the Nanjing local government implemented several pollution control measures, including closing approximately 2630 construction sites, decreasing the outputs of heavy industrial factories (such as iron and steel plants and petrochemical plants) by 20%, and banning high emission vehicles (such as trucks and heavy-duty vehicles). Industrial plants emitting large amounts of pollutants were also closed in 22 nearby cities in order to cooperate with the Nanjing local government 41 .
Magnetic measurements. The blank filters and corresponding PM 2.5 samples were weighed, after which each filter was placed in a nonmagnetic plastic cylinder with a volume of 11.15 cm 3 . The magnetic susceptibility at a low frequency (976 Hz) (χ LF ) of each filter was measured and then mass normalized to obtain the χ LF value using an MFK1-FA Kappabridge system (AGICO, Brno, Czech Republic) at room temperature. Anhysteretic remanent magnetization (ARM) was applied using a DTECH 2000 AF demagnetizer (ASC Scientific, Carlsbad, CA, USA), a peak alternating field of 100 mT, and a direct current bias field of 0.04 mT. The measurements were expressed as the ARM susceptibility (χ ARM ), obtained by dividing the remanence by the steady field value. Isothermal remanent magnetization (IRM) experiments were performed using an MMPM10 pulse magnetizer (Magnetic Measurements, Aughton, UK). IRM was measured using a JR-6 dual speed spinner magnetometer (AGICO). The IRM measured in a field of 1 T is called the saturation isothermal remanent magnetization (SIRM) 33 . The IRM measured in a reverse field was expressed as a percentage of the reverse saturation of the SIRM, and the S-ratio and L-ratio were calculated using the equations S-ratio = −IRM −300 mT /SIRM and L-ratio = (SIRM + IRM −300 mT )/(SIRM + IRM −100 mT ). The 'hard' IRM (HIRM) was calculated using the equation: Mass-specific HIRM = [(SIRM + IRM −300mT )/2]/mass 19,55 . Hyteresis loops and magnetic hysteresis parameters, including the saturation magnetization (M s ), saturation remanent magnetization (M rs ), coercivity (B c ), and remanent coercivity (B cr ), of representative samples were measured at room temperature using an MMVFTB multi-purpose magnetic measurement system. High-temperature thermomagnetic analyses were performed using an MFK1-FA Kappabridge system equipped with a CS-3 high-temperature furnace (AGICO). Each sample was heated from room temperature to 700 °C and then cooled in an argon atmosphere to room temperature. The magnetic measurements were performed at the State Key Laboratory of Estuarine and Coastal Research, East China Normal University, Shanghai, China.
Total trace metal analysis. Each PM 2.5 sample was digested in ultra-pure concentrated HNO 3 and 30% H 2 O 2 to allow the total As, Cd, Co, Cr, Cu, Fe, Mn Ni, Pb, Sr, Ti, V, and Zn concentrations to be determined. The method was a slight modification of US Environmental Protection Agency method 3050B 41 . Briefly, about 1/16 of a quartz filter was cut into even smaller pieces (<2 mm 2 ) using a clean ceramic knife. The pieces were then immersed in 10 mL of ultra-pure concentrated HNO 3 (final acid to filter volume ratio of 1:1), ensuring that the filter pieces containing the PM 2.5 were fully in contact with the digestion solution. The mixture was digested at 105 °C for 5 h, after which a further 10 mL of concentrated HNO 3 was added and the mixture was digested again. The digest was allowed to evaporate almost to dryness and allowed to cool. After the addition of 2 mL of 30% H 2 O 2 , the solution was evaporated to <1 mL. The extract was then diluted to 50 mL with Milli-Q water. The concentrations of the elements of interest in the extracts were determined using an ICP-mass spectrometer (Perkin SCIENTIfIC RePoRTS | 7: 9865 | DOI:10.1038/s41598-017-08628-0 Elmer, Waltham, MA, USA). Quality assurance and quality control procedures included analyzing blank quartz microfiber filters, analytical duplicates, and standard reference material SRM 1648a (urban particulate matter) (US National Institute of Standards and Technology, Gaithersburg, MD, USA). The recovery rates were within ±10%, as recommended by the US Environmental Protection Agency 56 . The concentration of each trace element in each PM 2.5 sample was corrected by subtracting the mean concentration of the element in the blanks.
Stable isotope ratios in PM 2.5 . The stable Pb isotope ratios 206 Pb/ 204 Pb, 207 Pb/ 206 Pb, and 208 Pb/ 206 Pb in the total Pb extracted from each PM 2.5 sample were determined using ICP-mass spectrometry. The trace metal extract of each PM 2.5 sample was diluted with 0.1 M high-purity HNO 3 to a Pb concentration of 15 μg/L. The instrument parameters were 190 sweeps/reading, 1 reading/replicate, a dwell time of 40 ms for 204 Pb, and a dwell time of 25 ms for 206 Pb, 207 Pb, and 208 Pb. Each PM 2.5 sample was measured 10 times, and the mean Pb isotope ratio and the relative standard deviation were then calculated. The isotope intensities of the blank quartz microfiber filter samples were subtracted from the isotope intensities of each PM 2.5 sample. The common Pb isotope standard reference material SRM 981 (US National Institute of Standards and Technology) was analyzed as a quality control procedure. The SRM 981 extract was diluted to a Pb concentration of 15 μg/L and run after every five samples to provide ratio correction factors to allow mass bias correction factors for the isotope ratios to be determined. The relative standard deviation for 10 replicate PM 2.5 samples was generally <0.5% for both the 207 Pb/ 206 Pb and the 208 Pb/ 206 Pb ratios.
Back trajectory analysis. The Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT-4.8) model, developed by the US National Oceanic and Air Administration Air Resources Laboratory, was used to calculate 72-h air mass back trajectories using 3-hourly archived meteorological data from the US National Centers for Environmental Prediction Global Data Assimilation System 57 . Back trajectories were calculated for both sampling periods at heights of 100, 500, and 1000 m above ground level.
Statistical analyses. The enrichment factor (EF) was calculated to evaluate the degrees to which the PM 2.5 samples were polluted with trace metals 58,59 . The EF was calculated using Eq. (1) in the Supplementary Information. One-way analysis of variance, Pearson's correlation coefficient analysis, and principal components analysis to evaluate the data and multiple linear regressions were performed using SPSS version 16.0 for Windows. Multiple linear regression analysis was used to develop a linear model according to Eq. (2) (see the Supplementary  Information). The data were partitioned at random into two sets: 75% for the training set and 25% for the test set.