IntroductionAmbient fine particulate matter (PM2.5) has long been associated with respiratory and cardiovascular morbidity and mortality (Dockery et al., 1993; Franklin et al., 2007; Pope et al., 1995). Recently, PM2.5 and related health effects have been shown to vary across the USA (Bell et al., 2009; Sampson et al., 2013), highlighting the importance of researching the impact of compositional differences in ambient PM2.5. It is particularly important to understand differences in components that are relevant to human health, such as metallic and organic species (Melaku et al., 2008; Ravindra et al., 2001; Schaumann et al., 2004). As epidemiological evidence of PM2.5-associated health effects continues to be strengthened, toxicology studies to understand the mechanisms behind these outcomes and impacts of compositionally differing PM2.5 have advanced.
Toxicology studies allow for research into PM2.5 health effects while avoiding confounders present in many epidemiological studies, such as lifestyle and occupational factors (Jerrett et al., 2005). In order to capture the compositional complexity of PM2.5, ambient samples must be used; however, this requires collection of samples predominantly through concentrator systems or filter-based methods (Ghio & Huang, 2004). Concentrated ambient particles (CAPs) provide PM2.5 samples that maintain ratios of ambient mixtures while increasing the mass to allow for both in vitro and in vivo studies (Ghio & Huang, 2004). While CAPs provide a number of benefits to research, they require an expensive concentrator system that is fixed at a single sampling location, lacking the potential to study spatial differences in ambient PM2.5 concentration and composition (Matte et al., 2013).
Filter sampling allows for collection of ambient PM2.5 that may vary by location and source while utilizing a relatively low-cost method conducive to a variety of air sampling equipment systems (Kundu & Stone, 2014). An important consideration is the translation of PM2.5 on the filter, to a liquid suspension of PM2.5 that can be used in toxicology experiments. This process is integral to maximizing extraction efficiency while maintaining compositional integrity, so that the final extraction solution yields sufficient PM2.5 mass that remains representative of ambient PM2.5.
Preparation of PM2.5 is typically a multistep process involving removal from the filter into solution, recovery of dry PM2.5, and re-suspension into media appropriate for the toxicology application. A variety of extraction techniques have been implemented in toxicology research, which differ by the type of filter used for ambient collection, removal procedure and solvent, concentration method, and the media used for re-suspension. Table 1 summarizes a literature review of preparation procedures and emphasizes the variability of extraction methods that have been used.
Variation in extraction techniques between research groups creates a potential for bias, where findings may be dependent on the extraction procedures used rather than on the characteristics of ambient material (Bein & Wexler, 2014). Well-characterized extraction solutions would avoid these biases as the exact concentration and composition of PM2.5 used would be known, enabling a more accurate interpretation of exposure studies. Thus far, a limited number of toxicology studies using ambient PM2.5 have reported chemical characterization of both metals and organics in extraction solutions (Huang et al., 2014; Lauer et al., 2009; Verma et al., 2012). Here a novel method was developed to measure compositional differences in PM2.5 between collected ambient material and the corresponding extraction solutions.
ResultsSampling location differencesMasses for PM2.5, metals, and organics were determined for each of the five locations. PM2.5 mass was determined at three stages: (1) ambient material collected over the sampling period (“PMamb”); (2) recovered material measured based on pre- and post-removal filter weights (“PMrem”); and (3) recovered material concentrated and re-suspended into DMEM (“PMext”). Metals and organics masses were determined at stage 1 (“metalsamb, organicsamb”) and 3 (“metalsext, organicsext”). Ambient mass collected varied between sampling sites, but trends between PMamb, PMrem, and PMext were similar across locations (Figure 1).
Metalsamb and metalsext (Figure 2A) were calculated by summing the masses of all species analyzed. Metalsamb did not correspond to PMamb, yet the highest PMamb sampling location also had the highest metalsamb. Metalsamb and metalsext also did not trend together indicating that extraction differences between sampling sites impacted metalsext. Interestingly, the lowest metalsext was observed at the location with the highest PMamb and metalsamb.
Organicsamb and organicsext were quantified by summing all species analyzed and variability was observed between all locations (Figure 2B). Similarly to ambient metals, organicsamb did not trend with PMamb. However, unlike metals, the location with the highest organicsamb was not from the sampling location with the highest PMamb. All organicsext were less than organicsamb and varied between sampling locations independent of PMamb. As with metalsext, the effect of extraction differences between sampling locations was observed in organicsext.
Extraction efficiencyAmbient masses of PM2.5, metals, and organics were compared to extracted masses to determine the percent extracted. Removal of total PM2.5 was 98.0±1.4% following sonication; however, extraction efficiencies following concentration and re-suspension in DMEM were found to be substantially lower at 80.2±0.8% of PMamb (Figure 3).
Overall extraction efficiency for metals (Table 3) was 47.2±22.3%, with extraction efficiencies for specific metals ranging from 0.7 (for Ce) to 73.4% (for Na). High variability in extraction efficiency was also observed for specific metals between sampling locations (i.e., SD±35.8% for Ni). All averaged extraction efficiencies were less than 100% removal except for three trace metals that were present in DMEM (Ca, Mg, and P); these components were excluded from calculations of total metals. Contributions to total ambient metals for Ca, Mg, and P were 2.2, 1.1, and 0.0%, respectively.
Extraction efficiencies for organics (Table 3) are displayed for the five compounds detected in both ambient and extracted samples: 1-methyl phenanthrene (1MP), acenaphthylene (Acy), benzo[b]fluoranthene (BbFl), benzo[ghi]perylene (BghiPer), and indeno[1,2,3-cd]pyrene (Ipyr). Variability of efficiency was observed between species (Ipyr to 1MP: 17.0–101.5%) as well as between sampling locations for individual species (i.e., BghiPer SD±43.9%). Extraction efficiency for total organics measured was 24.8±14.5%. All hopanes (n=10), steranes (n=4), and other organic compounds (n=15) measured were found at varying concentrations in ambient samples but were below the limit of detection in all extraction solutions, suggesting near complete loss during the extraction process.
Expected masses for constituents were calculated for extraction samples based upon PM2.5 mass and ambient composition. The expected values were compared to actual masses recorded through analysis of extraction solution (Tables 4 and 5).
Relationship of constituents to total PM<sub>2.5</sub>The relative contribution of measured metals and organics, as a percent of PM2.5 mass, was determined for both ambient and extracted samples (Figure 4). In ambient samples, the contribution of metals to PMamb was 120 times higher than the contribution of organics. However, in extracted samples, the contribution of metals increased to over 705 times higher than organics.
CorrelationsAmbient to extractedCorrelations of PM2.5 to metal and organic components were calculated to compare how PM2.5 mass and constituents were related for both ambient and extracted samples (Table 6). Statistically significant positive correlations were observed between PMamb and Fe and Zn, while none of the extracted constituents exhibited a statistically significant correlation with PMext. Marked differences were observed between ambient and extracted correlations of constituents to PM2.5 mass (reported as ambient: extracted) for metals (0.911: −0.219), organics (0.805: −0.083), and a number of specific constituents including Al (0.967: −0.299), Cr (0.725: −0.350), Zn (0.975: −0.166), and Acy (0.734: −0.243).
To determine how specific components of PM2.5 related between ambient and extracted samples, correlation coefficients were determined for each component (Table 6). Several extracted constituent values were negatively correlated with ambient measurements of the same constituent, including: total metals, total organics, and specific components such as Al, S, and Sr. Positive correlations were observed for a number of components including: Cs, Fe, Mn, Sn, 1MP, and BbFl. Statistically significant positive correlations between ambient and extracted values were present for PM2.5 mass and Tb.
Extraction percent to ambient characteristicsCorrelations between calculated extraction percentages of specific components to total PM2.5 were made to investigate potential trends in extraction efficiency based upon PMamb (Table 6). The percent of total mass extracted was significantly positively correlated with PMamb. While both the percent of extracted metals and organics were negatively correlated with PMamb. These correlations suggest that as ambient PM2.5 mass increases, the efficiency of extraction for total metals and organics decreases. Similar trends were seen in individual constituents including: Al, Cu, Fe, Pb, and Acy. However, not all components measured had negative correlations between extraction efficiency and total PM2.5, these included: Cd, Cr, Sn, and BbFl.
DiscussionPM<sub>2.5</sub> massThe percent of mass removed via sonication was consistent across sampling locations, which is a characteristic necessary to avoid a methods bias. Less inter-filter variability was seen in removal efficiency than has previously been reported, where efficiency ranged from 59% to 95% (Imrich et al., 2000). Increased consistency in removal efficiency in this research is likely due to more deliberate selection of solvents based on anticipated chemical characteristics of PM2.5 components, as well as refinement of sonication methods. Extraction protocols using water as the sole solvent are common, and while effective for removal of water-soluble components and approximately 75% of PM2.5 mass, water is not effective for extraction of nonpolar species, including many organics compounds (Hawthorne et al., 2000; Longhin et al., 2013; Watterson et al., 2007). It should be noted that a portion of studies neglect to report removal percentages, limiting inter-method comparisons. Based solely on mass removal from the filter, the methods outlined here maintained a high PM2.5 yield and were consistent between filters and sampling locations.
The significant positive correlation between ambient and extracted PM2.5 mass suggests that ambient mass loadings do not impact the extraction of total PM2.5. Consistent extraction independent of mass makes the outlined methods translatable to many regions and sampling timescales. The methods are also effective in reducing release of filter material into the extracted solutions, as no significant loss of mass was observed with blank sonicated filters. The lack of observed fiber loss is likely due to a combination of the filter type used and decreased time and intensity of sonication. Previous methods have utilized potentially destructive probe sonication or extended sonication times, which can necessitate filtering of samples to remove fibers but also introduces a potential loss of PM2.5 (Godri et al., 2011; Huang et al., 2014; Riva et al., 2011; Van Winkle et al., 2015).
Components of PM<sub>2.5</sub>Differences in extraction efficiency between total PM2.5 and constituents of PM2.5 demonstrate a key limitation of filter extraction methods, discussed below. Importantly, a vast majority of the PAHs (15 of the 20 analyzed), hopanes, and steranes were not extracted at any of the locations. While loss in organics was not unexpected due to the volatility of the compounds (EPA, 2014), quantifying the shift from ambient contributions is useful to establish differences from filter samples.
A number of individual components, total organics, and total metals were inversely related to corresponding ambient masses, a result of decreased extraction efficiencies as mass increased. Ideally, ambient and extracted components would be equally correlated to total mass, indicating that composition of extracted PM2.5 was similar to that of ambient PM2.5. However, positive correlations in ambient samples alone suggest that the relative composition of the ambient material is changed during the extraction process. Additionally, as ambient PM2.5 mass increases, the percentage of metals and organics extracted decreases.
Loss between removal and concentrationTranslating the removal solution into dry particulate material is an imperative step before re-suspension into cell culture media to create an extraction solution for toxicology experiments. In this study, this process created a significant loss of mass and presumably loss of compounds that were volatile or soluble in the removal solution (9:1 methanol in water). However, characterization was only performed following re-suspension in cell culture media; additional characterizations of the removal and extraction solutions could identify at what point in the process the losses occurred. Better characterization of these losses is essential to accurate research, as extraction percentages are frequently reported as total mass removed from filter, without consideration of losses that occur during the subsequent preparation steps.
Impacts on toxicology applicationsStudies using ambient PM2.5 extracted from filters are an integral component for assessing biological impacts of PM2.5 both in vitro and in vivo. In some cases, responses are correlated to ambient concentrations, without regard for changes that occur during the extraction process (de Kok et al., 2005). The shift in relative contributions of specific components to total PM2.5 demonstrates that the resultant extraction solution in this work is not representative of the ambient mixtures. Recently, different extraction methods were found to result in distinct biological impacts (Van Winkle et al., 2015). Identifying the specific components of PM2.5 that are not representatively extracted by protocols can suggest mechanisms responsible for the varying biological responses. In this work we identified the loss of numerous health-relevant compounds including: Cr, Fe, Ni, Pb, Zn and 10 of the 16 EPA Priority PAHs (Chen & Lippmann, 2009). There is a need to further understand what effects these losses have on subsequent toxicology analyses.
Extraction efficiencies of components of PM2.5 were shown in this research to vary between sampling locations. Similar investigations found that extraction efficiencies were influenced by the source mixture and therefore composition of PM2.5 (Bein & Wexler, 2014). These findings are particularly important for studies examining multiple sampling locations or the impacts of mixed sources. With inconsistent extraction efficiency, the variation of ambient samples will be obscured or lost, and toxicology results will not be representative of the actual exposure of interest. An additional concern is the finding that extraction efficiencies of metals and organics are inversely related to total PM2.5 loadings. High mass loadings are necessary in toxicology studies to provide adequate material for exposures, but it is ineffective to collect such loadings when they decrease the yield of metal and organic species.
In this study, extraction was performed on samples with spatially varying ambient PM2.5 concentration and composition. The impacts of temporal and seasonal variation in PM2.5 composition on extraction efficiencies were not investigated here, but future research in these areas would strengthen the correspondence to studies using temporally variant ambient samples. Furthermore, results from this study are only generalizable to the methods utilized, and efficiencies will differ based upon the extraction procedures employed (Bein & Wexler, 2015). In this study, PM2.5 was re-suspended into cell culture media; to accurately determine how composition may be impacted by re-suspension in different toxicology medium (i.e. saline, PBS, or water), further studies should be conducted. However, based on this and previous works, it is clear that complete extraction has not been achieved using any current methodology; therefore, extraction solutions will differ compositionally from ambient source material, and the issues highlighted by this extraction protocol are likewise of concern with other methods (Akhtar et al., 2014; Happo et al., 2010).
Characterization of PM2.5 was performed only on ambient material and final extraction solutions; thus, changes in composition during the intermediate stages are unknown. Identifying specific steps in extraction procedure that is most impactful on the recovery of PM2.5 components would help to establish refined procedures that maintain ambient compositions. While this work has begun to uncover compositional differences, it examined only a subset of the key components of PM2.5 other substantial contributors to mass including inorganic ions and total elemental and organic carbon would further elucidate the compositional changes following extraction.
ConclusionsThis research has outlined a method for the extraction of PM2.5 from filter samples, which was effective in high mass recovery while maintaining filter integrity. Comparison of ambient and extracted samples suggests that the method was more effective in recovering metals in the extraction solution compared to organics. To the authors’ knowledge, only one study has performed a well-characterized extraction solution analysis, and this research highlighted the variance in extraction of components of PM2.5 based on the extraction procedures implemented (Bein & Wexler, 2015). However, this current work is the first to compare components measured in PM2.5 filter extract with those measured on collocated ambient filters from multiple sampling locations. The narrow understanding of alteration to PM2.5 composition as a result of extraction is a limitation that persists throughout a vast majority of toxicology research using PM2.5 collected on filters. Further awareness of the underlying mechanisms for the observed compositional shifts, in addition to the adoption of standardized extraction techniques that more efficiently extract all components of PM2.5, would allow for biological impact studies that are more readily translatable to ambient exposures, and would facilitate comparisons between studies.