Evaluation of anthropogenic secondary organic aerosol tracers from aromatic hydrocarbons

Products of secondary organic aerosol (SOA) from aromatic volatile organic compounds (VOCs) – 2,3-dihydroxy-4-oxopentanoic acid, dicarboxylic acids, nitromonoaromatics, and furandiones – were evaluated for their potential to serve as anthropogenic SOA tracers with respect to their (1) ambient concentrations and detectability in PM2.5 in Iowa City, IA, USA; (2) gas–particle partitioning behaviour; and (3) source specificity by way of correlations with primary and secondary source tracers and literature review. A widely used tracer for toluene-derived SOA, 2,3-dihydroxy-4-oxopentanoic acid was only detected in the particle phase (Fp = 1) at low but consistently measurable ambient concentrations (averaging 0.3 ng m−3). Four aromatic dicarboxylic acids were detected at relatively higher concentrations (9.1–34.5 ng m−3), of which phthalic acid was the most abundant. Phthalic acid had a low particlephase fraction (Fp = 0.26) likely due to quantitation interferences from phthalic anhydride, while 4-methylphthalic acid was predominantly in the particle phase (Fp = 0.82). Phthalic acid and 4-methylphthalic acid were both highly correlated with 2,3-dihydroxy-4-oxopentanoic acid (rs = 0.73, p = 0.003; rs = 0.80, p < 0.001, respectively), suggesting that they were derived from aromatic VOCs. Isophthalic and terephthalic acids, however, were detected only in the particle phase (Fp = 1), and correlations suggested association with primary emission sources. Nitromonoaromatics were dominated by particle-phase concentrations of 4-nitrocatechol (1.6 ng m−3) and 4-methyl-5-nitrocatechol (1.6 ng m−3) that were associated with biomass burning. Meanwhile, 4-hydroxy-3-nitrobenzyl alcohol was detected in a lower concentration (0.06 ng m−3) in the particle phase only (Fp = 1) and is known as a product of toluene photooxidation. Furandiones in the atmosphere have only been attributed to the photooxidation of aromatic hydrocarbons; however the substantial partitioning toward the gas phase (Fp≤ 0.16) and their water sensitivity limit their application as tracers. The outcome of this study is the demonstration that 2,3-dihydroxy-4-oxopentanoic acid, phthalic acid, 4-methylphthalic acid, and 4-hydroxy-3-nitrobenzyl alcohol are good candidates for tracing SOA from aromatic VOCs.

Abstract.Products of secondary organic aerosol (SOA) from aromatic volatile organic compounds (VOCs) -2,3-dihydroxy-4-oxopentanoic acid, dicarboxylic acids, nitromonoaromatics, and furandiones -were evaluated for their potential to serve as anthropogenic SOA tracers with respect to their (1) ambient concentrations and detectability in PM 2.5 in Iowa City, IA, USA; (2) gas-particle partitioning behaviour; and (3) source specificity by way of correlations with primary and secondary source tracers and literature review.A widely used tracer for toluene-derived SOA, 2,3-dihydroxy-4-oxopentanoic acid was only detected in the particle phase (F p = 1) at low but consistently measurable ambient concentrations (averaging 0.3 ng m −3 ).Four aromatic dicarboxylic acids were detected at relatively higher concentrations (9.1-34.5 ng m −3 ), of which phthalic acid was the most abundant.Phthalic acid had a low particlephase fraction (F p = 0.26) likely due to quantitation interferences from phthalic anhydride, while 4-methylphthalic acid was predominantly in the particle phase (F p = 0.82).Phthalic acid and 4-methylphthalic acid were both highly correlated with 2,3-dihydroxy-4-oxopentanoic acid (r s = 0.73, p = 0.003; r s = 0.80, p < 0.001, respectively), suggesting that they were derived from aromatic VOCs.Isophthalic and terephthalic acids, however, were detected only in the particle phase (F p = 1), and correlations suggested association with primary emission sources.Nitromonoaromatics were dominated by particle-phase concentrations of 4-nitrocatechol (1.6 ng m −3 ) and 4-methyl-5-nitrocatechol (1.6 ng m −3 ) that were associated with biomass burning.Meanwhile, 4-hydroxy-3-nitrobenzyl alcohol was detected in a lower concentration (0.06 ng m −3 ) in the particle phase only (F p = 1) and is known as a product of toluene photooxidation.Furandiones in the atmosphere have only been attributed to the photooxidation of aromatic hydrocarbons; however the substantial partitioning toward the gas phase (F p ≤ 0.16) and their water sensitivity limit their application as tracers.The outcome of this study is the demonstration that 2,3-dihydroxy-4-oxopentanoic acid, phthalic acid, 4-methylphthalic acid, and 4-hydroxy-3-nitrobenzyl alcohol are good candidates for tracing SOA from aromatic VOCs.

Introduction
Secondary organic aerosol (SOA) accounts for a large but undefined fraction of organic matter in PM 2.5 , forming through the photooxidation of biogenic and anthropogenic volatile organic compounds (VOCs) in the gas phase yielding lowvapour-pressure products that partition into the particle phase (Kroll and Seinfeld, 2008;Hallquist et al., 2009).The global fluxes of anthropogenic SOA are poorly constrained and highly uncertain, with a wide range of estimates from 2 to 25 Tg yr −1 (Volkamer et al., 2006;Henze et al., 2008).Measurements suggest that anthropogenic precursors form more SOA than predicted by models (Heald et al., 2005;Volkamer et al., 2006;Matsui et al., 2009), likely due to incomplete model representation of SOA formation pathways (Henze et al., 2008), partitioning (Donahue et al., 2006), ambient conditions (Ng et al., 2007), and precursors (Robinson et al., 2007;Fu et al., 2008).
A tracer-based approach has been useful in identifying aerosol sources and source apportionment (Schauer et al., 1996).SOA can be linked to its precursor VOC following the SOA tracer approach introduced by Kleindienst et al. (2007) in which ambient concentration of tracers (or the sum of thereof) are converted to secondary organic carbon Published by Copernicus Publications on behalf of the European Geosciences Union.
(SOC) or SOA mass yields using tracer-to-OC or tracer-to-SOA ratios, respectively, that were determined in chamber studies.For biogenic SOA, relatively well defined and established tracers are employed, such as methyltetrols for isoprene and β-caryophyllinic acid for β-caryophyllene (Kleindienst et al., 2007).In contrast, the tracer-based approach for aromatic SOA relies on a single molecule (2,3-dihydroxy-4oxopentanoic acid (DHOPA)) that is derived from toluene (Kleindienst et al., 2007).Advancing the tracer-based approach to anthropogenic SOA apportionment should involve expanding the number of available tracers, particularly those that form from aromatic VOCs other than toluene.
Chamber experiments have been conducted to identify SOA products formed during the photooxidation of aromatic precursors associated with anthropogenic sources, such as benzene, toluene, ethylbenzene, xylenes (BTEX), and lowmolecular-weight polycyclic aromatic hydrocarbons (PAHs).Furandiones have been identified as a major product of aromatic VOC photooxidation in the presence of NO x (Bandow et al., 1985;Forstner et al., 1997;Hamilton et al., 2003;Koehler et al., 2004).Nitromonoaromatics (e.g.nitrophenols, methyl-nitrophenols, nitrocatechols, and nitrosalicylic acids) are likewise products of aromatic VOC photooxidation in the presence of NO x (Forstner et al., 1997;Jang and Kamens, 2001;Hamilton et al., 2005;Sato et al., 2012;Irei et al., 2015), but some of these species have also been detected in the primary emission from vehicles (Tremp et al., 1993) and biomass burning (Iinuma et al., 2010).While nitromonoaromatics have been quantified in ambient aerosol (Dron et al., 2008;Kitanovski et al., 2012;Kahnt et al., 2013), the extent of their formation from primary and/or secondary sources has yet to be determined.Phthalic acid is a product of naphthalene photooxidation (Kautzman et al., 2010) and has been proposed as a tracer for naphthalene and methylnaphthalenes in PM 2.5 (Kleindienst et al., 2012).However, phthalic acid has also been observed in the emission of motor exhaust (Kawamura and Kaplan, 1987), and thus both primary and secondary sources can contribute to its ambient concentration.These three classes of compounds can be potentially used a tracers for SOA; however, further ambient studies are needed to evaluate their detectability, ambient concentrations, and origins.
There are many desired characteristics for a molecule to be used as a source tracer.First, it should be unique to the source of origin.For example DHOPA was previously identified as a unique product of toluene photooxidation in the presence of NO x (Kleindienst et al., 2004), and methyltetrols are unique to isoprene (Claeys et al., 2004a).Second, the tracer should be formed in reasonably high yields so it has sufficiently high concentrations in the atmosphere to allow for reliable quantification.Third, the tracer needs to be reasonably stable in the atmosphere, so that it is conserved between formation and collection at a receptor location.Fourth, an efficient SOA tracer should have a low vapour pressure so that it primarily partitioned to the particle phase, which min-imizes possible underestimation from loss to the gas phase.Thus, an effective SOA tracer will exhibit source specificity, consistent detectability, atmospheric stability, and partitioning to the aerosol phase.
In this work, we examine and evaluate the efficacy of nitromonoaromatics, furandiones, and aromatic dicarboxylic acid isomers as potential SOA tracers in terms of their ambient concentration, gas-particle partitioning, and source specificity through correlations with established tracers, including levoglucosan for biomass burning (Simoneit et al., 1999), hopanes for vehicular emissions (Schauer et al., 1999), and DHOPA for anthropogenic SOA (Kleindienst et al., 2004).Sample preparation procedures were optimized for the simultaneous extraction of primary and potential secondary source tracers, which were then quantified by gas chromatography-mass spectrometry (GC-MS).These methods were applied to measure the ambient concentrations and gas-particle distributions for analytes in fine particulate matter (PM 2.5 ) collected in Iowa City, IA, in the autumn of 2015.November was chosen for this study because, in a prior study at this site, biogenic SOA tracers were detected in this month (Jayarathne et al., 2016) and aromatic SOA tracers have a less pronounced seasonal variation than those that are biogenic (Shen et al., 2015;Ding et al., 2012;Lewandowski et al., 2008).Developing and evaluating these tracers provide additional tools for better understanding the contribution of aromatic VOCs to ambient aerosol and will help to expand the current knowledge about the composition and sources of ambient aerosol, particularly in urban and peri-urban environments.
2 Experimental methods 2.1 Field sampling Gas and particle (PM 2.5 ) samples were simultaneously collected daily for the period 4-15 November 2015 in Iowa City, IA, USA (41.6572 • N, 91.5035 • W).The sampler was installed on a wooden platform, and the inlet was positioned 3.5 m above ground level.The sampling site was surrounded by an agricultural field and a university parking lot.Sample collection was preformed using a medium-volume URG air sampler (3000B, URG Corp.) with a cyclone (URG) operating at a flow rate of 90 L min −1 .Air flow rate was monitored before and after sampling using a rotameter (Gilmont Instruments).PM 2.5 samples were collected on 90 mm quartz fibre filters (QFF; Pallflex ® Tissuquartz ™ , Pall Life Sciences) that were pre-cleaned by baking for 18 h at 550 • C. Gas samples were collected on 52 mm polyurethane foam (PUF) plugs placed after the filter holder (URG-2000-30-52PC).PUF plugs were pre-cleaned using acetone (HPLC grade, Sigma-Aldrich), hexanes, and acetonitrile (Optima-Fisher Scientific-Fisher Chemical) by a repeated compression extraction apparatus adapted from Rogge et al. (2011).This apparatus is composed of a thick-walled borosilicate glass cylinder equipped with a polytetrafluoroethylene (PTFE) valve and PTFE plunger that was used to compress the solvent out of the PUF.
Samples were collected for 23 h, and filter changing was performed at 08:00 (local time).After sampling, filters were transferred to Petri dishes, lined with pre-baked aluminum foil, and sealed with Teflon tape.PUF samples were transferred to pre-baked wide-mouth glass jars, capped with a Teflon-lined cap, and sealed with Teflon tape.Sampled filters and PUF were transported to the laboratory and stored frozen at −20 • C until analysis.One field blank was collected for every five samples following the same described procedure, except no air was pulled through the system.

Extraction
All glassware used in this experiment was first baked (500 • C for 5 h) to remove organic contaminants and then silanized (using 5 % solution of dichlorodimethylsilane in toluene) to minimize the sorption of analytes to the glass surface (Kitanovski et al., 2012).Filters and PUF were spiked with isotopically labelled internal standards, representing the different classes of organic compounds reported in this study.Adding internal standards prior to extraction corrects for loss of analyte during the extraction process, provided the internal standard adequately represents the chemical and physical properties of the analyte.Specifically, internal standards and their corresponding analytes were 3-nitrosalicylic acid-D 3 and 5-nitrosalicylic acid; 4-nitrophenol (4NP)-D 4 and other nitromonoaromatics; maleic anhydride-D 2 for 2,5furandione, and succinic anhydride-2,2,3,3-D 4 for the three other furandiones; levoglucosan 13 C 6 for levoglucosan; ketopinic acid (KPA) for DHOPA; phthalic acid-D 4 for aromatic dicarboxylic acids; and acenaphthene-D 10 , pyrene-D 10 , benzo[a]anthracene-D 12 , coronene-D 12 for PAHs.The use of KPA as an internal standard for DHOPA builds upon prior work by Kleindienst et al. (2007).
Filters were extracted sequentially with three 10 mL portions of acetonitrile using ultrasonication (Branson 5510, 137 W) for 15 min at 60 sonics per minute.The combined three extracts were reduced to 2 mL by rotary evaporation at 30 • C, 120 rpm, and 200 mbar.The reduced extracts were filtered with a 0.25 µm PTFE syringe filters and stored frozen at −20 • C until analysis.Immediately prior to analysis, the extracts were evaporated to 100 µL under a gentle stream of ultra-pure nitrogen at 30 • C. PUF samples were extracted by three cycles of repeated compression using acetonitrile; extracts were then combined, evaporated, filtered, and reduced to the final volume using the same conditions as filter extracts.

Instrumental analysis
OC and elemental carbon (EC) were measured by thermaloptical analysis (Sunset Laboratory Inc.) on a 1 cm 2 filter portion, following Schauer et al. (2003).
Organic species were analysed using an Agilent 7890A GC, coupled with 5975C MS (Agilent Technologies).2,3-Dihydroxy-4-oxopentanoic acid (Toronto Research Chemicals), phthalic acid isomers, levoglucosan, and biogenic SOA tracers were trimethylsilylated with N,Obis(trimethylsilyl)trifluoroacetamide (BSTFA-TMCS; Fluka Analytical 99 %) prior to analysis.The silylation reaction was performed by evaporating 10 µL of the extract under a gentle stream of nitrogen to dryness, adding 20 µL of the silylation agent and 10 µL of pyridine (Burdick & Jackson, Anhydrous), and heating to 70 • C for 3 h.A 2 µL aliquot of the silanized extract was introduced to the GC-MS equipped with a DB-5 column, electron ionization (EI) source (70 eV), and a GC inlet temperature of 300 • C. Nitromonoaromatics were also silylated using the same reagent but under different conditions, in which 20 µL of the extract was evaporated to dryness under a gentle stream of nitrogen, 10 µL of the silylation agent was added, and then the mixture was capped and heated for 90 min at 100 • C. The different silylation protocol used for nitromonoaromatics yielded more symmetrical peak shapes and higher intensities than the derivatization method used for levoglucosan and phthalic acid isomers that resulted in asymmetrical nitromonoaromatic peaks with low intensities.The GC injection volume was 1 µL, and the inlet conditions, column type, and MS parameters matched those previously described.Furandiones were analysed using the method developed in our previous work (Al-Naiema et al., 2017), and PAHs were analysed using a DB-5 column as described elsewhere (Al-Naiema et al., 2015).
Responses of the analytes were normalized to the corresponding isotopically labelled internal standards and quantified using a linear calibration curve with a squared correlation coefficient (R 2 ) ≥ 0.995.Analytical uncertainties were propagated from the standard deviation of the field blank value and 10 % of the measured concentration.For analytes not detected in the field blank, instrument detection limits were used in error propagation.All measurements were fieldblank-subtracted.Due to low recoveries of furandiones from PUF, gas-phase concentrations of furandiones were corrected for the recoveries of the authentic standards.

Particle-phase fraction calculation and model
The fraction of a species in the particle phase (F p ) was calculated from the ratio of concentration in the particle phase to the total concentration (sum of gas and particle), following Eq.(1). (1) F p was modelled using the gas-particle partitioning coefficient (Eq.2) from absorptive partitioning theory developed by Pankow (1994), and following Yatavelli et al. (2014): where C OA is the concentration of the organic aerosol (µg m −3 ) and k om is the partitioning coefficient (m 3 µg −1 ) described as in which R is an ideal gas constant (8.2 × 10 −5 m 3 atm mol −1 K −1 ); T is temperature (averaging 9 • C during this study); 10 6 is a unit conversion factor (µg g −1 ); P 0 L is the sub-cooled vapour pressure (atm), obtained from the Estimation Program Interface suite ™ version 4.11 from the Environmental Protection Agency (EPA); MW is the molar mass (g mol −1 ); and E is the activity coefficient (set equal to 0.3, 1, and 3).

Validation of gas-particle partitioning
PAHs with two to five rings span a range of high to low volatility, respectively.The accuracy of the measured gasparticle distributions were evaluated with PAHs that have been extensively discussed in the literature.The average fractions of PAH in the particle phase (F p ) measured in Iowa City, IA, USA (Fig. 1), were ∼ 5 % for 10-14 carbon atoms (two-three rings: naphthalene, acenaphthene, and anthracene), 14 % for 16 carbon atoms (four rings: pyrene), 59 % for 18 carbons (four rings: benzo(ghi)fluoranthene), and > 98 % for 20 carbon atoms (five rings: picene).The predicted F p values estimated using the Pankow absorption model (1994) following Eqs.( 1) and ( 2) and using parameters in Table S1 in the Supplement follow the same trend (Fig. 1, dashed line), with a systematic underestimation for the predicted F p by ≤ 12 % for most PAHs and 20 % for 18 carbons.Such an underestimation has been widely documented in comparison of theory to ambient partitioning studies and is attributed to the omission of PAH sorption on elemental carbon in the model (Dachs and Eisenreich, 2000;He and Balasubramanian, 2009;Wang et al., 2011).Ambient studies of gas-particle partitioning are influenced by many factors such as ambient temperatures (Terzi and Samara, 2004), relative humidity (Pankow et al., 1993), and sampling technique (reviewed by Kim and Kim, 2015), confounding direct comparisons between this and other studies.Overall, the general trends observed herein are consistent with prior studies that report PAHs with two-three aromatic rings (C 10 -C 14 ) mainly in the gas phase (F p ≥ 0.93), those with five or more aromatic rings (≥ C 20 ) mainly in the particle phase (F p ≥ 0.9), and those with four aromatic rings (C 16 and C 18 ) partitioned between the two phases depending on their chemical structure and atmospheric conditions (Yamasaki et al., 1982;Williams et al., 2010;Ma et al., 2011;Kim and Kim, 2015).
Gas-phase sampling using QFF without a denuder upfront is subject to artefacts caused by vapour adsorption on the filter, resulting in underestimation of the concentration of the species in the gas phase, particularly for low-MW PAHs (Delgado-Saborit et al., 2014).However, comparing the partitioning trend in this study to those sampled with a denuder during the same season (Possanzini et al., 2004) shows less than 5 % discrepancies for the low (C ≤ 12) and high (C ≥ 18) molecular weight PAHs, while for C 14 and C 16 our F p measurements were lower by 14 and 6 %, respectively.These results show a slight underestimation of F p rather than an overestimation, which would be expected if vapour adsorption on the QFF significantly impacted gas-particle partitioning results.We estimate that the uncertainties associated with our gas-particle partitioning measurements are ≤ 5 % for species predominantly in the particle phase (F p > 0.9) or gas phase (F p < 0.1) and are in the range of 14 % for semivolatile species (0.1 < F p < 0.9).

GC-MS identification
2,3-Dihydroxy-4-oxopentanoic acid (DHOPA, also known as T-3) has been identified as a product of toluene photooxidation by Kleindienst et al. (2004), and their chemical ionization mass spectrum has been used to identify this tracer in other studies.To support identification by the more common EI, the corresponding mass spectrum of its trimethylsilylated (TMS) derivative is given in Fig. 2. The most abundant ions are m/z 73 and 147, corresponding to Si(CH 3 ) 3 + and (CH 3 ) 2 Si = O Si(CH 3 ) 3 + fragments, respectively; however, these are common to the BSTFA-TMCS silylation reagent.Ions at m/z 277, 349, 321, and 364 are unique to DHOPA and are recommended for quantification.Here, the m/z 277 ion was used for quantification due to the high relative abundance and low background, and the other ions were used qualitatively.This mass spectrum obtained from a pure standard builds upon the previous EI mass spectrum for the DHOPA in an aerosol sample by Hu et al. (2008) that included some spectral interferences from adipic acid that coeluted.

Ambient concentration and gas-particle partitioning
The average mass concentration of DHOPA ranged from 0.14 to 0.50 ng m −3 and averaged 0.29 ± 0.12 ng m −3 (Fig. 3a).DHOPA was detected only in the particle phase (Table 2), although the 22.1 ± 13.5 % extraction recoveries of this species from PUF limited the sensitivity of gas-phase measurements.Nonetheless, it is reasonable to conclude that this species does not appreciably partition to the gas phase.The average concentrations of DHOPA in Iowa City were within the range of those observed in Bondville, IL, in autumn (Lewandowski et al., 2008) but were lower by a factor of 45 (on average) than what was detected in the Pearl River Delta, China, for the same season (Ding et al., 2012).Although an authentic standard for DHOPA was not previously available, prior measurements were based upon a surrogate standard response and are subject to bias.
The contribution of toluene SOA to OC was estimated based on the SOA tracer method introduced by Kleindienst et al. (2007), where the DHOPA mass fraction of SOC from toluene was 0.0079 ± 0.0026.Following this estimation method, toluene SOC was estimated to contribute 36.5 ± 15.0 ngC m −3 .The contribution of the estimated SOC to the total OC in this study ranged 0.3-7 % and averaged 2.2 ± 1.6 %.In other studies, the concentration of toluene SOC was variable and influenced by seasonal variations and local emission sources (Kleindienst et al., 2007;Peng et al., 2013).Our estimated SOC levels were less than half of those observed in the rural Midwestern United States previously (0.09 µg m −3 ) during the same season; however the contribution of the estimated toluene SOC to the total OC was 6 % (Lewandowski et al., 2008), which is within the upper end of the range observed in this study.Although toluene SOC concentrations were much higher in the Pearl River Delta (1.65 µg m −3 ) (Ding et al., 2012), its relative contribution to OC (7 %) was comparable.Because toluene is only one of many aromatic VOC precursors to SOA, additional tracers are needed to better evaluate the impact of aromatic VOCs on SOA.

Benzene dicarboxylic acids
Three isomers of benzene dicarboxylic acid and one methyl derivative were detected in all PM samples.The total (gas www.atmos-chem-phys.net/17/2053/2017/plus particle) concentration of phthalic acid (PhA), the most abundant isomer, ranged from 4.9 to 21.5 ng m −3 and averaged 13.0 ± 4.3 ng m −3 , while isophthalic acid (i-PhA), terephthalic acid (t-PhA) and 4-methylphthalic acid (4M-PhA) had increasingly lower concentrations in the range of 0.2 to 6.6 ng m −3 (Fig. 3b).Similar relative abundancies for these species were observed in other studies, with PhA consistently being the predominant isomer (Fraser et al., 2003;Mirivel et al., 2011;Mkoma and Kawamura, 2013).The relatively high ambient mass concentrations of these dicarboxylic acid isomers at levels that allow for consistent detection make them promising candidates for tracing aromatic SOA.
The majority of PhA was estimated to be in the gas phase (F p = 0.26), in contrast to i-PhA and t-PhA (F p = 1) and 4-M-PhA (F p = 0.82, Table 2).Vapour pressure values and partitioning theory (Table S1 in the Supplement) cannot explain the observed lower fraction of PhA in the particle phase compared to i-PhA and t-PhA (F p = 1).Instead, the gasphase measurement of PhA is expected to be positively biased due to interference by phthalic anhydride, which yields identical products to PhA when derivatized, hydrolysed, or exposed to high temperatures (like those encountered in GC analysis).For example, under the conditions employed in this study, phthalic anhydride and PhA have identical GC retention time and silylated MS spectra (Fig. S1 in the Supplement).Phthalic anhydride is a gas-phase product of naphthalene photooxidation (Chan et al., 2009;Kautzman et al., 2010) and has much higher vapour pressure (7.5 × 10 −6 atm) than that of PhA (8.9 × 10 −8 atm) (EPA, 2012).As such, phthalic anhydride will partition to a greater extent to the gas phase, which is supported by the absorption model estimations (F p = 4.9 × 10 −5 ), shown in Table S1.Thus, PhA concentrations reported here and in prior studies that involve the use of GC inlet temperatures ≥ 150 • C, derivatization, or hydrolysis (which is common in liquid chromatography) reflect the sum of PhA and phthalic anhydride.Because phthalic anhydride is primarily in the gas phase, this causes gas-phase PhA concentrations to be overestimated and F p estimates for PhA to be erroneously low.An accurate determination of F p for PhA requires collection and analysis of acid and anhydrides separately by in situ derivatization on veratrylaminecoated glass fibre filters (OSHA, 1991).
Although PhA and 4M-PhA can be emitted directly from primary sources such as motor vehicle engines (Kawamura and Kaplan, 1987), there is a lack of evidence for significant primary source contributions to these species in ambient air.In contrast, naphthalene, a precursor for secondary formation of PhA (Kautzman et al., 2010;Kleindienst et al., 2012), was found to be the most abundant PAH from many combustion sources (Oanh et al., 1999;Al-Naiema et al., 2015).As shown in Table S3, the concentrations of PhA and 4M-PhA in the particle phase are highly and significantly correlated with DHOPA (r s = 0.73, p = 0.003; r s = 0.79, p = 0.001, respectively), but they do not correlate with hopane (r s = 0.19, p = 0.529), a fossil fuel combustion biomarker.These correlation data indicate that the probable origin of these two acids is secondary reactions, rather than primary emissions.Although i-PhA has a strong correlation with DHOPA, it also correlates highly and significantly with biomass burning products (e.g.levoglucosan, 4-nitrocatechol, and 4-methyl-5-nitrocatechol) and moderately with hopane.The possibility of multiple sources of i-PhA limits its application as a tracer for anthropogenic SOA.Terephthalic acid correlates strongly with biomass burning tracers and with hopane, and there is no evidence supporting secondary formation; hence, t-PhA is not a valid SOA tracer candidate.The relatively high concentrations of PhA and 4M-PhA detected in the particle phase and their strong correlations with DHOPA suggest that these compounds are useful SOA tracers for naphthalene and methylnaphthalene photooxidation, respectively.

Analytical method performance
While many techniques for quantifying nitrophenols in various sample matrices have been developed and were reviewed by Harrison et al. (2005), our goal was to quantify these compounds in parallel to other primary and secondary source tracers using GC-MS and single-filter extraction protocol.
Using GC with a DB-5 column, a baseline separation was achieved for 10 nitromonoaromatic analytes as trimethylsilylated esters (Fig. 4), with highly symmetrical and narrow peak shapes.Mass spectrometry was used for identification by comparison of retention times and mass spectra, and quantification was done based on base peak area (Table S2).
Nitromonoaromatic mass spectra (Table S2) included mass fragments with m/z [M-60] + (from the loss of NO 2 and CH 3 ), where M is molecular ion for the trimethylsilylated ester.Save for nitroguiacols and 4-methyl-5-nitrocatechol, nitromonoaromatics' mass spectra included a mass fragment of [M-15] + (loss of CH 3 ).The mass spectra for the coeluting peaks (Fig. 4) indicate that potential interferences for the 4NP-D 4 , 4NP, and 4M-2NP are not appreciably strong (< 1 %), and thus interferences are expected to be negligible.
There is potential for 4M-3NP to interfere with detection of 4M-2NP, because the former shows a relatively strong signal for m/z 210 (at 38 % of the base peak signal) that is used to quantify the latter; however 4M-3NP was not detected in this study, so no interference is expected in this dataset.
The performance of the GC-MS method was evaluated with respect to linearity, detection and quantification limits of target analytes, and extraction efficiency (Table 1).The normalized response for the nitromonoaromatics was linear (R 2 ≥ 0.996) from 10 to 5000 µg L −1 with a constant internal standard concentration of 10 000 µg L −1 .This wide range of linearity indicates the suitability of this method to determine nitromonoaromatics in different applications.The limit of detection ranges from 2.7 to 14.9 µg L −1 for the 10 nitromonoaromatic compounds.These method detection limits are higher than those obtained from liquid chromatography coupled with tandem mass spectrometry (0.1-0.25 µg L −1 ) (Kitanovski et al., 2012) but are sufficient to detect the investigated species in ambient air.Filter extraction recoveries averaged 99.4 ± 3.8 %, demonstrating high accuracy and precision of the filter extraction with acetonitrile and reduction in volume under reduced pressure with rotary evaporation.For PUF extraction, very high recovery (> 97 %) was achieved for most compounds (Table 1); however two nitrocatechols, 4NC and 4M-5NC, had significantly lower (< 50 %) and much more variable (relative standard deviation (RSD): 34-60 %) recoveries.Similarly low recoveries of 4NC and 4M-5NC have been reported previously (Hawthorne et al., 1989), which is attributed to the strong interactions of phenols with the polymeric chains of the PUF.Consequently, gas-phase measurements of 4NC and 4M-5NC are biased low and subject to high uncertainty, such that their levels and gas-particle partitioning are not reported.Otherwise, the extraction and analysis method provides high accuracy and reliable precision for nitromonoaromatics from filters and PUF.

Ambient concentration, gas-particle partitioning, and potential sources
Total concentrations of eight nitromonoaromatics ranged from 0.7 to 17 ng m −3 in the particle phase and from 0.6 to 40 ng m −3 in the gas phase (Figs. 5 and S3).Average concentrations and F p are summarized in Table 2, with daily F p shown in Fig. S2.
A number of nitromonoaromatics were likely derived from biomass burning, as evidenced by correlations with a biomass burning marker (levoglucosan) in this and prior studies.Nitrocatechols were the most abundant particlephase species within this compound class, with average concentrations (±standard deviation) of 1.6 ± 2.9 ng m −3 and 1.6 ± 1.8 ng m −3 for 4NC and 4M-5NC, respectively.These two species have been previously associated with biomass burning in PM 10 , via their correlations with levoglucosan (Iinuma et al., 2010;Kahnt et al., 2013).The strong correlation of these two species with levoglucosan extends to PM 2.5 in Iowa City (Fig. 5) with very high correlations with levoglucosan for 4NC (r s = 0.90, p < 0.001) and 4M-5NC (r s = 0.85, p < 0.001).Although nitrocatechol can be formed from the toluene photooxidation (Lin et al., 2015), 4NC correlates weakly with DHOPA (r s ≤ 0.2) lacking statistical significance (Table S3), suggesting that toluene photooxidation is negligible in relation to biomass burning.Similarly, 5NSA (F p = 0.73) was highly correlated with levoglucosan (r s = 0.76, p = 0.002) but moderately correlated with DHOPA (r s = 0.49, p = 0.078), also suggesting its primary origin to be biomass burning (Kitanovski et al., 2012;Zhang et al., 2013) rather than photooxidation (Jang and Kamens, 2001), in agreement with prior studies.Consequently, these three species are characteristic of biomass burning, rather than anthropogenic SOA.
Nitroguaiacol was detected in low concentrations relative to other nitromonoaromatics.The concentrations of 4NG in PM 2.5 ranged from below the detection limit (BDL) to 0.11 ng m −3 , with a frequency of detection of 86 % (Table 2).Similarly low concentrations were also reported elsewhere (Kitanovski et al., 2012).In the gas phase, 4NG was not detected on most of the days, except for 14-16 November, when gas concentrations reached 0.5-2.1 ng m −3 (Fig. S2).On 14 November, outdoor festivities, barbecues, and slowmoving traffic occurred near the sampling site.The possibility of multiple sources and the low ambient concentrations suggest that this is not a suitable tracer for anthropogenic SOA.
4H-3NB was detected only in the particle phase (F p = 1), with a frequency of 71 % and relatively low concentrations ranging from BDL to 0.2 ng m −3 .4H-3NB was identified as a low-abundance product of toluene photooxidation with hydroxyl radicals (Hamilton et al., 2005;Fang et al., 2011).Other than toluene photooxidation, there are no other known emission sources for 4H-3NB.The specificity of 4H-3NB is supported by the lack of correlation with other biogenic or anthropogenic tracers (Table S3).Because it is detected only in the particle phase and is likely specific to toluene photooxidation, it has potential to be a unique nitromonoaromatic tracer for anthropogenic VOC photooxidation.However, due to the small number of samples and the frequency of detection for this tracer, further investigation is recommended to evaluate its detectability in other environments and source specificity.
to 7.3 ng m −3 and averaging 1.8 ± 2.1 ng m −3 .Likewise, two methyl-nitrophenol isomers (4M-2NP and 2M-4NP) averaged 0.3 ± 1.6 ng m −3 and 5.3 ± 8.5 ng m −3 , respectively (Table 2 and Fig. S3).The higher concentration of 4M-2NP with the higher standard deviation is largely driven by the aforementioned local source influences on 14 November (32.5 ng m −3 ), shown in Fig. S2.These three nitromonoaromatics showed substantial partitioning in the gas phase, with F p ≤ 0.33 (Table 2).The very high correlation of 4NP with 2M-4NP (r s = 0.90, p < 0.001) and with 4M-2NP (r s = 0.81, p < 0.001) indicates a similar source of origin.These three compounds have previously been shown to be products from the photooxidation of the monoaromatic compounds in the presence of NO x (Forstner et al., 1997;Harrison et al., 2005;Sato et al., 2007) as well as components of vehicle emissions (Tremp et al., 1993).However, no significant correlations were observed between these tracers with hopane or DHOPA.Because of their lack of source specificity and the significant partitioning in the gas phase, these three nitrophenols are not recommended for use as tracers of anthropogenic SOA.

Furandiones
Ambient gas and particle concentrations for the sum of four furandiones and their F p are shown in Fig. 6, with individual species data in Fig. S4.The total furandiones concentration detected in the particle phase ranged from 0.3 to 4.3 ng m −3 and averaged 1.6 ± 1.1 ng m −3 .These concentrations were lower than those detected in our previous study (9.3 ± 3.0 ng m −3 ) (Al-Naiema et al., 2017), which is likely due to the rainy and foggy weather in the autumn of 2015.In the presence of water, anhydrides undergo hydration and ring opening to form the carboxylic acid derivatives.The relative rate of hydrolysis for 2,5-furandione (FD) and 3-methyl-2,5-furandione (MFD) are 6 times higher than dihydro-2,5-furandione (DFD) and dihydro-3-methyl-2,5,-  (Trivedi and Culbertson, 1982).The highest stability against water hydrolysis might explain the higher concentration of DFD detected in this study compared to other furandiones.The sum of the gas-phase concentration for the furandiones (DFD, MFD, and DMFD) averaged 18.0 ± 10.7 ng m −3 .Furandiones were almost entirely in the gas phase (Fig. S4).The measured F p values were 0.31 for DFD, 0.08 for MFD, and 0.05 for DMFD, while this value is not reported for FD, which showed poor extraction recovery (< 10 %) from the PUF.Low F p values are expected for furandiones due to their high vapour pressures (Table S1).The measured F p values were substantially higher than those predicted by Pankow's absorption model by 2 orders of magnitude (Tables 2 and S1).It is possible that higher-thanpredicted F p values were driven by furandione adsorption on the front filter or breakthrough from the PUF (Chuang et al., 1987).
Although no sources other than photooxidation of anthropogenic VOCs are known to influence the atmospheric concentration of furandiones (Forstner et al., 1997;Hamilton et al., 2005), only a moderate correlation (r s = 0.50, p = 0.064) was observed between the particle concentrations of furandiones with DHOPA (Table S3).This may be due to the fact that DHOPA is a tracer specific to toluene, while furandiones can also form from other aromatic VOCs (Forstner et al., 1997).Overall, we conclude that furandiones hold a significant importance to serve as indicators for atmospherically processed aromatic VOCs due to their source specificity; however the substantial partitioning toward the gas phase and their water sensitivity limit their application as SOA tracers.

Conclusions
This study evaluates, for the first time, the influence of the source specificity, ambient concentration, and gas-particle partitioning on the efficacy of the use of nitromonoaromatics, benzene dicarboxylic acids, furandiones, and DHOPA as tracers for SOA from anthropogenic VOCs.First and foremost, DHOPA was detected consistently and only in particle phase and is specific to toluene photooxidation, making it a good tracer for toluene SOA despite its relatively low concentrations.Second, PhA is the most abundant benzene dicarboxylic acid isomer and correlates highly with DHOPA.Similarly, 4M-PhA correlates highly with DHOPA.Although the measured F p values suggest partitioning to the gas phase for these two species, this is likely due to instrumental interferences from the corresponding anhydrides.Their particlephase concentration, nonetheless, are expected to be useful in tracing naphthalene-derived SOA.Third, 4H-3NB was detected only in the particle phase and found to be specific to toluene photooxidation at low levels of NO x .Because of their unique sources, detectability, and partitioning towards the particle phase, these species are expected to pro-vide much needed insight to SOA from anthropogenic origins, which can support a better understanding of the sources of atmospheric aerosols.
While the above-described species are proposed as tracers of anthropogenic SOA, structurally similar compounds are largely associated with primary sources and are not suitable tracers of SOA.For example, t-PhA, 4NC, 4M-5NC, and 5NSA were highly correlated with levoglucosan and known to be biomass burning products.Other species such as furandiones hold significant potential to be used as an indicator of processed aromatic VOCs in the atmosphere due to their source specificity but are not recommended as SOA tracers because of their substantial partitioning in the gas phase and water sensitivity.These findings underscore the importance of evaluating and quantifying potential SOA tracers on an individual species level, as some species within a compound class may provide source specificity, while others do not.Given the limited time and geographic distribution for the samples analysed in this study, further investigation is needed to realize the value of these compounds as tracers of anthropogenic SOA more broadly.

Data availability
All ambient measurements used in this paper are available as Supplemental data.
The Supplement related to this article is available online at doi:10.5194/acp-17-2053-2017-supplement.
Competing interests.The authors declare that they have no conflict of interest.

Figure 1 .
Figure 1.Fraction of PAHs in the particle phase as a function of the number of carbon atoms.Circles represent the daily measured F p (n = 14), the red line represents average F p , and the dotted line (black) is the predicted F p using the absorption model (Table S1).Select PAHs were used compare ambient measurements and model estimation: C 10 is naphthalene, C 12 is acenaphthene, C 14 is anthracene, C 16 is pyrene, C 18 is benzo(ghi)fluoranthene, C 20 is benzo(b)fluoranthene, and C 22 is picene.

Table 2 .
Summary table of ambient concentrations of organic species in gas and particle phases, measured fraction in the particle phase (F